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" |
199c2a9c | 27 | #include "disk-io.h" |
dc17ff8f | 28 | |
6352b91d MX |
29 | static struct kmem_cache *btrfs_ordered_extent_cache; |
30 | ||
e6dcd2dc | 31 | static u64 entry_end(struct btrfs_ordered_extent *entry) |
dc17ff8f | 32 | { |
e6dcd2dc CM |
33 | if (entry->file_offset + entry->len < entry->file_offset) |
34 | return (u64)-1; | |
35 | return entry->file_offset + entry->len; | |
dc17ff8f CM |
36 | } |
37 | ||
d352ac68 CM |
38 | /* returns NULL if the insertion worked, or it returns the node it did find |
39 | * in the tree | |
40 | */ | |
e6dcd2dc CM |
41 | static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset, |
42 | struct rb_node *node) | |
dc17ff8f | 43 | { |
d397712b CM |
44 | struct rb_node **p = &root->rb_node; |
45 | struct rb_node *parent = NULL; | |
e6dcd2dc | 46 | struct btrfs_ordered_extent *entry; |
dc17ff8f | 47 | |
d397712b | 48 | while (*p) { |
dc17ff8f | 49 | parent = *p; |
e6dcd2dc | 50 | entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node); |
dc17ff8f | 51 | |
e6dcd2dc | 52 | if (file_offset < entry->file_offset) |
dc17ff8f | 53 | p = &(*p)->rb_left; |
e6dcd2dc | 54 | else if (file_offset >= entry_end(entry)) |
dc17ff8f CM |
55 | p = &(*p)->rb_right; |
56 | else | |
57 | return parent; | |
58 | } | |
59 | ||
60 | rb_link_node(node, parent, p); | |
61 | rb_insert_color(node, root); | |
62 | return NULL; | |
63 | } | |
64 | ||
43c04fb1 JM |
65 | static void ordered_data_tree_panic(struct inode *inode, int errno, |
66 | u64 offset) | |
67 | { | |
68 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
69 | btrfs_panic(fs_info, errno, "Inconsistency in ordered tree at offset " | |
c1c9ff7c | 70 | "%llu\n", offset); |
43c04fb1 JM |
71 | } |
72 | ||
d352ac68 CM |
73 | /* |
74 | * look for a given offset in the tree, and if it can't be found return the | |
75 | * first lesser offset | |
76 | */ | |
e6dcd2dc CM |
77 | static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset, |
78 | struct rb_node **prev_ret) | |
dc17ff8f | 79 | { |
d397712b | 80 | struct rb_node *n = root->rb_node; |
dc17ff8f | 81 | struct rb_node *prev = NULL; |
e6dcd2dc CM |
82 | struct rb_node *test; |
83 | struct btrfs_ordered_extent *entry; | |
84 | struct btrfs_ordered_extent *prev_entry = NULL; | |
dc17ff8f | 85 | |
d397712b | 86 | while (n) { |
e6dcd2dc | 87 | entry = rb_entry(n, struct btrfs_ordered_extent, rb_node); |
dc17ff8f CM |
88 | prev = n; |
89 | prev_entry = entry; | |
dc17ff8f | 90 | |
e6dcd2dc | 91 | if (file_offset < entry->file_offset) |
dc17ff8f | 92 | n = n->rb_left; |
e6dcd2dc | 93 | else if (file_offset >= entry_end(entry)) |
dc17ff8f CM |
94 | n = n->rb_right; |
95 | else | |
96 | return n; | |
97 | } | |
98 | if (!prev_ret) | |
99 | return NULL; | |
100 | ||
d397712b | 101 | while (prev && file_offset >= entry_end(prev_entry)) { |
e6dcd2dc CM |
102 | test = rb_next(prev); |
103 | if (!test) | |
104 | break; | |
105 | prev_entry = rb_entry(test, struct btrfs_ordered_extent, | |
106 | rb_node); | |
107 | if (file_offset < entry_end(prev_entry)) | |
108 | break; | |
109 | ||
110 | prev = test; | |
111 | } | |
112 | if (prev) | |
113 | prev_entry = rb_entry(prev, struct btrfs_ordered_extent, | |
114 | rb_node); | |
d397712b | 115 | while (prev && file_offset < entry_end(prev_entry)) { |
e6dcd2dc CM |
116 | test = rb_prev(prev); |
117 | if (!test) | |
118 | break; | |
119 | prev_entry = rb_entry(test, struct btrfs_ordered_extent, | |
120 | rb_node); | |
121 | prev = test; | |
dc17ff8f CM |
122 | } |
123 | *prev_ret = prev; | |
124 | return NULL; | |
125 | } | |
126 | ||
d352ac68 CM |
127 | /* |
128 | * helper to check if a given offset is inside a given entry | |
129 | */ | |
e6dcd2dc CM |
130 | static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset) |
131 | { | |
132 | if (file_offset < entry->file_offset || | |
133 | entry->file_offset + entry->len <= file_offset) | |
134 | return 0; | |
135 | return 1; | |
136 | } | |
137 | ||
4b46fce2 JB |
138 | static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset, |
139 | u64 len) | |
140 | { | |
141 | if (file_offset + len <= entry->file_offset || | |
142 | entry->file_offset + entry->len <= file_offset) | |
143 | return 0; | |
144 | return 1; | |
145 | } | |
146 | ||
d352ac68 CM |
147 | /* |
148 | * look find the first ordered struct that has this offset, otherwise | |
149 | * the first one less than this offset | |
150 | */ | |
e6dcd2dc CM |
151 | static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree, |
152 | u64 file_offset) | |
dc17ff8f | 153 | { |
e6dcd2dc | 154 | struct rb_root *root = &tree->tree; |
c87fb6fd | 155 | struct rb_node *prev = NULL; |
dc17ff8f | 156 | struct rb_node *ret; |
e6dcd2dc CM |
157 | struct btrfs_ordered_extent *entry; |
158 | ||
159 | if (tree->last) { | |
160 | entry = rb_entry(tree->last, struct btrfs_ordered_extent, | |
161 | rb_node); | |
162 | if (offset_in_entry(entry, file_offset)) | |
163 | return tree->last; | |
164 | } | |
165 | ret = __tree_search(root, file_offset, &prev); | |
dc17ff8f | 166 | if (!ret) |
e6dcd2dc CM |
167 | ret = prev; |
168 | if (ret) | |
169 | tree->last = ret; | |
dc17ff8f CM |
170 | return ret; |
171 | } | |
172 | ||
eb84ae03 CM |
173 | /* allocate and add a new ordered_extent into the per-inode tree. |
174 | * file_offset is the logical offset in the file | |
175 | * | |
176 | * start is the disk block number of an extent already reserved in the | |
177 | * extent allocation tree | |
178 | * | |
179 | * len is the length of the extent | |
180 | * | |
