]>
Commit | Line | Data |
---|---|---|
d1310b2e CM |
1 | #include <linux/bitops.h> |
2 | #include <linux/slab.h> | |
3 | #include <linux/bio.h> | |
4 | #include <linux/mm.h> | |
d1310b2e CM |
5 | #include <linux/pagemap.h> |
6 | #include <linux/page-flags.h> | |
d1310b2e CM |
7 | #include <linux/spinlock.h> |
8 | #include <linux/blkdev.h> | |
9 | #include <linux/swap.h> | |
d1310b2e CM |
10 | #include <linux/writeback.h> |
11 | #include <linux/pagevec.h> | |
268bb0ce | 12 | #include <linux/prefetch.h> |
90a887c9 | 13 | #include <linux/cleancache.h> |
d1310b2e CM |
14 | #include "extent_io.h" |
15 | #include "extent_map.h" | |
902b22f3 DW |
16 | #include "ctree.h" |
17 | #include "btrfs_inode.h" | |
4a54c8c1 | 18 | #include "volumes.h" |
21adbd5c | 19 | #include "check-integrity.h" |
0b32f4bb | 20 | #include "locking.h" |
606686ee | 21 | #include "rcu-string.h" |
fe09e16c | 22 | #include "backref.h" |
d1310b2e | 23 | |
d1310b2e CM |
24 | static struct kmem_cache *extent_state_cache; |
25 | static struct kmem_cache *extent_buffer_cache; | |
9be3395b | 26 | static struct bio_set *btrfs_bioset; |
d1310b2e | 27 | |
6d49ba1b | 28 | #ifdef CONFIG_BTRFS_DEBUG |
d1310b2e CM |
29 | static LIST_HEAD(buffers); |
30 | static LIST_HEAD(states); | |
4bef0848 | 31 | |
d397712b | 32 | static DEFINE_SPINLOCK(leak_lock); |
6d49ba1b ES |
33 | |
34 | static inline | |
35 | void btrfs_leak_debug_add(struct list_head *new, struct list_head *head) | |
36 | { | |
37 | unsigned long flags; | |
38 | ||
39 | spin_lock_irqsave(&leak_lock, flags); | |
40 | list_add(new, head); | |
41 | spin_unlock_irqrestore(&leak_lock, flags); | |
42 | } | |
43 | ||
44 | static inline | |
45 | void btrfs_leak_debug_del(struct list_head *entry) | |
46 | { | |
47 | unsigned long flags; | |
48 | ||
49 | spin_lock_irqsave(&leak_lock, flags); | |
50 | list_del(entry); | |
51 | spin_unlock_irqrestore(&leak_lock, flags); | |
52 | } | |
53 | ||
54 | static inline | |
55 | void btrfs_leak_debug_check(void) | |
56 | { | |
57 | struct extent_state *state; | |
58 | struct extent_buffer *eb; | |
59 | ||
60 | while (!list_empty(&states)) { | |
61 | state = list_entry(states.next, struct extent_state, leak_list); | |
62 | printk(KERN_ERR "btrfs state leak: start %llu end %llu " | |
63 | "state %lu in tree %p refs %d\n", | |
c1c9ff7c GU |
64 | state->start, state->end, state->state, state->tree, |
65 | atomic_read(&state->refs)); | |
6d49ba1b ES |
66 | list_del(&state->leak_list); |
67 | kmem_cache_free(extent_state_cache, state); | |
68 | } | |
69 | ||
70 | while (!list_empty(&buffers)) { | |
71 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
72 | printk(KERN_ERR "btrfs buffer leak start %llu len %lu " | |
c1c9ff7c GU |
73 | "refs %d\n", |
74 | eb->start, eb->len, atomic_read(&eb->refs)); | |
6d49ba1b ES |
75 | list_del(&eb->leak_list); |
76 | kmem_cache_free(extent_buffer_cache, eb); | |
77 | } | |
78 | } | |
8d599ae1 | 79 | |
a5dee37d JB |
80 | #define btrfs_debug_check_extent_io_range(tree, start, end) \ |
81 | __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end)) | |
8d599ae1 | 82 | static inline void __btrfs_debug_check_extent_io_range(const char *caller, |
a5dee37d | 83 | struct extent_io_tree *tree, u64 start, u64 end) |
8d599ae1 | 84 | { |
a5dee37d JB |
85 | struct inode *inode; |
86 | u64 isize; | |
8d599ae1 | 87 | |
a5dee37d JB |
88 | if (!tree->mapping) |
89 | return; | |
90 | ||
91 | inode = tree->mapping->host; | |
92 | isize = i_size_read(inode); | |
8d599ae1 DS |
93 | if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) { |
94 | printk_ratelimited(KERN_DEBUG | |
95 | "btrfs: %s: ino %llu isize %llu odd range [%llu,%llu]\n", | |
c1c9ff7c | 96 | caller, btrfs_ino(inode), isize, start, end); |
8d599ae1 DS |
97 | } |
98 | } | |
6d49ba1b ES |
99 | #else |
100 | #define btrfs_leak_debug_add(new, head) do {} while (0) | |
101 | #define btrfs_leak_debug_del(entry) do {} while (0) | |
102 | #define btrfs_leak_debug_check() do {} while (0) | |
8d599ae1 | 103 | #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0) |
4bef0848 | 104 | #endif |
d1310b2e | 105 | |
d1310b2e CM |
106 | #define BUFFER_LRU_MAX 64 |
107 | ||
108 | struct tree_entry { | |
109 | u64 start; | |
110 | u64 end; | |
d1310b2e CM |
111 | struct rb_node rb_node; |
112 | }; | |
113 | ||
114 | struct extent_page_data { | |
115 | struct bio *bio; | |
116 | struct extent_io_tree *tree; | |
117 | get_extent_t *get_extent; | |
de0022b9 | 118 | unsigned long bio_flags; |
771ed689 CM |
119 | |
120 | /* tells writepage not to lock the state bits for this range | |
121 | * it still does the unlocking | |
122 | */ | |
ffbd517d CM |
123 | unsigned int extent_locked:1; |
124 | ||
125 | /* tells the submit_bio code to use a WRITE_SYNC */ | |
126 | unsigned int sync_io:1; | |
d1310b2e CM |
127 | }; |
128 | ||
0b32f4bb | 129 | static noinline void flush_write_bio(void *data); |
c2d904e0 JM |
130 | static inline struct btrfs_fs_info * |
131 | tree_fs_info(struct extent_io_tree *tree) | |
132 | { | |
a5dee37d JB |
133 | if (!tree->mapping) |
134 | return NULL; | |
c2d904e0 JM |
135 | return btrfs_sb(tree->mapping->host->i_sb); |
136 | } | |
0b32f4bb | 137 | |
d1310b2e CM |
138 | int __init extent_io_init(void) |
139 | { | |
837e1972 | 140 | extent_state_cache = kmem_cache_create("btrfs_extent_state", |
9601e3f6 CH |
141 | sizeof(struct extent_state), 0, |
142 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
143 | if (!extent_state_cache) |
144 | return -ENOMEM; | |
145 | ||
837e1972 | 146 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 CH |
147 | sizeof(struct extent_buffer), 0, |
148 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
149 | if (!extent_buffer_cache) |
150 | goto free_state_cache; | |
9be3395b CM |
151 | |
152 | btrfs_bioset = bioset_create(BIO_POOL_SIZE, | |
153 | offsetof(struct btrfs_io_bio, bio)); | |
154 | if (!btrfs_bioset) | |
155 | goto free_buffer_cache; | |
b208c2f7 DW |
156 | |
157 | if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE)) | |
158 | goto free_bioset; | |
159 | ||
d1310b2e CM |
160 | return 0; |
161 | ||
b208c2f7 DW |
162 | free_bioset: |
163 | bioset_free(btrfs_bioset); | |
164 | btrfs_bioset = NULL; | |
165 | ||
9be3395b CM |
166 | free_buffer_cache: |
167 | kmem_cache_destroy(extent_buffer_cache); | |
168 | extent_buffer_cache = NULL; | |
169 | ||
d1310b2e CM |
170 | free_state_cache: |
171 | kmem_cache_destroy(extent_state_cache); | |
9be3395b | 172 | extent_state_cache = NULL; |
d1310b2e CM |
173 | return -ENOMEM; |
174 | } | |
175 | ||
176 | void extent_io_exit(void) | |
177 | { | |
6d49ba1b | 178 | btrfs_leak_debug_check(); |
8c0a8537 KS |
179 | |
180 | /* | |
181 | * Make sure all delayed rcu free are flushed before we | |
182 | * destroy caches. | |
183 | */ | |
184 | rcu_barrier(); | |
d1310b2e CM |
185 | if (extent_state_cache) |
186 | kmem_cache_destroy(extent_state_cache); | |
187 | if (extent_buffer_cache) | |
188 | kmem_cache_destroy(extent_buffer_cache); | |
9be3395b CM |
189 | if (btrfs_bioset) |
190 | bioset_free(btrfs_bioset); | |
d1310b2e CM |
191 | } |
192 | ||
193 | void extent_io_tree_init(struct extent_io_tree *tree, | |
f993c883 | 194 | struct address_space *mapping) |
d1310b2e | 195 | { |
6bef4d31 | 196 | tree->state = RB_ROOT; |
d1310b2e CM |
197 | tree->ops = NULL; |
198 | tree->dirty_bytes = 0; | |
70dec807 | 199 | spin_lock_init(&tree->lock); |
d1310b2e | 200 | tree->mapping = mapping; |
d1310b2e | 201 | } |
d1310b2e | 202 | |
b2950863 | 203 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
204 | { |
205 | struct extent_state *state; | |
d1310b2e CM |
206 | |
207 | state = kmem_cache_alloc(extent_state_cache, mask); | |
2b114d1d | 208 | if (!state) |
d1310b2e CM |
209 | return state; |
210 | state->state = 0; | |
d1310b2e | 211 | state->private = 0; |
70dec807 | 212 | state->tree = NULL; |
6d49ba1b | 213 | btrfs_leak_debug_add(&state->leak_list, &states); |
d1310b2e CM |
214 | atomic_set(&state->refs, 1); |
215 | init_waitqueue_head(&state->wq); | |
143bede5 | 216 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
217 | return state; |
218 | } | |
d1310b2e | 219 | |
4845e44f | 220 | void free_extent_state(struct extent_state *state) |
d1310b2e | 221 | { |
d1310b2e CM |
222 | if (!state) |
223 | return; | |
224 | if (atomic_dec_and_test(&state->refs)) { | |
70dec807 | 225 | WARN_ON(state->tree); |
6d49ba1b | 226 | btrfs_leak_debug_del(&state->leak_list); |
143bede5 | 227 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
228 | kmem_cache_free(extent_state_cache, state); |
229 | } | |
230 | } | |
d1310b2e CM |
231 | |
232 | static struct rb_node *tree_insert(struct rb_root *root, u64 offset, | |
12cfbad9 FDBM |
233 | struct rb_node *node, |
234 | struct rb_node ***p_in, | |
235 | struct rb_node **parent_in) | |
d1310b2e | 236 | { |
d397712b CM |
237 | struct rb_node **p = &root->rb_node; |
238 | struct rb_node *parent = NULL; | |
d1310b2e CM |
239 | struct tree_entry *entry; |
240 | ||
12cfbad9 FDBM |
241 | if (p_in && parent_in) { |
242 | p = *p_in; | |
243 | parent = *parent_in; | |
244 | goto do_insert; | |
245 | } | |
246 | ||
d397712b | 247 | while (*p) { |
d1310b2e CM |
248 | parent = *p; |
249 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
250 | ||
251 | if (offset < entry->start) | |
252 | p = &(*p)->rb_left; | |
253 | else if (offset > entry->end) | |
254 | p = &(*p)->rb_right; | |
255 | else | |
256 | return parent; | |
257 | } | |
258 | ||
12cfbad9 | 259 | do_insert: |
d1310b2e CM |
260 | rb_link_node(node, parent, p); |
261 | rb_insert_color(node, root); | |
262 | return NULL; | |
263 | } | |
264 | ||
80ea96b1 | 265 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
12cfbad9 FDBM |
266 | struct rb_node **prev_ret, |
267 | struct rb_node **next_ret, | |
268 | struct rb_node ***p_ret, | |
269 | struct rb_node **parent_ret) | |
d1310b2e | 270 | { |
80ea96b1 | 271 | struct rb_root *root = &tree->state; |
12cfbad9 | 272 | struct rb_node **n = &root->rb_node; |
d1310b2e CM |
273 | struct rb_node *prev = NULL; |
274 | struct rb_node *orig_prev = NULL; | |
275 | struct tree_entry *entry; | |
276 | struct tree_entry *prev_entry = NULL; | |
277 | ||
12cfbad9 FDBM |
278 | while (*n) { |
279 | prev = *n; | |
280 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
d1310b2e CM |
281 | prev_entry = entry; |
282 | ||
283 | if (offset < entry->start) | |
12cfbad9 | 284 | n = &(*n)->rb_left; |
d1310b2e | 285 | else if (offset > entry->end) |
12cfbad9 | 286 | n = &(*n)->rb_right; |
d397712b | 287 | else |
12cfbad9 | 288 | return *n; |
d1310b2e CM |
289 | } |
290 | ||
12cfbad9 FDBM |
291 | if (p_ret) |
292 | *p_ret = n; | |
293 | if (parent_ret) | |
294 | *parent_ret = prev; | |
295 | ||
d1310b2e CM |
296 | if (prev_ret) { |
297 | orig_prev = prev; | |
d397712b | 298 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
299 | prev = rb_next(prev); |
300 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
301 | } | |
302 | *prev_ret = prev; | |
303 | prev = orig_prev; | |
304 | } | |
305 | ||
306 | if (next_ret) { | |
307 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 308 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
309 | prev = rb_prev(prev); |
310 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
311 | } | |
312 | *next_ret = prev; | |
313 | } | |
314 | return NULL; | |
315 | } | |
316 | ||
12cfbad9 FDBM |
317 | static inline struct rb_node * |
318 | tree_search_for_insert(struct extent_io_tree *tree, | |
319 | u64 offset, | |
320 | struct rb_node ***p_ret, | |
321 | struct rb_node **parent_ret) | |
d1310b2e | 322 | { |
70dec807 | 323 | struct rb_node *prev = NULL; |
d1310b2e | 324 | struct rb_node *ret; |
70dec807 | 325 | |
12cfbad9 | 326 | ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret); |
d397712b | 327 | if (!ret) |
d1310b2e CM |
328 | return prev; |
329 | return ret; | |
330 | } | |
331 | ||
12cfbad9 FDBM |
332 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
333 | u64 offset) | |
334 | { | |
335 | return tree_search_for_insert(tree, offset, NULL, NULL); | |
336 | } | |
337 | ||
9ed74f2d JB |
338 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, |
339 | struct extent_state *other) | |
340 | { | |
341 | if (tree->ops && tree->ops->merge_extent_hook) | |
342 | tree->ops->merge_extent_hook(tree->mapping->host, new, | |
343 | other); | |
344 | } | |
345 | ||
d1310b2e CM |
346 | /* |
347 | * utility function to look for merge candidates inside a given range. | |
348 | * Any extents with matching state are merged together into a single | |
349 | * extent in the tree. Extents with EXTENT_IO in their state field | |
350 | * are not merged because the end_io handlers need to be able to do | |
351 | * operations on them without sleeping (or doing allocations/splits). | |
352 | * | |
353 | * This should be called with the tree lock held. | |
354 | */ | |
1bf85046 JM |
355 | static void merge_state(struct extent_io_tree *tree, |
356 | struct extent_state *state) | |
d1310b2e CM |
357 | { |
358 | struct extent_state *other; | |
359 | struct rb_node *other_node; | |
360 | ||
5b21f2ed | 361 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
1bf85046 | 362 | return; |
d1310b2e CM |
363 | |
364 | other_node = rb_prev(&state->rb_node); | |
365 | if (other_node) { | |
366 | other = rb_entry(other_node, struct extent_state, rb_node); | |
367 | if (other->end == state->start - 1 && | |
368 | other->state == state->state) { | |
9ed74f2d | 369 | merge_cb(tree, state, other); |
d1310b2e | 370 | state->start = other->start; |
70dec807 | 371 | other->tree = NULL; |
d1310b2e CM |
372 | rb_erase(&other->rb_node, &tree->state); |
373 | free_extent_state(other); | |
374 | } | |
375 | } | |
376 | other_node = rb_next(&state->rb_node); | |
377 | if (other_node) { | |
378 | other = rb_entry(other_node, struct extent_state, rb_node); | |
379 | if (other->start == state->end + 1 && | |
380 | other->state == state->state) { | |
9ed74f2d | 381 | merge_cb(tree, state, other); |
df98b6e2 JB |
382 | state->end = other->end; |
383 | other->tree = NULL; | |
384 | rb_erase(&other->rb_node, &tree->state); | |
385 | free_extent_state(other); | |
d1310b2e CM |
386 | } |
387 | } | |
d1310b2e CM |
388 | } |
389 | ||
1bf85046 | 390 | static void set_state_cb(struct extent_io_tree *tree, |
41074888 | 391 | struct extent_state *state, unsigned long *bits) |
291d673e | 392 | { |
1bf85046 JM |
393 | if (tree->ops && tree->ops->set_bit_hook) |
394 | tree->ops->set_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
395 | } |
396 | ||
397 | static void clear_state_cb(struct extent_io_tree *tree, | |
41074888 | 398 | struct extent_state *state, unsigned long *bits) |
291d673e | 399 | { |
9ed74f2d JB |
400 | if (tree->ops && tree->ops->clear_bit_hook) |
401 | tree->ops->clear_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
402 | } |
403 | ||
3150b699 | 404 | static void set_state_bits(struct extent_io_tree *tree, |
41074888 | 405 | struct extent_state *state, unsigned long *bits); |
3150b699 | 406 | |
d1310b2e CM |
407 | /* |
408 | * insert an extent_state struct into the tree. 'bits' are set on the | |
409 | * struct before it is inserted. | |
410 | * | |
411 | * This may return -EEXIST if the extent is already there, in which case the | |
412 | * state struct is freed. | |
413 | * | |
414 | * The tree lock is not taken internally. This is a utility function and | |
415 | * probably isn't what you want to call (see set/clear_extent_bit). | |
416 | */ | |
417 | static int insert_state(struct extent_io_tree *tree, | |
418 | struct extent_state *state, u64 start, u64 end, | |
12cfbad9 FDBM |
419 | struct rb_node ***p, |
420 | struct rb_node **parent, | |
41074888 | 421 | unsigned long *bits) |
d1310b2e CM |
422 | { |
423 | struct rb_node *node; | |
424 | ||
31b1a2bd JL |
425 | if (end < start) |
426 | WARN(1, KERN_ERR "btrfs end < start %llu %llu\n", | |
c1c9ff7c | 427 | end, start); |
d1310b2e CM |
428 | state->start = start; |
429 | state->end = end; | |
9ed74f2d | 430 | |
3150b699 XG |
431 | set_state_bits(tree, state, bits); |
432 | ||
12cfbad9 | 433 | node = tree_insert(&tree->state, end, &state->rb_node, p, parent); |
d1310b2e CM |
434 | if (node) { |
435 | struct extent_state *found; | |
436 | found = rb_entry(node, struct extent_state, rb_node); | |
d397712b | 437 | printk(KERN_ERR "btrfs found node %llu %llu on insert of " |
c1c9ff7c GU |
438 | "%llu %llu\n", |
439 | found->start, found->end, start, end); | |
d1310b2e CM |
440 | return -EEXIST; |
441 | } | |
70dec807 | 442 | state->tree = tree; |
d1310b2e CM |
443 | merge_state(tree, state); |
444 | return 0; | |
445 | } | |
446 | ||
1bf85046 | 447 | static void split_cb(struct extent_io_tree *tree, struct extent_state *orig, |
9ed74f2d JB |
448 | u64 split) |
449 | { | |
450 | if (tree->ops && tree->ops->split_extent_hook) | |
1bf85046 | 451 | tree->ops->split_extent_hook(tree->mapping->host, orig, split); |
9ed74f2d JB |
452 | } |
453 | ||
d1310b2e CM |
454 | /* |
455 | * split a given extent state struct in two, inserting the preallocated | |
456 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
457 | * offset inside 'orig' where it should be split. | |
458 | * | |
459 | * Before calling, | |
460 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
461 | * are two extent state structs in the tree: | |
462 | * prealloc: [orig->start, split - 1] | |
463 | * orig: [ split, orig->end ] | |
464 | * | |
465 | * The tree locks are not taken by this function. They need to be held | |
466 | * by the caller. | |
467 | */ | |
468 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
469 | struct extent_state *prealloc, u64 split) | |
470 | { | |
471 | struct rb_node *node; | |
9ed74f2d JB |
472 | |
473 | split_cb(tree, orig, split); | |
474 | ||
d1310b2e CM |
475 | prealloc->start = orig->start; |
476 | prealloc->end = split - 1; | |
477 | prealloc->state = orig->state; | |
478 | orig->start = split; | |
479 | ||
12cfbad9 FDBM |
480 | node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node, |
481 | NULL, NULL); | |
d1310b2e | 482 | if (node) { |
d1310b2e CM |
483 | free_extent_state(prealloc); |
484 | return -EEXIST; | |
485 | } | |
70dec807 | 486 | prealloc->tree = tree; |
d1310b2e CM |
487 | return 0; |
488 | } | |
489 | ||
cdc6a395 LZ |
490 | static struct extent_state *next_state(struct extent_state *state) |
491 | { | |
492 | struct rb_node *next = rb_next(&state->rb_node); | |
493 | if (next) | |
494 | return rb_entry(next, struct extent_state, rb_node); | |
495 | else | |
496 | return NULL; | |
497 | } | |
498 | ||
d1310b2e CM |
499 | /* |
500 | * utility function to clear some bits in an extent state struct. | |
1b303fc0 | 501 | * it will optionally wake up any one waiting on this state (wake == 1). |
d1310b2e CM |
502 | * |
503 | * If no bits are set on the state struct after clearing things, the | |
504 | * struct is freed and removed from the tree | |
505 | */ | |
cdc6a395 LZ |
506 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
507 | struct extent_state *state, | |
41074888 | 508 | unsigned long *bits, int wake) |
d1310b2e | 509 | { |
cdc6a395 | 510 | struct extent_state *next; |
41074888 | 511 | unsigned long bits_to_clear = *bits & ~EXTENT_CTLBITS; |
d1310b2e | 512 | |
0ca1f7ce | 513 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
514 | u64 range = state->end - state->start + 1; |
515 | WARN_ON(range > tree->dirty_bytes); | |
516 | tree->dirty_bytes -= range; | |
517 | } | |
291d673e | 518 | clear_state_cb(tree, state, bits); |
32c00aff | 519 | state->state &= ~bits_to_clear; |
d1310b2e CM |
520 | if (wake) |
521 | wake_up(&state->wq); | |
0ca1f7ce | 522 | if (state->state == 0) { |
cdc6a395 | 523 | next = next_state(state); |
70dec807 | 524 | if (state->tree) { |
d1310b2e | 525 | rb_erase(&state->rb_node, &tree->state); |
70dec807 | 526 | state->tree = NULL; |
d1310b2e CM |
527 | free_extent_state(state); |
528 | } else { | |
529 | WARN_ON(1); | |
530 | } | |
531 | } else { | |
532 | merge_state(tree, state); | |
cdc6a395 | 533 | next = next_state(state); |
d1310b2e | 534 | } |
cdc6a395 | 535 | return next; |
d1310b2e CM |
536 | } |
537 | ||
8233767a XG |
538 | static struct extent_state * |
539 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
540 | { | |
541 | if (!prealloc) | |
542 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
543 | ||
544 | return prealloc; | |
545 | } | |
546 | ||
48a3b636 | 547 | static void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
c2d904e0 JM |
548 | { |
549 | btrfs_panic(tree_fs_info(tree), err, "Locking error: " | |
550 | "Extent tree was modified by another " | |
551 | "thread while locked."); | |
552 | } | |
553 | ||
d1310b2e CM |
554 | /* |
555 | * clear some bits on a range in the tree. This may require splitting | |
556 | * or inserting elements in the tree, so the gfp mask is used to | |
557 | * indicate which allocations or sleeping are allowed. | |
558 | * | |
559 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
560 | * the given range from the tree regardless of state (ie for truncate). | |
561 | * | |
562 | * the range [start, end] is inclusive. | |
563 | * | |
6763af84 | 564 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e CM |
565 | */ |
566 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
41074888 | 567 | unsigned long bits, int wake, int delete, |
2c64c53d CM |
568 | struct extent_state **cached_state, |
569 | gfp_t mask) | |
d1310b2e CM |
570 | { |
571 | struct extent_state *state; | |
2c64c53d | 572 | struct extent_state *cached; |
d1310b2e CM |
573 | struct extent_state *prealloc = NULL; |
574 | struct rb_node *node; | |
5c939df5 | 575 | u64 last_end; |
d1310b2e | 576 | int err; |
2ac55d41 | 577 | int clear = 0; |
d1310b2e | 578 | |
a5dee37d | 579 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 580 | |
7ee9e440 JB |
581 | if (bits & EXTENT_DELALLOC) |
582 | bits |= EXTENT_NORESERVE; | |
583 | ||
0ca1f7ce YZ |
584 | if (delete) |
585 | bits |= ~EXTENT_CTLBITS; | |
586 | bits |= EXTENT_FIRST_DELALLOC; | |
587 | ||
2ac55d41 JB |
588 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
589 | clear = 1; | |
d1310b2e CM |
590 | again: |
591 | if (!prealloc && (mask & __GFP_WAIT)) { | |
592 | prealloc = alloc_extent_state(mask); | |
593 | if (!prealloc) | |
594 | return -ENOMEM; | |
595 | } | |
596 | ||
cad321ad | 597 | spin_lock(&tree->lock); |
2c64c53d CM |
598 | if (cached_state) { |
599 | cached = *cached_state; | |
2ac55d41 JB |
600 | |
601 | if (clear) { | |
602 | *cached_state = NULL; | |
603 | cached_state = NULL; | |
604 | } | |
605 | ||
df98b6e2 JB |
606 | if (cached && cached->tree && cached->start <= start && |
607 | cached->end > start) { | |
2ac55d41 JB |
608 | if (clear) |
609 | atomic_dec(&cached->refs); | |
2c64c53d | 610 | state = cached; |
42daec29 | 611 | goto hit_next; |
2c64c53d | 612 | } |
2ac55d41 JB |
613 | if (clear) |
614 | free_extent_state(cached); | |
2c64c53d | 615 | } |
d1310b2e CM |
616 | /* |
617 | * this search will find the extents that end after | |
618 | * our range starts | |
619 | */ | |
80ea96b1 | 620 | node = tree_search(tree, start); |
d1310b2e CM |
621 | if (!node) |
622 | goto out; | |
623 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 624 | hit_next: |
d1310b2e CM |
625 | if (state->start > end) |
626 | goto out; | |
627 | WARN_ON(state->end < start); | |
5c939df5 | 628 | last_end = state->end; |
d1310b2e | 629 | |
0449314a | 630 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
631 | if (!(state->state & bits)) { |
632 | state = next_state(state); | |
0449314a | 633 | goto next; |
cdc6a395 | 634 | } |
0449314a | 635 | |
d1310b2e CM |
636 | /* |
637 | * | ---- desired range ---- | | |
638 | * | state | or | |
639 | * | ------------- state -------------- | | |
640 | * | |
641 | * We need to split the extent we found, and may flip | |
642 | * bits on second half. | |
643 | * | |
644 | * If the extent we found extends past our range, we | |
645 | * just split and search again. It'll get split again | |
646 | * the next time though. | |
647 | * | |
648 | * If the extent we found is inside our range, we clear | |
649 | * the desired bit on it. | |
650 | */ | |
651 | ||
652 | if (state->start < start) { | |
8233767a XG |
653 | prealloc = alloc_extent_state_atomic(prealloc); |
654 | BUG_ON(!prealloc); | |
d1310b2e | 655 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
656 | if (err) |
