]>
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; | |
144 | changeset->bytes_changed += state->end - state->start + 1; | |
145 | ret = ulist_add(changeset->range_changed, state->start, state->end, | |
146 | GFP_ATOMIC); | |
147 | /* ENOMEM */ | |
148 | BUG_ON(ret < 0); | |
149 | } | |
150 | ||
0b32f4bb | 151 | static noinline void flush_write_bio(void *data); |
c2d904e0 JM |
152 | static inline struct btrfs_fs_info * |
153 | tree_fs_info(struct extent_io_tree *tree) | |
154 | { | |
a5dee37d JB |
155 | if (!tree->mapping) |
156 | return NULL; | |
c2d904e0 JM |
157 | return btrfs_sb(tree->mapping->host->i_sb); |
158 | } | |
0b32f4bb | 159 | |
d1310b2e CM |
160 | int __init extent_io_init(void) |
161 | { | |
837e1972 | 162 | extent_state_cache = kmem_cache_create("btrfs_extent_state", |
9601e3f6 CH |
163 | sizeof(struct extent_state), 0, |
164 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
165 | if (!extent_state_cache) |
166 | return -ENOMEM; | |
167 | ||
837e1972 | 168 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 CH |
169 | sizeof(struct extent_buffer), 0, |
170 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
171 | if (!extent_buffer_cache) |
172 | goto free_state_cache; | |
9be3395b CM |
173 | |
174 | btrfs_bioset = bioset_create(BIO_POOL_SIZE, | |
175 | offsetof(struct btrfs_io_bio, bio)); | |
176 | if (!btrfs_bioset) | |
177 | goto free_buffer_cache; | |
b208c2f7 DW |
178 | |
179 | if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE)) | |
180 | goto free_bioset; | |
181 | ||
d1310b2e CM |
182 | return 0; |
183 | ||
b208c2f7 DW |
184 | free_bioset: |
185 | bioset_free(btrfs_bioset); | |
186 | btrfs_bioset = NULL; | |
187 | ||
9be3395b CM |
188 | free_buffer_cache: |
189 | kmem_cache_destroy(extent_buffer_cache); | |
190 | extent_buffer_cache = NULL; | |
191 | ||
d1310b2e CM |
192 | free_state_cache: |
193 | kmem_cache_destroy(extent_state_cache); | |
9be3395b | 194 | extent_state_cache = NULL; |
d1310b2e CM |
195 | return -ENOMEM; |
196 | } | |
197 | ||
198 | void extent_io_exit(void) | |
199 | { | |
6d49ba1b | 200 | btrfs_leak_debug_check(); |
8c0a8537 KS |
201 | |
202 | /* | |
203 | * Make sure all delayed rcu free are flushed before we | |
204 | * destroy caches. | |
205 | */ | |
206 | rcu_barrier(); | |
d1310b2e CM |
207 | if (extent_state_cache) |
208 | kmem_cache_destroy(extent_state_cache); | |
209 | if (extent_buffer_cache) | |
210 | kmem_cache_destroy(extent_buffer_cache); | |
9be3395b CM |
211 | if (btrfs_bioset) |
212 | bioset_free(btrfs_bioset); | |
d1310b2e CM |
213 | } |
214 | ||
215 | void extent_io_tree_init(struct extent_io_tree *tree, | |
f993c883 | 216 | struct address_space *mapping) |
d1310b2e | 217 | { |
6bef4d31 | 218 | tree->state = RB_ROOT; |
d1310b2e CM |
219 | tree->ops = NULL; |
220 | tree->dirty_bytes = 0; | |
70dec807 | 221 | spin_lock_init(&tree->lock); |
d1310b2e | 222 | tree->mapping = mapping; |
d1310b2e | 223 | } |
d1310b2e | 224 | |
b2950863 | 225 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
226 | { |
227 | struct extent_state *state; | |
d1310b2e CM |
228 | |
229 | state = kmem_cache_alloc(extent_state_cache, mask); | |
2b114d1d | 230 | if (!state) |
d1310b2e CM |
231 | return state; |
232 | state->state = 0; | |
d1310b2e | 233 | state->private = 0; |
27a3507d | 234 | RB_CLEAR_NODE(&state->rb_node); |
6d49ba1b | 235 | btrfs_leak_debug_add(&state->leak_list, &states); |
d1310b2e CM |
236 | atomic_set(&state->refs, 1); |
237 | init_waitqueue_head(&state->wq); | |
143bede5 | 238 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
239 | return state; |
240 | } | |
d1310b2e | 241 | |
4845e44f | 242 | void free_extent_state(struct extent_state *state) |
d1310b2e | 243 | { |
d1310b2e CM |
244 | if (!state) |
245 | return; | |
246 | if (atomic_dec_and_test(&state->refs)) { | |
27a3507d | 247 | WARN_ON(extent_state_in_tree(state)); |
6d49ba1b | 248 | btrfs_leak_debug_del(&state->leak_list); |
143bede5 | 249 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
250 | kmem_cache_free(extent_state_cache, state); |
251 | } | |
252 | } | |
d1310b2e | 253 | |
f2071b21 FM |
254 | static struct rb_node *tree_insert(struct rb_root *root, |
255 | struct rb_node *search_start, | |
256 | u64 offset, | |
12cfbad9 FDBM |
257 | struct rb_node *node, |
258 | struct rb_node ***p_in, | |
259 | struct rb_node **parent_in) | |
d1310b2e | 260 | { |
f2071b21 | 261 | struct rb_node **p; |
d397712b | 262 | struct rb_node *parent = NULL; |
d1310b2e CM |
263 | struct tree_entry *entry; |
264 | ||
12cfbad9 FDBM |
265 | if (p_in && parent_in) { |
266 | p = *p_in; | |
267 | parent = *parent_in; | |
268 | goto do_insert; | |
269 | } | |
270 | ||
f2071b21 | 271 | p = search_start ? &search_start : &root->rb_node; |
d397712b | 272 | while (*p) { |
d1310b2e CM |
273 | parent = *p; |
274 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
275 | ||
276 | if (offset < entry->start) | |
277 | p = &(*p)->rb_left; | |
278 | else if (offset > entry->end) | |
279 | p = &(*p)->rb_right; | |
280 | else | |
281 | return parent; | |
282 | } | |
283 | ||
12cfbad9 | 284 | do_insert: |
d1310b2e CM |
285 | rb_link_node(node, parent, p); |
286 | rb_insert_color(node, root); | |
287 | return NULL; | |
288 | } | |
289 | ||
80ea96b1 | 290 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
12cfbad9 FDBM |
291 | struct rb_node **prev_ret, |
292 | struct rb_node **next_ret, | |
293 | struct rb_node ***p_ret, | |
294 | struct rb_node **parent_ret) | |
d1310b2e | 295 | { |
80ea96b1 | 296 | struct rb_root *root = &tree->state; |
12cfbad9 | 297 | struct rb_node **n = &root->rb_node; |
d1310b2e CM |
298 | struct rb_node *prev = NULL; |
299 | struct rb_node *orig_prev = NULL; | |
300 | struct tree_entry *entry; | |
301 | struct tree_entry *prev_entry = NULL; | |
302 | ||
12cfbad9 FDBM |
303 | while (*n) { |
304 | prev = *n; | |
305 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
d1310b2e CM |
306 | prev_entry = entry; |
307 | ||
308 | if (offset < entry->start) | |
12cfbad9 | 309 | n = &(*n)->rb_left; |
d1310b2e | 310 | else if (offset > entry->end) |
12cfbad9 | 311 | n = &(*n)->rb_right; |
d397712b | 312 | else |
12cfbad9 | 313 | return *n; |
d1310b2e CM |
314 | } |
315 | ||
12cfbad9 FDBM |
316 | if (p_ret) |
317 | *p_ret = n; | |
318 | if (parent_ret) | |
319 | *parent_ret = prev; | |
320 | ||
d1310b2e CM |
321 | if (prev_ret) { |
322 | orig_prev = prev; | |
d397712b | 323 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
324 | prev = rb_next(prev); |
325 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
326 | } | |
327 | *prev_ret = prev; | |
328 | prev = orig_prev; | |
329 | } | |
330 | ||
331 | if (next_ret) { | |
332 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 333 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
334 | prev = rb_prev(prev); |
335 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
336 | } | |
337 | *next_ret = prev; | |
338 | } | |
339 | return NULL; | |
340 | } | |
341 | ||
12cfbad9 FDBM |
342 | static inline struct rb_node * |
343 | tree_search_for_insert(struct extent_io_tree *tree, | |
344 | u64 offset, | |
345 | struct rb_node ***p_ret, | |
346 | struct rb_node **parent_ret) | |
d1310b2e | 347 | { |
70dec807 | 348 | struct rb_node *prev = NULL; |
d1310b2e | 349 | struct rb_node *ret; |
70dec807 | 350 | |
12cfbad9 | 351 | ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret); |
d397712b | 352 | if (!ret) |
d1310b2e CM |
353 | return prev; |
354 | return ret; | |
355 | } | |
356 | ||
12cfbad9 FDBM |
357 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
358 | u64 offset) | |
359 | { | |
360 | return tree_search_for_insert(tree, offset, NULL, NULL); | |
361 | } | |
362 | ||
9ed74f2d JB |
363 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, |
364 | struct extent_state *other) | |
365 | { | |
366 | if (tree->ops && tree->ops->merge_extent_hook) | |
367 | tree->ops->merge_extent_hook(tree->mapping->host, new, | |
368 | other); | |
369 | } | |
370 | ||
d1310b2e CM |
371 | /* |
372 | * utility function to look for merge candidates inside a given range. | |
373 | * Any extents with matching state are merged together into a single | |
374 | * extent in the tree. Extents with EXTENT_IO in their state field | |
375 | * are not merged because the end_io handlers need to be able to do | |
376 | * operations on them without sleeping (or doing allocations/splits). | |
377 | * | |
378 | * This should be called with the tree lock held. | |
379 | */ | |
1bf85046 JM |
380 | static void merge_state(struct extent_io_tree *tree, |
381 | struct extent_state *state) | |
d1310b2e CM |
382 | { |
383 | struct extent_state *other; | |
384 | struct rb_node *other_node; | |
385 | ||
5b21f2ed | 386 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
1bf85046 | 387 | return; |
d1310b2e CM |
388 | |
389 | other_node = rb_prev(&state->rb_node); | |
390 | if (other_node) { | |
391 | other = rb_entry(other_node, struct extent_state, rb_node); | |
392 | if (other->end == state->start - 1 && | |
393 | other->state == state->state) { | |
9ed74f2d | 394 | merge_cb(tree, state, other); |
d1310b2e | 395 | state->start = other->start; |
d1310b2e | 396 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 397 | RB_CLEAR_NODE(&other->rb_node); |
d1310b2e CM |
398 | free_extent_state(other); |
399 | } | |
400 | } | |
401 | other_node = rb_next(&state->rb_node); | |
402 | if (other_node) { | |
403 | other = rb_entry(other_node, struct extent_state, rb_node); | |
404 | if (other->start == state->end + 1 && | |
405 | other->state == state->state) { | |
9ed74f2d | 406 | merge_cb(tree, state, other); |
df98b6e2 | 407 | state->end = other->end; |
df98b6e2 | 408 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 409 | RB_CLEAR_NODE(&other->rb_node); |
df98b6e2 | 410 | free_extent_state(other); |
d1310b2e CM |
411 | } |
412 | } | |
d1310b2e CM |
413 | } |
414 | ||
1bf85046 | 415 | static void set_state_cb(struct extent_io_tree *tree, |
9ee49a04 | 416 | struct extent_state *state, unsigned *bits) |
291d673e | 417 | { |
1bf85046 JM |
418 | if (tree->ops && tree->ops->set_bit_hook) |
419 | tree->ops->set_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
420 | } |
421 | ||
422 | static void clear_state_cb(struct extent_io_tree *tree, | |
9ee49a04 | 423 | struct extent_state *state, unsigned *bits) |
291d673e | 424 | { |
9ed74f2d JB |
425 | if (tree->ops && tree->ops->clear_bit_hook) |
426 | tree->ops->clear_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
427 | } |
428 | ||
3150b699 | 429 | static void set_state_bits(struct extent_io_tree *tree, |
d38ed27f QW |
430 | struct extent_state *state, unsigned *bits, |
431 | struct extent_changeset *changeset); | |
3150b699 | 432 | |
d1310b2e CM |
433 | /* |
434 | * insert an extent_state struct into the tree. 'bits' are set on the | |
435 | * struct before it is inserted. | |
436 | * | |
437 | * This may return -EEXIST if the extent is already there, in which case the | |
438 | * state struct is freed. | |
439 | * | |
440 | * The tree lock is not taken internally. This is a utility function and | |
441 | * probably isn't what you want to call (see set/clear_extent_bit). | |
442 | */ | |
443 | static int insert_state(struct extent_io_tree *tree, | |
444 | struct extent_state *state, u64 start, u64 end, | |
12cfbad9 FDBM |
445 | struct rb_node ***p, |
446 | struct rb_node **parent, | |
d38ed27f | 447 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e CM |
448 | { |
449 | struct rb_node *node; | |
450 | ||
31b1a2bd | 451 | if (end < start) |
efe120a0 | 452 | WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n", |
c1c9ff7c | 453 | end, start); |
d1310b2e CM |
454 | state->start = start; |
455 | state->end = end; | |
9ed74f2d | 456 | |
d38ed27f | 457 | set_state_bits(tree, state, bits, changeset); |
3150b699 | 458 | |
f2071b21 | 459 | node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent); |
d1310b2e CM |
460 | if (node) { |
461 | struct extent_state *found; | |
462 | found = rb_entry(node, struct extent_state, rb_node); | |
efe120a0 | 463 | printk(KERN_ERR "BTRFS: found node %llu %llu on insert of " |
c1c9ff7c GU |
464 | "%llu %llu\n", |
465 | found->start, found->end, start, end); | |
d1310b2e CM |
466 | return -EEXIST; |
467 | } | |
468 | merge_state(tree, state); | |
469 | return 0; | |
470 | } | |
471 | ||
1bf85046 | 472 | static void split_cb(struct extent_io_tree *tree, struct extent_state *orig, |
9ed74f2d JB |
473 | u64 split) |
474 | { | |
475 | if (tree->ops && tree->ops->split_extent_hook) | |
1bf85046 | 476 | tree->ops->split_extent_hook(tree->mapping->host, orig, split); |
9ed74f2d JB |
477 | } |
478 | ||
d1310b2e CM |
479 | /* |
480 | * split a given extent state struct in two, inserting the preallocated | |
481 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
482 | * offset inside 'orig' where it should be split. | |
483 | * | |
484 | * Before calling, | |
485 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
486 | * are two extent state structs in the tree: | |
487 | * prealloc: [orig->start, split - 1] | |
488 | * orig: [ split, orig->end ] | |
489 | * | |
490 | * The tree locks are not taken by this function. They need to be held | |
491 | * by the caller. | |
492 | */ | |
493 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
494 | struct extent_state *prealloc, u64 split) | |
495 | { | |
496 | struct rb_node *node; | |
9ed74f2d JB |
497 | |
498 | split_cb(tree, orig, split); | |
499 | ||
d1310b2e CM |
500 | prealloc->start = orig->start; |
501 | prealloc->end = split - 1; | |
502 | prealloc->state = orig->state; | |
503 | orig->start = split; | |
504 | ||
f2071b21 FM |
505 | node = tree_insert(&tree->state, &orig->rb_node, prealloc->end, |
506 | &prealloc->rb_node, NULL, NULL); | |
d1310b2e | 507 | if (node) { |
d1310b2e CM |
508 | free_extent_state(prealloc); |
509 | return -EEXIST; | |
510 | } | |
511 | return 0; | |
512 | } | |
513 | ||
cdc6a395 LZ |
514 | static struct extent_state *next_state(struct extent_state *state) |
515 | { | |
516 | struct rb_node *next = rb_next(&state->rb_node); | |
517 | if (next) | |