eb84ae03 CM |
181 | * The tree is given a single reference on the ordered extent that was |
182 | * inserted. | |
183 | */ | |
4b46fce2 JB |
184 | static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset, |
185 | u64 start, u64 len, u64 disk_len, | |
261507a0 | 186 | int type, int dio, int compress_type) |
dc17ff8f | 187 | { |
199c2a9c | 188 | struct btrfs_root *root = BTRFS_I(inode)->root; |
dc17ff8f | 189 | struct btrfs_ordered_inode_tree *tree; |
e6dcd2dc CM |
190 | struct rb_node *node; |
191 | struct btrfs_ordered_extent *entry; | |
dc17ff8f | 192 | |
e6dcd2dc | 193 | tree = &BTRFS_I(inode)->ordered_tree; |
6352b91d | 194 | entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS); |
dc17ff8f CM |
195 | if (!entry) |
196 | return -ENOMEM; | |
197 | ||
e6dcd2dc CM |
198 | entry->file_offset = file_offset; |
199 | entry->start = start; | |
200 | entry->len = len; | |
2ab28f32 JB |
201 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) && |
202 | !(type == BTRFS_ORDERED_NOCOW)) | |
203 | entry->csum_bytes_left = disk_len; | |
c8b97818 | 204 | entry->disk_len = disk_len; |
8b62b72b | 205 | entry->bytes_left = len; |
5fd02043 | 206 | entry->inode = igrab(inode); |
261507a0 | 207 | entry->compress_type = compress_type; |
77cef2ec | 208 | entry->truncated_len = (u64)-1; |
d899e052 | 209 | if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE) |
80ff3856 | 210 | set_bit(type, &entry->flags); |
3eaa2885 | 211 | |
4b46fce2 JB |
212 | if (dio) |
213 | set_bit(BTRFS_ORDERED_DIRECT, &entry->flags); | |
214 | ||
e6dcd2dc CM |
215 | /* one ref for the tree */ |
216 | atomic_set(&entry->refs, 1); | |
217 | init_waitqueue_head(&entry->wait); | |
218 | INIT_LIST_HEAD(&entry->list); | |
3eaa2885 | 219 | INIT_LIST_HEAD(&entry->root_extent_list); |
9afab882 MX |
220 | INIT_LIST_HEAD(&entry->work_list); |
221 | init_completion(&entry->completion); | |
2ab28f32 | 222 | INIT_LIST_HEAD(&entry->log_list); |
dc17ff8f | 223 | |
1abe9b8a | 224 | trace_btrfs_ordered_extent_add(inode, entry); |
225 | ||
5fd02043 | 226 | spin_lock_irq(&tree->lock); |
e6dcd2dc CM |
227 | node = tree_insert(&tree->tree, file_offset, |
228 | &entry->rb_node); | |
43c04fb1 JM |
229 | if (node) |
230 | ordered_data_tree_panic(inode, -EEXIST, file_offset); | |
5fd02043 | 231 | spin_unlock_irq(&tree->lock); |
d397712b | 232 | |
199c2a9c | 233 | spin_lock(&root->ordered_extent_lock); |
3eaa2885 | 234 | list_add_tail(&entry->root_extent_list, |
199c2a9c MX |
235 | &root->ordered_extents); |
236 | root->nr_ordered_extents++; | |
237 | if (root->nr_ordered_extents == 1) { | |
238 | spin_lock(&root->fs_info->ordered_root_lock); | |
239 | BUG_ON(!list_empty(&root->ordered_root)); | |
240 | list_add_tail(&root->ordered_root, | |
241 | &root->fs_info->ordered_roots); | |
242 | spin_unlock(&root->fs_info->ordered_root_lock); | |
243 | } | |
244 | spin_unlock(&root->ordered_extent_lock); | |
3eaa2885 | 245 | |
dc17ff8f CM |
246 | return 0; |
247 | } | |
248 | ||
4b46fce2 JB |
249 | int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset, |
250 | u64 start, u64 len, u64 disk_len, int type) | |
251 | { | |
252 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, | |
261507a0 LZ |
253 | disk_len, type, 0, |
254 | BTRFS_COMPRESS_NONE); | |
4b46fce2 JB |
255 | } |
256 | ||
257 | int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset, | |
258 | u64 start, u64 len, u64 disk_len, int type) | |
259 | { | |
260 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, | |
261507a0 LZ |
261 | disk_len, type, 1, |
262 | BTRFS_COMPRESS_NONE); | |
263 | } | |
264 | ||
265 | int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset, | |
266 | u64 start, u64 len, u64 disk_len, | |
267 | int type, int compress_type) | |
268 | { | |
269 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, | |
270 | disk_len, type, 0, | |
271 | compress_type); | |
4b46fce2 JB |
272 | } |
273 | ||
eb84ae03 CM |
274 | /* |
275 | * Add a struct btrfs_ordered_sum into the list of checksums to be inserted | |
3edf7d33 CM |
276 | * when an ordered extent is finished. If the list covers more than one |
277 | * ordered extent, it is split across multiples. | |
eb84ae03 | 278 | */ |
143bede5 JM |
279 | void btrfs_add_ordered_sum(struct inode *inode, |
280 | struct btrfs_ordered_extent *entry, | |
281 | struct btrfs_ordered_sum *sum) | |
dc17ff8f | 282 | { |
e6dcd2dc | 283 | struct btrfs_ordered_inode_tree *tree; |
dc17ff8f | 284 | |
e6dcd2dc | 285 | tree = &BTRFS_I(inode)->ordered_tree; |
5fd02043 | 286 | spin_lock_irq(&tree->lock); |
e6dcd2dc | 287 | list_add_tail(&sum->list, &entry->list); |
2ab28f32 JB |
288 | WARN_ON(entry->csum_bytes_left < sum->len); |
289 | entry->csum_bytes_left -= sum->len; | |
290 | if (entry->csum_bytes_left == 0) | |
291 | wake_up(&entry->wait); | |
5fd02043 | 292 | spin_unlock_irq(&tree->lock); |
dc17ff8f CM |
293 | } |
294 | ||
163cf09c CM |
295 | /* |
296 | * this is used to account for finished IO across a given range | |
297 | * of the file. The IO may span ordered extents. If | |
298 | * a given ordered_extent is completely done, 1 is returned, otherwise | |
299 | * 0. | |
300 | * | |
301 | * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used | |
302 | * to make sure this function only returns 1 once for a given ordered extent. | |
303 | * | |
304 | * file_offset is updated to one byte past the range that is recorded as | |
305 | * complete. This allows you to walk forward in the file. | |
306 | */ | |
307 | int btrfs_dec_test_first_ordered_pending(struct inode *inode, | |
308 | struct btrfs_ordered_extent **cached, | |
