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; |
d899e052 | 208 | if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE) |
80ff3856 | 209 | set_bit(type, &entry->flags); |
3eaa2885 | 210 | |
4b46fce2 JB |
211 | if (dio) |
212 | set_bit(BTRFS_ORDERED_DIRECT, &entry->flags); | |
213 | ||
e6dcd2dc CM |
214 | /* one ref for the tree */ |
215 | atomic_set(&entry->refs, 1); | |
216 | init_waitqueue_head(&entry->wait); | |
217 | INIT_LIST_HEAD(&entry->list); | |
3eaa2885 | 218 | INIT_LIST_HEAD(&entry->root_extent_list); |
9afab882 MX |
219 | INIT_LIST_HEAD(&entry->work_list); |
220 | init_completion(&entry->completion); | |
2ab28f32 | 221 | INIT_LIST_HEAD(&entry->log_list); |
dc17ff8f | 222 | |
1abe9b8a | 223 | trace_btrfs_ordered_extent_add(inode, entry); |
224 | ||
5fd02043 | 225 | spin_lock_irq(&tree->lock); |
e6dcd2dc CM |
226 | node = tree_insert(&tree->tree, file_offset, |
227 | &entry->rb_node); | |
43c04fb1 JM |
228 | if (node) |
229 | ordered_data_tree_panic(inode, -EEXIST, file_offset); | |
5fd02043 | 230 | spin_unlock_irq(&tree->lock); |
d397712b | 231 | |
199c2a9c | 232 | spin_lock(&root->ordered_extent_lock); |
3eaa2885 | 233 | list_add_tail(&entry->root_extent_list, |
199c2a9c MX |
234 | &root->ordered_extents); |
235 | root->nr_ordered_extents++; | |
236 | if (root->nr_ordered_extents == 1) { | |
237 | spin_lock(&root->fs_info->ordered_root_lock); | |
238 | BUG_ON(!list_empty(&root->ordered_root)); | |
239 | list_add_tail(&root->ordered_root, | |
240 | &root->fs_info->ordered_roots); | |
241 | spin_unlock(&root->fs_info->ordered_root_lock); | |
242 | } | |
243 | spin_unlock(&root->ordered_extent_lock); | |
3eaa2885 | 244 | |
dc17ff8f CM |
245 | return 0; |
246 | } | |
247 | ||
4b46fce2 JB |
248 | int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset, |
249 | u64 start, u64 len, u64 disk_len, int type) | |
250 | { | |
251 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, | |
261507a0 LZ |
252 | disk_len, type, 0, |
253 | BTRFS_COMPRESS_NONE); | |
4b46fce2 JB |
254 | } |
255 | ||
256 | int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset, | |
257 | u64 start, u64 len, u64 disk_len, int type) | |
258 | { | |
259 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, | |
261507a0 LZ |
260 | disk_len, type, 1, |
261 | BTRFS_COMPRESS_NONE); | |
262 | } | |
263 | ||
264 | int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset, | |
265 | u64 start, u64 len, u64 disk_len, | |
266 | int type, int compress_type) | |
267 | { | |
268 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, | |
269 | disk_len, type, 0, | |
270 | compress_type); | |
4b46fce2 JB |
271 | } |
272 | ||
eb84ae03 CM |
273 | /* |
274 | * Add a struct btrfs_ordered_sum into the list of checksums to be inserted | |
3edf7d33 CM |
275 | * when an ordered extent is finished. If the list covers more than one |
276 | * ordered extent, it is split across multiples. | |
eb84ae03 | 277 | */ |
143bede5 JM |
278 | void btrfs_add_ordered_sum(struct inode *inode, |
279 | struct btrfs_ordered_extent *entry, | |
280 | struct btrfs_ordered_sum *sum) | |
dc17ff8f | 281 | { |
e6dcd2dc | 282 | struct btrfs_ordered_inode_tree *tree; |
dc17ff8f | 283 | |
e6dcd2dc | 284 | tree = &BTRFS_I(inode)->ordered_tree; |
5fd02043 | 285 | spin_lock_irq(&tree->lock); |
e6dcd2dc | 286 | list_add_tail(&sum->list, &entry->list); |
2ab28f32 JB |
287 | WARN_ON(entry->csum_bytes_left < sum->len); |
288 | entry->csum_bytes_left -= sum->len; | |
289 | if (entry->csum_bytes_left == 0) | |
290 | wake_up(&entry->wait); | |
5fd02043 | 291 | spin_unlock_irq(&tree->lock); |
dc17ff8f CM |
292 | } |
293 | ||
163cf09c CM |
294 | /* |
295 | * this is used to account for finished IO across a given range | |
296 | * of the file. The IO may span ordered extents. If | |
297 | * a given ordered_extent is completely done, 1 is returned, otherwise | |
298 | * 0. | |
299 | * | |
300 | * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used | |
301 | * to make sure this function only returns 1 once for a given ordered extent. | |
302 | * | |
303 | * file_offset is updated to one byte past the range that is recorded as | |
304 | * complete. This allows you to walk forward in the file. | |
305 | */ | |
306 | int btrfs_dec_test_first_ordered_pending(struct inode *inode, | |
307 | struct btrfs_ordered_extent **cached, | |
5fd02043 | 308 | u64 *file_offset, u64 io_size, int uptodate) |
163cf09c CM |
309 | { |
310 | struct btrfs_ordered_inode_tree *tree; | |
311 | struct rb_node *node; | |
