]>
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/spinlock.h> | |
8 | #include <linux/blkdev.h> | |
9 | #include <linux/swap.h> | |
10 | #include <linux/writeback.h> | |
11 | #include <linux/pagevec.h> | |
12 | #include <linux/prefetch.h> | |
13 | #include <linux/cleancache.h> | |
14 | #include "extent_io.h" | |
15 | #include "extent_map.h" | |
16 | #include "ctree.h" | |
17 | #include "btrfs_inode.h" | |
18 | #include "volumes.h" | |
19 | #include "check-integrity.h" | |
20 | #include "locking.h" | |
21 | #include "rcu-string.h" | |
22 | #include "backref.h" | |
23 | ||
24 | static struct kmem_cache *extent_state_cache; | |
25 | static struct kmem_cache *extent_buffer_cache; | |
26 | static struct bio_set *btrfs_bioset; | |
27 | ||
28 | static inline bool extent_state_in_tree(const struct extent_state *state) | |
29 | { | |
30 | return !RB_EMPTY_NODE(&state->rb_node); | |
31 | } | |
32 | ||
33 | #ifdef CONFIG_BTRFS_DEBUG | |
34 | static LIST_HEAD(buffers); | |
35 | static LIST_HEAD(states); | |
36 | ||
37 | static DEFINE_SPINLOCK(leak_lock); | |
38 | ||
39 | static inline | |
40 | void btrfs_leak_debug_add(struct list_head *new, struct list_head *head) | |
41 | { | |
42 | unsigned long flags; | |
43 | ||
44 | spin_lock_irqsave(&leak_lock, flags); | |
45 | list_add(new, head); | |
46 | spin_unlock_irqrestore(&leak_lock, flags); | |
47 | } | |
48 | ||
49 | static inline | |
50 | void btrfs_leak_debug_del(struct list_head *entry) | |
51 | { | |
52 | unsigned long flags; | |
53 | ||
54 | spin_lock_irqsave(&leak_lock, flags); | |
55 | list_del(entry); | |
56 | spin_unlock_irqrestore(&leak_lock, flags); | |
57 | } | |
58 | ||
59 | static inline | |
60 | void btrfs_leak_debug_check(void) | |
61 | { | |
62 | struct extent_state *state; | |
63 | struct extent_buffer *eb; | |
64 | ||
65 | while (!list_empty(&states)) { | |
66 | state = list_entry(states.next, struct extent_state, leak_list); | |
67 | pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n", | |
68 | state->start, state->end, state->state, | |
69 | extent_state_in_tree(state), | |
70 | atomic_read(&state->refs)); | |
71 | list_del(&state->leak_list); | |
72 | kmem_cache_free(extent_state_cache, state); | |
73 | } | |
74 | ||
75 | while (!list_empty(&buffers)) { | |
76 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
77 | printk(KERN_ERR "BTRFS: buffer leak start %llu len %lu " | |
78 | "refs %d\n", | |
79 | eb->start, eb->len, atomic_read(&eb->refs)); | |
80 | list_del(&eb->leak_list); | |
81 | kmem_cache_free(extent_buffer_cache, eb); | |
82 | } | |
83 | } | |
84 | ||
85 | #define btrfs_debug_check_extent_io_range(tree, start, end) \ | |
86 | __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end)) | |
87 | static inline void __btrfs_debug_check_extent_io_range(const char *caller, | |
88 | struct extent_io_tree *tree, u64 start, u64 end) | |
89 | { | |
90 | struct inode *inode; | |
91 | u64 isize; | |
92 | ||
93 | if (!tree->mapping) | |
94 | return; | |
95 | ||
96 | inode = tree->mapping->host; | |
97 | isize = i_size_read(inode); | |
98 | if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) { | |
99 | btrfs_debug_rl(BTRFS_I(inode)->root->fs_info, | |
100 | "%s: ino %llu isize %llu odd range [%llu,%llu]", | |
101 | caller, btrfs_ino(inode), isize, start, end); | |
102 | } | |
103 | } | |
104 | #else | |
105 | #define btrfs_leak_debug_add(new, head) do {} while (0) | |
106 | #define btrfs_leak_debug_del(entry) do {} while (0) | |
107 | #define btrfs_leak_debug_check() do {} while (0) | |
108 | #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0) | |
109 | #endif | |
110 | ||
111 | #define BUFFER_LRU_MAX 64 | |
112 | ||
113 | struct tree_entry { | |
114 | u64 start; | |
115 | u64 end; | |
116 | struct rb_node rb_node; | |
117 | }; | |
118 | ||
119 | struct extent_page_data { | |
120 | struct bio *bio; | |
121 | struct extent_io_tree *tree; | |
122 | get_extent_t *get_extent; | |
123 | unsigned long bio_flags; | |
124 | ||
125 | /* tells writepage not to lock the state bits for this range | |
126 | * it still does the unlocking | |
127 | */ | |
128 | unsigned int extent_locked:1; | |
129 | ||
130 | /* tells the submit_bio code to use a WRITE_SYNC */ | |
131 | unsigned int sync_io:1; | |
132 | }; | |
133 | ||
134 | static void add_extent_changeset(struct extent_state *state, unsigned bits, | |
135 | struct extent_changeset *changeset, | |
136 | int set) | |
137 | { | |
138 | int ret; | |
139 | ||
140 | if (!changeset) | |
141 | return; | |
142 | if (set && (state->state & bits) == bits) | |
143 | return; | |
144 | if (!set && (state->state & bits) == 0) | |
145 | return; | |
146 | changeset->bytes_changed += state->end - state->start + 1; | |
147 | ret = ulist_add(changeset->range_changed, state->start, state->end, | |
148 | GFP_ATOMIC); | |
149 | /* ENOMEM */ | |
150 | BUG_ON(ret < 0); | |
151 | } | |
152 | ||
153 | static noinline void flush_write_bio(void *data); | |
154 | static inline struct btrfs_fs_info * | |
155 | tree_fs_info(struct extent_io_tree *tree) | |
156 | { | |
157 | if (!tree->mapping) | |
158 | return NULL; | |
159 | return btrfs_sb(tree->mapping->host->i_sb); | |
160 | } | |
161 | ||
162 | int __init extent_io_init(void) | |
163 | { | |
164 | extent_state_cache = kmem_cache_create("btrfs_extent_state", | |
165 | sizeof(struct extent_state), 0, | |
166 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
167 | if (!extent_state_cache) | |
168 | return -ENOMEM; | |
169 | ||
170 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", | |
171 | sizeof(struct extent_buffer), 0, | |
172 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
173 | if (!extent_buffer_cache) | |
174 | goto free_state_cache; | |
175 | ||
176 | btrfs_bioset = bioset_create(BIO_POOL_SIZE, | |
177 | offsetof(struct btrfs_io_bio, bio)); | |
178 | if (!btrfs_bioset) | |
179 | goto free_buffer_cache; | |
180 | ||
181 | if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE)) | |
182 | goto free_bioset; | |
183 | ||
184 | return 0; | |
185 | ||
186 | free_bioset: | |
187 | bioset_free(btrfs_bioset); | |
188 | btrfs_bioset = NULL; | |
189 | ||
190 | free_buffer_cache: | |
191 | kmem_cache_destroy(extent_buffer_cache); | |
192 | extent_buffer_cache = NULL; | |
193 | ||
194 | free_state_cache: | |
195 | kmem_cache_destroy(extent_state_cache); | |
196 | extent_state_cache = NULL; | |
197 | return -ENOMEM; | |
198 | } | |
199 | ||
200 | void extent_io_exit(void) | |
201 | { | |
202 | btrfs_leak_debug_check(); | |
203 | ||
204 | /* | |
205 | * Make sure all delayed rcu free are flushed before we | |
206 | * destroy caches. | |
207 | */ | |
208 | rcu_barrier(); | |
209 | kmem_cache_destroy(extent_state_cache); | |
210 | kmem_cache_destroy(extent_buffer_cache); | |
211 | if (btrfs_bioset) | |
212 | bioset_free(btrfs_bioset); | |
213 | } | |
214 | ||
215 | void extent_io_tree_init(struct extent_io_tree *tree, | |
216 | struct address_space *mapping) | |
217 | { | |
218 | tree->state = RB_ROOT; | |
219 | tree->ops = NULL; | |
220 | tree->dirty_bytes = 0; | |
221 | spin_lock_init(&tree->lock); | |
222 | tree->mapping = mapping; | |
223 | } | |
224 | ||
225 | static struct extent_state *alloc_extent_state(gfp_t mask) | |
226 | { | |
227 | struct extent_state *state; | |
228 | ||
229 | state = kmem_cache_alloc(extent_state_cache, mask); | |
230 | if (!state) | |
231 | return state; | |
232 | state->state = 0; | |
233 | state->failrec = NULL; | |
234 | RB_CLEAR_NODE(&state->rb_node); | |
235 | btrfs_leak_debug_add(&state->leak_list, &states); | |
236 | atomic_set(&state->refs, 1); | |
237 | init_waitqueue_head(&state->wq); | |
238 | trace_alloc_extent_state(state, mask, _RET_IP_); | |
239 | return state; | |
240 | } | |
241 | ||
242 | void free_extent_state(struct extent_state *state) | |
243 | { | |
244 | if (!state) | |
245 | return; | |
246 | if (atomic_dec_and_test(&state->refs)) { | |
247 | WARN_ON(extent_state_in_tree(state)); | |
248 | btrfs_leak_debug_del(&state->leak_list); | |
249 | trace_free_extent_state(state, _RET_IP_); | |
250 | kmem_cache_free(extent_state_cache, state); | |
251 | } | |
252 | } | |
253 | ||
254 | static struct rb_node *tree_insert(struct rb_root *root, | |
255 | struct rb_node *search_start, | |
256 | u64 offset, | |
257 | struct rb_node *node, | |
258 | struct rb_node ***p_in, | |
259 | struct rb_node **parent_in) | |
260 | { | |
261 | struct rb_node **p; | |
262 | struct rb_node *parent = NULL; | |
263 | struct tree_entry *entry; | |
264 | ||
265 | if (p_in && parent_in) { | |
266 | p = *p_in; | |
267 | parent = *parent_in; | |
268 | goto do_insert; | |
269 | } | |
270 | ||
271 | p = search_start ? &search_start : &root->rb_node; | |
272 | while (*p) { | |
273 | parent = *p; | |
274 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
275 | ||
276 | if (offset < entry->start) | |
277 | p = &(*p)->rb_left; | |
278 | else if (offset > entry->end) | |
279 | p = &(*p)->rb_right; | |
280 | else | |
281 | return parent; | |
282 | } | |
283 | ||
284 | do_insert: | |
285 | rb_link_node(node, parent, p); | |
286 | rb_insert_color(node, root); | |
287 | return NULL; | |
288 | } | |
289 | ||
290 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, | |
291 | struct rb_node **prev_ret, | |
292 | struct rb_node **next_ret, | |
293 | struct rb_node ***p_ret, | |
294 | struct rb_node **parent_ret) | |
295 | { | |
296 | struct rb_root *root = &tree->state; | |
297 | struct rb_node **n = &root->rb_node; | |
298 | struct rb_node *prev = NULL; | |
299 | struct rb_node *orig_prev = NULL; | |
300 | struct tree_entry *entry; | |
301 | struct tree_entry *prev_entry = NULL; | |
302 | ||
303 | while (*n) { | |
304 | prev = *n; | |
305 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
306 | prev_entry = entry; | |
307 | ||
308 | if (offset < entry->start) | |
309 | n = &(*n)->rb_left; | |
310 | else if (offset > entry->end) | |
311 | n = &(*n)->rb_right; | |
312 | else | |
313 | return *n; | |
314 | } | |
315 | ||
316 | if (p_ret) | |
317 | *p_ret = n; | |
318 | if (parent_ret) | |
319 | *parent_ret = prev; | |
320 | ||
321 | if (prev_ret) { | |
322 | orig_prev = prev; | |
323 | while (prev && offset > prev_entry->end) { | |
324 | prev = rb_next(prev); | |
325 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
326 | } | |
327 | *prev_ret = prev; | |
328 | prev = orig_prev; | |
329 | } | |
330 | ||
331 | if (next_ret) { | |
332 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
333 | while (prev && offset < prev_entry->start) { | |
334 | prev = rb_prev(prev); | |
335 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
336 | } | |
337 | *next_ret = prev; | |
338 | } | |
339 | return NULL; | |
340 | } | |
341 | ||
342 | static inline struct rb_node * | |
343 | tree_search_for_insert(struct extent_io_tree *tree, | |
344 | u64 offset, | |
345 | struct rb_node ***p_ret, | |
346 | struct rb_node **parent_ret) | |
347 | { | |
348 | struct rb_node *prev = NULL; | |
349 | struct rb_node *ret; | |
350 | ||
351 | ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret); | |
352 | if (!ret) | |
353 | return prev; | |
354 | return ret; | |
355 | } | |
356 | ||
357 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, | |
358 | u64 offset) | |
359 | { | |
360 | return tree_search_for_insert(tree, offset, NULL, NULL); | |
361 | } | |
362 | ||
363 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, | |
364 | struct extent_state *other) | |
365 | { | |
366 | if (tree->ops && tree->ops->merge_extent_hook) | |
367 | tree->ops->merge_extent_hook(tree->mapping->host, new, | |
368 | other); | |
369 | } | |
370 | ||
371 | /* | |
372 | * utility function to look for merge candidates inside a given range. | |
373 | * Any extents with matching state are merged together into a single | |
374 | * extent in the tree. Extents with EXTENT_IO in their state field | |
375 | * are not merged because the end_io handlers need to be able to do | |
376 | * operations on them without sleeping (or doing allocations/splits). | |
377 | * | |
378 | * This should be called with the tree lock held. | |
379 | */ | |
380 | static void merge_state(struct extent_io_tree *tree, | |
381 | struct extent_state *state) | |
382 | { | |
383 | struct extent_state *other; | |
384 | struct rb_node *other_node; | |
385 | ||
386 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) | |
387 | return; | |
388 | ||
389 | other_node = rb_prev(&state->rb_node); | |
390 | if (other_node) { | |
391 | other = rb_entry(other_node, struct extent_state, rb_node); | |
392 | if (other->end == state->start - 1 && | |
393 | other->state == state->state) { | |
394 | merge_cb(tree, state, other); | |
395 | state->start = other->start; | |
396 | rb_erase(&other->rb_node, &tree->state); | |
397 | RB_CLEAR_NODE(&other->rb_node); | |
398 | free_extent_state(other); | |
399 | } | |
400 | } | |
401 | other_node = rb_next(&state->rb_node); | |
402 | if (other_node) { | |
403 | other = rb_entry(other_node, struct extent_state, rb_node); | |
404 | if (other->start == state->end + 1 && | |
405 | other->state == state->state) { | |
406 | merge_cb(tree, state, other); | |
407 | state->end = other->end; | |
408 | rb_erase(&other->rb_node, &tree->state); | |
409 | RB_CLEAR_NODE(&other->rb_node); | |
410 | free_extent_state(other); | |
411 | } | |
412 | } | |
413 | } | |
414 | ||
415 | static void set_state_cb(struct extent_io_tree *tree, | |
416 | struct extent_state *state, unsigned *bits) | |
417 | { | |
418 | if (tree->ops && tree->ops->set_bit_hook) | |
419 | tree->ops->set_bit_hook(tree->mapping->host, state, bits); | |
420 | } | |
421 | ||
422 | static void clear_state_cb(struct extent_io_tree *tree, | |
423 | struct extent_state *state, unsigned *bits) | |
424 | { | |
425 | if (tree->ops && tree->ops->clear_bit_hook) | |
426 | tree->ops->clear_bit_hook(tree->mapping->host, state, bits); | |
427 | } | |
428 | ||
429 | static void set_state_bits(struct extent_io_tree *tree, | |
430 | struct extent_state *state, unsigned *bits, | |
431 | struct extent_changeset *changeset); | |
432 | ||
433 | /* | |
434 | * insert an extent_state struct into the tree. 'bits' are set on the | |
435 | * struct before it is inserted. | |
436 | * | |
437 | * This may return -EEXIST if the extent is already there, in which case the | |
438 | * state struct is freed. | |
439 | * | |
440 | * The tree lock is not taken internally. This is a utility function and | |
441 | * probably isn't what you want to call (see set/clear_extent_bit). | |
442 | */ | |
443 | static int insert_state(struct extent_io_tree *tree, | |
444 | struct extent_state *state, u64 start, u64 end, | |
445 | struct rb_node ***p, | |
446 | struct rb_node **parent, | |
447 | unsigned *bits, struct extent_changeset *changeset) | |
448 | { | |
449 | struct rb_node *node; | |
450 | ||
451 | if (end < start) | |
452 | WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n", | |
453 | end, start); | |
454 | state->start = start; | |
455 | state->end = end; | |
456 | ||
457 | set_state_bits(tree, state, bits, changeset); | |
458 | ||
459 | node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent); | |
460 | if (node) { | |
461 | struct extent_state *found; | |
462 | found = rb_entry(node, struct extent_state, rb_node); | |
463 | printk(KERN_ERR "BTRFS: found node %llu %llu on insert of " | |
464 | "%llu %llu\n", | |
465 | found->start, found->end, start, end); | |
466 | return -EEXIST; | |
467 | } | |
468 | merge_state(tree, state); | |
469 | return 0; | |
470 | } | |
471 | ||
472 | static void split_cb(struct extent_io_tree *tree, struct extent_state *orig, | |
473 | u64 split) | |
474 | { | |
475 | if (tree->ops && tree->ops->split_extent_hook) | |
476 | tree->ops->split_extent_hook(tree->mapping->host, orig, split); | |
477 | } | |
478 | ||
479 | /* | |
480 | * split a given extent state struct in two, inserting the preallocated | |
481 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
482 | * offset inside 'orig' where it should be split. | |
483 | * | |
484 | * Before calling, | |
485 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
486 | * are two extent state structs in the tree: | |
487 | * prealloc: [orig->start, split - 1] | |
488 | * orig: [ split, orig->end ] | |
489 | * | |
490 | * The tree locks are not taken by this function. They need to be held | |
491 | * by the caller. | |
492 | */ | |
493 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
494 | struct extent_state *prealloc, u64 split) | |
495 | { | |
496 | struct rb_node *node; | |
497 | ||
498 | split_cb(tree, orig, split); | |
499 | ||
500 | prealloc->start = orig->start; | |
501 | prealloc->end = split - 1; | |
502 | prealloc->state = orig->state; | |
503 | orig->start = split; | |
504 | ||
505 | node = tree_insert(&tree->state, &orig->rb_node, prealloc->end, | |
506 | &prealloc->rb_node, NULL, NULL); | |
507 | if (node) { | |
508 | free_extent_state(prealloc); | |
509 | return -EEXIST; | |
510 | } | |
511 | return 0; | |
512 | } | |
513 | ||
514 | static struct extent_state *next_state(struct extent_state *state) | |
515 | { | |
516 | struct rb_node *next = rb_next(&state->rb_node); | |
517 | if (next) | |
518 | return rb_entry(next, struct extent_state, rb_node); | |
519 | else | |
520 | return NULL; | |
521 | } | |
522 | ||
523 | /* | |
524 | * utility function to clear some bits in an extent state struct. | |
525 | * it will optionally wake up any one waiting on this state (wake == 1). | |
526 | * | |
527 | * If no bits are set on the state struct after clearing things, the | |
528 | * struct is freed and removed from the tree | |
529 | */ | |
530 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, | |
531 | struct extent_state *state, | |
532 | unsigned *bits, int wake, | |
533 | struct extent_changeset *changeset) | |
534 | { | |
535 | struct extent_state *next; | |
536 | unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS; | |
537 | ||
538 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { | |
539 | u64 range = state->end - state->start + 1; | |
540 | WARN_ON(range > tree->dirty_bytes); | |
541 | tree->dirty_bytes -= range; | |
542 | } | |
543 | clear_state_cb(tree, state, bits); | |
544 | add_extent_changeset(state, bits_to_clear, changeset, 0); | |
545 | state->state &= ~bits_to_clear; | |
546 | if (wake) | |
547 | wake_up(&state->wq); | |
548 | if (state->state == 0) { | |
549 | next = next_state(state); | |
550 | if (extent_state_in_tree(state)) { | |
551 | rb_erase(&state->rb_node, &tree->state); | |
552 | RB_CLEAR_NODE(&state->rb_node); | |
553 | free_extent_state(state); | |
554 | } else { | |
555 | WARN_ON(1); | |
556 | } | |
557 | } else { | |
558 | merge_state(tree, state); | |
559 | next = next_state(state); | |
560 | } | |
561 | return next; | |
562 | } | |
563 | ||
564 | static struct extent_state * | |
565 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
566 | { | |
567 | if (!prealloc) | |
568 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
569 | ||
570 | return prealloc; | |
571 | } | |
572 | ||
573 | static void extent_io_tree_panic(struct extent_io_tree *tree, int err) | |
574 | { | |
575 | btrfs_panic(tree_fs_info(tree), err, "Locking error: " | |
576 | "Extent tree was modified by another " | |
577 | "thread while locked."); | |
578 | } | |
579 | ||
580 | /* | |
581 | * clear some bits on a range in the tree. This may require splitting | |
582 | * or inserting elements in the tree, so the gfp mask is used to | |
583 | * indicate which allocations or sleeping are allowed. | |
584 | * | |
585 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
586 | * the given range from the tree regardless of state (ie for truncate). | |
587 | * | |
588 | * the range [start, end] is inclusive. | |
589 | * | |
590 | * This takes the tree lock, and returns 0 on success and < 0 on error. | |
591 | */ | |
592 | static int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
593 | unsigned bits, int wake, int delete, | |
594 | struct extent_state **cached_state, | |
595 | gfp_t mask, struct extent_changeset *changeset) | |
596 | { | |
597 | struct extent_state *state; | |
598 | struct extent_state *cached; | |
599 | struct extent_state *prealloc = NULL; | |
600 | struct rb_node *node; | |
601 | u64 last_end; | |
602 | int err; | |
603 | int clear = 0; | |
604 | ||
605 | btrfs_debug_check_extent_io_range(tree, start, end); | |
606 | ||
607 | if (bits & EXTENT_DELALLOC) | |
608 | bits |= EXTENT_NORESERVE; | |
609 | ||
610 | if (delete) | |
611 | bits |= ~EXTENT_CTLBITS; | |
612 | bits |= EXTENT_FIRST_DELALLOC; | |
613 | ||
614 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) | |
615 | clear = 1; | |
616 | again: | |
617 | if (!prealloc && gfpflags_allow_blocking(mask)) { | |
618 | /* | |
619 | * Don't care for allocation failure here because we might end | |
620 | * up not needing the pre-allocated extent state at all, which | |
621 | * is the case if we only have in the tree extent states that | |
622 | * cover our input range and don't cover too any other range. | |
623 | * If we end up needing a new extent state we allocate it later. | |
624 | */ | |
625 | prealloc = alloc_extent_state(mask); | |
626 | } | |
627 | ||
628 | spin_lock(&tree->lock); | |
629 | if (cached_state) { | |
630 | cached = *cached_state; | |
631 | ||
632 | if (clear) { | |
633 | *cached_state = NULL; | |
634 | cached_state = NULL; | |
635 | } | |
636 | ||
637 | if (cached && extent_state_in_tree(cached) && | |
638 | cached->start <= start && cached->end > start) { | |
639 | if (clear) | |
640 | atomic_dec(&cached->refs); | |
641 | state = cached; | |
642 | goto hit_next; | |
643 | } | |
644 | if (clear) | |
645 | free_extent_state(cached); | |
646 | } | |
647 | /* | |
648 | * this search will find the extents that end after | |
649 | * our range starts | |
650 | */ | |
651 | node = tree_search(tree, start); | |
652 | if (!node) | |
653 | goto out; | |
654 | state = rb_entry(node, struct extent_state, rb_node); | |
655 | hit_next: | |
656 | if (state->start > end) | |
657 | goto out; | |
658 | WARN_ON(state->end < start); | |
659 | last_end = state->end; | |
660 | ||
661 | /* the state doesn't have the wanted bits, go ahead */ | |
662 | if (!(state->state & bits)) { | |
663 | state = next_state(state); | |
664 | goto next; | |
665 | } | |
666 | ||
667 | /* | |
668 | * | ---- desired range ---- | | |
669 | * | state | or | |
670 | * | ------------- state -------------- | | |
671 | * | |
672 | * We need to split the extent we found, and may flip | |
673 | * bits on second half. | |
674 | * | |
675 | * If the extent we found extends past our range, we | |
676 | * just split and search again. It'll get split again | |
677 | * the next time though. | |
678 | * | |
679 | * If the extent we found is inside our range, we clear | |
680 | * the desired bit on it. | |
681 | */ | |
682 | ||
683 | if (state->start < start) { | |
684 | prealloc = alloc_extent_state_atomic(prealloc); | |
685 | BUG_ON(!prealloc); | |
686 | err = split_state(tree, state, prealloc, start); | |
687 | if (err) | |
688 | extent_io_tree_panic(tree, err); | |
689 | ||
690 | prealloc = NULL; | |
691 | if (err) | |
692 | goto out; | |
693 | if (state->end <= end) { | |
694 | state = clear_state_bit(tree, state, &bits, wake, | |
695 | changeset); | |
696 | goto next; | |
697 | } | |
698 | goto search_again; | |
699 | } | |
700 | /* | |
701 | * | ---- desired range ---- | | |
702 | * | state | | |
703 | * We need to split the extent, and clear the bit | |
704 | * on the first half | |
705 | */ | |
706 | if (state->start <= end && state->end > end) { | |
707 | prealloc = alloc_extent_state_atomic(prealloc); | |
708 | BUG_ON(!prealloc); | |
709 | err = split_state(tree, state, prealloc, end + 1); | |
710 | if (err) | |
711 | extent_io_tree_panic(tree, err); | |
712 | ||
713 | if (wake) | |
714 | wake_up(&state->wq); | |
715 | ||
716 | clear_state_bit(tree, prealloc, &bits, wake, changeset); | |
717 | ||
718 | prealloc = NULL; | |
719 | goto out; | |
720 | } | |
721 | ||
722 | state = clear_state_bit(tree, state, &bits, wake, changeset); | |
723 | next: | |
724 | if (last_end == (u64)-1) | |
725 | goto out; | |
726 | start = last_end + 1; | |
727 | if (start <= end && state && !need_resched()) | |
728 | goto hit_next; | |
729 | goto search_again; | |
730 | ||
731 | out: | |
732 | spin_unlock(&tree->lock); | |
733 | if (prealloc) | |
734 | free_extent_state(prealloc); | |
735 | ||
736 | return 0; | |
737 | ||
738 | search_again: | |
739 | if (start > end) | |
740 | goto out; | |
741 | spin_unlock(&tree->lock); | |
