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 CM |
1287 | /* wrappers around set/clear extent bit */ |
1288 | int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
1289 | gfp_t mask) | |
1290 | { | |
3fbe5c02 | 1291 | return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL, |
2c64c53d | 1292 | NULL, mask); |
d1310b2e | 1293 | } |
d1310b2e CM |
1294 | |
1295 | int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1296 | unsigned bits, gfp_t mask) |
d1310b2e | 1297 | { |
3fbe5c02 | 1298 | return set_extent_bit(tree, start, end, bits, NULL, |
2c64c53d | 1299 | NULL, mask); |
d1310b2e | 1300 | } |
d1310b2e | 1301 | |
d38ed27f QW |
1302 | int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
1303 | unsigned bits, gfp_t mask, | |
1304 | struct extent_changeset *changeset) | |
1305 | { | |
1306 | /* | |
1307 | * We don't support EXTENT_LOCKED yet, as current changeset will | |
1308 | * record any bits changed, so for EXTENT_LOCKED case, it will | |
1309 | * either fail with -EEXIST or changeset will record the whole | |
1310 | * range. | |
1311 | */ | |
1312 | BUG_ON(bits & EXTENT_LOCKED); | |
1313 | ||
1314 | return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, mask, | |
1315 | changeset); | |
1316 | } | |
1317 | ||
fefdc557 QW |
1318 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
1319 | unsigned bits, int wake, int delete, | |
1320 | struct extent_state **cached, gfp_t mask) | |
1321 | { | |
1322 | return __clear_extent_bit(tree, start, end, bits, wake, delete, | |
1323 | cached, mask, NULL); | |
1324 | } | |
1325 | ||
d1310b2e | 1326 | int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
9ee49a04 | 1327 | unsigned bits, gfp_t mask) |
d1310b2e | 1328 | { |
0f31871f FM |
1329 | int wake = 0; |
1330 | ||
1331 | if (bits & EXTENT_LOCKED) | |
1332 | wake = 1; | |
1333 | ||
1334 | return clear_extent_bit(tree, start, end, bits, wake, 0, NULL, mask); | |
d1310b2e | 1335 | } |
d1310b2e | 1336 | |
fefdc557 QW |
1337 | int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
1338 | unsigned bits, gfp_t mask, | |
1339 | struct extent_changeset *changeset) | |
1340 | { | |
1341 | /* | |
1342 | * Don't support EXTENT_LOCKED case, same reason as | |
1343 | * set_record_extent_bits(). | |
1344 | */ | |
1345 | BUG_ON(bits & EXTENT_LOCKED); | |
1346 | ||
1347 | return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask, | |
1348 | changeset); | |
1349 | } | |
1350 | ||
d1310b2e | 1351 | int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, |
2ac55d41 | 1352 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
1353 | { |
1354 | return set_extent_bit(tree, start, end, | |
fee187d9 | 1355 | EXTENT_DELALLOC | EXTENT_UPTODATE, |
3fbe5c02 | 1356 | NULL, cached_state, mask); |
d1310b2e | 1357 | } |
d1310b2e | 1358 | |
9e8a4a8b LB |
1359 | int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end, |
1360 | struct extent_state **cached_state, gfp_t mask) | |
1361 | { | |
1362 | return set_extent_bit(tree, start, end, | |
1363 | EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG, | |
1364 | NULL, cached_state, mask); | |
1365 | } | |
1366 | ||
d1310b2e CM |
1367 | int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, |
1368 | gfp_t mask) | |
1369 | { | |
1370 | return clear_extent_bit(tree, start, end, | |
32c00aff | 1371 | EXTENT_DIRTY | EXTENT_DELALLOC | |
0ca1f7ce | 1372 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask); |
d1310b2e | 1373 | } |
d1310b2e CM |
1374 | |
1375 | int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, | |
1376 | gfp_t mask) | |
1377 | { | |
3fbe5c02 | 1378 | return set_extent_bit(tree, start, end, EXTENT_NEW, NULL, |
2c64c53d | 1379 | NULL, mask); |
d1310b2e | 1380 | } |
d1310b2e | 1381 | |
d1310b2e | 1382 | int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
507903b8 | 1383 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e | 1384 | { |
6b67a320 | 1385 | return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL, |
3fbe5c02 | 1386 | cached_state, mask); |
d1310b2e | 1387 | } |
d1310b2e | 1388 | |
5fd02043 JB |
1389 | int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
1390 | struct extent_state **cached_state, gfp_t mask) | |
d1310b2e | 1391 | { |
2c64c53d | 1392 | return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, |
2ac55d41 | 1393 | cached_state, mask); |
d1310b2e | 1394 | } |
d1310b2e | 1395 | |
d352ac68 CM |
1396 | /* |
1397 | * either insert or lock state struct between start and end use mask to tell | |
1398 | * us if waiting is desired. | |
1399 | */ | |
1edbb734 | 1400 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
9ee49a04 | 1401 | unsigned bits, struct extent_state **cached_state) |
d1310b2e CM |
1402 | { |
1403 | int err; | |
1404 | u64 failed_start; | |
9ee49a04 | 1405 | |
d1310b2e | 1406 | while (1) { |
3fbe5c02 JM |
1407 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits, |
1408 | EXTENT_LOCKED, &failed_start, | |
d38ed27f | 1409 | cached_state, GFP_NOFS, NULL); |
d0082371 | 1410 | if (err == -EEXIST) { |
d1310b2e CM |
1411 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1412 | start = failed_start; | |
d0082371 | 1413 | } else |
d1310b2e | 1414 | break; |
d1310b2e CM |
1415 | WARN_ON(start > end); |
1416 | } | |
1417 | return err; | |
1418 | } | |
d1310b2e | 1419 | |
d0082371 | 1420 | int lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
1edbb734 | 1421 | { |
d0082371 | 1422 | return lock_extent_bits(tree, start, end, 0, NULL); |
1edbb734 CM |
1423 | } |
1424 | ||
d0082371 | 1425 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1426 | { |
1427 | int err; | |
1428 | u64 failed_start; | |
1429 | ||
3fbe5c02 | 1430 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
d38ed27f | 1431 | &failed_start, NULL, GFP_NOFS, NULL); |
6643558d YZ |
1432 | if (err == -EEXIST) { |
1433 | if (failed_start > start) | |
1434 | clear_extent_bit(tree, start, failed_start - 1, | |
d0082371 | 1435 | EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS); |
25179201 | 1436 | return 0; |
6643558d | 1437 | } |
25179201 JB |
1438 | return 1; |
1439 | } | |
25179201 | 1440 | |
2c64c53d CM |
1441 | int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end, |
1442 | struct extent_state **cached, gfp_t mask) | |
1443 | { | |
1444 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, | |
1445 | mask); | |
1446 | } | |
1447 | ||
d0082371 | 1448 | int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e | 1449 | { |
2c64c53d | 1450 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL, |
d0082371 | 1451 | GFP_NOFS); |
d1310b2e | 1452 | } |
d1310b2e | 1453 | |
4adaa611 CM |
1454 | int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
1455 | { | |
1456 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1457 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1458 | struct page *page; | |
1459 | ||
1460 | while (index <= end_index) { | |
1461 | page = find_get_page(inode->i_mapping, index); | |
1462 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1463 | clear_page_dirty_for_io(page); | |
1464 | page_cache_release(page); | |
1465 | index++; | |
1466 | } | |
1467 | return 0; | |
1468 | } | |
1469 | ||
1470 | int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) | |
1471 | { | |
1472 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1473 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1474 | struct page *page; | |
1475 | ||
1476 | while (index <= end_index) { | |
1477 | page = find_get_page(inode->i_mapping, index); | |
1478 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
4adaa611 | 1479 | __set_page_dirty_nobuffers(page); |
8d38633c | 1480 | account_page_redirty(page); |
4adaa611 CM |
1481 | page_cache_release(page); |
1482 | index++; | |
1483 | } | |
1484 | return 0; | |
1485 | } | |
1486 | ||
d1310b2e CM |
1487 | /* |
1488 | * helper function to set both pages and extents in the tree writeback | |
1489 | */ | |
b2950863 | 1490 | static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e CM |
1491 | { |
1492 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1493 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1494 | struct page *page; | |
1495 | ||
1496 | while (index <= end_index) { | |
1497 | page = find_get_page(tree->mapping, index); | |
79787eaa | 1498 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ |
d1310b2e CM |
1499 | set_page_writeback(page); |
1500 | page_cache_release(page); | |
1501 | index++; | |
1502 | } | |
d1310b2e CM |
1503 | return 0; |
1504 | } | |
d1310b2e | 1505 | |
d352ac68 CM |
1506 | /* find the first state struct with 'bits' set after 'start', and |
1507 | * return it. tree->lock must be held. NULL will returned if | |
1508 | * nothing was found after 'start' | |
1509 | */ | |
48a3b636 ES |
1510 | static struct extent_state * |
1511 | find_first_extent_bit_state(struct extent_io_tree *tree, | |
9ee49a04 | 1512 | u64 start, unsigned bits) |
d7fc640e CM |
1513 | { |
1514 | struct rb_node *node; | |
1515 | struct extent_state *state; | |
1516 | ||
1517 | /* | |
1518 | * this search will find all the extents that end after | |
1519 | * our range starts. | |
1520 | */ | |
1521 | node = tree_search(tree, start); | |
d397712b | 1522 | if (!node) |
d7fc640e | 1523 | goto out; |
d7fc640e | 1524 | |
d397712b | 1525 | while (1) { |
d7fc640e | 1526 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1527 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1528 | return state; |
d397712b | 1529 | |
d7fc640e CM |
1530 | node = rb_next(node); |
1531 | if (!node) | |
1532 | break; | |
1533 | } | |
1534 | out: | |
1535 | return NULL; | |
1536 | } | |
d7fc640e | 1537 | |
69261c4b XG |
1538 | /* |
1539 | * find the first offset in the io tree with 'bits' set. zero is | |
1540 | * returned if we find something, and *start_ret and *end_ret are | |
1541 | * set to reflect the state struct that was found. | |
1542 | * | |
477d7eaf | 1543 | * If nothing was found, 1 is returned. If found something, return 0. |
69261c4b XG |
1544 | */ |
1545 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
9ee49a04 | 1546 | u64 *start_ret, u64 *end_ret, unsigned bits, |
e6138876 | 1547 | struct extent_state **cached_state) |
69261c4b XG |
1548 | { |
1549 | struct extent_state *state; | |
e6138876 | 1550 | struct rb_node *n; |
69261c4b XG |
1551 | int ret = 1; |
1552 | ||
1553 | spin_lock(&tree->lock); | |
e6138876 JB |
1554 | if (cached_state && *cached_state) { |
1555 | state = *cached_state; | |
27a3507d | 1556 | if (state->end == start - 1 && extent_state_in_tree(state)) { |
e6138876 JB |
1557 | n = rb_next(&state->rb_node); |
1558 | while (n) { | |
1559 | state = rb_entry(n, struct extent_state, | |
1560 | rb_node); | |
1561 | if (state->state & bits) | |
1562 | goto got_it; | |
1563 | n = rb_next(n); | |
1564 | } | |
1565 | free_extent_state(*cached_state); | |
1566 | *cached_state = NULL; | |
1567 | goto out; | |
1568 | } | |
1569 | free_extent_state(*cached_state); | |
1570 | *cached_state = NULL; | |
1571 | } | |
1572 | ||
69261c4b | 1573 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1574 | got_it: |
69261c4b | 1575 | if (state) { |
e38e2ed7 | 1576 | cache_state_if_flags(state, cached_state, 0); |
69261c4b XG |
1577 | *start_ret = state->start; |
1578 | *end_ret = state->end; | |
1579 | ret = 0; | |
1580 | } | |
e6138876 | 1581 | out: |
69261c4b XG |
1582 | spin_unlock(&tree->lock); |
1583 | return ret; | |
1584 | } | |
1585 | ||
d352ac68 CM |
1586 | /* |
1587 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1588 | * more than 'max_bytes'. start and end are used to return the range, | |
1589 | * | |
1590 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1591 | */ | |
c8b97818 | 1592 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1593 | u64 *start, u64 *end, u64 max_bytes, |
1594 | struct extent_state **cached_state) | |
d1310b2e CM |
1595 | { |
1596 | struct rb_node *node; | |
1597 | struct extent_state *state; | |
1598 | u64 cur_start = *start; | |
1599 | u64 found = 0; | |
1600 | u64 total_bytes = 0; | |
1601 | ||
cad321ad | 1602 | spin_lock(&tree->lock); |
c8b97818 | 1603 | |
d1310b2e CM |
1604 | /* |
1605 | * this search will find all the extents that end after | |
1606 | * our range starts. | |
1607 | */ | |
80ea96b1 | 1608 | node = tree_search(tree, cur_start); |
2b114d1d | 1609 | if (!node) { |
3b951516 CM |
1610 | if (!found) |
1611 | *end = (u64)-1; | |
d1310b2e CM |
1612 | goto out; |
1613 | } | |
1614 | ||
d397712b | 1615 | while (1) { |
d1310b2e | 1616 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1617 | if (found && (state->start != cur_start || |
1618 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1619 | goto out; |
1620 | } | |
1621 | if (!(state->state & EXTENT_DELALLOC)) { | |
1622 | if (!found) | |
1623 | *end = state->end; | |
1624 | goto out; | |
1625 | } | |
c2a128d2 | 1626 | if (!found) { |
d1310b2e | 1627 | *start = state->start; |
c2a128d2 JB |
1628 | *cached_state = state; |
1629 | atomic_inc(&state->refs); | |
1630 | } | |
d1310b2e CM |
1631 | found++; |
1632 | *end = state->end; | |
1633 | cur_start = state->end + 1; | |
1634 | node = rb_next(node); | |
d1310b2e | 1635 | total_bytes += state->end - state->start + 1; |
7bf811a5 | 1636 | if (total_bytes >= max_bytes) |
573aecaf | 1637 | break; |
573aecaf | 1638 | if (!node) |
d1310b2e CM |
1639 | break; |
1640 | } | |
1641 | out: | |
cad321ad | 1642 | spin_unlock(&tree->lock); |
d1310b2e CM |
1643 | return found; |
1644 | } | |
1645 | ||
143bede5 JM |
1646 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1647 | struct page *locked_page, | |
1648 | u64 start, u64 end) | |
c8b97818 CM |
1649 | { |
1650 | int ret; | |
1651 | struct page *pages[16]; | |
1652 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1653 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1654 | unsigned long nr_pages = end_index - index + 1; | |
1655 | int i; | |
1656 | ||
1657 | if (index == locked_page->index && end_index == index) | |
143bede5 | 1658 | return; |
c8b97818 | 1659 | |
d397712b | 1660 | while (nr_pages > 0) { |
c8b97818 | 1661 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1662 | min_t(unsigned long, nr_pages, |
1663 | ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1664 | for (i = 0; i < ret; i++) { |
1665 | if (pages[i] != locked_page) | |
1666 | unlock_page(pages[i]); | |
1667 | page_cache_release(pages[i]); | |
1668 | } | |
1669 | nr_pages -= ret; | |
1670 | index += ret; | |
1671 | cond_resched(); | |
1672 | } | |
c8b97818 CM |
1673 | } |
1674 | ||
1675 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1676 | struct page *locked_page, | |
1677 | u64 delalloc_start, | |
1678 | u64 delalloc_end) | |
1679 | { | |
1680 | unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT; | |
1681 | unsigned long start_index = index; | |
1682 | unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT; | |
1683 | unsigned long pages_locked = 0; | |
1684 | struct page *pages[16]; | |
1685 | unsigned long nrpages; | |
1686 | int ret; | |
1687 | int i; | |
1688 | ||
1689 | /* the caller is responsible for locking the start index */ | |
1690 | if (index == locked_page->index && index == end_index) | |
1691 | return 0; | |
1692 | ||
1693 | /* skip the page at the start index */ | |
1694 | nrpages = end_index - index + 1; | |
d397712b | 1695 | while (nrpages > 0) { |
c8b97818 | 1696 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1697 | min_t(unsigned long, |
1698 | nrpages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1699 | if (ret == 0) { |
1700 | ret = -EAGAIN; | |
1701 | goto done; | |
1702 | } | |
1703 | /* now we have an array of pages, lock them all */ | |
1704 | for (i = 0; i < ret; i++) { | |
1705 | /* | |
1706 | * the caller is taking responsibility for | |
1707 | * locked_page | |
1708 | */ | |
771ed689 | 1709 | if (pages[i] != locked_page) { |
c8b97818 | 1710 | lock_page(pages[i]); |
f2b1c41c CM |
1711 | if (!PageDirty(pages[i]) || |
1712 | pages[i]->mapping != inode->i_mapping) { | |
771ed689 CM |
1713 | ret = -EAGAIN; |
1714 | unlock_page(pages[i]); | |
1715 | page_cache_release(pages[i]); | |
1716 | goto done; | |
1717 | } | |
1718 | } | |
c8b97818 | 1719 | page_cache_release(pages[i]); |
771ed689 | 1720 | pages_locked++; |
c8b97818 | 1721 | } |
c8b97818 CM |
1722 | nrpages -= ret; |
1723 | index += ret; | |
1724 | cond_resched(); | |
1725 | } | |
1726 | ret = 0; | |
1727 | done: | |
1728 | if (ret && pages_locked) { | |
1729 | __unlock_for_delalloc(inode, locked_page, | |
1730 | delalloc_start, | |
1731 | ((u64)(start_index + pages_locked - 1)) << | |
1732 | PAGE_CACHE_SHIFT); | |
1733 | } | |
1734 | return ret; | |
1735 | } | |
1736 | ||
1737 | /* | |
1738 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1739 | * more than 'max_bytes'. start and end are used to return the range, | |
1740 | * | |
1741 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1742 | */ | |
294e30fe JB |
1743 | STATIC u64 find_lock_delalloc_range(struct inode *inode, |
1744 | struct extent_io_tree *tree, | |
1745 | struct page *locked_page, u64 *start, | |
1746 | u64 *end, u64 max_bytes) | |
c8b97818 CM |
1747 | { |
1748 | u64 delalloc_start; | |
1749 | u64 delalloc_end; | |
1750 | u64 found; | |
9655d298 | 1751 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1752 | int ret; |
1753 | int loops = 0; | |
1754 | ||
1755 | again: | |
1756 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1757 | delalloc_start = *start; | |
1758 | delalloc_end = 0; | |
1759 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1760 | max_bytes, &cached_state); |
70b99e69 | 1761 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1762 | *start = delalloc_start; |
1763 | *end = delalloc_end; | |
c2a128d2 | 1764 | free_extent_state(cached_state); |
385fe0be | 1765 | return 0; |
c8b97818 CM |
1766 | } |
1767 | ||
70b99e69 CM |
1768 | /* |
1769 | * start comes from the offset of locked_page. We have to lock | |
1770 | * pages in order, so we can't process delalloc bytes before | |
1771 | * locked_page | |
1772 | */ | |
d397712b | 1773 | if (delalloc_start < *start) |
70b99e69 | 1774 | delalloc_start = *start; |
70b99e69 | 1775 | |
c8b97818 CM |
1776 | /* |
1777 | * make sure to limit the number of pages we try to lock down | |
c8b97818 | 1778 | */ |
7bf811a5 JB |
1779 | if (delalloc_end + 1 - delalloc_start > max_bytes) |
1780 | delalloc_end = delalloc_start + max_bytes - 1; | |
d397712b | 1781 | |
c8b97818 CM |
1782 | /* step two, lock all the pages after the page that has start */ |
1783 | ret = lock_delalloc_pages(inode, locked_page, | |
1784 | delalloc_start, delalloc_end); | |
1785 | if (ret == -EAGAIN) { | |
1786 | /* some of the pages are gone, lets avoid looping by | |
1787 | * shortening the size of the delalloc range we're searching | |
1788 | */ | |
9655d298 | 1789 | free_extent_state(cached_state); |
7d788742 | 1790 | cached_state = NULL; |
c8b97818 | 1791 | if (!loops) { |
7bf811a5 | 1792 | max_bytes = PAGE_CACHE_SIZE; |
c8b97818 CM |
1793 | loops = 1; |
1794 | goto again; | |
1795 | } else { | |
1796 | found = 0; | |
1797 | goto out_failed; | |
1798 | } | |
1799 | } | |
79787eaa | 1800 | BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */ |
c8b97818 CM |
1801 | |
1802 | /* step three, lock the state bits for the whole range */ | |
d0082371 | 1803 | lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state); |
c8b97818 CM |
1804 | |
1805 | /* then test to make sure it is all still delalloc */ | |
1806 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1807 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1808 | if (!ret) { |
9655d298 CM |
1809 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
1810 | &cached_state, GFP_NOFS); | |
c8b97818 CM |
1811 | __unlock_for_delalloc(inode, locked_page, |
1812 | delalloc_start, delalloc_end); | |
1813 | cond_resched(); | |
1814 | goto again; | |
1815 | } | |
9655d298 | 1816 | free_extent_state(cached_state); |
c8b97818 CM |
1817 | *start = delalloc_start; |
1818 | *end = delalloc_end; | |
1819 | out_failed: | |
1820 | return found; | |
1821 | } | |
1822 | ||
c2790a2e JB |
1823 | int extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end, |
1824 | struct page *locked_page, | |
9ee49a04 | 1825 | unsigned clear_bits, |
c2790a2e | 1826 | unsigned long page_ops) |
c8b97818 | 1827 | { |
c2790a2e | 1828 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
c8b97818 CM |
1829 | int ret; |
1830 | struct page *pages[16]; | |
1831 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1832 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1833 | unsigned long nr_pages = end_index - index + 1; | |
1834 | int i; | |
771ed689 | 1835 | |
2c64c53d | 1836 | clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); |
c2790a2e | 1837 | if (page_ops == 0) |
771ed689 | 1838 | return 0; |
c8b97818 | 1839 | |
704de49d FM |
1840 | if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0) |
1841 | mapping_set_error(inode->i_mapping, -EIO); | |
1842 | ||
d397712b | 1843 | while (nr_pages > 0) { |
c8b97818 | 1844 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1845 | min_t(unsigned long, |