657 | extent_io_tree_panic(tree, err); | |
658 | ||
d1310b2e CM |
659 | prealloc = NULL; |
660 | if (err) | |
661 | goto out; | |
662 | if (state->end <= end) { | |
d1ac6e41 LB |
663 | state = clear_state_bit(tree, state, &bits, wake); |
664 | goto next; | |
d1310b2e CM |
665 | } |
666 | goto search_again; | |
667 | } | |
668 | /* | |
669 | * | ---- desired range ---- | | |
670 | * | state | | |
671 | * We need to split the extent, and clear the bit | |
672 | * on the first half | |
673 | */ | |
674 | if (state->start <= end && state->end > end) { | |
8233767a XG |
675 | prealloc = alloc_extent_state_atomic(prealloc); |
676 | BUG_ON(!prealloc); | |
d1310b2e | 677 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
678 | if (err) |
679 | extent_io_tree_panic(tree, err); | |
680 | ||
d1310b2e CM |
681 | if (wake) |
682 | wake_up(&state->wq); | |
42daec29 | 683 | |
6763af84 | 684 | clear_state_bit(tree, prealloc, &bits, wake); |
9ed74f2d | 685 | |
d1310b2e CM |
686 | prealloc = NULL; |
687 | goto out; | |
688 | } | |
42daec29 | 689 | |
cdc6a395 | 690 | state = clear_state_bit(tree, state, &bits, wake); |
0449314a | 691 | next: |
5c939df5 YZ |
692 | if (last_end == (u64)-1) |
693 | goto out; | |
694 | start = last_end + 1; | |
cdc6a395 | 695 | if (start <= end && state && !need_resched()) |
692e5759 | 696 | goto hit_next; |
d1310b2e CM |
697 | goto search_again; |
698 | ||
699 | out: | |
cad321ad | 700 | spin_unlock(&tree->lock); |
d1310b2e CM |
701 | if (prealloc) |
702 | free_extent_state(prealloc); | |
703 | ||
6763af84 | 704 | return 0; |
d1310b2e CM |
705 | |
706 | search_again: | |
707 | if (start > end) | |
708 | goto out; | |
cad321ad | 709 | spin_unlock(&tree->lock); |
d1310b2e CM |
710 | if (mask & __GFP_WAIT) |
711 | cond_resched(); | |
712 | goto again; | |
713 | } | |
d1310b2e | 714 | |
143bede5 JM |
715 | static void wait_on_state(struct extent_io_tree *tree, |
716 | struct extent_state *state) | |
641f5219 CH |
717 | __releases(tree->lock) |
718 | __acquires(tree->lock) | |
d1310b2e CM |
719 | { |
720 | DEFINE_WAIT(wait); | |
721 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 722 | spin_unlock(&tree->lock); |
d1310b2e | 723 | schedule(); |
cad321ad | 724 | spin_lock(&tree->lock); |
d1310b2e | 725 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
726 | } |
727 | ||
728 | /* | |
729 | * waits for one or more bits to clear on a range in the state tree. | |
730 | * The range [start, end] is inclusive. | |
731 | * The tree lock is taken by this function | |
732 | */ | |
41074888 DS |
733 | static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
734 | unsigned long bits) | |
d1310b2e CM |
735 | { |
736 | struct extent_state *state; | |
737 | struct rb_node *node; | |
738 | ||
a5dee37d | 739 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 740 | |
cad321ad | 741 | spin_lock(&tree->lock); |
d1310b2e CM |
742 | again: |
743 | while (1) { | |
744 | /* | |
745 | * this search will find all the extents that end after | |
746 | * our range starts | |
747 | */ | |
80ea96b1 | 748 | node = tree_search(tree, start); |
d1310b2e CM |
749 | if (!node) |
750 | break; | |
751 | ||
752 | state = rb_entry(node, struct extent_state, rb_node); | |
753 | ||
754 | if (state->start > end) | |
755 | goto out; | |
756 | ||
757 | if (state->state & bits) { | |
758 | start = state->start; | |
759 | atomic_inc(&state->refs); | |
760 | wait_on_state(tree, state); | |
761 | free_extent_state(state); | |
762 | goto again; | |
763 | } | |
764 | start = state->end + 1; | |
765 | ||
766 | if (start > end) | |
767 | break; | |
768 | ||
ded91f08 | 769 | cond_resched_lock(&tree->lock); |
d1310b2e CM |
770 | } |
771 | out: | |
cad321ad | 772 | spin_unlock(&tree->lock); |
d1310b2e | 773 | } |
d1310b2e | 774 | |
1bf85046 | 775 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 776 | struct extent_state *state, |
41074888 | 777 | unsigned long *bits) |
d1310b2e | 778 | { |
41074888 | 779 | unsigned long bits_to_set = *bits & ~EXTENT_CTLBITS; |
9ed74f2d | 780 | |
1bf85046 | 781 | set_state_cb(tree, state, bits); |
0ca1f7ce | 782 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
783 | u64 range = state->end - state->start + 1; |
784 | tree->dirty_bytes += range; | |
785 | } | |
0ca1f7ce | 786 | state->state |= bits_to_set; |
d1310b2e CM |
787 | } |
788 | ||
2c64c53d CM |
789 | static void cache_state(struct extent_state *state, |
790 | struct extent_state **cached_ptr) | |
791 | { | |
792 | if (cached_ptr && !(*cached_ptr)) { | |
793 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) { | |
794 | *cached_ptr = state; | |
795 | atomic_inc(&state->refs); | |
796 | } | |
797 | } | |
798 | } | |
799 | ||
d1310b2e | 800 | /* |
1edbb734 CM |
801 | * set some bits on a range in the tree. This may require allocations or |
802 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 803 | * |
1edbb734 CM |
804 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
805 | * part of the range already has the desired bits set. The start of the | |
806 | * existing range is returned in failed_start in this case. | |
d1310b2e | 807 | * |
1edbb734 | 808 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 809 | */ |
1edbb734 | 810 | |
3fbe5c02 JM |
811 | static int __must_check |
812 | __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
41074888 DS |
813 | unsigned long bits, unsigned long exclusive_bits, |
814 | u64 *failed_start, struct extent_state **cached_state, | |
815 | gfp_t mask) | |
d1310b2e CM |
816 | { |
817 | struct extent_state *state; | |
818 | struct extent_state *prealloc = NULL; | |
819 | struct rb_node *node; | |
12cfbad9 FDBM |
820 | struct rb_node **p; |
821 | struct rb_node *parent; | |
d1310b2e | 822 | int err = 0; |
d1310b2e CM |
823 | u64 last_start; |
824 | u64 last_end; | |
42daec29 | 825 | |
a5dee37d | 826 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 827 | |
0ca1f7ce | 828 | bits |= EXTENT_FIRST_DELALLOC; |
d1310b2e CM |
829 | again: |
830 | if (!prealloc && (mask & __GFP_WAIT)) { | |
831 | prealloc = alloc_extent_state(mask); | |
8233767a | 832 | BUG_ON(!prealloc); |
d1310b2e CM |
833 | } |
834 | ||
cad321ad | 835 | spin_lock(&tree->lock); |
9655d298 CM |
836 | if (cached_state && *cached_state) { |
837 | state = *cached_state; | |
df98b6e2 JB |
838 | if (state->start <= start && state->end > start && |
839 | state->tree) { | |
9655d298 CM |
840 | node = &state->rb_node; |
841 | goto hit_next; | |
842 | } | |
843 | } | |
d1310b2e CM |
844 | /* |
845 | * this search will find all the extents that end after | |
846 | * our range starts. | |
847 | */ | |
12cfbad9 | 848 | node = tree_search_for_insert(tree, start, &p, &parent); |
d1310b2e | 849 | if (!node) { |
8233767a XG |
850 | prealloc = alloc_extent_state_atomic(prealloc); |
851 | BUG_ON(!prealloc); | |
12cfbad9 FDBM |
852 | err = insert_state(tree, prealloc, start, end, |
853 | &p, &parent, &bits); | |
c2d904e0 JM |
854 | if (err) |
855 | extent_io_tree_panic(tree, err); | |
856 | ||
c42ac0bc | 857 | cache_state(prealloc, cached_state); |
d1310b2e | 858 | prealloc = NULL; |
d1310b2e CM |
859 | goto out; |
860 | } | |
d1310b2e | 861 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 862 | hit_next: |
d1310b2e CM |
863 | last_start = state->start; |
864 | last_end = state->end; | |
865 | ||
866 | /* | |
867 | * | ---- desired range ---- | | |
868 | * | state | | |
869 | * | |
870 | * Just lock what we found and keep going | |
871 | */ | |
872 | if (state->start == start && state->end <= end) { | |
1edbb734 | 873 | if (state->state & exclusive_bits) { |
d1310b2e CM |
874 | *failed_start = state->start; |
875 | err = -EEXIST; | |
876 | goto out; | |
877 | } | |
42daec29 | 878 | |
1bf85046 | 879 | set_state_bits(tree, state, &bits); |
2c64c53d | 880 | cache_state(state, cached_state); |
d1310b2e | 881 | merge_state(tree, state); |
5c939df5 YZ |
882 | if (last_end == (u64)-1) |
883 | goto out; | |
884 | start = last_end + 1; | |
d1ac6e41 LB |
885 | state = next_state(state); |
886 | if (start < end && state && state->start == start && | |
887 | !need_resched()) | |
888 | goto hit_next; | |
d1310b2e CM |
889 | goto search_again; |
890 | } | |
891 | ||
892 | /* | |
893 | * | ---- desired range ---- | | |
894 | * | state | | |
895 | * or | |
896 | * | ------------- state -------------- | | |
897 | * | |
898 | * We need to split the extent we found, and may flip bits on | |
899 | * second half. | |
900 | * | |
901 | * If the extent we found extends past our | |
902 | * range, we just split and search again. It'll get split | |
903 | * again the next time though. | |
904 | * | |
905 | * If the extent we found is inside our range, we set the | |
906 | * desired bit on it. | |
907 | */ | |
908 | if (state->start < start) { | |
1edbb734 | 909 | if (state->state & exclusive_bits) { |
d1310b2e CM |
910 | *failed_start = start; |
911 | err = -EEXIST; | |
912 | goto out; | |
913 | } | |
8233767a XG |
914 | |
915 | prealloc = alloc_extent_state_atomic(prealloc); | |
916 | BUG_ON(!prealloc); | |
d1310b2e | 917 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
918 | if (err) |
919 | extent_io_tree_panic(tree, err); | |
920 | ||
d1310b2e CM |
921 | prealloc = NULL; |
922 | if (err) | |
923 | goto out; | |
924 | if (state->end <= end) { | |
1bf85046 | 925 | set_state_bits(tree, state, &bits); |
2c64c53d | 926 | cache_state(state, cached_state); |
d1310b2e | 927 | merge_state(tree, state); |
5c939df5 YZ |
928 | if (last_end == (u64)-1) |
929 | goto out; | |
930 | start = last_end + 1; | |
d1ac6e41 LB |
931 | state = next_state(state); |
932 | if (start < end && state && state->start == start && | |
933 | !need_resched()) | |
934 | goto hit_next; | |
d1310b2e CM |
935 | } |
936 | goto search_again; | |
937 | } | |
938 | /* | |
939 | * | ---- desired range ---- | | |
940 | * | state | or | state | | |
941 | * | |
942 | * There's a hole, we need to insert something in it and | |
943 | * ignore the extent we found. | |
944 | */ | |
945 | if (state->start > start) { | |
946 | u64 this_end; | |
947 | if (end < last_start) | |
948 | this_end = end; | |
949 | else | |
d397712b | 950 | this_end = last_start - 1; |
8233767a XG |
951 | |
952 | prealloc = alloc_extent_state_atomic(prealloc); | |
953 | BUG_ON(!prealloc); | |
c7f895a2 XG |
954 | |
955 | /* | |
956 | * Avoid to free 'prealloc' if it can be merged with | |
957 | * the later extent. | |
958 | */ | |
d1310b2e | 959 | err = insert_state(tree, prealloc, start, this_end, |
12cfbad9 | 960 | NULL, NULL, &bits); |
c2d904e0 JM |
961 | if (err) |
962 | extent_io_tree_panic(tree, err); | |
963 | ||
9ed74f2d JB |
964 | cache_state(prealloc, cached_state); |
965 | prealloc = NULL; | |
d1310b2e CM |
966 | start = this_end + 1; |
967 | goto search_again; | |
968 | } | |
969 | /* | |
970 | * | ---- desired range ---- | | |
971 | * | state | | |
972 | * We need to split the extent, and set the bit | |
973 | * on the first half | |
974 | */ | |
975 | if (state->start <= end && state->end > end) { | |
1edbb734 | 976 | if (state->state & exclusive_bits) { |
d1310b2e CM |
977 | *failed_start = start; |
978 | err = -EEXIST; | |
979 | goto out; | |
980 | } | |
8233767a XG |
981 | |
982 | prealloc = alloc_extent_state_atomic(prealloc); | |
983 | BUG_ON(!prealloc); | |
d1310b2e | 984 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
985 | if (err) |
986 | extent_io_tree_panic(tree, err); | |
d1310b2e | 987 | |
1bf85046 | 988 | set_state_bits(tree, prealloc, &bits); |
2c64c53d | 989 | cache_state(prealloc, cached_state); |
d1310b2e CM |
990 | merge_state(tree, prealloc); |
991 | prealloc = NULL; | |
992 | goto out; | |
993 | } | |
994 | ||
995 | goto search_again; | |
996 | ||
997 | out: | |
cad321ad | 998 | spin_unlock(&tree->lock); |
d1310b2e CM |
999 | if (prealloc) |
1000 | free_extent_state(prealloc); | |
1001 | ||
1002 | return err; | |
1003 | ||
1004 | search_again: | |
1005 | if (start > end) | |
1006 | goto out; | |
cad321ad | 1007 | spin_unlock(&tree->lock); |
d1310b2e CM |
1008 | if (mask & __GFP_WAIT) |
1009 | cond_resched(); | |
1010 | goto again; | |
1011 | } | |
d1310b2e | 1012 | |
41074888 DS |
1013 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
1014 | unsigned long bits, u64 * failed_start, | |
1015 | struct extent_state **cached_state, gfp_t mask) | |
3fbe5c02 JM |
1016 | { |
1017 | return __set_extent_bit(tree, start, end, bits, 0, failed_start, | |
1018 | cached_state, mask); | |
1019 | } | |
1020 | ||
1021 | ||
462d6fac | 1022 | /** |
10983f2e LB |
1023 | * convert_extent_bit - convert all bits in a given range from one bit to |
1024 | * another | |
462d6fac JB |
1025 | * @tree: the io tree to search |
1026 | * @start: the start offset in bytes | |
1027 | * @end: the end offset in bytes (inclusive) | |
1028 | * @bits: the bits to set in this range | |
1029 | * @clear_bits: the bits to clear in this range | |
e6138876 | 1030 | * @cached_state: state that we're going to cache |
462d6fac JB |
1031 | * @mask: the allocation mask |
1032 | * | |
1033 | * This will go through and set bits for the given range. If any states exist | |
1034 | * already in this range they are set with the given bit and cleared of the | |
1035 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
1036 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
1037 | * boundary bits like LOCK. | |
1038 | */ | |
1039 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
41074888 | 1040 | unsigned long bits, unsigned long clear_bits, |
e6138876 | 1041 | struct extent_state **cached_state, gfp_t mask) |
462d6fac JB |
1042 | { |
1043 | struct extent_state *state; | |
1044 | struct extent_state *prealloc = NULL; | |
1045 | struct rb_node *node; | |
12cfbad9 FDBM |
1046 | struct rb_node **p; |
1047 | struct rb_node *parent; | |
462d6fac JB |
1048 | int err = 0; |
1049 | u64 last_start; | |
1050 | u64 last_end; | |
1051 | ||
a5dee37d | 1052 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 1053 | |
462d6fac JB |
1054 | again: |
1055 | if (!prealloc && (mask & __GFP_WAIT)) { | |
1056 | prealloc = alloc_extent_state(mask); | |
1057 | if (!prealloc) | |
1058 | return -ENOMEM; | |
1059 | } | |
1060 | ||
1061 | spin_lock(&tree->lock); | |
e6138876 JB |
1062 | if (cached_state && *cached_state) { |
1063 | state = *cached_state; | |
1064 | if (state->start <= start && state->end > start && | |
1065 | state->tree) { | |
1066 | node = &state->rb_node; | |
1067 | goto hit_next; | |
1068 | } | |
1069 | } | |
1070 | ||
462d6fac JB |
1071 | /* |
1072 | * this search will find all the extents that end after | |
1073 | * our range starts. | |
1074 | */ | |
12cfbad9 | 1075 | node = tree_search_for_insert(tree, start, &p, &parent); |
462d6fac JB |
1076 | if (!node) { |
1077 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1078 | if (!prealloc) { |
1079 | err = -ENOMEM; | |
1080 | goto out; | |
1081 | } | |
12cfbad9 FDBM |
1082 | err = insert_state(tree, prealloc, start, end, |
1083 | &p, &parent, &bits); | |
c2d904e0 JM |
1084 | if (err) |
1085 | extent_io_tree_panic(tree, err); | |
c42ac0bc FDBM |
1086 | cache_state(prealloc, cached_state); |
1087 | prealloc = NULL; | |
462d6fac JB |
1088 | goto out; |
1089 | } | |
1090 | state = rb_entry(node, struct extent_state, rb_node); | |
1091 | hit_next: | |
1092 | last_start = state->start; | |
1093 | last_end = state->end; | |
1094 | ||
1095 | /* | |
1096 | * | ---- desired range ---- | | |
1097 | * | state | | |
1098 | * | |
1099 | * Just lock what we found and keep going | |
1100 | */ | |
1101 | if (state->start == start && state->end <= end) { | |
462d6fac | 1102 | set_state_bits(tree, state, &bits); |
e6138876 | 1103 | cache_state(state, cached_state); |
d1ac6e41 | 1104 | state = clear_state_bit(tree, state, &clear_bits, 0); |
462d6fac JB |
1105 | if (last_end == (u64)-1) |
1106 | goto out; | |
462d6fac | 1107 | start = last_end + 1; |
d1ac6e41 LB |
1108 | if (start < end && state && state->start == start && |
1109 | !need_resched()) | |
1110 | goto hit_next; | |
462d6fac JB |
1111 | goto search_again; |
1112 | } | |
1113 | ||
1114 | /* | |
1115 | * | ---- desired range ---- | | |
1116 | * | state | | |
1117 | * or | |
1118 | * | ------------- state -------------- | | |
1119 | * | |
1120 | * We need to split the extent we found, and may flip bits on | |
1121 | * second half. | |
1122 | * | |
1123 | * If the extent we found extends past our | |
1124 | * range, we just split and search again. It'll get split | |
1125 | * again the next time though. | |
1126 | * | |
1127 | * If the extent we found is inside our range, we set the | |
1128 | * desired bit on it. | |
1129 | */ | |
1130 | if (state->start < start) { | |
1131 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1132 | if (!prealloc) { |
1133 | err = -ENOMEM; | |
1134 | goto out; | |
1135 | } | |
462d6fac | 1136 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1137 | if (err) |
1138 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1139 | prealloc = NULL; |
1140 | if (err) | |
1141 | goto out; | |
1142 | if (state->end <= end) { | |
1143 | set_state_bits(tree, state, &bits); | |
e6138876 | 1144 | cache_state(state, cached_state); |
d1ac6e41 | 1145 | state = clear_state_bit(tree, state, &clear_bits, 0); |
462d6fac JB |
1146 | if (last_end == (u64)-1) |
1147 | goto out; | |
1148 | start = last_end + 1; | |
d1ac6e41 LB |
1149 | if (start < end && state && state->start == start && |
1150 | !need_resched()) | |
1151 | goto hit_next; | |
462d6fac JB |
1152 | } |
1153 | goto search_again; | |
1154 | } | |
1155 | /* | |
1156 | * | ---- desired range ---- | | |
1157 | * | state | or | state | | |
1158 | * | |
1159 | * There's a hole, we need to insert something in it and | |
1160 | * ignore the extent we found. | |
1161 | */ | |
1162 | if (state->start > start) { | |
1163 | u64 this_end; | |
1164 | if (end < last_start) | |
1165 | this_end = end; | |
1166 | else | |
1167 | this_end = last_start - 1; | |
1168 | ||
1169 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1170 | if (!prealloc) { |
1171 | err = -ENOMEM; | |
1172 | goto out; | |
1173 | } | |
462d6fac JB |
1174 | |
1175 | /* | |
1176 | * Avoid to free 'prealloc' if it can be merged with | |
1177 | * the later extent. | |
1178 | */ | |
1179 | err = insert_state(tree, prealloc, start, this_end, | |
12cfbad9 | 1180 | NULL, NULL, &bits); |
c2d904e0 JM |
1181 | if (err) |
1182 | extent_io_tree_panic(tree, err); | |
e6138876 | 1183 | cache_state(prealloc, cached_state); |
462d6fac JB |
1184 | prealloc = NULL; |
1185 | start = this_end + 1; | |
1186 | goto search_again; | |
1187 | } | |
1188 | /* | |
1189 | * | ---- desired range ---- | | |
1190 | * | state | | |
1191 | * We need to split the extent, and set the bit | |
1192 | * on the first half | |
1193 | */ | |
1194 | if (state->start <= end && state->end > end) { | |
1195 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1196 | if (!prealloc) { |
1197 | err = -ENOMEM; | |
1198 | goto out; | |
1199 | } | |
462d6fac JB |
1200 | |
1201 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1202 | if (err) |
1203 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1204 | |
1205 | set_state_bits(tree, prealloc, &bits); | |
e6138876 | 1206 | cache_state(prealloc, cached_state); |
462d6fac | 1207 | clear_state_bit(tree, prealloc, &clear_bits, 0); |
462d6fac JB |
1208 | prealloc = NULL; |
1209 | goto out; | |
1210 | } | |
1211 | ||
1212 | goto search_again; | |
1213 | ||
1214 | out: | |
1215 | spin_unlock(&tree->lock); | |
1216 | if (prealloc) | |
1217 | free_extent_state(prealloc); | |
1218 | ||
1219 | return err; | |
1220 | ||
1221 | search_again: | |
1222 | if (start > end) | |
1223 | goto out; | |
1224 | spin_unlock(&tree->lock); | |
1225 | if (mask & __GFP_WAIT) | |
1226 | cond_resched(); | |
1227 | goto again; | |
1228 | } | |
1229 | ||
d1310b2e CM |
1230 | /* wrappers around set/clear extent bit */ |
1231 | int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
1232 | gfp_t mask) | |
1233 | { | |
3fbe5c02 | 1234 | return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL, |
2c64c53d | 1235 | NULL, mask); |
d1310b2e | 1236 | } |
d1310b2e CM |
1237 | |
1238 | int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
41074888 | 1239 | unsigned long bits, gfp_t mask) |
d1310b2e | 1240 | { |
3fbe5c02 | 1241 | return set_extent_bit(tree, start, end, bits, NULL, |
2c64c53d | 1242 | NULL, mask); |
d1310b2e | 1243 | } |
d1310b2e CM |
1244 | |
1245 | int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
41074888 | 1246 | unsigned long bits, gfp_t mask) |
d1310b2e | 1247 | { |
2c64c53d | 1248 | return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask); |
d1310b2e | 1249 | } |
d1310b2e CM |
1250 | |
1251 | int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, | |
2ac55d41 | 1252 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
1253 | { |
1254 | return set_extent_bit(tree, start, end, | |
fee187d9 | 1255 | EXTENT_DELALLOC | EXTENT_UPTODATE, |
3fbe5c02 | 1256 | NULL, cached_state, mask); |
d1310b2e | 1257 | } |
d1310b2e | 1258 | |
9e8a4a8b LB |
1259 | int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end, |
1260 | struct extent_state **cached_state, gfp_t mask) | |
1261 | { | |
1262 | return set_extent_bit(tree, start, end, | |
1263 | EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG, | |
1264 | NULL, cached_state, mask); | |
1265 | } | |
1266 | ||
d1310b2e CM |
1267 | int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, |
1268 | gfp_t mask) | |
1269 | { | |
1270 | return clear_extent_bit(tree, start, end, | |
32c00aff | 1271 | EXTENT_DIRTY | EXTENT_DELALLOC | |
0ca1f7ce | 1272 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask); |
d1310b2e | 1273 | } |
d1310b2e CM |
1274 | |
1275 | int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, | |
1276 | gfp_t mask) | |
1277 | { | |
3fbe5c02 | 1278 | return set_extent_bit(tree, start, end, EXTENT_NEW, NULL, |
2c64c53d | 1279 | NULL, mask); |
d1310b2e | 1280 | } |
d1310b2e | 1281 | |
d1310b2e | 1282 | int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
507903b8 | 1283 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e | 1284 | { |
6b67a320 | 1285 | return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL, |
3fbe5c02 | 1286 | cached_state, mask); |
d1310b2e | 1287 | } |
d1310b2e | 1288 | |
5fd02043 JB |
1289 | int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
1290 | struct extent_state **cached_state, gfp_t mask) | |
d1310b2e | 1291 | { |
2c64c53d | 1292 | return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, |
2ac55d41 | 1293 | cached_state, mask); |
d1310b2e | 1294 | } |
d1310b2e | 1295 | |
d352ac68 CM |
1296 | /* |
1297 | * either insert or lock state struct between start and end use mask to tell | |
1298 | * us if waiting is desired. | |
1299 | */ | |
1edbb734 | 1300 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
41074888 | 1301 | unsigned long bits, struct extent_state **cached_state) |
d1310b2e CM |
1302 | { |
1303 | int err; | |
1304 | u64 failed_start; | |
1305 | while (1) { | |
3fbe5c02 JM |
1306 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits, |
1307 | EXTENT_LOCKED, &failed_start, | |
1308 | cached_state, GFP_NOFS); | |
d0082371 | 1309 | if (err == -EEXIST) { |
d1310b2e CM |
1310 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1311 | start = failed_start; | |
d0082371 | 1312 | } else |
d1310b2e | 1313 | break; |
d1310b2e CM |
1314 | WARN_ON(start > end); |
1315 | } | |
1316 | return err; | |
1317 | } | |
d1310b2e | 1318 | |
d0082371 | 1319 | int lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
1edbb734 | 1320 | { |
d0082371 | 1321 | return lock_extent_bits(tree, start, end, 0, NULL); |
1edbb734 CM |
1322 | } |
1323 | ||
d0082371 | 1324 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1325 | { |
1326 | int err; | |
1327 | u64 failed_start; | |
1328 | ||
3fbe5c02 JM |
1329 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
1330 | &failed_start, NULL, GFP_NOFS); | |
6643558d YZ |
1331 | if (err == -EEXIST) { |
1332 | if (failed_start > start) | |
1333 | clear_extent_bit(tree, start, failed_start - 1, | |
d0082371 | 1334 | EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS); |
25179201 | 1335 | return 0; |
6643558d | 1336 | } |
25179201 JB |
1337 | return 1; |
1338 | } | |
25179201 | 1339 | |
2c64c53d CM |
1340 | int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end, |
1341 | struct extent_state **cached, gfp_t mask) | |
1342 | { | |
