]>
Commit | Line | Data |
---|---|---|
1 | #include <linux/bitops.h> | |
2 | #include <linux/slab.h> | |
3 | #include <linux/bio.h> | |
4 | #include <linux/mm.h> | |
5 | #include <linux/pagemap.h> | |
6 | #include <linux/page-flags.h> | |
7 | #include <linux/module.h> | |
8 | #include <linux/spinlock.h> | |
9 | #include <linux/blkdev.h> | |
10 | #include <linux/swap.h> | |
11 | #include <linux/writeback.h> | |
12 | #include <linux/pagevec.h> | |
13 | #include <linux/prefetch.h> | |
14 | #include <linux/cleancache.h> | |
15 | #include "extent_io.h" | |
16 | #include "extent_map.h" | |
17 | #include "compat.h" | |
18 | #include "ctree.h" | |
19 | #include "btrfs_inode.h" | |
20 | #include "volumes.h" | |
21 | #include "check-integrity.h" | |
22 | #include "locking.h" | |
23 | #include "rcu-string.h" | |
24 | ||
25 | static struct kmem_cache *extent_state_cache; | |
26 | static struct kmem_cache *extent_buffer_cache; | |
27 | ||
28 | static LIST_HEAD(buffers); | |
29 | static LIST_HEAD(states); | |
30 | ||
31 | #define LEAK_DEBUG 0 | |
32 | #if LEAK_DEBUG | |
33 | static DEFINE_SPINLOCK(leak_lock); | |
34 | #endif | |
35 | ||
36 | #define BUFFER_LRU_MAX 64 | |
37 | ||
38 | struct tree_entry { | |
39 | u64 start; | |
40 | u64 end; | |
41 | struct rb_node rb_node; | |
42 | }; | |
43 | ||
44 | struct extent_page_data { | |
45 | struct bio *bio; | |
46 | struct extent_io_tree *tree; | |
47 | get_extent_t *get_extent; | |
48 | unsigned long bio_flags; | |
49 | ||
50 | /* tells writepage not to lock the state bits for this range | |
51 | * it still does the unlocking | |
52 | */ | |
53 | unsigned int extent_locked:1; | |
54 | ||
55 | /* tells the submit_bio code to use a WRITE_SYNC */ | |
56 | unsigned int sync_io:1; | |
57 | }; | |
58 | ||
59 | static noinline void flush_write_bio(void *data); | |
60 | static inline struct btrfs_fs_info * | |
61 | tree_fs_info(struct extent_io_tree *tree) | |
62 | { | |
63 | return btrfs_sb(tree->mapping->host->i_sb); | |
64 | } | |
65 | ||
66 | int __init extent_io_init(void) | |
67 | { | |
68 | extent_state_cache = kmem_cache_create("btrfs_extent_state", | |
69 | sizeof(struct extent_state), 0, | |
70 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
71 | if (!extent_state_cache) | |
72 | return -ENOMEM; | |
73 | ||
74 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", | |
75 | sizeof(struct extent_buffer), 0, | |
76 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
77 | if (!extent_buffer_cache) | |
78 | goto free_state_cache; | |
79 | return 0; | |
80 | ||
81 | free_state_cache: | |
82 | kmem_cache_destroy(extent_state_cache); | |
83 | return -ENOMEM; | |
84 | } | |
85 | ||
86 | void extent_io_exit(void) | |
87 | { | |
88 | struct extent_state *state; | |
89 | struct extent_buffer *eb; | |
90 | ||
91 | while (!list_empty(&states)) { | |
92 | state = list_entry(states.next, struct extent_state, leak_list); | |
93 | printk(KERN_ERR "btrfs state leak: start %llu end %llu " | |
94 | "state %lu in tree %p refs %d\n", | |
95 | (unsigned long long)state->start, | |
96 | (unsigned long long)state->end, | |
97 | state->state, state->tree, atomic_read(&state->refs)); | |
98 | list_del(&state->leak_list); | |
99 | kmem_cache_free(extent_state_cache, state); | |
100 | ||
101 | } | |
102 | ||
103 | while (!list_empty(&buffers)) { | |
104 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
105 | printk(KERN_ERR "btrfs buffer leak start %llu len %lu " | |
106 | "refs %d\n", (unsigned long long)eb->start, | |
107 | eb->len, atomic_read(&eb->refs)); | |
108 | list_del(&eb->leak_list); | |
109 | kmem_cache_free(extent_buffer_cache, eb); | |
110 | } | |
111 | if (extent_state_cache) | |
112 | kmem_cache_destroy(extent_state_cache); | |
113 | if (extent_buffer_cache) | |
114 | kmem_cache_destroy(extent_buffer_cache); | |
115 | } | |
116 | ||
117 | void extent_io_tree_init(struct extent_io_tree *tree, | |
118 | struct address_space *mapping) | |
119 | { | |
120 | tree->state = RB_ROOT; | |
121 | INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC); | |
122 | tree->ops = NULL; | |
123 | tree->dirty_bytes = 0; | |
124 | spin_lock_init(&tree->lock); | |
125 | spin_lock_init(&tree->buffer_lock); | |
126 | tree->mapping = mapping; | |
127 | } | |
128 | ||
129 | static struct extent_state *alloc_extent_state(gfp_t mask) | |
130 | { | |
131 | struct extent_state *state; | |
132 | #if LEAK_DEBUG | |
133 | unsigned long flags; | |
134 | #endif | |
135 | ||
136 | state = kmem_cache_alloc(extent_state_cache, mask); | |
137 | if (!state) | |
138 | return state; | |
139 | state->state = 0; | |
140 | state->private = 0; | |
141 | state->tree = NULL; | |
142 | #if LEAK_DEBUG | |
143 | spin_lock_irqsave(&leak_lock, flags); | |
144 | list_add(&state->leak_list, &states); | |
145 | spin_unlock_irqrestore(&leak_lock, flags); | |
146 | #endif | |
147 | atomic_set(&state->refs, 1); | |
148 | init_waitqueue_head(&state->wq); | |
149 | trace_alloc_extent_state(state, mask, _RET_IP_); | |
150 | return state; | |
151 | } | |
152 | ||
153 | void free_extent_state(struct extent_state *state) | |
154 | { | |
155 | if (!state) | |
156 | return; | |
157 | if (atomic_dec_and_test(&state->refs)) { | |
158 | #if LEAK_DEBUG | |
159 | unsigned long flags; | |
160 | #endif | |
161 | WARN_ON(state->tree); | |
162 | #if LEAK_DEBUG | |
163 | spin_lock_irqsave(&leak_lock, flags); | |
164 | list_del(&state->leak_list); | |
165 | spin_unlock_irqrestore(&leak_lock, flags); | |
166 | #endif | |
167 | trace_free_extent_state(state, _RET_IP_); | |
168 | kmem_cache_free(extent_state_cache, state); | |
169 | } | |
170 | } | |
171 | ||
172 | static struct rb_node *tree_insert(struct rb_root *root, u64 offset, | |
173 | struct rb_node *node) | |
174 | { | |
175 | struct rb_node **p = &root->rb_node; | |
176 | struct rb_node *parent = NULL; | |
177 | struct tree_entry *entry; | |
178 | ||
179 | while (*p) { | |
180 | parent = *p; | |
181 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
182 | ||
183 | if (offset < entry->start) | |
184 | p = &(*p)->rb_left; | |
185 | else if (offset > entry->end) | |
186 | p = &(*p)->rb_right; | |
187 | else | |
188 | return parent; | |
189 | } | |
190 | ||
191 | rb_link_node(node, parent, p); | |
192 | rb_insert_color(node, root); | |
193 | return NULL; | |
194 | } | |
195 | ||
196 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, | |
197 | struct rb_node **prev_ret, | |
198 | struct rb_node **next_ret) | |
199 | { | |
200 | struct rb_root *root = &tree->state; | |
201 | struct rb_node *n = root->rb_node; | |
202 | struct rb_node *prev = NULL; | |
203 | struct rb_node *orig_prev = NULL; | |
204 | struct tree_entry *entry; | |
205 | struct tree_entry *prev_entry = NULL; | |
206 | ||
207 | while (n) { | |
208 | entry = rb_entry(n, struct tree_entry, rb_node); | |
209 | prev = n; | |
210 | prev_entry = entry; | |
211 | ||
212 | if (offset < entry->start) | |
213 | n = n->rb_left; | |
214 | else if (offset > entry->end) | |
215 | n = n->rb_right; | |
216 | else | |
217 | return n; | |
218 | } | |
219 | ||
220 | if (prev_ret) { | |
221 | orig_prev = prev; | |
222 | while (prev && offset > prev_entry->end) { | |
223 | prev = rb_next(prev); | |
224 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
225 | } | |
226 | *prev_ret = prev; | |
227 | prev = orig_prev; | |
228 | } | |
229 | ||
230 | if (next_ret) { | |
231 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
232 | while (prev && offset < prev_entry->start) { | |
233 | prev = rb_prev(prev); | |
234 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
235 | } | |
236 | *next_ret = prev; | |
237 | } | |
238 | return NULL; | |
239 | } | |
240 | ||
241 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, | |
242 | u64 offset) | |
243 | { | |
244 | struct rb_node *prev = NULL; | |
245 | struct rb_node *ret; | |
246 | ||
247 | ret = __etree_search(tree, offset, &prev, NULL); | |
248 | if (!ret) | |
249 | return prev; | |
250 | return ret; | |
251 | } | |
252 | ||
253 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, | |
254 | struct extent_state *other) | |
255 | { | |
256 | if (tree->ops && tree->ops->merge_extent_hook) | |
257 | tree->ops->merge_extent_hook(tree->mapping->host, new, | |
258 | other); | |
259 | } | |
260 | ||
261 | /* | |
262 | * utility function to look for merge candidates inside a given range. | |
263 | * Any extents with matching state are merged together into a single | |
264 | * extent in the tree. Extents with EXTENT_IO in their state field | |
265 | * are not merged because the end_io handlers need to be able to do | |
266 | * operations on them without sleeping (or doing allocations/splits). | |
267 | * | |
268 | * This should be called with the tree lock held. | |
269 | */ | |
270 | static void merge_state(struct extent_io_tree *tree, | |
271 | struct extent_state *state) | |
272 | { | |
273 | struct extent_state *other; | |
274 | struct rb_node *other_node; | |
275 | ||
276 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) | |
277 | return; | |
278 | ||
279 | other_node = rb_prev(&state->rb_node); | |
280 | if (other_node) { | |
281 | other = rb_entry(other_node, struct extent_state, rb_node); | |
282 | if (other->end == state->start - 1 && | |
283 | other->state == state->state) { | |
284 | merge_cb(tree, state, other); | |
285 | state->start = other->start; | |
286 | other->tree = NULL; | |
287 | rb_erase(&other->rb_node, &tree->state); | |
288 | free_extent_state(other); | |
289 | } | |
290 | } | |
291 | other_node = rb_next(&state->rb_node); | |
292 | if (other_node) { | |
293 | other = rb_entry(other_node, struct extent_state, rb_node); | |
294 | if (other->start == state->end + 1 && | |
295 | other->state == state->state) { | |
296 | merge_cb(tree, state, other); | |
297 | state->end = other->end; | |
298 | other->tree = NULL; | |
299 | rb_erase(&other->rb_node, &tree->state); | |
300 | free_extent_state(other); | |
301 | } | |
302 | } | |
303 | } | |
304 | ||
305 | static void set_state_cb(struct extent_io_tree *tree, | |
306 | struct extent_state *state, int *bits) | |
307 | { | |
308 | if (tree->ops && tree->ops->set_bit_hook) | |
309 | tree->ops->set_bit_hook(tree->mapping->host, state, bits); | |
310 | } | |
311 | ||
312 | static void clear_state_cb(struct extent_io_tree *tree, | |
313 | struct extent_state *state, int *bits) | |
314 | { | |
315 | if (tree->ops && tree->ops->clear_bit_hook) | |
316 | tree->ops->clear_bit_hook(tree->mapping->host, state, bits); | |
317 | } | |
318 | ||
319 | static void set_state_bits(struct extent_io_tree *tree, | |
320 | struct extent_state *state, int *bits); | |
321 | ||
322 | /* | |
323 | * insert an extent_state struct into the tree. 'bits' are set on the | |
324 | * struct before it is inserted. | |
325 | * | |
326 | * This may return -EEXIST if the extent is already there, in which case the | |
327 | * state struct is freed. | |
328 | * | |
329 | * The tree lock is not taken internally. This is a utility function and | |
330 | * probably isn't what you want to call (see set/clear_extent_bit). | |
331 | */ | |
332 | static int insert_state(struct extent_io_tree *tree, | |
333 | struct extent_state *state, u64 start, u64 end, | |
334 | int *bits) | |
335 | { | |
336 | struct rb_node *node; | |
337 | ||
338 | if (end < start) { | |
339 | printk(KERN_ERR "btrfs end < start %llu %llu\n", | |
340 | (unsigned long long)end, | |
341 | (unsigned long long)start); | |
342 | WARN_ON(1); | |
343 | } | |
344 | state->start = start; | |
345 | state->end = end; | |
346 | ||
347 | set_state_bits(tree, state, bits); | |
348 | ||
349 | node = tree_insert(&tree->state, end, &state->rb_node); | |
350 | if (node) { | |
351 | struct extent_state *found; | |
352 | found = rb_entry(node, struct extent_state, rb_node); | |
353 | printk(KERN_ERR "btrfs found node %llu %llu on insert of " | |
354 | "%llu %llu\n", (unsigned long long)found->start, | |
355 | (unsigned long long)found->end, | |
356 | (unsigned long long)start, (unsigned long long)end); | |
357 | return -EEXIST; | |
358 | } | |
359 | state->tree = tree; | |
360 | merge_state(tree, state); | |
361 | return 0; | |
362 | } | |
363 | ||
364 | static void split_cb(struct extent_io_tree *tree, struct extent_state *orig, | |
365 | u64 split) | |
366 | { | |
367 | if (tree->ops && tree->ops->split_extent_hook) | |
368 | tree->ops->split_extent_hook(tree->mapping->host, orig, split); | |
369 | } | |
370 | ||
371 | /* | |
372 | * split a given extent state struct in two, inserting the preallocated | |
373 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
374 | * offset inside 'orig' where it should be split. | |
375 | * | |
376 | * Before calling, | |
377 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
378 | * are two extent state structs in the tree: | |
379 | * prealloc: [orig->start, split - 1] | |
380 | * orig: [ split, orig->end ] | |
381 | * | |
382 | * The tree locks are not taken by this function. They need to be held | |
383 | * by the caller. | |
384 | */ | |
385 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
386 | struct extent_state *prealloc, u64 split) | |
387 | { | |
388 | struct rb_node *node; | |
389 | ||
390 | split_cb(tree, orig, split); | |
391 | ||
392 | prealloc->start = orig->start; | |
393 | prealloc->end = split - 1; | |
394 | prealloc->state = orig->state; | |
395 | orig->start = split; | |
396 | ||
397 | node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node); | |
398 | if (node) { | |
399 | free_extent_state(prealloc); | |
400 | return -EEXIST; | |
401 | } | |
402 | prealloc->tree = tree; | |
403 | return 0; | |
404 | } | |
405 | ||
406 | static struct extent_state *next_state(struct extent_state *state) | |
407 | { | |
408 | struct rb_node *next = rb_next(&state->rb_node); | |
409 | if (next) | |
410 | return rb_entry(next, struct extent_state, rb_node); | |
411 | else | |
412 | return NULL; | |
413 | } | |
414 | ||
415 | /* | |
416 | * utility function to clear some bits in an extent state struct. | |
417 | * it will optionally wake up any one waiting on this state (wake == 1). | |
418 | * | |
419 | * If no bits are set on the state struct after clearing things, the | |
420 | * struct is freed and removed from the tree | |
421 | */ | |
422 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, | |
423 | struct extent_state *state, | |
424 | int *bits, int wake) | |
425 | { | |
426 | struct extent_state *next; | |
427 | int bits_to_clear = *bits & ~EXTENT_CTLBITS; | |
428 | ||
429 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { | |
430 | u64 range = state->end - state->start + 1; | |
431 | WARN_ON(range > tree->dirty_bytes); | |
432 | tree->dirty_bytes -= range; | |
433 | } | |
434 | clear_state_cb(tree, state, bits); | |
435 | state->state &= ~bits_to_clear; | |
436 | if (wake) | |
437 | wake_up(&state->wq); | |
438 | if (state->state == 0) { | |
439 | next = next_state(state); | |
440 | if (state->tree) { | |
441 | rb_erase(&state->rb_node, &tree->state); | |
442 | state->tree = NULL; | |
443 | free_extent_state(state); | |
444 | } else { | |
445 | WARN_ON(1); | |
446 | } | |
447 | } else { | |
448 | merge_state(tree, state); | |
449 | next = next_state(state); | |
450 | } | |
451 | return next; | |
452 | } | |
453 | ||
454 | static struct extent_state * | |
455 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
456 | { | |
457 | if (!prealloc) | |
458 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
459 | ||
460 | return prealloc; | |
461 | } | |
462 | ||
463 | void extent_io_tree_panic(struct extent_io_tree *tree, int err) | |
464 | { | |
465 | btrfs_panic(tree_fs_info(tree), err, "Locking error: " | |
466 | "Extent tree was modified by another " | |
467 | "thread while locked."); | |
468 | } | |
469 | ||
470 | /* | |
471 | * clear some bits on a range in the tree. This may require splitting | |
472 | * or inserting elements in the tree, so the gfp mask is used to | |
473 | * indicate which allocations or sleeping are allowed. | |
474 | * | |
475 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
476 | * the given range from the tree regardless of state (ie for truncate). | |
477 | * | |
478 | * the range [start, end] is inclusive. | |
479 | * | |
480 | * This takes the tree lock, and returns 0 on success and < 0 on error. | |
481 | */ | |
482 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
483 | int bits, int wake, int delete, | |
484 | struct extent_state **cached_state, | |
485 | gfp_t mask) | |
486 | { | |
487 | struct extent_state *state; | |
488 | struct extent_state *cached; | |
489 | struct extent_state *prealloc = NULL; | |
490 | struct rb_node *node; | |
491 | u64 last_end; | |
492 | int err; | |
493 | int clear = 0; | |
494 | ||
495 | if (delete) | |
496 | bits |= ~EXTENT_CTLBITS; | |
497 | bits |= EXTENT_FIRST_DELALLOC; | |
498 | ||
499 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) | |
500 | clear = 1; | |
501 | again: | |
502 | if (!prealloc && (mask & __GFP_WAIT)) { | |
503 | prealloc = alloc_extent_state(mask); | |
504 | if (!prealloc) | |
505 | return -ENOMEM; | |
506 | } | |
507 | ||
508 | spin_lock(&tree->lock); | |
509 | if (cached_state) { | |
510 | cached = *cached_state; | |
511 | ||
512 | if (clear) { | |
513 | *cached_state = NULL; | |
514 | cached_state = NULL; | |
515 | } | |
516 | ||
517 | if (cached && cached->tree && cached->start <= start && | |
518 | cached->end > start) { | |
519 | if (clear) | |
520 | atomic_dec(&cached->refs); | |
521 | state = cached; | |
522 | goto hit_next; | |
523 | } | |
524 | if (clear) | |
525 | free_extent_state(cached); | |
526 | } | |
527 | /* | |
528 | * this search will find the extents that end after | |
529 | * our range starts | |
530 | */ | |
531 | node = tree_search(tree, start); | |
532 | if (!node) | |
533 | goto out; | |
534 | state = rb_entry(node, struct extent_state, rb_node); | |
535 | hit_next: | |
536 | if (state->start > end) | |
537 | goto out; | |
538 | WARN_ON(state->end < start); | |
539 | last_end = state->end; | |
540 | ||
541 | /* the state doesn't have the wanted bits, go ahead */ | |
542 | if (!(state->state & bits)) { | |
543 | state = next_state(state); | |
544 | goto next; | |
545 | } | |
546 | ||
547 | /* | |
548 | * | ---- desired range ---- | | |
549 | * | state | or | |
550 | * | ------------- state -------------- | | |
551 | * | |
552 | * We need to split the extent we found, and may flip | |
553 | * bits on second half. | |
554 | * | |
555 | * If the extent we found extends past our range, we | |
556 | * just split and search again. It'll get split again | |
557 | * the next time though. | |
558 | * | |
559 | * If the extent we found is inside our range, we clear | |
560 | * the desired bit on it. | |
561 | */ | |
562 | ||
563 | if (state->start < start) { | |
564 | prealloc = alloc_extent_state_atomic(prealloc); | |
565 | BUG_ON(!prealloc); | |
566 | err = split_state(tree, state, prealloc, start); | |
567 | if (err) | |
568 | extent_io_tree_panic(tree, err); | |
569 | ||
570 | prealloc = NULL; | |
571 | if (err) | |
572 | goto out; | |
573 | if (state->end <= end) { | |
574 | state = clear_state_bit(tree, state, &bits, wake); | |
575 | goto next; | |
576 | } | |
577 | goto search_again; | |
578 | } | |
579 | /* | |
580 | * | ---- desired range ---- | | |
581 | * | state | | |
582 | * We need to split the extent, and clear the bit | |
583 | * on the first half | |
584 | */ | |
585 | if (state->start <= end && state->end > end) { | |
586 | prealloc = alloc_extent_state_atomic(prealloc); | |
587 | BUG_ON(!prealloc); | |
588 | err = split_state(tree, state, prealloc, end + 1); | |
589 | if (err) | |
590 | extent_io_tree_panic(tree, err); | |
591 | ||
592 | if (wake) | |
593 | wake_up(&state->wq); | |
594 | ||
595 | clear_state_bit(tree, prealloc, &bits, wake); | |
596 | ||
597 | prealloc = NULL; | |
598 | goto out; | |
599 | } | |
600 | ||
601 | state = clear_state_bit(tree, state, &bits, wake); | |
602 | next: | |
603 | if (last_end == (u64)-1) | |
604 | goto out; | |
605 | start = last_end + 1; | |
606 | if (start <= end && state && !need_resched()) | |
607 | goto hit_next; | |
608 | goto search_again; | |
609 | ||
610 | out: | |
611 | spin_unlock(&tree->lock); | |
612 | if (prealloc) | |
613 | free_extent_state(prealloc); | |
614 | ||
615 | return 0; | |
616 | ||
617 | search_again: | |
618 | if (start > end) | |
619 | goto out; | |
620 | spin_unlock(&tree->lock); | |
621 | if (mask & __GFP_WAIT) | |
622 | cond_resched(); | |
623 | goto again; | |
624 | } | |
625 | ||
626 | static void wait_on_state(struct extent_io_tree *tree, | |
627 | struct extent_state *state) | |
628 | __releases(tree->lock) | |
629 | __acquires(tree->lock) | |
630 | { | |
631 | DEFINE_WAIT(wait); | |
632 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
633 | spin_unlock(&tree->lock); | |
634 | schedule(); | |
635 | spin_lock(&tree->lock); | |
636 | finish_wait(&state->wq, &wait); | |
637 | } | |
638 | ||
639 | /* | |
640 | * waits for one or more bits to clear on a range in the state tree. | |
641 | * The range [start, end] is inclusive. | |
