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 | |
6352b91d MX |
28 | static struct kmem_cache *btrfs_ordered_extent_cache; |
29 | ||
e6dcd2dc | 30 | static u64 entry_end(struct btrfs_ordered_extent *entry) |
dc17ff8f | 31 | { |
e6dcd2dc CM |
32 | if (entry->file_offset + entry->len < entry->file_offset) |
33 | return (u64)-1; | |
34 | return entry->file_offset + entry->len; | |
dc17ff8f CM |
35 | } |
36 | ||
d352ac68 CM |
37 | /* returns NULL if the insertion worked, or it returns the node it did find |
38 | * in the tree | |
39 | */ | |
e6dcd2dc CM |
40 | static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset, |
41 | struct rb_node *node) | |
dc17ff8f | 42 | { |
d397712b CM |
43 | struct rb_node **p = &root->rb_node; |
44 | struct rb_node *parent = NULL; | |
e6dcd2dc | 45 | struct btrfs_ordered_extent *entry; |
dc17ff8f | 46 | |
d397712b | 47 | while (*p) { |
dc17ff8f | 48 | parent = *p; |
e6dcd2dc | 49 | entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node); |
dc17ff8f | 50 | |
e6dcd2dc | 51 | if (file_offset < entry->file_offset) |
dc17ff8f | 52 | p = &(*p)->rb_left; |
e6dcd2dc | 53 | else if (file_offset >= entry_end(entry)) |
dc17ff8f CM |
54 | p = &(*p)->rb_right; |
55 | else | |
56 | return parent; | |
57 | } | |
58 | ||
59 | rb_link_node(node, parent, p); | |
60 | rb_insert_color(node, root); | |
61 | return NULL; | |
62 | } | |
63 | ||
43c04fb1 JM |
64 | static void ordered_data_tree_panic(struct inode *inode, int errno, |
65 | u64 offset) | |
66 | { | |
67 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
68 | btrfs_panic(fs_info, errno, "Inconsistency in ordered tree at offset " | |
69 | "%llu\n", (unsigned long long)offset); | |
70 | } | |
71 | ||
d352ac68 CM |
72 | /* |
73 | * look for a given offset in the tree, and if it can't be found return the | |
74 | * first lesser offset | |
75 | */ | |
e6dcd2dc CM |
76 | static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset, |
77 | struct rb_node **prev_ret) | |
dc17ff8f | 78 | { |
d397712b | 79 | struct rb_node *n = root->rb_node; |
dc17ff8f | 80 | struct rb_node *prev = NULL; |
e6dcd2dc CM |
81 | struct rb_node *test; |
82 | struct btrfs_ordered_extent *entry; | |
83 | struct btrfs_ordered_extent *prev_entry = NULL; | |
dc17ff8f | 84 | |
d397712b | 85 | while (n) { |
e6dcd2dc | 86 | entry = rb_entry(n, struct btrfs_ordered_extent, rb_node); |
dc17ff8f CM |
87 | prev = n; |
88 | prev_entry = entry; | |
dc17ff8f | 89 | |
e6dcd2dc | 90 | if (file_offset < entry->file_offset) |
dc17ff8f | 91 | n = n->rb_left; |
e6dcd2dc | 92 | else if (file_offset >= entry_end(entry)) |
dc17ff8f CM |
93 | n = n->rb_right; |
94 | else | |
95 | return n; | |
96 | } | |
97 | if (!prev_ret) | |
98 | return NULL; | |
99 | ||
d397712b | 100 | while (prev && file_offset >= entry_end(prev_entry)) { |
e6dcd2dc CM |
101 | test = rb_next(prev); |
102 | if (!test) | |
103 | break; | |
104 | prev_entry = rb_entry(test, struct btrfs_ordered_extent, | |
105 | rb_node); | |
106 | if (file_offset < entry_end(prev_entry)) | |
107 | break; | |
108 | ||
109 | prev = test; | |
110 | } | |
111 | if (prev) | |
112 | prev_entry = rb_entry(prev, struct btrfs_ordered_extent, | |
113 | rb_node); | |
d397712b | 114 | while (prev && file_offset < entry_end(prev_entry)) { |
e6dcd2dc CM |
115 | test = rb_prev(prev); |
116 | if (!test) | |
117 | break; | |
118 | prev_entry = rb_entry(test, struct btrfs_ordered_extent, | |
119 | rb_node); | |
120 | prev = test; | |
dc17ff8f CM |
121 | } |
122 | *prev_ret = prev; | |
123 | return NULL; | |
124 | } | |
125 | ||
d352ac68 CM |
126 | /* |
127 | * helper to check if a given offset is inside a given entry | |
128 | */ | |
e6dcd2dc CM |
129 | static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset) |
130 | { | |
131 | if (file_offset < entry->file_offset || | |
132 | entry->file_offset + entry->len <= file_offset) | |
133 | return 0; | |
134 | return 1; | |
135 | } | |
136 | ||
4b46fce2 JB |
137 | static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset, |
138 | u64 len) | |
139 | { | |
140 | if (file_offset + len <= entry->file_offset || | |
141 | entry->file_offset + entry->len <= file_offset) | |
142 | return 0; | |
143 | return 1; | |
144 | } | |
145 | ||
d352ac68 CM |
146 | /* |
147 | * look find the first ordered struct that has this offset, otherwise | |
148 | * the first one less than this offset | |
149 | */ | |
e6dcd2dc CM |
150 | static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree, |
151 | u64 file_offset) | |
dc17ff8f | 152 | { |
e6dcd2dc | 153 | struct rb_root *root = &tree->tree; |
c87fb6fd | 154 | struct rb_node *prev = NULL; |
dc17ff8f | 155 | struct rb_node *ret; |
e6dcd2dc CM |
156 | struct btrfs_ordered_extent *entry; |
157 | ||
158 | if (tree->last) { | |
159 | entry = rb_entry(tree->last, struct btrfs_ordered_extent, | |
160 | rb_node); | |
161 | if (offset_in_entry(entry, file_offset)) | |
162 | return tree->last; | |
163 | } | |
164 | ret = __tree_search(root, file_offset, &prev); | |
dc17ff8f | 165 | if (!ret) |
e6dcd2dc CM |
166 | ret = prev; |
167 | if (ret) | |
168 | tree->last = ret; | |
dc17ff8f CM |
169 | return ret; |
170 | } | |
171 | ||
eb84ae03 CM |