518 | return rb_entry(next, struct extent_state, rb_node); | |
519 | else | |
520 | return NULL; | |
521 | } | |
522 | ||
d1310b2e CM |
523 | /* |
524 | * utility function to clear some bits in an extent state struct. | |
1b303fc0 | 525 | * it will optionally wake up any one waiting on this state (wake == 1). |
d1310b2e CM |
526 | * |
527 | * If no bits are set on the state struct after clearing things, the | |
528 | * struct is freed and removed from the tree | |
529 | */ | |
cdc6a395 LZ |
530 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
531 | struct extent_state *state, | |
9ee49a04 | 532 | unsigned *bits, int wake) |
d1310b2e | 533 | { |
cdc6a395 | 534 | struct extent_state *next; |
9ee49a04 | 535 | unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS; |
d1310b2e | 536 | |
0ca1f7ce | 537 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
538 | u64 range = state->end - state->start + 1; |
539 | WARN_ON(range > tree->dirty_bytes); | |
540 | tree->dirty_bytes -= range; | |
541 | } | |
291d673e | 542 | clear_state_cb(tree, state, bits); |
32c00aff | 543 | state->state &= ~bits_to_clear; |
d1310b2e CM |
544 | if (wake) |
545 | wake_up(&state->wq); | |
0ca1f7ce | 546 | if (state->state == 0) { |
cdc6a395 | 547 | next = next_state(state); |
27a3507d | 548 | if (extent_state_in_tree(state)) { |
d1310b2e | 549 | rb_erase(&state->rb_node, &tree->state); |
27a3507d | 550 | RB_CLEAR_NODE(&state->rb_node); |
d1310b2e CM |
551 | free_extent_state(state); |
552 | } else { | |
553 | WARN_ON(1); | |
554 | } | |
555 | } else { | |
556 | merge_state(tree, state); | |
cdc6a395 | 557 | next = next_state(state); |
d1310b2e | 558 | } |
cdc6a395 | 559 | return next; |
d1310b2e CM |
560 | } |
561 | ||
8233767a XG |
562 | static struct extent_state * |
563 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
564 | { | |
565 | if (!prealloc) | |
566 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
567 | ||
568 | return prealloc; | |
569 | } | |
570 | ||
48a3b636 | 571 | static void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
c2d904e0 JM |
572 | { |
573 | btrfs_panic(tree_fs_info(tree), err, "Locking error: " | |
574 | "Extent tree was modified by another " | |
575 | "thread while locked."); | |
576 | } | |
577 | ||
d1310b2e CM |
578 | /* |
579 | * clear some bits on a range in the tree. This may require splitting | |
580 | * or inserting elements in the tree, so the gfp mask is used to | |
581 | * indicate which allocations or sleeping are allowed. | |
582 | * | |
583 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
584 | * the given range from the tree regardless of state (ie for truncate). | |
585 | * | |
586 | * the range [start, end] is inclusive. | |
587 | * | |
6763af84 | 588 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e CM |
589 | */ |
590 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 591 | unsigned bits, int wake, int delete, |
2c64c53d CM |
592 | struct extent_state **cached_state, |
593 | gfp_t mask) | |
d1310b2e CM |
594 | { |
595 | struct extent_state *state; | |
2c64c53d | 596 | struct extent_state *cached; |
d1310b2e CM |
597 | struct extent_state *prealloc = NULL; |
598 | struct rb_node *node; | |
5c939df5 | 599 | u64 last_end; |
d1310b2e | 600 | int err; |
2ac55d41 | 601 | int clear = 0; |
d1310b2e | 602 | |
a5dee37d | 603 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 604 | |
7ee9e440 JB |
605 | if (bits & EXTENT_DELALLOC) |
606 | bits |= EXTENT_NORESERVE; | |
607 | ||
0ca1f7ce YZ |
608 | if (delete) |
609 | bits |= ~EXTENT_CTLBITS; | |
610 | bits |= EXTENT_FIRST_DELALLOC; | |
611 | ||
2ac55d41 JB |
612 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
613 | clear = 1; | |
d1310b2e CM |
614 | again: |
615 | if (!prealloc && (mask & __GFP_WAIT)) { | |
c7bc6319 FM |
616 | /* |
617 | * Don't care for allocation failure here because we might end | |
618 | * up not needing the pre-allocated extent state at all, which | |
619 | * is the case if we only have in the tree extent states that | |
620 | * cover our input range and don't cover too any other range. | |
621 | * If we end up needing a new extent state we allocate it later. | |
622 | */ | |
d1310b2e | 623 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
624 | } |
625 | ||
cad321ad | 626 | spin_lock(&tree->lock); |
2c64c53d CM |
627 | if (cached_state) { |
628 | cached = *cached_state; | |
2ac55d41 JB |
629 | |
630 | if (clear) { | |
631 | *cached_state = NULL; | |
632 | cached_state = NULL; | |
633 | } | |
634 | ||
27a3507d FM |
635 | if (cached && extent_state_in_tree(cached) && |
636 | cached->start <= start && cached->end > start) { | |
2ac55d41 JB |
637 | if (clear) |
638 | atomic_dec(&cached->refs); | |
2c64c53d | 639 | state = cached; |
42daec29 | 640 | goto hit_next; |
2c64c53d | 641 | } |
2ac55d41 JB |
642 | if (clear) |
643 | free_extent_state(cached); | |
2c64c53d | 644 | } |
d1310b2e CM |
645 | /* |
646 | * this search will find the extents that end after | |
647 | * our range starts | |
648 | */ | |
80ea96b1 | 649 | node = tree_search(tree, start); |
d1310b2e CM |
650 | if (!node) |
651 | goto out; | |
652 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 653 | hit_next: |
d1310b2e CM |
654 | if (state->start > end) |
655 | goto out; | |
656 | WARN_ON(state->end < start); | |
5c939df5 | 657 | last_end = state->end; |
d1310b2e | 658 | |
0449314a | 659 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
660 | if (!(state->state & bits)) { |
661 | state = next_state(state); | |
0449314a | 662 | goto next; |
cdc6a395 | 663 | } |
0449314a | 664 | |
d1310b2e CM |
665 | /* |
666 | * | ---- desired range ---- | | |
667 | * | state | or | |
668 | * | ------------- state -------------- | | |
669 | * | |
670 | * We need to split the extent we found, and may flip | |
671 | * bits on second half. | |
672 | * | |
673 | * If the extent we found extends past our range, we | |
674 | * just split and search again. It'll get split again | |
675 | * the next time though. | |
676 | * | |
677 | * If the extent we found is inside our range, we clear | |
678 | * the desired bit on it. | |
679 | */ | |
680 | ||
681 | if (state->start < start) { | |
8233767a XG |
682 | prealloc = alloc_extent_state_atomic(prealloc); |
683 | BUG_ON(!prealloc); | |
d1310b2e | 684 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
685 | if (err) |
686 | extent_io_tree_panic(tree, err); | |
687 | ||
d1310b2e CM |
688 | prealloc = NULL; |
689 | if (err) | |
690 | goto out; | |
691 | if (state->end <= end) { | |
d1ac6e41 LB |
692 | state = clear_state_bit(tree, state, &bits, wake); |
693 | goto next; | |
d1310b2e CM |
694 | } |
695 | goto search_again; | |
696 | } | |
697 | /* | |
698 | * | ---- desired range ---- | | |
699 | * | state | | |
700 | * We need to split the extent, and clear the bit | |
701 | * on the first half | |
702 | */ | |
703 | if (state->start <= end && state->end > end) { | |
8233767a XG |
704 | prealloc = alloc_extent_state_atomic(prealloc); |
705 | BUG_ON(!prealloc); | |
d1310b2e | 706 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
707 | if (err) |
708 | extent_io_tree_panic(tree, err); | |
709 | ||
d1310b2e CM |
710 | if (wake) |
711 | wake_up(&state->wq); | |
42daec29 | 712 | |
6763af84 | 713 | clear_state_bit(tree, prealloc, &bits, wake); |
9ed74f2d | 714 | |
d1310b2e CM |
715 | prealloc = NULL; |
716 | goto out; | |
717 | } | |
42daec29 | 718 | |
cdc6a395 | 719 | state = clear_state_bit(tree, state, &bits, wake); |
0449314a | 720 | next: |
5c939df5 YZ |
721 | if (last_end == (u64)-1) |
722 | goto out; | |
723 | start = last_end + 1; | |
cdc6a395 | 724 | if (start <= end && state && !need_resched()) |
692e5759 | 725 | goto hit_next; |
d1310b2e CM |
726 | goto search_again; |
727 | ||
728 | out: | |
cad321ad | 729 | spin_unlock(&tree->lock); |
d1310b2e CM |
730 | if (prealloc) |
731 | free_extent_state(prealloc); | |
732 | ||
6763af84 | 733 | return 0; |
d1310b2e CM |
734 | |
735 | search_again: | |
736 | if (start > end) | |
737 | goto out; | |
cad321ad | 738 | spin_unlock(&tree->lock); |
d1310b2e CM |
739 | if (mask & __GFP_WAIT) |
740 | cond_resched(); | |
741 | goto again; | |
742 | } | |
d1310b2e | 743 | |
143bede5 JM |
744 | static void wait_on_state(struct extent_io_tree *tree, |
745 | struct extent_state *state) | |
641f5219 CH |
746 | __releases(tree->lock) |
747 | __acquires(tree->lock) | |
d1310b2e CM |
748 | { |
749 | DEFINE_WAIT(wait); | |
750 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 751 | spin_unlock(&tree->lock); |
d1310b2e | 752 | schedule(); |
cad321ad | 753 | spin_lock(&tree->lock); |
d1310b2e | 754 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
755 | } |
756 | ||
757 | /* | |
758 | * waits for one or more bits to clear on a range in the state tree. | |
759 | * The range [start, end] is inclusive. | |
760 | * The tree lock is taken by this function | |
761 | */ | |
41074888 DS |
762 | static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
763 | unsigned long bits) | |
d1310b2e CM |
764 | { |
765 | struct extent_state *state; | |
766 | struct rb_node *node; | |
767 | ||
a5dee37d | 768 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 769 | |
cad321ad | 770 | spin_lock(&tree->lock); |
d1310b2e CM |
771 | again: |
772 | while (1) { | |
773 | /* | |
774 | * this search will find all the extents that end after | |
775 | * our range starts | |
776 | */ | |
80ea96b1 | 777 | node = tree_search(tree, start); |
c50d3e71 | 778 | process_node: |
d1310b2e CM |
779 | if (!node) |
780 | break; | |
781 | ||
782 | state = rb_entry(node, struct extent_state, rb_node); | |
783 | ||
784 | if (state->start > end) | |
785 | goto out; | |
786 | ||
787 | if (state->state & bits) { | |
788 | start = state->start; | |
789 | atomic_inc(&state->refs); | |
790 | wait_on_state(tree, state); | |
791 | free_extent_state(state); | |
792 | goto again; | |
793 | } | |
794 | start = state->end + 1; | |
795 | ||
796 | if (start > end) | |
797 | break; | |
798 | ||
c50d3e71 FM |
799 | if (!cond_resched_lock(&tree->lock)) { |
800 | node = rb_next(node); | |
801 | goto process_node; | |
802 | } | |
d1310b2e CM |
803 | } |
804 | out: | |
cad321ad | 805 | spin_unlock(&tree->lock); |
d1310b2e | 806 | } |
d1310b2e | 807 | |
1bf85046 | 808 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 809 | struct extent_state *state, |
d38ed27f | 810 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e | 811 | { |
9ee49a04 | 812 | unsigned bits_to_set = *bits & ~EXTENT_CTLBITS; |
9ed74f2d | 813 | |
1bf85046 | 814 | set_state_cb(tree, state, bits); |
0ca1f7ce | 815 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
816 | u64 range = state->end - state->start + 1; |
817 | tree->dirty_bytes += range; | |
818 | } | |
d38ed27f | 819 | add_extent_changeset(state, bits_to_set, changeset, 1); |
0ca1f7ce | 820 | state->state |= bits_to_set; |
d1310b2e CM |
821 | } |
822 | ||
e38e2ed7 FM |
823 | static void cache_state_if_flags(struct extent_state *state, |
824 | struct extent_state **cached_ptr, | |
9ee49a04 | 825 | unsigned flags) |
2c64c53d CM |
826 | { |
827 | if (cached_ptr && !(*cached_ptr)) { | |
e38e2ed7 | 828 | if (!flags || (state->state & flags)) { |
2c64c53d CM |
829 | *cached_ptr = state; |
830 | atomic_inc(&state->refs); | |
831 | } | |
832 | } | |
833 | } | |
834 | ||
e38e2ed7 FM |
835 | static void cache_state(struct extent_state *state, |
836 | struct extent_state **cached_ptr) | |
837 | { | |
838 | return cache_state_if_flags(state, cached_ptr, | |
839 | EXTENT_IOBITS | EXTENT_BOUNDARY); | |
840 | } | |
841 | ||
d1310b2e | 842 | /* |
1edbb734 CM |
843 | * set some bits on a range in the tree. This may require allocations or |
844 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 845 | * |
1edbb734 CM |
846 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
847 | * part of the range already has the desired bits set. The start of the | |
848 | * existing range is returned in failed_start in this case. | |
d1310b2e | 849 | * |
1edbb734 | 850 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 851 | */ |
1edbb734 | 852 | |
3fbe5c02 JM |
853 | static int __must_check |
854 | __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 855 | unsigned bits, unsigned exclusive_bits, |
41074888 | 856 | u64 *failed_start, struct extent_state **cached_state, |
d38ed27f | 857 | gfp_t mask, struct extent_changeset *changeset) |
d1310b2e CM |
858 | { |
859 | struct extent_state *state; | |
860 | struct extent_state *prealloc = NULL; | |
861 | struct rb_node *node; | |
12cfbad9 FDBM |
862 | struct rb_node **p; |
863 | struct rb_node *parent; | |
d1310b2e | 864 | int err = 0; |
d1310b2e CM |
865 | u64 last_start; |
866 | u64 last_end; | |
42daec29 | 867 | |
a5dee37d | 868 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 869 | |
0ca1f7ce | 870 | bits |= EXTENT_FIRST_DELALLOC; |
d1310b2e CM |
871 | again: |
872 | if (!prealloc && (mask & __GFP_WAIT)) { | |
873 | prealloc = alloc_extent_state(mask); | |
8233767a | 874 | BUG_ON(!prealloc); |
d1310b2e CM |
875 | } |
876 | ||
cad321ad | 877 | spin_lock(&tree->lock); |
9655d298 CM |
878 | if (cached_state && *cached_state) { |
879 | state = *cached_state; | |
df98b6e2 | 880 | if (state->start <= start && state->end > start && |
27a3507d | 881 | extent_state_in_tree(state)) { |
9655d298 CM |
882 | node = &state->rb_node; |
883 | goto hit_next; | |
884 | } | |
885 | } | |
d1310b2e CM |
886 | /* |
887 | * this search will find all the extents that end after | |
888 | * our range starts. | |
889 | */ | |
12cfbad9 | 890 | node = tree_search_for_insert(tree, start, &p, &parent); |
d1310b2e | 891 | if (!node) { |
8233767a XG |
892 | prealloc = alloc_extent_state_atomic(prealloc); |
893 | BUG_ON(!prealloc); | |
12cfbad9 | 894 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 895 | &p, &parent, &bits, changeset); |
c2d904e0 JM |
896 | if (err) |
897 | extent_io_tree_panic(tree, err); | |
898 | ||
c42ac0bc | 899 | cache_state(prealloc, cached_state); |
d1310b2e | 900 | prealloc = NULL; |
d1310b2e CM |
901 | goto out; |
902 | } | |
d1310b2e | 903 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 904 | hit_next: |
d1310b2e CM |