642 | * The tree lock is taken by this function | |
643 | */ | |
644 | void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits) | |
645 | { | |
646 | struct extent_state *state; | |
647 | struct rb_node *node; | |
648 | ||
649 | spin_lock(&tree->lock); | |
650 | again: | |
651 | while (1) { | |
652 | /* | |
653 | * this search will find all the extents that end after | |
654 | * our range starts | |
655 | */ | |
656 | node = tree_search(tree, start); | |
657 | if (!node) | |
658 | break; | |
659 | ||
660 | state = rb_entry(node, struct extent_state, rb_node); | |
661 | ||
662 | if (state->start > end) | |
663 | goto out; | |
664 | ||
665 | if (state->state & bits) { | |
666 | start = state->start; | |
667 | atomic_inc(&state->refs); | |
668 | wait_on_state(tree, state); | |
669 | free_extent_state(state); | |
670 | goto again; | |
671 | } | |
672 | start = state->end + 1; | |
673 | ||
674 | if (start > end) | |
675 | break; | |
676 | ||
677 | cond_resched_lock(&tree->lock); | |
678 | } | |
679 | out: | |
680 | spin_unlock(&tree->lock); | |
681 | } | |
682 | ||
683 | static void set_state_bits(struct extent_io_tree *tree, | |
684 | struct extent_state *state, | |
685 | int *bits) | |
686 | { | |
687 | int bits_to_set = *bits & ~EXTENT_CTLBITS; | |
688 | ||
689 | set_state_cb(tree, state, bits); | |
690 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { | |
691 | u64 range = state->end - state->start + 1; | |
692 | tree->dirty_bytes += range; | |
693 | } | |
694 | state->state |= bits_to_set; | |
695 | } | |
696 | ||
697 | static void cache_state(struct extent_state *state, | |
698 | struct extent_state **cached_ptr) | |
699 | { | |
700 | if (cached_ptr && !(*cached_ptr)) { | |
701 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) { | |
702 | *cached_ptr = state; | |
703 | atomic_inc(&state->refs); | |
704 | } | |
705 | } | |
706 | } | |
707 | ||
708 | static void uncache_state(struct extent_state **cached_ptr) | |
709 | { | |
710 | if (cached_ptr && (*cached_ptr)) { | |
711 | struct extent_state *state = *cached_ptr; | |
712 | *cached_ptr = NULL; | |
713 | free_extent_state(state); | |
714 | } | |
715 | } | |
716 | ||
717 | /* | |
718 | * set some bits on a range in the tree. This may require allocations or | |
719 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
720 | * | |
721 | * If any of the exclusive bits are set, this will fail with -EEXIST if some | |
722 | * part of the range already has the desired bits set. The start of the | |
723 | * existing range is returned in failed_start in this case. | |
724 | * | |
725 | * [start, end] is inclusive This takes the tree lock. | |
726 | */ | |
727 | ||
728 | static int __must_check | |
729 | __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
730 | int bits, int exclusive_bits, u64 *failed_start, | |
731 | struct extent_state **cached_state, gfp_t mask) | |
732 | { | |
733 | struct extent_state *state; | |
734 | struct extent_state *prealloc = NULL; | |
735 | struct rb_node *node; | |
736 | int err = 0; | |
737 | u64 last_start; | |
738 | u64 last_end; | |
739 | ||
740 | bits |= EXTENT_FIRST_DELALLOC; | |
741 | again: | |
742 | if (!prealloc && (mask & __GFP_WAIT)) { | |
743 | prealloc = alloc_extent_state(mask); | |
744 | BUG_ON(!prealloc); | |
745 | } | |
746 | ||
747 | spin_lock(&tree->lock); | |
748 | if (cached_state && *cached_state) { | |
749 | state = *cached_state; | |
750 | if (state->start <= start && state->end > start && | |
751 | state->tree) { | |
752 | node = &state->rb_node; | |
753 | goto hit_next; | |
754 | } | |
755 | } | |
756 | /* | |
757 | * this search will find all the extents that end after | |
758 | * our range starts. | |
759 | */ | |
760 | node = tree_search(tree, start); | |
761 | if (!node) { | |
762 | prealloc = alloc_extent_state_atomic(prealloc); | |
763 | BUG_ON(!prealloc); | |
764 | err = insert_state(tree, prealloc, start, end, &bits); | |
765 | if (err) | |
766 | extent_io_tree_panic(tree, err); | |
767 | ||
768 | prealloc = NULL; | |
769 | goto out; | |
770 | } | |
771 | state = rb_entry(node, struct extent_state, rb_node); | |
772 | hit_next: | |
773 | last_start = state->start; | |
774 | last_end = state->end; | |
775 | ||
776 | /* | |
777 | * | ---- desired range ---- | | |
778 | * | state | | |
779 | * | |
780 | * Just lock what we found and keep going | |
781 | */ | |
782 | if (state->start == start && state->end <= end) { | |
783 | if (state->state & exclusive_bits) { | |
784 | *failed_start = state->start; | |
785 | err = -EEXIST; | |
786 | goto out; | |
787 | } | |
788 | ||
789 | set_state_bits(tree, state, &bits); | |
790 | cache_state(state, cached_state); | |
791 | merge_state(tree, state); | |
792 | if (last_end == (u64)-1) | |
793 | goto out; | |
794 | start = last_end + 1; | |
795 | state = next_state(state); | |
796 | if (start < end && state && state->start == start && | |
797 | !need_resched()) | |
798 | goto hit_next; | |
799 | goto search_again; | |
800 | } | |
801 | ||
802 | /* | |
803 | * | ---- desired range ---- | | |
804 | * | state | | |
805 | * or | |
806 | * | ------------- state -------------- | | |
807 | * | |
808 | * We need to split the extent we found, and may flip bits on | |
809 | * second half. | |
810 | * | |
811 | * If the extent we found extends past our | |
812 | * range, we just split and search again. It'll get split | |
813 | * again the next time though. | |
814 | * | |
815 | * If the extent we found is inside our range, we set the | |
816 | * desired bit on it. | |
817 | */ | |
818 | if (state->start < start) { | |
819 | if (state->state & exclusive_bits) { | |
820 | *failed_start = start; | |
821 | err = -EEXIST; | |
822 | goto out; | |
823 | } | |
824 | ||
825 | prealloc = alloc_extent_state_atomic(prealloc); | |
826 | BUG_ON(!prealloc); | |
827 | err = split_state(tree, state, prealloc, start); | |
828 | if (err) | |
829 | extent_io_tree_panic(tree, err); | |
830 | ||
831 | prealloc = NULL; | |
832 | if (err) | |
833 | goto out; | |
834 | if (state->end <= end) { | |
835 | set_state_bits(tree, state, &bits); | |
836 | cache_state(state, cached_state); | |
837 | merge_state(tree, state); | |
838 | if (last_end == (u64)-1) | |
839 | goto out; | |
840 | start = last_end + 1; | |
841 | state = next_state(state); | |
842 | if (start < end && state && state->start == start && | |
843 | !need_resched()) | |
844 | goto hit_next; | |
845 | } | |
846 | goto search_again; | |
847 | } | |
848 | /* | |
849 | * | ---- desired range ---- | | |
850 | * | state | or | state | | |
851 | * | |
852 | * There's a hole, we need to insert something in it and | |
853 | * ignore the extent we found. | |
854 | */ | |
855 | if (state->start > start) { | |
856 | u64 this_end; | |
857 | if (end < last_start) | |
858 | this_end = end; | |
859 | else | |
860 | this_end = last_start - 1; | |
861 | ||
862 | prealloc = alloc_extent_state_atomic(prealloc); | |
863 | BUG_ON(!prealloc); | |
864 | ||
865 | /* | |
866 | * Avoid to free 'prealloc' if it can be merged with | |
867 | * the later extent. | |
868 | */ | |
869 | err = insert_state(tree, prealloc, start, this_end, | |
870 | &bits); | |
871 | if (err) | |
872 | extent_io_tree_panic(tree, err); | |
873 | ||
874 | cache_state(prealloc, cached_state); | |
875 | prealloc = NULL; | |
876 | start = this_end + 1; | |
877 | goto search_again; | |
878 | } | |
879 | /* | |
880 | * | ---- desired range ---- | | |
881 | * | state | | |
882 | * We need to split the extent, and set the bit | |
883 | * on the first half | |
884 | */ | |
885 | if (state->start <= end && state->end > end) { | |
886 | if (state->state & exclusive_bits) { | |
887 | *failed_start = start; | |
888 | err = -EEXIST; | |
889 | goto out; | |
890 | } | |
891 | ||
892 | prealloc = alloc_extent_state_atomic(prealloc); | |
893 | BUG_ON(!prealloc); | |
894 | err = split_state(tree, state, prealloc, end + 1); | |
895 | if (err) | |
896 | extent_io_tree_panic(tree, err); | |
897 | ||
898 | set_state_bits(tree, prealloc, &bits); | |
899 | cache_state(prealloc, cached_state); | |
900 | merge_state(tree, prealloc); | |
901 | prealloc = NULL; | |
902 | goto out; | |
903 | } | |
904 | ||
905 | goto search_again; | |
906 | ||
907 | out: | |
908 | spin_unlock(&tree->lock); | |
909 | if (prealloc) | |
910 | free_extent_state(prealloc); | |
911 | ||
912 | return err; | |
913 | ||
914 | search_again: | |
915 | if (start > end) | |
916 | goto out; | |
917 | spin_unlock(&tree->lock); | |
918 | if (mask & __GFP_WAIT) | |
919 | cond_resched(); | |
920 | goto again; | |
921 | } | |
922 | ||
923 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits, | |
924 | u64 *failed_start, struct extent_state **cached_state, | |
925 | gfp_t mask) | |
926 | { | |
927 | return __set_extent_bit(tree, start, end, bits, 0, failed_start, | |
928 | cached_state, mask); | |
929 | } | |
930 | ||
931 | ||
932 | /** | |
933 | * convert_extent_bit - convert all bits in a given range from one bit to | |
934 | * another | |
935 | * @tree: the io tree to search | |
936 | * @start: the start offset in bytes | |
937 | * @end: the end offset in bytes (inclusive) | |
938 | * @bits: the bits to set in this range | |
939 | * @clear_bits: the bits to clear in this range | |
940 | * @cached_state: state that we're going to cache | |
941 | * @mask: the allocation mask | |
942 | * | |
943 | * This will go through and set bits for the given range. If any states exist | |
944 | * already in this range they are set with the given bit and cleared of the | |
945 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
946 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
947 | * boundary bits like LOCK. | |
948 | */ | |
949 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
950 | int bits, int clear_bits, | |
951 | struct extent_state **cached_state, gfp_t mask) | |
952 | { | |
953 | struct extent_state *state; | |
954 | struct extent_state *prealloc = NULL; | |
955 | struct rb_node *node; | |
956 | int err = 0; | |
957 | u64 last_start; | |
958 | u64 last_end; | |
959 | ||
960 | again: | |
961 | if (!prealloc && (mask & __GFP_WAIT)) { | |
962 | prealloc = alloc_extent_state(mask); | |
963 | if (!prealloc) | |
964 | return -ENOMEM; | |
965 | } | |
966 | ||
967 | spin_lock(&tree->lock); | |
968 | if (cached_state && *cached_state) { | |
969 | state = *cached_state; | |
970 | if (state->start <= start && state->end > start && | |
971 | state->tree) { | |
972 | node = &state->rb_node; | |
973 | goto hit_next; | |
974 | } | |
975 | } | |
976 | ||
977 | /* | |
978 | * this search will find all the extents that end after | |
979 | * our range starts. | |
980 | */ | |
981 | node = tree_search(tree, start); | |
982 | if (!node) { | |
983 | prealloc = alloc_extent_state_atomic(prealloc); | |
984 | if (!prealloc) { | |
985 | err = -ENOMEM; | |
986 | goto out; | |
987 | } | |
988 | err = insert_state(tree, prealloc, start, end, &bits); | |
989 | prealloc = NULL; | |
990 | if (err) | |
991 | extent_io_tree_panic(tree, err); | |
992 | goto out; | |
993 | } | |
994 | state = rb_entry(node, struct extent_state, rb_node); | |
995 | hit_next: | |
996 | last_start = state->start; | |
997 | last_end = state->end; | |
998 | ||
999 | /* | |
1000 | * | ---- desired range ---- | | |
1001 | * | state | | |
1002 | * | |
1003 | * Just lock what we found and keep going | |
1004 | */ | |
1005 | if (state->start == start && state->end <= end) { | |
1006 | set_state_bits(tree, state, &bits); | |
1007 | cache_state(state, cached_state); | |
1008 | state = clear_state_bit(tree, state, &clear_bits, 0); | |
1009 | if (last_end == (u64)-1) | |
1010 | goto out; | |
1011 | start = last_end + 1; | |
1012 | if (start < end && state && state->start == start && | |
1013 | !need_resched()) | |
1014 | goto hit_next; | |
1015 | goto search_again; | |
1016 | } | |
1017 | ||
1018 | /* | |
1019 | * | ---- desired range ---- | | |
1020 | * | state | | |
1021 | * or | |
1022 | * | ------------- state -------------- | | |
1023 | * | |
1024 | * We need to split the extent we found, and may flip bits on | |
1025 | * second half. | |
1026 | * | |
1027 | * If the extent we found extends past our | |
1028 | * range, we just split and search again. It'll get split | |
1029 | * again the next time though. | |
1030 | * | |
1031 | * If the extent we found is inside our range, we set the | |
1032 | * desired bit on it. | |
1033 | */ | |
1034 | if (state->start < start) { | |
1035 | prealloc = alloc_extent_state_atomic(prealloc); | |
1036 | if (!prealloc) { | |
1037 | err = -ENOMEM; | |
1038 | goto out; | |
1039 | } | |
1040 | err = split_state(tree, state, prealloc, start); | |
1041 | if (err) | |
1042 | extent_io_tree_panic(tree, err); | |
1043 | prealloc = NULL; | |
1044 | if (err) | |
1045 | goto out; | |
1046 | if (state->end <= end) { | |
1047 | set_state_bits(tree, state, &bits); | |
1048 | cache_state(state, cached_state); | |
1049 | state = clear_state_bit(tree, state, &clear_bits, 0); | |
1050 | if (last_end == (u64)-1) | |
1051 | goto out; | |
1052 | start = last_end + 1; | |
1053 | if (start < end && state && state->start == start && | |
1054 | !need_resched()) | |
1055 | goto hit_next; | |
1056 | } | |
1057 | goto search_again; | |
1058 | } | |
1059 | /* | |
1060 | * | ---- desired range ---- | | |
1061 | * | state | or | state | | |
1062 | * | |
1063 | * There's a hole, we need to insert something in it and | |
1064 | * ignore the extent we found. | |
1065 | */ | |
1066 | if (state->start > start) { | |
1067 | u64 this_end; | |
1068 | if (end < last_start) | |
1069 | this_end = end; | |
1070 | else | |
1071 | this_end = last_start - 1; | |
1072 | ||
1073 | prealloc = alloc_extent_state_atomic(prealloc); | |
1074 | if (!prealloc) { | |
1075 | err = -ENOMEM; | |
1076 | goto out; | |
1077 | } | |
1078 | ||
1079 | /* | |
1080 | * Avoid to free 'prealloc' if it can be merged with | |
1081 | * the later extent. | |
1082 | */ | |
1083 | err = insert_state(tree, prealloc, start, this_end, | |
1084 | &bits); | |
1085 | if (err) | |
1086 | extent_io_tree_panic(tree, err); | |
1087 | cache_state(prealloc, cached_state); | |
1088 | prealloc = NULL; | |
1089 | start = this_end + 1; | |
1090 | goto search_again; | |
1091 | } | |
1092 | /* | |
1093 | * | ---- desired range ---- | | |
1094 | * | state | | |
1095 | * We need to split the extent, and set the bit | |
1096 | * on the first half | |
1097 | */ | |
1098 | if (state->start <= end && state->end > end) { | |
1099 | prealloc = alloc_extent_state_atomic(prealloc); | |
1100 | if (!prealloc) { | |
1101 | err = -ENOMEM; | |
1102 | goto out; | |
1103 | } | |
1104 | ||
1105 | err = split_state(tree, state, prealloc, end + 1); | |
1106 | if (err) | |
1107 | extent_io_tree_panic(tree, err); | |
1108 | ||
1109 | set_state_bits(tree, prealloc, &bits); | |
1110 | cache_state(prealloc, cached_state); | |
1111 | clear_state_bit(tree, prealloc, &clear_bits, 0); | |
1112 | prealloc = NULL; | |
1113 | goto out; | |
1114 | } | |
1115 | ||
1116 | goto search_again; | |
1117 | ||
1118 | out: | |
1119 | spin_unlock(&tree->lock); | |
1120 | if (prealloc) | |
1121 | free_extent_state(prealloc); | |
1122 | ||
1123 | return err; | |
1124 | ||
1125 | search_again: | |
1126 | if (start > end) | |
1127 | goto out; | |
1128 | spin_unlock(&tree->lock); | |
1129 | if (mask & __GFP_WAIT) | |
1130 | cond_resched(); | |
1131 | goto again; | |
1132 | } | |
1133 | ||
1134 | /* wrappers around set/clear extent bit */ | |
1135 | int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
1136 | gfp_t mask) | |
1137 | { | |
1138 | return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL, | |
1139 | NULL, mask); | |
1140 | } | |
1141 | ||
1142 | int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
1143 | int bits, gfp_t mask) | |
1144 | { | |
1145 | return set_extent_bit(tree, start, end, bits, NULL, | |
1146 | NULL, mask); | |
1147 | } | |
1148 | ||
1149 | int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
1150 | int bits, gfp_t mask) | |
1151 | { | |
1152 | return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask); | |
1153 | } | |
1154 | ||
1155 | int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, | |
1156 | struct extent_state **cached_state, gfp_t mask) | |
1157 | { | |
1158 | return set_extent_bit(tree, start, end, | |
1159 | EXTENT_DELALLOC | EXTENT_UPTODATE, | |
1160 | NULL, cached_state, mask); | |
1161 | } | |
1162 | ||
1163 | int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end, | |
1164 | struct extent_state **cached_state, gfp_t mask) | |
1165 | { | |
1166 | return set_extent_bit(tree, start, end, | |
1167 | EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG, | |
1168 | NULL, cached_state, mask); | |
1169 | } | |
1170 | ||
1171 | int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
1172 | gfp_t mask) | |
1173 | { | |
1174 | return clear_extent_bit(tree, start, end, | |
1175 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
1176 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask); | |
1177 | } | |
1178 | ||
1179 | int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, | |
1180 | gfp_t mask) | |
1181 | { | |
1182 | return set_extent_bit(tree, start, end, EXTENT_NEW, NULL, | |
1183 | NULL, mask); | |
1184 | } | |
1185 | ||
1186 | int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, | |
1187 | struct extent_state **cached_state, gfp_t mask) | |
1188 | { | |
1189 | return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, | |
1190 | cached_state, mask); | |
1191 | } | |
1192 | ||
1193 | int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, | |
1194 | struct extent_state **cached_state, gfp_t mask) | |
1195 | { | |
1196 | return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, | |
1197 | cached_state, mask); | |
1198 | } | |
1199 | ||
1200 | /* | |
1201 | * either insert or lock state struct between start and end use mask to tell | |
1202 | * us if waiting is desired. | |
1203 | */ | |
1204 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
1205 | int bits, struct extent_state **cached_state) | |
1206 | { | |
1207 | int err; | |
1208 | u64 failed_start; | |
1209 | while (1) { | |
1210 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits, | |
1211 | EXTENT_LOCKED, &failed_start, | |
1212 | cached_state, GFP_NOFS); | |
1213 | if (err == -EEXIST) { | |
1214 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); | |
1215 | start = failed_start; | |
1216 | } else | |
1217 | break; | |
1218 | WARN_ON(start > end); | |
1219 | } | |
1220 | return err; | |
1221 | } | |
1222 | ||
1223 | int lock_extent(struct extent_io_tree *tree, u64 start, u64 end) | |
1224 | { | |
1225 | return lock_extent_bits(tree, start, end, 0, NULL); | |
1226 | } | |
1227 | ||
1228 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) | |
1229 | { | |
1230 | int err; | |
1231 | u64 failed_start; | |
1232 | ||
1233 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, | |
1234 | &failed_start, NULL, GFP_NOFS); | |
1235 | if (err == -EEXIST) { | |
1236 | if (failed_start > start) | |
1237 | clear_extent_bit(tree, start, failed_start - 1, | |
1238 | EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS); | |
1239 | return 0; | |
1240 | } | |
1241 | return 1; | |
1242 | } | |
1243 | ||
1244 | int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end, | |
1245 | struct extent_state **cached, gfp_t mask) | |
1246 | { | |
1247 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, | |
1248 | mask); | |
1249 | } | |
1250 | ||
1251 | int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end) | |
1252 | { | |
1253 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL, | |
1254 | GFP_NOFS); | |
1255 | } | |
1256 | ||
1257 | /* | |
1258 | * helper function to set both pages and extents in the tree writeback | |
1259 | */ | |
1260 | static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) | |
1261 | { | |
1262 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1263 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1264 | struct page *page; | |
1265 | ||
1266 | while (index <= end_index) { | |
1267 | page = find_get_page(tree->mapping, index); | |
1268 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1269 | set_page_writeback(page); | |
1270 | page_cache_release(page); | |
1271 | index++; | |
1272 | } | |
1273 | return 0; | |
1274 | } | |
1275 | ||
1276 | /* find the first state struct with 'bits' set after 'start', and | |
1277 | * return it. tree->lock must be held. NULL will returned if | |
1278 | * nothing was found after 'start' | |
1279 | */ | |
1280 | struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree, | |
1281 | u64 start, int bits) | |
1282 | { | |
1283 | struct rb_node *node; | |
1284 | struct extent_state *state; | |
1285 | ||
1286 | /* | |
1287 | * this search will find all the extents that end after | |
1288 | * our range starts. | |
1289 | */ | |
1290 | node = tree_search(tree, start); | |
1291 | if (!node) | |
1292 | goto out; | |
1293 | ||
1294 | while (1) { | |
1295 | state = rb_entry(node, struct extent_state, rb_node); | |