5fd02043 | 309 | u64 *file_offset, u64 io_size, int uptodate) |
163cf09c CM |
310 | { |
311 | struct btrfs_ordered_inode_tree *tree; | |
312 | struct rb_node *node; | |
313 | struct btrfs_ordered_extent *entry = NULL; | |
314 | int ret; | |
5fd02043 | 315 | unsigned long flags; |
163cf09c CM |
316 | u64 dec_end; |
317 | u64 dec_start; | |
318 | u64 to_dec; | |
319 | ||
320 | tree = &BTRFS_I(inode)->ordered_tree; | |
5fd02043 | 321 | spin_lock_irqsave(&tree->lock, flags); |
163cf09c CM |
322 | node = tree_search(tree, *file_offset); |
323 | if (!node) { | |
324 | ret = 1; | |
325 | goto out; | |
326 | } | |
327 | ||
328 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
329 | if (!offset_in_entry(entry, *file_offset)) { | |
330 | ret = 1; | |
331 | goto out; | |
332 | } | |
333 | ||
334 | dec_start = max(*file_offset, entry->file_offset); | |
335 | dec_end = min(*file_offset + io_size, entry->file_offset + | |
336 | entry->len); | |
337 | *file_offset = dec_end; | |
338 | if (dec_start > dec_end) { | |
339 | printk(KERN_CRIT "bad ordering dec_start %llu end %llu\n", | |
c1c9ff7c | 340 | dec_start, dec_end); |
163cf09c CM |
341 | } |
342 | to_dec = dec_end - dec_start; | |
343 | if (to_dec > entry->bytes_left) { | |
344 | printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n", | |
c1c9ff7c | 345 | entry->bytes_left, to_dec); |
163cf09c CM |
346 | } |
347 | entry->bytes_left -= to_dec; | |
5fd02043 JB |
348 | if (!uptodate) |
349 | set_bit(BTRFS_ORDERED_IOERR, &entry->flags); | |
350 | ||
163cf09c CM |
351 | if (entry->bytes_left == 0) |
352 | ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); | |
353 | else | |
354 | ret = 1; | |
355 | out: | |
356 | if (!ret && cached && entry) { | |
357 | *cached = entry; | |
358 | atomic_inc(&entry->refs); | |
359 | } | |
5fd02043 | 360 | spin_unlock_irqrestore(&tree->lock, flags); |
163cf09c CM |
361 | return ret == 0; |
362 | } | |
363 | ||
eb84ae03 CM |
364 | /* |
365 | * this is used to account for finished IO across a given range | |
366 | * of the file. The IO should not span ordered extents. If | |
367 | * a given ordered_extent is completely done, 1 is returned, otherwise | |
368 | * 0. | |
369 | * | |
370 | * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used | |
371 | * to make sure this function only returns 1 once for a given ordered extent. | |
372 | */ | |
e6dcd2dc | 373 | int btrfs_dec_test_ordered_pending(struct inode *inode, |
5a1a3df1 | 374 | struct btrfs_ordered_extent **cached, |
5fd02043 | 375 | u64 file_offset, u64 io_size, int uptodate) |
dc17ff8f | 376 | { |
e6dcd2dc | 377 | struct btrfs_ordered_inode_tree *tree; |
dc17ff8f | 378 | struct rb_node *node; |
5a1a3df1 | 379 | struct btrfs_ordered_extent *entry = NULL; |
5fd02043 | 380 | unsigned long flags; |
e6dcd2dc CM |
381 | int ret; |
382 | ||
383 | tree = &BTRFS_I(inode)->ordered_tree; | |
5fd02043 JB |
384 | spin_lock_irqsave(&tree->lock, flags); |
385 | if (cached && *cached) { | |
386 | entry = *cached; | |
387 | goto have_entry; | |
388 | } | |
389 | ||
e6dcd2dc | 390 | node = tree_search(tree, file_offset); |
dc17ff8f | 391 | if (!node) { |
e6dcd2dc CM |
392 | ret = 1; |
393 | goto out; | |
dc17ff8f CM |
394 | } |
395 | ||
e6dcd2dc | 396 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
5fd02043 | 397 | have_entry: |
e6dcd2dc CM |
398 | if (!offset_in_entry(entry, file_offset)) { |
399 | ret = 1; | |
400 | goto out; | |
dc17ff8f | 401 | } |
e6dcd2dc | 402 | |
8b62b72b CM |
403 | if (io_size > entry->bytes_left) { |
404 | printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n", | |
c1c9ff7c | 405 | entry->bytes_left, io_size); |
8b62b72b CM |
406 | } |
407 | entry->bytes_left -= io_size; | |
5fd02043 JB |
408 | if (!uptodate) |
409 | set_bit(BTRFS_ORDERED_IOERR, &entry->flags); | |
410 | ||
8b62b72b | 411 | if (entry->bytes_left == 0) |
e6dcd2dc | 412 | ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); |
8b62b72b CM |
413 | else |
414 | ret = 1; | |
e6dcd2dc | 415 | out: |
5a1a3df1 JB |
416 | if (!ret && cached && entry) { |
417 | *cached = entry; | |
418 | atomic_inc(&entry->refs); | |
419 | } | |
5fd02043 | 420 | spin_unlock_irqrestore(&tree->lock, flags); |
e6dcd2dc CM |
421 | return ret == 0; |
422 | } | |
dc17ff8f | 423 | |
2ab28f32 JB |
424 | /* Needs to either be called under a log transaction or the log_mutex */ |
425 | void btrfs_get_logged_extents(struct btrfs_root *log, struct inode *inode) | |
426 | { | |
427 | struct btrfs_ordered_inode_tree *tree; | |
428 | struct btrfs_ordered_extent *ordered; | |
429 | struct rb_node *n; | |
430 | int index = log->log_transid % 2; | |
431 | ||
432 | tree = &BTRFS_I(inode)->ordered_tree; | |
433 | spin_lock_irq(&tree->lock); | |
434 | for (n = rb_first(&tree->tree); n; n = rb_next(n)) { | |
435 | ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node); | |
436 | spin_lock(&log->log_extents_lock[index]); | |
437 | if (list_empty(&ordered->log_list)) { | |
438 | list_add_tail(&ordered->log_list, &log->logged_list[index]); | |
439 | atomic_inc(&ordered->refs); | |
440 | } | |
441 | spin_unlock(&log->log_extents_lock[index]); | |
442 | } | |
443 | spin_unlock_irq(&tree->lock); | |
444 | } | |
445 | ||
446 | void btrfs_wait_logged_extents(struct btrfs_root *log, u64 transid) | |
447 | { | |
448 | struct btrfs_ordered_extent *ordered; | |
449 | int index = transid % 2; | |
450 | ||
451 | spin_lock_irq(&log->log_extents_lock[index]); | |
452 | while (!list_empty(&log->logged_list[index])) { | |
453 | ordered = list_first_entry(&log->logged_list[index], | |