312 | struct btrfs_ordered_extent *entry = NULL; | |
313 | int ret; | |
5fd02043 | 314 | unsigned long flags; |
163cf09c CM |
315 | u64 dec_end; |
316 | u64 dec_start; | |
317 | u64 to_dec; | |
318 | ||
319 | tree = &BTRFS_I(inode)->ordered_tree; | |
5fd02043 | 320 | spin_lock_irqsave(&tree->lock, flags); |
163cf09c CM |
321 | node = tree_search(tree, *file_offset); |
322 | if (!node) { | |
323 | ret = 1; | |
324 | goto out; | |
325 | } | |
326 | ||
327 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
328 | if (!offset_in_entry(entry, *file_offset)) { | |
329 | ret = 1; | |
330 | goto out; | |
331 | } | |
332 | ||
333 | dec_start = max(*file_offset, entry->file_offset); | |
334 | dec_end = min(*file_offset + io_size, entry->file_offset + | |
335 | entry->len); | |
336 | *file_offset = dec_end; | |
337 | if (dec_start > dec_end) { | |
338 | printk(KERN_CRIT "bad ordering dec_start %llu end %llu\n", | |
c1c9ff7c | 339 | dec_start, dec_end); |
163cf09c CM |
340 | } |
341 | to_dec = dec_end - dec_start; | |
342 | if (to_dec > entry->bytes_left) { | |
343 | printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n", | |
c1c9ff7c | 344 | entry->bytes_left, to_dec); |
163cf09c CM |
345 | } |
346 | entry->bytes_left -= to_dec; | |
5fd02043 JB |
347 | if (!uptodate) |
348 | set_bit(BTRFS_ORDERED_IOERR, &entry->flags); | |
349 | ||
163cf09c CM |
350 | if (entry->bytes_left == 0) |
351 | ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); | |
352 | else | |
353 | ret = 1; | |
354 | out: | |
355 | if (!ret && cached && entry) { | |
356 | *cached = entry; | |
357 | atomic_inc(&entry->refs); | |
358 | } | |
5fd02043 | 359 | spin_unlock_irqrestore(&tree->lock, flags); |
163cf09c CM |
360 | return ret == 0; |
361 | } | |
362 | ||
eb84ae03 CM |
363 | /* |
364 | * this is used to account for finished IO across a given range | |
365 | * of the file. The IO should not span ordered extents. If | |
366 | * a given ordered_extent is completely done, 1 is returned, otherwise | |
367 | * 0. | |
368 | * | |
369 | * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used | |
370 | * to make sure this function only returns 1 once for a given ordered extent. | |
371 | */ | |
e6dcd2dc | 372 | int btrfs_dec_test_ordered_pending(struct inode *inode, |
5a1a3df1 | 373 | struct btrfs_ordered_extent **cached, |
5fd02043 | 374 | u64 file_offset, u64 io_size, int uptodate) |
dc17ff8f | 375 | { |
e6dcd2dc | 376 | struct btrfs_ordered_inode_tree *tree; |
dc17ff8f | 377 | struct rb_node *node; |
5a1a3df1 | 378 | struct btrfs_ordered_extent *entry = NULL; |
5fd02043 | 379 | unsigned long flags; |
e6dcd2dc CM |
380 | int ret; |
381 | ||
382 | tree = &BTRFS_I(inode)->ordered_tree; | |
5fd02043 JB |
383 | spin_lock_irqsave(&tree->lock, flags); |
384 | if (cached && *cached) { | |
385 | entry = *cached; | |
386 | goto have_entry; | |
387 | } | |
388 | ||
e6dcd2dc | 389 | node = tree_search(tree, file_offset); |
dc17ff8f | 390 | if (!node) { |
e6dcd2dc CM |
391 | ret = 1; |
392 | goto out; | |
dc17ff8f CM |
393 | } |
394 | ||
e6dcd2dc | 395 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
5fd02043 | 396 | have_entry: |
e6dcd2dc CM |
397 | if (!offset_in_entry(entry, file_offset)) { |
398 | ret = 1; | |
399 | goto out; | |
dc17ff8f | 400 | } |
e6dcd2dc | 401 | |
8b62b72b CM |
402 | if (io_size > entry->bytes_left) { |
403 | printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n", | |
c1c9ff7c | 404 | entry->bytes_left, io_size); |
8b62b72b CM |
405 | } |
406 | entry->bytes_left -= io_size; | |
5fd02043 JB |
407 | if (!uptodate) |
408 | set_bit(BTRFS_ORDERED_IOERR, &entry->flags); | |
409 | ||
8b62b72b | 410 | if (entry->bytes_left == 0) |
e6dcd2dc | 411 | ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); |
8b62b72b CM |
412 | else |
413 | ret = 1; | |
e6dcd2dc | 414 | out: |
5a1a3df1 JB |
415 | if (!ret && cached && entry) { |
416 | *cached = entry; | |
417 | atomic_inc(&entry->refs); | |
418 | } | |
5fd02043 | 419 | spin_unlock_irqrestore(&tree->lock, flags); |
e6dcd2dc CM |
420 | return ret == 0; |
421 | } | |
dc17ff8f | 422 | |
2ab28f32 JB |
423 | /* Needs to either be called under a log transaction or the log_mutex */ |
424 | void btrfs_get_logged_extents(struct btrfs_root *log, struct inode *inode) | |
425 | { | |
426 | struct btrfs_ordered_inode_tree *tree; | |
427 | struct btrfs_ordered_extent *ordered; | |
428 | struct rb_node *n; | |
429 | int index = log->log_transid % 2; | |
430 | ||
431 | tree = &BTRFS_I(inode)->ordered_tree; | |