1296 | if (state->end >= start && (state->state & bits)) | |
1297 | return state; | |
1298 | ||
1299 | node = rb_next(node); | |
1300 | if (!node) | |
1301 | break; | |
1302 | } | |
1303 | out: | |
1304 | return NULL; | |
1305 | } | |
1306 | ||
1307 | /* | |
1308 | * find the first offset in the io tree with 'bits' set. zero is | |
1309 | * returned if we find something, and *start_ret and *end_ret are | |
1310 | * set to reflect the state struct that was found. | |
1311 | * | |
1312 | * If nothing was found, 1 is returned. If found something, return 0. | |
1313 | */ | |
1314 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
1315 | u64 *start_ret, u64 *end_ret, int bits, | |
1316 | struct extent_state **cached_state) | |
1317 | { | |
1318 | struct extent_state *state; | |
1319 | struct rb_node *n; | |
1320 | int ret = 1; | |
1321 | ||
1322 | spin_lock(&tree->lock); | |
1323 | if (cached_state && *cached_state) { | |
1324 | state = *cached_state; | |
1325 | if (state->end == start - 1 && state->tree) { | |
1326 | n = rb_next(&state->rb_node); | |
1327 | while (n) { | |
1328 | state = rb_entry(n, struct extent_state, | |
1329 | rb_node); | |
1330 | if (state->state & bits) | |
1331 | goto got_it; | |
1332 | n = rb_next(n); | |
1333 | } | |
1334 | free_extent_state(*cached_state); | |
1335 | *cached_state = NULL; | |
1336 | goto out; | |
1337 | } | |
1338 | free_extent_state(*cached_state); | |
1339 | *cached_state = NULL; | |
1340 | } | |
1341 | ||
1342 | state = find_first_extent_bit_state(tree, start, bits); | |
1343 | got_it: | |
1344 | if (state) { | |
1345 | cache_state(state, cached_state); | |
1346 | *start_ret = state->start; | |
1347 | *end_ret = state->end; | |
1348 | ret = 0; | |
1349 | } | |
1350 | out: | |
1351 | spin_unlock(&tree->lock); | |
1352 | return ret; | |
1353 | } | |
1354 | ||
1355 | /* | |
1356 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1357 | * more than 'max_bytes'. start and end are used to return the range, | |
1358 | * | |
1359 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1360 | */ | |
1361 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, | |
1362 | u64 *start, u64 *end, u64 max_bytes, | |
1363 | struct extent_state **cached_state) | |
1364 | { | |
1365 | struct rb_node *node; | |
1366 | struct extent_state *state; | |
1367 | u64 cur_start = *start; | |
1368 | u64 found = 0; | |
1369 | u64 total_bytes = 0; | |
1370 | ||
1371 | spin_lock(&tree->lock); | |
1372 | ||
1373 | /* | |
1374 | * this search will find all the extents that end after | |
1375 | * our range starts. | |
1376 | */ | |
1377 | node = tree_search(tree, cur_start); | |
1378 | if (!node) { | |
1379 | if (!found) | |
1380 | *end = (u64)-1; | |
1381 | goto out; | |
1382 | } | |
1383 | ||
1384 | while (1) { | |
1385 | state = rb_entry(node, struct extent_state, rb_node); | |
1386 | if (found && (state->start != cur_start || | |
1387 | (state->state & EXTENT_BOUNDARY))) { | |
1388 | goto out; | |
1389 | } | |
1390 | if (!(state->state & EXTENT_DELALLOC)) { | |
1391 | if (!found) | |
1392 | *end = state->end; | |
1393 | goto out; | |
1394 | } | |
1395 | if (!found) { | |
1396 | *start = state->start; | |
1397 | *cached_state = state; | |
1398 | atomic_inc(&state->refs); | |
1399 | } | |
1400 | found++; | |
1401 | *end = state->end; | |
1402 | cur_start = state->end + 1; | |
1403 | node = rb_next(node); | |
1404 | if (!node) | |
1405 | break; | |
1406 | total_bytes += state->end - state->start + 1; | |
1407 | if (total_bytes >= max_bytes) | |
1408 | break; | |
1409 | } | |
1410 | out: | |
1411 | spin_unlock(&tree->lock); | |
1412 | return found; | |
1413 | } | |
1414 | ||
1415 | static noinline void __unlock_for_delalloc(struct inode *inode, | |
1416 | struct page *locked_page, | |
1417 | u64 start, u64 end) | |
1418 | { | |
1419 | int ret; | |
1420 | struct page *pages[16]; | |
1421 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1422 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1423 | unsigned long nr_pages = end_index - index + 1; | |
1424 | int i; | |
1425 | ||
1426 | if (index == locked_page->index && end_index == index) | |
1427 | return; | |
1428 | ||
1429 | while (nr_pages > 0) { | |
1430 | ret = find_get_pages_contig(inode->i_mapping, index, | |
1431 | min_t(unsigned long, nr_pages, | |
1432 | ARRAY_SIZE(pages)), pages); | |
1433 | for (i = 0; i < ret; i++) { | |
1434 | if (pages[i] != locked_page) | |
1435 | unlock_page(pages[i]); | |
1436 | page_cache_release(pages[i]); | |
1437 | } | |
1438 | nr_pages -= ret; | |
1439 | index += ret; | |
1440 | cond_resched(); | |
1441 | } | |
1442 | } | |
1443 | ||
1444 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1445 | struct page *locked_page, | |
1446 | u64 delalloc_start, | |
1447 | u64 delalloc_end) | |
1448 | { | |
1449 | unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT; | |
1450 | unsigned long start_index = index; | |
1451 | unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT; | |
1452 | unsigned long pages_locked = 0; | |
1453 | struct page *pages[16]; | |
1454 | unsigned long nrpages; | |
1455 | int ret; | |
1456 | int i; | |
1457 | ||
1458 | /* the caller is responsible for locking the start index */ | |
1459 | if (index == locked_page->index && index == end_index) | |
1460 | return 0; | |
1461 | ||
1462 | /* skip the page at the start index */ | |
1463 | nrpages = end_index - index + 1; | |
1464 | while (nrpages > 0) { | |
1465 | ret = find_get_pages_contig(inode->i_mapping, index, | |
1466 | min_t(unsigned long, | |
1467 | nrpages, ARRAY_SIZE(pages)), pages); | |
1468 | if (ret == 0) { | |
1469 | ret = -EAGAIN; | |
1470 | goto done; | |
1471 | } | |
1472 | /* now we have an array of pages, lock them all */ | |
1473 | for (i = 0; i < ret; i++) { | |
1474 | /* | |
1475 | * the caller is taking responsibility for | |
1476 | * locked_page | |
1477 | */ | |
1478 | if (pages[i] != locked_page) { | |
1479 | lock_page(pages[i]); | |
1480 | if (!PageDirty(pages[i]) || | |
1481 | pages[i]->mapping != inode->i_mapping) { | |
1482 | ret = -EAGAIN; | |
1483 | unlock_page(pages[i]); | |
1484 | page_cache_release(pages[i]); | |
1485 | goto done; | |
1486 | } | |
1487 | } | |
1488 | page_cache_release(pages[i]); | |
1489 | pages_locked++; | |
1490 | } | |
1491 | nrpages -= ret; | |
1492 | index += ret; | |
1493 | cond_resched(); | |
1494 | } | |
1495 | ret = 0; | |
1496 | done: | |
1497 | if (ret && pages_locked) { | |
1498 | __unlock_for_delalloc(inode, locked_page, | |
1499 | delalloc_start, | |
1500 | ((u64)(start_index + pages_locked - 1)) << | |
1501 | PAGE_CACHE_SHIFT); | |
1502 | } | |
1503 | return ret; | |
1504 | } | |
1505 | ||
1506 | /* | |
1507 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1508 | * more than 'max_bytes'. start and end are used to return the range, | |
1509 | * | |
1510 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1511 | */ | |
1512 | static noinline u64 find_lock_delalloc_range(struct inode *inode, | |
1513 | struct extent_io_tree *tree, | |
1514 | struct page *locked_page, | |
1515 | u64 *start, u64 *end, | |
1516 | u64 max_bytes) | |
1517 | { | |
1518 | u64 delalloc_start; | |
1519 | u64 delalloc_end; | |
1520 | u64 found; | |
1521 | struct extent_state *cached_state = NULL; | |
1522 | int ret; | |
1523 | int loops = 0; | |
1524 | ||
1525 | again: | |
1526 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1527 | delalloc_start = *start; | |
1528 | delalloc_end = 0; | |
1529 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
1530 | max_bytes, &cached_state); | |
1531 | if (!found || delalloc_end <= *start) { | |
1532 | *start = delalloc_start; | |
1533 | *end = delalloc_end; | |
1534 | free_extent_state(cached_state); | |
1535 | return found; | |
1536 | } | |
1537 | ||
1538 | /* | |
1539 | * start comes from the offset of locked_page. We have to lock | |
1540 | * pages in order, so we can't process delalloc bytes before | |
1541 | * locked_page | |
1542 | */ | |
1543 | if (delalloc_start < *start) | |
1544 | delalloc_start = *start; | |
1545 | ||
1546 | /* | |
1547 | * make sure to limit the number of pages we try to lock down | |
1548 | * if we're looping. | |
1549 | */ | |
1550 | if (delalloc_end + 1 - delalloc_start > max_bytes && loops) | |
1551 | delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1; | |
1552 | ||
1553 | /* step two, lock all the pages after the page that has start */ | |
1554 | ret = lock_delalloc_pages(inode, locked_page, | |
1555 | delalloc_start, delalloc_end); | |
1556 | if (ret == -EAGAIN) { | |
1557 | /* some of the pages are gone, lets avoid looping by | |
1558 | * shortening the size of the delalloc range we're searching | |
1559 | */ | |
1560 | free_extent_state(cached_state); | |
1561 | if (!loops) { | |
1562 | unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1); | |
1563 | max_bytes = PAGE_CACHE_SIZE - offset; | |
1564 | loops = 1; | |
1565 | goto again; | |
1566 | } else { | |
1567 | found = 0; | |
1568 | goto out_failed; | |
1569 | } | |
1570 | } | |
1571 | BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */ | |
1572 | ||
1573 | /* step three, lock the state bits for the whole range */ | |
1574 | lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state); | |
1575 | ||
1576 | /* then test to make sure it is all still delalloc */ | |
1577 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
1578 | EXTENT_DELALLOC, 1, cached_state); | |
1579 | if (!ret) { | |
1580 | unlock_extent_cached(tree, delalloc_start, delalloc_end, | |
1581 | &cached_state, GFP_NOFS); | |
1582 | __unlock_for_delalloc(inode, locked_page, | |
1583 | delalloc_start, delalloc_end); | |
1584 | cond_resched(); | |
1585 | goto again; | |
1586 | } | |
1587 | free_extent_state(cached_state); | |
1588 | *start = delalloc_start; | |
1589 | *end = delalloc_end; | |
1590 | out_failed: | |
1591 | return found; | |
1592 | } | |
1593 | ||
1594 | int extent_clear_unlock_delalloc(struct inode *inode, | |
1595 | struct extent_io_tree *tree, | |
1596 | u64 start, u64 end, struct page *locked_page, | |
1597 | unsigned long op) | |
1598 | { | |
1599 | int ret; | |
1600 | struct page *pages[16]; | |
1601 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1602 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1603 | unsigned long nr_pages = end_index - index + 1; | |
1604 | int i; | |
1605 | int clear_bits = 0; | |
1606 | ||
1607 | if (op & EXTENT_CLEAR_UNLOCK) | |
1608 | clear_bits |= EXTENT_LOCKED; | |
1609 | if (op & EXTENT_CLEAR_DIRTY) | |
1610 | clear_bits |= EXTENT_DIRTY; | |
1611 | ||
1612 | if (op & EXTENT_CLEAR_DELALLOC) | |
1613 | clear_bits |= EXTENT_DELALLOC; | |
1614 | ||
1615 | clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); | |
1616 | if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | | |
1617 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK | | |
1618 | EXTENT_SET_PRIVATE2))) | |
1619 | return 0; | |
1620 | ||
1621 | while (nr_pages > 0) { | |
1622 | ret = find_get_pages_contig(inode->i_mapping, index, | |
1623 | min_t(unsigned long, | |
1624 | nr_pages, ARRAY_SIZE(pages)), pages); | |
1625 | for (i = 0; i < ret; i++) { | |
1626 | ||
1627 | if (op & EXTENT_SET_PRIVATE2) | |
1628 | SetPagePrivate2(pages[i]); | |
1629 | ||
1630 | if (pages[i] == locked_page) { | |
1631 | page_cache_release(pages[i]); | |
1632 | continue; | |
1633 | } | |
1634 | if (op & EXTENT_CLEAR_DIRTY) | |
1635 | clear_page_dirty_for_io(pages[i]); | |
1636 | if (op & EXTENT_SET_WRITEBACK) | |
1637 | set_page_writeback(pages[i]); | |
1638 | if (op & EXTENT_END_WRITEBACK) | |
1639 | end_page_writeback(pages[i]); | |
1640 | if (op & EXTENT_CLEAR_UNLOCK_PAGE) | |
1641 | unlock_page(pages[i]); | |
1642 | page_cache_release(pages[i]); | |
1643 | } | |
1644 | nr_pages -= ret; | |
1645 | index += ret; | |
1646 | cond_resched(); | |
1647 | } | |
1648 | return 0; | |
1649 | } | |
1650 | ||
1651 | /* | |
1652 | * count the number of bytes in the tree that have a given bit(s) | |
1653 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1654 | * cached. The total number found is returned. | |
1655 | */ | |
1656 | u64 count_range_bits(struct extent_io_tree *tree, | |
1657 | u64 *start, u64 search_end, u64 max_bytes, | |
1658 | unsigned long bits, int contig) | |
1659 | { | |
1660 | struct rb_node *node; | |
1661 | struct extent_state *state; | |
1662 | u64 cur_start = *start; | |
1663 | u64 total_bytes = 0; | |
1664 | u64 last = 0; | |
1665 | int found = 0; | |
1666 | ||
1667 | if (search_end <= cur_start) { | |
1668 | WARN_ON(1); | |
1669 | return 0; | |
1670 | } | |
1671 | ||
1672 | spin_lock(&tree->lock); | |
1673 | if (cur_start == 0 && bits == EXTENT_DIRTY) { | |
1674 | total_bytes = tree->dirty_bytes; | |
1675 | goto out; | |
1676 | } | |
1677 | /* | |
1678 | * this search will find all the extents that end after | |
1679 | * our range starts. | |
1680 | */ | |
1681 | node = tree_search(tree, cur_start); | |
1682 | if (!node) | |
1683 | goto out; | |
1684 | ||
1685 | while (1) { | |
1686 | state = rb_entry(node, struct extent_state, rb_node); | |
1687 | if (state->start > search_end) | |
1688 | break; | |
1689 | if (contig && found && state->start > last + 1) | |
1690 | break; | |
1691 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
1692 | total_bytes += min(search_end, state->end) + 1 - | |
1693 | max(cur_start, state->start); | |
1694 | if (total_bytes >= max_bytes) | |
1695 | break; | |
1696 | if (!found) { | |
1697 | *start = max(cur_start, state->start); | |
1698 | found = 1; | |
1699 | } | |
1700 | last = state->end; | |
1701 | } else if (contig && found) { | |
1702 | break; | |
1703 | } | |
1704 | node = rb_next(node); | |
1705 | if (!node) | |
1706 | break; | |
1707 | } | |
1708 | out: | |
1709 | spin_unlock(&tree->lock); | |
1710 | return total_bytes; | |
1711 | } | |
1712 | ||
1713 | /* | |
1714 | * set the private field for a given byte offset in the tree. If there isn't | |
1715 | * an extent_state there already, this does nothing. | |
1716 | */ | |
1717 | int set_state_private(struct extent_io_tree *tree, u64 start, u64 private) | |
1718 | { | |
1719 | struct rb_node *node; | |
1720 | struct extent_state *state; | |
1721 | int ret = 0; | |
1722 | ||
1723 | spin_lock(&tree->lock); | |
1724 | /* | |
1725 | * this search will find all the extents that end after | |
1726 | * our range starts. | |
1727 | */ | |
1728 | node = tree_search(tree, start); | |
1729 | if (!node) { | |
1730 | ret = -ENOENT; | |
1731 | goto out; | |
1732 | } | |
1733 | state = rb_entry(node, struct extent_state, rb_node); | |
1734 | if (state->start != start) { | |
1735 | ret = -ENOENT; | |
1736 | goto out; | |
1737 | } | |
1738 | state->private = private; | |
1739 | out: | |
1740 | spin_unlock(&tree->lock); | |
1741 | return ret; | |
1742 | } | |
1743 | ||
1744 | int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private) | |
1745 | { | |
1746 | struct rb_node *node; | |
1747 | struct extent_state *state; | |
1748 | int ret = 0; | |
1749 | ||
1750 | spin_lock(&tree->lock); | |
1751 | /* | |
1752 | * this search will find all the extents that end after | |
1753 | * our range starts. | |
1754 | */ | |
1755 | node = tree_search(tree, start); | |
1756 | if (!node) { | |
1757 | ret = -ENOENT; | |
1758 | goto out; | |
1759 | } | |
1760 | state = rb_entry(node, struct extent_state, rb_node); | |
1761 | if (state->start != start) { | |
1762 | ret = -ENOENT; | |
1763 | goto out; | |
1764 | } | |
1765 | *private = state->private; | |
1766 | out: | |
1767 | spin_unlock(&tree->lock); | |
1768 | return ret; | |
1769 | } | |
1770 | ||
1771 | /* | |
1772 | * searches a range in the state tree for a given mask. | |
1773 | * If 'filled' == 1, this returns 1 only if every extent in the tree | |
1774 | * has the bits set. Otherwise, 1 is returned if any bit in the | |
1775 | * range is found set. | |
1776 | */ | |
1777 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
1778 | int bits, int filled, struct extent_state *cached) | |
1779 | { | |
1780 | struct extent_state *state = NULL; | |
1781 | struct rb_node *node; | |
1782 | int bitset = 0; | |
1783 | ||
1784 | spin_lock(&tree->lock); | |
1785 | if (cached && cached->tree && cached->start <= start && | |
1786 | cached->end > start) | |
1787 | node = &cached->rb_node; | |
1788 | else | |
1789 | node = tree_search(tree, start); | |
1790 | while (node && start <= end) { | |
1791 | state = rb_entry(node, struct extent_state, rb_node); | |
1792 | ||
1793 | if (filled && state->start > start) { | |
1794 | bitset = 0; | |
1795 | break; | |
1796 | } | |
1797 | ||
1798 | if (state->start > end) | |
1799 | break; | |
1800 | ||
1801 | if (state->state & bits) { | |
1802 | bitset = 1; | |
1803 | if (!filled) | |
1804 | break; | |
1805 | } else if (filled) { | |
1806 | bitset = 0; | |
1807 | break; | |
1808 | } | |
1809 | ||
1810 | if (state->end == (u64)-1) | |
1811 | break; | |
1812 | ||
1813 | start = state->end + 1; | |
1814 | if (start > end) | |
1815 | break; | |
1816 | node = rb_next(node); | |
1817 | if (!node) { | |
1818 | if (filled) | |
1819 | bitset = 0; | |
1820 | break; | |
1821 | } | |
1822 | } | |
1823 | spin_unlock(&tree->lock); | |
1824 | return bitset; | |
1825 | } | |
1826 | ||
1827 | /* | |
1828 | * helper function to set a given page up to date if all the | |
1829 | * extents in the tree for that page are up to date | |
1830 | */ | |
1831 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) | |
1832 | { | |
1833 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1834 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
1835 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) | |
1836 | SetPageUptodate(page); | |
1837 | } | |
1838 | ||
1839 | /* | |
1840 | * helper function to unlock a page if all the extents in the tree | |
1841 | * for that page are unlocked | |
1842 | */ | |
1843 | static void check_page_locked(struct extent_io_tree *tree, struct page *page) | |
1844 | { | |
1845 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1846 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
1847 | if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) | |
1848 | unlock_page(page); | |
1849 | } | |
1850 | ||
1851 | /* | |
1852 | * helper function to end page writeback if all the extents | |
1853 | * in the tree for that page are done with writeback | |
1854 | */ | |
1855 | static void check_page_writeback(struct extent_io_tree *tree, | |
1856 | struct page *page) | |
1857 | { | |
1858 | end_page_writeback(page); | |
1859 | } | |
1860 | ||
1861 | /* | |
1862 | * When IO fails, either with EIO or csum verification fails, we | |
1863 | * try other mirrors that might have a good copy of the data. This | |
1864 | * io_failure_record is used to record state as we go through all the | |
1865 | * mirrors. If another mirror has good data, the page is set up to date | |
1866 | * and things continue. If a good mirror can't be found, the original | |
1867 | * bio end_io callback is called to indicate things have failed. | |
1868 | */ | |
1869 | struct io_failure_record { | |
1870 | struct page *page; | |
1871 | u64 start; | |
1872 | u64 len; | |
1873 | u64 logical; | |
1874 | unsigned long bio_flags; | |
1875 | int this_mirror; | |
1876 | int failed_mirror; | |
1877 | int in_validation; | |
1878 | }; | |
1879 | ||
1880 | static int free_io_failure(struct inode *inode, struct io_failure_record *rec, | |
1881 | int did_repair) | |
1882 | { | |
1883 | int ret; | |
1884 | int err = 0; | |
1885 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
1886 | ||
1887 | set_state_private(failure_tree, rec->start, 0); | |
1888 | ret = clear_extent_bits(failure_tree, rec->start, | |
1889 | rec->start + rec->len - 1, | |
1890 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1891 | if (ret) | |
1892 | err = ret; | |
1893 | ||
1894 | if (did_repair) { | |
1895 | ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start, | |
1896 | rec->start + rec->len - 1, | |
1897 | EXTENT_DAMAGED, GFP_NOFS); | |
1898 | if (ret && !err) | |
1899 | err = ret; | |
1900 | } | |
1901 | ||
1902 | kfree(rec); | |
1903 | return err; | |
1904 | } | |
1905 | ||
1906 | static void repair_io_failure_callback(struct bio *bio, int err) | |
1907 | { | |
1908 | complete(bio->bi_private); | |
1909 | } | |
1910 | ||
1911 | /* | |
1912 | * this bypasses the standard btrfs submit functions deliberately, as | |
1913 | * the standard behavior is to write all copies in a raid setup. here we only | |
1914 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
1915 | * submit_bio directly. | |
1916 | * to avoid any synchonization issues, wait for the data after writing, which | |
1917 | * actually prevents the read that triggered the error from finishing. | |
1918 | * currently, there can be no more than two copies of every data bit. thus, | |
1919 | * exactly one rewrite is required. | |
1920 | */ | |
1921 | int repair_io_failure(struct btrfs_mapping_tree *map_tree, u64 start, | |
1922 | u64 length, u64 logical, struct page *page, | |
1923 | int mirror_num) | |
1924 | { | |
1925 | struct bio *bio; | |
1926 | struct btrfs_device *dev; | |
