Commit | Line | Data |
---|---|---|
dc17ff8f CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
19 | #include <linux/gfp.h> | |
20 | #include <linux/slab.h> | |
d6bfde87 | 21 | #include <linux/blkdev.h> |
f421950f CM |
22 | #include <linux/writeback.h> |
23 | #include <linux/pagevec.h> | |
dc17ff8f CM |
24 | #include "ctree.h" |
25 | #include "transaction.h" | |
26 | #include "btrfs_inode.h" | |
e6dcd2dc | 27 | #include "extent_io.h" |
dc17ff8f | 28 | |
e6dcd2dc | 29 | static u64 entry_end(struct btrfs_ordered_extent *entry) |
dc17ff8f | 30 | { |
e6dcd2dc CM |
31 | if (entry->file_offset + entry->len < entry->file_offset) |
32 | return (u64)-1; | |
33 | return entry->file_offset + entry->len; | |
dc17ff8f CM |
34 | } |
35 | ||
d352ac68 CM |
36 | /* returns NULL if the insertion worked, or it returns the node it did find |
37 | * in the tree | |
38 | */ | |
e6dcd2dc CM |
39 | static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset, |
40 | struct rb_node *node) | |
dc17ff8f | 41 | { |
d397712b CM |
42 | struct rb_node **p = &root->rb_node; |
43 | struct rb_node *parent = NULL; | |
e6dcd2dc | 44 | struct btrfs_ordered_extent *entry; |
dc17ff8f | 45 | |
d397712b | 46 | while (*p) { |
dc17ff8f | 47 | parent = *p; |
e6dcd2dc | 48 | entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node); |
dc17ff8f | 49 | |
e6dcd2dc | 50 | if (file_offset < entry->file_offset) |
dc17ff8f | 51 | p = &(*p)->rb_left; |
e6dcd2dc | 52 | else if (file_offset >= entry_end(entry)) |
dc17ff8f CM |
53 | p = &(*p)->rb_right; |
54 | else | |
55 | return parent; | |
56 | } | |
57 | ||
58 | rb_link_node(node, parent, p); | |
59 | rb_insert_color(node, root); | |
60 | return NULL; | |
61 | } | |
62 | ||
d352ac68 CM |
63 | /* |
64 | * look for a given offset in the tree, and if it can't be found return the | |
65 | * first lesser offset | |
66 | */ | |
e6dcd2dc CM |
67 | static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset, |
68 | struct rb_node **prev_ret) | |
dc17ff8f | 69 | { |
d397712b | 70 | struct rb_node *n = root->rb_node; |
dc17ff8f | 71 | struct rb_node *prev = NULL; |
e6dcd2dc CM |
72 | struct rb_node *test; |
73 | struct btrfs_ordered_extent *entry; | |
74 | struct btrfs_ordered_extent *prev_entry = NULL; | |
dc17ff8f | 75 | |
d397712b | 76 | while (n) { |
e6dcd2dc | 77 | entry = rb_entry(n, struct btrfs_ordered_extent, rb_node); |
dc17ff8f CM |
78 | prev = n; |
79 | prev_entry = entry; | |
dc17ff8f | 80 | |
e6dcd2dc | 81 | if (file_offset < entry->file_offset) |
dc17ff8f | 82 | n = n->rb_left; |
e6dcd2dc | 83 | else if (file_offset >= entry_end(entry)) |
dc17ff8f CM |
84 | n = n->rb_right; |
85 | else | |
86 | return n; | |
87 | } | |
88 | if (!prev_ret) | |
89 | return NULL; | |
90 | ||
d397712b | 91 | while (prev && file_offset >= entry_end(prev_entry)) { |
e6dcd2dc CM |
92 | test = rb_next(prev); |
93 | if (!test) | |
94 | break; | |
95 | prev_entry = rb_entry(test, struct btrfs_ordered_extent, | |
96 | rb_node); | |
97 | if (file_offset < entry_end(prev_entry)) | |
98 | break; | |
99 | ||
100 | prev = test; | |
101 | } | |
102 | if (prev) | |
103 | prev_entry = rb_entry(prev, struct btrfs_ordered_extent, | |
104 | rb_node); | |
d397712b | 105 | while (prev && file_offset < entry_end(prev_entry)) { |
e6dcd2dc CM |
106 | test = rb_prev(prev); |
107 | if (!test) | |
108 | break; | |
109 | prev_entry = rb_entry(test, struct btrfs_ordered_extent, | |