172 | /* allocate and add a new ordered_extent into the per-inode tree. |
173 | * file_offset is the logical offset in the file | |
174 | * | |
175 | * start is the disk block number of an extent already reserved in the | |
176 | * extent allocation tree | |
177 | * | |
178 | * len is the length of the extent | |
179 | * | |
eb84ae03 CM |
180 | * The tree is given a single reference on the ordered extent that was |
181 | * inserted. | |
182 | */ | |
4b46fce2 JB |
183 | static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset, |
184 | u64 start, u64 len, u64 disk_len, | |
261507a0 | 185 | int type, int dio, int compress_type) |
dc17ff8f | 186 | { |
dc17ff8f | 187 | struct btrfs_ordered_inode_tree *tree; |
e6dcd2dc CM |
188 | struct rb_node *node; |
189 | struct btrfs_ordered_extent *entry; | |
dc17ff8f | 190 | |
e6dcd2dc | 191 | tree = &BTRFS_I(inode)->ordered_tree; |
6352b91d | 192 | entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS); |
dc17ff8f CM |
193 | if (!entry) |
194 | return -ENOMEM; | |
195 | ||
e6dcd2dc CM |
196 | entry->file_offset = file_offset; |
197 | entry->start = start; | |
198 | entry->len = len; | |
2ab28f32 JB |
199 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) && |
200 | !(type == BTRFS_ORDERED_NOCOW)) | |
201 | entry->csum_bytes_left = disk_len; | |
c8b97818 | 202 | entry->disk_len = disk_len; |
8b62b72b | 203 | entry->bytes_left = len; |
5fd02043 | 204 | entry->inode = igrab(inode); |
261507a0 | 205 | entry->compress_type = compress_type; |
d899e052 | 206 | if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE) |
80ff3856 | 207 | set_bit(type, &entry->flags); |
3eaa2885 | 208 | |
4b46fce2 JB |
209 | if (dio) |
210 | set_bit(BTRFS_ORDERED_DIRECT, &entry->flags); | |
211 | ||
e6dcd2dc CM |
212 | /* one ref for the tree */ |
213 | atomic_set(&entry->refs, 1); | |
214 | init_waitqueue_head(&entry->wait); | |
215 | INIT_LIST_HEAD(&entry->list); | |
3eaa2885 | 216 | INIT_LIST_HEAD(&entry->root_extent_list); |
9afab882 MX |
217 | INIT_LIST_HEAD(&entry->work_list); |
218 | init_completion(&entry->completion); | |
2ab28f32 | 219 | INIT_LIST_HEAD(&entry->log_list); |
dc17ff8f | 220 | |
1abe9b8a | 221 | trace_btrfs_ordered_extent_add(inode, entry); |
222 | ||
5fd02043 | 223 | spin_lock_irq(&tree->lock); |
e6dcd2dc CM |
224 | node = tree_insert(&tree->tree, file_offset, |
225 | &entry->rb_node); | |
43c04fb1 JM |
226 | if (node) |
227 | ordered_data_tree_panic(inode, -EEXIST, file_offset); | |
5fd02043 | 228 | spin_unlock_irq(&tree->lock); |
d397712b | 229 | |
3eaa2885 CM |
230 | spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); |
231 | list_add_tail(&entry->root_extent_list, | |
232 | &BTRFS_I(inode)->root->fs_info->ordered_extents); | |
233 | spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); | |
234 | ||
dc17ff8f CM |
235 | return 0; |
236 | } | |
237 | ||
4b46fce2 JB |
238 | int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset, |
239 | u64 start, u64 len, u64 disk_len, int type) | |
240 | { | |
241 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, | |
261507a0 LZ |
242 | disk_len, type, 0, |
243 | BTRFS_COMPRESS_NONE); | |
4b46fce2 JB |
244 | } |
245 | ||
246 | int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset, | |
247 | u64 start, u64 len, u64 disk_len, int type) | |
248 | { | |
249 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, | |
261507a0 LZ |
250 | disk_len, type, 1, |
251 | BTRFS_COMPRESS_NONE); | |
252 | } | |
253 | ||
254 | int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset, | |
255 | u64 start, u64 len, u64 disk_len, | |
256 | int type, int compress_type) | |
257 | { | |
258 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, | |
259 | disk_len, type, 0, | |
260 | compress_type); | |
4b46fce2 JB |
261 | } |
262 | ||
eb84ae03 CM |
263 | /* |
264 | * Add a struct btrfs_ordered_sum into the list of checksums to be inserted | |
3edf7d33 CM |
265 | * when an ordered extent is finished. If the list covers more than one |
266 | * ordered extent, it is split across multiples. | |
eb84ae03 | 267 | */ |
143bede5 JM |
268 | void btrfs_add_ordered_sum(struct inode *inode, |
269 | struct btrfs_ordered_extent *entry, | |
270 | struct btrfs_ordered_sum *sum) | |
dc17ff8f | 271 | { |
e6dcd2dc | 272 | struct btrfs_ordered_inode_tree *tree; |
dc17ff8f | 273 | |
e6dcd2dc | 274 | tree = &BTRFS_I(inode)->ordered_tree; |
5fd02043 | 275 | spin_lock_irq(&tree->lock); |
e6dcd2dc | 276 | list_add_tail(&sum->list, &entry->list); |
2ab28f32 JB |
277 | WARN_ON(entry->csum_bytes_left < sum->len); |
278 | entry->csum_bytes_left -= sum->len; | |
279 | if (entry->csum_bytes_left == 0) | |
280 | wake_up(&entry->wait); | |
5fd02043 | 281 | spin_unlock_irq(&tree->lock); |
dc17ff8f CM |
282 | } |
283 | ||
163cf09c CM |
284 | /* |
285 | * this is used to account for finished IO across a given range | |
286 | * of the file. The IO may span ordered extents. If | |
287 | * a given ordered_extent is completely done, 1 is returned, otherwise | |
288 | * 0. | |
289 | * | |
290 | * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used | |
291 | * to make sure this function only returns 1 once for a given ordered extent. | |
292 | * | |
293 | * file_offset is updated to one byte past the range that is recorded as | |