432 | spin_lock_irq(&tree->lock); | |
433 | for (n = rb_first(&tree->tree); n; n = rb_next(n)) { | |
434 | ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node); | |
435 | spin_lock(&log->log_extents_lock[index]); | |
436 | if (list_empty(&ordered->log_list)) { | |
437 | list_add_tail(&ordered->log_list, &log->logged_list[index]); | |
438 | atomic_inc(&ordered->refs); | |
439 | } | |
440 | spin_unlock(&log->log_extents_lock[index]); | |
441 | } | |
442 | spin_unlock_irq(&tree->lock); | |
443 | } | |
444 | ||
445 | void btrfs_wait_logged_extents(struct btrfs_root *log, u64 transid) | |
446 | { | |
447 | struct btrfs_ordered_extent *ordered; | |
448 | int index = transid % 2; | |
449 | ||
450 | spin_lock_irq(&log->log_extents_lock[index]); | |
451 | while (!list_empty(&log->logged_list[index])) { | |
452 | ordered = list_first_entry(&log->logged_list[index], | |
453 | struct btrfs_ordered_extent, | |
454 | log_list); | |
455 | list_del_init(&ordered->log_list); | |
456 | spin_unlock_irq(&log->log_extents_lock[index]); | |
457 | wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE, | |
458 | &ordered->flags)); | |
459 | btrfs_put_ordered_extent(ordered); | |
460 | spin_lock_irq(&log->log_extents_lock[index]); | |
461 | } | |
462 | spin_unlock_irq(&log->log_extents_lock[index]); | |
463 | } | |
464 | ||
465 | void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid) | |
466 | { | |
467 | struct btrfs_ordered_extent *ordered; | |
468 | int index = transid % 2; | |
469 | ||
470 | spin_lock_irq(&log->log_extents_lock[index]); | |
471 | while (!list_empty(&log->logged_list[index])) { | |
472 | ordered = list_first_entry(&log->logged_list[index], | |
473 | struct btrfs_ordered_extent, | |
474 | log_list); | |
475 | list_del_init(&ordered->log_list); | |
476 | spin_unlock_irq(&log->log_extents_lock[index]); | |
477 | btrfs_put_ordered_extent(ordered); | |
478 | spin_lock_irq(&log->log_extents_lock[index]); | |
479 | } | |
480 | spin_unlock_irq(&log->log_extents_lock[index]); | |
481 | } | |
482 | ||
eb84ae03 CM |
483 | /* |
484 | * used to drop a reference on an ordered extent. This will free | |
485 | * the extent if the last reference is dropped | |
486 | */ | |
143bede5 | 487 | void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry) |
e6dcd2dc | 488 | { |
ba1da2f4 CM |
489 | struct list_head *cur; |
490 | struct btrfs_ordered_sum *sum; | |
491 | ||
1abe9b8a | 492 | trace_btrfs_ordered_extent_put(entry->inode, entry); |
493 | ||
ba1da2f4 | 494 | if (atomic_dec_and_test(&entry->refs)) { |
5fd02043 JB |
495 | if (entry->inode) |
496 | btrfs_add_delayed_iput(entry->inode); | |
d397712b | 497 | while (!list_empty(&entry->list)) { |
ba1da2f4 CM |
498 | cur = entry->list.next; |
499 | sum = list_entry(cur, struct btrfs_ordered_sum, list); | |
500 | list_del(&sum->list); | |
501 | kfree(sum); | |
502 | } | |
6352b91d | 503 | kmem_cache_free(btrfs_ordered_extent_cache, entry); |
ba1da2f4 | 504 | } |
dc17ff8f | 505 | } |
cee36a03 | 506 | |
eb84ae03 CM |
507 | /* |
508 | * remove an ordered extent from the tree. No references are dropped | |
5fd02043 | 509 | * and waiters are woken up. |
eb84ae03 | 510 | */ |
5fd02043 JB |
511 | void btrfs_remove_ordered_extent(struct inode *inode, |
512 | struct btrfs_ordered_extent *entry) | |
cee36a03 | 513 | { |
e6dcd2dc | 514 | struct btrfs_ordered_inode_tree *tree; |
287a0ab9 | 515 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cee36a03 | 516 | struct rb_node *node; |
cee36a03 | 517 | |
e6dcd2dc | 518 | tree = &BTRFS_I(inode)->ordered_tree; |
5fd02043 | 519 | spin_lock_irq(&tree->lock); |
e6dcd2dc | 520 | node = &entry->rb_node; |
cee36a03 | 521 | rb_erase(node, &tree->tree); |
e6dcd2dc CM |
522 | tree->last = NULL; |
523 | set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags); | |
5fd02043 | 524 | spin_unlock_irq(&tree->lock); |
3eaa2885 | 525 | |
199c2a9c | 526 | spin_lock(&root->ordered_extent_lock); |
3eaa2885 | 527 | list_del_init(&entry->root_extent_list); |
199c2a9c | 528 | root->nr_ordered_extents--; |
5a3f23d5 | 529 | |
1abe9b8a | 530 | trace_btrfs_ordered_extent_remove(inode, entry); |
531 | ||
5a3f23d5 CM |
532 | /* |
533 | * we have no more ordered extents for this inode and | |
534 | * no dirty pages. We can safely remove it from the | |
535 | * list of ordered extents | |
536 | */ | |
537 | if (RB_EMPTY_ROOT(&tree->tree) && | |
538 | !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { | |
539 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
540 | } | |
199c2a9c MX |
