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" |
ebb8765b | 28 | #include "compression.h" |
dc17ff8f | 29 | |
6352b91d MX |
30 | static struct kmem_cache *btrfs_ordered_extent_cache; |
31 | ||
e6dcd2dc | 32 | static u64 entry_end(struct btrfs_ordered_extent *entry) |
dc17ff8f | 33 | { |
e6dcd2dc CM |
34 | if (entry->file_offset + entry->len < entry->file_offset) |
35 | return (u64)-1; | |
36 | return entry->file_offset + entry->len; | |
dc17ff8f CM |
37 | } |
38 | ||
d352ac68 CM |
39 | /* returns NULL if the insertion worked, or it returns the node it did find |
40 | * in the tree | |
41 | */ | |
e6dcd2dc CM |
42 | static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset, |
43 | struct rb_node *node) | |
dc17ff8f | 44 | { |
d397712b CM |
45 | struct rb_node **p = &root->rb_node; |
46 | struct rb_node *parent = NULL; | |
e6dcd2dc | 47 | struct btrfs_ordered_extent *entry; |
dc17ff8f | 48 | |
d397712b | 49 | while (*p) { |
dc17ff8f | 50 | parent = *p; |
e6dcd2dc | 51 | entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node); |
dc17ff8f | 52 | |
e6dcd2dc | 53 | if (file_offset < entry->file_offset) |
dc17ff8f | 54 | p = &(*p)->rb_left; |
e6dcd2dc | 55 | else if (file_offset >= entry_end(entry)) |
dc17ff8f CM |
56 | p = &(*p)->rb_right; |
57 | else | |
58 | return parent; | |
59 | } | |
60 | ||
61 | rb_link_node(node, parent, p); | |
62 | rb_insert_color(node, root); | |
63 | return NULL; | |
64 | } | |
65 | ||
43c04fb1 JM |
66 | static void ordered_data_tree_panic(struct inode *inode, int errno, |
67 | u64 offset) | |
68 | { | |
69 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
5d163e0e JM |
70 | btrfs_panic(fs_info, errno, |
71 | "Inconsistency in ordered tree at offset %llu", offset); | |
43c04fb1 JM |
72 | } |
73 | ||
d352ac68 CM |
74 | /* |
75 | * look for a given offset in the tree, and if it can't be found return the | |
76 | * first lesser offset | |
77 | */ | |
e6dcd2dc CM |
78 | static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset, |
79 | struct rb_node **prev_ret) | |
dc17ff8f | 80 | { |
d397712b | 81 | struct rb_node *n = root->rb_node; |
dc17ff8f | 82 | struct rb_node *prev = NULL; |
e6dcd2dc CM |
83 | struct rb_node *test; |
84 | struct btrfs_ordered_extent *entry; | |
85 | struct btrfs_ordered_extent *prev_entry = NULL; | |
dc17ff8f | 86 | |
d397712b | 87 | while (n) { |
e6dcd2dc | 88 | entry = rb_entry(n, struct btrfs_ordered_extent, rb_node); |
dc17ff8f CM |
89 | prev = n; |
90 | prev_entry = entry; | |
dc17ff8f | 91 | |
e6dcd2dc | 92 | if (file_offset < entry->file_offset) |
dc17ff8f | 93 | n = n->rb_left; |
e6dcd2dc | 94 | else if (file_offset >= entry_end(entry)) |
dc17ff8f CM |
95 | n = n->rb_right; |
96 | else | |
97 | return n; | |
98 | } | |
99 | if (!prev_ret) | |
100 | return NULL; | |
101 | ||
d397712b | 102 | while (prev && file_offset >= entry_end(prev_entry)) { |
e6dcd2dc CM |
103 | test = rb_next(prev); |
104 | if (!test) | |
105 | break; | |
106 | prev_entry = rb_entry(test, struct btrfs_ordered_extent, | |
107 | rb_node); | |
108 | if (file_offset < entry_end(prev_entry)) | |
109 | break; | |
110 | ||
111 | prev = test; | |
112 | } | |
113 | if (prev) | |
114 | prev_entry = rb_entry(prev, struct btrfs_ordered_extent, | |
115 | rb_node); | |
d397712b | 116 | while (prev && file_offset < entry_end(prev_entry)) { |
e6dcd2dc CM |
117 | test = rb_prev(prev); |
118 | if (!test) | |
119 | break; | |
120 | prev_entry = rb_entry(test, struct btrfs_ordered_extent, | |
121 | rb_node); | |
122 | prev = test; | |
dc17ff8f CM |
123 | } |
124 | *prev_ret = prev; | |
125 | return NULL; | |
126 | } | |
127 | ||
d352ac68 CM |
128 | /* |
129 | * helper to check if a given offset is inside a given entry | |
130 | */ | |
e6dcd2dc CM |
131 | static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset) |
132 | { | |
133 | if (file_offset < entry->file_offset || | |
134 | entry->file_offset + entry->len <= file_offset) | |
135 | return 0; | |
136 | return 1; | |
137 | } | |
138 | ||
4b46fce2 JB |
139 | static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset, |
140 | u64 len) | |
141 | { | |
142 | if (file_offset + len <= entry->file_offset || | |
143 | entry->file_offset + entry->len <= file_offset) | |
144 | return 0; | |
145 | return 1; | |
146 | } | |
147 | ||
d352ac68 CM |
148 | /* |
149 | * look find the first ordered struct that has this offset, otherwise | |
150 | * the first one less than this offset | |
151 | */ | |
e6dcd2dc CM |
152 | static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree, |
153 | u64 file_offset) | |
dc17ff8f | 154 | { |
e6dcd2dc | 155 | struct rb_root *root = &tree->tree; |
c87fb6fd | 156 | struct rb_node *prev = NULL; |
dc17ff8f | 157 | struct rb_node *ret; |
e6dcd2dc CM |
158 | struct btrfs_ordered_extent *entry; |
159 | ||
160 | if (tree->last) { | |
161 | entry = rb_entry(tree->last, struct btrfs_ordered_extent, | |
162 | rb_node); | |
163 | if (offset_in_entry(entry, file_offset)) | |
164 | return tree->last; | |
165 | } | |
166 | ret = __tree_search(root, file_offset, &prev); | |
dc17ff8f | 167 | if (!ret) |
e6dcd2dc CM |
168 | ret = prev; |
169 | if (ret) | |
170 | tree->last = ret; | |
dc17ff8f CM |
171 | return ret; |
172 | } | |
173 | ||
eb84ae03 CM |
174 | /* allocate and add a new ordered_extent into the per-inode tree. |
175 | * file_offset is the logical offset in the file | |
176 | * | |
177 | * start is the disk block number of an extent already reserved in the | |
178 | * extent allocation tree | |
179 | * | |
180 | * len is the length of the extent | |
181 | * | |
eb84ae03 CM |