294 | * complete. This allows you to walk forward in the file. | |
295 | */ | |
296 | int btrfs_dec_test_first_ordered_pending(struct inode *inode, | |
297 | struct btrfs_ordered_extent **cached, | |
5fd02043 | 298 | u64 *file_offset, u64 io_size, int uptodate) |
163cf09c CM |
299 | { |
300 | struct btrfs_ordered_inode_tree *tree; | |
301 | struct rb_node *node; | |
302 | struct btrfs_ordered_extent *entry = NULL; | |
303 | int ret; | |
5fd02043 | 304 | unsigned long flags; |
163cf09c CM |
305 | u64 dec_end; |
306 | u64 dec_start; | |
307 | u64 to_dec; | |
308 | ||
309 | tree = &BTRFS_I(inode)->ordered_tree; | |
5fd02043 | 310 | spin_lock_irqsave(&tree->lock, flags); |
163cf09c CM |
311 | node = tree_search(tree, *file_offset); |
312 | if (!node) { | |
313 | ret = 1; | |
314 | goto out; | |
315 | } | |
316 | ||
317 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
318 | if (!offset_in_entry(entry, *file_offset)) { | |
319 | ret = 1; | |
320 | goto out; | |
321 | } | |
322 | ||
323 | dec_start = max(*file_offset, entry->file_offset); | |
324 | dec_end = min(*file_offset + io_size, entry->file_offset + | |
325 | entry->len); | |
326 | *file_offset = dec_end; | |
327 | if (dec_start > dec_end) { | |
328 | printk(KERN_CRIT "bad ordering dec_start %llu end %llu\n", | |
329 | (unsigned long long)dec_start, | |
330 | (unsigned long long)dec_end); | |
331 | } | |
332 | to_dec = dec_end - dec_start; | |
333 | if (to_dec > entry->bytes_left) { | |
334 | printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n", | |
335 | (unsigned long long)entry->bytes_left, | |
336 | (unsigned long long)to_dec); | |
337 | } | |
338 | entry->bytes_left -= to_dec; | |
5fd02043 JB |
339 | if (!uptodate) |
340 | set_bit(BTRFS_ORDERED_IOERR, &entry->flags); | |
341 | ||
163cf09c CM |
342 | if (entry->bytes_left == 0) |
343 | ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); | |
344 | else | |
345 | ret = 1; | |
346 | out: | |
347 | if (!ret && cached && entry) { | |
348 | *cached = entry; | |
349 | atomic_inc(&entry->refs); | |
350 | } | |
5fd02043 | 351 | spin_unlock_irqrestore(&tree->lock, flags); |
163cf09c CM |
352 | return ret == 0; |
353 | } | |
354 | ||
eb84ae03 CM |
355 | /* |
356 | * this is used to account for finished IO across a given range | |
357 | * of the file. The IO should not span ordered extents. If | |
358 | * a given ordered_extent is completely done, 1 is returned, otherwise | |
359 | * 0. | |
360 | * | |
361 | * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used | |
362 | * to make sure this function only returns 1 once for a given ordered extent. | |
363 | */ | |
e6dcd2dc | 364 | int btrfs_dec_test_ordered_pending(struct inode *inode, |
5a1a3df1 | 365 | struct btrfs_ordered_extent **cached, |
5fd02043 | 366 | u64 file_offset, u64 io_size, int uptodate) |
dc17ff8f | 367 | { |
e6dcd2dc | 368 | struct btrfs_ordered_inode_tree *tree; |
dc17ff8f | 369 | struct rb_node *node; |
5a1a3df1 | 370 | struct btrfs_ordered_extent *entry = NULL; |
5fd02043 | 371 | unsigned long flags; |
e6dcd2dc CM |
372 | int ret; |
373 | ||
374 | tree = &BTRFS_I(inode)->ordered_tree; | |
5fd02043 JB |
375 | spin_lock_irqsave(&tree->lock, flags); |
376 | if (cached && *cached) { | |
377 | entry = *cached; | |
378 | goto have_entry; | |
379 | } | |
380 | ||
e6dcd2dc | 381 | node = tree_search(tree, file_offset); |
dc17ff8f | 382 | if (!node) { |
e6dcd2dc CM |
383 | ret = 1; |
384 | goto out; | |
dc17ff8f CM |
385 | } |
386 | ||
e6dcd2dc | 387 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
5fd02043 | 388 | have_entry: |
e6dcd2dc CM |
389 | if (!offset_in_entry(entry, file_offset)) { |
390 | ret = 1; | |
391 | goto out; | |
dc17ff8f | 392 | } |
e6dcd2dc | 393 | |
8b62b72b CM |
394 | if (io_size > entry->bytes_left) { |
395 | printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n", | |
396 | (unsigned long long)entry->bytes_left, | |
397 | (unsigned long long)io_size); | |
398 | } | |
399 | entry->bytes_left -= io_size; | |
5fd02043 JB |
400 | if (!uptodate) |
401 | set_bit(BTRFS_ORDERED_IOERR, &entry->flags); | |
402 | ||
8b62b72b | 403 | if (entry->bytes_left == 0) |
e6dcd2dc | 404 | ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); |
8b62b72b CM |
405 | else |
406 | ret = 1; | |
e6dcd2dc | 407 | out: |
5a1a3df1 JB |
408 | if (!ret && cached && entry) { |
409 | *cached = entry; | |
410 | atomic_inc(&entry->refs); | |
411 | } | |
5fd02043 | 412 | spin_unlock_irqrestore(&tree->lock, flags); |
e6dcd2dc CM |
413 | return ret == 0; |
414 | } | |
dc17ff8f | 415 | |
2ab28f32 JB |
416 | /* Needs to either be called under a log transaction or the log_mutex */ |
417 | void btrfs_get_logged_extents(struct btrfs_root *log, struct inode *inode) | |
418 | { | |
419 | struct btrfs_ordered_inode_tree *tree; | |
420 | struct btrfs_ordered_extent *ordered; | |
421 | struct rb_node *n; | |
422 | int index = log->log_transid % 2; | |
423 | ||
424 | tree = &BTRFS_I(inode)->ordered_tree; | |
425 | spin_lock_irq(&tree->lock); | |
426 | for (n = rb_first(&tree->tree); n; n = rb_next(n)) { | |
427 | ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node); | |
428 | spin_lock(&log->log_extents_lock[index]); | |