110 | rb_node); | |
111 | prev = test; | |
dc17ff8f CM |
112 | } |
113 | *prev_ret = prev; | |
114 | return NULL; | |
115 | } | |
116 | ||
d352ac68 CM |
117 | /* |
118 | * helper to check if a given offset is inside a given entry | |
119 | */ | |
e6dcd2dc CM |
120 | static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset) |
121 | { | |
122 | if (file_offset < entry->file_offset || | |
123 | entry->file_offset + entry->len <= file_offset) | |
124 | return 0; | |
125 | return 1; | |
126 | } | |
127 | ||
d352ac68 CM |
128 | /* |
129 | * look find the first ordered struct that has this offset, otherwise | |
130 | * the first one less than this offset | |
131 | */ | |
e6dcd2dc CM |
132 | static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree, |
133 | u64 file_offset) | |
dc17ff8f | 134 | { |
e6dcd2dc | 135 | struct rb_root *root = &tree->tree; |
dc17ff8f CM |
136 | struct rb_node *prev; |
137 | struct rb_node *ret; | |
e6dcd2dc CM |
138 | struct btrfs_ordered_extent *entry; |
139 | ||
140 | if (tree->last) { | |
141 | entry = rb_entry(tree->last, struct btrfs_ordered_extent, | |
142 | rb_node); | |
143 | if (offset_in_entry(entry, file_offset)) | |
144 | return tree->last; | |
145 | } | |
146 | ret = __tree_search(root, file_offset, &prev); | |
dc17ff8f | 147 | if (!ret) |
e6dcd2dc CM |
148 | ret = prev; |
149 | if (ret) | |
150 | tree->last = ret; | |
dc17ff8f CM |
151 | return ret; |
152 | } | |
153 | ||
eb84ae03 CM |
154 | /* allocate and add a new ordered_extent into the per-inode tree. |
155 | * file_offset is the logical offset in the file | |
156 | * | |
157 | * start is the disk block number of an extent already reserved in the | |
158 | * extent allocation tree | |
159 | * | |
160 | * len is the length of the extent | |
161 | * | |
162 | * This also sets the EXTENT_ORDERED bit on the range in the inode. | |
163 | * | |
164 | * The tree is given a single reference on the ordered extent that was | |
165 | * inserted. | |
166 | */ | |
e6dcd2dc | 167 | int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset, |
80ff3856 | 168 | u64 start, u64 len, u64 disk_len, int type) |
dc17ff8f | 169 | { |
dc17ff8f | 170 | struct btrfs_ordered_inode_tree *tree; |
e6dcd2dc CM |
171 | struct rb_node *node; |
172 | struct btrfs_ordered_extent *entry; | |
dc17ff8f | 173 | |
e6dcd2dc CM |
174 | tree = &BTRFS_I(inode)->ordered_tree; |
175 | entry = kzalloc(sizeof(*entry), GFP_NOFS); | |
dc17ff8f CM |
176 | if (!entry) |
177 | return -ENOMEM; | |
178 | ||
e6dcd2dc CM |
179 | mutex_lock(&tree->mutex); |
180 | entry->file_offset = file_offset; | |
181 | entry->start = start; | |
182 | entry->len = len; | |
c8b97818 | 183 | entry->disk_len = disk_len; |
3eaa2885 | 184 | entry->inode = inode; |
d899e052 | 185 | if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE) |
80ff3856 | 186 | set_bit(type, &entry->flags); |
3eaa2885 | 187 | |
e6dcd2dc CM |
188 | /* one ref for the tree */ |
189 | atomic_set(&entry->refs, 1); | |
190 | init_waitqueue_head(&entry->wait); | |
191 | INIT_LIST_HEAD(&entry->list); | |
3eaa2885 | 192 | INIT_LIST_HEAD(&entry->root_extent_list); |
dc17ff8f | 193 | |
e6dcd2dc CM |
194 | node = tree_insert(&tree->tree, file_offset, |
195 | &entry->rb_node); | |
d397712b CM |
196 | BUG_ON(node); |
197 | ||
e6dcd2dc CM |
198 | set_extent_ordered(&BTRFS_I(inode)->io_tree, file_offset, |