905 | last_start = state->start; |
906 | last_end = state->end; | |
907 | ||
908 | /* | |
909 | * | ---- desired range ---- | | |
910 | * | state | | |
911 | * | |
912 | * Just lock what we found and keep going | |
913 | */ | |
914 | if (state->start == start && state->end <= end) { | |
1edbb734 | 915 | if (state->state & exclusive_bits) { |
d1310b2e CM |
916 | *failed_start = state->start; |
917 | err = -EEXIST; | |
918 | goto out; | |
919 | } | |
42daec29 | 920 | |
d38ed27f | 921 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 922 | cache_state(state, cached_state); |
d1310b2e | 923 | merge_state(tree, state); |
5c939df5 YZ |
924 | if (last_end == (u64)-1) |
925 | goto out; | |
926 | start = last_end + 1; | |
d1ac6e41 LB |
927 | state = next_state(state); |
928 | if (start < end && state && state->start == start && | |
929 | !need_resched()) | |
930 | goto hit_next; | |
d1310b2e CM |
931 | goto search_again; |
932 | } | |
933 | ||
934 | /* | |
935 | * | ---- desired range ---- | | |
936 | * | state | | |
937 | * or | |
938 | * | ------------- state -------------- | | |
939 | * | |
940 | * We need to split the extent we found, and may flip bits on | |
941 | * second half. | |
942 | * | |
943 | * If the extent we found extends past our | |
944 | * range, we just split and search again. It'll get split | |
945 | * again the next time though. | |
946 | * | |
947 | * If the extent we found is inside our range, we set the | |
948 | * desired bit on it. | |
949 | */ | |
950 | if (state->start < start) { | |
1edbb734 | 951 | if (state->state & exclusive_bits) { |
d1310b2e CM |
952 | *failed_start = start; |
953 | err = -EEXIST; | |
954 | goto out; | |
955 | } | |
8233767a XG |
956 | |
957 | prealloc = alloc_extent_state_atomic(prealloc); | |
958 | BUG_ON(!prealloc); | |
d1310b2e | 959 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
960 | if (err) |
961 | extent_io_tree_panic(tree, err); | |
962 | ||
d1310b2e CM |
963 | prealloc = NULL; |
964 | if (err) | |
965 | goto out; | |
966 | if (state->end <= end) { | |
d38ed27f | 967 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 968 | cache_state(state, cached_state); |
d1310b2e | 969 | merge_state(tree, state); |
5c939df5 YZ |
970 | if (last_end == (u64)-1) |
971 | goto out; | |
972 | start = last_end + 1; | |
d1ac6e41 LB |
973 | state = next_state(state); |
974 | if (start < end && state && state->start == start && | |
975 | !need_resched()) | |
976 | goto hit_next; | |
d1310b2e CM |
977 | } |
978 | goto search_again; | |
979 | } | |
980 | /* | |
981 | * | ---- desired range ---- | | |
982 | * | state | or | state | | |
983 | * | |
984 | * There's a hole, we need to insert something in it and | |
985 | * ignore the extent we found. | |
986 | */ | |
987 | if (state->start > start) { | |
988 | u64 this_end; | |
989 | if (end < last_start) | |
990 | this_end = end; | |
991 | else | |
d397712b | 992 | this_end = last_start - 1; |
8233767a XG |
993 | |
994 | prealloc = alloc_extent_state_atomic(prealloc); | |
995 | BUG_ON(!prealloc); | |
c7f895a2 XG |
996 | |
997 | /* | |
998 | * Avoid to free 'prealloc' if it can be merged with | |
999 | * the later extent. | |
1000 | */ | |
d1310b2e | 1001 | err = insert_state(tree, prealloc, start, this_end, |
d38ed27f | 1002 | NULL, NULL, &bits, changeset); |
c2d904e0 JM |
1003 | if (err) |
1004 | extent_io_tree_panic(tree, err); | |
1005 | ||
9ed74f2d JB |
1006 | cache_state(prealloc, cached_state); |
1007 | prealloc = NULL; | |
d1310b2e CM |
1008 | start = this_end + 1; |
1009 | goto search_again; | |
1010 | } | |
1011 | /* | |
1012 | * | ---- desired range ---- | | |
1013 | * | state | | |
1014 | * We need to split the extent, and set the bit | |
1015 | * on the first half | |
1016 | */ | |
1017 | if (state->start <= end && state->end > end) { | |
1edbb734 | 1018 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1019 | *failed_start = start; |
1020 | err = -EEXIST; | |
1021 | goto out; | |
1022 | } | |
8233767a XG |
1023 | |
1024 | prealloc = alloc_extent_state_atomic(prealloc); | |
1025 | BUG_ON(!prealloc); | |
d1310b2e | 1026 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
1027 | if (err) |
1028 | extent_io_tree_panic(tree, err); | |
d1310b2e | 1029 | |
d38ed27f | 1030 | set_state_bits(tree, prealloc, &bits, changeset); |
2c64c53d | 1031 | cache_state(prealloc, cached_state); |
d1310b2e CM |
1032 | merge_state(tree, prealloc); |
1033 | prealloc = NULL; | |
1034 | goto out; | |
1035 | } | |
1036 | ||
1037 | goto search_again; | |
1038 | ||
1039 | out: | |
cad321ad | 1040 | spin_unlock(&tree->lock); |
d1310b2e CM |
1041 | if (prealloc) |
1042 | free_extent_state(prealloc); | |
1043 | ||
1044 | return err; | |
1045 | ||
1046 | search_again: | |
1047 | if (start > end) | |
1048 | goto out; | |
cad321ad | 1049 | spin_unlock(&tree->lock); |
d1310b2e CM |
1050 | if (mask & __GFP_WAIT) |
1051 | cond_resched(); | |
1052 | goto again; | |
1053 | } | |
d1310b2e | 1054 | |
41074888 | 1055 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
9ee49a04 | 1056 | unsigned bits, u64 * failed_start, |
41074888 | 1057 | struct extent_state **cached_state, gfp_t mask) |
3fbe5c02 JM |
1058 | { |
1059 | return __set_extent_bit(tree, start, end, bits, 0, failed_start, | |
d38ed27f | 1060 | cached_state, mask, NULL); |
3fbe5c02 JM |
1061 | } |
1062 | ||
1063 | ||
462d6fac | 1064 | /** |
10983f2e LB |
1065 | * convert_extent_bit - convert all bits in a given range from one bit to |
1066 | * another | |
462d6fac JB |
1067 | * @tree: the io tree to search |
1068 | * @start: the start offset in bytes | |
1069 | * @end: the end offset in bytes (inclusive) | |
1070 | * @bits: the bits to set in this range | |
1071 | * @clear_bits: the bits to clear in this range | |
e6138876 | 1072 | * @cached_state: state that we're going to cache |
462d6fac JB |
1073 | * @mask: the allocation mask |
1074 | * | |
1075 | * This will go through and set bits for the given range. If any states exist | |
1076 | * already in this range they are set with the given bit and cleared of the | |
1077 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
1078 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
1079 | * boundary bits like LOCK. | |
1080 | */ | |
1081 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1082 | unsigned bits, unsigned clear_bits, |
e6138876 | 1083 | struct extent_state **cached_state, gfp_t mask) |
462d6fac JB |
1084 | { |
1085 | struct extent_state *state; | |
1086 | struct extent_state *prealloc = NULL; | |
1087 | struct rb_node *node; | |
12cfbad9 FDBM |
1088 | struct rb_node **p; |
1089 | struct rb_node *parent; | |
462d6fac JB |
1090 | int err = 0; |
1091 | u64 last_start; | |
1092 | u64 last_end; | |
c8fd3de7 | 1093 | bool first_iteration = true; |
462d6fac | 1094 | |
a5dee37d | 1095 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 1096 | |
462d6fac JB |
1097 | again: |
1098 | if (!prealloc && (mask & __GFP_WAIT)) { | |
c8fd3de7 FM |
1099 | /* |
1100 | * Best effort, don't worry if extent state allocation fails | |
1101 | * here for the first iteration. We might have a cached state | |
1102 | * that matches exactly the target range, in which case no | |
1103 | * extent state allocations are needed. We'll only know this | |
1104 | * after locking the tree. | |
1105 | */ | |
462d6fac | 1106 | prealloc = alloc_extent_state(mask); |
c8fd3de7 | 1107 | if (!prealloc && !first_iteration) |
462d6fac JB |
1108 | return -ENOMEM; |
1109 | } | |
1110 | ||
1111 | spin_lock(&tree->lock); | |
e6138876 JB |
1112 | if (cached_state && *cached_state) { |
1113 | state = *cached_state; | |
1114 | if (state->start <= start && state->end > start && | |
27a3507d | 1115 | extent_state_in_tree(state)) { |
e6138876 JB |
1116 | node = &state->rb_node; |
1117 | goto hit_next; | |
1118 | } | |
1119 | } | |
1120 | ||
462d6fac JB |
1121 | /* |
1122 | * this search will find all the extents that end after | |
1123 | * our range starts. | |
1124 | */ | |
12cfbad9 | 1125 | node = tree_search_for_insert(tree, start, &p, &parent); |
462d6fac JB |
1126 | if (!node) { |
1127 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1128 | if (!prealloc) { |
1129 | err = -ENOMEM; | |
1130 | goto out; | |
1131 | } | |
12cfbad9 | 1132 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 1133 | &p, &parent, &bits, NULL); |
c2d904e0 JM |
1134 | if (err) |
1135 | extent_io_tree_panic(tree, err); | |
c42ac0bc FDBM |
1136 | cache_state(prealloc, cached_state); |
1137 | prealloc = NULL; | |
462d6fac JB |
1138 | goto out; |
1139 | } | |
1140 | state = rb_entry(node, struct extent_state, rb_node); | |
1141 | hit_next: | |
1142 | last_start = state->start; | |
1143 | last_end = state->end; | |
1144 | ||
1145 | /* | |
1146 | * | ---- desired range ---- | | |
1147 | * | state | | |
1148 | * | |
1149 | * Just lock what we found and keep going | |
1150 | */ | |
1151 | if (state->start == start && state->end <= end) { | |
d38ed27f | 1152 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1153 | cache_state(state, cached_state); |
d1ac6e41 | 1154 | state = clear_state_bit(tree, state, &clear_bits, 0); |
462d6fac JB |
1155 | if (last_end == (u64)-1) |
1156 | goto out; | |
462d6fac | 1157 | start = last_end + 1; |
d1ac6e41 LB |
1158 | if (start < end && state && state->start == start && |
1159 | !need_resched()) | |
1160 | goto hit_next; | |
462d6fac JB |
1161 | goto search_again; |
1162 | } | |
1163 | ||
1164 | /* | |
1165 | * | ---- desired range ---- | | |
1166 | * | state | | |
1167 | * or | |
1168 | * | ------------- state -------------- | | |
1169 | * | |
1170 | * We need to split the extent we found, and may flip bits on | |
1171 | * second half. | |
1172 | * | |
1173 | * If the extent we found extends past our | |
1174 | * range, we just split and search again. It'll get split | |
1175 | * again the next time though. | |
1176 | * | |
1177 | * If the extent we found is inside our range, we set the | |
1178 | * desired bit on it. | |
1179 | */ | |
1180 | if (state->start < start) { | |
1181 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1182 | if (!prealloc) { |
1183 | err = -ENOMEM; | |
1184 | goto out; | |
1185 | } | |
462d6fac | 1186 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1187 | if (err) |
1188 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1189 | prealloc = NULL; |
1190 | if (err) | |
1191 | goto out; | |
1192 | if (state->end <= end) { | |
d38ed27f | 1193 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1194 | cache_state(state, cached_state); |
d1ac6e41 | 1195 | state = clear_state_bit(tree, state, &clear_bits, 0); |
462d6fac JB |
1196 | if (last_end == (u64)-1) |
1197 | goto out; | |
1198 | start = last_end + 1; | |
d1ac6e41 LB |
1199 | if (start < end && state && state->start == start && |
1200 | !need_resched()) | |
1201 | goto hit_next; | |
462d6fac JB |
1202 | } |
1203 | goto search_again; | |
1204 | } | |
1205 | /* | |
1206 | * | ---- desired range ---- | | |
1207 | * | state | or | state | | |
1208 | * | |
1209 | * There's a hole, we need to insert something in it and | |
1210 | * ignore the extent we found. | |
1211 | */ | |
1212 | if (state->start > start) { | |
1213 | u64 this_end; | |
1214 | if (end < last_start) | |
1215 | this_end = end; | |
1216 | else | |
1217 | this_end = last_start - 1; | |
1218 | ||
1219 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1220 | if (!prealloc) { |
1221 | err = -ENOMEM; | |
1222 | goto out; | |
1223 | } | |
462d6fac JB |
1224 | |
1225 | /* | |
1226 | * Avoid to free 'prealloc' if it can be merged with | |
1227 | * the later extent. | |
1228 | */ | |
1229 | err = insert_state(tree, prealloc, start, this_end, | |
d38ed27f | 1230 | NULL, NULL, &bits, NULL); |
c2d904e0 JM |
1231 | if (err) |
1232 | extent_io_tree_panic(tree, err); | |
e6138876 | 1233 | cache_state(prealloc, cached_state); |
462d6fac JB |
1234 | prealloc = NULL; |
1235 | start = this_end + 1; | |
1236 | goto search_again; | |
1237 | } | |
1238 | /* | |
1239 | * | ---- desired range ---- | | |
1240 | * | state | | |
1241 | * We need to split the extent, and set the bit | |
1242 | * on the first half | |
1243 | */ | |
1244 | if (state->start <= end && state->end > end) { | |
1245 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1246 | if (!prealloc) { |
1247 | err = -ENOMEM; | |
1248 | goto out; | |
1249 | } | |
462d6fac JB |
1250 | |
1251 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1252 | if (err) |
1253 | extent_io_tree_panic(tree, err); | |
462d6fac | 1254 | |
d38ed27f | 1255 | set_state_bits(tree, prealloc, &bits, NULL); |
e6138876 | 1256 | cache_state(prealloc, cached_state); |
462d6fac | 1257 | clear_state_bit(tree, prealloc, &clear_bits, 0); |
462d6fac JB |
1258 | prealloc = NULL; |
1259 | goto out; | |
1260 | } | |
1261 | ||
1262 | goto search_again; | |
1263 | ||
1264 | out: | |
1265 | spin_unlock(&tree->lock); | |
1266 | if (prealloc) | |
1267 | free_extent_state(prealloc); | |
1268 | ||
1269 | return err; | |
1270 | ||
1271 | search_again: | |
1272 | if (start > end) | |
1273 | goto out; | |
1274 | spin_unlock(&tree->lock); | |
1275 | if (mask & __GFP_WAIT) | |
1276 | cond_resched(); | |
c8fd3de7 | 1277 | first_iteration = false; |
462d6fac JB |
1278 | goto again; |
1279 | } | |
1280 | ||
d1310b2e CM |
1281 | /* wrappers around set/clear extent bit */ |
1282 | int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
1283 | gfp_t mask) | |
1284 | { | |
3fbe5c02 | 1285 | return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL, |
2c64c53d | 1286 | NULL, mask); |
d1310b2e | 1287 | } |
d1310b2e CM |
1288 | |
1289 | int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1290 | unsigned bits, gfp_t mask) |
d1310b2e | 1291 | { |
3fbe5c02 | 1292 | return set_extent_bit(tree, start, end, bits, NULL, |
2c64c53d | 1293 | NULL, mask); |
d1310b2e | 1294 | } |
d1310b2e | 1295 | |
d38ed27f QW |
1296 | int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
1297 | unsigned bits, gfp_t mask, | |
1298 | struct extent_changeset *changeset) | |
1299 | { | |
1300 | /* | |
1301 | * We don't support EXTENT_LOCKED yet, as current changeset will | |