541 | |
542 | if (!root->nr_ordered_extents) { | |
543 | spin_lock(&root->fs_info->ordered_root_lock); | |
544 | BUG_ON(list_empty(&root->ordered_root)); | |
545 | list_del_init(&root->ordered_root); | |
546 | spin_unlock(&root->fs_info->ordered_root_lock); | |
547 | } | |
548 | spin_unlock(&root->ordered_extent_lock); | |
e6dcd2dc | 549 | wake_up(&entry->wait); |
cee36a03 CM |
550 | } |
551 | ||
9afab882 MX |
552 | static void btrfs_run_ordered_extent_work(struct btrfs_work *work) |
553 | { | |
554 | struct btrfs_ordered_extent *ordered; | |
555 | ||
556 | ordered = container_of(work, struct btrfs_ordered_extent, flush_work); | |
557 | btrfs_start_ordered_extent(ordered->inode, ordered, 1); | |
558 | complete(&ordered->completion); | |
559 | } | |
560 | ||
d352ac68 CM |
561 | /* |
562 | * wait for all the ordered extents in a root. This is done when balancing | |
563 | * space between drives. | |
564 | */ | |
6bbe3a9c | 565 | void btrfs_wait_ordered_extents(struct btrfs_root *root, int delay_iput) |
3eaa2885 | 566 | { |
9afab882 | 567 | struct list_head splice, works; |
9afab882 | 568 | struct btrfs_ordered_extent *ordered, *next; |
3eaa2885 CM |
569 | struct inode *inode; |
570 | ||
571 | INIT_LIST_HEAD(&splice); | |
9afab882 | 572 | INIT_LIST_HEAD(&works); |
3eaa2885 | 573 | |
db1d607d | 574 | mutex_lock(&root->fs_info->ordered_operations_mutex); |
199c2a9c MX |
575 | spin_lock(&root->ordered_extent_lock); |
576 | list_splice_init(&root->ordered_extents, &splice); | |
5b21f2ed | 577 | while (!list_empty(&splice)) { |
199c2a9c MX |
578 | ordered = list_first_entry(&splice, struct btrfs_ordered_extent, |
579 | root_extent_list); | |
580 | list_move_tail(&ordered->root_extent_list, | |
581 | &root->ordered_extents); | |
3eaa2885 | 582 | /* |
5b21f2ed | 583 | * the inode may be getting freed (in sys_unlink path). |
3eaa2885 | 584 | */ |
5b21f2ed | 585 | inode = igrab(ordered->inode); |
199c2a9c MX |
586 | if (!inode) { |
587 | cond_resched_lock(&root->ordered_extent_lock); | |
588 | continue; | |
589 | } | |
5b21f2ed | 590 | |
199c2a9c MX |
591 | atomic_inc(&ordered->refs); |
592 | spin_unlock(&root->ordered_extent_lock); | |
3eaa2885 | 593 | |
199c2a9c MX |
594 | ordered->flush_work.func = btrfs_run_ordered_extent_work; |
595 | list_add_tail(&ordered->work_list, &works); | |
596 | btrfs_queue_worker(&root->fs_info->flush_workers, | |
597 | &ordered->flush_work); | |
3eaa2885 | 598 | |
9afab882 | 599 | cond_resched(); |
199c2a9c | 600 | spin_lock(&root->ordered_extent_lock); |
3eaa2885 | 601 | } |
199c2a9c | 602 | spin_unlock(&root->ordered_extent_lock); |
9afab882 MX |
603 | |
604 | list_for_each_entry_safe(ordered, next, &works, work_list) { | |
605 | list_del_init(&ordered->work_list); | |
606 | wait_for_completion(&ordered->completion); | |
607 | ||
608 | inode = ordered->inode; | |
609 | btrfs_put_ordered_extent(ordered); | |
610 | if (delay_iput) | |
611 | btrfs_add_delayed_iput(inode); | |
612 | else | |
613 | iput(inode); | |
614 | ||
615 | cond_resched(); | |
616 | } | |
db1d607d | 617 | mutex_unlock(&root->fs_info->ordered_operations_mutex); |
3eaa2885 CM |
618 | } |
619 | ||
199c2a9c MX |
620 | void btrfs_wait_all_ordered_extents(struct btrfs_fs_info *fs_info, |
621 | int delay_iput) | |
622 | { | |
623 | struct btrfs_root *root; | |
624 | struct list_head splice; | |
625 | ||
626 | INIT_LIST_HEAD(&splice); | |
627 | ||
628 | spin_lock(&fs_info->ordered_root_lock); | |
629 | list_splice_init(&fs_info->ordered_roots, &splice); | |
630 | while (!list_empty(&splice)) { | |
631 | root = list_first_entry(&splice, struct btrfs_root, | |
632 | ordered_root); | |
633 | root = btrfs_grab_fs_root(root); | |
634 | BUG_ON(!root); | |
635 | list_move_tail(&root->ordered_root, | |
636 | &fs_info->ordered_roots); | |
637 | spin_unlock(&fs_info->ordered_root_lock); | |
638 | ||
639 | btrfs_wait_ordered_extents(root, delay_iput); | |
640 | btrfs_put_fs_root(root); | |
641 | ||
642 | spin_lock(&fs_info->ordered_root_lock); | |
643 | } | |
644 | spin_unlock(&fs_info->ordered_root_lock); | |
645 | } | |
646 | ||
5a3f23d5 CM |
647 | /* |
648 | * this is used during transaction commit to write all the inodes | |
649 | * added to the ordered operation list. These files must be fully on | |
650 | * disk before the transaction commits. | |
651 | * | |
652 | * we have two modes here, one is to just start the IO via filemap_flush | |
653 | * and the other is to wait for all the io. When we wait, we have an | |