1846 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 | 1847 | for (i = 0; i < ret; i++) { |
8b62b72b | 1848 | |
c2790a2e | 1849 | if (page_ops & PAGE_SET_PRIVATE2) |
8b62b72b CM |
1850 | SetPagePrivate2(pages[i]); |
1851 | ||
c8b97818 CM |
1852 | if (pages[i] == locked_page) { |
1853 | page_cache_release(pages[i]); | |
1854 | continue; | |
1855 | } | |
c2790a2e | 1856 | if (page_ops & PAGE_CLEAR_DIRTY) |
c8b97818 | 1857 | clear_page_dirty_for_io(pages[i]); |
c2790a2e | 1858 | if (page_ops & PAGE_SET_WRITEBACK) |
c8b97818 | 1859 | set_page_writeback(pages[i]); |
704de49d FM |
1860 | if (page_ops & PAGE_SET_ERROR) |
1861 | SetPageError(pages[i]); | |
c2790a2e | 1862 | if (page_ops & PAGE_END_WRITEBACK) |
c8b97818 | 1863 | end_page_writeback(pages[i]); |
c2790a2e | 1864 | if (page_ops & PAGE_UNLOCK) |
771ed689 | 1865 | unlock_page(pages[i]); |
c8b97818 CM |
1866 | page_cache_release(pages[i]); |
1867 | } | |
1868 | nr_pages -= ret; | |
1869 | index += ret; | |
1870 | cond_resched(); | |
1871 | } | |
1872 | return 0; | |
1873 | } | |
c8b97818 | 1874 | |
d352ac68 CM |
1875 | /* |
1876 | * count the number of bytes in the tree that have a given bit(s) | |
1877 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1878 | * cached. The total number found is returned. | |
1879 | */ | |
d1310b2e CM |
1880 | u64 count_range_bits(struct extent_io_tree *tree, |
1881 | u64 *start, u64 search_end, u64 max_bytes, | |
9ee49a04 | 1882 | unsigned bits, int contig) |
d1310b2e CM |
1883 | { |
1884 | struct rb_node *node; | |
1885 | struct extent_state *state; | |
1886 | u64 cur_start = *start; | |
1887 | u64 total_bytes = 0; | |
ec29ed5b | 1888 | u64 last = 0; |
d1310b2e CM |
1889 | int found = 0; |
1890 | ||
fae7f21c | 1891 | if (WARN_ON(search_end <= cur_start)) |
d1310b2e | 1892 | return 0; |
d1310b2e | 1893 | |
cad321ad | 1894 | spin_lock(&tree->lock); |
d1310b2e CM |
1895 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1896 | total_bytes = tree->dirty_bytes; | |
1897 | goto out; | |
1898 | } | |
1899 | /* | |
1900 | * this search will find all the extents that end after | |
1901 | * our range starts. | |
1902 | */ | |
80ea96b1 | 1903 | node = tree_search(tree, cur_start); |
d397712b | 1904 | if (!node) |
d1310b2e | 1905 | goto out; |
d1310b2e | 1906 | |
d397712b | 1907 | while (1) { |
d1310b2e CM |
1908 | state = rb_entry(node, struct extent_state, rb_node); |
1909 | if (state->start > search_end) | |
1910 | break; | |
ec29ed5b CM |
1911 | if (contig && found && state->start > last + 1) |
1912 | break; | |
1913 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1914 | total_bytes += min(search_end, state->end) + 1 - |
1915 | max(cur_start, state->start); | |
1916 | if (total_bytes >= max_bytes) | |
1917 | break; | |
1918 | if (!found) { | |
af60bed2 | 1919 | *start = max(cur_start, state->start); |
d1310b2e CM |
1920 | found = 1; |
1921 | } | |
ec29ed5b CM |
1922 | last = state->end; |
1923 | } else if (contig && found) { | |
1924 | break; | |
d1310b2e CM |
1925 | } |
1926 | node = rb_next(node); | |
1927 | if (!node) | |
1928 | break; | |
1929 | } | |
1930 | out: | |
cad321ad | 1931 | spin_unlock(&tree->lock); |
d1310b2e CM |
1932 | return total_bytes; |
1933 | } | |
b2950863 | 1934 | |
d352ac68 CM |
1935 | /* |
1936 | * set the private field for a given byte offset in the tree. If there isn't | |
1937 | * an extent_state there already, this does nothing. | |
1938 | */ | |
171170c1 | 1939 | static int set_state_private(struct extent_io_tree *tree, u64 start, u64 private) |
d1310b2e CM |
1940 | { |
1941 | struct rb_node *node; | |
1942 | struct extent_state *state; | |
1943 | int ret = 0; | |
1944 | ||
cad321ad | 1945 | spin_lock(&tree->lock); |
d1310b2e CM |
1946 | /* |
1947 | * this search will find all the extents that end after | |
1948 | * our range starts. | |
1949 | */ | |
80ea96b1 | 1950 | node = tree_search(tree, start); |
2b114d1d | 1951 | if (!node) { |
d1310b2e CM |
1952 | ret = -ENOENT; |
1953 | goto out; | |
1954 | } | |
1955 | state = rb_entry(node, struct extent_state, rb_node); | |
1956 | if (state->start != start) { | |
1957 | ret = -ENOENT; | |
1958 | goto out; | |
1959 | } | |
1960 | state->private = private; | |
1961 | out: | |
cad321ad | 1962 | spin_unlock(&tree->lock); |
d1310b2e CM |
1963 | return ret; |
1964 | } | |
1965 | ||
1966 | int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private) | |
1967 | { | |
1968 | struct rb_node *node; | |
1969 | struct extent_state *state; | |
1970 | int ret = 0; | |
1971 | ||
cad321ad | 1972 | spin_lock(&tree->lock); |
d1310b2e CM |
1973 | /* |
1974 | * this search will find all the extents that end after | |
1975 | * our range starts. | |
1976 | */ | |
80ea96b1 | 1977 | node = tree_search(tree, start); |
2b114d1d | 1978 | if (!node) { |
d1310b2e CM |
1979 | ret = -ENOENT; |
1980 | goto out; | |
1981 | } | |
1982 | state = rb_entry(node, struct extent_state, rb_node); | |
1983 | if (state->start != start) { | |
1984 | ret = -ENOENT; | |
1985 | goto out; | |
1986 | } | |
1987 | *private = state->private; | |
1988 | out: | |
cad321ad | 1989 | spin_unlock(&tree->lock); |
d1310b2e CM |
1990 | return ret; |
1991 | } | |
1992 | ||
1993 | /* | |
1994 | * searches a range in the state tree for a given mask. | |
70dec807 | 1995 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1996 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1997 | * range is found set. | |
1998 | */ | |
1999 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 2000 | unsigned bits, int filled, struct extent_state *cached) |
d1310b2e CM |
2001 | { |
2002 | struct extent_state *state = NULL; | |
2003 | struct rb_node *node; | |
2004 | int bitset = 0; | |
d1310b2e | 2005 | |
cad321ad | 2006 | spin_lock(&tree->lock); |
27a3507d | 2007 | if (cached && extent_state_in_tree(cached) && cached->start <= start && |
df98b6e2 | 2008 | cached->end > start) |
9655d298 CM |
2009 | node = &cached->rb_node; |
2010 | else | |
2011 | node = tree_search(tree, start); | |
d1310b2e CM |
2012 | while (node && start <= end) { |
2013 | state = rb_entry(node, struct extent_state, rb_node); | |
2014 | ||
2015 | if (filled && state->start > start) { | |
2016 | bitset = 0; | |
2017 | break; | |
2018 | } | |
2019 | ||
2020 | if (state->start > end) | |
2021 | break; | |
2022 | ||
2023 | if (state->state & bits) { | |
2024 | bitset = 1; | |
2025 | if (!filled) | |
2026 | break; | |
2027 | } else if (filled) { | |
2028 | bitset = 0; | |
2029 | break; | |
2030 | } | |
46562cec CM |
2031 | |
2032 | if (state->end == (u64)-1) | |
2033 | break; | |
2034 | ||
d1310b2e CM |
2035 | start = state->end + 1; |
2036 | if (start > end) | |
2037 | break; | |
2038 | node = rb_next(node); | |
2039 | if (!node) { | |
2040 | if (filled) | |
2041 | bitset = 0; | |
2042 | break; | |
2043 | } | |
2044 | } | |
cad321ad | 2045 | spin_unlock(&tree->lock); |
d1310b2e CM |
2046 | return bitset; |
2047 | } | |
d1310b2e CM |
2048 | |
2049 | /* | |
2050 | * helper function to set a given page up to date if all the | |
2051 | * extents in the tree for that page are up to date | |
2052 | */ | |
143bede5 | 2053 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e | 2054 | { |
4eee4fa4 | 2055 | u64 start = page_offset(page); |
d1310b2e | 2056 | u64 end = start + PAGE_CACHE_SIZE - 1; |
9655d298 | 2057 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 2058 | SetPageUptodate(page); |
d1310b2e CM |
2059 | } |
2060 | ||
8b110e39 | 2061 | int free_io_failure(struct inode *inode, struct io_failure_record *rec) |
4a54c8c1 JS |
2062 | { |
2063 | int ret; | |
2064 | int err = 0; | |
2065 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
2066 | ||
2067 | set_state_private(failure_tree, rec->start, 0); | |
2068 | ret = clear_extent_bits(failure_tree, rec->start, | |
2069 | rec->start + rec->len - 1, | |
2070 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
2071 | if (ret) | |
2072 | err = ret; | |
2073 | ||
53b381b3 DW |
2074 | ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start, |
2075 | rec->start + rec->len - 1, | |
2076 | EXTENT_DAMAGED, GFP_NOFS); | |
2077 | if (ret && !err) | |
2078 | err = ret; | |
4a54c8c1 JS |
2079 | |
2080 | kfree(rec); | |
2081 | return err; | |
2082 | } | |
2083 | ||
4a54c8c1 JS |
2084 | /* |
2085 | * this bypasses the standard btrfs submit functions deliberately, as | |
2086 | * the standard behavior is to write all copies in a raid setup. here we only | |
2087 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
2088 | * submit_bio directly. | |
3ec706c8 | 2089 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
2090 | * actually prevents the read that triggered the error from finishing. |
2091 | * currently, there can be no more than two copies of every data bit. thus, | |
2092 | * exactly one rewrite is required. | |
2093 | */ | |
1203b681 MX |
2094 | int repair_io_failure(struct inode *inode, u64 start, u64 length, u64 logical, |
2095 | struct page *page, unsigned int pg_offset, int mirror_num) | |
4a54c8c1 | 2096 | { |
1203b681 | 2097 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; |
4a54c8c1 JS |
2098 | struct bio *bio; |
2099 | struct btrfs_device *dev; | |
4a54c8c1 JS |
2100 | u64 map_length = 0; |
2101 | u64 sector; | |
2102 | struct btrfs_bio *bbio = NULL; | |
53b381b3 | 2103 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
4a54c8c1 JS |
2104 | int ret; |
2105 | ||
908960c6 | 2106 | ASSERT(!(fs_info->sb->s_flags & MS_RDONLY)); |
4a54c8c1 JS |
2107 | BUG_ON(!mirror_num); |
2108 | ||
53b381b3 DW |
2109 | /* we can't repair anything in raid56 yet */ |
2110 | if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num)) | |
2111 | return 0; | |
2112 | ||
9be3395b | 2113 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
4a54c8c1 JS |
2114 | if (!bio) |
2115 | return -EIO; | |
4f024f37 | 2116 | bio->bi_iter.bi_size = 0; |
4a54c8c1 JS |
2117 | map_length = length; |
2118 | ||
3ec706c8 | 2119 | ret = btrfs_map_block(fs_info, WRITE, logical, |
4a54c8c1 JS |
2120 | &map_length, &bbio, mirror_num); |
2121 | if (ret) { | |
2122 | bio_put(bio); | |
2123 | return -EIO; | |
2124 | } | |
2125 | BUG_ON(mirror_num != bbio->mirror_num); | |
2126 | sector = bbio->stripes[mirror_num-1].physical >> 9; | |
4f024f37 | 2127 | bio->bi_iter.bi_sector = sector; |
4a54c8c1 | 2128 | dev = bbio->stripes[mirror_num-1].dev; |
6e9606d2 | 2129 | btrfs_put_bbio(bbio); |
4a54c8c1 JS |
2130 | if (!dev || !dev->bdev || !dev->writeable) { |
2131 | bio_put(bio); | |
2132 | return -EIO; | |
2133 | } | |
2134 | bio->bi_bdev = dev->bdev; | |
ffdd2018 | 2135 | bio_add_page(bio, page, length, pg_offset); |
4a54c8c1 | 2136 | |
33879d45 | 2137 | if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) { |
4a54c8c1 JS |
2138 | /* try to remap that extent elsewhere? */ |
2139 | bio_put(bio); | |
442a4f63 | 2140 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
2141 | return -EIO; |
2142 | } | |
2143 | ||
b14af3b4 DS |
2144 | btrfs_info_rl_in_rcu(fs_info, |
2145 | "read error corrected: ino %llu off %llu (dev %s sector %llu)", | |
1203b681 MX |
2146 | btrfs_ino(inode), start, |
2147 | rcu_str_deref(dev->name), sector); | |
4a54c8c1 JS |
2148 | bio_put(bio); |
2149 | return 0; | |
2150 | } | |
2151 | ||
ea466794 JB |
2152 | int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb, |
2153 | int mirror_num) | |
2154 | { | |
ea466794 JB |
2155 | u64 start = eb->start; |
2156 | unsigned long i, num_pages = num_extent_pages(eb->start, eb->len); | |
d95603b2 | 2157 | int ret = 0; |
ea466794 | 2158 | |
908960c6 ID |
2159 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
2160 | return -EROFS; | |
2161 | ||
ea466794 | 2162 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 2163 | struct page *p = eb->pages[i]; |
1203b681 MX |
2164 | |
2165 | ret = repair_io_failure(root->fs_info->btree_inode, start, | |
2166 | PAGE_CACHE_SIZE, start, p, | |
2167 | start - page_offset(p), mirror_num); | |
ea466794 JB |
2168 | if (ret) |
2169 | break; | |
2170 | start += PAGE_CACHE_SIZE; | |
2171 | } | |
2172 | ||
2173 | return ret; | |
2174 | } | |
2175 | ||
4a54c8c1 JS |
2176 | /* |
2177 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2178 | * to see if we need to process or clean up an io_failure_record | |
2179 | */ | |
8b110e39 MX |
2180 | int clean_io_failure(struct inode *inode, u64 start, struct page *page, |
2181 | unsigned int pg_offset) | |
4a54c8c1 JS |
2182 | { |
2183 | u64 private; | |
2184 | u64 private_failure; | |
2185 | struct io_failure_record *failrec; | |
908960c6 | 2186 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; |
4a54c8c1 JS |
2187 | struct extent_state *state; |
2188 | int num_copies; | |
4a54c8c1 | 2189 | int ret; |
4a54c8c1 JS |
2190 | |
2191 | private = 0; | |
2192 | ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
2193 | (u64)-1, 1, EXTENT_DIRTY, 0); | |
2194 | if (!ret) | |
2195 | return 0; | |
2196 | ||
2197 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start, | |
2198 | &private_failure); | |
2199 | if (ret) | |
2200 | return 0; | |
2201 | ||
2202 | failrec = (struct io_failure_record *)(unsigned long) private_failure; | |
2203 | BUG_ON(!failrec->this_mirror); | |
2204 | ||
2205 | if (failrec->in_validation) { | |
2206 | /* there was no real error, just free the record */ | |
2207 | pr_debug("clean_io_failure: freeing dummy error at %llu\n", | |
2208 | failrec->start); | |
4a54c8c1 JS |
2209 | goto out; |
2210 | } | |
908960c6 ID |
2211 | if (fs_info->sb->s_flags & MS_RDONLY) |
2212 | goto out; | |
4a54c8c1 JS |
2213 | |
2214 | spin_lock(&BTRFS_I(inode)->io_tree.lock); | |
2215 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, | |
2216 | failrec->start, | |
2217 | EXTENT_LOCKED); | |
2218 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); | |
2219 | ||
883d0de4 MX |
2220 | if (state && state->start <= failrec->start && |
2221 | state->end >= failrec->start + failrec->len - 1) { | |
3ec706c8 SB |
2222 | num_copies = btrfs_num_copies(fs_info, failrec->logical, |
2223 | failrec->len); | |
4a54c8c1 | 2224 | if (num_copies > 1) { |
1203b681 | 2225 | repair_io_failure(inode, start, failrec->len, |
454ff3de | 2226 | failrec->logical, page, |
1203b681 | 2227 | pg_offset, failrec->failed_mirror); |
4a54c8c1 JS |
2228 | } |
2229 | } | |
2230 | ||
2231 | out: | |
454ff3de | 2232 | free_io_failure(inode, failrec); |
4a54c8c1 | 2233 | |
454ff3de | 2234 | return 0; |
4a54c8c1 JS |
2235 | } |
2236 | ||
f612496b MX |
2237 | /* |
2238 | * Can be called when | |
2239 | * - hold extent lock | |
2240 | * - under ordered extent | |
2241 | * - the inode is freeing | |
2242 | */ | |
2243 | void btrfs_free_io_failure_record(struct inode *inode, u64 start, u64 end) | |
2244 | { | |
2245 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
2246 | struct io_failure_record *failrec; | |
2247 | struct extent_state *state, *next; | |
2248 | ||
2249 | if (RB_EMPTY_ROOT(&failure_tree->state)) | |
2250 | return; | |
2251 | ||
2252 | spin_lock(&failure_tree->lock); | |
2253 | state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY); | |
2254 | while (state) { | |
2255 | if (state->start > end) | |
2256 | break; | |
2257 | ||
2258 | ASSERT(state->end <= end); | |
2259 | ||
2260 | next = next_state(state); | |
2261 | ||
6e1103a6 | 2262 | failrec = (struct io_failure_record *)(unsigned long)state->private; |
f612496b MX |
2263 | free_extent_state(state); |
2264 | kfree(failrec); | |
2265 | ||
2266 | state = next; | |
2267 | } | |
2268 | spin_unlock(&failure_tree->lock); | |
2269 | } | |
2270 | ||
2fe6303e MX |
2271 | int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end, |
2272 | struct io_failure_record **failrec_ret) | |
4a54c8c1 | 2273 | { |
2fe6303e | 2274 | struct io_failure_record *failrec; |
4a54c8c1 JS |
2275 | u64 private; |
2276 | struct extent_map *em; | |
4a54c8c1 JS |
2277 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2278 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2279 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
4a54c8c1 | 2280 | int ret; |
4a54c8c1 JS |
2281 | u64 logical; |
2282 | ||
4a54c8c1 JS |
2283 | ret = get_state_private(failure_tree, start, &private); |
2284 | if (ret) { | |
2285 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2286 | if (!failrec) | |
2287 | return -ENOMEM; | |
2fe6303e | 2288 | |
4a54c8c1 JS |
2289 | failrec->start = start; |
2290 | failrec->len = end - start + 1; | |
2291 | failrec->this_mirror = 0; | |
2292 | failrec->bio_flags = 0; | |
2293 | failrec->in_validation = 0; | |
2294 | ||
2295 | read_lock(&em_tree->lock); | |
2296 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2297 | if (!em) { | |
2298 | read_unlock(&em_tree->lock); | |
2299 | kfree(failrec); | |
2300 | return -EIO; | |
2301 | } | |
2302 | ||
68ba990f | 2303 | if (em->start > start || em->start + em->len <= start) { |
4a54c8c1 JS |
2304 | free_extent_map(em); |
2305 | em = NULL; | |
2306 | } | |
2307 | read_unlock(&em_tree->lock); | |
7a2d6a64 | 2308 | if (!em) { |
4a54c8c1 JS |
2309 | kfree(failrec); |
2310 | return -EIO; | |
2311 | } | |
2fe6303e | 2312 | |
4a54c8c1 JS |
2313 | logical = start - em->start; |
2314 | logical = em->block_start + logical; | |
2315 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2316 | logical = em->block_start; | |
2317 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2318 | extent_set_compress_type(&failrec->bio_flags, | |
2319 | em->compress_type); | |
2320 | } | |
2fe6303e MX |
2321 | |
2322 | pr_debug("Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu\n", | |
2323 | logical, start, failrec->len); | |
2324 | ||
4a54c8c1 JS |
2325 | failrec->logical = logical; |
2326 | free_extent_map(em); | |
2327 | ||
2328 | /* set the bits in the private failure tree */ | |
2329 | ret = set_extent_bits(failure_tree, start, end, | |
2330 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
2331 | if (ret >= 0) | |
2332 | ret = set_state_private(failure_tree, start, | |
2333 | (u64)(unsigned long)failrec); | |
2334 | /* set the bits in the inode's tree */ | |
2335 | if (ret >= 0) | |
2336 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED, | |
2337 | GFP_NOFS); | |
2338 | if (ret < 0) { | |
2339 | kfree(failrec); | |
2340 | return ret; | |
2341 | } | |
2342 | } else { | |
2343 | failrec = (struct io_failure_record *)(unsigned long)private; | |
2fe6303e | 2344 | pr_debug("Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d\n", |
4a54c8c1 JS |
2345 | failrec->logical, failrec->start, failrec->len, |
2346 | failrec->in_validation); | |
2347 | /* | |
2348 | * when data can be on disk more than twice, add to failrec here | |
2349 | * (e.g. with a list for failed_mirror) to make | |
2350 | * clean_io_failure() clean all those errors at once. | |
2351 | */ | |
2352 | } | |
2fe6303e MX |
2353 | |
2354 | *failrec_ret = failrec; | |
2355 | ||
2356 | return 0; | |
2357 | } | |
2358 | ||
2359 | int btrfs_check_repairable(struct inode *inode, struct bio *failed_bio, | |
2360 | struct io_failure_record *failrec, int failed_mirror) | |
2361 | { | |
2362 | int num_copies; | |
2363 | ||
5d964051 SB |
2364 | num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info, |
2365 | failrec->logical, failrec->len); | |
4a54c8c1 JS |
2366 | if (num_copies == 1) { |
2367 | /* | |
2368 | * we only have a single copy of the data, so don't bother with | |
2369 | * all the retry and error correction code that follows. no | |
2370 | * matter what the error is, it is very likely to persist. | |
2371 | */ | |
2fe6303e | 2372 | pr_debug("Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n", |
09a7f7a2 | 2373 | num_copies, failrec->this_mirror, failed_mirror); |
2fe6303e | 2374 | return 0; |
4a54c8c1 JS |
2375 | } |
2376 | ||
4a54c8c1 JS |
2377 | /* |
2378 | * there are two premises: | |
2379 | * a) deliver good data to the caller | |
2380 | * b) correct the bad sectors on disk | |
2381 | */ | |
2382 | if (failed_bio->bi_vcnt > 1) { | |
2383 | /* | |
2384 | * to fulfill b), we need to know the exact failing sectors, as | |
2385 | * we don't want to rewrite any more than the failed ones. thus, | |
2386 | * we need separate read requests for the failed bio | |
2387 | * | |
2388 | * if the following BUG_ON triggers, our validation request got | |
2389 | * merged. we need separate requests for our algorithm to work. | |
2390 | */ | |
2391 | BUG_ON(failrec->in_validation); | |
2392 | failrec->in_validation = 1; | |
2393 | failrec->this_mirror = failed_mirror; | |
4a54c8c1 JS |
2394 | } else { |
2395 | /* | |
2396 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2397 | * of the failed sector and if we succeed, we have setup | |
2398 | * everything for repair_io_failure to do the rest for us. | |
2399 | */ | |
2400 | if (failrec->in_validation) { | |
2401 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2402 | failrec->in_validation = 0; | |
2403 | failrec->this_mirror = 0; | |
2404 | } | |
2405 | failrec->failed_mirror = failed_mirror; | |
2406 | failrec->this_mirror++; | |
2407 | if (failrec->this_mirror == failed_mirror) | |
2408 | failrec->this_mirror++; | |
4a54c8c1 JS |
2409 | } |
2410 | ||
facc8a22 | 2411 | if (failrec->this_mirror > num_copies) { |
2fe6303e | 2412 | pr_debug("Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n", |