199 | entry_end(entry) - 1, GFP_NOFS); | |
1b1e2135 | 200 | |
3eaa2885 CM |
201 | spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); |
202 | list_add_tail(&entry->root_extent_list, | |
203 | &BTRFS_I(inode)->root->fs_info->ordered_extents); | |
204 | spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); | |
205 | ||
e6dcd2dc CM |
206 | mutex_unlock(&tree->mutex); |
207 | BUG_ON(node); | |
dc17ff8f CM |
208 | return 0; |
209 | } | |
210 | ||
eb84ae03 CM |
211 | /* |
212 | * Add a struct btrfs_ordered_sum into the list of checksums to be inserted | |
3edf7d33 CM |
213 | * when an ordered extent is finished. If the list covers more than one |
214 | * ordered extent, it is split across multiples. | |
eb84ae03 | 215 | */ |
3edf7d33 CM |
216 | int btrfs_add_ordered_sum(struct inode *inode, |
217 | struct btrfs_ordered_extent *entry, | |
218 | struct btrfs_ordered_sum *sum) | |
dc17ff8f | 219 | { |
e6dcd2dc | 220 | struct btrfs_ordered_inode_tree *tree; |
dc17ff8f | 221 | |
e6dcd2dc CM |
222 | tree = &BTRFS_I(inode)->ordered_tree; |
223 | mutex_lock(&tree->mutex); | |
e6dcd2dc CM |
224 | list_add_tail(&sum->list, &entry->list); |
225 | mutex_unlock(&tree->mutex); | |
226 | return 0; | |
dc17ff8f CM |
227 | } |
228 | ||
eb84ae03 CM |
229 | /* |
230 | * this is used to account for finished IO across a given range | |
231 | * of the file. The IO should not span ordered extents. If | |
232 | * a given ordered_extent is completely done, 1 is returned, otherwise | |
233 | * 0. | |
234 | * | |
235 | * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used | |
236 | * to make sure this function only returns 1 once for a given ordered extent. | |
237 | */ | |
e6dcd2dc CM |
238 | int btrfs_dec_test_ordered_pending(struct inode *inode, |
239 | u64 file_offset, u64 io_size) | |
dc17ff8f | 240 | { |
e6dcd2dc | 241 | struct btrfs_ordered_inode_tree *tree; |
dc17ff8f | 242 | struct rb_node *node; |
e6dcd2dc CM |
243 | struct btrfs_ordered_extent *entry; |
244 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
245 | int ret; | |
246 | ||
247 | tree = &BTRFS_I(inode)->ordered_tree; | |
248 | mutex_lock(&tree->mutex); | |
249 | clear_extent_ordered(io_tree, file_offset, file_offset + io_size - 1, | |
250 | GFP_NOFS); | |
251 | node = tree_search(tree, file_offset); | |
dc17ff8f | 252 | if (!node) { |
e6dcd2dc CM |
253 | ret = 1; |
254 | goto out; | |
dc17ff8f CM |
255 | } |
256 | ||
e6dcd2dc CM |
257 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
258 | if (!offset_in_entry(entry, file_offset)) { | |
259 | ret = 1; | |
260 | goto out; | |
dc17ff8f | 261 | } |
e6dcd2dc CM |
262 | |
263 | ret = test_range_bit(io_tree, entry->file_offset, | |
264 | entry->file_offset + entry->len - 1, | |
265 | EXTENT_ORDERED, 0); | |
e6dcd2dc CM |
266 | if (ret == 0) |
267 | ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); | |
268 | out: | |
269 | mutex_unlock(&tree->mutex); | |
270 | return ret == 0; | |
271 | } | |
dc17ff8f | 272 | |
eb84ae03 CM |
273 | /* |
274 | * used to drop a reference on an ordered extent. This will free | |
275 | * the extent if the last reference is dropped | |
276 | */ | |
e6dcd2dc CM |
277 | int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry) |
278 | { | |
ba1da2f4 CM |
279 | struct list_head *cur; |
280 | struct btrfs_ordered_sum *sum; | |
281 | ||
282 | if (atomic_dec_and_test(&entry->refs)) { | |
