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 | |
27a3507d FM |
28 | static inline bool extent_state_in_tree(const struct extent_state *state) |
29 | { | |
30 | return !RB_EMPTY_NODE(&state->rb_node); | |
31 | } | |
32 | ||
6d49ba1b | 33 | #ifdef CONFIG_BTRFS_DEBUG |
d1310b2e CM |
34 | static LIST_HEAD(buffers); |
35 | static LIST_HEAD(states); | |
4bef0848 | 36 | |
d397712b | 37 | static DEFINE_SPINLOCK(leak_lock); |
6d49ba1b ES |
38 | |
39 | static inline | |
40 | void btrfs_leak_debug_add(struct list_head *new, struct list_head *head) | |
41 | { | |
42 | unsigned long flags; | |
43 | ||
44 | spin_lock_irqsave(&leak_lock, flags); | |
45 | list_add(new, head); | |
46 | spin_unlock_irqrestore(&leak_lock, flags); | |
47 | } | |
48 | ||
49 | static inline | |
50 | void btrfs_leak_debug_del(struct list_head *entry) | |
51 | { | |
52 | unsigned long flags; | |
53 | ||
54 | spin_lock_irqsave(&leak_lock, flags); | |
55 | list_del(entry); | |
56 | spin_unlock_irqrestore(&leak_lock, flags); | |
57 | } | |
58 | ||
59 | static inline | |
60 | void btrfs_leak_debug_check(void) | |
61 | { | |
62 | struct extent_state *state; | |
63 | struct extent_buffer *eb; | |
64 | ||
65 | while (!list_empty(&states)) { | |
66 | state = list_entry(states.next, struct extent_state, leak_list); | |
9ee49a04 | 67 | pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n", |
27a3507d FM |
68 | state->start, state->end, state->state, |
69 | extent_state_in_tree(state), | |
c1c9ff7c | 70 | atomic_read(&state->refs)); |
6d49ba1b ES |
71 | list_del(&state->leak_list); |
72 | kmem_cache_free(extent_state_cache, state); | |
73 | } | |
74 | ||
75 | while (!list_empty(&buffers)) { | |
76 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
efe120a0 | 77 | printk(KERN_ERR "BTRFS: buffer leak start %llu len %lu " |
c1c9ff7c GU |
78 | "refs %d\n", |
79 | eb->start, eb->len, atomic_read(&eb->refs)); | |
6d49ba1b ES |
80 | list_del(&eb->leak_list); |
81 | kmem_cache_free(extent_buffer_cache, eb); | |
82 | } | |
83 | } | |
8d599ae1 | 84 | |
a5dee37d JB |
85 | #define btrfs_debug_check_extent_io_range(tree, start, end) \ |
86 | __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end)) | |
8d599ae1 | 87 | static inline void __btrfs_debug_check_extent_io_range(const char *caller, |
a5dee37d | 88 | struct extent_io_tree *tree, u64 start, u64 end) |
8d599ae1 | 89 | { |
a5dee37d JB |
90 | struct inode *inode; |
91 | u64 isize; | |
8d599ae1 | 92 | |
a5dee37d JB |
93 | if (!tree->mapping) |
94 | return; | |
8d599ae1 | 95 | |
a5dee37d JB |
96 | inode = tree->mapping->host; |
97 | isize = i_size_read(inode); | |
8d599ae1 | 98 | if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) { |
94647322 DS |
99 | btrfs_debug_rl(BTRFS_I(inode)->root->fs_info, |
100 | "%s: ino %llu isize %llu odd range [%llu,%llu]", | |
c1c9ff7c | 101 | caller, btrfs_ino(inode), isize, start, end); |
8d599ae1 DS |
102 | } |
103 | } | |
6d49ba1b ES |
104 | #else |
105 | #define btrfs_leak_debug_add(new, head) do {} while (0) | |
106 | #define btrfs_leak_debug_del(entry) do {} while (0) | |
107 | #define btrfs_leak_debug_check() do {} while (0) | |
8d599ae1 | 108 | #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0) |
4bef0848 | 109 | #endif |
d1310b2e | 110 | |
d1310b2e CM |
111 | #define BUFFER_LRU_MAX 64 |
112 | ||
113 | struct tree_entry { | |
114 | u64 start; | |
115 | u64 end; | |
d1310b2e CM |
116 | struct rb_node rb_node; |
117 | }; | |
118 | ||
119 | struct extent_page_data { | |
120 | struct bio *bio; | |
121 | struct extent_io_tree *tree; | |
122 | get_extent_t *get_extent; | |
de0022b9 | 123 | unsigned long bio_flags; |
771ed689 CM |
124 | |
125 | /* tells writepage not to lock the state bits for this range | |
126 | * it still does the unlocking | |
127 | */ | |
ffbd517d CM |
128 | unsigned int extent_locked:1; |
129 | ||
130 | /* tells the submit_bio code to use a WRITE_SYNC */ | |
131 | unsigned int sync_io:1; | |
d1310b2e CM |
132 | }; |
133 | ||
d38ed27f QW |
134 | static void add_extent_changeset(struct extent_state *state, unsigned bits, |
135 | struct extent_changeset *changeset, | |
136 | int set) | |
137 | { | |
138 | int ret; | |
139 | ||
140 | if (!changeset) | |
141 | return; | |
142 | if (set && (state->state & bits) == bits) | |
143 | return; | |
fefdc557 QW |
144 | if (!set && (state->state & bits) == 0) |
145 | return; | |
d38ed27f QW |
146 | changeset->bytes_changed += state->end - state->start + 1; |
147 | ret = ulist_add(changeset->range_changed, state->start, state->end, | |
148 | GFP_ATOMIC); | |
149 | /* ENOMEM */ | |
150 | BUG_ON(ret < 0); | |
151 | } | |
152 | ||
0b32f4bb | 153 | static noinline void flush_write_bio(void *data); |
c2d904e0 JM |
154 | static inline struct btrfs_fs_info * |
155 | tree_fs_info(struct extent_io_tree *tree) | |
156 | { | |
a5dee37d JB |
157 | if (!tree->mapping) |
158 | return NULL; | |
c2d904e0 JM |
159 | return btrfs_sb(tree->mapping->host->i_sb); |
160 | } | |
0b32f4bb | 161 | |
d1310b2e CM |
162 | int __init extent_io_init(void) |
163 | { | |
837e1972 | 164 | extent_state_cache = kmem_cache_create("btrfs_extent_state", |
9601e3f6 CH |
165 | sizeof(struct extent_state), 0, |
166 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
167 | if (!extent_state_cache) |
168 | return -ENOMEM; | |
169 | ||
837e1972 | 170 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 CH |
171 | sizeof(struct extent_buffer), 0, |
172 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
173 | if (!extent_buffer_cache) |
174 | goto free_state_cache; | |
9be3395b CM |
175 | |
176 | btrfs_bioset = bioset_create(BIO_POOL_SIZE, | |
177 | offsetof(struct btrfs_io_bio, bio)); | |
178 | if (!btrfs_bioset) | |
179 | goto free_buffer_cache; | |
b208c2f7 DW |
180 | |
181 | if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE)) | |
182 | goto free_bioset; | |
183 | ||
d1310b2e CM |
184 | return 0; |
185 | ||
b208c2f7 DW |
186 | free_bioset: |
187 | bioset_free(btrfs_bioset); | |
188 | btrfs_bioset = NULL; | |
189 | ||
9be3395b CM |
190 | free_buffer_cache: |
191 | kmem_cache_destroy(extent_buffer_cache); | |
192 | extent_buffer_cache = NULL; | |
193 | ||
d1310b2e CM |
194 | free_state_cache: |
195 | kmem_cache_destroy(extent_state_cache); | |
9be3395b | 196 | extent_state_cache = NULL; |
d1310b2e CM |
197 | return -ENOMEM; |
198 | } | |
199 | ||
200 | void extent_io_exit(void) | |
201 | { | |
6d49ba1b | 202 | btrfs_leak_debug_check(); |
8c0a8537 KS |
203 | |
204 | /* | |
205 | * Make sure all delayed rcu free are flushed before we | |
206 | * destroy caches. | |
207 | */ | |
208 | rcu_barrier(); | |
d1310b2e CM |
209 | if (extent_state_cache) |
210 | kmem_cache_destroy(extent_state_cache); | |
211 | if (extent_buffer_cache) | |
212 | kmem_cache_destroy(extent_buffer_cache); | |
9be3395b CM |
213 | if (btrfs_bioset) |
214 | bioset_free(btrfs_bioset); | |
d1310b2e CM |
215 | } |
216 | ||
217 | void extent_io_tree_init(struct extent_io_tree *tree, | |
f993c883 | 218 | struct address_space *mapping) |
d1310b2e | 219 | { |
6bef4d31 | 220 | tree->state = RB_ROOT; |
d1310b2e CM |
221 | tree->ops = NULL; |
222 | tree->dirty_bytes = 0; | |
70dec807 | 223 | spin_lock_init(&tree->lock); |
d1310b2e | 224 | tree->mapping = mapping; |
d1310b2e | 225 | } |
d1310b2e | 226 | |
b2950863 | 227 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
228 | { |
229 | struct extent_state *state; | |
d1310b2e CM |
230 | |
231 | state = kmem_cache_alloc(extent_state_cache, mask); | |
2b114d1d | 232 | if (!state) |
d1310b2e CM |
233 | return state; |
234 | state->state = 0; | |
d1310b2e | 235 | state->private = 0; |
27a3507d | 236 | RB_CLEAR_NODE(&state->rb_node); |
6d49ba1b | 237 | btrfs_leak_debug_add(&state->leak_list, &states); |
d1310b2e CM |
238 | atomic_set(&state->refs, 1); |
239 | init_waitqueue_head(&state->wq); | |
143bede5 | 240 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
241 | return state; |
242 | } | |
d1310b2e | 243 | |
4845e44f | 244 | void free_extent_state(struct extent_state *state) |
d1310b2e | 245 | { |
d1310b2e CM |
246 | if (!state) |
247 | return; | |
248 | if (atomic_dec_and_test(&state->refs)) { | |
27a3507d | 249 | WARN_ON(extent_state_in_tree(state)); |
6d49ba1b | 250 | btrfs_leak_debug_del(&state->leak_list); |
143bede5 | 251 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
252 | kmem_cache_free(extent_state_cache, state); |
253 | } | |
254 | } | |
d1310b2e | 255 | |
f2071b21 FM |
256 | static struct rb_node *tree_insert(struct rb_root *root, |
257 | struct rb_node *search_start, | |
258 | u64 offset, | |
12cfbad9 FDBM |
259 | struct rb_node *node, |
260 | struct rb_node ***p_in, | |
261 | struct rb_node **parent_in) | |
d1310b2e | 262 | { |
f2071b21 | 263 | struct rb_node **p; |
d397712b | 264 | struct rb_node *parent = NULL; |
d1310b2e CM |
265 | struct tree_entry *entry; |
266 | ||
12cfbad9 FDBM |
267 | if (p_in && parent_in) { |
268 | p = *p_in; | |
269 | parent = *parent_in; | |
270 | goto do_insert; | |
271 | } | |
272 | ||
f2071b21 | 273 | p = search_start ? &search_start : &root->rb_node; |
d397712b | 274 | while (*p) { |
d1310b2e CM |
275 | parent = *p; |
276 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
277 | ||
278 | if (offset < entry->start) | |
279 | p = &(*p)->rb_left; | |
280 | else if (offset > entry->end) | |
281 | p = &(*p)->rb_right; | |
282 | else | |
283 | return parent; | |
284 | } | |
285 | ||
12cfbad9 | 286 | do_insert: |
d1310b2e CM |
287 | rb_link_node(node, parent, p); |
288 | rb_insert_color(node, root); | |
289 | return NULL; | |
290 | } | |
291 | ||
80ea96b1 | 292 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
12cfbad9 FDBM |
293 | struct rb_node **prev_ret, |
294 | struct rb_node **next_ret, | |
295 | struct rb_node ***p_ret, | |
296 | struct rb_node **parent_ret) | |
d1310b2e | 297 | { |
80ea96b1 | 298 | struct rb_root *root = &tree->state; |
12cfbad9 | 299 | struct rb_node **n = &root->rb_node; |
d1310b2e CM |
300 | struct rb_node *prev = NULL; |
301 | struct rb_node *orig_prev = NULL; | |
302 | struct tree_entry *entry; | |
303 | struct tree_entry *prev_entry = NULL; | |
304 | ||
12cfbad9 FDBM |
305 | while (*n) { |
306 | prev = *n; | |
307 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
d1310b2e CM |
308 | prev_entry = entry; |
309 | ||
310 | if (offset < entry->start) | |
12cfbad9 | 311 | n = &(*n)->rb_left; |
d1310b2e | 312 | else if (offset > entry->end) |
12cfbad9 | 313 | n = &(*n)->rb_right; |
d397712b | 314 | else |
12cfbad9 | 315 | return *n; |
d1310b2e CM |
316 | } |
317 | ||
12cfbad9 FDBM |
318 | if (p_ret) |
319 | *p_ret = n; | |
320 | if (parent_ret) | |
321 | *parent_ret = prev; | |
322 | ||
d1310b2e CM |
323 | if (prev_ret) { |
324 | orig_prev = prev; | |
d397712b | 325 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
326 | prev = rb_next(prev); |
327 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
328 | } | |
329 | *prev_ret = prev; | |
330 | prev = orig_prev; | |
331 | } | |
332 | ||
333 | if (next_ret) { | |
334 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 335 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
336 | prev = rb_prev(prev); |
337 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
338 | } | |
339 | *next_ret = prev; | |
340 | } | |
341 | return NULL; | |
342 | } | |
343 | ||
12cfbad9 FDBM |
344 | static inline struct rb_node * |
345 | tree_search_for_insert(struct extent_io_tree *tree, | |
346 | u64 offset, | |
347 | struct rb_node ***p_ret, | |
348 | struct rb_node **parent_ret) | |
d1310b2e | 349 | { |
70dec807 | 350 | struct rb_node *prev = NULL; |
d1310b2e | 351 | struct rb_node *ret; |
70dec807 | 352 | |
12cfbad9 | 353 | ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret); |
d397712b | 354 | if (!ret) |
d1310b2e CM |
355 | return prev; |
356 | return ret; | |
357 | } | |
358 | ||
12cfbad9 FDBM |
359 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
360 | u64 offset) | |
361 | { | |
362 | return tree_search_for_insert(tree, offset, NULL, NULL); | |
363 | } | |
364 | ||
9ed74f2d JB |
365 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, |
366 | struct extent_state *other) | |
367 | { | |
368 | if (tree->ops && tree->ops->merge_extent_hook) | |
369 | tree->ops->merge_extent_hook(tree->mapping->host, new, | |
370 | other); | |
371 | } | |
372 | ||
d1310b2e CM |
373 | /* |
374 | * utility function to look for merge candidates inside a given range. | |
375 | * Any extents with matching state are merged together into a single | |
376 | * extent in the tree. Extents with EXTENT_IO in their state field | |
377 | * are not merged because the end_io handlers need to be able to do | |
378 | * operations on them without sleeping (or doing allocations/splits). | |
379 | * | |
380 | * This should be called with the tree lock held. | |
381 | */ | |
1bf85046 JM |
382 | static void merge_state(struct extent_io_tree *tree, |
383 | struct extent_state *state) | |
d1310b2e CM |
384 | { |
385 | struct extent_state *other; | |
386 | struct rb_node *other_node; | |
387 | ||
5b21f2ed | 388 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
1bf85046 | 389 | return; |
d1310b2e CM |
390 | |
391 | other_node = rb_prev(&state->rb_node); | |
392 | if (other_node) { | |
393 | other = rb_entry(other_node, struct extent_state, rb_node); | |
394 | if (other->end == state->start - 1 && | |
395 | other->state == state->state) { | |
9ed74f2d | 396 | merge_cb(tree, state, other); |
d1310b2e | 397 | state->start = other->start; |
d1310b2e | 398 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 399 | RB_CLEAR_NODE(&other->rb_node); |
d1310b2e CM |
400 | free_extent_state(other); |
401 | } | |
402 | } | |
403 | other_node = rb_next(&state->rb_node); | |
404 | if (other_node) { | |
405 | other = rb_entry(other_node, struct extent_state, rb_node); | |
406 | if (other->start == state->end + 1 && | |
407 | other->state == state->state) { | |
9ed74f2d | 408 | merge_cb(tree, state, other); |
df98b6e2 | 409 | state->end = other->end; |
df98b6e2 | 410 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 411 | RB_CLEAR_NODE(&other->rb_node); |
df98b6e2 | 412 | free_extent_state(other); |
d1310b2e CM |
413 | } |
414 | } | |
d1310b2e CM |
415 | } |
416 | ||
1bf85046 | 417 | static void set_state_cb(struct extent_io_tree *tree, |
9ee49a04 | 418 | struct extent_state *state, unsigned *bits) |
291d673e | 419 | { |
1bf85046 JM |
420 | if (tree->ops && tree->ops->set_bit_hook) |
421 | tree->ops->set_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
422 | } |
423 | ||
424 | static void clear_state_cb(struct extent_io_tree *tree, | |
9ee49a04 | 425 | struct extent_state *state, unsigned *bits) |
291d673e | 426 | { |
9ed74f2d JB |
427 | if (tree->ops && tree->ops->clear_bit_hook) |
428 | tree->ops->clear_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
429 | } |
430 | ||
3150b699 | 431 | static void set_state_bits(struct extent_io_tree *tree, |
d38ed27f QW |
432 | struct extent_state *state, unsigned *bits, |
433 | struct extent_changeset *changeset); | |
3150b699 | 434 | |
d1310b2e CM |
435 | /* |
436 | * insert an extent_state struct into the tree. 'bits' are set on the | |
437 | * struct before it is inserted. | |
438 | * | |
439 | * This may return -EEXIST if the extent is already there, in which case the | |
440 | * state struct is freed. | |
441 | * | |
442 | * The tree lock is not taken internally. This is a utility function and | |
443 | * probably isn't what you want to call (see set/clear_extent_bit). | |
444 | */ | |
445 | static int insert_state(struct extent_io_tree *tree, | |
446 | struct extent_state *state, u64 start, u64 end, | |
12cfbad9 FDBM |
447 | struct rb_node ***p, |
448 | struct rb_node **parent, | |
d38ed27f | 449 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e CM |
450 | { |
451 | struct rb_node *node; | |
452 | ||
31b1a2bd | 453 | if (end < start) |
efe120a0 | 454 | WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n", |
c1c9ff7c | 455 | end, start); |
d1310b2e CM |
456 | state->start = start; |
457 | state->end = end; | |
9ed74f2d | 458 | |
d38ed27f | 459 | set_state_bits(tree, state, bits, changeset); |
3150b699 | 460 | |
f2071b21 | 461 | node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent); |
d1310b2e CM |
462 | if (node) { |
463 | struct extent_state *found; | |
464 | found = rb_entry(node, struct extent_state, rb_node); | |
efe120a0 | 465 | printk(KERN_ERR "BTRFS: found node %llu %llu on insert of " |
c1c9ff7c GU |
466 | "%llu %llu\n", |
467 | found->start, found->end, start, end); | |
d1310b2e CM |
468 | return -EEXIST; |
469 | } | |
470 | merge_state(tree, state); | |
471 | return 0; | |
472 | } | |
473 | ||
1bf85046 | 474 | static void split_cb(struct extent_io_tree *tree, struct extent_state *orig, |
9ed74f2d JB |
475 | u64 split) |
476 | { | |
477 | if (tree->ops && tree->ops->split_extent_hook) | |
1bf85046 | 478 | tree->ops->split_extent_hook(tree->mapping->host, orig, split); |
9ed74f2d JB |
479 | } |
480 | ||
d1310b2e CM |
481 | /* |
482 | * split a given extent state struct in two, inserting the preallocated | |
483 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
484 | * offset inside 'orig' where it should be split. | |
485 | * | |
486 | * Before calling, | |
487 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
488 | * are two extent state structs in the tree: | |
489 | * prealloc: [orig->start, split - 1] | |
490 | * orig: [ split, orig->end ] | |
491 | * | |
492 | * The tree locks are not taken by this function. They need to be held | |
493 | * by the caller. | |
494 | */ | |
495 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
496 | struct extent_state *prealloc, u64 split) | |
497 | { | |
498 | struct rb_node *node; | |
9ed74f2d JB |
499 | |
500 | split_cb(tree, orig, split); | |
501 | ||
d1310b2e CM |
502 | prealloc->start = orig->start; |
503 | prealloc->end = split - 1; | |
504 | prealloc->state = orig->state; | |
505 | orig->start = split; | |
506 | ||
f2071b21 FM |
507 | node = tree_insert(&tree->state, &orig->rb_node, prealloc->end, |
508 | &prealloc->rb_node, NULL, NULL); | |
d1310b2e | 509 | if (node) { |
d1310b2e CM |
510 | free_extent_state(prealloc); |
511 | return -EEXIST; | |
512 | } | |
513 | return 0; | |
514 | } | |
515 | ||
cdc6a395 LZ |
516 | static struct extent_state *next_state(struct extent_state *state) |
517 | { | |
518 | struct rb_node *next = rb_next(&state->rb_node); | |
519 | if (next) | |
520 | return rb_entry(next, struct extent_state, rb_node); | |
521 | else | |
522 | return NULL; | |
523 | } | |
524 | ||
d1310b2e CM |
525 | /* |
526 | * utility function to clear some bits in an extent state struct. | |
1b303fc0 | 527 | * it will optionally wake up any one waiting on this state (wake == 1). |
d1310b2e CM |
528 | * |
529 | * If no bits are set on the state struct after clearing things, the | |
530 | * struct is freed and removed from the tree | |
531 | */ | |
cdc6a395 LZ |
532 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
533 | struct extent_state *state, | |
fefdc557 QW |
534 | unsigned *bits, int wake, |
535 | struct extent_changeset *changeset) | |
d1310b2e | 536 | { |
cdc6a395 | 537 | struct extent_state *next; |
9ee49a04 | 538 | unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS; |
d1310b2e | 539 | |
0ca1f7ce | 540 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
541 | u64 range = state->end - state->start + 1; |
542 | WARN_ON(range > tree->dirty_bytes); | |
543 | tree->dirty_bytes -= range; | |
544 | } | |
291d673e | 545 | clear_state_cb(tree, state, bits); |
fefdc557 | 546 | add_extent_changeset(state, bits_to_clear, changeset, 0); |
32c00aff | 547 | state->state &= ~bits_to_clear; |
d1310b2e CM |
548 | if (wake) |
549 | wake_up(&state->wq); | |
0ca1f7ce | 550 | if (state->state == 0) { |
cdc6a395 | 551 | next = next_state(state); |
27a3507d | 552 | if (extent_state_in_tree(state)) { |
d1310b2e | 553 | rb_erase(&state->rb_node, &tree->state); |
27a3507d | 554 | RB_CLEAR_NODE(&state->rb_node); |
d1310b2e CM |
555 | free_extent_state(state); |
556 | } else { | |
557 | WARN_ON(1); | |
558 | } | |
559 | } else { | |
560 | merge_state(tree, state); | |
cdc6a395 | 561 | next = next_state(state); |
d1310b2e | 562 | } |
cdc6a395 | 563 | return next; |
d1310b2e CM |
564 | } |
565 | ||
8233767a XG |
566 | static struct extent_state * |
567 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
568 | { | |
569 | if (!prealloc) | |
570 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
571 | ||
572 | return prealloc; | |
573 | } | |
574 | ||
48a3b636 | 575 | static void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
c2d904e0 JM |
576 | { |
577 | btrfs_panic(tree_fs_info(tree), err, "Locking error: " | |
578 | "Extent tree was modified by another " | |
579 | "thread while locked."); | |
580 | } | |
581 | ||
d1310b2e CM |
582 | /* |
583 | * clear some bits on a range in the tree. This may require splitting | |
584 | * or inserting elements in the tree, so the gfp mask is used to | |
585 | * indicate which allocations or sleeping are allowed. | |
586 | * | |
587 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
588 | * the given range from the tree regardless of state (ie for truncate). | |
589 | * | |
590 | * the range [start, end] is inclusive. | |
591 | * | |
6763af84 | 592 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e | 593 | */ |
fefdc557 QW |
594 | static int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
595 | unsigned bits, int wake, int delete, | |
596 | struct extent_state **cached_state, | |
597 | gfp_t mask, struct extent_changeset *changeset) | |
d1310b2e CM |
598 | { |
599 | struct extent_state *state; | |
2c64c53d | 600 | struct extent_state *cached; |
d1310b2e CM |
601 | struct extent_state *prealloc = NULL; |
602 | struct rb_node *node; | |
5c939df5 | 603 | u64 last_end; |
d1310b2e | 604 | int err; |
2ac55d41 | 605 | int clear = 0; |
d1310b2e | 606 | |
a5dee37d | 607 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 608 | |
7ee9e440 JB |
609 | if (bits & EXTENT_DELALLOC) |
610 | bits |= EXTENT_NORESERVE; | |
611 | ||
0ca1f7ce YZ |
612 | if (delete) |
613 | bits |= ~EXTENT_CTLBITS; | |
614 | bits |= EXTENT_FIRST_DELALLOC; | |
615 | ||
2ac55d41 JB |
616 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
617 | clear = 1; | |
d1310b2e | 618 | again: |
d0164adc | 619 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
c7bc6319 FM |
620 | /* |
621 | * Don't care for allocation failure here because we might end | |
622 | * up not needing the pre-allocated extent state at all, which | |
623 | * is the case if we only have in the tree extent states that | |
624 | * cover our input range and don't cover too any other range. | |
625 | * If we end up needing a new extent state we allocate it later. | |
626 | */ | |
d1310b2e | 627 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
628 | } |
629 | ||
cad321ad | 630 | spin_lock(&tree->lock); |
2c64c53d CM |
631 | if (cached_state) { |
632 | cached = *cached_state; | |
2ac55d41 JB |
633 | |
634 | if (clear) { | |
635 | *cached_state = NULL; | |
636 | cached_state = NULL; | |
637 | } | |
638 | ||
27a3507d FM |
639 | if (cached && extent_state_in_tree(cached) && |
640 | cached->start <= start && cached->end > start) { | |
2ac55d41 JB |
641 | if (clear) |
642 | atomic_dec(&cached->refs); | |
2c64c53d | 643 | state = cached; |
42daec29 | 644 | goto hit_next; |
2c64c53d | 645 | } |
2ac55d41 JB |
646 | if (clear) |
647 | free_extent_state(cached); | |
2c64c53d | 648 | } |
d1310b2e CM |
649 | /* |
650 | * this search will find the extents that end after | |
651 | * our range starts | |
652 | */ | |
80ea96b1 | 653 | node = tree_search(tree, start); |
d1310b2e CM |
654 | if (!node) |
655 | goto out; | |
656 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 657 | hit_next: |
d1310b2e CM |
658 | if (state->start > end) |
659 | goto out; | |
660 | WARN_ON(state->end < start); | |
5c939df5 | 661 | last_end = state->end; |
d1310b2e | 662 | |
0449314a | 663 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
664 | if (!(state->state & bits)) { |
665 | state = next_state(state); | |
0449314a | 666 | goto next; |
cdc6a395 | 667 | } |
0449314a | 668 | |
d1310b2e CM |
669 | /* |
670 | * | ---- desired range ---- | | |
671 | * | state | or | |
672 | * | ------------- state -------------- | | |
673 | * | |
674 | * We need to split the extent we found, and may flip | |
675 | * bits on second half. | |
676 | * | |
677 | * If the extent we found extends past our range, we | |
678 | * just split and search again. It'll get split again | |
679 | * the next time though. | |
680 | * | |
681 | * If the extent we found is inside our range, we clear | |
682 | * the desired bit on it. | |
683 | */ | |
684 | ||
685 | if (state->start < start) { | |
8233767a XG |
686 | prealloc = alloc_extent_state_atomic(prealloc); |
687 | BUG_ON(!prealloc); | |
d1310b2e | 688 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
689 | if (err) |
690 | extent_io_tree_panic(tree, err); | |
691 | ||
d1310b2e CM |
692 | prealloc = NULL; |
693 | if (err) | |
694 | goto out; | |
695 | if (state->end <= end) { | |
fefdc557 QW |
696 | state = clear_state_bit(tree, state, &bits, wake, |
697 | changeset); | |
d1ac6e41 | 698 | goto next; |
d1310b2e CM |
699 | } |
700 | goto search_again; | |
701 | } | |
702 | /* | |
703 | * | ---- desired range ---- | | |
704 | * | state | | |
705 | * We need to split the extent, and clear the bit | |
706 | * on the first half | |
707 | */ | |
708 | if (state->start <= end && state->end > end) { | |
8233767a XG |
709 | prealloc = alloc_extent_state_atomic(prealloc); |
710 | BUG_ON(!prealloc); | |
d1310b2e | 711 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
712 | if (err) |
713 | extent_io_tree_panic(tree, err); | |
714 | ||
d1310b2e CM |
715 | if (wake) |
716 | wake_up(&state->wq); | |
42daec29 | 717 | |
fefdc557 | 718 | clear_state_bit(tree, prealloc, &bits, wake, changeset); |
9ed74f2d | 719 | |
d1310b2e CM |
720 | prealloc = NULL; |
721 | goto out; | |
722 | } | |
42daec29 | 723 | |
fefdc557 | 724 | state = clear_state_bit(tree, state, &bits, wake, changeset); |
0449314a | 725 | next: |
5c939df5 YZ |
726 | if (last_end == (u64)-1) |
727 | goto out; | |
728 | start = last_end + 1; | |
cdc6a395 | 729 | if (start <= end && state && !need_resched()) |
692e5759 | 730 | goto hit_next; |
d1310b2e CM |
731 | goto search_again; |
732 | ||
733 | out: | |
cad321ad | 734 | spin_unlock(&tree->lock); |
d1310b2e CM |
735 | if (prealloc) |
736 | free_extent_state(prealloc); | |
737 | ||
6763af84 | 738 | return 0; |
d1310b2e CM |
739 | |
740 | search_again: | |
741 | if (start > end) | |
742 | goto out; | |
cad321ad | 743 | spin_unlock(&tree->lock); |
d0164adc | 744 | if (gfpflags_allow_blocking(mask)) |
d1310b2e CM |
745 | cond_resched(); |
746 | goto again; | |
747 | } | |
d1310b2e | 748 | |
143bede5 JM |
749 | static void wait_on_state(struct extent_io_tree *tree, |
750 | struct extent_state *state) | |
641f5219 CH |
751 | __releases(tree->lock) |
752 | __acquires(tree->lock) | |
d1310b2e CM |