454 | struct btrfs_ordered_extent, | |
455 | log_list); | |
456 | list_del_init(&ordered->log_list); | |
457 | spin_unlock_irq(&log->log_extents_lock[index]); | |
458 | wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE, | |
459 | &ordered->flags)); | |
460 | btrfs_put_ordered_extent(ordered); | |
461 | spin_lock_irq(&log->log_extents_lock[index]); | |
462 | } | |
463 | spin_unlock_irq(&log->log_extents_lock[index]); | |
464 | } | |
465 | ||
466 | void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid) | |
467 | { | |
468 | struct btrfs_ordered_extent *ordered; | |
469 | int index = transid % 2; | |
470 | ||
471 | spin_lock_irq(&log->log_extents_lock[index]); | |
472 | while (!list_empty(&log->logged_list[index])) { | |
473 | ordered = list_first_entry(&log->logged_list[index], | |
474 | struct btrfs_ordered_extent, | |
475 | log_list); | |
476 | list_del_init(&ordered->log_list); | |
477 | spin_unlock_irq(&log->log_extents_lock[index]); | |
478 | btrfs_put_ordered_extent(ordered); | |
479 | spin_lock_irq(&log->log_extents_lock[index]); | |
480 | } | |
481 | spin_unlock_irq(&log->log_extents_lock[index]); | |
482 | } | |
483 | ||
eb84ae03 CM |
484 | /* |
485 | * used to drop a reference on an ordered extent. This will free | |
486 | * the extent if the last reference is dropped | |
487 | */ | |
143bede5 | 488 | void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry) |
e6dcd2dc | 489 | { |
ba1da2f4 CM |
490 | struct list_head *cur; |
491 | struct btrfs_ordered_sum *sum; | |
492 | ||
1abe9b8a | 493 | trace_btrfs_ordered_extent_put(entry->inode, entry); |
494 | ||
ba1da2f4 | 495 | if (atomic_dec_and_test(&entry->refs)) { |
5fd02043 JB |
496 | if (entry->inode) |
497 | btrfs_add_delayed_iput(entry->inode); | |
d397712b | 498 | while (!list_empty(&entry->list)) { |
ba1da2f4 CM |
499 | cur = entry->list.next; |
500 | sum = list_entry(cur, struct btrfs_ordered_sum, list); | |
501 | list_del(&sum->list); | |
502 | kfree(sum); | |
503 | } | |
6352b91d | 504 | kmem_cache_free(btrfs_ordered_extent_cache, entry); |
ba1da2f4 | 505 | } |
dc17ff8f | 506 | } |
cee36a03 | 507 | |
eb84ae03 CM |
508 | /* |
509 | * remove an ordered extent from the tree. No references are dropped | |
5fd02043 | 510 | * and waiters are woken up. |
eb84ae03 | 511 | */ |
5fd02043 JB |
512 | void btrfs_remove_ordered_extent(struct inode *inode, |
513 | struct btrfs_ordered_extent *entry) | |
cee36a03 | 514 | { |
e6dcd2dc | 515 | struct btrfs_ordered_inode_tree *tree; |
287a0ab9 | 516 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cee36a03 | 517 | struct rb_node *node; |
cee36a03 | 518 | |
e6dcd2dc | 519 | tree = &BTRFS_I(inode)->ordered_tree; |
5fd02043 | 520 | spin_lock_irq(&tree->lock); |
e6dcd2dc | 521 | node = &entry->rb_node; |
cee36a03 | 522 | rb_erase(node, &tree->tree); |
e6dcd2dc CM |
523 | tree->last = NULL; |
524 | set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags); | |
5fd02043 | 525 | spin_unlock_irq(&tree->lock); |
3eaa2885 | 526 | |
199c2a9c | 527 | spin_lock(&root->ordered_extent_lock); |
3eaa2885 | 528 | list_del_init(&entry->root_extent_list); |
199c2a9c | 529 | root->nr_ordered_extents--; |
5a3f23d5 | 530 | |
1abe9b8a | 531 | trace_btrfs_ordered_extent_remove(inode, entry); |
532 | ||
5a3f23d5 CM |
533 | /* |
534 | * we have no more ordered extents for this inode and | |
535 | * no dirty pages. We can safely remove it from the | |
536 | * list of ordered extents | |
537 | */ | |
538 | if (RB_EMPTY_ROOT(&tree->tree) && | |
539 | !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { | |
93858769 | 540 | spin_lock(&root->fs_info->ordered_root_lock); |
5a3f23d5 | 541 | list_del_init(&BTRFS_I(inode)->ordered_operations); |
93858769 | 542 | spin_unlock(&root->fs_info->ordered_root_lock); |
5a3f23d5 | 543 | } |
199c2a9c MX |
544 | |
545 | if (!root->nr_ordered_extents) { | |
546 | spin_lock(&root->fs_info->ordered_root_lock); | |
547 | BUG_ON(list_empty(&root->ordered_root)); | |
548 | list_del_init(&root->ordered_root); | |
549 | spin_unlock(&root->fs_info->ordered_root_lock); | |
550 | } | |
551 | spin_unlock(&root->ordered_extent_lock); | |
e6dcd2dc | 552 | wake_up(&entry->wait); |
cee36a03 CM |
553 | } |
554 | ||
9afab882 MX |
555 | static void btrfs_run_ordered_extent_work(struct btrfs_work *work) |
556 | { | |
557 | struct btrfs_ordered_extent *ordered; | |
558 | ||
559 | ordered = container_of(work, struct btrfs_ordered_extent, flush_work); | |
560 | btrfs_start_ordered_extent(ordered->inode, ordered, 1); | |
561 | complete(&ordered->completion); | |
562 | } | |
563 | ||
d352ac68 CM |
564 | /* |
565 | * wait for all the ordered extents in a root. This is done when balancing | |
566 | * space between drives. | |
567 | */ | |
b0244199 | 568 | int btrfs_wait_ordered_extents(struct btrfs_root *root, int nr) |
3eaa2885 | 569 | { |
9afab882 | 570 | struct list_head splice, works; |
9afab882 | 571 | struct btrfs_ordered_extent *ordered, *next; |
b0244199 | 572 | int count = 0; |
3eaa2885 CM |
573 | |
574 | INIT_LIST_HEAD(&splice); | |
9afab882 | 575 | INIT_LIST_HEAD(&works); |
3eaa2885 | 576 | |
db1d607d | 577 | mutex_lock(&root->fs_info->ordered_operations_mutex); |
199c2a9c MX |
578 | spin_lock(&root->ordered_extent_lock); |
579 | list_splice_init(&root->ordered_extents, &splice); | |
b0244199 | 580 | while (!list_empty(&splice) && nr) { |
199c2a9c MX |
581 | ordered = list_first_entry(&splice, struct btrfs_ordered_extent, |
582 | root_extent_list); | |
583 | list_move_tail(&ordered->root_extent_list, | |
584 | &root->ordered_extents); | |
199c2a9c MX |