d397712b | 283 | while (!list_empty(&entry->list)) { |
ba1da2f4 CM |
284 | cur = entry->list.next; |
285 | sum = list_entry(cur, struct btrfs_ordered_sum, list); | |
286 | list_del(&sum->list); | |
287 | kfree(sum); | |
288 | } | |
e6dcd2dc | 289 | kfree(entry); |
ba1da2f4 | 290 | } |
e6dcd2dc | 291 | return 0; |
dc17ff8f | 292 | } |
cee36a03 | 293 | |
eb84ae03 CM |
294 | /* |
295 | * remove an ordered extent from the tree. No references are dropped | |
296 | * but, anyone waiting on this extent is woken up. | |
297 | */ | |
e6dcd2dc CM |
298 | int btrfs_remove_ordered_extent(struct inode *inode, |
299 | struct btrfs_ordered_extent *entry) | |
cee36a03 | 300 | { |
e6dcd2dc | 301 | struct btrfs_ordered_inode_tree *tree; |
cee36a03 | 302 | struct rb_node *node; |
cee36a03 | 303 | |
e6dcd2dc CM |
304 | tree = &BTRFS_I(inode)->ordered_tree; |
305 | mutex_lock(&tree->mutex); | |
306 | node = &entry->rb_node; | |
cee36a03 | 307 | rb_erase(node, &tree->tree); |
e6dcd2dc CM |
308 | tree->last = NULL; |
309 | set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags); | |
3eaa2885 CM |
310 | |
311 | spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); | |
312 | list_del_init(&entry->root_extent_list); | |
313 | spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); | |
314 | ||
e6dcd2dc CM |
315 | mutex_unlock(&tree->mutex); |
316 | wake_up(&entry->wait); | |
317 | return 0; | |
cee36a03 CM |
318 | } |
319 | ||
d352ac68 CM |
320 | /* |
321 | * wait for all the ordered extents in a root. This is done when balancing | |
322 | * space between drives. | |
323 | */ | |
7ea394f1 | 324 | int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only) |
3eaa2885 CM |
325 | { |
326 | struct list_head splice; | |
327 | struct list_head *cur; | |
328 | struct btrfs_ordered_extent *ordered; | |
329 | struct inode *inode; | |
330 | ||
331 | INIT_LIST_HEAD(&splice); | |
332 | ||
333 | spin_lock(&root->fs_info->ordered_extent_lock); | |
334 | list_splice_init(&root->fs_info->ordered_extents, &splice); | |
5b21f2ed | 335 | while (!list_empty(&splice)) { |
3eaa2885 CM |
336 | cur = splice.next; |
337 | ordered = list_entry(cur, struct btrfs_ordered_extent, | |
338 | root_extent_list); | |
7ea394f1 | 339 | if (nocow_only && |
d899e052 YZ |
340 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) && |
341 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) { | |
5b21f2ed ZY |
342 | list_move(&ordered->root_extent_list, |
343 | &root->fs_info->ordered_extents); | |
7ea394f1 YZ |
344 | cond_resched_lock(&root->fs_info->ordered_extent_lock); |
345 | continue; | |
346 | } | |
347 | ||
3eaa2885 CM |
348 | list_del_init(&ordered->root_extent_list); |
349 | atomic_inc(&ordered->refs); | |
3eaa2885 CM |
350 | |
351 | /* | |
5b21f2ed | 352 | * the inode may be getting freed (in sys_unlink path). |
3eaa2885 | 353 | */ |
5b21f2ed ZY |
354 | inode = igrab(ordered->inode); |
355 | ||
3eaa2885 CM |
356 | spin_unlock(&root->fs_info->ordered_extent_lock); |
357 | ||
5b21f2ed ZY |
358 | if (inode) { |
359 | btrfs_start_ordered_extent(inode, ordered, 1); | |
360 | btrfs_put_ordered_extent(ordered); | |
361 | iput(inode); | |
362 | } else { | |
363 | btrfs_put_ordered_extent(ordered); | |
364 | } | |
3eaa2885 CM |
365 | |
366 | spin_lock(&root->fs_info->ordered_extent_lock); | |
367 | } | |
368 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
369 | return 0; | |
370 | } | |
371 | ||