1343 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, | |
1344 | mask); | |
1345 | } | |
1346 | ||
d0082371 | 1347 | int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e | 1348 | { |
2c64c53d | 1349 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL, |
d0082371 | 1350 | GFP_NOFS); |
d1310b2e | 1351 | } |
d1310b2e | 1352 | |
4adaa611 CM |
1353 | int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
1354 | { | |
1355 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1356 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1357 | struct page *page; | |
1358 | ||
1359 | while (index <= end_index) { | |
1360 | page = find_get_page(inode->i_mapping, index); | |
1361 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1362 | clear_page_dirty_for_io(page); | |
1363 | page_cache_release(page); | |
1364 | index++; | |
1365 | } | |
1366 | return 0; | |
1367 | } | |
1368 | ||
1369 | int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) | |
1370 | { | |
1371 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1372 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1373 | struct page *page; | |
1374 | ||
1375 | while (index <= end_index) { | |
1376 | page = find_get_page(inode->i_mapping, index); | |
1377 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1378 | account_page_redirty(page); | |
1379 | __set_page_dirty_nobuffers(page); | |
1380 | page_cache_release(page); | |
1381 | index++; | |
1382 | } | |
1383 | return 0; | |
1384 | } | |
1385 | ||
d1310b2e CM |
1386 | /* |
1387 | * helper function to set both pages and extents in the tree writeback | |
1388 | */ | |
b2950863 | 1389 | static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e CM |
1390 | { |
1391 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1392 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1393 | struct page *page; | |
1394 | ||
1395 | while (index <= end_index) { | |
1396 | page = find_get_page(tree->mapping, index); | |
79787eaa | 1397 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ |
d1310b2e CM |
1398 | set_page_writeback(page); |
1399 | page_cache_release(page); | |
1400 | index++; | |
1401 | } | |
d1310b2e CM |
1402 | return 0; |
1403 | } | |
d1310b2e | 1404 | |
d352ac68 CM |
1405 | /* find the first state struct with 'bits' set after 'start', and |
1406 | * return it. tree->lock must be held. NULL will returned if | |
1407 | * nothing was found after 'start' | |
1408 | */ | |
48a3b636 ES |
1409 | static struct extent_state * |
1410 | find_first_extent_bit_state(struct extent_io_tree *tree, | |
41074888 | 1411 | u64 start, unsigned long bits) |
d7fc640e CM |
1412 | { |
1413 | struct rb_node *node; | |
1414 | struct extent_state *state; | |
1415 | ||
1416 | /* | |
1417 | * this search will find all the extents that end after | |
1418 | * our range starts. | |
1419 | */ | |
1420 | node = tree_search(tree, start); | |
d397712b | 1421 | if (!node) |
d7fc640e | 1422 | goto out; |
d7fc640e | 1423 | |
d397712b | 1424 | while (1) { |
d7fc640e | 1425 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1426 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1427 | return state; |
d397712b | 1428 | |
d7fc640e CM |
1429 | node = rb_next(node); |
1430 | if (!node) | |
1431 | break; | |
1432 | } | |
1433 | out: | |
1434 | return NULL; | |
1435 | } | |
d7fc640e | 1436 | |
69261c4b XG |
1437 | /* |
1438 | * find the first offset in the io tree with 'bits' set. zero is | |
1439 | * returned if we find something, and *start_ret and *end_ret are | |
1440 | * set to reflect the state struct that was found. | |
1441 | * | |
477d7eaf | 1442 | * If nothing was found, 1 is returned. If found something, return 0. |
69261c4b XG |
1443 | */ |
1444 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
41074888 | 1445 | u64 *start_ret, u64 *end_ret, unsigned long bits, |
e6138876 | 1446 | struct extent_state **cached_state) |
69261c4b XG |
1447 | { |
1448 | struct extent_state *state; | |
e6138876 | 1449 | struct rb_node *n; |
69261c4b XG |
1450 | int ret = 1; |
1451 | ||
1452 | spin_lock(&tree->lock); | |
e6138876 JB |
1453 | if (cached_state && *cached_state) { |
1454 | state = *cached_state; | |
1455 | if (state->end == start - 1 && state->tree) { | |
1456 | n = rb_next(&state->rb_node); | |
1457 | while (n) { | |
1458 | state = rb_entry(n, struct extent_state, | |
1459 | rb_node); | |
1460 | if (state->state & bits) | |
1461 | goto got_it; | |
1462 | n = rb_next(n); | |
1463 | } | |
1464 | free_extent_state(*cached_state); | |
1465 | *cached_state = NULL; | |
1466 | goto out; | |
1467 | } | |
1468 | free_extent_state(*cached_state); | |
1469 | *cached_state = NULL; | |
1470 | } | |
1471 | ||
69261c4b | 1472 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1473 | got_it: |
69261c4b | 1474 | if (state) { |
e6138876 | 1475 | cache_state(state, cached_state); |
69261c4b XG |
1476 | *start_ret = state->start; |
1477 | *end_ret = state->end; | |
1478 | ret = 0; | |
1479 | } | |
e6138876 | 1480 | out: |
69261c4b XG |
1481 | spin_unlock(&tree->lock); |
1482 | return ret; | |
1483 | } | |
1484 | ||
d352ac68 CM |
1485 | /* |
1486 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1487 | * more than 'max_bytes'. start and end are used to return the range, | |
1488 | * | |
1489 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1490 | */ | |
c8b97818 | 1491 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1492 | u64 *start, u64 *end, u64 max_bytes, |
1493 | struct extent_state **cached_state) | |
d1310b2e CM |
1494 | { |
1495 | struct rb_node *node; | |
1496 | struct extent_state *state; | |
1497 | u64 cur_start = *start; | |
1498 | u64 found = 0; | |
1499 | u64 total_bytes = 0; | |
1500 | ||
cad321ad | 1501 | spin_lock(&tree->lock); |
c8b97818 | 1502 | |
d1310b2e CM |
1503 | /* |
1504 | * this search will find all the extents that end after | |
1505 | * our range starts. | |
1506 | */ | |
80ea96b1 | 1507 | node = tree_search(tree, cur_start); |
2b114d1d | 1508 | if (!node) { |
3b951516 CM |
1509 | if (!found) |
1510 | *end = (u64)-1; | |
d1310b2e CM |
1511 | goto out; |
1512 | } | |
1513 | ||
d397712b | 1514 | while (1) { |
d1310b2e | 1515 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1516 | if (found && (state->start != cur_start || |
1517 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1518 | goto out; |
1519 | } | |
1520 | if (!(state->state & EXTENT_DELALLOC)) { | |
1521 | if (!found) | |
1522 | *end = state->end; | |
1523 | goto out; | |
1524 | } | |
c2a128d2 | 1525 | if (!found) { |
d1310b2e | 1526 | *start = state->start; |
c2a128d2 JB |
1527 | *cached_state = state; |
1528 | atomic_inc(&state->refs); | |
1529 | } | |
d1310b2e CM |
1530 | found++; |
1531 | *end = state->end; | |
1532 | cur_start = state->end + 1; | |
1533 | node = rb_next(node); | |
d1310b2e | 1534 | total_bytes += state->end - state->start + 1; |
7bf811a5 | 1535 | if (total_bytes >= max_bytes) |
573aecaf | 1536 | break; |
573aecaf | 1537 | if (!node) |
d1310b2e CM |
1538 | break; |
1539 | } | |
1540 | out: | |
cad321ad | 1541 | spin_unlock(&tree->lock); |
d1310b2e CM |
1542 | return found; |
1543 | } | |
1544 | ||
143bede5 JM |
1545 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1546 | struct page *locked_page, | |
1547 | u64 start, u64 end) | |
c8b97818 CM |
1548 | { |
1549 | int ret; | |
1550 | struct page *pages[16]; | |
1551 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1552 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1553 | unsigned long nr_pages = end_index - index + 1; | |
1554 | int i; | |
1555 | ||
1556 | if (index == locked_page->index && end_index == index) | |
143bede5 | 1557 | return; |
c8b97818 | 1558 | |
d397712b | 1559 | while (nr_pages > 0) { |
c8b97818 | 1560 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1561 | min_t(unsigned long, nr_pages, |
1562 | ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1563 | for (i = 0; i < ret; i++) { |
1564 | if (pages[i] != locked_page) | |
1565 | unlock_page(pages[i]); | |
1566 | page_cache_release(pages[i]); | |
1567 | } | |
1568 | nr_pages -= ret; | |
1569 | index += ret; | |
1570 | cond_resched(); | |
1571 | } | |
c8b97818 CM |
1572 | } |
1573 | ||
1574 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1575 | struct page *locked_page, | |
1576 | u64 delalloc_start, | |
1577 | u64 delalloc_end) | |
1578 | { | |
1579 | unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT; | |
1580 | unsigned long start_index = index; | |
1581 | unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT; | |
1582 | unsigned long pages_locked = 0; | |
1583 | struct page *pages[16]; | |
1584 | unsigned long nrpages; | |
1585 | int ret; | |
1586 | int i; | |
1587 | ||
1588 | /* the caller is responsible for locking the start index */ | |
1589 | if (index == locked_page->index && index == end_index) | |
1590 | return 0; | |
1591 | ||
1592 | /* skip the page at the start index */ | |
1593 | nrpages = end_index - index + 1; | |
d397712b | 1594 | while (nrpages > 0) { |
c8b97818 | 1595 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1596 | min_t(unsigned long, |
1597 | nrpages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1598 | if (ret == 0) { |
1599 | ret = -EAGAIN; | |
1600 | goto done; | |
1601 | } | |
1602 | /* now we have an array of pages, lock them all */ | |
1603 | for (i = 0; i < ret; i++) { | |
1604 | /* | |
1605 | * the caller is taking responsibility for | |
1606 | * locked_page | |
1607 | */ | |
771ed689 | 1608 | if (pages[i] != locked_page) { |
c8b97818 | 1609 | lock_page(pages[i]); |
f2b1c41c CM |
1610 | if (!PageDirty(pages[i]) || |
1611 | pages[i]->mapping != inode->i_mapping) { | |
771ed689 CM |
1612 | ret = -EAGAIN; |
1613 | unlock_page(pages[i]); | |
1614 | page_cache_release(pages[i]); | |
1615 | goto done; | |
1616 | } | |
1617 | } | |
c8b97818 | 1618 | page_cache_release(pages[i]); |
771ed689 | 1619 | pages_locked++; |
c8b97818 | 1620 | } |
c8b97818 CM |
1621 | nrpages -= ret; |
1622 | index += ret; | |
1623 | cond_resched(); | |
1624 | } | |
1625 | ret = 0; | |
1626 | done: | |
1627 | if (ret && pages_locked) { | |
1628 | __unlock_for_delalloc(inode, locked_page, | |
1629 | delalloc_start, | |
1630 | ((u64)(start_index + pages_locked - 1)) << | |
1631 | PAGE_CACHE_SHIFT); | |
1632 | } | |
1633 | return ret; | |
1634 | } | |
1635 | ||
1636 | /* | |
1637 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1638 | * more than 'max_bytes'. start and end are used to return the range, | |
1639 | * | |
1640 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1641 | */ | |
294e30fe JB |
1642 | STATIC u64 find_lock_delalloc_range(struct inode *inode, |
1643 | struct extent_io_tree *tree, | |
1644 | struct page *locked_page, u64 *start, | |
1645 | u64 *end, u64 max_bytes) | |
c8b97818 CM |
1646 | { |
1647 | u64 delalloc_start; | |
1648 | u64 delalloc_end; | |
1649 | u64 found; | |
9655d298 | 1650 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1651 | int ret; |
1652 | int loops = 0; | |
1653 | ||
1654 | again: | |
1655 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1656 | delalloc_start = *start; | |
1657 | delalloc_end = 0; | |
1658 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1659 | max_bytes, &cached_state); |
70b99e69 | 1660 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1661 | *start = delalloc_start; |
1662 | *end = delalloc_end; | |
c2a128d2 | 1663 | free_extent_state(cached_state); |
385fe0be | 1664 | return 0; |
c8b97818 CM |
1665 | } |
1666 | ||
70b99e69 CM |
1667 | /* |
1668 | * start comes from the offset of locked_page. We have to lock | |
1669 | * pages in order, so we can't process delalloc bytes before | |
1670 | * locked_page | |
1671 | */ | |
d397712b | 1672 | if (delalloc_start < *start) |
70b99e69 | 1673 | delalloc_start = *start; |
70b99e69 | 1674 | |
c8b97818 CM |
1675 | /* |
1676 | * make sure to limit the number of pages we try to lock down | |
c8b97818 | 1677 | */ |
7bf811a5 JB |
1678 | if (delalloc_end + 1 - delalloc_start > max_bytes) |
1679 | delalloc_end = delalloc_start + max_bytes - 1; | |
d397712b | 1680 | |
c8b97818 CM |
1681 | /* step two, lock all the pages after the page that has start */ |
1682 | ret = lock_delalloc_pages(inode, locked_page, | |
1683 | delalloc_start, delalloc_end); | |
1684 | if (ret == -EAGAIN) { | |
1685 | /* some of the pages are gone, lets avoid looping by | |
1686 | * shortening the size of the delalloc range we're searching | |
1687 | */ | |
9655d298 | 1688 | free_extent_state(cached_state); |
c8b97818 | 1689 | if (!loops) { |
7bf811a5 | 1690 | max_bytes = PAGE_CACHE_SIZE; |
c8b97818 CM |
1691 | loops = 1; |
1692 | goto again; | |
1693 | } else { | |
1694 | found = 0; | |
1695 | goto out_failed; | |
1696 | } | |
1697 | } | |
79787eaa | 1698 | BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */ |
c8b97818 CM |
1699 | |
1700 | /* step three, lock the state bits for the whole range */ | |
d0082371 | 1701 | lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state); |
c8b97818 CM |
1702 | |
1703 | /* then test to make sure it is all still delalloc */ | |
1704 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1705 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1706 | if (!ret) { |
9655d298 CM |
1707 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
1708 | &cached_state, GFP_NOFS); | |
c8b97818 CM |
1709 | __unlock_for_delalloc(inode, locked_page, |
1710 | delalloc_start, delalloc_end); | |
1711 | cond_resched(); | |
1712 | goto again; | |
1713 | } | |
9655d298 | 1714 | free_extent_state(cached_state); |
c8b97818 CM |
1715 | *start = delalloc_start; |
1716 | *end = delalloc_end; | |
1717 | out_failed: | |
1718 | return found; | |
1719 | } | |
1720 | ||
c2790a2e JB |
1721 | int extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end, |
1722 | struct page *locked_page, | |
1723 | unsigned long clear_bits, | |
1724 | unsigned long page_ops) | |
c8b97818 | 1725 | { |
c2790a2e | 1726 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
c8b97818 CM |
1727 | int ret; |
1728 | struct page *pages[16]; | |
1729 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1730 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1731 | unsigned long nr_pages = end_index - index + 1; | |
1732 | int i; | |
771ed689 | 1733 | |
2c64c53d | 1734 | clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); |
c2790a2e | 1735 | if (page_ops == 0) |
771ed689 | 1736 | return 0; |
c8b97818 | 1737 | |
d397712b | 1738 | while (nr_pages > 0) { |
c8b97818 | 1739 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1740 | min_t(unsigned long, |
1741 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 | 1742 | for (i = 0; i < ret; i++) { |
8b62b72b | 1743 | |
c2790a2e | 1744 | if (page_ops & PAGE_SET_PRIVATE2) |
8b62b72b CM |
1745 | SetPagePrivate2(pages[i]); |
1746 | ||
c8b97818 CM |
1747 | if (pages[i] == locked_page) { |
1748 | page_cache_release(pages[i]); | |
1749 | continue; | |
1750 | } | |
c2790a2e | 1751 | if (page_ops & PAGE_CLEAR_DIRTY) |
c8b97818 | 1752 | clear_page_dirty_for_io(pages[i]); |
c2790a2e | 1753 | if (page_ops & PAGE_SET_WRITEBACK) |
c8b97818 | 1754 | set_page_writeback(pages[i]); |
c2790a2e | 1755 | if (page_ops & PAGE_END_WRITEBACK) |
c8b97818 | 1756 | end_page_writeback(pages[i]); |
c2790a2e | 1757 | if (page_ops & PAGE_UNLOCK) |
771ed689 | 1758 | unlock_page(pages[i]); |
c8b97818 CM |
1759 | page_cache_release(pages[i]); |
1760 | } | |
1761 | nr_pages -= ret; | |
1762 | index += ret; | |
1763 | cond_resched(); | |
1764 | } | |
1765 | return 0; | |
1766 | } | |
c8b97818 | 1767 | |
d352ac68 CM |
1768 | /* |
1769 | * count the number of bytes in the tree that have a given bit(s) | |
1770 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1771 | * cached. The total number found is returned. | |
1772 | */ | |
d1310b2e CM |
1773 | u64 count_range_bits(struct extent_io_tree *tree, |
1774 | u64 *start, u64 search_end, u64 max_bytes, | |
ec29ed5b | 1775 | unsigned long bits, int contig) |
d1310b2e CM |
1776 | { |
1777 | struct rb_node *node; | |
1778 | struct extent_state *state; | |
1779 | u64 cur_start = *start; | |
1780 | u64 total_bytes = 0; | |
ec29ed5b | 1781 | u64 last = 0; |
d1310b2e CM |
1782 | int found = 0; |
1783 | ||
fae7f21c | 1784 | if (WARN_ON(search_end <= cur_start)) |
d1310b2e | 1785 | return 0; |
d1310b2e | 1786 | |
cad321ad | 1787 | spin_lock(&tree->lock); |
d1310b2e CM |
1788 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1789 | total_bytes = tree->dirty_bytes; | |
1790 | goto out; | |
1791 | } | |
1792 | /* | |
1793 | * this search will find all the extents that end after | |
1794 | * our range starts. | |
1795 | */ | |
80ea96b1 | 1796 | node = tree_search(tree, cur_start); |
d397712b | 1797 | if (!node) |
d1310b2e | 1798 | goto out; |
d1310b2e | 1799 | |
d397712b | 1800 | while (1) { |
d1310b2e CM |
1801 | state = rb_entry(node, struct extent_state, rb_node); |
1802 | if (state->start > search_end) | |
1803 | break; | |
ec29ed5b CM |
1804 | if (contig && found && state->start > last + 1) |
1805 | break; | |
1806 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1807 | total_bytes += min(search_end, state->end) + 1 - |
1808 | max(cur_start, state->start); | |
1809 | if (total_bytes >= max_bytes) | |
1810 | break; | |
1811 | if (!found) { | |
af60bed2 | 1812 | *start = max(cur_start, state->start); |
d1310b2e CM |
1813 | found = 1; |
1814 | } | |
ec29ed5b CM |
1815 | last = state->end; |
1816 | } else if (contig && found) { | |
1817 | break; | |
d1310b2e CM |
1818 | } |
1819 | node = rb_next(node); | |
1820 | if (!node) | |
1821 | break; | |
1822 | } | |
1823 | out: | |
cad321ad | 1824 | spin_unlock(&tree->lock); |
d1310b2e CM |
1825 | return total_bytes; |
1826 | } | |
b2950863 | 1827 | |
d352ac68 CM |
1828 | /* |
1829 | * set the private field for a given byte offset in the tree. If there isn't | |
1830 | * an extent_state there already, this does nothing. | |
1831 | */ | |
171170c1 | 1832 | static int set_state_private(struct extent_io_tree *tree, u64 start, u64 private) |
d1310b2e CM |
1833 | { |
1834 | struct rb_node *node; | |
1835 | struct extent_state *state; | |
1836 | int ret = 0; | |
1837 | ||
cad321ad | 1838 | spin_lock(&tree->lock); |
d1310b2e CM |
1839 | /* |
1840 | * this search will find all the extents that end after | |
1841 | * our range starts. | |
1842 | */ | |
80ea96b1 | 1843 | node = tree_search(tree, start); |
2b114d1d | 1844 | if (!node) { |
d1310b2e CM |
1845 | ret = -ENOENT; |
1846 | goto out; | |
1847 | } | |
1848 | state = rb_entry(node, struct extent_state, rb_node); | |
1849 | if (state->start != start) { | |
1850 | ret = -ENOENT; | |
1851 | goto out; | |
1852 | } | |
1853 | state->private = private; | |
1854 | out: | |
cad321ad | 1855 | spin_unlock(&tree->lock); |
d1310b2e CM |
1856 | return ret; |
1857 | } | |
1858 | ||
1859 | int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private) | |
1860 | { | |
1861 | struct rb_node *node; | |
1862 | struct extent_state *state; | |
1863 | int ret = 0; | |
1864 | ||
cad321ad | 1865 | spin_lock(&tree->lock); |
d1310b2e CM |
1866 | /* |
1867 | * this search will find all the extents that end after | |
1868 | * our range starts. | |
1869 | */ | |
80ea96b1 | 1870 | node = tree_search(tree, start); |
2b114d1d | 1871 | if (!node) { |
d1310b2e CM |
1872 | ret = -ENOENT; |
1873 | goto out; | |
1874 | } | |
1875 | state = rb_entry(node, struct extent_state, rb_node); | |
1876 | if (state->start != start) { | |
1877 | ret = -ENOENT; | |
1878 | goto out; | |
1879 | } | |
1880 | *private = state->private; | |
1881 | out: | |
cad321ad | 1882 | spin_unlock(&tree->lock); |
d1310b2e CM |
1883 | return ret; |
1884 | } | |
1885 | ||
1886 | /* | |
1887 | * searches a range in the state tree for a given mask. | |
70dec807 | 1888 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1889 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1890 | * range is found set. | |
1891 | */ | |
1892 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
41074888 | 1893 | unsigned long bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1894 | { |
1895 | struct extent_state *state = NULL; | |
1896 | struct rb_node *node; | |
1897 | int bitset = 0; | |
d1310b2e | 1898 | |
cad321ad | 1899 | spin_lock(&tree->lock); |
df98b6e2 JB |
1900 | if (cached && cached->tree && cached->start <= start && |
1901 | cached->end > start) | |
9655d298 CM |
1902 | node = &cached->rb_node; |
1903 | else | |
1904 | node = tree_search(tree, start); | |
d1310b2e CM |
1905 | while (node && start <= end) { |
1906 | state = rb_entry(node, struct extent_state, rb_node); | |
1907 | ||
1908 | if (filled && state->start > start) { | |
1909 | bitset = 0; | |
1910 | break; | |
1911 | } | |
1912 | ||
1913 | if (state->start > end) | |
1914 | break; | |
1915 | ||
1916 | if (state->state & bits) { | |
1917 | bitset = 1; | |
1918 | if (!filled) | |
1919 | break; | |
1920 | } else if (filled) { | |
1921 | bitset = 0; | |
1922 | break; | |
1923 | } | |
46562cec CM |
1924 | |
1925 | if (state->end == (u64)-1) | |
1926 | break; | |
1927 | ||
d1310b2e CM |
1928 | start = state->end + 1; |
1929 | if (start > end) | |
1930 | break; | |
1931 | node = rb_next(node); | |
1932 | if (!node) { | |
1933 | if (filled) | |
1934 | bitset = 0; | |
1935 | break; | |
1936 | } | |
1937 | } | |
cad321ad | 1938 | spin_unlock(&tree->lock); |
d1310b2e CM |
1939 | return bitset; |
1940 | } | |
d1310b2e CM |
1941 | |
1942 | /* | |
1943 | * helper function to set a given page up to date if all the | |
1944 | * extents in the tree for that page are up to date | |
1945 | */ | |
143bede5 | 1946 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e | 1947 | { |
4eee4fa4 | 1948 | u64 start = page_offset(page); |
d1310b2e | 1949 | u64 end = start + PAGE_CACHE_SIZE - 1; |
9655d298 | 1950 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 1951 | SetPageUptodate(page); |
d1310b2e CM |
1952 | } |
1953 | ||
4a54c8c1 JS |
1954 | /* |
1955 | * When IO fails, either with EIO or csum verification fails, we | |
1956 | * try other mirrors that might have a good copy of the data. This | |
1957 | * io_failure_record is used to record state as we go through all the | |
1958 | * mirrors. If another mirror has good data, the page is set up to date | |
1959 | * and things continue. If a good mirror can't be found, the original | |
1960 | * bio end_io callback is called to indicate things have failed. | |
1961 | */ | |
1962 | struct io_failure_record { | |
1963 | struct page *page; | |
1964 | u64 start; | |
1965 | u64 len; | |
1966 | u64 logical; | |
1967 | unsigned long bio_flags; | |
1968 | int this_mirror; | |
1969 | int failed_mirror; | |
1970 | int in_validation; | |
1971 | }; | |
1972 | ||
1973 | static int free_io_failure(struct inode *inode, struct io_failure_record *rec, | |
1974 | int did_repair) | |
1975 | { | |
1976 | int ret; | |
1977 | int err = 0; | |
1978 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
1979 | ||
1980 | set_state_private(failure_tree, rec->start, 0); | |
1981 | ret = clear_extent_bits(failure_tree, rec->start, | |
1982 | rec->start + rec->len - 1, | |
1983 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1984 | if (ret) | |
1985 | err = ret; | |
1986 | ||
53b381b3 DW |
1987 | ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start, |
1988 | rec->start + rec->len - 1, | |
1989 | EXTENT_DAMAGED, GFP_NOFS); | |
1990 | if (ret && !err) | |
1991 | err = ret; | |
4a54c8c1 JS |
1992 | |
1993 | kfree(rec); | |
1994 | return err; | |
1995 | } | |
1996 | ||
4a54c8c1 JS |
1997 | /* |
1998 | * this bypasses the standard btrfs submit functions deliberately, as | |
1999 | * the standard behavior is to write all copies in a raid setup. here we only | |
2000 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
2001 | * submit_bio directly. | |
3ec706c8 | 2002 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
2003 | * actually prevents the read that triggered the error from finishing. |
2004 | * currently, there can be no more than two copies of every data bit. thus, | |
2005 | * exactly one rewrite is required. | |
2006 | */ | |
3ec706c8 | 2007 | int repair_io_failure(struct btrfs_fs_info *fs_info, u64 start, |
4a54c8c1 JS |
2008 | u64 length, u64 logical, struct page *page, |
2009 | int mirror_num) | |
2010 | { | |
2011 | struct bio *bio; | |
2012 | struct btrfs_device *dev; | |
4a54c8c1 JS |
2013 | u64 map_length = 0; |
2014 | u64 sector; | |