585 | atomic_inc(&ordered->refs); |
586 | spin_unlock(&root->ordered_extent_lock); | |
3eaa2885 | 587 | |
199c2a9c MX |
588 | ordered->flush_work.func = btrfs_run_ordered_extent_work; |
589 | list_add_tail(&ordered->work_list, &works); | |
590 | btrfs_queue_worker(&root->fs_info->flush_workers, | |
591 | &ordered->flush_work); | |
3eaa2885 | 592 | |
9afab882 | 593 | cond_resched(); |
199c2a9c | 594 | spin_lock(&root->ordered_extent_lock); |
b0244199 MX |
595 | if (nr != -1) |
596 | nr--; | |
597 | count++; | |
3eaa2885 | 598 | } |
b0244199 | 599 | list_splice_tail(&splice, &root->ordered_extents); |
199c2a9c | 600 | spin_unlock(&root->ordered_extent_lock); |
9afab882 MX |
601 | |
602 | list_for_each_entry_safe(ordered, next, &works, work_list) { | |
603 | list_del_init(&ordered->work_list); | |
604 | wait_for_completion(&ordered->completion); | |
9afab882 | 605 | btrfs_put_ordered_extent(ordered); |
9afab882 MX |
606 | cond_resched(); |
607 | } | |
db1d607d | 608 | mutex_unlock(&root->fs_info->ordered_operations_mutex); |
b0244199 MX |
609 | |
610 | return count; | |
3eaa2885 CM |
611 | } |
612 | ||
b0244199 | 613 | void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr) |
199c2a9c MX |
614 | { |
615 | struct btrfs_root *root; | |
616 | struct list_head splice; | |
b0244199 | 617 | int done; |
199c2a9c MX |
618 | |
619 | INIT_LIST_HEAD(&splice); | |
620 | ||
621 | spin_lock(&fs_info->ordered_root_lock); | |
622 | list_splice_init(&fs_info->ordered_roots, &splice); | |
b0244199 | 623 | while (!list_empty(&splice) && nr) { |
199c2a9c MX |
624 | root = list_first_entry(&splice, struct btrfs_root, |
625 | ordered_root); | |
626 | root = btrfs_grab_fs_root(root); | |
627 | BUG_ON(!root); | |
628 | list_move_tail(&root->ordered_root, | |
629 | &fs_info->ordered_roots); | |
630 | spin_unlock(&fs_info->ordered_root_lock); | |
631 | ||
b0244199 | 632 | done = btrfs_wait_ordered_extents(root, nr); |
199c2a9c MX |
633 | btrfs_put_fs_root(root); |
634 | ||
635 | spin_lock(&fs_info->ordered_root_lock); | |
b0244199 MX |
636 | if (nr != -1) { |
637 | nr -= done; | |
638 | WARN_ON(nr < 0); | |
639 | } | |
199c2a9c | 640 | } |
931aa877 | 641 | list_splice_tail(&splice, &fs_info->ordered_roots); |
199c2a9c MX |
642 | spin_unlock(&fs_info->ordered_root_lock); |
643 | } | |
644 | ||
5a3f23d5 CM |
645 | /* |
646 | * this is used during transaction commit to write all the inodes | |
647 | * added to the ordered operation list. These files must be fully on | |
648 | * disk before the transaction commits. | |
649 | * | |
650 | * we have two modes here, one is to just start the IO via filemap_flush | |
651 | * and the other is to wait for all the io. When we wait, we have an | |
652 | * extra check to make sure the ordered operation list really is empty | |
653 | * before we return | |
654 | */ | |
569e0f35 JB |
655 | int btrfs_run_ordered_operations(struct btrfs_trans_handle *trans, |
656 | struct btrfs_root *root, int wait) | |
5a3f23d5 CM |
657 | { |
658 | struct btrfs_inode *btrfs_inode; | |
659 | struct inode *inode; | |
569e0f35 | 660 | struct btrfs_transaction *cur_trans = trans->transaction; |
5a3f23d5 | 661 | struct list_head splice; |
25287e0a MX |
662 | struct list_head works; |
663 | struct btrfs_delalloc_work *work, *next; | |
664 | int ret = 0; | |
5a3f23d5 CM |
665 | |
666 | INIT_LIST_HEAD(&splice); | |
25287e0a | 667 | INIT_LIST_HEAD(&works); |
5a3f23d5 | 668 | |
9ffba8cd | 669 | mutex_lock(&root->fs_info->ordered_extent_flush_mutex); |
199c2a9c | 670 | spin_lock(&root->fs_info->ordered_root_lock); |
569e0f35 | 671 | list_splice_init(&cur_trans->ordered_operations, &splice); |
5a3f23d5 CM |
672 | while (!list_empty(&splice)) { |
673 | btrfs_inode = list_entry(splice.next, struct btrfs_inode, | |
674 | ordered_operations); | |
5a3f23d5 CM |
675 | inode = &btrfs_inode->vfs_inode; |
676 | ||
677 | list_del_init(&btrfs_inode->ordered_operations); | |
678 | ||
679 | /* | |
680 | * the inode may be getting freed (in sys_unlink path). | |
681 | */ | |
682 | inode = igrab(inode); | |
25287e0a MX |
683 | if (!inode) |
684 | continue; | |
5b947f1b MX |
685 | |
686 | if (!wait) | |
687 | list_add_tail(&BTRFS_I(inode)->ordered_operations, | |
569e0f35 | 688 | &cur_trans->ordered_operations); |
199c2a9c | 689 | spin_unlock(&root->fs_info->ordered_root_lock); |
5a3f23d5 | 690 | |
25287e0a MX |
691 | work = btrfs_alloc_delalloc_work(inode, wait, 1); |
692 | if (!work) { | |
199c2a9c | 693 | spin_lock(&root->fs_info->ordered_root_lock); |
25287e0a MX |
694 | if (list_empty(&BTRFS_I(inode)->ordered_operations)) |
695 | list_add_tail(&btrfs_inode->ordered_operations, | |
696 | &splice); | |
25287e0a | 697 | list_splice_tail(&splice, |
569e0f35 | 698 | &cur_trans->ordered_operations); |
199c2a9c | 699 | spin_unlock(&root->fs_info->ordered_root_lock); |
25287e0a MX |
700 | ret = -ENOMEM; |
701 | goto out; | |
5a3f23d5 | 702 | } |
25287e0a MX |
703 | list_add_tail(&work->list, &works); |
704 | btrfs_queue_worker(&root->fs_info->flush_workers, | |
705 | &work->work); | |
5a3f23d5 CM |
706 | |
707 | cond_resched(); | |
199c2a9c | 708 | spin_lock(&root->fs_info->ordered_root_lock); |
5a3f23d5 | 709 | } |
199c2a9c | 710 | spin_unlock(&root->fs_info->ordered_root_lock); |
25287e0a MX |
711 | out: |
712 | list_for_each_entry_safe(work, next, &works, list) { | |
713 | list_del_init(&work->list); | |
714 | btrfs_wait_and_free_delalloc_work(work); | |
715 | } | |
9ffba8cd | 716 | mutex_unlock(&root->fs_info->ordered_extent_flush_mutex); |
25287e0a | 717 | return ret; |
5a3f23d5 CM |
718 | } |
719 | ||
eb84ae03 CM |