1302 | * record any bits changed, so for EXTENT_LOCKED case, it will | |
1303 | * either fail with -EEXIST or changeset will record the whole | |
1304 | * range. | |
1305 | */ | |
1306 | BUG_ON(bits & EXTENT_LOCKED); | |
1307 | ||
1308 | return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, mask, | |
1309 | changeset); | |
1310 | } | |
1311 | ||
d1310b2e | 1312 | int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
9ee49a04 | 1313 | unsigned bits, gfp_t mask) |
d1310b2e | 1314 | { |
0f31871f FM |
1315 | int wake = 0; |
1316 | ||
1317 | if (bits & EXTENT_LOCKED) | |
1318 | wake = 1; | |
1319 | ||
1320 | return clear_extent_bit(tree, start, end, bits, wake, 0, NULL, mask); | |
d1310b2e | 1321 | } |
d1310b2e CM |
1322 | |
1323 | int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, | |
2ac55d41 | 1324 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
1325 | { |
1326 | return set_extent_bit(tree, start, end, | |
fee187d9 | 1327 | EXTENT_DELALLOC | EXTENT_UPTODATE, |
3fbe5c02 | 1328 | NULL, cached_state, mask); |
d1310b2e | 1329 | } |
d1310b2e | 1330 | |
9e8a4a8b LB |
1331 | int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end, |
1332 | struct extent_state **cached_state, gfp_t mask) | |
1333 | { | |
1334 | return set_extent_bit(tree, start, end, | |
1335 | EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG, | |
1336 | NULL, cached_state, mask); | |
1337 | } | |
1338 | ||
d1310b2e CM |
1339 | int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, |
1340 | gfp_t mask) | |
1341 | { | |
1342 | return clear_extent_bit(tree, start, end, | |
32c00aff | 1343 | EXTENT_DIRTY | EXTENT_DELALLOC | |
0ca1f7ce | 1344 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask); |
d1310b2e | 1345 | } |
d1310b2e CM |
1346 | |
1347 | int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, | |
1348 | gfp_t mask) | |
1349 | { | |
3fbe5c02 | 1350 | return set_extent_bit(tree, start, end, EXTENT_NEW, NULL, |
2c64c53d | 1351 | NULL, mask); |
d1310b2e | 1352 | } |
d1310b2e | 1353 | |
d1310b2e | 1354 | int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
507903b8 | 1355 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e | 1356 | { |
6b67a320 | 1357 | return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL, |
3fbe5c02 | 1358 | cached_state, mask); |
d1310b2e | 1359 | } |
d1310b2e | 1360 | |
5fd02043 JB |
1361 | int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
1362 | struct extent_state **cached_state, gfp_t mask) | |
d1310b2e | 1363 | { |
2c64c53d | 1364 | return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, |
2ac55d41 | 1365 | cached_state, mask); |
d1310b2e | 1366 | } |
d1310b2e | 1367 | |
d352ac68 CM |
1368 | /* |
1369 | * either insert or lock state struct between start and end use mask to tell | |
1370 | * us if waiting is desired. | |
1371 | */ | |
1edbb734 | 1372 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
9ee49a04 | 1373 | unsigned bits, struct extent_state **cached_state) |
d1310b2e CM |
1374 | { |
1375 | int err; | |
1376 | u64 failed_start; | |
9ee49a04 | 1377 | |
d1310b2e | 1378 | while (1) { |
3fbe5c02 JM |
1379 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits, |
1380 | EXTENT_LOCKED, &failed_start, | |
d38ed27f | 1381 | cached_state, GFP_NOFS, NULL); |
d0082371 | 1382 | if (err == -EEXIST) { |
d1310b2e CM |
1383 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1384 | start = failed_start; | |
d0082371 | 1385 | } else |
d1310b2e | 1386 | break; |
d1310b2e CM |
1387 | WARN_ON(start > end); |
1388 | } | |
1389 | return err; | |
1390 | } | |
d1310b2e | 1391 | |
d0082371 | 1392 | int lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
1edbb734 | 1393 | { |
d0082371 | 1394 | return lock_extent_bits(tree, start, end, 0, NULL); |
1edbb734 CM |
1395 | } |
1396 | ||
d0082371 | 1397 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1398 | { |
1399 | int err; | |
1400 | u64 failed_start; | |
1401 | ||
3fbe5c02 | 1402 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
d38ed27f | 1403 | &failed_start, NULL, GFP_NOFS, NULL); |
6643558d YZ |
1404 | if (err == -EEXIST) { |
1405 | if (failed_start > start) | |
1406 | clear_extent_bit(tree, start, failed_start - 1, | |
d0082371 | 1407 | EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS); |
25179201 | 1408 | return 0; |
6643558d | 1409 | } |
25179201 JB |
1410 | return 1; |
1411 | } | |
25179201 | 1412 | |
2c64c53d CM |
1413 | int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end, |
1414 | struct extent_state **cached, gfp_t mask) | |
1415 | { | |
1416 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, | |
1417 | mask); | |
1418 | } | |
1419 | ||
d0082371 | 1420 | int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e | 1421 | { |
2c64c53d | 1422 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL, |
d0082371 | 1423 | GFP_NOFS); |
d1310b2e | 1424 | } |
d1310b2e | 1425 | |
4adaa611 CM |
1426 | int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
1427 | { | |
1428 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1429 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1430 | struct page *page; | |
1431 | ||
1432 | while (index <= end_index) { | |
1433 | page = find_get_page(inode->i_mapping, index); | |
1434 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1435 | clear_page_dirty_for_io(page); | |
1436 | page_cache_release(page); | |
1437 | index++; | |
1438 | } | |
1439 | return 0; | |
1440 | } | |
1441 | ||
1442 | int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) | |
1443 | { | |
1444 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1445 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1446 | struct page *page; | |
1447 | ||
1448 | while (index <= end_index) { | |
1449 | page = find_get_page(inode->i_mapping, index); | |
1450 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
4adaa611 | 1451 | __set_page_dirty_nobuffers(page); |
8d38633c | 1452 | account_page_redirty(page); |
4adaa611 CM |
1453 | page_cache_release(page); |
1454 | index++; | |
1455 | } | |
1456 | return 0; | |
1457 | } | |
1458 | ||
d1310b2e CM |
1459 | /* |
1460 | * helper function to set both pages and extents in the tree writeback | |
1461 | */ | |
b2950863 | 1462 | static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e CM |
1463 | { |
1464 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1465 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1466 | struct page *page; | |
1467 | ||
1468 | while (index <= end_index) { | |
1469 | page = find_get_page(tree->mapping, index); | |
79787eaa | 1470 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ |
d1310b2e CM |
1471 | set_page_writeback(page); |
1472 | page_cache_release(page); | |
1473 | index++; | |
1474 | } | |
d1310b2e CM |
1475 | return 0; |
1476 | } | |
d1310b2e | 1477 | |
d352ac68 CM |
1478 | /* find the first state struct with 'bits' set after 'start', and |
1479 | * return it. tree->lock must be held. NULL will returned if | |
1480 | * nothing was found after 'start' | |
1481 | */ | |
48a3b636 ES |
1482 | static struct extent_state * |
1483 | find_first_extent_bit_state(struct extent_io_tree *tree, | |
9ee49a04 | 1484 | u64 start, unsigned bits) |
d7fc640e CM |
1485 | { |
1486 | struct rb_node *node; | |
1487 | struct extent_state *state; | |
1488 | ||
1489 | /* | |
1490 | * this search will find all the extents that end after | |
1491 | * our range starts. | |
1492 | */ | |
1493 | node = tree_search(tree, start); | |
d397712b | 1494 | if (!node) |
d7fc640e | 1495 | goto out; |
d7fc640e | 1496 | |
d397712b | 1497 | while (1) { |
d7fc640e | 1498 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1499 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1500 | return state; |
d397712b | 1501 | |
d7fc640e CM |
1502 | node = rb_next(node); |
1503 | if (!node) | |
1504 | break; | |
1505 | } | |
1506 | out: | |
1507 | return NULL; | |
1508 | } | |
d7fc640e | 1509 | |
69261c4b XG |
1510 | /* |
1511 | * find the first offset in the io tree with 'bits' set. zero is | |
1512 | * returned if we find something, and *start_ret and *end_ret are | |
1513 | * set to reflect the state struct that was found. | |
1514 | * | |
477d7eaf | 1515 | * If nothing was found, 1 is returned. If found something, return 0. |
69261c4b XG |
1516 | */ |
1517 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
9ee49a04 | 1518 | u64 *start_ret, u64 *end_ret, unsigned bits, |
e6138876 | 1519 | struct extent_state **cached_state) |
69261c4b XG |
1520 | { |
1521 | struct extent_state *state; | |
e6138876 | 1522 | struct rb_node *n; |
69261c4b XG |
1523 | int ret = 1; |
1524 | ||
1525 | spin_lock(&tree->lock); | |
e6138876 JB |
1526 | if (cached_state && *cached_state) { |
1527 | state = *cached_state; | |
27a3507d | 1528 | if (state->end == start - 1 && extent_state_in_tree(state)) { |
e6138876 JB |
1529 | n = rb_next(&state->rb_node); |
1530 | while (n) { | |
1531 | state = rb_entry(n, struct extent_state, | |
1532 | rb_node); | |
1533 | if (state->state & bits) | |
1534 | goto got_it; | |
1535 | n = rb_next(n); | |
1536 | } | |
1537 | free_extent_state(*cached_state); | |
1538 | *cached_state = NULL; | |
1539 | goto out; | |
1540 | } | |
1541 | free_extent_state(*cached_state); | |
1542 | *cached_state = NULL; | |
1543 | } | |
1544 | ||
69261c4b | 1545 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1546 | got_it: |
69261c4b | 1547 | if (state) { |
e38e2ed7 | 1548 | cache_state_if_flags(state, cached_state, 0); |
69261c4b XG |
1549 | *start_ret = state->start; |
1550 | *end_ret = state->end; | |
1551 | ret = 0; | |
1552 | } | |
e6138876 | 1553 | out: |
69261c4b XG |
1554 | spin_unlock(&tree->lock); |
1555 | return ret; | |
1556 | } | |
1557 | ||
d352ac68 CM |
1558 | /* |
1559 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1560 | * more than 'max_bytes'. start and end are used to return the range, | |
1561 | * | |
1562 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1563 | */ | |
c8b97818 | 1564 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1565 | u64 *start, u64 *end, u64 max_bytes, |
1566 | struct extent_state **cached_state) | |
d1310b2e CM |
1567 | { |
1568 | struct rb_node *node; | |
1569 | struct extent_state *state; | |
1570 | u64 cur_start = *start; | |
1571 | u64 found = 0; | |
1572 | u64 total_bytes = 0; | |
1573 | ||
cad321ad | 1574 | spin_lock(&tree->lock); |
c8b97818 | 1575 | |
d1310b2e CM |
1576 | /* |
1577 | * this search will find all the extents that end after | |
1578 | * our range starts. | |
1579 | */ | |
80ea96b1 | 1580 | node = tree_search(tree, cur_start); |
2b114d1d | 1581 | if (!node) { |
3b951516 CM |
1582 | if (!found) |
1583 | *end = (u64)-1; | |
d1310b2e CM |
1584 | goto out; |
1585 | } | |
1586 | ||
d397712b | 1587 | while (1) { |
d1310b2e | 1588 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1589 | if (found && (state->start != cur_start || |
1590 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1591 | goto out; |
1592 | } | |
1593 | if (!(state->state & EXTENT_DELALLOC)) { | |
1594 | if (!found) | |
1595 | *end = state->end; | |
1596 | goto out; | |
1597 | } | |
c2a128d2 | 1598 | if (!found) { |
d1310b2e | 1599 | *start = state->start; |
c2a128d2 JB |
1600 | *cached_state = state; |
1601 | atomic_inc(&state->refs); | |
1602 | } | |
d1310b2e CM |
1603 | found++; |
1604 | *end = state->end; | |
1605 | cur_start = state->end + 1; | |
1606 | node = rb_next(node); | |
d1310b2e | 1607 | total_bytes += state->end - state->start + 1; |
7bf811a5 | 1608 | if (total_bytes >= max_bytes) |
573aecaf | 1609 | break; |
573aecaf | 1610 | if (!node) |
d1310b2e CM |
1611 | break; |
1612 | } | |
1613 | out: | |
cad321ad | 1614 | spin_unlock(&tree->lock); |
d1310b2e CM |
1615 | return found; |
1616 | } | |
1617 | ||
143bede5 JM |
1618 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1619 | struct page *locked_page, | |
1620 | u64 start, u64 end) | |
c8b97818 CM |
1621 | { |
1622 | int ret; | |
1623 | struct page *pages[16]; | |
1624 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1625 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1626 | unsigned long nr_pages = end_index - index + 1; | |
1627 | int i; | |
1628 | ||
1629 | if (index == locked_page->index && end_index == index) | |
143bede5 | 1630 | return; |
c8b97818 | 1631 | |
d397712b | 1632 | while (nr_pages > 0) { |
c8b97818 | 1633 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1634 | min_t(unsigned long, nr_pages, |
1635 | ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1636 | for (i = 0; i < ret; i++) { |
1637 | if (pages[i] != locked_page) | |
1638 | unlock_page(pages[i]); | |
1639 | page_cache_release(pages[i]); | |
1640 | } | |
1641 | nr_pages -= ret; | |
1642 | index += ret; | |
1643 | cond_resched(); | |
1644 | } | |
c8b97818 CM |
1645 | } |
1646 | ||
1647 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1648 | struct page *locked_page, | |
1649 | u64 delalloc_start, | |
1650 | u64 delalloc_end) | |
1651 | { | |
1652 | unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT; | |
1653 | unsigned long start_index = index; | |
1654 | unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT; | |
1655 | unsigned long pages_locked = 0; | |
1656 | struct page *pages[16]; | |
1657 | unsigned long nrpages; | |
1658 | int ret; | |
1659 | int i; | |
1660 | ||
1661 | /* the caller is responsible for locking the start index */ | |
1662 | if (index == locked_page->index && index == end_index) | |
1663 | return 0; | |
1664 | ||
1665 | /* skip the page at the start index */ | |
1666 | nrpages = end_index - index + 1; | |
d397712b | 1667 | while (nrpages > 0) { |
c8b97818 | 1668 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1669 | min_t(unsigned long, |
1670 | nrpages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1671 | if (ret == 0) { |
1672 | ret = -EAGAIN; | |
1673 | goto done; | |
1674 | } | |
1675 | /* now we have an array of pages, lock them all */ | |
1676 | for (i = 0; i < ret; i++) { | |
1677 | /* | |
1678 | * the caller is taking responsibility for | |
1679 | * locked_page | |
1680 | */ | |
771ed689 | 1681 | if (pages[i] != locked_page) { |
c8b97818 | 1682 | lock_page(pages[i]); |