742 | if (gfpflags_allow_blocking(mask)) | |
743 | cond_resched(); | |
744 | goto again; | |
745 | } | |
746 | ||
747 | static void wait_on_state(struct extent_io_tree *tree, | |
748 | struct extent_state *state) | |
749 | __releases(tree->lock) | |
750 | __acquires(tree->lock) | |
751 | { | |
752 | DEFINE_WAIT(wait); | |
753 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
754 | spin_unlock(&tree->lock); | |
755 | schedule(); | |
756 | spin_lock(&tree->lock); | |
757 | finish_wait(&state->wq, &wait); | |
758 | } | |
759 | ||
760 | /* | |
761 | * waits for one or more bits to clear on a range in the state tree. | |
762 | * The range [start, end] is inclusive. | |
763 | * The tree lock is taken by this function | |
764 | */ | |
765 | static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
766 | unsigned long bits) | |
767 | { | |
768 | struct extent_state *state; | |
769 | struct rb_node *node; | |
770 | ||
771 | btrfs_debug_check_extent_io_range(tree, start, end); | |
772 | ||
773 | spin_lock(&tree->lock); | |
774 | again: | |
775 | while (1) { | |
776 | /* | |
777 | * this search will find all the extents that end after | |
778 | * our range starts | |
779 | */ | |
780 | node = tree_search(tree, start); | |
781 | process_node: | |
782 | if (!node) | |
783 | break; | |
784 | ||
785 | state = rb_entry(node, struct extent_state, rb_node); | |
786 | ||
787 | if (state->start > end) | |
788 | goto out; | |
789 | ||
790 | if (state->state & bits) { | |
791 | start = state->start; | |
792 | atomic_inc(&state->refs); | |
793 | wait_on_state(tree, state); | |
794 | free_extent_state(state); | |
795 | goto again; | |
796 | } | |
797 | start = state->end + 1; | |
798 | ||
799 | if (start > end) | |
800 | break; | |
801 | ||
802 | if (!cond_resched_lock(&tree->lock)) { | |
803 | node = rb_next(node); | |
804 | goto process_node; | |
805 | } | |
806 | } | |
807 | out: | |
808 | spin_unlock(&tree->lock); | |
809 | } | |
810 | ||
811 | static void set_state_bits(struct extent_io_tree *tree, | |
812 | struct extent_state *state, | |
813 | unsigned *bits, struct extent_changeset *changeset) | |
814 | { | |
815 | unsigned bits_to_set = *bits & ~EXTENT_CTLBITS; | |
816 | ||
817 | set_state_cb(tree, state, bits); | |
818 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { | |
819 | u64 range = state->end - state->start + 1; | |
820 | tree->dirty_bytes += range; | |
821 | } | |
822 | add_extent_changeset(state, bits_to_set, changeset, 1); | |
823 | state->state |= bits_to_set; | |
824 | } | |
825 | ||
826 | static void cache_state_if_flags(struct extent_state *state, | |
827 | struct extent_state **cached_ptr, | |
828 | unsigned flags) | |
829 | { | |
830 | if (cached_ptr && !(*cached_ptr)) { | |
831 | if (!flags || (state->state & flags)) { | |
832 | *cached_ptr = state; | |
833 | atomic_inc(&state->refs); | |
834 | } | |
835 | } | |
836 | } | |
837 | ||
838 | static void cache_state(struct extent_state *state, | |
839 | struct extent_state **cached_ptr) | |
840 | { | |
841 | return cache_state_if_flags(state, cached_ptr, | |
842 | EXTENT_IOBITS | EXTENT_BOUNDARY); | |
843 | } | |
844 | ||
845 | /* | |
846 | * set some bits on a range in the tree. This may require allocations or | |
847 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
848 | * | |
849 | * If any of the exclusive bits are set, this will fail with -EEXIST if some | |
850 | * part of the range already has the desired bits set. The start of the | |
851 | * existing range is returned in failed_start in this case. | |
852 | * | |
853 | * [start, end] is inclusive This takes the tree lock. | |
854 | */ | |
855 | ||
856 | static int __must_check | |
857 | __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
858 | unsigned bits, unsigned exclusive_bits, | |
859 | u64 *failed_start, struct extent_state **cached_state, | |
860 | gfp_t mask, struct extent_changeset *changeset) | |
861 | { | |
862 | struct extent_state *state; | |
863 | struct extent_state *prealloc = NULL; | |
864 | struct rb_node *node; | |
865 | struct rb_node **p; | |
866 | struct rb_node *parent; | |
867 | int err = 0; | |
868 | u64 last_start; | |
869 | u64 last_end; | |
870 | ||
871 | btrfs_debug_check_extent_io_range(tree, start, end); | |
872 | ||
873 | bits |= EXTENT_FIRST_DELALLOC; | |
874 | again: | |
875 | if (!prealloc && gfpflags_allow_blocking(mask)) { | |
876 | prealloc = alloc_extent_state(mask); | |
877 | BUG_ON(!prealloc); | |
878 | } | |
879 | ||
880 | spin_lock(&tree->lock); | |
881 | if (cached_state && *cached_state) { | |
882 | state = *cached_state; | |
883 | if (state->start <= start && state->end > start && | |
884 | extent_state_in_tree(state)) { | |
885 | node = &state->rb_node; | |
886 | goto hit_next; | |
887 | } | |
888 | } | |
889 | /* | |
890 | * this search will find all the extents that end after | |
891 | * our range starts. | |
892 | */ | |
893 | node = tree_search_for_insert(tree, start, &p, &parent); | |
894 | if (!node) { | |
895 | prealloc = alloc_extent_state_atomic(prealloc); | |
896 | BUG_ON(!prealloc); | |
897 | err = insert_state(tree, prealloc, start, end, | |
898 | &p, &parent, &bits, changeset); | |
899 | if (err) | |
900 | extent_io_tree_panic(tree, err); | |
901 | ||
902 | cache_state(prealloc, cached_state); | |
903 | prealloc = NULL; | |
904 | goto out; | |
905 | } | |
906 | state = rb_entry(node, struct extent_state, rb_node); | |
907 | hit_next: | |
908 | last_start = state->start; | |
909 | last_end = state->end; | |
910 | ||
911 | /* | |
912 | * | ---- desired range ---- | | |
913 | * | state | | |
914 | * | |
915 | * Just lock what we found and keep going | |
916 | */ | |
917 | if (state->start == start && state->end <= end) { | |
918 | if (state->state & exclusive_bits) { | |
919 | *failed_start = state->start; | |
920 | err = -EEXIST; | |
921 | goto out; | |
922 | } | |
923 | ||
924 | set_state_bits(tree, state, &bits, changeset); | |
925 | cache_state(state, cached_state); | |
926 | merge_state(tree, state); | |
927 | if (last_end == (u64)-1) | |
928 | goto out; | |
929 | start = last_end + 1; | |
930 | state = next_state(state); | |
931 | if (start < end && state && state->start == start && | |
932 | !need_resched()) | |
933 | goto hit_next; | |
934 | goto search_again; | |
935 | } | |
936 | ||
937 | /* | |
938 | * | ---- desired range ---- | | |
939 | * | state | | |
940 | * or | |
941 | * | ------------- state -------------- | | |
942 | * | |
943 | * We need to split the extent we found, and may flip bits on | |
944 | * second half. | |
945 | * | |
946 | * If the extent we found extends past our | |
947 | * range, we just split and search again. It'll get split | |
948 | * again the next time though. | |
949 | * | |
950 | * If the extent we found is inside our range, we set the | |
951 | * desired bit on it. | |
952 | */ | |
953 | if (state->start < start) { | |
954 | if (state->state & exclusive_bits) { | |
955 | *failed_start = start; | |
956 | err = -EEXIST; | |
957 | goto out; | |
958 | } | |
959 | ||
960 | prealloc = alloc_extent_state_atomic(prealloc); | |
961 | BUG_ON(!prealloc); | |
962 | err = split_state(tree, state, prealloc, start); | |
963 | if (err) | |
964 | extent_io_tree_panic(tree, err); | |
965 | ||
966 | prealloc = NULL; | |
967 | if (err) | |
968 | goto out; | |
969 | if (state->end <= end) { | |
970 | set_state_bits(tree, state, &bits, changeset); | |
971 | cache_state(state, cached_state); | |
972 | merge_state(tree, state); | |
973 | if (last_end == (u64)-1) | |
974 | goto out; | |
975 | start = last_end + 1; | |
976 | state = next_state(state); | |
977 | if (start < end && state && state->start == start && | |
978 | !need_resched()) | |
979 | goto hit_next; | |
980 | } | |
981 | goto search_again; | |
982 | } | |
983 | /* | |
984 | * | ---- desired range ---- | | |
985 | * | state | or | state | | |
986 | * | |
987 | * There's a hole, we need to insert something in it and | |
988 | * ignore the extent we found. | |
989 | */ | |
990 | if (state->start > start) { | |
991 | u64 this_end; | |
992 | if (end < last_start) | |
993 | this_end = end; | |
994 | else | |
995 | this_end = last_start - 1; | |
996 | ||
997 | prealloc = alloc_extent_state_atomic(prealloc); | |
998 | BUG_ON(!prealloc); | |
999 | ||
1000 | /* | |
1001 | * Avoid to free 'prealloc' if it can be merged with | |
1002 | * the later extent. | |
1003 | */ | |
1004 | err = insert_state(tree, prealloc, start, this_end, | |
1005 | NULL, NULL, &bits, changeset); | |
1006 | if (err) | |
1007 | extent_io_tree_panic(tree, err); | |
1008 | ||
1009 | cache_state(prealloc, cached_state); | |
1010 | prealloc = NULL; | |
1011 | start = this_end + 1; | |
1012 | goto search_again; | |
1013 | } | |
1014 | /* | |
1015 | * | ---- desired range ---- | | |
1016 | * | state | | |
1017 | * We need to split the extent, and set the bit | |
1018 | * on the first half | |
1019 | */ | |
1020 | if (state->start <= end && state->end > end) { | |
1021 | if (state->state & exclusive_bits) { | |
1022 | *failed_start = start; | |
1023 | err = -EEXIST; | |
1024 | goto out; | |
1025 | } | |
1026 | ||
1027 | prealloc = alloc_extent_state_atomic(prealloc); | |
1028 | BUG_ON(!prealloc); | |
1029 | err = split_state(tree, state, prealloc, end + 1); | |
1030 | if (err) | |
1031 | extent_io_tree_panic(tree, err); | |
1032 | ||
1033 | set_state_bits(tree, prealloc, &bits, changeset); | |
1034 | cache_state(prealloc, cached_state); | |
1035 | merge_state(tree, prealloc); | |
1036 | prealloc = NULL; | |
1037 | goto out; | |
1038 | } | |
1039 | ||
1040 | goto search_again; | |
1041 | ||
1042 | out: | |
1043 | spin_unlock(&tree->lock); | |
1044 | if (prealloc) | |
1045 | free_extent_state(prealloc); | |
1046 | ||
1047 | return err; | |
1048 | ||
1049 | search_again: | |
1050 | if (start > end) | |
1051 | goto out; | |
1052 | spin_unlock(&tree->lock); | |
1053 | if (gfpflags_allow_blocking(mask)) | |
1054 | cond_resched(); | |
1055 | goto again; | |
1056 | } | |
1057 | ||
1058 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
1059 | unsigned bits, u64 * failed_start, | |
1060 | struct extent_state **cached_state, gfp_t mask) | |
1061 | { | |
1062 | return __set_extent_bit(tree, start, end, bits, 0, failed_start, | |
1063 | cached_state, mask, NULL); | |
1064 | } | |
1065 | ||
1066 | ||
1067 | /** | |
1068 | * convert_extent_bit - convert all bits in a given range from one bit to | |
1069 | * another | |
1070 | * @tree: the io tree to search | |
1071 | * @start: the start offset in bytes | |
1072 | * @end: the end offset in bytes (inclusive) | |
1073 | * @bits: the bits to set in this range | |
1074 | * @clear_bits: the bits to clear in this range | |
1075 | * @cached_state: state that we're going to cache | |
1076 | * @mask: the allocation mask | |
1077 | * | |
1078 | * This will go through and set bits for the given range. If any states exist | |
1079 | * already in this range they are set with the given bit and cleared of the | |
1080 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
1081 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
1082 | * boundary bits like LOCK. | |
1083 | */ | |
1084 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
1085 | unsigned bits, unsigned clear_bits, | |
1086 | struct extent_state **cached_state, gfp_t mask) | |
1087 | { | |
1088 | struct extent_state *state; | |
1089 | struct extent_state *prealloc = NULL; | |
1090 | struct rb_node *node; | |
1091 | struct rb_node **p; | |
1092 | struct rb_node *parent; | |
1093 | int err = 0; | |
1094 | u64 last_start; | |
1095 | u64 last_end; | |
1096 | bool first_iteration = true; | |
1097 | ||
1098 | btrfs_debug_check_extent_io_range(tree, start, end); | |
1099 | ||
1100 | again: | |
1101 | if (!prealloc && gfpflags_allow_blocking(mask)) { | |
1102 | /* | |
1103 | * Best effort, don't worry if extent state allocation fails | |
1104 | * here for the first iteration. We might have a cached state | |
1105 | * that matches exactly the target range, in which case no | |
1106 | * extent state allocations are needed. We'll only know this | |
1107 | * after locking the tree. | |
1108 | */ | |
1109 | prealloc = alloc_extent_state(mask); | |
1110 | if (!prealloc && !first_iteration) | |
1111 | return -ENOMEM; | |
1112 | } | |
1113 | ||
1114 | spin_lock(&tree->lock); | |
1115 | if (cached_state && *cached_state) { | |
1116 | state = *cached_state; | |
1117 | if (state->start <= start && state->end > start && | |
1118 | extent_state_in_tree(state)) { | |
1119 | node = &state->rb_node; | |
1120 | goto hit_next; | |
1121 | } | |
1122 | } | |
1123 | ||
1124 | /* | |
1125 | * this search will find all the extents that end after | |
1126 | * our range starts. | |
1127 | */ | |
1128 | node = tree_search_for_insert(tree, start, &p, &parent); | |
1129 | if (!node) { | |
1130 | prealloc = alloc_extent_state_atomic(prealloc); | |
1131 | if (!prealloc) { | |
1132 | err = -ENOMEM; | |
1133 | goto out; | |
1134 | } | |
1135 | err = insert_state(tree, prealloc, start, end, | |
1136 | &p, &parent, &bits, NULL); | |
1137 | if (err) | |
1138 | extent_io_tree_panic(tree, err); | |
1139 | cache_state(prealloc, cached_state); | |
1140 | prealloc = NULL; | |
1141 | goto out; | |
1142 | } | |
1143 | state = rb_entry(node, struct extent_state, rb_node); | |
1144 | hit_next: | |
1145 | last_start = state->start; | |
1146 | last_end = state->end; | |
1147 | ||
1148 | /* | |
1149 | * | ---- desired range ---- | | |
1150 | * | state | | |
1151 | * | |
1152 | * Just lock what we found and keep going | |
1153 | */ | |
1154 | if (state->start == start && state->end <= end) { | |
1155 | set_state_bits(tree, state, &bits, NULL); | |
1156 | cache_state(state, cached_state); | |
1157 | state = clear_state_bit(tree, state, &clear_bits, 0, NULL); | |
1158 | if (last_end == (u64)-1) | |
1159 | goto out; | |
1160 | start = last_end + 1; | |
1161 | if (start < end && state && state->start == start && | |
1162 | !need_resched()) | |
1163 | goto hit_next; | |
1164 | goto search_again; | |
1165 | } | |
1166 | ||
1167 | /* | |
1168 | * | ---- desired range ---- | | |
1169 | * | state | | |
1170 | * or | |
1171 | * | ------------- state -------------- | | |
1172 | * | |
1173 | * We need to split the extent we found, and may flip bits on | |
1174 | * second half. | |
1175 | * | |
1176 | * If the extent we found extends past our | |
1177 | * range, we just split and search again. It'll get split | |
1178 | * again the next time though. | |
1179 | * | |
1180 | * If the extent we found is inside our range, we set the | |
1181 | * desired bit on it. | |
1182 | */ | |
1183 | if (state->start < start) { | |
1184 | prealloc = alloc_extent_state_atomic(prealloc); | |
1185 | if (!prealloc) { | |
1186 | err = -ENOMEM; | |
1187 | goto out; | |
1188 | } | |
1189 | err = split_state(tree, state, prealloc, start); | |
1190 | if (err) | |
1191 | extent_io_tree_panic(tree, err); | |
1192 | prealloc = NULL; | |
1193 | if (err) | |
1194 | goto out; | |
1195 | if (state->end <= end) { | |
1196 | set_state_bits(tree, state, &bits, NULL); | |
1197 | cache_state(state, cached_state); | |
1198 | state = clear_state_bit(tree, state, &clear_bits, 0, | |
1199 | NULL); | |
1200 | if (last_end == (u64)-1) | |
1201 | goto out; | |
1202 | start = last_end + 1; | |
1203 | if (start < end && state && state->start == start && | |
1204 | !need_resched()) | |
1205 | goto hit_next; | |
1206 | } | |
1207 | goto search_again; | |
1208 | } | |
1209 | /* | |
1210 | * | ---- desired range ---- | | |
1211 | * | state | or | state | | |
1212 | * | |
1213 | * There's a hole, we need to insert something in it and | |
1214 | * ignore the extent we found. | |
1215 | */ | |
1216 | if (state->start > start) { | |
1217 | u64 this_end; | |
1218 | if (end < last_start) | |
1219 | this_end = end; | |
1220 | else | |
1221 | this_end = last_start - 1; | |
1222 | ||
1223 | prealloc = alloc_extent_state_atomic(prealloc); | |
1224 | if (!prealloc) { | |
1225 | err = -ENOMEM; | |
1226 | goto out; | |
1227 | } | |
1228 | ||
1229 | /* | |
1230 | * Avoid to free 'prealloc' if it can be merged with | |
1231 | * the later extent. | |
1232 | */ | |
1233 | err = insert_state(tree, prealloc, start, this_end, | |
1234 | NULL, NULL, &bits, NULL); | |
1235 | if (err) | |
1236 | extent_io_tree_panic(tree, err); | |
1237 | cache_state(prealloc, cached_state); | |
1238 | prealloc = NULL; | |
1239 | start = this_end + 1; | |
1240 | goto search_again; | |
1241 | } | |
1242 | /* | |
1243 | * | ---- desired range ---- | | |
1244 | * | state | | |
1245 | * We need to split the extent, and set the bit | |
1246 | * on the first half | |
1247 | */ | |
1248 | if (state->start <= end && state->end > end) { | |
1249 | prealloc = alloc_extent_state_atomic(prealloc); | |
1250 | if (!prealloc) { | |
1251 | err = -ENOMEM; | |
1252 | goto out; | |
1253 | } | |
1254 | ||
1255 | err = split_state(tree, state, prealloc, end + 1); | |
1256 | if (err) | |
1257 | extent_io_tree_panic(tree, err); | |
1258 | ||
1259 | set_state_bits(tree, prealloc, &bits, NULL); | |
1260 | cache_state(prealloc, cached_state); | |
1261 | clear_state_bit(tree, prealloc, &clear_bits, 0, NULL); | |
1262 | prealloc = NULL; | |
1263 | goto out; | |
1264 | } | |
1265 | ||
1266 | goto search_again; | |
1267 | ||
1268 | out: | |
1269 | spin_unlock(&tree->lock); | |
1270 | if (prealloc) | |
1271 | free_extent_state(prealloc); | |
1272 | ||
1273 | return err; | |
1274 | ||
1275 | search_again: | |
1276 | if (start > end) | |
1277 | goto out; | |
1278 | spin_unlock(&tree->lock); | |
1279 | if (gfpflags_allow_blocking(mask)) | |
1280 | cond_resched(); | |
1281 | first_iteration = false; | |
1282 | goto again; | |
1283 | } | |
1284 | ||
1285 | /* wrappers around set/clear extent bit */ | |
1286 | int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
1287 | unsigned bits, gfp_t mask, | |
1288 | struct extent_changeset *changeset) | |
1289 | { | |
1290 | /* | |
1291 | * We don't support EXTENT_LOCKED yet, as current changeset will | |
1292 | * record any bits changed, so for EXTENT_LOCKED case, it will | |
1293 | * either fail with -EEXIST or changeset will record the whole | |
1294 | * range. | |
1295 | */ | |
1296 | BUG_ON(bits & EXTENT_LOCKED); | |
1297 | ||
1298 | return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, mask, | |
1299 | changeset); | |
1300 | } | |
1301 | ||
1302 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
1303 | unsigned bits, int wake, int delete, | |
1304 | struct extent_state **cached, gfp_t mask) | |
1305 | { | |
1306 | return __clear_extent_bit(tree, start, end, bits, wake, delete, | |
1307 | cached, mask, NULL); | |
1308 | } | |
1309 | ||
1310 | int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
1311 | unsigned bits, gfp_t mask, | |
1312 | struct extent_changeset *changeset) | |
1313 | { | |
1314 | /* | |
1315 | * Don't support EXTENT_LOCKED case, same reason as | |
1316 | * set_record_extent_bits(). | |
1317 | */ | |
1318 | BUG_ON(bits & EXTENT_LOCKED); | |
1319 | ||
1320 | return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask, | |
1321 | changeset); | |
1322 | } | |
1323 | ||
1324 | /* | |
1325 | * either insert or lock state struct between start and end use mask to tell | |
1326 | * us if waiting is desired. | |
1327 | */ | |
1328 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
1329 | struct extent_state **cached_state) | |
1330 | { | |
1331 | int err; | |
1332 | u64 failed_start; | |
1333 | ||
1334 | while (1) { | |
1335 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, | |
1336 | EXTENT_LOCKED, &failed_start, | |
1337 | cached_state, GFP_NOFS, NULL); | |
1338 | if (err == -EEXIST) { | |
1339 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); | |
1340 | start = failed_start; | |
1341 | } else | |
1342 | break; | |
1343 | WARN_ON(start > end); | |
1344 | } | |
1345 | return err; | |
1346 | } | |
1347 | ||
1348 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) | |
1349 | { | |
1350 | int err; | |
1351 | u64 failed_start; | |
1352 | ||
1353 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, | |
1354 | &failed_start, NULL, GFP_NOFS, NULL); | |
1355 | if (err == -EEXIST) { | |
1356 | if (failed_start > start) | |
1357 | clear_extent_bit(tree, start, failed_start - 1, | |
1358 | EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS); | |
1359 | return 0; | |
1360 | } | |
1361 | return 1; | |
1362 | } | |
1363 | ||
1364 | void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) | |
1365 | { | |
1366 | unsigned long index = start >> PAGE_SHIFT; | |
1367 | unsigned long end_index = end >> PAGE_SHIFT; | |
1368 | struct page *page; | |
1369 | ||
1370 | while (index <= end_index) { | |
1371 | page = find_get_page(inode->i_mapping, index); | |
1372 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1373 | clear_page_dirty_for_io(page); | |
1374 | put_page(page); | |
1375 | index++; | |
1376 | } | |
1377 | } | |
1378 | ||
1379 | void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) | |
1380 | { | |
1381 | unsigned long index = start >> PAGE_SHIFT; | |
1382 | unsigned long end_index = end >> PAGE_SHIFT; | |
1383 | struct page *page; | |
1384 | ||
1385 | while (index <= end_index) { | |
1386 | page = find_get_page(inode->i_mapping, index); | |
1387 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1388 | __set_page_dirty_nobuffers(page); | |
1389 | account_page_redirty(page); | |
1390 | put_page(page); | |
1391 | index++; | |
1392 | } | |
1393 | } | |
1394 | ||
1395 | /* | |
1396 | * helper function to set both pages and extents in the tree writeback | |
1397 | */ | |
1398 | static void set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) | |
1399 | { | |
1400 | unsigned long index = start >> PAGE_SHIFT; | |
1401 | unsigned long end_index = end >> PAGE_SHIFT; | |
1402 | struct page *page; | |
1403 | ||
1404 | while (index <= end_index) { | |
1405 | page = find_get_page(tree->mapping, index); | |
1406 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1407 | set_page_writeback(page); | |
1408 | put_page(page); | |
1409 | index++; | |
1410 | } | |
1411 | } | |
1412 | ||
1413 | /* find the first state struct with 'bits' set after 'start', and | |
1414 | * return it. tree->lock must be held. NULL will returned if | |
1415 | * nothing was found after 'start' | |
1416 | */ | |
1417 | static struct extent_state * | |
1418 | find_first_extent_bit_state(struct extent_io_tree *tree, | |
1419 | u64 start, unsigned bits) | |
1420 | { | |
1421 | struct rb_node *node; | |
1422 | struct extent_state *state; | |
1423 | ||
1424 | /* | |
1425 | * this search will find all the extents that end after | |
1426 | * our range starts. | |
1427 | */ | |
1428 | node = tree_search(tree, start); | |
1429 | if (!node) | |
1430 | goto out; | |
1431 | ||
1432 | while (1) { | |
1433 | state = rb_entry(node, struct extent_state, rb_node); | |
1434 | if (state->end >= start && (state->state & bits)) | |
1435 | return state; | |
1436 | ||
1437 | node = rb_next(node); | |
1438 | if (!node) | |
1439 | break; | |
1440 | } | |
1441 | out: | |
1442 | return NULL; | |
1443 | } | |
1444 | ||
1445 | /* | |
1446 | * find the first offset in the io tree with 'bits' set. zero is | |
1447 | * returned if we find something, and *start_ret and *end_ret are | |
1448 | * set to reflect the state struct that was found. | |
1449 | * | |
1450 | * If nothing was found, 1 is returned. If found something, return 0. | |
1451 | */ | |
1452 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
1453 | u64 *start_ret, u64 *end_ret, unsigned bits, | |
1454 | struct extent_state **cached_state) | |
1455 | { | |
1456 | struct extent_state *state; | |
1457 | struct rb_node *n; | |
1458 | int ret = 1; | |
1459 | ||
1460 | spin_lock(&tree->lock); | |
1461 | if (cached_state && *cached_state) { | |
1462 | state = *cached_state; | |
1463 | if (state->end == start - 1 && extent_state_in_tree(state)) { | |
1464 | n = rb_next(&state->rb_node); | |
1465 | while (n) { | |
1466 | state = rb_entry(n, struct extent_state, | |
1467 | rb_node); | |
1468 | if (state->state & bits) | |
1469 | goto got_it; | |
1470 | n = rb_next(n); | |
1471 | } | |
1472 | free_extent_state(*cached_state); | |
1473 | *cached_state = NULL; | |
1474 | goto out; | |
1475 | } | |
1476 | free_extent_state(*cached_state); | |
1477 | *cached_state = NULL; | |
1478 | } | |
1479 | ||
1480 | state = find_first_extent_bit_state(tree, start, bits); | |
1481 | got_it: | |
1482 | if (state) { | |
1483 | cache_state_if_flags(state, cached_state, 0); | |