1927 | DECLARE_COMPLETION_ONSTACK(compl); | |
1928 | u64 map_length = 0; | |
1929 | u64 sector; | |
1930 | struct btrfs_bio *bbio = NULL; | |
1931 | int ret; | |
1932 | ||
1933 | BUG_ON(!mirror_num); | |
1934 | ||
1935 | bio = bio_alloc(GFP_NOFS, 1); | |
1936 | if (!bio) | |
1937 | return -EIO; | |
1938 | bio->bi_private = &compl; | |
1939 | bio->bi_end_io = repair_io_failure_callback; | |
1940 | bio->bi_size = 0; | |
1941 | map_length = length; | |
1942 | ||
1943 | ret = btrfs_map_block(map_tree, WRITE, logical, | |
1944 | &map_length, &bbio, mirror_num); | |
1945 | if (ret) { | |
1946 | bio_put(bio); | |
1947 | return -EIO; | |
1948 | } | |
1949 | BUG_ON(mirror_num != bbio->mirror_num); | |
1950 | sector = bbio->stripes[mirror_num-1].physical >> 9; | |
1951 | bio->bi_sector = sector; | |
1952 | dev = bbio->stripes[mirror_num-1].dev; | |
1953 | kfree(bbio); | |
1954 | if (!dev || !dev->bdev || !dev->writeable) { | |
1955 | bio_put(bio); | |
1956 | return -EIO; | |
1957 | } | |
1958 | bio->bi_bdev = dev->bdev; | |
1959 | bio_add_page(bio, page, length, start-page_offset(page)); | |
1960 | btrfsic_submit_bio(WRITE_SYNC, bio); | |
1961 | wait_for_completion(&compl); | |
1962 | ||
1963 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) { | |
1964 | /* try to remap that extent elsewhere? */ | |
1965 | bio_put(bio); | |
1966 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); | |
1967 | return -EIO; | |
1968 | } | |
1969 | ||
1970 | printk_ratelimited_in_rcu(KERN_INFO "btrfs read error corrected: ino %lu off %llu " | |
1971 | "(dev %s sector %llu)\n", page->mapping->host->i_ino, | |
1972 | start, rcu_str_deref(dev->name), sector); | |
1973 | ||
1974 | bio_put(bio); | |
1975 | return 0; | |
1976 | } | |
1977 | ||
1978 | int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb, | |
1979 | int mirror_num) | |
1980 | { | |
1981 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
1982 | u64 start = eb->start; | |
1983 | unsigned long i, num_pages = num_extent_pages(eb->start, eb->len); | |
1984 | int ret = 0; | |
1985 | ||
1986 | for (i = 0; i < num_pages; i++) { | |
1987 | struct page *p = extent_buffer_page(eb, i); | |
1988 | ret = repair_io_failure(map_tree, start, PAGE_CACHE_SIZE, | |
1989 | start, p, mirror_num); | |
1990 | if (ret) | |
1991 | break; | |
1992 | start += PAGE_CACHE_SIZE; | |
1993 | } | |
1994 | ||
1995 | return ret; | |
1996 | } | |
1997 | ||
1998 | /* | |
1999 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2000 | * to see if we need to process or clean up an io_failure_record | |
2001 | */ | |
2002 | static int clean_io_failure(u64 start, struct page *page) | |
2003 | { | |
2004 | u64 private; | |
2005 | u64 private_failure; | |
2006 | struct io_failure_record *failrec; | |
2007 | struct btrfs_mapping_tree *map_tree; | |
2008 | struct extent_state *state; | |
2009 | int num_copies; | |
2010 | int did_repair = 0; | |
2011 | int ret; | |
2012 | struct inode *inode = page->mapping->host; | |
2013 | ||
2014 | private = 0; | |
2015 | ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
2016 | (u64)-1, 1, EXTENT_DIRTY, 0); | |
2017 | if (!ret) | |
2018 | return 0; | |
2019 | ||
2020 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start, | |
2021 | &private_failure); | |
2022 | if (ret) | |
2023 | return 0; | |
2024 | ||
2025 | failrec = (struct io_failure_record *)(unsigned long) private_failure; | |
2026 | BUG_ON(!failrec->this_mirror); | |
2027 | ||
2028 | if (failrec->in_validation) { | |
2029 | /* there was no real error, just free the record */ | |
2030 | pr_debug("clean_io_failure: freeing dummy error at %llu\n", | |
2031 | failrec->start); | |
2032 | did_repair = 1; | |
2033 | goto out; | |
2034 | } | |
2035 | ||
2036 | spin_lock(&BTRFS_I(inode)->io_tree.lock); | |
2037 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, | |
2038 | failrec->start, | |
2039 | EXTENT_LOCKED); | |
2040 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); | |
2041 | ||
2042 | if (state && state->start == failrec->start) { | |
2043 | map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree; | |
2044 | num_copies = btrfs_num_copies(map_tree, failrec->logical, | |
2045 | failrec->len); | |
2046 | if (num_copies > 1) { | |
2047 | ret = repair_io_failure(map_tree, start, failrec->len, | |
2048 | failrec->logical, page, | |
2049 | failrec->failed_mirror); | |
2050 | did_repair = !ret; | |
2051 | } | |
2052 | } | |
2053 | ||
2054 | out: | |
2055 | if (!ret) | |
2056 | ret = free_io_failure(inode, failrec, did_repair); | |
2057 | ||
2058 | return ret; | |
2059 | } | |
2060 | ||
2061 | /* | |
2062 | * this is a generic handler for readpage errors (default | |
2063 | * readpage_io_failed_hook). if other copies exist, read those and write back | |
2064 | * good data to the failed position. does not investigate in remapping the | |
2065 | * failed extent elsewhere, hoping the device will be smart enough to do this as | |
2066 | * needed | |
2067 | */ | |
2068 | ||
2069 | static int bio_readpage_error(struct bio *failed_bio, struct page *page, | |
2070 | u64 start, u64 end, int failed_mirror, | |
2071 | struct extent_state *state) | |
2072 | { | |
2073 | struct io_failure_record *failrec = NULL; | |
2074 | u64 private; | |
2075 | struct extent_map *em; | |
2076 | struct inode *inode = page->mapping->host; | |
2077 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
2078 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2079 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
2080 | struct bio *bio; | |
2081 | int num_copies; | |
2082 | int ret; | |
2083 | int read_mode; | |
2084 | u64 logical; | |
2085 | ||
2086 | BUG_ON(failed_bio->bi_rw & REQ_WRITE); | |
2087 | ||
2088 | ret = get_state_private(failure_tree, start, &private); | |
2089 | if (ret) { | |
2090 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2091 | if (!failrec) | |
2092 | return -ENOMEM; | |
2093 | failrec->start = start; | |
2094 | failrec->len = end - start + 1; | |
2095 | failrec->this_mirror = 0; | |
2096 | failrec->bio_flags = 0; | |
2097 | failrec->in_validation = 0; | |
2098 | ||
2099 | read_lock(&em_tree->lock); | |
2100 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2101 | if (!em) { | |
2102 | read_unlock(&em_tree->lock); | |
2103 | kfree(failrec); | |
2104 | return -EIO; | |
2105 | } | |
2106 | ||
2107 | if (em->start > start || em->start + em->len < start) { | |
2108 | free_extent_map(em); | |
2109 | em = NULL; | |
2110 | } | |
2111 | read_unlock(&em_tree->lock); | |
2112 | ||
2113 | if (!em) { | |
2114 | kfree(failrec); | |
2115 | return -EIO; | |
2116 | } | |
2117 | logical = start - em->start; | |
2118 | logical = em->block_start + logical; | |
2119 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2120 | logical = em->block_start; | |
2121 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2122 | extent_set_compress_type(&failrec->bio_flags, | |
2123 | em->compress_type); | |
2124 | } | |
2125 | pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, " | |
2126 | "len=%llu\n", logical, start, failrec->len); | |
2127 | failrec->logical = logical; | |
2128 | free_extent_map(em); | |
2129 | ||
2130 | /* set the bits in the private failure tree */ | |
2131 | ret = set_extent_bits(failure_tree, start, end, | |
2132 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
2133 | if (ret >= 0) | |
2134 | ret = set_state_private(failure_tree, start, | |
2135 | (u64)(unsigned long)failrec); | |
2136 | /* set the bits in the inode's tree */ | |
2137 | if (ret >= 0) | |
2138 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED, | |
2139 | GFP_NOFS); | |
2140 | if (ret < 0) { | |
2141 | kfree(failrec); | |
2142 | return ret; | |
2143 | } | |
2144 | } else { | |
2145 | failrec = (struct io_failure_record *)(unsigned long)private; | |
2146 | pr_debug("bio_readpage_error: (found) logical=%llu, " | |
2147 | "start=%llu, len=%llu, validation=%d\n", | |
2148 | failrec->logical, failrec->start, failrec->len, | |
2149 | failrec->in_validation); | |
2150 | /* | |
2151 | * when data can be on disk more than twice, add to failrec here | |
2152 | * (e.g. with a list for failed_mirror) to make | |
2153 | * clean_io_failure() clean all those errors at once. | |
2154 | */ | |
2155 | } | |
2156 | num_copies = btrfs_num_copies( | |
2157 | &BTRFS_I(inode)->root->fs_info->mapping_tree, | |
2158 | failrec->logical, failrec->len); | |
2159 | if (num_copies == 1) { | |
2160 | /* | |
2161 | * we only have a single copy of the data, so don't bother with | |
2162 | * all the retry and error correction code that follows. no | |
2163 | * matter what the error is, it is very likely to persist. | |
2164 | */ | |
2165 | pr_debug("bio_readpage_error: cannot repair, num_copies == 1. " | |
2166 | "state=%p, num_copies=%d, next_mirror %d, " | |
2167 | "failed_mirror %d\n", state, num_copies, | |
2168 | failrec->this_mirror, failed_mirror); | |
2169 | free_io_failure(inode, failrec, 0); | |
2170 | return -EIO; | |
2171 | } | |
2172 | ||
2173 | if (!state) { | |
2174 | spin_lock(&tree->lock); | |
2175 | state = find_first_extent_bit_state(tree, failrec->start, | |
2176 | EXTENT_LOCKED); | |
2177 | if (state && state->start != failrec->start) | |
2178 | state = NULL; | |
2179 | spin_unlock(&tree->lock); | |
2180 | } | |
2181 | ||
2182 | /* | |
2183 | * there are two premises: | |
2184 | * a) deliver good data to the caller | |
2185 | * b) correct the bad sectors on disk | |
2186 | */ | |
2187 | if (failed_bio->bi_vcnt > 1) { | |
2188 | /* | |
2189 | * to fulfill b), we need to know the exact failing sectors, as | |
2190 | * we don't want to rewrite any more than the failed ones. thus, | |
2191 | * we need separate read requests for the failed bio | |
2192 | * | |
2193 | * if the following BUG_ON triggers, our validation request got | |
2194 | * merged. we need separate requests for our algorithm to work. | |
2195 | */ | |
2196 | BUG_ON(failrec->in_validation); | |
2197 | failrec->in_validation = 1; | |
2198 | failrec->this_mirror = failed_mirror; | |
2199 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; | |
2200 | } else { | |
2201 | /* | |
2202 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2203 | * of the failed sector and if we succeed, we have setup | |
2204 | * everything for repair_io_failure to do the rest for us. | |
2205 | */ | |
2206 | if (failrec->in_validation) { | |
2207 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2208 | failrec->in_validation = 0; | |
2209 | failrec->this_mirror = 0; | |
2210 | } | |
2211 | failrec->failed_mirror = failed_mirror; | |
2212 | failrec->this_mirror++; | |
2213 | if (failrec->this_mirror == failed_mirror) | |
2214 | failrec->this_mirror++; | |
2215 | read_mode = READ_SYNC; | |
2216 | } | |
2217 | ||
2218 | if (!state || failrec->this_mirror > num_copies) { | |
2219 | pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, " | |
2220 | "next_mirror %d, failed_mirror %d\n", state, | |
2221 | num_copies, failrec->this_mirror, failed_mirror); | |
2222 | free_io_failure(inode, failrec, 0); | |
2223 | return -EIO; | |
2224 | } | |
2225 | ||
2226 | bio = bio_alloc(GFP_NOFS, 1); | |
2227 | if (!bio) { | |
2228 | free_io_failure(inode, failrec, 0); | |
2229 | return -EIO; | |
2230 | } | |
2231 | bio->bi_private = state; | |
2232 | bio->bi_end_io = failed_bio->bi_end_io; | |
2233 | bio->bi_sector = failrec->logical >> 9; | |
2234 | bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; | |
2235 | bio->bi_size = 0; | |
2236 | ||
2237 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); | |
2238 | ||
2239 | pr_debug("bio_readpage_error: submitting new read[%#x] to " | |
2240 | "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode, | |
2241 | failrec->this_mirror, num_copies, failrec->in_validation); | |
2242 | ||
2243 | ret = tree->ops->submit_bio_hook(inode, read_mode, bio, | |
2244 | failrec->this_mirror, | |
2245 | failrec->bio_flags, 0); | |
2246 | return ret; | |
2247 | } | |
2248 | ||
2249 | /* lots and lots of room for performance fixes in the end_bio funcs */ | |
2250 | ||
2251 | int end_extent_writepage(struct page *page, int err, u64 start, u64 end) | |
2252 | { | |
2253 | int uptodate = (err == 0); | |
2254 | struct extent_io_tree *tree; | |
2255 | int ret; | |
2256 | ||
2257 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
2258 | ||
2259 | if (tree->ops && tree->ops->writepage_end_io_hook) { | |
2260 | ret = tree->ops->writepage_end_io_hook(page, start, | |
2261 | end, NULL, uptodate); | |
2262 | if (ret) | |
2263 | uptodate = 0; | |
2264 | } | |
2265 | ||
2266 | if (!uptodate) { | |
2267 | ClearPageUptodate(page); | |
2268 | SetPageError(page); | |
2269 | } | |
2270 | return 0; | |
2271 | } | |
2272 | ||
2273 | /* | |
2274 | * after a writepage IO is done, we need to: | |
2275 | * clear the uptodate bits on error | |
2276 | * clear the writeback bits in the extent tree for this IO | |
2277 | * end_page_writeback if the page has no more pending IO | |
2278 | * | |
2279 | * Scheduling is not allowed, so the extent state tree is expected | |
2280 | * to have one and only one object corresponding to this IO. | |
2281 | */ | |
2282 | static void end_bio_extent_writepage(struct bio *bio, int err) | |
2283 | { | |
2284 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
2285 | struct extent_io_tree *tree; | |
2286 | u64 start; | |
2287 | u64 end; | |
2288 | int whole_page; | |
2289 | ||
2290 | do { | |
2291 | struct page *page = bvec->bv_page; | |
2292 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
2293 | ||
2294 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + | |
2295 | bvec->bv_offset; | |
2296 | end = start + bvec->bv_len - 1; | |
2297 | ||
2298 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
2299 | whole_page = 1; | |
2300 | else | |
2301 | whole_page = 0; | |
2302 | ||
2303 | if (--bvec >= bio->bi_io_vec) | |
2304 | prefetchw(&bvec->bv_page->flags); | |
2305 | ||
2306 | if (end_extent_writepage(page, err, start, end)) | |
2307 | continue; | |
2308 | ||
2309 | if (whole_page) | |
2310 | end_page_writeback(page); | |
2311 | else | |
2312 | check_page_writeback(tree, page); | |
2313 | } while (bvec >= bio->bi_io_vec); | |
2314 | ||
2315 | bio_put(bio); | |
2316 | } | |
2317 | ||
2318 | /* | |
2319 | * after a readpage IO is done, we need to: | |
2320 | * clear the uptodate bits on error | |
2321 | * set the uptodate bits if things worked | |
2322 | * set the page up to date if all extents in the tree are uptodate | |
2323 | * clear the lock bit in the extent tree | |
2324 | * unlock the page if there are no other extents locked for it | |
2325 | * | |
2326 | * Scheduling is not allowed, so the extent state tree is expected | |
2327 | * to have one and only one object corresponding to this IO. | |
2328 | */ | |
2329 | static void end_bio_extent_readpage(struct bio *bio, int err) | |
2330 | { | |
2331 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
2332 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; | |
2333 | struct bio_vec *bvec = bio->bi_io_vec; | |
2334 | struct extent_io_tree *tree; | |
2335 | u64 start; | |
2336 | u64 end; | |
2337 | int whole_page; | |
2338 | int mirror; | |
2339 | int ret; | |
2340 | ||
2341 | if (err) | |
2342 | uptodate = 0; | |
2343 | ||
2344 | do { | |
2345 | struct page *page = bvec->bv_page; | |
2346 | struct extent_state *cached = NULL; | |
2347 | struct extent_state *state; | |
2348 | ||
2349 | pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, " | |
2350 | "mirror=%ld\n", (u64)bio->bi_sector, err, | |
2351 | (long int)bio->bi_bdev); | |
2352 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
2353 | ||
2354 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + | |
2355 | bvec->bv_offset; | |
2356 | end = start + bvec->bv_len - 1; | |
2357 | ||
2358 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
2359 | whole_page = 1; | |
2360 | else | |
2361 | whole_page = 0; | |
2362 | ||
2363 | if (++bvec <= bvec_end) | |
2364 | prefetchw(&bvec->bv_page->flags); | |
2365 | ||
2366 | spin_lock(&tree->lock); | |
2367 | state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED); | |
2368 | if (state && state->start == start) { | |
2369 | /* | |
2370 | * take a reference on the state, unlock will drop | |
2371 | * the ref | |
2372 | */ | |
2373 | cache_state(state, &cached); | |
2374 | } | |
2375 | spin_unlock(&tree->lock); | |
2376 | ||
2377 | mirror = (int)(unsigned long)bio->bi_bdev; | |
2378 | if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) { | |
2379 | ret = tree->ops->readpage_end_io_hook(page, start, end, | |
2380 | state, mirror); | |
2381 | if (ret) | |
2382 | uptodate = 0; | |
2383 | else | |
2384 | clean_io_failure(start, page); | |
2385 | } | |
2386 | ||
2387 | if (!uptodate && tree->ops && tree->ops->readpage_io_failed_hook) { | |
2388 | ret = tree->ops->readpage_io_failed_hook(page, mirror); | |
2389 | if (!ret && !err && | |
2390 | test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
2391 | uptodate = 1; | |
2392 | } else if (!uptodate) { | |
2393 | /* | |
2394 | * The generic bio_readpage_error handles errors the | |
2395 | * following way: If possible, new read requests are | |
2396 | * created and submitted and will end up in | |
2397 | * end_bio_extent_readpage as well (if we're lucky, not | |
2398 | * in the !uptodate case). In that case it returns 0 and | |
2399 | * we just go on with the next page in our bio. If it | |
2400 | * can't handle the error it will return -EIO and we | |
2401 | * remain responsible for that page. | |
2402 | */ | |
2403 | ret = bio_readpage_error(bio, page, start, end, mirror, NULL); | |
2404 | if (ret == 0) { | |
2405 | uptodate = | |
2406 | test_bit(BIO_UPTODATE, &bio->bi_flags); | |
2407 | if (err) | |
2408 | uptodate = 0; | |
2409 | uncache_state(&cached); | |
2410 | continue; | |
2411 | } | |
2412 | } | |
2413 | ||
2414 | if (uptodate && tree->track_uptodate) { | |
2415 | set_extent_uptodate(tree, start, end, &cached, | |
2416 | GFP_ATOMIC); | |
2417 | } | |
2418 | unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC); | |
2419 | ||
2420 | if (whole_page) { | |
2421 | if (uptodate) { | |
2422 | SetPageUptodate(page); | |
2423 | } else { | |
2424 | ClearPageUptodate(page); | |
2425 | SetPageError(page); | |
2426 | } | |
2427 | unlock_page(page); | |
2428 | } else { | |
2429 | if (uptodate) { | |
2430 | check_page_uptodate(tree, page); | |
2431 | } else { | |
2432 | ClearPageUptodate(page); | |
2433 | SetPageError(page); | |
2434 | } | |
2435 | check_page_locked(tree, page); | |
2436 | } | |
2437 | } while (bvec <= bvec_end); | |
2438 | ||
2439 | bio_put(bio); | |
2440 | } | |
2441 | ||
2442 | struct bio * | |
2443 | btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
2444 | gfp_t gfp_flags) | |
2445 | { | |
2446 | struct bio *bio; | |
2447 | ||
2448 | bio = bio_alloc(gfp_flags, nr_vecs); | |
2449 | ||
2450 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
2451 | while (!bio && (nr_vecs /= 2)) | |
2452 | bio = bio_alloc(gfp_flags, nr_vecs); | |
2453 | } | |
2454 | ||
2455 | if (bio) { | |
2456 | bio->bi_size = 0; | |
2457 | bio->bi_bdev = bdev; | |
2458 | bio->bi_sector = first_sector; | |
2459 | } | |
2460 | return bio; | |
2461 | } | |
2462 | ||
2463 | /* | |
2464 | * Since writes are async, they will only return -ENOMEM. | |
2465 | * Reads can return the full range of I/O error conditions. | |
2466 | */ | |
2467 | static int __must_check submit_one_bio(int rw, struct bio *bio, | |
2468 | int mirror_num, unsigned long bio_flags) | |
2469 | { | |
2470 | int ret = 0; | |
2471 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
2472 | struct page *page = bvec->bv_page; | |
2473 | struct extent_io_tree *tree = bio->bi_private; | |
2474 | u64 start; | |
2475 | ||
2476 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset; | |
2477 | ||
2478 | bio->bi_private = NULL; | |
2479 | ||
2480 | bio_get(bio); | |
2481 | ||
2482 | if (tree->ops && tree->ops->submit_bio_hook) | |
2483 | ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio, | |
2484 | mirror_num, bio_flags, start); | |
2485 | else | |
2486 | btrfsic_submit_bio(rw, bio); | |
2487 | ||
2488 | if (bio_flagged(bio, BIO_EOPNOTSUPP)) | |
2489 | ret = -EOPNOTSUPP; | |
2490 | bio_put(bio); | |
2491 | return ret; | |
2492 | } | |
2493 | ||
2494 | static int merge_bio(struct extent_io_tree *tree, struct page *page, | |
2495 | unsigned long offset, size_t size, struct bio *bio, | |
2496 | unsigned long bio_flags) | |
2497 | { | |
2498 | int ret = 0; | |
2499 | if (tree->ops && tree->ops->merge_bio_hook) | |
2500 | ret = tree->ops->merge_bio_hook(page, offset, size, bio, | |
2501 | bio_flags); | |
2502 | BUG_ON(ret < 0); | |
2503 | return ret; | |
2504 | ||
2505 | } | |
2506 | ||
2507 | static int submit_extent_page(int rw, struct extent_io_tree *tree, | |
2508 | struct page *page, sector_t sector, | |
2509 | size_t size, unsigned long offset, | |
2510 | struct block_device *bdev, | |
2511 | struct bio **bio_ret, | |
2512 | unsigned long max_pages, | |
2513 | bio_end_io_t end_io_func, | |
2514 | int mirror_num, | |
2515 | unsigned long prev_bio_flags, | |