2015 | struct btrfs_bio *bbio = NULL; | |
53b381b3 | 2016 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
4a54c8c1 JS |
2017 | int ret; |
2018 | ||
908960c6 | 2019 | ASSERT(!(fs_info->sb->s_flags & MS_RDONLY)); |
4a54c8c1 JS |
2020 | BUG_ON(!mirror_num); |
2021 | ||
53b381b3 DW |
2022 | /* we can't repair anything in raid56 yet */ |
2023 | if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num)) | |
2024 | return 0; | |
2025 | ||
9be3395b | 2026 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
4a54c8c1 JS |
2027 | if (!bio) |
2028 | return -EIO; | |
4a54c8c1 JS |
2029 | bio->bi_size = 0; |
2030 | map_length = length; | |
2031 | ||
3ec706c8 | 2032 | ret = btrfs_map_block(fs_info, WRITE, logical, |
4a54c8c1 JS |
2033 | &map_length, &bbio, mirror_num); |
2034 | if (ret) { | |
2035 | bio_put(bio); | |
2036 | return -EIO; | |
2037 | } | |
2038 | BUG_ON(mirror_num != bbio->mirror_num); | |
2039 | sector = bbio->stripes[mirror_num-1].physical >> 9; | |
2040 | bio->bi_sector = sector; | |
2041 | dev = bbio->stripes[mirror_num-1].dev; | |
2042 | kfree(bbio); | |
2043 | if (!dev || !dev->bdev || !dev->writeable) { | |
2044 | bio_put(bio); | |
2045 | return -EIO; | |
2046 | } | |
2047 | bio->bi_bdev = dev->bdev; | |
4eee4fa4 | 2048 | bio_add_page(bio, page, length, start - page_offset(page)); |
4a54c8c1 | 2049 | |
c170bbb4 | 2050 | if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) { |
4a54c8c1 JS |
2051 | /* try to remap that extent elsewhere? */ |
2052 | bio_put(bio); | |
442a4f63 | 2053 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
2054 | return -EIO; |
2055 | } | |
2056 | ||
d5b025d5 | 2057 | printk_ratelimited_in_rcu(KERN_INFO "btrfs read error corrected: ino %lu off %llu " |
606686ee JB |
2058 | "(dev %s sector %llu)\n", page->mapping->host->i_ino, |
2059 | start, rcu_str_deref(dev->name), sector); | |
4a54c8c1 JS |
2060 | |
2061 | bio_put(bio); | |
2062 | return 0; | |
2063 | } | |
2064 | ||
ea466794 JB |
2065 | int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb, |
2066 | int mirror_num) | |
2067 | { | |
ea466794 JB |
2068 | u64 start = eb->start; |
2069 | unsigned long i, num_pages = num_extent_pages(eb->start, eb->len); | |
d95603b2 | 2070 | int ret = 0; |
ea466794 | 2071 | |
908960c6 ID |
2072 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
2073 | return -EROFS; | |
2074 | ||
ea466794 JB |
2075 | for (i = 0; i < num_pages; i++) { |
2076 | struct page *p = extent_buffer_page(eb, i); | |
3ec706c8 | 2077 | ret = repair_io_failure(root->fs_info, start, PAGE_CACHE_SIZE, |
ea466794 JB |
2078 | start, p, mirror_num); |
2079 | if (ret) | |
2080 | break; | |
2081 | start += PAGE_CACHE_SIZE; | |
2082 | } | |
2083 | ||
2084 | return ret; | |
2085 | } | |
2086 | ||
4a54c8c1 JS |
2087 | /* |
2088 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2089 | * to see if we need to process or clean up an io_failure_record | |
2090 | */ | |
2091 | static int clean_io_failure(u64 start, struct page *page) | |
2092 | { | |
2093 | u64 private; | |
2094 | u64 private_failure; | |
2095 | struct io_failure_record *failrec; | |
908960c6 ID |
2096 | struct inode *inode = page->mapping->host; |
2097 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
4a54c8c1 JS |
2098 | struct extent_state *state; |
2099 | int num_copies; | |
2100 | int did_repair = 0; | |
2101 | int ret; | |
4a54c8c1 JS |
2102 | |
2103 | private = 0; | |
2104 | ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
2105 | (u64)-1, 1, EXTENT_DIRTY, 0); | |
2106 | if (!ret) | |
2107 | return 0; | |
2108 | ||
2109 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start, | |
2110 | &private_failure); | |
2111 | if (ret) | |
2112 | return 0; | |
2113 | ||
2114 | failrec = (struct io_failure_record *)(unsigned long) private_failure; | |
2115 | BUG_ON(!failrec->this_mirror); | |
2116 | ||
2117 | if (failrec->in_validation) { | |
2118 | /* there was no real error, just free the record */ | |
2119 | pr_debug("clean_io_failure: freeing dummy error at %llu\n", | |
2120 | failrec->start); | |
2121 | did_repair = 1; | |
2122 | goto out; | |
2123 | } | |
908960c6 ID |
2124 | if (fs_info->sb->s_flags & MS_RDONLY) |
2125 | goto out; | |
4a54c8c1 JS |
2126 | |
2127 | spin_lock(&BTRFS_I(inode)->io_tree.lock); | |
2128 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, | |
2129 | failrec->start, | |
2130 | EXTENT_LOCKED); | |
2131 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); | |
2132 | ||
883d0de4 MX |
2133 | if (state && state->start <= failrec->start && |
2134 | state->end >= failrec->start + failrec->len - 1) { | |
3ec706c8 SB |
2135 | num_copies = btrfs_num_copies(fs_info, failrec->logical, |
2136 | failrec->len); | |
4a54c8c1 | 2137 | if (num_copies > 1) { |
3ec706c8 | 2138 | ret = repair_io_failure(fs_info, start, failrec->len, |
4a54c8c1 JS |
2139 | failrec->logical, page, |
2140 | failrec->failed_mirror); | |
2141 | did_repair = !ret; | |
2142 | } | |
53b381b3 | 2143 | ret = 0; |
4a54c8c1 JS |
2144 | } |
2145 | ||
2146 | out: | |
2147 | if (!ret) | |
2148 | ret = free_io_failure(inode, failrec, did_repair); | |
2149 | ||
2150 | return ret; | |
2151 | } | |
2152 | ||
2153 | /* | |
2154 | * this is a generic handler for readpage errors (default | |
2155 | * readpage_io_failed_hook). if other copies exist, read those and write back | |
2156 | * good data to the failed position. does not investigate in remapping the | |
2157 | * failed extent elsewhere, hoping the device will be smart enough to do this as | |
2158 | * needed | |
2159 | */ | |
2160 | ||
facc8a22 MX |
2161 | static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset, |
2162 | struct page *page, u64 start, u64 end, | |
2163 | int failed_mirror) | |
4a54c8c1 JS |
2164 | { |
2165 | struct io_failure_record *failrec = NULL; | |
2166 | u64 private; | |
2167 | struct extent_map *em; | |
2168 | struct inode *inode = page->mapping->host; | |
2169 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
2170 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2171 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
2172 | struct bio *bio; | |
facc8a22 MX |
2173 | struct btrfs_io_bio *btrfs_failed_bio; |
2174 | struct btrfs_io_bio *btrfs_bio; | |
4a54c8c1 JS |
2175 | int num_copies; |
2176 | int ret; | |
2177 | int read_mode; | |
2178 | u64 logical; | |
2179 | ||
2180 | BUG_ON(failed_bio->bi_rw & REQ_WRITE); | |
2181 | ||
2182 | ret = get_state_private(failure_tree, start, &private); | |
2183 | if (ret) { | |
2184 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2185 | if (!failrec) | |
2186 | return -ENOMEM; | |
2187 | failrec->start = start; | |
2188 | failrec->len = end - start + 1; | |
2189 | failrec->this_mirror = 0; | |
2190 | failrec->bio_flags = 0; | |
2191 | failrec->in_validation = 0; | |
2192 | ||
2193 | read_lock(&em_tree->lock); | |
2194 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2195 | if (!em) { | |
2196 | read_unlock(&em_tree->lock); | |
2197 | kfree(failrec); | |
2198 | return -EIO; | |
2199 | } | |
2200 | ||
68ba990f | 2201 | if (em->start > start || em->start + em->len <= start) { |
4a54c8c1 JS |
2202 | free_extent_map(em); |
2203 | em = NULL; | |
2204 | } | |
2205 | read_unlock(&em_tree->lock); | |
2206 | ||
7a2d6a64 | 2207 | if (!em) { |
4a54c8c1 JS |
2208 | kfree(failrec); |
2209 | return -EIO; | |
2210 | } | |
2211 | logical = start - em->start; | |
2212 | logical = em->block_start + logical; | |
2213 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2214 | logical = em->block_start; | |
2215 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2216 | extent_set_compress_type(&failrec->bio_flags, | |
2217 | em->compress_type); | |
2218 | } | |
2219 | pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, " | |
2220 | "len=%llu\n", logical, start, failrec->len); | |
2221 | failrec->logical = logical; | |
2222 | free_extent_map(em); | |
2223 | ||
2224 | /* set the bits in the private failure tree */ | |
2225 | ret = set_extent_bits(failure_tree, start, end, | |
2226 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
2227 | if (ret >= 0) | |
2228 | ret = set_state_private(failure_tree, start, | |
2229 | (u64)(unsigned long)failrec); | |
2230 | /* set the bits in the inode's tree */ | |
2231 | if (ret >= 0) | |
2232 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED, | |
2233 | GFP_NOFS); | |
2234 | if (ret < 0) { | |
2235 | kfree(failrec); | |
2236 | return ret; | |
2237 | } | |
2238 | } else { | |
2239 | failrec = (struct io_failure_record *)(unsigned long)private; | |
2240 | pr_debug("bio_readpage_error: (found) logical=%llu, " | |
2241 | "start=%llu, len=%llu, validation=%d\n", | |
2242 | failrec->logical, failrec->start, failrec->len, | |
2243 | failrec->in_validation); | |
2244 | /* | |
2245 | * when data can be on disk more than twice, add to failrec here | |
2246 | * (e.g. with a list for failed_mirror) to make | |
2247 | * clean_io_failure() clean all those errors at once. | |
2248 | */ | |
2249 | } | |
5d964051 SB |
2250 | num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info, |
2251 | failrec->logical, failrec->len); | |
4a54c8c1 JS |
2252 | if (num_copies == 1) { |
2253 | /* | |
2254 | * we only have a single copy of the data, so don't bother with | |
2255 | * all the retry and error correction code that follows. no | |
2256 | * matter what the error is, it is very likely to persist. | |
2257 | */ | |
09a7f7a2 MX |
2258 | pr_debug("bio_readpage_error: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n", |
2259 | num_copies, failrec->this_mirror, failed_mirror); | |
4a54c8c1 JS |
2260 | free_io_failure(inode, failrec, 0); |
2261 | return -EIO; | |
2262 | } | |
2263 | ||
4a54c8c1 JS |
2264 | /* |
2265 | * there are two premises: | |
2266 | * a) deliver good data to the caller | |
2267 | * b) correct the bad sectors on disk | |
2268 | */ | |
2269 | if (failed_bio->bi_vcnt > 1) { | |
2270 | /* | |
2271 | * to fulfill b), we need to know the exact failing sectors, as | |
2272 | * we don't want to rewrite any more than the failed ones. thus, | |
2273 | * we need separate read requests for the failed bio | |
2274 | * | |
2275 | * if the following BUG_ON triggers, our validation request got | |
2276 | * merged. we need separate requests for our algorithm to work. | |
2277 | */ | |
2278 | BUG_ON(failrec->in_validation); | |
2279 | failrec->in_validation = 1; | |
2280 | failrec->this_mirror = failed_mirror; | |
2281 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; | |
2282 | } else { | |
2283 | /* | |
2284 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2285 | * of the failed sector and if we succeed, we have setup | |
2286 | * everything for repair_io_failure to do the rest for us. | |
2287 | */ | |
2288 | if (failrec->in_validation) { | |
2289 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2290 | failrec->in_validation = 0; | |
2291 | failrec->this_mirror = 0; | |
2292 | } | |
2293 | failrec->failed_mirror = failed_mirror; | |
2294 | failrec->this_mirror++; | |
2295 | if (failrec->this_mirror == failed_mirror) | |
2296 | failrec->this_mirror++; | |
2297 | read_mode = READ_SYNC; | |
2298 | } | |
2299 | ||
facc8a22 MX |
2300 | if (failrec->this_mirror > num_copies) { |
2301 | pr_debug("bio_readpage_error: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n", | |
4a54c8c1 JS |
2302 | num_copies, failrec->this_mirror, failed_mirror); |
2303 | free_io_failure(inode, failrec, 0); | |
2304 | return -EIO; | |
2305 | } | |
2306 | ||
9be3395b | 2307 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
e627ee7b TI |
2308 | if (!bio) { |
2309 | free_io_failure(inode, failrec, 0); | |
2310 | return -EIO; | |
2311 | } | |
4a54c8c1 JS |
2312 | bio->bi_end_io = failed_bio->bi_end_io; |
2313 | bio->bi_sector = failrec->logical >> 9; | |
2314 | bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; | |
2315 | bio->bi_size = 0; | |
2316 | ||
facc8a22 MX |
2317 | btrfs_failed_bio = btrfs_io_bio(failed_bio); |
2318 | if (btrfs_failed_bio->csum) { | |
2319 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2320 | u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); | |
2321 | ||
2322 | btrfs_bio = btrfs_io_bio(bio); | |
2323 | btrfs_bio->csum = btrfs_bio->csum_inline; | |
2324 | phy_offset >>= inode->i_sb->s_blocksize_bits; | |
2325 | phy_offset *= csum_size; | |
2326 | memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + phy_offset, | |
2327 | csum_size); | |
2328 | } | |
2329 | ||
4a54c8c1 JS |
2330 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); |
2331 | ||
2332 | pr_debug("bio_readpage_error: submitting new read[%#x] to " | |
2333 | "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode, | |
2334 | failrec->this_mirror, num_copies, failrec->in_validation); | |
2335 | ||
013bd4c3 TI |
2336 | ret = tree->ops->submit_bio_hook(inode, read_mode, bio, |
2337 | failrec->this_mirror, | |
2338 | failrec->bio_flags, 0); | |
2339 | return ret; | |
4a54c8c1 JS |
2340 | } |
2341 | ||
d1310b2e CM |
2342 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2343 | ||
87826df0 JM |
2344 | int end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
2345 | { | |
2346 | int uptodate = (err == 0); | |
2347 | struct extent_io_tree *tree; | |
2348 | int ret; | |
2349 | ||
2350 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
2351 | ||
2352 | if (tree->ops && tree->ops->writepage_end_io_hook) { | |
2353 | ret = tree->ops->writepage_end_io_hook(page, start, | |
2354 | end, NULL, uptodate); | |
2355 | if (ret) | |
2356 | uptodate = 0; | |
2357 | } | |
2358 | ||
87826df0 | 2359 | if (!uptodate) { |
87826df0 JM |
2360 | ClearPageUptodate(page); |
2361 | SetPageError(page); | |
2362 | } | |
2363 | return 0; | |
2364 | } | |
2365 | ||
d1310b2e CM |
2366 | /* |
2367 | * after a writepage IO is done, we need to: | |
2368 | * clear the uptodate bits on error | |
2369 | * clear the writeback bits in the extent tree for this IO | |
2370 | * end_page_writeback if the page has no more pending IO | |
2371 | * | |
2372 | * Scheduling is not allowed, so the extent state tree is expected | |
2373 | * to have one and only one object corresponding to this IO. | |
2374 | */ | |
d1310b2e | 2375 | static void end_bio_extent_writepage(struct bio *bio, int err) |
d1310b2e | 2376 | { |
d1310b2e | 2377 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
d1310b2e CM |
2378 | u64 start; |
2379 | u64 end; | |
d1310b2e | 2380 | |
d1310b2e CM |
2381 | do { |
2382 | struct page *page = bvec->bv_page; | |
902b22f3 | 2383 | |
17a5adcc AO |
2384 | /* We always issue full-page reads, but if some block |
2385 | * in a page fails to read, blk_update_request() will | |
2386 | * advance bv_offset and adjust bv_len to compensate. | |
2387 | * Print a warning for nonzero offsets, and an error | |
2388 | * if they don't add up to a full page. */ | |
2389 | if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) | |
2390 | printk("%s page write in btrfs with offset %u and length %u\n", | |
2391 | bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE | |
2392 | ? KERN_ERR "partial" : KERN_INFO "incomplete", | |
2393 | bvec->bv_offset, bvec->bv_len); | |
d1310b2e | 2394 | |
17a5adcc AO |
2395 | start = page_offset(page); |
2396 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
d1310b2e CM |
2397 | |
2398 | if (--bvec >= bio->bi_io_vec) | |
2399 | prefetchw(&bvec->bv_page->flags); | |
1259ab75 | 2400 | |
87826df0 JM |
2401 | if (end_extent_writepage(page, err, start, end)) |
2402 | continue; | |
70dec807 | 2403 | |
17a5adcc | 2404 | end_page_writeback(page); |
d1310b2e | 2405 | } while (bvec >= bio->bi_io_vec); |
2b1f55b0 | 2406 | |
d1310b2e | 2407 | bio_put(bio); |
d1310b2e CM |
2408 | } |
2409 | ||
883d0de4 MX |
2410 | static void |
2411 | endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len, | |
2412 | int uptodate) | |
2413 | { | |
2414 | struct extent_state *cached = NULL; | |
2415 | u64 end = start + len - 1; | |
2416 | ||
2417 | if (uptodate && tree->track_uptodate) | |
2418 | set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC); | |
2419 | unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC); | |
2420 | } | |
2421 | ||
d1310b2e CM |
2422 | /* |
2423 | * after a readpage IO is done, we need to: | |
2424 | * clear the uptodate bits on error | |
2425 | * set the uptodate bits if things worked | |
2426 | * set the page up to date if all extents in the tree are uptodate | |
2427 | * clear the lock bit in the extent tree | |
2428 | * unlock the page if there are no other extents locked for it | |
2429 | * | |
2430 | * Scheduling is not allowed, so the extent state tree is expected | |
2431 | * to have one and only one object corresponding to this IO. | |
2432 | */ | |
d1310b2e | 2433 | static void end_bio_extent_readpage(struct bio *bio, int err) |
d1310b2e CM |
2434 | { |
2435 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
4125bf76 CM |
2436 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
2437 | struct bio_vec *bvec = bio->bi_io_vec; | |
facc8a22 | 2438 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); |
902b22f3 | 2439 | struct extent_io_tree *tree; |
facc8a22 | 2440 | u64 offset = 0; |
d1310b2e CM |
2441 | u64 start; |
2442 | u64 end; | |
facc8a22 | 2443 | u64 len; |
883d0de4 MX |
2444 | u64 extent_start = 0; |
2445 | u64 extent_len = 0; | |
5cf1ab56 | 2446 | int mirror; |
d1310b2e CM |
2447 | int ret; |
2448 | ||
d20f7043 CM |
2449 | if (err) |
2450 | uptodate = 0; | |
2451 | ||
d1310b2e CM |
2452 | do { |
2453 | struct page *page = bvec->bv_page; | |
a71754fc | 2454 | struct inode *inode = page->mapping->host; |
507903b8 | 2455 | |
be3940c0 | 2456 | pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, " |
9be3395b CM |
2457 | "mirror=%lu\n", (u64)bio->bi_sector, err, |
2458 | io_bio->mirror_num); | |
a71754fc | 2459 | tree = &BTRFS_I(inode)->io_tree; |
902b22f3 | 2460 | |
17a5adcc AO |
2461 | /* We always issue full-page reads, but if some block |
2462 | * in a page fails to read, blk_update_request() will | |
2463 | * advance bv_offset and adjust bv_len to compensate. | |
2464 | * Print a warning for nonzero offsets, and an error | |
2465 | * if they don't add up to a full page. */ | |
2466 | if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) | |
2467 | printk("%s page read in btrfs with offset %u and length %u\n", | |
2468 | bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE | |
2469 | ? KERN_ERR "partial" : KERN_INFO "incomplete", | |
2470 | bvec->bv_offset, bvec->bv_len); | |
d1310b2e | 2471 | |
17a5adcc AO |
2472 | start = page_offset(page); |
2473 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
facc8a22 | 2474 | len = bvec->bv_len; |
d1310b2e | 2475 | |
4125bf76 | 2476 | if (++bvec <= bvec_end) |
d1310b2e CM |
2477 | prefetchw(&bvec->bv_page->flags); |
2478 | ||
9be3395b | 2479 | mirror = io_bio->mirror_num; |
f2a09da9 MX |
2480 | if (likely(uptodate && tree->ops && |
2481 | tree->ops->readpage_end_io_hook)) { | |
facc8a22 MX |
2482 | ret = tree->ops->readpage_end_io_hook(io_bio, offset, |
2483 | page, start, end, | |
2484 | mirror); | |
5ee0844d | 2485 | if (ret) |
d1310b2e | 2486 | uptodate = 0; |
5ee0844d | 2487 | else |
4a54c8c1 | 2488 | clean_io_failure(start, page); |
d1310b2e | 2489 | } |
ea466794 | 2490 | |
f2a09da9 MX |
2491 | if (likely(uptodate)) |
2492 | goto readpage_ok; | |
2493 | ||
2494 | if (tree->ops && tree->ops->readpage_io_failed_hook) { | |
5cf1ab56 | 2495 | ret = tree->ops->readpage_io_failed_hook(page, mirror); |
ea466794 JB |
2496 | if (!ret && !err && |
2497 | test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
2498 | uptodate = 1; | |
f2a09da9 | 2499 | } else { |
f4a8e656 JS |
2500 | /* |
2501 | * The generic bio_readpage_error handles errors the | |
2502 | * following way: If possible, new read requests are | |
2503 | * created and submitted and will end up in | |
2504 | * end_bio_extent_readpage as well (if we're lucky, not | |
2505 | * in the !uptodate case). In that case it returns 0 and | |
2506 | * we just go on with the next page in our bio. If it | |
2507 | * can't handle the error it will return -EIO and we | |
2508 | * remain responsible for that page. | |
2509 | */ | |
facc8a22 MX |
2510 | ret = bio_readpage_error(bio, offset, page, start, end, |
2511 | mirror); | |
7e38326f | 2512 | if (ret == 0) { |
3b951516 CM |
2513 | uptodate = |
2514 | test_bit(BIO_UPTODATE, &bio->bi_flags); | |
d20f7043 CM |
2515 | if (err) |
2516 | uptodate = 0; | |
7e38326f CM |
2517 | continue; |
2518 | } | |
2519 | } | |
f2a09da9 | 2520 | readpage_ok: |
883d0de4 | 2521 | if (likely(uptodate)) { |
a71754fc JB |
2522 | loff_t i_size = i_size_read(inode); |
2523 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
2524 | unsigned offset; | |
2525 | ||
2526 | /* Zero out the end if this page straddles i_size */ | |
2527 | offset = i_size & (PAGE_CACHE_SIZE-1); | |
2528 | if (page->index == end_index && offset) | |
2529 | zero_user_segment(page, offset, PAGE_CACHE_SIZE); | |
17a5adcc | 2530 | SetPageUptodate(page); |
70dec807 | 2531 | } else { |
17a5adcc AO |
2532 | ClearPageUptodate(page); |
2533 | SetPageError(page); | |
70dec807 | 2534 | } |
17a5adcc | 2535 | unlock_page(page); |
facc8a22 | 2536 | offset += len; |
883d0de4 MX |
2537 | |
2538 | if (unlikely(!uptodate)) { | |
2539 | if (extent_len) { | |
2540 | endio_readpage_release_extent(tree, | |
2541 | extent_start, | |
2542 | extent_len, 1); | |
2543 | extent_start = 0; | |
2544 | extent_len = 0; | |
2545 | } | |
2546 | endio_readpage_release_extent(tree, start, | |
2547 | end - start + 1, 0); | |
2548 | } else if (!extent_len) { | |
2549 | extent_start = start; | |
2550 | extent_len = end + 1 - start; | |
2551 | } else if (extent_start + extent_len == start) { | |
2552 | extent_len += end + 1 - start; | |
2553 | } else { | |
2554 | endio_readpage_release_extent(tree, extent_start, | |
2555 | extent_len, uptodate); | |
2556 | extent_start = start; | |
2557 | extent_len = end + 1 - start; | |
2558 | } | |
4125bf76 | 2559 | } while (bvec <= bvec_end); |
d1310b2e | 2560 | |
883d0de4 MX |
2561 | if (extent_len) |
2562 | endio_readpage_release_extent(tree, extent_start, extent_len, | |
2563 | uptodate); | |
facc8a22 MX |
2564 | if (io_bio->end_io) |
2565 | io_bio->end_io(io_bio, err); | |
d1310b2e | 2566 | bio_put(bio); |
d1310b2e CM |
2567 | } |
2568 | ||
9be3395b CM |
2569 | /* |
2570 | * this allocates from the btrfs_bioset. We're returning a bio right now | |
2571 | * but you can call btrfs_io_bio for the appropriate container_of magic | |
2572 | */ | |
88f794ed MX |
2573 | struct bio * |
2574 | btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
2575 | gfp_t gfp_flags) | |
d1310b2e | 2576 | { |
facc8a22 | 2577 | struct btrfs_io_bio *btrfs_bio; |
d1310b2e CM |
2578 | struct bio *bio; |
2579 | ||
9be3395b | 2580 | bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset); |
d1310b2e CM |
2581 | |
2582 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
9be3395b CM |
2583 | while (!bio && (nr_vecs /= 2)) { |
2584 | bio = bio_alloc_bioset(gfp_flags, | |
2585 | nr_vecs, btrfs_bioset); | |
2586 | } | |
d1310b2e CM |
2587 | } |
2588 | ||
2589 | if (bio) { | |
e1c4b745 | 2590 | bio->bi_size = 0; |
d1310b2e CM |
2591 | bio->bi_bdev = bdev; |
2592 | bio->bi_sector = first_sector; | |
facc8a22 MX |
2593 | btrfs_bio = btrfs_io_bio(bio); |
2594 | btrfs_bio->csum = NULL; | |
2595 | btrfs_bio->csum_allocated = NULL; | |
2596 | btrfs_bio->end_io = NULL; | |
d1310b2e CM |
2597 | } |
2598 | return bio; | |
2599 | } | |
2600 | ||
9be3395b CM |
2601 | struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask) |
2602 | { | |
2603 | return bio_clone_bioset(bio, gfp_mask, btrfs_bioset); | |
2604 | } | |
2605 | ||
2606 | ||
2607 | /* this also allocates from the btrfs_bioset */ | |
2608 | struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs) | |
2609 | { | |
facc8a22 MX |
2610 | struct btrfs_io_bio *btrfs_bio; |
2611 | struct bio *bio; | |
2612 | ||
2613 | bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset); | |
2614 | if (bio) { | |
2615 | btrfs_bio = btrfs_io_bio(bio); | |
2616 | btrfs_bio->csum = NULL; | |
2617 | btrfs_bio->csum_allocated = NULL; | |
2618 | btrfs_bio->end_io = NULL; | |
2619 | } | |
2620 | return bio; | |
9be3395b CM |
2621 | } |
2622 | ||
2623 | ||
355808c2 JM |
2624 | static int __must_check submit_one_bio(int rw, struct bio *bio, |
2625 | int mirror_num, unsigned long bio_flags) | |
d1310b2e | 2626 | { |