1484 | *start_ret = state->start; | |
1485 | *end_ret = state->end; | |
1486 | ret = 0; | |
1487 | } | |
1488 | out: | |
1489 | spin_unlock(&tree->lock); | |
1490 | return ret; | |
1491 | } | |
1492 | ||
1493 | /* | |
1494 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1495 | * more than 'max_bytes'. start and end are used to return the range, | |
1496 | * | |
1497 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1498 | */ | |
1499 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, | |
1500 | u64 *start, u64 *end, u64 max_bytes, | |
1501 | struct extent_state **cached_state) | |
1502 | { | |
1503 | struct rb_node *node; | |
1504 | struct extent_state *state; | |
1505 | u64 cur_start = *start; | |
1506 | u64 found = 0; | |
1507 | u64 total_bytes = 0; | |
1508 | ||
1509 | spin_lock(&tree->lock); | |
1510 | ||
1511 | /* | |
1512 | * this search will find all the extents that end after | |
1513 | * our range starts. | |
1514 | */ | |
1515 | node = tree_search(tree, cur_start); | |
1516 | if (!node) { | |
1517 | if (!found) | |
1518 | *end = (u64)-1; | |
1519 | goto out; | |
1520 | } | |
1521 | ||
1522 | while (1) { | |
1523 | state = rb_entry(node, struct extent_state, rb_node); | |
1524 | if (found && (state->start != cur_start || | |
1525 | (state->state & EXTENT_BOUNDARY))) { | |
1526 | goto out; | |
1527 | } | |
1528 | if (!(state->state & EXTENT_DELALLOC)) { | |
1529 | if (!found) | |
1530 | *end = state->end; | |
1531 | goto out; | |
1532 | } | |
1533 | if (!found) { | |
1534 | *start = state->start; | |
1535 | *cached_state = state; | |
1536 | atomic_inc(&state->refs); | |
1537 | } | |
1538 | found++; | |
1539 | *end = state->end; | |
1540 | cur_start = state->end + 1; | |
1541 | node = rb_next(node); | |
1542 | total_bytes += state->end - state->start + 1; | |
1543 | if (total_bytes >= max_bytes) | |
1544 | break; | |
1545 | if (!node) | |
1546 | break; | |
1547 | } | |
1548 | out: | |
1549 | spin_unlock(&tree->lock); | |
1550 | return found; | |
1551 | } | |
1552 | ||
1553 | static noinline void __unlock_for_delalloc(struct inode *inode, | |
1554 | struct page *locked_page, | |
1555 | u64 start, u64 end) | |
1556 | { | |
1557 | int ret; | |
1558 | struct page *pages[16]; | |
1559 | unsigned long index = start >> PAGE_SHIFT; | |
1560 | unsigned long end_index = end >> PAGE_SHIFT; | |
1561 | unsigned long nr_pages = end_index - index + 1; | |
1562 | int i; | |
1563 | ||
1564 | if (index == locked_page->index && end_index == index) | |
1565 | return; | |
1566 | ||
1567 | while (nr_pages > 0) { | |
1568 | ret = find_get_pages_contig(inode->i_mapping, index, | |
1569 | min_t(unsigned long, nr_pages, | |
1570 | ARRAY_SIZE(pages)), pages); | |
1571 | for (i = 0; i < ret; i++) { | |
1572 | if (pages[i] != locked_page) | |
1573 | unlock_page(pages[i]); | |
1574 | put_page(pages[i]); | |
1575 | } | |
1576 | nr_pages -= ret; | |
1577 | index += ret; | |
1578 | cond_resched(); | |
1579 | } | |
1580 | } | |
1581 | ||
1582 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1583 | struct page *locked_page, | |
1584 | u64 delalloc_start, | |
1585 | u64 delalloc_end) | |
1586 | { | |
1587 | unsigned long index = delalloc_start >> PAGE_SHIFT; | |
1588 | unsigned long start_index = index; | |
1589 | unsigned long end_index = delalloc_end >> PAGE_SHIFT; | |
1590 | unsigned long pages_locked = 0; | |
1591 | struct page *pages[16]; | |
1592 | unsigned long nrpages; | |
1593 | int ret; | |
1594 | int i; | |
1595 | ||
1596 | /* the caller is responsible for locking the start index */ | |
1597 | if (index == locked_page->index && index == end_index) | |
1598 | return 0; | |
1599 | ||
1600 | /* skip the page at the start index */ | |
1601 | nrpages = end_index - index + 1; | |
1602 | while (nrpages > 0) { | |
1603 | ret = find_get_pages_contig(inode->i_mapping, index, | |
1604 | min_t(unsigned long, | |
1605 | nrpages, ARRAY_SIZE(pages)), pages); | |
1606 | if (ret == 0) { | |
1607 | ret = -EAGAIN; | |
1608 | goto done; | |
1609 | } | |
1610 | /* now we have an array of pages, lock them all */ | |
1611 | for (i = 0; i < ret; i++) { | |
1612 | /* | |
1613 | * the caller is taking responsibility for | |
1614 | * locked_page | |
1615 | */ | |
1616 | if (pages[i] != locked_page) { | |
1617 | lock_page(pages[i]); | |
1618 | if (!PageDirty(pages[i]) || | |
1619 | pages[i]->mapping != inode->i_mapping) { | |
1620 | ret = -EAGAIN; | |
1621 | unlock_page(pages[i]); | |
1622 | put_page(pages[i]); | |
1623 | goto done; | |
1624 | } | |
1625 | } | |
1626 | put_page(pages[i]); | |
1627 | pages_locked++; | |
1628 | } | |
1629 | nrpages -= ret; | |
1630 | index += ret; | |
1631 | cond_resched(); | |
1632 | } | |
1633 | ret = 0; | |
1634 | done: | |
1635 | if (ret && pages_locked) { | |
1636 | __unlock_for_delalloc(inode, locked_page, | |
1637 | delalloc_start, | |
1638 | ((u64)(start_index + pages_locked - 1)) << | |
1639 | PAGE_SHIFT); | |
1640 | } | |
1641 | return ret; | |
1642 | } | |
1643 | ||
1644 | /* | |
1645 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1646 | * more than 'max_bytes'. start and end are used to return the range, | |
1647 | * | |
1648 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1649 | */ | |
1650 | STATIC u64 find_lock_delalloc_range(struct inode *inode, | |
1651 | struct extent_io_tree *tree, | |
1652 | struct page *locked_page, u64 *start, | |
1653 | u64 *end, u64 max_bytes) | |
1654 | { | |
1655 | u64 delalloc_start; | |
1656 | u64 delalloc_end; | |
1657 | u64 found; | |
1658 | struct extent_state *cached_state = NULL; | |
1659 | int ret; | |
1660 | int loops = 0; | |
1661 | ||
1662 | again: | |
1663 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1664 | delalloc_start = *start; | |
1665 | delalloc_end = 0; | |
1666 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
1667 | max_bytes, &cached_state); | |
1668 | if (!found || delalloc_end <= *start) { | |
1669 | *start = delalloc_start; | |
1670 | *end = delalloc_end; | |
1671 | free_extent_state(cached_state); | |
1672 | return 0; | |
1673 | } | |
1674 | ||
1675 | /* | |
1676 | * start comes from the offset of locked_page. We have to lock | |
1677 | * pages in order, so we can't process delalloc bytes before | |
1678 | * locked_page | |
1679 | */ | |
1680 | if (delalloc_start < *start) | |
1681 | delalloc_start = *start; | |
1682 | ||
1683 | /* | |
1684 | * make sure to limit the number of pages we try to lock down | |
1685 | */ | |
1686 | if (delalloc_end + 1 - delalloc_start > max_bytes) | |
1687 | delalloc_end = delalloc_start + max_bytes - 1; | |
1688 | ||
1689 | /* step two, lock all the pages after the page that has start */ | |
1690 | ret = lock_delalloc_pages(inode, locked_page, | |
1691 | delalloc_start, delalloc_end); | |
1692 | if (ret == -EAGAIN) { | |
1693 | /* some of the pages are gone, lets avoid looping by | |
1694 | * shortening the size of the delalloc range we're searching | |
1695 | */ | |
1696 | free_extent_state(cached_state); | |
1697 | cached_state = NULL; | |
1698 | if (!loops) { | |
1699 | max_bytes = PAGE_SIZE; | |
1700 | loops = 1; | |
1701 | goto again; | |
1702 | } else { | |
1703 | found = 0; | |
1704 | goto out_failed; | |
1705 | } | |
1706 | } | |
1707 | BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */ | |
1708 | ||
1709 | /* step three, lock the state bits for the whole range */ | |
1710 | lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state); | |
1711 | ||
1712 | /* then test to make sure it is all still delalloc */ | |
1713 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
1714 | EXTENT_DELALLOC, 1, cached_state); | |
1715 | if (!ret) { | |
1716 | unlock_extent_cached(tree, delalloc_start, delalloc_end, | |
1717 | &cached_state, GFP_NOFS); | |
1718 | __unlock_for_delalloc(inode, locked_page, | |
1719 | delalloc_start, delalloc_end); | |
1720 | cond_resched(); | |
1721 | goto again; | |
1722 | } | |
1723 | free_extent_state(cached_state); | |
1724 | *start = delalloc_start; | |
1725 | *end = delalloc_end; | |
1726 | out_failed: | |
1727 | return found; | |
1728 | } | |
1729 | ||
1730 | void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end, | |
1731 | struct page *locked_page, | |
1732 | unsigned clear_bits, | |
1733 | unsigned long page_ops) | |
1734 | { | |
1735 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
1736 | int ret; | |
1737 | struct page *pages[16]; | |
1738 | unsigned long index = start >> PAGE_SHIFT; | |
1739 | unsigned long end_index = end >> PAGE_SHIFT; | |
1740 | unsigned long nr_pages = end_index - index + 1; | |
1741 | int i; | |
1742 | ||
1743 | clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); | |
1744 | if (page_ops == 0) | |
1745 | return; | |
1746 | ||
1747 | if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0) | |
1748 | mapping_set_error(inode->i_mapping, -EIO); | |
1749 | ||
1750 | while (nr_pages > 0) { | |
1751 | ret = find_get_pages_contig(inode->i_mapping, index, | |
1752 | min_t(unsigned long, | |
1753 | nr_pages, ARRAY_SIZE(pages)), pages); | |
1754 | for (i = 0; i < ret; i++) { | |
1755 | ||
1756 | if (page_ops & PAGE_SET_PRIVATE2) | |
1757 | SetPagePrivate2(pages[i]); | |
1758 | ||
1759 | if (pages[i] == locked_page) { | |
1760 | put_page(pages[i]); | |
1761 | continue; | |
1762 | } | |
1763 | if (page_ops & PAGE_CLEAR_DIRTY) | |
1764 | clear_page_dirty_for_io(pages[i]); | |
1765 | if (page_ops & PAGE_SET_WRITEBACK) | |
1766 | set_page_writeback(pages[i]); | |
1767 | if (page_ops & PAGE_SET_ERROR) | |
1768 | SetPageError(pages[i]); | |
1769 | if (page_ops & PAGE_END_WRITEBACK) | |
1770 | end_page_writeback(pages[i]); | |
1771 | if (page_ops & PAGE_UNLOCK) | |
1772 | unlock_page(pages[i]); | |
1773 | put_page(pages[i]); | |
1774 | } | |
1775 | nr_pages -= ret; | |
1776 | index += ret; | |
1777 | cond_resched(); | |
1778 | } | |
1779 | } | |
1780 | ||
1781 | /* | |
1782 | * count the number of bytes in the tree that have a given bit(s) | |
1783 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1784 | * cached. The total number found is returned. | |
1785 | */ | |
1786 | u64 count_range_bits(struct extent_io_tree *tree, | |
1787 | u64 *start, u64 search_end, u64 max_bytes, | |
1788 | unsigned bits, int contig) | |
1789 | { | |
1790 | struct rb_node *node; | |
1791 | struct extent_state *state; | |
1792 | u64 cur_start = *start; | |
1793 | u64 total_bytes = 0; | |
1794 | u64 last = 0; | |
1795 | int found = 0; | |
1796 | ||
1797 | if (WARN_ON(search_end <= cur_start)) | |
1798 | return 0; | |
1799 | ||
1800 | spin_lock(&tree->lock); | |
1801 | if (cur_start == 0 && bits == EXTENT_DIRTY) { | |
1802 | total_bytes = tree->dirty_bytes; | |
1803 | goto out; | |
1804 | } | |
1805 | /* | |
1806 | * this search will find all the extents that end after | |
1807 | * our range starts. | |
1808 | */ | |
1809 | node = tree_search(tree, cur_start); | |
1810 | if (!node) | |
1811 | goto out; | |
1812 | ||
1813 | while (1) { | |
1814 | state = rb_entry(node, struct extent_state, rb_node); | |
1815 | if (state->start > search_end) | |
1816 | break; | |
1817 | if (contig && found && state->start > last + 1) | |
1818 | break; | |
1819 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
1820 | total_bytes += min(search_end, state->end) + 1 - | |
1821 | max(cur_start, state->start); | |
1822 | if (total_bytes >= max_bytes) | |
1823 | break; | |
1824 | if (!found) { | |
1825 | *start = max(cur_start, state->start); | |
1826 | found = 1; | |
1827 | } | |
1828 | last = state->end; | |
1829 | } else if (contig && found) { | |
1830 | break; | |
1831 | } | |
1832 | node = rb_next(node); | |
1833 | if (!node) | |
1834 | break; | |
1835 | } | |
1836 | out: | |
1837 | spin_unlock(&tree->lock); | |
1838 | return total_bytes; | |
1839 | } | |
1840 | ||
1841 | /* | |
1842 | * set the private field for a given byte offset in the tree. If there isn't | |
1843 | * an extent_state there already, this does nothing. | |
1844 | */ | |
1845 | static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start, | |
1846 | struct io_failure_record *failrec) | |
1847 | { | |
1848 | struct rb_node *node; | |
1849 | struct extent_state *state; | |
1850 | int ret = 0; | |
1851 | ||
1852 | spin_lock(&tree->lock); | |
1853 | /* | |
1854 | * this search will find all the extents that end after | |
1855 | * our range starts. | |
1856 | */ | |
1857 | node = tree_search(tree, start); | |
1858 | if (!node) { | |
1859 | ret = -ENOENT; | |
1860 | goto out; | |
1861 | } | |
1862 | state = rb_entry(node, struct extent_state, rb_node); | |
1863 | if (state->start != start) { | |
1864 | ret = -ENOENT; | |
1865 | goto out; | |
1866 | } | |
1867 | state->failrec = failrec; | |
1868 | out: | |
1869 | spin_unlock(&tree->lock); | |
1870 | return ret; | |
1871 | } | |
1872 | ||
1873 | static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start, | |
1874 | struct io_failure_record **failrec) | |
1875 | { | |
1876 | struct rb_node *node; | |
1877 | struct extent_state *state; | |
1878 | int ret = 0; | |
1879 | ||
1880 | spin_lock(&tree->lock); | |
1881 | /* | |
1882 | * this search will find all the extents that end after | |
1883 | * our range starts. | |
1884 | */ | |
1885 | node = tree_search(tree, start); | |
1886 | if (!node) { | |
1887 | ret = -ENOENT; | |
1888 | goto out; | |
1889 | } | |
1890 | state = rb_entry(node, struct extent_state, rb_node); | |
1891 | if (state->start != start) { | |
1892 | ret = -ENOENT; | |
1893 | goto out; | |
1894 | } | |
1895 | *failrec = state->failrec; | |
1896 | out: | |
1897 | spin_unlock(&tree->lock); | |
1898 | return ret; | |
1899 | } | |
1900 | ||
1901 | /* | |
1902 | * searches a range in the state tree for a given mask. | |
1903 | * If 'filled' == 1, this returns 1 only if every extent in the tree | |
1904 | * has the bits set. Otherwise, 1 is returned if any bit in the | |
1905 | * range is found set. | |
1906 | */ | |
1907 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
1908 | unsigned bits, int filled, struct extent_state *cached) | |
1909 | { | |
1910 | struct extent_state *state = NULL; | |
1911 | struct rb_node *node; | |
1912 | int bitset = 0; | |
1913 | ||
1914 | spin_lock(&tree->lock); | |
1915 | if (cached && extent_state_in_tree(cached) && cached->start <= start && | |
1916 | cached->end > start) | |
1917 | node = &cached->rb_node; | |
1918 | else | |
1919 | node = tree_search(tree, start); | |
1920 | while (node && start <= end) { | |
1921 | state = rb_entry(node, struct extent_state, rb_node); | |
1922 | ||
1923 | if (filled && state->start > start) { | |
1924 | bitset = 0; | |
1925 | break; | |
1926 | } | |
1927 | ||
1928 | if (state->start > end) | |
1929 | break; | |
1930 | ||
1931 | if (state->state & bits) { | |
1932 | bitset = 1; | |
1933 | if (!filled) | |
1934 | break; | |
1935 | } else if (filled) { | |
1936 | bitset = 0; | |
1937 | break; | |
1938 | } | |
1939 | ||
1940 | if (state->end == (u64)-1) | |
1941 | break; | |
1942 | ||
1943 | start = state->end + 1; | |
1944 | if (start > end) | |
1945 | break; | |
1946 | node = rb_next(node); | |
1947 | if (!node) { | |
1948 | if (filled) | |
1949 | bitset = 0; | |
1950 | break; | |
1951 | } | |
1952 | } | |
1953 | spin_unlock(&tree->lock); | |
1954 | return bitset; | |
1955 | } | |
1956 | ||
1957 | /* | |
1958 | * helper function to set a given page up to date if all the | |
1959 | * extents in the tree for that page are up to date | |
1960 | */ | |
1961 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) | |
1962 | { | |
1963 | u64 start = page_offset(page); | |
1964 | u64 end = start + PAGE_SIZE - 1; | |
1965 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) | |
1966 | SetPageUptodate(page); | |
1967 | } | |
1968 | ||
1969 | int free_io_failure(struct inode *inode, struct io_failure_record *rec) | |
1970 | { | |
1971 | int ret; | |
1972 | int err = 0; | |
1973 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
1974 | ||
1975 | set_state_failrec(failure_tree, rec->start, NULL); | |
1976 | ret = clear_extent_bits(failure_tree, rec->start, | |
1977 | rec->start + rec->len - 1, | |
1978 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1979 | if (ret) | |
1980 | err = ret; | |
1981 | ||
1982 | ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start, | |
1983 | rec->start + rec->len - 1, | |
1984 | EXTENT_DAMAGED, GFP_NOFS); | |
1985 | if (ret && !err) | |
1986 | err = ret; | |
1987 | ||
1988 | kfree(rec); | |
1989 | return err; | |
1990 | } | |
1991 | ||
1992 | /* | |
1993 | * this bypasses the standard btrfs submit functions deliberately, as | |
1994 | * the standard behavior is to write all copies in a raid setup. here we only | |
1995 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
1996 | * submit_bio directly. | |
1997 | * to avoid any synchronization issues, wait for the data after writing, which | |
1998 | * actually prevents the read that triggered the error from finishing. | |
1999 | * currently, there can be no more than two copies of every data bit. thus, | |
2000 | * exactly one rewrite is required. | |
2001 | */ | |
2002 | int repair_io_failure(struct inode *inode, u64 start, u64 length, u64 logical, | |
2003 | struct page *page, unsigned int pg_offset, int mirror_num) | |
2004 | { | |
2005 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2006 | struct bio *bio; | |
2007 | struct btrfs_device *dev; | |
2008 | u64 map_length = 0; | |
2009 | u64 sector; | |
2010 | struct btrfs_bio *bbio = NULL; | |
2011 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
2012 | int ret; | |
2013 | ||
2014 | ASSERT(!(fs_info->sb->s_flags & MS_RDONLY)); | |
2015 | BUG_ON(!mirror_num); | |
2016 | ||
2017 | /* we can't repair anything in raid56 yet */ | |
2018 | if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num)) | |
2019 | return 0; | |
2020 | ||
2021 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); | |
2022 | if (!bio) | |
2023 | return -EIO; | |
2024 | bio->bi_iter.bi_size = 0; | |
2025 | map_length = length; | |
2026 | ||
2027 | ret = btrfs_map_block(fs_info, WRITE, logical, | |
2028 | &map_length, &bbio, mirror_num); | |
2029 | if (ret) { | |
2030 | bio_put(bio); | |
2031 | return -EIO; | |
2032 | } | |
2033 | BUG_ON(mirror_num != bbio->mirror_num); | |
2034 | sector = bbio->stripes[mirror_num-1].physical >> 9; | |
2035 | bio->bi_iter.bi_sector = sector; | |
2036 | dev = bbio->stripes[mirror_num-1].dev; | |
2037 | btrfs_put_bbio(bbio); | |
2038 | if (!dev || !dev->bdev || !dev->writeable) { | |
2039 | bio_put(bio); | |
2040 | return -EIO; | |
2041 | } | |
2042 | bio->bi_bdev = dev->bdev; | |
2043 | bio_add_page(bio, page, length, pg_offset); | |
2044 | ||
2045 | if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) { | |
2046 | /* try to remap that extent elsewhere? */ | |
2047 | bio_put(bio); | |
2048 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); | |
2049 | return -EIO; | |
2050 | } | |
2051 | ||
2052 | btrfs_info_rl_in_rcu(fs_info, | |
2053 | "read error corrected: ino %llu off %llu (dev %s sector %llu)", | |
2054 | btrfs_ino(inode), start, | |
2055 | rcu_str_deref(dev->name), sector); | |
2056 | bio_put(bio); | |
2057 | return 0; | |
2058 | } | |
2059 | ||
2060 | int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb, | |
2061 | int mirror_num) | |
2062 | { | |
2063 | u64 start = eb->start; | |
2064 | unsigned long i, num_pages = num_extent_pages(eb->start, eb->len); | |
2065 | int ret = 0; | |
2066 | ||
2067 | if (root->fs_info->sb->s_flags & MS_RDONLY) | |
2068 | return -EROFS; | |
2069 | ||
2070 | for (i = 0; i < num_pages; i++) { | |
2071 | struct page *p = eb->pages[i]; | |
2072 | ||
2073 | ret = repair_io_failure(root->fs_info->btree_inode, start, | |
2074 | PAGE_SIZE, start, p, | |
2075 | start - page_offset(p), mirror_num); | |
2076 | if (ret) | |
2077 | break; | |
2078 | start += PAGE_SIZE; | |
2079 | } | |
2080 | ||
2081 | return ret; | |
2082 | } | |
2083 | ||
2084 | /* | |
2085 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2086 | * to see if we need to process or clean up an io_failure_record | |
2087 | */ | |
2088 | int clean_io_failure(struct inode *inode, u64 start, struct page *page, | |
2089 | unsigned int pg_offset) | |
2090 | { | |
2091 | u64 private; | |
2092 | struct io_failure_record *failrec; | |
2093 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2094 | struct extent_state *state; | |
2095 | int num_copies; | |
2096 | int ret; | |
2097 | ||
2098 | private = 0; | |
2099 | ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
2100 | (u64)-1, 1, EXTENT_DIRTY, 0); | |
2101 | if (!ret) | |
2102 | return 0; | |
2103 | ||
2104 | ret = get_state_failrec(&BTRFS_I(inode)->io_failure_tree, start, | |
2105 | &failrec); | |
2106 | if (ret) | |
2107 | return 0; | |
2108 | ||
2109 | BUG_ON(!failrec->this_mirror); | |
2110 | ||
2111 | if (failrec->in_validation) { | |
2112 | /* there was no real error, just free the record */ | |
2113 | pr_debug("clean_io_failure: freeing dummy error at %llu\n", | |
2114 | failrec->start); | |
2115 | goto out; | |
2116 | } | |
2117 | if (fs_info->sb->s_flags & MS_RDONLY) | |
2118 | goto out; | |
2119 | ||
2120 | spin_lock(&BTRFS_I(inode)->io_tree.lock); | |
2121 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, | |
2122 | failrec->start, | |
2123 | EXTENT_LOCKED); | |
2124 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); | |
2125 | ||
2126 | if (state && state->start <= failrec->start && | |
2127 | state->end >= failrec->start + failrec->len - 1) { | |
2128 | num_copies = btrfs_num_copies(fs_info, failrec->logical, | |
2129 | failrec->len); | |
2130 | if (num_copies > 1) { | |
2131 | repair_io_failure(inode, start, failrec->len, | |
2132 | failrec->logical, page, | |
2133 | pg_offset, failrec->failed_mirror); | |
2134 | } | |
2135 | } | |
2136 | ||
2137 | out: | |
2138 | free_io_failure(inode, failrec); | |
2139 | ||
2140 | return 0; | |
2141 | } | |
2142 | ||
2143 | /* | |
2144 | * Can be called when | |
2145 | * - hold extent lock | |
2146 | * - under ordered extent | |
2147 | * - the inode is freeing | |
2148 | */ | |
2149 | void btrfs_free_io_failure_record(struct inode *inode, u64 start, u64 end) | |
2150 | { | |
2151 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
2152 | struct io_failure_record *failrec; | |
2153 | struct extent_state *state, *next; | |
2154 | ||
2155 | if (RB_EMPTY_ROOT(&failure_tree->state)) | |
2156 | return; | |
2157 | ||
2158 | spin_lock(&failure_tree->lock); | |
2159 | state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY); | |
2160 | while (state) { | |
2161 | if (state->start > end) | |
2162 | break; | |
2163 | ||
2164 | ASSERT(state->end <= end); | |
2165 | ||
2166 | next = next_state(state); | |
2167 | ||
2168 | failrec = state->failrec; | |
2169 | free_extent_state(state); | |
2170 | kfree(failrec); | |
2171 | ||
2172 | state = next; | |
2173 | } | |
2174 | spin_unlock(&failure_tree->lock); | |
2175 | } | |
2176 | ||
2177 | int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end, | |
2178 | struct io_failure_record **failrec_ret) | |
2179 | { | |
2180 | struct io_failure_record *failrec; | |
2181 | struct extent_map *em; | |
2182 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
2183 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2184 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
2185 | int ret; | |
2186 | u64 logical; | |
2187 | ||
2188 | ret = get_state_failrec(failure_tree, start, &failrec); | |
2189 | if (ret) { | |
2190 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2191 | if (!failrec) | |
2192 | return -ENOMEM; | |
2193 | ||
2194 | failrec->start = start; | |
2195 | failrec->len = end - start + 1; | |
2196 | failrec->this_mirror = 0; | |
2197 | failrec->bio_flags = 0; | |
2198 | failrec->in_validation = 0; | |
2199 | ||
2200 | read_lock(&em_tree->lock); | |
2201 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2202 | if (!em) { | |
2203 | read_unlock(&em_tree->lock); | |
2204 | kfree(failrec); | |
2205 | return -EIO; | |
2206 | } | |
2207 | ||
2208 | if (em->start > start || em->start + em->len <= start) { | |
2209 | free_extent_map(em); | |
2210 | em = NULL; | |
2211 | } | |
2212 | read_unlock(&em_tree->lock); | |
2213 | if (!em) { | |
2214 | kfree(failrec); | |
2215 | return -EIO; | |
2216 | } | |
2217 | ||
2218 | logical = start - em->start; | |
2219 | logical = em->block_start + logical; | |
2220 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2221 | logical = em->block_start; | |
2222 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2223 | extent_set_compress_type(&failrec->bio_flags, | |
2224 | em->compress_type); | |