654 | * extra check to make sure the ordered operation list really is empty | |
655 | * before we return | |
656 | */ | |
569e0f35 JB |
657 | int btrfs_run_ordered_operations(struct btrfs_trans_handle *trans, |
658 | struct btrfs_root *root, int wait) | |
5a3f23d5 CM |
659 | { |
660 | struct btrfs_inode *btrfs_inode; | |
661 | struct inode *inode; | |
569e0f35 | 662 | struct btrfs_transaction *cur_trans = trans->transaction; |
5a3f23d5 | 663 | struct list_head splice; |
25287e0a MX |
664 | struct list_head works; |
665 | struct btrfs_delalloc_work *work, *next; | |
666 | int ret = 0; | |
5a3f23d5 CM |
667 | |
668 | INIT_LIST_HEAD(&splice); | |
25287e0a | 669 | INIT_LIST_HEAD(&works); |
5a3f23d5 | 670 | |
9ffba8cd | 671 | mutex_lock(&root->fs_info->ordered_extent_flush_mutex); |
199c2a9c | 672 | spin_lock(&root->fs_info->ordered_root_lock); |
569e0f35 | 673 | list_splice_init(&cur_trans->ordered_operations, &splice); |
5a3f23d5 CM |
674 | while (!list_empty(&splice)) { |
675 | btrfs_inode = list_entry(splice.next, struct btrfs_inode, | |
676 | ordered_operations); | |
5a3f23d5 CM |
677 | inode = &btrfs_inode->vfs_inode; |
678 | ||
679 | list_del_init(&btrfs_inode->ordered_operations); | |
680 | ||
681 | /* | |
682 | * the inode may be getting freed (in sys_unlink path). | |
683 | */ | |
684 | inode = igrab(inode); | |
25287e0a MX |
685 | if (!inode) |
686 | continue; | |
5b947f1b MX |
687 | |
688 | if (!wait) | |
689 | list_add_tail(&BTRFS_I(inode)->ordered_operations, | |
569e0f35 | 690 | &cur_trans->ordered_operations); |
199c2a9c | 691 | spin_unlock(&root->fs_info->ordered_root_lock); |
5a3f23d5 | 692 | |
25287e0a MX |
693 | work = btrfs_alloc_delalloc_work(inode, wait, 1); |
694 | if (!work) { | |
199c2a9c | 695 | spin_lock(&root->fs_info->ordered_root_lock); |
25287e0a MX |
696 | if (list_empty(&BTRFS_I(inode)->ordered_operations)) |
697 | list_add_tail(&btrfs_inode->ordered_operations, | |
698 | &splice); | |
25287e0a | 699 | list_splice_tail(&splice, |
569e0f35 | 700 | &cur_trans->ordered_operations); |
199c2a9c | 701 | spin_unlock(&root->fs_info->ordered_root_lock); |
25287e0a MX |
702 | ret = -ENOMEM; |
703 | goto out; | |
5a3f23d5 | 704 | } |
25287e0a MX |
705 | list_add_tail(&work->list, &works); |
706 | btrfs_queue_worker(&root->fs_info->flush_workers, | |
707 | &work->work); | |
5a3f23d5 CM |
708 | |
709 | cond_resched(); | |
199c2a9c | 710 | spin_lock(&root->fs_info->ordered_root_lock); |
5a3f23d5 | 711 | } |
199c2a9c | 712 | spin_unlock(&root->fs_info->ordered_root_lock); |
25287e0a MX |
713 | out: |
714 | list_for_each_entry_safe(work, next, &works, list) { | |
715 | list_del_init(&work->list); | |
716 | btrfs_wait_and_free_delalloc_work(work); | |
717 | } | |
9ffba8cd | 718 | mutex_unlock(&root->fs_info->ordered_extent_flush_mutex); |
25287e0a | 719 | return ret; |
5a3f23d5 CM |
720 | } |
721 | ||
eb84ae03 CM |
722 | /* |
723 | * Used to start IO or wait for a given ordered extent to finish. | |
724 | * | |
725 | * If wait is one, this effectively waits on page writeback for all the pages | |
726 | * in the extent, and it waits on the io completion code to insert | |
727 | * metadata into the btree corresponding to the extent | |
728 | */ | |
729 | void btrfs_start_ordered_extent(struct inode *inode, | |
730 | struct btrfs_ordered_extent *entry, | |
731 | int wait) | |
e6dcd2dc CM |
732 | { |
733 | u64 start = entry->file_offset; | |
734 | u64 end = start + entry->len - 1; | |
e1b81e67 | 735 | |
1abe9b8a | 736 | trace_btrfs_ordered_extent_start(inode, entry); |
737 | ||
eb84ae03 CM |
738 | /* |
739 | * pages in the range can be dirty, clean or writeback. We | |
740 | * start IO on any dirty ones so the wait doesn't stall waiting | |
b2570314 | 741 | * for the flusher thread to find them |
eb84ae03 | 742 | */ |
4b46fce2 JB |
743 | if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags)) |
744 | filemap_fdatawrite_range(inode->i_mapping, start, end); | |
c8b97818 | 745 | if (wait) { |
e6dcd2dc CM |
746 | wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE, |
747 | &entry->flags)); | |
c8b97818 | 748 | } |
e6dcd2dc | 749 | } |
cee36a03 | 750 | |
eb84ae03 CM |
751 | /* |
752 | * Used to wait on ordered extents across a large range of bytes. | |
753 | */ | |
143bede5 | 754 | void btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len) |
e6dcd2dc CM |
755 | { |
756 | u64 end; | |
e5a2217e | 757 | u64 orig_end; |
e6dcd2dc | 758 | struct btrfs_ordered_extent *ordered; |
e5a2217e CM |
759 | |
760 | if (start + len < start) { | |