429 | if (list_empty(&ordered->log_list)) { | |
430 | list_add_tail(&ordered->log_list, &log->logged_list[index]); | |
431 | atomic_inc(&ordered->refs); | |
432 | } | |
433 | spin_unlock(&log->log_extents_lock[index]); | |
434 | } | |
435 | spin_unlock_irq(&tree->lock); | |
436 | } | |
437 | ||
438 | void btrfs_wait_logged_extents(struct btrfs_root *log, u64 transid) | |
439 | { | |
440 | struct btrfs_ordered_extent *ordered; | |
441 | int index = transid % 2; | |
442 | ||
443 | spin_lock_irq(&log->log_extents_lock[index]); | |
444 | while (!list_empty(&log->logged_list[index])) { | |
445 | ordered = list_first_entry(&log->logged_list[index], | |
446 | struct btrfs_ordered_extent, | |
447 | log_list); | |
448 | list_del_init(&ordered->log_list); | |
449 | spin_unlock_irq(&log->log_extents_lock[index]); | |
450 | wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE, | |
451 | &ordered->flags)); | |
452 | btrfs_put_ordered_extent(ordered); | |
453 | spin_lock_irq(&log->log_extents_lock[index]); | |
454 | } | |
455 | spin_unlock_irq(&log->log_extents_lock[index]); | |
456 | } | |
457 | ||
458 | void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid) | |
459 | { | |
460 | struct btrfs_ordered_extent *ordered; | |
461 | int index = transid % 2; | |
462 | ||
463 | spin_lock_irq(&log->log_extents_lock[index]); | |
464 | while (!list_empty(&log->logged_list[index])) { | |
465 | ordered = list_first_entry(&log->logged_list[index], | |
466 | struct btrfs_ordered_extent, | |
467 | log_list); | |
468 | list_del_init(&ordered->log_list); | |
469 | spin_unlock_irq(&log->log_extents_lock[index]); | |
470 | btrfs_put_ordered_extent(ordered); | |
471 | spin_lock_irq(&log->log_extents_lock[index]); | |
472 | } | |
473 | spin_unlock_irq(&log->log_extents_lock[index]); | |
474 | } | |
475 | ||
eb84ae03 CM |
476 | /* |
477 | * used to drop a reference on an ordered extent. This will free | |
478 | * the extent if the last reference is dropped | |
479 | */ | |
143bede5 | 480 | void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry) |
e6dcd2dc | 481 | { |
ba1da2f4 CM |
482 | struct list_head *cur; |
483 | struct btrfs_ordered_sum *sum; | |
484 | ||
1abe9b8a | 485 | trace_btrfs_ordered_extent_put(entry->inode, entry); |
486 | ||
ba1da2f4 | 487 | if (atomic_dec_and_test(&entry->refs)) { |
5fd02043 JB |
488 | if (entry->inode) |
489 | btrfs_add_delayed_iput(entry->inode); | |
d397712b | 490 | while (!list_empty(&entry->list)) { |
ba1da2f4 CM |
491 | cur = entry->list.next; |
492 | sum = list_entry(cur, struct btrfs_ordered_sum, list); | |
493 | list_del(&sum->list); | |
494 | kfree(sum); | |
495 | } | |
6352b91d | 496 | kmem_cache_free(btrfs_ordered_extent_cache, entry); |
ba1da2f4 | 497 | } |
dc17ff8f | 498 | } |
cee36a03 | 499 | |
eb84ae03 CM |
500 | /* |
501 | * remove an ordered extent from the tree. No references are dropped | |
5fd02043 | 502 | * and waiters are woken up. |
eb84ae03 | 503 | */ |
5fd02043 JB |
504 | void btrfs_remove_ordered_extent(struct inode *inode, |
505 | struct btrfs_ordered_extent *entry) | |
cee36a03 | 506 | { |
e6dcd2dc | 507 | struct btrfs_ordered_inode_tree *tree; |
287a0ab9 | 508 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cee36a03 | 509 | struct rb_node *node; |
cee36a03 | 510 | |
e6dcd2dc | 511 | tree = &BTRFS_I(inode)->ordered_tree; |
5fd02043 | 512 | spin_lock_irq(&tree->lock); |
e6dcd2dc | 513 | node = &entry->rb_node; |
cee36a03 | 514 | rb_erase(node, &tree->tree); |
e6dcd2dc CM |
515 | tree->last = NULL; |
516 | set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags); | |
5fd02043 | 517 | spin_unlock_irq(&tree->lock); |
3eaa2885 | 518 | |
287a0ab9 | 519 | spin_lock(&root->fs_info->ordered_extent_lock); |
3eaa2885 | 520 | list_del_init(&entry->root_extent_list); |
5a3f23d5 | 521 | |
1abe9b8a | 522 | trace_btrfs_ordered_extent_remove(inode, entry); |
523 | ||
5a3f23d5 CM |
524 | /* |
525 | * we have no more ordered extents for this inode and | |
526 | * no dirty pages. We can safely remove it from the | |
527 | * list of ordered extents | |
528 | */ | |
529 | if (RB_EMPTY_ROOT(&tree->tree) && | |
530 | !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { | |
531 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
532 | } | |
287a0ab9 | 533 | spin_unlock(&root->fs_info->ordered_extent_lock); |
e6dcd2dc | 534 | wake_up(&entry->wait); |
cee36a03 CM |
535 | } |
536 | ||
9afab882 MX |
537 | static void btrfs_run_ordered_extent_work(struct btrfs_work *work) |
538 | { | |
539 | struct btrfs_ordered_extent *ordered; | |
540 | ||
541 | ordered = container_of(work, struct btrfs_ordered_extent, flush_work); | |
542 | btrfs_start_ordered_extent(ordered->inode, ordered, 1); | |
543 | complete(&ordered->completion); | |
544 | } | |
545 | ||
d352ac68 CM |
546 | /* |
547 | * wait for all the ordered extents in a root. This is done when balancing | |
548 | * space between drives. | |
549 | */ | |
6bbe3a9c | 550 | void btrfs_wait_ordered_extents(struct btrfs_root *root, int delay_iput) |
3eaa2885 | 551 | { |
9afab882 | 552 | struct list_head splice, works; |
3eaa2885 | 553 | struct list_head *cur; |
9afab882 | 554 | struct btrfs_ordered_extent *ordered, *next; |
3eaa2885 CM |
555 | struct inode *inode; |