753 | { |
754 | DEFINE_WAIT(wait); | |
755 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 756 | spin_unlock(&tree->lock); |
d1310b2e | 757 | schedule(); |
cad321ad | 758 | spin_lock(&tree->lock); |
d1310b2e | 759 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
760 | } |
761 | ||
762 | /* | |
763 | * waits for one or more bits to clear on a range in the state tree. | |
764 | * The range [start, end] is inclusive. | |
765 | * The tree lock is taken by this function | |
766 | */ | |
41074888 DS |
767 | static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
768 | unsigned long bits) | |
d1310b2e CM |
769 | { |
770 | struct extent_state *state; | |
771 | struct rb_node *node; | |
772 | ||
a5dee37d | 773 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 774 | |
cad321ad | 775 | spin_lock(&tree->lock); |
d1310b2e CM |
776 | again: |
777 | while (1) { | |
778 | /* | |
779 | * this search will find all the extents that end after | |
780 | * our range starts | |
781 | */ | |
80ea96b1 | 782 | node = tree_search(tree, start); |
c50d3e71 | 783 | process_node: |
d1310b2e CM |
784 | if (!node) |
785 | break; | |
786 | ||
787 | state = rb_entry(node, struct extent_state, rb_node); | |
788 | ||
789 | if (state->start > end) | |
790 | goto out; | |
791 | ||
792 | if (state->state & bits) { | |
793 | start = state->start; | |
794 | atomic_inc(&state->refs); | |
795 | wait_on_state(tree, state); | |
796 | free_extent_state(state); | |
797 | goto again; | |
798 | } | |
799 | start = state->end + 1; | |
800 | ||
801 | if (start > end) | |
802 | break; | |
803 | ||
c50d3e71 FM |
804 | if (!cond_resched_lock(&tree->lock)) { |
805 | node = rb_next(node); | |
806 | goto process_node; | |
807 | } | |
d1310b2e CM |
808 | } |
809 | out: | |
cad321ad | 810 | spin_unlock(&tree->lock); |
d1310b2e | 811 | } |
d1310b2e | 812 | |
1bf85046 | 813 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 814 | struct extent_state *state, |
d38ed27f | 815 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e | 816 | { |
9ee49a04 | 817 | unsigned bits_to_set = *bits & ~EXTENT_CTLBITS; |
9ed74f2d | 818 | |
1bf85046 | 819 | set_state_cb(tree, state, bits); |
0ca1f7ce | 820 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
821 | u64 range = state->end - state->start + 1; |
822 | tree->dirty_bytes += range; | |
823 | } | |
d38ed27f | 824 | add_extent_changeset(state, bits_to_set, changeset, 1); |
0ca1f7ce | 825 | state->state |= bits_to_set; |
d1310b2e CM |
826 | } |
827 | ||
e38e2ed7 FM |
828 | static void cache_state_if_flags(struct extent_state *state, |
829 | struct extent_state **cached_ptr, | |
9ee49a04 | 830 | unsigned flags) |
2c64c53d CM |
831 | { |
832 | if (cached_ptr && !(*cached_ptr)) { | |
e38e2ed7 | 833 | if (!flags || (state->state & flags)) { |
2c64c53d CM |
834 | *cached_ptr = state; |
835 | atomic_inc(&state->refs); | |
836 | } | |
837 | } | |
838 | } | |
839 | ||
e38e2ed7 FM |
840 | static void cache_state(struct extent_state *state, |
841 | struct extent_state **cached_ptr) | |
842 | { | |
843 | return cache_state_if_flags(state, cached_ptr, | |
844 | EXTENT_IOBITS | EXTENT_BOUNDARY); | |
845 | } | |
846 | ||
d1310b2e | 847 | /* |
1edbb734 CM |
848 | * set some bits on a range in the tree. This may require allocations or |
849 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 850 | * |
1edbb734 CM |
851 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
852 | * part of the range already has the desired bits set. The start of the | |
853 | * existing range is returned in failed_start in this case. | |
d1310b2e | 854 | * |
1edbb734 | 855 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 856 | */ |
1edbb734 | 857 | |
3fbe5c02 JM |
858 | static int __must_check |
859 | __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 860 | unsigned bits, unsigned exclusive_bits, |
41074888 | 861 | u64 *failed_start, struct extent_state **cached_state, |
d38ed27f | 862 | gfp_t mask, struct extent_changeset *changeset) |
d1310b2e CM |
863 | { |
864 | struct extent_state *state; | |
865 | struct extent_state *prealloc = NULL; | |
866 | struct rb_node *node; | |
12cfbad9 FDBM |
867 | struct rb_node **p; |
868 | struct rb_node *parent; | |
d1310b2e | 869 | int err = 0; |
d1310b2e CM |
870 | u64 last_start; |
871 | u64 last_end; | |
42daec29 | 872 | |
a5dee37d | 873 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 874 | |
0ca1f7ce | 875 | bits |= EXTENT_FIRST_DELALLOC; |
d1310b2e | 876 | again: |
d0164adc | 877 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
d1310b2e | 878 | prealloc = alloc_extent_state(mask); |
8233767a | 879 | BUG_ON(!prealloc); |
d1310b2e CM |
880 | } |
881 | ||
cad321ad | 882 | spin_lock(&tree->lock); |
9655d298 CM |
883 | if (cached_state && *cached_state) { |
884 | state = *cached_state; | |
df98b6e2 | 885 | if (state->start <= start && state->end > start && |
27a3507d | 886 | extent_state_in_tree(state)) { |
9655d298 CM |
887 | node = &state->rb_node; |
888 | goto hit_next; | |
889 | } | |
890 | } | |
d1310b2e CM |
891 | /* |
892 | * this search will find all the extents that end after | |
893 | * our range starts. | |
894 | */ | |
12cfbad9 | 895 | node = tree_search_for_insert(tree, start, &p, &parent); |
d1310b2e | 896 | if (!node) { |
8233767a XG |
897 | prealloc = alloc_extent_state_atomic(prealloc); |
898 | BUG_ON(!prealloc); | |
12cfbad9 | 899 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 900 | &p, &parent, &bits, changeset); |
c2d904e0 JM |
901 | if (err) |
902 | extent_io_tree_panic(tree, err); | |
903 | ||
c42ac0bc | 904 | cache_state(prealloc, cached_state); |
d1310b2e | 905 | prealloc = NULL; |
d1310b2e CM |
906 | goto out; |
907 | } | |
d1310b2e | 908 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 909 | hit_next: |
d1310b2e CM |
910 | last_start = state->start; |
911 | last_end = state->end; | |
912 | ||
913 | /* | |
914 | * | ---- desired range ---- | | |
915 | * | state | | |
916 | * | |
917 | * Just lock what we found and keep going | |
918 | */ | |
919 | if (state->start == start && state->end <= end) { | |
1edbb734 | 920 | if (state->state & exclusive_bits) { |
d1310b2e CM |
921 | *failed_start = state->start; |
922 | err = -EEXIST; | |
923 | goto out; | |
924 | } | |
42daec29 | 925 | |
d38ed27f | 926 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 927 | cache_state(state, cached_state); |
d1310b2e | 928 | merge_state(tree, state); |
5c939df5 YZ |
929 | if (last_end == (u64)-1) |
930 | goto out; | |
931 | start = last_end + 1; | |
d1ac6e41 LB |
932 | state = next_state(state); |
933 | if (start < end && state && state->start == start && | |
934 | !need_resched()) | |
935 | goto hit_next; | |
d1310b2e CM |
936 | goto search_again; |
937 | } | |
938 | ||
939 | /* | |
940 | * | ---- desired range ---- | | |
941 | * | state | | |
942 | * or | |
943 | * | ------------- state -------------- | | |
944 | * | |
945 | * We need to split the extent we found, and may flip bits on | |
946 | * second half. | |
947 | * | |
948 | * If the extent we found extends past our | |
949 | * range, we just split and search again. It'll get split | |
950 | * again the next time though. | |
951 | * | |
952 | * If the extent we found is inside our range, we set the | |
953 | * desired bit on it. | |
954 | */ | |
955 | if (state->start < start) { | |
1edbb734 | 956 | if (state->state & exclusive_bits) { |
d1310b2e CM |
957 | *failed_start = start; |
958 | err = -EEXIST; | |
959 | goto out; | |
960 | } | |
8233767a XG |
961 | |
962 | prealloc = alloc_extent_state_atomic(prealloc); | |
963 | BUG_ON(!prealloc); | |
d1310b2e | 964 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
965 | if (err) |
966 | extent_io_tree_panic(tree, err); | |
967 | ||
d1310b2e CM |
968 | prealloc = NULL; |
969 | if (err) | |
970 | goto out; | |
971 | if (state->end <= end) { | |
d38ed27f | 972 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 973 | cache_state(state, cached_state); |
d1310b2e | 974 | merge_state(tree, state); |
5c939df5 YZ |
975 | if (last_end == (u64)-1) |
976 | goto out; | |
977 | start = last_end + 1; | |
d1ac6e41 LB |
978 | state = next_state(state); |
979 | if (start < end && state && state->start == start && | |
980 | !need_resched()) | |
981 | goto hit_next; | |
d1310b2e CM |
982 | } |
983 | goto search_again; | |
984 | } | |
985 | /* | |
986 | * | ---- desired range ---- | | |
987 | * | state | or | state | | |
988 | * | |
989 | * There's a hole, we need to insert something in it and | |
990 | * ignore the extent we found. | |
991 | */ | |
992 | if (state->start > start) { | |
993 | u64 this_end; | |
994 | if (end < last_start) | |
995 | this_end = end; | |
996 | else | |
d397712b | 997 | this_end = last_start - 1; |
8233767a XG |
998 | |
999 | prealloc = alloc_extent_state_atomic(prealloc); | |
1000 | BUG_ON(!prealloc); | |
c7f895a2 XG |
1001 | |
1002 | /* | |
1003 | * Avoid to free 'prealloc' if it can be merged with | |
1004 | * the later extent. | |
1005 | */ | |
d1310b2e | 1006 | err = insert_state(tree, prealloc, start, this_end, |
d38ed27f | 1007 | NULL, NULL, &bits, changeset); |
c2d904e0 JM |
1008 | if (err) |
1009 | extent_io_tree_panic(tree, err); | |
1010 | ||
9ed74f2d JB |
1011 | cache_state(prealloc, cached_state); |
1012 | prealloc = NULL; | |
d1310b2e CM |
1013 | start = this_end + 1; |
1014 | goto search_again; | |
1015 | } | |
1016 | /* | |
1017 | * | ---- desired range ---- | | |
1018 | * | state | | |
1019 | * We need to split the extent, and set the bit | |
1020 | * on the first half | |
1021 | */ | |
1022 | if (state->start <= end && state->end > end) { | |
1edbb734 | 1023 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1024 | *failed_start = start; |
1025 | err = -EEXIST; | |
1026 | goto out; | |
1027 | } | |
8233767a XG |
1028 | |
1029 | prealloc = alloc_extent_state_atomic(prealloc); | |
1030 | BUG_ON(!prealloc); | |
d1310b2e | 1031 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
1032 | if (err) |
1033 | extent_io_tree_panic(tree, err); | |
d1310b2e | 1034 | |
d38ed27f | 1035 | set_state_bits(tree, prealloc, &bits, changeset); |
2c64c53d | 1036 | cache_state(prealloc, cached_state); |
d1310b2e CM |
1037 | merge_state(tree, prealloc); |
1038 | prealloc = NULL; | |
1039 | goto out; | |
1040 | } | |
1041 | ||
1042 | goto search_again; | |
1043 | ||
1044 | out: | |
cad321ad | 1045 | spin_unlock(&tree->lock); |
d1310b2e CM |
1046 | if (prealloc) |
1047 | free_extent_state(prealloc); | |
1048 | ||
1049 | return err; | |
1050 | ||
1051 | search_again: | |
1052 | if (start > end) | |
1053 | goto out; | |
cad321ad | 1054 | spin_unlock(&tree->lock); |
d0164adc | 1055 | if (gfpflags_allow_blocking(mask)) |
d1310b2e CM |
1056 | cond_resched(); |
1057 | goto again; | |
1058 | } | |
d1310b2e | 1059 | |
41074888 | 1060 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
9ee49a04 | 1061 | unsigned bits, u64 * failed_start, |
41074888 | 1062 | struct extent_state **cached_state, gfp_t mask) |
3fbe5c02 JM |
1063 | { |
1064 | return __set_extent_bit(tree, start, end, bits, 0, failed_start, | |
d38ed27f | 1065 | cached_state, mask, NULL); |
3fbe5c02 JM |
1066 | } |
1067 | ||
1068 | ||
462d6fac | 1069 | /** |
10983f2e LB |
1070 | * convert_extent_bit - convert all bits in a given range from one bit to |
1071 | * another | |
462d6fac JB |
1072 | * @tree: the io tree to search |
1073 | * @start: the start offset in bytes | |
1074 | * @end: the end offset in bytes (inclusive) | |
1075 | * @bits: the bits to set in this range | |
1076 | * @clear_bits: the bits to clear in this range | |
e6138876 | 1077 | * @cached_state: state that we're going to cache |
462d6fac JB |
1078 | * @mask: the allocation mask |
1079 | * | |
1080 | * This will go through and set bits for the given range. If any states exist | |
1081 | * already in this range they are set with the given bit and cleared of the | |
1082 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
1083 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
1084 | * boundary bits like LOCK. | |
1085 | */ | |
1086 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1087 | unsigned bits, unsigned clear_bits, |
e6138876 | 1088 | struct extent_state **cached_state, gfp_t mask) |
462d6fac JB |
1089 | { |
1090 | struct extent_state *state; | |
1091 | struct extent_state *prealloc = NULL; | |
1092 | struct rb_node *node; | |
12cfbad9 FDBM |
1093 | struct rb_node **p; |
1094 | struct rb_node *parent; | |
462d6fac JB |
1095 | int err = 0; |
1096 | u64 last_start; | |
1097 | u64 last_end; | |
c8fd3de7 | 1098 | bool first_iteration = true; |
462d6fac | 1099 | |
a5dee37d | 1100 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 1101 | |
462d6fac | 1102 | again: |
d0164adc | 1103 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
c8fd3de7 FM |
1104 | /* |
1105 | * Best effort, don't worry if extent state allocation fails | |
1106 | * here for the first iteration. We might have a cached state | |
1107 | * that matches exactly the target range, in which case no | |
1108 | * extent state allocations are needed. We'll only know this | |
1109 | * after locking the tree. | |
1110 | */ | |
462d6fac | 1111 | prealloc = alloc_extent_state(mask); |
c8fd3de7 | 1112 | if (!prealloc && !first_iteration) |
462d6fac JB |
1113 | return -ENOMEM; |
1114 | } | |
1115 | ||
1116 | spin_lock(&tree->lock); | |
e6138876 JB |
1117 | if (cached_state && *cached_state) { |
1118 | state = *cached_state; | |
1119 | if (state->start <= start && state->end > start && | |
27a3507d | 1120 | extent_state_in_tree(state)) { |
e6138876 JB |
1121 | node = &state->rb_node; |
1122 | goto hit_next; | |
1123 | } | |
1124 | } | |
1125 | ||
462d6fac JB |
1126 | /* |
1127 | * this search will find all the extents that end after | |
1128 | * our range starts. | |
1129 | */ | |
12cfbad9 | 1130 | node = tree_search_for_insert(tree, start, &p, &parent); |
462d6fac JB |
1131 | if (!node) { |
1132 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1133 | if (!prealloc) { |
1134 | err = -ENOMEM; | |
1135 | goto out; | |
1136 | } | |
12cfbad9 | 1137 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 1138 | &p, &parent, &bits, NULL); |
c2d904e0 JM |
1139 | if (err) |
1140 | extent_io_tree_panic(tree, err); | |
c42ac0bc FDBM |
1141 | cache_state(prealloc, cached_state); |
1142 | prealloc = NULL; | |
462d6fac JB |
1143 | goto out; |
1144 | } | |
1145 | state = rb_entry(node, struct extent_state, rb_node); | |
1146 | hit_next: | |
1147 | last_start = state->start; | |
1148 | last_end = state->end; | |
1149 | ||
1150 | /* | |
1151 | * | ---- desired range ---- | | |
1152 | * | state | | |
1153 | * | |
1154 | * Just lock what we found and keep going | |
1155 | */ | |
1156 | if (state->start == start && state->end <= end) { | |
d38ed27f | 1157 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1158 | cache_state(state, cached_state); |
fefdc557 | 1159 | state = clear_state_bit(tree, state, &clear_bits, 0, NULL); |
462d6fac JB |
1160 | if (last_end == (u64)-1) |
1161 | goto out; | |
462d6fac | 1162 | start = last_end + 1; |
d1ac6e41 LB |
1163 | if (start < end && state && state->start == start && |
1164 | !need_resched()) | |
1165 | goto hit_next; | |
462d6fac JB |
1166 | goto search_again; |
1167 | } | |
1168 | ||
1169 | /* | |
1170 | * | ---- desired range ---- | | |
1171 | * | state | | |
1172 | * or | |
1173 | * | ------------- state -------------- | | |
1174 | * | |
1175 | * We need to split the extent we found, and may flip bits on | |
1176 | * second half. | |
1177 | * | |
1178 | * If the extent we found extends past our | |
1179 | * range, we just split and search again. It'll get split | |
1180 | * again the next time though. | |
1181 | * | |
1182 | * If the extent we found is inside our range, we set the | |
1183 | * desired bit on it. | |
1184 | */ | |
1185 | if (state->start < start) { | |
1186 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1187 | if (!prealloc) { |
1188 | err = -ENOMEM; | |
1189 | goto out; | |
1190 | } | |
462d6fac | 1191 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1192 | if (err) |
1193 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1194 | prealloc = NULL; |
1195 | if (err) | |
1196 | goto out; | |
1197 | if (state->end <= end) { | |
d38ed27f | 1198 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1199 | cache_state(state, cached_state); |
fefdc557 QW |
1200 | state = clear_state_bit(tree, state, &clear_bits, 0, |
1201 | NULL); | |
462d6fac JB |
1202 | if (last_end == (u64)-1) |
1203 | goto out; | |
1204 | start = last_end + 1; | |
d1ac6e41 LB |
1205 | if (start < end && state && state->start == start && |
1206 | !need_resched()) | |
1207 | goto hit_next; | |
462d6fac JB |
1208 | } |
1209 | goto search_again; | |
1210 | } | |
1211 | /* | |
1212 | * | ---- desired range ---- | | |
1213 | * | state | or | state | | |
1214 | * | |
1215 | * There's a hole, we need to insert something in it and | |
1216 | * ignore the extent we found. | |
1217 | */ | |
1218 | if (state->start > start) { | |
1219 | u64 this_end; | |
1220 | if (end < last_start) | |
1221 | this_end = end; | |
1222 | else | |
1223 | this_end = last_start - 1; | |
1224 | ||
1225 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1226 | if (!prealloc) { |
1227 | err = -ENOMEM; | |
1228 | goto out; | |
1229 | } | |
462d6fac JB |
1230 | |
1231 | /* | |
1232 | * Avoid to free 'prealloc' if it can be merged with | |
1233 | * the later extent. | |
1234 | */ | |
1235 | err = insert_state(tree, prealloc, start, this_end, | |
d38ed27f | 1236 | NULL, NULL, &bits, NULL); |
c2d904e0 JM |
1237 | if (err) |
1238 | extent_io_tree_panic(tree, err); | |
e6138876 | 1239 | cache_state(prealloc, cached_state); |
462d6fac JB |
1240 | prealloc = NULL; |
1241 | start = this_end + 1; | |
1242 | goto search_again; | |
1243 | } | |
1244 | /* | |
1245 | * | ---- desired range ---- | | |
1246 | * | state | | |
1247 | * We need to split the extent, and set the bit | |
1248 | * on the first half | |
1249 | */ | |
1250 | if (state->start <= end && state->end > end) { | |
1251 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1252 | if (!prealloc) { |
1253 | err = -ENOMEM; | |
1254 | goto out; | |
1255 | } | |
462d6fac JB |
1256 | |
1257 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1258 | if (err) |
1259 | extent_io_tree_panic(tree, err); | |
462d6fac | 1260 | |
d38ed27f | 1261 | set_state_bits(tree, prealloc, &bits, NULL); |
e6138876 | 1262 | cache_state(prealloc, cached_state); |
fefdc557 | 1263 | clear_state_bit(tree, prealloc, &clear_bits, 0, NULL); |
462d6fac JB |
1264 | prealloc = NULL; |
1265 | goto out; | |
1266 | } | |
1267 | ||
1268 | goto search_again; | |
1269 | ||
1270 | out: | |
1271 | spin_unlock(&tree->lock); | |
1272 | if (prealloc) | |
1273 | free_extent_state(prealloc); | |
1274 | ||
1275 | return err; | |
1276 | ||
1277 | search_again: | |
1278 | if (start > end) | |
1279 | goto out; | |
1280 | spin_unlock(&tree->lock); | |
d0164adc | 1281 | if (gfpflags_allow_blocking(mask)) |
462d6fac | 1282 | cond_resched(); |
c8fd3de7 | 1283 | first_iteration = false; |
462d6fac JB |
1284 | goto again; |
1285 | } | |
1286 | ||
d1310b2e | 1287 | /* wrappers around set/clear extent bit */ |
d38ed27f QW |
1288 | int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
1289 | unsigned bits, gfp_t mask, | |
1290 | struct extent_changeset *changeset) | |
1291 | { | |
1292 | /* | |
1293 | * We don't support EXTENT_LOCKED yet, as current changeset will | |
1294 | * record any bits changed, so for EXTENT_LOCKED case, it will | |
1295 | * either fail with -EEXIST or changeset will record the whole | |
1296 | * range. | |
1297 | */ | |
1298 | BUG_ON(bits & EXTENT_LOCKED); | |
1299 | ||
1300 | return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, mask, | |
1301 | changeset); | |
1302 | } | |
1303 | ||
fefdc557 QW |
1304 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
1305 | unsigned bits, int wake, int delete, | |
1306 | struct extent_state **cached, gfp_t mask) | |
1307 | { | |
1308 | return __clear_extent_bit(tree, start, end, bits, wake, delete, | |
1309 | cached, mask, NULL); | |
1310 | } | |
1311 | ||
fefdc557 QW |
1312 | int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
1313 | unsigned bits, gfp_t mask, | |
1314 | struct extent_changeset *changeset) | |
1315 | { | |
1316 | /* | |
1317 | * Don't support EXTENT_LOCKED case, same reason as | |
1318 | * set_record_extent_bits(). | |
1319 | */ | |
1320 | BUG_ON(bits & EXTENT_LOCKED); | |
1321 | ||
1322 | return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask, | |
1323 | changeset); | |
1324 | } | |
1325 | ||
d352ac68 CM |
1326 | /* |
1327 | * either insert or lock state struct between start and end use mask to tell | |
1328 | * us if waiting is desired. | |
1329 | */ | |
1edbb734 | 1330 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
ff13db41 | 1331 | struct extent_state **cached_state) |
d1310b2e CM |
1332 | { |
1333 | int err; | |
1334 | u64 failed_start; | |
9ee49a04 | 1335 | |
d1310b2e | 1336 | while (1) { |
ff13db41 | 1337 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, |
3fbe5c02 | 1338 | EXTENT_LOCKED, &failed_start, |
d38ed27f | 1339 | cached_state, GFP_NOFS, NULL); |
d0082371 | 1340 | if (err == -EEXIST) { |
d1310b2e CM |
1341 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1342 | start = failed_start; | |
d0082371 | 1343 | } else |
d1310b2e | 1344 | break; |
d1310b2e CM |
1345 | WARN_ON(start > end); |
1346 | } | |
1347 | return err; | |
1348 | } | |
d1310b2e | 1349 | |
d0082371 | 1350 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1351 | { |
1352 | int err; | |
1353 | u64 failed_start; | |
1354 | ||
3fbe5c02 | 1355 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
d38ed27f | 1356 | &failed_start, NULL, GFP_NOFS, NULL); |
6643558d YZ |
1357 | if (err == -EEXIST) { |
1358 | if (failed_start > start) | |
1359 | clear_extent_bit(tree, start, failed_start - 1, | |
d0082371 | 1360 | EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS); |
25179201 | 1361 | return 0; |
6643558d | 1362 | } |
25179201 JB |
1363 | return 1; |
1364 | } | |
25179201 | 1365 | |
bd1fa4f0 | 1366 | void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 CM |
1367 | { |
1368 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1369 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1370 | struct page *page; | |
1371 | ||
1372 | while (index <= end_index) { | |
1373 | page = find_get_page(inode->i_mapping, index); | |
1374 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1375 | clear_page_dirty_for_io(page); | |
1376 | page_cache_release(page); | |
1377 | index++; | |
1378 | } | |
4adaa611 CM |
1379 | } |
1380 | ||
f6311572 | 1381 | void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 CM |
1382 | { |
1383 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1384 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1385 | struct page *page; | |
1386 | ||
1387 | while (index <= end_index) { | |
1388 | page = find_get_page(inode->i_mapping, index); | |
1389 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
4adaa611 | 1390 | __set_page_dirty_nobuffers(page); |
8d38633c | 1391 | account_page_redirty(page); |
4adaa611 CM |
1392 | page_cache_release(page); |
1393 | index++; | |
1394 | } | |
4adaa611 CM |
1395 | } |
1396 | ||
d1310b2e CM |
1397 | /* |
1398 | * helper function to set both pages and extents in the tree writeback | |
1399 | */ | |
35de6db2 | 1400 | static void set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e CM |
1401 | { |
1402 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1403 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1404 | struct page *page; | |
1405 | ||
1406 | while (index <= end_index) { | |
1407 | page = find_get_page(tree->mapping, index); | |
79787eaa | 1408 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ |
d1310b2e CM |
1409 | set_page_writeback(page); |
1410 | page_cache_release(page); | |
1411 | index++; | |
1412 | } | |
d1310b2e | 1413 | } |
d1310b2e | 1414 | |
d352ac68 CM |
1415 | /* find the first state struct with 'bits' set after 'start', and |
1416 | * return it. tree->lock must be held. NULL will returned if | |
1417 | * nothing was found after 'start' | |
1418 | */ | |
48a3b636 ES |
1419 | static struct extent_state * |
1420 | find_first_extent_bit_state(struct extent_io_tree *tree, | |
9ee49a04 | 1421 | u64 start, unsigned bits) |
d7fc640e CM |
1422 | { |
1423 | struct rb_node *node; | |
1424 | struct extent_state *state; | |
1425 | ||
1426 | /* | |
1427 | * this search will find all the extents that end after | |
1428 | * our range starts. | |
1429 | */ | |
1430 | node = tree_search(tree, start); | |
d397712b | 1431 | if (!node) |
d7fc640e | 1432 | goto out; |
d7fc640e | 1433 | |
d397712b | 1434 | while (1) { |
d7fc640e | 1435 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1436 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1437 | return state; |
d397712b | 1438 | |
d7fc640e CM |
1439 | node = rb_next(node); |
1440 | if (!node) | |
1441 | break; | |
1442 | } | |
1443 | out: | |
1444 | return NULL; | |
1445 | } | |
d7fc640e | 1446 | |
69261c4b XG |
1447 | /* |
1448 | * find the first offset in the io tree with 'bits' set. zero is | |
1449 | * returned if we find something, and *start_ret and *end_ret are | |
1450 | * set to reflect the state struct that was found. | |
1451 | * | |
477d7eaf | 1452 | * If nothing was found, 1 is returned. If found something, return 0. |
69261c4b XG |
1453 | */ |
1454 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
9ee49a04 | 1455 | u64 *start_ret, u64 *end_ret, unsigned bits, |
e6138876 | 1456 | struct extent_state **cached_state) |
69261c4b XG |
1457 | { |
1458 | struct extent_state *state; | |
e6138876 | 1459 | struct rb_node *n; |
69261c4b XG |
1460 | int ret = 1; |
1461 | ||
1462 | spin_lock(&tree->lock); | |
e6138876 JB |
1463 | if (cached_state && *cached_state) { |
1464 | state = *cached_state; | |
27a3507d | 1465 | if (state->end == start - 1 && extent_state_in_tree(state)) { |
e6138876 JB |
1466 | n = rb_next(&state->rb_node); |
1467 | while (n) { | |
1468 | state = rb_entry(n, struct extent_state, | |
1469 | rb_node); | |
1470 | if (state->state & bits) | |
1471 | goto got_it; | |
1472 | n = rb_next(n); | |
1473 | } | |
1474 | free_extent_state(*cached_state); | |
1475 | *cached_state = NULL; | |
1476 | goto out; | |
1477 | } | |
1478 | free_extent_state(*cached_state); | |
1479 | *cached_state = NULL; | |
1480 | } | |
1481 | ||
69261c4b | 1482 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1483 | got_it: |
69261c4b | 1484 | if (state) { |
e38e2ed7 | 1485 | cache_state_if_flags(state, cached_state, 0); |
69261c4b XG |
1486 | *start_ret = state->start; |
1487 | *end_ret = state->end; | |
1488 | ret = 0; | |
1489 | } | |
e6138876 | 1490 | out: |
69261c4b XG |
1491 | spin_unlock(&tree->lock); |
1492 | return ret; | |
1493 | } | |
1494 | ||
d352ac68 CM |
1495 | /* |
1496 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1497 | * more than 'max_bytes'. start and end are used to return the range, | |