182 | * The tree is given a single reference on the ordered extent that was |
183 | * inserted. | |
184 | */ | |
4b46fce2 JB |
185 | static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset, |
186 | u64 start, u64 len, u64 disk_len, | |
261507a0 | 187 | int type, int dio, int compress_type) |
dc17ff8f | 188 | { |
0b246afa | 189 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
199c2a9c | 190 | struct btrfs_root *root = BTRFS_I(inode)->root; |
dc17ff8f | 191 | struct btrfs_ordered_inode_tree *tree; |
e6dcd2dc CM |
192 | struct rb_node *node; |
193 | struct btrfs_ordered_extent *entry; | |
dc17ff8f | 194 | |
e6dcd2dc | 195 | tree = &BTRFS_I(inode)->ordered_tree; |
6352b91d | 196 | entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS); |
dc17ff8f CM |
197 | if (!entry) |
198 | return -ENOMEM; | |
199 | ||
e6dcd2dc CM |
200 | entry->file_offset = file_offset; |
201 | entry->start = start; | |
202 | entry->len = len; | |
c8b97818 | 203 | entry->disk_len = disk_len; |
8b62b72b | 204 | entry->bytes_left = len; |
5fd02043 | 205 | entry->inode = igrab(inode); |
261507a0 | 206 | entry->compress_type = compress_type; |
77cef2ec | 207 | entry->truncated_len = (u64)-1; |
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 | 214 | /* one ref for the tree */ |
e76edab7 | 215 | refcount_set(&entry->refs, 1); |
e6dcd2dc CM |
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); |
50d9aa99 | 222 | INIT_LIST_HEAD(&entry->trans_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) { | |
0b246afa | 238 | spin_lock(&fs_info->ordered_root_lock); |
199c2a9c | 239 | BUG_ON(!list_empty(&root->ordered_root)); |
0b246afa JM |
240 | list_add_tail(&root->ordered_root, &fs_info->ordered_roots); |
241 | spin_unlock(&fs_info->ordered_root_lock); | |
199c2a9c MX |
242 | } |
243 | spin_unlock(&root->ordered_extent_lock); | |
3eaa2885 | 244 | |
8b62f87b JB |
245 | /* |
246 | * We don't need the count_max_extents here, we can assume that all of | |
247 | * that work has been done at higher layers, so this is truly the | |
248 | * smallest the extent is going to get. | |
249 | */ | |
250 | spin_lock(&BTRFS_I(inode)->lock); | |
251 | btrfs_mod_outstanding_extents(BTRFS_I(inode), 1); | |
252 | spin_unlock(&BTRFS_I(inode)->lock); | |
253 | ||
dc17ff8f CM |
254 | return 0; |
255 | } | |
256 | ||
4b46fce2 JB |
257 | int btrfs_add_ordered_extent(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, 0, |
262 | BTRFS_COMPRESS_NONE); | |
4b46fce2 JB |
263 | } |
264 | ||
265 | int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset, | |
266 | u64 start, u64 len, u64 disk_len, int type) | |
267 | { | |
268 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, | |
261507a0 LZ |
269 | disk_len, type, 1, |
270 | BTRFS_COMPRESS_NONE); | |
271 | } | |
272 | ||
273 | int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset, | |
274 | u64 start, u64 len, u64 disk_len, | |
275 | int type, int compress_type) | |
276 | { | |
277 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, | |
278 | disk_len, type, 0, | |
279 | compress_type); | |
4b46fce2 JB |
280 | } |
281 | ||
eb84ae03 CM |
282 | /* |
283 | * Add a struct btrfs_ordered_sum into the list of checksums to be inserted | |
3edf7d33 CM |
284 | * when an ordered extent is finished. If the list covers more than one |
285 | * ordered extent, it is split across multiples. | |
eb84ae03 | 286 | */ |
143bede5 JM |
287 | void btrfs_add_ordered_sum(struct inode *inode, |
288 | struct btrfs_ordered_extent *entry, | |
289 | struct btrfs_ordered_sum *sum) | |
dc17ff8f | 290 | { |
e6dcd2dc | 291 | struct btrfs_ordered_inode_tree *tree; |
dc17ff8f | 292 | |
e6dcd2dc | 293 | tree = &BTRFS_I(inode)->ordered_tree; |
5fd02043 | 294 | spin_lock_irq(&tree->lock); |
e6dcd2dc | 295 | list_add_tail(&sum->list, &entry->list); |
5fd02043 | 296 | spin_unlock_irq(&tree->lock); |
dc17ff8f CM |
297 | } |
298 | ||
163cf09c CM |
299 | /* |
300 | * this is used to account for finished IO across a given range | |
301 | * of the file. The IO may span ordered extents. If | |
302 | * a given ordered_extent is completely done, 1 is returned, otherwise | |
303 | * 0. | |
304 | * | |
305 | * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used | |
306 | * to make sure this function only returns 1 once for a given ordered extent. | |
307 | * | |
308 | * file_offset is updated to one byte past the range that is recorded as | |
309 | * complete. This allows you to walk forward in the file. | |
310 | */ | |
311 | int btrfs_dec_test_first_ordered_pending(struct inode *inode, | |
312 | struct btrfs_ordered_extent **cached, | |
5fd02043 | 313 | u64 *file_offset, u64 io_size, int uptodate) |
163cf09c | 314 | { |
0b246afa | 315 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
163cf09c CM |
316 | struct btrfs_ordered_inode_tree *tree; |
317 | struct rb_node *node; | |
318 | struct btrfs_ordered_extent *entry = NULL; | |
319 | int ret; | |
5fd02043 | 320 | unsigned long flags; |
163cf09c CM |
321 | u64 dec_end; |
322 | u64 dec_start; | |
323 | u64 to_dec; | |
324 | ||
325 | tree = &BTRFS_I(inode)->ordered_tree; | |
5fd02043 | 326 | spin_lock_irqsave(&tree->lock, flags); |
163cf09c CM |
327 | node = tree_search(tree, *file_offset); |
328 | if (!node) { | |
329 | ret = 1; | |
330 | goto out; | |
331 | } | |
332 | ||
333 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
334 | if (!offset_in_entry(entry, *file_offset)) { | |
335 | ret = 1; | |
336 | goto out; | |
337 | } | |
338 | ||
339 | dec_start = max(*file_offset, entry->file_offset); | |
340 | dec_end = min(*file_offset + io_size, entry->file_offset + | |