4a54c8c1 | 2413 | num_copies, failrec->this_mirror, failed_mirror); |
2fe6303e | 2414 | return 0; |
4a54c8c1 JS |
2415 | } |
2416 | ||
2fe6303e MX |
2417 | return 1; |
2418 | } | |
2419 | ||
2420 | ||
2421 | struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio, | |
2422 | struct io_failure_record *failrec, | |
2423 | struct page *page, int pg_offset, int icsum, | |
8b110e39 | 2424 | bio_end_io_t *endio_func, void *data) |
2fe6303e MX |
2425 | { |
2426 | struct bio *bio; | |
2427 | struct btrfs_io_bio *btrfs_failed_bio; | |
2428 | struct btrfs_io_bio *btrfs_bio; | |
2429 | ||
9be3395b | 2430 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
2fe6303e MX |
2431 | if (!bio) |
2432 | return NULL; | |
2433 | ||
2434 | bio->bi_end_io = endio_func; | |
4f024f37 | 2435 | bio->bi_iter.bi_sector = failrec->logical >> 9; |
4a54c8c1 | 2436 | bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; |
4f024f37 | 2437 | bio->bi_iter.bi_size = 0; |
8b110e39 | 2438 | bio->bi_private = data; |
4a54c8c1 | 2439 | |
facc8a22 MX |
2440 | btrfs_failed_bio = btrfs_io_bio(failed_bio); |
2441 | if (btrfs_failed_bio->csum) { | |
2442 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2443 | u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); | |
2444 | ||
2445 | btrfs_bio = btrfs_io_bio(bio); | |
2446 | btrfs_bio->csum = btrfs_bio->csum_inline; | |
2fe6303e MX |
2447 | icsum *= csum_size; |
2448 | memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum, | |
facc8a22 MX |
2449 | csum_size); |
2450 | } | |
2451 | ||
2fe6303e MX |
2452 | bio_add_page(bio, page, failrec->len, pg_offset); |
2453 | ||
2454 | return bio; | |
2455 | } | |
2456 | ||
2457 | /* | |
2458 | * this is a generic handler for readpage errors (default | |
2459 | * readpage_io_failed_hook). if other copies exist, read those and write back | |
2460 | * good data to the failed position. does not investigate in remapping the | |
2461 | * failed extent elsewhere, hoping the device will be smart enough to do this as | |
2462 | * needed | |
2463 | */ | |
2464 | ||
2465 | static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset, | |
2466 | struct page *page, u64 start, u64 end, | |
2467 | int failed_mirror) | |
2468 | { | |
2469 | struct io_failure_record *failrec; | |
2470 | struct inode *inode = page->mapping->host; | |
2471 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2472 | struct bio *bio; | |
2473 | int read_mode; | |
2474 | int ret; | |
2475 | ||
2476 | BUG_ON(failed_bio->bi_rw & REQ_WRITE); | |
2477 | ||
2478 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
2479 | if (ret) | |
2480 | return ret; | |
2481 | ||
2482 | ret = btrfs_check_repairable(inode, failed_bio, failrec, failed_mirror); | |
2483 | if (!ret) { | |
2484 | free_io_failure(inode, failrec); | |
2485 | return -EIO; | |
2486 | } | |
2487 | ||
2488 | if (failed_bio->bi_vcnt > 1) | |
2489 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; | |
2490 | else | |
2491 | read_mode = READ_SYNC; | |
2492 | ||
2493 | phy_offset >>= inode->i_sb->s_blocksize_bits; | |
2494 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2495 | start - page_offset(page), | |
8b110e39 MX |
2496 | (int)phy_offset, failed_bio->bi_end_io, |
2497 | NULL); | |
2fe6303e MX |
2498 | if (!bio) { |
2499 | free_io_failure(inode, failrec); | |
2500 | return -EIO; | |
2501 | } | |
4a54c8c1 | 2502 | |
2fe6303e MX |
2503 | pr_debug("Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d\n", |
2504 | read_mode, failrec->this_mirror, failrec->in_validation); | |
4a54c8c1 | 2505 | |
013bd4c3 TI |
2506 | ret = tree->ops->submit_bio_hook(inode, read_mode, bio, |
2507 | failrec->this_mirror, | |
2508 | failrec->bio_flags, 0); | |
6c387ab2 | 2509 | if (ret) { |
454ff3de | 2510 | free_io_failure(inode, failrec); |
6c387ab2 MX |
2511 | bio_put(bio); |
2512 | } | |
2513 | ||
013bd4c3 | 2514 | return ret; |
4a54c8c1 JS |
2515 | } |
2516 | ||
d1310b2e CM |
2517 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2518 | ||
87826df0 JM |
2519 | int end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
2520 | { | |
2521 | int uptodate = (err == 0); | |
2522 | struct extent_io_tree *tree; | |
3e2426bd | 2523 | int ret = 0; |
87826df0 JM |
2524 | |
2525 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
2526 | ||
2527 | if (tree->ops && tree->ops->writepage_end_io_hook) { | |
2528 | ret = tree->ops->writepage_end_io_hook(page, start, | |
2529 | end, NULL, uptodate); | |
2530 | if (ret) | |
2531 | uptodate = 0; | |
2532 | } | |
2533 | ||
87826df0 | 2534 | if (!uptodate) { |
87826df0 JM |
2535 | ClearPageUptodate(page); |
2536 | SetPageError(page); | |
5dca6eea LB |
2537 | ret = ret < 0 ? ret : -EIO; |
2538 | mapping_set_error(page->mapping, ret); | |
87826df0 JM |
2539 | } |
2540 | return 0; | |
2541 | } | |
2542 | ||
d1310b2e CM |
2543 | /* |
2544 | * after a writepage IO is done, we need to: | |
2545 | * clear the uptodate bits on error | |
2546 | * clear the writeback bits in the extent tree for this IO | |
2547 | * end_page_writeback if the page has no more pending IO | |
2548 | * | |
2549 | * Scheduling is not allowed, so the extent state tree is expected | |
2550 | * to have one and only one object corresponding to this IO. | |
2551 | */ | |
4246a0b6 | 2552 | static void end_bio_extent_writepage(struct bio *bio) |
d1310b2e | 2553 | { |
2c30c71b | 2554 | struct bio_vec *bvec; |
d1310b2e CM |
2555 | u64 start; |
2556 | u64 end; | |
2c30c71b | 2557 | int i; |
d1310b2e | 2558 | |
2c30c71b | 2559 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2560 | struct page *page = bvec->bv_page; |
902b22f3 | 2561 | |
17a5adcc AO |
2562 | /* We always issue full-page reads, but if some block |
2563 | * in a page fails to read, blk_update_request() will | |
2564 | * advance bv_offset and adjust bv_len to compensate. | |
2565 | * Print a warning for nonzero offsets, and an error | |
2566 | * if they don't add up to a full page. */ | |
efe120a0 FH |
2567 | if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) { |
2568 | if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE) | |
2569 | btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info, | |
2570 | "partial page write in btrfs with offset %u and length %u", | |
2571 | bvec->bv_offset, bvec->bv_len); | |
2572 | else | |
2573 | btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info, | |
2574 | "incomplete page write in btrfs with offset %u and " | |
2575 | "length %u", | |
2576 | bvec->bv_offset, bvec->bv_len); | |
2577 | } | |
d1310b2e | 2578 | |
17a5adcc AO |
2579 | start = page_offset(page); |
2580 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
d1310b2e | 2581 | |
4246a0b6 | 2582 | if (end_extent_writepage(page, bio->bi_error, start, end)) |
87826df0 | 2583 | continue; |
70dec807 | 2584 | |
17a5adcc | 2585 | end_page_writeback(page); |
2c30c71b | 2586 | } |
2b1f55b0 | 2587 | |
d1310b2e | 2588 | bio_put(bio); |
d1310b2e CM |
2589 | } |
2590 | ||
883d0de4 MX |
2591 | static void |
2592 | endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len, | |
2593 | int uptodate) | |
2594 | { | |
2595 | struct extent_state *cached = NULL; | |
2596 | u64 end = start + len - 1; | |
2597 | ||
2598 | if (uptodate && tree->track_uptodate) | |
2599 | set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC); | |
2600 | unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC); | |
2601 | } | |
2602 | ||
d1310b2e CM |
2603 | /* |
2604 | * after a readpage IO is done, we need to: | |
2605 | * clear the uptodate bits on error | |
2606 | * set the uptodate bits if things worked | |
2607 | * set the page up to date if all extents in the tree are uptodate | |
2608 | * clear the lock bit in the extent tree | |
2609 | * unlock the page if there are no other extents locked for it | |
2610 | * | |
2611 | * Scheduling is not allowed, so the extent state tree is expected | |
2612 | * to have one and only one object corresponding to this IO. | |
2613 | */ | |
4246a0b6 | 2614 | static void end_bio_extent_readpage(struct bio *bio) |
d1310b2e | 2615 | { |
2c30c71b | 2616 | struct bio_vec *bvec; |
4246a0b6 | 2617 | int uptodate = !bio->bi_error; |
facc8a22 | 2618 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); |
902b22f3 | 2619 | struct extent_io_tree *tree; |
facc8a22 | 2620 | u64 offset = 0; |
d1310b2e CM |
2621 | u64 start; |
2622 | u64 end; | |
facc8a22 | 2623 | u64 len; |
883d0de4 MX |
2624 | u64 extent_start = 0; |
2625 | u64 extent_len = 0; | |
5cf1ab56 | 2626 | int mirror; |
d1310b2e | 2627 | int ret; |
2c30c71b | 2628 | int i; |
d1310b2e | 2629 | |
2c30c71b | 2630 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2631 | struct page *page = bvec->bv_page; |
a71754fc | 2632 | struct inode *inode = page->mapping->host; |
507903b8 | 2633 | |
be3940c0 | 2634 | pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, " |
4246a0b6 CH |
2635 | "mirror=%u\n", (u64)bio->bi_iter.bi_sector, |
2636 | bio->bi_error, io_bio->mirror_num); | |
a71754fc | 2637 | tree = &BTRFS_I(inode)->io_tree; |
902b22f3 | 2638 | |
17a5adcc AO |
2639 | /* We always issue full-page reads, but if some block |
2640 | * in a page fails to read, blk_update_request() will | |
2641 | * advance bv_offset and adjust bv_len to compensate. | |
2642 | * Print a warning for nonzero offsets, and an error | |
2643 | * if they don't add up to a full page. */ | |
efe120a0 FH |
2644 | if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) { |
2645 | if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE) | |
2646 | btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info, | |
2647 | "partial page read in btrfs with offset %u and length %u", | |
2648 | bvec->bv_offset, bvec->bv_len); | |
2649 | else | |
2650 | btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info, | |
2651 | "incomplete page read in btrfs with offset %u and " | |
2652 | "length %u", | |
2653 | bvec->bv_offset, bvec->bv_len); | |
2654 | } | |
d1310b2e | 2655 | |
17a5adcc AO |
2656 | start = page_offset(page); |
2657 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
facc8a22 | 2658 | len = bvec->bv_len; |
d1310b2e | 2659 | |
9be3395b | 2660 | mirror = io_bio->mirror_num; |
f2a09da9 MX |
2661 | if (likely(uptodate && tree->ops && |
2662 | tree->ops->readpage_end_io_hook)) { | |
facc8a22 MX |
2663 | ret = tree->ops->readpage_end_io_hook(io_bio, offset, |
2664 | page, start, end, | |
2665 | mirror); | |
5ee0844d | 2666 | if (ret) |
d1310b2e | 2667 | uptodate = 0; |
5ee0844d | 2668 | else |
1203b681 | 2669 | clean_io_failure(inode, start, page, 0); |
d1310b2e | 2670 | } |
ea466794 | 2671 | |
f2a09da9 MX |
2672 | if (likely(uptodate)) |
2673 | goto readpage_ok; | |
2674 | ||
2675 | if (tree->ops && tree->ops->readpage_io_failed_hook) { | |
5cf1ab56 | 2676 | ret = tree->ops->readpage_io_failed_hook(page, mirror); |
4246a0b6 | 2677 | if (!ret && !bio->bi_error) |
ea466794 | 2678 | uptodate = 1; |
f2a09da9 | 2679 | } else { |
f4a8e656 JS |
2680 | /* |
2681 | * The generic bio_readpage_error handles errors the | |
2682 | * following way: If possible, new read requests are | |
2683 | * created and submitted and will end up in | |
2684 | * end_bio_extent_readpage as well (if we're lucky, not | |
2685 | * in the !uptodate case). In that case it returns 0 and | |
2686 | * we just go on with the next page in our bio. If it | |
2687 | * can't handle the error it will return -EIO and we | |
2688 | * remain responsible for that page. | |
2689 | */ | |
facc8a22 MX |
2690 | ret = bio_readpage_error(bio, offset, page, start, end, |
2691 | mirror); | |
7e38326f | 2692 | if (ret == 0) { |
4246a0b6 | 2693 | uptodate = !bio->bi_error; |
38c1c2e4 | 2694 | offset += len; |
7e38326f CM |
2695 | continue; |
2696 | } | |
2697 | } | |
f2a09da9 | 2698 | readpage_ok: |
883d0de4 | 2699 | if (likely(uptodate)) { |
a71754fc JB |
2700 | loff_t i_size = i_size_read(inode); |
2701 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
a583c026 | 2702 | unsigned off; |
a71754fc JB |
2703 | |
2704 | /* Zero out the end if this page straddles i_size */ | |
a583c026 LB |
2705 | off = i_size & (PAGE_CACHE_SIZE-1); |
2706 | if (page->index == end_index && off) | |
2707 | zero_user_segment(page, off, PAGE_CACHE_SIZE); | |
17a5adcc | 2708 | SetPageUptodate(page); |
70dec807 | 2709 | } else { |
17a5adcc AO |
2710 | ClearPageUptodate(page); |
2711 | SetPageError(page); | |
70dec807 | 2712 | } |
17a5adcc | 2713 | unlock_page(page); |
facc8a22 | 2714 | offset += len; |
883d0de4 MX |
2715 | |
2716 | if (unlikely(!uptodate)) { | |
2717 | if (extent_len) { | |
2718 | endio_readpage_release_extent(tree, | |
2719 | extent_start, | |
2720 | extent_len, 1); | |
2721 | extent_start = 0; | |
2722 | extent_len = 0; | |
2723 | } | |
2724 | endio_readpage_release_extent(tree, start, | |
2725 | end - start + 1, 0); | |
2726 | } else if (!extent_len) { | |
2727 | extent_start = start; | |
2728 | extent_len = end + 1 - start; | |
2729 | } else if (extent_start + extent_len == start) { | |
2730 | extent_len += end + 1 - start; | |
2731 | } else { | |
2732 | endio_readpage_release_extent(tree, extent_start, | |
2733 | extent_len, uptodate); | |
2734 | extent_start = start; | |
2735 | extent_len = end + 1 - start; | |
2736 | } | |
2c30c71b | 2737 | } |
d1310b2e | 2738 | |
883d0de4 MX |
2739 | if (extent_len) |
2740 | endio_readpage_release_extent(tree, extent_start, extent_len, | |
2741 | uptodate); | |
facc8a22 | 2742 | if (io_bio->end_io) |
4246a0b6 | 2743 | io_bio->end_io(io_bio, bio->bi_error); |
d1310b2e | 2744 | bio_put(bio); |
d1310b2e CM |
2745 | } |
2746 | ||
9be3395b CM |
2747 | /* |
2748 | * this allocates from the btrfs_bioset. We're returning a bio right now | |
2749 | * but you can call btrfs_io_bio for the appropriate container_of magic | |
2750 | */ | |
88f794ed MX |
2751 | struct bio * |
2752 | btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
2753 | gfp_t gfp_flags) | |
d1310b2e | 2754 | { |
facc8a22 | 2755 | struct btrfs_io_bio *btrfs_bio; |
d1310b2e CM |
2756 | struct bio *bio; |
2757 | ||
9be3395b | 2758 | bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset); |
d1310b2e CM |
2759 | |
2760 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
9be3395b CM |
2761 | while (!bio && (nr_vecs /= 2)) { |
2762 | bio = bio_alloc_bioset(gfp_flags, | |
2763 | nr_vecs, btrfs_bioset); | |
2764 | } | |
d1310b2e CM |
2765 | } |
2766 | ||
2767 | if (bio) { | |
2768 | bio->bi_bdev = bdev; | |
4f024f37 | 2769 | bio->bi_iter.bi_sector = first_sector; |
facc8a22 MX |
2770 | btrfs_bio = btrfs_io_bio(bio); |
2771 | btrfs_bio->csum = NULL; | |
2772 | btrfs_bio->csum_allocated = NULL; | |
2773 | btrfs_bio->end_io = NULL; | |
d1310b2e CM |
2774 | } |
2775 | return bio; | |
2776 | } | |
2777 | ||
9be3395b CM |
2778 | struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask) |
2779 | { | |
23ea8e5a MX |
2780 | struct btrfs_io_bio *btrfs_bio; |
2781 | struct bio *new; | |
9be3395b | 2782 | |
23ea8e5a MX |
2783 | new = bio_clone_bioset(bio, gfp_mask, btrfs_bioset); |
2784 | if (new) { | |
2785 | btrfs_bio = btrfs_io_bio(new); | |
2786 | btrfs_bio->csum = NULL; | |
2787 | btrfs_bio->csum_allocated = NULL; | |
2788 | btrfs_bio->end_io = NULL; | |
3a9508b0 CM |
2789 | |
2790 | #ifdef CONFIG_BLK_CGROUP | |
da2f0f74 CM |
2791 | /* FIXME, put this into bio_clone_bioset */ |
2792 | if (bio->bi_css) | |
2793 | bio_associate_blkcg(new, bio->bi_css); | |
3a9508b0 | 2794 | #endif |
23ea8e5a MX |
2795 | } |
2796 | return new; | |
2797 | } | |
9be3395b CM |
2798 | |
2799 | /* this also allocates from the btrfs_bioset */ | |
2800 | struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs) | |
2801 | { | |
facc8a22 MX |
2802 | struct btrfs_io_bio *btrfs_bio; |
2803 | struct bio *bio; | |
2804 | ||
2805 | bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset); | |
2806 | if (bio) { | |
2807 | btrfs_bio = btrfs_io_bio(bio); | |
2808 | btrfs_bio->csum = NULL; | |
2809 | btrfs_bio->csum_allocated = NULL; | |
2810 | btrfs_bio->end_io = NULL; | |
2811 | } | |
2812 | return bio; | |
9be3395b CM |
2813 | } |
2814 | ||
2815 | ||
355808c2 JM |
2816 | static int __must_check submit_one_bio(int rw, struct bio *bio, |
2817 | int mirror_num, unsigned long bio_flags) | |
d1310b2e | 2818 | { |
d1310b2e | 2819 | int ret = 0; |
70dec807 CM |
2820 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
2821 | struct page *page = bvec->bv_page; | |
2822 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 2823 | u64 start; |
70dec807 | 2824 | |
4eee4fa4 | 2825 | start = page_offset(page) + bvec->bv_offset; |
70dec807 | 2826 | |
902b22f3 | 2827 | bio->bi_private = NULL; |
d1310b2e CM |
2828 | |
2829 | bio_get(bio); | |
2830 | ||
065631f6 | 2831 | if (tree->ops && tree->ops->submit_bio_hook) |
6b82ce8d | 2832 | ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio, |
eaf25d93 | 2833 | mirror_num, bio_flags, start); |
0b86a832 | 2834 | else |
21adbd5c | 2835 | btrfsic_submit_bio(rw, bio); |
4a54c8c1 | 2836 | |
d1310b2e CM |
2837 | bio_put(bio); |
2838 | return ret; | |
2839 | } | |
2840 | ||
64a16701 | 2841 | static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page, |
3444a972 JM |
2842 | unsigned long offset, size_t size, struct bio *bio, |
2843 | unsigned long bio_flags) | |
2844 | { | |
2845 | int ret = 0; | |
2846 | if (tree->ops && tree->ops->merge_bio_hook) | |
64a16701 | 2847 | ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio, |
3444a972 JM |
2848 | bio_flags); |
2849 | BUG_ON(ret < 0); | |
2850 | return ret; | |
2851 | ||
2852 | } | |
2853 | ||
d1310b2e | 2854 | static int submit_extent_page(int rw, struct extent_io_tree *tree, |
da2f0f74 | 2855 | struct writeback_control *wbc, |
d1310b2e CM |
2856 | struct page *page, sector_t sector, |
2857 | size_t size, unsigned long offset, | |
2858 | struct block_device *bdev, | |
2859 | struct bio **bio_ret, | |
2860 | unsigned long max_pages, | |
f188591e | 2861 | bio_end_io_t end_io_func, |
c8b97818 CM |
2862 | int mirror_num, |
2863 | unsigned long prev_bio_flags, | |
005efedf FM |
2864 | unsigned long bio_flags, |
2865 | bool force_bio_submit) | |
d1310b2e CM |
2866 | { |
2867 | int ret = 0; | |
2868 | struct bio *bio; | |
c8b97818 | 2869 | int contig = 0; |
c8b97818 | 2870 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; |
5b050f04 | 2871 | size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE); |
d1310b2e CM |
2872 | |
2873 | if (bio_ret && *bio_ret) { | |
2874 | bio = *bio_ret; | |
c8b97818 | 2875 | if (old_compressed) |
4f024f37 | 2876 | contig = bio->bi_iter.bi_sector == sector; |
c8b97818 | 2877 | else |
f73a1c7d | 2878 | contig = bio_end_sector(bio) == sector; |
c8b97818 CM |
2879 | |
2880 | if (prev_bio_flags != bio_flags || !contig || | |
005efedf | 2881 | force_bio_submit || |
64a16701 | 2882 | merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) || |
c8b97818 CM |
2883 | bio_add_page(bio, page, page_size, offset) < page_size) { |
2884 | ret = submit_one_bio(rw, bio, mirror_num, | |
2885 | prev_bio_flags); | |
289454ad NA |
2886 | if (ret < 0) { |
2887 | *bio_ret = NULL; | |
79787eaa | 2888 | return ret; |
289454ad | 2889 | } |
d1310b2e CM |
2890 | bio = NULL; |
2891 | } else { | |
da2f0f74 CM |
2892 | if (wbc) |
2893 | wbc_account_io(wbc, page, page_size); | |
d1310b2e CM |
2894 | return 0; |
2895 | } | |
2896 | } | |
c8b97818 | 2897 | |
b54ffb73 KO |
2898 | bio = btrfs_bio_alloc(bdev, sector, BIO_MAX_PAGES, |
2899 | GFP_NOFS | __GFP_HIGH); | |
5df67083 TI |
2900 | if (!bio) |
2901 | return -ENOMEM; | |
70dec807 | 2902 | |
c8b97818 | 2903 | bio_add_page(bio, page, page_size, offset); |
d1310b2e CM |
2904 | bio->bi_end_io = end_io_func; |
2905 | bio->bi_private = tree; | |
da2f0f74 CM |
2906 | if (wbc) { |
2907 | wbc_init_bio(wbc, bio); | |
2908 | wbc_account_io(wbc, page, page_size); | |
2909 | } | |
70dec807 | 2910 | |
d397712b | 2911 | if (bio_ret) |
d1310b2e | 2912 | *bio_ret = bio; |
d397712b | 2913 | else |
c8b97818 | 2914 | ret = submit_one_bio(rw, bio, mirror_num, bio_flags); |
d1310b2e CM |
2915 | |
2916 | return ret; | |
2917 | } | |
2918 | ||
48a3b636 ES |
2919 | static void attach_extent_buffer_page(struct extent_buffer *eb, |
2920 | struct page *page) | |
d1310b2e CM |
2921 | { |
2922 | if (!PagePrivate(page)) { | |
2923 | SetPagePrivate(page); | |
d1310b2e | 2924 | page_cache_get(page); |
4f2de97a JB |
2925 | set_page_private(page, (unsigned long)eb); |
2926 | } else { | |
2927 | WARN_ON(page->private != (unsigned long)eb); | |
d1310b2e CM |
2928 | } |
2929 | } | |
2930 | ||
4f2de97a | 2931 | void set_page_extent_mapped(struct page *page) |
d1310b2e | 2932 | { |
4f2de97a JB |
2933 | if (!PagePrivate(page)) { |
2934 | SetPagePrivate(page); | |
2935 | page_cache_get(page); | |
2936 | set_page_private(page, EXTENT_PAGE_PRIVATE); | |
2937 | } | |
d1310b2e CM |
2938 | } |
2939 | ||
125bac01 MX |
2940 | static struct extent_map * |
2941 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
2942 | u64 start, u64 len, get_extent_t *get_extent, | |
2943 | struct extent_map **em_cached) | |
2944 | { | |
2945 | struct extent_map *em; | |
2946 | ||
2947 | if (em_cached && *em_cached) { | |
2948 | em = *em_cached; | |