1498 | * | |
1499 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1500 | */ | |
c8b97818 | 1501 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1502 | u64 *start, u64 *end, u64 max_bytes, |
1503 | struct extent_state **cached_state) | |
d1310b2e CM |
1504 | { |
1505 | struct rb_node *node; | |
1506 | struct extent_state *state; | |
1507 | u64 cur_start = *start; | |
1508 | u64 found = 0; | |
1509 | u64 total_bytes = 0; | |
1510 | ||
cad321ad | 1511 | spin_lock(&tree->lock); |
c8b97818 | 1512 | |
d1310b2e CM |
1513 | /* |
1514 | * this search will find all the extents that end after | |
1515 | * our range starts. | |
1516 | */ | |
80ea96b1 | 1517 | node = tree_search(tree, cur_start); |
2b114d1d | 1518 | if (!node) { |
3b951516 CM |
1519 | if (!found) |
1520 | *end = (u64)-1; | |
d1310b2e CM |
1521 | goto out; |
1522 | } | |
1523 | ||
d397712b | 1524 | while (1) { |
d1310b2e | 1525 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1526 | if (found && (state->start != cur_start || |
1527 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1528 | goto out; |
1529 | } | |
1530 | if (!(state->state & EXTENT_DELALLOC)) { | |
1531 | if (!found) | |
1532 | *end = state->end; | |
1533 | goto out; | |
1534 | } | |
c2a128d2 | 1535 | if (!found) { |
d1310b2e | 1536 | *start = state->start; |
c2a128d2 JB |
1537 | *cached_state = state; |
1538 | atomic_inc(&state->refs); | |
1539 | } | |
d1310b2e CM |
1540 | found++; |
1541 | *end = state->end; | |
1542 | cur_start = state->end + 1; | |
1543 | node = rb_next(node); | |
d1310b2e | 1544 | total_bytes += state->end - state->start + 1; |
7bf811a5 | 1545 | if (total_bytes >= max_bytes) |
573aecaf | 1546 | break; |
573aecaf | 1547 | if (!node) |
d1310b2e CM |
1548 | break; |
1549 | } | |
1550 | out: | |
cad321ad | 1551 | spin_unlock(&tree->lock); |
d1310b2e CM |
1552 | return found; |
1553 | } | |
1554 | ||
143bede5 JM |
1555 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1556 | struct page *locked_page, | |
1557 | u64 start, u64 end) | |
c8b97818 CM |
1558 | { |
1559 | int ret; | |
1560 | struct page *pages[16]; | |
1561 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1562 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1563 | unsigned long nr_pages = end_index - index + 1; | |
1564 | int i; | |
1565 | ||
1566 | if (index == locked_page->index && end_index == index) | |
143bede5 | 1567 | return; |
c8b97818 | 1568 | |
d397712b | 1569 | while (nr_pages > 0) { |
c8b97818 | 1570 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1571 | min_t(unsigned long, nr_pages, |
1572 | ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1573 | for (i = 0; i < ret; i++) { |
1574 | if (pages[i] != locked_page) | |
1575 | unlock_page(pages[i]); | |
1576 | page_cache_release(pages[i]); | |
1577 | } | |
1578 | nr_pages -= ret; | |
1579 | index += ret; | |
1580 | cond_resched(); | |
1581 | } | |
c8b97818 CM |
1582 | } |
1583 | ||
1584 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1585 | struct page *locked_page, | |
1586 | u64 delalloc_start, | |
1587 | u64 delalloc_end) | |
1588 | { | |
1589 | unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT; | |
1590 | unsigned long start_index = index; | |
1591 | unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT; | |
1592 | unsigned long pages_locked = 0; | |
1593 | struct page *pages[16]; | |
1594 | unsigned long nrpages; | |
1595 | int ret; | |
1596 | int i; | |
1597 | ||
1598 | /* the caller is responsible for locking the start index */ | |
1599 | if (index == locked_page->index && index == end_index) | |
1600 | return 0; | |
1601 | ||
1602 | /* skip the page at the start index */ | |
1603 | nrpages = end_index - index + 1; | |
d397712b | 1604 | while (nrpages > 0) { |
c8b97818 | 1605 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1606 | min_t(unsigned long, |
1607 | nrpages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1608 | if (ret == 0) { |
1609 | ret = -EAGAIN; | |
1610 | goto done; | |
1611 | } | |
1612 | /* now we have an array of pages, lock them all */ | |
1613 | for (i = 0; i < ret; i++) { | |
1614 | /* | |
1615 | * the caller is taking responsibility for | |
1616 | * locked_page | |
1617 | */ | |
771ed689 | 1618 | if (pages[i] != locked_page) { |
c8b97818 | 1619 | lock_page(pages[i]); |
f2b1c41c CM |
1620 | if (!PageDirty(pages[i]) || |
1621 | pages[i]->mapping != inode->i_mapping) { | |
771ed689 CM |
1622 | ret = -EAGAIN; |
1623 | unlock_page(pages[i]); | |
1624 | page_cache_release(pages[i]); | |
1625 | goto done; | |
1626 | } | |
1627 | } | |
c8b97818 | 1628 | page_cache_release(pages[i]); |
771ed689 | 1629 | pages_locked++; |
c8b97818 | 1630 | } |
c8b97818 CM |
1631 | nrpages -= ret; |
1632 | index += ret; | |
1633 | cond_resched(); | |
1634 | } | |
1635 | ret = 0; | |
1636 | done: | |
1637 | if (ret && pages_locked) { | |
1638 | __unlock_for_delalloc(inode, locked_page, | |
1639 | delalloc_start, | |
1640 | ((u64)(start_index + pages_locked - 1)) << | |
1641 | PAGE_CACHE_SHIFT); | |
1642 | } | |
1643 | return ret; | |
1644 | } | |
1645 | ||
1646 | /* | |
1647 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1648 | * more than 'max_bytes'. start and end are used to return the range, | |
1649 | * | |
1650 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1651 | */ | |
294e30fe JB |
1652 | STATIC u64 find_lock_delalloc_range(struct inode *inode, |
1653 | struct extent_io_tree *tree, | |
1654 | struct page *locked_page, u64 *start, | |
1655 | u64 *end, u64 max_bytes) | |
c8b97818 CM |
1656 | { |
1657 | u64 delalloc_start; | |
1658 | u64 delalloc_end; | |
1659 | u64 found; | |
9655d298 | 1660 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1661 | int ret; |
1662 | int loops = 0; | |
1663 | ||
1664 | again: | |
1665 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1666 | delalloc_start = *start; | |
1667 | delalloc_end = 0; | |
1668 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1669 | max_bytes, &cached_state); |
70b99e69 | 1670 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1671 | *start = delalloc_start; |
1672 | *end = delalloc_end; | |
c2a128d2 | 1673 | free_extent_state(cached_state); |
385fe0be | 1674 | return 0; |
c8b97818 CM |
1675 | } |
1676 | ||
70b99e69 CM |
1677 | /* |
1678 | * start comes from the offset of locked_page. We have to lock | |
1679 | * pages in order, so we can't process delalloc bytes before | |
1680 | * locked_page | |
1681 | */ | |
d397712b | 1682 | if (delalloc_start < *start) |
70b99e69 | 1683 | delalloc_start = *start; |
70b99e69 | 1684 | |
c8b97818 CM |
1685 | /* |
1686 | * make sure to limit the number of pages we try to lock down | |
c8b97818 | 1687 | */ |
7bf811a5 JB |
1688 | if (delalloc_end + 1 - delalloc_start > max_bytes) |
1689 | delalloc_end = delalloc_start + max_bytes - 1; | |
d397712b | 1690 | |
c8b97818 CM |
1691 | /* step two, lock all the pages after the page that has start */ |
1692 | ret = lock_delalloc_pages(inode, locked_page, | |
1693 | delalloc_start, delalloc_end); | |
1694 | if (ret == -EAGAIN) { | |
1695 | /* some of the pages are gone, lets avoid looping by | |
1696 | * shortening the size of the delalloc range we're searching | |
1697 | */ | |
9655d298 | 1698 | free_extent_state(cached_state); |
7d788742 | 1699 | cached_state = NULL; |
c8b97818 | 1700 | if (!loops) { |
7bf811a5 | 1701 | max_bytes = PAGE_CACHE_SIZE; |
c8b97818 CM |
1702 | loops = 1; |
1703 | goto again; | |
1704 | } else { | |
1705 | found = 0; | |
1706 | goto out_failed; | |
1707 | } | |
1708 | } | |
79787eaa | 1709 | BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */ |
c8b97818 CM |
1710 | |
1711 | /* step three, lock the state bits for the whole range */ | |
ff13db41 | 1712 | lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state); |
c8b97818 CM |
1713 | |
1714 | /* then test to make sure it is all still delalloc */ | |
1715 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1716 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1717 | if (!ret) { |
9655d298 CM |
1718 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
1719 | &cached_state, GFP_NOFS); | |
c8b97818 CM |
1720 | __unlock_for_delalloc(inode, locked_page, |
1721 | delalloc_start, delalloc_end); | |
1722 | cond_resched(); | |
1723 | goto again; | |
1724 | } | |
9655d298 | 1725 | free_extent_state(cached_state); |
c8b97818 CM |
1726 | *start = delalloc_start; |
1727 | *end = delalloc_end; | |
1728 | out_failed: | |
1729 | return found; | |
1730 | } | |
1731 | ||
a9d93e17 | 1732 | void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end, |
c2790a2e | 1733 | struct page *locked_page, |
9ee49a04 | 1734 | unsigned clear_bits, |
c2790a2e | 1735 | unsigned long page_ops) |
c8b97818 | 1736 | { |
c2790a2e | 1737 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
c8b97818 CM |
1738 | int ret; |
1739 | struct page *pages[16]; | |
1740 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1741 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1742 | unsigned long nr_pages = end_index - index + 1; | |
1743 | int i; | |
771ed689 | 1744 | |
2c64c53d | 1745 | clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); |
c2790a2e | 1746 | if (page_ops == 0) |
a9d93e17 | 1747 | return; |
c8b97818 | 1748 | |
704de49d FM |
1749 | if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0) |
1750 | mapping_set_error(inode->i_mapping, -EIO); | |
1751 | ||
d397712b | 1752 | while (nr_pages > 0) { |
c8b97818 | 1753 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1754 | min_t(unsigned long, |
1755 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 | 1756 | for (i = 0; i < ret; i++) { |
8b62b72b | 1757 | |
c2790a2e | 1758 | if (page_ops & PAGE_SET_PRIVATE2) |
8b62b72b CM |
1759 | SetPagePrivate2(pages[i]); |
1760 | ||
c8b97818 CM |
1761 | if (pages[i] == locked_page) { |
1762 | page_cache_release(pages[i]); | |
1763 | continue; | |
1764 | } | |
c2790a2e | 1765 | if (page_ops & PAGE_CLEAR_DIRTY) |
c8b97818 | 1766 | clear_page_dirty_for_io(pages[i]); |
c2790a2e | 1767 | if (page_ops & PAGE_SET_WRITEBACK) |
c8b97818 | 1768 | set_page_writeback(pages[i]); |
704de49d FM |
1769 | if (page_ops & PAGE_SET_ERROR) |
1770 | SetPageError(pages[i]); | |
c2790a2e | 1771 | if (page_ops & PAGE_END_WRITEBACK) |
c8b97818 | 1772 | end_page_writeback(pages[i]); |
c2790a2e | 1773 | if (page_ops & PAGE_UNLOCK) |
771ed689 | 1774 | unlock_page(pages[i]); |
c8b97818 CM |
1775 | page_cache_release(pages[i]); |
1776 | } | |
1777 | nr_pages -= ret; | |
1778 | index += ret; | |
1779 | cond_resched(); | |
1780 | } | |
c8b97818 | 1781 | } |
c8b97818 | 1782 | |
d352ac68 CM |
1783 | /* |
1784 | * count the number of bytes in the tree that have a given bit(s) | |
1785 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1786 | * cached. The total number found is returned. | |
1787 | */ | |
d1310b2e CM |
1788 | u64 count_range_bits(struct extent_io_tree *tree, |
1789 | u64 *start, u64 search_end, u64 max_bytes, | |
9ee49a04 | 1790 | unsigned bits, int contig) |
d1310b2e CM |
1791 | { |
1792 | struct rb_node *node; | |
1793 | struct extent_state *state; | |
1794 | u64 cur_start = *start; | |
1795 | u64 total_bytes = 0; | |
ec29ed5b | 1796 | u64 last = 0; |
d1310b2e CM |
1797 | int found = 0; |
1798 | ||
fae7f21c | 1799 | if (WARN_ON(search_end <= cur_start)) |
d1310b2e | 1800 | return 0; |
d1310b2e | 1801 | |
cad321ad | 1802 | spin_lock(&tree->lock); |
d1310b2e CM |
1803 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1804 | total_bytes = tree->dirty_bytes; | |
1805 | goto out; | |
1806 | } | |
1807 | /* | |
1808 | * this search will find all the extents that end after | |
1809 | * our range starts. | |
1810 | */ | |
80ea96b1 | 1811 | node = tree_search(tree, cur_start); |
d397712b | 1812 | if (!node) |
d1310b2e | 1813 | goto out; |
d1310b2e | 1814 | |
d397712b | 1815 | while (1) { |
d1310b2e CM |
1816 | state = rb_entry(node, struct extent_state, rb_node); |
1817 | if (state->start > search_end) | |
1818 | break; | |
ec29ed5b CM |
1819 | if (contig && found && state->start > last + 1) |
1820 | break; | |
1821 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1822 | total_bytes += min(search_end, state->end) + 1 - |
1823 | max(cur_start, state->start); | |
1824 | if (total_bytes >= max_bytes) | |
1825 | break; | |
1826 | if (!found) { | |
af60bed2 | 1827 | *start = max(cur_start, state->start); |
d1310b2e CM |
1828 | found = 1; |
1829 | } | |
ec29ed5b CM |
1830 | last = state->end; |
1831 | } else if (contig && found) { | |
1832 | break; | |
d1310b2e CM |
1833 | } |
1834 | node = rb_next(node); | |
1835 | if (!node) | |
1836 | break; | |
1837 | } | |
1838 | out: | |
cad321ad | 1839 | spin_unlock(&tree->lock); |
d1310b2e CM |
1840 | return total_bytes; |
1841 | } | |
b2950863 | 1842 | |
d352ac68 CM |
1843 | /* |
1844 | * set the private field for a given byte offset in the tree. If there isn't | |
1845 | * an extent_state there already, this does nothing. | |
1846 | */ | |
171170c1 | 1847 | static int set_state_private(struct extent_io_tree *tree, u64 start, u64 private) |
d1310b2e CM |
1848 | { |
1849 | struct rb_node *node; | |
1850 | struct extent_state *state; | |
1851 | int ret = 0; | |
1852 | ||
cad321ad | 1853 | spin_lock(&tree->lock); |
d1310b2e CM |
1854 | /* |
1855 | * this search will find all the extents that end after | |
1856 | * our range starts. | |
1857 | */ | |
80ea96b1 | 1858 | node = tree_search(tree, start); |
2b114d1d | 1859 | if (!node) { |
d1310b2e CM |
1860 | ret = -ENOENT; |
1861 | goto out; | |
1862 | } | |
1863 | state = rb_entry(node, struct extent_state, rb_node); | |
1864 | if (state->start != start) { | |
1865 | ret = -ENOENT; | |
1866 | goto out; | |
1867 | } | |
1868 | state->private = private; | |
1869 | out: | |
cad321ad | 1870 | spin_unlock(&tree->lock); |
d1310b2e CM |
1871 | return ret; |
1872 | } | |
1873 | ||
1874 | int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private) | |
1875 | { | |
1876 | struct rb_node *node; | |
1877 | struct extent_state *state; | |
1878 | int ret = 0; | |
1879 | ||
cad321ad | 1880 | spin_lock(&tree->lock); |
d1310b2e CM |
1881 | /* |
1882 | * this search will find all the extents that end after | |
1883 | * our range starts. | |
1884 | */ | |
80ea96b1 | 1885 | node = tree_search(tree, start); |
2b114d1d | 1886 | if (!node) { |
d1310b2e CM |
1887 | ret = -ENOENT; |
1888 | goto out; | |
1889 | } | |
1890 | state = rb_entry(node, struct extent_state, rb_node); | |
1891 | if (state->start != start) { | |
1892 | ret = -ENOENT; | |
1893 | goto out; | |
1894 | } | |
1895 | *private = state->private; | |
1896 | out: | |
cad321ad | 1897 | spin_unlock(&tree->lock); |
d1310b2e CM |
1898 | return ret; |
1899 | } | |
1900 | ||
1901 | /* | |
1902 | * searches a range in the state tree for a given mask. | |
70dec807 | 1903 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1904 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1905 | * range is found set. | |
1906 | */ | |
1907 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1908 | unsigned bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1909 | { |
1910 | struct extent_state *state = NULL; | |
1911 | struct rb_node *node; | |
1912 | int bitset = 0; | |
d1310b2e | 1913 | |
cad321ad | 1914 | spin_lock(&tree->lock); |
27a3507d | 1915 | if (cached && extent_state_in_tree(cached) && cached->start <= start && |
df98b6e2 | 1916 | cached->end > start) |
9655d298 CM |
1917 | node = &cached->rb_node; |
1918 | else | |
1919 | node = tree_search(tree, start); | |
d1310b2e CM |
1920 | while (node && start <= end) { |
1921 | state = rb_entry(node, struct extent_state, rb_node); | |
1922 | ||
1923 | if (filled && state->start > start) { | |
1924 | bitset = 0; | |
1925 | break; | |
1926 | } | |
1927 | ||
1928 | if (state->start > end) | |
1929 | break; | |
1930 | ||
1931 | if (state->state & bits) { | |
1932 | bitset = 1; | |
1933 | if (!filled) | |
1934 | break; | |
1935 | } else if (filled) { | |
1936 | bitset = 0; | |
1937 | break; | |
1938 | } | |
46562cec CM |
1939 | |
1940 | if (state->end == (u64)-1) | |
1941 | break; | |
1942 | ||
d1310b2e CM |
1943 | start = state->end + 1; |
1944 | if (start > end) | |
1945 | break; | |
1946 | node = rb_next(node); | |
1947 | if (!node) { | |
1948 | if (filled) | |
1949 | bitset = 0; | |
1950 | break; | |
1951 | } | |
1952 | } | |
cad321ad | 1953 | spin_unlock(&tree->lock); |
d1310b2e CM |
1954 | return bitset; |
1955 | } | |
d1310b2e CM |
1956 | |
1957 | /* | |
1958 | * helper function to set a given page up to date if all the | |
1959 | * extents in the tree for that page are up to date | |
1960 | */ | |
143bede5 | 1961 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e | 1962 | { |
4eee4fa4 | 1963 | u64 start = page_offset(page); |
d1310b2e | 1964 | u64 end = start + PAGE_CACHE_SIZE - 1; |
9655d298 | 1965 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 1966 | SetPageUptodate(page); |
d1310b2e CM |
1967 | } |
1968 | ||
8b110e39 | 1969 | int free_io_failure(struct inode *inode, struct io_failure_record *rec) |
4a54c8c1 JS |
1970 | { |
1971 | int ret; | |
1972 | int err = 0; | |
1973 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
1974 | ||
1975 | set_state_private(failure_tree, rec->start, 0); | |
1976 | ret = clear_extent_bits(failure_tree, rec->start, | |
1977 | rec->start + rec->len - 1, | |
1978 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1979 | if (ret) | |
1980 | err = ret; | |
1981 | ||
53b381b3 DW |
1982 | ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start, |
1983 | rec->start + rec->len - 1, | |
1984 | EXTENT_DAMAGED, GFP_NOFS); | |
1985 | if (ret && !err) | |
1986 | err = ret; | |
4a54c8c1 JS |
1987 | |
1988 | kfree(rec); | |
1989 | return err; | |
1990 | } | |
1991 | ||
4a54c8c1 JS |
1992 | /* |
1993 | * this bypasses the standard btrfs submit functions deliberately, as | |
1994 | * the standard behavior is to write all copies in a raid setup. here we only | |
1995 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
1996 | * submit_bio directly. | |
3ec706c8 | 1997 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
1998 | * actually prevents the read that triggered the error from finishing. |
1999 | * currently, there can be no more than two copies of every data bit. thus, | |
2000 | * exactly one rewrite is required. | |
2001 | */ | |
1203b681 MX |
2002 | int repair_io_failure(struct inode *inode, u64 start, u64 length, u64 logical, |
2003 | struct page *page, unsigned int pg_offset, int mirror_num) | |
4a54c8c1 | 2004 | { |
1203b681 | 2005 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; |
4a54c8c1 JS |
2006 | struct bio *bio; |
2007 | struct btrfs_device *dev; | |
4a54c8c1 JS |
2008 | u64 map_length = 0; |
2009 | u64 sector; | |
2010 | struct btrfs_bio *bbio = NULL; | |
53b381b3 | 2011 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
4a54c8c1 JS |
2012 | int ret; |
2013 | ||
908960c6 | 2014 | ASSERT(!(fs_info->sb->s_flags & MS_RDONLY)); |
4a54c8c1 JS |
2015 | BUG_ON(!mirror_num); |
2016 | ||
53b381b3 DW |
2017 | /* we can't repair anything in raid56 yet */ |
2018 | if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num)) | |
2019 | return 0; | |
2020 | ||
9be3395b | 2021 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
4a54c8c1 JS |
2022 | if (!bio) |
2023 | return -EIO; | |
4f024f37 | 2024 | bio->bi_iter.bi_size = 0; |
4a54c8c1 JS |
2025 | map_length = length; |
2026 | ||
3ec706c8 | 2027 | ret = btrfs_map_block(fs_info, WRITE, logical, |
4a54c8c1 JS |
2028 | &map_length, &bbio, mirror_num); |
2029 | if (ret) { | |
2030 | bio_put(bio); | |
2031 | return -EIO; | |
2032 | } | |
2033 | BUG_ON(mirror_num != bbio->mirror_num); | |
2034 | sector = bbio->stripes[mirror_num-1].physical >> 9; | |
4f024f37 | 2035 | bio->bi_iter.bi_sector = sector; |
4a54c8c1 | 2036 | dev = bbio->stripes[mirror_num-1].dev; |
6e9606d2 | 2037 | btrfs_put_bbio(bbio); |
4a54c8c1 JS |
2038 | if (!dev || !dev->bdev || !dev->writeable) { |
2039 | bio_put(bio); | |
2040 | return -EIO; | |
2041 | } | |
2042 | bio->bi_bdev = dev->bdev; | |
ffdd2018 | 2043 | bio_add_page(bio, page, length, pg_offset); |
4a54c8c1 | 2044 | |
33879d45 | 2045 | if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) { |
4a54c8c1 JS |
2046 | /* try to remap that extent elsewhere? */ |
2047 | bio_put(bio); | |
442a4f63 | 2048 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
2049 | return -EIO; |
2050 | } | |
2051 | ||
b14af3b4 DS |
2052 | btrfs_info_rl_in_rcu(fs_info, |
2053 | "read error corrected: ino %llu off %llu (dev %s sector %llu)", | |
1203b681 MX |
2054 | btrfs_ino(inode), start, |
2055 | rcu_str_deref(dev->name), sector); | |
4a54c8c1 JS |
2056 | bio_put(bio); |
2057 | return 0; | |
2058 | } | |
2059 | ||
ea466794 JB |
2060 | int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb, |
2061 | int mirror_num) | |
2062 | { | |
ea466794 JB |
2063 | u64 start = eb->start; |
2064 | unsigned long i, num_pages = num_extent_pages(eb->start, eb->len); | |
d95603b2 | 2065 | int ret = 0; |
ea466794 | 2066 | |
908960c6 ID |
2067 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
2068 | return -EROFS; | |
2069 | ||
ea466794 | 2070 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 2071 | struct page *p = eb->pages[i]; |
1203b681 MX |
2072 | |
2073 | ret = repair_io_failure(root->fs_info->btree_inode, start, | |
2074 | PAGE_CACHE_SIZE, start, p, | |
2075 | start - page_offset(p), mirror_num); | |
ea466794 JB |
2076 | if (ret) |
2077 | break; | |
2078 | start += PAGE_CACHE_SIZE; | |
2079 | } | |
2080 | ||
2081 | return ret; | |
2082 | } | |
2083 | ||
4a54c8c1 JS |
2084 | /* |
2085 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2086 | * to see if we need to process or clean up an io_failure_record | |
2087 | */ | |
8b110e39 MX |
2088 | int clean_io_failure(struct inode *inode, u64 start, struct page *page, |
2089 | unsigned int pg_offset) | |
4a54c8c1 JS |
2090 | { |
2091 | u64 private; | |
2092 | u64 private_failure; | |
2093 | struct io_failure_record *failrec; | |
908960c6 | 2094 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; |
4a54c8c1 JS |
2095 | struct extent_state *state; |
2096 | int num_copies; | |
4a54c8c1 | 2097 | int ret; |
4a54c8c1 JS |
2098 | |
2099 | private = 0; | |
2100 | ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
2101 | (u64)-1, 1, EXTENT_DIRTY, 0); | |
2102 | if (!ret) | |
2103 | return 0; | |
2104 | ||
2105 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start, | |
2106 | &private_failure); | |
2107 | if (ret) | |
2108 | return 0; | |
2109 | ||
2110 | failrec = (struct io_failure_record *)(unsigned long) private_failure; | |
2111 | BUG_ON(!failrec->this_mirror); | |
2112 | ||
2113 | if (failrec->in_validation) { | |
2114 | /* there was no real error, just free the record */ | |
2115 | pr_debug("clean_io_failure: freeing dummy error at %llu\n", | |
2116 | failrec->start); | |
4a54c8c1 JS |
2117 | goto out; |
2118 | } | |
908960c6 ID |
2119 | if (fs_info->sb->s_flags & MS_RDONLY) |
2120 | goto out; | |
4a54c8c1 JS |
2121 | |
2122 | spin_lock(&BTRFS_I(inode)->io_tree.lock); | |
2123 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, | |
2124 | failrec->start, | |
2125 | EXTENT_LOCKED); | |
2126 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); | |
2127 | ||
883d0de4 MX |
2128 | if (state && state->start <= failrec->start && |
2129 | state->end >= failrec->start + failrec->len - 1) { | |
3ec706c8 SB |
2130 | num_copies = btrfs_num_copies(fs_info, failrec->logical, |
2131 | failrec->len); | |
4a54c8c1 | 2132 | if (num_copies > 1) { |
1203b681 | 2133 | repair_io_failure(inode, start, failrec->len, |
454ff3de | 2134 | failrec->logical, page, |
1203b681 | 2135 | pg_offset, failrec->failed_mirror); |
4a54c8c1 JS |
2136 | } |
2137 | } | |
2138 | ||
2139 | out: | |
454ff3de | 2140 | free_io_failure(inode, failrec); |
4a54c8c1 | 2141 | |
454ff3de | 2142 | return 0; |
4a54c8c1 JS |
2143 | } |
2144 | ||
f612496b MX |
2145 | /* |
2146 | * Can be called when | |
2147 | * - hold extent lock | |
2148 | * - under ordered extent | |
2149 | * - the inode is freeing | |
2150 | */ | |
2151 | void btrfs_free_io_failure_record(struct inode *inode, u64 start, u64 end) | |
2152 | { | |
2153 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
2154 | struct io_failure_record *failrec; | |
2155 | struct extent_state *state, *next; | |
2156 | ||
2157 | if (RB_EMPTY_ROOT(&failure_tree->state)) | |
2158 | return; | |
2159 | ||
2160 | spin_lock(&failure_tree->lock); | |
2161 | state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY); | |
2162 | while (state) { | |
2163 | if (state->start > end) | |
2164 | break; | |
2165 | ||
2166 | ASSERT(state->end <= end); | |
2167 | ||
2168 | next = next_state(state); | |
2169 | ||
6e1103a6 | 2170 | failrec = (struct io_failure_record *)(unsigned long)state->private; |
f612496b MX |
2171 | free_extent_state(state); |
2172 | kfree(failrec); | |
2173 | ||
2174 | state = next; | |
2175 | } | |
2176 | spin_unlock(&failure_tree->lock); | |
2177 | } | |
2178 | ||
2fe6303e MX |
2179 | int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end, |
2180 | struct io_failure_record **failrec_ret) | |
4a54c8c1 | 2181 | { |
2fe6303e | 2182 | struct io_failure_record *failrec; |
4a54c8c1 JS |
2183 | u64 private; |
2184 | struct extent_map *em; | |
4a54c8c1 JS |
2185 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2186 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2187 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
4a54c8c1 | 2188 | int ret; |
4a54c8c1 JS |
2189 | u64 logical; |
2190 | ||
4a54c8c1 JS |
2191 | ret = get_state_private(failure_tree, start, &private); |
2192 | if (ret) { | |
2193 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2194 | if (!failrec) | |
2195 | return -ENOMEM; | |
2fe6303e | 2196 | |
4a54c8c1 JS |
2197 | failrec->start = start; |
2198 | failrec->len = end - start + 1; | |
2199 | failrec->this_mirror = 0; | |
2200 | failrec->bio_flags = 0; | |
2201 | failrec->in_validation = 0; | |
2202 | ||
2203 | read_lock(&em_tree->lock); | |
2204 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2205 | if (!em) { | |
2206 | read_unlock(&em_tree->lock); | |
2207 | kfree(failrec); | |
2208 | return -EIO; | |
2209 | } | |
2210 | ||
68ba990f | 2211 | if (em->start > start || em->start + em->len <= start) { |
4a54c8c1 JS |
2212 | free_extent_map(em); |
2213 | em = NULL; | |
2214 | } | |
2215 | read_unlock(&em_tree->lock); | |
7a2d6a64 | 2216 | if (!em) { |
4a54c8c1 JS |
2217 | kfree(failrec); |
2218 | return -EIO; | |
2219 | } | |
2fe6303e | 2220 | |
4a54c8c1 JS |
2221 | logical = start - em->start; |
2222 | logical = em->block_start + logical; | |
2223 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2224 | logical = em->block_start; | |
2225 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2226 | extent_set_compress_type(&failrec->bio_flags, | |
2227 | em->compress_type); | |
2228 | } | |
2fe6303e MX |
2229 | |
2230 | pr_debug("Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu\n", | |
2231 | logical, start, failrec->len); | |
2232 | ||
4a54c8c1 JS |