556 | ||
557 | INIT_LIST_HEAD(&splice); | |
9afab882 | 558 | INIT_LIST_HEAD(&works); |
3eaa2885 CM |
559 | |
560 | spin_lock(&root->fs_info->ordered_extent_lock); | |
561 | list_splice_init(&root->fs_info->ordered_extents, &splice); | |
5b21f2ed | 562 | while (!list_empty(&splice)) { |
3eaa2885 CM |
563 | cur = splice.next; |
564 | ordered = list_entry(cur, struct btrfs_ordered_extent, | |
565 | root_extent_list); | |
566 | list_del_init(&ordered->root_extent_list); | |
567 | atomic_inc(&ordered->refs); | |
3eaa2885 CM |
568 | |
569 | /* | |
5b21f2ed | 570 | * the inode may be getting freed (in sys_unlink path). |
3eaa2885 | 571 | */ |
5b21f2ed ZY |
572 | inode = igrab(ordered->inode); |
573 | ||
3eaa2885 CM |
574 | spin_unlock(&root->fs_info->ordered_extent_lock); |
575 | ||
5b21f2ed | 576 | if (inode) { |
9afab882 MX |
577 | ordered->flush_work.func = btrfs_run_ordered_extent_work; |
578 | list_add_tail(&ordered->work_list, &works); | |
579 | btrfs_queue_worker(&root->fs_info->flush_workers, | |
580 | &ordered->flush_work); | |
5b21f2ed ZY |
581 | } else { |
582 | btrfs_put_ordered_extent(ordered); | |
583 | } | |
3eaa2885 | 584 | |
9afab882 | 585 | cond_resched(); |
3eaa2885 CM |
586 | spin_lock(&root->fs_info->ordered_extent_lock); |
587 | } | |
588 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
9afab882 MX |
589 | |
590 | list_for_each_entry_safe(ordered, next, &works, work_list) { | |
591 | list_del_init(&ordered->work_list); | |
592 | wait_for_completion(&ordered->completion); | |
593 | ||
594 | inode = ordered->inode; | |
595 | btrfs_put_ordered_extent(ordered); | |
596 | if (delay_iput) | |
597 | btrfs_add_delayed_iput(inode); | |
598 | else | |
599 | iput(inode); | |
600 | ||
601 | cond_resched(); | |
602 | } | |
3eaa2885 CM |
603 | } |
604 | ||
5a3f23d5 CM |
605 | /* |
606 | * this is used during transaction commit to write all the inodes | |
607 | * added to the ordered operation list. These files must be fully on | |
608 | * disk before the transaction commits. | |
609 | * | |
610 | * we have two modes here, one is to just start the IO via filemap_flush | |
611 | * and the other is to wait for all the io. When we wait, we have an | |
612 | * extra check to make sure the ordered operation list really is empty | |
613 | * before we return | |
614 | */ | |
569e0f35 JB |
615 | int btrfs_run_ordered_operations(struct btrfs_trans_handle *trans, |
616 | struct btrfs_root *root, int wait) | |
5a3f23d5 CM |
617 | { |
618 | struct btrfs_inode *btrfs_inode; | |
619 | struct inode *inode; | |
569e0f35 | 620 | struct btrfs_transaction *cur_trans = trans->transaction; |
5a3f23d5 | 621 | struct list_head splice; |
25287e0a MX |
622 | struct list_head works; |
623 | struct btrfs_delalloc_work *work, *next; | |
624 | int ret = 0; | |
5a3f23d5 CM |
625 | |
626 | INIT_LIST_HEAD(&splice); | |
25287e0a | 627 | INIT_LIST_HEAD(&works); |
5a3f23d5 CM |
628 | |
629 | mutex_lock(&root->fs_info->ordered_operations_mutex); | |
630 | spin_lock(&root->fs_info->ordered_extent_lock); | |
569e0f35 | 631 | list_splice_init(&cur_trans->ordered_operations, &splice); |
5a3f23d5 CM |
632 | while (!list_empty(&splice)) { |
633 | btrfs_inode = list_entry(splice.next, struct btrfs_inode, | |
634 | ordered_operations); | |
5a3f23d5 CM |
635 | inode = &btrfs_inode->vfs_inode; |
636 | ||
637 | list_del_init(&btrfs_inode->ordered_operations); | |
638 | ||
639 | /* | |
640 | * the inode may be getting freed (in sys_unlink path). | |
641 | */ | |
642 | inode = igrab(inode); | |
25287e0a MX |
643 | if (!inode) |
644 | continue; | |
5b947f1b MX |
645 | |
646 | if (!wait) | |
647 | list_add_tail(&BTRFS_I(inode)->ordered_operations, | |
569e0f35 | 648 | &cur_trans->ordered_operations); |
5a3f23d5 CM |
649 | spin_unlock(&root->fs_info->ordered_extent_lock); |
650 | ||
25287e0a MX |
651 | work = btrfs_alloc_delalloc_work(inode, wait, 1); |
652 | if (!work) { | |
5b947f1b | 653 | spin_lock(&root->fs_info->ordered_extent_lock); |
25287e0a MX |
654 | if (list_empty(&BTRFS_I(inode)->ordered_operations)) |
655 | list_add_tail(&btrfs_inode->ordered_operations, | |
656 | &splice); | |
25287e0a | 657 | list_splice_tail(&splice, |
569e0f35 | 658 | &cur_trans->ordered_operations); |
25287e0a MX |
659 | spin_unlock(&root->fs_info->ordered_extent_lock); |
660 | ret = -ENOMEM; | |
661 | goto out; | |
5a3f23d5 | 662 | } |
25287e0a MX |
663 | list_add_tail(&work->list, &works); |
664 | btrfs_queue_worker(&root->fs_info->flush_workers, | |
665 | &work->work); | |
5a3f23d5 CM |
666 | |
667 | cond_resched(); | |
668 | spin_lock(&root->fs_info->ordered_extent_lock); | |
669 | } | |
5a3f23d5 | 670 | spin_unlock(&root->fs_info->ordered_extent_lock); |
25287e0a MX |
671 | out: |
672 | list_for_each_entry_safe(work, next, &works, list) { | |
673 | list_del_init(&work->list); | |
674 | btrfs_wait_and_free_delalloc_work(work); | |
675 | } | |
5a3f23d5 | 676 | mutex_unlock(&root->fs_info->ordered_operations_mutex); |
25287e0a | 677 | return ret; |
5a3f23d5 CM |
678 | } |
679 | ||
eb84ae03 CM |
680 | /* |
681 | * Used to start IO or wait for a given ordered extent to finish. | |
682 | * | |
683 | * If wait is one, this effectively waits on page writeback for all the pages | |