cbc0e928 | 2949 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 MX |
2950 | start < extent_map_end(em)) { |
2951 | atomic_inc(&em->refs); | |
2952 | return em; | |
2953 | } | |
2954 | ||
2955 | free_extent_map(em); | |
2956 | *em_cached = NULL; | |
2957 | } | |
2958 | ||
2959 | em = get_extent(inode, page, pg_offset, start, len, 0); | |
2960 | if (em_cached && !IS_ERR_OR_NULL(em)) { | |
2961 | BUG_ON(*em_cached); | |
2962 | atomic_inc(&em->refs); | |
2963 | *em_cached = em; | |
2964 | } | |
2965 | return em; | |
2966 | } | |
d1310b2e CM |
2967 | /* |
2968 | * basic readpage implementation. Locked extent state structs are inserted | |
2969 | * into the tree that are removed when the IO is done (by the end_io | |
2970 | * handlers) | |
79787eaa | 2971 | * XXX JDM: This needs looking at to ensure proper page locking |
d1310b2e | 2972 | */ |
9974090b MX |
2973 | static int __do_readpage(struct extent_io_tree *tree, |
2974 | struct page *page, | |
2975 | get_extent_t *get_extent, | |
125bac01 | 2976 | struct extent_map **em_cached, |
9974090b | 2977 | struct bio **bio, int mirror_num, |
005efedf FM |
2978 | unsigned long *bio_flags, int rw, |
2979 | u64 *prev_em_start) | |
d1310b2e CM |
2980 | { |
2981 | struct inode *inode = page->mapping->host; | |
4eee4fa4 | 2982 | u64 start = page_offset(page); |
d1310b2e CM |
2983 | u64 page_end = start + PAGE_CACHE_SIZE - 1; |
2984 | u64 end; | |
2985 | u64 cur = start; | |
2986 | u64 extent_offset; | |
2987 | u64 last_byte = i_size_read(inode); | |
2988 | u64 block_start; | |
2989 | u64 cur_end; | |
2990 | sector_t sector; | |
2991 | struct extent_map *em; | |
2992 | struct block_device *bdev; | |
2993 | int ret; | |
2994 | int nr = 0; | |
4b384318 | 2995 | int parent_locked = *bio_flags & EXTENT_BIO_PARENT_LOCKED; |
306e16ce | 2996 | size_t pg_offset = 0; |
d1310b2e | 2997 | size_t iosize; |
c8b97818 | 2998 | size_t disk_io_size; |
d1310b2e | 2999 | size_t blocksize = inode->i_sb->s_blocksize; |
4b384318 | 3000 | unsigned long this_bio_flag = *bio_flags & EXTENT_BIO_PARENT_LOCKED; |
d1310b2e CM |
3001 | |
3002 | set_page_extent_mapped(page); | |
3003 | ||
9974090b | 3004 | end = page_end; |
90a887c9 DM |
3005 | if (!PageUptodate(page)) { |
3006 | if (cleancache_get_page(page) == 0) { | |
3007 | BUG_ON(blocksize != PAGE_SIZE); | |
9974090b | 3008 | unlock_extent(tree, start, end); |
90a887c9 DM |
3009 | goto out; |
3010 | } | |
3011 | } | |
3012 | ||
c8b97818 CM |
3013 | if (page->index == last_byte >> PAGE_CACHE_SHIFT) { |
3014 | char *userpage; | |
3015 | size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1); | |
3016 | ||
3017 | if (zero_offset) { | |
3018 | iosize = PAGE_CACHE_SIZE - zero_offset; | |
7ac687d9 | 3019 | userpage = kmap_atomic(page); |
c8b97818 CM |
3020 | memset(userpage + zero_offset, 0, iosize); |
3021 | flush_dcache_page(page); | |
7ac687d9 | 3022 | kunmap_atomic(userpage); |
c8b97818 CM |
3023 | } |
3024 | } | |
d1310b2e | 3025 | while (cur <= end) { |
c8f2f24b | 3026 | unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1; |
005efedf | 3027 | bool force_bio_submit = false; |
c8f2f24b | 3028 | |
d1310b2e CM |
3029 | if (cur >= last_byte) { |
3030 | char *userpage; | |
507903b8 AJ |
3031 | struct extent_state *cached = NULL; |
3032 | ||
306e16ce | 3033 | iosize = PAGE_CACHE_SIZE - pg_offset; |
7ac687d9 | 3034 | userpage = kmap_atomic(page); |
306e16ce | 3035 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3036 | flush_dcache_page(page); |
7ac687d9 | 3037 | kunmap_atomic(userpage); |
d1310b2e | 3038 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 3039 | &cached, GFP_NOFS); |
4b384318 MF |
3040 | if (!parent_locked) |
3041 | unlock_extent_cached(tree, cur, | |
3042 | cur + iosize - 1, | |
3043 | &cached, GFP_NOFS); | |
d1310b2e CM |
3044 | break; |
3045 | } | |
125bac01 MX |
3046 | em = __get_extent_map(inode, page, pg_offset, cur, |
3047 | end - cur + 1, get_extent, em_cached); | |
c704005d | 3048 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3049 | SetPageError(page); |
4b384318 MF |
3050 | if (!parent_locked) |
3051 | unlock_extent(tree, cur, end); | |
d1310b2e CM |
3052 | break; |
3053 | } | |
d1310b2e CM |
3054 | extent_offset = cur - em->start; |
3055 | BUG_ON(extent_map_end(em) <= cur); | |
3056 | BUG_ON(end < cur); | |
3057 | ||
261507a0 | 3058 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
4b384318 | 3059 | this_bio_flag |= EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
3060 | extent_set_compress_type(&this_bio_flag, |
3061 | em->compress_type); | |
3062 | } | |
c8b97818 | 3063 | |
d1310b2e CM |
3064 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
3065 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 3066 | iosize = ALIGN(iosize, blocksize); |
c8b97818 CM |
3067 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
3068 | disk_io_size = em->block_len; | |
3069 | sector = em->block_start >> 9; | |
3070 | } else { | |
3071 | sector = (em->block_start + extent_offset) >> 9; | |
3072 | disk_io_size = iosize; | |
3073 | } | |
d1310b2e CM |
3074 | bdev = em->bdev; |
3075 | block_start = em->block_start; | |
d899e052 YZ |
3076 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
3077 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
3078 | |
3079 | /* | |
3080 | * If we have a file range that points to a compressed extent | |
3081 | * and it's followed by a consecutive file range that points to | |
3082 | * to the same compressed extent (possibly with a different | |
3083 | * offset and/or length, so it either points to the whole extent | |
3084 | * or only part of it), we must make sure we do not submit a | |
3085 | * single bio to populate the pages for the 2 ranges because | |
3086 | * this makes the compressed extent read zero out the pages | |
3087 | * belonging to the 2nd range. Imagine the following scenario: | |
3088 | * | |
3089 | * File layout | |
3090 | * [0 - 8K] [8K - 24K] | |
3091 | * | | | |
3092 | * | | | |
3093 | * points to extent X, points to extent X, | |
3094 | * offset 4K, length of 8K offset 0, length 16K | |
3095 | * | |
3096 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
3097 | * | |
3098 | * If the bio to read the compressed extent covers both ranges, | |
3099 | * it will decompress extent X into the pages belonging to the | |
3100 | * first range and then it will stop, zeroing out the remaining | |
3101 | * pages that belong to the other range that points to extent X. | |
3102 | * So here we make sure we submit 2 bios, one for the first | |
3103 | * range and another one for the third range. Both will target | |
3104 | * the same physical extent from disk, but we can't currently | |
3105 | * make the compressed bio endio callback populate the pages | |
3106 | * for both ranges because each compressed bio is tightly | |
3107 | * coupled with a single extent map, and each range can have | |
3108 | * an extent map with a different offset value relative to the | |
3109 | * uncompressed data of our extent and different lengths. This | |
3110 | * is a corner case so we prioritize correctness over | |
3111 | * non-optimal behavior (submitting 2 bios for the same extent). | |
3112 | */ | |
3113 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
3114 | prev_em_start && *prev_em_start != (u64)-1 && | |
3115 | *prev_em_start != em->orig_start) | |
3116 | force_bio_submit = true; | |
3117 | ||
3118 | if (prev_em_start) | |
3119 | *prev_em_start = em->orig_start; | |
3120 | ||
d1310b2e CM |
3121 | free_extent_map(em); |
3122 | em = NULL; | |
3123 | ||
3124 | /* we've found a hole, just zero and go on */ | |
3125 | if (block_start == EXTENT_MAP_HOLE) { | |
3126 | char *userpage; | |
507903b8 AJ |
3127 | struct extent_state *cached = NULL; |
3128 | ||
7ac687d9 | 3129 | userpage = kmap_atomic(page); |
306e16ce | 3130 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3131 | flush_dcache_page(page); |
7ac687d9 | 3132 | kunmap_atomic(userpage); |
d1310b2e CM |
3133 | |
3134 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 | 3135 | &cached, GFP_NOFS); |
5e6ecb36 FM |
3136 | if (parent_locked) |
3137 | free_extent_state(cached); | |
3138 | else | |
3139 | unlock_extent_cached(tree, cur, | |
3140 | cur + iosize - 1, | |
3141 | &cached, GFP_NOFS); | |
d1310b2e | 3142 | cur = cur + iosize; |
306e16ce | 3143 | pg_offset += iosize; |
d1310b2e CM |
3144 | continue; |
3145 | } | |
3146 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
3147 | if (test_range_bit(tree, cur, cur_end, |
3148 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 3149 | check_page_uptodate(tree, page); |
4b384318 MF |
3150 | if (!parent_locked) |
3151 | unlock_extent(tree, cur, cur + iosize - 1); | |
d1310b2e | 3152 | cur = cur + iosize; |
306e16ce | 3153 | pg_offset += iosize; |
d1310b2e CM |
3154 | continue; |
3155 | } | |
70dec807 CM |
3156 | /* we have an inline extent but it didn't get marked up |
3157 | * to date. Error out | |
3158 | */ | |
3159 | if (block_start == EXTENT_MAP_INLINE) { | |
3160 | SetPageError(page); | |
4b384318 MF |
3161 | if (!parent_locked) |
3162 | unlock_extent(tree, cur, cur + iosize - 1); | |
70dec807 | 3163 | cur = cur + iosize; |
306e16ce | 3164 | pg_offset += iosize; |
70dec807 CM |
3165 | continue; |
3166 | } | |
d1310b2e | 3167 | |
c8f2f24b | 3168 | pnr -= page->index; |
da2f0f74 | 3169 | ret = submit_extent_page(rw, tree, NULL, page, |
306e16ce | 3170 | sector, disk_io_size, pg_offset, |
89642229 | 3171 | bdev, bio, pnr, |
c8b97818 CM |
3172 | end_bio_extent_readpage, mirror_num, |
3173 | *bio_flags, | |
005efedf FM |
3174 | this_bio_flag, |
3175 | force_bio_submit); | |
c8f2f24b JB |
3176 | if (!ret) { |
3177 | nr++; | |
3178 | *bio_flags = this_bio_flag; | |
3179 | } else { | |
d1310b2e | 3180 | SetPageError(page); |
4b384318 MF |
3181 | if (!parent_locked) |
3182 | unlock_extent(tree, cur, cur + iosize - 1); | |
edd33c99 | 3183 | } |
d1310b2e | 3184 | cur = cur + iosize; |
306e16ce | 3185 | pg_offset += iosize; |
d1310b2e | 3186 | } |
90a887c9 | 3187 | out: |
d1310b2e CM |
3188 | if (!nr) { |
3189 | if (!PageError(page)) | |
3190 | SetPageUptodate(page); | |
3191 | unlock_page(page); | |
3192 | } | |
3193 | return 0; | |
3194 | } | |
3195 | ||
9974090b MX |
3196 | static inline void __do_contiguous_readpages(struct extent_io_tree *tree, |
3197 | struct page *pages[], int nr_pages, | |
3198 | u64 start, u64 end, | |
3199 | get_extent_t *get_extent, | |
125bac01 | 3200 | struct extent_map **em_cached, |
9974090b | 3201 | struct bio **bio, int mirror_num, |
808f80b4 FM |
3202 | unsigned long *bio_flags, int rw, |
3203 | u64 *prev_em_start) | |
9974090b MX |
3204 | { |
3205 | struct inode *inode; | |
3206 | struct btrfs_ordered_extent *ordered; | |
3207 | int index; | |
3208 | ||
3209 | inode = pages[0]->mapping->host; | |
3210 | while (1) { | |
3211 | lock_extent(tree, start, end); | |
3212 | ordered = btrfs_lookup_ordered_range(inode, start, | |
3213 | end - start + 1); | |
3214 | if (!ordered) | |
3215 | break; | |
3216 | unlock_extent(tree, start, end); | |
3217 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3218 | btrfs_put_ordered_extent(ordered); | |
3219 | } | |
3220 | ||
3221 | for (index = 0; index < nr_pages; index++) { | |
125bac01 | 3222 | __do_readpage(tree, pages[index], get_extent, em_cached, bio, |
808f80b4 | 3223 | mirror_num, bio_flags, rw, prev_em_start); |
9974090b MX |
3224 | page_cache_release(pages[index]); |
3225 | } | |
3226 | } | |
3227 | ||
3228 | static void __extent_readpages(struct extent_io_tree *tree, | |
3229 | struct page *pages[], | |
3230 | int nr_pages, get_extent_t *get_extent, | |
125bac01 | 3231 | struct extent_map **em_cached, |
9974090b | 3232 | struct bio **bio, int mirror_num, |
808f80b4 FM |
3233 | unsigned long *bio_flags, int rw, |
3234 | u64 *prev_em_start) | |
9974090b | 3235 | { |
35a3621b | 3236 | u64 start = 0; |
9974090b MX |
3237 | u64 end = 0; |
3238 | u64 page_start; | |
3239 | int index; | |
35a3621b | 3240 | int first_index = 0; |
9974090b MX |
3241 | |
3242 | for (index = 0; index < nr_pages; index++) { | |
3243 | page_start = page_offset(pages[index]); | |
3244 | if (!end) { | |
3245 | start = page_start; | |
3246 | end = start + PAGE_CACHE_SIZE - 1; | |
3247 | first_index = index; | |
3248 | } else if (end + 1 == page_start) { | |
3249 | end += PAGE_CACHE_SIZE; | |
3250 | } else { | |
3251 | __do_contiguous_readpages(tree, &pages[first_index], | |
3252 | index - first_index, start, | |
125bac01 MX |
3253 | end, get_extent, em_cached, |
3254 | bio, mirror_num, bio_flags, | |
808f80b4 | 3255 | rw, prev_em_start); |
9974090b MX |
3256 | start = page_start; |
3257 | end = start + PAGE_CACHE_SIZE - 1; | |
3258 | first_index = index; | |
3259 | } | |
3260 | } | |
3261 | ||
3262 | if (end) | |
3263 | __do_contiguous_readpages(tree, &pages[first_index], | |
3264 | index - first_index, start, | |
125bac01 | 3265 | end, get_extent, em_cached, bio, |
808f80b4 FM |
3266 | mirror_num, bio_flags, rw, |
3267 | prev_em_start); | |
9974090b MX |
3268 | } |
3269 | ||
3270 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
3271 | struct page *page, | |
3272 | get_extent_t *get_extent, | |
3273 | struct bio **bio, int mirror_num, | |
3274 | unsigned long *bio_flags, int rw) | |
3275 | { | |
3276 | struct inode *inode = page->mapping->host; | |
3277 | struct btrfs_ordered_extent *ordered; | |
3278 | u64 start = page_offset(page); | |
3279 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
3280 | int ret; | |
3281 | ||
3282 | while (1) { | |
3283 | lock_extent(tree, start, end); | |
3284 | ordered = btrfs_lookup_ordered_extent(inode, start); | |
3285 | if (!ordered) | |
3286 | break; | |
3287 | unlock_extent(tree, start, end); | |
3288 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3289 | btrfs_put_ordered_extent(ordered); | |
3290 | } | |
3291 | ||
125bac01 | 3292 | ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num, |
005efedf | 3293 | bio_flags, rw, NULL); |
9974090b MX |
3294 | return ret; |
3295 | } | |
3296 | ||
d1310b2e | 3297 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, |
8ddc7d9c | 3298 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3299 | { |
3300 | struct bio *bio = NULL; | |
c8b97818 | 3301 | unsigned long bio_flags = 0; |
d1310b2e CM |
3302 | int ret; |
3303 | ||
8ddc7d9c | 3304 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, |
d4c7ca86 | 3305 | &bio_flags, READ); |
d1310b2e | 3306 | if (bio) |
8ddc7d9c | 3307 | ret = submit_one_bio(READ, bio, mirror_num, bio_flags); |
d1310b2e CM |
3308 | return ret; |
3309 | } | |
d1310b2e | 3310 | |
4b384318 MF |
3311 | int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page, |
3312 | get_extent_t *get_extent, int mirror_num) | |
3313 | { | |
3314 | struct bio *bio = NULL; | |
3315 | unsigned long bio_flags = EXTENT_BIO_PARENT_LOCKED; | |
3316 | int ret; | |
3317 | ||
3318 | ret = __do_readpage(tree, page, get_extent, NULL, &bio, mirror_num, | |
005efedf | 3319 | &bio_flags, READ, NULL); |
4b384318 MF |
3320 | if (bio) |
3321 | ret = submit_one_bio(READ, bio, mirror_num, bio_flags); | |
3322 | return ret; | |
3323 | } | |
3324 | ||
11c8349b CM |
3325 | static noinline void update_nr_written(struct page *page, |
3326 | struct writeback_control *wbc, | |
3327 | unsigned long nr_written) | |
3328 | { | |
3329 | wbc->nr_to_write -= nr_written; | |
3330 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && | |
3331 | wbc->range_start == 0 && wbc->range_end == LLONG_MAX)) | |
3332 | page->mapping->writeback_index = page->index + nr_written; | |
3333 | } | |
3334 | ||
d1310b2e | 3335 | /* |
40f76580 CM |
3336 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
3337 | * | |
3338 | * This returns 1 if our fill_delalloc function did all the work required | |
3339 | * to write the page (copy into inline extent). In this case the IO has | |
3340 | * been started and the page is already unlocked. | |
3341 | * | |
3342 | * This returns 0 if all went well (page still locked) | |
3343 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 3344 | */ |
40f76580 CM |
3345 | static noinline_for_stack int writepage_delalloc(struct inode *inode, |
3346 | struct page *page, struct writeback_control *wbc, | |
3347 | struct extent_page_data *epd, | |
3348 | u64 delalloc_start, | |
3349 | unsigned long *nr_written) | |
3350 | { | |
3351 | struct extent_io_tree *tree = epd->tree; | |
3352 | u64 page_end = delalloc_start + PAGE_CACHE_SIZE - 1; | |
3353 | u64 nr_delalloc; | |
3354 | u64 delalloc_to_write = 0; | |
3355 | u64 delalloc_end = 0; | |
3356 | int ret; | |
3357 | int page_started = 0; | |
3358 | ||
3359 | if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc) | |
3360 | return 0; | |
3361 | ||
3362 | while (delalloc_end < page_end) { | |
3363 | nr_delalloc = find_lock_delalloc_range(inode, tree, | |
3364 | page, | |
3365 | &delalloc_start, | |
3366 | &delalloc_end, | |
dcab6a3b | 3367 | BTRFS_MAX_EXTENT_SIZE); |
40f76580 CM |
3368 | if (nr_delalloc == 0) { |
3369 | delalloc_start = delalloc_end + 1; | |
3370 | continue; | |
3371 | } | |
3372 | ret = tree->ops->fill_delalloc(inode, page, | |
3373 | delalloc_start, | |
3374 | delalloc_end, | |
3375 | &page_started, | |
3376 | nr_written); | |
3377 | /* File system has been set read-only */ | |
3378 | if (ret) { | |
3379 | SetPageError(page); | |
3380 | /* fill_delalloc should be return < 0 for error | |
3381 | * but just in case, we use > 0 here meaning the | |
3382 | * IO is started, so we don't want to return > 0 | |
3383 | * unless things are going well. | |
3384 | */ | |
3385 | ret = ret < 0 ? ret : -EIO; | |
3386 | goto done; | |
3387 | } | |
3388 | /* | |
3389 | * delalloc_end is already one less than the total | |
3390 | * length, so we don't subtract one from | |
3391 | * PAGE_CACHE_SIZE | |
3392 | */ | |
3393 | delalloc_to_write += (delalloc_end - delalloc_start + | |
3394 | PAGE_CACHE_SIZE) >> | |
3395 | PAGE_CACHE_SHIFT; | |
3396 | delalloc_start = delalloc_end + 1; | |
3397 | } | |
3398 | if (wbc->nr_to_write < delalloc_to_write) { | |
3399 | int thresh = 8192; | |
3400 | ||
3401 | if (delalloc_to_write < thresh * 2) | |
3402 | thresh = delalloc_to_write; | |
3403 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3404 | thresh); | |
3405 | } | |
3406 | ||
3407 | /* did the fill delalloc function already unlock and start | |
3408 | * the IO? | |
3409 | */ | |
3410 | if (page_started) { | |
3411 | /* | |
3412 | * we've unlocked the page, so we can't update | |
3413 | * the mapping's writeback index, just update | |
3414 | * nr_to_write. | |
3415 | */ | |
3416 | wbc->nr_to_write -= *nr_written; | |
3417 | return 1; | |
3418 | } | |
3419 | ||
3420 | ret = 0; | |
3421 | ||
3422 | done: | |
3423 | return ret; | |
3424 | } | |
3425 | ||
3426 | /* | |
3427 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3428 | * and does the loop to map the page into extents and bios. | |
3429 | * | |
3430 | * We return 1 if the IO is started and the page is unlocked, | |
3431 | * 0 if all went well (page still locked) | |
3432 | * < 0 if there were errors (page still locked) | |
3433 | */ | |
3434 | static noinline_for_stack int __extent_writepage_io(struct inode *inode, | |
3435 | struct page *page, | |
3436 | struct writeback_control *wbc, | |
3437 | struct extent_page_data *epd, | |
3438 | loff_t i_size, | |
3439 | unsigned long nr_written, | |
3440 | int write_flags, int *nr_ret) | |
d1310b2e | 3441 | { |
d1310b2e | 3442 | struct extent_io_tree *tree = epd->tree; |
4eee4fa4 | 3443 | u64 start = page_offset(page); |
d1310b2e CM |
3444 | u64 page_end = start + PAGE_CACHE_SIZE - 1; |
3445 | u64 end; | |
3446 | u64 cur = start; | |
3447 | u64 extent_offset; | |
d1310b2e CM |
3448 | u64 block_start; |
3449 | u64 iosize; | |
3450 | sector_t sector; | |
2c64c53d | 3451 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
3452 | struct extent_map *em; |
3453 | struct block_device *bdev; | |
7f3c74fb | 3454 | size_t pg_offset = 0; |
d1310b2e | 3455 | size_t blocksize; |
40f76580 CM |
3456 | int ret = 0; |
3457 | int nr = 0; | |
3458 | bool compressed; | |
c8b97818 | 3459 | |
247e743c | 3460 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
3461 | ret = tree->ops->writepage_start_hook(page, start, |
3462 | page_end); | |
87826df0 JM |
3463 | if (ret) { |
3464 | /* Fixup worker will requeue */ | |
3465 | if (ret == -EBUSY) | |
3466 | wbc->pages_skipped++; | |
3467 | else | |
3468 | redirty_page_for_writepage(wbc, page); | |
40f76580 | 3469 | |
11c8349b | 3470 | update_nr_written(page, wbc, nr_written); |
247e743c | 3471 | unlock_page(page); |
40f76580 | 3472 | ret = 1; |
11c8349b | 3473 | goto done_unlocked; |
247e743c CM |
3474 | } |
3475 | } | |
3476 | ||
11c8349b CM |
3477 | /* |
3478 | * we don't want to touch the inode after unlocking the page, | |
3479 | * so we update the mapping writeback index now | |
3480 | */ | |
3481 | update_nr_written(page, wbc, nr_written + 1); | |
771ed689 | 3482 | |
d1310b2e | 3483 | end = page_end; |
40f76580 | 3484 | if (i_size <= start) { |
e6dcd2dc CM |
3485 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3486 | tree->ops->writepage_end_io_hook(page, start, | |