2225 | } | |
2226 | ||
2227 | pr_debug("Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu\n", | |
2228 | logical, start, failrec->len); | |
2229 | ||
2230 | failrec->logical = logical; | |
2231 | free_extent_map(em); | |
2232 | ||
2233 | /* set the bits in the private failure tree */ | |
2234 | ret = set_extent_bits(failure_tree, start, end, | |
2235 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
2236 | if (ret >= 0) | |
2237 | ret = set_state_failrec(failure_tree, start, failrec); | |
2238 | /* set the bits in the inode's tree */ | |
2239 | if (ret >= 0) | |
2240 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED, | |
2241 | GFP_NOFS); | |
2242 | if (ret < 0) { | |
2243 | kfree(failrec); | |
2244 | return ret; | |
2245 | } | |
2246 | } else { | |
2247 | pr_debug("Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d\n", | |
2248 | failrec->logical, failrec->start, failrec->len, | |
2249 | failrec->in_validation); | |
2250 | /* | |
2251 | * when data can be on disk more than twice, add to failrec here | |
2252 | * (e.g. with a list for failed_mirror) to make | |
2253 | * clean_io_failure() clean all those errors at once. | |
2254 | */ | |
2255 | } | |
2256 | ||
2257 | *failrec_ret = failrec; | |
2258 | ||
2259 | return 0; | |
2260 | } | |
2261 | ||
2262 | int btrfs_check_repairable(struct inode *inode, struct bio *failed_bio, | |
2263 | struct io_failure_record *failrec, int failed_mirror) | |
2264 | { | |
2265 | int num_copies; | |
2266 | ||
2267 | num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info, | |
2268 | failrec->logical, failrec->len); | |
2269 | if (num_copies == 1) { | |
2270 | /* | |
2271 | * we only have a single copy of the data, so don't bother with | |
2272 | * all the retry and error correction code that follows. no | |
2273 | * matter what the error is, it is very likely to persist. | |
2274 | */ | |
2275 | pr_debug("Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n", | |
2276 | num_copies, failrec->this_mirror, failed_mirror); | |
2277 | return 0; | |
2278 | } | |
2279 | ||
2280 | /* | |
2281 | * there are two premises: | |
2282 | * a) deliver good data to the caller | |
2283 | * b) correct the bad sectors on disk | |
2284 | */ | |
2285 | if (failed_bio->bi_vcnt > 1) { | |
2286 | /* | |
2287 | * to fulfill b), we need to know the exact failing sectors, as | |
2288 | * we don't want to rewrite any more than the failed ones. thus, | |
2289 | * we need separate read requests for the failed bio | |
2290 | * | |
2291 | * if the following BUG_ON triggers, our validation request got | |
2292 | * merged. we need separate requests for our algorithm to work. | |
2293 | */ | |
2294 | BUG_ON(failrec->in_validation); | |
2295 | failrec->in_validation = 1; | |
2296 | failrec->this_mirror = failed_mirror; | |
2297 | } else { | |
2298 | /* | |
2299 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2300 | * of the failed sector and if we succeed, we have setup | |
2301 | * everything for repair_io_failure to do the rest for us. | |
2302 | */ | |
2303 | if (failrec->in_validation) { | |
2304 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2305 | failrec->in_validation = 0; | |
2306 | failrec->this_mirror = 0; | |
2307 | } | |
2308 | failrec->failed_mirror = failed_mirror; | |
2309 | failrec->this_mirror++; | |
2310 | if (failrec->this_mirror == failed_mirror) | |
2311 | failrec->this_mirror++; | |
2312 | } | |
2313 | ||
2314 | if (failrec->this_mirror > num_copies) { | |
2315 | pr_debug("Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n", | |
2316 | num_copies, failrec->this_mirror, failed_mirror); | |
2317 | return 0; | |
2318 | } | |
2319 | ||
2320 | return 1; | |
2321 | } | |
2322 | ||
2323 | ||
2324 | struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio, | |
2325 | struct io_failure_record *failrec, | |
2326 | struct page *page, int pg_offset, int icsum, | |
2327 | bio_end_io_t *endio_func, void *data) | |
2328 | { | |
2329 | struct bio *bio; | |
2330 | struct btrfs_io_bio *btrfs_failed_bio; | |
2331 | struct btrfs_io_bio *btrfs_bio; | |
2332 | ||
2333 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); | |
2334 | if (!bio) | |
2335 | return NULL; | |
2336 | ||
2337 | bio->bi_end_io = endio_func; | |
2338 | bio->bi_iter.bi_sector = failrec->logical >> 9; | |
2339 | bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; | |
2340 | bio->bi_iter.bi_size = 0; | |
2341 | bio->bi_private = data; | |
2342 | ||
2343 | btrfs_failed_bio = btrfs_io_bio(failed_bio); | |
2344 | if (btrfs_failed_bio->csum) { | |
2345 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2346 | u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); | |
2347 | ||
2348 | btrfs_bio = btrfs_io_bio(bio); | |
2349 | btrfs_bio->csum = btrfs_bio->csum_inline; | |
2350 | icsum *= csum_size; | |
2351 | memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum, | |
2352 | csum_size); | |
2353 | } | |
2354 | ||
2355 | bio_add_page(bio, page, failrec->len, pg_offset); | |
2356 | ||
2357 | return bio; | |
2358 | } | |
2359 | ||
2360 | /* | |
2361 | * this is a generic handler for readpage errors (default | |
2362 | * readpage_io_failed_hook). if other copies exist, read those and write back | |
2363 | * good data to the failed position. does not investigate in remapping the | |
2364 | * failed extent elsewhere, hoping the device will be smart enough to do this as | |
2365 | * needed | |
2366 | */ | |
2367 | ||
2368 | static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset, | |
2369 | struct page *page, u64 start, u64 end, | |
2370 | int failed_mirror) | |
2371 | { | |
2372 | struct io_failure_record *failrec; | |
2373 | struct inode *inode = page->mapping->host; | |
2374 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2375 | struct bio *bio; | |
2376 | int read_mode; | |
2377 | int ret; | |
2378 | ||
2379 | BUG_ON(failed_bio->bi_rw & REQ_WRITE); | |
2380 | ||
2381 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
2382 | if (ret) | |
2383 | return ret; | |
2384 | ||
2385 | ret = btrfs_check_repairable(inode, failed_bio, failrec, failed_mirror); | |
2386 | if (!ret) { | |
2387 | free_io_failure(inode, failrec); | |
2388 | return -EIO; | |
2389 | } | |
2390 | ||
2391 | if (failed_bio->bi_vcnt > 1) | |
2392 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; | |
2393 | else | |
2394 | read_mode = READ_SYNC; | |
2395 | ||
2396 | phy_offset >>= inode->i_sb->s_blocksize_bits; | |
2397 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2398 | start - page_offset(page), | |
2399 | (int)phy_offset, failed_bio->bi_end_io, | |
2400 | NULL); | |
2401 | if (!bio) { | |
2402 | free_io_failure(inode, failrec); | |
2403 | return -EIO; | |
2404 | } | |
2405 | ||
2406 | pr_debug("Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d\n", | |
2407 | read_mode, failrec->this_mirror, failrec->in_validation); | |
2408 | ||
2409 | ret = tree->ops->submit_bio_hook(inode, read_mode, bio, | |
2410 | failrec->this_mirror, | |
2411 | failrec->bio_flags, 0); | |
2412 | if (ret) { | |
2413 | free_io_failure(inode, failrec); | |
2414 | bio_put(bio); | |
2415 | } | |
2416 | ||
2417 | return ret; | |
2418 | } | |
2419 | ||
2420 | /* lots and lots of room for performance fixes in the end_bio funcs */ | |
2421 | ||
2422 | void end_extent_writepage(struct page *page, int err, u64 start, u64 end) | |
2423 | { | |
2424 | int uptodate = (err == 0); | |
2425 | struct extent_io_tree *tree; | |
2426 | int ret = 0; | |
2427 | ||
2428 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
2429 | ||
2430 | if (tree->ops && tree->ops->writepage_end_io_hook) { | |
2431 | ret = tree->ops->writepage_end_io_hook(page, start, | |
2432 | end, NULL, uptodate); | |
2433 | if (ret) | |
2434 | uptodate = 0; | |
2435 | } | |
2436 | ||
2437 | if (!uptodate) { | |
2438 | ClearPageUptodate(page); | |
2439 | SetPageError(page); | |
2440 | ret = ret < 0 ? ret : -EIO; | |
2441 | mapping_set_error(page->mapping, ret); | |
2442 | } | |
2443 | } | |
2444 | ||
2445 | /* | |
2446 | * after a writepage IO is done, we need to: | |
2447 | * clear the uptodate bits on error | |
2448 | * clear the writeback bits in the extent tree for this IO | |
2449 | * end_page_writeback if the page has no more pending IO | |
2450 | * | |
2451 | * Scheduling is not allowed, so the extent state tree is expected | |
2452 | * to have one and only one object corresponding to this IO. | |
2453 | */ | |
2454 | static void end_bio_extent_writepage(struct bio *bio) | |
2455 | { | |
2456 | struct bio_vec *bvec; | |
2457 | u64 start; | |
2458 | u64 end; | |
2459 | int i; | |
2460 | ||
2461 | bio_for_each_segment_all(bvec, bio, i) { | |
2462 | struct page *page = bvec->bv_page; | |
2463 | ||
2464 | /* We always issue full-page reads, but if some block | |
2465 | * in a page fails to read, blk_update_request() will | |
2466 | * advance bv_offset and adjust bv_len to compensate. | |
2467 | * Print a warning for nonzero offsets, and an error | |
2468 | * if they don't add up to a full page. */ | |
2469 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { | |
2470 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
2471 | btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info, | |
2472 | "partial page write in btrfs with offset %u and length %u", | |
2473 | bvec->bv_offset, bvec->bv_len); | |
2474 | else | |
2475 | btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info, | |
2476 | "incomplete page write in btrfs with offset %u and " | |
2477 | "length %u", | |
2478 | bvec->bv_offset, bvec->bv_len); | |
2479 | } | |
2480 | ||
2481 | start = page_offset(page); | |
2482 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
2483 | ||
2484 | end_extent_writepage(page, bio->bi_error, start, end); | |
2485 | end_page_writeback(page); | |
2486 | } | |
2487 | ||
2488 | bio_put(bio); | |
2489 | } | |
2490 | ||
2491 | static void | |
2492 | endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len, | |
2493 | int uptodate) | |
2494 | { | |
2495 | struct extent_state *cached = NULL; | |
2496 | u64 end = start + len - 1; | |
2497 | ||
2498 | if (uptodate && tree->track_uptodate) | |
2499 | set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC); | |
2500 | unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC); | |
2501 | } | |
2502 | ||
2503 | /* | |
2504 | * after a readpage IO is done, we need to: | |
2505 | * clear the uptodate bits on error | |
2506 | * set the uptodate bits if things worked | |
2507 | * set the page up to date if all extents in the tree are uptodate | |
2508 | * clear the lock bit in the extent tree | |
2509 | * unlock the page if there are no other extents locked for it | |
2510 | * | |
2511 | * Scheduling is not allowed, so the extent state tree is expected | |
2512 | * to have one and only one object corresponding to this IO. | |
2513 | */ | |
2514 | static void end_bio_extent_readpage(struct bio *bio) | |
2515 | { | |
2516 | struct bio_vec *bvec; | |
2517 | int uptodate = !bio->bi_error; | |
2518 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
2519 | struct extent_io_tree *tree; | |
2520 | u64 offset = 0; | |
2521 | u64 start; | |
2522 | u64 end; | |
2523 | u64 len; | |
2524 | u64 extent_start = 0; | |
2525 | u64 extent_len = 0; | |
2526 | int mirror; | |
2527 | int ret; | |
2528 | int i; | |
2529 | ||
2530 | bio_for_each_segment_all(bvec, bio, i) { | |
2531 | struct page *page = bvec->bv_page; | |
2532 | struct inode *inode = page->mapping->host; | |
2533 | ||
2534 | pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, " | |
2535 | "mirror=%u\n", (u64)bio->bi_iter.bi_sector, | |
2536 | bio->bi_error, io_bio->mirror_num); | |
2537 | tree = &BTRFS_I(inode)->io_tree; | |
2538 | ||
2539 | /* We always issue full-page reads, but if some block | |
2540 | * in a page fails to read, blk_update_request() will | |
2541 | * advance bv_offset and adjust bv_len to compensate. | |
2542 | * Print a warning for nonzero offsets, and an error | |
2543 | * if they don't add up to a full page. */ | |
2544 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { | |
2545 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
2546 | btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info, | |
2547 | "partial page read in btrfs with offset %u and length %u", | |
2548 | bvec->bv_offset, bvec->bv_len); | |
2549 | else | |
2550 | btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info, | |
2551 | "incomplete page read in btrfs with offset %u and " | |
2552 | "length %u", | |
2553 | bvec->bv_offset, bvec->bv_len); | |
2554 | } | |
2555 | ||
2556 | start = page_offset(page); | |
2557 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
2558 | len = bvec->bv_len; | |
2559 | ||
2560 | mirror = io_bio->mirror_num; | |
2561 | if (likely(uptodate && tree->ops && | |
2562 | tree->ops->readpage_end_io_hook)) { | |
2563 | ret = tree->ops->readpage_end_io_hook(io_bio, offset, | |
2564 | page, start, end, | |
2565 | mirror); | |
2566 | if (ret) | |
2567 | uptodate = 0; | |
2568 | else | |
2569 | clean_io_failure(inode, start, page, 0); | |
2570 | } | |
2571 | ||
2572 | if (likely(uptodate)) | |
2573 | goto readpage_ok; | |
2574 | ||
2575 | if (tree->ops && tree->ops->readpage_io_failed_hook) { | |
2576 | ret = tree->ops->readpage_io_failed_hook(page, mirror); | |
2577 | if (!ret && !bio->bi_error) | |
2578 | uptodate = 1; | |
2579 | } else { | |
2580 | /* | |
2581 | * The generic bio_readpage_error handles errors the | |
2582 | * following way: If possible, new read requests are | |
2583 | * created and submitted and will end up in | |
2584 | * end_bio_extent_readpage as well (if we're lucky, not | |
2585 | * in the !uptodate case). In that case it returns 0 and | |
2586 | * we just go on with the next page in our bio. If it | |
2587 | * can't handle the error it will return -EIO and we | |
2588 | * remain responsible for that page. | |
2589 | */ | |
2590 | ret = bio_readpage_error(bio, offset, page, start, end, | |
2591 | mirror); | |
2592 | if (ret == 0) { | |
2593 | uptodate = !bio->bi_error; | |
2594 | offset += len; | |
2595 | continue; | |
2596 | } | |
2597 | } | |
2598 | readpage_ok: | |
2599 | if (likely(uptodate)) { | |
2600 | loff_t i_size = i_size_read(inode); | |
2601 | pgoff_t end_index = i_size >> PAGE_SHIFT; | |
2602 | unsigned off; | |
2603 | ||
2604 | /* Zero out the end if this page straddles i_size */ | |
2605 | off = i_size & (PAGE_SIZE-1); | |
2606 | if (page->index == end_index && off) | |
2607 | zero_user_segment(page, off, PAGE_SIZE); | |
2608 | SetPageUptodate(page); | |
2609 | } else { | |
2610 | ClearPageUptodate(page); | |
2611 | SetPageError(page); | |
2612 | } | |
2613 | unlock_page(page); | |
2614 | offset += len; | |
2615 | ||
2616 | if (unlikely(!uptodate)) { | |
2617 | if (extent_len) { | |
2618 | endio_readpage_release_extent(tree, | |
2619 | extent_start, | |
2620 | extent_len, 1); | |
2621 | extent_start = 0; | |
2622 | extent_len = 0; | |
2623 | } | |
2624 | endio_readpage_release_extent(tree, start, | |
2625 | end - start + 1, 0); | |
2626 | } else if (!extent_len) { | |
2627 | extent_start = start; | |
2628 | extent_len = end + 1 - start; | |
2629 | } else if (extent_start + extent_len == start) { | |
2630 | extent_len += end + 1 - start; | |
2631 | } else { | |
2632 | endio_readpage_release_extent(tree, extent_start, | |
2633 | extent_len, uptodate); | |
2634 | extent_start = start; | |
2635 | extent_len = end + 1 - start; | |
2636 | } | |
2637 | } | |
2638 | ||
2639 | if (extent_len) | |
2640 | endio_readpage_release_extent(tree, extent_start, extent_len, | |
2641 | uptodate); | |
2642 | if (io_bio->end_io) | |
2643 | io_bio->end_io(io_bio, bio->bi_error); | |
2644 | bio_put(bio); | |
2645 | } | |
2646 | ||
2647 | /* | |
2648 | * this allocates from the btrfs_bioset. We're returning a bio right now | |
2649 | * but you can call btrfs_io_bio for the appropriate container_of magic | |
2650 | */ | |
2651 | struct bio * | |
2652 | btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
2653 | gfp_t gfp_flags) | |
2654 | { | |
2655 | struct btrfs_io_bio *btrfs_bio; | |
2656 | struct bio *bio; | |
2657 | ||
2658 | bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset); | |
2659 | ||
2660 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
2661 | while (!bio && (nr_vecs /= 2)) { | |
2662 | bio = bio_alloc_bioset(gfp_flags, | |
2663 | nr_vecs, btrfs_bioset); | |
2664 | } | |
2665 | } | |
2666 | ||
2667 | if (bio) { | |
2668 | bio->bi_bdev = bdev; | |
2669 | bio->bi_iter.bi_sector = first_sector; | |
2670 | btrfs_bio = btrfs_io_bio(bio); | |
2671 | btrfs_bio->csum = NULL; | |
2672 | btrfs_bio->csum_allocated = NULL; | |
2673 | btrfs_bio->end_io = NULL; | |
2674 | } | |
2675 | return bio; | |
2676 | } | |
2677 | ||
2678 | struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask) | |
2679 | { | |
2680 | struct btrfs_io_bio *btrfs_bio; | |
2681 | struct bio *new; | |
2682 | ||
2683 | new = bio_clone_bioset(bio, gfp_mask, btrfs_bioset); | |
2684 | if (new) { | |
2685 | btrfs_bio = btrfs_io_bio(new); | |
2686 | btrfs_bio->csum = NULL; | |
2687 | btrfs_bio->csum_allocated = NULL; | |
2688 | btrfs_bio->end_io = NULL; | |
2689 | ||
2690 | #ifdef CONFIG_BLK_CGROUP | |
2691 | /* FIXME, put this into bio_clone_bioset */ | |
2692 | if (bio->bi_css) | |
2693 | bio_associate_blkcg(new, bio->bi_css); | |
2694 | #endif | |
2695 | } | |
2696 | return new; | |
2697 | } | |
2698 | ||
2699 | /* this also allocates from the btrfs_bioset */ | |
2700 | struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs) | |
2701 | { | |
2702 | struct btrfs_io_bio *btrfs_bio; | |
2703 | struct bio *bio; | |
2704 | ||
2705 | bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset); | |
2706 | if (bio) { | |
2707 | btrfs_bio = btrfs_io_bio(bio); | |
2708 | btrfs_bio->csum = NULL; | |
2709 | btrfs_bio->csum_allocated = NULL; | |
2710 | btrfs_bio->end_io = NULL; | |
2711 | } | |
2712 | return bio; | |
2713 | } | |
2714 | ||
2715 | ||
2716 | static int __must_check submit_one_bio(int rw, struct bio *bio, | |
2717 | int mirror_num, unsigned long bio_flags) | |
2718 | { | |
2719 | int ret = 0; | |
2720 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
2721 | struct page *page = bvec->bv_page; | |
2722 | struct extent_io_tree *tree = bio->bi_private; | |
2723 | u64 start; | |
2724 | ||
2725 | start = page_offset(page) + bvec->bv_offset; | |
2726 | ||
2727 | bio->bi_private = NULL; | |
2728 | ||
2729 | bio_get(bio); | |
2730 | ||
2731 | if (tree->ops && tree->ops->submit_bio_hook) | |
2732 | ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio, | |
2733 | mirror_num, bio_flags, start); | |
2734 | else | |
2735 | btrfsic_submit_bio(rw, bio); | |
2736 | ||
2737 | bio_put(bio); | |
2738 | return ret; | |
2739 | } | |
2740 | ||
2741 | static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page, | |
2742 | unsigned long offset, size_t size, struct bio *bio, | |
2743 | unsigned long bio_flags) | |
2744 | { | |
2745 | int ret = 0; | |
2746 | if (tree->ops && tree->ops->merge_bio_hook) | |
2747 | ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio, | |
2748 | bio_flags); | |
2749 | BUG_ON(ret < 0); | |
2750 | return ret; | |
2751 | ||
2752 | } | |
2753 | ||
2754 | static int submit_extent_page(int rw, struct extent_io_tree *tree, | |
2755 | struct writeback_control *wbc, | |
2756 | struct page *page, sector_t sector, | |
2757 | size_t size, unsigned long offset, | |
2758 | struct block_device *bdev, | |
2759 | struct bio **bio_ret, | |
2760 | unsigned long max_pages, | |
2761 | bio_end_io_t end_io_func, | |
2762 | int mirror_num, | |
2763 | unsigned long prev_bio_flags, | |
2764 | unsigned long bio_flags, | |
2765 | bool force_bio_submit) | |
2766 | { | |
2767 | int ret = 0; | |
2768 | struct bio *bio; | |
2769 | int contig = 0; | |
2770 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; | |
2771 | size_t page_size = min_t(size_t, size, PAGE_SIZE); | |
2772 | ||
2773 | if (bio_ret && *bio_ret) { | |
2774 | bio = *bio_ret; | |
2775 | if (old_compressed) | |
2776 | contig = bio->bi_iter.bi_sector == sector; | |
2777 | else | |
2778 | contig = bio_end_sector(bio) == sector; | |
2779 | ||
2780 | if (prev_bio_flags != bio_flags || !contig || | |
2781 | force_bio_submit || | |
2782 | merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) || | |
2783 | bio_add_page(bio, page, page_size, offset) < page_size) { | |
2784 | ret = submit_one_bio(rw, bio, mirror_num, | |
2785 | prev_bio_flags); | |
2786 | if (ret < 0) { | |
2787 | *bio_ret = NULL; | |
2788 | return ret; | |
2789 | } | |
2790 | bio = NULL; | |
2791 | } else { | |
2792 | if (wbc) | |
2793 | wbc_account_io(wbc, page, page_size); | |
2794 | return 0; | |
2795 | } | |
2796 | } | |
2797 | ||
2798 | bio = btrfs_bio_alloc(bdev, sector, BIO_MAX_PAGES, | |
2799 | GFP_NOFS | __GFP_HIGH); | |
2800 | if (!bio) | |
2801 | return -ENOMEM; | |
2802 | ||
2803 | bio_add_page(bio, page, page_size, offset); | |
2804 | bio->bi_end_io = end_io_func; | |
2805 | bio->bi_private = tree; | |
2806 | if (wbc) { | |
2807 | wbc_init_bio(wbc, bio); | |
2808 | wbc_account_io(wbc, page, page_size); | |
2809 | } | |
2810 | ||
2811 | if (bio_ret) | |
2812 | *bio_ret = bio; | |
2813 | else | |
2814 | ret = submit_one_bio(rw, bio, mirror_num, bio_flags); | |
2815 | ||
2816 | return ret; | |
2817 | } | |
2818 | ||
2819 | static void attach_extent_buffer_page(struct extent_buffer *eb, | |
2820 | struct page *page) | |
2821 | { | |
2822 | if (!PagePrivate(page)) { | |
2823 | SetPagePrivate(page); | |
2824 | get_page(page); | |
2825 | set_page_private(page, (unsigned long)eb); | |
2826 | } else { | |
2827 | WARN_ON(page->private != (unsigned long)eb); | |
2828 | } | |
2829 | } | |
2830 | ||
2831 | void set_page_extent_mapped(struct page *page) | |
2832 | { | |
2833 | if (!PagePrivate(page)) { | |
2834 | SetPagePrivate(page); | |
2835 | get_page(page); | |
2836 | set_page_private(page, EXTENT_PAGE_PRIVATE); | |
2837 | } | |
2838 | } | |
2839 | ||
2840 | static struct extent_map * | |
2841 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
2842 | u64 start, u64 len, get_extent_t *get_extent, | |
2843 | struct extent_map **em_cached) | |
2844 | { | |
2845 | struct extent_map *em; | |
2846 | ||
2847 | if (em_cached && *em_cached) { | |
2848 | em = *em_cached; | |
2849 | if (extent_map_in_tree(em) && start >= em->start && | |
2850 | start < extent_map_end(em)) { | |
2851 | atomic_inc(&em->refs); | |
2852 | return em; | |
2853 | } | |
2854 | ||
2855 | free_extent_map(em); | |
2856 | *em_cached = NULL; | |
2857 | } | |
2858 | ||
2859 | em = get_extent(inode, page, pg_offset, start, len, 0); | |
2860 | if (em_cached && !IS_ERR_OR_NULL(em)) { | |
2861 | BUG_ON(*em_cached); | |
2862 | atomic_inc(&em->refs); | |
2863 | *em_cached = em; | |
2864 | } | |
2865 | return em; | |
2866 | } | |
2867 | /* | |
2868 | * basic readpage implementation. Locked extent state structs are inserted | |
2869 | * into the tree that are removed when the IO is done (by the end_io | |
2870 | * handlers) | |
2871 | * XXX JDM: This needs looking at to ensure proper page locking | |
2872 | */ | |
2873 | static int __do_readpage(struct extent_io_tree *tree, | |
2874 | struct page *page, | |
2875 | get_extent_t *get_extent, | |
2876 | struct extent_map **em_cached, | |
2877 | struct bio **bio, int mirror_num, | |
2878 | unsigned long *bio_flags, int rw, | |
2879 | u64 *prev_em_start) | |
2880 | { | |
2881 | struct inode *inode = page->mapping->host; | |
2882 | u64 start = page_offset(page); | |
2883 | u64 page_end = start + PAGE_SIZE - 1; | |
2884 | u64 end; | |
2885 | u64 cur = start; | |
2886 | u64 extent_offset; | |
2887 | u64 last_byte = i_size_read(inode); | |
2888 | u64 block_start; | |
2889 | u64 cur_end; | |
2890 | sector_t sector; | |
2891 | struct extent_map *em; | |
2892 | struct block_device *bdev; | |
2893 | int ret; | |
2894 | int nr = 0; | |
2895 | size_t pg_offset = 0; | |
2896 | size_t iosize; | |
2897 | size_t disk_io_size; | |
2898 | size_t blocksize = inode->i_sb->s_blocksize; | |
2899 | unsigned long this_bio_flag = 0; | |
2900 | ||
2901 | set_page_extent_mapped(page); | |
2902 | ||
2903 | end = page_end; | |
2904 | if (!PageUptodate(page)) { | |
2905 | if (cleancache_get_page(page) == 0) { | |
2906 | BUG_ON(blocksize != PAGE_SIZE); | |
2907 | unlock_extent(tree, start, end); | |
2908 | goto out; | |
2909 | } | |
2910 | } | |
2911 | ||
2912 | if (page->index == last_byte >> PAGE_SHIFT) { | |
2913 | char *userpage; | |
2914 | size_t zero_offset = last_byte & (PAGE_SIZE - 1); | |
2915 | ||
2916 | if (zero_offset) { | |