2516 | unsigned long bio_flags) | |
2517 | { | |
2518 | int ret = 0; | |
2519 | struct bio *bio; | |
2520 | int nr; | |
2521 | int contig = 0; | |
2522 | int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED; | |
2523 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; | |
2524 | size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE); | |
2525 | ||
2526 | if (bio_ret && *bio_ret) { | |
2527 | bio = *bio_ret; | |
2528 | if (old_compressed) | |
2529 | contig = bio->bi_sector == sector; | |
2530 | else | |
2531 | contig = bio->bi_sector + (bio->bi_size >> 9) == | |
2532 | sector; | |
2533 | ||
2534 | if (prev_bio_flags != bio_flags || !contig || | |
2535 | merge_bio(tree, page, offset, page_size, bio, bio_flags) || | |
2536 | bio_add_page(bio, page, page_size, offset) < page_size) { | |
2537 | ret = submit_one_bio(rw, bio, mirror_num, | |
2538 | prev_bio_flags); | |
2539 | if (ret < 0) | |
2540 | return ret; | |
2541 | bio = NULL; | |
2542 | } else { | |
2543 | return 0; | |
2544 | } | |
2545 | } | |
2546 | if (this_compressed) | |
2547 | nr = BIO_MAX_PAGES; | |
2548 | else | |
2549 | nr = bio_get_nr_vecs(bdev); | |
2550 | ||
2551 | bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH); | |
2552 | if (!bio) | |
2553 | return -ENOMEM; | |
2554 | ||
2555 | bio_add_page(bio, page, page_size, offset); | |
2556 | bio->bi_end_io = end_io_func; | |
2557 | bio->bi_private = tree; | |
2558 | ||
2559 | if (bio_ret) | |
2560 | *bio_ret = bio; | |
2561 | else | |
2562 | ret = submit_one_bio(rw, bio, mirror_num, bio_flags); | |
2563 | ||
2564 | return ret; | |
2565 | } | |
2566 | ||
2567 | void attach_extent_buffer_page(struct extent_buffer *eb, struct page *page) | |
2568 | { | |
2569 | if (!PagePrivate(page)) { | |
2570 | SetPagePrivate(page); | |
2571 | page_cache_get(page); | |
2572 | set_page_private(page, (unsigned long)eb); | |
2573 | } else { | |
2574 | WARN_ON(page->private != (unsigned long)eb); | |
2575 | } | |
2576 | } | |
2577 | ||
2578 | void set_page_extent_mapped(struct page *page) | |
2579 | { | |
2580 | if (!PagePrivate(page)) { | |
2581 | SetPagePrivate(page); | |
2582 | page_cache_get(page); | |
2583 | set_page_private(page, EXTENT_PAGE_PRIVATE); | |
2584 | } | |
2585 | } | |
2586 | ||
2587 | /* | |
2588 | * basic readpage implementation. Locked extent state structs are inserted | |
2589 | * into the tree that are removed when the IO is done (by the end_io | |
2590 | * handlers) | |
2591 | * XXX JDM: This needs looking at to ensure proper page locking | |
2592 | */ | |
2593 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
2594 | struct page *page, | |
2595 | get_extent_t *get_extent, | |
2596 | struct bio **bio, int mirror_num, | |
2597 | unsigned long *bio_flags) | |
2598 | { | |
2599 | struct inode *inode = page->mapping->host; | |
2600 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2601 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2602 | u64 end; | |
2603 | u64 cur = start; | |
2604 | u64 extent_offset; | |
2605 | u64 last_byte = i_size_read(inode); | |
2606 | u64 block_start; | |
2607 | u64 cur_end; | |
2608 | sector_t sector; | |
2609 | struct extent_map *em; | |
2610 | struct block_device *bdev; | |
2611 | struct btrfs_ordered_extent *ordered; | |
2612 | int ret; | |
2613 | int nr = 0; | |
2614 | size_t pg_offset = 0; | |
2615 | size_t iosize; | |
2616 | size_t disk_io_size; | |
2617 | size_t blocksize = inode->i_sb->s_blocksize; | |
2618 | unsigned long this_bio_flag = 0; | |
2619 | ||
2620 | set_page_extent_mapped(page); | |
2621 | ||
2622 | if (!PageUptodate(page)) { | |
2623 | if (cleancache_get_page(page) == 0) { | |
2624 | BUG_ON(blocksize != PAGE_SIZE); | |
2625 | goto out; | |
2626 | } | |
2627 | } | |
2628 | ||
2629 | end = page_end; | |
2630 | while (1) { | |
2631 | lock_extent(tree, start, end); | |
2632 | ordered = btrfs_lookup_ordered_extent(inode, start); | |
2633 | if (!ordered) | |
2634 | break; | |
2635 | unlock_extent(tree, start, end); | |
2636 | btrfs_start_ordered_extent(inode, ordered, 1); | |
2637 | btrfs_put_ordered_extent(ordered); | |
2638 | } | |
2639 | ||
2640 | if (page->index == last_byte >> PAGE_CACHE_SHIFT) { | |
2641 | char *userpage; | |
2642 | size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1); | |
2643 | ||
2644 | if (zero_offset) { | |
2645 | iosize = PAGE_CACHE_SIZE - zero_offset; | |
2646 | userpage = kmap_atomic(page); | |
2647 | memset(userpage + zero_offset, 0, iosize); | |
2648 | flush_dcache_page(page); | |
2649 | kunmap_atomic(userpage); | |
2650 | } | |
2651 | } | |
2652 | while (cur <= end) { | |
2653 | if (cur >= last_byte) { | |
2654 | char *userpage; | |
2655 | struct extent_state *cached = NULL; | |
2656 | ||
2657 | iosize = PAGE_CACHE_SIZE - pg_offset; | |
2658 | userpage = kmap_atomic(page); | |
2659 | memset(userpage + pg_offset, 0, iosize); | |
2660 | flush_dcache_page(page); | |
2661 | kunmap_atomic(userpage); | |
2662 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
2663 | &cached, GFP_NOFS); | |
2664 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
2665 | &cached, GFP_NOFS); | |
2666 | break; | |
2667 | } | |
2668 | em = get_extent(inode, page, pg_offset, cur, | |
2669 | end - cur + 1, 0); | |
2670 | if (IS_ERR_OR_NULL(em)) { | |
2671 | SetPageError(page); | |
2672 | unlock_extent(tree, cur, end); | |
2673 | break; | |
2674 | } | |
2675 | extent_offset = cur - em->start; | |
2676 | BUG_ON(extent_map_end(em) <= cur); | |
2677 | BUG_ON(end < cur); | |
2678 | ||
2679 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2680 | this_bio_flag = EXTENT_BIO_COMPRESSED; | |
2681 | extent_set_compress_type(&this_bio_flag, | |
2682 | em->compress_type); | |
2683 | } | |
2684 | ||
2685 | iosize = min(extent_map_end(em) - cur, end - cur + 1); | |
2686 | cur_end = min(extent_map_end(em) - 1, end); | |
2687 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
2688 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { | |
2689 | disk_io_size = em->block_len; | |
2690 | sector = em->block_start >> 9; | |
2691 | } else { | |
2692 | sector = (em->block_start + extent_offset) >> 9; | |
2693 | disk_io_size = iosize; | |
2694 | } | |
2695 | bdev = em->bdev; | |
2696 | block_start = em->block_start; | |
2697 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
2698 | block_start = EXTENT_MAP_HOLE; | |
2699 | free_extent_map(em); | |
2700 | em = NULL; | |
2701 | ||
2702 | /* we've found a hole, just zero and go on */ | |
2703 | if (block_start == EXTENT_MAP_HOLE) { | |
2704 | char *userpage; | |
2705 | struct extent_state *cached = NULL; | |
2706 | ||
2707 | userpage = kmap_atomic(page); | |
2708 | memset(userpage + pg_offset, 0, iosize); | |
2709 | flush_dcache_page(page); | |
2710 | kunmap_atomic(userpage); | |
2711 | ||
2712 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
2713 | &cached, GFP_NOFS); | |
2714 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
2715 | &cached, GFP_NOFS); | |
2716 | cur = cur + iosize; | |
2717 | pg_offset += iosize; | |
2718 | continue; | |
2719 | } | |
2720 | /* the get_extent function already copied into the page */ | |
2721 | if (test_range_bit(tree, cur, cur_end, | |
2722 | EXTENT_UPTODATE, 1, NULL)) { | |
2723 | check_page_uptodate(tree, page); | |
2724 | unlock_extent(tree, cur, cur + iosize - 1); | |
2725 | cur = cur + iosize; | |
2726 | pg_offset += iosize; | |
2727 | continue; | |
2728 | } | |
2729 | /* we have an inline extent but it didn't get marked up | |
2730 | * to date. Error out | |
2731 | */ | |
2732 | if (block_start == EXTENT_MAP_INLINE) { | |
2733 | SetPageError(page); | |
2734 | unlock_extent(tree, cur, cur + iosize - 1); | |
2735 | cur = cur + iosize; | |
2736 | pg_offset += iosize; | |
2737 | continue; | |
2738 | } | |
2739 | ||
2740 | ret = 0; | |
2741 | if (tree->ops && tree->ops->readpage_io_hook) { | |
2742 | ret = tree->ops->readpage_io_hook(page, cur, | |
2743 | cur + iosize - 1); | |
2744 | } | |
2745 | if (!ret) { | |
2746 | unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1; | |
2747 | pnr -= page->index; | |
2748 | ret = submit_extent_page(READ, tree, page, | |
2749 | sector, disk_io_size, pg_offset, | |
2750 | bdev, bio, pnr, | |
2751 | end_bio_extent_readpage, mirror_num, | |
2752 | *bio_flags, | |
2753 | this_bio_flag); | |
2754 | if (!ret) { | |
2755 | nr++; | |
2756 | *bio_flags = this_bio_flag; | |
2757 | } | |
2758 | } | |
2759 | if (ret) { | |
2760 | SetPageError(page); | |
2761 | unlock_extent(tree, cur, cur + iosize - 1); | |
2762 | } | |
2763 | cur = cur + iosize; | |
2764 | pg_offset += iosize; | |
2765 | } | |
2766 | out: | |
2767 | if (!nr) { | |
2768 | if (!PageError(page)) | |
2769 | SetPageUptodate(page); | |
2770 | unlock_page(page); | |
2771 | } | |
2772 | return 0; | |
2773 | } | |
2774 | ||
2775 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, | |
2776 | get_extent_t *get_extent, int mirror_num) | |
2777 | { | |
2778 | struct bio *bio = NULL; | |
2779 | unsigned long bio_flags = 0; | |
2780 | int ret; | |
2781 | ||
2782 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, | |
2783 | &bio_flags); | |
2784 | if (bio) | |
2785 | ret = submit_one_bio(READ, bio, mirror_num, bio_flags); | |
2786 | return ret; | |
2787 | } | |
2788 | ||
2789 | static noinline void update_nr_written(struct page *page, | |
2790 | struct writeback_control *wbc, | |
2791 | unsigned long nr_written) | |
2792 | { | |
2793 | wbc->nr_to_write -= nr_written; | |
2794 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && | |
2795 | wbc->range_start == 0 && wbc->range_end == LLONG_MAX)) | |
2796 | page->mapping->writeback_index = page->index + nr_written; | |
2797 | } | |
2798 | ||
2799 | /* | |
2800 | * the writepage semantics are similar to regular writepage. extent | |
2801 | * records are inserted to lock ranges in the tree, and as dirty areas | |
2802 | * are found, they are marked writeback. Then the lock bits are removed | |
2803 | * and the end_io handler clears the writeback ranges | |
2804 | */ | |
2805 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
2806 | void *data) | |
2807 | { | |
2808 | struct inode *inode = page->mapping->host; | |
2809 | struct extent_page_data *epd = data; | |
2810 | struct extent_io_tree *tree = epd->tree; | |
2811 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2812 | u64 delalloc_start; | |
2813 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2814 | u64 end; | |
2815 | u64 cur = start; | |
2816 | u64 extent_offset; | |
2817 | u64 last_byte = i_size_read(inode); | |
2818 | u64 block_start; | |
2819 | u64 iosize; | |
2820 | sector_t sector; | |
2821 | struct extent_state *cached_state = NULL; | |
2822 | struct extent_map *em; | |
2823 | struct block_device *bdev; | |
2824 | int ret; | |
2825 | int nr = 0; | |
2826 | size_t pg_offset = 0; | |
2827 | size_t blocksize; | |
2828 | loff_t i_size = i_size_read(inode); | |
2829 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
2830 | u64 nr_delalloc; | |
2831 | u64 delalloc_end; | |
2832 | int page_started; | |
2833 | int compressed; | |
2834 | int write_flags; | |
2835 | unsigned long nr_written = 0; | |
2836 | bool fill_delalloc = true; | |
2837 | ||
2838 | if (wbc->sync_mode == WB_SYNC_ALL) | |
2839 | write_flags = WRITE_SYNC; | |
2840 | else | |
2841 | write_flags = WRITE; | |
2842 | ||
2843 | trace___extent_writepage(page, inode, wbc); | |
2844 | ||
2845 | WARN_ON(!PageLocked(page)); | |
2846 | ||
2847 | ClearPageError(page); | |
2848 | ||
2849 | pg_offset = i_size & (PAGE_CACHE_SIZE - 1); | |
2850 | if (page->index > end_index || | |
2851 | (page->index == end_index && !pg_offset)) { | |
2852 | page->mapping->a_ops->invalidatepage(page, 0); | |
2853 | unlock_page(page); | |
2854 | return 0; | |
2855 | } | |
2856 | ||
2857 | if (page->index == end_index) { | |
2858 | char *userpage; | |
2859 | ||
2860 | userpage = kmap_atomic(page); | |
2861 | memset(userpage + pg_offset, 0, | |
2862 | PAGE_CACHE_SIZE - pg_offset); | |
2863 | kunmap_atomic(userpage); | |
2864 | flush_dcache_page(page); | |
2865 | } | |
2866 | pg_offset = 0; | |
2867 | ||
2868 | set_page_extent_mapped(page); | |
2869 | ||
2870 | if (!tree->ops || !tree->ops->fill_delalloc) | |
2871 | fill_delalloc = false; | |
2872 | ||
2873 | delalloc_start = start; | |
2874 | delalloc_end = 0; | |
2875 | page_started = 0; | |
2876 | if (!epd->extent_locked && fill_delalloc) { | |
2877 | u64 delalloc_to_write = 0; | |
2878 | /* | |
2879 | * make sure the wbc mapping index is at least updated | |
2880 | * to this page. | |
2881 | */ | |
2882 | update_nr_written(page, wbc, 0); | |
2883 | ||
2884 | while (delalloc_end < page_end) { | |
2885 | nr_delalloc = find_lock_delalloc_range(inode, tree, | |
2886 | page, | |
2887 | &delalloc_start, | |
2888 | &delalloc_end, | |
2889 | 128 * 1024 * 1024); | |
2890 | if (nr_delalloc == 0) { | |
2891 | delalloc_start = delalloc_end + 1; | |
2892 | continue; | |
2893 | } | |
2894 | ret = tree->ops->fill_delalloc(inode, page, | |
2895 | delalloc_start, | |
2896 | delalloc_end, | |
2897 | &page_started, | |
2898 | &nr_written); | |
2899 | /* File system has been set read-only */ | |
2900 | if (ret) { | |
2901 | SetPageError(page); | |
2902 | goto done; | |
2903 | } | |
2904 | /* | |
2905 | * delalloc_end is already one less than the total | |
2906 | * length, so we don't subtract one from | |
2907 | * PAGE_CACHE_SIZE | |
2908 | */ | |
2909 | delalloc_to_write += (delalloc_end - delalloc_start + | |
2910 | PAGE_CACHE_SIZE) >> | |
2911 | PAGE_CACHE_SHIFT; | |
2912 | delalloc_start = delalloc_end + 1; | |
2913 | } | |
2914 | if (wbc->nr_to_write < delalloc_to_write) { | |
2915 | int thresh = 8192; | |
2916 | ||
2917 | if (delalloc_to_write < thresh * 2) | |
2918 | thresh = delalloc_to_write; | |
2919 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
2920 | thresh); | |
2921 | } | |
2922 | ||
2923 | /* did the fill delalloc function already unlock and start | |
2924 | * the IO? | |
2925 | */ | |
2926 | if (page_started) { | |
2927 | ret = 0; | |
2928 | /* | |
2929 | * we've unlocked the page, so we can't update | |
2930 | * the mapping's writeback index, just update | |
2931 | * nr_to_write. | |
2932 | */ | |
2933 | wbc->nr_to_write -= nr_written; | |
2934 | goto done_unlocked; | |
2935 | } | |
2936 | } | |
2937 | if (tree->ops && tree->ops->writepage_start_hook) { | |
2938 | ret = tree->ops->writepage_start_hook(page, start, | |
2939 | page_end); | |
2940 | if (ret) { | |
2941 | /* Fixup worker will requeue */ | |
2942 | if (ret == -EBUSY) | |
2943 | wbc->pages_skipped++; | |
2944 | else | |
2945 | redirty_page_for_writepage(wbc, page); | |
2946 | update_nr_written(page, wbc, nr_written); | |
2947 | unlock_page(page); | |
2948 | ret = 0; | |
2949 | goto done_unlocked; | |
2950 | } | |
2951 | } | |
2952 | ||
2953 | /* | |
2954 | * we don't want to touch the inode after unlocking the page, | |
2955 | * so we update the mapping writeback index now | |
2956 | */ | |
2957 | update_nr_written(page, wbc, nr_written + 1); | |
2958 | ||
2959 | end = page_end; | |
2960 | if (last_byte <= start) { | |
2961 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
2962 | tree->ops->writepage_end_io_hook(page, start, | |
2963 | page_end, NULL, 1); | |
2964 | goto done; | |
2965 | } | |
2966 | ||
2967 | blocksize = inode->i_sb->s_blocksize; | |
2968 | ||
2969 | while (cur <= end) { | |
2970 | if (cur >= last_byte) { | |
2971 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
2972 | tree->ops->writepage_end_io_hook(page, cur, | |
2973 | page_end, NULL, 1); | |
2974 | break; | |
2975 | } | |
2976 | em = epd->get_extent(inode, page, pg_offset, cur, | |
2977 | end - cur + 1, 1); | |
2978 | if (IS_ERR_OR_NULL(em)) { | |
2979 | SetPageError(page); | |
2980 | break; | |
2981 | } | |
2982 | ||
2983 | extent_offset = cur - em->start; | |
2984 | BUG_ON(extent_map_end(em) <= cur); | |
2985 | BUG_ON(end < cur); | |
2986 | iosize = min(extent_map_end(em) - cur, end - cur + 1); | |
2987 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
2988 | sector = (em->block_start + extent_offset) >> 9; | |
2989 | bdev = em->bdev; | |
2990 | block_start = em->block_start; | |
2991 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
2992 | free_extent_map(em); | |
2993 | em = NULL; | |
2994 | ||
2995 | /* | |
2996 | * compressed and inline extents are written through other | |
2997 | * paths in the FS | |
2998 | */ | |
2999 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
3000 | block_start == EXTENT_MAP_INLINE) { | |
3001 | /* | |
3002 | * end_io notification does not happen here for | |
3003 | * compressed extents | |
3004 | */ | |
3005 | if (!compressed && tree->ops && | |
3006 | tree->ops->writepage_end_io_hook) | |
3007 | tree->ops->writepage_end_io_hook(page, cur, | |
3008 | cur + iosize - 1, | |
3009 | NULL, 1); | |
3010 | else if (compressed) { | |
3011 | /* we don't want to end_page_writeback on | |
3012 | * a compressed extent. this happens | |
3013 | * elsewhere | |
3014 | */ | |
3015 | nr++; | |
3016 | } | |
3017 | ||
3018 | cur += iosize; | |
3019 | pg_offset += iosize; | |
3020 | continue; | |
3021 | } | |
3022 | /* leave this out until we have a page_mkwrite call */ | |
3023 | if (0 && !test_range_bit(tree, cur, cur + iosize - 1, | |
3024 | EXTENT_DIRTY, 0, NULL)) { | |
3025 | cur = cur + iosize; | |
3026 | pg_offset += iosize; | |
3027 | continue; | |
3028 | } | |
3029 | ||
3030 | if (tree->ops && tree->ops->writepage_io_hook) { | |
3031 | ret = tree->ops->writepage_io_hook(page, cur, | |
3032 | cur + iosize - 1); | |
3033 | } else { | |
3034 | ret = 0; | |
3035 | } | |
3036 | if (ret) { | |
3037 | SetPageError(page); | |
3038 | } else { | |
3039 | unsigned long max_nr = end_index + 1; | |
3040 | ||
3041 | set_range_writeback(tree, cur, cur + iosize - 1); | |
3042 | if (!PageWriteback(page)) { | |
3043 | printk(KERN_ERR "btrfs warning page %lu not " | |
3044 | "writeback, cur %llu end %llu\n", | |
3045 | page->index, (unsigned long long)cur, | |
3046 | (unsigned long long)end); | |
3047 | } | |
3048 | ||
3049 | ret = submit_extent_page(write_flags, tree, page, | |
3050 | sector, iosize, pg_offset, | |
3051 | bdev, &epd->bio, max_nr, | |
3052 | end_bio_extent_writepage, | |
3053 | 0, 0, 0); | |
3054 | if (ret) | |
3055 | SetPageError(page); | |
3056 | } | |
3057 | cur = cur + iosize; | |
3058 | pg_offset += iosize; | |
3059 | nr++; | |
3060 | } | |
3061 | done: | |
3062 | if (nr == 0) { | |
3063 | /* make sure the mapping tag for page dirty gets cleared */ | |
3064 | set_page_writeback(page); | |
3065 | end_page_writeback(page); | |
3066 | } | |
3067 | unlock_page(page); | |
3068 | ||
3069 | done_unlocked: | |
3070 | ||
3071 | /* drop our reference on any cached states */ | |
3072 | free_extent_state(cached_state); | |
3073 | return 0; | |
3074 | } | |
3075 | ||
3076 | static int eb_wait(void *word) | |
3077 | { | |
3078 | io_schedule(); | |
3079 | return 0; | |
3080 | } | |
3081 | ||
3082 | static void wait_on_extent_buffer_writeback(struct extent_buffer *eb) | |
3083 | { | |
3084 | wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait, | |
3085 | TASK_UNINTERRUPTIBLE); | |
3086 | } | |
3087 | ||
3088 | static int lock_extent_buffer_for_io(struct extent_buffer *eb, | |
3089 | struct btrfs_fs_info *fs_info, | |
3090 | struct extent_page_data *epd) | |
3091 | { | |
3092 | unsigned long i, num_pages; | |
3093 | int flush = 0; | |
3094 | int ret = 0; | |
3095 | ||
3096 | if (!btrfs_try_tree_write_lock(eb)) { | |
3097 | flush = 1; | |
3098 | flush_write_bio(epd); | |
3099 | btrfs_tree_lock(eb); | |
3100 | } | |
3101 | ||
3102 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3103 | btrfs_tree_unlock(eb); | |
3104 | if (!epd->sync_io) | |
3105 | return 0; | |
3106 | if (!flush) { | |
3107 | flush_write_bio(epd); | |
3108 | flush = 1; | |
3109 | } | |
3110 | while (1) { | |
3111 | wait_on_extent_buffer_writeback(eb); | |
3112 | btrfs_tree_lock(eb); | |
3113 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3114 | break; | |
3115 | btrfs_tree_unlock(eb); | |
3116 | } | |
3117 | } | |
3118 | ||
3119 | /* | |
3120 | * We need to do this to prevent races in people who check if the eb is | |
3121 | * under IO since we can end up having no IO bits set for a short period | |
3122 | * of time. | |
3123 | */ | |
3124 | spin_lock(&eb->refs_lock); | |
3125 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { | |
3126 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
3127 | spin_unlock(&eb->refs_lock); | |
3128 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); | |
3129 | spin_lock(&fs_info->delalloc_lock); | |
3130 | if (fs_info->dirty_metadata_bytes >= eb->len) | |
3131 | fs_info->dirty_metadata_bytes -= eb->len; | |
3132 | else | |
3133 | WARN_ON(1); | |
3134 | spin_unlock(&fs_info->delalloc_lock); | |
3135 | ret = 1; | |
3136 | } else { | |
3137 | spin_unlock(&eb->refs_lock); | |
3138 | } | |
3139 | ||
3140 | btrfs_tree_unlock(eb); | |
3141 | ||
3142 | if (!ret) | |
3143 | return ret; | |
3144 | ||
3145 | num_pages = num_extent_pages(eb->start, eb->len); | |
3146 | for (i = 0; i < num_pages; i++) { | |
3147 | struct page *p = extent_buffer_page(eb, i); | |