684 | * in the extent, and it waits on the io completion code to insert | |
685 | * metadata into the btree corresponding to the extent | |
686 | */ | |
687 | void btrfs_start_ordered_extent(struct inode *inode, | |
688 | struct btrfs_ordered_extent *entry, | |
689 | int wait) | |
e6dcd2dc CM |
690 | { |
691 | u64 start = entry->file_offset; | |
692 | u64 end = start + entry->len - 1; | |
e1b81e67 | 693 | |
1abe9b8a | 694 | trace_btrfs_ordered_extent_start(inode, entry); |
695 | ||
eb84ae03 CM |
696 | /* |
697 | * pages in the range can be dirty, clean or writeback. We | |
698 | * start IO on any dirty ones so the wait doesn't stall waiting | |
b2570314 | 699 | * for the flusher thread to find them |
eb84ae03 | 700 | */ |
4b46fce2 JB |
701 | if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags)) |
702 | filemap_fdatawrite_range(inode->i_mapping, start, end); | |
c8b97818 | 703 | if (wait) { |
e6dcd2dc CM |
704 | wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE, |
705 | &entry->flags)); | |
c8b97818 | 706 | } |
e6dcd2dc | 707 | } |
cee36a03 | 708 | |
eb84ae03 CM |
709 | /* |
710 | * Used to wait on ordered extents across a large range of bytes. | |
711 | */ | |
143bede5 | 712 | void btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len) |
e6dcd2dc CM |
713 | { |
714 | u64 end; | |
e5a2217e | 715 | u64 orig_end; |
e6dcd2dc | 716 | struct btrfs_ordered_extent *ordered; |
e5a2217e CM |
717 | |
718 | if (start + len < start) { | |
f421950f | 719 | orig_end = INT_LIMIT(loff_t); |
e5a2217e CM |
720 | } else { |
721 | orig_end = start + len - 1; | |
f421950f CM |
722 | if (orig_end > INT_LIMIT(loff_t)) |
723 | orig_end = INT_LIMIT(loff_t); | |
e5a2217e | 724 | } |
551ebb2d | 725 | |
e5a2217e CM |
726 | /* start IO across the range first to instantiate any delalloc |
727 | * extents | |
728 | */ | |
7ddf5a42 JB |
729 | filemap_fdatawrite_range(inode->i_mapping, start, orig_end); |
730 | ||
731 | /* | |
732 | * So with compression we will find and lock a dirty page and clear the | |
733 | * first one as dirty, setup an async extent, and immediately return | |
734 | * with the entire range locked but with nobody actually marked with | |
735 | * writeback. So we can't just filemap_write_and_wait_range() and | |
736 | * expect it to work since it will just kick off a thread to do the | |
737 | * actual work. So we need to call filemap_fdatawrite_range _again_ | |
738 | * since it will wait on the page lock, which won't be unlocked until | |
739 | * after the pages have been marked as writeback and so we're good to go | |
740 | * from there. We have to do this otherwise we'll miss the ordered | |
741 | * extents and that results in badness. Please Josef, do not think you | |
742 | * know better and pull this out at some point in the future, it is | |
743 | * right and you are wrong. | |
744 | */ | |
745 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
746 | &BTRFS_I(inode)->runtime_flags)) | |
747 | filemap_fdatawrite_range(inode->i_mapping, start, orig_end); | |
748 | ||
749 | filemap_fdatawait_range(inode->i_mapping, start, orig_end); | |
e5a2217e | 750 | |
f421950f | 751 | end = orig_end; |
d397712b | 752 | while (1) { |
e6dcd2dc | 753 | ordered = btrfs_lookup_first_ordered_extent(inode, end); |
d397712b | 754 | if (!ordered) |
e6dcd2dc | 755 | break; |
e5a2217e | 756 | if (ordered->file_offset > orig_end) { |
e6dcd2dc CM |
757 | btrfs_put_ordered_extent(ordered); |
758 | break; | |
759 | } | |
760 | if (ordered->file_offset + ordered->len < start) { | |
761 | btrfs_put_ordered_extent(ordered); | |
762 | break; | |
763 | } | |
e5a2217e | 764 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
765 | end = ordered->file_offset; |
766 | btrfs_put_ordered_extent(ordered); | |
e5a2217e | 767 | if (end == 0 || end == start) |
e6dcd2dc CM |
768 | break; |
769 | end--; | |
770 | } | |
cee36a03 CM |
771 | } |
772 | ||
eb84ae03 CM |
773 | /* |
774 | * find an ordered extent corresponding to file_offset. return NULL if | |
775 | * nothing is found, otherwise take a reference on the extent and return it | |
776 | */ | |
e6dcd2dc CM |
777 | struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode, |
778 | u64 file_offset) | |
779 | { | |
780 | struct btrfs_ordered_inode_tree *tree; | |
781 | struct rb_node *node; | |
782 | struct btrfs_ordered_extent *entry = NULL; | |
783 | ||
784 | tree = &BTRFS_I(inode)->ordered_tree; | |
5fd02043 | 785 | spin_lock_irq(&tree->lock); |
e6dcd2dc CM |
786 | node = tree_search(tree, file_offset); |
787 | if (!node) | |
788 | goto out; | |
789 | ||
790 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
791 | if (!offset_in_entry(entry, file_offset)) | |
792 | entry = NULL; | |
793 | if (entry) | |
794 | atomic_inc(&entry->refs); | |
795 | out: | |
5fd02043 | 796 | spin_unlock_irq(&tree->lock); |
e6dcd2dc CM |
797 | return entry; |
798 | } | |
799 | ||
4b46fce2 JB |
800 | /* Since the DIO code tries to lock a wide area we need to look for any ordered |
801 | * extents that exist in the range, rather than just the start of the range. | |
802 | */ | |
803 | struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode, | |
804 | u64 file_offset, | |
805 | u64 len) | |
806 | { | |