720 | /* |
721 | * Used to start IO or wait for a given ordered extent to finish. | |
722 | * | |
723 | * If wait is one, this effectively waits on page writeback for all the pages | |
724 | * in the extent, and it waits on the io completion code to insert | |
725 | * metadata into the btree corresponding to the extent | |
726 | */ | |
727 | void btrfs_start_ordered_extent(struct inode *inode, | |
728 | struct btrfs_ordered_extent *entry, | |
729 | int wait) | |
e6dcd2dc CM |
730 | { |
731 | u64 start = entry->file_offset; | |
732 | u64 end = start + entry->len - 1; | |
e1b81e67 | 733 | |
1abe9b8a | 734 | trace_btrfs_ordered_extent_start(inode, entry); |
735 | ||
eb84ae03 CM |
736 | /* |
737 | * pages in the range can be dirty, clean or writeback. We | |
738 | * start IO on any dirty ones so the wait doesn't stall waiting | |
b2570314 | 739 | * for the flusher thread to find them |
eb84ae03 | 740 | */ |
4b46fce2 JB |
741 | if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags)) |
742 | filemap_fdatawrite_range(inode->i_mapping, start, end); | |
c8b97818 | 743 | if (wait) { |
e6dcd2dc CM |
744 | wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE, |
745 | &entry->flags)); | |
c8b97818 | 746 | } |
e6dcd2dc | 747 | } |
cee36a03 | 748 | |
eb84ae03 CM |
749 | /* |
750 | * Used to wait on ordered extents across a large range of bytes. | |
751 | */ | |
0ef8b726 | 752 | int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len) |
e6dcd2dc | 753 | { |
0ef8b726 | 754 | int ret = 0; |
e6dcd2dc | 755 | u64 end; |
e5a2217e | 756 | u64 orig_end; |
e6dcd2dc | 757 | struct btrfs_ordered_extent *ordered; |
e5a2217e CM |
758 | |
759 | if (start + len < start) { | |
f421950f | 760 | orig_end = INT_LIMIT(loff_t); |
e5a2217e CM |
761 | } else { |
762 | orig_end = start + len - 1; | |
f421950f CM |
763 | if (orig_end > INT_LIMIT(loff_t)) |
764 | orig_end = INT_LIMIT(loff_t); | |
e5a2217e | 765 | } |
551ebb2d | 766 | |
e5a2217e CM |
767 | /* start IO across the range first to instantiate any delalloc |
768 | * extents | |
769 | */ | |
0ef8b726 JB |
770 | ret = filemap_fdatawrite_range(inode->i_mapping, start, orig_end); |
771 | if (ret) | |
772 | return ret; | |
7ddf5a42 JB |
773 | /* |
774 | * So with compression we will find and lock a dirty page and clear the | |
775 | * first one as dirty, setup an async extent, and immediately return | |
776 | * with the entire range locked but with nobody actually marked with | |
777 | * writeback. So we can't just filemap_write_and_wait_range() and | |
778 | * expect it to work since it will just kick off a thread to do the | |
779 | * actual work. So we need to call filemap_fdatawrite_range _again_ | |
780 | * since it will wait on the page lock, which won't be unlocked until | |
781 | * after the pages have been marked as writeback and so we're good to go | |
782 | * from there. We have to do this otherwise we'll miss the ordered | |
783 | * extents and that results in badness. Please Josef, do not think you | |
784 | * know better and pull this out at some point in the future, it is | |
785 | * right and you are wrong. | |
786 | */ | |
787 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
0ef8b726 JB |
788 | &BTRFS_I(inode)->runtime_flags)) { |
789 | ret = filemap_fdatawrite_range(inode->i_mapping, start, | |
790 | orig_end); | |
791 | if (ret) | |
792 | return ret; | |
793 | } | |
794 | ret = filemap_fdatawait_range(inode->i_mapping, start, orig_end); | |
795 | if (ret) | |
796 | return ret; | |
e5a2217e | 797 | |
f421950f | 798 | end = orig_end; |
d397712b | 799 | while (1) { |
e6dcd2dc | 800 | ordered = btrfs_lookup_first_ordered_extent(inode, end); |
d397712b | 801 | if (!ordered) |
e6dcd2dc | 802 | break; |
e5a2217e | 803 | if (ordered->file_offset > orig_end) { |
e6dcd2dc CM |
804 | btrfs_put_ordered_extent(ordered); |
805 | break; | |
806 | } | |
b52abf1e | 807 | if (ordered->file_offset + ordered->len <= start) { |
e6dcd2dc CM |
808 | btrfs_put_ordered_extent(ordered); |
809 | break; | |
810 | } | |
e5a2217e | 811 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc | 812 | end = ordered->file_offset; |
0ef8b726 JB |
813 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) |
814 | ret = -EIO; | |
e6dcd2dc | 815 | btrfs_put_ordered_extent(ordered); |
0ef8b726 | 816 | if (ret || end == 0 || end == start) |
e6dcd2dc CM |
817 | break; |
818 | end--; | |
819 | } | |
0ef8b726 | 820 | return ret; |
cee36a03 CM |
821 | } |
822 | ||
eb84ae03 CM |
823 | /* |
824 | * find an ordered extent corresponding to file_offset. return NULL if | |
825 | * nothing is found, otherwise take a reference on the extent and return it | |
826 | */ | |
e6dcd2dc CM |
827 | struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode, |
828 | u64 file_offset) | |
829 | { | |
830 | struct btrfs_ordered_inode_tree *tree; | |
831 | struct rb_node *node; | |
832 | struct btrfs_ordered_extent *entry = NULL; | |
833 | ||
834 | tree = &BTRFS_I(inode)->ordered_tree; | |
5fd02043 | 835 | spin_lock_irq(&tree->lock); |
e6dcd2dc CM |
836 | node = tree_search(tree, file_offset); |
837 | if (!node) | |
838 | goto out; | |
839 | ||
840 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
841 | if (!offset_in_entry(entry, file_offset)) | |
842 | entry = NULL; | |
843 | if (entry) | |
844 | atomic_inc(&entry->refs); | |
845 | out: | |
5fd02043 | 846 | spin_unlock_irq(&tree->lock); |
e6dcd2dc CM |
847 | return entry; |
848 | } | |
849 | ||
4b46fce2 JB |
850 | /* Since the DIO code tries to lock a wide area we need to look for any ordered |