2233 | failrec->logical = logical; |
2234 | free_extent_map(em); | |
2235 | ||
2236 | /* set the bits in the private failure tree */ | |
2237 | ret = set_extent_bits(failure_tree, start, end, | |
2238 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
2239 | if (ret >= 0) | |
2240 | ret = set_state_private(failure_tree, start, | |
2241 | (u64)(unsigned long)failrec); | |
2242 | /* set the bits in the inode's tree */ | |
2243 | if (ret >= 0) | |
2244 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED, | |
2245 | GFP_NOFS); | |
2246 | if (ret < 0) { | |
2247 | kfree(failrec); | |
2248 | return ret; | |
2249 | } | |
2250 | } else { | |
2251 | failrec = (struct io_failure_record *)(unsigned long)private; | |
2fe6303e | 2252 | pr_debug("Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d\n", |
4a54c8c1 JS |
2253 | failrec->logical, failrec->start, failrec->len, |
2254 | failrec->in_validation); | |
2255 | /* | |
2256 | * when data can be on disk more than twice, add to failrec here | |
2257 | * (e.g. with a list for failed_mirror) to make | |
2258 | * clean_io_failure() clean all those errors at once. | |
2259 | */ | |
2260 | } | |
2fe6303e MX |
2261 | |
2262 | *failrec_ret = failrec; | |
2263 | ||
2264 | return 0; | |
2265 | } | |
2266 | ||
2267 | int btrfs_check_repairable(struct inode *inode, struct bio *failed_bio, | |
2268 | struct io_failure_record *failrec, int failed_mirror) | |
2269 | { | |
2270 | int num_copies; | |
2271 | ||
5d964051 SB |
2272 | num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info, |
2273 | failrec->logical, failrec->len); | |
4a54c8c1 JS |
2274 | if (num_copies == 1) { |
2275 | /* | |
2276 | * we only have a single copy of the data, so don't bother with | |
2277 | * all the retry and error correction code that follows. no | |
2278 | * matter what the error is, it is very likely to persist. | |
2279 | */ | |
2fe6303e | 2280 | pr_debug("Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n", |
09a7f7a2 | 2281 | num_copies, failrec->this_mirror, failed_mirror); |
2fe6303e | 2282 | return 0; |
4a54c8c1 JS |
2283 | } |
2284 | ||
4a54c8c1 JS |
2285 | /* |
2286 | * there are two premises: | |
2287 | * a) deliver good data to the caller | |
2288 | * b) correct the bad sectors on disk | |
2289 | */ | |
2290 | if (failed_bio->bi_vcnt > 1) { | |
2291 | /* | |
2292 | * to fulfill b), we need to know the exact failing sectors, as | |
2293 | * we don't want to rewrite any more than the failed ones. thus, | |
2294 | * we need separate read requests for the failed bio | |
2295 | * | |
2296 | * if the following BUG_ON triggers, our validation request got | |
2297 | * merged. we need separate requests for our algorithm to work. | |
2298 | */ | |
2299 | BUG_ON(failrec->in_validation); | |
2300 | failrec->in_validation = 1; | |
2301 | failrec->this_mirror = failed_mirror; | |
4a54c8c1 JS |
2302 | } else { |
2303 | /* | |
2304 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2305 | * of the failed sector and if we succeed, we have setup | |
2306 | * everything for repair_io_failure to do the rest for us. | |
2307 | */ | |
2308 | if (failrec->in_validation) { | |
2309 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2310 | failrec->in_validation = 0; | |
2311 | failrec->this_mirror = 0; | |
2312 | } | |
2313 | failrec->failed_mirror = failed_mirror; | |
2314 | failrec->this_mirror++; | |
2315 | if (failrec->this_mirror == failed_mirror) | |
2316 | failrec->this_mirror++; | |
4a54c8c1 JS |
2317 | } |
2318 | ||
facc8a22 | 2319 | if (failrec->this_mirror > num_copies) { |
2fe6303e | 2320 | pr_debug("Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n", |
4a54c8c1 | 2321 | num_copies, failrec->this_mirror, failed_mirror); |
2fe6303e | 2322 | return 0; |
4a54c8c1 JS |
2323 | } |
2324 | ||
2fe6303e MX |
2325 | return 1; |
2326 | } | |
2327 | ||
2328 | ||
2329 | struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio, | |
2330 | struct io_failure_record *failrec, | |
2331 | struct page *page, int pg_offset, int icsum, | |
8b110e39 | 2332 | bio_end_io_t *endio_func, void *data) |
2fe6303e MX |
2333 | { |
2334 | struct bio *bio; | |
2335 | struct btrfs_io_bio *btrfs_failed_bio; | |
2336 | struct btrfs_io_bio *btrfs_bio; | |
2337 | ||
9be3395b | 2338 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
2fe6303e MX |
2339 | if (!bio) |
2340 | return NULL; | |
2341 | ||
2342 | bio->bi_end_io = endio_func; | |
4f024f37 | 2343 | bio->bi_iter.bi_sector = failrec->logical >> 9; |
4a54c8c1 | 2344 | bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; |
4f024f37 | 2345 | bio->bi_iter.bi_size = 0; |
8b110e39 | 2346 | bio->bi_private = data; |
4a54c8c1 | 2347 | |
facc8a22 MX |
2348 | btrfs_failed_bio = btrfs_io_bio(failed_bio); |
2349 | if (btrfs_failed_bio->csum) { | |
2350 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2351 | u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); | |
2352 | ||
2353 | btrfs_bio = btrfs_io_bio(bio); | |
2354 | btrfs_bio->csum = btrfs_bio->csum_inline; | |
2fe6303e MX |
2355 | icsum *= csum_size; |
2356 | memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum, | |
facc8a22 MX |
2357 | csum_size); |
2358 | } | |
2359 | ||
2fe6303e MX |
2360 | bio_add_page(bio, page, failrec->len, pg_offset); |
2361 | ||
2362 | return bio; | |
2363 | } | |
2364 | ||
2365 | /* | |
2366 | * this is a generic handler for readpage errors (default | |
2367 | * readpage_io_failed_hook). if other copies exist, read those and write back | |
2368 | * good data to the failed position. does not investigate in remapping the | |
2369 | * failed extent elsewhere, hoping the device will be smart enough to do this as | |
2370 | * needed | |
2371 | */ | |
2372 | ||
2373 | static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset, | |
2374 | struct page *page, u64 start, u64 end, | |
2375 | int failed_mirror) | |
2376 | { | |
2377 | struct io_failure_record *failrec; | |
2378 | struct inode *inode = page->mapping->host; | |
2379 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2380 | struct bio *bio; | |
2381 | int read_mode; | |
2382 | int ret; | |
2383 | ||
2384 | BUG_ON(failed_bio->bi_rw & REQ_WRITE); | |
2385 | ||
2386 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
2387 | if (ret) | |
2388 | return ret; | |
2389 | ||
2390 | ret = btrfs_check_repairable(inode, failed_bio, failrec, failed_mirror); | |
2391 | if (!ret) { | |
2392 | free_io_failure(inode, failrec); | |
2393 | return -EIO; | |
2394 | } | |
2395 | ||
2396 | if (failed_bio->bi_vcnt > 1) | |
2397 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; | |
2398 | else | |
2399 | read_mode = READ_SYNC; | |
2400 | ||
2401 | phy_offset >>= inode->i_sb->s_blocksize_bits; | |
2402 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2403 | start - page_offset(page), | |
8b110e39 MX |
2404 | (int)phy_offset, failed_bio->bi_end_io, |
2405 | NULL); | |
2fe6303e MX |
2406 | if (!bio) { |
2407 | free_io_failure(inode, failrec); | |
2408 | return -EIO; | |
2409 | } | |
4a54c8c1 | 2410 | |
2fe6303e MX |
2411 | pr_debug("Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d\n", |
2412 | read_mode, failrec->this_mirror, failrec->in_validation); | |
4a54c8c1 | 2413 | |
013bd4c3 TI |
2414 | ret = tree->ops->submit_bio_hook(inode, read_mode, bio, |
2415 | failrec->this_mirror, | |
2416 | failrec->bio_flags, 0); | |
6c387ab2 | 2417 | if (ret) { |
454ff3de | 2418 | free_io_failure(inode, failrec); |
6c387ab2 MX |
2419 | bio_put(bio); |
2420 | } | |
2421 | ||
013bd4c3 | 2422 | return ret; |
4a54c8c1 JS |
2423 | } |
2424 | ||
d1310b2e CM |
2425 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2426 | ||
b5227c07 | 2427 | void end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
87826df0 JM |
2428 | { |
2429 | int uptodate = (err == 0); | |
2430 | struct extent_io_tree *tree; | |
3e2426bd | 2431 | int ret = 0; |
87826df0 JM |
2432 | |
2433 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
2434 | ||
2435 | if (tree->ops && tree->ops->writepage_end_io_hook) { | |
2436 | ret = tree->ops->writepage_end_io_hook(page, start, | |
2437 | end, NULL, uptodate); | |
2438 | if (ret) | |
2439 | uptodate = 0; | |
2440 | } | |
2441 | ||
87826df0 | 2442 | if (!uptodate) { |
87826df0 JM |
2443 | ClearPageUptodate(page); |
2444 | SetPageError(page); | |
5dca6eea LB |
2445 | ret = ret < 0 ? ret : -EIO; |
2446 | mapping_set_error(page->mapping, ret); | |
87826df0 | 2447 | } |
87826df0 JM |
2448 | } |
2449 | ||
d1310b2e CM |
2450 | /* |
2451 | * after a writepage IO is done, we need to: | |
2452 | * clear the uptodate bits on error | |
2453 | * clear the writeback bits in the extent tree for this IO | |
2454 | * end_page_writeback if the page has no more pending IO | |
2455 | * | |
2456 | * Scheduling is not allowed, so the extent state tree is expected | |
2457 | * to have one and only one object corresponding to this IO. | |
2458 | */ | |
4246a0b6 | 2459 | static void end_bio_extent_writepage(struct bio *bio) |
d1310b2e | 2460 | { |
2c30c71b | 2461 | struct bio_vec *bvec; |
d1310b2e CM |
2462 | u64 start; |
2463 | u64 end; | |
2c30c71b | 2464 | int i; |
d1310b2e | 2465 | |
2c30c71b | 2466 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2467 | struct page *page = bvec->bv_page; |
902b22f3 | 2468 | |
17a5adcc AO |
2469 | /* We always issue full-page reads, but if some block |
2470 | * in a page fails to read, blk_update_request() will | |
2471 | * advance bv_offset and adjust bv_len to compensate. | |
2472 | * Print a warning for nonzero offsets, and an error | |
2473 | * if they don't add up to a full page. */ | |
efe120a0 FH |
2474 | if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) { |
2475 | if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE) | |
2476 | btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info, | |
2477 | "partial page write in btrfs with offset %u and length %u", | |
2478 | bvec->bv_offset, bvec->bv_len); | |
2479 | else | |
2480 | btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info, | |
2481 | "incomplete page write in btrfs with offset %u and " | |
2482 | "length %u", | |
2483 | bvec->bv_offset, bvec->bv_len); | |
2484 | } | |
d1310b2e | 2485 | |
17a5adcc AO |
2486 | start = page_offset(page); |
2487 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
d1310b2e | 2488 | |
b5227c07 | 2489 | end_extent_writepage(page, bio->bi_error, start, end); |
17a5adcc | 2490 | end_page_writeback(page); |
2c30c71b | 2491 | } |
2b1f55b0 | 2492 | |
d1310b2e | 2493 | bio_put(bio); |
d1310b2e CM |
2494 | } |
2495 | ||
883d0de4 MX |
2496 | static void |
2497 | endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len, | |
2498 | int uptodate) | |
2499 | { | |
2500 | struct extent_state *cached = NULL; | |
2501 | u64 end = start + len - 1; | |
2502 | ||
2503 | if (uptodate && tree->track_uptodate) | |
2504 | set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC); | |
2505 | unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC); | |
2506 | } | |
2507 | ||
d1310b2e CM |
2508 | /* |
2509 | * after a readpage IO is done, we need to: | |
2510 | * clear the uptodate bits on error | |
2511 | * set the uptodate bits if things worked | |
2512 | * set the page up to date if all extents in the tree are uptodate | |
2513 | * clear the lock bit in the extent tree | |
2514 | * unlock the page if there are no other extents locked for it | |
2515 | * | |
2516 | * Scheduling is not allowed, so the extent state tree is expected | |
2517 | * to have one and only one object corresponding to this IO. | |
2518 | */ | |
4246a0b6 | 2519 | static void end_bio_extent_readpage(struct bio *bio) |
d1310b2e | 2520 | { |
2c30c71b | 2521 | struct bio_vec *bvec; |
4246a0b6 | 2522 | int uptodate = !bio->bi_error; |
facc8a22 | 2523 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); |
902b22f3 | 2524 | struct extent_io_tree *tree; |
facc8a22 | 2525 | u64 offset = 0; |
d1310b2e CM |
2526 | u64 start; |
2527 | u64 end; | |
facc8a22 | 2528 | u64 len; |
883d0de4 MX |
2529 | u64 extent_start = 0; |
2530 | u64 extent_len = 0; | |
5cf1ab56 | 2531 | int mirror; |
d1310b2e | 2532 | int ret; |
2c30c71b | 2533 | int i; |
d1310b2e | 2534 | |
2c30c71b | 2535 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2536 | struct page *page = bvec->bv_page; |
a71754fc | 2537 | struct inode *inode = page->mapping->host; |
507903b8 | 2538 | |
be3940c0 | 2539 | pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, " |
4246a0b6 CH |
2540 | "mirror=%u\n", (u64)bio->bi_iter.bi_sector, |
2541 | bio->bi_error, io_bio->mirror_num); | |
a71754fc | 2542 | tree = &BTRFS_I(inode)->io_tree; |
902b22f3 | 2543 | |
17a5adcc AO |
2544 | /* We always issue full-page reads, but if some block |
2545 | * in a page fails to read, blk_update_request() will | |
2546 | * advance bv_offset and adjust bv_len to compensate. | |
2547 | * Print a warning for nonzero offsets, and an error | |
2548 | * if they don't add up to a full page. */ | |
efe120a0 FH |
2549 | if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) { |
2550 | if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE) | |
2551 | btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info, | |
2552 | "partial page read in btrfs with offset %u and length %u", | |
2553 | bvec->bv_offset, bvec->bv_len); | |
2554 | else | |
2555 | btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info, | |
2556 | "incomplete page read in btrfs with offset %u and " | |
2557 | "length %u", | |
2558 | bvec->bv_offset, bvec->bv_len); | |
2559 | } | |
d1310b2e | 2560 | |
17a5adcc AO |
2561 | start = page_offset(page); |
2562 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
facc8a22 | 2563 | len = bvec->bv_len; |
d1310b2e | 2564 | |
9be3395b | 2565 | mirror = io_bio->mirror_num; |
f2a09da9 MX |
2566 | if (likely(uptodate && tree->ops && |
2567 | tree->ops->readpage_end_io_hook)) { | |
facc8a22 MX |
2568 | ret = tree->ops->readpage_end_io_hook(io_bio, offset, |
2569 | page, start, end, | |
2570 | mirror); | |
5ee0844d | 2571 | if (ret) |
d1310b2e | 2572 | uptodate = 0; |
5ee0844d | 2573 | else |
1203b681 | 2574 | clean_io_failure(inode, start, page, 0); |
d1310b2e | 2575 | } |
ea466794 | 2576 | |
f2a09da9 MX |
2577 | if (likely(uptodate)) |
2578 | goto readpage_ok; | |
2579 | ||
2580 | if (tree->ops && tree->ops->readpage_io_failed_hook) { | |
5cf1ab56 | 2581 | ret = tree->ops->readpage_io_failed_hook(page, mirror); |
4246a0b6 | 2582 | if (!ret && !bio->bi_error) |
ea466794 | 2583 | uptodate = 1; |
f2a09da9 | 2584 | } else { |
f4a8e656 JS |
2585 | /* |
2586 | * The generic bio_readpage_error handles errors the | |
2587 | * following way: If possible, new read requests are | |
2588 | * created and submitted and will end up in | |
2589 | * end_bio_extent_readpage as well (if we're lucky, not | |
2590 | * in the !uptodate case). In that case it returns 0 and | |
2591 | * we just go on with the next page in our bio. If it | |
2592 | * can't handle the error it will return -EIO and we | |
2593 | * remain responsible for that page. | |
2594 | */ | |
facc8a22 MX |
2595 | ret = bio_readpage_error(bio, offset, page, start, end, |
2596 | mirror); | |
7e38326f | 2597 | if (ret == 0) { |
4246a0b6 | 2598 | uptodate = !bio->bi_error; |
38c1c2e4 | 2599 | offset += len; |
7e38326f CM |
2600 | continue; |
2601 | } | |
2602 | } | |
f2a09da9 | 2603 | readpage_ok: |
883d0de4 | 2604 | if (likely(uptodate)) { |
a71754fc JB |
2605 | loff_t i_size = i_size_read(inode); |
2606 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
a583c026 | 2607 | unsigned off; |
a71754fc JB |
2608 | |
2609 | /* Zero out the end if this page straddles i_size */ | |
a583c026 LB |
2610 | off = i_size & (PAGE_CACHE_SIZE-1); |
2611 | if (page->index == end_index && off) | |
2612 | zero_user_segment(page, off, PAGE_CACHE_SIZE); | |
17a5adcc | 2613 | SetPageUptodate(page); |
70dec807 | 2614 | } else { |
17a5adcc AO |
2615 | ClearPageUptodate(page); |
2616 | SetPageError(page); | |
70dec807 | 2617 | } |
17a5adcc | 2618 | unlock_page(page); |
facc8a22 | 2619 | offset += len; |
883d0de4 MX |
2620 | |
2621 | if (unlikely(!uptodate)) { | |
2622 | if (extent_len) { | |
2623 | endio_readpage_release_extent(tree, | |
2624 | extent_start, | |
2625 | extent_len, 1); | |
2626 | extent_start = 0; | |
2627 | extent_len = 0; | |
2628 | } | |
2629 | endio_readpage_release_extent(tree, start, | |
2630 | end - start + 1, 0); | |
2631 | } else if (!extent_len) { | |
2632 | extent_start = start; | |
2633 | extent_len = end + 1 - start; | |
2634 | } else if (extent_start + extent_len == start) { | |
2635 | extent_len += end + 1 - start; | |
2636 | } else { | |
2637 | endio_readpage_release_extent(tree, extent_start, | |
2638 | extent_len, uptodate); | |
2639 | extent_start = start; | |
2640 | extent_len = end + 1 - start; | |
2641 | } | |
2c30c71b | 2642 | } |
d1310b2e | 2643 | |
883d0de4 MX |
2644 | if (extent_len) |
2645 | endio_readpage_release_extent(tree, extent_start, extent_len, | |
2646 | uptodate); | |
facc8a22 | 2647 | if (io_bio->end_io) |
4246a0b6 | 2648 | io_bio->end_io(io_bio, bio->bi_error); |
d1310b2e | 2649 | bio_put(bio); |
d1310b2e CM |
2650 | } |
2651 | ||
9be3395b CM |
2652 | /* |
2653 | * this allocates from the btrfs_bioset. We're returning a bio right now | |
2654 | * but you can call btrfs_io_bio for the appropriate container_of magic | |
2655 | */ | |
88f794ed MX |
2656 | struct bio * |
2657 | btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
2658 | gfp_t gfp_flags) | |
d1310b2e | 2659 | { |
facc8a22 | 2660 | struct btrfs_io_bio *btrfs_bio; |
d1310b2e CM |
2661 | struct bio *bio; |
2662 | ||
9be3395b | 2663 | bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset); |
d1310b2e CM |
2664 | |
2665 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
9be3395b CM |
2666 | while (!bio && (nr_vecs /= 2)) { |
2667 | bio = bio_alloc_bioset(gfp_flags, | |
2668 | nr_vecs, btrfs_bioset); | |
2669 | } | |
d1310b2e CM |
2670 | } |
2671 | ||
2672 | if (bio) { | |
2673 | bio->bi_bdev = bdev; | |
4f024f37 | 2674 | bio->bi_iter.bi_sector = first_sector; |
facc8a22 MX |
2675 | btrfs_bio = btrfs_io_bio(bio); |
2676 | btrfs_bio->csum = NULL; | |
2677 | btrfs_bio->csum_allocated = NULL; | |
2678 | btrfs_bio->end_io = NULL; | |
d1310b2e CM |
2679 | } |
2680 | return bio; | |
2681 | } | |
2682 | ||
9be3395b CM |
2683 | struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask) |
2684 | { | |
23ea8e5a MX |
2685 | struct btrfs_io_bio *btrfs_bio; |
2686 | struct bio *new; | |
9be3395b | 2687 | |
23ea8e5a MX |
2688 | new = bio_clone_bioset(bio, gfp_mask, btrfs_bioset); |
2689 | if (new) { | |
2690 | btrfs_bio = btrfs_io_bio(new); | |
2691 | btrfs_bio->csum = NULL; | |
2692 | btrfs_bio->csum_allocated = NULL; | |
2693 | btrfs_bio->end_io = NULL; | |
3a9508b0 CM |
2694 | |
2695 | #ifdef CONFIG_BLK_CGROUP | |
da2f0f74 CM |
2696 | /* FIXME, put this into bio_clone_bioset */ |
2697 | if (bio->bi_css) | |
2698 | bio_associate_blkcg(new, bio->bi_css); | |
3a9508b0 | 2699 | #endif |
23ea8e5a MX |
2700 | } |
2701 | return new; | |
2702 | } | |
9be3395b CM |
2703 | |
2704 | /* this also allocates from the btrfs_bioset */ | |
2705 | struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs) | |
2706 | { | |
facc8a22 MX |
2707 | struct btrfs_io_bio *btrfs_bio; |
2708 | struct bio *bio; | |
2709 | ||
2710 | bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset); | |
2711 | if (bio) { | |
2712 | btrfs_bio = btrfs_io_bio(bio); | |
2713 | btrfs_bio->csum = NULL; | |
2714 | btrfs_bio->csum_allocated = NULL; | |
2715 | btrfs_bio->end_io = NULL; | |
2716 | } | |
2717 | return bio; | |
9be3395b CM |
2718 | } |
2719 | ||
2720 | ||
355808c2 JM |
2721 | static int __must_check submit_one_bio(int rw, struct bio *bio, |
2722 | int mirror_num, unsigned long bio_flags) | |
d1310b2e | 2723 | { |
d1310b2e | 2724 | int ret = 0; |
70dec807 CM |
2725 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
2726 | struct page *page = bvec->bv_page; | |
2727 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 2728 | u64 start; |
70dec807 | 2729 | |
4eee4fa4 | 2730 | start = page_offset(page) + bvec->bv_offset; |
70dec807 | 2731 | |
902b22f3 | 2732 | bio->bi_private = NULL; |
d1310b2e CM |
2733 | |
2734 | bio_get(bio); | |
2735 | ||
065631f6 | 2736 | if (tree->ops && tree->ops->submit_bio_hook) |
6b82ce8d | 2737 | ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio, |
eaf25d93 | 2738 | mirror_num, bio_flags, start); |
0b86a832 | 2739 | else |
21adbd5c | 2740 | btrfsic_submit_bio(rw, bio); |
4a54c8c1 | 2741 | |
d1310b2e CM |
2742 | bio_put(bio); |
2743 | return ret; | |
2744 | } | |
2745 | ||
64a16701 | 2746 | static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page, |
3444a972 JM |
2747 | unsigned long offset, size_t size, struct bio *bio, |
2748 | unsigned long bio_flags) | |
2749 | { | |
2750 | int ret = 0; | |
2751 | if (tree->ops && tree->ops->merge_bio_hook) | |
64a16701 | 2752 | ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio, |
3444a972 JM |
2753 | bio_flags); |
2754 | BUG_ON(ret < 0); | |
2755 | return ret; | |
2756 | ||
2757 | } | |
2758 | ||
d1310b2e | 2759 | static int submit_extent_page(int rw, struct extent_io_tree *tree, |
da2f0f74 | 2760 | struct writeback_control *wbc, |
d1310b2e CM |
2761 | struct page *page, sector_t sector, |
2762 | size_t size, unsigned long offset, | |
2763 | struct block_device *bdev, | |
2764 | struct bio **bio_ret, | |
2765 | unsigned long max_pages, | |
f188591e | 2766 | bio_end_io_t end_io_func, |
c8b97818 CM |
2767 | int mirror_num, |
2768 | unsigned long prev_bio_flags, | |
005efedf FM |
2769 | unsigned long bio_flags, |
2770 | bool force_bio_submit) | |
d1310b2e CM |
2771 | { |
2772 | int ret = 0; | |
2773 | struct bio *bio; | |
c8b97818 | 2774 | int contig = 0; |
c8b97818 | 2775 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; |
5b050f04 | 2776 | size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE); |
d1310b2e CM |
2777 | |
2778 | if (bio_ret && *bio_ret) { | |
2779 | bio = *bio_ret; | |
c8b97818 | 2780 | if (old_compressed) |
4f024f37 | 2781 | contig = bio->bi_iter.bi_sector == sector; |
c8b97818 | 2782 | else |
f73a1c7d | 2783 | contig = bio_end_sector(bio) == sector; |
c8b97818 CM |
2784 | |
2785 | if (prev_bio_flags != bio_flags || !contig || | |
005efedf | 2786 | force_bio_submit || |
64a16701 | 2787 | merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) || |
c8b97818 CM |
2788 | bio_add_page(bio, page, page_size, offset) < page_size) { |
2789 | ret = submit_one_bio(rw, bio, mirror_num, | |
2790 | prev_bio_flags); | |
289454ad NA |
2791 | if (ret < 0) { |
2792 | *bio_ret = NULL; | |
79787eaa | 2793 | return ret; |
289454ad | 2794 | } |
d1310b2e CM |
2795 | bio = NULL; |
2796 | } else { | |
da2f0f74 CM |
2797 | if (wbc) |
2798 | wbc_account_io(wbc, page, page_size); | |
d1310b2e CM |
2799 | return 0; |
2800 | } | |
2801 | } | |
c8b97818 | 2802 | |
b54ffb73 KO |
2803 | bio = btrfs_bio_alloc(bdev, sector, BIO_MAX_PAGES, |
2804 | GFP_NOFS | __GFP_HIGH); | |
5df67083 TI |
2805 | if (!bio) |
2806 | return -ENOMEM; | |
70dec807 | 2807 | |
c8b97818 | 2808 | bio_add_page(bio, page, page_size, offset); |
d1310b2e CM |
2809 | bio->bi_end_io = end_io_func; |
2810 | bio->bi_private = tree; | |
da2f0f74 CM |
2811 | if (wbc) { |
2812 | wbc_init_bio(wbc, bio); | |
2813 | wbc_account_io(wbc, page, page_size); | |
2814 | } | |
70dec807 | 2815 | |
d397712b | 2816 | if (bio_ret) |
d1310b2e | 2817 | *bio_ret = bio; |
d397712b | 2818 | else |
c8b97818 | 2819 | ret = submit_one_bio(rw, bio, mirror_num, bio_flags); |
d1310b2e CM |
2820 | |
2821 | return ret; | |
2822 | } | |
2823 | ||
48a3b636 ES |
2824 | static void attach_extent_buffer_page(struct extent_buffer *eb, |
2825 | struct page *page) | |
d1310b2e CM |
2826 | { |
2827 | if (!PagePrivate(page)) { | |
2828 | SetPagePrivate(page); | |
d1310b2e | 2829 | page_cache_get(page); |
4f2de97a JB |
2830 | set_page_private(page, (unsigned long)eb); |
2831 | } else { | |
2832 | WARN_ON(page->private != (unsigned long)eb); | |
d1310b2e CM |
2833 | } |
2834 | } | |
2835 | ||
4f2de97a | 2836 | void set_page_extent_mapped(struct page *page) |
d1310b2e | 2837 | { |
4f2de97a JB |
2838 | if (!PagePrivate(page)) { |
2839 | SetPagePrivate(page); | |
2840 | page_cache_get(page); | |
2841 | set_page_private(page, EXTENT_PAGE_PRIVATE); | |
2842 | } | |
d1310b2e CM |
2843 | } |
2844 | ||
125bac01 MX |
2845 | static struct extent_map * |
2846 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
2847 | u64 start, u64 len, get_extent_t *get_extent, | |
2848 | struct extent_map **em_cached) | |
2849 | { | |
2850 | struct extent_map *em; | |
2851 | ||
2852 | if (em_cached && *em_cached) { | |
2853 | em = *em_cached; | |
cbc0e928 | 2854 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 MX |
2855 | start < extent_map_end(em)) { |
2856 | atomic_inc(&em->refs); | |
2857 | return em; | |
2858 | } | |
2859 | ||
2860 | free_extent_map(em); | |
2861 | *em_cached = NULL; | |
2862 | } | |
2863 | ||
2864 | em = get_extent(inode, page, pg_offset, start, len, 0); | |
2865 | if (em_cached && !IS_ERR_OR_NULL(em)) { | |
2866 | BUG_ON(*em_cached); | |
2867 | atomic_inc(&em->refs); | |
2868 | *em_cached = em; | |
2869 | } | |
2870 | return em; | |
2871 | } | |
d1310b2e CM |
2872 | /* |
2873 | * basic readpage implementation. Locked extent state structs are inserted | |
2874 | * into the tree that are removed when the IO is done (by the end_io | |
2875 | * handlers) | |
79787eaa | 2876 | * XXX JDM: This needs looking at to ensure proper page locking |
d1310b2e | 2877 | */ |
9974090b MX |
2878 | static int __do_readpage(struct extent_io_tree *tree, |
2879 | struct page *page, | |
2880 | get_extent_t *get_extent, | |
125bac01 | 2881 | struct extent_map **em_cached, |
9974090b | 2882 | struct bio **bio, int mirror_num, |
005efedf FM |
2883 | unsigned long *bio_flags, int rw, |
2884 | u64 *prev_em_start) | |
d1310b2e CM |
2885 | { |
2886 | struct inode *inode = page->mapping->host; | |
4eee4fa4 | 2887 | u64 start = page_offset(page); |
d1310b2e CM |
2888 | u64 page_end = start + PAGE_CACHE_SIZE - 1; |
2889 | u64 end; | |
2890 | u64 cur = start; | |
2891 | u64 extent_offset; | |
2892 | u64 last_byte = i_size_read(inode); | |
2893 | u64 block_start; | |
2894 | u64 cur_end; | |
2895 | sector_t sector; | |
2896 | struct extent_map *em; | |
2897 | struct block_device *bdev; | |
2898 | int ret; | |
2899 | int nr = 0; | |
306e16ce | 2900 | size_t pg_offset = 0; |
d1310b2e | 2901 | size_t iosize; |
c8b97818 | 2902 | size_t disk_io_size; |
d1310b2e | 2903 | size_t blocksize = inode->i_sb->s_blocksize; |
7f042a83 | 2904 | unsigned long this_bio_flag = 0; |
d1310b2e CM |
2905 | |
2906 | set_page_extent_mapped(page); | |
2907 | ||
9974090b | 2908 | end = page_end; |
90a887c9 DM |
2909 | if (!PageUptodate(page)) { |
2910 | if (cleancache_get_page(page) == 0) { | |
2911 | BUG_ON(blocksize != PAGE_SIZE); | |
9974090b | 2912 | unlock_extent(tree, start, end); |
90a887c9 DM |
2913 | goto out; |
2914 | } | |
2915 | } | |
2916 | ||
c8b97818 CM |
2917 | if (page->index == last_byte >> PAGE_CACHE_SHIFT) { |