807 | struct btrfs_ordered_inode_tree *tree; | |
808 | struct rb_node *node; | |
809 | struct btrfs_ordered_extent *entry = NULL; | |
810 | ||
811 | tree = &BTRFS_I(inode)->ordered_tree; | |
5fd02043 | 812 | spin_lock_irq(&tree->lock); |
4b46fce2 JB |
813 | node = tree_search(tree, file_offset); |
814 | if (!node) { | |
815 | node = tree_search(tree, file_offset + len); | |
816 | if (!node) | |
817 | goto out; | |
818 | } | |
819 | ||
820 | while (1) { | |
821 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
822 | if (range_overlaps(entry, file_offset, len)) | |
823 | break; | |
824 | ||
825 | if (entry->file_offset >= file_offset + len) { | |
826 | entry = NULL; | |
827 | break; | |
828 | } | |
829 | entry = NULL; | |
830 | node = rb_next(node); | |
831 | if (!node) | |
832 | break; | |
833 | } | |
834 | out: | |
835 | if (entry) | |
836 | atomic_inc(&entry->refs); | |
5fd02043 | 837 | spin_unlock_irq(&tree->lock); |
4b46fce2 JB |
838 | return entry; |
839 | } | |
840 | ||
eb84ae03 CM |
841 | /* |
842 | * lookup and return any extent before 'file_offset'. NULL is returned | |
843 | * if none is found | |
844 | */ | |
e6dcd2dc | 845 | struct btrfs_ordered_extent * |
d397712b | 846 | btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset) |
e6dcd2dc CM |
847 | { |
848 | struct btrfs_ordered_inode_tree *tree; | |
849 | struct rb_node *node; | |
850 | struct btrfs_ordered_extent *entry = NULL; | |
851 | ||
852 | tree = &BTRFS_I(inode)->ordered_tree; | |
5fd02043 | 853 | spin_lock_irq(&tree->lock); |
e6dcd2dc CM |
854 | node = tree_search(tree, file_offset); |
855 | if (!node) | |
856 | goto out; | |
857 | ||
858 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
859 | atomic_inc(&entry->refs); | |
860 | out: | |
5fd02043 | 861 | spin_unlock_irq(&tree->lock); |
e6dcd2dc | 862 | return entry; |
81d7ed29 | 863 | } |
dbe674a9 | 864 | |
eb84ae03 CM |
865 | /* |
866 | * After an extent is done, call this to conditionally update the on disk | |
867 | * i_size. i_size is updated to cover any fully written part of the file. | |
868 | */ | |
c2167754 | 869 | int btrfs_ordered_update_i_size(struct inode *inode, u64 offset, |
dbe674a9 CM |
870 | struct btrfs_ordered_extent *ordered) |
871 | { | |
872 | struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; | |
dbe674a9 CM |
873 | u64 disk_i_size; |
874 | u64 new_i_size; | |
c2167754 | 875 | u64 i_size = i_size_read(inode); |
dbe674a9 | 876 | struct rb_node *node; |
c2167754 | 877 | struct rb_node *prev = NULL; |
dbe674a9 | 878 | struct btrfs_ordered_extent *test; |
c2167754 YZ |
879 | int ret = 1; |
880 | ||
881 | if (ordered) | |
882 | offset = entry_end(ordered); | |
a038fab0 YZ |
883 | else |
884 | offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize); | |
dbe674a9 | 885 | |
5fd02043 | 886 | spin_lock_irq(&tree->lock); |
dbe674a9 CM |
887 | disk_i_size = BTRFS_I(inode)->disk_i_size; |
888 | ||
c2167754 YZ |
889 | /* truncate file */ |
890 | if (disk_i_size > i_size) { | |
891 | BTRFS_I(inode)->disk_i_size = i_size; | |
892 | ret = 0; | |
893 | goto out; | |
894 | } | |
895 | ||
dbe674a9 CM |
896 | /* |
897 | * if the disk i_size is already at the inode->i_size, or | |
898 | * this ordered extent is inside the disk i_size, we're done | |
899 | */ | |
5d1f4020 JB |
900 | if (disk_i_size == i_size) |
901 | goto out; | |
902 | ||
903 | /* | |
904 | * We still need to update disk_i_size if outstanding_isize is greater | |
905 | * than disk_i_size. | |
906 | */ | |
907 | if (offset <= disk_i_size && | |
908 | (!ordered || ordered->outstanding_isize <= disk_i_size)) | |
dbe674a9 | 909 | goto out; |
dbe674a9 | 910 | |
dbe674a9 CM |
911 | /* |
912 | * walk backward from this ordered extent to disk_i_size. | |
913 | * if we find an ordered extent then we can't update disk i_size | |
914 | * yet | |
915 | */ | |
c2167754 YZ |
916 | if (ordered) { |
917 | node = rb_prev(&ordered->rb_node); | |
918 | } else { | |
919 | prev = tree_search(tree, offset); | |
920 | /* | |
921 | * we insert file extents without involving ordered struct, | |
922 | * so there should be no ordered struct cover this offset | |
923 | */ | |
924 | if (prev) { | |
925 | test = rb_entry(prev, struct btrfs_ordered_extent, | |
926 | rb_node); | |
927 | BUG_ON(offset_in_entry(test, offset)); | |
928 | } | |
929 | node = prev; | |
930 | } | |
5fd02043 | 931 | for (; node; node = rb_prev(node)) { |
dbe674a9 | 932 | test = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
5fd02043 JB |
933 | |
934 | /* We treat this entry as if it doesnt exist */ | |
935 | if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags)) | |
936 | continue; | |
dbe674a9 CM |
937 | if (test->file_offset + test->len <= disk_i_size) |
938 | break; | |
c2167754 | 939 | if (test->file_offset >= i_size) |
dbe674a9 | 940 | break; |
59fe4f41 | 941 | if (entry_end(test) > disk_i_size) { |
b9a8cc5b MX |
942 | /* |
943 | * we don't update disk_i_size now, so record this | |
944 | * undealt i_size. Or we will not know the real | |
945 | * i_size. | |
946 | */ | |
947 | if (test->outstanding_isize < offset) | |
948 | test->outstanding_isize = offset; | |
949 | if (ordered && | |
950 | ordered->outstanding_isize > | |
951 | test->outstanding_isize) | |