341 | entry->len); | |
342 | *file_offset = dec_end; | |
343 | if (dec_start > dec_end) { | |
0b246afa JM |
344 | btrfs_crit(fs_info, "bad ordering dec_start %llu end %llu", |
345 | dec_start, dec_end); | |
163cf09c CM |
346 | } |
347 | to_dec = dec_end - dec_start; | |
348 | if (to_dec > entry->bytes_left) { | |
0b246afa JM |
349 | btrfs_crit(fs_info, |
350 | "bad ordered accounting left %llu size %llu", | |
351 | entry->bytes_left, to_dec); | |
163cf09c CM |
352 | } |
353 | entry->bytes_left -= to_dec; | |
5fd02043 JB |
354 | if (!uptodate) |
355 | set_bit(BTRFS_ORDERED_IOERR, &entry->flags); | |
356 | ||
af7a6509 | 357 | if (entry->bytes_left == 0) { |
163cf09c | 358 | ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); |
a83342aa DS |
359 | /* |
360 | * Implicit memory barrier after test_and_set_bit | |
361 | */ | |
af7a6509 MX |
362 | if (waitqueue_active(&entry->wait)) |
363 | wake_up(&entry->wait); | |
364 | } else { | |
163cf09c | 365 | ret = 1; |
af7a6509 | 366 | } |
163cf09c CM |
367 | out: |
368 | if (!ret && cached && entry) { | |
369 | *cached = entry; | |
e76edab7 | 370 | refcount_inc(&entry->refs); |
163cf09c | 371 | } |
5fd02043 | 372 | spin_unlock_irqrestore(&tree->lock, flags); |
163cf09c CM |
373 | return ret == 0; |
374 | } | |
375 | ||
eb84ae03 CM |
376 | /* |
377 | * this is used to account for finished IO across a given range | |
378 | * of the file. The IO should not span ordered extents. If | |
379 | * a given ordered_extent is completely done, 1 is returned, otherwise | |
380 | * 0. | |
381 | * | |
382 | * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used | |
383 | * to make sure this function only returns 1 once for a given ordered extent. | |
384 | */ | |
e6dcd2dc | 385 | int btrfs_dec_test_ordered_pending(struct inode *inode, |
5a1a3df1 | 386 | struct btrfs_ordered_extent **cached, |
5fd02043 | 387 | u64 file_offset, u64 io_size, int uptodate) |
dc17ff8f | 388 | { |
e6dcd2dc | 389 | struct btrfs_ordered_inode_tree *tree; |
dc17ff8f | 390 | struct rb_node *node; |
5a1a3df1 | 391 | struct btrfs_ordered_extent *entry = NULL; |
5fd02043 | 392 | unsigned long flags; |
e6dcd2dc CM |
393 | int ret; |
394 | ||
395 | tree = &BTRFS_I(inode)->ordered_tree; | |
5fd02043 JB |
396 | spin_lock_irqsave(&tree->lock, flags); |
397 | if (cached && *cached) { | |
398 | entry = *cached; | |
399 | goto have_entry; | |
400 | } | |
401 | ||
e6dcd2dc | 402 | node = tree_search(tree, file_offset); |
dc17ff8f | 403 | if (!node) { |
e6dcd2dc CM |
404 | ret = 1; |
405 | goto out; | |
dc17ff8f CM |
406 | } |
407 | ||
e6dcd2dc | 408 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
5fd02043 | 409 | have_entry: |
e6dcd2dc CM |
410 | if (!offset_in_entry(entry, file_offset)) { |
411 | ret = 1; | |
412 | goto out; | |
dc17ff8f | 413 | } |
e6dcd2dc | 414 | |
8b62b72b | 415 | if (io_size > entry->bytes_left) { |
efe120a0 FH |
416 | btrfs_crit(BTRFS_I(inode)->root->fs_info, |
417 | "bad ordered accounting left %llu size %llu", | |
c1c9ff7c | 418 | entry->bytes_left, io_size); |
8b62b72b CM |
419 | } |
420 | entry->bytes_left -= io_size; | |
5fd02043 JB |
421 | if (!uptodate) |
422 | set_bit(BTRFS_ORDERED_IOERR, &entry->flags); | |
423 | ||
af7a6509 | 424 | if (entry->bytes_left == 0) { |
e6dcd2dc | 425 | ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); |
a83342aa DS |
426 | /* |
427 | * Implicit memory barrier after test_and_set_bit | |
428 | */ | |
af7a6509 MX |
429 | if (waitqueue_active(&entry->wait)) |
430 | wake_up(&entry->wait); | |
431 | } else { | |
8b62b72b | 432 | ret = 1; |
af7a6509 | 433 | } |
e6dcd2dc | 434 | out: |
5a1a3df1 JB |
435 | if (!ret && cached && entry) { |
436 | *cached = entry; | |
e76edab7 | 437 | refcount_inc(&entry->refs); |
5a1a3df1 | 438 | } |
5fd02043 | 439 | spin_unlock_irqrestore(&tree->lock, flags); |
e6dcd2dc CM |
440 | return ret == 0; |
441 | } | |
dc17ff8f | 442 | |
2ab28f32 | 443 | /* Needs to either be called under a log transaction or the log_mutex */ |
22346637 | 444 | void btrfs_get_logged_extents(struct btrfs_inode *inode, |
0870295b FM |
445 | struct list_head *logged_list, |
446 | const loff_t start, | |
447 | const loff_t end) | |
2ab28f32 JB |
448 | { |
449 | struct btrfs_ordered_inode_tree *tree; | |
450 | struct btrfs_ordered_extent *ordered; | |
451 | struct rb_node *n; | |
0870295b | 452 | struct rb_node *prev; |
2ab28f32 | 453 | |
22346637 | 454 | tree = &inode->ordered_tree; |
2ab28f32 | 455 | spin_lock_irq(&tree->lock); |
0870295b FM |
456 | n = __tree_search(&tree->tree, end, &prev); |
457 | if (!n) | |
458 | n = prev; | |
459 | for (; n; n = rb_prev(n)) { | |
2ab28f32 | 460 | ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node); |
0870295b FM |
461 | if (ordered->file_offset > end) |
462 | continue; | |
463 | if (entry_end(ordered) <= start) | |
464 | break; | |
4d884fce | 465 | if (test_and_set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags)) |
50d9aa99 | 466 | continue; |
0870295b | 467 | list_add(&ordered->log_list, logged_list); |
e76edab7 | 468 | refcount_inc(&ordered->refs); |
2ab28f32 JB |
469 | } |
470 | spin_unlock_irq(&tree->lock); | |
471 | } | |
472 | ||
827463c4 MX |
473 | void btrfs_put_logged_extents(struct list_head *logged_list) |
474 | { | |
475 | struct btrfs_ordered_extent *ordered; | |
476 | ||
477 | while (!list_empty(logged_list)) { | |
478 | ordered = list_first_entry(logged_list, | |
479 | struct btrfs_ordered_extent, | |
480 | log_list); | |
481 | list_del_init(&ordered->log_list); | |
482 | btrfs_put_ordered_extent(ordered); | |
483 | } | |