d1310b2e | 2627 | int ret = 0; |
70dec807 CM |
2628 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
2629 | struct page *page = bvec->bv_page; | |
2630 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 2631 | u64 start; |
70dec807 | 2632 | |
4eee4fa4 | 2633 | start = page_offset(page) + bvec->bv_offset; |
70dec807 | 2634 | |
902b22f3 | 2635 | bio->bi_private = NULL; |
d1310b2e CM |
2636 | |
2637 | bio_get(bio); | |
2638 | ||
065631f6 | 2639 | if (tree->ops && tree->ops->submit_bio_hook) |
6b82ce8d | 2640 | ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio, |
eaf25d93 | 2641 | mirror_num, bio_flags, start); |
0b86a832 | 2642 | else |
21adbd5c | 2643 | btrfsic_submit_bio(rw, bio); |
4a54c8c1 | 2644 | |
d1310b2e CM |
2645 | if (bio_flagged(bio, BIO_EOPNOTSUPP)) |
2646 | ret = -EOPNOTSUPP; | |
2647 | bio_put(bio); | |
2648 | return ret; | |
2649 | } | |
2650 | ||
64a16701 | 2651 | static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page, |
3444a972 JM |
2652 | unsigned long offset, size_t size, struct bio *bio, |
2653 | unsigned long bio_flags) | |
2654 | { | |
2655 | int ret = 0; | |
2656 | if (tree->ops && tree->ops->merge_bio_hook) | |
64a16701 | 2657 | ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio, |
3444a972 JM |
2658 | bio_flags); |
2659 | BUG_ON(ret < 0); | |
2660 | return ret; | |
2661 | ||
2662 | } | |
2663 | ||
d1310b2e CM |
2664 | static int submit_extent_page(int rw, struct extent_io_tree *tree, |
2665 | struct page *page, sector_t sector, | |
2666 | size_t size, unsigned long offset, | |
2667 | struct block_device *bdev, | |
2668 | struct bio **bio_ret, | |
2669 | unsigned long max_pages, | |
f188591e | 2670 | bio_end_io_t end_io_func, |
c8b97818 CM |
2671 | int mirror_num, |
2672 | unsigned long prev_bio_flags, | |
2673 | unsigned long bio_flags) | |
d1310b2e CM |
2674 | { |
2675 | int ret = 0; | |
2676 | struct bio *bio; | |
2677 | int nr; | |
c8b97818 CM |
2678 | int contig = 0; |
2679 | int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED; | |
2680 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; | |
5b050f04 | 2681 | size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE); |
d1310b2e CM |
2682 | |
2683 | if (bio_ret && *bio_ret) { | |
2684 | bio = *bio_ret; | |
c8b97818 CM |
2685 | if (old_compressed) |
2686 | contig = bio->bi_sector == sector; | |
2687 | else | |
f73a1c7d | 2688 | contig = bio_end_sector(bio) == sector; |
c8b97818 CM |
2689 | |
2690 | if (prev_bio_flags != bio_flags || !contig || | |
64a16701 | 2691 | merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) || |
c8b97818 CM |
2692 | bio_add_page(bio, page, page_size, offset) < page_size) { |
2693 | ret = submit_one_bio(rw, bio, mirror_num, | |
2694 | prev_bio_flags); | |
79787eaa JM |
2695 | if (ret < 0) |
2696 | return ret; | |
d1310b2e CM |
2697 | bio = NULL; |
2698 | } else { | |
2699 | return 0; | |
2700 | } | |
2701 | } | |
c8b97818 CM |
2702 | if (this_compressed) |
2703 | nr = BIO_MAX_PAGES; | |
2704 | else | |
2705 | nr = bio_get_nr_vecs(bdev); | |
2706 | ||
88f794ed | 2707 | bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH); |
5df67083 TI |
2708 | if (!bio) |
2709 | return -ENOMEM; | |
70dec807 | 2710 | |
c8b97818 | 2711 | bio_add_page(bio, page, page_size, offset); |
d1310b2e CM |
2712 | bio->bi_end_io = end_io_func; |
2713 | bio->bi_private = tree; | |
70dec807 | 2714 | |
d397712b | 2715 | if (bio_ret) |
d1310b2e | 2716 | *bio_ret = bio; |
d397712b | 2717 | else |
c8b97818 | 2718 | ret = submit_one_bio(rw, bio, mirror_num, bio_flags); |
d1310b2e CM |
2719 | |
2720 | return ret; | |
2721 | } | |
2722 | ||
48a3b636 ES |
2723 | static void attach_extent_buffer_page(struct extent_buffer *eb, |
2724 | struct page *page) | |
d1310b2e CM |
2725 | { |
2726 | if (!PagePrivate(page)) { | |
2727 | SetPagePrivate(page); | |
d1310b2e | 2728 | page_cache_get(page); |
4f2de97a JB |
2729 | set_page_private(page, (unsigned long)eb); |
2730 | } else { | |
2731 | WARN_ON(page->private != (unsigned long)eb); | |
d1310b2e CM |
2732 | } |
2733 | } | |
2734 | ||
4f2de97a | 2735 | void set_page_extent_mapped(struct page *page) |
d1310b2e | 2736 | { |
4f2de97a JB |
2737 | if (!PagePrivate(page)) { |
2738 | SetPagePrivate(page); | |
2739 | page_cache_get(page); | |
2740 | set_page_private(page, EXTENT_PAGE_PRIVATE); | |
2741 | } | |
d1310b2e CM |
2742 | } |
2743 | ||
125bac01 MX |
2744 | static struct extent_map * |
2745 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
2746 | u64 start, u64 len, get_extent_t *get_extent, | |
2747 | struct extent_map **em_cached) | |
2748 | { | |
2749 | struct extent_map *em; | |
2750 | ||
2751 | if (em_cached && *em_cached) { | |
2752 | em = *em_cached; | |
2753 | if (em->in_tree && start >= em->start && | |
2754 | start < extent_map_end(em)) { | |
2755 | atomic_inc(&em->refs); | |
2756 | return em; | |
2757 | } | |
2758 | ||
2759 | free_extent_map(em); | |
2760 | *em_cached = NULL; | |
2761 | } | |
2762 | ||
2763 | em = get_extent(inode, page, pg_offset, start, len, 0); | |
2764 | if (em_cached && !IS_ERR_OR_NULL(em)) { | |
2765 | BUG_ON(*em_cached); | |
2766 | atomic_inc(&em->refs); | |
2767 | *em_cached = em; | |
2768 | } | |
2769 | return em; | |
2770 | } | |
d1310b2e CM |
2771 | /* |
2772 | * basic readpage implementation. Locked extent state structs are inserted | |
2773 | * into the tree that are removed when the IO is done (by the end_io | |
2774 | * handlers) | |
79787eaa | 2775 | * XXX JDM: This needs looking at to ensure proper page locking |
d1310b2e | 2776 | */ |
9974090b MX |
2777 | static int __do_readpage(struct extent_io_tree *tree, |
2778 | struct page *page, | |
2779 | get_extent_t *get_extent, | |
125bac01 | 2780 | struct extent_map **em_cached, |
9974090b MX |
2781 | struct bio **bio, int mirror_num, |
2782 | unsigned long *bio_flags, int rw) | |
d1310b2e CM |
2783 | { |
2784 | struct inode *inode = page->mapping->host; | |
4eee4fa4 | 2785 | u64 start = page_offset(page); |
d1310b2e CM |
2786 | u64 page_end = start + PAGE_CACHE_SIZE - 1; |
2787 | u64 end; | |
2788 | u64 cur = start; | |
2789 | u64 extent_offset; | |
2790 | u64 last_byte = i_size_read(inode); | |
2791 | u64 block_start; | |
2792 | u64 cur_end; | |
2793 | sector_t sector; | |
2794 | struct extent_map *em; | |
2795 | struct block_device *bdev; | |
2796 | int ret; | |
2797 | int nr = 0; | |
4b384318 | 2798 | int parent_locked = *bio_flags & EXTENT_BIO_PARENT_LOCKED; |
306e16ce | 2799 | size_t pg_offset = 0; |
d1310b2e | 2800 | size_t iosize; |
c8b97818 | 2801 | size_t disk_io_size; |
d1310b2e | 2802 | size_t blocksize = inode->i_sb->s_blocksize; |
4b384318 | 2803 | unsigned long this_bio_flag = *bio_flags & EXTENT_BIO_PARENT_LOCKED; |
d1310b2e CM |
2804 | |
2805 | set_page_extent_mapped(page); | |
2806 | ||
9974090b | 2807 | end = page_end; |
90a887c9 DM |
2808 | if (!PageUptodate(page)) { |
2809 | if (cleancache_get_page(page) == 0) { | |
2810 | BUG_ON(blocksize != PAGE_SIZE); | |
9974090b | 2811 | unlock_extent(tree, start, end); |
90a887c9 DM |
2812 | goto out; |
2813 | } | |
2814 | } | |
2815 | ||
c8b97818 CM |
2816 | if (page->index == last_byte >> PAGE_CACHE_SHIFT) { |
2817 | char *userpage; | |
2818 | size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1); | |
2819 | ||
2820 | if (zero_offset) { | |
2821 | iosize = PAGE_CACHE_SIZE - zero_offset; | |
7ac687d9 | 2822 | userpage = kmap_atomic(page); |
c8b97818 CM |
2823 | memset(userpage + zero_offset, 0, iosize); |
2824 | flush_dcache_page(page); | |
7ac687d9 | 2825 | kunmap_atomic(userpage); |
c8b97818 CM |
2826 | } |
2827 | } | |
d1310b2e | 2828 | while (cur <= end) { |
c8f2f24b JB |
2829 | unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1; |
2830 | ||
d1310b2e CM |
2831 | if (cur >= last_byte) { |
2832 | char *userpage; | |
507903b8 AJ |
2833 | struct extent_state *cached = NULL; |
2834 | ||
306e16ce | 2835 | iosize = PAGE_CACHE_SIZE - pg_offset; |
7ac687d9 | 2836 | userpage = kmap_atomic(page); |
306e16ce | 2837 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2838 | flush_dcache_page(page); |
7ac687d9 | 2839 | kunmap_atomic(userpage); |
d1310b2e | 2840 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 2841 | &cached, GFP_NOFS); |
4b384318 MF |
2842 | if (!parent_locked) |
2843 | unlock_extent_cached(tree, cur, | |
2844 | cur + iosize - 1, | |
2845 | &cached, GFP_NOFS); | |
d1310b2e CM |
2846 | break; |
2847 | } | |
125bac01 MX |
2848 | em = __get_extent_map(inode, page, pg_offset, cur, |
2849 | end - cur + 1, get_extent, em_cached); | |
c704005d | 2850 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 2851 | SetPageError(page); |
4b384318 MF |
2852 | if (!parent_locked) |
2853 | unlock_extent(tree, cur, end); | |
d1310b2e CM |
2854 | break; |
2855 | } | |
d1310b2e CM |
2856 | extent_offset = cur - em->start; |
2857 | BUG_ON(extent_map_end(em) <= cur); | |
2858 | BUG_ON(end < cur); | |
2859 | ||
261507a0 | 2860 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
4b384318 | 2861 | this_bio_flag |= EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
2862 | extent_set_compress_type(&this_bio_flag, |
2863 | em->compress_type); | |
2864 | } | |
c8b97818 | 2865 | |
d1310b2e CM |
2866 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
2867 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 2868 | iosize = ALIGN(iosize, blocksize); |
c8b97818 CM |
2869 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
2870 | disk_io_size = em->block_len; | |
2871 | sector = em->block_start >> 9; | |
2872 | } else { | |
2873 | sector = (em->block_start + extent_offset) >> 9; | |
2874 | disk_io_size = iosize; | |
2875 | } | |
d1310b2e CM |
2876 | bdev = em->bdev; |
2877 | block_start = em->block_start; | |
d899e052 YZ |
2878 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2879 | block_start = EXTENT_MAP_HOLE; | |
d1310b2e CM |
2880 | free_extent_map(em); |
2881 | em = NULL; | |
2882 | ||
2883 | /* we've found a hole, just zero and go on */ | |
2884 | if (block_start == EXTENT_MAP_HOLE) { | |
2885 | char *userpage; | |
507903b8 AJ |
2886 | struct extent_state *cached = NULL; |
2887 | ||
7ac687d9 | 2888 | userpage = kmap_atomic(page); |
306e16ce | 2889 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2890 | flush_dcache_page(page); |
7ac687d9 | 2891 | kunmap_atomic(userpage); |
d1310b2e CM |
2892 | |
2893 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 AJ |
2894 | &cached, GFP_NOFS); |
2895 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
2896 | &cached, GFP_NOFS); | |
d1310b2e | 2897 | cur = cur + iosize; |
306e16ce | 2898 | pg_offset += iosize; |
d1310b2e CM |
2899 | continue; |
2900 | } | |
2901 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
2902 | if (test_range_bit(tree, cur, cur_end, |
2903 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 2904 | check_page_uptodate(tree, page); |
4b384318 MF |
2905 | if (!parent_locked) |
2906 | unlock_extent(tree, cur, cur + iosize - 1); | |
d1310b2e | 2907 | cur = cur + iosize; |
306e16ce | 2908 | pg_offset += iosize; |
d1310b2e CM |
2909 | continue; |
2910 | } | |
70dec807 CM |
2911 | /* we have an inline extent but it didn't get marked up |
2912 | * to date. Error out | |
2913 | */ | |
2914 | if (block_start == EXTENT_MAP_INLINE) { | |
2915 | SetPageError(page); | |
4b384318 MF |
2916 | if (!parent_locked) |
2917 | unlock_extent(tree, cur, cur + iosize - 1); | |
70dec807 | 2918 | cur = cur + iosize; |
306e16ce | 2919 | pg_offset += iosize; |
70dec807 CM |
2920 | continue; |
2921 | } | |
d1310b2e | 2922 | |
c8f2f24b | 2923 | pnr -= page->index; |
d4c7ca86 | 2924 | ret = submit_extent_page(rw, tree, page, |
306e16ce | 2925 | sector, disk_io_size, pg_offset, |
89642229 | 2926 | bdev, bio, pnr, |
c8b97818 CM |
2927 | end_bio_extent_readpage, mirror_num, |
2928 | *bio_flags, | |
2929 | this_bio_flag); | |
c8f2f24b JB |
2930 | if (!ret) { |
2931 | nr++; | |
2932 | *bio_flags = this_bio_flag; | |
2933 | } else { | |
d1310b2e | 2934 | SetPageError(page); |
4b384318 MF |
2935 | if (!parent_locked) |
2936 | unlock_extent(tree, cur, cur + iosize - 1); | |
edd33c99 | 2937 | } |
d1310b2e | 2938 | cur = cur + iosize; |
306e16ce | 2939 | pg_offset += iosize; |
d1310b2e | 2940 | } |
90a887c9 | 2941 | out: |
d1310b2e CM |
2942 | if (!nr) { |
2943 | if (!PageError(page)) | |
2944 | SetPageUptodate(page); | |
2945 | unlock_page(page); | |
2946 | } | |
2947 | return 0; | |
2948 | } | |
2949 | ||
9974090b MX |
2950 | static inline void __do_contiguous_readpages(struct extent_io_tree *tree, |
2951 | struct page *pages[], int nr_pages, | |
2952 | u64 start, u64 end, | |
2953 | get_extent_t *get_extent, | |
125bac01 | 2954 | struct extent_map **em_cached, |
9974090b MX |
2955 | struct bio **bio, int mirror_num, |
2956 | unsigned long *bio_flags, int rw) | |
2957 | { | |
2958 | struct inode *inode; | |
2959 | struct btrfs_ordered_extent *ordered; | |
2960 | int index; | |
2961 | ||
2962 | inode = pages[0]->mapping->host; | |
2963 | while (1) { | |
2964 | lock_extent(tree, start, end); | |
2965 | ordered = btrfs_lookup_ordered_range(inode, start, | |
2966 | end - start + 1); | |
2967 | if (!ordered) | |
2968 | break; | |
2969 | unlock_extent(tree, start, end); | |
2970 | btrfs_start_ordered_extent(inode, ordered, 1); | |
2971 | btrfs_put_ordered_extent(ordered); | |
2972 | } | |
2973 | ||
2974 | for (index = 0; index < nr_pages; index++) { | |
125bac01 MX |
2975 | __do_readpage(tree, pages[index], get_extent, em_cached, bio, |
2976 | mirror_num, bio_flags, rw); | |
9974090b MX |
2977 | page_cache_release(pages[index]); |
2978 | } | |
2979 | } | |
2980 | ||
2981 | static void __extent_readpages(struct extent_io_tree *tree, | |
2982 | struct page *pages[], | |
2983 | int nr_pages, get_extent_t *get_extent, | |
125bac01 | 2984 | struct extent_map **em_cached, |
9974090b MX |
2985 | struct bio **bio, int mirror_num, |
2986 | unsigned long *bio_flags, int rw) | |
2987 | { | |
35a3621b | 2988 | u64 start = 0; |
9974090b MX |
2989 | u64 end = 0; |
2990 | u64 page_start; | |
2991 | int index; | |
35a3621b | 2992 | int first_index = 0; |
9974090b MX |
2993 | |
2994 | for (index = 0; index < nr_pages; index++) { | |
2995 | page_start = page_offset(pages[index]); | |
2996 | if (!end) { | |
2997 | start = page_start; | |
2998 | end = start + PAGE_CACHE_SIZE - 1; | |
2999 | first_index = index; | |
3000 | } else if (end + 1 == page_start) { | |
3001 | end += PAGE_CACHE_SIZE; | |
3002 | } else { | |
3003 | __do_contiguous_readpages(tree, &pages[first_index], | |
3004 | index - first_index, start, | |
125bac01 MX |
3005 | end, get_extent, em_cached, |
3006 | bio, mirror_num, bio_flags, | |
3007 | rw); | |
9974090b MX |
3008 | start = page_start; |
3009 | end = start + PAGE_CACHE_SIZE - 1; | |
3010 | first_index = index; | |
3011 | } | |
3012 | } | |
3013 | ||
3014 | if (end) | |
3015 | __do_contiguous_readpages(tree, &pages[first_index], | |
3016 | index - first_index, start, | |
125bac01 | 3017 | end, get_extent, em_cached, bio, |
9974090b MX |
3018 | mirror_num, bio_flags, rw); |
3019 | } | |
3020 | ||
3021 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
3022 | struct page *page, | |
3023 | get_extent_t *get_extent, | |
3024 | struct bio **bio, int mirror_num, | |
3025 | unsigned long *bio_flags, int rw) | |
3026 | { | |
3027 | struct inode *inode = page->mapping->host; | |
3028 | struct btrfs_ordered_extent *ordered; | |
3029 | u64 start = page_offset(page); | |
3030 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
3031 | int ret; | |
3032 | ||
3033 | while (1) { | |
3034 | lock_extent(tree, start, end); | |
3035 | ordered = btrfs_lookup_ordered_extent(inode, start); | |
3036 | if (!ordered) | |
3037 | break; | |
3038 | unlock_extent(tree, start, end); | |
3039 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3040 | btrfs_put_ordered_extent(ordered); | |
3041 | } | |
3042 | ||
125bac01 MX |
3043 | ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num, |
3044 | bio_flags, rw); | |
9974090b MX |
3045 | return ret; |
3046 | } | |
3047 | ||
d1310b2e | 3048 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, |
8ddc7d9c | 3049 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3050 | { |
3051 | struct bio *bio = NULL; | |
c8b97818 | 3052 | unsigned long bio_flags = 0; |
d1310b2e CM |
3053 | int ret; |
3054 | ||
8ddc7d9c | 3055 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, |
d4c7ca86 | 3056 | &bio_flags, READ); |
d1310b2e | 3057 | if (bio) |
8ddc7d9c | 3058 | ret = submit_one_bio(READ, bio, mirror_num, bio_flags); |
d1310b2e CM |
3059 | return ret; |
3060 | } | |
d1310b2e | 3061 | |
4b384318 MF |
3062 | int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page, |
3063 | get_extent_t *get_extent, int mirror_num) | |
3064 | { | |
3065 | struct bio *bio = NULL; | |
3066 | unsigned long bio_flags = EXTENT_BIO_PARENT_LOCKED; | |
3067 | int ret; | |
3068 | ||
3069 | ret = __do_readpage(tree, page, get_extent, NULL, &bio, mirror_num, | |
3070 | &bio_flags, READ); | |
3071 | if (bio) | |
3072 | ret = submit_one_bio(READ, bio, mirror_num, bio_flags); | |
3073 | return ret; | |
3074 | } | |
3075 | ||
11c8349b CM |
3076 | static noinline void update_nr_written(struct page *page, |
3077 | struct writeback_control *wbc, | |
3078 | unsigned long nr_written) | |
3079 | { | |
3080 | wbc->nr_to_write -= nr_written; | |
3081 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && | |
3082 | wbc->range_start == 0 && wbc->range_end == LLONG_MAX)) | |
3083 | page->mapping->writeback_index = page->index + nr_written; | |
3084 | } | |
3085 | ||
d1310b2e CM |
3086 | /* |
3087 | * the writepage semantics are similar to regular writepage. extent | |
3088 | * records are inserted to lock ranges in the tree, and as dirty areas | |
3089 | * are found, they are marked writeback. Then the lock bits are removed | |
3090 | * and the end_io handler clears the writeback ranges | |
3091 | */ | |
3092 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
3093 | void *data) | |
3094 | { | |
3095 | struct inode *inode = page->mapping->host; | |
3096 | struct extent_page_data *epd = data; | |
3097 | struct extent_io_tree *tree = epd->tree; | |
4eee4fa4 | 3098 | u64 start = page_offset(page); |
d1310b2e CM |
3099 | u64 delalloc_start; |
3100 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
3101 | u64 end; | |
3102 | u64 cur = start; | |
3103 | u64 extent_offset; | |
3104 | u64 last_byte = i_size_read(inode); | |
3105 | u64 block_start; | |
3106 | u64 iosize; | |
3107 | sector_t sector; | |
2c64c53d | 3108 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
3109 | struct extent_map *em; |
3110 | struct block_device *bdev; | |
3111 | int ret; | |
3112 | int nr = 0; | |
7f3c74fb | 3113 | size_t pg_offset = 0; |
d1310b2e CM |
3114 | size_t blocksize; |
3115 | loff_t i_size = i_size_read(inode); | |
3116 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
3117 | u64 nr_delalloc; | |
3118 | u64 delalloc_end; | |
c8b97818 CM |
3119 | int page_started; |
3120 | int compressed; | |
ffbd517d | 3121 | int write_flags; |
771ed689 | 3122 | unsigned long nr_written = 0; |
9e487107 | 3123 | bool fill_delalloc = true; |
d1310b2e | 3124 | |
ffbd517d | 3125 | if (wbc->sync_mode == WB_SYNC_ALL) |
721a9602 | 3126 | write_flags = WRITE_SYNC; |
ffbd517d CM |
3127 | else |
3128 | write_flags = WRITE; | |
3129 | ||
1abe9b8a | 3130 | trace___extent_writepage(page, inode, wbc); |
3131 | ||
d1310b2e | 3132 | WARN_ON(!PageLocked(page)); |
bf0da8c1 CM |
3133 | |
3134 | ClearPageError(page); | |
3135 | ||
7f3c74fb | 3136 | pg_offset = i_size & (PAGE_CACHE_SIZE - 1); |
211c17f5 | 3137 | if (page->index > end_index || |
7f3c74fb | 3138 | (page->index == end_index && !pg_offset)) { |
d47992f8 | 3139 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE); |
d1310b2e CM |
3140 | unlock_page(page); |
3141 | return 0; | |
3142 | } | |
3143 | ||
3144 | if (page->index == end_index) { | |
3145 | char *userpage; | |
3146 | ||
7ac687d9 | 3147 | userpage = kmap_atomic(page); |
7f3c74fb CM |
3148 | memset(userpage + pg_offset, 0, |
3149 | PAGE_CACHE_SIZE - pg_offset); | |
7ac687d9 | 3150 | kunmap_atomic(userpage); |
211c17f5 | 3151 | flush_dcache_page(page); |
d1310b2e | 3152 | } |
7f3c74fb | 3153 | pg_offset = 0; |
d1310b2e CM |
3154 | |
3155 | set_page_extent_mapped(page); | |
3156 | ||
9e487107 JB |
3157 | if (!tree->ops || !tree->ops->fill_delalloc) |
3158 | fill_delalloc = false; | |
3159 | ||
d1310b2e CM |
3160 | delalloc_start = start; |
3161 | delalloc_end = 0; | |
c8b97818 | 3162 | page_started = 0; |
9e487107 | 3163 | if (!epd->extent_locked && fill_delalloc) { |
f85d7d6c | 3164 | u64 delalloc_to_write = 0; |
11c8349b CM |
3165 | /* |
3166 | * make sure the wbc mapping index is at least updated | |
3167 | * to this page. | |
3168 | */ | |
3169 | update_nr_written(page, wbc, 0); | |
3170 | ||
d397712b | 3171 | while (delalloc_end < page_end) { |
771ed689 | 3172 | nr_delalloc = find_lock_delalloc_range(inode, tree, |
c8b97818 CM |
3173 | page, |
3174 | &delalloc_start, | |
d1310b2e CM |
3175 | &delalloc_end, |
3176 | 128 * 1024 * 1024); | |
771ed689 CM |
3177 | if (nr_delalloc == 0) { |
3178 | delalloc_start = delalloc_end + 1; | |
3179 | continue; | |
3180 | } | |
013bd4c3 TI |
3181 | ret = tree->ops->fill_delalloc(inode, page, |
3182 | delalloc_start, | |
3183 | delalloc_end, | |
3184 | &page_started, | |
3185 | &nr_written); | |
79787eaa JM |
3186 | /* File system has been set read-only */ |
3187 | if (ret) { | |
3188 | SetPageError(page); | |
3189 | goto done; | |
3190 | } | |
f85d7d6c CM |
3191 | /* |
3192 | * delalloc_end is already one less than the total | |
3193 | * length, so we don't subtract one from | |
3194 | * PAGE_CACHE_SIZE | |
3195 | */ | |
3196 | delalloc_to_write += (delalloc_end - delalloc_start + | |
3197 | PAGE_CACHE_SIZE) >> | |
3198 | PAGE_CACHE_SHIFT; | |
d1310b2e | 3199 | delalloc_start = delalloc_end + 1; |
d1310b2e | 3200 | } |
f85d7d6c CM |
3201 | if (wbc->nr_to_write < delalloc_to_write) { |
3202 | int thresh = 8192; | |
3203 | ||
3204 | if (delalloc_to_write < thresh * 2) | |
3205 | thresh = delalloc_to_write; | |
3206 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3207 | thresh); | |
3208 | } | |
c8b97818 | 3209 | |
771ed689 CM |
3210 | /* did the fill delalloc function already unlock and start |
3211 | * the IO? | |
3212 | */ | |
3213 | if (page_started) { | |
3214 | ret = 0; | |
11c8349b CM |
3215 | /* |
3216 | * we've unlocked the page, so we can't update | |
3217 | * the mapping's writeback index, just update | |
3218 | * nr_to_write. | |
3219 | */ | |
3220 | wbc->nr_to_write -= nr_written; | |
3221 | goto done_unlocked; | |
771ed689 | 3222 | } |
c8b97818 | 3223 | } |
247e743c | 3224 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
3225 | ret = tree->ops->writepage_start_hook(page, start, |
3226 | page_end); | |
87826df0 JM |
3227 | if (ret) { |
3228 | /* Fixup worker will requeue */ | |
3229 | if (ret == -EBUSY) | |
3230 | wbc->pages_skipped++; | |
3231 | else | |
3232 | redirty_page_for_writepage(wbc, page); | |
11c8349b | 3233 | update_nr_written(page, wbc, nr_written); |
247e743c | 3234 | unlock_page(page); |
771ed689 | 3235 | ret = 0; |
11c8349b | 3236 | goto done_unlocked; |
247e743c CM |
3237 | } |
3238 | } | |
3239 | ||
11c8349b CM |
3240 | /* |
3241 | * we don't want to touch the inode after unlocking the page, | |
3242 | * so we update the mapping writeback index now | |
3243 | */ | |
3244 | update_nr_written(page, wbc, nr_written + 1); | |
771ed689 | 3245 | |
d1310b2e | 3246 | end = page_end; |
d1310b2e | 3247 | if (last_byte <= start) { |
e6dcd2dc CM |
3248 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3249 | tree->ops->writepage_end_io_hook(page, start, | |
3250 | page_end, NULL, 1); | |
d1310b2e CM |
3251 | goto done; |
3252 | } | |
3253 | ||
d1310b2e CM |
3254 | blocksize = inode->i_sb->s_blocksize; |
3255 | ||
3256 | while (cur <= end) { | |
3257 | if (cur >= last_byte) { | |
e6dcd2dc CM |
3258 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3259 | tree->ops->writepage_end_io_hook(page, cur, | |
3260 | page_end, NULL, 1); | |
d1310b2e CM |
3261 | break; |
3262 | } | |
7f3c74fb | 3263 | em = epd->get_extent(inode, page, pg_offset, cur, |
d1310b2e | 3264 | end - cur + 1, 1); |
c704005d | 3265 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e CM |
3266 | SetPageError(page); |
3267 | break; | |
3268 | } | |
3269 | ||
3270 | extent_offset = cur - em->start; | |