f421950f | 761 | orig_end = INT_LIMIT(loff_t); |
e5a2217e CM |
762 | } else { |
763 | orig_end = start + len - 1; | |
f421950f CM |
764 | if (orig_end > INT_LIMIT(loff_t)) |
765 | orig_end = INT_LIMIT(loff_t); | |
e5a2217e | 766 | } |
551ebb2d | 767 | |
e5a2217e CM |
768 | /* start IO across the range first to instantiate any delalloc |
769 | * extents | |
770 | */ | |
7ddf5a42 JB |
771 | filemap_fdatawrite_range(inode->i_mapping, start, orig_end); |
772 | ||
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, | |
788 | &BTRFS_I(inode)->runtime_flags)) | |
789 | filemap_fdatawrite_range(inode->i_mapping, start, orig_end); | |
790 | ||
791 | filemap_fdatawait_range(inode->i_mapping, start, orig_end); | |
e5a2217e | 792 | |
f421950f | 793 | end = orig_end; |
d397712b | 794 | while (1) { |
e6dcd2dc | 795 | ordered = btrfs_lookup_first_ordered_extent(inode, end); |
d397712b | 796 | if (!ordered) |
e6dcd2dc | 797 | break; |
e5a2217e | 798 | if (ordered->file_offset > orig_end) { |
e6dcd2dc CM |
799 | btrfs_put_ordered_extent(ordered); |
800 | break; | |
801 | } | |
802 | if (ordered->file_offset + ordered->len < start) { | |
803 | btrfs_put_ordered_extent(ordered); | |
804 | break; | |
805 | } | |
e5a2217e | 806 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
807 | end = ordered->file_offset; |
808 | btrfs_put_ordered_extent(ordered); | |
e5a2217e | 809 | if (end == 0 || end == start) |
e6dcd2dc CM |
810 | break; |
811 | end--; | |
812 | } | |
cee36a03 CM |
813 | } |
814 | ||
eb84ae03 CM |
815 | /* |
816 | * find an ordered extent corresponding to file_offset. return NULL if | |
817 | * nothing is found, otherwise take a reference on the extent and return it | |
818 | */ | |
e6dcd2dc CM |
819 | struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode, |
820 | u64 file_offset) | |
821 | { | |
822 | struct btrfs_ordered_inode_tree *tree; | |
823 | struct rb_node *node; | |
824 | struct btrfs_ordered_extent *entry = NULL; | |
825 | ||
826 | tree = &BTRFS_I(inode)->ordered_tree; | |
5fd02043 | 827 | spin_lock_irq(&tree->lock); |
e6dcd2dc CM |
828 | node = tree_search(tree, file_offset); |
829 | if (!node) | |
830 | goto out; | |
831 | ||
832 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
833 | if (!offset_in_entry(entry, file_offset)) | |
834 | entry = NULL; | |
835 | if (entry) | |
836 | atomic_inc(&entry->refs); | |
837 | out: | |
5fd02043 | 838 | spin_unlock_irq(&tree->lock); |
e6dcd2dc CM |
839 | return entry; |
840 | } | |
841 | ||
4b46fce2 JB |
842 | /* Since the DIO code tries to lock a wide area we need to look for any ordered |
843 | * extents that exist in the range, rather than just the start of the range. | |
844 | */ | |
845 | struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode, | |
846 | u64 file_offset, | |
847 | u64 len) | |
848 | { | |
849 | struct btrfs_ordered_inode_tree *tree; | |
850 | struct rb_node *node; | |
851 | struct btrfs_ordered_extent *entry = NULL; | |
852 | ||
853 | tree = &BTRFS_I(inode)->ordered_tree; | |
5fd02043 | 854 | spin_lock_irq(&tree->lock); |
4b46fce2 JB |
855 | node = tree_search(tree, file_offset); |
856 | if (!node) { | |
857 | node = tree_search(tree, file_offset + len); | |
858 | if (!node) | |
859 | goto out; | |
860 | } | |
861 | ||
862 | while (1) { | |
863 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
864 | if (range_overlaps(entry, file_offset, len)) | |
865 | break; | |
866 | ||
867 | if (entry->file_offset >= file_offset + len) { | |
868 | entry = NULL; | |
869 | break; | |
870 | } | |
871 | entry = NULL; | |
872 | node = rb_next(node); | |
873 | if (!node) | |
874 | break; | |
875 | } | |
876 | out: | |
877 | if (entry) | |
878 | atomic_inc(&entry->refs); | |
5fd02043 | 879 | spin_unlock_irq(&tree->lock); |
4b46fce2 JB |
880 | return entry; |
881 | } | |
882 | ||
eb84ae03 CM |
883 | /* |
884 | * lookup and return any extent before 'file_offset'. NULL is returned | |
885 | * if none is found | |
886 | */ | |
e6dcd2dc | 887 | struct btrfs_ordered_extent * |
d397712b | 888 | btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset) |
e6dcd2dc CM |
889 | { |
890 | struct btrfs_ordered_inode_tree *tree; | |
891 | struct rb_node *node; | |
892 | struct btrfs_ordered_extent *entry = NULL; | |
893 | ||
894 | tree = &BTRFS_I(inode)->ordered_tree; | |
5fd02043 | 895 | spin_lock_irq(&tree->lock); |
e6dcd2dc CM |
896 | node = tree_search(tree, file_offset); |
897 | if (!node) | |
898 | goto out; | |
899 | ||
900 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