f2b1c41c CM |
1683 | if (!PageDirty(pages[i]) || |
1684 | pages[i]->mapping != inode->i_mapping) { | |
771ed689 CM |
1685 | ret = -EAGAIN; |
1686 | unlock_page(pages[i]); | |
1687 | page_cache_release(pages[i]); | |
1688 | goto done; | |
1689 | } | |
1690 | } | |
c8b97818 | 1691 | page_cache_release(pages[i]); |
771ed689 | 1692 | pages_locked++; |
c8b97818 | 1693 | } |
c8b97818 CM |
1694 | nrpages -= ret; |
1695 | index += ret; | |
1696 | cond_resched(); | |
1697 | } | |
1698 | ret = 0; | |
1699 | done: | |
1700 | if (ret && pages_locked) { | |
1701 | __unlock_for_delalloc(inode, locked_page, | |
1702 | delalloc_start, | |
1703 | ((u64)(start_index + pages_locked - 1)) << | |
1704 | PAGE_CACHE_SHIFT); | |
1705 | } | |
1706 | return ret; | |
1707 | } | |
1708 | ||
1709 | /* | |
1710 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1711 | * more than 'max_bytes'. start and end are used to return the range, | |
1712 | * | |
1713 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1714 | */ | |
294e30fe JB |
1715 | STATIC u64 find_lock_delalloc_range(struct inode *inode, |
1716 | struct extent_io_tree *tree, | |
1717 | struct page *locked_page, u64 *start, | |
1718 | u64 *end, u64 max_bytes) | |
c8b97818 CM |
1719 | { |
1720 | u64 delalloc_start; | |
1721 | u64 delalloc_end; | |
1722 | u64 found; | |
9655d298 | 1723 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1724 | int ret; |
1725 | int loops = 0; | |
1726 | ||
1727 | again: | |
1728 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1729 | delalloc_start = *start; | |
1730 | delalloc_end = 0; | |
1731 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1732 | max_bytes, &cached_state); |
70b99e69 | 1733 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1734 | *start = delalloc_start; |
1735 | *end = delalloc_end; | |
c2a128d2 | 1736 | free_extent_state(cached_state); |
385fe0be | 1737 | return 0; |
c8b97818 CM |
1738 | } |
1739 | ||
70b99e69 CM |
1740 | /* |
1741 | * start comes from the offset of locked_page. We have to lock | |
1742 | * pages in order, so we can't process delalloc bytes before | |
1743 | * locked_page | |
1744 | */ | |
d397712b | 1745 | if (delalloc_start < *start) |
70b99e69 | 1746 | delalloc_start = *start; |
70b99e69 | 1747 | |
c8b97818 CM |
1748 | /* |
1749 | * make sure to limit the number of pages we try to lock down | |
c8b97818 | 1750 | */ |
7bf811a5 JB |
1751 | if (delalloc_end + 1 - delalloc_start > max_bytes) |
1752 | delalloc_end = delalloc_start + max_bytes - 1; | |
d397712b | 1753 | |
c8b97818 CM |
1754 | /* step two, lock all the pages after the page that has start */ |
1755 | ret = lock_delalloc_pages(inode, locked_page, | |
1756 | delalloc_start, delalloc_end); | |
1757 | if (ret == -EAGAIN) { | |
1758 | /* some of the pages are gone, lets avoid looping by | |
1759 | * shortening the size of the delalloc range we're searching | |
1760 | */ | |
9655d298 | 1761 | free_extent_state(cached_state); |
7d788742 | 1762 | cached_state = NULL; |
c8b97818 | 1763 | if (!loops) { |
7bf811a5 | 1764 | max_bytes = PAGE_CACHE_SIZE; |
c8b97818 CM |
1765 | loops = 1; |
1766 | goto again; | |
1767 | } else { | |
1768 | found = 0; | |
1769 | goto out_failed; | |
1770 | } | |
1771 | } | |
79787eaa | 1772 | BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */ |
c8b97818 CM |
1773 | |
1774 | /* step three, lock the state bits for the whole range */ | |
d0082371 | 1775 | lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state); |
c8b97818 CM |
1776 | |
1777 | /* then test to make sure it is all still delalloc */ | |
1778 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1779 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1780 | if (!ret) { |
9655d298 CM |
1781 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
1782 | &cached_state, GFP_NOFS); | |
c8b97818 CM |
1783 | __unlock_for_delalloc(inode, locked_page, |
1784 | delalloc_start, delalloc_end); | |
1785 | cond_resched(); | |
1786 | goto again; | |
1787 | } | |
9655d298 | 1788 | free_extent_state(cached_state); |
c8b97818 CM |
1789 | *start = delalloc_start; |
1790 | *end = delalloc_end; | |
1791 | out_failed: | |
1792 | return found; | |
1793 | } | |
1794 | ||
c2790a2e JB |
1795 | int extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end, |
1796 | struct page *locked_page, | |
9ee49a04 | 1797 | unsigned clear_bits, |
c2790a2e | 1798 | unsigned long page_ops) |
c8b97818 | 1799 | { |
c2790a2e | 1800 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
c8b97818 CM |
1801 | int ret; |
1802 | struct page *pages[16]; | |
1803 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1804 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1805 | unsigned long nr_pages = end_index - index + 1; | |
1806 | int i; | |
771ed689 | 1807 | |
2c64c53d | 1808 | clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); |
c2790a2e | 1809 | if (page_ops == 0) |
771ed689 | 1810 | return 0; |
c8b97818 | 1811 | |
704de49d FM |
1812 | if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0) |
1813 | mapping_set_error(inode->i_mapping, -EIO); | |
1814 | ||
d397712b | 1815 | while (nr_pages > 0) { |
c8b97818 | 1816 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1817 | min_t(unsigned long, |
1818 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 | 1819 | for (i = 0; i < ret; i++) { |
8b62b72b | 1820 | |
c2790a2e | 1821 | if (page_ops & PAGE_SET_PRIVATE2) |
8b62b72b CM |
1822 | SetPagePrivate2(pages[i]); |
1823 | ||
c8b97818 CM |
1824 | if (pages[i] == locked_page) { |
1825 | page_cache_release(pages[i]); | |
1826 | continue; | |
1827 | } | |
c2790a2e | 1828 | if (page_ops & PAGE_CLEAR_DIRTY) |
c8b97818 | 1829 | clear_page_dirty_for_io(pages[i]); |
c2790a2e | 1830 | if (page_ops & PAGE_SET_WRITEBACK) |
c8b97818 | 1831 | set_page_writeback(pages[i]); |
704de49d FM |
1832 | if (page_ops & PAGE_SET_ERROR) |
1833 | SetPageError(pages[i]); | |
c2790a2e | 1834 | if (page_ops & PAGE_END_WRITEBACK) |
c8b97818 | 1835 | end_page_writeback(pages[i]); |
c2790a2e | 1836 | if (page_ops & PAGE_UNLOCK) |
771ed689 | 1837 | unlock_page(pages[i]); |
c8b97818 CM |
1838 | page_cache_release(pages[i]); |
1839 | } | |
1840 | nr_pages -= ret; | |
1841 | index += ret; | |
1842 | cond_resched(); | |
1843 | } | |
1844 | return 0; | |
1845 | } | |
c8b97818 | 1846 | |
d352ac68 CM |
1847 | /* |
1848 | * count the number of bytes in the tree that have a given bit(s) | |
1849 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1850 | * cached. The total number found is returned. | |
1851 | */ | |
d1310b2e CM |
1852 | u64 count_range_bits(struct extent_io_tree *tree, |
1853 | u64 *start, u64 search_end, u64 max_bytes, | |
9ee49a04 | 1854 | unsigned bits, int contig) |
d1310b2e CM |
1855 | { |
1856 | struct rb_node *node; | |
1857 | struct extent_state *state; | |
1858 | u64 cur_start = *start; | |
1859 | u64 total_bytes = 0; | |
ec29ed5b | 1860 | u64 last = 0; |
d1310b2e CM |
1861 | int found = 0; |
1862 | ||
fae7f21c | 1863 | if (WARN_ON(search_end <= cur_start)) |
d1310b2e | 1864 | return 0; |
d1310b2e | 1865 | |
cad321ad | 1866 | spin_lock(&tree->lock); |
d1310b2e CM |
1867 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1868 | total_bytes = tree->dirty_bytes; | |
1869 | goto out; | |
1870 | } | |
1871 | /* | |
1872 | * this search will find all the extents that end after | |
1873 | * our range starts. | |
1874 | */ | |
80ea96b1 | 1875 | node = tree_search(tree, cur_start); |
d397712b | 1876 | if (!node) |
d1310b2e | 1877 | goto out; |
d1310b2e | 1878 | |
d397712b | 1879 | while (1) { |
d1310b2e CM |
1880 | state = rb_entry(node, struct extent_state, rb_node); |
1881 | if (state->start > search_end) | |
1882 | break; | |
ec29ed5b CM |
1883 | if (contig && found && state->start > last + 1) |
1884 | break; | |
1885 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1886 | total_bytes += min(search_end, state->end) + 1 - |
1887 | max(cur_start, state->start); | |
1888 | if (total_bytes >= max_bytes) | |
1889 | break; | |
1890 | if (!found) { | |
af60bed2 | 1891 | *start = max(cur_start, state->start); |
d1310b2e CM |
1892 | found = 1; |
1893 | } | |
ec29ed5b CM |
1894 | last = state->end; |
1895 | } else if (contig && found) { | |
1896 | break; | |
d1310b2e CM |
1897 | } |
1898 | node = rb_next(node); | |
1899 | if (!node) | |
1900 | break; | |
1901 | } | |
1902 | out: | |
cad321ad | 1903 | spin_unlock(&tree->lock); |
d1310b2e CM |
1904 | return total_bytes; |
1905 | } | |
b2950863 | 1906 | |
d352ac68 CM |
1907 | /* |
1908 | * set the private field for a given byte offset in the tree. If there isn't | |
1909 | * an extent_state there already, this does nothing. | |
1910 | */ | |
171170c1 | 1911 | static int set_state_private(struct extent_io_tree *tree, u64 start, u64 private) |
d1310b2e CM |
1912 | { |
1913 | struct rb_node *node; | |
1914 | struct extent_state *state; | |
1915 | int ret = 0; | |
1916 | ||
cad321ad | 1917 | spin_lock(&tree->lock); |
d1310b2e CM |
1918 | /* |
1919 | * this search will find all the extents that end after | |
1920 | * our range starts. | |
1921 | */ | |
80ea96b1 | 1922 | node = tree_search(tree, start); |
2b114d1d | 1923 | if (!node) { |
d1310b2e CM |
1924 | ret = -ENOENT; |
1925 | goto out; | |
1926 | } | |
1927 | state = rb_entry(node, struct extent_state, rb_node); | |
1928 | if (state->start != start) { | |
1929 | ret = -ENOENT; | |
1930 | goto out; | |
1931 | } | |
1932 | state->private = private; | |
1933 | out: | |
cad321ad | 1934 | spin_unlock(&tree->lock); |
d1310b2e CM |
1935 | return ret; |
1936 | } | |
1937 | ||
1938 | int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private) | |
1939 | { | |
1940 | struct rb_node *node; | |
1941 | struct extent_state *state; | |
1942 | int ret = 0; | |
1943 | ||
cad321ad | 1944 | spin_lock(&tree->lock); |
d1310b2e CM |
1945 | /* |
1946 | * this search will find all the extents that end after | |
1947 | * our range starts. | |
1948 | */ | |
80ea96b1 | 1949 | node = tree_search(tree, start); |
2b114d1d | 1950 | if (!node) { |
d1310b2e CM |
1951 | ret = -ENOENT; |
1952 | goto out; | |
1953 | } | |
1954 | state = rb_entry(node, struct extent_state, rb_node); | |
1955 | if (state->start != start) { | |
1956 | ret = -ENOENT; | |
1957 | goto out; | |
1958 | } | |
1959 | *private = state->private; | |
1960 | out: | |
cad321ad | 1961 | spin_unlock(&tree->lock); |
d1310b2e CM |
1962 | return ret; |
1963 | } | |
1964 | ||
1965 | /* | |
1966 | * searches a range in the state tree for a given mask. | |
70dec807 | 1967 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1968 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1969 | * range is found set. | |
1970 | */ | |
1971 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1972 | unsigned bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1973 | { |
1974 | struct extent_state *state = NULL; | |
1975 | struct rb_node *node; | |
1976 | int bitset = 0; | |
d1310b2e | 1977 | |
cad321ad | 1978 | spin_lock(&tree->lock); |
27a3507d | 1979 | if (cached && extent_state_in_tree(cached) && cached->start <= start && |
df98b6e2 | 1980 | cached->end > start) |
9655d298 CM |
1981 | node = &cached->rb_node; |
1982 | else | |
1983 | node = tree_search(tree, start); | |
d1310b2e CM |
1984 | while (node && start <= end) { |
1985 | state = rb_entry(node, struct extent_state, rb_node); | |
1986 | ||
1987 | if (filled && state->start > start) { | |
1988 | bitset = 0; | |
1989 | break; | |
1990 | } | |
1991 | ||
1992 | if (state->start > end) | |
1993 | break; | |
1994 | ||
1995 | if (state->state & bits) { | |
1996 | bitset = 1; | |
1997 | if (!filled) | |
1998 | break; | |
1999 | } else if (filled) { | |
2000 | bitset = 0; | |
2001 | break; | |
2002 | } | |
46562cec CM |
2003 | |
2004 | if (state->end == (u64)-1) | |
2005 | break; | |
2006 | ||
d1310b2e CM |
2007 | start = state->end + 1; |
2008 | if (start > end) | |
2009 | break; | |
2010 | node = rb_next(node); | |
2011 | if (!node) { | |
2012 | if (filled) | |
2013 | bitset = 0; | |
2014 | break; | |
2015 | } | |
2016 | } | |
cad321ad | 2017 | spin_unlock(&tree->lock); |
d1310b2e CM |
2018 | return bitset; |
2019 | } | |
d1310b2e CM |
2020 | |
2021 | /* | |
2022 | * helper function to set a given page up to date if all the | |
2023 | * extents in the tree for that page are up to date | |
2024 | */ | |
143bede5 | 2025 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e | 2026 | { |
4eee4fa4 | 2027 | u64 start = page_offset(page); |
d1310b2e | 2028 | u64 end = start + PAGE_CACHE_SIZE - 1; |
9655d298 | 2029 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 2030 | SetPageUptodate(page); |
d1310b2e CM |
2031 | } |
2032 | ||
8b110e39 | 2033 | int free_io_failure(struct inode *inode, struct io_failure_record *rec) |
4a54c8c1 JS |
2034 | { |
2035 | int ret; | |
2036 | int err = 0; | |
2037 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
2038 | ||
2039 | set_state_private(failure_tree, rec->start, 0); | |
2040 | ret = clear_extent_bits(failure_tree, rec->start, | |
2041 | rec->start + rec->len - 1, | |
2042 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
2043 | if (ret) | |
2044 | err = ret; | |
2045 | ||
53b381b3 DW |
2046 | ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start, |
2047 | rec->start + rec->len - 1, | |
2048 | EXTENT_DAMAGED, GFP_NOFS); | |
2049 | if (ret && !err) | |
2050 | err = ret; | |
4a54c8c1 JS |
2051 | |
2052 | kfree(rec); | |
2053 | return err; | |
2054 | } | |
2055 | ||
4a54c8c1 JS |
2056 | /* |
2057 | * this bypasses the standard btrfs submit functions deliberately, as | |
2058 | * the standard behavior is to write all copies in a raid setup. here we only | |
2059 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
2060 | * submit_bio directly. | |
3ec706c8 | 2061 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
2062 | * actually prevents the read that triggered the error from finishing. |
2063 | * currently, there can be no more than two copies of every data bit. thus, | |
2064 | * exactly one rewrite is required. | |
2065 | */ | |
1203b681 MX |
2066 | int repair_io_failure(struct inode *inode, u64 start, u64 length, u64 logical, |