484 | } | |
485 | ||
486 | void btrfs_submit_logged_extents(struct list_head *logged_list, | |
487 | struct btrfs_root *log) | |
488 | { | |
489 | int index = log->log_transid % 2; | |
490 | ||
491 | spin_lock_irq(&log->log_extents_lock[index]); | |
492 | list_splice_tail(logged_list, &log->logged_list[index]); | |
493 | spin_unlock_irq(&log->log_extents_lock[index]); | |
494 | } | |
495 | ||
50d9aa99 JB |
496 | void btrfs_wait_logged_extents(struct btrfs_trans_handle *trans, |
497 | struct btrfs_root *log, u64 transid) | |
2ab28f32 JB |
498 | { |
499 | struct btrfs_ordered_extent *ordered; | |
500 | int index = transid % 2; | |
501 | ||
502 | spin_lock_irq(&log->log_extents_lock[index]); | |
503 | while (!list_empty(&log->logged_list[index])) { | |
161c3549 | 504 | struct inode *inode; |
2ab28f32 JB |
505 | ordered = list_first_entry(&log->logged_list[index], |
506 | struct btrfs_ordered_extent, | |
507 | log_list); | |
508 | list_del_init(&ordered->log_list); | |
161c3549 | 509 | inode = ordered->inode; |
2ab28f32 | 510 | spin_unlock_irq(&log->log_extents_lock[index]); |
98ce2ded LB |
511 | |
512 | if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) && | |
513 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) { | |
98ce2ded LB |
514 | u64 start = ordered->file_offset; |
515 | u64 end = ordered->file_offset + ordered->len - 1; | |
516 | ||
517 | WARN_ON(!inode); | |
518 | filemap_fdatawrite_range(inode->i_mapping, start, end); | |
519 | } | |
2ab28f32 JB |
520 | wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE, |
521 | &ordered->flags)); | |
98ce2ded | 522 | |
7558c8bc | 523 | /* |
161c3549 JB |
524 | * In order to keep us from losing our ordered extent |
525 | * information when committing the transaction we have to make | |
526 | * sure that any logged extents are completed when we go to | |
527 | * commit the transaction. To do this we simply increase the | |
528 | * current transactions pending_ordered counter and decrement it | |
529 | * when the ordered extent completes. | |
7558c8bc | 530 | */ |
161c3549 JB |
531 | if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) { |
532 | struct btrfs_ordered_inode_tree *tree; | |
533 | ||
534 | tree = &BTRFS_I(inode)->ordered_tree; | |
535 | spin_lock_irq(&tree->lock); | |
536 | if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) { | |
537 | set_bit(BTRFS_ORDERED_PENDING, &ordered->flags); | |
538 | atomic_inc(&trans->transaction->pending_ordered); | |
539 | } | |
540 | spin_unlock_irq(&tree->lock); | |
541 | } | |
542 | btrfs_put_ordered_extent(ordered); | |
2ab28f32 JB |
543 | spin_lock_irq(&log->log_extents_lock[index]); |
544 | } | |
545 | spin_unlock_irq(&log->log_extents_lock[index]); | |
546 | } | |
547 | ||
548 | void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid) | |
549 | { | |
550 | struct btrfs_ordered_extent *ordered; | |
551 | int index = transid % 2; | |
552 | ||
553 | spin_lock_irq(&log->log_extents_lock[index]); | |
554 | while (!list_empty(&log->logged_list[index])) { | |
555 | ordered = list_first_entry(&log->logged_list[index], | |
556 | struct btrfs_ordered_extent, | |
557 | log_list); | |
558 | list_del_init(&ordered->log_list); | |
559 | spin_unlock_irq(&log->log_extents_lock[index]); | |
560 | btrfs_put_ordered_extent(ordered); | |
561 | spin_lock_irq(&log->log_extents_lock[index]); | |
562 | } | |
563 | spin_unlock_irq(&log->log_extents_lock[index]); | |
564 | } | |
565 | ||
eb84ae03 CM |
566 | /* |
567 | * used to drop a reference on an ordered extent. This will free | |
568 | * the extent if the last reference is dropped | |
569 | */ | |
143bede5 | 570 | void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry) |
e6dcd2dc | 571 | { |
ba1da2f4 CM |
572 | struct list_head *cur; |
573 | struct btrfs_ordered_sum *sum; | |
574 | ||
1abe9b8a | 575 | trace_btrfs_ordered_extent_put(entry->inode, entry); |
576 | ||
e76edab7 | 577 | if (refcount_dec_and_test(&entry->refs)) { |
61de718f FM |
578 | ASSERT(list_empty(&entry->log_list)); |
579 | ASSERT(list_empty(&entry->trans_list)); | |
580 | ASSERT(list_empty(&entry->root_extent_list)); | |
581 | ASSERT(RB_EMPTY_NODE(&entry->rb_node)); | |
5fd02043 JB |
582 | if (entry->inode) |
583 | btrfs_add_delayed_iput(entry->inode); | |
d397712b | 584 | while (!list_empty(&entry->list)) { |
ba1da2f4 CM |
585 | cur = entry->list.next; |
586 | sum = list_entry(cur, struct btrfs_ordered_sum, list); | |
587 | list_del(&sum->list); | |
588 | kfree(sum); | |
589 | } | |
6352b91d | 590 | kmem_cache_free(btrfs_ordered_extent_cache, entry); |
ba1da2f4 | 591 | } |
dc17ff8f | 592 | } |
cee36a03 | 593 | |
eb84ae03 CM |
594 | /* |
595 | * remove an ordered extent from the tree. No references are dropped | |
5fd02043 | 596 | * and waiters are woken up. |
eb84ae03 | 597 | */ |
5fd02043 JB |
598 | void btrfs_remove_ordered_extent(struct inode *inode, |
599 | struct btrfs_ordered_extent *entry) | |
cee36a03 | 600 | { |
0b246afa | 601 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 602 | struct btrfs_ordered_inode_tree *tree; |
8b62f87b JB |
603 | struct btrfs_inode *btrfs_inode = BTRFS_I(inode); |
604 | struct btrfs_root *root = btrfs_inode->root; | |
cee36a03 | 605 | struct rb_node *node; |
161c3549 | 606 | bool dec_pending_ordered = false; |
cee36a03 | 607 | |
8b62f87b JB |
608 | /* This is paired with btrfs_add_ordered_extent. */ |
609 | spin_lock(&btrfs_inode->lock); | |
610 | btrfs_mod_outstanding_extents(btrfs_inode, -1); | |
611 | spin_unlock(&btrfs_inode->lock); | |
612 | if (root != fs_info->tree_root) | |
613 | btrfs_delalloc_release_metadata(btrfs_inode, entry->len); | |