2918 | char *userpage; | |
2919 | size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1); | |
2920 | ||
2921 | if (zero_offset) { | |
2922 | iosize = PAGE_CACHE_SIZE - zero_offset; | |
7ac687d9 | 2923 | userpage = kmap_atomic(page); |
c8b97818 CM |
2924 | memset(userpage + zero_offset, 0, iosize); |
2925 | flush_dcache_page(page); | |
7ac687d9 | 2926 | kunmap_atomic(userpage); |
c8b97818 CM |
2927 | } |
2928 | } | |
d1310b2e | 2929 | while (cur <= end) { |
c8f2f24b | 2930 | unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1; |
005efedf | 2931 | bool force_bio_submit = false; |
c8f2f24b | 2932 | |
d1310b2e CM |
2933 | if (cur >= last_byte) { |
2934 | char *userpage; | |
507903b8 AJ |
2935 | struct extent_state *cached = NULL; |
2936 | ||
306e16ce | 2937 | iosize = PAGE_CACHE_SIZE - pg_offset; |
7ac687d9 | 2938 | userpage = kmap_atomic(page); |
306e16ce | 2939 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2940 | flush_dcache_page(page); |
7ac687d9 | 2941 | kunmap_atomic(userpage); |
d1310b2e | 2942 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 2943 | &cached, GFP_NOFS); |
7f042a83 FM |
2944 | unlock_extent_cached(tree, cur, |
2945 | cur + iosize - 1, | |
2946 | &cached, GFP_NOFS); | |
d1310b2e CM |
2947 | break; |
2948 | } | |
125bac01 MX |
2949 | em = __get_extent_map(inode, page, pg_offset, cur, |
2950 | end - cur + 1, get_extent, em_cached); | |
c704005d | 2951 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 2952 | SetPageError(page); |
7f042a83 | 2953 | unlock_extent(tree, cur, end); |
d1310b2e CM |
2954 | break; |
2955 | } | |
d1310b2e CM |
2956 | extent_offset = cur - em->start; |
2957 | BUG_ON(extent_map_end(em) <= cur); | |
2958 | BUG_ON(end < cur); | |
2959 | ||
261507a0 | 2960 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
4b384318 | 2961 | this_bio_flag |= EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
2962 | extent_set_compress_type(&this_bio_flag, |
2963 | em->compress_type); | |
2964 | } | |
c8b97818 | 2965 | |
d1310b2e CM |
2966 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
2967 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 2968 | iosize = ALIGN(iosize, blocksize); |
c8b97818 CM |
2969 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
2970 | disk_io_size = em->block_len; | |
2971 | sector = em->block_start >> 9; | |
2972 | } else { | |
2973 | sector = (em->block_start + extent_offset) >> 9; | |
2974 | disk_io_size = iosize; | |
2975 | } | |
d1310b2e CM |
2976 | bdev = em->bdev; |
2977 | block_start = em->block_start; | |
d899e052 YZ |
2978 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2979 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
2980 | |
2981 | /* | |
2982 | * If we have a file range that points to a compressed extent | |
2983 | * and it's followed by a consecutive file range that points to | |
2984 | * to the same compressed extent (possibly with a different | |
2985 | * offset and/or length, so it either points to the whole extent | |
2986 | * or only part of it), we must make sure we do not submit a | |
2987 | * single bio to populate the pages for the 2 ranges because | |
2988 | * this makes the compressed extent read zero out the pages | |
2989 | * belonging to the 2nd range. Imagine the following scenario: | |
2990 | * | |
2991 | * File layout | |
2992 | * [0 - 8K] [8K - 24K] | |
2993 | * | | | |
2994 | * | | | |
2995 | * points to extent X, points to extent X, | |
2996 | * offset 4K, length of 8K offset 0, length 16K | |
2997 | * | |
2998 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
2999 | * | |
3000 | * If the bio to read the compressed extent covers both ranges, | |
3001 | * it will decompress extent X into the pages belonging to the | |
3002 | * first range and then it will stop, zeroing out the remaining | |
3003 | * pages that belong to the other range that points to extent X. | |
3004 | * So here we make sure we submit 2 bios, one for the first | |
3005 | * range and another one for the third range. Both will target | |
3006 | * the same physical extent from disk, but we can't currently | |
3007 | * make the compressed bio endio callback populate the pages | |
3008 | * for both ranges because each compressed bio is tightly | |
3009 | * coupled with a single extent map, and each range can have | |
3010 | * an extent map with a different offset value relative to the | |
3011 | * uncompressed data of our extent and different lengths. This | |
3012 | * is a corner case so we prioritize correctness over | |
3013 | * non-optimal behavior (submitting 2 bios for the same extent). | |
3014 | */ | |
3015 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
3016 | prev_em_start && *prev_em_start != (u64)-1 && | |
3017 | *prev_em_start != em->orig_start) | |
3018 | force_bio_submit = true; | |
3019 | ||
3020 | if (prev_em_start) | |
3021 | *prev_em_start = em->orig_start; | |
3022 | ||
d1310b2e CM |
3023 | free_extent_map(em); |
3024 | em = NULL; | |
3025 | ||
3026 | /* we've found a hole, just zero and go on */ | |
3027 | if (block_start == EXTENT_MAP_HOLE) { | |
3028 | char *userpage; | |
507903b8 AJ |
3029 | struct extent_state *cached = NULL; |
3030 | ||
7ac687d9 | 3031 | userpage = kmap_atomic(page); |
306e16ce | 3032 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3033 | flush_dcache_page(page); |
7ac687d9 | 3034 | kunmap_atomic(userpage); |
d1310b2e CM |
3035 | |
3036 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 | 3037 | &cached, GFP_NOFS); |
7f042a83 FM |
3038 | unlock_extent_cached(tree, cur, |
3039 | cur + iosize - 1, | |
3040 | &cached, GFP_NOFS); | |
d1310b2e | 3041 | cur = cur + iosize; |
306e16ce | 3042 | pg_offset += iosize; |
d1310b2e CM |
3043 | continue; |
3044 | } | |
3045 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
3046 | if (test_range_bit(tree, cur, cur_end, |
3047 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 3048 | check_page_uptodate(tree, page); |
7f042a83 | 3049 | unlock_extent(tree, cur, cur + iosize - 1); |
d1310b2e | 3050 | cur = cur + iosize; |
306e16ce | 3051 | pg_offset += iosize; |
d1310b2e CM |
3052 | continue; |
3053 | } | |
70dec807 CM |
3054 | /* we have an inline extent but it didn't get marked up |
3055 | * to date. Error out | |
3056 | */ | |
3057 | if (block_start == EXTENT_MAP_INLINE) { | |
3058 | SetPageError(page); | |
7f042a83 | 3059 | unlock_extent(tree, cur, cur + iosize - 1); |
70dec807 | 3060 | cur = cur + iosize; |
306e16ce | 3061 | pg_offset += iosize; |
70dec807 CM |
3062 | continue; |
3063 | } | |
d1310b2e | 3064 | |
c8f2f24b | 3065 | pnr -= page->index; |
da2f0f74 | 3066 | ret = submit_extent_page(rw, tree, NULL, page, |
306e16ce | 3067 | sector, disk_io_size, pg_offset, |
89642229 | 3068 | bdev, bio, pnr, |
c8b97818 CM |
3069 | end_bio_extent_readpage, mirror_num, |
3070 | *bio_flags, | |
005efedf FM |
3071 | this_bio_flag, |
3072 | force_bio_submit); | |
c8f2f24b JB |
3073 | if (!ret) { |
3074 | nr++; | |
3075 | *bio_flags = this_bio_flag; | |
3076 | } else { | |
d1310b2e | 3077 | SetPageError(page); |
7f042a83 | 3078 | unlock_extent(tree, cur, cur + iosize - 1); |
edd33c99 | 3079 | } |
d1310b2e | 3080 | cur = cur + iosize; |
306e16ce | 3081 | pg_offset += iosize; |
d1310b2e | 3082 | } |
90a887c9 | 3083 | out: |
d1310b2e CM |
3084 | if (!nr) { |
3085 | if (!PageError(page)) | |
3086 | SetPageUptodate(page); | |
3087 | unlock_page(page); | |
3088 | } | |
3089 | return 0; | |
3090 | } | |
3091 | ||
9974090b MX |
3092 | static inline void __do_contiguous_readpages(struct extent_io_tree *tree, |
3093 | struct page *pages[], int nr_pages, | |
3094 | u64 start, u64 end, | |
3095 | get_extent_t *get_extent, | |
125bac01 | 3096 | struct extent_map **em_cached, |
9974090b | 3097 | struct bio **bio, int mirror_num, |
808f80b4 FM |
3098 | unsigned long *bio_flags, int rw, |
3099 | u64 *prev_em_start) | |
9974090b MX |
3100 | { |
3101 | struct inode *inode; | |
3102 | struct btrfs_ordered_extent *ordered; | |
3103 | int index; | |
3104 | ||
3105 | inode = pages[0]->mapping->host; | |
3106 | while (1) { | |
3107 | lock_extent(tree, start, end); | |
3108 | ordered = btrfs_lookup_ordered_range(inode, start, | |
3109 | end - start + 1); | |
3110 | if (!ordered) | |
3111 | break; | |
3112 | unlock_extent(tree, start, end); | |
3113 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3114 | btrfs_put_ordered_extent(ordered); | |
3115 | } | |
3116 | ||
3117 | for (index = 0; index < nr_pages; index++) { | |
125bac01 | 3118 | __do_readpage(tree, pages[index], get_extent, em_cached, bio, |
808f80b4 | 3119 | mirror_num, bio_flags, rw, prev_em_start); |
9974090b MX |
3120 | page_cache_release(pages[index]); |
3121 | } | |
3122 | } | |
3123 | ||
3124 | static void __extent_readpages(struct extent_io_tree *tree, | |
3125 | struct page *pages[], | |
3126 | int nr_pages, get_extent_t *get_extent, | |
125bac01 | 3127 | struct extent_map **em_cached, |
9974090b | 3128 | struct bio **bio, int mirror_num, |
808f80b4 FM |
3129 | unsigned long *bio_flags, int rw, |
3130 | u64 *prev_em_start) | |
9974090b | 3131 | { |
35a3621b | 3132 | u64 start = 0; |
9974090b MX |
3133 | u64 end = 0; |
3134 | u64 page_start; | |
3135 | int index; | |
35a3621b | 3136 | int first_index = 0; |
9974090b MX |
3137 | |
3138 | for (index = 0; index < nr_pages; index++) { | |
3139 | page_start = page_offset(pages[index]); | |
3140 | if (!end) { | |
3141 | start = page_start; | |
3142 | end = start + PAGE_CACHE_SIZE - 1; | |
3143 | first_index = index; | |
3144 | } else if (end + 1 == page_start) { | |
3145 | end += PAGE_CACHE_SIZE; | |
3146 | } else { | |
3147 | __do_contiguous_readpages(tree, &pages[first_index], | |
3148 | index - first_index, start, | |
125bac01 MX |
3149 | end, get_extent, em_cached, |
3150 | bio, mirror_num, bio_flags, | |
808f80b4 | 3151 | rw, prev_em_start); |
9974090b MX |
3152 | start = page_start; |
3153 | end = start + PAGE_CACHE_SIZE - 1; | |
3154 | first_index = index; | |
3155 | } | |
3156 | } | |
3157 | ||
3158 | if (end) | |
3159 | __do_contiguous_readpages(tree, &pages[first_index], | |
3160 | index - first_index, start, | |
125bac01 | 3161 | end, get_extent, em_cached, bio, |
808f80b4 FM |
3162 | mirror_num, bio_flags, rw, |
3163 | prev_em_start); | |
9974090b MX |
3164 | } |
3165 | ||
3166 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
3167 | struct page *page, | |
3168 | get_extent_t *get_extent, | |
3169 | struct bio **bio, int mirror_num, | |
3170 | unsigned long *bio_flags, int rw) | |
3171 | { | |
3172 | struct inode *inode = page->mapping->host; | |
3173 | struct btrfs_ordered_extent *ordered; | |
3174 | u64 start = page_offset(page); | |
3175 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
3176 | int ret; | |
3177 | ||
3178 | while (1) { | |
3179 | lock_extent(tree, start, end); | |
3180 | ordered = btrfs_lookup_ordered_extent(inode, start); | |
3181 | if (!ordered) | |
3182 | break; | |
3183 | unlock_extent(tree, start, end); | |
3184 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3185 | btrfs_put_ordered_extent(ordered); | |
3186 | } | |
3187 | ||
125bac01 | 3188 | ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num, |
005efedf | 3189 | bio_flags, rw, NULL); |
9974090b MX |
3190 | return ret; |
3191 | } | |
3192 | ||
d1310b2e | 3193 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, |
8ddc7d9c | 3194 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3195 | { |
3196 | struct bio *bio = NULL; | |
c8b97818 | 3197 | unsigned long bio_flags = 0; |
d1310b2e CM |
3198 | int ret; |
3199 | ||
8ddc7d9c | 3200 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, |
d4c7ca86 | 3201 | &bio_flags, READ); |
d1310b2e | 3202 | if (bio) |
8ddc7d9c | 3203 | ret = submit_one_bio(READ, bio, mirror_num, bio_flags); |
d1310b2e CM |
3204 | return ret; |
3205 | } | |
d1310b2e | 3206 | |
11c8349b CM |
3207 | static noinline void update_nr_written(struct page *page, |
3208 | struct writeback_control *wbc, | |
3209 | unsigned long nr_written) | |
3210 | { | |
3211 | wbc->nr_to_write -= nr_written; | |
3212 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && | |
3213 | wbc->range_start == 0 && wbc->range_end == LLONG_MAX)) | |
3214 | page->mapping->writeback_index = page->index + nr_written; | |
3215 | } | |
3216 | ||
d1310b2e | 3217 | /* |
40f76580 CM |
3218 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
3219 | * | |
3220 | * This returns 1 if our fill_delalloc function did all the work required | |
3221 | * to write the page (copy into inline extent). In this case the IO has | |
3222 | * been started and the page is already unlocked. | |
3223 | * | |
3224 | * This returns 0 if all went well (page still locked) | |
3225 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 3226 | */ |
40f76580 CM |
3227 | static noinline_for_stack int writepage_delalloc(struct inode *inode, |
3228 | struct page *page, struct writeback_control *wbc, | |
3229 | struct extent_page_data *epd, | |
3230 | u64 delalloc_start, | |
3231 | unsigned long *nr_written) | |
3232 | { | |
3233 | struct extent_io_tree *tree = epd->tree; | |
3234 | u64 page_end = delalloc_start + PAGE_CACHE_SIZE - 1; | |
3235 | u64 nr_delalloc; | |
3236 | u64 delalloc_to_write = 0; | |
3237 | u64 delalloc_end = 0; | |
3238 | int ret; | |
3239 | int page_started = 0; | |
3240 | ||
3241 | if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc) | |
3242 | return 0; | |
3243 | ||
3244 | while (delalloc_end < page_end) { | |
3245 | nr_delalloc = find_lock_delalloc_range(inode, tree, | |
3246 | page, | |
3247 | &delalloc_start, | |
3248 | &delalloc_end, | |
dcab6a3b | 3249 | BTRFS_MAX_EXTENT_SIZE); |
40f76580 CM |
3250 | if (nr_delalloc == 0) { |
3251 | delalloc_start = delalloc_end + 1; | |
3252 | continue; | |
3253 | } | |
3254 | ret = tree->ops->fill_delalloc(inode, page, | |
3255 | delalloc_start, | |
3256 | delalloc_end, | |
3257 | &page_started, | |
3258 | nr_written); | |
3259 | /* File system has been set read-only */ | |
3260 | if (ret) { | |
3261 | SetPageError(page); | |
3262 | /* fill_delalloc should be return < 0 for error | |
3263 | * but just in case, we use > 0 here meaning the | |
3264 | * IO is started, so we don't want to return > 0 | |
3265 | * unless things are going well. | |
3266 | */ | |
3267 | ret = ret < 0 ? ret : -EIO; | |
3268 | goto done; | |
3269 | } | |
3270 | /* | |
3271 | * delalloc_end is already one less than the total | |
3272 | * length, so we don't subtract one from | |
3273 | * PAGE_CACHE_SIZE | |
3274 | */ | |
3275 | delalloc_to_write += (delalloc_end - delalloc_start + | |
3276 | PAGE_CACHE_SIZE) >> | |
3277 | PAGE_CACHE_SHIFT; | |
3278 | delalloc_start = delalloc_end + 1; | |
3279 | } | |
3280 | if (wbc->nr_to_write < delalloc_to_write) { | |
3281 | int thresh = 8192; | |
3282 | ||
3283 | if (delalloc_to_write < thresh * 2) | |
3284 | thresh = delalloc_to_write; | |
3285 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3286 | thresh); | |
3287 | } | |
3288 | ||
3289 | /* did the fill delalloc function already unlock and start | |
3290 | * the IO? | |
3291 | */ | |
3292 | if (page_started) { | |
3293 | /* | |
3294 | * we've unlocked the page, so we can't update | |
3295 | * the mapping's writeback index, just update | |
3296 | * nr_to_write. | |
3297 | */ | |
3298 | wbc->nr_to_write -= *nr_written; | |
3299 | return 1; | |
3300 | } | |
3301 | ||
3302 | ret = 0; | |
3303 | ||
3304 | done: | |
3305 | return ret; | |
3306 | } | |
3307 | ||
3308 | /* | |
3309 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3310 | * and does the loop to map the page into extents and bios. | |
3311 | * | |
3312 | * We return 1 if the IO is started and the page is unlocked, | |
3313 | * 0 if all went well (page still locked) | |
3314 | * < 0 if there were errors (page still locked) | |
3315 | */ | |
3316 | static noinline_for_stack int __extent_writepage_io(struct inode *inode, | |
3317 | struct page *page, | |
3318 | struct writeback_control *wbc, | |
3319 | struct extent_page_data *epd, | |
3320 | loff_t i_size, | |
3321 | unsigned long nr_written, | |
3322 | int write_flags, int *nr_ret) | |
d1310b2e | 3323 | { |
d1310b2e | 3324 | struct extent_io_tree *tree = epd->tree; |
4eee4fa4 | 3325 | u64 start = page_offset(page); |
d1310b2e CM |
3326 | u64 page_end = start + PAGE_CACHE_SIZE - 1; |
3327 | u64 end; | |
3328 | u64 cur = start; | |
3329 | u64 extent_offset; | |
d1310b2e CM |
3330 | u64 block_start; |
3331 | u64 iosize; | |
3332 | sector_t sector; | |
2c64c53d | 3333 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
3334 | struct extent_map *em; |
3335 | struct block_device *bdev; | |
7f3c74fb | 3336 | size_t pg_offset = 0; |
d1310b2e | 3337 | size_t blocksize; |
40f76580 CM |
3338 | int ret = 0; |
3339 | int nr = 0; | |
3340 | bool compressed; | |
c8b97818 | 3341 | |
247e743c | 3342 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
3343 | ret = tree->ops->writepage_start_hook(page, start, |
3344 | page_end); | |
87826df0 JM |
3345 | if (ret) { |
3346 | /* Fixup worker will requeue */ | |
3347 | if (ret == -EBUSY) | |
3348 | wbc->pages_skipped++; | |
3349 | else | |
3350 | redirty_page_for_writepage(wbc, page); | |
40f76580 | 3351 | |
11c8349b | 3352 | update_nr_written(page, wbc, nr_written); |
247e743c | 3353 | unlock_page(page); |
40f76580 | 3354 | ret = 1; |
11c8349b | 3355 | goto done_unlocked; |
247e743c CM |
3356 | } |
3357 | } | |
3358 | ||
11c8349b CM |
3359 | /* |
3360 | * we don't want to touch the inode after unlocking the page, | |
3361 | * so we update the mapping writeback index now | |
3362 | */ | |
3363 | update_nr_written(page, wbc, nr_written + 1); | |
771ed689 | 3364 | |
d1310b2e | 3365 | end = page_end; |
40f76580 | 3366 | if (i_size <= start) { |
e6dcd2dc CM |
3367 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3368 | tree->ops->writepage_end_io_hook(page, start, | |
3369 | page_end, NULL, 1); | |
d1310b2e CM |
3370 | goto done; |
3371 | } | |
3372 | ||
d1310b2e CM |
3373 | blocksize = inode->i_sb->s_blocksize; |
3374 | ||
3375 | while (cur <= end) { | |
40f76580 CM |
3376 | u64 em_end; |
3377 | if (cur >= i_size) { | |
e6dcd2dc CM |
3378 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3379 | tree->ops->writepage_end_io_hook(page, cur, | |
3380 | page_end, NULL, 1); | |
d1310b2e CM |
3381 | break; |
3382 | } | |
7f3c74fb | 3383 | em = epd->get_extent(inode, page, pg_offset, cur, |
d1310b2e | 3384 | end - cur + 1, 1); |
c704005d | 3385 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3386 | SetPageError(page); |
61391d56 | 3387 | ret = PTR_ERR_OR_ZERO(em); |
d1310b2e CM |
3388 | break; |
3389 | } | |
3390 | ||
3391 | extent_offset = cur - em->start; | |
40f76580 CM |
3392 | em_end = extent_map_end(em); |
3393 | BUG_ON(em_end <= cur); | |
d1310b2e | 3394 | BUG_ON(end < cur); |
40f76580 | 3395 | iosize = min(em_end - cur, end - cur + 1); |
fda2832f | 3396 | iosize = ALIGN(iosize, blocksize); |
d1310b2e CM |
3397 | sector = (em->block_start + extent_offset) >> 9; |
3398 | bdev = em->bdev; | |
3399 | block_start = em->block_start; | |
c8b97818 | 3400 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
3401 | free_extent_map(em); |
3402 | em = NULL; | |
3403 | ||
c8b97818 CM |
3404 | /* |
3405 | * compressed and inline extents are written through other | |
3406 | * paths in the FS | |
3407 | */ | |
3408 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3409 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
3410 | /* |
3411 | * end_io notification does not happen here for | |
3412 | * compressed extents | |
3413 | */ | |
3414 | if (!compressed && tree->ops && | |
3415 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
3416 | tree->ops->writepage_end_io_hook(page, cur, |
3417 | cur + iosize - 1, | |
3418 | NULL, 1); | |
c8b97818 CM |
3419 | else if (compressed) { |
3420 | /* we don't want to end_page_writeback on | |
3421 | * a compressed extent. this happens | |
3422 | * elsewhere | |
3423 | */ | |
3424 | nr++; | |
3425 | } | |
3426 | ||
3427 | cur += iosize; | |
7f3c74fb | 3428 | pg_offset += iosize; |
d1310b2e CM |
3429 | continue; |
3430 | } | |
c8b97818 | 3431 | |
d1310b2e CM |
3432 | if (tree->ops && tree->ops->writepage_io_hook) { |
3433 | ret = tree->ops->writepage_io_hook(page, cur, | |
3434 | cur + iosize - 1); | |
3435 | } else { | |
3436 | ret = 0; | |
3437 | } | |
1259ab75 | 3438 | if (ret) { |
d1310b2e | 3439 | SetPageError(page); |
1259ab75 | 3440 | } else { |
40f76580 | 3441 | unsigned long max_nr = (i_size >> PAGE_CACHE_SHIFT) + 1; |
7f3c74fb | 3442 | |
d1310b2e CM |
3443 | set_range_writeback(tree, cur, cur + iosize - 1); |
3444 | if (!PageWriteback(page)) { | |
efe120a0 FH |
3445 | btrfs_err(BTRFS_I(inode)->root->fs_info, |
3446 | "page %lu not writeback, cur %llu end %llu", | |
c1c9ff7c | 3447 | page->index, cur, end); |
d1310b2e CM |
3448 | } |
3449 | ||
da2f0f74 | 3450 | ret = submit_extent_page(write_flags, tree, wbc, page, |
ffbd517d CM |
3451 | sector, iosize, pg_offset, |
3452 | bdev, &epd->bio, max_nr, | |
c8b97818 | 3453 | end_bio_extent_writepage, |
005efedf | 3454 | 0, 0, 0, false); |
d1310b2e CM |
3455 | if (ret) |
3456 | SetPageError(page); | |
3457 | } | |
3458 | cur = cur + iosize; | |
7f3c74fb | 3459 | pg_offset += iosize; |
d1310b2e CM |
3460 | nr++; |
3461 | } | |
40f76580 CM |
3462 | done: |
3463 | *nr_ret = nr; | |
3464 | ||
3465 | done_unlocked: | |
3466 | ||
3467 | /* drop our reference on any cached states */ | |
3468 | free_extent_state(cached_state); | |
3469 | return ret; | |
3470 | } | |
3471 | ||
3472 | /* | |
3473 | * the writepage semantics are similar to regular writepage. extent | |
3474 | * records are inserted to lock ranges in the tree, and as dirty areas | |
3475 | * are found, they are marked writeback. Then the lock bits are removed | |
3476 | * and the end_io handler clears the writeback ranges | |
3477 | */ | |
3478 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
3479 | void *data) | |
3480 | { | |
3481 | struct inode *inode = page->mapping->host; | |
3482 | struct extent_page_data *epd = data; | |
3483 | u64 start = page_offset(page); | |
3484 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
3485 | int ret; | |
3486 | int nr = 0; | |
3487 | size_t pg_offset = 0; | |
3488 | loff_t i_size = i_size_read(inode); | |
3489 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
3490 | int write_flags; | |
3491 | unsigned long nr_written = 0; | |
3492 | ||
3493 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3494 | write_flags = WRITE_SYNC; | |
3495 | else | |
3496 | write_flags = WRITE; | |
3497 | ||
3498 | trace___extent_writepage(page, inode, wbc); | |
3499 | ||
3500 | WARN_ON(!PageLocked(page)); | |
3501 | ||
3502 | ClearPageError(page); | |
3503 | ||
3504 | pg_offset = i_size & (PAGE_CACHE_SIZE - 1); | |
3505 | if (page->index > end_index || | |
3506 | (page->index == end_index && !pg_offset)) { | |
3507 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE); | |
3508 | unlock_page(page); | |
3509 | return 0; | |
3510 | } | |
3511 | ||
3512 | if (page->index == end_index) { | |
3513 | char *userpage; | |
3514 | ||
3515 | userpage = kmap_atomic(page); | |
3516 | memset(userpage + pg_offset, 0, | |
3517 | PAGE_CACHE_SIZE - pg_offset); | |
3518 | kunmap_atomic(userpage); | |
3519 | flush_dcache_page(page); | |
3520 | } | |
3521 | ||
3522 | pg_offset = 0; | |
3523 | ||
3524 | set_page_extent_mapped(page); | |
3525 | ||
3526 | ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written); | |
3527 | if (ret == 1) | |
3528 | goto done_unlocked; | |
3529 | if (ret) | |
3530 | goto done; | |
3531 | ||
3532 | ret = __extent_writepage_io(inode, page, wbc, epd, | |
3533 | i_size, nr_written, write_flags, &nr); | |
3534 | if (ret == 1) | |
3535 | goto done_unlocked; | |
3536 | ||
d1310b2e CM |
3537 | done: |
3538 | if (nr == 0) { | |
3539 | /* make sure the mapping tag for page dirty gets cleared */ | |
3540 | set_page_writeback(page); | |
3541 | end_page_writeback(page); | |
3542 | } | |
61391d56 FM |
3543 | if (PageError(page)) { |
3544 | ret = ret < 0 ? ret : -EIO; | |
3545 | end_extent_writepage(page, ret, start, page_end); | |
3546 | } | |
d1310b2e | 3547 | unlock_page(page); |
40f76580 | 3548 | return ret; |
771ed689 | 3549 | |
11c8349b | 3550 | done_unlocked: |
d1310b2e CM |
3551 | return 0; |
3552 | } | |
3553 | ||
fd8b2b61 | 3554 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 3555 | { |
74316201 N |
3556 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
3557 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
3558 | } |
3559 | ||
0e378df1 CM |
3560 | static noinline_for_stack int |
3561 | lock_extent_buffer_for_io(struct extent_buffer *eb, | |
3562 | struct btrfs_fs_info *fs_info, | |
3563 | struct extent_page_data *epd) | |
0b32f4bb JB |
3564 | { |
3565 | unsigned long i, num_pages; | |
3566 | int flush = 0; | |
3567 | int ret = 0; | |
3568 | ||
3569 | if (!btrfs_try_tree_write_lock(eb)) { | |
3570 | flush = 1; | |
3571 | flush_write_bio(epd); | |
3572 | btrfs_tree_lock(eb); | |
3573 | } | |
3574 | ||
3575 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3576 | btrfs_tree_unlock(eb); | |
3577 | if (!epd->sync_io) | |
3578 | return 0; | |
3579 | if (!flush) { | |
3580 | flush_write_bio(epd); | |
3581 | flush = 1; | |
3582 | } | |
a098d8e8 CM |
3583 | while (1) { |
3584 | wait_on_extent_buffer_writeback(eb); | |
3585 | btrfs_tree_lock(eb); | |
3586 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3587 | break; | |
0b32f4bb | 3588 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3589 | } |
3590 | } | |
3591 | ||
51561ffe JB |
3592 | /* |
3593 | * We need to do this to prevent races in people who check if the eb is | |
3594 | * under IO since we can end up having no IO bits set for a short period | |
3595 | * of time. | |
3596 | */ | |
3597 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
3598 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3599 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 3600 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 3601 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
e2d84521 MX |
3602 | __percpu_counter_add(&fs_info->dirty_metadata_bytes, |
3603 | -eb->len, | |
3604 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 3605 | ret = 1; |
51561ffe JB |
3606 | } else { |
3607 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
3608 | } |
3609 | ||
3610 | btrfs_tree_unlock(eb); | |
3611 | ||
3612 | if (!ret) | |
3613 | return ret; | |
3614 | ||
3615 | num_pages = num_extent_pages(eb->start, eb->len); | |
3616 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 3617 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3618 | |
3619 | if (!trylock_page(p)) { | |
3620 | if (!flush) { | |
3621 | flush_write_bio(epd); | |
3622 | flush = 1; | |
3623 | } | |
3624 | lock_page(p); | |
3625 | } | |
3626 | } | |
3627 | ||
3628 | return ret; | |
3629 | } | |
3630 | ||
3631 | static void end_extent_buffer_writeback(struct extent_buffer *eb) | |
3632 | { | |
3633 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
4e857c58 | 3634 | smp_mb__after_atomic(); |
0b32f4bb JB |