2067 | struct page *page, unsigned int pg_offset, int mirror_num) | |
4a54c8c1 | 2068 | { |
1203b681 | 2069 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; |
4a54c8c1 JS |
2070 | struct bio *bio; |
2071 | struct btrfs_device *dev; | |
4a54c8c1 JS |
2072 | u64 map_length = 0; |
2073 | u64 sector; | |
2074 | struct btrfs_bio *bbio = NULL; | |
53b381b3 | 2075 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
4a54c8c1 JS |
2076 | int ret; |
2077 | ||
908960c6 | 2078 | ASSERT(!(fs_info->sb->s_flags & MS_RDONLY)); |
4a54c8c1 JS |
2079 | BUG_ON(!mirror_num); |
2080 | ||
53b381b3 DW |
2081 | /* we can't repair anything in raid56 yet */ |
2082 | if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num)) | |
2083 | return 0; | |
2084 | ||
9be3395b | 2085 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
4a54c8c1 JS |
2086 | if (!bio) |
2087 | return -EIO; | |
4f024f37 | 2088 | bio->bi_iter.bi_size = 0; |
4a54c8c1 JS |
2089 | map_length = length; |
2090 | ||
3ec706c8 | 2091 | ret = btrfs_map_block(fs_info, WRITE, logical, |
4a54c8c1 JS |
2092 | &map_length, &bbio, mirror_num); |
2093 | if (ret) { | |
2094 | bio_put(bio); | |
2095 | return -EIO; | |
2096 | } | |
2097 | BUG_ON(mirror_num != bbio->mirror_num); | |
2098 | sector = bbio->stripes[mirror_num-1].physical >> 9; | |
4f024f37 | 2099 | bio->bi_iter.bi_sector = sector; |
4a54c8c1 | 2100 | dev = bbio->stripes[mirror_num-1].dev; |
6e9606d2 | 2101 | btrfs_put_bbio(bbio); |
4a54c8c1 JS |
2102 | if (!dev || !dev->bdev || !dev->writeable) { |
2103 | bio_put(bio); | |
2104 | return -EIO; | |
2105 | } | |
2106 | bio->bi_bdev = dev->bdev; | |
ffdd2018 | 2107 | bio_add_page(bio, page, length, pg_offset); |
4a54c8c1 | 2108 | |
33879d45 | 2109 | if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) { |
4a54c8c1 JS |
2110 | /* try to remap that extent elsewhere? */ |
2111 | bio_put(bio); | |
442a4f63 | 2112 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
2113 | return -EIO; |
2114 | } | |
2115 | ||
b14af3b4 DS |
2116 | btrfs_info_rl_in_rcu(fs_info, |
2117 | "read error corrected: ino %llu off %llu (dev %s sector %llu)", | |
1203b681 MX |
2118 | btrfs_ino(inode), start, |
2119 | rcu_str_deref(dev->name), sector); | |
4a54c8c1 JS |
2120 | bio_put(bio); |
2121 | return 0; | |
2122 | } | |
2123 | ||
ea466794 JB |
2124 | int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb, |
2125 | int mirror_num) | |
2126 | { | |
ea466794 JB |
2127 | u64 start = eb->start; |
2128 | unsigned long i, num_pages = num_extent_pages(eb->start, eb->len); | |
d95603b2 | 2129 | int ret = 0; |
ea466794 | 2130 | |
908960c6 ID |
2131 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
2132 | return -EROFS; | |
2133 | ||
ea466794 | 2134 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 2135 | struct page *p = eb->pages[i]; |
1203b681 MX |
2136 | |
2137 | ret = repair_io_failure(root->fs_info->btree_inode, start, | |
2138 | PAGE_CACHE_SIZE, start, p, | |
2139 | start - page_offset(p), mirror_num); | |
ea466794 JB |
2140 | if (ret) |
2141 | break; | |
2142 | start += PAGE_CACHE_SIZE; | |
2143 | } | |
2144 | ||
2145 | return ret; | |
2146 | } | |
2147 | ||
4a54c8c1 JS |
2148 | /* |
2149 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2150 | * to see if we need to process or clean up an io_failure_record | |
2151 | */ | |
8b110e39 MX |
2152 | int clean_io_failure(struct inode *inode, u64 start, struct page *page, |
2153 | unsigned int pg_offset) | |
4a54c8c1 JS |
2154 | { |
2155 | u64 private; | |
2156 | u64 private_failure; | |
2157 | struct io_failure_record *failrec; | |
908960c6 | 2158 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; |
4a54c8c1 JS |
2159 | struct extent_state *state; |
2160 | int num_copies; | |
4a54c8c1 | 2161 | int ret; |
4a54c8c1 JS |
2162 | |
2163 | private = 0; | |
2164 | ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
2165 | (u64)-1, 1, EXTENT_DIRTY, 0); | |
2166 | if (!ret) | |
2167 | return 0; | |
2168 | ||
2169 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start, | |
2170 | &private_failure); | |
2171 | if (ret) | |
2172 | return 0; | |
2173 | ||
2174 | failrec = (struct io_failure_record *)(unsigned long) private_failure; | |
2175 | BUG_ON(!failrec->this_mirror); | |
2176 | ||
2177 | if (failrec->in_validation) { | |
2178 | /* there was no real error, just free the record */ | |
2179 | pr_debug("clean_io_failure: freeing dummy error at %llu\n", | |
2180 | failrec->start); | |
4a54c8c1 JS |
2181 | goto out; |
2182 | } | |
908960c6 ID |
2183 | if (fs_info->sb->s_flags & MS_RDONLY) |
2184 | goto out; | |
4a54c8c1 JS |
2185 | |
2186 | spin_lock(&BTRFS_I(inode)->io_tree.lock); | |
2187 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, | |
2188 | failrec->start, | |
2189 | EXTENT_LOCKED); | |
2190 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); | |
2191 | ||
883d0de4 MX |
2192 | if (state && state->start <= failrec->start && |
2193 | state->end >= failrec->start + failrec->len - 1) { | |
3ec706c8 SB |
2194 | num_copies = btrfs_num_copies(fs_info, failrec->logical, |
2195 | failrec->len); | |
4a54c8c1 | 2196 | if (num_copies > 1) { |
1203b681 | 2197 | repair_io_failure(inode, start, failrec->len, |
454ff3de | 2198 | failrec->logical, page, |
1203b681 | 2199 | pg_offset, failrec->failed_mirror); |
4a54c8c1 JS |
2200 | } |
2201 | } | |
2202 | ||
2203 | out: | |
454ff3de | 2204 | free_io_failure(inode, failrec); |
4a54c8c1 | 2205 | |
454ff3de | 2206 | return 0; |
4a54c8c1 JS |
2207 | } |
2208 | ||
f612496b MX |
2209 | /* |
2210 | * Can be called when | |
2211 | * - hold extent lock | |
2212 | * - under ordered extent | |
2213 | * - the inode is freeing | |
2214 | */ | |
2215 | void btrfs_free_io_failure_record(struct inode *inode, u64 start, u64 end) | |
2216 | { | |
2217 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
2218 | struct io_failure_record *failrec; | |
2219 | struct extent_state *state, *next; | |
2220 | ||
2221 | if (RB_EMPTY_ROOT(&failure_tree->state)) | |
2222 | return; | |
2223 | ||
2224 | spin_lock(&failure_tree->lock); | |
2225 | state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY); | |
2226 | while (state) { | |
2227 | if (state->start > end) | |
2228 | break; | |
2229 | ||
2230 | ASSERT(state->end <= end); | |
2231 | ||
2232 | next = next_state(state); | |
2233 | ||
6e1103a6 | 2234 | failrec = (struct io_failure_record *)(unsigned long)state->private; |
f612496b MX |
2235 | free_extent_state(state); |
2236 | kfree(failrec); | |
2237 | ||
2238 | state = next; | |
2239 | } | |
2240 | spin_unlock(&failure_tree->lock); | |
2241 | } | |
2242 | ||
2fe6303e MX |
2243 | int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end, |
2244 | struct io_failure_record **failrec_ret) | |
4a54c8c1 | 2245 | { |
2fe6303e | 2246 | struct io_failure_record *failrec; |
4a54c8c1 JS |
2247 | u64 private; |
2248 | struct extent_map *em; | |
4a54c8c1 JS |
2249 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2250 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2251 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
4a54c8c1 | 2252 | int ret; |
4a54c8c1 JS |
2253 | u64 logical; |
2254 | ||
4a54c8c1 JS |
2255 | ret = get_state_private(failure_tree, start, &private); |
2256 | if (ret) { | |
2257 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2258 | if (!failrec) | |
2259 | return -ENOMEM; | |
2fe6303e | 2260 | |
4a54c8c1 JS |
2261 | failrec->start = start; |
2262 | failrec->len = end - start + 1; | |
2263 | failrec->this_mirror = 0; | |
2264 | failrec->bio_flags = 0; | |
2265 | failrec->in_validation = 0; | |
2266 | ||
2267 | read_lock(&em_tree->lock); | |
2268 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2269 | if (!em) { | |
2270 | read_unlock(&em_tree->lock); | |
2271 | kfree(failrec); | |
2272 | return -EIO; | |
2273 | } | |
2274 | ||
68ba990f | 2275 | if (em->start > start || em->start + em->len <= start) { |
4a54c8c1 JS |
2276 | free_extent_map(em); |
2277 | em = NULL; | |
2278 | } | |
2279 | read_unlock(&em_tree->lock); | |
7a2d6a64 | 2280 | if (!em) { |
4a54c8c1 JS |
2281 | kfree(failrec); |
2282 | return -EIO; | |
2283 | } | |
2fe6303e | 2284 | |
4a54c8c1 JS |
2285 | logical = start - em->start; |
2286 | logical = em->block_start + logical; | |
2287 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2288 | logical = em->block_start; | |
2289 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2290 | extent_set_compress_type(&failrec->bio_flags, | |
2291 | em->compress_type); | |
2292 | } | |
2fe6303e MX |
2293 | |
2294 | pr_debug("Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu\n", | |
2295 | logical, start, failrec->len); | |
2296 | ||
4a54c8c1 JS |
2297 | failrec->logical = logical; |
2298 | free_extent_map(em); | |
2299 | ||
2300 | /* set the bits in the private failure tree */ | |
2301 | ret = set_extent_bits(failure_tree, start, end, | |
2302 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
2303 | if (ret >= 0) | |
2304 | ret = set_state_private(failure_tree, start, | |
2305 | (u64)(unsigned long)failrec); | |
2306 | /* set the bits in the inode's tree */ | |
2307 | if (ret >= 0) | |
2308 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED, | |
2309 | GFP_NOFS); | |
2310 | if (ret < 0) { | |
2311 | kfree(failrec); | |
2312 | return ret; | |
2313 | } | |
2314 | } else { | |
2315 | failrec = (struct io_failure_record *)(unsigned long)private; | |
2fe6303e | 2316 | pr_debug("Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d\n", |
4a54c8c1 JS |
2317 | failrec->logical, failrec->start, failrec->len, |
2318 | failrec->in_validation); | |
2319 | /* | |
2320 | * when data can be on disk more than twice, add to failrec here | |
2321 | * (e.g. with a list for failed_mirror) to make | |
2322 | * clean_io_failure() clean all those errors at once. | |
2323 | */ | |
2324 | } | |
2fe6303e MX |
2325 | |
2326 | *failrec_ret = failrec; | |
2327 | ||
2328 | return 0; | |
2329 | } | |
2330 | ||
2331 | int btrfs_check_repairable(struct inode *inode, struct bio *failed_bio, | |
2332 | struct io_failure_record *failrec, int failed_mirror) | |
2333 | { | |
2334 | int num_copies; | |
2335 | ||
5d964051 SB |
2336 | num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info, |
2337 | failrec->logical, failrec->len); | |
4a54c8c1 JS |
2338 | if (num_copies == 1) { |
2339 | /* | |
2340 | * we only have a single copy of the data, so don't bother with | |
2341 | * all the retry and error correction code that follows. no | |
2342 | * matter what the error is, it is very likely to persist. | |
2343 | */ | |
2fe6303e | 2344 | pr_debug("Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n", |
09a7f7a2 | 2345 | num_copies, failrec->this_mirror, failed_mirror); |
2fe6303e | 2346 | return 0; |
4a54c8c1 JS |
2347 | } |
2348 | ||
4a54c8c1 JS |
2349 | /* |
2350 | * there are two premises: | |
2351 | * a) deliver good data to the caller | |
2352 | * b) correct the bad sectors on disk | |
2353 | */ | |
2354 | if (failed_bio->bi_vcnt > 1) { | |
2355 | /* | |
2356 | * to fulfill b), we need to know the exact failing sectors, as | |
2357 | * we don't want to rewrite any more than the failed ones. thus, | |
2358 | * we need separate read requests for the failed bio | |
2359 | * | |
2360 | * if the following BUG_ON triggers, our validation request got | |
2361 | * merged. we need separate requests for our algorithm to work. | |
2362 | */ | |
2363 | BUG_ON(failrec->in_validation); | |
2364 | failrec->in_validation = 1; | |
2365 | failrec->this_mirror = failed_mirror; | |
4a54c8c1 JS |
2366 | } else { |
2367 | /* | |
2368 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2369 | * of the failed sector and if we succeed, we have setup | |
2370 | * everything for repair_io_failure to do the rest for us. | |
2371 | */ | |
2372 | if (failrec->in_validation) { | |
2373 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2374 | failrec->in_validation = 0; | |
2375 | failrec->this_mirror = 0; | |
2376 | } | |
2377 | failrec->failed_mirror = failed_mirror; | |
2378 | failrec->this_mirror++; | |
2379 | if (failrec->this_mirror == failed_mirror) | |
2380 | failrec->this_mirror++; | |
4a54c8c1 JS |
2381 | } |
2382 | ||
facc8a22 | 2383 | if (failrec->this_mirror > num_copies) { |
2fe6303e | 2384 | pr_debug("Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n", |
4a54c8c1 | 2385 | num_copies, failrec->this_mirror, failed_mirror); |
2fe6303e | 2386 | return 0; |
4a54c8c1 JS |
2387 | } |
2388 | ||
2fe6303e MX |
2389 | return 1; |
2390 | } | |
2391 | ||
2392 | ||
2393 | struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio, | |
2394 | struct io_failure_record *failrec, | |
2395 | struct page *page, int pg_offset, int icsum, | |
8b110e39 | 2396 | bio_end_io_t *endio_func, void *data) |
2fe6303e MX |
2397 | { |
2398 | struct bio *bio; | |
2399 | struct btrfs_io_bio *btrfs_failed_bio; | |
2400 | struct btrfs_io_bio *btrfs_bio; | |
2401 | ||
9be3395b | 2402 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
2fe6303e MX |
2403 | if (!bio) |
2404 | return NULL; | |
2405 | ||
2406 | bio->bi_end_io = endio_func; | |
4f024f37 | 2407 | bio->bi_iter.bi_sector = failrec->logical >> 9; |
4a54c8c1 | 2408 | bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; |
4f024f37 | 2409 | bio->bi_iter.bi_size = 0; |
8b110e39 | 2410 | bio->bi_private = data; |
4a54c8c1 | 2411 | |
facc8a22 MX |
2412 | btrfs_failed_bio = btrfs_io_bio(failed_bio); |
2413 | if (btrfs_failed_bio->csum) { | |
2414 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2415 | u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); | |
2416 | ||
2417 | btrfs_bio = btrfs_io_bio(bio); | |
2418 | btrfs_bio->csum = btrfs_bio->csum_inline; | |
2fe6303e MX |
2419 | icsum *= csum_size; |
2420 | memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum, | |
facc8a22 MX |
2421 | csum_size); |
2422 | } | |
2423 | ||
2fe6303e MX |
2424 | bio_add_page(bio, page, failrec->len, pg_offset); |
2425 | ||
2426 | return bio; | |
2427 | } | |
2428 | ||
2429 | /* | |
2430 | * this is a generic handler for readpage errors (default | |
2431 | * readpage_io_failed_hook). if other copies exist, read those and write back | |