2917 | iosize = PAGE_SIZE - zero_offset; | |
2918 | userpage = kmap_atomic(page); | |
2919 | memset(userpage + zero_offset, 0, iosize); | |
2920 | flush_dcache_page(page); | |
2921 | kunmap_atomic(userpage); | |
2922 | } | |
2923 | } | |
2924 | while (cur <= end) { | |
2925 | unsigned long pnr = (last_byte >> PAGE_SHIFT) + 1; | |
2926 | bool force_bio_submit = false; | |
2927 | ||
2928 | if (cur >= last_byte) { | |
2929 | char *userpage; | |
2930 | struct extent_state *cached = NULL; | |
2931 | ||
2932 | iosize = PAGE_SIZE - pg_offset; | |
2933 | userpage = kmap_atomic(page); | |
2934 | memset(userpage + pg_offset, 0, iosize); | |
2935 | flush_dcache_page(page); | |
2936 | kunmap_atomic(userpage); | |
2937 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
2938 | &cached, GFP_NOFS); | |
2939 | unlock_extent_cached(tree, cur, | |
2940 | cur + iosize - 1, | |
2941 | &cached, GFP_NOFS); | |
2942 | break; | |
2943 | } | |
2944 | em = __get_extent_map(inode, page, pg_offset, cur, | |
2945 | end - cur + 1, get_extent, em_cached); | |
2946 | if (IS_ERR_OR_NULL(em)) { | |
2947 | SetPageError(page); | |
2948 | unlock_extent(tree, cur, end); | |
2949 | break; | |
2950 | } | |
2951 | extent_offset = cur - em->start; | |
2952 | BUG_ON(extent_map_end(em) <= cur); | |
2953 | BUG_ON(end < cur); | |
2954 | ||
2955 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2956 | this_bio_flag |= EXTENT_BIO_COMPRESSED; | |
2957 | extent_set_compress_type(&this_bio_flag, | |
2958 | em->compress_type); | |
2959 | } | |
2960 | ||
2961 | iosize = min(extent_map_end(em) - cur, end - cur + 1); | |
2962 | cur_end = min(extent_map_end(em) - 1, end); | |
2963 | iosize = ALIGN(iosize, blocksize); | |
2964 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { | |
2965 | disk_io_size = em->block_len; | |
2966 | sector = em->block_start >> 9; | |
2967 | } else { | |
2968 | sector = (em->block_start + extent_offset) >> 9; | |
2969 | disk_io_size = iosize; | |
2970 | } | |
2971 | bdev = em->bdev; | |
2972 | block_start = em->block_start; | |
2973 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
2974 | block_start = EXTENT_MAP_HOLE; | |
2975 | ||
2976 | /* | |
2977 | * If we have a file range that points to a compressed extent | |
2978 | * and it's followed by a consecutive file range that points to | |
2979 | * to the same compressed extent (possibly with a different | |
2980 | * offset and/or length, so it either points to the whole extent | |
2981 | * or only part of it), we must make sure we do not submit a | |
2982 | * single bio to populate the pages for the 2 ranges because | |
2983 | * this makes the compressed extent read zero out the pages | |
2984 | * belonging to the 2nd range. Imagine the following scenario: | |
2985 | * | |
2986 | * File layout | |
2987 | * [0 - 8K] [8K - 24K] | |
2988 | * | | | |
2989 | * | | | |
2990 | * points to extent X, points to extent X, | |
2991 | * offset 4K, length of 8K offset 0, length 16K | |
2992 | * | |
2993 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
2994 | * | |
2995 | * If the bio to read the compressed extent covers both ranges, | |
2996 | * it will decompress extent X into the pages belonging to the | |
2997 | * first range and then it will stop, zeroing out the remaining | |
2998 | * pages that belong to the other range that points to extent X. | |
2999 | * So here we make sure we submit 2 bios, one for the first | |
3000 | * range and another one for the third range. Both will target | |
3001 | * the same physical extent from disk, but we can't currently | |
3002 | * make the compressed bio endio callback populate the pages | |
3003 | * for both ranges because each compressed bio is tightly | |
3004 | * coupled with a single extent map, and each range can have | |
3005 | * an extent map with a different offset value relative to the | |
3006 | * uncompressed data of our extent and different lengths. This | |
3007 | * is a corner case so we prioritize correctness over | |
3008 | * non-optimal behavior (submitting 2 bios for the same extent). | |
3009 | */ | |
3010 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
3011 | prev_em_start && *prev_em_start != (u64)-1 && | |
3012 | *prev_em_start != em->orig_start) | |
3013 | force_bio_submit = true; | |
3014 | ||
3015 | if (prev_em_start) | |
3016 | *prev_em_start = em->orig_start; | |
3017 | ||
3018 | free_extent_map(em); | |
3019 | em = NULL; | |
3020 | ||
3021 | /* we've found a hole, just zero and go on */ | |
3022 | if (block_start == EXTENT_MAP_HOLE) { | |
3023 | char *userpage; | |
3024 | struct extent_state *cached = NULL; | |
3025 | ||
3026 | userpage = kmap_atomic(page); | |
3027 | memset(userpage + pg_offset, 0, iosize); | |
3028 | flush_dcache_page(page); | |
3029 | kunmap_atomic(userpage); | |
3030 | ||
3031 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
3032 | &cached, GFP_NOFS); | |
3033 | unlock_extent_cached(tree, cur, | |
3034 | cur + iosize - 1, | |
3035 | &cached, GFP_NOFS); | |
3036 | cur = cur + iosize; | |
3037 | pg_offset += iosize; | |
3038 | continue; | |
3039 | } | |
3040 | /* the get_extent function already copied into the page */ | |
3041 | if (test_range_bit(tree, cur, cur_end, | |
3042 | EXTENT_UPTODATE, 1, NULL)) { | |
3043 | check_page_uptodate(tree, page); | |
3044 | unlock_extent(tree, cur, cur + iosize - 1); | |
3045 | cur = cur + iosize; | |
3046 | pg_offset += iosize; | |
3047 | continue; | |
3048 | } | |
3049 | /* we have an inline extent but it didn't get marked up | |
3050 | * to date. Error out | |
3051 | */ | |
3052 | if (block_start == EXTENT_MAP_INLINE) { | |
3053 | SetPageError(page); | |
3054 | unlock_extent(tree, cur, cur + iosize - 1); | |
3055 | cur = cur + iosize; | |
3056 | pg_offset += iosize; | |
3057 | continue; | |
3058 | } | |
3059 | ||
3060 | pnr -= page->index; | |
3061 | ret = submit_extent_page(rw, tree, NULL, page, | |
3062 | sector, disk_io_size, pg_offset, | |
3063 | bdev, bio, pnr, | |
3064 | end_bio_extent_readpage, mirror_num, | |
3065 | *bio_flags, | |
3066 | this_bio_flag, | |
3067 | force_bio_submit); | |
3068 | if (!ret) { | |
3069 | nr++; | |
3070 | *bio_flags = this_bio_flag; | |
3071 | } else { | |
3072 | SetPageError(page); | |
3073 | unlock_extent(tree, cur, cur + iosize - 1); | |
3074 | } | |
3075 | cur = cur + iosize; | |
3076 | pg_offset += iosize; | |
3077 | } | |
3078 | out: | |
3079 | if (!nr) { | |
3080 | if (!PageError(page)) | |
3081 | SetPageUptodate(page); | |
3082 | unlock_page(page); | |
3083 | } | |
3084 | return 0; | |
3085 | } | |
3086 | ||
3087 | static inline void __do_contiguous_readpages(struct extent_io_tree *tree, | |
3088 | struct page *pages[], int nr_pages, | |
3089 | u64 start, u64 end, | |
3090 | get_extent_t *get_extent, | |
3091 | struct extent_map **em_cached, | |
3092 | struct bio **bio, int mirror_num, | |
3093 | unsigned long *bio_flags, int rw, | |
3094 | u64 *prev_em_start) | |
3095 | { | |
3096 | struct inode *inode; | |
3097 | struct btrfs_ordered_extent *ordered; | |
3098 | int index; | |
3099 | ||
3100 | inode = pages[0]->mapping->host; | |
3101 | while (1) { | |
3102 | lock_extent(tree, start, end); | |
3103 | ordered = btrfs_lookup_ordered_range(inode, start, | |
3104 | end - start + 1); | |
3105 | if (!ordered) | |
3106 | break; | |
3107 | unlock_extent(tree, start, end); | |
3108 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3109 | btrfs_put_ordered_extent(ordered); | |
3110 | } | |
3111 | ||
3112 | for (index = 0; index < nr_pages; index++) { | |
3113 | __do_readpage(tree, pages[index], get_extent, em_cached, bio, | |
3114 | mirror_num, bio_flags, rw, prev_em_start); | |
3115 | put_page(pages[index]); | |
3116 | } | |
3117 | } | |
3118 | ||
3119 | static void __extent_readpages(struct extent_io_tree *tree, | |
3120 | struct page *pages[], | |
3121 | int nr_pages, get_extent_t *get_extent, | |
3122 | struct extent_map **em_cached, | |
3123 | struct bio **bio, int mirror_num, | |
3124 | unsigned long *bio_flags, int rw, | |
3125 | u64 *prev_em_start) | |
3126 | { | |
3127 | u64 start = 0; | |
3128 | u64 end = 0; | |
3129 | u64 page_start; | |
3130 | int index; | |
3131 | int first_index = 0; | |
3132 | ||
3133 | for (index = 0; index < nr_pages; index++) { | |
3134 | page_start = page_offset(pages[index]); | |
3135 | if (!end) { | |
3136 | start = page_start; | |
3137 | end = start + PAGE_SIZE - 1; | |
3138 | first_index = index; | |
3139 | } else if (end + 1 == page_start) { | |
3140 | end += PAGE_SIZE; | |
3141 | } else { | |
3142 | __do_contiguous_readpages(tree, &pages[first_index], | |
3143 | index - first_index, start, | |
3144 | end, get_extent, em_cached, | |
3145 | bio, mirror_num, bio_flags, | |
3146 | rw, prev_em_start); | |
3147 | start = page_start; | |
3148 | end = start + PAGE_SIZE - 1; | |
3149 | first_index = index; | |
3150 | } | |
3151 | } | |
3152 | ||
3153 | if (end) | |
3154 | __do_contiguous_readpages(tree, &pages[first_index], | |
3155 | index - first_index, start, | |
3156 | end, get_extent, em_cached, bio, | |
3157 | mirror_num, bio_flags, rw, | |
3158 | prev_em_start); | |
3159 | } | |
3160 | ||
3161 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
3162 | struct page *page, | |
3163 | get_extent_t *get_extent, | |
3164 | struct bio **bio, int mirror_num, | |
3165 | unsigned long *bio_flags, int rw) | |
3166 | { | |
3167 | struct inode *inode = page->mapping->host; | |
3168 | struct btrfs_ordered_extent *ordered; | |
3169 | u64 start = page_offset(page); | |
3170 | u64 end = start + PAGE_SIZE - 1; | |
3171 | int ret; | |
3172 | ||
3173 | while (1) { | |
3174 | lock_extent(tree, start, end); | |
3175 | ordered = btrfs_lookup_ordered_range(inode, start, | |
3176 | PAGE_SIZE); | |
3177 | if (!ordered) | |
3178 | break; | |
3179 | unlock_extent(tree, start, end); | |
3180 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3181 | btrfs_put_ordered_extent(ordered); | |
3182 | } | |
3183 | ||
3184 | ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num, | |
3185 | bio_flags, rw, NULL); | |
3186 | return ret; | |
3187 | } | |
3188 | ||
3189 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, | |
3190 | get_extent_t *get_extent, int mirror_num) | |
3191 | { | |
3192 | struct bio *bio = NULL; | |
3193 | unsigned long bio_flags = 0; | |
3194 | int ret; | |
3195 | ||
3196 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, | |
3197 | &bio_flags, READ); | |
3198 | if (bio) | |
3199 | ret = submit_one_bio(READ, bio, mirror_num, bio_flags); | |
3200 | return ret; | |
3201 | } | |
3202 | ||
3203 | static noinline void update_nr_written(struct page *page, | |
3204 | struct writeback_control *wbc, | |
3205 | unsigned long nr_written) | |
3206 | { | |
3207 | wbc->nr_to_write -= nr_written; | |
3208 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && | |
3209 | wbc->range_start == 0 && wbc->range_end == LLONG_MAX)) | |
3210 | page->mapping->writeback_index = page->index + nr_written; | |
3211 | } | |
3212 | ||
3213 | /* | |
3214 | * helper for __extent_writepage, doing all of the delayed allocation setup. | |
3215 | * | |
3216 | * This returns 1 if our fill_delalloc function did all the work required | |
3217 | * to write the page (copy into inline extent). In this case the IO has | |
3218 | * been started and the page is already unlocked. | |
3219 | * | |
3220 | * This returns 0 if all went well (page still locked) | |
3221 | * This returns < 0 if there were errors (page still locked) | |
3222 | */ | |
3223 | static noinline_for_stack int writepage_delalloc(struct inode *inode, | |
3224 | struct page *page, struct writeback_control *wbc, | |
3225 | struct extent_page_data *epd, | |
3226 | u64 delalloc_start, | |
3227 | unsigned long *nr_written) | |
3228 | { | |
3229 | struct extent_io_tree *tree = epd->tree; | |
3230 | u64 page_end = delalloc_start + PAGE_SIZE - 1; | |
3231 | u64 nr_delalloc; | |
3232 | u64 delalloc_to_write = 0; | |
3233 | u64 delalloc_end = 0; | |
3234 | int ret; | |
3235 | int page_started = 0; | |
3236 | ||
3237 | if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc) | |
3238 | return 0; | |
3239 | ||
3240 | while (delalloc_end < page_end) { | |
3241 | nr_delalloc = find_lock_delalloc_range(inode, tree, | |
3242 | page, | |
3243 | &delalloc_start, | |
3244 | &delalloc_end, | |
3245 | BTRFS_MAX_EXTENT_SIZE); | |
3246 | if (nr_delalloc == 0) { | |
3247 | delalloc_start = delalloc_end + 1; | |
3248 | continue; | |
3249 | } | |
3250 | ret = tree->ops->fill_delalloc(inode, page, | |
3251 | delalloc_start, | |
3252 | delalloc_end, | |
3253 | &page_started, | |
3254 | nr_written); | |
3255 | /* File system has been set read-only */ | |
3256 | if (ret) { | |
3257 | SetPageError(page); | |
3258 | /* fill_delalloc should be return < 0 for error | |
3259 | * but just in case, we use > 0 here meaning the | |
3260 | * IO is started, so we don't want to return > 0 | |
3261 | * unless things are going well. | |
3262 | */ | |
3263 | ret = ret < 0 ? ret : -EIO; | |
3264 | goto done; | |
3265 | } | |
3266 | /* | |
3267 | * delalloc_end is already one less than the total length, so | |
3268 | * we don't subtract one from PAGE_SIZE | |
3269 | */ | |
3270 | delalloc_to_write += (delalloc_end - delalloc_start + | |
3271 | PAGE_SIZE) >> PAGE_SHIFT; | |
3272 | delalloc_start = delalloc_end + 1; | |
3273 | } | |
3274 | if (wbc->nr_to_write < delalloc_to_write) { | |
3275 | int thresh = 8192; | |
3276 | ||
3277 | if (delalloc_to_write < thresh * 2) | |
3278 | thresh = delalloc_to_write; | |
3279 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3280 | thresh); | |
3281 | } | |
3282 | ||
3283 | /* did the fill delalloc function already unlock and start | |
3284 | * the IO? | |
3285 | */ | |
3286 | if (page_started) { | |
3287 | /* | |
3288 | * we've unlocked the page, so we can't update | |
3289 | * the mapping's writeback index, just update | |
3290 | * nr_to_write. | |
3291 | */ | |
3292 | wbc->nr_to_write -= *nr_written; | |
3293 | return 1; | |
3294 | } | |
3295 | ||
3296 | ret = 0; | |
3297 | ||
3298 | done: | |
3299 | return ret; | |
3300 | } | |
3301 | ||
3302 | /* | |
3303 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3304 | * and does the loop to map the page into extents and bios. | |
3305 | * | |
3306 | * We return 1 if the IO is started and the page is unlocked, | |
3307 | * 0 if all went well (page still locked) | |
3308 | * < 0 if there were errors (page still locked) | |
3309 | */ | |
3310 | static noinline_for_stack int __extent_writepage_io(struct inode *inode, | |
3311 | struct page *page, | |
3312 | struct writeback_control *wbc, | |
3313 | struct extent_page_data *epd, | |
3314 | loff_t i_size, | |
3315 | unsigned long nr_written, | |
3316 | int write_flags, int *nr_ret) | |
3317 | { | |
3318 | struct extent_io_tree *tree = epd->tree; | |
3319 | u64 start = page_offset(page); | |
3320 | u64 page_end = start + PAGE_SIZE - 1; | |
3321 | u64 end; | |
3322 | u64 cur = start; | |
3323 | u64 extent_offset; | |
3324 | u64 block_start; | |
3325 | u64 iosize; | |
3326 | sector_t sector; | |
3327 | struct extent_state *cached_state = NULL; | |
3328 | struct extent_map *em; | |
3329 | struct block_device *bdev; | |
3330 | size_t pg_offset = 0; | |
3331 | size_t blocksize; | |
3332 | int ret = 0; | |
3333 | int nr = 0; | |
3334 | bool compressed; | |
3335 | ||
3336 | if (tree->ops && tree->ops->writepage_start_hook) { | |
3337 | ret = tree->ops->writepage_start_hook(page, start, | |
3338 | page_end); | |
3339 | if (ret) { | |
3340 | /* Fixup worker will requeue */ | |
3341 | if (ret == -EBUSY) | |
3342 | wbc->pages_skipped++; | |
3343 | else | |
3344 | redirty_page_for_writepage(wbc, page); | |
3345 | ||
3346 | update_nr_written(page, wbc, nr_written); | |
3347 | unlock_page(page); | |
3348 | ret = 1; | |
3349 | goto done_unlocked; | |
3350 | } | |
3351 | } | |
3352 | ||
3353 | /* | |
3354 | * we don't want to touch the inode after unlocking the page, | |
3355 | * so we update the mapping writeback index now | |
3356 | */ | |
3357 | update_nr_written(page, wbc, nr_written + 1); | |
3358 | ||
3359 | end = page_end; | |
3360 | if (i_size <= start) { | |
3361 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
3362 | tree->ops->writepage_end_io_hook(page, start, | |
3363 | page_end, NULL, 1); | |
3364 | goto done; | |
3365 | } | |
3366 | ||
3367 | blocksize = inode->i_sb->s_blocksize; | |
3368 | ||
3369 | while (cur <= end) { | |
3370 | u64 em_end; | |
3371 | unsigned long max_nr; | |
3372 | ||
3373 | if (cur >= i_size) { | |
3374 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
3375 | tree->ops->writepage_end_io_hook(page, cur, | |
3376 | page_end, NULL, 1); | |
3377 | break; | |
3378 | } | |
3379 | em = epd->get_extent(inode, page, pg_offset, cur, | |
3380 | end - cur + 1, 1); | |
3381 | if (IS_ERR_OR_NULL(em)) { | |
3382 | SetPageError(page); | |
3383 | ret = PTR_ERR_OR_ZERO(em); | |
3384 | break; | |
3385 | } | |
3386 | ||
3387 | extent_offset = cur - em->start; | |
3388 | em_end = extent_map_end(em); | |
3389 | BUG_ON(em_end <= cur); | |
3390 | BUG_ON(end < cur); | |
3391 | iosize = min(em_end - cur, end - cur + 1); | |
3392 | iosize = ALIGN(iosize, blocksize); | |
3393 | sector = (em->block_start + extent_offset) >> 9; | |
3394 | bdev = em->bdev; | |
3395 | block_start = em->block_start; | |
3396 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
3397 | free_extent_map(em); | |
3398 | em = NULL; | |
3399 | ||
3400 | /* | |
3401 | * compressed and inline extents are written through other | |
3402 | * paths in the FS | |
3403 | */ | |
3404 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
3405 | block_start == EXTENT_MAP_INLINE) { | |
3406 | /* | |
3407 | * end_io notification does not happen here for | |
3408 | * compressed extents | |
3409 | */ | |
3410 | if (!compressed && tree->ops && | |
3411 | tree->ops->writepage_end_io_hook) | |
3412 | tree->ops->writepage_end_io_hook(page, cur, | |
3413 | cur + iosize - 1, | |
3414 | NULL, 1); | |
3415 | else if (compressed) { | |
3416 | /* we don't want to end_page_writeback on | |
3417 | * a compressed extent. this happens | |
3418 | * elsewhere | |
3419 | */ | |
3420 | nr++; | |
3421 | } | |
3422 | ||
3423 | cur += iosize; | |
3424 | pg_offset += iosize; | |
3425 | continue; | |
3426 | } | |
3427 | ||
3428 | max_nr = (i_size >> PAGE_SHIFT) + 1; | |
3429 | ||
3430 | set_range_writeback(tree, cur, cur + iosize - 1); | |
3431 | if (!PageWriteback(page)) { | |
3432 | btrfs_err(BTRFS_I(inode)->root->fs_info, | |
3433 | "page %lu not writeback, cur %llu end %llu", | |
3434 | page->index, cur, end); | |
3435 | } | |
3436 | ||
3437 | ret = submit_extent_page(write_flags, tree, wbc, page, | |
3438 | sector, iosize, pg_offset, | |
3439 | bdev, &epd->bio, max_nr, | |
3440 | end_bio_extent_writepage, | |
3441 | 0, 0, 0, false); | |
3442 | if (ret) | |
3443 | SetPageError(page); | |
3444 | ||
3445 | cur = cur + iosize; | |
3446 | pg_offset += iosize; | |
3447 | nr++; | |
3448 | } | |
3449 | done: | |
3450 | *nr_ret = nr; | |
3451 | ||
3452 | done_unlocked: | |
3453 | ||
3454 | /* drop our reference on any cached states */ | |
3455 | free_extent_state(cached_state); | |
3456 | return ret; | |
3457 | } | |
3458 | ||
3459 | /* | |
3460 | * the writepage semantics are similar to regular writepage. extent | |
3461 | * records are inserted to lock ranges in the tree, and as dirty areas | |
3462 | * are found, they are marked writeback. Then the lock bits are removed | |
3463 | * and the end_io handler clears the writeback ranges | |
3464 | */ | |
3465 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
3466 | void *data) | |
3467 | { | |
3468 | struct inode *inode = page->mapping->host; | |
3469 | struct extent_page_data *epd = data; | |
3470 | u64 start = page_offset(page); | |
3471 | u64 page_end = start + PAGE_SIZE - 1; | |
3472 | int ret; | |
3473 | int nr = 0; | |
3474 | size_t pg_offset = 0; | |
3475 | loff_t i_size = i_size_read(inode); | |
3476 | unsigned long end_index = i_size >> PAGE_SHIFT; | |
3477 | int write_flags; | |
3478 | unsigned long nr_written = 0; | |
3479 | ||
3480 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3481 | write_flags = WRITE_SYNC; | |
3482 | else | |
3483 | write_flags = WRITE; | |
3484 | ||
3485 | trace___extent_writepage(page, inode, wbc); | |
3486 | ||
3487 | WARN_ON(!PageLocked(page)); | |
3488 | ||
3489 | ClearPageError(page); | |
3490 | ||
3491 | pg_offset = i_size & (PAGE_SIZE - 1); | |
3492 | if (page->index > end_index || | |
3493 | (page->index == end_index && !pg_offset)) { | |
3494 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); | |
3495 | unlock_page(page); | |
3496 | return 0; | |
3497 | } | |
3498 | ||
3499 | if (page->index == end_index) { | |
3500 | char *userpage; | |
3501 | ||
3502 | userpage = kmap_atomic(page); | |
3503 | memset(userpage + pg_offset, 0, | |
3504 | PAGE_SIZE - pg_offset); | |
3505 | kunmap_atomic(userpage); | |
3506 | flush_dcache_page(page); | |
3507 | } | |
3508 | ||
3509 | pg_offset = 0; | |
3510 | ||
3511 | set_page_extent_mapped(page); | |
3512 | ||
3513 | ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written); | |
3514 | if (ret == 1) | |
3515 | goto done_unlocked; | |
3516 | if (ret) | |
3517 | goto done; | |
3518 | ||
3519 | ret = __extent_writepage_io(inode, page, wbc, epd, | |
3520 | i_size, nr_written, write_flags, &nr); | |
3521 | if (ret == 1) | |
3522 | goto done_unlocked; | |
3523 | ||
3524 | done: | |
3525 | if (nr == 0) { | |
3526 | /* make sure the mapping tag for page dirty gets cleared */ | |
3527 | set_page_writeback(page); | |
3528 | end_page_writeback(page); | |
3529 | } | |
3530 | if (PageError(page)) { | |
3531 | ret = ret < 0 ? ret : -EIO; | |
3532 | end_extent_writepage(page, ret, start, page_end); | |
3533 | } | |
3534 | unlock_page(page); | |
3535 | return ret; | |
3536 | ||
3537 | done_unlocked: | |
3538 | return 0; | |
3539 | } | |
3540 | ||
3541 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) | |
3542 | { | |
3543 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, | |
3544 | TASK_UNINTERRUPTIBLE); | |
3545 | } | |
3546 | ||
3547 | static noinline_for_stack int | |
3548 | lock_extent_buffer_for_io(struct extent_buffer *eb, | |
3549 | struct btrfs_fs_info *fs_info, | |
3550 | struct extent_page_data *epd) | |
3551 | { | |
3552 | unsigned long i, num_pages; | |
3553 | int flush = 0; | |
3554 | int ret = 0; | |
3555 | ||
3556 | if (!btrfs_try_tree_write_lock(eb)) { | |
3557 | flush = 1; | |
3558 | flush_write_bio(epd); | |
3559 | btrfs_tree_lock(eb); | |
3560 | } | |
3561 | ||
3562 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3563 | btrfs_tree_unlock(eb); | |
3564 | if (!epd->sync_io) | |
3565 | return 0; | |
3566 | if (!flush) { | |
3567 | flush_write_bio(epd); | |
3568 | flush = 1; | |
3569 | } | |
3570 | while (1) { | |
3571 | wait_on_extent_buffer_writeback(eb); | |
3572 | btrfs_tree_lock(eb); | |
3573 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3574 | break; | |
3575 | btrfs_tree_unlock(eb); | |
3576 | } | |
3577 | } | |
3578 | ||
3579 | /* | |
3580 | * We need to do this to prevent races in people who check if the eb is | |
3581 | * under IO since we can end up having no IO bits set for a short period | |
3582 | * of time. | |
3583 | */ | |
3584 | spin_lock(&eb->refs_lock); | |
3585 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { | |
3586 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
3587 | spin_unlock(&eb->refs_lock); | |
3588 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); | |
3589 | __percpu_counter_add(&fs_info->dirty_metadata_bytes, | |
3590 | -eb->len, | |
3591 | fs_info->dirty_metadata_batch); | |
3592 | ret = 1; | |
3593 | } else { | |
3594 | spin_unlock(&eb->refs_lock); | |
3595 | } | |
3596 | ||
3597 | btrfs_tree_unlock(eb); | |
3598 | ||
3599 | if (!ret) | |
3600 | return ret; | |
3601 | ||
3602 | num_pages = num_extent_pages(eb->start, eb->len); | |
3603 | for (i = 0; i < num_pages; i++) { | |
3604 | struct page *p = eb->pages[i]; | |
3605 | ||
3606 | if (!trylock_page(p)) { | |
3607 | if (!flush) { | |
3608 | flush_write_bio(epd); | |
3609 | flush = 1; | |
3610 | } | |
3611 | lock_page(p); | |
3612 | } | |
3613 | } | |
3614 | ||
3615 | return ret; | |
3616 | } | |
3617 | ||
3618 | static void end_extent_buffer_writeback(struct extent_buffer *eb) | |
3619 | { | |
3620 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
3621 | smp_mb__after_atomic(); | |
3622 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); | |
3623 | } | |
3624 | ||
3625 | static void set_btree_ioerr(struct page *page) | |