3635 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); |
3636 | } | |
3637 | ||
656f30db FM |
3638 | static void set_btree_ioerr(struct page *page) |
3639 | { | |
3640 | struct extent_buffer *eb = (struct extent_buffer *)page->private; | |
3641 | struct btrfs_inode *btree_ino = BTRFS_I(eb->fs_info->btree_inode); | |
3642 | ||
3643 | SetPageError(page); | |
3644 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) | |
3645 | return; | |
3646 | ||
3647 | /* | |
3648 | * If writeback for a btree extent that doesn't belong to a log tree | |
3649 | * failed, increment the counter transaction->eb_write_errors. | |
3650 | * We do this because while the transaction is running and before it's | |
3651 | * committing (when we call filemap_fdata[write|wait]_range against | |
3652 | * the btree inode), we might have | |
3653 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
3654 | * returns an error or an error happens during writeback, when we're | |
3655 | * committing the transaction we wouldn't know about it, since the pages | |
3656 | * can be no longer dirty nor marked anymore for writeback (if a | |
3657 | * subsequent modification to the extent buffer didn't happen before the | |
3658 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
3659 | * able to find the pages tagged with SetPageError at transaction | |
3660 | * commit time. So if this happens we must abort the transaction, | |
3661 | * otherwise we commit a super block with btree roots that point to | |
3662 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
3663 | * or the content of some node/leaf from a past generation that got | |
3664 | * cowed or deleted and is no longer valid. | |
3665 | * | |
3666 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
3667 | * not be enough - we need to distinguish between log tree extents vs | |
3668 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
3669 | * will catch and clear such errors in the mapping - and that call might | |
3670 | * be from a log sync and not from a transaction commit. Also, checking | |
3671 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
3672 | * not done and would not be reliable - the eb might have been released | |
3673 | * from memory and reading it back again means that flag would not be | |
3674 | * set (since it's a runtime flag, not persisted on disk). | |
3675 | * | |
3676 | * Using the flags below in the btree inode also makes us achieve the | |
3677 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
3678 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
3679 | * is called, the writeback for all dirty pages had already finished | |
3680 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
3681 | * filemap_fdatawait_range() would return success, as it could not know | |
3682 | * that writeback errors happened (the pages were no longer tagged for | |
3683 | * writeback). | |
3684 | */ | |
3685 | switch (eb->log_index) { | |
3686 | case -1: | |
3687 | set_bit(BTRFS_INODE_BTREE_ERR, &btree_ino->runtime_flags); | |
3688 | break; | |
3689 | case 0: | |
3690 | set_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags); | |
3691 | break; | |
3692 | case 1: | |
3693 | set_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags); | |
3694 | break; | |
3695 | default: | |
3696 | BUG(); /* unexpected, logic error */ | |
3697 | } | |
3698 | } | |
3699 | ||
4246a0b6 | 3700 | static void end_bio_extent_buffer_writepage(struct bio *bio) |
0b32f4bb | 3701 | { |
2c30c71b | 3702 | struct bio_vec *bvec; |
0b32f4bb | 3703 | struct extent_buffer *eb; |
2c30c71b | 3704 | int i, done; |
0b32f4bb | 3705 | |
2c30c71b | 3706 | bio_for_each_segment_all(bvec, bio, i) { |
0b32f4bb JB |
3707 | struct page *page = bvec->bv_page; |
3708 | ||
0b32f4bb JB |
3709 | eb = (struct extent_buffer *)page->private; |
3710 | BUG_ON(!eb); | |
3711 | done = atomic_dec_and_test(&eb->io_pages); | |
3712 | ||
4246a0b6 CH |
3713 | if (bio->bi_error || |
3714 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { | |
0b32f4bb | 3715 | ClearPageUptodate(page); |
656f30db | 3716 | set_btree_ioerr(page); |
0b32f4bb JB |
3717 | } |
3718 | ||
3719 | end_page_writeback(page); | |
3720 | ||
3721 | if (!done) | |
3722 | continue; | |
3723 | ||
3724 | end_extent_buffer_writeback(eb); | |
2c30c71b | 3725 | } |
0b32f4bb JB |
3726 | |
3727 | bio_put(bio); | |
0b32f4bb JB |
3728 | } |
3729 | ||
0e378df1 | 3730 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb JB |
3731 | struct btrfs_fs_info *fs_info, |
3732 | struct writeback_control *wbc, | |
3733 | struct extent_page_data *epd) | |
3734 | { | |
3735 | struct block_device *bdev = fs_info->fs_devices->latest_bdev; | |
f28491e0 | 3736 | struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree; |
0b32f4bb JB |
3737 | u64 offset = eb->start; |
3738 | unsigned long i, num_pages; | |
de0022b9 | 3739 | unsigned long bio_flags = 0; |
d4c7ca86 | 3740 | int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META; |
d7dbe9e7 | 3741 | int ret = 0; |
0b32f4bb | 3742 | |
656f30db | 3743 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); |
0b32f4bb JB |
3744 | num_pages = num_extent_pages(eb->start, eb->len); |
3745 | atomic_set(&eb->io_pages, num_pages); | |
de0022b9 JB |
3746 | if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID) |
3747 | bio_flags = EXTENT_BIO_TREE_LOG; | |
3748 | ||
0b32f4bb | 3749 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3750 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3751 | |
3752 | clear_page_dirty_for_io(p); | |
3753 | set_page_writeback(p); | |
da2f0f74 | 3754 | ret = submit_extent_page(rw, tree, wbc, p, offset >> 9, |
0b32f4bb JB |
3755 | PAGE_CACHE_SIZE, 0, bdev, &epd->bio, |
3756 | -1, end_bio_extent_buffer_writepage, | |
005efedf | 3757 | 0, epd->bio_flags, bio_flags, false); |
de0022b9 | 3758 | epd->bio_flags = bio_flags; |
0b32f4bb | 3759 | if (ret) { |
656f30db | 3760 | set_btree_ioerr(p); |
55e3bd2e | 3761 | end_page_writeback(p); |
0b32f4bb JB |
3762 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
3763 | end_extent_buffer_writeback(eb); | |
3764 | ret = -EIO; | |
3765 | break; | |
3766 | } | |
3767 | offset += PAGE_CACHE_SIZE; | |
3768 | update_nr_written(p, wbc, 1); | |
3769 | unlock_page(p); | |
3770 | } | |
3771 | ||
3772 | if (unlikely(ret)) { | |
3773 | for (; i < num_pages; i++) { | |
bbf65cf0 | 3774 | struct page *p = eb->pages[i]; |
81465028 | 3775 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
3776 | unlock_page(p); |
3777 | } | |
3778 | } | |
3779 | ||
3780 | return ret; | |
3781 | } | |
3782 | ||
3783 | int btree_write_cache_pages(struct address_space *mapping, | |
3784 | struct writeback_control *wbc) | |
3785 | { | |
3786 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; | |
3787 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; | |
3788 | struct extent_buffer *eb, *prev_eb = NULL; | |
3789 | struct extent_page_data epd = { | |
3790 | .bio = NULL, | |
3791 | .tree = tree, | |
3792 | .extent_locked = 0, | |
3793 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
de0022b9 | 3794 | .bio_flags = 0, |
0b32f4bb JB |
3795 | }; |
3796 | int ret = 0; | |
3797 | int done = 0; | |
3798 | int nr_to_write_done = 0; | |
3799 | struct pagevec pvec; | |
3800 | int nr_pages; | |
3801 | pgoff_t index; | |
3802 | pgoff_t end; /* Inclusive */ | |
3803 | int scanned = 0; | |
3804 | int tag; | |
3805 | ||
3806 | pagevec_init(&pvec, 0); | |
3807 | if (wbc->range_cyclic) { | |
3808 | index = mapping->writeback_index; /* Start from prev offset */ | |
3809 | end = -1; | |
3810 | } else { | |
3811 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
3812 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
3813 | scanned = 1; | |
3814 | } | |
3815 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3816 | tag = PAGECACHE_TAG_TOWRITE; | |
3817 | else | |
3818 | tag = PAGECACHE_TAG_DIRTY; | |
3819 | retry: | |
3820 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3821 | tag_pages_for_writeback(mapping, index, end); | |
3822 | while (!done && !nr_to_write_done && (index <= end) && | |
3823 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, | |
3824 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
3825 | unsigned i; | |
3826 | ||
3827 | scanned = 1; | |
3828 | for (i = 0; i < nr_pages; i++) { | |
3829 | struct page *page = pvec.pages[i]; | |
3830 | ||
3831 | if (!PagePrivate(page)) | |
3832 | continue; | |
3833 | ||
3834 | if (!wbc->range_cyclic && page->index > end) { | |
3835 | done = 1; | |
3836 | break; | |
3837 | } | |
3838 | ||
b5bae261 JB |
3839 | spin_lock(&mapping->private_lock); |
3840 | if (!PagePrivate(page)) { | |
3841 | spin_unlock(&mapping->private_lock); | |
3842 | continue; | |
3843 | } | |
3844 | ||
0b32f4bb | 3845 | eb = (struct extent_buffer *)page->private; |
b5bae261 JB |
3846 | |
3847 | /* | |
3848 | * Shouldn't happen and normally this would be a BUG_ON | |
3849 | * but no sense in crashing the users box for something | |
3850 | * we can survive anyway. | |
3851 | */ | |
fae7f21c | 3852 | if (WARN_ON(!eb)) { |
b5bae261 | 3853 | spin_unlock(&mapping->private_lock); |
0b32f4bb JB |
3854 | continue; |
3855 | } | |
3856 | ||
b5bae261 JB |
3857 | if (eb == prev_eb) { |
3858 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 3859 | continue; |
b5bae261 | 3860 | } |
0b32f4bb | 3861 | |
b5bae261 JB |
3862 | ret = atomic_inc_not_zero(&eb->refs); |
3863 | spin_unlock(&mapping->private_lock); | |
3864 | if (!ret) | |
0b32f4bb | 3865 | continue; |
0b32f4bb JB |
3866 | |
3867 | prev_eb = eb; | |
3868 | ret = lock_extent_buffer_for_io(eb, fs_info, &epd); | |
3869 | if (!ret) { | |
3870 | free_extent_buffer(eb); | |
3871 | continue; | |
3872 | } | |
3873 | ||
3874 | ret = write_one_eb(eb, fs_info, wbc, &epd); | |
3875 | if (ret) { | |
3876 | done = 1; | |
3877 | free_extent_buffer(eb); | |
3878 | break; | |
3879 | } | |
3880 | free_extent_buffer(eb); | |
3881 | ||
3882 | /* | |
3883 | * the filesystem may choose to bump up nr_to_write. | |
3884 | * We have to make sure to honor the new nr_to_write | |
3885 | * at any time | |
3886 | */ | |
3887 | nr_to_write_done = wbc->nr_to_write <= 0; | |
3888 | } | |
3889 | pagevec_release(&pvec); | |
3890 | cond_resched(); | |
3891 | } | |
3892 | if (!scanned && !done) { | |
3893 | /* | |
3894 | * We hit the last page and there is more work to be done: wrap | |
3895 | * back to the start of the file | |
3896 | */ | |
3897 | scanned = 1; | |
3898 | index = 0; | |
3899 | goto retry; | |
3900 | } | |
3901 | flush_write_bio(&epd); | |
3902 | return ret; | |
3903 | } | |
3904 | ||
d1310b2e | 3905 | /** |
4bef0848 | 3906 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
3907 | * @mapping: address space structure to write |
3908 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
3909 | * @writepage: function called for each page | |
3910 | * @data: data passed to writepage function | |
3911 | * | |
3912 | * If a page is already under I/O, write_cache_pages() skips it, even | |
3913 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
3914 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
3915 | * and msync() need to guarantee that all the data which was dirty at the time | |
3916 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
3917 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
3918 | * existing IO to complete. | |
3919 | */ | |
b2950863 | 3920 | static int extent_write_cache_pages(struct extent_io_tree *tree, |
4bef0848 CM |
3921 | struct address_space *mapping, |
3922 | struct writeback_control *wbc, | |
d2c3f4f6 CM |
3923 | writepage_t writepage, void *data, |
3924 | void (*flush_fn)(void *)) | |
d1310b2e | 3925 | { |
7fd1a3f7 | 3926 | struct inode *inode = mapping->host; |
d1310b2e CM |
3927 | int ret = 0; |
3928 | int done = 0; | |
61391d56 | 3929 | int err = 0; |
f85d7d6c | 3930 | int nr_to_write_done = 0; |
d1310b2e CM |
3931 | struct pagevec pvec; |
3932 | int nr_pages; | |
3933 | pgoff_t index; | |
3934 | pgoff_t end; /* Inclusive */ | |
3935 | int scanned = 0; | |
f7aaa06b | 3936 | int tag; |
d1310b2e | 3937 | |
7fd1a3f7 JB |
3938 | /* |
3939 | * We have to hold onto the inode so that ordered extents can do their | |
3940 | * work when the IO finishes. The alternative to this is failing to add | |
3941 | * an ordered extent if the igrab() fails there and that is a huge pain | |
3942 | * to deal with, so instead just hold onto the inode throughout the | |
3943 | * writepages operation. If it fails here we are freeing up the inode | |
3944 | * anyway and we'd rather not waste our time writing out stuff that is | |
3945 | * going to be truncated anyway. | |
3946 | */ | |
3947 | if (!igrab(inode)) | |
3948 | return 0; | |
3949 | ||
d1310b2e CM |
3950 | pagevec_init(&pvec, 0); |
3951 | if (wbc->range_cyclic) { | |
3952 | index = mapping->writeback_index; /* Start from prev offset */ | |
3953 | end = -1; | |
3954 | } else { | |
3955 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
3956 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
3957 | scanned = 1; |
3958 | } | |
f7aaa06b JB |
3959 | if (wbc->sync_mode == WB_SYNC_ALL) |
3960 | tag = PAGECACHE_TAG_TOWRITE; | |
3961 | else | |
3962 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 3963 | retry: |
f7aaa06b JB |
3964 | if (wbc->sync_mode == WB_SYNC_ALL) |
3965 | tag_pages_for_writeback(mapping, index, end); | |
f85d7d6c | 3966 | while (!done && !nr_to_write_done && (index <= end) && |
f7aaa06b JB |
3967 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
3968 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
d1310b2e CM |
3969 | unsigned i; |
3970 | ||
3971 | scanned = 1; | |
3972 | for (i = 0; i < nr_pages; i++) { | |
3973 | struct page *page = pvec.pages[i]; | |
3974 | ||
3975 | /* | |
3976 | * At this point we hold neither mapping->tree_lock nor | |
3977 | * lock on the page itself: the page may be truncated or | |
3978 | * invalidated (changing page->mapping to NULL), or even | |
3979 | * swizzled back from swapper_space to tmpfs file | |
3980 | * mapping | |
3981 | */ | |
c8f2f24b JB |
3982 | if (!trylock_page(page)) { |
3983 | flush_fn(data); | |
3984 | lock_page(page); | |
01d658f2 | 3985 | } |
d1310b2e CM |
3986 | |
3987 | if (unlikely(page->mapping != mapping)) { | |
3988 | unlock_page(page); | |
3989 | continue; | |
3990 | } | |
3991 | ||
3992 | if (!wbc->range_cyclic && page->index > end) { | |
3993 | done = 1; | |
3994 | unlock_page(page); | |
3995 | continue; | |
3996 | } | |
3997 | ||
d2c3f4f6 | 3998 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 CM |
3999 | if (PageWriteback(page)) |
4000 | flush_fn(data); | |
d1310b2e | 4001 | wait_on_page_writeback(page); |
d2c3f4f6 | 4002 | } |
d1310b2e CM |
4003 | |
4004 | if (PageWriteback(page) || | |
4005 | !clear_page_dirty_for_io(page)) { | |
4006 | unlock_page(page); | |
4007 | continue; | |
4008 | } | |
4009 | ||
4010 | ret = (*writepage)(page, wbc, data); | |
4011 | ||
4012 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
4013 | unlock_page(page); | |
4014 | ret = 0; | |
4015 | } | |
61391d56 FM |
4016 | if (!err && ret < 0) |
4017 | err = ret; | |
f85d7d6c CM |
4018 | |
4019 | /* | |
4020 | * the filesystem may choose to bump up nr_to_write. | |
4021 | * We have to make sure to honor the new nr_to_write | |
4022 | * at any time | |
4023 | */ | |
4024 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
4025 | } |
4026 | pagevec_release(&pvec); | |
4027 | cond_resched(); | |
4028 | } | |
61391d56 | 4029 | if (!scanned && !done && !err) { |
d1310b2e CM |
4030 | /* |
4031 | * We hit the last page and there is more work to be done: wrap | |
4032 | * back to the start of the file | |
4033 | */ | |
4034 | scanned = 1; | |
4035 | index = 0; | |
4036 | goto retry; | |
4037 | } | |
7fd1a3f7 | 4038 | btrfs_add_delayed_iput(inode); |
61391d56 | 4039 | return err; |
d1310b2e | 4040 | } |
d1310b2e | 4041 | |
ffbd517d | 4042 | static void flush_epd_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 4043 | { |
d2c3f4f6 | 4044 | if (epd->bio) { |
355808c2 JM |
4045 | int rw = WRITE; |
4046 | int ret; | |
4047 | ||
ffbd517d | 4048 | if (epd->sync_io) |
355808c2 JM |
4049 | rw = WRITE_SYNC; |
4050 | ||
de0022b9 | 4051 | ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags); |
79787eaa | 4052 | BUG_ON(ret < 0); /* -ENOMEM */ |
d2c3f4f6 CM |
4053 | epd->bio = NULL; |
4054 | } | |
4055 | } | |
4056 | ||
ffbd517d CM |
4057 | static noinline void flush_write_bio(void *data) |
4058 | { | |
4059 | struct extent_page_data *epd = data; | |
4060 | flush_epd_write_bio(epd); | |
4061 | } | |
4062 | ||
d1310b2e CM |
4063 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, |
4064 | get_extent_t *get_extent, | |
4065 | struct writeback_control *wbc) | |
4066 | { | |
4067 | int ret; | |
d1310b2e CM |
4068 | struct extent_page_data epd = { |
4069 | .bio = NULL, | |
4070 | .tree = tree, | |
4071 | .get_extent = get_extent, | |
771ed689 | 4072 | .extent_locked = 0, |
ffbd517d | 4073 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
de0022b9 | 4074 | .bio_flags = 0, |
d1310b2e | 4075 | }; |
d1310b2e | 4076 | |
d1310b2e CM |
4077 | ret = __extent_writepage(page, wbc, &epd); |
4078 | ||
ffbd517d | 4079 | flush_epd_write_bio(&epd); |
d1310b2e CM |
4080 | return ret; |
4081 | } | |
d1310b2e | 4082 | |
771ed689 CM |
4083 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, |
4084 | u64 start, u64 end, get_extent_t *get_extent, | |
4085 | int mode) | |
4086 | { | |
4087 | int ret = 0; | |
4088 | struct address_space *mapping = inode->i_mapping; | |
4089 | struct page *page; | |
4090 | unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >> | |
4091 | PAGE_CACHE_SHIFT; | |
4092 | ||
4093 | struct extent_page_data epd = { | |
4094 | .bio = NULL, | |
4095 | .tree = tree, | |
4096 | .get_extent = get_extent, | |
4097 | .extent_locked = 1, | |
ffbd517d | 4098 | .sync_io = mode == WB_SYNC_ALL, |
de0022b9 | 4099 | .bio_flags = 0, |
771ed689 CM |
4100 | }; |
4101 | struct writeback_control wbc_writepages = { | |
771ed689 | 4102 | .sync_mode = mode, |
771ed689 CM |
4103 | .nr_to_write = nr_pages * 2, |
4104 | .range_start = start, | |
4105 | .range_end = end + 1, | |
4106 | }; | |
4107 | ||
d397712b | 4108 | while (start <= end) { |
771ed689 CM |
4109 | page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); |
4110 | if (clear_page_dirty_for_io(page)) | |
4111 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
4112 | else { | |
4113 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
4114 | tree->ops->writepage_end_io_hook(page, start, | |
4115 | start + PAGE_CACHE_SIZE - 1, | |
4116 | NULL, 1); | |
4117 | unlock_page(page); | |
4118 | } | |
4119 | page_cache_release(page); | |
4120 | start += PAGE_CACHE_SIZE; | |
4121 | } | |
4122 | ||
ffbd517d | 4123 | flush_epd_write_bio(&epd); |
771ed689 CM |
4124 | return ret; |
4125 | } | |
d1310b2e CM |
4126 | |
4127 | int extent_writepages(struct extent_io_tree *tree, | |
4128 | struct address_space *mapping, | |
4129 | get_extent_t *get_extent, | |
4130 | struct writeback_control *wbc) | |
4131 | { | |
4132 | int ret = 0; | |
4133 | struct extent_page_data epd = { | |
4134 | .bio = NULL, | |
4135 | .tree = tree, | |
4136 | .get_extent = get_extent, | |
771ed689 | 4137 | .extent_locked = 0, |
ffbd517d | 4138 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
de0022b9 | 4139 | .bio_flags = 0, |
d1310b2e CM |
4140 | }; |
4141 | ||
4bef0848 | 4142 | ret = extent_write_cache_pages(tree, mapping, wbc, |
d2c3f4f6 CM |
4143 | __extent_writepage, &epd, |
4144 | flush_write_bio); | |
ffbd517d | 4145 | flush_epd_write_bio(&epd); |
d1310b2e CM |
4146 | return ret; |
4147 | } | |
d1310b2e CM |
4148 | |
4149 | int extent_readpages(struct extent_io_tree *tree, | |
4150 | struct address_space *mapping, | |
4151 | struct list_head *pages, unsigned nr_pages, | |
4152 | get_extent_t get_extent) | |
4153 | { | |
4154 | struct bio *bio = NULL; | |
4155 | unsigned page_idx; | |
c8b97818 | 4156 | unsigned long bio_flags = 0; |
67c9684f LB |
4157 | struct page *pagepool[16]; |
4158 | struct page *page; | |
125bac01 | 4159 | struct extent_map *em_cached = NULL; |
67c9684f | 4160 | int nr = 0; |
808f80b4 | 4161 | u64 prev_em_start = (u64)-1; |
d1310b2e | 4162 | |
d1310b2e | 4163 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
67c9684f | 4164 | page = list_entry(pages->prev, struct page, lru); |
d1310b2e CM |
4165 | |
4166 | prefetchw(&page->flags); | |
4167 | list_del(&page->lru); | |
67c9684f | 4168 | if (add_to_page_cache_lru(page, mapping, |
43e817a1 | 4169 | page->index, GFP_NOFS)) { |
67c9684f LB |
4170 | page_cache_release(page); |
4171 | continue; | |
d1310b2e | 4172 | } |
67c9684f LB |
4173 | |
4174 | pagepool[nr++] = page; | |
4175 | if (nr < ARRAY_SIZE(pagepool)) | |
4176 | continue; | |
125bac01 | 4177 | __extent_readpages(tree, pagepool, nr, get_extent, &em_cached, |
808f80b4 | 4178 | &bio, 0, &bio_flags, READ, &prev_em_start); |
67c9684f | 4179 | nr = 0; |
d1310b2e | 4180 | } |
9974090b | 4181 | if (nr) |
125bac01 | 4182 | __extent_readpages(tree, pagepool, nr, get_extent, &em_cached, |
808f80b4 | 4183 | &bio, 0, &bio_flags, READ, &prev_em_start); |
67c9684f | 4184 | |
125bac01 MX |
4185 | if (em_cached) |
4186 | free_extent_map(em_cached); | |
4187 | ||
d1310b2e CM |
4188 | BUG_ON(!list_empty(pages)); |
4189 | if (bio) | |
79787eaa | 4190 | return submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
4191 | return 0; |
4192 | } | |
d1310b2e CM |
4193 | |
4194 | /* | |
4195 | * basic invalidatepage code, this waits on any locked or writeback | |
4196 | * ranges corresponding to the page, and then deletes any extent state | |
4197 | * records from the tree | |
4198 | */ | |
4199 | int extent_invalidatepage(struct extent_io_tree *tree, | |
4200 | struct page *page, unsigned long offset) | |
4201 | { | |
2ac55d41 | 4202 | struct extent_state *cached_state = NULL; |
4eee4fa4 | 4203 | u64 start = page_offset(page); |
d1310b2e CM |
4204 | u64 end = start + PAGE_CACHE_SIZE - 1; |
4205 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
4206 | ||
fda2832f | 4207 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
4208 | if (start > end) |
4209 | return 0; | |
4210 | ||
ff13db41 | 4211 | lock_extent_bits(tree, start, end, &cached_state); |
1edbb734 | 4212 | wait_on_page_writeback(page); |
d1310b2e | 4213 | clear_extent_bit(tree, start, end, |
32c00aff JB |
4214 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
4215 | EXTENT_DO_ACCOUNTING, | |
2ac55d41 | 4216 | 1, 1, &cached_state, GFP_NOFS); |
d1310b2e CM |
4217 | return 0; |
4218 | } | |
d1310b2e | 4219 | |
7b13b7b1 CM |
4220 | /* |
4221 | * a helper for releasepage, this tests for areas of the page that | |
4222 | * are locked or under IO and drops the related state bits if it is safe | |
4223 | * to drop the page. | |
4224 | */ | |
48a3b636 ES |
4225 | static int try_release_extent_state(struct extent_map_tree *map, |
4226 | struct extent_io_tree *tree, | |
4227 | struct page *page, gfp_t mask) | |
7b13b7b1 | 4228 | { |
4eee4fa4 | 4229 | u64 start = page_offset(page); |
7b13b7b1 CM |
4230 | u64 end = start + PAGE_CACHE_SIZE - 1; |
4231 | int ret = 1; | |
4232 | ||
211f90e6 | 4233 | if (test_range_bit(tree, start, end, |
8b62b72b | 4234 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
4235 | ret = 0; |
4236 | else { | |
4237 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
4238 | mask = GFP_NOFS; | |
11ef160f CM |
4239 | /* |
4240 | * at this point we can safely clear everything except the | |
4241 | * locked bit and the nodatasum bit | |
4242 | */ | |
e3f24cc5 | 4243 | ret = clear_extent_bit(tree, start, end, |
11ef160f CM |
4244 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
4245 | 0, 0, NULL, mask); | |
e3f24cc5 CM |
4246 | |
4247 | /* if clear_extent_bit failed for enomem reasons, | |
4248 | * we can't allow the release to continue. | |
4249 | */ | |
4250 | if (ret < 0) | |
4251 | ret = 0; | |
4252 | else | |
4253 | ret = 1; | |
7b13b7b1 CM |
4254 | } |
4255 | return ret; | |
4256 | } | |
7b13b7b1 | 4257 | |
d1310b2e CM |
4258 | /* |
4259 | * a helper for releasepage. As long as there are no locked extents | |
4260 | * in the range corresponding to the page, both state records and extent | |
4261 | * map records are removed | |
4262 | */ | |
4263 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
4264 | struct extent_io_tree *tree, struct page *page, |
4265 | gfp_t mask) | |
d1310b2e CM |
4266 | { |
4267 | struct extent_map *em; | |
4eee4fa4 | 4268 | u64 start = page_offset(page); |
d1310b2e | 4269 | u64 end = start + PAGE_CACHE_SIZE - 1; |
7b13b7b1 | 4270 | |
d0164adc | 4271 | if (gfpflags_allow_blocking(mask) && |
ee22184b | 4272 | page->mapping->host->i_size > SZ_16M) { |
39b5637f | 4273 | u64 len; |
70dec807 | 4274 | while (start <= end) { |
39b5637f | 4275 | len = end - start + 1; |
890871be | 4276 | write_lock(&map->lock); |
39b5637f | 4277 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 4278 | if (!em) { |
890871be | 4279 | write_unlock(&map->lock); |
70dec807 CM |
4280 | break; |
4281 | } | |
7f3c74fb CM |
4282 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
4283 | em->start != start) { | |
890871be | 4284 | write_unlock(&map->lock); |
70dec807 CM |
4285 | free_extent_map(em); |
4286 | break; | |
4287 | } | |
4288 | if (!test_range_bit(tree, em->start, | |
4289 | extent_map_end(em) - 1, | |
8b62b72b | 4290 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 4291 | 0, NULL)) { |
70dec807 CM |
4292 | remove_extent_mapping(map, em); |
4293 | /* once for the rb tree */ | |
4294 | free_extent_map(em); | |
4295 | } | |
4296 | start = extent_map_end(em); | |
890871be | 4297 | write_unlock(&map->lock); |
70dec807 CM |
4298 | |
4299 | /* once for us */ | |
d1310b2e CM |
4300 | free_extent_map(em); |
4301 | } | |
d1310b2e | 4302 | } |
7b13b7b1 | 4303 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 4304 | } |
d1310b2e | 4305 | |
ec29ed5b CM |
4306 | /* |
4307 | * helper function for fiemap, which doesn't want to see any holes. | |
4308 | * This maps until we find something past 'last' | |
4309 | */ | |
4310 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
4311 | u64 offset, | |
4312 | u64 last, | |
4313 | get_extent_t *get_extent) | |
4314 | { | |
4315 | u64 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
4316 | struct extent_map *em; | |
4317 | u64 len; | |
4318 | ||
4319 | if (offset >= last) | |
4320 | return NULL; | |
4321 | ||
67871254 | 4322 | while (1) { |
ec29ed5b CM |
4323 | len = last - offset; |
4324 | if (len == 0) | |
4325 | break; | |
fda2832f | 4326 | len = ALIGN(len, sectorsize); |
ec29ed5b | 4327 | em = get_extent(inode, NULL, 0, offset, len, 0); |
c704005d | 4328 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
4329 | return em; |
4330 | ||
4331 | /* if this isn't a hole return it */ | |
4332 | if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) && | |
4333 | em->block_start != EXTENT_MAP_HOLE) { | |
4334 | return em; | |
4335 | } | |
4336 | ||
4337 | /* this is a hole, advance to the next extent */ | |
4338 | offset = extent_map_end(em); | |
4339 | free_extent_map(em); | |