3487 | page_end, NULL, 1); | |
d1310b2e CM |
3488 | goto done; |
3489 | } | |
3490 | ||
d1310b2e CM |
3491 | blocksize = inode->i_sb->s_blocksize; |
3492 | ||
3493 | while (cur <= end) { | |
40f76580 CM |
3494 | u64 em_end; |
3495 | if (cur >= i_size) { | |
e6dcd2dc CM |
3496 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3497 | tree->ops->writepage_end_io_hook(page, cur, | |
3498 | page_end, NULL, 1); | |
d1310b2e CM |
3499 | break; |
3500 | } | |
7f3c74fb | 3501 | em = epd->get_extent(inode, page, pg_offset, cur, |
d1310b2e | 3502 | end - cur + 1, 1); |
c704005d | 3503 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3504 | SetPageError(page); |
61391d56 | 3505 | ret = PTR_ERR_OR_ZERO(em); |
d1310b2e CM |
3506 | break; |
3507 | } | |
3508 | ||
3509 | extent_offset = cur - em->start; | |
40f76580 CM |
3510 | em_end = extent_map_end(em); |
3511 | BUG_ON(em_end <= cur); | |
d1310b2e | 3512 | BUG_ON(end < cur); |
40f76580 | 3513 | iosize = min(em_end - cur, end - cur + 1); |
fda2832f | 3514 | iosize = ALIGN(iosize, blocksize); |
d1310b2e CM |
3515 | sector = (em->block_start + extent_offset) >> 9; |
3516 | bdev = em->bdev; | |
3517 | block_start = em->block_start; | |
c8b97818 | 3518 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
3519 | free_extent_map(em); |
3520 | em = NULL; | |
3521 | ||
c8b97818 CM |
3522 | /* |
3523 | * compressed and inline extents are written through other | |
3524 | * paths in the FS | |
3525 | */ | |
3526 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3527 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
3528 | /* |
3529 | * end_io notification does not happen here for | |
3530 | * compressed extents | |
3531 | */ | |
3532 | if (!compressed && tree->ops && | |
3533 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
3534 | tree->ops->writepage_end_io_hook(page, cur, |
3535 | cur + iosize - 1, | |
3536 | NULL, 1); | |
c8b97818 CM |
3537 | else if (compressed) { |
3538 | /* we don't want to end_page_writeback on | |
3539 | * a compressed extent. this happens | |
3540 | * elsewhere | |
3541 | */ | |
3542 | nr++; | |
3543 | } | |
3544 | ||
3545 | cur += iosize; | |
7f3c74fb | 3546 | pg_offset += iosize; |
d1310b2e CM |
3547 | continue; |
3548 | } | |
c8b97818 | 3549 | |
d1310b2e CM |
3550 | if (tree->ops && tree->ops->writepage_io_hook) { |
3551 | ret = tree->ops->writepage_io_hook(page, cur, | |
3552 | cur + iosize - 1); | |
3553 | } else { | |
3554 | ret = 0; | |
3555 | } | |
1259ab75 | 3556 | if (ret) { |
d1310b2e | 3557 | SetPageError(page); |
1259ab75 | 3558 | } else { |
40f76580 | 3559 | unsigned long max_nr = (i_size >> PAGE_CACHE_SHIFT) + 1; |
7f3c74fb | 3560 | |
d1310b2e CM |
3561 | set_range_writeback(tree, cur, cur + iosize - 1); |
3562 | if (!PageWriteback(page)) { | |
efe120a0 FH |
3563 | btrfs_err(BTRFS_I(inode)->root->fs_info, |
3564 | "page %lu not writeback, cur %llu end %llu", | |
c1c9ff7c | 3565 | page->index, cur, end); |
d1310b2e CM |
3566 | } |
3567 | ||
da2f0f74 | 3568 | ret = submit_extent_page(write_flags, tree, wbc, page, |
ffbd517d CM |
3569 | sector, iosize, pg_offset, |
3570 | bdev, &epd->bio, max_nr, | |
c8b97818 | 3571 | end_bio_extent_writepage, |
005efedf | 3572 | 0, 0, 0, false); |
d1310b2e CM |
3573 | if (ret) |
3574 | SetPageError(page); | |
3575 | } | |
3576 | cur = cur + iosize; | |
7f3c74fb | 3577 | pg_offset += iosize; |
d1310b2e CM |
3578 | nr++; |
3579 | } | |
40f76580 CM |
3580 | done: |
3581 | *nr_ret = nr; | |
3582 | ||
3583 | done_unlocked: | |
3584 | ||
3585 | /* drop our reference on any cached states */ | |
3586 | free_extent_state(cached_state); | |
3587 | return ret; | |
3588 | } | |
3589 | ||
3590 | /* | |
3591 | * the writepage semantics are similar to regular writepage. extent | |
3592 | * records are inserted to lock ranges in the tree, and as dirty areas | |
3593 | * are found, they are marked writeback. Then the lock bits are removed | |
3594 | * and the end_io handler clears the writeback ranges | |
3595 | */ | |
3596 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
3597 | void *data) | |
3598 | { | |
3599 | struct inode *inode = page->mapping->host; | |
3600 | struct extent_page_data *epd = data; | |
3601 | u64 start = page_offset(page); | |
3602 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
3603 | int ret; | |
3604 | int nr = 0; | |
3605 | size_t pg_offset = 0; | |
3606 | loff_t i_size = i_size_read(inode); | |
3607 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
3608 | int write_flags; | |
3609 | unsigned long nr_written = 0; | |
3610 | ||
3611 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3612 | write_flags = WRITE_SYNC; | |
3613 | else | |
3614 | write_flags = WRITE; | |
3615 | ||
3616 | trace___extent_writepage(page, inode, wbc); | |
3617 | ||
3618 | WARN_ON(!PageLocked(page)); | |
3619 | ||
3620 | ClearPageError(page); | |
3621 | ||
3622 | pg_offset = i_size & (PAGE_CACHE_SIZE - 1); | |
3623 | if (page->index > end_index || | |
3624 | (page->index == end_index && !pg_offset)) { | |
3625 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE); | |
3626 | unlock_page(page); | |
3627 | return 0; | |
3628 | } | |
3629 | ||
3630 | if (page->index == end_index) { | |
3631 | char *userpage; | |
3632 | ||
3633 | userpage = kmap_atomic(page); | |
3634 | memset(userpage + pg_offset, 0, | |
3635 | PAGE_CACHE_SIZE - pg_offset); | |
3636 | kunmap_atomic(userpage); | |
3637 | flush_dcache_page(page); | |
3638 | } | |
3639 | ||
3640 | pg_offset = 0; | |
3641 | ||
3642 | set_page_extent_mapped(page); | |
3643 | ||
3644 | ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written); | |
3645 | if (ret == 1) | |
3646 | goto done_unlocked; | |
3647 | if (ret) | |
3648 | goto done; | |
3649 | ||
3650 | ret = __extent_writepage_io(inode, page, wbc, epd, | |
3651 | i_size, nr_written, write_flags, &nr); | |
3652 | if (ret == 1) | |
3653 | goto done_unlocked; | |
3654 | ||
d1310b2e CM |
3655 | done: |
3656 | if (nr == 0) { | |
3657 | /* make sure the mapping tag for page dirty gets cleared */ | |
3658 | set_page_writeback(page); | |
3659 | end_page_writeback(page); | |
3660 | } | |
61391d56 FM |
3661 | if (PageError(page)) { |
3662 | ret = ret < 0 ? ret : -EIO; | |
3663 | end_extent_writepage(page, ret, start, page_end); | |
3664 | } | |
d1310b2e | 3665 | unlock_page(page); |
40f76580 | 3666 | return ret; |
771ed689 | 3667 | |
11c8349b | 3668 | done_unlocked: |
d1310b2e CM |
3669 | return 0; |
3670 | } | |
3671 | ||
fd8b2b61 | 3672 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 3673 | { |
74316201 N |
3674 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
3675 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
3676 | } |
3677 | ||
0e378df1 CM |
3678 | static noinline_for_stack int |
3679 | lock_extent_buffer_for_io(struct extent_buffer *eb, | |
3680 | struct btrfs_fs_info *fs_info, | |
3681 | struct extent_page_data *epd) | |
0b32f4bb JB |
3682 | { |
3683 | unsigned long i, num_pages; | |
3684 | int flush = 0; | |
3685 | int ret = 0; | |
3686 | ||
3687 | if (!btrfs_try_tree_write_lock(eb)) { | |
3688 | flush = 1; | |
3689 | flush_write_bio(epd); | |
3690 | btrfs_tree_lock(eb); | |
3691 | } | |
3692 | ||
3693 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3694 | btrfs_tree_unlock(eb); | |
3695 | if (!epd->sync_io) | |
3696 | return 0; | |
3697 | if (!flush) { | |
3698 | flush_write_bio(epd); | |
3699 | flush = 1; | |
3700 | } | |
a098d8e8 CM |
3701 | while (1) { |
3702 | wait_on_extent_buffer_writeback(eb); | |
3703 | btrfs_tree_lock(eb); | |
3704 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3705 | break; | |
0b32f4bb | 3706 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3707 | } |
3708 | } | |
3709 | ||
51561ffe JB |
3710 | /* |
3711 | * We need to do this to prevent races in people who check if the eb is | |
3712 | * under IO since we can end up having no IO bits set for a short period | |
3713 | * of time. | |
3714 | */ | |
3715 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
3716 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3717 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 3718 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 3719 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
e2d84521 MX |
3720 | __percpu_counter_add(&fs_info->dirty_metadata_bytes, |
3721 | -eb->len, | |
3722 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 3723 | ret = 1; |
51561ffe JB |
3724 | } else { |
3725 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
3726 | } |
3727 | ||
3728 | btrfs_tree_unlock(eb); | |
3729 | ||
3730 | if (!ret) | |
3731 | return ret; | |
3732 | ||
3733 | num_pages = num_extent_pages(eb->start, eb->len); | |
3734 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 3735 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3736 | |
3737 | if (!trylock_page(p)) { | |
3738 | if (!flush) { | |
3739 | flush_write_bio(epd); | |
3740 | flush = 1; | |
3741 | } | |
3742 | lock_page(p); | |
3743 | } | |
3744 | } | |
3745 | ||
3746 | return ret; | |
3747 | } | |
3748 | ||
3749 | static void end_extent_buffer_writeback(struct extent_buffer *eb) | |
3750 | { | |
3751 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
4e857c58 | 3752 | smp_mb__after_atomic(); |
0b32f4bb JB |
3753 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); |
3754 | } | |
3755 | ||
656f30db FM |
3756 | static void set_btree_ioerr(struct page *page) |
3757 | { | |
3758 | struct extent_buffer *eb = (struct extent_buffer *)page->private; | |
3759 | struct btrfs_inode *btree_ino = BTRFS_I(eb->fs_info->btree_inode); | |
3760 | ||
3761 | SetPageError(page); | |
3762 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) | |
3763 | return; | |
3764 | ||
3765 | /* | |
3766 | * If writeback for a btree extent that doesn't belong to a log tree | |
3767 | * failed, increment the counter transaction->eb_write_errors. | |
3768 | * We do this because while the transaction is running and before it's | |
3769 | * committing (when we call filemap_fdata[write|wait]_range against | |
3770 | * the btree inode), we might have | |
3771 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
3772 | * returns an error or an error happens during writeback, when we're | |
3773 | * committing the transaction we wouldn't know about it, since the pages | |
3774 | * can be no longer dirty nor marked anymore for writeback (if a | |
3775 | * subsequent modification to the extent buffer didn't happen before the | |
3776 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
3777 | * able to find the pages tagged with SetPageError at transaction | |
3778 | * commit time. So if this happens we must abort the transaction, | |
3779 | * otherwise we commit a super block with btree roots that point to | |
3780 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
3781 | * or the content of some node/leaf from a past generation that got | |
3782 | * cowed or deleted and is no longer valid. | |
3783 | * | |
3784 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
3785 | * not be enough - we need to distinguish between log tree extents vs | |
3786 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
3787 | * will catch and clear such errors in the mapping - and that call might | |
3788 | * be from a log sync and not from a transaction commit. Also, checking | |
3789 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
3790 | * not done and would not be reliable - the eb might have been released | |
3791 | * from memory and reading it back again means that flag would not be | |
3792 | * set (since it's a runtime flag, not persisted on disk). | |
3793 | * | |
3794 | * Using the flags below in the btree inode also makes us achieve the | |
3795 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
3796 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
3797 | * is called, the writeback for all dirty pages had already finished | |
3798 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
3799 | * filemap_fdatawait_range() would return success, as it could not know | |
3800 | * that writeback errors happened (the pages were no longer tagged for | |
3801 | * writeback). | |
3802 | */ | |
3803 | switch (eb->log_index) { | |
3804 | case -1: | |
3805 | set_bit(BTRFS_INODE_BTREE_ERR, &btree_ino->runtime_flags); | |
3806 | break; | |
3807 | case 0: | |
3808 | set_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags); | |
3809 | break; | |
3810 | case 1: | |
3811 | set_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags); | |
3812 | break; | |
3813 | default: | |
3814 | BUG(); /* unexpected, logic error */ | |
3815 | } | |
3816 | } | |
3817 | ||
4246a0b6 | 3818 | static void end_bio_extent_buffer_writepage(struct bio *bio) |
0b32f4bb | 3819 | { |
2c30c71b | 3820 | struct bio_vec *bvec; |
0b32f4bb | 3821 | struct extent_buffer *eb; |
2c30c71b | 3822 | int i, done; |
0b32f4bb | 3823 | |
2c30c71b | 3824 | bio_for_each_segment_all(bvec, bio, i) { |
0b32f4bb JB |
3825 | struct page *page = bvec->bv_page; |
3826 | ||
0b32f4bb JB |
3827 | eb = (struct extent_buffer *)page->private; |
3828 | BUG_ON(!eb); | |
3829 | done = atomic_dec_and_test(&eb->io_pages); | |
3830 | ||
4246a0b6 CH |
3831 | if (bio->bi_error || |
3832 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { | |
0b32f4bb | 3833 | ClearPageUptodate(page); |
656f30db | 3834 | set_btree_ioerr(page); |
0b32f4bb JB |
3835 | } |
3836 | ||
3837 | end_page_writeback(page); | |
3838 | ||
3839 | if (!done) | |
3840 | continue; | |
3841 | ||
3842 | end_extent_buffer_writeback(eb); | |
2c30c71b | 3843 | } |
0b32f4bb JB |
3844 | |
3845 | bio_put(bio); | |
0b32f4bb JB |
3846 | } |
3847 | ||
0e378df1 | 3848 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb JB |
3849 | struct btrfs_fs_info *fs_info, |
3850 | struct writeback_control *wbc, | |
3851 | struct extent_page_data *epd) | |
3852 | { | |
3853 | struct block_device *bdev = fs_info->fs_devices->latest_bdev; | |
f28491e0 | 3854 | struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree; |
0b32f4bb JB |
3855 | u64 offset = eb->start; |
3856 | unsigned long i, num_pages; | |
de0022b9 | 3857 | unsigned long bio_flags = 0; |
d4c7ca86 | 3858 | int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META; |
d7dbe9e7 | 3859 | int ret = 0; |
0b32f4bb | 3860 | |
656f30db | 3861 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); |
0b32f4bb JB |
3862 | num_pages = num_extent_pages(eb->start, eb->len); |
3863 | atomic_set(&eb->io_pages, num_pages); | |
de0022b9 JB |
3864 | if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID) |
3865 | bio_flags = EXTENT_BIO_TREE_LOG; | |
3866 | ||
0b32f4bb | 3867 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3868 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3869 | |
3870 | clear_page_dirty_for_io(p); | |
3871 | set_page_writeback(p); | |
da2f0f74 | 3872 | ret = submit_extent_page(rw, tree, wbc, p, offset >> 9, |
0b32f4bb JB |
3873 | PAGE_CACHE_SIZE, 0, bdev, &epd->bio, |
3874 | -1, end_bio_extent_buffer_writepage, | |
005efedf | 3875 | 0, epd->bio_flags, bio_flags, false); |
de0022b9 | 3876 | epd->bio_flags = bio_flags; |
0b32f4bb | 3877 | if (ret) { |
656f30db | 3878 | set_btree_ioerr(p); |
55e3bd2e | 3879 | end_page_writeback(p); |
0b32f4bb JB |
3880 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
3881 | end_extent_buffer_writeback(eb); | |
3882 | ret = -EIO; | |
3883 | break; | |
3884 | } | |
3885 | offset += PAGE_CACHE_SIZE; | |
3886 | update_nr_written(p, wbc, 1); | |
3887 | unlock_page(p); | |
3888 | } | |
3889 | ||
3890 | if (unlikely(ret)) { | |
3891 | for (; i < num_pages; i++) { | |
bbf65cf0 | 3892 | struct page *p = eb->pages[i]; |
81465028 | 3893 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
3894 | unlock_page(p); |
3895 | } | |
3896 | } | |
3897 | ||
3898 | return ret; | |
3899 | } | |
3900 | ||
3901 | int btree_write_cache_pages(struct address_space *mapping, | |
3902 | struct writeback_control *wbc) | |
3903 | { | |
3904 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; | |
3905 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; | |
3906 | struct extent_buffer *eb, *prev_eb = NULL; | |
3907 | struct extent_page_data epd = { | |
3908 | .bio = NULL, | |
3909 | .tree = tree, | |
3910 | .extent_locked = 0, | |
3911 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
de0022b9 | 3912 | .bio_flags = 0, |
0b32f4bb JB |
3913 | }; |
3914 | int ret = 0; | |
3915 | int done = 0; | |
3916 | int nr_to_write_done = 0; | |
3917 | struct pagevec pvec; | |
3918 | int nr_pages; | |
3919 | pgoff_t index; | |
3920 | pgoff_t end; /* Inclusive */ | |
3921 | int scanned = 0; | |
3922 | int tag; | |
3923 | ||
3924 | pagevec_init(&pvec, 0); | |
3925 | if (wbc->range_cyclic) { | |
3926 | index = mapping->writeback_index; /* Start from prev offset */ | |
3927 | end = -1; | |
3928 | } else { | |
3929 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
3930 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
3931 | scanned = 1; | |
3932 | } | |
3933 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3934 | tag = PAGECACHE_TAG_TOWRITE; | |
3935 | else | |
3936 | tag = PAGECACHE_TAG_DIRTY; | |
3937 | retry: | |
3938 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3939 | tag_pages_for_writeback(mapping, index, end); | |
3940 | while (!done && !nr_to_write_done && (index <= end) && | |
3941 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, | |
3942 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
3943 | unsigned i; | |
3944 | ||
3945 | scanned = 1; | |
3946 | for (i = 0; i < nr_pages; i++) { | |
3947 | struct page *page = pvec.pages[i]; | |
3948 | ||
3949 | if (!PagePrivate(page)) | |
3950 | continue; | |
3951 | ||
3952 | if (!wbc->range_cyclic && page->index > end) { | |
3953 | done = 1; | |
3954 | break; | |
3955 | } | |
3956 | ||
b5bae261 JB |
3957 | spin_lock(&mapping->private_lock); |
3958 | if (!PagePrivate(page)) { | |
3959 | spin_unlock(&mapping->private_lock); | |
3960 | continue; | |
3961 | } | |
3962 | ||
0b32f4bb | 3963 | eb = (struct extent_buffer *)page->private; |
b5bae261 JB |
3964 | |
3965 | /* | |
3966 | * Shouldn't happen and normally this would be a BUG_ON | |
3967 | * but no sense in crashing the users box for something | |
3968 | * we can survive anyway. | |
3969 | */ | |
fae7f21c | 3970 | if (WARN_ON(!eb)) { |
b5bae261 | 3971 | spin_unlock(&mapping->private_lock); |
0b32f4bb JB |
3972 | continue; |
3973 | } | |
3974 | ||
b5bae261 JB |
3975 | if (eb == prev_eb) { |
3976 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 3977 | continue; |
b5bae261 | 3978 | } |
0b32f4bb | 3979 | |
b5bae261 JB |
3980 | ret = atomic_inc_not_zero(&eb->refs); |
3981 | spin_unlock(&mapping->private_lock); | |
3982 | if (!ret) | |
0b32f4bb | 3983 | continue; |
0b32f4bb JB |
3984 | |
3985 | prev_eb = eb; | |
3986 | ret = lock_extent_buffer_for_io(eb, fs_info, &epd); | |
3987 | if (!ret) { | |
3988 | free_extent_buffer(eb); | |
3989 | continue; | |
3990 | } | |
3991 | ||
3992 | ret = write_one_eb(eb, fs_info, wbc, &epd); | |
3993 | if (ret) { | |
3994 | done = 1; | |
3995 | free_extent_buffer(eb); | |
3996 | break; | |
3997 | } | |
3998 | free_extent_buffer(eb); | |
3999 | ||
4000 | /* | |
4001 | * the filesystem may choose to bump up nr_to_write. | |
4002 | * We have to make sure to honor the new nr_to_write | |
4003 | * at any time | |
4004 | */ | |
4005 | nr_to_write_done = wbc->nr_to_write <= 0; | |
4006 | } | |
4007 | pagevec_release(&pvec); | |
4008 | cond_resched(); | |
4009 | } | |
4010 | if (!scanned && !done) { | |
4011 | /* | |
4012 | * We hit the last page and there is more work to be done: wrap | |
4013 | * back to the start of the file | |
4014 | */ | |
4015 | scanned = 1; | |
4016 | index = 0; | |
4017 | goto retry; | |
4018 | } | |
4019 | flush_write_bio(&epd); | |
4020 | return ret; | |
4021 | } | |
4022 | ||
d1310b2e | 4023 | /** |
4bef0848 | 4024 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
4025 | * @mapping: address space structure to write |
4026 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
4027 | * @writepage: function called for each page | |
4028 | * @data: data passed to writepage function | |
4029 | * | |
4030 | * If a page is already under I/O, write_cache_pages() skips it, even | |
4031 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
4032 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
4033 | * and msync() need to guarantee that all the data which was dirty at the time | |
4034 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
4035 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
4036 | * existing IO to complete. | |
4037 | */ | |
b2950863 | 4038 | static int extent_write_cache_pages(struct extent_io_tree *tree, |
4bef0848 CM |
4039 | struct address_space *mapping, |
4040 | struct writeback_control *wbc, | |
d2c3f4f6 CM |
4041 | writepage_t writepage, void *data, |
4042 | void (*flush_fn)(void *)) | |
d1310b2e | 4043 | { |
7fd1a3f7 | 4044 | struct inode *inode = mapping->host; |
d1310b2e CM |
4045 | int ret = 0; |
4046 | int done = 0; | |
61391d56 | 4047 | int err = 0; |
f85d7d6c | 4048 | int nr_to_write_done = 0; |
d1310b2e CM |
4049 | struct pagevec pvec; |