614 | ||
615 | tree = &btrfs_inode->ordered_tree; | |
5fd02043 | 616 | spin_lock_irq(&tree->lock); |
e6dcd2dc | 617 | node = &entry->rb_node; |
cee36a03 | 618 | rb_erase(node, &tree->tree); |
61de718f | 619 | RB_CLEAR_NODE(node); |
1b8e7e45 FDBM |
620 | if (tree->last == node) |
621 | tree->last = NULL; | |
e6dcd2dc | 622 | set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags); |
161c3549 JB |
623 | if (test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags)) |
624 | dec_pending_ordered = true; | |
5fd02043 | 625 | spin_unlock_irq(&tree->lock); |
3eaa2885 | 626 | |
161c3549 JB |
627 | /* |
628 | * The current running transaction is waiting on us, we need to let it | |
629 | * know that we're complete and wake it up. | |
630 | */ | |
631 | if (dec_pending_ordered) { | |
632 | struct btrfs_transaction *trans; | |
633 | ||
634 | /* | |
635 | * The checks for trans are just a formality, it should be set, | |
636 | * but if it isn't we don't want to deref/assert under the spin | |
637 | * lock, so be nice and check if trans is set, but ASSERT() so | |
638 | * if it isn't set a developer will notice. | |
639 | */ | |
0b246afa JM |
640 | spin_lock(&fs_info->trans_lock); |
641 | trans = fs_info->running_transaction; | |
161c3549 | 642 | if (trans) |
9b64f57d | 643 | refcount_inc(&trans->use_count); |
0b246afa | 644 | spin_unlock(&fs_info->trans_lock); |
161c3549 JB |
645 | |
646 | ASSERT(trans); | |
647 | if (trans) { | |
648 | if (atomic_dec_and_test(&trans->pending_ordered)) | |
649 | wake_up(&trans->pending_wait); | |
650 | btrfs_put_transaction(trans); | |
651 | } | |
652 | } | |
653 | ||
199c2a9c | 654 | spin_lock(&root->ordered_extent_lock); |
3eaa2885 | 655 | list_del_init(&entry->root_extent_list); |
199c2a9c | 656 | root->nr_ordered_extents--; |
5a3f23d5 | 657 | |
1abe9b8a | 658 | trace_btrfs_ordered_extent_remove(inode, entry); |
659 | ||
199c2a9c | 660 | if (!root->nr_ordered_extents) { |
0b246afa | 661 | spin_lock(&fs_info->ordered_root_lock); |
199c2a9c MX |
662 | BUG_ON(list_empty(&root->ordered_root)); |
663 | list_del_init(&root->ordered_root); | |
0b246afa | 664 | spin_unlock(&fs_info->ordered_root_lock); |
199c2a9c MX |
665 | } |
666 | spin_unlock(&root->ordered_extent_lock); | |
e6dcd2dc | 667 | wake_up(&entry->wait); |
cee36a03 CM |
668 | } |
669 | ||
d458b054 | 670 | static void btrfs_run_ordered_extent_work(struct btrfs_work *work) |
9afab882 MX |
671 | { |
672 | struct btrfs_ordered_extent *ordered; | |
673 | ||
674 | ordered = container_of(work, struct btrfs_ordered_extent, flush_work); | |
675 | btrfs_start_ordered_extent(ordered->inode, ordered, 1); | |
676 | complete(&ordered->completion); | |
677 | } | |
678 | ||
d352ac68 CM |
679 | /* |
680 | * wait for all the ordered extents in a root. This is done when balancing | |
681 | * space between drives. | |
682 | */ | |
6374e57a | 683 | u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr, |
578def7c | 684 | const u64 range_start, const u64 range_len) |
3eaa2885 | 685 | { |
0b246afa | 686 | struct btrfs_fs_info *fs_info = root->fs_info; |
578def7c FM |
687 | LIST_HEAD(splice); |
688 | LIST_HEAD(skipped); | |
689 | LIST_HEAD(works); | |
9afab882 | 690 | struct btrfs_ordered_extent *ordered, *next; |
6374e57a | 691 | u64 count = 0; |
578def7c | 692 | const u64 range_end = range_start + range_len; |
3eaa2885 | 693 | |
31f3d255 | 694 | mutex_lock(&root->ordered_extent_mutex); |
199c2a9c MX |
695 | spin_lock(&root->ordered_extent_lock); |
696 | list_splice_init(&root->ordered_extents, &splice); | |
b0244199 | 697 | while (!list_empty(&splice) && nr) { |
199c2a9c MX |
698 | ordered = list_first_entry(&splice, struct btrfs_ordered_extent, |
699 | root_extent_list); | |
578def7c FM |
700 | |
701 | if (range_end <= ordered->start || | |
702 | ordered->start + ordered->disk_len <= range_start) { | |
703 | list_move_tail(&ordered->root_extent_list, &skipped); | |
704 | cond_resched_lock(&root->ordered_extent_lock); | |
705 | continue; | |
706 | } | |
707 | ||
199c2a9c MX |
708 | list_move_tail(&ordered->root_extent_list, |
709 | &root->ordered_extents); | |
e76edab7 | 710 | refcount_inc(&ordered->refs); |
199c2a9c | 711 | spin_unlock(&root->ordered_extent_lock); |
3eaa2885 | 712 | |
a44903ab | 713 | btrfs_init_work(&ordered->flush_work, |
9e0af237 | 714 | btrfs_flush_delalloc_helper, |
a44903ab | 715 | btrfs_run_ordered_extent_work, NULL, NULL); |
199c2a9c | 716 | list_add_tail(&ordered->work_list, &works); |
0b246afa | 717 | btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work); |
3eaa2885 | 718 | |
9afab882 | 719 | cond_resched(); |
199c2a9c | 720 | spin_lock(&root->ordered_extent_lock); |
6374e57a | 721 | if (nr != U64_MAX) |
b0244199 MX |
722 | nr--; |
723 | count++; | |
3eaa2885 | 724 | } |
578def7c | 725 | list_splice_tail(&skipped, &root->ordered_extents); |
b0244199 | 726 | list_splice_tail(&splice, &root->ordered_extents); |
199c2a9c | 727 | spin_unlock(&root->ordered_extent_lock); |
9afab882 MX |
728 | |
729 | list_for_each_entry_safe(ordered, next, &works, work_list) { | |
730 | list_del_init(&ordered->work_list); | |
731 | wait_for_completion(&ordered->completion); | |
9afab882 | 732 | btrfs_put_ordered_extent(ordered); |
9afab882 MX |
733 | cond_resched(); |
734 | } | |
31f3d255 | 735 | mutex_unlock(&root->ordered_extent_mutex); |
b0244199 MX |
736 | |
737 | return count; | |
3eaa2885 CM |
738 | } |
739 | ||
6374e57a CM |
740 | u64 btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr, |
741 | const u64 range_start, const u64 range_len) | |
199c2a9c MX |