3148 | ||
3149 | if (!trylock_page(p)) { | |
3150 | if (!flush) { | |
3151 | flush_write_bio(epd); | |
3152 | flush = 1; | |
3153 | } | |
3154 | lock_page(p); | |
3155 | } | |
3156 | } | |
3157 | ||
3158 | return ret; | |
3159 | } | |
3160 | ||
3161 | static void end_extent_buffer_writeback(struct extent_buffer *eb) | |
3162 | { | |
3163 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
3164 | smp_mb__after_clear_bit(); | |
3165 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); | |
3166 | } | |
3167 | ||
3168 | static void end_bio_extent_buffer_writepage(struct bio *bio, int err) | |
3169 | { | |
3170 | int uptodate = err == 0; | |
3171 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
3172 | struct extent_buffer *eb; | |
3173 | int done; | |
3174 | ||
3175 | do { | |
3176 | struct page *page = bvec->bv_page; | |
3177 | ||
3178 | bvec--; | |
3179 | eb = (struct extent_buffer *)page->private; | |
3180 | BUG_ON(!eb); | |
3181 | done = atomic_dec_and_test(&eb->io_pages); | |
3182 | ||
3183 | if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) { | |
3184 | set_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
3185 | ClearPageUptodate(page); | |
3186 | SetPageError(page); | |
3187 | } | |
3188 | ||
3189 | end_page_writeback(page); | |
3190 | ||
3191 | if (!done) | |
3192 | continue; | |
3193 | ||
3194 | end_extent_buffer_writeback(eb); | |
3195 | } while (bvec >= bio->bi_io_vec); | |
3196 | ||
3197 | bio_put(bio); | |
3198 | ||
3199 | } | |
3200 | ||
3201 | static int write_one_eb(struct extent_buffer *eb, | |
3202 | struct btrfs_fs_info *fs_info, | |
3203 | struct writeback_control *wbc, | |
3204 | struct extent_page_data *epd) | |
3205 | { | |
3206 | struct block_device *bdev = fs_info->fs_devices->latest_bdev; | |
3207 | u64 offset = eb->start; | |
3208 | unsigned long i, num_pages; | |
3209 | unsigned long bio_flags = 0; | |
3210 | int rw = (epd->sync_io ? WRITE_SYNC : WRITE); | |
3211 | int ret = 0; | |
3212 | ||
3213 | clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
3214 | num_pages = num_extent_pages(eb->start, eb->len); | |
3215 | atomic_set(&eb->io_pages, num_pages); | |
3216 | if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID) | |
3217 | bio_flags = EXTENT_BIO_TREE_LOG; | |
3218 | ||
3219 | for (i = 0; i < num_pages; i++) { | |
3220 | struct page *p = extent_buffer_page(eb, i); | |
3221 | ||
3222 | clear_page_dirty_for_io(p); | |
3223 | set_page_writeback(p); | |
3224 | ret = submit_extent_page(rw, eb->tree, p, offset >> 9, | |
3225 | PAGE_CACHE_SIZE, 0, bdev, &epd->bio, | |
3226 | -1, end_bio_extent_buffer_writepage, | |
3227 | 0, epd->bio_flags, bio_flags); | |
3228 | epd->bio_flags = bio_flags; | |
3229 | if (ret) { | |
3230 | set_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
3231 | SetPageError(p); | |
3232 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) | |
3233 | end_extent_buffer_writeback(eb); | |
3234 | ret = -EIO; | |
3235 | break; | |
3236 | } | |
3237 | offset += PAGE_CACHE_SIZE; | |
3238 | update_nr_written(p, wbc, 1); | |
3239 | unlock_page(p); | |
3240 | } | |
3241 | ||
3242 | if (unlikely(ret)) { | |
3243 | for (; i < num_pages; i++) { | |
3244 | struct page *p = extent_buffer_page(eb, i); | |
3245 | unlock_page(p); | |
3246 | } | |
3247 | } | |
3248 | ||
3249 | return ret; | |
3250 | } | |
3251 | ||
3252 | int btree_write_cache_pages(struct address_space *mapping, | |
3253 | struct writeback_control *wbc) | |
3254 | { | |
3255 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; | |
3256 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; | |
3257 | struct extent_buffer *eb, *prev_eb = NULL; | |
3258 | struct extent_page_data epd = { | |
3259 | .bio = NULL, | |
3260 | .tree = tree, | |
3261 | .extent_locked = 0, | |
3262 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
3263 | .bio_flags = 0, | |
3264 | }; | |
3265 | int ret = 0; | |
3266 | int done = 0; | |
3267 | int nr_to_write_done = 0; | |
3268 | struct pagevec pvec; | |
3269 | int nr_pages; | |
3270 | pgoff_t index; | |
3271 | pgoff_t end; /* Inclusive */ | |
3272 | int scanned = 0; | |
3273 | int tag; | |
3274 | ||
3275 | pagevec_init(&pvec, 0); | |
3276 | if (wbc->range_cyclic) { | |
3277 | index = mapping->writeback_index; /* Start from prev offset */ | |
3278 | end = -1; | |
3279 | } else { | |
3280 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
3281 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
3282 | scanned = 1; | |
3283 | } | |
3284 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3285 | tag = PAGECACHE_TAG_TOWRITE; | |
3286 | else | |
3287 | tag = PAGECACHE_TAG_DIRTY; | |
3288 | retry: | |
3289 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3290 | tag_pages_for_writeback(mapping, index, end); | |
3291 | while (!done && !nr_to_write_done && (index <= end) && | |
3292 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, | |
3293 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
3294 | unsigned i; | |
3295 | ||
3296 | scanned = 1; | |
3297 | for (i = 0; i < nr_pages; i++) { | |
3298 | struct page *page = pvec.pages[i]; | |
3299 | ||
3300 | if (!PagePrivate(page)) | |
3301 | continue; | |
3302 | ||
3303 | if (!wbc->range_cyclic && page->index > end) { | |
3304 | done = 1; | |
3305 | break; | |
3306 | } | |
3307 | ||
3308 | spin_lock(&mapping->private_lock); | |
3309 | if (!PagePrivate(page)) { | |
3310 | spin_unlock(&mapping->private_lock); | |
3311 | continue; | |
3312 | } | |
3313 | ||
3314 | eb = (struct extent_buffer *)page->private; | |
3315 | ||
3316 | /* | |
3317 | * Shouldn't happen and normally this would be a BUG_ON | |
3318 | * but no sense in crashing the users box for something | |
3319 | * we can survive anyway. | |
3320 | */ | |
3321 | if (!eb) { | |
3322 | spin_unlock(&mapping->private_lock); | |
3323 | WARN_ON(1); | |
3324 | continue; | |
3325 | } | |
3326 | ||
3327 | if (eb == prev_eb) { | |
3328 | spin_unlock(&mapping->private_lock); | |
3329 | continue; | |
3330 | } | |
3331 | ||
3332 | ret = atomic_inc_not_zero(&eb->refs); | |
3333 | spin_unlock(&mapping->private_lock); | |
3334 | if (!ret) | |
3335 | continue; | |
3336 | ||
3337 | prev_eb = eb; | |
3338 | ret = lock_extent_buffer_for_io(eb, fs_info, &epd); | |
3339 | if (!ret) { | |
3340 | free_extent_buffer(eb); | |
3341 | continue; | |
3342 | } | |
3343 | ||
3344 | ret = write_one_eb(eb, fs_info, wbc, &epd); | |
3345 | if (ret) { | |
3346 | done = 1; | |
3347 | free_extent_buffer(eb); | |
3348 | break; | |
3349 | } | |
3350 | free_extent_buffer(eb); | |
3351 | ||
3352 | /* | |
3353 | * the filesystem may choose to bump up nr_to_write. | |
3354 | * We have to make sure to honor the new nr_to_write | |
3355 | * at any time | |
3356 | */ | |
3357 | nr_to_write_done = wbc->nr_to_write <= 0; | |
3358 | } | |
3359 | pagevec_release(&pvec); | |
3360 | cond_resched(); | |
3361 | } | |
3362 | if (!scanned && !done) { | |
3363 | /* | |
3364 | * We hit the last page and there is more work to be done: wrap | |
3365 | * back to the start of the file | |
3366 | */ | |
3367 | scanned = 1; | |
3368 | index = 0; | |
3369 | goto retry; | |
3370 | } | |
3371 | flush_write_bio(&epd); | |
3372 | return ret; | |
3373 | } | |
3374 | ||
3375 | /** | |
3376 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. | |
3377 | * @mapping: address space structure to write | |
3378 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
3379 | * @writepage: function called for each page | |
3380 | * @data: data passed to writepage function | |
3381 | * | |
3382 | * If a page is already under I/O, write_cache_pages() skips it, even | |
3383 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
3384 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
3385 | * and msync() need to guarantee that all the data which was dirty at the time | |
3386 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
3387 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
3388 | * existing IO to complete. | |
3389 | */ | |
3390 | static int extent_write_cache_pages(struct extent_io_tree *tree, | |
3391 | struct address_space *mapping, | |
3392 | struct writeback_control *wbc, | |
3393 | writepage_t writepage, void *data, | |
3394 | void (*flush_fn)(void *)) | |
3395 | { | |
3396 | struct inode *inode = mapping->host; | |
3397 | int ret = 0; | |
3398 | int done = 0; | |
3399 | int nr_to_write_done = 0; | |
3400 | struct pagevec pvec; | |
3401 | int nr_pages; | |
3402 | pgoff_t index; | |
3403 | pgoff_t end; /* Inclusive */ | |
3404 | int scanned = 0; | |
3405 | int tag; | |
3406 | ||
3407 | /* | |
3408 | * We have to hold onto the inode so that ordered extents can do their | |
3409 | * work when the IO finishes. The alternative to this is failing to add | |
3410 | * an ordered extent if the igrab() fails there and that is a huge pain | |
3411 | * to deal with, so instead just hold onto the inode throughout the | |
3412 | * writepages operation. If it fails here we are freeing up the inode | |
3413 | * anyway and we'd rather not waste our time writing out stuff that is | |
3414 | * going to be truncated anyway. | |
3415 | */ | |
3416 | if (!igrab(inode)) | |
3417 | return 0; | |
3418 | ||
3419 | pagevec_init(&pvec, 0); | |
3420 | if (wbc->range_cyclic) { | |
3421 | index = mapping->writeback_index; /* Start from prev offset */ | |
3422 | end = -1; | |
3423 | } else { | |
3424 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
3425 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
3426 | scanned = 1; | |
3427 | } | |
3428 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3429 | tag = PAGECACHE_TAG_TOWRITE; | |
3430 | else | |
3431 | tag = PAGECACHE_TAG_DIRTY; | |
3432 | retry: | |
3433 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3434 | tag_pages_for_writeback(mapping, index, end); | |
3435 | while (!done && !nr_to_write_done && (index <= end) && | |
3436 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, | |
3437 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
3438 | unsigned i; | |
3439 | ||
3440 | scanned = 1; | |
3441 | for (i = 0; i < nr_pages; i++) { | |
3442 | struct page *page = pvec.pages[i]; | |
3443 | ||
3444 | /* | |
3445 | * At this point we hold neither mapping->tree_lock nor | |
3446 | * lock on the page itself: the page may be truncated or | |
3447 | * invalidated (changing page->mapping to NULL), or even | |
3448 | * swizzled back from swapper_space to tmpfs file | |
3449 | * mapping | |
3450 | */ | |
3451 | if (tree->ops && | |
3452 | tree->ops->write_cache_pages_lock_hook) { | |
3453 | tree->ops->write_cache_pages_lock_hook(page, | |
3454 | data, flush_fn); | |
3455 | } else { | |
3456 | if (!trylock_page(page)) { | |
3457 | flush_fn(data); | |
3458 | lock_page(page); | |
3459 | } | |
3460 | } | |
3461 | ||
3462 | if (unlikely(page->mapping != mapping)) { | |
3463 | unlock_page(page); | |
3464 | continue; | |
3465 | } | |
3466 | ||
3467 | if (!wbc->range_cyclic && page->index > end) { | |
3468 | done = 1; | |
3469 | unlock_page(page); | |
3470 | continue; | |
3471 | } | |
3472 | ||
3473 | if (wbc->sync_mode != WB_SYNC_NONE) { | |
3474 | if (PageWriteback(page)) | |
3475 | flush_fn(data); | |
3476 | wait_on_page_writeback(page); | |
3477 | } | |
3478 | ||
3479 | if (PageWriteback(page) || | |
3480 | !clear_page_dirty_for_io(page)) { | |
3481 | unlock_page(page); | |
3482 | continue; | |
3483 | } | |
3484 | ||
3485 | ret = (*writepage)(page, wbc, data); | |
3486 | ||
3487 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
3488 | unlock_page(page); | |
3489 | ret = 0; | |
3490 | } | |
3491 | if (ret) | |
3492 | done = 1; | |
3493 | ||
3494 | /* | |
3495 | * the filesystem may choose to bump up nr_to_write. | |
3496 | * We have to make sure to honor the new nr_to_write | |
3497 | * at any time | |
3498 | */ | |
3499 | nr_to_write_done = wbc->nr_to_write <= 0; | |
3500 | } | |
3501 | pagevec_release(&pvec); | |
3502 | cond_resched(); | |
3503 | } | |
3504 | if (!scanned && !done) { | |
3505 | /* | |
3506 | * We hit the last page and there is more work to be done: wrap | |
3507 | * back to the start of the file | |
3508 | */ | |
3509 | scanned = 1; | |
3510 | index = 0; | |
3511 | goto retry; | |
3512 | } | |
3513 | btrfs_add_delayed_iput(inode); | |
3514 | return ret; | |
3515 | } | |
3516 | ||
3517 | static void flush_epd_write_bio(struct extent_page_data *epd) | |
3518 | { | |
3519 | if (epd->bio) { | |
3520 | int rw = WRITE; | |
3521 | int ret; | |
3522 | ||
3523 | if (epd->sync_io) | |
3524 | rw = WRITE_SYNC; | |
3525 | ||
3526 | ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags); | |
3527 | BUG_ON(ret < 0); /* -ENOMEM */ | |
3528 | epd->bio = NULL; | |
3529 | } | |
3530 | } | |
3531 | ||
3532 | static noinline void flush_write_bio(void *data) | |
3533 | { | |
3534 | struct extent_page_data *epd = data; | |
3535 | flush_epd_write_bio(epd); | |
3536 | } | |
3537 | ||
3538 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, | |
3539 | get_extent_t *get_extent, | |
3540 | struct writeback_control *wbc) | |
3541 | { | |
3542 | int ret; | |
3543 | struct extent_page_data epd = { | |
3544 | .bio = NULL, | |
3545 | .tree = tree, | |
3546 | .get_extent = get_extent, | |
3547 | .extent_locked = 0, | |
3548 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
3549 | .bio_flags = 0, | |
3550 | }; | |
3551 | ||
3552 | ret = __extent_writepage(page, wbc, &epd); | |
3553 | ||
3554 | flush_epd_write_bio(&epd); | |
3555 | return ret; | |
3556 | } | |
3557 | ||
3558 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, | |
3559 | u64 start, u64 end, get_extent_t *get_extent, | |
3560 | int mode) | |
3561 | { | |
3562 | int ret = 0; | |
3563 | struct address_space *mapping = inode->i_mapping; | |
3564 | struct page *page; | |
3565 | unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >> | |
3566 | PAGE_CACHE_SHIFT; | |
3567 | ||
3568 | struct extent_page_data epd = { | |
3569 | .bio = NULL, | |
3570 | .tree = tree, | |
3571 | .get_extent = get_extent, | |
3572 | .extent_locked = 1, | |
3573 | .sync_io = mode == WB_SYNC_ALL, | |
3574 | .bio_flags = 0, | |
3575 | }; | |
3576 | struct writeback_control wbc_writepages = { | |
3577 | .sync_mode = mode, | |
3578 | .nr_to_write = nr_pages * 2, | |
3579 | .range_start = start, | |
3580 | .range_end = end + 1, | |
3581 | }; | |
3582 | ||
3583 | while (start <= end) { | |
3584 | page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); | |
3585 | if (clear_page_dirty_for_io(page)) | |
3586 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
3587 | else { | |
3588 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
3589 | tree->ops->writepage_end_io_hook(page, start, | |
3590 | start + PAGE_CACHE_SIZE - 1, | |
3591 | NULL, 1); | |
3592 | unlock_page(page); | |
3593 | } | |
3594 | page_cache_release(page); | |
3595 | start += PAGE_CACHE_SIZE; | |
3596 | } | |
3597 | ||
3598 | flush_epd_write_bio(&epd); | |
3599 | return ret; | |
3600 | } | |
3601 | ||
3602 | int extent_writepages(struct extent_io_tree *tree, | |
3603 | struct address_space *mapping, | |
3604 | get_extent_t *get_extent, | |
3605 | struct writeback_control *wbc) | |
3606 | { | |
3607 | int ret = 0; | |
3608 | struct extent_page_data epd = { | |
3609 | .bio = NULL, | |
3610 | .tree = tree, | |
3611 | .get_extent = get_extent, | |
3612 | .extent_locked = 0, | |
3613 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
3614 | .bio_flags = 0, | |
3615 | }; | |
3616 | ||
3617 | ret = extent_write_cache_pages(tree, mapping, wbc, | |
3618 | __extent_writepage, &epd, | |
3619 | flush_write_bio); | |
3620 | flush_epd_write_bio(&epd); | |
3621 | return ret; | |
3622 | } | |
3623 | ||
3624 | int extent_readpages(struct extent_io_tree *tree, | |
3625 | struct address_space *mapping, | |
3626 | struct list_head *pages, unsigned nr_pages, | |
3627 | get_extent_t get_extent) | |
3628 | { | |
3629 | struct bio *bio = NULL; | |
3630 | unsigned page_idx; | |
3631 | unsigned long bio_flags = 0; | |
3632 | struct page *pagepool[16]; | |
3633 | struct page *page; | |
3634 | int i = 0; | |
3635 | int nr = 0; | |
3636 | ||
3637 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { | |
3638 | page = list_entry(pages->prev, struct page, lru); | |
3639 | ||
3640 | prefetchw(&page->flags); | |
3641 | list_del(&page->lru); | |
3642 | if (add_to_page_cache_lru(page, mapping, | |
3643 | page->index, GFP_NOFS)) { | |
3644 | page_cache_release(page); | |
3645 | continue; | |
3646 | } | |
3647 | ||
3648 | pagepool[nr++] = page; | |
3649 | if (nr < ARRAY_SIZE(pagepool)) | |
3650 | continue; | |
3651 | for (i = 0; i < nr; i++) { | |
3652 | __extent_read_full_page(tree, pagepool[i], get_extent, | |
3653 | &bio, 0, &bio_flags); | |
3654 | page_cache_release(pagepool[i]); | |
3655 | } | |
3656 | nr = 0; | |
3657 | } | |
3658 | for (i = 0; i < nr; i++) { | |
3659 | __extent_read_full_page(tree, pagepool[i], get_extent, | |
3660 | &bio, 0, &bio_flags); | |
3661 | page_cache_release(pagepool[i]); | |
3662 | } | |
3663 | ||
3664 | BUG_ON(!list_empty(pages)); | |
3665 | if (bio) | |
3666 | return submit_one_bio(READ, bio, 0, bio_flags); | |
3667 | return 0; | |
3668 | } | |
3669 | ||
3670 | /* | |
3671 | * basic invalidatepage code, this waits on any locked or writeback | |
3672 | * ranges corresponding to the page, and then deletes any extent state | |
3673 | * records from the tree | |
3674 | */ | |
3675 | int extent_invalidatepage(struct extent_io_tree *tree, | |
3676 | struct page *page, unsigned long offset) | |
3677 | { | |
3678 | struct extent_state *cached_state = NULL; | |
3679 | u64 start = ((u64)page->index << PAGE_CACHE_SHIFT); | |
3680 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
3681 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
3682 | ||
3683 | start += (offset + blocksize - 1) & ~(blocksize - 1); | |
3684 | if (start > end) | |
3685 | return 0; | |
3686 | ||
3687 | lock_extent_bits(tree, start, end, 0, &cached_state); | |
3688 | wait_on_page_writeback(page); | |
3689 | clear_extent_bit(tree, start, end, | |
3690 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | | |
3691 | EXTENT_DO_ACCOUNTING, | |
3692 | 1, 1, &cached_state, GFP_NOFS); | |
3693 | return 0; | |
3694 | } | |
3695 | ||
3696 | /* | |
3697 | * a helper for releasepage, this tests for areas of the page that | |
3698 | * are locked or under IO and drops the related state bits if it is safe | |
3699 | * to drop the page. | |
3700 | */ | |
3701 | int try_release_extent_state(struct extent_map_tree *map, | |
3702 | struct extent_io_tree *tree, struct page *page, | |
3703 | gfp_t mask) | |
3704 | { | |
3705 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
3706 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
3707 | int ret = 1; | |
3708 | ||
3709 | if (test_range_bit(tree, start, end, | |
3710 | EXTENT_IOBITS, 0, NULL)) | |
3711 | ret = 0; | |
3712 | else { | |
3713 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
3714 | mask = GFP_NOFS; | |
3715 | /* | |
3716 | * at this point we can safely clear everything except the | |
3717 | * locked bit and the nodatasum bit | |
3718 | */ | |
3719 | ret = clear_extent_bit(tree, start, end, | |
3720 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), | |
3721 | 0, 0, NULL, mask); | |
3722 | ||
3723 | /* if clear_extent_bit failed for enomem reasons, | |
3724 | * we can't allow the release to continue. | |
3725 | */ | |
3726 | if (ret < 0) | |
3727 | ret = 0; | |
3728 | else | |
3729 | ret = 1; | |
3730 | } | |
3731 | return ret; | |
3732 | } | |
3733 | ||
3734 | /* | |
3735 | * a helper for releasepage. As long as there are no locked extents | |
3736 | * in the range corresponding to the page, both state records and extent | |
3737 | * map records are removed | |
3738 | */ | |
3739 | int try_release_extent_mapping(struct extent_map_tree *map, | |
3740 | struct extent_io_tree *tree, struct page *page, | |
3741 | gfp_t mask) | |
3742 | { | |
3743 | struct extent_map *em; | |
3744 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
3745 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
3746 | ||
3747 | if ((mask & __GFP_WAIT) && | |
3748 | page->mapping->host->i_size > 16 * 1024 * 1024) { | |
3749 | u64 len; | |
3750 | while (start <= end) { | |
3751 | len = end - start + 1; | |
3752 | write_lock(&map->lock); | |
3753 | em = lookup_extent_mapping(map, start, len); | |
3754 | if (!em) { | |
3755 | write_unlock(&map->lock); | |
3756 | break; | |
3757 | } | |
3758 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || | |
3759 | em->start != start) { | |
3760 | write_unlock(&map->lock); | |
3761 | free_extent_map(em); | |
3762 | break; | |
3763 | } | |
3764 | if (!test_range_bit(tree, em->start, | |
3765 | extent_map_end(em) - 1, | |
3766 | EXTENT_LOCKED | EXTENT_WRITEBACK, | |
3767 | 0, NULL)) { | |
3768 | remove_extent_mapping(map, em); | |
3769 | /* once for the rb tree */ | |
3770 | free_extent_map(em); | |
3771 | } | |
3772 | start = extent_map_end(em); | |
3773 | write_unlock(&map->lock); | |
3774 | ||
3775 | /* once for us */ | |
3776 | free_extent_map(em); | |
3777 | } | |
3778 | } | |