eb84ae03 CM |
372 | /* |
373 | * Used to start IO or wait for a given ordered extent to finish. | |
374 | * | |
375 | * If wait is one, this effectively waits on page writeback for all the pages | |
376 | * in the extent, and it waits on the io completion code to insert | |
377 | * metadata into the btree corresponding to the extent | |
378 | */ | |
379 | void btrfs_start_ordered_extent(struct inode *inode, | |
380 | struct btrfs_ordered_extent *entry, | |
381 | int wait) | |
e6dcd2dc CM |
382 | { |
383 | u64 start = entry->file_offset; | |
384 | u64 end = start + entry->len - 1; | |
e1b81e67 | 385 | |
eb84ae03 CM |
386 | /* |
387 | * pages in the range can be dirty, clean or writeback. We | |
388 | * start IO on any dirty ones so the wait doesn't stall waiting | |
389 | * for pdflush to find them | |
390 | */ | |
771ed689 | 391 | btrfs_fdatawrite_range(inode->i_mapping, start, end, WB_SYNC_ALL); |
c8b97818 | 392 | if (wait) { |
e6dcd2dc CM |
393 | wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE, |
394 | &entry->flags)); | |
c8b97818 | 395 | } |
e6dcd2dc | 396 | } |
cee36a03 | 397 | |
eb84ae03 CM |
398 | /* |
399 | * Used to wait on ordered extents across a large range of bytes. | |
400 | */ | |
cb843a6f | 401 | int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len) |
e6dcd2dc CM |
402 | { |
403 | u64 end; | |
e5a2217e CM |
404 | u64 orig_end; |
405 | u64 wait_end; | |
e6dcd2dc | 406 | struct btrfs_ordered_extent *ordered; |
e5a2217e CM |
407 | |
408 | if (start + len < start) { | |
f421950f | 409 | orig_end = INT_LIMIT(loff_t); |
e5a2217e CM |
410 | } else { |
411 | orig_end = start + len - 1; | |
f421950f CM |
412 | if (orig_end > INT_LIMIT(loff_t)) |
413 | orig_end = INT_LIMIT(loff_t); | |
e5a2217e | 414 | } |
f421950f | 415 | wait_end = orig_end; |
4a096752 | 416 | again: |
e5a2217e CM |
417 | /* start IO across the range first to instantiate any delalloc |
418 | * extents | |
419 | */ | |
f421950f CM |
420 | btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_NONE); |
421 | ||
771ed689 CM |
422 | /* The compression code will leave pages locked but return from |
423 | * writepage without setting the page writeback. Starting again | |
424 | * with WB_SYNC_ALL will end up waiting for the IO to actually start. | |
425 | */ | |
426 | btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_ALL); | |
427 | ||
f421950f CM |
428 | btrfs_wait_on_page_writeback_range(inode->i_mapping, |
429 | start >> PAGE_CACHE_SHIFT, | |
430 | orig_end >> PAGE_CACHE_SHIFT); | |
e5a2217e | 431 | |
f421950f | 432 | end = orig_end; |
d397712b | 433 | while (1) { |
e6dcd2dc | 434 | ordered = btrfs_lookup_first_ordered_extent(inode, end); |
d397712b | 435 | if (!ordered) |
e6dcd2dc | 436 | break; |
e5a2217e | 437 | if (ordered->file_offset > orig_end) { |
e6dcd2dc CM |
438 | btrfs_put_ordered_extent(ordered); |
439 | break; | |
440 | } | |
441 | if (ordered->file_offset + ordered->len < start) { | |
442 | btrfs_put_ordered_extent(ordered); | |
443 | break; | |
444 | } | |
e5a2217e | 445 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
446 | end = ordered->file_offset; |
447 | btrfs_put_ordered_extent(ordered); | |
e5a2217e | 448 | if (end == 0 || end == start) |
e6dcd2dc CM |
449 | break; |
450 | end--; | |
451 | } | |
4a096752 CM |
452 | if (test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end, |
453 | EXTENT_ORDERED | EXTENT_DELALLOC, 0)) { | |