742 | { |
743 | struct btrfs_root *root; | |
744 | struct list_head splice; | |
6374e57a CM |
745 | u64 total_done = 0; |
746 | u64 done; | |
199c2a9c MX |
747 | |
748 | INIT_LIST_HEAD(&splice); | |
749 | ||
8b9d83cd | 750 | mutex_lock(&fs_info->ordered_operations_mutex); |
199c2a9c MX |
751 | spin_lock(&fs_info->ordered_root_lock); |
752 | list_splice_init(&fs_info->ordered_roots, &splice); | |
b0244199 | 753 | while (!list_empty(&splice) && nr) { |
199c2a9c MX |
754 | root = list_first_entry(&splice, struct btrfs_root, |
755 | ordered_root); | |
756 | root = btrfs_grab_fs_root(root); | |
757 | BUG_ON(!root); | |
758 | list_move_tail(&root->ordered_root, | |
759 | &fs_info->ordered_roots); | |
760 | spin_unlock(&fs_info->ordered_root_lock); | |
761 | ||
578def7c FM |
762 | done = btrfs_wait_ordered_extents(root, nr, |
763 | range_start, range_len); | |
199c2a9c | 764 | btrfs_put_fs_root(root); |
f0e9b7d6 | 765 | total_done += done; |
199c2a9c MX |
766 | |
767 | spin_lock(&fs_info->ordered_root_lock); | |
6374e57a | 768 | if (nr != U64_MAX) { |
b0244199 | 769 | nr -= done; |
b0244199 | 770 | } |
199c2a9c | 771 | } |
931aa877 | 772 | list_splice_tail(&splice, &fs_info->ordered_roots); |
199c2a9c | 773 | spin_unlock(&fs_info->ordered_root_lock); |
8b9d83cd | 774 | mutex_unlock(&fs_info->ordered_operations_mutex); |
f0e9b7d6 FM |
775 | |
776 | return total_done; | |
199c2a9c MX |
777 | } |
778 | ||
eb84ae03 CM |
779 | /* |
780 | * Used to start IO or wait for a given ordered extent to finish. | |
781 | * | |
782 | * If wait is one, this effectively waits on page writeback for all the pages | |
783 | * in the extent, and it waits on the io completion code to insert | |
784 | * metadata into the btree corresponding to the extent | |
785 | */ | |
786 | void btrfs_start_ordered_extent(struct inode *inode, | |
787 | struct btrfs_ordered_extent *entry, | |
788 | int wait) | |
e6dcd2dc CM |
789 | { |
790 | u64 start = entry->file_offset; | |
791 | u64 end = start + entry->len - 1; | |
e1b81e67 | 792 | |
1abe9b8a | 793 | trace_btrfs_ordered_extent_start(inode, entry); |
794 | ||
eb84ae03 CM |
795 | /* |
796 | * pages in the range can be dirty, clean or writeback. We | |
797 | * start IO on any dirty ones so the wait doesn't stall waiting | |
b2570314 | 798 | * for the flusher thread to find them |
eb84ae03 | 799 | */ |
4b46fce2 JB |
800 | if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags)) |
801 | filemap_fdatawrite_range(inode->i_mapping, start, end); | |
c8b97818 | 802 | if (wait) { |
e6dcd2dc CM |
803 | wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE, |
804 | &entry->flags)); | |
c8b97818 | 805 | } |
e6dcd2dc | 806 | } |
cee36a03 | 807 | |
eb84ae03 CM |
808 | /* |
809 | * Used to wait on ordered extents across a large range of bytes. | |
810 | */ | |
0ef8b726 | 811 | int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len) |
e6dcd2dc | 812 | { |
0ef8b726 | 813 | int ret = 0; |
28aeeac1 | 814 | int ret_wb = 0; |
e6dcd2dc | 815 | u64 end; |
e5a2217e | 816 | u64 orig_end; |
e6dcd2dc | 817 | struct btrfs_ordered_extent *ordered; |
e5a2217e CM |
818 | |
819 | if (start + len < start) { | |
f421950f | 820 | orig_end = INT_LIMIT(loff_t); |
e5a2217e CM |
821 | } else { |
822 | orig_end = start + len - 1; | |
f421950f CM |
823 | if (orig_end > INT_LIMIT(loff_t)) |
824 | orig_end = INT_LIMIT(loff_t); | |
e5a2217e | 825 | } |
551ebb2d | 826 | |
e5a2217e CM |
827 | /* start IO across the range first to instantiate any delalloc |
828 | * extents | |
829 | */ | |
728404da | 830 | ret = btrfs_fdatawrite_range(inode, start, orig_end); |
0ef8b726 JB |
831 | if (ret) |
832 | return ret; | |
728404da | 833 | |
28aeeac1 FM |
834 | /* |
835 | * If we have a writeback error don't return immediately. Wait first | |
836 | * for any ordered extents that haven't completed yet. This is to make | |
837 | * sure no one can dirty the same page ranges and call writepages() | |
838 | * before the ordered extents complete - to avoid failures (-EEXIST) | |
839 | * when adding the new ordered extents to the ordered tree. | |
840 | */ | |
841 | ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end); | |
e5a2217e | 842 | |
f421950f | 843 | end = orig_end; |
d397712b | 844 | while (1) { |
e6dcd2dc | 845 | ordered = btrfs_lookup_first_ordered_extent(inode, end); |
d397712b | 846 | if (!ordered) |
e6dcd2dc | 847 | break; |
e5a2217e | 848 | if (ordered->file_offset > orig_end) { |
e6dcd2dc CM |
849 | btrfs_put_ordered_extent(ordered); |
850 | break; | |
851 | } | |
b52abf1e | 852 | if (ordered->file_offset + ordered->len <= start) { |
e6dcd2dc CM |
853 | btrfs_put_ordered_extent(ordered); |
854 | break; | |
855 | } | |
e5a2217e | 856 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc | 857 | end = ordered->file_offset; |
0ef8b726 JB |
858 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) |
859 | ret = -EIO; | |
e6dcd2dc | 860 | btrfs_put_ordered_extent(ordered); |
0ef8b726 | 861 | if (ret || end == 0 || end == start) |
e6dcd2dc CM |
862 | break; |
863 | end--; | |
864 | } | |
28aeeac1 | 865 | return ret_wb ? ret_wb : ret; |
cee36a03 CM |
866 | } |
867 | ||
eb84ae03 CM |
868 | /* |
869 | * find an ordered extent corresponding to file_offset. return NULL if | |
870 | * nothing is found, otherwise take a reference on the extent and return it | |
871 | */ | |
e6dcd2dc CM |
872 | struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode, |
873 | u64 file_offset) | |
874 | { | |
875 | struct btrfs_ordered_inode_tree *tree; | |
876 | struct rb_node *node; | |
877 | struct btrfs_ordered_extent *entry = NULL; | |