3271 | BUG_ON(extent_map_end(em) <= cur); | |
3272 | BUG_ON(end < cur); | |
3273 | iosize = min(extent_map_end(em) - cur, end - cur + 1); | |
fda2832f | 3274 | iosize = ALIGN(iosize, blocksize); |
d1310b2e CM |
3275 | sector = (em->block_start + extent_offset) >> 9; |
3276 | bdev = em->bdev; | |
3277 | block_start = em->block_start; | |
c8b97818 | 3278 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
3279 | free_extent_map(em); |
3280 | em = NULL; | |
3281 | ||
c8b97818 CM |
3282 | /* |
3283 | * compressed and inline extents are written through other | |
3284 | * paths in the FS | |
3285 | */ | |
3286 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3287 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
3288 | /* |
3289 | * end_io notification does not happen here for | |
3290 | * compressed extents | |
3291 | */ | |
3292 | if (!compressed && tree->ops && | |
3293 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
3294 | tree->ops->writepage_end_io_hook(page, cur, |
3295 | cur + iosize - 1, | |
3296 | NULL, 1); | |
c8b97818 CM |
3297 | else if (compressed) { |
3298 | /* we don't want to end_page_writeback on | |
3299 | * a compressed extent. this happens | |
3300 | * elsewhere | |
3301 | */ | |
3302 | nr++; | |
3303 | } | |
3304 | ||
3305 | cur += iosize; | |
7f3c74fb | 3306 | pg_offset += iosize; |
d1310b2e CM |
3307 | continue; |
3308 | } | |
d1310b2e CM |
3309 | /* leave this out until we have a page_mkwrite call */ |
3310 | if (0 && !test_range_bit(tree, cur, cur + iosize - 1, | |
9655d298 | 3311 | EXTENT_DIRTY, 0, NULL)) { |
d1310b2e | 3312 | cur = cur + iosize; |
7f3c74fb | 3313 | pg_offset += iosize; |
d1310b2e CM |
3314 | continue; |
3315 | } | |
c8b97818 | 3316 | |
d1310b2e CM |
3317 | if (tree->ops && tree->ops->writepage_io_hook) { |
3318 | ret = tree->ops->writepage_io_hook(page, cur, | |
3319 | cur + iosize - 1); | |
3320 | } else { | |
3321 | ret = 0; | |
3322 | } | |
1259ab75 | 3323 | if (ret) { |
d1310b2e | 3324 | SetPageError(page); |
1259ab75 | 3325 | } else { |
d1310b2e | 3326 | unsigned long max_nr = end_index + 1; |
7f3c74fb | 3327 | |
d1310b2e CM |
3328 | set_range_writeback(tree, cur, cur + iosize - 1); |
3329 | if (!PageWriteback(page)) { | |
d397712b CM |
3330 | printk(KERN_ERR "btrfs warning page %lu not " |
3331 | "writeback, cur %llu end %llu\n", | |
c1c9ff7c | 3332 | page->index, cur, end); |
d1310b2e CM |
3333 | } |
3334 | ||
ffbd517d CM |
3335 | ret = submit_extent_page(write_flags, tree, page, |
3336 | sector, iosize, pg_offset, | |
3337 | bdev, &epd->bio, max_nr, | |
c8b97818 CM |
3338 | end_bio_extent_writepage, |
3339 | 0, 0, 0); | |
d1310b2e CM |
3340 | if (ret) |
3341 | SetPageError(page); | |
3342 | } | |
3343 | cur = cur + iosize; | |
7f3c74fb | 3344 | pg_offset += iosize; |
d1310b2e CM |
3345 | nr++; |
3346 | } | |
3347 | done: | |
3348 | if (nr == 0) { | |
3349 | /* make sure the mapping tag for page dirty gets cleared */ | |
3350 | set_page_writeback(page); | |
3351 | end_page_writeback(page); | |
3352 | } | |
d1310b2e | 3353 | unlock_page(page); |
771ed689 | 3354 | |
11c8349b CM |
3355 | done_unlocked: |
3356 | ||
2c64c53d CM |
3357 | /* drop our reference on any cached states */ |
3358 | free_extent_state(cached_state); | |
d1310b2e CM |
3359 | return 0; |
3360 | } | |
3361 | ||
0b32f4bb JB |
3362 | static int eb_wait(void *word) |
3363 | { | |
3364 | io_schedule(); | |
3365 | return 0; | |
3366 | } | |
3367 | ||
fd8b2b61 | 3368 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb JB |
3369 | { |
3370 | wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait, | |
3371 | TASK_UNINTERRUPTIBLE); | |
3372 | } | |
3373 | ||
3374 | static int lock_extent_buffer_for_io(struct extent_buffer *eb, | |
3375 | struct btrfs_fs_info *fs_info, | |
3376 | struct extent_page_data *epd) | |
3377 | { | |
3378 | unsigned long i, num_pages; | |
3379 | int flush = 0; | |
3380 | int ret = 0; | |
3381 | ||
3382 | if (!btrfs_try_tree_write_lock(eb)) { | |
3383 | flush = 1; | |
3384 | flush_write_bio(epd); | |
3385 | btrfs_tree_lock(eb); | |
3386 | } | |
3387 | ||
3388 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3389 | btrfs_tree_unlock(eb); | |
3390 | if (!epd->sync_io) | |
3391 | return 0; | |
3392 | if (!flush) { | |
3393 | flush_write_bio(epd); | |
3394 | flush = 1; | |
3395 | } | |
a098d8e8 CM |
3396 | while (1) { |
3397 | wait_on_extent_buffer_writeback(eb); | |
3398 | btrfs_tree_lock(eb); | |
3399 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3400 | break; | |
0b32f4bb | 3401 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3402 | } |
3403 | } | |
3404 | ||
51561ffe JB |
3405 | /* |
3406 | * We need to do this to prevent races in people who check if the eb is | |
3407 | * under IO since we can end up having no IO bits set for a short period | |
3408 | * of time. | |
3409 | */ | |
3410 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
3411 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3412 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 3413 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 3414 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
e2d84521 MX |
3415 | __percpu_counter_add(&fs_info->dirty_metadata_bytes, |
3416 | -eb->len, | |
3417 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 3418 | ret = 1; |
51561ffe JB |
3419 | } else { |
3420 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
3421 | } |
3422 | ||
3423 | btrfs_tree_unlock(eb); | |
3424 | ||
3425 | if (!ret) | |
3426 | return ret; | |
3427 | ||
3428 | num_pages = num_extent_pages(eb->start, eb->len); | |
3429 | for (i = 0; i < num_pages; i++) { | |
3430 | struct page *p = extent_buffer_page(eb, i); | |
3431 | ||
3432 | if (!trylock_page(p)) { | |
3433 | if (!flush) { | |
3434 | flush_write_bio(epd); | |
3435 | flush = 1; | |
3436 | } | |
3437 | lock_page(p); | |
3438 | } | |
3439 | } | |
3440 | ||
3441 | return ret; | |
3442 | } | |
3443 | ||
3444 | static void end_extent_buffer_writeback(struct extent_buffer *eb) | |
3445 | { | |
3446 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
3447 | smp_mb__after_clear_bit(); | |
3448 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); | |
3449 | } | |
3450 | ||
3451 | static void end_bio_extent_buffer_writepage(struct bio *bio, int err) | |
3452 | { | |
3453 | int uptodate = err == 0; | |
3454 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
3455 | struct extent_buffer *eb; | |
3456 | int done; | |
3457 | ||
3458 | do { | |
3459 | struct page *page = bvec->bv_page; | |
3460 | ||
3461 | bvec--; | |
3462 | eb = (struct extent_buffer *)page->private; | |
3463 | BUG_ON(!eb); | |
3464 | done = atomic_dec_and_test(&eb->io_pages); | |
3465 | ||
3466 | if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) { | |
3467 | set_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
3468 | ClearPageUptodate(page); | |
3469 | SetPageError(page); | |
3470 | } | |
3471 | ||
3472 | end_page_writeback(page); | |
3473 | ||
3474 | if (!done) | |
3475 | continue; | |
3476 | ||
3477 | end_extent_buffer_writeback(eb); | |
3478 | } while (bvec >= bio->bi_io_vec); | |
3479 | ||
3480 | bio_put(bio); | |
3481 | ||
3482 | } | |
3483 | ||
3484 | static int write_one_eb(struct extent_buffer *eb, | |
3485 | struct btrfs_fs_info *fs_info, | |
3486 | struct writeback_control *wbc, | |
3487 | struct extent_page_data *epd) | |
3488 | { | |
3489 | struct block_device *bdev = fs_info->fs_devices->latest_bdev; | |
f28491e0 | 3490 | struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree; |
0b32f4bb JB |
3491 | u64 offset = eb->start; |
3492 | unsigned long i, num_pages; | |
de0022b9 | 3493 | unsigned long bio_flags = 0; |
d4c7ca86 | 3494 | int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META; |
d7dbe9e7 | 3495 | int ret = 0; |
0b32f4bb JB |
3496 | |
3497 | clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
3498 | num_pages = num_extent_pages(eb->start, eb->len); | |
3499 | atomic_set(&eb->io_pages, num_pages); | |
de0022b9 JB |
3500 | if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID) |
3501 | bio_flags = EXTENT_BIO_TREE_LOG; | |
3502 | ||
0b32f4bb JB |
3503 | for (i = 0; i < num_pages; i++) { |
3504 | struct page *p = extent_buffer_page(eb, i); | |
3505 | ||
3506 | clear_page_dirty_for_io(p); | |
3507 | set_page_writeback(p); | |
f28491e0 | 3508 | ret = submit_extent_page(rw, tree, p, offset >> 9, |
0b32f4bb JB |
3509 | PAGE_CACHE_SIZE, 0, bdev, &epd->bio, |
3510 | -1, end_bio_extent_buffer_writepage, | |
de0022b9 JB |
3511 | 0, epd->bio_flags, bio_flags); |
3512 | epd->bio_flags = bio_flags; | |
0b32f4bb JB |
3513 | if (ret) { |
3514 | set_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
3515 | SetPageError(p); | |
3516 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) | |
3517 | end_extent_buffer_writeback(eb); | |
3518 | ret = -EIO; | |
3519 | break; | |
3520 | } | |
3521 | offset += PAGE_CACHE_SIZE; | |
3522 | update_nr_written(p, wbc, 1); | |
3523 | unlock_page(p); | |
3524 | } | |
3525 | ||
3526 | if (unlikely(ret)) { | |
3527 | for (; i < num_pages; i++) { | |
3528 | struct page *p = extent_buffer_page(eb, i); | |
3529 | unlock_page(p); | |
3530 | } | |
3531 | } | |
3532 | ||
3533 | return ret; | |
3534 | } | |
3535 | ||
3536 | int btree_write_cache_pages(struct address_space *mapping, | |
3537 | struct writeback_control *wbc) | |
3538 | { | |
3539 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; | |
3540 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; | |
3541 | struct extent_buffer *eb, *prev_eb = NULL; | |
3542 | struct extent_page_data epd = { | |
3543 | .bio = NULL, | |
3544 | .tree = tree, | |
3545 | .extent_locked = 0, | |
3546 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
de0022b9 | 3547 | .bio_flags = 0, |
0b32f4bb JB |
3548 | }; |
3549 | int ret = 0; | |
3550 | int done = 0; | |
3551 | int nr_to_write_done = 0; | |
3552 | struct pagevec pvec; | |
3553 | int nr_pages; | |
3554 | pgoff_t index; | |
3555 | pgoff_t end; /* Inclusive */ | |
3556 | int scanned = 0; | |
3557 | int tag; | |
3558 | ||
3559 | pagevec_init(&pvec, 0); | |
3560 | if (wbc->range_cyclic) { | |
3561 | index = mapping->writeback_index; /* Start from prev offset */ | |
3562 | end = -1; | |
3563 | } else { | |
3564 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
3565 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
3566 | scanned = 1; | |
3567 | } | |
3568 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3569 | tag = PAGECACHE_TAG_TOWRITE; | |
3570 | else | |
3571 | tag = PAGECACHE_TAG_DIRTY; | |
3572 | retry: | |
3573 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3574 | tag_pages_for_writeback(mapping, index, end); | |
3575 | while (!done && !nr_to_write_done && (index <= end) && | |
3576 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, | |
3577 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
3578 | unsigned i; | |
3579 | ||
3580 | scanned = 1; | |
3581 | for (i = 0; i < nr_pages; i++) { | |
3582 | struct page *page = pvec.pages[i]; | |
3583 | ||
3584 | if (!PagePrivate(page)) | |
3585 | continue; | |
3586 | ||
3587 | if (!wbc->range_cyclic && page->index > end) { | |
3588 | done = 1; | |
3589 | break; | |
3590 | } | |
3591 | ||
b5bae261 JB |
3592 | spin_lock(&mapping->private_lock); |
3593 | if (!PagePrivate(page)) { | |
3594 | spin_unlock(&mapping->private_lock); | |
3595 | continue; | |
3596 | } | |
3597 | ||
0b32f4bb | 3598 | eb = (struct extent_buffer *)page->private; |
b5bae261 JB |
3599 | |
3600 | /* | |
3601 | * Shouldn't happen and normally this would be a BUG_ON | |
3602 | * but no sense in crashing the users box for something | |
3603 | * we can survive anyway. | |
3604 | */ | |
fae7f21c | 3605 | if (WARN_ON(!eb)) { |
b5bae261 | 3606 | spin_unlock(&mapping->private_lock); |
0b32f4bb JB |
3607 | continue; |
3608 | } | |
3609 | ||
b5bae261 JB |
3610 | if (eb == prev_eb) { |
3611 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 3612 | continue; |
b5bae261 | 3613 | } |
0b32f4bb | 3614 | |
b5bae261 JB |
3615 | ret = atomic_inc_not_zero(&eb->refs); |
3616 | spin_unlock(&mapping->private_lock); | |
3617 | if (!ret) | |
0b32f4bb | 3618 | continue; |
0b32f4bb JB |
3619 | |
3620 | prev_eb = eb; | |
3621 | ret = lock_extent_buffer_for_io(eb, fs_info, &epd); | |
3622 | if (!ret) { | |
3623 | free_extent_buffer(eb); | |
3624 | continue; | |
3625 | } | |
3626 | ||
3627 | ret = write_one_eb(eb, fs_info, wbc, &epd); | |
3628 | if (ret) { | |
3629 | done = 1; | |
3630 | free_extent_buffer(eb); | |
3631 | break; | |
3632 | } | |
3633 | free_extent_buffer(eb); | |
3634 | ||
3635 | /* | |
3636 | * the filesystem may choose to bump up nr_to_write. | |
3637 | * We have to make sure to honor the new nr_to_write | |
3638 | * at any time | |
3639 | */ | |
3640 | nr_to_write_done = wbc->nr_to_write <= 0; | |
3641 | } | |
3642 | pagevec_release(&pvec); | |
3643 | cond_resched(); | |
3644 | } | |
3645 | if (!scanned && !done) { | |
3646 | /* | |
3647 | * We hit the last page and there is more work to be done: wrap | |
3648 | * back to the start of the file | |
3649 | */ | |
3650 | scanned = 1; | |
3651 | index = 0; | |
3652 | goto retry; | |
3653 | } | |
3654 | flush_write_bio(&epd); | |
3655 | return ret; | |
3656 | } | |
3657 | ||
d1310b2e | 3658 | /** |
4bef0848 | 3659 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
3660 | * @mapping: address space structure to write |
3661 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
3662 | * @writepage: function called for each page | |
3663 | * @data: data passed to writepage function | |
3664 | * | |
3665 | * If a page is already under I/O, write_cache_pages() skips it, even | |
3666 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
3667 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
3668 | * and msync() need to guarantee that all the data which was dirty at the time | |
3669 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
3670 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
3671 | * existing IO to complete. | |
3672 | */ | |
b2950863 | 3673 | static int extent_write_cache_pages(struct extent_io_tree *tree, |
4bef0848 CM |
3674 | struct address_space *mapping, |
3675 | struct writeback_control *wbc, | |
d2c3f4f6 CM |
3676 | writepage_t writepage, void *data, |
3677 | void (*flush_fn)(void *)) | |
d1310b2e | 3678 | { |
7fd1a3f7 | 3679 | struct inode *inode = mapping->host; |
d1310b2e CM |
3680 | int ret = 0; |
3681 | int done = 0; | |
f85d7d6c | 3682 | int nr_to_write_done = 0; |
d1310b2e CM |
3683 | struct pagevec pvec; |
3684 | int nr_pages; | |
3685 | pgoff_t index; | |
3686 | pgoff_t end; /* Inclusive */ | |
3687 | int scanned = 0; | |
f7aaa06b | 3688 | int tag; |
d1310b2e | 3689 | |
7fd1a3f7 JB |
3690 | /* |
3691 | * We have to hold onto the inode so that ordered extents can do their | |
3692 | * work when the IO finishes. The alternative to this is failing to add | |
3693 | * an ordered extent if the igrab() fails there and that is a huge pain | |
3694 | * to deal with, so instead just hold onto the inode throughout the | |
3695 | * writepages operation. If it fails here we are freeing up the inode | |
3696 | * anyway and we'd rather not waste our time writing out stuff that is | |
3697 | * going to be truncated anyway. | |
3698 | */ | |
3699 | if (!igrab(inode)) | |
3700 | return 0; | |
3701 | ||
d1310b2e CM |
3702 | pagevec_init(&pvec, 0); |
3703 | if (wbc->range_cyclic) { | |
3704 | index = mapping->writeback_index; /* Start from prev offset */ | |
3705 | end = -1; | |
3706 | } else { | |
3707 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
3708 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
3709 | scanned = 1; |
3710 | } | |
f7aaa06b JB |
3711 | if (wbc->sync_mode == WB_SYNC_ALL) |
3712 | tag = PAGECACHE_TAG_TOWRITE; | |
3713 | else | |
3714 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 3715 | retry: |
f7aaa06b JB |
3716 | if (wbc->sync_mode == WB_SYNC_ALL) |
3717 | tag_pages_for_writeback(mapping, index, end); | |
f85d7d6c | 3718 | while (!done && !nr_to_write_done && (index <= end) && |
f7aaa06b JB |
3719 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
3720 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
d1310b2e CM |
3721 | unsigned i; |
3722 | ||
3723 | scanned = 1; | |
3724 | for (i = 0; i < nr_pages; i++) { | |
3725 | struct page *page = pvec.pages[i]; | |
3726 | ||
3727 | /* | |
3728 | * At this point we hold neither mapping->tree_lock nor | |
3729 | * lock on the page itself: the page may be truncated or | |
3730 | * invalidated (changing page->mapping to NULL), or even | |
3731 | * swizzled back from swapper_space to tmpfs file | |
3732 | * mapping | |
3733 | */ | |
c8f2f24b JB |
3734 | if (!trylock_page(page)) { |
3735 | flush_fn(data); | |
3736 | lock_page(page); | |
01d658f2 | 3737 | } |
d1310b2e CM |
3738 | |
3739 | if (unlikely(page->mapping != mapping)) { | |
3740 | unlock_page(page); | |
3741 | continue; | |
3742 | } | |
3743 | ||
3744 | if (!wbc->range_cyclic && page->index > end) { | |
3745 | done = 1; | |
3746 | unlock_page(page); | |
3747 | continue; | |
3748 | } | |
3749 | ||
d2c3f4f6 | 3750 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 CM |
3751 | if (PageWriteback(page)) |
3752 | flush_fn(data); | |
d1310b2e | 3753 | wait_on_page_writeback(page); |
d2c3f4f6 | 3754 | } |
d1310b2e CM |
3755 | |
3756 | if (PageWriteback(page) || | |
3757 | !clear_page_dirty_for_io(page)) { | |
3758 | unlock_page(page); | |
3759 | continue; | |
3760 | } | |
3761 | ||
3762 | ret = (*writepage)(page, wbc, data); | |
3763 | ||
3764 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
3765 | unlock_page(page); | |
3766 | ret = 0; | |
3767 | } | |
f85d7d6c | 3768 | if (ret) |
d1310b2e | 3769 | done = 1; |
f85d7d6c CM |
3770 | |
3771 | /* | |
3772 | * the filesystem may choose to bump up nr_to_write. | |
3773 | * We have to make sure to honor the new nr_to_write | |
3774 | * at any time | |
3775 | */ | |
3776 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
3777 | } |
3778 | pagevec_release(&pvec); | |
3779 | cond_resched(); | |
3780 | } | |
3781 | if (!scanned && !done) { | |
3782 | /* | |
3783 | * We hit the last page and there is more work to be done: wrap | |
3784 | * back to the start of the file | |
3785 | */ | |
3786 | scanned = 1; | |
3787 | index = 0; | |
3788 | goto retry; | |
3789 | } | |
7fd1a3f7 | 3790 | btrfs_add_delayed_iput(inode); |
d1310b2e CM |
3791 | return ret; |
3792 | } | |
d1310b2e | 3793 | |
ffbd517d | 3794 | static void flush_epd_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 3795 | { |
d2c3f4f6 | 3796 | if (epd->bio) { |
355808c2 JM |
3797 | int rw = WRITE; |
3798 | int ret; | |
3799 | ||
ffbd517d | 3800 | if (epd->sync_io) |
355808c2 JM |
3801 | rw = WRITE_SYNC; |
3802 | ||
de0022b9 | 3803 | ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags); |
79787eaa | 3804 | BUG_ON(ret < 0); /* -ENOMEM */ |
d2c3f4f6 CM |
3805 | epd->bio = NULL; |
3806 | } | |
3807 | } | |
3808 | ||
ffbd517d CM |
3809 | static noinline void flush_write_bio(void *data) |
3810 | { | |
3811 | struct extent_page_data *epd = data; | |
3812 | flush_epd_write_bio(epd); | |
3813 | } | |
3814 | ||
d1310b2e CM |
3815 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, |
3816 | get_extent_t *get_extent, | |
3817 | struct writeback_control *wbc) | |
3818 | { | |
3819 | int ret; | |
d1310b2e CM |
3820 | struct extent_page_data epd = { |
3821 | .bio = NULL, | |
3822 | .tree = tree, | |
3823 | .get_extent = get_extent, | |
771ed689 | 3824 | .extent_locked = 0, |
ffbd517d | 3825 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
de0022b9 | 3826 | .bio_flags = 0, |
d1310b2e | 3827 | }; |
d1310b2e | 3828 | |
d1310b2e CM |
3829 | ret = __extent_writepage(page, wbc, &epd); |
3830 | ||
ffbd517d | 3831 | flush_epd_write_bio(&epd); |
d1310b2e CM |
3832 | return ret; |
3833 | } | |
d1310b2e | 3834 | |
771ed689 CM |
3835 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, |
3836 | u64 start, u64 end, get_extent_t *get_extent, | |
3837 | int mode) | |
3838 | { | |
3839 | int ret = 0; | |
3840 | struct address_space *mapping = inode->i_mapping; | |
3841 | struct page *page; | |
3842 | unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >> | |
3843 | PAGE_CACHE_SHIFT; | |
3844 | ||
3845 | struct extent_page_data epd = { | |
3846 | .bio = NULL, | |
3847 | .tree = tree, | |
3848 | .get_extent = get_extent, | |
3849 | .extent_locked = 1, | |
ffbd517d | 3850 | .sync_io = mode == WB_SYNC_ALL, |
de0022b9 | 3851 | .bio_flags = 0, |
771ed689 CM |
3852 | }; |
3853 | struct writeback_control wbc_writepages = { | |
771ed689 | 3854 | .sync_mode = mode, |
771ed689 CM |
3855 | .nr_to_write = nr_pages * 2, |
3856 | .range_start = start, | |
3857 | .range_end = end + 1, | |
3858 | }; | |
3859 | ||
d397712b | 3860 | while (start <= end) { |
771ed689 CM |
3861 | page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); |
3862 | if (clear_page_dirty_for_io(page)) | |
3863 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
3864 | else { | |
3865 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
3866 | tree->ops->writepage_end_io_hook(page, start, | |
3867 | start + PAGE_CACHE_SIZE - 1, | |
3868 | NULL, 1); | |
3869 | unlock_page(page); | |
3870 | } | |
3871 | page_cache_release(page); | |
3872 | start += PAGE_CACHE_SIZE; | |
3873 | } | |
3874 | ||
ffbd517d | 3875 | flush_epd_write_bio(&epd); |
771ed689 CM |
3876 | return ret; |
3877 | } | |
d1310b2e CM |
3878 | |
3879 | int extent_writepages(struct extent_io_tree *tree, | |
3880 | struct address_space *mapping, | |
3881 | get_extent_t *get_extent, | |
3882 | struct writeback_control *wbc) | |
3883 | { | |
3884 | int ret = 0; | |
3885 | struct extent_page_data epd = { | |
3886 | .bio = NULL, | |
3887 | .tree = tree, | |
3888 | .get_extent = get_extent, | |
771ed689 | 3889 | .extent_locked = 0, |
ffbd517d | 3890 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
de0022b9 | 3891 | .bio_flags = 0, |
d1310b2e CM |
3892 | }; |
3893 | ||
4bef0848 | 3894 | ret = extent_write_cache_pages(tree, mapping, wbc, |
d2c3f4f6 CM |
3895 | __extent_writepage, &epd, |
3896 | flush_write_bio); | |
ffbd517d | 3897 | flush_epd_write_bio(&epd); |
d1310b2e CM |
3898 | return ret; |
3899 | } | |
d1310b2e CM |
3900 | |
3901 | int extent_readpages(struct extent_io_tree *tree, | |
3902 | struct address_space *mapping, | |
3903 | struct list_head *pages, unsigned nr_pages, | |
3904 | get_extent_t get_extent) | |
3905 | { | |
3906 | struct bio *bio = NULL; | |
3907 | unsigned page_idx; | |
c8b97818 | 3908 | unsigned long bio_flags = 0; |
67c9684f LB |
3909 | struct page *pagepool[16]; |
3910 | struct page *page; | |
125bac01 | 3911 | struct extent_map *em_cached = NULL; |
67c9684f | 3912 | int nr = 0; |
d1310b2e | 3913 | |
d1310b2e | 3914 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
67c9684f | 3915 | page = list_entry(pages->prev, struct page, lru); |
d1310b2e CM |
3916 | |
3917 | prefetchw(&page->flags); | |
3918 | list_del(&page->lru); | |
67c9684f | 3919 | if (add_to_page_cache_lru(page, mapping, |
43e817a1 | 3920 | page->index, GFP_NOFS)) { |
67c9684f LB |
3921 | page_cache_release(page); |
3922 | continue; | |
d1310b2e | 3923 | } |
67c9684f LB |
3924 | |
3925 | pagepool[nr++] = page; | |
3926 | if (nr < ARRAY_SIZE(pagepool)) | |
3927 | continue; | |
125bac01 | 3928 | __extent_readpages(tree, pagepool, nr, get_extent, &em_cached, |
9974090b | 3929 | &bio, 0, &bio_flags, READ); |
67c9684f | 3930 | nr = 0; |
d1310b2e | 3931 | } |
9974090b | 3932 | if (nr) |
125bac01 | 3933 | __extent_readpages(tree, pagepool, nr, get_extent, &em_cached, |
9974090b | 3934 | &bio, 0, &bio_flags, READ); |
67c9684f | 3935 | |
125bac01 MX |
3936 | if (em_cached) |
3937 | free_extent_map(em_cached); | |
3938 | ||
d1310b2e CM |
3939 | BUG_ON(!list_empty(pages)); |
3940 | if (bio) | |
79787eaa | 3941 | return submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
3942 | return 0; |
3943 | } | |
d1310b2e CM |
3944 | |
3945 | /* | |
3946 | * basic invalidatepage code, this waits on any locked or writeback | |
3947 | * ranges corresponding to the page, and then deletes any extent state | |
3948 | * records from the tree | |
3949 | */ | |
3950 | int extent_invalidatepage(struct extent_io_tree *tree, | |
3951 | struct page *page, unsigned long offset) | |
3952 | { | |
2ac55d41 | 3953 | struct extent_state *cached_state = NULL; |
4eee4fa4 | 3954 | u64 start = page_offset(page); |