4340 | if (offset >= last) | |
4341 | break; | |
4342 | } | |
4343 | return NULL; | |
4344 | } | |
4345 | ||
1506fcc8 YS |
4346 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4347 | __u64 start, __u64 len, get_extent_t *get_extent) | |
4348 | { | |
975f84fe | 4349 | int ret = 0; |
1506fcc8 YS |
4350 | u64 off = start; |
4351 | u64 max = start + len; | |
4352 | u32 flags = 0; | |
975f84fe JB |
4353 | u32 found_type; |
4354 | u64 last; | |
ec29ed5b | 4355 | u64 last_for_get_extent = 0; |
1506fcc8 | 4356 | u64 disko = 0; |
ec29ed5b | 4357 | u64 isize = i_size_read(inode); |
975f84fe | 4358 | struct btrfs_key found_key; |
1506fcc8 | 4359 | struct extent_map *em = NULL; |
2ac55d41 | 4360 | struct extent_state *cached_state = NULL; |
975f84fe | 4361 | struct btrfs_path *path; |
dc046b10 | 4362 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1506fcc8 | 4363 | int end = 0; |
ec29ed5b CM |
4364 | u64 em_start = 0; |
4365 | u64 em_len = 0; | |
4366 | u64 em_end = 0; | |
1506fcc8 YS |
4367 | |
4368 | if (len == 0) | |
4369 | return -EINVAL; | |
4370 | ||
975f84fe JB |
4371 | path = btrfs_alloc_path(); |
4372 | if (!path) | |
4373 | return -ENOMEM; | |
4374 | path->leave_spinning = 1; | |
4375 | ||
2c91943b QW |
4376 | start = round_down(start, BTRFS_I(inode)->root->sectorsize); |
4377 | len = round_up(max, BTRFS_I(inode)->root->sectorsize) - start; | |
4d479cf0 | 4378 | |
ec29ed5b CM |
4379 | /* |
4380 | * lookup the last file extent. We're not using i_size here | |
4381 | * because there might be preallocation past i_size | |
4382 | */ | |
dc046b10 JB |
4383 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1, |
4384 | 0); | |
975f84fe JB |
4385 | if (ret < 0) { |
4386 | btrfs_free_path(path); | |
4387 | return ret; | |
4388 | } | |
4389 | WARN_ON(!ret); | |
4390 | path->slots[0]--; | |
975f84fe | 4391 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
962a298f | 4392 | found_type = found_key.type; |
975f84fe | 4393 | |
ec29ed5b | 4394 | /* No extents, but there might be delalloc bits */ |
33345d01 | 4395 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 4396 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
4397 | /* have to trust i_size as the end */ |
4398 | last = (u64)-1; | |
4399 | last_for_get_extent = isize; | |
4400 | } else { | |
4401 | /* | |
4402 | * remember the start of the last extent. There are a | |
4403 | * bunch of different factors that go into the length of the | |
4404 | * extent, so its much less complex to remember where it started | |
4405 | */ | |
4406 | last = found_key.offset; | |
4407 | last_for_get_extent = last + 1; | |
975f84fe | 4408 | } |
fe09e16c | 4409 | btrfs_release_path(path); |
975f84fe | 4410 | |
ec29ed5b CM |
4411 | /* |
4412 | * we might have some extents allocated but more delalloc past those | |
4413 | * extents. so, we trust isize unless the start of the last extent is | |
4414 | * beyond isize | |
4415 | */ | |
4416 | if (last < isize) { | |
4417 | last = (u64)-1; | |
4418 | last_for_get_extent = isize; | |
4419 | } | |
4420 | ||
ff13db41 | 4421 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
d0082371 | 4422 | &cached_state); |
ec29ed5b | 4423 | |
4d479cf0 | 4424 | em = get_extent_skip_holes(inode, start, last_for_get_extent, |
ec29ed5b | 4425 | get_extent); |
1506fcc8 YS |
4426 | if (!em) |
4427 | goto out; | |
4428 | if (IS_ERR(em)) { | |
4429 | ret = PTR_ERR(em); | |
4430 | goto out; | |
4431 | } | |
975f84fe | 4432 | |
1506fcc8 | 4433 | while (!end) { |
b76bb701 | 4434 | u64 offset_in_extent = 0; |
ea8efc74 CM |
4435 | |
4436 | /* break if the extent we found is outside the range */ | |
4437 | if (em->start >= max || extent_map_end(em) < off) | |
4438 | break; | |
4439 | ||
4440 | /* | |
4441 | * get_extent may return an extent that starts before our | |
4442 | * requested range. We have to make sure the ranges | |
4443 | * we return to fiemap always move forward and don't | |
4444 | * overlap, so adjust the offsets here | |
4445 | */ | |
4446 | em_start = max(em->start, off); | |
1506fcc8 | 4447 | |
ea8efc74 CM |
4448 | /* |
4449 | * record the offset from the start of the extent | |
b76bb701 JB |
4450 | * for adjusting the disk offset below. Only do this if the |
4451 | * extent isn't compressed since our in ram offset may be past | |
4452 | * what we have actually allocated on disk. | |
ea8efc74 | 4453 | */ |
b76bb701 JB |
4454 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
4455 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 4456 | em_end = extent_map_end(em); |
ea8efc74 | 4457 | em_len = em_end - em_start; |
1506fcc8 YS |
4458 | disko = 0; |
4459 | flags = 0; | |
4460 | ||
ea8efc74 CM |
4461 | /* |
4462 | * bump off for our next call to get_extent | |
4463 | */ | |
4464 | off = extent_map_end(em); | |
4465 | if (off >= max) | |
4466 | end = 1; | |
4467 | ||
93dbfad7 | 4468 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
4469 | end = 1; |
4470 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 4471 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
4472 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
4473 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 4474 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
4475 | flags |= (FIEMAP_EXTENT_DELALLOC | |
4476 | FIEMAP_EXTENT_UNKNOWN); | |
dc046b10 JB |
4477 | } else if (fieinfo->fi_extents_max) { |
4478 | u64 bytenr = em->block_start - | |
4479 | (em->start - em->orig_start); | |
fe09e16c | 4480 | |
ea8efc74 | 4481 | disko = em->block_start + offset_in_extent; |
fe09e16c LB |
4482 | |
4483 | /* | |
4484 | * As btrfs supports shared space, this information | |
4485 | * can be exported to userspace tools via | |
dc046b10 JB |
4486 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 |
4487 | * then we're just getting a count and we can skip the | |
4488 | * lookup stuff. | |
fe09e16c | 4489 | */ |
dc046b10 JB |
4490 | ret = btrfs_check_shared(NULL, root->fs_info, |
4491 | root->objectid, | |
4492 | btrfs_ino(inode), bytenr); | |
4493 | if (ret < 0) | |
fe09e16c | 4494 | goto out_free; |
dc046b10 | 4495 | if (ret) |
fe09e16c | 4496 | flags |= FIEMAP_EXTENT_SHARED; |
dc046b10 | 4497 | ret = 0; |
1506fcc8 YS |
4498 | } |
4499 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
4500 | flags |= FIEMAP_EXTENT_ENCODED; | |
0d2b2372 JB |
4501 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
4502 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
1506fcc8 | 4503 | |
1506fcc8 YS |
4504 | free_extent_map(em); |
4505 | em = NULL; | |
ec29ed5b CM |
4506 | if ((em_start >= last) || em_len == (u64)-1 || |
4507 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
4508 | flags |= FIEMAP_EXTENT_LAST; |
4509 | end = 1; | |
4510 | } | |
4511 | ||
ec29ed5b CM |
4512 | /* now scan forward to see if this is really the last extent. */ |
4513 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
4514 | get_extent); | |
4515 | if (IS_ERR(em)) { | |
4516 | ret = PTR_ERR(em); | |
4517 | goto out; | |
4518 | } | |
4519 | if (!em) { | |
975f84fe JB |
4520 | flags |= FIEMAP_EXTENT_LAST; |
4521 | end = 1; | |
4522 | } | |
ec29ed5b CM |
4523 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, |
4524 | em_len, flags); | |
26e726af CS |
4525 | if (ret) { |
4526 | if (ret == 1) | |
4527 | ret = 0; | |
ec29ed5b | 4528 | goto out_free; |
26e726af | 4529 | } |
1506fcc8 YS |
4530 | } |
4531 | out_free: | |
4532 | free_extent_map(em); | |
4533 | out: | |
fe09e16c | 4534 | btrfs_free_path(path); |
a52f4cd2 | 4535 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
2ac55d41 | 4536 | &cached_state, GFP_NOFS); |
1506fcc8 YS |
4537 | return ret; |
4538 | } | |
4539 | ||
727011e0 CM |
4540 | static void __free_extent_buffer(struct extent_buffer *eb) |
4541 | { | |
6d49ba1b | 4542 | btrfs_leak_debug_del(&eb->leak_list); |
727011e0 CM |
4543 | kmem_cache_free(extent_buffer_cache, eb); |
4544 | } | |
4545 | ||
a26e8c9f | 4546 | int extent_buffer_under_io(struct extent_buffer *eb) |
db7f3436 JB |
4547 | { |
4548 | return (atomic_read(&eb->io_pages) || | |
4549 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4550 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4551 | } | |
4552 | ||
4553 | /* | |
4554 | * Helper for releasing extent buffer page. | |
4555 | */ | |
a50924e3 | 4556 | static void btrfs_release_extent_buffer_page(struct extent_buffer *eb) |
db7f3436 JB |
4557 | { |
4558 | unsigned long index; | |
db7f3436 JB |
4559 | struct page *page; |
4560 | int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4561 | ||
4562 | BUG_ON(extent_buffer_under_io(eb)); | |
4563 | ||
a50924e3 DS |
4564 | index = num_extent_pages(eb->start, eb->len); |
4565 | if (index == 0) | |
db7f3436 JB |
4566 | return; |
4567 | ||
4568 | do { | |
4569 | index--; | |
fb85fc9a | 4570 | page = eb->pages[index]; |
5d2361db FL |
4571 | if (!page) |
4572 | continue; | |
4573 | if (mapped) | |
db7f3436 | 4574 | spin_lock(&page->mapping->private_lock); |
5d2361db FL |
4575 | /* |
4576 | * We do this since we'll remove the pages after we've | |
4577 | * removed the eb from the radix tree, so we could race | |
4578 | * and have this page now attached to the new eb. So | |
4579 | * only clear page_private if it's still connected to | |
4580 | * this eb. | |
4581 | */ | |
4582 | if (PagePrivate(page) && | |
4583 | page->private == (unsigned long)eb) { | |
4584 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4585 | BUG_ON(PageDirty(page)); | |
4586 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 4587 | /* |
5d2361db FL |
4588 | * We need to make sure we haven't be attached |
4589 | * to a new eb. | |
db7f3436 | 4590 | */ |
5d2361db FL |
4591 | ClearPagePrivate(page); |
4592 | set_page_private(page, 0); | |
4593 | /* One for the page private */ | |
db7f3436 JB |
4594 | page_cache_release(page); |
4595 | } | |
5d2361db FL |
4596 | |
4597 | if (mapped) | |
4598 | spin_unlock(&page->mapping->private_lock); | |
4599 | ||
4600 | /* One for when we alloced the page */ | |
4601 | page_cache_release(page); | |
a50924e3 | 4602 | } while (index != 0); |
db7f3436 JB |
4603 | } |
4604 | ||
4605 | /* | |
4606 | * Helper for releasing the extent buffer. | |
4607 | */ | |
4608 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4609 | { | |
a50924e3 | 4610 | btrfs_release_extent_buffer_page(eb); |
db7f3436 JB |
4611 | __free_extent_buffer(eb); |
4612 | } | |
4613 | ||
f28491e0 JB |
4614 | static struct extent_buffer * |
4615 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 4616 | unsigned long len) |
d1310b2e CM |
4617 | { |
4618 | struct extent_buffer *eb = NULL; | |
4619 | ||
d1b5c567 | 4620 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
4621 | eb->start = start; |
4622 | eb->len = len; | |
f28491e0 | 4623 | eb->fs_info = fs_info; |
815a51c7 | 4624 | eb->bflags = 0; |
bd681513 CM |
4625 | rwlock_init(&eb->lock); |
4626 | atomic_set(&eb->write_locks, 0); | |
4627 | atomic_set(&eb->read_locks, 0); | |
4628 | atomic_set(&eb->blocking_readers, 0); | |
4629 | atomic_set(&eb->blocking_writers, 0); | |
4630 | atomic_set(&eb->spinning_readers, 0); | |
4631 | atomic_set(&eb->spinning_writers, 0); | |
5b25f70f | 4632 | eb->lock_nested = 0; |
bd681513 CM |
4633 | init_waitqueue_head(&eb->write_lock_wq); |
4634 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 4635 | |
6d49ba1b ES |
4636 | btrfs_leak_debug_add(&eb->leak_list, &buffers); |
4637 | ||
3083ee2e | 4638 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 4639 | atomic_set(&eb->refs, 1); |
0b32f4bb | 4640 | atomic_set(&eb->io_pages, 0); |
727011e0 | 4641 | |
b8dae313 DS |
4642 | /* |
4643 | * Sanity checks, currently the maximum is 64k covered by 16x 4k pages | |
4644 | */ | |
4645 | BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE | |
4646 | > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
4647 | BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
d1310b2e CM |
4648 | |
4649 | return eb; | |
4650 | } | |
4651 | ||
815a51c7 JS |
4652 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) |
4653 | { | |
4654 | unsigned long i; | |
4655 | struct page *p; | |
4656 | struct extent_buffer *new; | |
4657 | unsigned long num_pages = num_extent_pages(src->start, src->len); | |
4658 | ||
3f556f78 | 4659 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
4660 | if (new == NULL) |
4661 | return NULL; | |
4662 | ||
4663 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4664 | p = alloc_page(GFP_NOFS); |
db7f3436 JB |
4665 | if (!p) { |
4666 | btrfs_release_extent_buffer(new); | |
4667 | return NULL; | |
4668 | } | |
815a51c7 JS |
4669 | attach_extent_buffer_page(new, p); |
4670 | WARN_ON(PageDirty(p)); | |
4671 | SetPageUptodate(p); | |
4672 | new->pages[i] = p; | |
4673 | } | |
4674 | ||
4675 | copy_extent_buffer(new, src, 0, 0, src->len); | |
4676 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); | |
4677 | set_bit(EXTENT_BUFFER_DUMMY, &new->bflags); | |
4678 | ||
4679 | return new; | |
4680 | } | |
4681 | ||
0f331229 OS |
4682 | struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
4683 | u64 start, unsigned long len) | |
815a51c7 JS |
4684 | { |
4685 | struct extent_buffer *eb; | |
3f556f78 | 4686 | unsigned long num_pages; |
815a51c7 JS |
4687 | unsigned long i; |
4688 | ||
0f331229 | 4689 | num_pages = num_extent_pages(start, len); |
3f556f78 DS |
4690 | |
4691 | eb = __alloc_extent_buffer(fs_info, start, len); | |
815a51c7 JS |
4692 | if (!eb) |
4693 | return NULL; | |
4694 | ||
4695 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4696 | eb->pages[i] = alloc_page(GFP_NOFS); |
815a51c7 JS |
4697 | if (!eb->pages[i]) |
4698 | goto err; | |
4699 | } | |
4700 | set_extent_buffer_uptodate(eb); | |
4701 | btrfs_set_header_nritems(eb, 0); | |
4702 | set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4703 | ||
4704 | return eb; | |
4705 | err: | |
84167d19 SB |
4706 | for (; i > 0; i--) |
4707 | __free_page(eb->pages[i - 1]); | |
815a51c7 JS |
4708 | __free_extent_buffer(eb); |
4709 | return NULL; | |
4710 | } | |
4711 | ||
0f331229 OS |
4712 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
4713 | u64 start) | |
4714 | { | |
4715 | unsigned long len; | |
4716 | ||
4717 | if (!fs_info) { | |
4718 | /* | |
4719 | * Called only from tests that don't always have a fs_info | |
4720 | * available, but we know that nodesize is 4096 | |
4721 | */ | |
4722 | len = 4096; | |
4723 | } else { | |
4724 | len = fs_info->tree_root->nodesize; | |
4725 | } | |
4726 | ||
4727 | return __alloc_dummy_extent_buffer(fs_info, start, len); | |
4728 | } | |
4729 | ||
0b32f4bb JB |
4730 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
4731 | { | |
242e18c7 | 4732 | int refs; |
0b32f4bb JB |
4733 | /* the ref bit is tricky. We have to make sure it is set |
4734 | * if we have the buffer dirty. Otherwise the | |
4735 | * code to free a buffer can end up dropping a dirty | |
4736 | * page | |
4737 | * | |
4738 | * Once the ref bit is set, it won't go away while the | |
4739 | * buffer is dirty or in writeback, and it also won't | |
4740 | * go away while we have the reference count on the | |
4741 | * eb bumped. | |
4742 | * | |
4743 | * We can't just set the ref bit without bumping the | |
4744 | * ref on the eb because free_extent_buffer might | |
4745 | * see the ref bit and try to clear it. If this happens | |
4746 | * free_extent_buffer might end up dropping our original | |
4747 | * ref by mistake and freeing the page before we are able | |
4748 | * to add one more ref. | |
4749 | * | |
4750 | * So bump the ref count first, then set the bit. If someone | |
4751 | * beat us to it, drop the ref we added. | |
4752 | */ | |
242e18c7 CM |
4753 | refs = atomic_read(&eb->refs); |
4754 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
4755 | return; | |
4756 | ||
594831c4 JB |
4757 | spin_lock(&eb->refs_lock); |
4758 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 4759 | atomic_inc(&eb->refs); |
594831c4 | 4760 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
4761 | } |
4762 | ||
2457aec6 MG |
4763 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
4764 | struct page *accessed) | |
5df4235e JB |
4765 | { |
4766 | unsigned long num_pages, i; | |
4767 | ||
0b32f4bb JB |
4768 | check_buffer_tree_ref(eb); |
4769 | ||
5df4235e JB |
4770 | num_pages = num_extent_pages(eb->start, eb->len); |
4771 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a DS |
4772 | struct page *p = eb->pages[i]; |
4773 | ||
2457aec6 MG |
4774 | if (p != accessed) |
4775 | mark_page_accessed(p); | |
5df4235e JB |
4776 | } |
4777 | } | |
4778 | ||
f28491e0 JB |
4779 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
4780 | u64 start) | |
452c75c3 CS |
4781 | { |
4782 | struct extent_buffer *eb; | |
4783 | ||
4784 | rcu_read_lock(); | |
f28491e0 JB |
4785 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
4786 | start >> PAGE_CACHE_SHIFT); | |
452c75c3 CS |
4787 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
4788 | rcu_read_unlock(); | |
062c19e9 FM |
4789 | /* |
4790 | * Lock our eb's refs_lock to avoid races with | |
4791 | * free_extent_buffer. When we get our eb it might be flagged | |
4792 | * with EXTENT_BUFFER_STALE and another task running | |
4793 | * free_extent_buffer might have seen that flag set, | |
4794 | * eb->refs == 2, that the buffer isn't under IO (dirty and | |
4795 | * writeback flags not set) and it's still in the tree (flag | |
4796 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process | |
4797 | * of decrementing the extent buffer's reference count twice. | |
4798 | * So here we could race and increment the eb's reference count, | |
4799 | * clear its stale flag, mark it as dirty and drop our reference | |
4800 | * before the other task finishes executing free_extent_buffer, | |
4801 | * which would later result in an attempt to free an extent | |
4802 | * buffer that is dirty. | |
4803 | */ | |
4804 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
4805 | spin_lock(&eb->refs_lock); | |
4806 | spin_unlock(&eb->refs_lock); | |
4807 | } | |
2457aec6 | 4808 | mark_extent_buffer_accessed(eb, NULL); |
452c75c3 CS |
4809 | return eb; |
4810 | } | |
4811 | rcu_read_unlock(); | |
4812 | ||
4813 | return NULL; | |
4814 | } | |
4815 | ||
faa2dbf0 JB |
4816 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
4817 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
ce3e6984 | 4818 | u64 start) |
faa2dbf0 JB |
4819 | { |
4820 | struct extent_buffer *eb, *exists = NULL; | |
4821 | int ret; | |
4822 | ||
4823 | eb = find_extent_buffer(fs_info, start); | |
4824 | if (eb) | |
4825 | return eb; | |
3f556f78 | 4826 | eb = alloc_dummy_extent_buffer(fs_info, start); |
faa2dbf0 JB |
4827 | if (!eb) |
4828 | return NULL; | |
4829 | eb->fs_info = fs_info; | |
4830 | again: | |
4831 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); | |
4832 | if (ret) | |
4833 | goto free_eb; | |
4834 | spin_lock(&fs_info->buffer_lock); | |
4835 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
4836 | start >> PAGE_CACHE_SHIFT, eb); | |
4837 | spin_unlock(&fs_info->buffer_lock); | |
4838 | radix_tree_preload_end(); | |
4839 | if (ret == -EEXIST) { | |
4840 | exists = find_extent_buffer(fs_info, start); | |
4841 | if (exists) | |
4842 | goto free_eb; | |
4843 | else | |
4844 | goto again; | |
4845 | } | |
4846 | check_buffer_tree_ref(eb); | |
4847 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
4848 | ||
4849 | /* | |
4850 | * We will free dummy extent buffer's if they come into | |
4851 | * free_extent_buffer with a ref count of 2, but if we are using this we | |
4852 | * want the buffers to stay in memory until we're done with them, so | |
4853 | * bump the ref count again. | |
4854 | */ | |
4855 | atomic_inc(&eb->refs); | |
4856 | return eb; | |
4857 | free_eb: | |
4858 | btrfs_release_extent_buffer(eb); | |
4859 | return exists; | |
4860 | } | |
4861 | #endif | |
4862 | ||
f28491e0 | 4863 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
ce3e6984 | 4864 | u64 start) |
d1310b2e | 4865 | { |
ce3e6984 | 4866 | unsigned long len = fs_info->tree_root->nodesize; |
d1310b2e CM |
4867 | unsigned long num_pages = num_extent_pages(start, len); |
4868 | unsigned long i; | |
4869 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
4870 | struct extent_buffer *eb; | |
6af118ce | 4871 | struct extent_buffer *exists = NULL; |
d1310b2e | 4872 | struct page *p; |
f28491e0 | 4873 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
d1310b2e | 4874 | int uptodate = 1; |
19fe0a8b | 4875 | int ret; |
d1310b2e | 4876 | |
f28491e0 | 4877 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 4878 | if (eb) |
6af118ce | 4879 | return eb; |
6af118ce | 4880 | |
23d79d81 | 4881 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 4882 | if (!eb) |
d1310b2e CM |
4883 | return NULL; |
4884 | ||
727011e0 | 4885 | for (i = 0; i < num_pages; i++, index++) { |
d1b5c567 | 4886 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
4804b382 | 4887 | if (!p) |
6af118ce | 4888 | goto free_eb; |
4f2de97a JB |
4889 | |
4890 | spin_lock(&mapping->private_lock); | |
4891 | if (PagePrivate(p)) { | |
4892 | /* | |
4893 | * We could have already allocated an eb for this page | |
4894 | * and attached one so lets see if we can get a ref on | |
4895 | * the existing eb, and if we can we know it's good and | |
4896 | * we can just return that one, else we know we can just | |
4897 | * overwrite page->private. | |
4898 | */ | |
4899 | exists = (struct extent_buffer *)p->private; | |
4900 | if (atomic_inc_not_zero(&exists->refs)) { | |
4901 | spin_unlock(&mapping->private_lock); | |
4902 | unlock_page(p); | |
17de39ac | 4903 | page_cache_release(p); |
2457aec6 | 4904 | mark_extent_buffer_accessed(exists, p); |
4f2de97a JB |
4905 | goto free_eb; |
4906 | } | |
5ca64f45 | 4907 | exists = NULL; |
4f2de97a | 4908 | |
0b32f4bb | 4909 | /* |
4f2de97a JB |
4910 | * Do this so attach doesn't complain and we need to |
4911 | * drop the ref the old guy had. | |
4912 | */ | |
4913 | ClearPagePrivate(p); | |
0b32f4bb | 4914 | WARN_ON(PageDirty(p)); |
4f2de97a | 4915 | page_cache_release(p); |
d1310b2e | 4916 | } |
4f2de97a JB |
4917 | attach_extent_buffer_page(eb, p); |
4918 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 4919 | WARN_ON(PageDirty(p)); |
727011e0 | 4920 | eb->pages[i] = p; |
d1310b2e CM |
4921 | if (!PageUptodate(p)) |
4922 | uptodate = 0; | |
eb14ab8e CM |
4923 | |
4924 | /* | |
4925 | * see below about how we avoid a nasty race with release page | |
4926 | * and why we unlock later | |
4927 | */ | |
d1310b2e CM |
4928 | } |
4929 | if (uptodate) | |
b4ce94de | 4930 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 4931 | again: |
19fe0a8b MX |
4932 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); |
4933 | if (ret) | |
4934 | goto free_eb; | |
4935 | ||
f28491e0 JB |
4936 | spin_lock(&fs_info->buffer_lock); |
4937 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
4938 | start >> PAGE_CACHE_SHIFT, eb); | |
4939 | spin_unlock(&fs_info->buffer_lock); | |
452c75c3 | 4940 | radix_tree_preload_end(); |
19fe0a8b | 4941 | if (ret == -EEXIST) { |
f28491e0 | 4942 | exists = find_extent_buffer(fs_info, start); |
452c75c3 CS |
4943 | if (exists) |
4944 | goto free_eb; | |
4945 | else | |
115391d2 | 4946 | goto again; |
6af118ce | 4947 | } |
6af118ce | 4948 | /* add one reference for the tree */ |
0b32f4bb | 4949 | check_buffer_tree_ref(eb); |
34b41ace | 4950 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
4951 | |
4952 | /* | |
4953 | * there is a race where release page may have | |
4954 | * tried to find this extent buffer in the radix | |
4955 | * but failed. It will tell the VM it is safe to | |
4956 | * reclaim the, and it will clear the page private bit. | |
4957 | * We must make sure to set the page private bit properly | |
4958 | * after the extent buffer is in the radix tree so | |
4959 | * it doesn't get lost | |
4960 | */ | |
727011e0 CM |
4961 | SetPageChecked(eb->pages[0]); |
4962 | for (i = 1; i < num_pages; i++) { | |
fb85fc9a | 4963 | p = eb->pages[i]; |
727011e0 CM |
4964 | ClearPageChecked(p); |
4965 | unlock_page(p); | |
4966 | } | |
4967 | unlock_page(eb->pages[0]); | |
d1310b2e CM |
4968 | return eb; |
4969 | ||
6af118ce | 4970 | free_eb: |
5ca64f45 | 4971 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
4972 | for (i = 0; i < num_pages; i++) { |
4973 | if (eb->pages[i]) | |
4974 | unlock_page(eb->pages[i]); | |
4975 | } | |
eb14ab8e | 4976 | |
897ca6e9 | 4977 | btrfs_release_extent_buffer(eb); |
6af118ce | 4978 | return exists; |
d1310b2e | 4979 | } |
d1310b2e | 4980 | |
3083ee2e JB |
4981 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
4982 | { | |
4983 | struct extent_buffer *eb = | |
4984 | container_of(head, struct extent_buffer, rcu_head); | |
4985 | ||
4986 | __free_extent_buffer(eb); | |
4987 | } | |
4988 | ||
3083ee2e | 4989 | /* Expects to have eb->eb_lock already held */ |
f7a52a40 | 4990 | static int release_extent_buffer(struct extent_buffer *eb) |
3083ee2e JB |
4991 | { |
4992 | WARN_ON(atomic_read(&eb->refs) == 0); | |
4993 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 4994 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 4995 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 4996 | |
815a51c7 | 4997 | spin_unlock(&eb->refs_lock); |
3083ee2e | 4998 | |
f28491e0 JB |
4999 | spin_lock(&fs_info->buffer_lock); |
5000 | radix_tree_delete(&fs_info->buffer_radix, | |
815a51c7 | 5001 | eb->start >> PAGE_CACHE_SHIFT); |
f28491e0 | 5002 | spin_unlock(&fs_info->buffer_lock); |
34b41ace JB |
5003 | } else { |
5004 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 5005 | } |
3083ee2e JB |
5006 | |
5007 | /* Should be safe to release our pages at this point */ | |
a50924e3 | 5008 | btrfs_release_extent_buffer_page(eb); |
bcb7e449 JB |
5009 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
5010 | if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) { | |
5011 | __free_extent_buffer(eb); | |
5012 | return 1; | |
5013 | } | |
5014 | #endif | |
3083ee2e | 5015 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 5016 | return 1; |
3083ee2e JB |
5017 | } |
5018 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
5019 | |
5020 | return 0; | |
3083ee2e JB |
5021 | } |
5022 | ||
d1310b2e CM |
5023 | void free_extent_buffer(struct extent_buffer *eb) |
5024 | { | |
242e18c7 CM |
5025 | int refs; |
5026 | int old; | |
d1310b2e CM |
5027 | if (!eb) |
5028 | return; | |
5029 | ||
242e18c7 CM |
5030 | while (1) { |
5031 | refs = atomic_read(&eb->refs); | |
5032 | if (refs <= 3) | |
5033 | break; | |
5034 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
5035 | if (old == refs) | |
5036 | return; | |
5037 | } | |
5038 | ||
3083ee2e | 5039 | spin_lock(&eb->refs_lock); |
815a51c7 JS |
5040 | if (atomic_read(&eb->refs) == 2 && |
5041 | test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) | |
5042 | atomic_dec(&eb->refs); | |
5043 | ||
3083ee2e JB |
5044 | if (atomic_read(&eb->refs) == 2 && |
5045 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 5046 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