3779 | return try_release_extent_state(map, tree, page, mask); | |
3780 | } | |
3781 | ||
3782 | /* | |
3783 | * helper function for fiemap, which doesn't want to see any holes. | |
3784 | * This maps until we find something past 'last' | |
3785 | */ | |
3786 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
3787 | u64 offset, | |
3788 | u64 last, | |
3789 | get_extent_t *get_extent) | |
3790 | { | |
3791 | u64 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
3792 | struct extent_map *em; | |
3793 | u64 len; | |
3794 | ||
3795 | if (offset >= last) | |
3796 | return NULL; | |
3797 | ||
3798 | while(1) { | |
3799 | len = last - offset; | |
3800 | if (len == 0) | |
3801 | break; | |
3802 | len = (len + sectorsize - 1) & ~(sectorsize - 1); | |
3803 | em = get_extent(inode, NULL, 0, offset, len, 0); | |
3804 | if (IS_ERR_OR_NULL(em)) | |
3805 | return em; | |
3806 | ||
3807 | /* if this isn't a hole return it */ | |
3808 | if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) && | |
3809 | em->block_start != EXTENT_MAP_HOLE) { | |
3810 | return em; | |
3811 | } | |
3812 | ||
3813 | /* this is a hole, advance to the next extent */ | |
3814 | offset = extent_map_end(em); | |
3815 | free_extent_map(em); | |
3816 | if (offset >= last) | |
3817 | break; | |
3818 | } | |
3819 | return NULL; | |
3820 | } | |
3821 | ||
3822 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, | |
3823 | __u64 start, __u64 len, get_extent_t *get_extent) | |
3824 | { | |
3825 | int ret = 0; | |
3826 | u64 off = start; | |
3827 | u64 max = start + len; | |
3828 | u32 flags = 0; | |
3829 | u32 found_type; | |
3830 | u64 last; | |
3831 | u64 last_for_get_extent = 0; | |
3832 | u64 disko = 0; | |
3833 | u64 isize = i_size_read(inode); | |
3834 | struct btrfs_key found_key; | |
3835 | struct extent_map *em = NULL; | |
3836 | struct extent_state *cached_state = NULL; | |
3837 | struct btrfs_path *path; | |
3838 | struct btrfs_file_extent_item *item; | |
3839 | int end = 0; | |
3840 | u64 em_start = 0; | |
3841 | u64 em_len = 0; | |
3842 | u64 em_end = 0; | |
3843 | unsigned long emflags; | |
3844 | ||
3845 | if (len == 0) | |
3846 | return -EINVAL; | |
3847 | ||
3848 | path = btrfs_alloc_path(); | |
3849 | if (!path) | |
3850 | return -ENOMEM; | |
3851 | path->leave_spinning = 1; | |
3852 | ||
3853 | start = ALIGN(start, BTRFS_I(inode)->root->sectorsize); | |
3854 | len = ALIGN(len, BTRFS_I(inode)->root->sectorsize); | |
3855 | ||
3856 | /* | |
3857 | * lookup the last file extent. We're not using i_size here | |
3858 | * because there might be preallocation past i_size | |
3859 | */ | |
3860 | ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root, | |
3861 | path, btrfs_ino(inode), -1, 0); | |
3862 | if (ret < 0) { | |
3863 | btrfs_free_path(path); | |
3864 | return ret; | |
3865 | } | |
3866 | WARN_ON(!ret); | |
3867 | path->slots[0]--; | |
3868 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3869 | struct btrfs_file_extent_item); | |
3870 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
3871 | found_type = btrfs_key_type(&found_key); | |
3872 | ||
3873 | /* No extents, but there might be delalloc bits */ | |
3874 | if (found_key.objectid != btrfs_ino(inode) || | |
3875 | found_type != BTRFS_EXTENT_DATA_KEY) { | |
3876 | /* have to trust i_size as the end */ | |
3877 | last = (u64)-1; | |
3878 | last_for_get_extent = isize; | |
3879 | } else { | |
3880 | /* | |
3881 | * remember the start of the last extent. There are a | |
3882 | * bunch of different factors that go into the length of the | |
3883 | * extent, so its much less complex to remember where it started | |
3884 | */ | |
3885 | last = found_key.offset; | |
3886 | last_for_get_extent = last + 1; | |
3887 | } | |
3888 | btrfs_free_path(path); | |
3889 | ||
3890 | /* | |
3891 | * we might have some extents allocated but more delalloc past those | |
3892 | * extents. so, we trust isize unless the start of the last extent is | |
3893 | * beyond isize | |
3894 | */ | |
3895 | if (last < isize) { | |
3896 | last = (u64)-1; | |
3897 | last_for_get_extent = isize; | |
3898 | } | |
3899 | ||
3900 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0, | |
3901 | &cached_state); | |
3902 | ||
3903 | em = get_extent_skip_holes(inode, start, last_for_get_extent, | |
3904 | get_extent); | |
3905 | if (!em) | |
3906 | goto out; | |
3907 | if (IS_ERR(em)) { | |
3908 | ret = PTR_ERR(em); | |
3909 | goto out; | |
3910 | } | |
3911 | ||
3912 | while (!end) { | |
3913 | u64 offset_in_extent; | |
3914 | ||
3915 | /* break if the extent we found is outside the range */ | |
3916 | if (em->start >= max || extent_map_end(em) < off) | |
3917 | break; | |
3918 | ||
3919 | /* | |
3920 | * get_extent may return an extent that starts before our | |
3921 | * requested range. We have to make sure the ranges | |
3922 | * we return to fiemap always move forward and don't | |
3923 | * overlap, so adjust the offsets here | |
3924 | */ | |
3925 | em_start = max(em->start, off); | |
3926 | ||
3927 | /* | |
3928 | * record the offset from the start of the extent | |
3929 | * for adjusting the disk offset below | |
3930 | */ | |
3931 | offset_in_extent = em_start - em->start; | |
3932 | em_end = extent_map_end(em); | |
3933 | em_len = em_end - em_start; | |
3934 | emflags = em->flags; | |
3935 | disko = 0; | |
3936 | flags = 0; | |
3937 | ||
3938 | /* | |
3939 | * bump off for our next call to get_extent | |
3940 | */ | |
3941 | off = extent_map_end(em); | |
3942 | if (off >= max) | |
3943 | end = 1; | |
3944 | ||
3945 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { | |
3946 | end = 1; | |
3947 | flags |= FIEMAP_EXTENT_LAST; | |
3948 | } else if (em->block_start == EXTENT_MAP_INLINE) { | |
3949 | flags |= (FIEMAP_EXTENT_DATA_INLINE | | |
3950 | FIEMAP_EXTENT_NOT_ALIGNED); | |
3951 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { | |
3952 | flags |= (FIEMAP_EXTENT_DELALLOC | | |
3953 | FIEMAP_EXTENT_UNKNOWN); | |
3954 | } else { | |
3955 | disko = em->block_start + offset_in_extent; | |
3956 | } | |
3957 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
3958 | flags |= FIEMAP_EXTENT_ENCODED; | |
3959 | ||
3960 | free_extent_map(em); | |
3961 | em = NULL; | |
3962 | if ((em_start >= last) || em_len == (u64)-1 || | |
3963 | (last == (u64)-1 && isize <= em_end)) { | |
3964 | flags |= FIEMAP_EXTENT_LAST; | |
3965 | end = 1; | |
3966 | } | |
3967 | ||
3968 | /* now scan forward to see if this is really the last extent. */ | |
3969 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
3970 | get_extent); | |
3971 | if (IS_ERR(em)) { | |
3972 | ret = PTR_ERR(em); | |
3973 | goto out; | |
3974 | } | |
3975 | if (!em) { | |
3976 | flags |= FIEMAP_EXTENT_LAST; | |
3977 | end = 1; | |
3978 | } | |
3979 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, | |
3980 | em_len, flags); | |
3981 | if (ret) | |
3982 | goto out_free; | |
3983 | } | |
3984 | out_free: | |
3985 | free_extent_map(em); | |
3986 | out: | |
3987 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len, | |
3988 | &cached_state, GFP_NOFS); | |
3989 | return ret; | |
3990 | } | |
3991 | ||
3992 | static void __free_extent_buffer(struct extent_buffer *eb) | |
3993 | { | |
3994 | #if LEAK_DEBUG | |
3995 | unsigned long flags; | |
3996 | spin_lock_irqsave(&leak_lock, flags); | |
3997 | list_del(&eb->leak_list); | |
3998 | spin_unlock_irqrestore(&leak_lock, flags); | |
3999 | #endif | |
4000 | if (eb->pages && eb->pages != eb->inline_pages) | |
4001 | kfree(eb->pages); | |
4002 | kmem_cache_free(extent_buffer_cache, eb); | |
4003 | } | |
4004 | ||
4005 | static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree, | |
4006 | u64 start, | |
4007 | unsigned long len, | |
4008 | gfp_t mask) | |
4009 | { | |
4010 | struct extent_buffer *eb = NULL; | |
4011 | #if LEAK_DEBUG | |
4012 | unsigned long flags; | |
4013 | #endif | |
4014 | ||
4015 | eb = kmem_cache_zalloc(extent_buffer_cache, mask); | |
4016 | if (eb == NULL) | |
4017 | return NULL; | |
4018 | eb->start = start; | |
4019 | eb->len = len; | |
4020 | eb->tree = tree; | |
4021 | eb->bflags = 0; | |
4022 | rwlock_init(&eb->lock); | |
4023 | atomic_set(&eb->write_locks, 0); | |
4024 | atomic_set(&eb->read_locks, 0); | |
4025 | atomic_set(&eb->blocking_readers, 0); | |
4026 | atomic_set(&eb->blocking_writers, 0); | |
4027 | atomic_set(&eb->spinning_readers, 0); | |
4028 | atomic_set(&eb->spinning_writers, 0); | |
4029 | eb->lock_nested = 0; | |
4030 | init_waitqueue_head(&eb->write_lock_wq); | |
4031 | init_waitqueue_head(&eb->read_lock_wq); | |
4032 | ||
4033 | #if LEAK_DEBUG | |
4034 | spin_lock_irqsave(&leak_lock, flags); | |
4035 | list_add(&eb->leak_list, &buffers); | |
4036 | spin_unlock_irqrestore(&leak_lock, flags); | |
4037 | #endif | |
4038 | spin_lock_init(&eb->refs_lock); | |
4039 | atomic_set(&eb->refs, 1); | |
4040 | atomic_set(&eb->io_pages, 0); | |
4041 | ||
4042 | if (len > MAX_INLINE_EXTENT_BUFFER_SIZE) { | |
4043 | struct page **pages; | |
4044 | int num_pages = (len + PAGE_CACHE_SIZE - 1) >> | |
4045 | PAGE_CACHE_SHIFT; | |
4046 | pages = kzalloc(num_pages, mask); | |
4047 | if (!pages) { | |
4048 | __free_extent_buffer(eb); | |
4049 | return NULL; | |
4050 | } | |
4051 | eb->pages = pages; | |
4052 | } else { | |
4053 | eb->pages = eb->inline_pages; | |
4054 | } | |
4055 | ||
4056 | return eb; | |
4057 | } | |
4058 | ||
4059 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) | |
4060 | { | |
4061 | unsigned long i; | |
4062 | struct page *p; | |
4063 | struct extent_buffer *new; | |
4064 | unsigned long num_pages = num_extent_pages(src->start, src->len); | |
4065 | ||
4066 | new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_ATOMIC); | |
4067 | if (new == NULL) | |
4068 | return NULL; | |
4069 | ||
4070 | for (i = 0; i < num_pages; i++) { | |
4071 | p = alloc_page(GFP_ATOMIC); | |
4072 | BUG_ON(!p); | |
4073 | attach_extent_buffer_page(new, p); | |
4074 | WARN_ON(PageDirty(p)); | |
4075 | SetPageUptodate(p); | |
4076 | new->pages[i] = p; | |
4077 | } | |
4078 | ||
4079 | copy_extent_buffer(new, src, 0, 0, src->len); | |
4080 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); | |
4081 | set_bit(EXTENT_BUFFER_DUMMY, &new->bflags); | |
4082 | ||
4083 | return new; | |
4084 | } | |
4085 | ||
4086 | struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len) | |
4087 | { | |
4088 | struct extent_buffer *eb; | |
4089 | unsigned long num_pages = num_extent_pages(0, len); | |
4090 | unsigned long i; | |
4091 | ||
4092 | eb = __alloc_extent_buffer(NULL, start, len, GFP_ATOMIC); | |
4093 | if (!eb) | |
4094 | return NULL; | |
4095 | ||
4096 | for (i = 0; i < num_pages; i++) { | |
4097 | eb->pages[i] = alloc_page(GFP_ATOMIC); | |
4098 | if (!eb->pages[i]) | |
4099 | goto err; | |
4100 | } | |
4101 | set_extent_buffer_uptodate(eb); | |
4102 | btrfs_set_header_nritems(eb, 0); | |
4103 | set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4104 | ||
4105 | return eb; | |
4106 | err: | |
4107 | for (i--; i > 0; i--) | |
4108 | __free_page(eb->pages[i]); | |
4109 | __free_extent_buffer(eb); | |
4110 | return NULL; | |
4111 | } | |
4112 | ||
4113 | static int extent_buffer_under_io(struct extent_buffer *eb) | |
4114 | { | |
4115 | return (atomic_read(&eb->io_pages) || | |
4116 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4117 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4118 | } | |
4119 | ||
4120 | /* | |
4121 | * Helper for releasing extent buffer page. | |
4122 | */ | |
4123 | static void btrfs_release_extent_buffer_page(struct extent_buffer *eb, | |
4124 | unsigned long start_idx) | |
4125 | { | |
4126 | unsigned long index; | |
4127 | unsigned long num_pages; | |
4128 | struct page *page; | |
4129 | int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4130 | ||
4131 | BUG_ON(extent_buffer_under_io(eb)); | |
4132 | ||
4133 | num_pages = num_extent_pages(eb->start, eb->len); | |
4134 | index = start_idx + num_pages; | |
4135 | if (start_idx >= index) | |
4136 | return; | |
4137 | ||
4138 | do { | |
4139 | index--; | |
4140 | page = extent_buffer_page(eb, index); | |
4141 | if (page && mapped) { | |
4142 | spin_lock(&page->mapping->private_lock); | |
4143 | /* | |
4144 | * We do this since we'll remove the pages after we've | |
4145 | * removed the eb from the radix tree, so we could race | |
4146 | * and have this page now attached to the new eb. So | |
4147 | * only clear page_private if it's still connected to | |
4148 | * this eb. | |
4149 | */ | |
4150 | if (PagePrivate(page) && | |
4151 | page->private == (unsigned long)eb) { | |
4152 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4153 | BUG_ON(PageDirty(page)); | |
4154 | BUG_ON(PageWriteback(page)); | |
4155 | /* | |
4156 | * We need to make sure we haven't be attached | |
4157 | * to a new eb. | |
4158 | */ | |
4159 | ClearPagePrivate(page); | |
4160 | set_page_private(page, 0); | |
4161 | /* One for the page private */ | |
4162 | page_cache_release(page); | |
4163 | } | |
4164 | spin_unlock(&page->mapping->private_lock); | |
4165 | ||
4166 | } | |
4167 | if (page) { | |
4168 | /* One for when we alloced the page */ | |
4169 | page_cache_release(page); | |
4170 | } | |
4171 | } while (index != start_idx); | |
4172 | } | |
4173 | ||
4174 | /* | |
4175 | * Helper for releasing the extent buffer. | |
4176 | */ | |
4177 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4178 | { | |
4179 | btrfs_release_extent_buffer_page(eb, 0); | |
4180 | __free_extent_buffer(eb); | |
4181 | } | |
4182 | ||
4183 | static void check_buffer_tree_ref(struct extent_buffer *eb) | |
4184 | { | |
4185 | /* the ref bit is tricky. We have to make sure it is set | |
4186 | * if we have the buffer dirty. Otherwise the | |
4187 | * code to free a buffer can end up dropping a dirty | |
4188 | * page | |
4189 | * | |
4190 | * Once the ref bit is set, it won't go away while the | |
4191 | * buffer is dirty or in writeback, and it also won't | |
4192 | * go away while we have the reference count on the | |
4193 | * eb bumped. | |
4194 | * | |
4195 | * We can't just set the ref bit without bumping the | |
4196 | * ref on the eb because free_extent_buffer might | |
4197 | * see the ref bit and try to clear it. If this happens | |
4198 | * free_extent_buffer might end up dropping our original | |
4199 | * ref by mistake and freeing the page before we are able | |
4200 | * to add one more ref. | |
4201 | * | |
4202 | * So bump the ref count first, then set the bit. If someone | |
4203 | * beat us to it, drop the ref we added. | |
4204 | */ | |
4205 | spin_lock(&eb->refs_lock); | |
4206 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
4207 | atomic_inc(&eb->refs); | |
4208 | spin_unlock(&eb->refs_lock); | |
4209 | } | |
4210 | ||
4211 | static void mark_extent_buffer_accessed(struct extent_buffer *eb) | |
4212 | { | |
4213 | unsigned long num_pages, i; | |
4214 | ||
4215 | check_buffer_tree_ref(eb); | |
4216 | ||
4217 | num_pages = num_extent_pages(eb->start, eb->len); | |
4218 | for (i = 0; i < num_pages; i++) { | |
4219 | struct page *p = extent_buffer_page(eb, i); | |
4220 | mark_page_accessed(p); | |
4221 | } | |
4222 | } | |
4223 | ||
4224 | struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree, | |
4225 | u64 start, unsigned long len) | |
4226 | { | |
4227 | unsigned long num_pages = num_extent_pages(start, len); | |
4228 | unsigned long i; | |
4229 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
4230 | struct extent_buffer *eb; | |
4231 | struct extent_buffer *exists = NULL; | |
4232 | struct page *p; | |
4233 | struct address_space *mapping = tree->mapping; | |
4234 | int uptodate = 1; | |
4235 | int ret; | |
4236 | ||
4237 | rcu_read_lock(); | |
4238 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); | |
4239 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
4240 | rcu_read_unlock(); | |
4241 | mark_extent_buffer_accessed(eb); | |
4242 | return eb; | |
4243 | } | |
4244 | rcu_read_unlock(); | |
4245 | ||
4246 | eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS); | |
4247 | if (!eb) | |
4248 | return NULL; | |
4249 | ||
4250 | for (i = 0; i < num_pages; i++, index++) { | |
4251 | p = find_or_create_page(mapping, index, GFP_NOFS); | |
4252 | if (!p) | |
4253 | goto free_eb; | |
4254 | ||
4255 | spin_lock(&mapping->private_lock); | |
4256 | if (PagePrivate(p)) { | |
4257 | /* | |
4258 | * We could have already allocated an eb for this page | |
4259 | * and attached one so lets see if we can get a ref on | |
4260 | * the existing eb, and if we can we know it's good and | |
4261 | * we can just return that one, else we know we can just | |
4262 | * overwrite page->private. | |
4263 | */ | |
4264 | exists = (struct extent_buffer *)p->private; | |
4265 | if (atomic_inc_not_zero(&exists->refs)) { | |
4266 | spin_unlock(&mapping->private_lock); | |
4267 | unlock_page(p); | |
4268 | page_cache_release(p); | |
4269 | mark_extent_buffer_accessed(exists); | |
4270 | goto free_eb; | |
4271 | } | |
4272 | ||
4273 | /* | |
4274 | * Do this so attach doesn't complain and we need to | |
4275 | * drop the ref the old guy had. | |
4276 | */ | |
4277 | ClearPagePrivate(p); | |
4278 | WARN_ON(PageDirty(p)); | |
4279 | page_cache_release(p); | |
4280 | } | |
4281 | attach_extent_buffer_page(eb, p); | |
4282 | spin_unlock(&mapping->private_lock); | |
4283 | WARN_ON(PageDirty(p)); | |
4284 | mark_page_accessed(p); | |
4285 | eb->pages[i] = p; | |
4286 | if (!PageUptodate(p)) | |
4287 | uptodate = 0; | |
4288 | ||
4289 | /* | |
4290 | * see below about how we avoid a nasty race with release page | |
4291 | * and why we unlock later | |
4292 | */ | |
4293 | } | |
4294 | if (uptodate) | |
4295 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
4296 | again: | |
4297 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); | |
4298 | if (ret) | |
4299 | goto free_eb; | |
4300 | ||
4301 | spin_lock(&tree->buffer_lock); | |
4302 | ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb); | |
4303 | if (ret == -EEXIST) { | |
4304 | exists = radix_tree_lookup(&tree->buffer, | |
4305 | start >> PAGE_CACHE_SHIFT); | |
4306 | if (!atomic_inc_not_zero(&exists->refs)) { | |
4307 | spin_unlock(&tree->buffer_lock); | |
4308 | radix_tree_preload_end(); | |
4309 | exists = NULL; | |
4310 | goto again; | |
4311 | } | |
4312 | spin_unlock(&tree->buffer_lock); | |
4313 | radix_tree_preload_end(); | |
4314 | mark_extent_buffer_accessed(exists); | |
4315 | goto free_eb; | |
4316 | } | |
4317 | /* add one reference for the tree */ | |
4318 | check_buffer_tree_ref(eb); | |
4319 | spin_unlock(&tree->buffer_lock); | |
4320 | radix_tree_preload_end(); | |
4321 | ||
4322 | /* | |
4323 | * there is a race where release page may have | |
4324 | * tried to find this extent buffer in the radix | |
4325 | * but failed. It will tell the VM it is safe to | |
4326 | * reclaim the, and it will clear the page private bit. | |
4327 | * We must make sure to set the page private bit properly | |
4328 | * after the extent buffer is in the radix tree so | |
4329 | * it doesn't get lost | |
4330 | */ | |
4331 | SetPageChecked(eb->pages[0]); | |
4332 | for (i = 1; i < num_pages; i++) { | |
4333 | p = extent_buffer_page(eb, i); | |
4334 | ClearPageChecked(p); | |
4335 | unlock_page(p); | |
4336 | } | |
4337 | unlock_page(eb->pages[0]); | |
4338 | return eb; | |
4339 | ||
4340 | free_eb: | |
4341 | for (i = 0; i < num_pages; i++) { | |
4342 | if (eb->pages[i]) | |
4343 | unlock_page(eb->pages[i]); | |
4344 | } | |
4345 | ||
4346 | WARN_ON(!atomic_dec_and_test(&eb->refs)); | |
4347 | btrfs_release_extent_buffer(eb); | |
4348 | return exists; | |
4349 | } | |
4350 | ||
4351 | struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree, | |
4352 | u64 start, unsigned long len) | |
4353 | { | |
4354 | struct extent_buffer *eb; | |
4355 | ||
4356 | rcu_read_lock(); | |
4357 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); | |
4358 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
4359 | rcu_read_unlock(); | |
4360 | mark_extent_buffer_accessed(eb); | |
4361 | return eb; | |
4362 | } | |
4363 | rcu_read_unlock(); | |
4364 | ||
4365 | return NULL; | |
4366 | } | |
4367 | ||
4368 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) | |
4369 | { | |
4370 | struct extent_buffer *eb = | |
4371 | container_of(head, struct extent_buffer, rcu_head); | |
4372 | ||
4373 | __free_extent_buffer(eb); | |
4374 | } | |
4375 | ||
4376 | /* Expects to have eb->eb_lock already held */ | |
4377 | static int release_extent_buffer(struct extent_buffer *eb, gfp_t mask) | |
4378 | { | |
4379 | WARN_ON(atomic_read(&eb->refs) == 0); | |
4380 | if (atomic_dec_and_test(&eb->refs)) { | |
4381 | if (test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) { | |
4382 | spin_unlock(&eb->refs_lock); | |
4383 | } else { | |
4384 | struct extent_io_tree *tree = eb->tree; | |
4385 | ||
4386 | spin_unlock(&eb->refs_lock); | |
4387 | ||
4388 | spin_lock(&tree->buffer_lock); | |
4389 | radix_tree_delete(&tree->buffer, | |
4390 | eb->start >> PAGE_CACHE_SHIFT); | |
4391 | spin_unlock(&tree->buffer_lock); | |
4392 | } | |
4393 | ||