d1310b2e CM |
3955 | u64 end = start + PAGE_CACHE_SIZE - 1; |
3956 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
3957 | ||
fda2832f | 3958 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
3959 | if (start > end) |
3960 | return 0; | |
3961 | ||
d0082371 | 3962 | lock_extent_bits(tree, start, end, 0, &cached_state); |
1edbb734 | 3963 | wait_on_page_writeback(page); |
d1310b2e | 3964 | clear_extent_bit(tree, start, end, |
32c00aff JB |
3965 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
3966 | EXTENT_DO_ACCOUNTING, | |
2ac55d41 | 3967 | 1, 1, &cached_state, GFP_NOFS); |
d1310b2e CM |
3968 | return 0; |
3969 | } | |
d1310b2e | 3970 | |
7b13b7b1 CM |
3971 | /* |
3972 | * a helper for releasepage, this tests for areas of the page that | |
3973 | * are locked or under IO and drops the related state bits if it is safe | |
3974 | * to drop the page. | |
3975 | */ | |
48a3b636 ES |
3976 | static int try_release_extent_state(struct extent_map_tree *map, |
3977 | struct extent_io_tree *tree, | |
3978 | struct page *page, gfp_t mask) | |
7b13b7b1 | 3979 | { |
4eee4fa4 | 3980 | u64 start = page_offset(page); |
7b13b7b1 CM |
3981 | u64 end = start + PAGE_CACHE_SIZE - 1; |
3982 | int ret = 1; | |
3983 | ||
211f90e6 | 3984 | if (test_range_bit(tree, start, end, |
8b62b72b | 3985 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
3986 | ret = 0; |
3987 | else { | |
3988 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
3989 | mask = GFP_NOFS; | |
11ef160f CM |
3990 | /* |
3991 | * at this point we can safely clear everything except the | |
3992 | * locked bit and the nodatasum bit | |
3993 | */ | |
e3f24cc5 | 3994 | ret = clear_extent_bit(tree, start, end, |
11ef160f CM |
3995 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
3996 | 0, 0, NULL, mask); | |
e3f24cc5 CM |
3997 | |
3998 | /* if clear_extent_bit failed for enomem reasons, | |
3999 | * we can't allow the release to continue. | |
4000 | */ | |
4001 | if (ret < 0) | |
4002 | ret = 0; | |
4003 | else | |
4004 | ret = 1; | |
7b13b7b1 CM |
4005 | } |
4006 | return ret; | |
4007 | } | |
7b13b7b1 | 4008 | |
d1310b2e CM |
4009 | /* |
4010 | * a helper for releasepage. As long as there are no locked extents | |
4011 | * in the range corresponding to the page, both state records and extent | |
4012 | * map records are removed | |
4013 | */ | |
4014 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
4015 | struct extent_io_tree *tree, struct page *page, |
4016 | gfp_t mask) | |
d1310b2e CM |
4017 | { |
4018 | struct extent_map *em; | |
4eee4fa4 | 4019 | u64 start = page_offset(page); |
d1310b2e | 4020 | u64 end = start + PAGE_CACHE_SIZE - 1; |
7b13b7b1 | 4021 | |
70dec807 CM |
4022 | if ((mask & __GFP_WAIT) && |
4023 | page->mapping->host->i_size > 16 * 1024 * 1024) { | |
39b5637f | 4024 | u64 len; |
70dec807 | 4025 | while (start <= end) { |
39b5637f | 4026 | len = end - start + 1; |
890871be | 4027 | write_lock(&map->lock); |
39b5637f | 4028 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 4029 | if (!em) { |
890871be | 4030 | write_unlock(&map->lock); |
70dec807 CM |
4031 | break; |
4032 | } | |
7f3c74fb CM |
4033 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
4034 | em->start != start) { | |
890871be | 4035 | write_unlock(&map->lock); |
70dec807 CM |
4036 | free_extent_map(em); |
4037 | break; | |
4038 | } | |
4039 | if (!test_range_bit(tree, em->start, | |
4040 | extent_map_end(em) - 1, | |
8b62b72b | 4041 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 4042 | 0, NULL)) { |
70dec807 CM |
4043 | remove_extent_mapping(map, em); |
4044 | /* once for the rb tree */ | |
4045 | free_extent_map(em); | |
4046 | } | |
4047 | start = extent_map_end(em); | |
890871be | 4048 | write_unlock(&map->lock); |
70dec807 CM |
4049 | |
4050 | /* once for us */ | |
d1310b2e CM |
4051 | free_extent_map(em); |
4052 | } | |
d1310b2e | 4053 | } |
7b13b7b1 | 4054 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 4055 | } |
d1310b2e | 4056 | |
ec29ed5b CM |
4057 | /* |
4058 | * helper function for fiemap, which doesn't want to see any holes. | |
4059 | * This maps until we find something past 'last' | |
4060 | */ | |
4061 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
4062 | u64 offset, | |
4063 | u64 last, | |
4064 | get_extent_t *get_extent) | |
4065 | { | |
4066 | u64 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
4067 | struct extent_map *em; | |
4068 | u64 len; | |
4069 | ||
4070 | if (offset >= last) | |
4071 | return NULL; | |
4072 | ||
67871254 | 4073 | while (1) { |
ec29ed5b CM |
4074 | len = last - offset; |
4075 | if (len == 0) | |
4076 | break; | |
fda2832f | 4077 | len = ALIGN(len, sectorsize); |
ec29ed5b | 4078 | em = get_extent(inode, NULL, 0, offset, len, 0); |
c704005d | 4079 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
4080 | return em; |
4081 | ||
4082 | /* if this isn't a hole return it */ | |
4083 | if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) && | |
4084 | em->block_start != EXTENT_MAP_HOLE) { | |
4085 | return em; | |
4086 | } | |
4087 | ||
4088 | /* this is a hole, advance to the next extent */ | |
4089 | offset = extent_map_end(em); | |
4090 | free_extent_map(em); | |
4091 | if (offset >= last) | |
4092 | break; | |
4093 | } | |
4094 | return NULL; | |
4095 | } | |
4096 | ||
fe09e16c LB |
4097 | static noinline int count_ext_ref(u64 inum, u64 offset, u64 root_id, void *ctx) |
4098 | { | |
4099 | unsigned long cnt = *((unsigned long *)ctx); | |
4100 | ||
4101 | cnt++; | |
4102 | *((unsigned long *)ctx) = cnt; | |
4103 | ||
4104 | /* Now we're sure that the extent is shared. */ | |
4105 | if (cnt > 1) | |
4106 | return 1; | |
4107 | return 0; | |
4108 | } | |
4109 | ||
1506fcc8 YS |
4110 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4111 | __u64 start, __u64 len, get_extent_t *get_extent) | |
4112 | { | |
975f84fe | 4113 | int ret = 0; |
1506fcc8 YS |
4114 | u64 off = start; |
4115 | u64 max = start + len; | |
4116 | u32 flags = 0; | |
975f84fe JB |
4117 | u32 found_type; |
4118 | u64 last; | |
ec29ed5b | 4119 | u64 last_for_get_extent = 0; |
1506fcc8 | 4120 | u64 disko = 0; |
ec29ed5b | 4121 | u64 isize = i_size_read(inode); |
975f84fe | 4122 | struct btrfs_key found_key; |
1506fcc8 | 4123 | struct extent_map *em = NULL; |
2ac55d41 | 4124 | struct extent_state *cached_state = NULL; |
975f84fe | 4125 | struct btrfs_path *path; |
1506fcc8 | 4126 | int end = 0; |
ec29ed5b CM |
4127 | u64 em_start = 0; |
4128 | u64 em_len = 0; | |
4129 | u64 em_end = 0; | |
1506fcc8 YS |
4130 | |
4131 | if (len == 0) | |
4132 | return -EINVAL; | |
4133 | ||
975f84fe JB |
4134 | path = btrfs_alloc_path(); |
4135 | if (!path) | |
4136 | return -ENOMEM; | |
4137 | path->leave_spinning = 1; | |
4138 | ||
4d479cf0 JB |
4139 | start = ALIGN(start, BTRFS_I(inode)->root->sectorsize); |
4140 | len = ALIGN(len, BTRFS_I(inode)->root->sectorsize); | |
4141 | ||
ec29ed5b CM |
4142 | /* |
4143 | * lookup the last file extent. We're not using i_size here | |
4144 | * because there might be preallocation past i_size | |
4145 | */ | |
975f84fe | 4146 | ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root, |
33345d01 | 4147 | path, btrfs_ino(inode), -1, 0); |
975f84fe JB |
4148 | if (ret < 0) { |
4149 | btrfs_free_path(path); | |
4150 | return ret; | |
4151 | } | |
4152 | WARN_ON(!ret); | |
4153 | path->slots[0]--; | |
975f84fe JB |
4154 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
4155 | found_type = btrfs_key_type(&found_key); | |
4156 | ||
ec29ed5b | 4157 | /* No extents, but there might be delalloc bits */ |
33345d01 | 4158 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 4159 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
4160 | /* have to trust i_size as the end */ |
4161 | last = (u64)-1; | |
4162 | last_for_get_extent = isize; | |
4163 | } else { | |
4164 | /* | |
4165 | * remember the start of the last extent. There are a | |
4166 | * bunch of different factors that go into the length of the | |
4167 | * extent, so its much less complex to remember where it started | |
4168 | */ | |
4169 | last = found_key.offset; | |
4170 | last_for_get_extent = last + 1; | |
975f84fe | 4171 | } |
fe09e16c | 4172 | btrfs_release_path(path); |
975f84fe | 4173 | |
ec29ed5b CM |
4174 | /* |
4175 | * we might have some extents allocated but more delalloc past those | |
4176 | * extents. so, we trust isize unless the start of the last extent is | |
4177 | * beyond isize | |
4178 | */ | |
4179 | if (last < isize) { | |
4180 | last = (u64)-1; | |
4181 | last_for_get_extent = isize; | |
4182 | } | |
4183 | ||
a52f4cd2 | 4184 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0, |
d0082371 | 4185 | &cached_state); |
ec29ed5b | 4186 | |
4d479cf0 | 4187 | em = get_extent_skip_holes(inode, start, last_for_get_extent, |
ec29ed5b | 4188 | get_extent); |
1506fcc8 YS |
4189 | if (!em) |
4190 | goto out; | |
4191 | if (IS_ERR(em)) { | |
4192 | ret = PTR_ERR(em); | |
4193 | goto out; | |
4194 | } | |
975f84fe | 4195 | |
1506fcc8 | 4196 | while (!end) { |
b76bb701 | 4197 | u64 offset_in_extent = 0; |
ea8efc74 CM |
4198 | |
4199 | /* break if the extent we found is outside the range */ | |
4200 | if (em->start >= max || extent_map_end(em) < off) | |
4201 | break; | |
4202 | ||
4203 | /* | |
4204 | * get_extent may return an extent that starts before our | |
4205 | * requested range. We have to make sure the ranges | |
4206 | * we return to fiemap always move forward and don't | |
4207 | * overlap, so adjust the offsets here | |
4208 | */ | |
4209 | em_start = max(em->start, off); | |
1506fcc8 | 4210 | |
ea8efc74 CM |
4211 | /* |
4212 | * record the offset from the start of the extent | |
b76bb701 JB |
4213 | * for adjusting the disk offset below. Only do this if the |
4214 | * extent isn't compressed since our in ram offset may be past | |
4215 | * what we have actually allocated on disk. | |
ea8efc74 | 4216 | */ |
b76bb701 JB |
4217 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
4218 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 4219 | em_end = extent_map_end(em); |
ea8efc74 | 4220 | em_len = em_end - em_start; |
1506fcc8 YS |
4221 | disko = 0; |
4222 | flags = 0; | |
4223 | ||
ea8efc74 CM |
4224 | /* |
4225 | * bump off for our next call to get_extent | |
4226 | */ | |
4227 | off = extent_map_end(em); | |
4228 | if (off >= max) | |
4229 | end = 1; | |
4230 | ||
93dbfad7 | 4231 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
4232 | end = 1; |
4233 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 4234 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
4235 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
4236 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 4237 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
4238 | flags |= (FIEMAP_EXTENT_DELALLOC | |
4239 | FIEMAP_EXTENT_UNKNOWN); | |
93dbfad7 | 4240 | } else { |
fe09e16c LB |
4241 | unsigned long ref_cnt = 0; |
4242 | ||
ea8efc74 | 4243 | disko = em->block_start + offset_in_extent; |
fe09e16c LB |
4244 | |
4245 | /* | |
4246 | * As btrfs supports shared space, this information | |
4247 | * can be exported to userspace tools via | |
4248 | * flag FIEMAP_EXTENT_SHARED. | |
4249 | */ | |
4250 | ret = iterate_inodes_from_logical( | |
4251 | em->block_start, | |
4252 | BTRFS_I(inode)->root->fs_info, | |
4253 | path, count_ext_ref, &ref_cnt); | |
4254 | if (ret < 0 && ret != -ENOENT) | |
4255 | goto out_free; | |
4256 | ||
4257 | if (ref_cnt > 1) | |
4258 | flags |= FIEMAP_EXTENT_SHARED; | |
1506fcc8 YS |
4259 | } |
4260 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
4261 | flags |= FIEMAP_EXTENT_ENCODED; | |
4262 | ||
1506fcc8 YS |
4263 | free_extent_map(em); |
4264 | em = NULL; | |
ec29ed5b CM |
4265 | if ((em_start >= last) || em_len == (u64)-1 || |
4266 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
4267 | flags |= FIEMAP_EXTENT_LAST; |
4268 | end = 1; | |
4269 | } | |
4270 | ||
ec29ed5b CM |
4271 | /* now scan forward to see if this is really the last extent. */ |
4272 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
4273 | get_extent); | |
4274 | if (IS_ERR(em)) { | |
4275 | ret = PTR_ERR(em); | |
4276 | goto out; | |
4277 | } | |
4278 | if (!em) { | |
975f84fe JB |
4279 | flags |= FIEMAP_EXTENT_LAST; |
4280 | end = 1; | |
4281 | } | |
ec29ed5b CM |
4282 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, |
4283 | em_len, flags); | |
4284 | if (ret) | |
4285 | goto out_free; | |
1506fcc8 YS |
4286 | } |
4287 | out_free: | |
4288 | free_extent_map(em); | |
4289 | out: | |
fe09e16c | 4290 | btrfs_free_path(path); |
a52f4cd2 | 4291 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
2ac55d41 | 4292 | &cached_state, GFP_NOFS); |
1506fcc8 YS |
4293 | return ret; |
4294 | } | |
4295 | ||
727011e0 CM |
4296 | static void __free_extent_buffer(struct extent_buffer *eb) |
4297 | { | |
6d49ba1b | 4298 | btrfs_leak_debug_del(&eb->leak_list); |
727011e0 CM |
4299 | kmem_cache_free(extent_buffer_cache, eb); |
4300 | } | |
4301 | ||
db7f3436 JB |
4302 | static int extent_buffer_under_io(struct extent_buffer *eb) |
4303 | { | |
4304 | return (atomic_read(&eb->io_pages) || | |
4305 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4306 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4307 | } | |
4308 | ||
4309 | /* | |
4310 | * Helper for releasing extent buffer page. | |
4311 | */ | |
4312 | static void btrfs_release_extent_buffer_page(struct extent_buffer *eb, | |
4313 | unsigned long start_idx) | |
4314 | { | |
4315 | unsigned long index; | |
4316 | unsigned long num_pages; | |
4317 | struct page *page; | |
4318 | int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4319 | ||
4320 | BUG_ON(extent_buffer_under_io(eb)); | |
4321 | ||
4322 | num_pages = num_extent_pages(eb->start, eb->len); | |
4323 | index = start_idx + num_pages; | |
4324 | if (start_idx >= index) | |
4325 | return; | |
4326 | ||
4327 | do { | |
4328 | index--; | |
4329 | page = extent_buffer_page(eb, index); | |
4330 | if (page && mapped) { | |
4331 | spin_lock(&page->mapping->private_lock); | |
4332 | /* | |
4333 | * We do this since we'll remove the pages after we've | |
4334 | * removed the eb from the radix tree, so we could race | |
4335 | * and have this page now attached to the new eb. So | |
4336 | * only clear page_private if it's still connected to | |
4337 | * this eb. | |
4338 | */ | |
4339 | if (PagePrivate(page) && | |
4340 | page->private == (unsigned long)eb) { | |
4341 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4342 | BUG_ON(PageDirty(page)); | |
4343 | BUG_ON(PageWriteback(page)); | |
4344 | /* | |
4345 | * We need to make sure we haven't be attached | |
4346 | * to a new eb. | |
4347 | */ | |
4348 | ClearPagePrivate(page); | |
4349 | set_page_private(page, 0); | |
4350 | /* One for the page private */ | |
4351 | page_cache_release(page); | |
4352 | } | |
4353 | spin_unlock(&page->mapping->private_lock); | |
4354 | ||
4355 | } | |
4356 | if (page) { | |
4357 | /* One for when we alloced the page */ | |
4358 | page_cache_release(page); | |
4359 | } | |
4360 | } while (index != start_idx); | |
4361 | } | |
4362 | ||
4363 | /* | |
4364 | * Helper for releasing the extent buffer. | |
4365 | */ | |
4366 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4367 | { | |
4368 | btrfs_release_extent_buffer_page(eb, 0); | |
4369 | __free_extent_buffer(eb); | |
4370 | } | |
4371 | ||
f28491e0 JB |
4372 | static struct extent_buffer * |
4373 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
4374 | unsigned long len, gfp_t mask) | |
d1310b2e CM |
4375 | { |
4376 | struct extent_buffer *eb = NULL; | |
4377 | ||
d1310b2e | 4378 | eb = kmem_cache_zalloc(extent_buffer_cache, mask); |
91ca338d TI |
4379 | if (eb == NULL) |
4380 | return NULL; | |
d1310b2e CM |
4381 | eb->start = start; |
4382 | eb->len = len; | |
f28491e0 | 4383 | eb->fs_info = fs_info; |
815a51c7 | 4384 | eb->bflags = 0; |
bd681513 CM |
4385 | rwlock_init(&eb->lock); |
4386 | atomic_set(&eb->write_locks, 0); | |
4387 | atomic_set(&eb->read_locks, 0); | |
4388 | atomic_set(&eb->blocking_readers, 0); | |
4389 | atomic_set(&eb->blocking_writers, 0); | |
4390 | atomic_set(&eb->spinning_readers, 0); | |
4391 | atomic_set(&eb->spinning_writers, 0); | |
5b25f70f | 4392 | eb->lock_nested = 0; |
bd681513 CM |
4393 | init_waitqueue_head(&eb->write_lock_wq); |
4394 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 4395 | |
6d49ba1b ES |
4396 | btrfs_leak_debug_add(&eb->leak_list, &buffers); |
4397 | ||
3083ee2e | 4398 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 4399 | atomic_set(&eb->refs, 1); |
0b32f4bb | 4400 | atomic_set(&eb->io_pages, 0); |
727011e0 | 4401 | |
b8dae313 DS |
4402 | /* |
4403 | * Sanity checks, currently the maximum is 64k covered by 16x 4k pages | |
4404 | */ | |
4405 | BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE | |
4406 | > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
4407 | BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
d1310b2e CM |
4408 | |
4409 | return eb; | |
4410 | } | |
4411 | ||
815a51c7 JS |
4412 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) |
4413 | { | |
4414 | unsigned long i; | |
4415 | struct page *p; | |
4416 | struct extent_buffer *new; | |
4417 | unsigned long num_pages = num_extent_pages(src->start, src->len); | |
4418 | ||
9ec72677 | 4419 | new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_NOFS); |
815a51c7 JS |
4420 | if (new == NULL) |
4421 | return NULL; | |
4422 | ||
4423 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4424 | p = alloc_page(GFP_NOFS); |
db7f3436 JB |
4425 | if (!p) { |
4426 | btrfs_release_extent_buffer(new); | |
4427 | return NULL; | |
4428 | } | |
815a51c7 JS |
4429 | attach_extent_buffer_page(new, p); |
4430 | WARN_ON(PageDirty(p)); | |
4431 | SetPageUptodate(p); | |
4432 | new->pages[i] = p; | |
4433 | } | |
4434 | ||
4435 | copy_extent_buffer(new, src, 0, 0, src->len); | |
4436 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); | |
4437 | set_bit(EXTENT_BUFFER_DUMMY, &new->bflags); | |
4438 | ||
4439 | return new; | |
4440 | } | |
4441 | ||
4442 | struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len) | |
4443 | { | |
4444 | struct extent_buffer *eb; | |
4445 | unsigned long num_pages = num_extent_pages(0, len); | |
4446 | unsigned long i; | |
4447 | ||
9ec72677 | 4448 | eb = __alloc_extent_buffer(NULL, start, len, GFP_NOFS); |
815a51c7 JS |
4449 | if (!eb) |
4450 | return NULL; | |
4451 | ||
4452 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4453 | eb->pages[i] = alloc_page(GFP_NOFS); |
815a51c7 JS |
4454 | if (!eb->pages[i]) |
4455 | goto err; | |
4456 | } | |
4457 | set_extent_buffer_uptodate(eb); | |
4458 | btrfs_set_header_nritems(eb, 0); | |
4459 | set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4460 | ||
4461 | return eb; | |
4462 | err: | |
84167d19 SB |
4463 | for (; i > 0; i--) |
4464 | __free_page(eb->pages[i - 1]); | |
815a51c7 JS |
4465 | __free_extent_buffer(eb); |
4466 | return NULL; | |
4467 | } | |
4468 | ||
0b32f4bb JB |
4469 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
4470 | { | |
242e18c7 | 4471 | int refs; |
0b32f4bb JB |
4472 | /* the ref bit is tricky. We have to make sure it is set |
4473 | * if we have the buffer dirty. Otherwise the | |
4474 | * code to free a buffer can end up dropping a dirty | |
4475 | * page | |
4476 | * | |
4477 | * Once the ref bit is set, it won't go away while the | |
4478 | * buffer is dirty or in writeback, and it also won't | |
4479 | * go away while we have the reference count on the | |
4480 | * eb bumped. | |
4481 | * | |
4482 | * We can't just set the ref bit without bumping the | |
4483 | * ref on the eb because free_extent_buffer might | |
4484 | * see the ref bit and try to clear it. If this happens | |
4485 | * free_extent_buffer might end up dropping our original | |
4486 | * ref by mistake and freeing the page before we are able | |
4487 | * to add one more ref. | |
4488 | * | |
4489 | * So bump the ref count first, then set the bit. If someone | |
4490 | * beat us to it, drop the ref we added. | |
4491 | */ | |
242e18c7 CM |
4492 | refs = atomic_read(&eb->refs); |
4493 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
4494 | return; | |
4495 | ||
594831c4 JB |
4496 | spin_lock(&eb->refs_lock); |
4497 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 4498 | atomic_inc(&eb->refs); |
594831c4 | 4499 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
4500 | } |
4501 | ||
5df4235e JB |
4502 | static void mark_extent_buffer_accessed(struct extent_buffer *eb) |
4503 | { | |
4504 | unsigned long num_pages, i; | |
4505 | ||
0b32f4bb JB |
4506 | check_buffer_tree_ref(eb); |
4507 | ||
5df4235e JB |
4508 | num_pages = num_extent_pages(eb->start, eb->len); |
4509 | for (i = 0; i < num_pages; i++) { | |
4510 | struct page *p = extent_buffer_page(eb, i); | |
4511 | mark_page_accessed(p); | |
4512 | } | |
4513 | } | |
4514 | ||
f28491e0 JB |
4515 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
4516 | u64 start) | |
452c75c3 CS |
4517 | { |
4518 | struct extent_buffer *eb; | |
4519 | ||
4520 | rcu_read_lock(); | |
f28491e0 JB |
4521 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
4522 | start >> PAGE_CACHE_SHIFT); | |
452c75c3 CS |
4523 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
4524 | rcu_read_unlock(); | |
4525 | mark_extent_buffer_accessed(eb); | |
4526 | return eb; | |
4527 | } | |
4528 | rcu_read_unlock(); | |
4529 | ||
4530 | return NULL; | |
4531 | } | |
4532 | ||
f28491e0 | 4533 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
727011e0 | 4534 | u64 start, unsigned long len) |
d1310b2e CM |
4535 | { |
4536 | unsigned long num_pages = num_extent_pages(start, len); | |
4537 | unsigned long i; | |
4538 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
4539 | struct extent_buffer *eb; | |
6af118ce | 4540 | struct extent_buffer *exists = NULL; |
d1310b2e | 4541 | struct page *p; |
f28491e0 | 4542 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
d1310b2e | 4543 | int uptodate = 1; |
19fe0a8b | 4544 | int ret; |
d1310b2e | 4545 | |
f28491e0 | 4546 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 4547 | if (eb) |
6af118ce | 4548 | return eb; |
6af118ce | 4549 | |
f28491e0 | 4550 | eb = __alloc_extent_buffer(fs_info, start, len, GFP_NOFS); |
2b114d1d | 4551 | if (!eb) |
d1310b2e CM |
4552 | return NULL; |
4553 | ||
727011e0 | 4554 | for (i = 0; i < num_pages; i++, index++) { |
a6591715 | 4555 | p = find_or_create_page(mapping, index, GFP_NOFS); |
4804b382 | 4556 | if (!p) |
6af118ce | 4557 | goto free_eb; |
4f2de97a JB |
4558 | |
4559 | spin_lock(&mapping->private_lock); | |
4560 | if (PagePrivate(p)) { | |
4561 | /* | |
4562 | * We could have already allocated an eb for this page | |
4563 | * and attached one so lets see if we can get a ref on | |
4564 | * the existing eb, and if we can we know it's good and | |
4565 | * we can just return that one, else we know we can just | |
4566 | * overwrite page->private. | |
4567 | */ | |
4568 | exists = (struct extent_buffer *)p->private; | |
4569 | if (atomic_inc_not_zero(&exists->refs)) { | |
4570 | spin_unlock(&mapping->private_lock); | |
4571 | unlock_page(p); | |
17de39ac | 4572 | page_cache_release(p); |
5df4235e | 4573 | mark_extent_buffer_accessed(exists); |
4f2de97a JB |
4574 | goto free_eb; |
4575 | } | |
4576 | ||
0b32f4bb | 4577 | /* |
4f2de97a JB |
4578 | * Do this so attach doesn't complain and we need to |
4579 | * drop the ref the old guy had. | |
4580 | */ | |
4581 | ClearPagePrivate(p); | |
0b32f4bb | 4582 | WARN_ON(PageDirty(p)); |
4f2de97a | 4583 | page_cache_release(p); |
d1310b2e | 4584 | } |
4f2de97a JB |
4585 | attach_extent_buffer_page(eb, p); |
4586 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 4587 | WARN_ON(PageDirty(p)); |
d1310b2e | 4588 | mark_page_accessed(p); |
727011e0 | 4589 | eb->pages[i] = p; |
d1310b2e CM |
4590 | if (!PageUptodate(p)) |
4591 | uptodate = 0; | |
eb14ab8e CM |
4592 | |
4593 | /* | |
4594 | * see below about how we avoid a nasty race with release page | |
4595 | * and why we unlock later | |
4596 | */ | |
d1310b2e CM |
4597 | } |
4598 | if (uptodate) | |
b4ce94de | 4599 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 4600 | again: |
19fe0a8b MX |
4601 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); |
4602 | if (ret) | |
4603 | goto free_eb; | |
4604 | ||
f28491e0 JB |
4605 | spin_lock(&fs_info->buffer_lock); |