5047 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5048 | atomic_dec(&eb->refs); | |
5049 | ||
5050 | /* | |
5051 | * I know this is terrible, but it's temporary until we stop tracking | |
5052 | * the uptodate bits and such for the extent buffers. | |
5053 | */ | |
f7a52a40 | 5054 | release_extent_buffer(eb); |
3083ee2e JB |
5055 | } |
5056 | ||
5057 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
5058 | { | |
5059 | if (!eb) | |
d1310b2e CM |
5060 | return; |
5061 | ||
3083ee2e JB |
5062 | spin_lock(&eb->refs_lock); |
5063 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
5064 | ||
0b32f4bb | 5065 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
5066 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5067 | atomic_dec(&eb->refs); | |
f7a52a40 | 5068 | release_extent_buffer(eb); |
d1310b2e | 5069 | } |
d1310b2e | 5070 | |
1d4284bd | 5071 | void clear_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5072 | { |
d1310b2e CM |
5073 | unsigned long i; |
5074 | unsigned long num_pages; | |
5075 | struct page *page; | |
5076 | ||
d1310b2e CM |
5077 | num_pages = num_extent_pages(eb->start, eb->len); |
5078 | ||
5079 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 5080 | page = eb->pages[i]; |
b9473439 | 5081 | if (!PageDirty(page)) |
d2c3f4f6 CM |
5082 | continue; |
5083 | ||
a61e6f29 | 5084 | lock_page(page); |
eb14ab8e CM |
5085 | WARN_ON(!PagePrivate(page)); |
5086 | ||
d1310b2e | 5087 | clear_page_dirty_for_io(page); |
0ee0fda0 | 5088 | spin_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
5089 | if (!PageDirty(page)) { |
5090 | radix_tree_tag_clear(&page->mapping->page_tree, | |
5091 | page_index(page), | |
5092 | PAGECACHE_TAG_DIRTY); | |
5093 | } | |
0ee0fda0 | 5094 | spin_unlock_irq(&page->mapping->tree_lock); |
bf0da8c1 | 5095 | ClearPageError(page); |
a61e6f29 | 5096 | unlock_page(page); |
d1310b2e | 5097 | } |
0b32f4bb | 5098 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 5099 | } |
d1310b2e | 5100 | |
0b32f4bb | 5101 | int set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e CM |
5102 | { |
5103 | unsigned long i; | |
5104 | unsigned long num_pages; | |
b9473439 | 5105 | int was_dirty = 0; |
d1310b2e | 5106 | |
0b32f4bb JB |
5107 | check_buffer_tree_ref(eb); |
5108 | ||
b9473439 | 5109 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 5110 | |
d1310b2e | 5111 | num_pages = num_extent_pages(eb->start, eb->len); |
3083ee2e | 5112 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
5113 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
5114 | ||
b9473439 | 5115 | for (i = 0; i < num_pages; i++) |
fb85fc9a | 5116 | set_page_dirty(eb->pages[i]); |
b9473439 | 5117 | return was_dirty; |
d1310b2e | 5118 | } |
d1310b2e | 5119 | |
69ba3927 | 5120 | void clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 CM |
5121 | { |
5122 | unsigned long i; | |
5123 | struct page *page; | |
5124 | unsigned long num_pages; | |
5125 | ||
b4ce94de | 5126 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
0b32f4bb | 5127 | num_pages = num_extent_pages(eb->start, eb->len); |
1259ab75 | 5128 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5129 | page = eb->pages[i]; |
33958dc6 CM |
5130 | if (page) |
5131 | ClearPageUptodate(page); | |
1259ab75 | 5132 | } |
1259ab75 CM |
5133 | } |
5134 | ||
09c25a8c | 5135 | void set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e CM |
5136 | { |
5137 | unsigned long i; | |
5138 | struct page *page; | |
5139 | unsigned long num_pages; | |
5140 | ||
0b32f4bb | 5141 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 5142 | num_pages = num_extent_pages(eb->start, eb->len); |
d1310b2e | 5143 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5144 | page = eb->pages[i]; |
d1310b2e CM |
5145 | SetPageUptodate(page); |
5146 | } | |
d1310b2e | 5147 | } |
d1310b2e | 5148 | |
0b32f4bb | 5149 | int extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 5150 | { |
0b32f4bb | 5151 | return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 5152 | } |
d1310b2e CM |
5153 | |
5154 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
bb82ab88 | 5155 | struct extent_buffer *eb, u64 start, int wait, |
f188591e | 5156 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
5157 | { |
5158 | unsigned long i; | |
5159 | unsigned long start_i; | |
5160 | struct page *page; | |
5161 | int err; | |
5162 | int ret = 0; | |
ce9adaa5 CM |
5163 | int locked_pages = 0; |
5164 | int all_uptodate = 1; | |
d1310b2e | 5165 | unsigned long num_pages; |
727011e0 | 5166 | unsigned long num_reads = 0; |
a86c12c7 | 5167 | struct bio *bio = NULL; |
c8b97818 | 5168 | unsigned long bio_flags = 0; |
a86c12c7 | 5169 | |
b4ce94de | 5170 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
5171 | return 0; |
5172 | ||
d1310b2e CM |
5173 | if (start) { |
5174 | WARN_ON(start < eb->start); | |
5175 | start_i = (start >> PAGE_CACHE_SHIFT) - | |
5176 | (eb->start >> PAGE_CACHE_SHIFT); | |
5177 | } else { | |
5178 | start_i = 0; | |
5179 | } | |
5180 | ||
5181 | num_pages = num_extent_pages(eb->start, eb->len); | |
5182 | for (i = start_i; i < num_pages; i++) { | |
fb85fc9a | 5183 | page = eb->pages[i]; |
bb82ab88 | 5184 | if (wait == WAIT_NONE) { |
2db04966 | 5185 | if (!trylock_page(page)) |
ce9adaa5 | 5186 | goto unlock_exit; |
d1310b2e CM |
5187 | } else { |
5188 | lock_page(page); | |
5189 | } | |
ce9adaa5 | 5190 | locked_pages++; |
727011e0 CM |
5191 | if (!PageUptodate(page)) { |
5192 | num_reads++; | |
ce9adaa5 | 5193 | all_uptodate = 0; |
727011e0 | 5194 | } |
ce9adaa5 CM |
5195 | } |
5196 | if (all_uptodate) { | |
5197 | if (start_i == 0) | |
b4ce94de | 5198 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
5199 | goto unlock_exit; |
5200 | } | |
5201 | ||
656f30db | 5202 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 5203 | eb->read_mirror = 0; |
0b32f4bb | 5204 | atomic_set(&eb->io_pages, num_reads); |
ce9adaa5 | 5205 | for (i = start_i; i < num_pages; i++) { |
fb85fc9a | 5206 | page = eb->pages[i]; |
ce9adaa5 | 5207 | if (!PageUptodate(page)) { |
f188591e | 5208 | ClearPageError(page); |
a86c12c7 | 5209 | err = __extent_read_full_page(tree, page, |
f188591e | 5210 | get_extent, &bio, |
d4c7ca86 JB |
5211 | mirror_num, &bio_flags, |
5212 | READ | REQ_META); | |
d397712b | 5213 | if (err) |
d1310b2e | 5214 | ret = err; |
d1310b2e CM |
5215 | } else { |
5216 | unlock_page(page); | |
5217 | } | |
5218 | } | |
5219 | ||
355808c2 | 5220 | if (bio) { |
d4c7ca86 JB |
5221 | err = submit_one_bio(READ | REQ_META, bio, mirror_num, |
5222 | bio_flags); | |
79787eaa JM |
5223 | if (err) |
5224 | return err; | |
355808c2 | 5225 | } |
a86c12c7 | 5226 | |
bb82ab88 | 5227 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 5228 | return ret; |
d397712b | 5229 | |
d1310b2e | 5230 | for (i = start_i; i < num_pages; i++) { |
fb85fc9a | 5231 | page = eb->pages[i]; |
d1310b2e | 5232 | wait_on_page_locked(page); |
d397712b | 5233 | if (!PageUptodate(page)) |
d1310b2e | 5234 | ret = -EIO; |
d1310b2e | 5235 | } |
d397712b | 5236 | |
d1310b2e | 5237 | return ret; |
ce9adaa5 CM |
5238 | |
5239 | unlock_exit: | |
5240 | i = start_i; | |
d397712b | 5241 | while (locked_pages > 0) { |
fb85fc9a | 5242 | page = eb->pages[i]; |
ce9adaa5 CM |
5243 | i++; |
5244 | unlock_page(page); | |
5245 | locked_pages--; | |
5246 | } | |
5247 | return ret; | |
d1310b2e | 5248 | } |
d1310b2e CM |
5249 | |
5250 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
5251 | unsigned long start, | |
5252 | unsigned long len) | |
5253 | { | |
5254 | size_t cur; | |
5255 | size_t offset; | |
5256 | struct page *page; | |
5257 | char *kaddr; | |
5258 | char *dst = (char *)dstv; | |
5259 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5260 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
5261 | |
5262 | WARN_ON(start > eb->len); | |
5263 | WARN_ON(start + len > eb->start + eb->len); | |
5264 | ||
778746b5 | 5265 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5266 | |
d397712b | 5267 | while (len > 0) { |
fb85fc9a | 5268 | page = eb->pages[i]; |
d1310b2e CM |
5269 | |
5270 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
a6591715 | 5271 | kaddr = page_address(page); |
d1310b2e | 5272 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
5273 | |
5274 | dst += cur; | |
5275 | len -= cur; | |
5276 | offset = 0; | |
5277 | i++; | |
5278 | } | |
5279 | } | |
d1310b2e | 5280 | |
550ac1d8 GH |
5281 | int read_extent_buffer_to_user(struct extent_buffer *eb, void __user *dstv, |
5282 | unsigned long start, | |
5283 | unsigned long len) | |
5284 | { | |
5285 | size_t cur; | |
5286 | size_t offset; | |
5287 | struct page *page; | |
5288 | char *kaddr; | |
5289 | char __user *dst = (char __user *)dstv; | |
5290 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5291 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5292 | int ret = 0; | |
5293 | ||
5294 | WARN_ON(start > eb->len); | |
5295 | WARN_ON(start + len > eb->start + eb->len); | |
5296 | ||
5297 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); | |
5298 | ||
5299 | while (len > 0) { | |
fb85fc9a | 5300 | page = eb->pages[i]; |
550ac1d8 GH |
5301 | |
5302 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
5303 | kaddr = page_address(page); | |
5304 | if (copy_to_user(dst, kaddr + offset, cur)) { | |
5305 | ret = -EFAULT; | |
5306 | break; | |
5307 | } | |
5308 | ||
5309 | dst += cur; | |
5310 | len -= cur; | |
5311 | offset = 0; | |
5312 | i++; | |
5313 | } | |
5314 | ||
5315 | return ret; | |
5316 | } | |
5317 | ||
d1310b2e | 5318 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, |
a6591715 | 5319 | unsigned long min_len, char **map, |
d1310b2e | 5320 | unsigned long *map_start, |
a6591715 | 5321 | unsigned long *map_len) |
d1310b2e CM |
5322 | { |
5323 | size_t offset = start & (PAGE_CACHE_SIZE - 1); | |
5324 | char *kaddr; | |
5325 | struct page *p; | |
5326 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5327 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5328 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
5329 | PAGE_CACHE_SHIFT; | |
5330 | ||
5331 | if (i != end_i) | |
5332 | return -EINVAL; | |
5333 | ||
5334 | if (i == 0) { | |
5335 | offset = start_offset; | |
5336 | *map_start = 0; | |
5337 | } else { | |
5338 | offset = 0; | |
5339 | *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; | |
5340 | } | |
d397712b | 5341 | |
d1310b2e | 5342 | if (start + min_len > eb->len) { |
31b1a2bd | 5343 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, " |
c1c9ff7c GU |
5344 | "wanted %lu %lu\n", |
5345 | eb->start, eb->len, start, min_len); | |
85026533 | 5346 | return -EINVAL; |
d1310b2e CM |
5347 | } |
5348 | ||
fb85fc9a | 5349 | p = eb->pages[i]; |
a6591715 | 5350 | kaddr = page_address(p); |
d1310b2e CM |
5351 | *map = kaddr + offset; |
5352 | *map_len = PAGE_CACHE_SIZE - offset; | |
5353 | return 0; | |
5354 | } | |
d1310b2e | 5355 | |
d1310b2e CM |
5356 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, |
5357 | unsigned long start, | |
5358 | unsigned long len) | |
5359 | { | |
5360 | size_t cur; | |
5361 | size_t offset; | |
5362 | struct page *page; | |
5363 | char *kaddr; | |
5364 | char *ptr = (char *)ptrv; | |
5365 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5366 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5367 | int ret = 0; | |
5368 | ||
5369 | WARN_ON(start > eb->len); | |
5370 | WARN_ON(start + len > eb->start + eb->len); | |
5371 | ||
778746b5 | 5372 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5373 | |
d397712b | 5374 | while (len > 0) { |
fb85fc9a | 5375 | page = eb->pages[i]; |
d1310b2e CM |
5376 | |
5377 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
5378 | ||
a6591715 | 5379 | kaddr = page_address(page); |
d1310b2e | 5380 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
5381 | if (ret) |
5382 | break; | |
5383 | ||
5384 | ptr += cur; | |
5385 | len -= cur; | |
5386 | offset = 0; | |
5387 | i++; | |
5388 | } | |
5389 | return ret; | |
5390 | } | |
d1310b2e CM |
5391 | |
5392 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
5393 | unsigned long start, unsigned long len) | |
5394 | { | |
5395 | size_t cur; | |
5396 | size_t offset; | |
5397 | struct page *page; | |
5398 | char *kaddr; | |
5399 | char *src = (char *)srcv; | |
5400 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5401 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5402 | ||
5403 | WARN_ON(start > eb->len); | |
5404 | WARN_ON(start + len > eb->start + eb->len); | |
5405 | ||
778746b5 | 5406 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5407 | |
d397712b | 5408 | while (len > 0) { |
fb85fc9a | 5409 | page = eb->pages[i]; |
d1310b2e CM |
5410 | WARN_ON(!PageUptodate(page)); |
5411 | ||
5412 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 5413 | kaddr = page_address(page); |
d1310b2e | 5414 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
5415 | |
5416 | src += cur; | |
5417 | len -= cur; | |
5418 | offset = 0; | |
5419 | i++; | |
5420 | } | |
5421 | } | |
d1310b2e CM |
5422 | |
5423 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
5424 | unsigned long start, unsigned long len) | |
5425 | { | |
5426 | size_t cur; | |
5427 | size_t offset; | |
5428 | struct page *page; | |
5429 | char *kaddr; | |
5430 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5431 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5432 | ||
5433 | WARN_ON(start > eb->len); | |
5434 | WARN_ON(start + len > eb->start + eb->len); | |
5435 | ||
778746b5 | 5436 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5437 | |
d397712b | 5438 | while (len > 0) { |
fb85fc9a | 5439 | page = eb->pages[i]; |
d1310b2e CM |
5440 | WARN_ON(!PageUptodate(page)); |
5441 | ||
5442 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 5443 | kaddr = page_address(page); |
d1310b2e | 5444 | memset(kaddr + offset, c, cur); |
d1310b2e CM |
5445 | |
5446 | len -= cur; | |
5447 | offset = 0; | |
5448 | i++; | |
5449 | } | |
5450 | } | |
d1310b2e CM |
5451 | |
5452 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
5453 | unsigned long dst_offset, unsigned long src_offset, | |
5454 | unsigned long len) | |
5455 | { | |
5456 | u64 dst_len = dst->len; | |
5457 | size_t cur; | |
5458 | size_t offset; | |
5459 | struct page *page; | |
5460 | char *kaddr; | |
5461 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5462 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
5463 | ||
5464 | WARN_ON(src->len != dst_len); | |
5465 | ||
5466 | offset = (start_offset + dst_offset) & | |
778746b5 | 5467 | (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5468 | |
d397712b | 5469 | while (len > 0) { |
fb85fc9a | 5470 | page = dst->pages[i]; |
d1310b2e CM |
5471 | WARN_ON(!PageUptodate(page)); |
5472 | ||
5473 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
5474 | ||
a6591715 | 5475 | kaddr = page_address(page); |
d1310b2e | 5476 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
5477 | |
5478 | src_offset += cur; | |
5479 | len -= cur; | |
5480 | offset = 0; | |
5481 | i++; | |
5482 | } | |
5483 | } | |
d1310b2e | 5484 | |
3e1e8bb7 OS |
5485 | /* |
5486 | * The extent buffer bitmap operations are done with byte granularity because | |
5487 | * bitmap items are not guaranteed to be aligned to a word and therefore a | |
5488 | * single word in a bitmap may straddle two pages in the extent buffer. | |
5489 | */ | |
5490 | #define BIT_BYTE(nr) ((nr) / BITS_PER_BYTE) | |
5491 | #define BYTE_MASK ((1 << BITS_PER_BYTE) - 1) | |
5492 | #define BITMAP_FIRST_BYTE_MASK(start) \ | |
5493 | ((BYTE_MASK << ((start) & (BITS_PER_BYTE - 1))) & BYTE_MASK) | |
5494 | #define BITMAP_LAST_BYTE_MASK(nbits) \ | |
5495 | (BYTE_MASK >> (-(nbits) & (BITS_PER_BYTE - 1))) | |
5496 | ||
5497 | /* | |
5498 | * eb_bitmap_offset() - calculate the page and offset of the byte containing the | |
5499 | * given bit number | |
5500 | * @eb: the extent buffer | |
5501 | * @start: offset of the bitmap item in the extent buffer | |
5502 | * @nr: bit number | |
5503 | * @page_index: return index of the page in the extent buffer that contains the | |
5504 | * given bit number | |
5505 | * @page_offset: return offset into the page given by page_index | |
5506 | * | |
5507 | * This helper hides the ugliness of finding the byte in an extent buffer which | |
5508 | * contains a given bit. | |
5509 | */ | |
5510 | static inline void eb_bitmap_offset(struct extent_buffer *eb, | |
5511 | unsigned long start, unsigned long nr, | |
5512 | unsigned long *page_index, | |
5513 | size_t *page_offset) | |
5514 | { | |
5515 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5516 | size_t byte_offset = BIT_BYTE(nr); | |
5517 | size_t offset; | |
5518 | ||
5519 | /* | |
5520 | * The byte we want is the offset of the extent buffer + the offset of | |
5521 | * the bitmap item in the extent buffer + the offset of the byte in the | |
5522 | * bitmap item. | |
5523 | */ | |
5524 | offset = start_offset + start + byte_offset; | |
5525 | ||
5526 | *page_index = offset >> PAGE_CACHE_SHIFT; | |
5527 | *page_offset = offset & (PAGE_CACHE_SIZE - 1); | |
5528 | } | |
5529 | ||
5530 | /** | |
5531 | * extent_buffer_test_bit - determine whether a bit in a bitmap item is set | |
5532 | * @eb: the extent buffer | |
5533 | * @start: offset of the bitmap item in the extent buffer | |
5534 | * @nr: bit number to test | |
5535 | */ | |
5536 | int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start, | |
5537 | unsigned long nr) | |
5538 | { | |
5539 | char *kaddr; | |
5540 | struct page *page; | |
5541 | unsigned long i; | |
5542 | size_t offset; | |
5543 | ||
5544 | eb_bitmap_offset(eb, start, nr, &i, &offset); | |
5545 | page = eb->pages[i]; | |
5546 | WARN_ON(!PageUptodate(page)); | |
5547 | kaddr = page_address(page); | |
5548 | return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1))); | |
5549 | } | |
5550 | ||
5551 | /** | |
5552 | * extent_buffer_bitmap_set - set an area of a bitmap | |
5553 | * @eb: the extent buffer | |
5554 | * @start: offset of the bitmap item in the extent buffer | |
5555 | * @pos: bit number of the first bit | |
5556 | * @len: number of bits to set | |
5557 | */ | |
5558 | void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start, | |
5559 | unsigned long pos, unsigned long len) | |
5560 | { | |
5561 | char *kaddr; | |
5562 | struct page *page; | |
5563 | unsigned long i; | |
5564 | size_t offset; | |
5565 | const unsigned int size = pos + len; | |
5566 | int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
5567 | unsigned int mask_to_set = BITMAP_FIRST_BYTE_MASK(pos); | |
5568 | ||
5569 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5570 | page = eb->pages[i]; | |
5571 | WARN_ON(!PageUptodate(page)); | |
5572 | kaddr = page_address(page); | |
5573 | ||
5574 | while (len >= bits_to_set) { | |
5575 | kaddr[offset] |= mask_to_set; | |
5576 | len -= bits_to_set; | |
5577 | bits_to_set = BITS_PER_BYTE; | |
5578 | mask_to_set = ~0U; | |
5579 | if (++offset >= PAGE_CACHE_SIZE && len > 0) { | |
5580 | offset = 0; | |
5581 | page = eb->pages[++i]; | |
5582 | WARN_ON(!PageUptodate(page)); | |
5583 | kaddr = page_address(page); | |
5584 | } | |
5585 | } | |
5586 | if (len) { | |
5587 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
5588 | kaddr[offset] |= mask_to_set; | |
5589 | } | |
5590 | } | |
5591 | ||
5592 | ||
5593 | /** | |
5594 | * extent_buffer_bitmap_clear - clear an area of a bitmap | |
5595 | * @eb: the extent buffer | |
5596 | * @start: offset of the bitmap item in the extent buffer | |
5597 | * @pos: bit number of the first bit | |
5598 | * @len: number of bits to clear | |
5599 | */ | |
5600 | void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start, | |
5601 | unsigned long pos, unsigned long len) | |
5602 | { | |
5603 | char *kaddr; | |
5604 | struct page *page; | |
5605 | unsigned long i; | |
5606 | size_t offset; | |
5607 | const unsigned int size = pos + len; | |
5608 | int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
5609 | unsigned int mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos); | |
5610 | ||
5611 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5612 | page = eb->pages[i]; | |
5613 | WARN_ON(!PageUptodate(page)); | |
5614 | kaddr = page_address(page); | |
5615 | ||
5616 | while (len >= bits_to_clear) { | |
5617 | kaddr[offset] &= ~mask_to_clear; | |
5618 | len -= bits_to_clear; | |
5619 | bits_to_clear = BITS_PER_BYTE; | |
5620 | mask_to_clear = ~0U; | |
5621 | if (++offset >= PAGE_CACHE_SIZE && len > 0) { | |
5622 | offset = 0; | |
5623 | page = eb->pages[++i]; | |
5624 | WARN_ON(!PageUptodate(page)); | |
5625 | kaddr = page_address(page); | |
5626 | } | |
5627 | } | |
5628 | if (len) { | |
5629 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
5630 | kaddr[offset] &= ~mask_to_clear; | |
5631 | } | |
5632 | } | |
5633 | ||
3387206f ST |
5634 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
5635 | { | |
5636 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
5637 | return distance < len; | |
5638 | } | |
5639 | ||
d1310b2e CM |
5640 | static void copy_pages(struct page *dst_page, struct page *src_page, |
5641 | unsigned long dst_off, unsigned long src_off, | |
5642 | unsigned long len) | |
5643 | { | |
a6591715 | 5644 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 5645 | char *src_kaddr; |
727011e0 | 5646 | int must_memmove = 0; |
d1310b2e | 5647 | |
3387206f | 5648 | if (dst_page != src_page) { |
a6591715 | 5649 | src_kaddr = page_address(src_page); |
3387206f | 5650 | } else { |
d1310b2e | 5651 | src_kaddr = dst_kaddr; |
727011e0 CM |
5652 | if (areas_overlap(src_off, dst_off, len)) |
5653 | must_memmove = 1; | |
3387206f | 5654 | } |
d1310b2e | 5655 | |
727011e0 CM |
5656 | if (must_memmove) |
5657 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
5658 | else | |
5659 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
5660 | } |
5661 | ||
5662 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5663 | unsigned long src_offset, unsigned long len) | |
5664 | { | |
5665 | size_t cur; | |
5666 | size_t dst_off_in_page; | |
5667 | size_t src_off_in_page; | |
5668 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5669 | unsigned long dst_i; | |
5670 | unsigned long src_i; | |
5671 | ||
5672 | if (src_offset + len > dst->len) { | |
f14d104d DS |
5673 | btrfs_err(dst->fs_info, |
5674 | "memmove bogus src_offset %lu move " | |
5675 | "len %lu dst len %lu", src_offset, len, dst->len); | |
d1310b2e CM |
5676 | BUG_ON(1); |
5677 | } | |
5678 | if (dst_offset + len > dst->len) { | |
f14d104d DS |
5679 | btrfs_err(dst->fs_info, |
5680 | "memmove bogus dst_offset %lu move " | |
5681 | "len %lu dst len %lu", dst_offset, len, dst->len); | |
d1310b2e CM |
5682 | BUG_ON(1); |
5683 | } | |
5684 | ||
d397712b | 5685 | while (len > 0) { |
d1310b2e | 5686 | dst_off_in_page = (start_offset + dst_offset) & |
778746b5 | 5687 | (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5688 | src_off_in_page = (start_offset + src_offset) & |
778746b5 | 5689 | (PAGE_CACHE_SIZE - 1); |
d1310b2e CM |
5690 | |
5691 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
5692 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
5693 | ||
5694 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
5695 | src_off_in_page)); | |
5696 | cur = min_t(unsigned long, cur, | |
5697 | (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); | |
5698 | ||
fb85fc9a | 5699 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5700 | dst_off_in_page, src_off_in_page, cur); |
5701 | ||
5702 | src_offset += cur; | |
5703 | dst_offset += cur; | |
5704 | len -= cur; | |
5705 | } | |
5706 | } | |
d1310b2e CM |
5707 | |
5708 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5709 | unsigned long src_offset, unsigned long len) | |
5710 | { | |
5711 | size_t cur; | |
5712 | size_t dst_off_in_page; | |
5713 | size_t src_off_in_page; | |
5714 | unsigned long dst_end = dst_offset + len - 1; | |
5715 | unsigned long src_end = src_offset + len - 1; | |
5716 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5717 | unsigned long dst_i; | |
5718 | unsigned long src_i; | |
5719 | ||
5720 | if (src_offset + len > dst->len) { | |
f14d104d DS |
5721 | btrfs_err(dst->fs_info, "memmove bogus src_offset %lu move " |
5722 | "len %lu len %lu", src_offset, len, dst->len); | |
d1310b2e CM |
5723 | BUG_ON(1); |
5724 | } | |
5725 | if (dst_offset + len > dst->len) { | |
f14d104d DS |
5726 | btrfs_err(dst->fs_info, "memmove bogus dst_offset %lu move " |
5727 | "len %lu len %lu", dst_offset, len, dst->len); | |
d1310b2e CM |
5728 | BUG_ON(1); |
5729 | } | |
727011e0 | 5730 | if (dst_offset < src_offset) { |
d1310b2e CM |
5731 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
5732 | return; | |
5733 | } | |
d397712b | 5734 | while (len > 0) { |
d1310b2e CM |
5735 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; |
5736 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
5737 | ||
5738 | dst_off_in_page = (start_offset + dst_end) & | |
778746b5 | 5739 | (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5740 | src_off_in_page = (start_offset + src_end) & |
778746b5 | 5741 | (PAGE_CACHE_SIZE - 1); |
d1310b2e CM |
5742 | |
5743 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
5744 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 5745 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5746 | dst_off_in_page - cur + 1, |
5747 | src_off_in_page - cur + 1, cur); | |
5748 | ||
5749 | dst_end -= cur; | |
5750 | src_end -= cur; | |
5751 | len -= cur; | |
5752 | } | |
5753 | } | |
6af118ce | 5754 | |
f7a52a40 | 5755 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 5756 | { |
6af118ce | 5757 | struct extent_buffer *eb; |
6af118ce | 5758 | |
3083ee2e JB |
5759 | /* |
5760 | * We need to make sure noboody is attaching this page to an eb right | |
5761 | * now. | |
5762 | */ | |
5763 | spin_lock(&page->mapping->private_lock); | |
5764 | if (!PagePrivate(page)) { | |
5765 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 5766 | return 1; |
45f49bce | 5767 | } |
6af118ce | 5768 | |
3083ee2e JB |
5769 | eb = (struct extent_buffer *)page->private; |
5770 | BUG_ON(!eb); | |
19fe0a8b MX |
5771 | |
5772 | /* | |
3083ee2e JB |
5773 | * This is a little awful but should be ok, we need to make sure that |
5774 | * the eb doesn't disappear out from under us while we're looking at | |
5775 | * this page. | |
19fe0a8b | 5776 | */ |
3083ee2e | 5777 | spin_lock(&eb->refs_lock); |
0b32f4bb | 5778 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
5779 | spin_unlock(&eb->refs_lock); |
5780 | spin_unlock(&page->mapping->private_lock); | |
5781 | return 0; | |
b9473439 | 5782 | } |
3083ee2e | 5783 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 5784 | |
19fe0a8b | 5785 | /* |
3083ee2e JB |
5786 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
5787 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 5788 | */ |
3083ee2e JB |
5789 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
5790 | spin_unlock(&eb->refs_lock); | |
5791 | return 0; | |
b9473439 | 5792 | } |
19fe0a8b | 5793 | |
f7a52a40 | 5794 | return release_extent_buffer(eb); |
6af118ce | 5795 | } |