2432 | * good data to the failed position. does not investigate in remapping the | |
2433 | * failed extent elsewhere, hoping the device will be smart enough to do this as | |
2434 | * needed | |
2435 | */ | |
2436 | ||
2437 | static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset, | |
2438 | struct page *page, u64 start, u64 end, | |
2439 | int failed_mirror) | |
2440 | { | |
2441 | struct io_failure_record *failrec; | |
2442 | struct inode *inode = page->mapping->host; | |
2443 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2444 | struct bio *bio; | |
2445 | int read_mode; | |
2446 | int ret; | |
2447 | ||
2448 | BUG_ON(failed_bio->bi_rw & REQ_WRITE); | |
2449 | ||
2450 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
2451 | if (ret) | |
2452 | return ret; | |
2453 | ||
2454 | ret = btrfs_check_repairable(inode, failed_bio, failrec, failed_mirror); | |
2455 | if (!ret) { | |
2456 | free_io_failure(inode, failrec); | |
2457 | return -EIO; | |
2458 | } | |
2459 | ||
2460 | if (failed_bio->bi_vcnt > 1) | |
2461 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; | |
2462 | else | |
2463 | read_mode = READ_SYNC; | |
2464 | ||
2465 | phy_offset >>= inode->i_sb->s_blocksize_bits; | |
2466 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2467 | start - page_offset(page), | |
8b110e39 MX |
2468 | (int)phy_offset, failed_bio->bi_end_io, |
2469 | NULL); | |
2fe6303e MX |
2470 | if (!bio) { |
2471 | free_io_failure(inode, failrec); | |
2472 | return -EIO; | |
2473 | } | |
4a54c8c1 | 2474 | |
2fe6303e MX |
2475 | pr_debug("Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d\n", |
2476 | read_mode, failrec->this_mirror, failrec->in_validation); | |
4a54c8c1 | 2477 | |
013bd4c3 TI |
2478 | ret = tree->ops->submit_bio_hook(inode, read_mode, bio, |
2479 | failrec->this_mirror, | |
2480 | failrec->bio_flags, 0); | |
6c387ab2 | 2481 | if (ret) { |
454ff3de | 2482 | free_io_failure(inode, failrec); |
6c387ab2 MX |
2483 | bio_put(bio); |
2484 | } | |
2485 | ||
013bd4c3 | 2486 | return ret; |
4a54c8c1 JS |
2487 | } |
2488 | ||
d1310b2e CM |
2489 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2490 | ||
87826df0 JM |
2491 | int end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
2492 | { | |
2493 | int uptodate = (err == 0); | |
2494 | struct extent_io_tree *tree; | |
3e2426bd | 2495 | int ret = 0; |
87826df0 JM |
2496 | |
2497 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
2498 | ||
2499 | if (tree->ops && tree->ops->writepage_end_io_hook) { | |
2500 | ret = tree->ops->writepage_end_io_hook(page, start, | |
2501 | end, NULL, uptodate); | |
2502 | if (ret) | |
2503 | uptodate = 0; | |
2504 | } | |
2505 | ||
87826df0 | 2506 | if (!uptodate) { |
87826df0 JM |
2507 | ClearPageUptodate(page); |
2508 | SetPageError(page); | |
5dca6eea LB |
2509 | ret = ret < 0 ? ret : -EIO; |
2510 | mapping_set_error(page->mapping, ret); | |
87826df0 JM |
2511 | } |
2512 | return 0; | |
2513 | } | |
2514 | ||
d1310b2e CM |
2515 | /* |
2516 | * after a writepage IO is done, we need to: | |
2517 | * clear the uptodate bits on error | |
2518 | * clear the writeback bits in the extent tree for this IO | |
2519 | * end_page_writeback if the page has no more pending IO | |
2520 | * | |
2521 | * Scheduling is not allowed, so the extent state tree is expected | |
2522 | * to have one and only one object corresponding to this IO. | |
2523 | */ | |
4246a0b6 | 2524 | static void end_bio_extent_writepage(struct bio *bio) |
d1310b2e | 2525 | { |
2c30c71b | 2526 | struct bio_vec *bvec; |
d1310b2e CM |
2527 | u64 start; |
2528 | u64 end; | |
2c30c71b | 2529 | int i; |
d1310b2e | 2530 | |
2c30c71b | 2531 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2532 | struct page *page = bvec->bv_page; |
902b22f3 | 2533 | |
17a5adcc AO |
2534 | /* We always issue full-page reads, but if some block |
2535 | * in a page fails to read, blk_update_request() will | |
2536 | * advance bv_offset and adjust bv_len to compensate. | |
2537 | * Print a warning for nonzero offsets, and an error | |
2538 | * if they don't add up to a full page. */ | |
efe120a0 FH |
2539 | if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) { |
2540 | if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE) | |
2541 | btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info, | |
2542 | "partial page write in btrfs with offset %u and length %u", | |
2543 | bvec->bv_offset, bvec->bv_len); | |
2544 | else | |
2545 | btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info, | |
2546 | "incomplete page write in btrfs with offset %u and " | |
2547 | "length %u", | |
2548 | bvec->bv_offset, bvec->bv_len); | |
2549 | } | |
d1310b2e | 2550 | |
17a5adcc AO |
2551 | start = page_offset(page); |
2552 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
d1310b2e | 2553 | |
4246a0b6 | 2554 | if (end_extent_writepage(page, bio->bi_error, start, end)) |
87826df0 | 2555 | continue; |
70dec807 | 2556 | |
17a5adcc | 2557 | end_page_writeback(page); |
2c30c71b | 2558 | } |
2b1f55b0 | 2559 | |
d1310b2e | 2560 | bio_put(bio); |
d1310b2e CM |
2561 | } |
2562 | ||
883d0de4 MX |
2563 | static void |
2564 | endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len, | |
2565 | int uptodate) | |
2566 | { | |
2567 | struct extent_state *cached = NULL; | |
2568 | u64 end = start + len - 1; | |
2569 | ||
2570 | if (uptodate && tree->track_uptodate) | |
2571 | set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC); | |
2572 | unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC); | |
2573 | } | |
2574 | ||
d1310b2e CM |
2575 | /* |
2576 | * after a readpage IO is done, we need to: | |
2577 | * clear the uptodate bits on error | |
2578 | * set the uptodate bits if things worked | |
2579 | * set the page up to date if all extents in the tree are uptodate | |
2580 | * clear the lock bit in the extent tree | |
2581 | * unlock the page if there are no other extents locked for it | |
2582 | * | |
2583 | * Scheduling is not allowed, so the extent state tree is expected | |
2584 | * to have one and only one object corresponding to this IO. | |
2585 | */ | |
4246a0b6 | 2586 | static void end_bio_extent_readpage(struct bio *bio) |
d1310b2e | 2587 | { |
2c30c71b | 2588 | struct bio_vec *bvec; |
4246a0b6 | 2589 | int uptodate = !bio->bi_error; |
facc8a22 | 2590 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); |
902b22f3 | 2591 | struct extent_io_tree *tree; |
facc8a22 | 2592 | u64 offset = 0; |
d1310b2e CM |
2593 | u64 start; |
2594 | u64 end; | |
facc8a22 | 2595 | u64 len; |
883d0de4 MX |
2596 | u64 extent_start = 0; |
2597 | u64 extent_len = 0; | |
5cf1ab56 | 2598 | int mirror; |
d1310b2e | 2599 | int ret; |
2c30c71b | 2600 | int i; |
d1310b2e | 2601 | |
2c30c71b | 2602 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2603 | struct page *page = bvec->bv_page; |
a71754fc | 2604 | struct inode *inode = page->mapping->host; |
507903b8 | 2605 | |
be3940c0 | 2606 | pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, " |
4246a0b6 CH |
2607 | "mirror=%u\n", (u64)bio->bi_iter.bi_sector, |
2608 | bio->bi_error, io_bio->mirror_num); | |
a71754fc | 2609 | tree = &BTRFS_I(inode)->io_tree; |
902b22f3 | 2610 | |
17a5adcc AO |
2611 | /* We always issue full-page reads, but if some block |
2612 | * in a page fails to read, blk_update_request() will | |
2613 | * advance bv_offset and adjust bv_len to compensate. | |
2614 | * Print a warning for nonzero offsets, and an error | |
2615 | * if they don't add up to a full page. */ | |
efe120a0 FH |
2616 | if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) { |
2617 | if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE) | |
2618 | btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info, | |
2619 | "partial page read in btrfs with offset %u and length %u", | |
2620 | bvec->bv_offset, bvec->bv_len); | |
2621 | else | |
2622 | btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info, | |
2623 | "incomplete page read in btrfs with offset %u and " | |
2624 | "length %u", | |
2625 | bvec->bv_offset, bvec->bv_len); | |
2626 | } | |
d1310b2e | 2627 | |
17a5adcc AO |
2628 | start = page_offset(page); |
2629 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
facc8a22 | 2630 | len = bvec->bv_len; |
d1310b2e | 2631 | |
9be3395b | 2632 | mirror = io_bio->mirror_num; |
f2a09da9 MX |
2633 | if (likely(uptodate && tree->ops && |
2634 | tree->ops->readpage_end_io_hook)) { | |
facc8a22 MX |
2635 | ret = tree->ops->readpage_end_io_hook(io_bio, offset, |
2636 | page, start, end, | |
2637 | mirror); | |
5ee0844d | 2638 | if (ret) |
d1310b2e | 2639 | uptodate = 0; |
5ee0844d | 2640 | else |
1203b681 | 2641 | clean_io_failure(inode, start, page, 0); |
d1310b2e | 2642 | } |
ea466794 | 2643 | |
f2a09da9 MX |
2644 | if (likely(uptodate)) |
2645 | goto readpage_ok; | |
2646 | ||
2647 | if (tree->ops && tree->ops->readpage_io_failed_hook) { | |
5cf1ab56 | 2648 | ret = tree->ops->readpage_io_failed_hook(page, mirror); |
4246a0b6 | 2649 | if (!ret && !bio->bi_error) |
ea466794 | 2650 | uptodate = 1; |
f2a09da9 | 2651 | } else { |
f4a8e656 JS |
2652 | /* |
2653 | * The generic bio_readpage_error handles errors the | |
2654 | * following way: If possible, new read requests are | |
2655 | * created and submitted and will end up in | |
2656 | * end_bio_extent_readpage as well (if we're lucky, not | |
2657 | * in the !uptodate case). In that case it returns 0 and | |
2658 | * we just go on with the next page in our bio. If it | |
2659 | * can't handle the error it will return -EIO and we | |
2660 | * remain responsible for that page. | |
2661 | */ | |
facc8a22 MX |
2662 | ret = bio_readpage_error(bio, offset, page, start, end, |
2663 | mirror); | |
7e38326f | 2664 | if (ret == 0) { |
4246a0b6 | 2665 | uptodate = !bio->bi_error; |
38c1c2e4 | 2666 | offset += len; |
7e38326f CM |
2667 | continue; |
2668 | } | |
2669 | } | |
f2a09da9 | 2670 | readpage_ok: |
883d0de4 | 2671 | if (likely(uptodate)) { |
a71754fc JB |
2672 | loff_t i_size = i_size_read(inode); |
2673 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
a583c026 | 2674 | unsigned off; |
a71754fc JB |
2675 | |
2676 | /* Zero out the end if this page straddles i_size */ | |
a583c026 LB |
2677 | off = i_size & (PAGE_CACHE_SIZE-1); |
2678 | if (page->index == end_index && off) | |
2679 | zero_user_segment(page, off, PAGE_CACHE_SIZE); | |
17a5adcc | 2680 | SetPageUptodate(page); |
70dec807 | 2681 | } else { |
17a5adcc AO |
2682 | ClearPageUptodate(page); |
2683 | SetPageError(page); | |
70dec807 | 2684 | } |
17a5adcc | 2685 | unlock_page(page); |
facc8a22 | 2686 | offset += len; |
883d0de4 MX |
2687 | |
2688 | if (unlikely(!uptodate)) { | |
2689 | if (extent_len) { | |
2690 | endio_readpage_release_extent(tree, | |
2691 | extent_start, | |
2692 | extent_len, 1); | |
2693 | extent_start = 0; | |
2694 | extent_len = 0; | |
2695 | } | |
2696 | endio_readpage_release_extent(tree, start, | |
2697 | end - start + 1, 0); | |
2698 | } else if (!extent_len) { | |
2699 | extent_start = start; | |
2700 | extent_len = end + 1 - start; | |
2701 | } else if (extent_start + extent_len == start) { | |
2702 | extent_len += end + 1 - start; | |
2703 | } else { | |
2704 | endio_readpage_release_extent(tree, extent_start, | |
2705 | extent_len, uptodate); | |
2706 | extent_start = start; | |
2707 | extent_len = end + 1 - start; | |
2708 | } | |
2c30c71b | 2709 | } |
d1310b2e | 2710 | |
883d0de4 MX |
2711 | if (extent_len) |
2712 | endio_readpage_release_extent(tree, extent_start, extent_len, | |
2713 | uptodate); | |
facc8a22 | 2714 | if (io_bio->end_io) |
4246a0b6 | 2715 | io_bio->end_io(io_bio, bio->bi_error); |
d1310b2e | 2716 | bio_put(bio); |
d1310b2e CM |
2717 | } |
2718 | ||
9be3395b CM |
2719 | /* |
2720 | * this allocates from the btrfs_bioset. We're returning a bio right now | |
2721 | * but you can call btrfs_io_bio for the appropriate container_of magic | |
2722 | */ | |
88f794ed MX |
2723 | struct bio * |
2724 | btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
2725 | gfp_t gfp_flags) | |
d1310b2e | 2726 | { |
facc8a22 | 2727 | struct btrfs_io_bio *btrfs_bio; |
d1310b2e CM |
2728 | struct bio *bio; |
2729 | ||
9be3395b | 2730 | bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset); |
d1310b2e CM |
2731 | |
2732 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
9be3395b CM |
2733 | while (!bio && (nr_vecs /= 2)) { |
2734 | bio = bio_alloc_bioset(gfp_flags, | |
2735 | nr_vecs, btrfs_bioset); | |
2736 | } | |
d1310b2e CM |
2737 | } |
2738 | ||
2739 | if (bio) { | |
2740 | bio->bi_bdev = bdev; | |
4f024f37 | 2741 | bio->bi_iter.bi_sector = first_sector; |
facc8a22 MX |
2742 | btrfs_bio = btrfs_io_bio(bio); |
2743 | btrfs_bio->csum = NULL; | |
2744 | btrfs_bio->csum_allocated = NULL; | |
2745 | btrfs_bio->end_io = NULL; | |
d1310b2e CM |
2746 | } |
2747 | return bio; | |
2748 | } | |
2749 | ||
9be3395b CM |
2750 | struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask) |
2751 | { | |
23ea8e5a MX |
2752 | struct btrfs_io_bio *btrfs_bio; |
2753 | struct bio *new; | |
9be3395b | 2754 | |
23ea8e5a MX |
2755 | new = bio_clone_bioset(bio, gfp_mask, btrfs_bioset); |
2756 | if (new) { | |
2757 | btrfs_bio = btrfs_io_bio(new); | |
2758 | btrfs_bio->csum = NULL; | |
2759 | btrfs_bio->csum_allocated = NULL; | |
2760 | btrfs_bio->end_io = NULL; | |
3a9508b0 CM |
2761 | |
2762 | #ifdef CONFIG_BLK_CGROUP | |
da2f0f74 CM |
2763 | /* FIXME, put this into bio_clone_bioset */ |
2764 | if (bio->bi_css) | |
2765 | bio_associate_blkcg(new, bio->bi_css); | |
3a9508b0 | 2766 | #endif |
23ea8e5a MX |
2767 | } |
2768 | return new; | |
2769 | } | |
9be3395b CM |
2770 | |
2771 | /* this also allocates from the btrfs_bioset */ | |
2772 | struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs) | |
2773 | { | |
facc8a22 MX |
2774 | struct btrfs_io_bio *btrfs_bio; |
2775 | struct bio *bio; | |
2776 | ||
2777 | bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset); | |
2778 | if (bio) { | |
2779 | btrfs_bio = btrfs_io_bio(bio); | |
2780 | btrfs_bio->csum = NULL; | |
2781 | btrfs_bio->csum_allocated = NULL; | |
2782 | btrfs_bio->end_io = NULL; | |
2783 | } | |