771ed689 | 454 | schedule_timeout(1); |
4a096752 CM |
455 | goto again; |
456 | } | |
cb843a6f | 457 | return 0; |
cee36a03 CM |
458 | } |
459 | ||
eb84ae03 CM |
460 | /* |
461 | * find an ordered extent corresponding to file_offset. return NULL if | |
462 | * nothing is found, otherwise take a reference on the extent and return it | |
463 | */ | |
e6dcd2dc CM |
464 | struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode, |
465 | u64 file_offset) | |
466 | { | |
467 | struct btrfs_ordered_inode_tree *tree; | |
468 | struct rb_node *node; | |
469 | struct btrfs_ordered_extent *entry = NULL; | |
470 | ||
471 | tree = &BTRFS_I(inode)->ordered_tree; | |
472 | mutex_lock(&tree->mutex); | |
473 | node = tree_search(tree, file_offset); | |
474 | if (!node) | |
475 | goto out; | |
476 | ||
477 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
478 | if (!offset_in_entry(entry, file_offset)) | |
479 | entry = NULL; | |
480 | if (entry) | |
481 | atomic_inc(&entry->refs); | |
482 | out: | |
483 | mutex_unlock(&tree->mutex); | |
484 | return entry; | |
485 | } | |
486 | ||
eb84ae03 CM |
487 | /* |
488 | * lookup and return any extent before 'file_offset'. NULL is returned | |
489 | * if none is found | |
490 | */ | |
e6dcd2dc | 491 | struct btrfs_ordered_extent * |
d397712b | 492 | btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset) |
e6dcd2dc CM |
493 | { |
494 | struct btrfs_ordered_inode_tree *tree; | |
495 | struct rb_node *node; | |
496 | struct btrfs_ordered_extent *entry = NULL; | |
497 | ||
498 | tree = &BTRFS_I(inode)->ordered_tree; | |
499 | mutex_lock(&tree->mutex); | |
500 | node = tree_search(tree, file_offset); | |
501 | if (!node) | |
502 | goto out; | |
503 | ||
504 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
505 | atomic_inc(&entry->refs); | |
506 | out: | |
507 | mutex_unlock(&tree->mutex); | |
508 | return entry; | |
81d7ed29 | 509 | } |
dbe674a9 | 510 | |
eb84ae03 CM |
511 | /* |
512 | * After an extent is done, call this to conditionally update the on disk | |
513 | * i_size. i_size is updated to cover any fully written part of the file. | |
514 | */ | |
dbe674a9 CM |
515 | int btrfs_ordered_update_i_size(struct inode *inode, |
516 | struct btrfs_ordered_extent *ordered) | |
517 | { | |
518 | struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; | |
519 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
520 | u64 disk_i_size; | |
521 | u64 new_i_size; | |
522 | u64 i_size_test; | |
523 | struct rb_node *node; | |
524 | struct btrfs_ordered_extent *test; | |
525 | ||
526 | mutex_lock(&tree->mutex); | |
527 | disk_i_size = BTRFS_I(inode)->disk_i_size; | |
528 | ||
529 | /* | |
530 | * if the disk i_size is already at the inode->i_size, or | |
531 | * this ordered extent is inside the disk i_size, we're done | |
532 | */ | |
533 | if (disk_i_size >= inode->i_size || | |
534 | ordered->file_offset + ordered->len <= disk_i_size) { | |
535 | goto out; | |
536 | } | |
537 | ||
538 | /* | |
539 | * we can't update the disk_isize if there are delalloc bytes | |
540 | * between disk_i_size and this ordered extent | |
541 | */ | |
542 | if (test_range_bit(io_tree, disk_i_size, | |
543 | ordered->file_offset + ordered->len - 1, | |
544 | EXTENT_DELALLOC, 0)) { | |
545 | goto out; | |
546 | } | |
547 | /* | |
548 | * walk backward from this ordered extent to disk_i_size. | |