3626 | { | |
3627 | struct extent_buffer *eb = (struct extent_buffer *)page->private; | |
3628 | struct btrfs_inode *btree_ino = BTRFS_I(eb->fs_info->btree_inode); | |
3629 | ||
3630 | SetPageError(page); | |
3631 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) | |
3632 | return; | |
3633 | ||
3634 | /* | |
3635 | * If writeback for a btree extent that doesn't belong to a log tree | |
3636 | * failed, increment the counter transaction->eb_write_errors. | |
3637 | * We do this because while the transaction is running and before it's | |
3638 | * committing (when we call filemap_fdata[write|wait]_range against | |
3639 | * the btree inode), we might have | |
3640 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
3641 | * returns an error or an error happens during writeback, when we're | |
3642 | * committing the transaction we wouldn't know about it, since the pages | |
3643 | * can be no longer dirty nor marked anymore for writeback (if a | |
3644 | * subsequent modification to the extent buffer didn't happen before the | |
3645 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
3646 | * able to find the pages tagged with SetPageError at transaction | |
3647 | * commit time. So if this happens we must abort the transaction, | |
3648 | * otherwise we commit a super block with btree roots that point to | |
3649 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
3650 | * or the content of some node/leaf from a past generation that got | |
3651 | * cowed or deleted and is no longer valid. | |
3652 | * | |
3653 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
3654 | * not be enough - we need to distinguish between log tree extents vs | |
3655 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
3656 | * will catch and clear such errors in the mapping - and that call might | |
3657 | * be from a log sync and not from a transaction commit. Also, checking | |
3658 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
3659 | * not done and would not be reliable - the eb might have been released | |
3660 | * from memory and reading it back again means that flag would not be | |
3661 | * set (since it's a runtime flag, not persisted on disk). | |
3662 | * | |
3663 | * Using the flags below in the btree inode also makes us achieve the | |
3664 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
3665 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
3666 | * is called, the writeback for all dirty pages had already finished | |
3667 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
3668 | * filemap_fdatawait_range() would return success, as it could not know | |
3669 | * that writeback errors happened (the pages were no longer tagged for | |
3670 | * writeback). | |
3671 | */ | |
3672 | switch (eb->log_index) { | |
3673 | case -1: | |
3674 | set_bit(BTRFS_INODE_BTREE_ERR, &btree_ino->runtime_flags); | |
3675 | break; | |
3676 | case 0: | |
3677 | set_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags); | |
3678 | break; | |
3679 | case 1: | |
3680 | set_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags); | |
3681 | break; | |
3682 | default: | |
3683 | BUG(); /* unexpected, logic error */ | |
3684 | } | |
3685 | } | |
3686 | ||
3687 | static void end_bio_extent_buffer_writepage(struct bio *bio) | |
3688 | { | |
3689 | struct bio_vec *bvec; | |
3690 | struct extent_buffer *eb; | |
3691 | int i, done; | |
3692 | ||
3693 | bio_for_each_segment_all(bvec, bio, i) { | |
3694 | struct page *page = bvec->bv_page; | |
3695 | ||
3696 | eb = (struct extent_buffer *)page->private; | |
3697 | BUG_ON(!eb); | |
3698 | done = atomic_dec_and_test(&eb->io_pages); | |
3699 | ||
3700 | if (bio->bi_error || | |
3701 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { | |
3702 | ClearPageUptodate(page); | |
3703 | set_btree_ioerr(page); | |
3704 | } | |
3705 | ||
3706 | end_page_writeback(page); | |
3707 | ||
3708 | if (!done) | |
3709 | continue; | |
3710 | ||
3711 | end_extent_buffer_writeback(eb); | |
3712 | } | |
3713 | ||
3714 | bio_put(bio); | |
3715 | } | |
3716 | ||
3717 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, | |
3718 | struct btrfs_fs_info *fs_info, | |
3719 | struct writeback_control *wbc, | |
3720 | struct extent_page_data *epd) | |
3721 | { | |
3722 | struct block_device *bdev = fs_info->fs_devices->latest_bdev; | |
3723 | struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree; | |
3724 | u64 offset = eb->start; | |
3725 | unsigned long i, num_pages; | |
3726 | unsigned long bio_flags = 0; | |
3727 | int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META; | |
3728 | int ret = 0; | |
3729 | ||
3730 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); | |
3731 | num_pages = num_extent_pages(eb->start, eb->len); | |
3732 | atomic_set(&eb->io_pages, num_pages); | |
3733 | if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID) | |
3734 | bio_flags = EXTENT_BIO_TREE_LOG; | |
3735 | ||
3736 | for (i = 0; i < num_pages; i++) { | |
3737 | struct page *p = eb->pages[i]; | |
3738 | ||
3739 | clear_page_dirty_for_io(p); | |
3740 | set_page_writeback(p); | |
3741 | ret = submit_extent_page(rw, tree, wbc, p, offset >> 9, | |
3742 | PAGE_SIZE, 0, bdev, &epd->bio, | |
3743 | -1, end_bio_extent_buffer_writepage, | |
3744 | 0, epd->bio_flags, bio_flags, false); | |
3745 | epd->bio_flags = bio_flags; | |
3746 | if (ret) { | |
3747 | set_btree_ioerr(p); | |
3748 | end_page_writeback(p); | |
3749 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) | |
3750 | end_extent_buffer_writeback(eb); | |
3751 | ret = -EIO; | |
3752 | break; | |
3753 | } | |
3754 | offset += PAGE_SIZE; | |
3755 | update_nr_written(p, wbc, 1); | |
3756 | unlock_page(p); | |
3757 | } | |
3758 | ||
3759 | if (unlikely(ret)) { | |
3760 | for (; i < num_pages; i++) { | |
3761 | struct page *p = eb->pages[i]; | |
3762 | clear_page_dirty_for_io(p); | |
3763 | unlock_page(p); | |
3764 | } | |
3765 | } | |
3766 | ||
3767 | return ret; | |
3768 | } | |
3769 | ||
3770 | int btree_write_cache_pages(struct address_space *mapping, | |
3771 | struct writeback_control *wbc) | |
3772 | { | |
3773 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; | |
3774 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; | |
3775 | struct extent_buffer *eb, *prev_eb = NULL; | |
3776 | struct extent_page_data epd = { | |
3777 | .bio = NULL, | |
3778 | .tree = tree, | |
3779 | .extent_locked = 0, | |
3780 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
3781 | .bio_flags = 0, | |
3782 | }; | |
3783 | int ret = 0; | |
3784 | int done = 0; | |
3785 | int nr_to_write_done = 0; | |
3786 | struct pagevec pvec; | |
3787 | int nr_pages; | |
3788 | pgoff_t index; | |
3789 | pgoff_t end; /* Inclusive */ | |
3790 | int scanned = 0; | |
3791 | int tag; | |
3792 | ||
3793 | pagevec_init(&pvec, 0); | |
3794 | if (wbc->range_cyclic) { | |
3795 | index = mapping->writeback_index; /* Start from prev offset */ | |
3796 | end = -1; | |
3797 | } else { | |
3798 | index = wbc->range_start >> PAGE_SHIFT; | |
3799 | end = wbc->range_end >> PAGE_SHIFT; | |
3800 | scanned = 1; | |
3801 | } | |
3802 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3803 | tag = PAGECACHE_TAG_TOWRITE; | |
3804 | else | |
3805 | tag = PAGECACHE_TAG_DIRTY; | |
3806 | retry: | |
3807 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3808 | tag_pages_for_writeback(mapping, index, end); | |
3809 | while (!done && !nr_to_write_done && (index <= end) && | |
3810 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, | |
3811 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
3812 | unsigned i; | |
3813 | ||
3814 | scanned = 1; | |
3815 | for (i = 0; i < nr_pages; i++) { | |
3816 | struct page *page = pvec.pages[i]; | |
3817 | ||
3818 | if (!PagePrivate(page)) | |
3819 | continue; | |
3820 | ||
3821 | if (!wbc->range_cyclic && page->index > end) { | |
3822 | done = 1; | |
3823 | break; | |
3824 | } | |
3825 | ||
3826 | spin_lock(&mapping->private_lock); | |
3827 | if (!PagePrivate(page)) { | |
3828 | spin_unlock(&mapping->private_lock); | |
3829 | continue; | |
3830 | } | |
3831 | ||
3832 | eb = (struct extent_buffer *)page->private; | |
3833 | ||
3834 | /* | |
3835 | * Shouldn't happen and normally this would be a BUG_ON | |
3836 | * but no sense in crashing the users box for something | |
3837 | * we can survive anyway. | |
3838 | */ | |
3839 | if (WARN_ON(!eb)) { | |
3840 | spin_unlock(&mapping->private_lock); | |
3841 | continue; | |
3842 | } | |
3843 | ||
3844 | if (eb == prev_eb) { | |
3845 | spin_unlock(&mapping->private_lock); | |
3846 | continue; | |
3847 | } | |
3848 | ||
3849 | ret = atomic_inc_not_zero(&eb->refs); | |
3850 | spin_unlock(&mapping->private_lock); | |
3851 | if (!ret) | |
3852 | continue; | |
3853 | ||
3854 | prev_eb = eb; | |
3855 | ret = lock_extent_buffer_for_io(eb, fs_info, &epd); | |
3856 | if (!ret) { | |
3857 | free_extent_buffer(eb); | |
3858 | continue; | |
3859 | } | |
3860 | ||
3861 | ret = write_one_eb(eb, fs_info, wbc, &epd); | |
3862 | if (ret) { | |
3863 | done = 1; | |
3864 | free_extent_buffer(eb); | |
3865 | break; | |
3866 | } | |
3867 | free_extent_buffer(eb); | |
3868 | ||
3869 | /* | |
3870 | * the filesystem may choose to bump up nr_to_write. | |
3871 | * We have to make sure to honor the new nr_to_write | |
3872 | * at any time | |
3873 | */ | |
3874 | nr_to_write_done = wbc->nr_to_write <= 0; | |
3875 | } | |
3876 | pagevec_release(&pvec); | |
3877 | cond_resched(); | |
3878 | } | |
3879 | if (!scanned && !done) { | |
3880 | /* | |
3881 | * We hit the last page and there is more work to be done: wrap | |
3882 | * back to the start of the file | |
3883 | */ | |
3884 | scanned = 1; | |
3885 | index = 0; | |
3886 | goto retry; | |
3887 | } | |
3888 | flush_write_bio(&epd); | |
3889 | return ret; | |
3890 | } | |
3891 | ||
3892 | /** | |
3893 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. | |
3894 | * @mapping: address space structure to write | |
3895 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
3896 | * @writepage: function called for each page | |
3897 | * @data: data passed to writepage function | |
3898 | * | |
3899 | * If a page is already under I/O, write_cache_pages() skips it, even | |
3900 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
3901 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
3902 | * and msync() need to guarantee that all the data which was dirty at the time | |
3903 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
3904 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
3905 | * existing IO to complete. | |
3906 | */ | |
3907 | static int extent_write_cache_pages(struct extent_io_tree *tree, | |
3908 | struct address_space *mapping, | |
3909 | struct writeback_control *wbc, | |
3910 | writepage_t writepage, void *data, | |
3911 | void (*flush_fn)(void *)) | |
3912 | { | |
3913 | struct inode *inode = mapping->host; | |
3914 | int ret = 0; | |
3915 | int done = 0; | |
3916 | int err = 0; | |
3917 | int nr_to_write_done = 0; | |
3918 | struct pagevec pvec; | |
3919 | int nr_pages; | |
3920 | pgoff_t index; | |
3921 | pgoff_t end; /* Inclusive */ | |
3922 | int scanned = 0; | |
3923 | int tag; | |
3924 | ||
3925 | /* | |
3926 | * We have to hold onto the inode so that ordered extents can do their | |
3927 | * work when the IO finishes. The alternative to this is failing to add | |
3928 | * an ordered extent if the igrab() fails there and that is a huge pain | |
3929 | * to deal with, so instead just hold onto the inode throughout the | |
3930 | * writepages operation. If it fails here we are freeing up the inode | |
3931 | * anyway and we'd rather not waste our time writing out stuff that is | |
3932 | * going to be truncated anyway. | |
3933 | */ | |
3934 | if (!igrab(inode)) | |
3935 | return 0; | |
3936 | ||
3937 | pagevec_init(&pvec, 0); | |
3938 | if (wbc->range_cyclic) { | |
3939 | index = mapping->writeback_index; /* Start from prev offset */ | |
3940 | end = -1; | |
3941 | } else { | |
3942 | index = wbc->range_start >> PAGE_SHIFT; | |
3943 | end = wbc->range_end >> PAGE_SHIFT; | |
3944 | scanned = 1; | |
3945 | } | |
3946 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3947 | tag = PAGECACHE_TAG_TOWRITE; | |
3948 | else | |
3949 | tag = PAGECACHE_TAG_DIRTY; | |
3950 | retry: | |
3951 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3952 | tag_pages_for_writeback(mapping, index, end); | |
3953 | while (!done && !nr_to_write_done && (index <= end) && | |
3954 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, | |
3955 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
3956 | unsigned i; | |
3957 | ||
3958 | scanned = 1; | |
3959 | for (i = 0; i < nr_pages; i++) { | |
3960 | struct page *page = pvec.pages[i]; | |
3961 | ||
3962 | /* | |
3963 | * At this point we hold neither mapping->tree_lock nor | |
3964 | * lock on the page itself: the page may be truncated or | |
3965 | * invalidated (changing page->mapping to NULL), or even | |
3966 | * swizzled back from swapper_space to tmpfs file | |
3967 | * mapping | |
3968 | */ | |
3969 | if (!trylock_page(page)) { | |
3970 | flush_fn(data); | |
3971 | lock_page(page); | |
3972 | } | |
3973 | ||
3974 | if (unlikely(page->mapping != mapping)) { | |
3975 | unlock_page(page); | |
3976 | continue; | |
3977 | } | |
3978 | ||
3979 | if (!wbc->range_cyclic && page->index > end) { | |
3980 | done = 1; | |
3981 | unlock_page(page); | |
3982 | continue; | |
3983 | } | |
3984 | ||
3985 | if (wbc->sync_mode != WB_SYNC_NONE) { | |
3986 | if (PageWriteback(page)) | |
3987 | flush_fn(data); | |
3988 | wait_on_page_writeback(page); | |
3989 | } | |
3990 | ||
3991 | if (PageWriteback(page) || | |
3992 | !clear_page_dirty_for_io(page)) { | |
3993 | unlock_page(page); | |
3994 | continue; | |
3995 | } | |
3996 | ||
3997 | ret = (*writepage)(page, wbc, data); | |
3998 | ||
3999 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
4000 | unlock_page(page); | |
4001 | ret = 0; | |
4002 | } | |
4003 | if (!err && ret < 0) | |
4004 | err = ret; | |
4005 | ||
4006 | /* | |
4007 | * the filesystem may choose to bump up nr_to_write. | |
4008 | * We have to make sure to honor the new nr_to_write | |
4009 | * at any time | |
4010 | */ | |
4011 | nr_to_write_done = wbc->nr_to_write <= 0; | |
4012 | } | |
4013 | pagevec_release(&pvec); | |
4014 | cond_resched(); | |
4015 | } | |
4016 | if (!scanned && !done && !err) { | |
4017 | /* | |
4018 | * We hit the last page and there is more work to be done: wrap | |
4019 | * back to the start of the file | |
4020 | */ | |
4021 | scanned = 1; | |
4022 | index = 0; | |
4023 | goto retry; | |
4024 | } | |
4025 | btrfs_add_delayed_iput(inode); | |
4026 | return err; | |
4027 | } | |
4028 | ||
4029 | static void flush_epd_write_bio(struct extent_page_data *epd) | |
4030 | { | |
4031 | if (epd->bio) { | |
4032 | int rw = WRITE; | |
4033 | int ret; | |
4034 | ||
4035 | if (epd->sync_io) | |
4036 | rw = WRITE_SYNC; | |
4037 | ||
4038 | ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags); | |
4039 | BUG_ON(ret < 0); /* -ENOMEM */ | |
4040 | epd->bio = NULL; | |
4041 | } | |
4042 | } | |
4043 | ||
4044 | static noinline void flush_write_bio(void *data) | |
4045 | { | |
4046 | struct extent_page_data *epd = data; | |
4047 | flush_epd_write_bio(epd); | |
4048 | } | |
4049 | ||
4050 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, | |
4051 | get_extent_t *get_extent, | |
4052 | struct writeback_control *wbc) | |
4053 | { | |
4054 | int ret; | |
4055 | struct extent_page_data epd = { | |
4056 | .bio = NULL, | |
4057 | .tree = tree, | |
4058 | .get_extent = get_extent, | |
4059 | .extent_locked = 0, | |
4060 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
4061 | .bio_flags = 0, | |
4062 | }; | |
4063 | ||
4064 | ret = __extent_writepage(page, wbc, &epd); | |
4065 | ||
4066 | flush_epd_write_bio(&epd); | |
4067 | return ret; | |
4068 | } | |
4069 | ||
4070 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, | |
4071 | u64 start, u64 end, get_extent_t *get_extent, | |
4072 | int mode) | |
4073 | { | |
4074 | int ret = 0; | |
4075 | struct address_space *mapping = inode->i_mapping; | |
4076 | struct page *page; | |
4077 | unsigned long nr_pages = (end - start + PAGE_SIZE) >> | |
4078 | PAGE_SHIFT; | |
4079 | ||
4080 | struct extent_page_data epd = { | |
4081 | .bio = NULL, | |
4082 | .tree = tree, | |
4083 | .get_extent = get_extent, | |
4084 | .extent_locked = 1, | |
4085 | .sync_io = mode == WB_SYNC_ALL, | |
4086 | .bio_flags = 0, | |
4087 | }; | |
4088 | struct writeback_control wbc_writepages = { | |
4089 | .sync_mode = mode, | |
4090 | .nr_to_write = nr_pages * 2, | |
4091 | .range_start = start, | |
4092 | .range_end = end + 1, | |
4093 | }; | |
4094 | ||
4095 | while (start <= end) { | |
4096 | page = find_get_page(mapping, start >> PAGE_SHIFT); | |
4097 | if (clear_page_dirty_for_io(page)) | |
4098 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
4099 | else { | |
4100 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
4101 | tree->ops->writepage_end_io_hook(page, start, | |
4102 | start + PAGE_SIZE - 1, | |
4103 | NULL, 1); | |
4104 | unlock_page(page); | |
4105 | } | |
4106 | put_page(page); | |
4107 | start += PAGE_SIZE; | |
4108 | } | |
4109 | ||
4110 | flush_epd_write_bio(&epd); | |
4111 | return ret; | |
4112 | } | |
4113 | ||
4114 | int extent_writepages(struct extent_io_tree *tree, | |
4115 | struct address_space *mapping, | |
4116 | get_extent_t *get_extent, | |
4117 | struct writeback_control *wbc) | |
4118 | { | |
4119 | int ret = 0; | |
4120 | struct extent_page_data epd = { | |
4121 | .bio = NULL, | |
4122 | .tree = tree, | |
4123 | .get_extent = get_extent, | |
4124 | .extent_locked = 0, | |
4125 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
4126 | .bio_flags = 0, | |
4127 | }; | |
4128 | ||
4129 | ret = extent_write_cache_pages(tree, mapping, wbc, | |
4130 | __extent_writepage, &epd, | |
4131 | flush_write_bio); | |
4132 | flush_epd_write_bio(&epd); | |
4133 | return ret; | |
4134 | } | |
4135 | ||
4136 | int extent_readpages(struct extent_io_tree *tree, | |
4137 | struct address_space *mapping, | |
4138 | struct list_head *pages, unsigned nr_pages, | |
4139 | get_extent_t get_extent) | |
4140 | { | |
4141 | struct bio *bio = NULL; | |
4142 | unsigned page_idx; | |
4143 | unsigned long bio_flags = 0; | |
4144 | struct page *pagepool[16]; | |
4145 | struct page *page; | |
4146 | struct extent_map *em_cached = NULL; | |
4147 | int nr = 0; | |
4148 | u64 prev_em_start = (u64)-1; | |
4149 | ||
4150 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { | |
4151 | page = list_entry(pages->prev, struct page, lru); | |
4152 | ||
4153 | prefetchw(&page->flags); | |
4154 | list_del(&page->lru); | |
4155 | if (add_to_page_cache_lru(page, mapping, | |
4156 | page->index, GFP_NOFS)) { | |
4157 | put_page(page); | |
4158 | continue; | |
4159 | } | |
4160 | ||
4161 | pagepool[nr++] = page; | |
4162 | if (nr < ARRAY_SIZE(pagepool)) | |
4163 | continue; | |
4164 | __extent_readpages(tree, pagepool, nr, get_extent, &em_cached, | |
4165 | &bio, 0, &bio_flags, READ, &prev_em_start); | |
4166 | nr = 0; | |
4167 | } | |
4168 | if (nr) | |
4169 | __extent_readpages(tree, pagepool, nr, get_extent, &em_cached, | |
4170 | &bio, 0, &bio_flags, READ, &prev_em_start); | |
4171 | ||
4172 | if (em_cached) | |
4173 | free_extent_map(em_cached); | |
4174 | ||
4175 | BUG_ON(!list_empty(pages)); | |
4176 | if (bio) | |
4177 | return submit_one_bio(READ, bio, 0, bio_flags); | |
4178 | return 0; | |
4179 | } | |
4180 | ||
4181 | /* | |
4182 | * basic invalidatepage code, this waits on any locked or writeback | |
4183 | * ranges corresponding to the page, and then deletes any extent state | |
4184 | * records from the tree | |
4185 | */ | |
4186 | int extent_invalidatepage(struct extent_io_tree *tree, | |
4187 | struct page *page, unsigned long offset) | |
4188 | { | |
4189 | struct extent_state *cached_state = NULL; | |
4190 | u64 start = page_offset(page); | |
4191 | u64 end = start + PAGE_SIZE - 1; | |
4192 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
4193 | ||
4194 | start += ALIGN(offset, blocksize); | |
4195 | if (start > end) | |
4196 | return 0; | |
4197 | ||
4198 | lock_extent_bits(tree, start, end, &cached_state); | |
4199 | wait_on_page_writeback(page); | |
4200 | clear_extent_bit(tree, start, end, | |
4201 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | | |
4202 | EXTENT_DO_ACCOUNTING, | |
4203 | 1, 1, &cached_state, GFP_NOFS); | |
4204 | return 0; | |
4205 | } | |
4206 | ||
4207 | /* | |
4208 | * a helper for releasepage, this tests for areas of the page that | |
4209 | * are locked or under IO and drops the related state bits if it is safe | |
4210 | * to drop the page. | |
4211 | */ | |
4212 | static int try_release_extent_state(struct extent_map_tree *map, | |
4213 | struct extent_io_tree *tree, | |
4214 | struct page *page, gfp_t mask) | |
4215 | { | |
4216 | u64 start = page_offset(page); | |
4217 | u64 end = start + PAGE_SIZE - 1; | |
4218 | int ret = 1; | |
4219 | ||
4220 | if (test_range_bit(tree, start, end, | |
4221 | EXTENT_IOBITS, 0, NULL)) | |
4222 | ret = 0; | |
4223 | else { | |
4224 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
4225 | mask = GFP_NOFS; | |
4226 | /* | |
4227 | * at this point we can safely clear everything except the | |
4228 | * locked bit and the nodatasum bit | |
4229 | */ | |
4230 | ret = clear_extent_bit(tree, start, end, | |
4231 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), | |
4232 | 0, 0, NULL, mask); | |
4233 | ||
4234 | /* if clear_extent_bit failed for enomem reasons, | |
4235 | * we can't allow the release to continue. | |
4236 | */ | |
4237 | if (ret < 0) | |
4238 | ret = 0; | |
4239 | else | |
4240 | ret = 1; | |
4241 | } | |
4242 | return ret; | |
4243 | } | |
4244 | ||
4245 | /* | |
4246 | * a helper for releasepage. As long as there are no locked extents | |
4247 | * in the range corresponding to the page, both state records and extent | |
4248 | * map records are removed | |
4249 | */ | |
4250 | int try_release_extent_mapping(struct extent_map_tree *map, | |
4251 | struct extent_io_tree *tree, struct page *page, | |
4252 | gfp_t mask) | |
4253 | { | |
4254 | struct extent_map *em; | |
4255 | u64 start = page_offset(page); | |
4256 | u64 end = start + PAGE_SIZE - 1; | |
4257 | ||
4258 | if (gfpflags_allow_blocking(mask) && | |
4259 | page->mapping->host->i_size > SZ_16M) { | |
4260 | u64 len; | |
4261 | while (start <= end) { | |
4262 | len = end - start + 1; | |
4263 | write_lock(&map->lock); | |
4264 | em = lookup_extent_mapping(map, start, len); | |
4265 | if (!em) { | |
4266 | write_unlock(&map->lock); | |
4267 | break; | |
4268 | } | |
4269 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || | |
4270 | em->start != start) { | |
4271 | write_unlock(&map->lock); | |
4272 | free_extent_map(em); | |
4273 | break; | |
4274 | } | |
4275 | if (!test_range_bit(tree, em->start, | |
4276 | extent_map_end(em) - 1, | |
4277 | EXTENT_LOCKED | EXTENT_WRITEBACK, | |
4278 | 0, NULL)) { | |
4279 | remove_extent_mapping(map, em); | |
4280 | /* once for the rb tree */ | |
4281 | free_extent_map(em); | |
4282 | } | |
4283 | start = extent_map_end(em); | |
4284 | write_unlock(&map->lock); | |
4285 | ||
4286 | /* once for us */ | |
4287 | free_extent_map(em); | |
4288 | } | |
4289 | } | |
4290 | return try_release_extent_state(map, tree, page, mask); | |
4291 | } | |
4292 | ||
4293 | /* | |
4294 | * helper function for fiemap, which doesn't want to see any holes. | |
4295 | * This maps until we find something past 'last' | |
4296 | */ | |
4297 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
4298 | u64 offset, | |
4299 | u64 last, | |
4300 | get_extent_t *get_extent) | |
4301 | { | |
4302 | u64 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
4303 | struct extent_map *em; | |
4304 | u64 len; | |
4305 | ||
4306 | if (offset >= last) | |
4307 | return NULL; | |
4308 | ||
4309 | while (1) { | |
4310 | len = last - offset; | |
4311 | if (len == 0) | |
4312 | break; | |
4313 | len = ALIGN(len, sectorsize); | |
4314 | em = get_extent(inode, NULL, 0, offset, len, 0); | |
4315 | if (IS_ERR_OR_NULL(em)) | |
4316 | return em; | |
4317 | ||
4318 | /* if this isn't a hole return it */ | |
4319 | if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) && | |
4320 | em->block_start != EXTENT_MAP_HOLE) { | |
4321 | return em; | |
4322 | } | |
4323 | ||
4324 | /* this is a hole, advance to the next extent */ | |
4325 | offset = extent_map_end(em); | |
4326 | free_extent_map(em); | |