952 | test->outstanding_isize = | |
953 | ordered->outstanding_isize; | |
dbe674a9 | 954 | goto out; |
5fd02043 | 955 | } |
dbe674a9 | 956 | } |
b9a8cc5b | 957 | new_i_size = min_t(u64, offset, i_size); |
dbe674a9 CM |
958 | |
959 | /* | |
b9a8cc5b MX |
960 | * Some ordered extents may completed before the current one, and |
961 | * we hold the real i_size in ->outstanding_isize. | |
dbe674a9 | 962 | */ |
b9a8cc5b MX |
963 | if (ordered && ordered->outstanding_isize > new_i_size) |
964 | new_i_size = min_t(u64, ordered->outstanding_isize, i_size); | |
dbe674a9 | 965 | BTRFS_I(inode)->disk_i_size = new_i_size; |
c2167754 | 966 | ret = 0; |
dbe674a9 | 967 | out: |
c2167754 | 968 | /* |
5fd02043 JB |
969 | * We need to do this because we can't remove ordered extents until |
970 | * after the i_disk_size has been updated and then the inode has been | |
971 | * updated to reflect the change, so we need to tell anybody who finds | |
972 | * this ordered extent that we've already done all the real work, we | |
973 | * just haven't completed all the other work. | |
c2167754 YZ |
974 | */ |
975 | if (ordered) | |
5fd02043 JB |
976 | set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags); |
977 | spin_unlock_irq(&tree->lock); | |
c2167754 | 978 | return ret; |
dbe674a9 | 979 | } |
ba1da2f4 | 980 | |
eb84ae03 CM |
981 | /* |
982 | * search the ordered extents for one corresponding to 'offset' and | |
983 | * try to find a checksum. This is used because we allow pages to | |
984 | * be reclaimed before their checksum is actually put into the btree | |
985 | */ | |
d20f7043 CM |
986 | int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr, |
987 | u32 *sum) | |
ba1da2f4 CM |
988 | { |
989 | struct btrfs_ordered_sum *ordered_sum; | |
990 | struct btrfs_sector_sum *sector_sums; | |
991 | struct btrfs_ordered_extent *ordered; | |
992 | struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; | |
3edf7d33 CM |
993 | unsigned long num_sectors; |
994 | unsigned long i; | |
995 | u32 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
ba1da2f4 | 996 | int ret = 1; |
ba1da2f4 CM |
997 | |
998 | ordered = btrfs_lookup_ordered_extent(inode, offset); | |
999 | if (!ordered) | |
1000 | return 1; | |
1001 | ||
5fd02043 | 1002 | spin_lock_irq(&tree->lock); |
c6e30871 | 1003 | list_for_each_entry_reverse(ordered_sum, &ordered->list, list) { |
d20f7043 | 1004 | if (disk_bytenr >= ordered_sum->bytenr) { |
3edf7d33 | 1005 | num_sectors = ordered_sum->len / sectorsize; |
ed98b56a | 1006 | sector_sums = ordered_sum->sums; |
3edf7d33 | 1007 | for (i = 0; i < num_sectors; i++) { |
d20f7043 | 1008 | if (sector_sums[i].bytenr == disk_bytenr) { |
3edf7d33 CM |
1009 | *sum = sector_sums[i].sum; |
1010 | ret = 0; | |
1011 | goto out; | |
1012 | } | |
1013 | } | |
ba1da2f4 CM |
1014 | } |
1015 | } | |
1016 | out: | |
5fd02043 | 1017 | spin_unlock_irq(&tree->lock); |
89642229 | 1018 | btrfs_put_ordered_extent(ordered); |
ba1da2f4 CM |
1019 | return ret; |
1020 | } | |
1021 | ||
f421950f | 1022 | |
5a3f23d5 CM |
1023 | /* |
1024 | * add a given inode to the list of inodes that must be fully on | |
1025 | * disk before a transaction commit finishes. | |
1026 | * | |
1027 | * This basically gives us the ext3 style data=ordered mode, and it is mostly | |
1028 | * used to make sure renamed files are fully on disk. | |
1029 | * | |
1030 | * It is a noop if the inode is already fully on disk. | |
1031 | * | |
1032 | * If trans is not null, we'll do a friendly check for a transaction that | |
1033 | * is already flushing things and force the IO down ourselves. | |
1034 | */ | |
143bede5 JM |
1035 | void btrfs_add_ordered_operation(struct btrfs_trans_handle *trans, |
1036 | struct btrfs_root *root, struct inode *inode) | |
5a3f23d5 | 1037 | { |
569e0f35 | 1038 | struct btrfs_transaction *cur_trans = trans->transaction; |
5a3f23d5 CM |
1039 | u64 last_mod; |
1040 | ||
1041 | last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans); | |
1042 | ||
1043 | /* | |
1044 | * if this file hasn't been changed since the last transaction | |
1045 | * commit, we can safely return without doing anything | |
1046 | */ | |
1047 | if (last_mod < root->fs_info->last_trans_committed) | |
143bede5 | 1048 | return; |
5a3f23d5 | 1049 | |
5a3f23d5 CM |
1050 | spin_lock(&root->fs_info->ordered_extent_lock); |
1051 | if (list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
1052 | list_add_tail(&BTRFS_I(inode)->ordered_operations, | |
569e0f35 | 1053 | &cur_trans->ordered_operations); |
5a3f23d5 CM |
1054 | } |
1055 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
5a3f23d5 | 1056 | } |
6352b91d MX |
1057 | |
1058 | int __init ordered_data_init(void) | |
1059 | { | |
1060 | btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent", | |
1061 | sizeof(struct btrfs_ordered_extent), 0, | |
1062 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, | |
1063 | NULL); | |
1064 | if (!btrfs_ordered_extent_cache) | |
1065 | return -ENOMEM; | |
25287e0a | 1066 | |
6352b91d MX |
1067 | return 0; |
1068 | } | |
1069 | ||
1070 | void ordered_data_exit(void) | |
1071 | { | |
1072 | if (btrfs_ordered_extent_cache) | |
1073 | kmem_cache_destroy(btrfs_ordered_extent_cache); | |
1074 | } |