549 | * if we find an ordered extent then we can't update disk i_size | |
550 | * yet | |
551 | */ | |
ba1da2f4 | 552 | node = &ordered->rb_node; |
d397712b | 553 | while (1) { |
ba1da2f4 | 554 | node = rb_prev(node); |
dbe674a9 CM |
555 | if (!node) |
556 | break; | |
557 | test = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
558 | if (test->file_offset + test->len <= disk_i_size) | |
559 | break; | |
560 | if (test->file_offset >= inode->i_size) | |
561 | break; | |
562 | if (test->file_offset >= disk_i_size) | |
563 | goto out; | |
564 | } | |
565 | new_i_size = min_t(u64, entry_end(ordered), i_size_read(inode)); | |
566 | ||
567 | /* | |
568 | * at this point, we know we can safely update i_size to at least | |
569 | * the offset from this ordered extent. But, we need to | |
570 | * walk forward and see if ios from higher up in the file have | |
571 | * finished. | |
572 | */ | |
573 | node = rb_next(&ordered->rb_node); | |
574 | i_size_test = 0; | |
575 | if (node) { | |
576 | /* | |
577 | * do we have an area where IO might have finished | |
578 | * between our ordered extent and the next one. | |
579 | */ | |
580 | test = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
d397712b | 581 | if (test->file_offset > entry_end(ordered)) |
b48652c1 | 582 | i_size_test = test->file_offset; |
dbe674a9 CM |
583 | } else { |
584 | i_size_test = i_size_read(inode); | |
585 | } | |
586 | ||
587 | /* | |
588 | * i_size_test is the end of a region after this ordered | |
589 | * extent where there are no ordered extents. As long as there | |
590 | * are no delalloc bytes in this area, it is safe to update | |
591 | * disk_i_size to the end of the region. | |
592 | */ | |
593 | if (i_size_test > entry_end(ordered) && | |
b48652c1 | 594 | !test_range_bit(io_tree, entry_end(ordered), i_size_test - 1, |
dbe674a9 CM |
595 | EXTENT_DELALLOC, 0)) { |
596 | new_i_size = min_t(u64, i_size_test, i_size_read(inode)); | |
597 | } | |
598 | BTRFS_I(inode)->disk_i_size = new_i_size; | |
599 | out: | |
600 | mutex_unlock(&tree->mutex); | |
601 | return 0; | |
602 | } | |
ba1da2f4 | 603 | |
eb84ae03 CM |
604 | /* |
605 | * search the ordered extents for one corresponding to 'offset' and | |
606 | * try to find a checksum. This is used because we allow pages to | |
607 | * be reclaimed before their checksum is actually put into the btree | |
608 | */ | |
d20f7043 CM |
609 | int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr, |
610 | u32 *sum) | |
ba1da2f4 CM |
611 | { |
612 | struct btrfs_ordered_sum *ordered_sum; | |
613 | struct btrfs_sector_sum *sector_sums; | |
614 | struct btrfs_ordered_extent *ordered; | |
615 | struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; | |
616 | struct list_head *cur; | |
3edf7d33 CM |
617 | unsigned long num_sectors; |
618 | unsigned long i; | |
619 | u32 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
ba1da2f4 | 620 | int ret = 1; |
ba1da2f4 CM |
621 | |
622 | ordered = btrfs_lookup_ordered_extent(inode, offset); | |
623 | if (!ordered) | |
624 | return 1; | |
625 | ||
626 | mutex_lock(&tree->mutex); | |
627 | list_for_each_prev(cur, &ordered->list) { | |
628 | ordered_sum = list_entry(cur, struct btrfs_ordered_sum, list); | |
d20f7043 | 629 | if (disk_bytenr >= ordered_sum->bytenr) { |
3edf7d33 | 630 | num_sectors = ordered_sum->len / sectorsize; |
ed98b56a | 631 | sector_sums = ordered_sum->sums; |