4327 | if (offset >= last) | |
4328 | break; | |
4329 | } | |
4330 | return NULL; | |
4331 | } | |
4332 | ||
4333 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, | |
4334 | __u64 start, __u64 len, get_extent_t *get_extent) | |
4335 | { | |
4336 | int ret = 0; | |
4337 | u64 off = start; | |
4338 | u64 max = start + len; | |
4339 | u32 flags = 0; | |
4340 | u32 found_type; | |
4341 | u64 last; | |
4342 | u64 last_for_get_extent = 0; | |
4343 | u64 disko = 0; | |
4344 | u64 isize = i_size_read(inode); | |
4345 | struct btrfs_key found_key; | |
4346 | struct extent_map *em = NULL; | |
4347 | struct extent_state *cached_state = NULL; | |
4348 | struct btrfs_path *path; | |
4349 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4350 | int end = 0; | |
4351 | u64 em_start = 0; | |
4352 | u64 em_len = 0; | |
4353 | u64 em_end = 0; | |
4354 | ||
4355 | if (len == 0) | |
4356 | return -EINVAL; | |
4357 | ||
4358 | path = btrfs_alloc_path(); | |
4359 | if (!path) | |
4360 | return -ENOMEM; | |
4361 | path->leave_spinning = 1; | |
4362 | ||
4363 | start = round_down(start, BTRFS_I(inode)->root->sectorsize); | |
4364 | len = round_up(max, BTRFS_I(inode)->root->sectorsize) - start; | |
4365 | ||
4366 | /* | |
4367 | * lookup the last file extent. We're not using i_size here | |
4368 | * because there might be preallocation past i_size | |
4369 | */ | |
4370 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1, | |
4371 | 0); | |
4372 | if (ret < 0) { | |
4373 | btrfs_free_path(path); | |
4374 | return ret; | |
4375 | } | |
4376 | WARN_ON(!ret); | |
4377 | path->slots[0]--; | |
4378 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
4379 | found_type = found_key.type; | |
4380 | ||
4381 | /* No extents, but there might be delalloc bits */ | |
4382 | if (found_key.objectid != btrfs_ino(inode) || | |
4383 | found_type != BTRFS_EXTENT_DATA_KEY) { | |
4384 | /* have to trust i_size as the end */ | |
4385 | last = (u64)-1; | |
4386 | last_for_get_extent = isize; | |
4387 | } else { | |
4388 | /* | |
4389 | * remember the start of the last extent. There are a | |
4390 | * bunch of different factors that go into the length of the | |
4391 | * extent, so its much less complex to remember where it started | |
4392 | */ | |
4393 | last = found_key.offset; | |
4394 | last_for_get_extent = last + 1; | |
4395 | } | |
4396 | btrfs_release_path(path); | |
4397 | ||
4398 | /* | |
4399 | * we might have some extents allocated but more delalloc past those | |
4400 | * extents. so, we trust isize unless the start of the last extent is | |
4401 | * beyond isize | |
4402 | */ | |
4403 | if (last < isize) { | |
4404 | last = (u64)-1; | |
4405 | last_for_get_extent = isize; | |
4406 | } | |
4407 | ||
4408 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, | |
4409 | &cached_state); | |
4410 | ||
4411 | em = get_extent_skip_holes(inode, start, last_for_get_extent, | |
4412 | get_extent); | |
4413 | if (!em) | |
4414 | goto out; | |
4415 | if (IS_ERR(em)) { | |
4416 | ret = PTR_ERR(em); | |
4417 | goto out; | |
4418 | } | |
4419 | ||
4420 | while (!end) { | |
4421 | u64 offset_in_extent = 0; | |
4422 | ||
4423 | /* break if the extent we found is outside the range */ | |
4424 | if (em->start >= max || extent_map_end(em) < off) | |
4425 | break; | |
4426 | ||
4427 | /* | |
4428 | * get_extent may return an extent that starts before our | |
4429 | * requested range. We have to make sure the ranges | |
4430 | * we return to fiemap always move forward and don't | |
4431 | * overlap, so adjust the offsets here | |
4432 | */ | |
4433 | em_start = max(em->start, off); | |
4434 | ||
4435 | /* | |
4436 | * record the offset from the start of the extent | |
4437 | * for adjusting the disk offset below. Only do this if the | |
4438 | * extent isn't compressed since our in ram offset may be past | |
4439 | * what we have actually allocated on disk. | |
4440 | */ | |
4441 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
4442 | offset_in_extent = em_start - em->start; | |
4443 | em_end = extent_map_end(em); | |
4444 | em_len = em_end - em_start; | |
4445 | disko = 0; | |
4446 | flags = 0; | |
4447 | ||
4448 | /* | |
4449 | * bump off for our next call to get_extent | |
4450 | */ | |
4451 | off = extent_map_end(em); | |
4452 | if (off >= max) | |
4453 | end = 1; | |
4454 | ||
4455 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { | |
4456 | end = 1; | |
4457 | flags |= FIEMAP_EXTENT_LAST; | |
4458 | } else if (em->block_start == EXTENT_MAP_INLINE) { | |
4459 | flags |= (FIEMAP_EXTENT_DATA_INLINE | | |
4460 | FIEMAP_EXTENT_NOT_ALIGNED); | |
4461 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { | |
4462 | flags |= (FIEMAP_EXTENT_DELALLOC | | |
4463 | FIEMAP_EXTENT_UNKNOWN); | |
4464 | } else if (fieinfo->fi_extents_max) { | |
4465 | u64 bytenr = em->block_start - | |
4466 | (em->start - em->orig_start); | |
4467 | ||
4468 | disko = em->block_start + offset_in_extent; | |
4469 | ||
4470 | /* | |
4471 | * As btrfs supports shared space, this information | |
4472 | * can be exported to userspace tools via | |
4473 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 | |
4474 | * then we're just getting a count and we can skip the | |
4475 | * lookup stuff. | |
4476 | */ | |
4477 | ret = btrfs_check_shared(NULL, root->fs_info, | |
4478 | root->objectid, | |
4479 | btrfs_ino(inode), bytenr); | |
4480 | if (ret < 0) | |
4481 | goto out_free; | |
4482 | if (ret) | |
4483 | flags |= FIEMAP_EXTENT_SHARED; | |
4484 | ret = 0; | |
4485 | } | |
4486 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
4487 | flags |= FIEMAP_EXTENT_ENCODED; | |
4488 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
4489 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
4490 | ||
4491 | free_extent_map(em); | |
4492 | em = NULL; | |
4493 | if ((em_start >= last) || em_len == (u64)-1 || | |
4494 | (last == (u64)-1 && isize <= em_end)) { | |
4495 | flags |= FIEMAP_EXTENT_LAST; | |
4496 | end = 1; | |
4497 | } | |
4498 | ||
4499 | /* now scan forward to see if this is really the last extent. */ | |
4500 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
4501 | get_extent); | |
4502 | if (IS_ERR(em)) { | |
4503 | ret = PTR_ERR(em); | |
4504 | goto out; | |
4505 | } | |
4506 | if (!em) { | |
4507 | flags |= FIEMAP_EXTENT_LAST; | |
4508 | end = 1; | |
4509 | } | |
4510 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, | |
4511 | em_len, flags); | |
4512 | if (ret) { | |
4513 | if (ret == 1) | |
4514 | ret = 0; | |
4515 | goto out_free; | |
4516 | } | |
4517 | } | |
4518 | out_free: | |
4519 | free_extent_map(em); | |
4520 | out: | |
4521 | btrfs_free_path(path); | |
4522 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1, | |
4523 | &cached_state, GFP_NOFS); | |
4524 | return ret; | |
4525 | } | |
4526 | ||
4527 | static void __free_extent_buffer(struct extent_buffer *eb) | |
4528 | { | |
4529 | btrfs_leak_debug_del(&eb->leak_list); | |
4530 | kmem_cache_free(extent_buffer_cache, eb); | |
4531 | } | |
4532 | ||
4533 | int extent_buffer_under_io(struct extent_buffer *eb) | |
4534 | { | |
4535 | return (atomic_read(&eb->io_pages) || | |
4536 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4537 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4538 | } | |
4539 | ||
4540 | /* | |
4541 | * Helper for releasing extent buffer page. | |
4542 | */ | |
4543 | static void btrfs_release_extent_buffer_page(struct extent_buffer *eb) | |
4544 | { | |
4545 | unsigned long index; | |
4546 | struct page *page; | |
4547 | int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4548 | ||
4549 | BUG_ON(extent_buffer_under_io(eb)); | |
4550 | ||
4551 | index = num_extent_pages(eb->start, eb->len); | |
4552 | if (index == 0) | |
4553 | return; | |
4554 | ||
4555 | do { | |
4556 | index--; | |
4557 | page = eb->pages[index]; | |
4558 | if (!page) | |
4559 | continue; | |
4560 | if (mapped) | |
4561 | spin_lock(&page->mapping->private_lock); | |
4562 | /* | |
4563 | * We do this since we'll remove the pages after we've | |
4564 | * removed the eb from the radix tree, so we could race | |
4565 | * and have this page now attached to the new eb. So | |
4566 | * only clear page_private if it's still connected to | |
4567 | * this eb. | |
4568 | */ | |
4569 | if (PagePrivate(page) && | |
4570 | page->private == (unsigned long)eb) { | |
4571 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4572 | BUG_ON(PageDirty(page)); | |
4573 | BUG_ON(PageWriteback(page)); | |
4574 | /* | |
4575 | * We need to make sure we haven't be attached | |
4576 | * to a new eb. | |
4577 | */ | |
4578 | ClearPagePrivate(page); | |
4579 | set_page_private(page, 0); | |
4580 | /* One for the page private */ | |
4581 | put_page(page); | |
4582 | } | |
4583 | ||
4584 | if (mapped) | |
4585 | spin_unlock(&page->mapping->private_lock); | |
4586 | ||
4587 | /* One for when we alloced the page */ | |
4588 | put_page(page); | |
4589 | } while (index != 0); | |
4590 | } | |
4591 | ||
4592 | /* | |
4593 | * Helper for releasing the extent buffer. | |
4594 | */ | |
4595 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4596 | { | |
4597 | btrfs_release_extent_buffer_page(eb); | |
4598 | __free_extent_buffer(eb); | |
4599 | } | |
4600 | ||
4601 | static struct extent_buffer * | |
4602 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
4603 | unsigned long len) | |
4604 | { | |
4605 | struct extent_buffer *eb = NULL; | |
4606 | ||
4607 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); | |
4608 | eb->start = start; | |
4609 | eb->len = len; | |
4610 | eb->fs_info = fs_info; | |
4611 | eb->bflags = 0; | |
4612 | rwlock_init(&eb->lock); | |
4613 | atomic_set(&eb->write_locks, 0); | |
4614 | atomic_set(&eb->read_locks, 0); | |
4615 | atomic_set(&eb->blocking_readers, 0); | |
4616 | atomic_set(&eb->blocking_writers, 0); | |
4617 | atomic_set(&eb->spinning_readers, 0); | |
4618 | atomic_set(&eb->spinning_writers, 0); | |
4619 | eb->lock_nested = 0; | |
4620 | init_waitqueue_head(&eb->write_lock_wq); | |
4621 | init_waitqueue_head(&eb->read_lock_wq); | |
4622 | ||
4623 | btrfs_leak_debug_add(&eb->leak_list, &buffers); | |
4624 | ||
4625 | spin_lock_init(&eb->refs_lock); | |
4626 | atomic_set(&eb->refs, 1); | |
4627 | atomic_set(&eb->io_pages, 0); | |
4628 | ||
4629 | /* | |
4630 | * Sanity checks, currently the maximum is 64k covered by 16x 4k pages | |
4631 | */ | |
4632 | BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE | |
4633 | > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
4634 | BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
4635 | ||
4636 | return eb; | |
4637 | } | |
4638 | ||
4639 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) | |
4640 | { | |
4641 | unsigned long i; | |
4642 | struct page *p; | |
4643 | struct extent_buffer *new; | |
4644 | unsigned long num_pages = num_extent_pages(src->start, src->len); | |
4645 | ||
4646 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); | |
4647 | if (new == NULL) | |
4648 | return NULL; | |
4649 | ||
4650 | for (i = 0; i < num_pages; i++) { | |
4651 | p = alloc_page(GFP_NOFS); | |
4652 | if (!p) { | |
4653 | btrfs_release_extent_buffer(new); | |
4654 | return NULL; | |
4655 | } | |
4656 | attach_extent_buffer_page(new, p); | |
4657 | WARN_ON(PageDirty(p)); | |
4658 | SetPageUptodate(p); | |
4659 | new->pages[i] = p; | |
4660 | } | |
4661 | ||
4662 | copy_extent_buffer(new, src, 0, 0, src->len); | |
4663 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); | |
4664 | set_bit(EXTENT_BUFFER_DUMMY, &new->bflags); | |
4665 | ||
4666 | return new; | |
4667 | } | |
4668 | ||
4669 | struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, | |
4670 | u64 start, unsigned long len) | |
4671 | { | |
4672 | struct extent_buffer *eb; | |
4673 | unsigned long num_pages; | |
4674 | unsigned long i; | |
4675 | ||
4676 | num_pages = num_extent_pages(start, len); | |
4677 | ||
4678 | eb = __alloc_extent_buffer(fs_info, start, len); | |
4679 | if (!eb) | |
4680 | return NULL; | |
4681 | ||
4682 | for (i = 0; i < num_pages; i++) { | |
4683 | eb->pages[i] = alloc_page(GFP_NOFS); | |
4684 | if (!eb->pages[i]) | |
4685 | goto err; | |
4686 | } | |
4687 | set_extent_buffer_uptodate(eb); | |
4688 | btrfs_set_header_nritems(eb, 0); | |
4689 | set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4690 | ||
4691 | return eb; | |
4692 | err: | |
4693 | for (; i > 0; i--) | |
4694 | __free_page(eb->pages[i - 1]); | |
4695 | __free_extent_buffer(eb); | |
4696 | return NULL; | |
4697 | } | |
4698 | ||
4699 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, | |
4700 | u64 start) | |
4701 | { | |
4702 | unsigned long len; | |
4703 | ||
4704 | if (!fs_info) { | |
4705 | /* | |
4706 | * Called only from tests that don't always have a fs_info | |
4707 | * available, but we know that nodesize is 4096 | |
4708 | */ | |
4709 | len = 4096; | |
4710 | } else { | |
4711 | len = fs_info->tree_root->nodesize; | |
4712 | } | |
4713 | ||
4714 | return __alloc_dummy_extent_buffer(fs_info, start, len); | |
4715 | } | |
4716 | ||
4717 | static void check_buffer_tree_ref(struct extent_buffer *eb) | |
4718 | { | |
4719 | int refs; | |
4720 | /* the ref bit is tricky. We have to make sure it is set | |
4721 | * if we have the buffer dirty. Otherwise the | |
4722 | * code to free a buffer can end up dropping a dirty | |
4723 | * page | |
4724 | * | |
4725 | * Once the ref bit is set, it won't go away while the | |
4726 | * buffer is dirty or in writeback, and it also won't | |
4727 | * go away while we have the reference count on the | |
4728 | * eb bumped. | |
4729 | * | |
4730 | * We can't just set the ref bit without bumping the | |
4731 | * ref on the eb because free_extent_buffer might | |
4732 | * see the ref bit and try to clear it. If this happens | |
4733 | * free_extent_buffer might end up dropping our original | |
4734 | * ref by mistake and freeing the page before we are able | |
4735 | * to add one more ref. | |
4736 | * | |
4737 | * So bump the ref count first, then set the bit. If someone | |
4738 | * beat us to it, drop the ref we added. | |
4739 | */ | |
4740 | refs = atomic_read(&eb->refs); | |
4741 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
4742 | return; | |
4743 | ||
4744 | spin_lock(&eb->refs_lock); | |
4745 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
4746 | atomic_inc(&eb->refs); | |
4747 | spin_unlock(&eb->refs_lock); | |
4748 | } | |
4749 | ||
4750 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, | |
4751 | struct page *accessed) | |
4752 | { | |
4753 | unsigned long num_pages, i; | |
4754 | ||
4755 | check_buffer_tree_ref(eb); | |
4756 | ||
4757 | num_pages = num_extent_pages(eb->start, eb->len); | |
4758 | for (i = 0; i < num_pages; i++) { | |
4759 | struct page *p = eb->pages[i]; | |
4760 | ||
4761 | if (p != accessed) | |
4762 | mark_page_accessed(p); | |
4763 | } | |
4764 | } | |
4765 | ||
4766 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, | |
4767 | u64 start) | |
4768 | { | |
4769 | struct extent_buffer *eb; | |
4770 | ||
4771 | rcu_read_lock(); | |
4772 | eb = radix_tree_lookup(&fs_info->buffer_radix, | |
4773 | start >> PAGE_SHIFT); | |
4774 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
4775 | rcu_read_unlock(); | |
4776 | /* | |
4777 | * Lock our eb's refs_lock to avoid races with | |
4778 | * free_extent_buffer. When we get our eb it might be flagged | |
4779 | * with EXTENT_BUFFER_STALE and another task running | |
4780 | * free_extent_buffer might have seen that flag set, | |
4781 | * eb->refs == 2, that the buffer isn't under IO (dirty and | |
4782 | * writeback flags not set) and it's still in the tree (flag | |
4783 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process | |
4784 | * of decrementing the extent buffer's reference count twice. | |
4785 | * So here we could race and increment the eb's reference count, | |
4786 | * clear its stale flag, mark it as dirty and drop our reference | |
4787 | * before the other task finishes executing free_extent_buffer, | |
4788 | * which would later result in an attempt to free an extent | |
4789 | * buffer that is dirty. | |
4790 | */ | |
4791 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
4792 | spin_lock(&eb->refs_lock); | |
4793 | spin_unlock(&eb->refs_lock); | |
4794 | } | |
4795 | mark_extent_buffer_accessed(eb, NULL); | |
4796 | return eb; | |
4797 | } | |
4798 | rcu_read_unlock(); | |
4799 | ||
4800 | return NULL; | |
4801 | } | |
4802 | ||
4803 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS | |
4804 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
4805 | u64 start) | |
4806 | { | |
4807 | struct extent_buffer *eb, *exists = NULL; | |
4808 | int ret; | |
4809 | ||
4810 | eb = find_extent_buffer(fs_info, start); | |
4811 | if (eb) | |
4812 | return eb; | |
4813 | eb = alloc_dummy_extent_buffer(fs_info, start); | |
4814 | if (!eb) | |
4815 | return NULL; | |
4816 | eb->fs_info = fs_info; | |
4817 | again: | |
4818 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); | |
4819 | if (ret) | |
4820 | goto free_eb; | |
4821 | spin_lock(&fs_info->buffer_lock); | |
4822 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
4823 | start >> PAGE_SHIFT, eb); | |
4824 | spin_unlock(&fs_info->buffer_lock); | |
4825 | radix_tree_preload_end(); | |
4826 | if (ret == -EEXIST) { | |
4827 | exists = find_extent_buffer(fs_info, start); | |
4828 | if (exists) | |
4829 | goto free_eb; | |
4830 | else | |
4831 | goto again; | |
4832 | } | |
4833 | check_buffer_tree_ref(eb); | |
4834 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
4835 | ||
4836 | /* | |
4837 | * We will free dummy extent buffer's if they come into | |
4838 | * free_extent_buffer with a ref count of 2, but if we are using this we | |
4839 | * want the buffers to stay in memory until we're done with them, so | |
4840 | * bump the ref count again. | |
4841 | */ | |
4842 | atomic_inc(&eb->refs); | |
4843 | return eb; | |
4844 | free_eb: | |
4845 | btrfs_release_extent_buffer(eb); | |
4846 | return exists; | |
4847 | } | |
4848 | #endif | |
4849 | ||
4850 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, | |
4851 | u64 start) | |
4852 | { | |
4853 | unsigned long len = fs_info->tree_root->nodesize; | |
4854 | unsigned long num_pages = num_extent_pages(start, len); | |
4855 | unsigned long i; | |
4856 | unsigned long index = start >> PAGE_SHIFT; | |
4857 | struct extent_buffer *eb; | |
4858 | struct extent_buffer *exists = NULL; | |
4859 | struct page *p; | |
4860 | struct address_space *mapping = fs_info->btree_inode->i_mapping; | |
4861 | int uptodate = 1; | |
4862 | int ret; | |
4863 | ||
4864 | eb = find_extent_buffer(fs_info, start); | |
4865 | if (eb) | |
4866 | return eb; | |
4867 | ||
4868 | eb = __alloc_extent_buffer(fs_info, start, len); | |
4869 | if (!eb) | |
4870 | return NULL; | |
4871 | ||
4872 | for (i = 0; i < num_pages; i++, index++) { | |
4873 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); | |
4874 | if (!p) | |
4875 | goto free_eb; | |
4876 | ||
4877 | spin_lock(&mapping->private_lock); | |
4878 | if (PagePrivate(p)) { | |
4879 | /* | |
4880 | * We could have already allocated an eb for this page | |
4881 | * and attached one so lets see if we can get a ref on | |
4882 | * the existing eb, and if we can we know it's good and | |
4883 | * we can just return that one, else we know we can just | |
4884 | * overwrite page->private. | |
4885 | */ | |
4886 | exists = (struct extent_buffer *)p->private; | |
4887 | if (atomic_inc_not_zero(&exists->refs)) { | |
4888 | spin_unlock(&mapping->private_lock); | |
4889 | unlock_page(p); | |
4890 | put_page(p); | |
4891 | mark_extent_buffer_accessed(exists, p); | |
4892 | goto free_eb; | |
4893 | } | |
4894 | exists = NULL; | |
4895 | ||
4896 | /* | |
4897 | * Do this so attach doesn't complain and we need to | |
4898 | * drop the ref the old guy had. | |
4899 | */ | |
4900 | ClearPagePrivate(p); | |
4901 | WARN_ON(PageDirty(p)); | |
4902 | put_page(p); | |
4903 | } | |
4904 | attach_extent_buffer_page(eb, p); | |
4905 | spin_unlock(&mapping->private_lock); | |
4906 | WARN_ON(PageDirty(p)); | |
4907 | eb->pages[i] = p; | |
4908 | if (!PageUptodate(p)) | |
4909 | uptodate = 0; | |
4910 | ||
4911 | /* | |
4912 | * see below about how we avoid a nasty race with release page | |
4913 | * and why we unlock later | |
4914 | */ | |
4915 | } | |
4916 | if (uptodate) | |
4917 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
4918 | again: | |
4919 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); | |
4920 | if (ret) | |
4921 | goto free_eb; | |
4922 | ||
4923 | spin_lock(&fs_info->buffer_lock); | |
4924 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
4925 | start >> PAGE_SHIFT, eb); | |
4926 | spin_unlock(&fs_info->buffer_lock); | |
4927 | radix_tree_preload_end(); | |
4928 | if (ret == -EEXIST) { | |
4929 | exists = find_extent_buffer(fs_info, start); | |
4930 | if (exists) | |
4931 | goto free_eb; | |
4932 | else | |
4933 | goto again; | |
4934 | } | |
4935 | /* add one reference for the tree */ | |
4936 | check_buffer_tree_ref(eb); | |
4937 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
4938 | ||
4939 | /* | |
4940 | * there is a race where release page may have | |
4941 | * tried to find this extent buffer in the radix | |
4942 | * but failed. It will tell the VM it is safe to | |
4943 | * reclaim the, and it will clear the page private bit. | |
4944 | * We must make sure to set the page private bit properly | |
4945 | * after the extent buffer is in the radix tree so | |
4946 | * it doesn't get lost | |
4947 | */ | |
4948 | SetPageChecked(eb->pages[0]); | |
4949 | for (i = 1; i < num_pages; i++) { | |
4950 | p = eb->pages[i]; | |
4951 | ClearPageChecked(p); | |
4952 | unlock_page(p); | |
4953 | } | |
4954 | unlock_page(eb->pages[0]); | |
4955 | return eb; | |
4956 | ||
4957 | free_eb: | |
4958 | WARN_ON(!atomic_dec_and_test(&eb->refs)); | |
4959 | for (i = 0; i < num_pages; i++) { | |
4960 | if (eb->pages[i]) | |
4961 | unlock_page(eb->pages[i]); | |
4962 | } | |
4963 | ||
4964 | btrfs_release_extent_buffer(eb); | |
4965 | return exists; | |
4966 | } | |
4967 | ||
4968 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) | |
4969 | { | |
4970 | struct extent_buffer *eb = | |
4971 | container_of(head, struct extent_buffer, rcu_head); | |
4972 | ||
4973 | __free_extent_buffer(eb); | |
4974 | } | |
4975 | ||
4976 | /* Expects to have eb->eb_lock already held */ | |
4977 | static int release_extent_buffer(struct extent_buffer *eb) | |
4978 | { | |
4979 | WARN_ON(atomic_read(&eb->refs) == 0); | |
4980 | if (atomic_dec_and_test(&eb->refs)) { | |
4981 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { | |
4982 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
4983 | ||
4984 | spin_unlock(&eb->refs_lock); | |
4985 | ||
4986 | spin_lock(&fs_info->buffer_lock); | |
4987 | radix_tree_delete(&fs_info->buffer_radix, | |
4988 | eb->start >> PAGE_SHIFT); | |
4989 | spin_unlock(&fs_info->buffer_lock); | |
4990 | } else { | |
4991 | spin_unlock(&eb->refs_lock); | |
4992 | } | |
4993 | ||
4994 | /* Should be safe to release our pages at this point */ | |
4995 | btrfs_release_extent_buffer_page(eb); | |
4996 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS | |
4997 | if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) { | |
4998 | __free_extent_buffer(eb); | |
4999 | return 1; | |
5000 | } | |
5001 | #endif | |
5002 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); | |
5003 | return 1; | |
5004 | } | |
5005 | spin_unlock(&eb->refs_lock); | |
5006 | ||
5007 | return 0; | |
5008 | } | |
5009 | ||
5010 | void free_extent_buffer(struct extent_buffer *eb) | |
5011 | { | |
5012 | int refs; | |
5013 | int old; | |
5014 | if (!eb) | |
5015 | return; | |
5016 | ||
5017 | while (1) { | |
5018 | refs = atomic_read(&eb->refs); | |
5019 | if (refs <= 3) | |
5020 | break; | |
5021 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
5022 | if (old == refs) | |
5023 | return; | |
5024 | } | |
5025 | ||
5026 | spin_lock(&eb->refs_lock); | |
5027 | if (atomic_read(&eb->refs) == 2 && | |
5028 | test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) | |
5029 | atomic_dec(&eb->refs); | |
5030 | ||
5031 | if (atomic_read(&eb->refs) == 2 && | |
5032 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
5033 | !extent_buffer_under_io(eb) && | |
5034 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
5035 | atomic_dec(&eb->refs); | |
5036 | ||
5037 | /* | |
5038 | * I know this is terrible, but it's temporary until we stop tracking | |