851 | * extents that exist in the range, rather than just the start of the range. | |
852 | */ | |
853 | struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode, | |
854 | u64 file_offset, | |
855 | u64 len) | |
856 | { | |
857 | struct btrfs_ordered_inode_tree *tree; | |
858 | struct rb_node *node; | |
859 | struct btrfs_ordered_extent *entry = NULL; | |
860 | ||
861 | tree = &BTRFS_I(inode)->ordered_tree; | |
5fd02043 | 862 | spin_lock_irq(&tree->lock); |
4b46fce2 JB |
863 | node = tree_search(tree, file_offset); |
864 | if (!node) { | |
865 | node = tree_search(tree, file_offset + len); | |
866 | if (!node) | |
867 | goto out; | |
868 | } | |
869 | ||
870 | while (1) { | |
871 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
872 | if (range_overlaps(entry, file_offset, len)) | |
873 | break; | |
874 | ||
875 | if (entry->file_offset >= file_offset + len) { | |
876 | entry = NULL; | |
877 | break; | |
878 | } | |
879 | entry = NULL; | |
880 | node = rb_next(node); | |
881 | if (!node) | |
882 | break; | |
883 | } | |
884 | out: | |
885 | if (entry) | |
886 | atomic_inc(&entry->refs); | |
5fd02043 | 887 | spin_unlock_irq(&tree->lock); |
4b46fce2 JB |
888 | return entry; |
889 | } | |
890 | ||
eb84ae03 CM |
891 | /* |
892 | * lookup and return any extent before 'file_offset'. NULL is returned | |
893 | * if none is found | |
894 | */ | |
e6dcd2dc | 895 | struct btrfs_ordered_extent * |
d397712b | 896 | btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset) |
e6dcd2dc CM |
897 | { |
898 | struct btrfs_ordered_inode_tree *tree; | |
899 | struct rb_node *node; | |
900 | struct btrfs_ordered_extent *entry = NULL; | |
901 | ||
902 | tree = &BTRFS_I(inode)->ordered_tree; | |
5fd02043 | 903 | spin_lock_irq(&tree->lock); |
e6dcd2dc CM |
904 | node = tree_search(tree, file_offset); |
905 | if (!node) | |
906 | goto out; | |
907 | ||
908 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
909 | atomic_inc(&entry->refs); | |
910 | out: | |
5fd02043 | 911 | spin_unlock_irq(&tree->lock); |
e6dcd2dc | 912 | return entry; |
81d7ed29 | 913 | } |
dbe674a9 | 914 | |
eb84ae03 CM |
915 | /* |
916 | * After an extent is done, call this to conditionally update the on disk | |
917 | * i_size. i_size is updated to cover any fully written part of the file. | |
918 | */ | |
c2167754 | 919 | int btrfs_ordered_update_i_size(struct inode *inode, u64 offset, |
dbe674a9 CM |
920 | struct btrfs_ordered_extent *ordered) |
921 | { | |
922 | struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; | |
dbe674a9 CM |
923 | u64 disk_i_size; |
924 | u64 new_i_size; | |
c2167754 | 925 | u64 i_size = i_size_read(inode); |
dbe674a9 | 926 | struct rb_node *node; |
c2167754 | 927 | struct rb_node *prev = NULL; |
dbe674a9 | 928 | struct btrfs_ordered_extent *test; |
c2167754 YZ |
929 | int ret = 1; |
930 | ||
77cef2ec JB |
931 | spin_lock_irq(&tree->lock); |
932 | if (ordered) { | |
c2167754 | 933 | offset = entry_end(ordered); |
77cef2ec JB |
934 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags)) |
935 | offset = min(offset, | |
936 | ordered->file_offset + | |
937 | ordered->truncated_len); | |
938 | } else { | |
a038fab0 | 939 | offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize); |
77cef2ec | 940 | } |
dbe674a9 CM |
941 | disk_i_size = BTRFS_I(inode)->disk_i_size; |
942 | ||
c2167754 YZ |
943 | /* truncate file */ |
944 | if (disk_i_size > i_size) { | |
945 | BTRFS_I(inode)->disk_i_size = i_size; | |
946 | ret = 0; | |
947 | goto out; | |
948 | } | |
949 | ||
dbe674a9 CM |
950 | /* |
951 | * if the disk i_size is already at the inode->i_size, or | |
952 | * this ordered extent is inside the disk i_size, we're done | |
953 | */ | |
5d1f4020 JB |
954 | if (disk_i_size == i_size) |
955 | goto out; | |
956 | ||
957 | /* | |
958 | * We still need to update disk_i_size if outstanding_isize is greater | |
959 | * than disk_i_size. | |
960 | */ | |
961 | if (offset <= disk_i_size && | |
962 | (!ordered || ordered->outstanding_isize <= disk_i_size)) | |
dbe674a9 | 963 | goto out; |
dbe674a9 | 964 | |
dbe674a9 CM |
965 | /* |
966 | * walk backward from this ordered extent to disk_i_size. | |
967 | * if we find an ordered extent then we can't update disk i_size | |
968 | * yet | |
969 | */ | |
c2167754 YZ |
970 | if (ordered) { |
971 | node = rb_prev(&ordered->rb_node); | |
972 | } else { | |
973 | prev = tree_search(tree, offset); | |
974 | /* | |
975 | * we insert file extents without involving ordered struct, | |
976 | * so there should be no ordered struct cover this offset | |
977 | */ | |
978 | if (prev) { | |
979 | test = rb_entry(prev, struct btrfs_ordered_extent, | |
980 | rb_node); | |
981 | BUG_ON(offset_in_entry(test, offset)); | |
982 | } | |
983 | node = prev; | |
984 | } | |
5fd02043 | 985 | for (; node; node = rb_prev(node)) { |
dbe674a9 | 986 | test = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
5fd02043 JB |
987 | |
988 | /* We treat this entry as if it doesnt exist */ | |
989 | if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags)) | |
990 | continue; | |
dbe674a9 CM |
991 | if (test->file_offset + test->len <= disk_i_size) |
992 | break; | |
c2167754 | 993 | if (test->file_offset >= i_size) |
dbe674a9 | 994 | break; |
59fe4f41 | 995 | if (entry_end(test) > disk_i_size) { |
b9a8cc5b MX |
996 | /* |
997 | * we don't update disk_i_size now, so record this | |
998 | * undealt i_size. Or we will not know the real | |
999 | * i_size. | |
1000 | */ | |
1001 | if (test->outstanding_isize < offset) | |
1002 | test->outstanding_isize = offset; | |
1003 | if (ordered && | |
1004 | ordered->outstanding_isize > | |