4050 | int nr_pages; | |
4051 | pgoff_t index; | |
4052 | pgoff_t end; /* Inclusive */ | |
4053 | int scanned = 0; | |
f7aaa06b | 4054 | int tag; |
d1310b2e | 4055 | |
7fd1a3f7 JB |
4056 | /* |
4057 | * We have to hold onto the inode so that ordered extents can do their | |
4058 | * work when the IO finishes. The alternative to this is failing to add | |
4059 | * an ordered extent if the igrab() fails there and that is a huge pain | |
4060 | * to deal with, so instead just hold onto the inode throughout the | |
4061 | * writepages operation. If it fails here we are freeing up the inode | |
4062 | * anyway and we'd rather not waste our time writing out stuff that is | |
4063 | * going to be truncated anyway. | |
4064 | */ | |
4065 | if (!igrab(inode)) | |
4066 | return 0; | |
4067 | ||
d1310b2e CM |
4068 | pagevec_init(&pvec, 0); |
4069 | if (wbc->range_cyclic) { | |
4070 | index = mapping->writeback_index; /* Start from prev offset */ | |
4071 | end = -1; | |
4072 | } else { | |
4073 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
4074 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
4075 | scanned = 1; |
4076 | } | |
f7aaa06b JB |
4077 | if (wbc->sync_mode == WB_SYNC_ALL) |
4078 | tag = PAGECACHE_TAG_TOWRITE; | |
4079 | else | |
4080 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 4081 | retry: |
f7aaa06b JB |
4082 | if (wbc->sync_mode == WB_SYNC_ALL) |
4083 | tag_pages_for_writeback(mapping, index, end); | |
f85d7d6c | 4084 | while (!done && !nr_to_write_done && (index <= end) && |
f7aaa06b JB |
4085 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
4086 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
d1310b2e CM |
4087 | unsigned i; |
4088 | ||
4089 | scanned = 1; | |
4090 | for (i = 0; i < nr_pages; i++) { | |
4091 | struct page *page = pvec.pages[i]; | |
4092 | ||
4093 | /* | |
4094 | * At this point we hold neither mapping->tree_lock nor | |
4095 | * lock on the page itself: the page may be truncated or | |
4096 | * invalidated (changing page->mapping to NULL), or even | |
4097 | * swizzled back from swapper_space to tmpfs file | |
4098 | * mapping | |
4099 | */ | |
c8f2f24b JB |
4100 | if (!trylock_page(page)) { |
4101 | flush_fn(data); | |
4102 | lock_page(page); | |
01d658f2 | 4103 | } |
d1310b2e CM |
4104 | |
4105 | if (unlikely(page->mapping != mapping)) { | |
4106 | unlock_page(page); | |
4107 | continue; | |
4108 | } | |
4109 | ||
4110 | if (!wbc->range_cyclic && page->index > end) { | |
4111 | done = 1; | |
4112 | unlock_page(page); | |
4113 | continue; | |
4114 | } | |
4115 | ||
d2c3f4f6 | 4116 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 CM |
4117 | if (PageWriteback(page)) |
4118 | flush_fn(data); | |
d1310b2e | 4119 | wait_on_page_writeback(page); |
d2c3f4f6 | 4120 | } |
d1310b2e CM |
4121 | |
4122 | if (PageWriteback(page) || | |
4123 | !clear_page_dirty_for_io(page)) { | |
4124 | unlock_page(page); | |
4125 | continue; | |
4126 | } | |
4127 | ||
4128 | ret = (*writepage)(page, wbc, data); | |
4129 | ||
4130 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
4131 | unlock_page(page); | |
4132 | ret = 0; | |
4133 | } | |
61391d56 FM |
4134 | if (!err && ret < 0) |
4135 | err = ret; | |
f85d7d6c CM |
4136 | |
4137 | /* | |
4138 | * the filesystem may choose to bump up nr_to_write. | |
4139 | * We have to make sure to honor the new nr_to_write | |
4140 | * at any time | |
4141 | */ | |
4142 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
4143 | } |
4144 | pagevec_release(&pvec); | |
4145 | cond_resched(); | |
4146 | } | |
61391d56 | 4147 | if (!scanned && !done && !err) { |
d1310b2e CM |
4148 | /* |
4149 | * We hit the last page and there is more work to be done: wrap | |
4150 | * back to the start of the file | |
4151 | */ | |
4152 | scanned = 1; | |
4153 | index = 0; | |
4154 | goto retry; | |
4155 | } | |
7fd1a3f7 | 4156 | btrfs_add_delayed_iput(inode); |
61391d56 | 4157 | return err; |
d1310b2e | 4158 | } |
d1310b2e | 4159 | |
ffbd517d | 4160 | static void flush_epd_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 4161 | { |
d2c3f4f6 | 4162 | if (epd->bio) { |
355808c2 JM |
4163 | int rw = WRITE; |
4164 | int ret; | |
4165 | ||
ffbd517d | 4166 | if (epd->sync_io) |
355808c2 JM |
4167 | rw = WRITE_SYNC; |
4168 | ||
de0022b9 | 4169 | ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags); |
79787eaa | 4170 | BUG_ON(ret < 0); /* -ENOMEM */ |
d2c3f4f6 CM |
4171 | epd->bio = NULL; |
4172 | } | |
4173 | } | |
4174 | ||
ffbd517d CM |
4175 | static noinline void flush_write_bio(void *data) |
4176 | { | |
4177 | struct extent_page_data *epd = data; | |
4178 | flush_epd_write_bio(epd); | |
4179 | } | |
4180 | ||
d1310b2e CM |
4181 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, |
4182 | get_extent_t *get_extent, | |
4183 | struct writeback_control *wbc) | |
4184 | { | |
4185 | int ret; | |
d1310b2e CM |
4186 | struct extent_page_data epd = { |
4187 | .bio = NULL, | |
4188 | .tree = tree, | |
4189 | .get_extent = get_extent, | |
771ed689 | 4190 | .extent_locked = 0, |
ffbd517d | 4191 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
de0022b9 | 4192 | .bio_flags = 0, |
d1310b2e | 4193 | }; |
d1310b2e | 4194 | |
d1310b2e CM |
4195 | ret = __extent_writepage(page, wbc, &epd); |
4196 | ||
ffbd517d | 4197 | flush_epd_write_bio(&epd); |
d1310b2e CM |
4198 | return ret; |
4199 | } | |
d1310b2e | 4200 | |
771ed689 CM |
4201 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, |
4202 | u64 start, u64 end, get_extent_t *get_extent, | |
4203 | int mode) | |
4204 | { | |
4205 | int ret = 0; | |
4206 | struct address_space *mapping = inode->i_mapping; | |
4207 | struct page *page; | |
4208 | unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >> | |
4209 | PAGE_CACHE_SHIFT; | |
4210 | ||
4211 | struct extent_page_data epd = { | |
4212 | .bio = NULL, | |
4213 | .tree = tree, | |
4214 | .get_extent = get_extent, | |
4215 | .extent_locked = 1, | |
ffbd517d | 4216 | .sync_io = mode == WB_SYNC_ALL, |
de0022b9 | 4217 | .bio_flags = 0, |
771ed689 CM |
4218 | }; |
4219 | struct writeback_control wbc_writepages = { | |
771ed689 | 4220 | .sync_mode = mode, |
771ed689 CM |
4221 | .nr_to_write = nr_pages * 2, |
4222 | .range_start = start, | |
4223 | .range_end = end + 1, | |
4224 | }; | |
4225 | ||
d397712b | 4226 | while (start <= end) { |
771ed689 CM |
4227 | page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); |
4228 | if (clear_page_dirty_for_io(page)) | |
4229 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
4230 | else { | |
4231 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
4232 | tree->ops->writepage_end_io_hook(page, start, | |
4233 | start + PAGE_CACHE_SIZE - 1, | |
4234 | NULL, 1); | |
4235 | unlock_page(page); | |
4236 | } | |
4237 | page_cache_release(page); | |
4238 | start += PAGE_CACHE_SIZE; | |
4239 | } | |
4240 | ||
ffbd517d | 4241 | flush_epd_write_bio(&epd); |
771ed689 CM |
4242 | return ret; |
4243 | } | |
d1310b2e CM |
4244 | |
4245 | int extent_writepages(struct extent_io_tree *tree, | |
4246 | struct address_space *mapping, | |
4247 | get_extent_t *get_extent, | |
4248 | struct writeback_control *wbc) | |
4249 | { | |
4250 | int ret = 0; | |
4251 | struct extent_page_data epd = { | |
4252 | .bio = NULL, | |
4253 | .tree = tree, | |
4254 | .get_extent = get_extent, | |
771ed689 | 4255 | .extent_locked = 0, |
ffbd517d | 4256 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
de0022b9 | 4257 | .bio_flags = 0, |
d1310b2e CM |
4258 | }; |
4259 | ||
4bef0848 | 4260 | ret = extent_write_cache_pages(tree, mapping, wbc, |
d2c3f4f6 CM |
4261 | __extent_writepage, &epd, |
4262 | flush_write_bio); | |
ffbd517d | 4263 | flush_epd_write_bio(&epd); |
d1310b2e CM |
4264 | return ret; |
4265 | } | |
d1310b2e CM |
4266 | |
4267 | int extent_readpages(struct extent_io_tree *tree, | |
4268 | struct address_space *mapping, | |
4269 | struct list_head *pages, unsigned nr_pages, | |
4270 | get_extent_t get_extent) | |
4271 | { | |
4272 | struct bio *bio = NULL; | |
4273 | unsigned page_idx; | |
c8b97818 | 4274 | unsigned long bio_flags = 0; |
67c9684f LB |
4275 | struct page *pagepool[16]; |
4276 | struct page *page; | |
125bac01 | 4277 | struct extent_map *em_cached = NULL; |
67c9684f | 4278 | int nr = 0; |
808f80b4 | 4279 | u64 prev_em_start = (u64)-1; |
d1310b2e | 4280 | |
d1310b2e | 4281 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
67c9684f | 4282 | page = list_entry(pages->prev, struct page, lru); |
d1310b2e CM |
4283 | |
4284 | prefetchw(&page->flags); | |
4285 | list_del(&page->lru); | |
67c9684f | 4286 | if (add_to_page_cache_lru(page, mapping, |
43e817a1 | 4287 | page->index, GFP_NOFS)) { |
67c9684f LB |
4288 | page_cache_release(page); |
4289 | continue; | |
d1310b2e | 4290 | } |
67c9684f LB |
4291 | |
4292 | pagepool[nr++] = page; | |
4293 | if (nr < ARRAY_SIZE(pagepool)) | |
4294 | continue; | |
125bac01 | 4295 | __extent_readpages(tree, pagepool, nr, get_extent, &em_cached, |
808f80b4 | 4296 | &bio, 0, &bio_flags, READ, &prev_em_start); |
67c9684f | 4297 | nr = 0; |
d1310b2e | 4298 | } |
9974090b | 4299 | if (nr) |
125bac01 | 4300 | __extent_readpages(tree, pagepool, nr, get_extent, &em_cached, |
808f80b4 | 4301 | &bio, 0, &bio_flags, READ, &prev_em_start); |
67c9684f | 4302 | |
125bac01 MX |
4303 | if (em_cached) |
4304 | free_extent_map(em_cached); | |
4305 | ||
d1310b2e CM |
4306 | BUG_ON(!list_empty(pages)); |
4307 | if (bio) | |
79787eaa | 4308 | return submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
4309 | return 0; |
4310 | } | |
d1310b2e CM |
4311 | |
4312 | /* | |
4313 | * basic invalidatepage code, this waits on any locked or writeback | |
4314 | * ranges corresponding to the page, and then deletes any extent state | |
4315 | * records from the tree | |
4316 | */ | |
4317 | int extent_invalidatepage(struct extent_io_tree *tree, | |
4318 | struct page *page, unsigned long offset) | |
4319 | { | |
2ac55d41 | 4320 | struct extent_state *cached_state = NULL; |
4eee4fa4 | 4321 | u64 start = page_offset(page); |
d1310b2e CM |
4322 | u64 end = start + PAGE_CACHE_SIZE - 1; |
4323 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
4324 | ||
fda2832f | 4325 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
4326 | if (start > end) |
4327 | return 0; | |
4328 | ||
d0082371 | 4329 | lock_extent_bits(tree, start, end, 0, &cached_state); |
1edbb734 | 4330 | wait_on_page_writeback(page); |
d1310b2e | 4331 | clear_extent_bit(tree, start, end, |
32c00aff JB |
4332 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
4333 | EXTENT_DO_ACCOUNTING, | |
2ac55d41 | 4334 | 1, 1, &cached_state, GFP_NOFS); |
d1310b2e CM |
4335 | return 0; |
4336 | } | |
d1310b2e | 4337 | |
7b13b7b1 CM |
4338 | /* |
4339 | * a helper for releasepage, this tests for areas of the page that | |
4340 | * are locked or under IO and drops the related state bits if it is safe | |
4341 | * to drop the page. | |
4342 | */ | |
48a3b636 ES |
4343 | static int try_release_extent_state(struct extent_map_tree *map, |
4344 | struct extent_io_tree *tree, | |
4345 | struct page *page, gfp_t mask) | |
7b13b7b1 | 4346 | { |
4eee4fa4 | 4347 | u64 start = page_offset(page); |
7b13b7b1 CM |
4348 | u64 end = start + PAGE_CACHE_SIZE - 1; |
4349 | int ret = 1; | |
4350 | ||
211f90e6 | 4351 | if (test_range_bit(tree, start, end, |
8b62b72b | 4352 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
4353 | ret = 0; |
4354 | else { | |
4355 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
4356 | mask = GFP_NOFS; | |
11ef160f CM |
4357 | /* |
4358 | * at this point we can safely clear everything except the | |
4359 | * locked bit and the nodatasum bit | |
4360 | */ | |
e3f24cc5 | 4361 | ret = clear_extent_bit(tree, start, end, |
11ef160f CM |
4362 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
4363 | 0, 0, NULL, mask); | |
e3f24cc5 CM |
4364 | |
4365 | /* if clear_extent_bit failed for enomem reasons, | |
4366 | * we can't allow the release to continue. | |
4367 | */ | |
4368 | if (ret < 0) | |
4369 | ret = 0; | |
4370 | else | |
4371 | ret = 1; | |
7b13b7b1 CM |
4372 | } |
4373 | return ret; | |
4374 | } | |
7b13b7b1 | 4375 | |
d1310b2e CM |
4376 | /* |
4377 | * a helper for releasepage. As long as there are no locked extents | |
4378 | * in the range corresponding to the page, both state records and extent | |
4379 | * map records are removed | |
4380 | */ | |
4381 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
4382 | struct extent_io_tree *tree, struct page *page, |
4383 | gfp_t mask) | |
d1310b2e CM |
4384 | { |
4385 | struct extent_map *em; | |
4eee4fa4 | 4386 | u64 start = page_offset(page); |
d1310b2e | 4387 | u64 end = start + PAGE_CACHE_SIZE - 1; |
7b13b7b1 | 4388 | |
d0164adc | 4389 | if (gfpflags_allow_blocking(mask) && |
70dec807 | 4390 | page->mapping->host->i_size > 16 * 1024 * 1024) { |
39b5637f | 4391 | u64 len; |
70dec807 | 4392 | while (start <= end) { |
39b5637f | 4393 | len = end - start + 1; |
890871be | 4394 | write_lock(&map->lock); |
39b5637f | 4395 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 4396 | if (!em) { |
890871be | 4397 | write_unlock(&map->lock); |
70dec807 CM |
4398 | break; |
4399 | } | |
7f3c74fb CM |
4400 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
4401 | em->start != start) { | |
890871be | 4402 | write_unlock(&map->lock); |
70dec807 CM |
4403 | free_extent_map(em); |
4404 | break; | |
4405 | } | |
4406 | if (!test_range_bit(tree, em->start, | |
4407 | extent_map_end(em) - 1, | |
8b62b72b | 4408 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 4409 | 0, NULL)) { |
70dec807 CM |
4410 | remove_extent_mapping(map, em); |
4411 | /* once for the rb tree */ | |
4412 | free_extent_map(em); | |
4413 | } | |
4414 | start = extent_map_end(em); | |
890871be | 4415 | write_unlock(&map->lock); |
70dec807 CM |
4416 | |
4417 | /* once for us */ | |
d1310b2e CM |
4418 | free_extent_map(em); |
4419 | } | |
d1310b2e | 4420 | } |
7b13b7b1 | 4421 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 4422 | } |
d1310b2e | 4423 | |
ec29ed5b CM |
4424 | /* |
4425 | * helper function for fiemap, which doesn't want to see any holes. | |
4426 | * This maps until we find something past 'last' | |
4427 | */ | |
4428 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
4429 | u64 offset, | |
4430 | u64 last, | |
4431 | get_extent_t *get_extent) | |
4432 | { | |
4433 | u64 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
4434 | struct extent_map *em; | |
4435 | u64 len; | |
4436 | ||
4437 | if (offset >= last) | |
4438 | return NULL; | |
4439 | ||
67871254 | 4440 | while (1) { |
ec29ed5b CM |
4441 | len = last - offset; |
4442 | if (len == 0) | |
4443 | break; | |
fda2832f | 4444 | len = ALIGN(len, sectorsize); |
ec29ed5b | 4445 | em = get_extent(inode, NULL, 0, offset, len, 0); |
c704005d | 4446 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
4447 | return em; |
4448 | ||
4449 | /* if this isn't a hole return it */ | |
4450 | if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) && | |
4451 | em->block_start != EXTENT_MAP_HOLE) { | |
4452 | return em; | |
4453 | } | |
4454 | ||
4455 | /* this is a hole, advance to the next extent */ | |
4456 | offset = extent_map_end(em); | |
4457 | free_extent_map(em); | |
4458 | if (offset >= last) | |
4459 | break; | |
4460 | } | |
4461 | return NULL; | |
4462 | } | |
4463 | ||
1506fcc8 YS |
4464 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4465 | __u64 start, __u64 len, get_extent_t *get_extent) | |
4466 | { | |
975f84fe | 4467 | int ret = 0; |
1506fcc8 YS |
4468 | u64 off = start; |
4469 | u64 max = start + len; | |
4470 | u32 flags = 0; | |
975f84fe JB |
4471 | u32 found_type; |
4472 | u64 last; | |
ec29ed5b | 4473 | u64 last_for_get_extent = 0; |
1506fcc8 | 4474 | u64 disko = 0; |
ec29ed5b | 4475 | u64 isize = i_size_read(inode); |
975f84fe | 4476 | struct btrfs_key found_key; |
1506fcc8 | 4477 | struct extent_map *em = NULL; |
2ac55d41 | 4478 | struct extent_state *cached_state = NULL; |
975f84fe | 4479 | struct btrfs_path *path; |
dc046b10 | 4480 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1506fcc8 | 4481 | int end = 0; |
ec29ed5b CM |
4482 | u64 em_start = 0; |
4483 | u64 em_len = 0; | |
4484 | u64 em_end = 0; | |
1506fcc8 YS |
4485 | |
4486 | if (len == 0) | |
4487 | return -EINVAL; | |
4488 | ||
975f84fe JB |
4489 | path = btrfs_alloc_path(); |
4490 | if (!path) | |
4491 | return -ENOMEM; | |
4492 | path->leave_spinning = 1; | |
4493 | ||
2c91943b QW |
4494 | start = round_down(start, BTRFS_I(inode)->root->sectorsize); |
4495 | len = round_up(max, BTRFS_I(inode)->root->sectorsize) - start; | |
4d479cf0 | 4496 | |
ec29ed5b CM |
4497 | /* |
4498 | * lookup the last file extent. We're not using i_size here | |
4499 | * because there might be preallocation past i_size | |
4500 | */ | |
dc046b10 JB |
4501 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1, |
4502 | 0); | |
975f84fe JB |
4503 | if (ret < 0) { |
4504 | btrfs_free_path(path); | |
4505 | return ret; | |
4506 | } | |
4507 | WARN_ON(!ret); | |
4508 | path->slots[0]--; | |
975f84fe | 4509 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
962a298f | 4510 | found_type = found_key.type; |
975f84fe | 4511 | |
ec29ed5b | 4512 | /* No extents, but there might be delalloc bits */ |
33345d01 | 4513 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 4514 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
4515 | /* have to trust i_size as the end */ |
4516 | last = (u64)-1; | |
4517 | last_for_get_extent = isize; | |
4518 | } else { | |
4519 | /* | |
4520 | * remember the start of the last extent. There are a | |
4521 | * bunch of different factors that go into the length of the | |
4522 | * extent, so its much less complex to remember where it started | |
4523 | */ | |
4524 | last = found_key.offset; | |
4525 | last_for_get_extent = last + 1; | |
975f84fe | 4526 | } |
fe09e16c | 4527 | btrfs_release_path(path); |
975f84fe | 4528 | |
ec29ed5b CM |
4529 | /* |
4530 | * we might have some extents allocated but more delalloc past those | |
4531 | * extents. so, we trust isize unless the start of the last extent is | |
4532 | * beyond isize | |
4533 | */ | |
4534 | if (last < isize) { | |
4535 | last = (u64)-1; | |
4536 | last_for_get_extent = isize; | |
4537 | } | |
4538 | ||
a52f4cd2 | 4539 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0, |
d0082371 | 4540 | &cached_state); |
ec29ed5b | 4541 | |
4d479cf0 | 4542 | em = get_extent_skip_holes(inode, start, last_for_get_extent, |
ec29ed5b | 4543 | get_extent); |
1506fcc8 YS |
4544 | if (!em) |
4545 | goto out; | |
4546 | if (IS_ERR(em)) { | |
4547 | ret = PTR_ERR(em); | |
4548 | goto out; | |
4549 | } | |
975f84fe | 4550 | |
1506fcc8 | 4551 | while (!end) { |
b76bb701 | 4552 | u64 offset_in_extent = 0; |
ea8efc74 CM |
4553 | |
4554 | /* break if the extent we found is outside the range */ | |
4555 | if (em->start >= max || extent_map_end(em) < off) | |
4556 | break; | |
4557 | ||
4558 | /* | |
4559 | * get_extent may return an extent that starts before our | |
4560 | * requested range. We have to make sure the ranges | |
4561 | * we return to fiemap always move forward and don't | |
4562 | * overlap, so adjust the offsets here | |
4563 | */ | |
4564 | em_start = max(em->start, off); | |
1506fcc8 | 4565 | |
ea8efc74 CM |
4566 | /* |
4567 | * record the offset from the start of the extent | |
b76bb701 JB |
4568 | * for adjusting the disk offset below. Only do this if the |
4569 | * extent isn't compressed since our in ram offset may be past | |
4570 | * what we have actually allocated on disk. | |
ea8efc74 | 4571 | */ |
b76bb701 JB |
4572 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
4573 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 4574 | em_end = extent_map_end(em); |
ea8efc74 | 4575 | em_len = em_end - em_start; |
1506fcc8 YS |
4576 | disko = 0; |
4577 | flags = 0; | |
4578 | ||
ea8efc74 CM |
4579 | /* |
4580 | * bump off for our next call to get_extent | |
4581 | */ | |
4582 | off = extent_map_end(em); | |
4583 | if (off >= max) | |
4584 | end = 1; | |
4585 | ||
93dbfad7 | 4586 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
4587 | end = 1; |
4588 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 4589 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
4590 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
4591 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 4592 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
4593 | flags |= (FIEMAP_EXTENT_DELALLOC | |
4594 | FIEMAP_EXTENT_UNKNOWN); | |
dc046b10 JB |
4595 | } else if (fieinfo->fi_extents_max) { |
4596 | u64 bytenr = em->block_start - | |
4597 | (em->start - em->orig_start); | |
fe09e16c | 4598 | |
ea8efc74 | 4599 | disko = em->block_start + offset_in_extent; |
fe09e16c LB |
4600 | |
4601 | /* | |
4602 | * As btrfs supports shared space, this information | |
4603 | * can be exported to userspace tools via | |
dc046b10 JB |
4604 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 |
4605 | * then we're just getting a count and we can skip the | |
4606 | * lookup stuff. | |
fe09e16c | 4607 | */ |
dc046b10 JB |
4608 | ret = btrfs_check_shared(NULL, root->fs_info, |