5039 | * the uptodate bits and such for the extent buffers. | |
5040 | */ | |
5041 | release_extent_buffer(eb); | |
5042 | } | |
5043 | ||
5044 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
5045 | { | |
5046 | if (!eb) | |
5047 | return; | |
5048 | ||
5049 | spin_lock(&eb->refs_lock); | |
5050 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
5051 | ||
5052 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && | |
5053 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
5054 | atomic_dec(&eb->refs); | |
5055 | release_extent_buffer(eb); | |
5056 | } | |
5057 | ||
5058 | void clear_extent_buffer_dirty(struct extent_buffer *eb) | |
5059 | { | |
5060 | unsigned long i; | |
5061 | unsigned long num_pages; | |
5062 | struct page *page; | |
5063 | ||
5064 | num_pages = num_extent_pages(eb->start, eb->len); | |
5065 | ||
5066 | for (i = 0; i < num_pages; i++) { | |
5067 | page = eb->pages[i]; | |
5068 | if (!PageDirty(page)) | |
5069 | continue; | |
5070 | ||
5071 | lock_page(page); | |
5072 | WARN_ON(!PagePrivate(page)); | |
5073 | ||
5074 | clear_page_dirty_for_io(page); | |
5075 | spin_lock_irq(&page->mapping->tree_lock); | |
5076 | if (!PageDirty(page)) { | |
5077 | radix_tree_tag_clear(&page->mapping->page_tree, | |
5078 | page_index(page), | |
5079 | PAGECACHE_TAG_DIRTY); | |
5080 | } | |
5081 | spin_unlock_irq(&page->mapping->tree_lock); | |
5082 | ClearPageError(page); | |
5083 | unlock_page(page); | |
5084 | } | |
5085 | WARN_ON(atomic_read(&eb->refs) == 0); | |
5086 | } | |
5087 | ||
5088 | int set_extent_buffer_dirty(struct extent_buffer *eb) | |
5089 | { | |
5090 | unsigned long i; | |
5091 | unsigned long num_pages; | |
5092 | int was_dirty = 0; | |
5093 | ||
5094 | check_buffer_tree_ref(eb); | |
5095 | ||
5096 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); | |
5097 | ||
5098 | num_pages = num_extent_pages(eb->start, eb->len); | |
5099 | WARN_ON(atomic_read(&eb->refs) == 0); | |
5100 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); | |
5101 | ||
5102 | for (i = 0; i < num_pages; i++) | |
5103 | set_page_dirty(eb->pages[i]); | |
5104 | return was_dirty; | |
5105 | } | |
5106 | ||
5107 | void clear_extent_buffer_uptodate(struct extent_buffer *eb) | |
5108 | { | |
5109 | unsigned long i; | |
5110 | struct page *page; | |
5111 | unsigned long num_pages; | |
5112 | ||
5113 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
5114 | num_pages = num_extent_pages(eb->start, eb->len); | |
5115 | for (i = 0; i < num_pages; i++) { | |
5116 | page = eb->pages[i]; | |
5117 | if (page) | |
5118 | ClearPageUptodate(page); | |
5119 | } | |
5120 | } | |
5121 | ||
5122 | void set_extent_buffer_uptodate(struct extent_buffer *eb) | |
5123 | { | |
5124 | unsigned long i; | |
5125 | struct page *page; | |
5126 | unsigned long num_pages; | |
5127 | ||
5128 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
5129 | num_pages = num_extent_pages(eb->start, eb->len); | |
5130 | for (i = 0; i < num_pages; i++) { | |
5131 | page = eb->pages[i]; | |
5132 | SetPageUptodate(page); | |
5133 | } | |
5134 | } | |
5135 | ||
5136 | int extent_buffer_uptodate(struct extent_buffer *eb) | |
5137 | { | |
5138 | return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
5139 | } | |
5140 | ||
5141 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
5142 | struct extent_buffer *eb, u64 start, int wait, | |
5143 | get_extent_t *get_extent, int mirror_num) | |
5144 | { | |
5145 | unsigned long i; | |
5146 | unsigned long start_i; | |
5147 | struct page *page; | |
5148 | int err; | |
5149 | int ret = 0; | |
5150 | int locked_pages = 0; | |
5151 | int all_uptodate = 1; | |
5152 | unsigned long num_pages; | |
5153 | unsigned long num_reads = 0; | |
5154 | struct bio *bio = NULL; | |
5155 | unsigned long bio_flags = 0; | |
5156 | ||
5157 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) | |
5158 | return 0; | |
5159 | ||
5160 | if (start) { | |
5161 | WARN_ON(start < eb->start); | |
5162 | start_i = (start >> PAGE_SHIFT) - | |
5163 | (eb->start >> PAGE_SHIFT); | |
5164 | } else { | |
5165 | start_i = 0; | |
5166 | } | |
5167 | ||
5168 | num_pages = num_extent_pages(eb->start, eb->len); | |
5169 | for (i = start_i; i < num_pages; i++) { | |
5170 | page = eb->pages[i]; | |
5171 | if (wait == WAIT_NONE) { | |
5172 | if (!trylock_page(page)) | |
5173 | goto unlock_exit; | |
5174 | } else { | |
5175 | lock_page(page); | |
5176 | } | |
5177 | locked_pages++; | |
5178 | if (!PageUptodate(page)) { | |
5179 | num_reads++; | |
5180 | all_uptodate = 0; | |
5181 | } | |
5182 | } | |
5183 | if (all_uptodate) { | |
5184 | if (start_i == 0) | |
5185 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
5186 | goto unlock_exit; | |
5187 | } | |
5188 | ||
5189 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); | |
5190 | eb->read_mirror = 0; | |
5191 | atomic_set(&eb->io_pages, num_reads); | |
5192 | for (i = start_i; i < num_pages; i++) { | |
5193 | page = eb->pages[i]; | |
5194 | if (!PageUptodate(page)) { | |
5195 | ClearPageError(page); | |
5196 | err = __extent_read_full_page(tree, page, | |
5197 | get_extent, &bio, | |
5198 | mirror_num, &bio_flags, | |
5199 | READ | REQ_META); | |
5200 | if (err) | |
5201 | ret = err; | |
5202 | } else { | |
5203 | unlock_page(page); | |
5204 | } | |
5205 | } | |
5206 | ||
5207 | if (bio) { | |
5208 | err = submit_one_bio(READ | REQ_META, bio, mirror_num, | |
5209 | bio_flags); | |
5210 | if (err) | |
5211 | return err; | |
5212 | } | |
5213 | ||
5214 | if (ret || wait != WAIT_COMPLETE) | |
5215 | return ret; | |
5216 | ||
5217 | for (i = start_i; i < num_pages; i++) { | |
5218 | page = eb->pages[i]; | |
5219 | wait_on_page_locked(page); | |
5220 | if (!PageUptodate(page)) | |
5221 | ret = -EIO; | |
5222 | } | |
5223 | ||
5224 | return ret; | |
5225 | ||
5226 | unlock_exit: | |
5227 | i = start_i; | |
5228 | while (locked_pages > 0) { | |
5229 | page = eb->pages[i]; | |
5230 | i++; | |
5231 | unlock_page(page); | |
5232 | locked_pages--; | |
5233 | } | |
5234 | return ret; | |
5235 | } | |
5236 | ||
5237 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
5238 | unsigned long start, | |
5239 | unsigned long len) | |
5240 | { | |
5241 | size_t cur; | |
5242 | size_t offset; | |
5243 | struct page *page; | |
5244 | char *kaddr; | |
5245 | char *dst = (char *)dstv; | |
5246 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); | |
5247 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
5248 | ||
5249 | WARN_ON(start > eb->len); | |
5250 | WARN_ON(start + len > eb->start + eb->len); | |
5251 | ||
5252 | offset = (start_offset + start) & (PAGE_SIZE - 1); | |
5253 | ||
5254 | while (len > 0) { | |
5255 | page = eb->pages[i]; | |
5256 | ||
5257 | cur = min(len, (PAGE_SIZE - offset)); | |
5258 | kaddr = page_address(page); | |
5259 | memcpy(dst, kaddr + offset, cur); | |
5260 | ||
5261 | dst += cur; | |
5262 | len -= cur; | |
5263 | offset = 0; | |
5264 | i++; | |
5265 | } | |
5266 | } | |
5267 | ||
5268 | int read_extent_buffer_to_user(struct extent_buffer *eb, void __user *dstv, | |
5269 | unsigned long start, | |
5270 | unsigned long len) | |
5271 | { | |
5272 | size_t cur; | |
5273 | size_t offset; | |
5274 | struct page *page; | |
5275 | char *kaddr; | |
5276 | char __user *dst = (char __user *)dstv; | |
5277 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); | |
5278 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
5279 | int ret = 0; | |
5280 | ||
5281 | WARN_ON(start > eb->len); | |
5282 | WARN_ON(start + len > eb->start + eb->len); | |
5283 | ||
5284 | offset = (start_offset + start) & (PAGE_SIZE - 1); | |
5285 | ||
5286 | while (len > 0) { | |
5287 | page = eb->pages[i]; | |
5288 | ||
5289 | cur = min(len, (PAGE_SIZE - offset)); | |
5290 | kaddr = page_address(page); | |
5291 | if (copy_to_user(dst, kaddr + offset, cur)) { | |
5292 | ret = -EFAULT; | |
5293 | break; | |
5294 | } | |
5295 | ||
5296 | dst += cur; | |
5297 | len -= cur; | |
5298 | offset = 0; | |
5299 | i++; | |
5300 | } | |
5301 | ||
5302 | return ret; | |
5303 | } | |
5304 | ||
5305 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
5306 | unsigned long min_len, char **map, | |
5307 | unsigned long *map_start, | |
5308 | unsigned long *map_len) | |
5309 | { | |
5310 | size_t offset = start & (PAGE_SIZE - 1); | |
5311 | char *kaddr; | |
5312 | struct page *p; | |
5313 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); | |
5314 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
5315 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
5316 | PAGE_SHIFT; | |
5317 | ||
5318 | if (i != end_i) | |
5319 | return -EINVAL; | |
5320 | ||
5321 | if (i == 0) { | |
5322 | offset = start_offset; | |
5323 | *map_start = 0; | |
5324 | } else { | |
5325 | offset = 0; | |
5326 | *map_start = ((u64)i << PAGE_SHIFT) - start_offset; | |
5327 | } | |
5328 | ||
5329 | if (start + min_len > eb->len) { | |
5330 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, " | |
5331 | "wanted %lu %lu\n", | |
5332 | eb->start, eb->len, start, min_len); | |
5333 | return -EINVAL; | |
5334 | } | |
5335 | ||
5336 | p = eb->pages[i]; | |
5337 | kaddr = page_address(p); | |
5338 | *map = kaddr + offset; | |
5339 | *map_len = PAGE_SIZE - offset; | |
5340 | return 0; | |
5341 | } | |
5342 | ||
5343 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, | |
5344 | unsigned long start, | |
5345 | unsigned long len) | |
5346 | { | |
5347 | size_t cur; | |
5348 | size_t offset; | |
5349 | struct page *page; | |
5350 | char *kaddr; | |
5351 | char *ptr = (char *)ptrv; | |
5352 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); | |
5353 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
5354 | int ret = 0; | |
5355 | ||
5356 | WARN_ON(start > eb->len); | |
5357 | WARN_ON(start + len > eb->start + eb->len); | |
5358 | ||
5359 | offset = (start_offset + start) & (PAGE_SIZE - 1); | |
5360 | ||
5361 | while (len > 0) { | |
5362 | page = eb->pages[i]; | |
5363 | ||
5364 | cur = min(len, (PAGE_SIZE - offset)); | |
5365 | ||
5366 | kaddr = page_address(page); | |
5367 | ret = memcmp(ptr, kaddr + offset, cur); | |
5368 | if (ret) | |
5369 | break; | |
5370 | ||
5371 | ptr += cur; | |
5372 | len -= cur; | |
5373 | offset = 0; | |
5374 | i++; | |
5375 | } | |
5376 | return ret; | |
5377 | } | |
5378 | ||
5379 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
5380 | unsigned long start, unsigned long len) | |
5381 | { | |
5382 | size_t cur; | |
5383 | size_t offset; | |
5384 | struct page *page; | |
5385 | char *kaddr; | |
5386 | char *src = (char *)srcv; | |
5387 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); | |
5388 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
5389 | ||
5390 | WARN_ON(start > eb->len); | |
5391 | WARN_ON(start + len > eb->start + eb->len); | |
5392 | ||
5393 | offset = (start_offset + start) & (PAGE_SIZE - 1); | |
5394 | ||
5395 | while (len > 0) { | |
5396 | page = eb->pages[i]; | |
5397 | WARN_ON(!PageUptodate(page)); | |
5398 | ||
5399 | cur = min(len, PAGE_SIZE - offset); | |
5400 | kaddr = page_address(page); | |
5401 | memcpy(kaddr + offset, src, cur); | |
5402 | ||
5403 | src += cur; | |
5404 | len -= cur; | |
5405 | offset = 0; | |
5406 | i++; | |
5407 | } | |
5408 | } | |
5409 | ||
5410 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
5411 | unsigned long start, unsigned long len) | |
5412 | { | |
5413 | size_t cur; | |
5414 | size_t offset; | |
5415 | struct page *page; | |
5416 | char *kaddr; | |
5417 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); | |
5418 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
5419 | ||
5420 | WARN_ON(start > eb->len); | |
5421 | WARN_ON(start + len > eb->start + eb->len); | |
5422 | ||
5423 | offset = (start_offset + start) & (PAGE_SIZE - 1); | |
5424 | ||
5425 | while (len > 0) { | |
5426 | page = eb->pages[i]; | |
5427 | WARN_ON(!PageUptodate(page)); | |
5428 | ||
5429 | cur = min(len, PAGE_SIZE - offset); | |
5430 | kaddr = page_address(page); | |
5431 | memset(kaddr + offset, c, cur); | |
5432 | ||
5433 | len -= cur; | |
5434 | offset = 0; | |
5435 | i++; | |
5436 | } | |
5437 | } | |
5438 | ||
5439 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
5440 | unsigned long dst_offset, unsigned long src_offset, | |
5441 | unsigned long len) | |
5442 | { | |
5443 | u64 dst_len = dst->len; | |
5444 | size_t cur; | |
5445 | size_t offset; | |
5446 | struct page *page; | |
5447 | char *kaddr; | |
5448 | size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1); | |
5449 | unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT; | |
5450 | ||
5451 | WARN_ON(src->len != dst_len); | |
5452 | ||
5453 | offset = (start_offset + dst_offset) & | |
5454 | (PAGE_SIZE - 1); | |
5455 | ||
5456 | while (len > 0) { | |
5457 | page = dst->pages[i]; | |
5458 | WARN_ON(!PageUptodate(page)); | |
5459 | ||
5460 | cur = min(len, (unsigned long)(PAGE_SIZE - offset)); | |
5461 | ||
5462 | kaddr = page_address(page); | |
5463 | read_extent_buffer(src, kaddr + offset, src_offset, cur); | |
5464 | ||
5465 | src_offset += cur; | |
5466 | len -= cur; | |
5467 | offset = 0; | |
5468 | i++; | |
5469 | } | |
5470 | } | |
5471 | ||
5472 | /* | |
5473 | * The extent buffer bitmap operations are done with byte granularity because | |
5474 | * bitmap items are not guaranteed to be aligned to a word and therefore a | |
5475 | * single word in a bitmap may straddle two pages in the extent buffer. | |
5476 | */ | |
5477 | #define BIT_BYTE(nr) ((nr) / BITS_PER_BYTE) | |
5478 | #define BYTE_MASK ((1 << BITS_PER_BYTE) - 1) | |
5479 | #define BITMAP_FIRST_BYTE_MASK(start) \ | |
5480 | ((BYTE_MASK << ((start) & (BITS_PER_BYTE - 1))) & BYTE_MASK) | |
5481 | #define BITMAP_LAST_BYTE_MASK(nbits) \ | |
5482 | (BYTE_MASK >> (-(nbits) & (BITS_PER_BYTE - 1))) | |
5483 | ||
5484 | /* | |
5485 | * eb_bitmap_offset() - calculate the page and offset of the byte containing the | |
5486 | * given bit number | |
5487 | * @eb: the extent buffer | |
5488 | * @start: offset of the bitmap item in the extent buffer | |
5489 | * @nr: bit number | |
5490 | * @page_index: return index of the page in the extent buffer that contains the | |
5491 | * given bit number | |
5492 | * @page_offset: return offset into the page given by page_index | |
5493 | * | |
5494 | * This helper hides the ugliness of finding the byte in an extent buffer which | |
5495 | * contains a given bit. | |
5496 | */ | |
5497 | static inline void eb_bitmap_offset(struct extent_buffer *eb, | |
5498 | unsigned long start, unsigned long nr, | |
5499 | unsigned long *page_index, | |
5500 | size_t *page_offset) | |
5501 | { | |
5502 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); | |
5503 | size_t byte_offset = BIT_BYTE(nr); | |
5504 | size_t offset; | |
5505 | ||
5506 | /* | |
5507 | * The byte we want is the offset of the extent buffer + the offset of | |
5508 | * the bitmap item in the extent buffer + the offset of the byte in the | |
5509 | * bitmap item. | |
5510 | */ | |
5511 | offset = start_offset + start + byte_offset; | |
5512 | ||
5513 | *page_index = offset >> PAGE_SHIFT; | |
5514 | *page_offset = offset & (PAGE_SIZE - 1); | |
5515 | } | |
5516 | ||
5517 | /** | |
5518 | * extent_buffer_test_bit - determine whether a bit in a bitmap item is set | |
5519 | * @eb: the extent buffer | |
5520 | * @start: offset of the bitmap item in the extent buffer | |
5521 | * @nr: bit number to test | |
5522 | */ | |
5523 | int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start, | |
5524 | unsigned long nr) | |
5525 | { | |
5526 | char *kaddr; | |
5527 | struct page *page; | |
5528 | unsigned long i; | |
5529 | size_t offset; | |
5530 | ||
5531 | eb_bitmap_offset(eb, start, nr, &i, &offset); | |
5532 | page = eb->pages[i]; | |
5533 | WARN_ON(!PageUptodate(page)); | |
5534 | kaddr = page_address(page); | |
5535 | return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1))); | |
5536 | } | |
5537 | ||
5538 | /** | |
5539 | * extent_buffer_bitmap_set - set an area of a bitmap | |
5540 | * @eb: the extent buffer | |
5541 | * @start: offset of the bitmap item in the extent buffer | |
5542 | * @pos: bit number of the first bit | |
5543 | * @len: number of bits to set | |
5544 | */ | |
5545 | void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start, | |
5546 | unsigned long pos, unsigned long len) | |
5547 | { | |
5548 | char *kaddr; | |
5549 | struct page *page; | |
5550 | unsigned long i; | |
5551 | size_t offset; | |
5552 | const unsigned int size = pos + len; | |
5553 | int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
5554 | unsigned int mask_to_set = BITMAP_FIRST_BYTE_MASK(pos); | |
5555 | ||
5556 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5557 | page = eb->pages[i]; | |
5558 | WARN_ON(!PageUptodate(page)); | |
5559 | kaddr = page_address(page); | |
5560 | ||
5561 | while (len >= bits_to_set) { | |
5562 | kaddr[offset] |= mask_to_set; | |
5563 | len -= bits_to_set; | |
5564 | bits_to_set = BITS_PER_BYTE; | |
5565 | mask_to_set = ~0U; | |
5566 | if (++offset >= PAGE_SIZE && len > 0) { | |
5567 | offset = 0; | |
5568 | page = eb->pages[++i]; | |
5569 | WARN_ON(!PageUptodate(page)); | |
5570 | kaddr = page_address(page); | |
5571 | } | |
5572 | } | |
5573 | if (len) { | |
5574 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
5575 | kaddr[offset] |= mask_to_set; | |
5576 | } | |
5577 | } | |
5578 | ||
5579 | ||
5580 | /** | |
5581 | * extent_buffer_bitmap_clear - clear an area of a bitmap | |
5582 | * @eb: the extent buffer | |
5583 | * @start: offset of the bitmap item in the extent buffer | |
5584 | * @pos: bit number of the first bit | |
5585 | * @len: number of bits to clear | |
5586 | */ | |
5587 | void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start, | |
5588 | unsigned long pos, unsigned long len) | |
5589 | { | |
5590 | char *kaddr; | |
5591 | struct page *page; | |
5592 | unsigned long i; | |
5593 | size_t offset; | |
5594 | const unsigned int size = pos + len; | |
5595 | int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
5596 | unsigned int mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos); | |
5597 | ||
5598 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5599 | page = eb->pages[i]; | |
5600 | WARN_ON(!PageUptodate(page)); | |
5601 | kaddr = page_address(page); | |
5602 | ||
5603 | while (len >= bits_to_clear) { | |
5604 | kaddr[offset] &= ~mask_to_clear; | |
5605 | len -= bits_to_clear; | |
5606 | bits_to_clear = BITS_PER_BYTE; | |
5607 | mask_to_clear = ~0U; | |
5608 | if (++offset >= PAGE_SIZE && len > 0) { | |
5609 | offset = 0; | |
5610 | page = eb->pages[++i]; | |
5611 | WARN_ON(!PageUptodate(page)); | |
5612 | kaddr = page_address(page); | |
5613 | } | |
5614 | } | |
5615 | if (len) { | |
5616 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
5617 | kaddr[offset] &= ~mask_to_clear; | |
5618 | } | |
5619 | } | |
5620 | ||
5621 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) | |
5622 | { | |
5623 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
5624 | return distance < len; | |
5625 | } | |
5626 | ||
5627 | static void copy_pages(struct page *dst_page, struct page *src_page, | |
5628 | unsigned long dst_off, unsigned long src_off, | |
5629 | unsigned long len) | |
5630 | { | |
5631 | char *dst_kaddr = page_address(dst_page); | |
5632 | char *src_kaddr; | |
5633 | int must_memmove = 0; | |
5634 | ||
5635 | if (dst_page != src_page) { | |
5636 | src_kaddr = page_address(src_page); | |
5637 | } else { | |
5638 | src_kaddr = dst_kaddr; | |
5639 | if (areas_overlap(src_off, dst_off, len)) | |
5640 | must_memmove = 1; | |
5641 | } | |
5642 | ||
5643 | if (must_memmove) | |
5644 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
5645 | else | |
5646 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
5647 | } | |
5648 | ||
5649 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5650 | unsigned long src_offset, unsigned long len) | |
5651 | { | |
5652 | size_t cur; | |
5653 | size_t dst_off_in_page; | |
5654 | size_t src_off_in_page; | |
5655 | size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1); | |
5656 | unsigned long dst_i; | |
5657 | unsigned long src_i; | |
5658 | ||
5659 | if (src_offset + len > dst->len) { | |
5660 | btrfs_err(dst->fs_info, | |
5661 | "memmove bogus src_offset %lu move " | |
5662 | "len %lu dst len %lu", src_offset, len, dst->len); | |
5663 | BUG_ON(1); | |
5664 | } | |
5665 | if (dst_offset + len > dst->len) { | |
5666 | btrfs_err(dst->fs_info, | |
5667 | "memmove bogus dst_offset %lu move " | |
5668 | "len %lu dst len %lu", dst_offset, len, dst->len); | |
5669 | BUG_ON(1); | |
5670 | } | |
5671 | ||
5672 | while (len > 0) { | |
5673 | dst_off_in_page = (start_offset + dst_offset) & | |
5674 | (PAGE_SIZE - 1); | |
5675 | src_off_in_page = (start_offset + src_offset) & | |
5676 | (PAGE_SIZE - 1); | |
5677 | ||
5678 | dst_i = (start_offset + dst_offset) >> PAGE_SHIFT; | |
5679 | src_i = (start_offset + src_offset) >> PAGE_SHIFT; | |
5680 | ||
5681 | cur = min(len, (unsigned long)(PAGE_SIZE - | |
5682 | src_off_in_page)); | |
5683 | cur = min_t(unsigned long, cur, | |
5684 | (unsigned long)(PAGE_SIZE - dst_off_in_page)); | |
5685 | ||
5686 | copy_pages(dst->pages[dst_i], dst->pages[src_i], | |
5687 | dst_off_in_page, src_off_in_page, cur); | |
5688 | ||
5689 | src_offset += cur; | |
5690 | dst_offset += cur; | |
5691 | len -= cur; | |
5692 | } | |
5693 | } | |
5694 | ||
5695 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5696 | unsigned long src_offset, unsigned long len) | |
5697 | { | |
5698 | size_t cur; | |
5699 | size_t dst_off_in_page; | |
5700 | size_t src_off_in_page; | |
5701 | unsigned long dst_end = dst_offset + len - 1; | |
5702 | unsigned long src_end = src_offset + len - 1; | |
5703 | size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1); | |
5704 | unsigned long dst_i; | |
5705 | unsigned long src_i; | |
5706 | ||
5707 | if (src_offset + len > dst->len) { | |
5708 | btrfs_err(dst->fs_info, "memmove bogus src_offset %lu move " | |
5709 | "len %lu len %lu", src_offset, len, dst->len); | |
5710 | BUG_ON(1); | |
5711 | } | |
5712 | if (dst_offset + len > dst->len) { | |
5713 | btrfs_err(dst->fs_info, "memmove bogus dst_offset %lu move " | |
5714 | "len %lu len %lu", dst_offset, len, dst->len); | |
5715 | BUG_ON(1); | |
5716 | } | |
5717 | if (dst_offset < src_offset) { | |
5718 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); | |
5719 | return; | |
5720 | } | |
5721 | while (len > 0) { | |
5722 | dst_i = (start_offset + dst_end) >> PAGE_SHIFT; | |
5723 | src_i = (start_offset + src_end) >> PAGE_SHIFT; | |
5724 | ||
5725 | dst_off_in_page = (start_offset + dst_end) & | |
5726 | (PAGE_SIZE - 1); | |
5727 | src_off_in_page = (start_offset + src_end) & | |
5728 | (PAGE_SIZE - 1); | |
5729 | ||
5730 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
5731 | cur = min(cur, dst_off_in_page + 1); | |
5732 | copy_pages(dst->pages[dst_i], dst->pages[src_i], | |
5733 | dst_off_in_page - cur + 1, | |
5734 | src_off_in_page - cur + 1, cur); | |
5735 | ||
5736 | dst_end -= cur; | |
5737 | src_end -= cur; | |
5738 | len -= cur; | |
5739 | } | |
5740 | } | |
5741 | ||
5742 | int try_release_extent_buffer(struct page *page) | |
5743 | { | |
5744 | struct extent_buffer *eb; | |
5745 | ||
5746 | /* | |
5747 | * We need to make sure noboody is attaching this page to an eb right | |
5748 | * now. | |
5749 | */ | |
5750 | spin_lock(&page->mapping->private_lock); | |
5751 | if (!PagePrivate(page)) { | |
5752 | spin_unlock(&page->mapping->private_lock); | |
5753 | return 1; | |
5754 | } | |
5755 | ||
5756 | eb = (struct extent_buffer *)page->private; | |
5757 | BUG_ON(!eb); | |
5758 | ||
5759 | /* | |
5760 | * This is a little awful but should be ok, we need to make sure that | |
5761 | * the eb doesn't disappear out from under us while we're looking at | |
5762 | * this page. | |
5763 | */ | |
5764 | spin_lock(&eb->refs_lock); | |
5765 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { | |
5766 | spin_unlock(&eb->refs_lock); | |
5767 | spin_unlock(&page->mapping->private_lock); | |
5768 | return 0; | |
5769 | } | |
5770 | spin_unlock(&page->mapping->private_lock); | |
5771 | ||
5772 | /* | |
5773 | * If tree ref isn't set then we know the ref on this eb is a real ref, | |
5774 | * so just return, this page will likely be freed soon anyway. | |
5775 | */ | |
5776 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { | |
5777 | spin_unlock(&eb->refs_lock); | |
5778 | return 0; | |
5779 | } | |
5780 | ||
5781 | return release_extent_buffer(eb); | |
5782 | } |