901 | atomic_inc(&entry->refs); | |
902 | out: | |
5fd02043 | 903 | spin_unlock_irq(&tree->lock); |
e6dcd2dc | 904 | return entry; |
81d7ed29 | 905 | } |
dbe674a9 | 906 | |
eb84ae03 CM |
907 | /* |
908 | * After an extent is done, call this to conditionally update the on disk | |
909 | * i_size. i_size is updated to cover any fully written part of the file. | |
910 | */ | |
c2167754 | 911 | int btrfs_ordered_update_i_size(struct inode *inode, u64 offset, |
dbe674a9 CM |
912 | struct btrfs_ordered_extent *ordered) |
913 | { | |
914 | struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; | |
dbe674a9 CM |
915 | u64 disk_i_size; |
916 | u64 new_i_size; | |
c2167754 | 917 | u64 i_size = i_size_read(inode); |
dbe674a9 | 918 | struct rb_node *node; |
c2167754 | 919 | struct rb_node *prev = NULL; |
dbe674a9 | 920 | struct btrfs_ordered_extent *test; |
c2167754 YZ |
921 | int ret = 1; |
922 | ||
923 | if (ordered) | |
924 | offset = entry_end(ordered); | |
a038fab0 YZ |
925 | else |
926 | offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize); | |
dbe674a9 | 927 | |
5fd02043 | 928 | spin_lock_irq(&tree->lock); |
dbe674a9 CM |
929 | disk_i_size = BTRFS_I(inode)->disk_i_size; |
930 | ||
c2167754 YZ |
931 | /* truncate file */ |
932 | if (disk_i_size > i_size) { | |
933 | BTRFS_I(inode)->disk_i_size = i_size; | |
934 | ret = 0; | |
935 | goto out; | |
936 | } | |
937 | ||
dbe674a9 CM |
938 | /* |
939 | * if the disk i_size is already at the inode->i_size, or | |
940 | * this ordered extent is inside the disk i_size, we're done | |
941 | */ | |
5d1f4020 JB |
942 | if (disk_i_size == i_size) |
943 | goto out; | |
944 | ||
945 | /* | |
946 | * We still need to update disk_i_size if outstanding_isize is greater | |
947 | * than disk_i_size. | |
948 | */ | |
949 | if (offset <= disk_i_size && | |
950 | (!ordered || ordered->outstanding_isize <= disk_i_size)) | |
dbe674a9 | 951 | goto out; |
dbe674a9 | 952 | |
dbe674a9 CM |
953 | /* |
954 | * walk backward from this ordered extent to disk_i_size. | |
955 | * if we find an ordered extent then we can't update disk i_size | |
956 | * yet | |
957 | */ | |
c2167754 YZ |
958 | if (ordered) { |
959 | node = rb_prev(&ordered->rb_node); | |
960 | } else { | |
961 | prev = tree_search(tree, offset); | |
962 | /* | |
963 | * we insert file extents without involving ordered struct, | |
964 | * so there should be no ordered struct cover this offset | |
965 | */ | |
966 | if (prev) { | |
967 | test = rb_entry(prev, struct btrfs_ordered_extent, | |
968 | rb_node); | |
969 | BUG_ON(offset_in_entry(test, offset)); | |
970 | } | |
971 | node = prev; | |
972 | } | |
5fd02043 | 973 | for (; node; node = rb_prev(node)) { |
dbe674a9 | 974 | test = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
5fd02043 JB |
975 | |
976 | /* We treat this entry as if it doesnt exist */ | |
977 | if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags)) | |
978 | continue; | |
dbe674a9 CM |
979 | if (test->file_offset + test->len <= disk_i_size) |
980 | break; | |
c2167754 | 981 | if (test->file_offset >= i_size) |
dbe674a9 | 982 | break; |
59fe4f41 | 983 | if (entry_end(test) > disk_i_size) { |
b9a8cc5b MX |
984 | /* |
985 | * we don't update disk_i_size now, so record this | |
986 | * undealt i_size. Or we will not know the real | |
987 | * i_size. | |
988 | */ | |
989 | if (test->outstanding_isize < offset) | |
990 | test->outstanding_isize = offset; | |
991 | if (ordered && | |
992 | ordered->outstanding_isize > | |
993 | test->outstanding_isize) | |
994 | test->outstanding_isize = | |
995 | ordered->outstanding_isize; | |
dbe674a9 | 996 | goto out; |
5fd02043 | 997 | } |
dbe674a9 | 998 | } |
b9a8cc5b | 999 | new_i_size = min_t(u64, offset, i_size); |
dbe674a9 CM |
1000 | |
1001 | /* | |
b9a8cc5b MX |
1002 | * Some ordered extents may completed before the current one, and |
1003 | * we hold the real i_size in ->outstanding_isize. | |
dbe674a9 | 1004 | */ |
b9a8cc5b MX |
1005 | if (ordered && ordered->outstanding_isize > new_i_size) |
1006 | new_i_size = min_t(u64, ordered->outstanding_isize, i_size); | |
dbe674a9 | 1007 | BTRFS_I(inode)->disk_i_size = new_i_size; |
c2167754 | 1008 | ret = 0; |
dbe674a9 | 1009 | out: |
c2167754 | 1010 | /* |
5fd02043 JB |
1011 | * We need to do this because we can't remove ordered extents until |
1012 | * after the i_disk_size has been updated and then the inode has been | |
1013 | * updated to reflect the change, so we need to tell anybody who finds | |