3edf7d33 | 632 | for (i = 0; i < num_sectors; i++) { |
d20f7043 | 633 | if (sector_sums[i].bytenr == disk_bytenr) { |
3edf7d33 CM |
634 | *sum = sector_sums[i].sum; |
635 | ret = 0; | |
636 | goto out; | |
637 | } | |
638 | } | |
ba1da2f4 CM |
639 | } |
640 | } | |
641 | out: | |
642 | mutex_unlock(&tree->mutex); | |
89642229 | 643 | btrfs_put_ordered_extent(ordered); |
ba1da2f4 CM |
644 | return ret; |
645 | } | |
646 | ||
f421950f CM |
647 | |
648 | /** | |
649 | * taken from mm/filemap.c because it isn't exported | |
650 | * | |
651 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range | |
652 | * @mapping: address space structure to write | |
653 | * @start: offset in bytes where the range starts | |
654 | * @end: offset in bytes where the range ends (inclusive) | |
655 | * @sync_mode: enable synchronous operation | |
656 | * | |
657 | * Start writeback against all of a mapping's dirty pages that lie | |
658 | * within the byte offsets <start, end> inclusive. | |
659 | * | |
660 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as | |
661 | * opposed to a regular memory cleansing writeback. The difference between | |
662 | * these two operations is that if a dirty page/buffer is encountered, it must | |
663 | * be waited upon, and not just skipped over. | |
664 | */ | |
665 | int btrfs_fdatawrite_range(struct address_space *mapping, loff_t start, | |
666 | loff_t end, int sync_mode) | |
667 | { | |
668 | struct writeback_control wbc = { | |
669 | .sync_mode = sync_mode, | |
670 | .nr_to_write = mapping->nrpages * 2, | |
671 | .range_start = start, | |
672 | .range_end = end, | |
673 | .for_writepages = 1, | |
674 | }; | |
675 | return btrfs_writepages(mapping, &wbc); | |
676 | } | |
677 | ||
678 | /** | |
679 | * taken from mm/filemap.c because it isn't exported | |
680 | * | |
681 | * wait_on_page_writeback_range - wait for writeback to complete | |
682 | * @mapping: target address_space | |
683 | * @start: beginning page index | |
684 | * @end: ending page index | |
685 | * | |
686 | * Wait for writeback to complete against pages indexed by start->end | |
687 | * inclusive | |
688 | */ | |
689 | int btrfs_wait_on_page_writeback_range(struct address_space *mapping, | |
690 | pgoff_t start, pgoff_t end) | |
691 | { | |
692 | struct pagevec pvec; | |
693 | int nr_pages; | |
694 | int ret = 0; | |
695 | pgoff_t index; | |
696 | ||
697 | if (end < start) | |
698 | return 0; | |
699 | ||
700 | pagevec_init(&pvec, 0); | |
701 | index = start; | |
702 | while ((index <= end) && | |
703 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
704 | PAGECACHE_TAG_WRITEBACK, | |
705 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { | |
706 | unsigned i; | |
707 | ||
708 | for (i = 0; i < nr_pages; i++) { | |
709 | struct page *page = pvec.pages[i]; | |
710 | ||
711 | /* until radix tree lookup accepts end_index */ | |
712 | if (page->index > end) | |
713 | continue; | |
714 | ||
715 | wait_on_page_writeback(page); | |
716 | if (PageError(page)) | |
717 | ret = -EIO; | |
718 | } | |
719 | pagevec_release(&pvec); | |
720 | cond_resched(); | |
721 | } | |
722 | ||
723 | /* Check for outstanding write errors */ | |
724 | if (test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
725 | ret = -ENOSPC; | |
726 | if (test_and_clear_bit(AS_EIO, &mapping->flags)) | |
727 | ret = -EIO; | |
728 | ||
729 | return ret; | |
730 | } |