4394 | /* Should be safe to release our pages at this point */ | |
4395 | btrfs_release_extent_buffer_page(eb, 0); | |
4396 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); | |
4397 | return 1; | |
4398 | } | |
4399 | spin_unlock(&eb->refs_lock); | |
4400 | ||
4401 | return 0; | |
4402 | } | |
4403 | ||
4404 | void free_extent_buffer(struct extent_buffer *eb) | |
4405 | { | |
4406 | if (!eb) | |
4407 | return; | |
4408 | ||
4409 | spin_lock(&eb->refs_lock); | |
4410 | if (atomic_read(&eb->refs) == 2 && | |
4411 | test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) | |
4412 | atomic_dec(&eb->refs); | |
4413 | ||
4414 | if (atomic_read(&eb->refs) == 2 && | |
4415 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
4416 | !extent_buffer_under_io(eb) && | |
4417 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
4418 | atomic_dec(&eb->refs); | |
4419 | ||
4420 | /* | |
4421 | * I know this is terrible, but it's temporary until we stop tracking | |
4422 | * the uptodate bits and such for the extent buffers. | |
4423 | */ | |
4424 | release_extent_buffer(eb, GFP_ATOMIC); | |
4425 | } | |
4426 | ||
4427 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
4428 | { | |
4429 | if (!eb) | |
4430 | return; | |
4431 | ||
4432 | spin_lock(&eb->refs_lock); | |
4433 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
4434 | ||
4435 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && | |
4436 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
4437 | atomic_dec(&eb->refs); | |
4438 | release_extent_buffer(eb, GFP_NOFS); | |
4439 | } | |
4440 | ||
4441 | void clear_extent_buffer_dirty(struct extent_buffer *eb) | |
4442 | { | |
4443 | unsigned long i; | |
4444 | unsigned long num_pages; | |
4445 | struct page *page; | |
4446 | ||
4447 | num_pages = num_extent_pages(eb->start, eb->len); | |
4448 | ||
4449 | for (i = 0; i < num_pages; i++) { | |
4450 | page = extent_buffer_page(eb, i); | |
4451 | if (!PageDirty(page)) | |
4452 | continue; | |
4453 | ||
4454 | lock_page(page); | |
4455 | WARN_ON(!PagePrivate(page)); | |
4456 | ||
4457 | clear_page_dirty_for_io(page); | |
4458 | spin_lock_irq(&page->mapping->tree_lock); | |
4459 | if (!PageDirty(page)) { | |
4460 | radix_tree_tag_clear(&page->mapping->page_tree, | |
4461 | page_index(page), | |
4462 | PAGECACHE_TAG_DIRTY); | |
4463 | } | |
4464 | spin_unlock_irq(&page->mapping->tree_lock); | |
4465 | ClearPageError(page); | |
4466 | unlock_page(page); | |
4467 | } | |
4468 | WARN_ON(atomic_read(&eb->refs) == 0); | |
4469 | } | |
4470 | ||
4471 | int set_extent_buffer_dirty(struct extent_buffer *eb) | |
4472 | { | |
4473 | unsigned long i; | |
4474 | unsigned long num_pages; | |
4475 | int was_dirty = 0; | |
4476 | ||
4477 | check_buffer_tree_ref(eb); | |
4478 | ||
4479 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); | |
4480 | ||
4481 | num_pages = num_extent_pages(eb->start, eb->len); | |
4482 | WARN_ON(atomic_read(&eb->refs) == 0); | |
4483 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); | |
4484 | ||
4485 | for (i = 0; i < num_pages; i++) | |
4486 | set_page_dirty(extent_buffer_page(eb, i)); | |
4487 | return was_dirty; | |
4488 | } | |
4489 | ||
4490 | static int range_straddles_pages(u64 start, u64 len) | |
4491 | { | |
4492 | if (len < PAGE_CACHE_SIZE) | |
4493 | return 1; | |
4494 | if (start & (PAGE_CACHE_SIZE - 1)) | |
4495 | return 1; | |
4496 | if ((start + len) & (PAGE_CACHE_SIZE - 1)) | |
4497 | return 1; | |
4498 | return 0; | |
4499 | } | |
4500 | ||
4501 | int clear_extent_buffer_uptodate(struct extent_buffer *eb) | |
4502 | { | |
4503 | unsigned long i; | |
4504 | struct page *page; | |
4505 | unsigned long num_pages; | |
4506 | ||
4507 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
4508 | num_pages = num_extent_pages(eb->start, eb->len); | |
4509 | for (i = 0; i < num_pages; i++) { | |
4510 | page = extent_buffer_page(eb, i); | |
4511 | if (page) | |
4512 | ClearPageUptodate(page); | |
4513 | } | |
4514 | return 0; | |
4515 | } | |
4516 | ||
4517 | int set_extent_buffer_uptodate(struct extent_buffer *eb) | |
4518 | { | |
4519 | unsigned long i; | |
4520 | struct page *page; | |
4521 | unsigned long num_pages; | |
4522 | ||
4523 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
4524 | num_pages = num_extent_pages(eb->start, eb->len); | |
4525 | for (i = 0; i < num_pages; i++) { | |
4526 | page = extent_buffer_page(eb, i); | |
4527 | SetPageUptodate(page); | |
4528 | } | |
4529 | return 0; | |
4530 | } | |
4531 | ||
4532 | int extent_range_uptodate(struct extent_io_tree *tree, | |
4533 | u64 start, u64 end) | |
4534 | { | |
4535 | struct page *page; | |
4536 | int ret; | |
4537 | int pg_uptodate = 1; | |
4538 | int uptodate; | |
4539 | unsigned long index; | |
4540 | ||
4541 | if (range_straddles_pages(start, end - start + 1)) { | |
4542 | ret = test_range_bit(tree, start, end, | |
4543 | EXTENT_UPTODATE, 1, NULL); | |
4544 | if (ret) | |
4545 | return 1; | |
4546 | } | |
4547 | while (start <= end) { | |
4548 | index = start >> PAGE_CACHE_SHIFT; | |
4549 | page = find_get_page(tree->mapping, index); | |
4550 | if (!page) | |
4551 | return 1; | |
4552 | uptodate = PageUptodate(page); | |
4553 | page_cache_release(page); | |
4554 | if (!uptodate) { | |
4555 | pg_uptodate = 0; | |
4556 | break; | |
4557 | } | |
4558 | start += PAGE_CACHE_SIZE; | |
4559 | } | |
4560 | return pg_uptodate; | |
4561 | } | |
4562 | ||
4563 | int extent_buffer_uptodate(struct extent_buffer *eb) | |
4564 | { | |
4565 | return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
4566 | } | |
4567 | ||
4568 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
4569 | struct extent_buffer *eb, u64 start, int wait, | |
4570 | get_extent_t *get_extent, int mirror_num) | |
4571 | { | |
4572 | unsigned long i; | |
4573 | unsigned long start_i; | |
4574 | struct page *page; | |
4575 | int err; | |
4576 | int ret = 0; | |
4577 | int locked_pages = 0; | |
4578 | int all_uptodate = 1; | |
4579 | unsigned long num_pages; | |
4580 | unsigned long num_reads = 0; | |
4581 | struct bio *bio = NULL; | |
4582 | unsigned long bio_flags = 0; | |
4583 | ||
4584 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) | |
4585 | return 0; | |
4586 | ||
4587 | if (start) { | |
4588 | WARN_ON(start < eb->start); | |
4589 | start_i = (start >> PAGE_CACHE_SHIFT) - | |
4590 | (eb->start >> PAGE_CACHE_SHIFT); | |
4591 | } else { | |
4592 | start_i = 0; | |
4593 | } | |
4594 | ||
4595 | num_pages = num_extent_pages(eb->start, eb->len); | |
4596 | for (i = start_i; i < num_pages; i++) { | |
4597 | page = extent_buffer_page(eb, i); | |
4598 | if (wait == WAIT_NONE) { | |
4599 | if (!trylock_page(page)) | |
4600 | goto unlock_exit; | |
4601 | } else { | |
4602 | lock_page(page); | |
4603 | } | |
4604 | locked_pages++; | |
4605 | if (!PageUptodate(page)) { | |
4606 | num_reads++; | |
4607 | all_uptodate = 0; | |
4608 | } | |
4609 | } | |
4610 | if (all_uptodate) { | |
4611 | if (start_i == 0) | |
4612 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
4613 | goto unlock_exit; | |
4614 | } | |
4615 | ||
4616 | clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
4617 | eb->read_mirror = 0; | |
4618 | atomic_set(&eb->io_pages, num_reads); | |
4619 | for (i = start_i; i < num_pages; i++) { | |
4620 | page = extent_buffer_page(eb, i); | |
4621 | if (!PageUptodate(page)) { | |
4622 | ClearPageError(page); | |
4623 | err = __extent_read_full_page(tree, page, | |
4624 | get_extent, &bio, | |
4625 | mirror_num, &bio_flags); | |
4626 | if (err) | |
4627 | ret = err; | |
4628 | } else { | |
4629 | unlock_page(page); | |
4630 | } | |
4631 | } | |
4632 | ||
4633 | if (bio) { | |
4634 | err = submit_one_bio(READ, bio, mirror_num, bio_flags); | |
4635 | if (err) | |
4636 | return err; | |
4637 | } | |
4638 | ||
4639 | if (ret || wait != WAIT_COMPLETE) | |
4640 | return ret; | |
4641 | ||
4642 | for (i = start_i; i < num_pages; i++) { | |
4643 | page = extent_buffer_page(eb, i); | |
4644 | wait_on_page_locked(page); | |
4645 | if (!PageUptodate(page)) | |
4646 | ret = -EIO; | |
4647 | } | |
4648 | ||
4649 | return ret; | |
4650 | ||
4651 | unlock_exit: | |
4652 | i = start_i; | |
4653 | while (locked_pages > 0) { | |
4654 | page = extent_buffer_page(eb, i); | |
4655 | i++; | |
4656 | unlock_page(page); | |
4657 | locked_pages--; | |
4658 | } | |
4659 | return ret; | |
4660 | } | |
4661 | ||
4662 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
4663 | unsigned long start, | |
4664 | unsigned long len) | |
4665 | { | |
4666 | size_t cur; | |
4667 | size_t offset; | |
4668 | struct page *page; | |
4669 | char *kaddr; | |
4670 | char *dst = (char *)dstv; | |
4671 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4672 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4673 | ||
4674 | WARN_ON(start > eb->len); | |
4675 | WARN_ON(start + len > eb->start + eb->len); | |
4676 | ||
4677 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4678 | ||
4679 | while (len > 0) { | |
4680 | page = extent_buffer_page(eb, i); | |
4681 | ||
4682 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
4683 | kaddr = page_address(page); | |
4684 | memcpy(dst, kaddr + offset, cur); | |
4685 | ||
4686 | dst += cur; | |
4687 | len -= cur; | |
4688 | offset = 0; | |
4689 | i++; | |
4690 | } | |
4691 | } | |
4692 | ||
4693 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
4694 | unsigned long min_len, char **map, | |
4695 | unsigned long *map_start, | |
4696 | unsigned long *map_len) | |
4697 | { | |
4698 | size_t offset = start & (PAGE_CACHE_SIZE - 1); | |
4699 | char *kaddr; | |
4700 | struct page *p; | |
4701 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4702 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4703 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
4704 | PAGE_CACHE_SHIFT; | |
4705 | ||
4706 | if (i != end_i) | |
4707 | return -EINVAL; | |
4708 | ||
4709 | if (i == 0) { | |
4710 | offset = start_offset; | |
4711 | *map_start = 0; | |
4712 | } else { | |
4713 | offset = 0; | |
4714 | *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; | |
4715 | } | |
4716 | ||
4717 | if (start + min_len > eb->len) { | |
4718 | printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, " | |
4719 | "wanted %lu %lu\n", (unsigned long long)eb->start, | |
4720 | eb->len, start, min_len); | |
4721 | WARN_ON(1); | |
4722 | return -EINVAL; | |
4723 | } | |
4724 | ||
4725 | p = extent_buffer_page(eb, i); | |
4726 | kaddr = page_address(p); | |
4727 | *map = kaddr + offset; | |
4728 | *map_len = PAGE_CACHE_SIZE - offset; | |
4729 | return 0; | |
4730 | } | |
4731 | ||
4732 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, | |
4733 | unsigned long start, | |
4734 | unsigned long len) | |
4735 | { | |
4736 | size_t cur; | |
4737 | size_t offset; | |
4738 | struct page *page; | |
4739 | char *kaddr; | |
4740 | char *ptr = (char *)ptrv; | |
4741 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4742 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4743 | int ret = 0; | |
4744 | ||
4745 | WARN_ON(start > eb->len); | |
4746 | WARN_ON(start + len > eb->start + eb->len); | |
4747 | ||
4748 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4749 | ||
4750 | while (len > 0) { | |
4751 | page = extent_buffer_page(eb, i); | |
4752 | ||
4753 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
4754 | ||
4755 | kaddr = page_address(page); | |
4756 | ret = memcmp(ptr, kaddr + offset, cur); | |
4757 | if (ret) | |
4758 | break; | |
4759 | ||
4760 | ptr += cur; | |
4761 | len -= cur; | |
4762 | offset = 0; | |
4763 | i++; | |
4764 | } | |
4765 | return ret; | |
4766 | } | |
4767 | ||
4768 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
4769 | unsigned long start, unsigned long len) | |
4770 | { | |
4771 | size_t cur; | |
4772 | size_t offset; | |
4773 | struct page *page; | |
4774 | char *kaddr; | |
4775 | char *src = (char *)srcv; | |
4776 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4777 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4778 | ||
4779 | WARN_ON(start > eb->len); | |
4780 | WARN_ON(start + len > eb->start + eb->len); | |
4781 | ||
4782 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4783 | ||
4784 | while (len > 0) { | |
4785 | page = extent_buffer_page(eb, i); | |
4786 | WARN_ON(!PageUptodate(page)); | |
4787 | ||
4788 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
4789 | kaddr = page_address(page); | |
4790 | memcpy(kaddr + offset, src, cur); | |
4791 | ||
4792 | src += cur; | |
4793 | len -= cur; | |
4794 | offset = 0; | |
4795 | i++; | |
4796 | } | |
4797 | } | |
4798 | ||
4799 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
4800 | unsigned long start, unsigned long len) | |
4801 | { | |
4802 | size_t cur; | |
4803 | size_t offset; | |
4804 | struct page *page; | |
4805 | char *kaddr; | |
4806 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4807 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4808 | ||
4809 | WARN_ON(start > eb->len); | |
4810 | WARN_ON(start + len > eb->start + eb->len); | |
4811 | ||
4812 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4813 | ||
4814 | while (len > 0) { | |
4815 | page = extent_buffer_page(eb, i); | |
4816 | WARN_ON(!PageUptodate(page)); | |
4817 | ||
4818 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
4819 | kaddr = page_address(page); | |
4820 | memset(kaddr + offset, c, cur); | |
4821 | ||
4822 | len -= cur; | |
4823 | offset = 0; | |
4824 | i++; | |
4825 | } | |
4826 | } | |
4827 | ||
4828 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
4829 | unsigned long dst_offset, unsigned long src_offset, | |
4830 | unsigned long len) | |
4831 | { | |
4832 | u64 dst_len = dst->len; | |
4833 | size_t cur; | |
4834 | size_t offset; | |
4835 | struct page *page; | |
4836 | char *kaddr; | |
4837 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4838 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
4839 | ||
4840 | WARN_ON(src->len != dst_len); | |
4841 | ||
4842 | offset = (start_offset + dst_offset) & | |
4843 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4844 | ||
4845 | while (len > 0) { | |
4846 | page = extent_buffer_page(dst, i); | |
4847 | WARN_ON(!PageUptodate(page)); | |
4848 | ||
4849 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
4850 | ||
4851 | kaddr = page_address(page); | |
4852 | read_extent_buffer(src, kaddr + offset, src_offset, cur); | |
4853 | ||
4854 | src_offset += cur; | |
4855 | len -= cur; | |
4856 | offset = 0; | |
4857 | i++; | |
4858 | } | |
4859 | } | |
4860 | ||
4861 | static void move_pages(struct page *dst_page, struct page *src_page, | |
4862 | unsigned long dst_off, unsigned long src_off, | |
4863 | unsigned long len) | |
4864 | { | |
4865 | char *dst_kaddr = page_address(dst_page); | |
4866 | if (dst_page == src_page) { | |
4867 | memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len); | |
4868 | } else { | |
4869 | char *src_kaddr = page_address(src_page); | |
4870 | char *p = dst_kaddr + dst_off + len; | |
4871 | char *s = src_kaddr + src_off + len; | |
4872 | ||
4873 | while (len--) | |
4874 | *--p = *--s; | |
4875 | } | |
4876 | } | |
4877 | ||
4878 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) | |
4879 | { | |
4880 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
4881 | return distance < len; | |
4882 | } | |
4883 | ||
4884 | static void copy_pages(struct page *dst_page, struct page *src_page, | |
4885 | unsigned long dst_off, unsigned long src_off, | |
4886 | unsigned long len) | |
4887 | { | |
4888 | char *dst_kaddr = page_address(dst_page); | |
4889 | char *src_kaddr; | |
4890 | int must_memmove = 0; | |
4891 | ||
4892 | if (dst_page != src_page) { | |
4893 | src_kaddr = page_address(src_page); | |
4894 | } else { | |
4895 | src_kaddr = dst_kaddr; | |
4896 | if (areas_overlap(src_off, dst_off, len)) | |
4897 | must_memmove = 1; | |
4898 | } | |
4899 | ||
4900 | if (must_memmove) | |
4901 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
4902 | else | |
4903 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
4904 | } | |
4905 | ||
4906 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
4907 | unsigned long src_offset, unsigned long len) | |
4908 | { | |
4909 | size_t cur; | |
4910 | size_t dst_off_in_page; | |
4911 | size_t src_off_in_page; | |
4912 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4913 | unsigned long dst_i; | |
4914 | unsigned long src_i; | |
4915 | ||
4916 | if (src_offset + len > dst->len) { | |
4917 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " | |
4918 | "len %lu dst len %lu\n", src_offset, len, dst->len); | |
4919 | BUG_ON(1); | |
4920 | } | |
4921 | if (dst_offset + len > dst->len) { | |
4922 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " | |
4923 | "len %lu dst len %lu\n", dst_offset, len, dst->len); | |
4924 | BUG_ON(1); | |
4925 | } | |
4926 | ||
4927 | while (len > 0) { | |
4928 | dst_off_in_page = (start_offset + dst_offset) & | |
4929 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4930 | src_off_in_page = (start_offset + src_offset) & | |
4931 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4932 | ||
4933 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
4934 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
4935 | ||
4936 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
4937 | src_off_in_page)); | |
4938 | cur = min_t(unsigned long, cur, | |
4939 | (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); | |
4940 | ||
4941 | copy_pages(extent_buffer_page(dst, dst_i), | |
4942 | extent_buffer_page(dst, src_i), | |
4943 | dst_off_in_page, src_off_in_page, cur); | |
4944 | ||
4945 | src_offset += cur; | |
4946 | dst_offset += cur; | |
4947 | len -= cur; | |
4948 | } | |
4949 | } | |
4950 | ||
4951 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
4952 | unsigned long src_offset, unsigned long len) | |
4953 | { | |
4954 | size_t cur; | |
4955 | size_t dst_off_in_page; | |
4956 | size_t src_off_in_page; | |
4957 | unsigned long dst_end = dst_offset + len - 1; | |
4958 | unsigned long src_end = src_offset + len - 1; | |
4959 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4960 | unsigned long dst_i; | |
4961 | unsigned long src_i; | |
4962 | ||
4963 | if (src_offset + len > dst->len) { | |
4964 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " | |
4965 | "len %lu len %lu\n", src_offset, len, dst->len); | |
4966 | BUG_ON(1); | |
4967 | } | |
4968 | if (dst_offset + len > dst->len) { | |
4969 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " | |
4970 | "len %lu len %lu\n", dst_offset, len, dst->len); | |
4971 | BUG_ON(1); | |
4972 | } | |
4973 | if (dst_offset < src_offset) { | |
4974 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); | |
4975 | return; | |
4976 | } | |
4977 | while (len > 0) { | |
4978 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; | |
4979 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
4980 | ||
4981 | dst_off_in_page = (start_offset + dst_end) & | |
4982 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4983 | src_off_in_page = (start_offset + src_end) & | |
4984 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4985 | ||
4986 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
4987 | cur = min(cur, dst_off_in_page + 1); | |
4988 | move_pages(extent_buffer_page(dst, dst_i), | |
4989 | extent_buffer_page(dst, src_i), | |
4990 | dst_off_in_page - cur + 1, | |
4991 | src_off_in_page - cur + 1, cur); | |
4992 | ||
4993 | dst_end -= cur; | |
4994 | src_end -= cur; | |
4995 | len -= cur; | |
4996 | } | |
4997 | } | |
4998 | ||
4999 | int try_release_extent_buffer(struct page *page, gfp_t mask) | |
5000 | { | |
5001 | struct extent_buffer *eb; | |
5002 | ||
5003 | /* | |
5004 | * We need to make sure noboody is attaching this page to an eb right | |
5005 | * now. | |
5006 | */ | |
5007 | spin_lock(&page->mapping->private_lock); | |
5008 | if (!PagePrivate(page)) { | |
5009 | spin_unlock(&page->mapping->private_lock); | |
5010 | return 1; | |
5011 | } | |
5012 | ||
5013 | eb = (struct extent_buffer *)page->private; | |
5014 | BUG_ON(!eb); | |
5015 | ||
5016 | /* | |
5017 | * This is a little awful but should be ok, we need to make sure that | |
5018 | * the eb doesn't disappear out from under us while we're looking at | |
5019 | * this page. | |
5020 | */ | |
5021 | spin_lock(&eb->refs_lock); | |
5022 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { | |
5023 | spin_unlock(&eb->refs_lock); | |
5024 | spin_unlock(&page->mapping->private_lock); | |
5025 | return 0; | |
5026 | } | |
5027 | spin_unlock(&page->mapping->private_lock); | |
5028 | ||
5029 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
5030 | mask = GFP_NOFS; | |
5031 | ||
5032 | /* | |
5033 | * If tree ref isn't set then we know the ref on this eb is a real ref, | |
5034 | * so just return, this page will likely be freed soon anyway. | |
5035 | */ | |
5036 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { | |
5037 | spin_unlock(&eb->refs_lock); | |
5038 | return 0; | |
5039 | } | |
5040 | ||
5041 | return release_extent_buffer(eb, mask); | |
5042 | } |