878 | ||
879 | tree = &BTRFS_I(inode)->ordered_tree; | |
5fd02043 | 880 | spin_lock_irq(&tree->lock); |
e6dcd2dc CM |
881 | node = tree_search(tree, file_offset); |
882 | if (!node) | |
883 | goto out; | |
884 | ||
885 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
886 | if (!offset_in_entry(entry, file_offset)) | |
887 | entry = NULL; | |
888 | if (entry) | |
e76edab7 | 889 | refcount_inc(&entry->refs); |
e6dcd2dc | 890 | out: |
5fd02043 | 891 | spin_unlock_irq(&tree->lock); |
e6dcd2dc CM |
892 | return entry; |
893 | } | |
894 | ||
4b46fce2 JB |
895 | /* Since the DIO code tries to lock a wide area we need to look for any ordered |
896 | * extents that exist in the range, rather than just the start of the range. | |
897 | */ | |
a776c6fa NB |
898 | struct btrfs_ordered_extent *btrfs_lookup_ordered_range( |
899 | struct btrfs_inode *inode, u64 file_offset, u64 len) | |
4b46fce2 JB |
900 | { |
901 | struct btrfs_ordered_inode_tree *tree; | |
902 | struct rb_node *node; | |
903 | struct btrfs_ordered_extent *entry = NULL; | |
904 | ||
a776c6fa | 905 | tree = &inode->ordered_tree; |
5fd02043 | 906 | spin_lock_irq(&tree->lock); |
4b46fce2 JB |
907 | node = tree_search(tree, file_offset); |
908 | if (!node) { | |
909 | node = tree_search(tree, file_offset + len); | |
910 | if (!node) | |
911 | goto out; | |
912 | } | |
913 | ||
914 | while (1) { | |
915 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
916 | if (range_overlaps(entry, file_offset, len)) | |
917 | break; | |
918 | ||
919 | if (entry->file_offset >= file_offset + len) { | |
920 | entry = NULL; | |
921 | break; | |
922 | } | |
923 | entry = NULL; | |
924 | node = rb_next(node); | |
925 | if (!node) | |
926 | break; | |
927 | } | |
928 | out: | |
929 | if (entry) | |
e76edab7 | 930 | refcount_inc(&entry->refs); |
5fd02043 | 931 | spin_unlock_irq(&tree->lock); |
4b46fce2 JB |
932 | return entry; |
933 | } | |
934 | ||
b659ef02 FM |
935 | bool btrfs_have_ordered_extents_in_range(struct inode *inode, |
936 | u64 file_offset, | |
937 | u64 len) | |
938 | { | |
939 | struct btrfs_ordered_extent *oe; | |
940 | ||
a776c6fa | 941 | oe = btrfs_lookup_ordered_range(BTRFS_I(inode), file_offset, len); |
b659ef02 FM |
942 | if (oe) { |
943 | btrfs_put_ordered_extent(oe); | |
944 | return true; | |
945 | } | |
946 | return false; | |
947 | } | |
948 | ||
eb84ae03 CM |
949 | /* |
950 | * lookup and return any extent before 'file_offset'. NULL is returned | |
951 | * if none is found | |
952 | */ | |
e6dcd2dc | 953 | struct btrfs_ordered_extent * |
d397712b | 954 | btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset) |
e6dcd2dc CM |
955 | { |
956 | struct btrfs_ordered_inode_tree *tree; | |
957 | struct rb_node *node; | |
958 | struct btrfs_ordered_extent *entry = NULL; | |
959 | ||
960 | tree = &BTRFS_I(inode)->ordered_tree; | |
5fd02043 | 961 | spin_lock_irq(&tree->lock); |
e6dcd2dc CM |
962 | node = tree_search(tree, file_offset); |
963 | if (!node) | |
964 | goto out; | |
965 | ||
966 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
e76edab7 | 967 | refcount_inc(&entry->refs); |
e6dcd2dc | 968 | out: |
5fd02043 | 969 | spin_unlock_irq(&tree->lock); |
e6dcd2dc | 970 | return entry; |
81d7ed29 | 971 | } |
dbe674a9 | 972 | |
eb84ae03 CM |
973 | /* |
974 | * After an extent is done, call this to conditionally update the on disk | |
975 | * i_size. i_size is updated to cover any fully written part of the file. | |
976 | */ | |
c2167754 | 977 | int btrfs_ordered_update_i_size(struct inode *inode, u64 offset, |
dbe674a9 CM |
978 | struct btrfs_ordered_extent *ordered) |
979 | { | |
980 | struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; | |
dbe674a9 CM |
981 | u64 disk_i_size; |
982 | u64 new_i_size; | |
c2167754 | 983 | u64 i_size = i_size_read(inode); |
dbe674a9 | 984 | struct rb_node *node; |
c2167754 | 985 | struct rb_node *prev = NULL; |
dbe674a9 | 986 | struct btrfs_ordered_extent *test; |
c2167754 | 987 | int ret = 1; |
c0d2f610 | 988 | u64 orig_offset = offset; |
c2167754 | 989 | |
77cef2ec JB |
990 | spin_lock_irq(&tree->lock); |
991 | if (ordered) { | |
c2167754 | 992 | offset = entry_end(ordered); |
77cef2ec JB |
993 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags)) |
994 | offset = min(offset, | |
995 | ordered->file_offset + | |
996 | ordered->truncated_len); | |
997 | } else { | |
da17066c | 998 | offset = ALIGN(offset, btrfs_inode_sectorsize(inode)); |
77cef2ec | 999 | } |
dbe674a9 CM |
1000 | disk_i_size = BTRFS_I(inode)->disk_i_size; |
1001 | ||
19fd2df5 LB |
1002 | /* |
1003 | * truncate file. | |
1004 | * If ordered is not NULL, then this is called from endio and | |
1005 | * disk_i_size will be updated by either truncate itself or any | |
1006 | * in-flight IOs which are inside the disk_i_size. | |
1007 | * | |
1008 | * Because btrfs_setsize() may set i_size with disk_i_size if truncate | |
1009 | * fails somehow, we need to make sure we have a precise disk_i_size by | |
1010 | * updating it as usual. | |
1011 | * | |
1012 | */ | |
1013 | if (!ordered && disk_i_size > i_size) { | |
c0d2f610 | 1014 | BTRFS_I(inode)->disk_i_size = orig_offset; |
c2167754 YZ |
1015 | ret = 0; |
1016 | goto out; | |
1017 | } | |
1018 | ||
dbe674a9 CM |
1019 | /* |
1020 | * if the disk i_size is already at the inode->i_size, or | |
1021 | * this ordered extent is inside the disk i_size, we're done | |
1022 | */ | |
5d1f4020 JB |
1023 | if (disk_i_size == i_size) |
1024 | goto out; | |
1025 | ||
1026 | /* | |
1027 | * We still need to update disk_i_size if outstanding_isize is greater | |