4606 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
4607 | start >> PAGE_CACHE_SHIFT, eb); | |
4608 | spin_unlock(&fs_info->buffer_lock); | |
452c75c3 | 4609 | radix_tree_preload_end(); |
19fe0a8b | 4610 | if (ret == -EEXIST) { |
f28491e0 | 4611 | exists = find_extent_buffer(fs_info, start); |
452c75c3 CS |
4612 | if (exists) |
4613 | goto free_eb; | |
4614 | else | |
115391d2 | 4615 | goto again; |
6af118ce | 4616 | } |
6af118ce | 4617 | /* add one reference for the tree */ |
0b32f4bb | 4618 | check_buffer_tree_ref(eb); |
34b41ace | 4619 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
4620 | |
4621 | /* | |
4622 | * there is a race where release page may have | |
4623 | * tried to find this extent buffer in the radix | |
4624 | * but failed. It will tell the VM it is safe to | |
4625 | * reclaim the, and it will clear the page private bit. | |
4626 | * We must make sure to set the page private bit properly | |
4627 | * after the extent buffer is in the radix tree so | |
4628 | * it doesn't get lost | |
4629 | */ | |
727011e0 CM |
4630 | SetPageChecked(eb->pages[0]); |
4631 | for (i = 1; i < num_pages; i++) { | |
4632 | p = extent_buffer_page(eb, i); | |
727011e0 CM |
4633 | ClearPageChecked(p); |
4634 | unlock_page(p); | |
4635 | } | |
4636 | unlock_page(eb->pages[0]); | |
d1310b2e CM |
4637 | return eb; |
4638 | ||
6af118ce | 4639 | free_eb: |
727011e0 CM |
4640 | for (i = 0; i < num_pages; i++) { |
4641 | if (eb->pages[i]) | |
4642 | unlock_page(eb->pages[i]); | |
4643 | } | |
eb14ab8e | 4644 | |
17de39ac | 4645 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
897ca6e9 | 4646 | btrfs_release_extent_buffer(eb); |
6af118ce | 4647 | return exists; |
d1310b2e | 4648 | } |
d1310b2e | 4649 | |
3083ee2e JB |
4650 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
4651 | { | |
4652 | struct extent_buffer *eb = | |
4653 | container_of(head, struct extent_buffer, rcu_head); | |
4654 | ||
4655 | __free_extent_buffer(eb); | |
4656 | } | |
4657 | ||
3083ee2e | 4658 | /* Expects to have eb->eb_lock already held */ |
f7a52a40 | 4659 | static int release_extent_buffer(struct extent_buffer *eb) |
3083ee2e JB |
4660 | { |
4661 | WARN_ON(atomic_read(&eb->refs) == 0); | |
4662 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 4663 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 4664 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 4665 | |
815a51c7 | 4666 | spin_unlock(&eb->refs_lock); |
3083ee2e | 4667 | |
f28491e0 JB |
4668 | spin_lock(&fs_info->buffer_lock); |
4669 | radix_tree_delete(&fs_info->buffer_radix, | |
815a51c7 | 4670 | eb->start >> PAGE_CACHE_SHIFT); |
f28491e0 | 4671 | spin_unlock(&fs_info->buffer_lock); |
34b41ace JB |
4672 | } else { |
4673 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 4674 | } |
3083ee2e JB |
4675 | |
4676 | /* Should be safe to release our pages at this point */ | |
4677 | btrfs_release_extent_buffer_page(eb, 0); | |
3083ee2e | 4678 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 4679 | return 1; |
3083ee2e JB |
4680 | } |
4681 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
4682 | |
4683 | return 0; | |
3083ee2e JB |
4684 | } |
4685 | ||
d1310b2e CM |
4686 | void free_extent_buffer(struct extent_buffer *eb) |
4687 | { | |
242e18c7 CM |
4688 | int refs; |
4689 | int old; | |
d1310b2e CM |
4690 | if (!eb) |
4691 | return; | |
4692 | ||
242e18c7 CM |
4693 | while (1) { |
4694 | refs = atomic_read(&eb->refs); | |
4695 | if (refs <= 3) | |
4696 | break; | |
4697 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
4698 | if (old == refs) | |
4699 | return; | |
4700 | } | |
4701 | ||
3083ee2e | 4702 | spin_lock(&eb->refs_lock); |
815a51c7 JS |
4703 | if (atomic_read(&eb->refs) == 2 && |
4704 | test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) | |
4705 | atomic_dec(&eb->refs); | |
4706 | ||
3083ee2e JB |
4707 | if (atomic_read(&eb->refs) == 2 && |
4708 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 4709 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
4710 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
4711 | atomic_dec(&eb->refs); | |
4712 | ||
4713 | /* | |
4714 | * I know this is terrible, but it's temporary until we stop tracking | |
4715 | * the uptodate bits and such for the extent buffers. | |
4716 | */ | |
f7a52a40 | 4717 | release_extent_buffer(eb); |
3083ee2e JB |
4718 | } |
4719 | ||
4720 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
4721 | { | |
4722 | if (!eb) | |
d1310b2e CM |
4723 | return; |
4724 | ||
3083ee2e JB |
4725 | spin_lock(&eb->refs_lock); |
4726 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
4727 | ||
0b32f4bb | 4728 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
4729 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
4730 | atomic_dec(&eb->refs); | |
f7a52a40 | 4731 | release_extent_buffer(eb); |
d1310b2e | 4732 | } |
d1310b2e | 4733 | |
1d4284bd | 4734 | void clear_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 4735 | { |
d1310b2e CM |
4736 | unsigned long i; |
4737 | unsigned long num_pages; | |
4738 | struct page *page; | |
4739 | ||
d1310b2e CM |
4740 | num_pages = num_extent_pages(eb->start, eb->len); |
4741 | ||
4742 | for (i = 0; i < num_pages; i++) { | |
4743 | page = extent_buffer_page(eb, i); | |
b9473439 | 4744 | if (!PageDirty(page)) |
d2c3f4f6 CM |
4745 | continue; |
4746 | ||
a61e6f29 | 4747 | lock_page(page); |
eb14ab8e CM |
4748 | WARN_ON(!PagePrivate(page)); |
4749 | ||
d1310b2e | 4750 | clear_page_dirty_for_io(page); |
0ee0fda0 | 4751 | spin_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
4752 | if (!PageDirty(page)) { |
4753 | radix_tree_tag_clear(&page->mapping->page_tree, | |
4754 | page_index(page), | |
4755 | PAGECACHE_TAG_DIRTY); | |
4756 | } | |
0ee0fda0 | 4757 | spin_unlock_irq(&page->mapping->tree_lock); |
bf0da8c1 | 4758 | ClearPageError(page); |
a61e6f29 | 4759 | unlock_page(page); |
d1310b2e | 4760 | } |
0b32f4bb | 4761 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 4762 | } |
d1310b2e | 4763 | |
0b32f4bb | 4764 | int set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e CM |
4765 | { |
4766 | unsigned long i; | |
4767 | unsigned long num_pages; | |
b9473439 | 4768 | int was_dirty = 0; |
d1310b2e | 4769 | |
0b32f4bb JB |
4770 | check_buffer_tree_ref(eb); |
4771 | ||
b9473439 | 4772 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 4773 | |
d1310b2e | 4774 | num_pages = num_extent_pages(eb->start, eb->len); |
3083ee2e | 4775 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
4776 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
4777 | ||
b9473439 | 4778 | for (i = 0; i < num_pages; i++) |
0b32f4bb | 4779 | set_page_dirty(extent_buffer_page(eb, i)); |
b9473439 | 4780 | return was_dirty; |
d1310b2e | 4781 | } |
d1310b2e | 4782 | |
0b32f4bb | 4783 | int clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 CM |
4784 | { |
4785 | unsigned long i; | |
4786 | struct page *page; | |
4787 | unsigned long num_pages; | |
4788 | ||
b4ce94de | 4789 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
0b32f4bb | 4790 | num_pages = num_extent_pages(eb->start, eb->len); |
1259ab75 CM |
4791 | for (i = 0; i < num_pages; i++) { |
4792 | page = extent_buffer_page(eb, i); | |
33958dc6 CM |
4793 | if (page) |
4794 | ClearPageUptodate(page); | |
1259ab75 CM |
4795 | } |
4796 | return 0; | |
4797 | } | |
4798 | ||
0b32f4bb | 4799 | int set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e CM |
4800 | { |
4801 | unsigned long i; | |
4802 | struct page *page; | |
4803 | unsigned long num_pages; | |
4804 | ||
0b32f4bb | 4805 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 4806 | num_pages = num_extent_pages(eb->start, eb->len); |
d1310b2e CM |
4807 | for (i = 0; i < num_pages; i++) { |
4808 | page = extent_buffer_page(eb, i); | |
d1310b2e CM |
4809 | SetPageUptodate(page); |
4810 | } | |
4811 | return 0; | |
4812 | } | |
d1310b2e | 4813 | |
0b32f4bb | 4814 | int extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 4815 | { |
0b32f4bb | 4816 | return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 4817 | } |
d1310b2e CM |
4818 | |
4819 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
bb82ab88 | 4820 | struct extent_buffer *eb, u64 start, int wait, |
f188591e | 4821 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
4822 | { |
4823 | unsigned long i; | |
4824 | unsigned long start_i; | |
4825 | struct page *page; | |
4826 | int err; | |
4827 | int ret = 0; | |
ce9adaa5 CM |
4828 | int locked_pages = 0; |
4829 | int all_uptodate = 1; | |
d1310b2e | 4830 | unsigned long num_pages; |
727011e0 | 4831 | unsigned long num_reads = 0; |
a86c12c7 | 4832 | struct bio *bio = NULL; |
c8b97818 | 4833 | unsigned long bio_flags = 0; |
a86c12c7 | 4834 | |
b4ce94de | 4835 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
4836 | return 0; |
4837 | ||
d1310b2e CM |
4838 | if (start) { |
4839 | WARN_ON(start < eb->start); | |
4840 | start_i = (start >> PAGE_CACHE_SHIFT) - | |
4841 | (eb->start >> PAGE_CACHE_SHIFT); | |
4842 | } else { | |
4843 | start_i = 0; | |
4844 | } | |
4845 | ||
4846 | num_pages = num_extent_pages(eb->start, eb->len); | |
4847 | for (i = start_i; i < num_pages; i++) { | |
4848 | page = extent_buffer_page(eb, i); | |
bb82ab88 | 4849 | if (wait == WAIT_NONE) { |
2db04966 | 4850 | if (!trylock_page(page)) |
ce9adaa5 | 4851 | goto unlock_exit; |
d1310b2e CM |
4852 | } else { |
4853 | lock_page(page); | |
4854 | } | |
ce9adaa5 | 4855 | locked_pages++; |
727011e0 CM |
4856 | if (!PageUptodate(page)) { |
4857 | num_reads++; | |
ce9adaa5 | 4858 | all_uptodate = 0; |
727011e0 | 4859 | } |
ce9adaa5 CM |
4860 | } |
4861 | if (all_uptodate) { | |
4862 | if (start_i == 0) | |
b4ce94de | 4863 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
4864 | goto unlock_exit; |
4865 | } | |
4866 | ||
ea466794 | 4867 | clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags); |
5cf1ab56 | 4868 | eb->read_mirror = 0; |
0b32f4bb | 4869 | atomic_set(&eb->io_pages, num_reads); |
ce9adaa5 CM |
4870 | for (i = start_i; i < num_pages; i++) { |
4871 | page = extent_buffer_page(eb, i); | |
ce9adaa5 | 4872 | if (!PageUptodate(page)) { |
f188591e | 4873 | ClearPageError(page); |
a86c12c7 | 4874 | err = __extent_read_full_page(tree, page, |
f188591e | 4875 | get_extent, &bio, |
d4c7ca86 JB |
4876 | mirror_num, &bio_flags, |
4877 | READ | REQ_META); | |
d397712b | 4878 | if (err) |
d1310b2e | 4879 | ret = err; |
d1310b2e CM |
4880 | } else { |
4881 | unlock_page(page); | |
4882 | } | |
4883 | } | |
4884 | ||
355808c2 | 4885 | if (bio) { |
d4c7ca86 JB |
4886 | err = submit_one_bio(READ | REQ_META, bio, mirror_num, |
4887 | bio_flags); | |
79787eaa JM |
4888 | if (err) |
4889 | return err; | |
355808c2 | 4890 | } |
a86c12c7 | 4891 | |
bb82ab88 | 4892 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 4893 | return ret; |
d397712b | 4894 | |
d1310b2e CM |
4895 | for (i = start_i; i < num_pages; i++) { |
4896 | page = extent_buffer_page(eb, i); | |
4897 | wait_on_page_locked(page); | |
d397712b | 4898 | if (!PageUptodate(page)) |
d1310b2e | 4899 | ret = -EIO; |
d1310b2e | 4900 | } |
d397712b | 4901 | |
d1310b2e | 4902 | return ret; |
ce9adaa5 CM |
4903 | |
4904 | unlock_exit: | |
4905 | i = start_i; | |
d397712b | 4906 | while (locked_pages > 0) { |
ce9adaa5 CM |
4907 | page = extent_buffer_page(eb, i); |
4908 | i++; | |
4909 | unlock_page(page); | |
4910 | locked_pages--; | |
4911 | } | |
4912 | return ret; | |
d1310b2e | 4913 | } |
d1310b2e CM |
4914 | |
4915 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
4916 | unsigned long start, | |
4917 | unsigned long len) | |
4918 | { | |
4919 | size_t cur; | |
4920 | size_t offset; | |
4921 | struct page *page; | |
4922 | char *kaddr; | |
4923 | char *dst = (char *)dstv; | |
4924 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4925 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
4926 | |
4927 | WARN_ON(start > eb->len); | |
4928 | WARN_ON(start + len > eb->start + eb->len); | |
4929 | ||
778746b5 | 4930 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 4931 | |
d397712b | 4932 | while (len > 0) { |
d1310b2e | 4933 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
4934 | |
4935 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
a6591715 | 4936 | kaddr = page_address(page); |
d1310b2e | 4937 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
4938 | |
4939 | dst += cur; | |
4940 | len -= cur; | |
4941 | offset = 0; | |
4942 | i++; | |
4943 | } | |
4944 | } | |
d1310b2e CM |
4945 | |
4946 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
a6591715 | 4947 | unsigned long min_len, char **map, |
d1310b2e | 4948 | unsigned long *map_start, |
a6591715 | 4949 | unsigned long *map_len) |
d1310b2e CM |
4950 | { |
4951 | size_t offset = start & (PAGE_CACHE_SIZE - 1); | |
4952 | char *kaddr; | |
4953 | struct page *p; | |
4954 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4955 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4956 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
4957 | PAGE_CACHE_SHIFT; | |
4958 | ||
4959 | if (i != end_i) | |
4960 | return -EINVAL; | |
4961 | ||
4962 | if (i == 0) { | |
4963 | offset = start_offset; | |
4964 | *map_start = 0; | |
4965 | } else { | |
4966 | offset = 0; | |
4967 | *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; | |
4968 | } | |
d397712b | 4969 | |
d1310b2e | 4970 | if (start + min_len > eb->len) { |
31b1a2bd | 4971 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, " |
c1c9ff7c GU |
4972 | "wanted %lu %lu\n", |
4973 | eb->start, eb->len, start, min_len); | |
85026533 | 4974 | return -EINVAL; |
d1310b2e CM |
4975 | } |
4976 | ||
4977 | p = extent_buffer_page(eb, i); | |
a6591715 | 4978 | kaddr = page_address(p); |
d1310b2e CM |
4979 | *map = kaddr + offset; |
4980 | *map_len = PAGE_CACHE_SIZE - offset; | |
4981 | return 0; | |
4982 | } | |
d1310b2e | 4983 | |
d1310b2e CM |
4984 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, |
4985 | unsigned long start, | |
4986 | unsigned long len) | |
4987 | { | |
4988 | size_t cur; | |
4989 | size_t offset; | |
4990 | struct page *page; | |
4991 | char *kaddr; | |
4992 | char *ptr = (char *)ptrv; | |
4993 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4994 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4995 | int ret = 0; | |
4996 | ||
4997 | WARN_ON(start > eb->len); | |
4998 | WARN_ON(start + len > eb->start + eb->len); | |
4999 | ||
778746b5 | 5000 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5001 | |
d397712b | 5002 | while (len > 0) { |
d1310b2e | 5003 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
5004 | |
5005 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
5006 | ||
a6591715 | 5007 | kaddr = page_address(page); |
d1310b2e | 5008 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
5009 | if (ret) |
5010 | break; | |
5011 | ||
5012 | ptr += cur; | |
5013 | len -= cur; | |
5014 | offset = 0; | |
5015 | i++; | |
5016 | } | |
5017 | return ret; | |
5018 | } | |
d1310b2e CM |
5019 | |
5020 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
5021 | unsigned long start, unsigned long len) | |
5022 | { | |
5023 | size_t cur; | |
5024 | size_t offset; | |
5025 | struct page *page; | |
5026 | char *kaddr; | |
5027 | char *src = (char *)srcv; | |
5028 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5029 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5030 | ||
5031 | WARN_ON(start > eb->len); | |
5032 | WARN_ON(start + len > eb->start + eb->len); | |
5033 | ||
778746b5 | 5034 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5035 | |
d397712b | 5036 | while (len > 0) { |
d1310b2e CM |
5037 | page = extent_buffer_page(eb, i); |
5038 | WARN_ON(!PageUptodate(page)); | |
5039 | ||
5040 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 5041 | kaddr = page_address(page); |
d1310b2e | 5042 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
5043 | |
5044 | src += cur; | |
5045 | len -= cur; | |
5046 | offset = 0; | |
5047 | i++; | |
5048 | } | |
5049 | } | |
d1310b2e CM |
5050 | |
5051 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
5052 | unsigned long start, unsigned long len) | |
5053 | { | |
5054 | size_t cur; | |
5055 | size_t offset; | |
5056 | struct page *page; | |
5057 | char *kaddr; | |
5058 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5059 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5060 | ||
5061 | WARN_ON(start > eb->len); | |
5062 | WARN_ON(start + len > eb->start + eb->len); | |
5063 | ||
778746b5 | 5064 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5065 | |
d397712b | 5066 | while (len > 0) { |
d1310b2e CM |
5067 | page = extent_buffer_page(eb, i); |
5068 | WARN_ON(!PageUptodate(page)); | |
5069 | ||
5070 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 5071 | kaddr = page_address(page); |
d1310b2e | 5072 | memset(kaddr + offset, c, cur); |
d1310b2e CM |
5073 | |
5074 | len -= cur; | |
5075 | offset = 0; | |
5076 | i++; | |
5077 | } | |
5078 | } | |
d1310b2e CM |
5079 | |
5080 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
5081 | unsigned long dst_offset, unsigned long src_offset, | |
5082 | unsigned long len) | |
5083 | { | |
5084 | u64 dst_len = dst->len; | |
5085 | size_t cur; | |
5086 | size_t offset; | |
5087 | struct page *page; | |
5088 | char *kaddr; | |
5089 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5090 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
5091 | ||
5092 | WARN_ON(src->len != dst_len); | |
5093 | ||
5094 | offset = (start_offset + dst_offset) & | |
778746b5 | 5095 | (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5096 | |
d397712b | 5097 | while (len > 0) { |
d1310b2e CM |
5098 | page = extent_buffer_page(dst, i); |
5099 | WARN_ON(!PageUptodate(page)); | |
5100 | ||
5101 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
5102 | ||
a6591715 | 5103 | kaddr = page_address(page); |
d1310b2e | 5104 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
5105 | |
5106 | src_offset += cur; | |
5107 | len -= cur; | |
5108 | offset = 0; | |
5109 | i++; | |
5110 | } | |
5111 | } | |
d1310b2e | 5112 | |
3387206f ST |
5113 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
5114 | { | |
5115 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
5116 | return distance < len; | |
5117 | } | |
5118 | ||
d1310b2e CM |
5119 | static void copy_pages(struct page *dst_page, struct page *src_page, |
5120 | unsigned long dst_off, unsigned long src_off, | |
5121 | unsigned long len) | |
5122 | { | |
a6591715 | 5123 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 5124 | char *src_kaddr; |
727011e0 | 5125 | int must_memmove = 0; |
d1310b2e | 5126 | |
3387206f | 5127 | if (dst_page != src_page) { |
a6591715 | 5128 | src_kaddr = page_address(src_page); |
3387206f | 5129 | } else { |
d1310b2e | 5130 | src_kaddr = dst_kaddr; |
727011e0 CM |
5131 | if (areas_overlap(src_off, dst_off, len)) |
5132 | must_memmove = 1; | |
3387206f | 5133 | } |
d1310b2e | 5134 | |
727011e0 CM |
5135 | if (must_memmove) |
5136 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
5137 | else | |
5138 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
5139 | } |
5140 | ||
5141 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5142 | unsigned long src_offset, unsigned long len) | |
5143 | { | |
5144 | size_t cur; | |
5145 | size_t dst_off_in_page; | |
5146 | size_t src_off_in_page; | |
5147 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5148 | unsigned long dst_i; | |
5149 | unsigned long src_i; | |
5150 | ||
5151 | if (src_offset + len > dst->len) { | |
d397712b CM |
5152 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
5153 | "len %lu dst len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
5154 | BUG_ON(1); |
5155 | } | |
5156 | if (dst_offset + len > dst->len) { | |
d397712b CM |
5157 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
5158 | "len %lu dst len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
5159 | BUG_ON(1); |
5160 | } | |
5161 | ||
d397712b | 5162 | while (len > 0) { |
d1310b2e | 5163 | dst_off_in_page = (start_offset + dst_offset) & |
778746b5 | 5164 | (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5165 | src_off_in_page = (start_offset + src_offset) & |
778746b5 | 5166 | (PAGE_CACHE_SIZE - 1); |
d1310b2e CM |
5167 | |
5168 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
5169 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
5170 | ||
5171 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
5172 | src_off_in_page)); | |
5173 | cur = min_t(unsigned long, cur, | |
5174 | (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); | |
5175 | ||
5176 | copy_pages(extent_buffer_page(dst, dst_i), | |
5177 | extent_buffer_page(dst, src_i), | |
5178 | dst_off_in_page, src_off_in_page, cur); | |
5179 | ||
5180 | src_offset += cur; | |
5181 | dst_offset += cur; | |
5182 | len -= cur; | |
5183 | } | |
5184 | } | |
d1310b2e CM |
5185 | |
5186 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5187 | unsigned long src_offset, unsigned long len) | |
5188 | { | |
5189 | size_t cur; | |
5190 | size_t dst_off_in_page; | |
5191 | size_t src_off_in_page; | |
5192 | unsigned long dst_end = dst_offset + len - 1; | |
5193 | unsigned long src_end = src_offset + len - 1; | |
5194 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5195 | unsigned long dst_i; | |
5196 | unsigned long src_i; | |
5197 | ||
5198 | if (src_offset + len > dst->len) { | |
d397712b CM |
5199 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
5200 | "len %lu len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
5201 | BUG_ON(1); |
5202 | } | |
5203 | if (dst_offset + len > dst->len) { | |
d397712b CM |
5204 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
5205 | "len %lu len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
5206 | BUG_ON(1); |
5207 | } | |
727011e0 | 5208 | if (dst_offset < src_offset) { |
d1310b2e CM |
5209 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
5210 | return; | |
5211 | } | |
d397712b | 5212 | while (len > 0) { |
d1310b2e CM |
5213 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; |
5214 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
5215 | ||
5216 | dst_off_in_page = (start_offset + dst_end) & | |
778746b5 | 5217 | (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5218 | src_off_in_page = (start_offset + src_end) & |
778746b5 | 5219 | (PAGE_CACHE_SIZE - 1); |
d1310b2e CM |
5220 | |
5221 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
5222 | cur = min(cur, dst_off_in_page + 1); | |
1877e1a7 | 5223 | copy_pages(extent_buffer_page(dst, dst_i), |
d1310b2e CM |
5224 | extent_buffer_page(dst, src_i), |
5225 | dst_off_in_page - cur + 1, | |
5226 | src_off_in_page - cur + 1, cur); | |
5227 | ||
5228 | dst_end -= cur; | |
5229 | src_end -= cur; | |
5230 | len -= cur; | |
5231 | } | |
5232 | } | |
6af118ce | 5233 | |
f7a52a40 | 5234 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 5235 | { |
6af118ce | 5236 | struct extent_buffer *eb; |
6af118ce | 5237 | |
3083ee2e JB |
5238 | /* |
5239 | * We need to make sure noboody is attaching this page to an eb right | |
5240 | * now. | |
5241 | */ | |
5242 | spin_lock(&page->mapping->private_lock); | |
5243 | if (!PagePrivate(page)) { | |
5244 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 5245 | return 1; |
45f49bce | 5246 | } |
6af118ce | 5247 | |
3083ee2e JB |
5248 | eb = (struct extent_buffer *)page->private; |
5249 | BUG_ON(!eb); | |
19fe0a8b MX |
5250 | |
5251 | /* | |
3083ee2e JB |
5252 | * This is a little awful but should be ok, we need to make sure that |
5253 | * the eb doesn't disappear out from under us while we're looking at | |
5254 | * this page. | |
19fe0a8b | 5255 | */ |
3083ee2e | 5256 | spin_lock(&eb->refs_lock); |
0b32f4bb | 5257 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
5258 | spin_unlock(&eb->refs_lock); |
5259 | spin_unlock(&page->mapping->private_lock); | |
5260 | return 0; | |
b9473439 | 5261 | } |
3083ee2e | 5262 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 5263 | |
19fe0a8b | 5264 | /* |
3083ee2e JB |
5265 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
5266 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 5267 | */ |
3083ee2e JB |
5268 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
5269 | spin_unlock(&eb->refs_lock); | |
5270 | return 0; | |
b9473439 | 5271 | } |
19fe0a8b | 5272 | |
f7a52a40 | 5273 | return release_extent_buffer(eb); |
6af118ce | 5274 | } |