4609 | root->objectid, | |
4610 | btrfs_ino(inode), bytenr); | |
4611 | if (ret < 0) | |
fe09e16c | 4612 | goto out_free; |
dc046b10 | 4613 | if (ret) |
fe09e16c | 4614 | flags |= FIEMAP_EXTENT_SHARED; |
dc046b10 | 4615 | ret = 0; |
1506fcc8 YS |
4616 | } |
4617 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
4618 | flags |= FIEMAP_EXTENT_ENCODED; | |
0d2b2372 JB |
4619 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
4620 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
1506fcc8 | 4621 | |
1506fcc8 YS |
4622 | free_extent_map(em); |
4623 | em = NULL; | |
ec29ed5b CM |
4624 | if ((em_start >= last) || em_len == (u64)-1 || |
4625 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
4626 | flags |= FIEMAP_EXTENT_LAST; |
4627 | end = 1; | |
4628 | } | |
4629 | ||
ec29ed5b CM |
4630 | /* now scan forward to see if this is really the last extent. */ |
4631 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
4632 | get_extent); | |
4633 | if (IS_ERR(em)) { | |
4634 | ret = PTR_ERR(em); | |
4635 | goto out; | |
4636 | } | |
4637 | if (!em) { | |
975f84fe JB |
4638 | flags |= FIEMAP_EXTENT_LAST; |
4639 | end = 1; | |
4640 | } | |
ec29ed5b CM |
4641 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, |
4642 | em_len, flags); | |
26e726af CS |
4643 | if (ret) { |
4644 | if (ret == 1) | |
4645 | ret = 0; | |
ec29ed5b | 4646 | goto out_free; |
26e726af | 4647 | } |
1506fcc8 YS |
4648 | } |
4649 | out_free: | |
4650 | free_extent_map(em); | |
4651 | out: | |
fe09e16c | 4652 | btrfs_free_path(path); |
a52f4cd2 | 4653 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
2ac55d41 | 4654 | &cached_state, GFP_NOFS); |
1506fcc8 YS |
4655 | return ret; |
4656 | } | |
4657 | ||
727011e0 CM |
4658 | static void __free_extent_buffer(struct extent_buffer *eb) |
4659 | { | |
6d49ba1b | 4660 | btrfs_leak_debug_del(&eb->leak_list); |
727011e0 CM |
4661 | kmem_cache_free(extent_buffer_cache, eb); |
4662 | } | |
4663 | ||
a26e8c9f | 4664 | int extent_buffer_under_io(struct extent_buffer *eb) |
db7f3436 JB |
4665 | { |
4666 | return (atomic_read(&eb->io_pages) || | |
4667 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4668 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4669 | } | |
4670 | ||
4671 | /* | |
4672 | * Helper for releasing extent buffer page. | |
4673 | */ | |
a50924e3 | 4674 | static void btrfs_release_extent_buffer_page(struct extent_buffer *eb) |
db7f3436 JB |
4675 | { |
4676 | unsigned long index; | |
db7f3436 JB |
4677 | struct page *page; |
4678 | int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4679 | ||
4680 | BUG_ON(extent_buffer_under_io(eb)); | |
4681 | ||
a50924e3 DS |
4682 | index = num_extent_pages(eb->start, eb->len); |
4683 | if (index == 0) | |
db7f3436 JB |
4684 | return; |
4685 | ||
4686 | do { | |
4687 | index--; | |
fb85fc9a | 4688 | page = eb->pages[index]; |
5d2361db FL |
4689 | if (!page) |
4690 | continue; | |
4691 | if (mapped) | |
db7f3436 | 4692 | spin_lock(&page->mapping->private_lock); |
5d2361db FL |
4693 | /* |
4694 | * We do this since we'll remove the pages after we've | |
4695 | * removed the eb from the radix tree, so we could race | |
4696 | * and have this page now attached to the new eb. So | |
4697 | * only clear page_private if it's still connected to | |
4698 | * this eb. | |
4699 | */ | |
4700 | if (PagePrivate(page) && | |
4701 | page->private == (unsigned long)eb) { | |
4702 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4703 | BUG_ON(PageDirty(page)); | |
4704 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 4705 | /* |
5d2361db FL |
4706 | * We need to make sure we haven't be attached |
4707 | * to a new eb. | |
db7f3436 | 4708 | */ |
5d2361db FL |
4709 | ClearPagePrivate(page); |
4710 | set_page_private(page, 0); | |
4711 | /* One for the page private */ | |
db7f3436 JB |
4712 | page_cache_release(page); |
4713 | } | |
5d2361db FL |
4714 | |
4715 | if (mapped) | |
4716 | spin_unlock(&page->mapping->private_lock); | |
4717 | ||
4718 | /* One for when we alloced the page */ | |
4719 | page_cache_release(page); | |
a50924e3 | 4720 | } while (index != 0); |
db7f3436 JB |
4721 | } |
4722 | ||
4723 | /* | |
4724 | * Helper for releasing the extent buffer. | |
4725 | */ | |
4726 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4727 | { | |
a50924e3 | 4728 | btrfs_release_extent_buffer_page(eb); |
db7f3436 JB |
4729 | __free_extent_buffer(eb); |
4730 | } | |
4731 | ||
f28491e0 JB |
4732 | static struct extent_buffer * |
4733 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 4734 | unsigned long len) |
d1310b2e CM |
4735 | { |
4736 | struct extent_buffer *eb = NULL; | |
4737 | ||
d1b5c567 | 4738 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
4739 | eb->start = start; |
4740 | eb->len = len; | |
f28491e0 | 4741 | eb->fs_info = fs_info; |
815a51c7 | 4742 | eb->bflags = 0; |
bd681513 CM |
4743 | rwlock_init(&eb->lock); |
4744 | atomic_set(&eb->write_locks, 0); | |
4745 | atomic_set(&eb->read_locks, 0); | |
4746 | atomic_set(&eb->blocking_readers, 0); | |
4747 | atomic_set(&eb->blocking_writers, 0); | |
4748 | atomic_set(&eb->spinning_readers, 0); | |
4749 | atomic_set(&eb->spinning_writers, 0); | |
5b25f70f | 4750 | eb->lock_nested = 0; |
bd681513 CM |
4751 | init_waitqueue_head(&eb->write_lock_wq); |
4752 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 4753 | |
6d49ba1b ES |
4754 | btrfs_leak_debug_add(&eb->leak_list, &buffers); |
4755 | ||
3083ee2e | 4756 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 4757 | atomic_set(&eb->refs, 1); |
0b32f4bb | 4758 | atomic_set(&eb->io_pages, 0); |
727011e0 | 4759 | |
b8dae313 DS |
4760 | /* |
4761 | * Sanity checks, currently the maximum is 64k covered by 16x 4k pages | |
4762 | */ | |
4763 | BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE | |
4764 | > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
4765 | BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
d1310b2e CM |
4766 | |
4767 | return eb; | |
4768 | } | |
4769 | ||
815a51c7 JS |
4770 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) |
4771 | { | |
4772 | unsigned long i; | |
4773 | struct page *p; | |
4774 | struct extent_buffer *new; | |
4775 | unsigned long num_pages = num_extent_pages(src->start, src->len); | |
4776 | ||
3f556f78 | 4777 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
4778 | if (new == NULL) |
4779 | return NULL; | |
4780 | ||
4781 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4782 | p = alloc_page(GFP_NOFS); |
db7f3436 JB |
4783 | if (!p) { |
4784 | btrfs_release_extent_buffer(new); | |
4785 | return NULL; | |
4786 | } | |
815a51c7 JS |
4787 | attach_extent_buffer_page(new, p); |
4788 | WARN_ON(PageDirty(p)); | |
4789 | SetPageUptodate(p); | |
4790 | new->pages[i] = p; | |
4791 | } | |
4792 | ||
4793 | copy_extent_buffer(new, src, 0, 0, src->len); | |
4794 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); | |
4795 | set_bit(EXTENT_BUFFER_DUMMY, &new->bflags); | |
4796 | ||
4797 | return new; | |
4798 | } | |
4799 | ||
3f556f78 DS |
4800 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
4801 | u64 start) | |
815a51c7 JS |
4802 | { |
4803 | struct extent_buffer *eb; | |
3f556f78 DS |
4804 | unsigned long len; |
4805 | unsigned long num_pages; | |
815a51c7 JS |
4806 | unsigned long i; |
4807 | ||
3f556f78 DS |
4808 | if (!fs_info) { |
4809 | /* | |
4810 | * Called only from tests that don't always have a fs_info | |
4811 | * available, but we know that nodesize is 4096 | |
4812 | */ | |
4813 | len = 4096; | |
4814 | } else { | |
4815 | len = fs_info->tree_root->nodesize; | |
4816 | } | |
4817 | num_pages = num_extent_pages(0, len); | |
4818 | ||
4819 | eb = __alloc_extent_buffer(fs_info, start, len); | |
815a51c7 JS |
4820 | if (!eb) |
4821 | return NULL; | |
4822 | ||
4823 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4824 | eb->pages[i] = alloc_page(GFP_NOFS); |
815a51c7 JS |
4825 | if (!eb->pages[i]) |
4826 | goto err; | |
4827 | } | |
4828 | set_extent_buffer_uptodate(eb); | |
4829 | btrfs_set_header_nritems(eb, 0); | |
4830 | set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4831 | ||
4832 | return eb; | |
4833 | err: | |
84167d19 SB |
4834 | for (; i > 0; i--) |
4835 | __free_page(eb->pages[i - 1]); | |
815a51c7 JS |
4836 | __free_extent_buffer(eb); |
4837 | return NULL; | |
4838 | } | |
4839 | ||
0b32f4bb JB |
4840 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
4841 | { | |
242e18c7 | 4842 | int refs; |
0b32f4bb JB |
4843 | /* the ref bit is tricky. We have to make sure it is set |
4844 | * if we have the buffer dirty. Otherwise the | |
4845 | * code to free a buffer can end up dropping a dirty | |
4846 | * page | |
4847 | * | |
4848 | * Once the ref bit is set, it won't go away while the | |
4849 | * buffer is dirty or in writeback, and it also won't | |
4850 | * go away while we have the reference count on the | |
4851 | * eb bumped. | |
4852 | * | |
4853 | * We can't just set the ref bit without bumping the | |
4854 | * ref on the eb because free_extent_buffer might | |
4855 | * see the ref bit and try to clear it. If this happens | |
4856 | * free_extent_buffer might end up dropping our original | |
4857 | * ref by mistake and freeing the page before we are able | |
4858 | * to add one more ref. | |
4859 | * | |
4860 | * So bump the ref count first, then set the bit. If someone | |
4861 | * beat us to it, drop the ref we added. | |
4862 | */ | |
242e18c7 CM |
4863 | refs = atomic_read(&eb->refs); |
4864 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
4865 | return; | |
4866 | ||
594831c4 JB |
4867 | spin_lock(&eb->refs_lock); |
4868 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 4869 | atomic_inc(&eb->refs); |
594831c4 | 4870 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
4871 | } |
4872 | ||
2457aec6 MG |
4873 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
4874 | struct page *accessed) | |
5df4235e JB |
4875 | { |
4876 | unsigned long num_pages, i; | |
4877 | ||
0b32f4bb JB |
4878 | check_buffer_tree_ref(eb); |
4879 | ||
5df4235e JB |
4880 | num_pages = num_extent_pages(eb->start, eb->len); |
4881 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a DS |
4882 | struct page *p = eb->pages[i]; |
4883 | ||
2457aec6 MG |
4884 | if (p != accessed) |
4885 | mark_page_accessed(p); | |
5df4235e JB |
4886 | } |
4887 | } | |
4888 | ||
f28491e0 JB |
4889 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
4890 | u64 start) | |
452c75c3 CS |
4891 | { |
4892 | struct extent_buffer *eb; | |
4893 | ||
4894 | rcu_read_lock(); | |
f28491e0 JB |
4895 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
4896 | start >> PAGE_CACHE_SHIFT); | |
452c75c3 CS |
4897 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
4898 | rcu_read_unlock(); | |
062c19e9 FM |
4899 | /* |
4900 | * Lock our eb's refs_lock to avoid races with | |
4901 | * free_extent_buffer. When we get our eb it might be flagged | |
4902 | * with EXTENT_BUFFER_STALE and another task running | |
4903 | * free_extent_buffer might have seen that flag set, | |
4904 | * eb->refs == 2, that the buffer isn't under IO (dirty and | |
4905 | * writeback flags not set) and it's still in the tree (flag | |
4906 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process | |
4907 | * of decrementing the extent buffer's reference count twice. | |
4908 | * So here we could race and increment the eb's reference count, | |
4909 | * clear its stale flag, mark it as dirty and drop our reference | |
4910 | * before the other task finishes executing free_extent_buffer, | |
4911 | * which would later result in an attempt to free an extent | |
4912 | * buffer that is dirty. | |
4913 | */ | |
4914 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
4915 | spin_lock(&eb->refs_lock); | |
4916 | spin_unlock(&eb->refs_lock); | |
4917 | } | |
2457aec6 | 4918 | mark_extent_buffer_accessed(eb, NULL); |
452c75c3 CS |
4919 | return eb; |
4920 | } | |
4921 | rcu_read_unlock(); | |
4922 | ||
4923 | return NULL; | |
4924 | } | |
4925 | ||
faa2dbf0 JB |
4926 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
4927 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
ce3e6984 | 4928 | u64 start) |
faa2dbf0 JB |
4929 | { |
4930 | struct extent_buffer *eb, *exists = NULL; | |
4931 | int ret; | |
4932 | ||
4933 | eb = find_extent_buffer(fs_info, start); | |
4934 | if (eb) | |
4935 | return eb; | |
3f556f78 | 4936 | eb = alloc_dummy_extent_buffer(fs_info, start); |
faa2dbf0 JB |
4937 | if (!eb) |
4938 | return NULL; | |
4939 | eb->fs_info = fs_info; | |
4940 | again: | |
4941 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); | |
4942 | if (ret) | |
4943 | goto free_eb; | |
4944 | spin_lock(&fs_info->buffer_lock); | |
4945 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
4946 | start >> PAGE_CACHE_SHIFT, eb); | |
4947 | spin_unlock(&fs_info->buffer_lock); | |
4948 | radix_tree_preload_end(); | |
4949 | if (ret == -EEXIST) { | |
4950 | exists = find_extent_buffer(fs_info, start); | |
4951 | if (exists) | |
4952 | goto free_eb; | |
4953 | else | |
4954 | goto again; | |
4955 | } | |
4956 | check_buffer_tree_ref(eb); | |
4957 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
4958 | ||
4959 | /* | |
4960 | * We will free dummy extent buffer's if they come into | |
4961 | * free_extent_buffer with a ref count of 2, but if we are using this we | |
4962 | * want the buffers to stay in memory until we're done with them, so | |
4963 | * bump the ref count again. | |
4964 | */ | |
4965 | atomic_inc(&eb->refs); | |
4966 | return eb; | |
4967 | free_eb: | |
4968 | btrfs_release_extent_buffer(eb); | |
4969 | return exists; | |
4970 | } | |
4971 | #endif | |
4972 | ||
f28491e0 | 4973 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
ce3e6984 | 4974 | u64 start) |
d1310b2e | 4975 | { |
ce3e6984 | 4976 | unsigned long len = fs_info->tree_root->nodesize; |
d1310b2e CM |
4977 | unsigned long num_pages = num_extent_pages(start, len); |
4978 | unsigned long i; | |
4979 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
4980 | struct extent_buffer *eb; | |
6af118ce | 4981 | struct extent_buffer *exists = NULL; |
d1310b2e | 4982 | struct page *p; |
f28491e0 | 4983 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
d1310b2e | 4984 | int uptodate = 1; |
19fe0a8b | 4985 | int ret; |
d1310b2e | 4986 | |
f28491e0 | 4987 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 4988 | if (eb) |
6af118ce | 4989 | return eb; |
6af118ce | 4990 | |
23d79d81 | 4991 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 4992 | if (!eb) |
d1310b2e CM |
4993 | return NULL; |
4994 | ||
727011e0 | 4995 | for (i = 0; i < num_pages; i++, index++) { |
d1b5c567 | 4996 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
4804b382 | 4997 | if (!p) |
6af118ce | 4998 | goto free_eb; |
4f2de97a JB |
4999 | |
5000 | spin_lock(&mapping->private_lock); | |
5001 | if (PagePrivate(p)) { | |
5002 | /* | |
5003 | * We could have already allocated an eb for this page | |
5004 | * and attached one so lets see if we can get a ref on | |
5005 | * the existing eb, and if we can we know it's good and | |
5006 | * we can just return that one, else we know we can just | |
5007 | * overwrite page->private. | |
5008 | */ | |
5009 | exists = (struct extent_buffer *)p->private; | |
5010 | if (atomic_inc_not_zero(&exists->refs)) { | |
5011 | spin_unlock(&mapping->private_lock); | |
5012 | unlock_page(p); | |
17de39ac | 5013 | page_cache_release(p); |
2457aec6 | 5014 | mark_extent_buffer_accessed(exists, p); |
4f2de97a JB |
5015 | goto free_eb; |
5016 | } | |
5ca64f45 | 5017 | exists = NULL; |
4f2de97a | 5018 | |
0b32f4bb | 5019 | /* |
4f2de97a JB |
5020 | * Do this so attach doesn't complain and we need to |
5021 | * drop the ref the old guy had. | |
5022 | */ | |
5023 | ClearPagePrivate(p); | |
0b32f4bb | 5024 | WARN_ON(PageDirty(p)); |
4f2de97a | 5025 | page_cache_release(p); |
d1310b2e | 5026 | } |
4f2de97a JB |
5027 | attach_extent_buffer_page(eb, p); |
5028 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 5029 | WARN_ON(PageDirty(p)); |
727011e0 | 5030 | eb->pages[i] = p; |
d1310b2e CM |
5031 | if (!PageUptodate(p)) |
5032 | uptodate = 0; | |
eb14ab8e CM |
5033 | |
5034 | /* | |
5035 | * see below about how we avoid a nasty race with release page | |
5036 | * and why we unlock later | |
5037 | */ | |
d1310b2e CM |
5038 | } |
5039 | if (uptodate) | |
b4ce94de | 5040 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 5041 | again: |
19fe0a8b MX |
5042 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); |
5043 | if (ret) | |
5044 | goto free_eb; | |
5045 | ||
f28491e0 JB |
5046 | spin_lock(&fs_info->buffer_lock); |
5047 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
5048 | start >> PAGE_CACHE_SHIFT, eb); | |
5049 | spin_unlock(&fs_info->buffer_lock); | |
452c75c3 | 5050 | radix_tree_preload_end(); |
19fe0a8b | 5051 | if (ret == -EEXIST) { |
f28491e0 | 5052 | exists = find_extent_buffer(fs_info, start); |
452c75c3 CS |
5053 | if (exists) |
5054 | goto free_eb; | |
5055 | else | |
115391d2 | 5056 | goto again; |
6af118ce | 5057 | } |
6af118ce | 5058 | /* add one reference for the tree */ |
0b32f4bb | 5059 | check_buffer_tree_ref(eb); |
34b41ace | 5060 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
5061 | |
5062 | /* | |
5063 | * there is a race where release page may have | |
5064 | * tried to find this extent buffer in the radix | |
5065 | * but failed. It will tell the VM it is safe to | |
5066 | * reclaim the, and it will clear the page private bit. | |
5067 | * We must make sure to set the page private bit properly | |
5068 | * after the extent buffer is in the radix tree so | |
5069 | * it doesn't get lost | |
5070 | */ | |
727011e0 CM |
5071 | SetPageChecked(eb->pages[0]); |
5072 | for (i = 1; i < num_pages; i++) { | |
fb85fc9a | 5073 | p = eb->pages[i]; |
727011e0 CM |
5074 | ClearPageChecked(p); |
5075 | unlock_page(p); | |
5076 | } | |
5077 | unlock_page(eb->pages[0]); | |
d1310b2e CM |
5078 | return eb; |
5079 | ||
6af118ce | 5080 | free_eb: |
5ca64f45 | 5081 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
5082 | for (i = 0; i < num_pages; i++) { |
5083 | if (eb->pages[i]) | |
5084 | unlock_page(eb->pages[i]); | |
5085 | } | |
eb14ab8e | 5086 | |
897ca6e9 | 5087 | btrfs_release_extent_buffer(eb); |
6af118ce | 5088 | return exists; |
d1310b2e | 5089 | } |
d1310b2e | 5090 | |
3083ee2e JB |
5091 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
5092 | { | |
5093 | struct extent_buffer *eb = | |
5094 | container_of(head, struct extent_buffer, rcu_head); | |
5095 | ||
5096 | __free_extent_buffer(eb); | |
5097 | } | |
5098 | ||
3083ee2e | 5099 | /* Expects to have eb->eb_lock already held */ |
f7a52a40 | 5100 | static int release_extent_buffer(struct extent_buffer *eb) |
3083ee2e JB |
5101 | { |
5102 | WARN_ON(atomic_read(&eb->refs) == 0); | |
5103 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 5104 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 5105 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 5106 | |
815a51c7 | 5107 | spin_unlock(&eb->refs_lock); |
3083ee2e | 5108 | |
f28491e0 JB |
5109 | spin_lock(&fs_info->buffer_lock); |
5110 | radix_tree_delete(&fs_info->buffer_radix, | |
815a51c7 | 5111 | eb->start >> PAGE_CACHE_SHIFT); |
f28491e0 | 5112 | spin_unlock(&fs_info->buffer_lock); |
34b41ace JB |
5113 | } else { |
5114 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 5115 | } |
3083ee2e JB |
5116 | |
5117 | /* Should be safe to release our pages at this point */ | |
a50924e3 | 5118 | btrfs_release_extent_buffer_page(eb); |
bcb7e449 JB |
5119 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
5120 | if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) { | |
5121 | __free_extent_buffer(eb); | |
5122 | return 1; | |
5123 | } | |
5124 | #endif | |
3083ee2e | 5125 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 5126 | return 1; |
3083ee2e JB |
5127 | } |
5128 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
5129 | |
5130 | return 0; | |
3083ee2e JB |
5131 | } |
5132 | ||
d1310b2e CM |
5133 | void free_extent_buffer(struct extent_buffer *eb) |
5134 | { | |
242e18c7 CM |
5135 | int refs; |
5136 | int old; | |
d1310b2e CM |
5137 | if (!eb) |
5138 | return; | |
5139 | ||
242e18c7 CM |
5140 | while (1) { |
5141 | refs = atomic_read(&eb->refs); | |
5142 | if (refs <= 3) | |
5143 | break; | |
5144 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
5145 | if (old == refs) | |
5146 | return; | |
5147 | } | |
5148 | ||
3083ee2e | 5149 | spin_lock(&eb->refs_lock); |
815a51c7 JS |
5150 | if (atomic_read(&eb->refs) == 2 && |
5151 | test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) | |
5152 | atomic_dec(&eb->refs); | |
5153 | ||
3083ee2e JB |
5154 | if (atomic_read(&eb->refs) == 2 && |
5155 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 5156 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
5157 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5158 | atomic_dec(&eb->refs); | |
5159 | ||
5160 | /* | |
5161 | * I know this is terrible, but it's temporary until we stop tracking | |
5162 | * the uptodate bits and such for the extent buffers. | |