1028 | * than disk_i_size. | |
1029 | */ | |
1030 | if (offset <= disk_i_size && | |
1031 | (!ordered || ordered->outstanding_isize <= disk_i_size)) | |
dbe674a9 | 1032 | goto out; |
dbe674a9 | 1033 | |
dbe674a9 CM |
1034 | /* |
1035 | * walk backward from this ordered extent to disk_i_size. | |
1036 | * if we find an ordered extent then we can't update disk i_size | |
1037 | * yet | |
1038 | */ | |
c2167754 YZ |
1039 | if (ordered) { |
1040 | node = rb_prev(&ordered->rb_node); | |
1041 | } else { | |
1042 | prev = tree_search(tree, offset); | |
1043 | /* | |
1044 | * we insert file extents without involving ordered struct, | |
1045 | * so there should be no ordered struct cover this offset | |
1046 | */ | |
1047 | if (prev) { | |
1048 | test = rb_entry(prev, struct btrfs_ordered_extent, | |
1049 | rb_node); | |
1050 | BUG_ON(offset_in_entry(test, offset)); | |
1051 | } | |
1052 | node = prev; | |
1053 | } | |
5fd02043 | 1054 | for (; node; node = rb_prev(node)) { |
dbe674a9 | 1055 | test = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
5fd02043 | 1056 | |
bb7ab3b9 | 1057 | /* We treat this entry as if it doesn't exist */ |
5fd02043 JB |
1058 | if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags)) |
1059 | continue; | |
62c821a8 LB |
1060 | |
1061 | if (entry_end(test) <= disk_i_size) | |
dbe674a9 | 1062 | break; |
c2167754 | 1063 | if (test->file_offset >= i_size) |
dbe674a9 | 1064 | break; |
62c821a8 LB |
1065 | |
1066 | /* | |
1067 | * We don't update disk_i_size now, so record this undealt | |
1068 | * i_size. Or we will not know the real i_size. | |
1069 | */ | |
1070 | if (test->outstanding_isize < offset) | |
1071 | test->outstanding_isize = offset; | |
1072 | if (ordered && | |
1073 | ordered->outstanding_isize > test->outstanding_isize) | |
1074 | test->outstanding_isize = ordered->outstanding_isize; | |
1075 | goto out; | |
dbe674a9 | 1076 | } |
b9a8cc5b | 1077 | new_i_size = min_t(u64, offset, i_size); |
dbe674a9 CM |
1078 | |
1079 | /* | |
b9a8cc5b MX |
1080 | * Some ordered extents may completed before the current one, and |
1081 | * we hold the real i_size in ->outstanding_isize. | |
dbe674a9 | 1082 | */ |
b9a8cc5b MX |
1083 | if (ordered && ordered->outstanding_isize > new_i_size) |
1084 | new_i_size = min_t(u64, ordered->outstanding_isize, i_size); | |
dbe674a9 | 1085 | BTRFS_I(inode)->disk_i_size = new_i_size; |
c2167754 | 1086 | ret = 0; |
dbe674a9 | 1087 | out: |
c2167754 | 1088 | /* |
5fd02043 JB |
1089 | * We need to do this because we can't remove ordered extents until |
1090 | * after the i_disk_size has been updated and then the inode has been | |
1091 | * updated to reflect the change, so we need to tell anybody who finds | |
1092 | * this ordered extent that we've already done all the real work, we | |
1093 | * just haven't completed all the other work. | |
c2167754 YZ |
1094 | */ |
1095 | if (ordered) | |
5fd02043 JB |
1096 | set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags); |
1097 | spin_unlock_irq(&tree->lock); | |
c2167754 | 1098 | return ret; |
dbe674a9 | 1099 | } |
ba1da2f4 | 1100 | |
eb84ae03 CM |
1101 | /* |
1102 | * search the ordered extents for one corresponding to 'offset' and | |
1103 | * try to find a checksum. This is used because we allow pages to | |
1104 | * be reclaimed before their checksum is actually put into the btree | |
1105 | */ | |
d20f7043 | 1106 | int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr, |
e4100d98 | 1107 | u32 *sum, int len) |
ba1da2f4 CM |
1108 | { |
1109 | struct btrfs_ordered_sum *ordered_sum; | |
ba1da2f4 CM |
1110 | struct btrfs_ordered_extent *ordered; |
1111 | struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; | |
3edf7d33 CM |
1112 | unsigned long num_sectors; |
1113 | unsigned long i; | |
da17066c | 1114 | u32 sectorsize = btrfs_inode_sectorsize(inode); |
e4100d98 | 1115 | int index = 0; |
ba1da2f4 CM |
1116 | |
1117 | ordered = btrfs_lookup_ordered_extent(inode, offset); | |
1118 | if (!ordered) | |
e4100d98 | 1119 | return 0; |
ba1da2f4 | 1120 | |
5fd02043 | 1121 | spin_lock_irq(&tree->lock); |
c6e30871 | 1122 | list_for_each_entry_reverse(ordered_sum, &ordered->list, list) { |
e4100d98 MX |
1123 | if (disk_bytenr >= ordered_sum->bytenr && |
1124 | disk_bytenr < ordered_sum->bytenr + ordered_sum->len) { | |
1125 | i = (disk_bytenr - ordered_sum->bytenr) >> | |
1126 | inode->i_sb->s_blocksize_bits; | |
e4100d98 MX |
1127 | num_sectors = ordered_sum->len >> |
1128 | inode->i_sb->s_blocksize_bits; | |
f51a4a18 MX |
1129 | num_sectors = min_t(int, len - index, num_sectors - i); |
1130 | memcpy(sum + index, ordered_sum->sums + i, | |
1131 | num_sectors); | |
1132 | ||
1133 | index += (int)num_sectors; | |
1134 | if (index == len) | |
1135 | goto out; | |
1136 | disk_bytenr += num_sectors * sectorsize; | |
ba1da2f4 CM |
1137 | } |
1138 | } | |
1139 | out: | |
5fd02043 | 1140 | spin_unlock_irq(&tree->lock); |
89642229 | 1141 | btrfs_put_ordered_extent(ordered); |
e4100d98 | 1142 | return index; |
ba1da2f4 CM |
1143 | } |
1144 | ||
6352b91d MX |
1145 | int __init ordered_data_init(void) |
1146 | { | |
1147 | btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent", | |
1148 | sizeof(struct btrfs_ordered_extent), 0, | |
fba4b697 | 1149 | SLAB_MEM_SPREAD, |
6352b91d MX |
1150 | NULL); |
1151 | if (!btrfs_ordered_extent_cache) | |
1152 | return -ENOMEM; | |
25287e0a | 1153 | |
6352b91d MX |
1154 | return 0; |
1155 | } | |
1156 | ||
1157 | void ordered_data_exit(void) | |
1158 | { | |
5598e900 | 1159 | kmem_cache_destroy(btrfs_ordered_extent_cache); |
6352b91d | 1160 | } |