1005 | test->outstanding_isize) | |
1006 | test->outstanding_isize = | |
1007 | ordered->outstanding_isize; | |
dbe674a9 | 1008 | goto out; |
5fd02043 | 1009 | } |
dbe674a9 | 1010 | } |
b9a8cc5b | 1011 | new_i_size = min_t(u64, offset, i_size); |
dbe674a9 CM |
1012 | |
1013 | /* | |
b9a8cc5b MX |
1014 | * Some ordered extents may completed before the current one, and |
1015 | * we hold the real i_size in ->outstanding_isize. | |
dbe674a9 | 1016 | */ |
b9a8cc5b MX |
1017 | if (ordered && ordered->outstanding_isize > new_i_size) |
1018 | new_i_size = min_t(u64, ordered->outstanding_isize, i_size); | |
dbe674a9 | 1019 | BTRFS_I(inode)->disk_i_size = new_i_size; |
c2167754 | 1020 | ret = 0; |
dbe674a9 | 1021 | out: |
c2167754 | 1022 | /* |
5fd02043 JB |
1023 | * We need to do this because we can't remove ordered extents until |
1024 | * after the i_disk_size has been updated and then the inode has been | |
1025 | * updated to reflect the change, so we need to tell anybody who finds | |
1026 | * this ordered extent that we've already done all the real work, we | |
1027 | * just haven't completed all the other work. | |
c2167754 YZ |
1028 | */ |
1029 | if (ordered) | |
5fd02043 JB |
1030 | set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags); |
1031 | spin_unlock_irq(&tree->lock); | |
c2167754 | 1032 | return ret; |
dbe674a9 | 1033 | } |
ba1da2f4 | 1034 | |
eb84ae03 CM |
1035 | /* |
1036 | * search the ordered extents for one corresponding to 'offset' and | |
1037 | * try to find a checksum. This is used because we allow pages to | |
1038 | * be reclaimed before their checksum is actually put into the btree | |
1039 | */ | |
d20f7043 | 1040 | int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr, |
e4100d98 | 1041 | u32 *sum, int len) |
ba1da2f4 CM |
1042 | { |
1043 | struct btrfs_ordered_sum *ordered_sum; | |
ba1da2f4 CM |
1044 | struct btrfs_ordered_extent *ordered; |
1045 | struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; | |
3edf7d33 CM |
1046 | unsigned long num_sectors; |
1047 | unsigned long i; | |
1048 | u32 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
e4100d98 | 1049 | int index = 0; |
ba1da2f4 CM |
1050 | |
1051 | ordered = btrfs_lookup_ordered_extent(inode, offset); | |
1052 | if (!ordered) | |
e4100d98 | 1053 | return 0; |
ba1da2f4 | 1054 | |
5fd02043 | 1055 | spin_lock_irq(&tree->lock); |
c6e30871 | 1056 | list_for_each_entry_reverse(ordered_sum, &ordered->list, list) { |
e4100d98 MX |
1057 | if (disk_bytenr >= ordered_sum->bytenr && |
1058 | disk_bytenr < ordered_sum->bytenr + ordered_sum->len) { | |
1059 | i = (disk_bytenr - ordered_sum->bytenr) >> | |
1060 | inode->i_sb->s_blocksize_bits; | |
e4100d98 MX |
1061 | num_sectors = ordered_sum->len >> |
1062 | inode->i_sb->s_blocksize_bits; | |
f51a4a18 MX |
1063 | num_sectors = min_t(int, len - index, num_sectors - i); |
1064 | memcpy(sum + index, ordered_sum->sums + i, | |
1065 | num_sectors); | |
1066 | ||
1067 | index += (int)num_sectors; | |
1068 | if (index == len) | |
1069 | goto out; | |
1070 | disk_bytenr += num_sectors * sectorsize; | |
ba1da2f4 CM |
1071 | } |
1072 | } | |
1073 | out: | |
5fd02043 | 1074 | spin_unlock_irq(&tree->lock); |
89642229 | 1075 | btrfs_put_ordered_extent(ordered); |
e4100d98 | 1076 | return index; |
ba1da2f4 CM |
1077 | } |
1078 | ||
f421950f | 1079 | |
5a3f23d5 CM |
1080 | /* |
1081 | * add a given inode to the list of inodes that must be fully on | |
1082 | * disk before a transaction commit finishes. | |
1083 | * | |
1084 | * This basically gives us the ext3 style data=ordered mode, and it is mostly | |
1085 | * used to make sure renamed files are fully on disk. | |
1086 | * | |
1087 | * It is a noop if the inode is already fully on disk. | |
1088 | * | |
1089 | * If trans is not null, we'll do a friendly check for a transaction that | |
1090 | * is already flushing things and force the IO down ourselves. | |
1091 | */ | |
143bede5 JM |
1092 | void btrfs_add_ordered_operation(struct btrfs_trans_handle *trans, |
1093 | struct btrfs_root *root, struct inode *inode) | |
5a3f23d5 | 1094 | { |
569e0f35 | 1095 | struct btrfs_transaction *cur_trans = trans->transaction; |
5a3f23d5 CM |
1096 | u64 last_mod; |
1097 | ||
1098 | last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans); | |
1099 | ||
1100 | /* | |
1101 | * if this file hasn't been changed since the last transaction | |
1102 | * commit, we can safely return without doing anything | |
1103 | */ | |
5ede859b | 1104 | if (last_mod <= root->fs_info->last_trans_committed) |
143bede5 | 1105 | return; |
5a3f23d5 | 1106 | |
199c2a9c | 1107 | spin_lock(&root->fs_info->ordered_root_lock); |
5a3f23d5 CM |
1108 | if (list_empty(&BTRFS_I(inode)->ordered_operations)) { |
1109 | list_add_tail(&BTRFS_I(inode)->ordered_operations, | |
569e0f35 | 1110 | &cur_trans->ordered_operations); |
5a3f23d5 | 1111 | } |
199c2a9c | 1112 | spin_unlock(&root->fs_info->ordered_root_lock); |
5a3f23d5 | 1113 | } |
6352b91d MX |
1114 | |
1115 | int __init ordered_data_init(void) | |
1116 | { | |
1117 | btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent", | |
1118 | sizeof(struct btrfs_ordered_extent), 0, | |
1119 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, | |
1120 | NULL); | |
1121 | if (!btrfs_ordered_extent_cache) | |
1122 | return -ENOMEM; | |
25287e0a | 1123 | |
6352b91d MX |
1124 | return 0; |
1125 | } | |
1126 | ||
1127 | void ordered_data_exit(void) | |
1128 | { | |
1129 | if (btrfs_ordered_extent_cache) | |
1130 | kmem_cache_destroy(btrfs_ordered_extent_cache); | |
1131 | } |