1014 | * this ordered extent that we've already done all the real work, we | |
1015 | * just haven't completed all the other work. | |
c2167754 YZ |
1016 | */ |
1017 | if (ordered) | |
5fd02043 JB |
1018 | set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags); |
1019 | spin_unlock_irq(&tree->lock); | |
c2167754 | 1020 | return ret; |
dbe674a9 | 1021 | } |
ba1da2f4 | 1022 | |
eb84ae03 CM |
1023 | /* |
1024 | * search the ordered extents for one corresponding to 'offset' and | |
1025 | * try to find a checksum. This is used because we allow pages to | |
1026 | * be reclaimed before their checksum is actually put into the btree | |
1027 | */ | |
d20f7043 | 1028 | int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr, |
e4100d98 | 1029 | u32 *sum, int len) |
ba1da2f4 CM |
1030 | { |
1031 | struct btrfs_ordered_sum *ordered_sum; | |
ba1da2f4 CM |
1032 | struct btrfs_ordered_extent *ordered; |
1033 | struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; | |
3edf7d33 CM |
1034 | unsigned long num_sectors; |
1035 | unsigned long i; | |
1036 | u32 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
e4100d98 | 1037 | int index = 0; |
ba1da2f4 CM |
1038 | |
1039 | ordered = btrfs_lookup_ordered_extent(inode, offset); | |
1040 | if (!ordered) | |
e4100d98 | 1041 | return 0; |
ba1da2f4 | 1042 | |
5fd02043 | 1043 | spin_lock_irq(&tree->lock); |
c6e30871 | 1044 | list_for_each_entry_reverse(ordered_sum, &ordered->list, list) { |
e4100d98 MX |
1045 | if (disk_bytenr >= ordered_sum->bytenr && |
1046 | disk_bytenr < ordered_sum->bytenr + ordered_sum->len) { | |
1047 | i = (disk_bytenr - ordered_sum->bytenr) >> | |
1048 | inode->i_sb->s_blocksize_bits; | |
e4100d98 MX |
1049 | num_sectors = ordered_sum->len >> |
1050 | inode->i_sb->s_blocksize_bits; | |
f51a4a18 MX |
1051 | num_sectors = min_t(int, len - index, num_sectors - i); |
1052 | memcpy(sum + index, ordered_sum->sums + i, | |
1053 | num_sectors); | |
1054 | ||
1055 | index += (int)num_sectors; | |
1056 | if (index == len) | |
1057 | goto out; | |
1058 | disk_bytenr += num_sectors * sectorsize; | |
ba1da2f4 CM |
1059 | } |
1060 | } | |
1061 | out: | |
5fd02043 | 1062 | spin_unlock_irq(&tree->lock); |
89642229 | 1063 | btrfs_put_ordered_extent(ordered); |
e4100d98 | 1064 | return index; |
ba1da2f4 CM |
1065 | } |
1066 | ||
f421950f | 1067 | |
5a3f23d5 CM |
1068 | /* |
1069 | * add a given inode to the list of inodes that must be fully on | |
1070 | * disk before a transaction commit finishes. | |
1071 | * | |
1072 | * This basically gives us the ext3 style data=ordered mode, and it is mostly | |
1073 | * used to make sure renamed files are fully on disk. | |
1074 | * | |
1075 | * It is a noop if the inode is already fully on disk. | |
1076 | * | |
1077 | * If trans is not null, we'll do a friendly check for a transaction that | |
1078 | * is already flushing things and force the IO down ourselves. | |
1079 | */ | |
143bede5 JM |
1080 | void btrfs_add_ordered_operation(struct btrfs_trans_handle *trans, |
1081 | struct btrfs_root *root, struct inode *inode) | |
5a3f23d5 | 1082 | { |
569e0f35 | 1083 | struct btrfs_transaction *cur_trans = trans->transaction; |
5a3f23d5 CM |
1084 | u64 last_mod; |
1085 | ||
1086 | last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans); | |
1087 | ||
1088 | /* | |
1089 | * if this file hasn't been changed since the last transaction | |
1090 | * commit, we can safely return without doing anything | |
1091 | */ | |
1092 | if (last_mod < root->fs_info->last_trans_committed) | |
143bede5 | 1093 | return; |
5a3f23d5 | 1094 | |
199c2a9c | 1095 | spin_lock(&root->fs_info->ordered_root_lock); |
5a3f23d5 CM |
1096 | if (list_empty(&BTRFS_I(inode)->ordered_operations)) { |
1097 | list_add_tail(&BTRFS_I(inode)->ordered_operations, | |
569e0f35 | 1098 | &cur_trans->ordered_operations); |
5a3f23d5 | 1099 | } |
199c2a9c | 1100 | spin_unlock(&root->fs_info->ordered_root_lock); |
5a3f23d5 | 1101 | } |
6352b91d MX |
1102 | |
1103 | int __init ordered_data_init(void) | |
1104 | { | |
1105 | btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent", | |
1106 | sizeof(struct btrfs_ordered_extent), 0, | |
1107 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, | |
1108 | NULL); | |
1109 | if (!btrfs_ordered_extent_cache) | |
1110 | return -ENOMEM; | |
25287e0a | 1111 | |
6352b91d MX |
1112 | return 0; |
1113 | } | |
1114 | ||
1115 | void ordered_data_exit(void) | |
1116 | { | |
1117 | if (btrfs_ordered_extent_cache) | |
1118 | kmem_cache_destroy(btrfs_ordered_extent_cache); | |
1119 | } |