5163 | */ | |
f7a52a40 | 5164 | release_extent_buffer(eb); |
3083ee2e JB |
5165 | } |
5166 | ||
5167 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
5168 | { | |
5169 | if (!eb) | |
d1310b2e CM |
5170 | return; |
5171 | ||
3083ee2e JB |
5172 | spin_lock(&eb->refs_lock); |
5173 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
5174 | ||
0b32f4bb | 5175 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
5176 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5177 | atomic_dec(&eb->refs); | |
f7a52a40 | 5178 | release_extent_buffer(eb); |
d1310b2e | 5179 | } |
d1310b2e | 5180 | |
1d4284bd | 5181 | void clear_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5182 | { |
d1310b2e CM |
5183 | unsigned long i; |
5184 | unsigned long num_pages; | |
5185 | struct page *page; | |
5186 | ||
d1310b2e CM |
5187 | num_pages = num_extent_pages(eb->start, eb->len); |
5188 | ||
5189 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 5190 | page = eb->pages[i]; |
b9473439 | 5191 | if (!PageDirty(page)) |
d2c3f4f6 CM |
5192 | continue; |
5193 | ||
a61e6f29 | 5194 | lock_page(page); |
eb14ab8e CM |
5195 | WARN_ON(!PagePrivate(page)); |
5196 | ||
d1310b2e | 5197 | clear_page_dirty_for_io(page); |
0ee0fda0 | 5198 | spin_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
5199 | if (!PageDirty(page)) { |
5200 | radix_tree_tag_clear(&page->mapping->page_tree, | |
5201 | page_index(page), | |
5202 | PAGECACHE_TAG_DIRTY); | |
5203 | } | |
0ee0fda0 | 5204 | spin_unlock_irq(&page->mapping->tree_lock); |
bf0da8c1 | 5205 | ClearPageError(page); |
a61e6f29 | 5206 | unlock_page(page); |
d1310b2e | 5207 | } |
0b32f4bb | 5208 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 5209 | } |
d1310b2e | 5210 | |
0b32f4bb | 5211 | int set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e CM |
5212 | { |
5213 | unsigned long i; | |
5214 | unsigned long num_pages; | |
b9473439 | 5215 | int was_dirty = 0; |
d1310b2e | 5216 | |
0b32f4bb JB |
5217 | check_buffer_tree_ref(eb); |
5218 | ||
b9473439 | 5219 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 5220 | |
d1310b2e | 5221 | num_pages = num_extent_pages(eb->start, eb->len); |
3083ee2e | 5222 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
5223 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
5224 | ||
b9473439 | 5225 | for (i = 0; i < num_pages; i++) |
fb85fc9a | 5226 | set_page_dirty(eb->pages[i]); |
b9473439 | 5227 | return was_dirty; |
d1310b2e | 5228 | } |
d1310b2e | 5229 | |
0b32f4bb | 5230 | int clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 CM |
5231 | { |
5232 | unsigned long i; | |
5233 | struct page *page; | |
5234 | unsigned long num_pages; | |
5235 | ||
b4ce94de | 5236 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
0b32f4bb | 5237 | num_pages = num_extent_pages(eb->start, eb->len); |
1259ab75 | 5238 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5239 | page = eb->pages[i]; |
33958dc6 CM |
5240 | if (page) |
5241 | ClearPageUptodate(page); | |
1259ab75 CM |
5242 | } |
5243 | return 0; | |
5244 | } | |
5245 | ||
09c25a8c | 5246 | void set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e CM |
5247 | { |
5248 | unsigned long i; | |
5249 | struct page *page; | |
5250 | unsigned long num_pages; | |
5251 | ||
0b32f4bb | 5252 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 5253 | num_pages = num_extent_pages(eb->start, eb->len); |
d1310b2e | 5254 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5255 | page = eb->pages[i]; |
d1310b2e CM |
5256 | SetPageUptodate(page); |
5257 | } | |
d1310b2e | 5258 | } |
d1310b2e | 5259 | |
0b32f4bb | 5260 | int extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 5261 | { |
0b32f4bb | 5262 | return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 5263 | } |
d1310b2e CM |
5264 | |
5265 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
bb82ab88 | 5266 | struct extent_buffer *eb, u64 start, int wait, |
f188591e | 5267 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
5268 | { |
5269 | unsigned long i; | |
5270 | unsigned long start_i; | |
5271 | struct page *page; | |
5272 | int err; | |
5273 | int ret = 0; | |
ce9adaa5 CM |
5274 | int locked_pages = 0; |
5275 | int all_uptodate = 1; | |
d1310b2e | 5276 | unsigned long num_pages; |
727011e0 | 5277 | unsigned long num_reads = 0; |
a86c12c7 | 5278 | struct bio *bio = NULL; |
c8b97818 | 5279 | unsigned long bio_flags = 0; |
a86c12c7 | 5280 | |
b4ce94de | 5281 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
5282 | return 0; |
5283 | ||
d1310b2e CM |
5284 | if (start) { |
5285 | WARN_ON(start < eb->start); | |
5286 | start_i = (start >> PAGE_CACHE_SHIFT) - | |
5287 | (eb->start >> PAGE_CACHE_SHIFT); | |
5288 | } else { | |
5289 | start_i = 0; | |
5290 | } | |
5291 | ||
5292 | num_pages = num_extent_pages(eb->start, eb->len); | |
5293 | for (i = start_i; i < num_pages; i++) { | |
fb85fc9a | 5294 | page = eb->pages[i]; |
bb82ab88 | 5295 | if (wait == WAIT_NONE) { |
2db04966 | 5296 | if (!trylock_page(page)) |
ce9adaa5 | 5297 | goto unlock_exit; |
d1310b2e CM |
5298 | } else { |
5299 | lock_page(page); | |
5300 | } | |
ce9adaa5 | 5301 | locked_pages++; |
727011e0 CM |
5302 | if (!PageUptodate(page)) { |
5303 | num_reads++; | |
ce9adaa5 | 5304 | all_uptodate = 0; |
727011e0 | 5305 | } |
ce9adaa5 CM |
5306 | } |
5307 | if (all_uptodate) { | |
5308 | if (start_i == 0) | |
b4ce94de | 5309 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
5310 | goto unlock_exit; |
5311 | } | |
5312 | ||
656f30db | 5313 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 5314 | eb->read_mirror = 0; |
0b32f4bb | 5315 | atomic_set(&eb->io_pages, num_reads); |
ce9adaa5 | 5316 | for (i = start_i; i < num_pages; i++) { |
fb85fc9a | 5317 | page = eb->pages[i]; |
ce9adaa5 | 5318 | if (!PageUptodate(page)) { |
f188591e | 5319 | ClearPageError(page); |
a86c12c7 | 5320 | err = __extent_read_full_page(tree, page, |
f188591e | 5321 | get_extent, &bio, |
d4c7ca86 JB |
5322 | mirror_num, &bio_flags, |
5323 | READ | REQ_META); | |
d397712b | 5324 | if (err) |
d1310b2e | 5325 | ret = err; |
d1310b2e CM |
5326 | } else { |
5327 | unlock_page(page); | |
5328 | } | |
5329 | } | |
5330 | ||
355808c2 | 5331 | if (bio) { |
d4c7ca86 JB |
5332 | err = submit_one_bio(READ | REQ_META, bio, mirror_num, |
5333 | bio_flags); | |
79787eaa JM |
5334 | if (err) |
5335 | return err; | |
355808c2 | 5336 | } |
a86c12c7 | 5337 | |
bb82ab88 | 5338 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 5339 | return ret; |
d397712b | 5340 | |
d1310b2e | 5341 | for (i = start_i; i < num_pages; i++) { |
fb85fc9a | 5342 | page = eb->pages[i]; |
d1310b2e | 5343 | wait_on_page_locked(page); |
d397712b | 5344 | if (!PageUptodate(page)) |
d1310b2e | 5345 | ret = -EIO; |
d1310b2e | 5346 | } |
d397712b | 5347 | |
d1310b2e | 5348 | return ret; |
ce9adaa5 CM |
5349 | |
5350 | unlock_exit: | |
5351 | i = start_i; | |
d397712b | 5352 | while (locked_pages > 0) { |
fb85fc9a | 5353 | page = eb->pages[i]; |
ce9adaa5 CM |
5354 | i++; |
5355 | unlock_page(page); | |
5356 | locked_pages--; | |
5357 | } | |
5358 | return ret; | |
d1310b2e | 5359 | } |
d1310b2e CM |
5360 | |
5361 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
5362 | unsigned long start, | |
5363 | unsigned long len) | |
5364 | { | |
5365 | size_t cur; | |
5366 | size_t offset; | |
5367 | struct page *page; | |
5368 | char *kaddr; | |
5369 | char *dst = (char *)dstv; | |
5370 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5371 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
5372 | |
5373 | WARN_ON(start > eb->len); | |
5374 | WARN_ON(start + len > eb->start + eb->len); | |
5375 | ||
778746b5 | 5376 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5377 | |
d397712b | 5378 | while (len > 0) { |
fb85fc9a | 5379 | page = eb->pages[i]; |
d1310b2e CM |
5380 | |
5381 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
a6591715 | 5382 | kaddr = page_address(page); |
d1310b2e | 5383 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
5384 | |
5385 | dst += cur; | |
5386 | len -= cur; | |
5387 | offset = 0; | |
5388 | i++; | |
5389 | } | |
5390 | } | |
d1310b2e | 5391 | |
550ac1d8 GH |
5392 | int read_extent_buffer_to_user(struct extent_buffer *eb, void __user *dstv, |
5393 | unsigned long start, | |
5394 | unsigned long len) | |
5395 | { | |
5396 | size_t cur; | |
5397 | size_t offset; | |
5398 | struct page *page; | |
5399 | char *kaddr; | |
5400 | char __user *dst = (char __user *)dstv; | |
5401 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5402 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5403 | int ret = 0; | |
5404 | ||
5405 | WARN_ON(start > eb->len); | |
5406 | WARN_ON(start + len > eb->start + eb->len); | |
5407 | ||
5408 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); | |
5409 | ||
5410 | while (len > 0) { | |
fb85fc9a | 5411 | page = eb->pages[i]; |
550ac1d8 GH |
5412 | |
5413 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
5414 | kaddr = page_address(page); | |
5415 | if (copy_to_user(dst, kaddr + offset, cur)) { | |
5416 | ret = -EFAULT; | |
5417 | break; | |
5418 | } | |
5419 | ||
5420 | dst += cur; | |
5421 | len -= cur; | |
5422 | offset = 0; | |
5423 | i++; | |
5424 | } | |
5425 | ||
5426 | return ret; | |
5427 | } | |
5428 | ||
d1310b2e | 5429 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, |
a6591715 | 5430 | unsigned long min_len, char **map, |
d1310b2e | 5431 | unsigned long *map_start, |
a6591715 | 5432 | unsigned long *map_len) |
d1310b2e CM |
5433 | { |
5434 | size_t offset = start & (PAGE_CACHE_SIZE - 1); | |
5435 | char *kaddr; | |
5436 | struct page *p; | |
5437 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5438 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5439 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
5440 | PAGE_CACHE_SHIFT; | |
5441 | ||
5442 | if (i != end_i) | |
5443 | return -EINVAL; | |
5444 | ||
5445 | if (i == 0) { | |
5446 | offset = start_offset; | |
5447 | *map_start = 0; | |
5448 | } else { | |
5449 | offset = 0; | |
5450 | *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; | |
5451 | } | |
d397712b | 5452 | |
d1310b2e | 5453 | if (start + min_len > eb->len) { |
31b1a2bd | 5454 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, " |
c1c9ff7c GU |
5455 | "wanted %lu %lu\n", |
5456 | eb->start, eb->len, start, min_len); | |
85026533 | 5457 | return -EINVAL; |
d1310b2e CM |
5458 | } |
5459 | ||
fb85fc9a | 5460 | p = eb->pages[i]; |
a6591715 | 5461 | kaddr = page_address(p); |
d1310b2e CM |
5462 | *map = kaddr + offset; |
5463 | *map_len = PAGE_CACHE_SIZE - offset; | |
5464 | return 0; | |
5465 | } | |
d1310b2e | 5466 | |
d1310b2e CM |
5467 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, |
5468 | unsigned long start, | |
5469 | unsigned long len) | |
5470 | { | |
5471 | size_t cur; | |
5472 | size_t offset; | |
5473 | struct page *page; | |
5474 | char *kaddr; | |
5475 | char *ptr = (char *)ptrv; | |
5476 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5477 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5478 | int ret = 0; | |
5479 | ||
5480 | WARN_ON(start > eb->len); | |
5481 | WARN_ON(start + len > eb->start + eb->len); | |
5482 | ||
778746b5 | 5483 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5484 | |
d397712b | 5485 | while (len > 0) { |
fb85fc9a | 5486 | page = eb->pages[i]; |
d1310b2e CM |
5487 | |
5488 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
5489 | ||
a6591715 | 5490 | kaddr = page_address(page); |
d1310b2e | 5491 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
5492 | if (ret) |
5493 | break; | |
5494 | ||
5495 | ptr += cur; | |
5496 | len -= cur; | |
5497 | offset = 0; | |
5498 | i++; | |
5499 | } | |
5500 | return ret; | |
5501 | } | |
d1310b2e CM |
5502 | |
5503 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
5504 | unsigned long start, unsigned long len) | |
5505 | { | |
5506 | size_t cur; | |
5507 | size_t offset; | |
5508 | struct page *page; | |
5509 | char *kaddr; | |
5510 | char *src = (char *)srcv; | |
5511 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5512 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5513 | ||
5514 | WARN_ON(start > eb->len); | |
5515 | WARN_ON(start + len > eb->start + eb->len); | |
5516 | ||
778746b5 | 5517 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5518 | |
d397712b | 5519 | while (len > 0) { |
fb85fc9a | 5520 | page = eb->pages[i]; |
d1310b2e CM |
5521 | WARN_ON(!PageUptodate(page)); |
5522 | ||
5523 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 5524 | kaddr = page_address(page); |
d1310b2e | 5525 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
5526 | |
5527 | src += cur; | |
5528 | len -= cur; | |
5529 | offset = 0; | |
5530 | i++; | |
5531 | } | |
5532 | } | |
d1310b2e CM |
5533 | |
5534 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
5535 | unsigned long start, unsigned long len) | |
5536 | { | |
5537 | size_t cur; | |
5538 | size_t offset; | |
5539 | struct page *page; | |
5540 | char *kaddr; | |
5541 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5542 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5543 | ||
5544 | WARN_ON(start > eb->len); | |
5545 | WARN_ON(start + len > eb->start + eb->len); | |
5546 | ||
778746b5 | 5547 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5548 | |
d397712b | 5549 | while (len > 0) { |
fb85fc9a | 5550 | page = eb->pages[i]; |
d1310b2e CM |
5551 | WARN_ON(!PageUptodate(page)); |
5552 | ||
5553 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 5554 | kaddr = page_address(page); |
d1310b2e | 5555 | memset(kaddr + offset, c, cur); |
d1310b2e CM |
5556 | |
5557 | len -= cur; | |
5558 | offset = 0; | |
5559 | i++; | |
5560 | } | |
5561 | } | |
d1310b2e CM |
5562 | |
5563 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
5564 | unsigned long dst_offset, unsigned long src_offset, | |
5565 | unsigned long len) | |
5566 | { | |
5567 | u64 dst_len = dst->len; | |
5568 | size_t cur; | |
5569 | size_t offset; | |
5570 | struct page *page; | |
5571 | char *kaddr; | |
5572 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5573 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
5574 | ||
5575 | WARN_ON(src->len != dst_len); | |
5576 | ||
5577 | offset = (start_offset + dst_offset) & | |
778746b5 | 5578 | (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5579 | |
d397712b | 5580 | while (len > 0) { |
fb85fc9a | 5581 | page = dst->pages[i]; |
d1310b2e CM |
5582 | WARN_ON(!PageUptodate(page)); |
5583 | ||
5584 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
5585 | ||
a6591715 | 5586 | kaddr = page_address(page); |
d1310b2e | 5587 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
5588 | |
5589 | src_offset += cur; | |
5590 | len -= cur; | |
5591 | offset = 0; | |
5592 | i++; | |
5593 | } | |
5594 | } | |
d1310b2e | 5595 | |
3387206f ST |
5596 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
5597 | { | |
5598 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
5599 | return distance < len; | |
5600 | } | |
5601 | ||
d1310b2e CM |
5602 | static void copy_pages(struct page *dst_page, struct page *src_page, |
5603 | unsigned long dst_off, unsigned long src_off, | |
5604 | unsigned long len) | |
5605 | { | |
a6591715 | 5606 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 5607 | char *src_kaddr; |
727011e0 | 5608 | int must_memmove = 0; |
d1310b2e | 5609 | |
3387206f | 5610 | if (dst_page != src_page) { |
a6591715 | 5611 | src_kaddr = page_address(src_page); |
3387206f | 5612 | } else { |
d1310b2e | 5613 | src_kaddr = dst_kaddr; |
727011e0 CM |
5614 | if (areas_overlap(src_off, dst_off, len)) |
5615 | must_memmove = 1; | |
3387206f | 5616 | } |
d1310b2e | 5617 | |
727011e0 CM |
5618 | if (must_memmove) |
5619 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
5620 | else | |
5621 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
5622 | } |
5623 | ||
5624 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5625 | unsigned long src_offset, unsigned long len) | |
5626 | { | |
5627 | size_t cur; | |
5628 | size_t dst_off_in_page; | |
5629 | size_t src_off_in_page; | |
5630 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5631 | unsigned long dst_i; | |
5632 | unsigned long src_i; | |
5633 | ||
5634 | if (src_offset + len > dst->len) { | |
f14d104d DS |
5635 | btrfs_err(dst->fs_info, |
5636 | "memmove bogus src_offset %lu move " | |
5637 | "len %lu dst len %lu", src_offset, len, dst->len); | |
d1310b2e CM |
5638 | BUG_ON(1); |
5639 | } | |
5640 | if (dst_offset + len > dst->len) { | |
f14d104d DS |
5641 | btrfs_err(dst->fs_info, |
5642 | "memmove bogus dst_offset %lu move " | |
5643 | "len %lu dst len %lu", dst_offset, len, dst->len); | |
d1310b2e CM |
5644 | BUG_ON(1); |
5645 | } | |
5646 | ||
d397712b | 5647 | while (len > 0) { |
d1310b2e | 5648 | dst_off_in_page = (start_offset + dst_offset) & |
778746b5 | 5649 | (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5650 | src_off_in_page = (start_offset + src_offset) & |
778746b5 | 5651 | (PAGE_CACHE_SIZE - 1); |
d1310b2e CM |
5652 | |
5653 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
5654 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
5655 | ||
5656 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
5657 | src_off_in_page)); | |
5658 | cur = min_t(unsigned long, cur, | |
5659 | (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); | |
5660 | ||
fb85fc9a | 5661 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5662 | dst_off_in_page, src_off_in_page, cur); |
5663 | ||
5664 | src_offset += cur; | |
5665 | dst_offset += cur; | |
5666 | len -= cur; | |
5667 | } | |
5668 | } | |
d1310b2e CM |
5669 | |
5670 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5671 | unsigned long src_offset, unsigned long len) | |
5672 | { | |
5673 | size_t cur; | |
5674 | size_t dst_off_in_page; | |
5675 | size_t src_off_in_page; | |
5676 | unsigned long dst_end = dst_offset + len - 1; | |
5677 | unsigned long src_end = src_offset + len - 1; | |
5678 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5679 | unsigned long dst_i; | |
5680 | unsigned long src_i; | |
5681 | ||
5682 | if (src_offset + len > dst->len) { | |
f14d104d DS |
5683 | btrfs_err(dst->fs_info, "memmove bogus src_offset %lu move " |
5684 | "len %lu len %lu", src_offset, len, dst->len); | |
d1310b2e CM |
5685 | BUG_ON(1); |
5686 | } | |
5687 | if (dst_offset + len > dst->len) { | |
f14d104d DS |
5688 | btrfs_err(dst->fs_info, "memmove bogus dst_offset %lu move " |
5689 | "len %lu len %lu", dst_offset, len, dst->len); | |
d1310b2e CM |
5690 | BUG_ON(1); |
5691 | } | |
727011e0 | 5692 | if (dst_offset < src_offset) { |
d1310b2e CM |
5693 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
5694 | return; | |
5695 | } | |
d397712b | 5696 | while (len > 0) { |
d1310b2e CM |
5697 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; |
5698 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
5699 | ||
5700 | dst_off_in_page = (start_offset + dst_end) & | |
778746b5 | 5701 | (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5702 | src_off_in_page = (start_offset + src_end) & |
778746b5 | 5703 | (PAGE_CACHE_SIZE - 1); |
d1310b2e CM |
5704 | |
5705 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
5706 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 5707 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5708 | dst_off_in_page - cur + 1, |
5709 | src_off_in_page - cur + 1, cur); | |
5710 | ||
5711 | dst_end -= cur; | |
5712 | src_end -= cur; | |
5713 | len -= cur; | |
5714 | } | |
5715 | } | |
6af118ce | 5716 | |
f7a52a40 | 5717 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 5718 | { |
6af118ce | 5719 | struct extent_buffer *eb; |
6af118ce | 5720 | |
3083ee2e JB |
5721 | /* |
5722 | * We need to make sure noboody is attaching this page to an eb right | |
5723 | * now. | |
5724 | */ | |
5725 | spin_lock(&page->mapping->private_lock); | |
5726 | if (!PagePrivate(page)) { | |
5727 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 5728 | return 1; |
45f49bce | 5729 | } |
6af118ce | 5730 | |
3083ee2e JB |
5731 | eb = (struct extent_buffer *)page->private; |
5732 | BUG_ON(!eb); | |
19fe0a8b MX |
5733 | |
5734 | /* | |
3083ee2e JB |
5735 | * This is a little awful but should be ok, we need to make sure that |
5736 | * the eb doesn't disappear out from under us while we're looking at | |
5737 | * this page. | |
19fe0a8b | 5738 | */ |
3083ee2e | 5739 | spin_lock(&eb->refs_lock); |
0b32f4bb | 5740 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
5741 | spin_unlock(&eb->refs_lock); |
5742 | spin_unlock(&page->mapping->private_lock); | |
5743 | return 0; | |
b9473439 | 5744 | } |
3083ee2e | 5745 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 5746 | |
19fe0a8b | 5747 | /* |
3083ee2e JB |
5748 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
5749 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 5750 | */ |
3083ee2e JB |
5751 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
5752 | spin_unlock(&eb->refs_lock); | |
5753 | return 0; | |
b9473439 | 5754 | } |
19fe0a8b | 5755 | |
f7a52a40 | 5756 | return release_extent_buffer(eb); |
6af118ce | 5757 | } |