2784 | return bio; | |
9be3395b CM |
2785 | } |
2786 | ||
2787 | ||
355808c2 JM |
2788 | static int __must_check submit_one_bio(int rw, struct bio *bio, |
2789 | int mirror_num, unsigned long bio_flags) | |
d1310b2e | 2790 | { |
d1310b2e | 2791 | int ret = 0; |
70dec807 CM |
2792 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
2793 | struct page *page = bvec->bv_page; | |
2794 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 2795 | u64 start; |
70dec807 | 2796 | |
4eee4fa4 | 2797 | start = page_offset(page) + bvec->bv_offset; |
70dec807 | 2798 | |
902b22f3 | 2799 | bio->bi_private = NULL; |
d1310b2e CM |
2800 | |
2801 | bio_get(bio); | |
2802 | ||
065631f6 | 2803 | if (tree->ops && tree->ops->submit_bio_hook) |
6b82ce8d | 2804 | ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio, |
eaf25d93 | 2805 | mirror_num, bio_flags, start); |
0b86a832 | 2806 | else |
21adbd5c | 2807 | btrfsic_submit_bio(rw, bio); |
4a54c8c1 | 2808 | |
d1310b2e CM |
2809 | bio_put(bio); |
2810 | return ret; | |
2811 | } | |
2812 | ||
64a16701 | 2813 | static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page, |
3444a972 JM |
2814 | unsigned long offset, size_t size, struct bio *bio, |
2815 | unsigned long bio_flags) | |
2816 | { | |
2817 | int ret = 0; | |
2818 | if (tree->ops && tree->ops->merge_bio_hook) | |
64a16701 | 2819 | ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio, |
3444a972 JM |
2820 | bio_flags); |
2821 | BUG_ON(ret < 0); | |
2822 | return ret; | |
2823 | ||
2824 | } | |
2825 | ||
d1310b2e | 2826 | static int submit_extent_page(int rw, struct extent_io_tree *tree, |
da2f0f74 | 2827 | struct writeback_control *wbc, |
d1310b2e CM |
2828 | struct page *page, sector_t sector, |
2829 | size_t size, unsigned long offset, | |
2830 | struct block_device *bdev, | |
2831 | struct bio **bio_ret, | |
2832 | unsigned long max_pages, | |
f188591e | 2833 | bio_end_io_t end_io_func, |
c8b97818 CM |
2834 | int mirror_num, |
2835 | unsigned long prev_bio_flags, | |
005efedf FM |
2836 | unsigned long bio_flags, |
2837 | bool force_bio_submit) | |
d1310b2e CM |
2838 | { |
2839 | int ret = 0; | |
2840 | struct bio *bio; | |
c8b97818 | 2841 | int contig = 0; |
c8b97818 | 2842 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; |
5b050f04 | 2843 | size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE); |
d1310b2e CM |
2844 | |
2845 | if (bio_ret && *bio_ret) { | |
2846 | bio = *bio_ret; | |
c8b97818 | 2847 | if (old_compressed) |
4f024f37 | 2848 | contig = bio->bi_iter.bi_sector == sector; |
c8b97818 | 2849 | else |
f73a1c7d | 2850 | contig = bio_end_sector(bio) == sector; |
c8b97818 CM |
2851 | |
2852 | if (prev_bio_flags != bio_flags || !contig || | |
005efedf | 2853 | force_bio_submit || |
64a16701 | 2854 | merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) || |
c8b97818 CM |
2855 | bio_add_page(bio, page, page_size, offset) < page_size) { |
2856 | ret = submit_one_bio(rw, bio, mirror_num, | |
2857 | prev_bio_flags); | |
289454ad NA |
2858 | if (ret < 0) { |
2859 | *bio_ret = NULL; | |
79787eaa | 2860 | return ret; |
289454ad | 2861 | } |
d1310b2e CM |
2862 | bio = NULL; |
2863 | } else { | |
da2f0f74 CM |
2864 | if (wbc) |
2865 | wbc_account_io(wbc, page, page_size); | |
d1310b2e CM |
2866 | return 0; |
2867 | } | |
2868 | } | |
c8b97818 | 2869 | |
b54ffb73 KO |
2870 | bio = btrfs_bio_alloc(bdev, sector, BIO_MAX_PAGES, |
2871 | GFP_NOFS | __GFP_HIGH); | |
5df67083 TI |
2872 | if (!bio) |
2873 | return -ENOMEM; | |
70dec807 | 2874 | |
c8b97818 | 2875 | bio_add_page(bio, page, page_size, offset); |
d1310b2e CM |
2876 | bio->bi_end_io = end_io_func; |
2877 | bio->bi_private = tree; | |
da2f0f74 CM |
2878 | if (wbc) { |
2879 | wbc_init_bio(wbc, bio); | |
2880 | wbc_account_io(wbc, page, page_size); | |
2881 | } | |
70dec807 | 2882 | |
d397712b | 2883 | if (bio_ret) |
d1310b2e | 2884 | *bio_ret = bio; |
d397712b | 2885 | else |
c8b97818 | 2886 | ret = submit_one_bio(rw, bio, mirror_num, bio_flags); |
d1310b2e CM |
2887 | |
2888 | return ret; | |
2889 | } | |
2890 | ||
48a3b636 ES |
2891 | static void attach_extent_buffer_page(struct extent_buffer *eb, |
2892 | struct page *page) | |
d1310b2e CM |
2893 | { |
2894 | if (!PagePrivate(page)) { | |
2895 | SetPagePrivate(page); | |
d1310b2e | 2896 | page_cache_get(page); |
4f2de97a JB |
2897 | set_page_private(page, (unsigned long)eb); |
2898 | } else { | |
2899 | WARN_ON(page->private != (unsigned long)eb); | |
d1310b2e CM |
2900 | } |
2901 | } | |
2902 | ||
4f2de97a | 2903 | void set_page_extent_mapped(struct page *page) |
d1310b2e | 2904 | { |
4f2de97a JB |
2905 | if (!PagePrivate(page)) { |
2906 | SetPagePrivate(page); | |
2907 | page_cache_get(page); | |
2908 | set_page_private(page, EXTENT_PAGE_PRIVATE); | |
2909 | } | |
d1310b2e CM |
2910 | } |
2911 | ||
125bac01 MX |
2912 | static struct extent_map * |
2913 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
2914 | u64 start, u64 len, get_extent_t *get_extent, | |
2915 | struct extent_map **em_cached) | |
2916 | { | |
2917 | struct extent_map *em; | |
2918 | ||
2919 | if (em_cached && *em_cached) { | |
2920 | em = *em_cached; | |
cbc0e928 | 2921 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 MX |
2922 | start < extent_map_end(em)) { |
2923 | atomic_inc(&em->refs); | |
2924 | return em; | |
2925 | } | |
2926 | ||
2927 | free_extent_map(em); | |
2928 | *em_cached = NULL; | |
2929 | } | |
2930 | ||
2931 | em = get_extent(inode, page, pg_offset, start, len, 0); | |
2932 | if (em_cached && !IS_ERR_OR_NULL(em)) { | |
2933 | BUG_ON(*em_cached); | |
2934 | atomic_inc(&em->refs); | |
2935 | *em_cached = em; | |
2936 | } | |
2937 | return em; | |
2938 | } | |
d1310b2e CM |
2939 | /* |
2940 | * basic readpage implementation. Locked extent state structs are inserted | |
2941 | * into the tree that are removed when the IO is done (by the end_io | |
2942 | * handlers) | |
79787eaa | 2943 | * XXX JDM: This needs looking at to ensure proper page locking |
d1310b2e | 2944 | */ |
9974090b MX |
2945 | static int __do_readpage(struct extent_io_tree *tree, |
2946 | struct page *page, | |
2947 | get_extent_t *get_extent, | |
125bac01 | 2948 | struct extent_map **em_cached, |
9974090b | 2949 | struct bio **bio, int mirror_num, |
005efedf FM |
2950 | unsigned long *bio_flags, int rw, |
2951 | u64 *prev_em_start) | |
d1310b2e CM |
2952 | { |
2953 | struct inode *inode = page->mapping->host; | |
4eee4fa4 | 2954 | u64 start = page_offset(page); |
d1310b2e CM |
2955 | u64 page_end = start + PAGE_CACHE_SIZE - 1; |
2956 | u64 end; | |
2957 | u64 cur = start; | |
2958 | u64 extent_offset; | |
2959 | u64 last_byte = i_size_read(inode); | |
2960 | u64 block_start; | |
2961 | u64 cur_end; | |
2962 | sector_t sector; | |
2963 | struct extent_map *em; | |
2964 | struct block_device *bdev; | |
2965 | int ret; | |
2966 | int nr = 0; | |
4b384318 | 2967 | int parent_locked = *bio_flags & EXTENT_BIO_PARENT_LOCKED; |
306e16ce | 2968 | size_t pg_offset = 0; |
d1310b2e | 2969 | size_t iosize; |
c8b97818 | 2970 | size_t disk_io_size; |
d1310b2e | 2971 | size_t blocksize = inode->i_sb->s_blocksize; |
4b384318 | 2972 | unsigned long this_bio_flag = *bio_flags & EXTENT_BIO_PARENT_LOCKED; |
d1310b2e CM |
2973 | |
2974 | set_page_extent_mapped(page); | |
2975 | ||
9974090b | 2976 | end = page_end; |
90a887c9 DM |
2977 | if (!PageUptodate(page)) { |
2978 | if (cleancache_get_page(page) == 0) { | |
2979 | BUG_ON(blocksize != PAGE_SIZE); | |
9974090b | 2980 | unlock_extent(tree, start, end); |
90a887c9 DM |
2981 | goto out; |
2982 | } | |
2983 | } | |
2984 | ||
c8b97818 CM |
2985 | if (page->index == last_byte >> PAGE_CACHE_SHIFT) { |
2986 | char *userpage; | |
2987 | size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1); | |
2988 | ||
2989 | if (zero_offset) { | |
2990 | iosize = PAGE_CACHE_SIZE - zero_offset; | |
7ac687d9 | 2991 | userpage = kmap_atomic(page); |
c8b97818 CM |
2992 | memset(userpage + zero_offset, 0, iosize); |
2993 | flush_dcache_page(page); | |
7ac687d9 | 2994 | kunmap_atomic(userpage); |
c8b97818 CM |
2995 | } |
2996 | } | |
d1310b2e | 2997 | while (cur <= end) { |
c8f2f24b | 2998 | unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1; |
005efedf | 2999 | bool force_bio_submit = false; |
c8f2f24b | 3000 | |
d1310b2e CM |
3001 | if (cur >= last_byte) { |
3002 | char *userpage; | |
507903b8 AJ |
3003 | struct extent_state *cached = NULL; |
3004 | ||
306e16ce | 3005 | iosize = PAGE_CACHE_SIZE - pg_offset; |
7ac687d9 | 3006 | userpage = kmap_atomic(page); |
306e16ce | 3007 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3008 | flush_dcache_page(page); |
7ac687d9 | 3009 | kunmap_atomic(userpage); |
d1310b2e | 3010 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 3011 | &cached, GFP_NOFS); |
4b384318 MF |
3012 | if (!parent_locked) |
3013 | unlock_extent_cached(tree, cur, | |
3014 | cur + iosize - 1, | |
3015 | &cached, GFP_NOFS); | |
d1310b2e CM |
3016 | break; |
3017 | } | |
125bac01 MX |
3018 | em = __get_extent_map(inode, page, pg_offset, cur, |
3019 | end - cur + 1, get_extent, em_cached); | |
c704005d | 3020 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3021 | SetPageError(page); |
4b384318 MF |
3022 | if (!parent_locked) |
3023 | unlock_extent(tree, cur, end); | |
d1310b2e CM |
3024 | break; |
3025 | } | |
d1310b2e CM |
3026 | extent_offset = cur - em->start; |
3027 | BUG_ON(extent_map_end(em) <= cur); | |
3028 | BUG_ON(end < cur); | |
3029 | ||
261507a0 | 3030 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
4b384318 | 3031 | this_bio_flag |= EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
3032 | extent_set_compress_type(&this_bio_flag, |
3033 | em->compress_type); | |
3034 | } | |
c8b97818 | 3035 | |
d1310b2e CM |
3036 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
3037 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 3038 | iosize = ALIGN(iosize, blocksize); |
c8b97818 CM |
3039 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
3040 | disk_io_size = em->block_len; | |
3041 | sector = em->block_start >> 9; | |
3042 | } else { | |
3043 | sector = (em->block_start + extent_offset) >> 9; | |
3044 | disk_io_size = iosize; | |
3045 | } | |
d1310b2e CM |
3046 | bdev = em->bdev; |
3047 | block_start = em->block_start; | |
d899e052 YZ |
3048 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
3049 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
3050 | |
3051 | /* | |
3052 | * If we have a file range that points to a compressed extent | |
3053 | * and it's followed by a consecutive file range that points to | |
3054 | * to the same compressed extent (possibly with a different | |
3055 | * offset and/or length, so it either points to the whole extent | |
3056 | * or only part of it), we must make sure we do not submit a | |
3057 | * single bio to populate the pages for the 2 ranges because | |
3058 | * this makes the compressed extent read zero out the pages | |
3059 | * belonging to the 2nd range. Imagine the following scenario: | |
3060 | * | |
3061 | * File layout | |
3062 | * [0 - 8K] [8K - 24K] | |
3063 | * | | | |
3064 | * | | | |
3065 | * points to extent X, points to extent X, | |
3066 | * offset 4K, length of 8K offset 0, length 16K | |
3067 | * | |
3068 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
3069 | * | |
3070 | * If the bio to read the compressed extent covers both ranges, | |
3071 | * it will decompress extent X into the pages belonging to the | |
3072 | * first range and then it will stop, zeroing out the remaining | |
3073 | * pages that belong to the other range that points to extent X. | |
3074 | * So here we make sure we submit 2 bios, one for the first | |
3075 | * range and another one for the third range. Both will target | |
3076 | * the same physical extent from disk, but we can't currently | |
3077 | * make the compressed bio endio callback populate the pages | |
3078 | * for both ranges because each compressed bio is tightly | |
3079 | * coupled with a single extent map, and each range can have | |
3080 | * an extent map with a different offset value relative to the | |
3081 | * uncompressed data of our extent and different lengths. This | |
3082 | * is a corner case so we prioritize correctness over | |
3083 | * non-optimal behavior (submitting 2 bios for the same extent). | |
3084 | */ | |
3085 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
3086 | prev_em_start && *prev_em_start != (u64)-1 && | |
3087 | *prev_em_start != em->orig_start) | |
3088 | force_bio_submit = true; | |
3089 | ||
3090 | if (prev_em_start) | |
3091 | *prev_em_start = em->orig_start; | |
3092 | ||
d1310b2e CM |
3093 | free_extent_map(em); |
3094 | em = NULL; | |
3095 | ||
3096 | /* we've found a hole, just zero and go on */ | |
3097 | if (block_start == EXTENT_MAP_HOLE) { | |
3098 | char *userpage; | |
507903b8 AJ |
3099 | struct extent_state *cached = NULL; |
3100 | ||
7ac687d9 | 3101 | userpage = kmap_atomic(page); |
306e16ce | 3102 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3103 | flush_dcache_page(page); |
7ac687d9 | 3104 | kunmap_atomic(userpage); |
d1310b2e CM |
3105 | |
3106 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 | 3107 | &cached, GFP_NOFS); |
5e6ecb36 FM |
3108 | if (parent_locked) |
3109 | free_extent_state(cached); | |
3110 | else | |
3111 | unlock_extent_cached(tree, cur, | |
3112 | cur + iosize - 1, | |
3113 | &cached, GFP_NOFS); | |
d1310b2e | 3114 | cur = cur + iosize; |
306e16ce | 3115 | pg_offset += iosize; |
d1310b2e CM |
3116 | continue; |
3117 | } | |
3118 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
3119 | if (test_range_bit(tree, cur, cur_end, |
3120 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 3121 | check_page_uptodate(tree, page); |
4b384318 MF |
3122 | if (!parent_locked) |
3123 | unlock_extent(tree, cur, cur + iosize - 1); | |
d1310b2e | 3124 | cur = cur + iosize; |
306e16ce | 3125 | pg_offset += iosize; |
d1310b2e CM |
3126 | continue; |
3127 | } | |
70dec807 CM |
3128 | /* we have an inline extent but it didn't get marked up |
3129 | * to date. Error out | |
3130 | */ | |
3131 | if (block_start == EXTENT_MAP_INLINE) { | |
3132 | SetPageError(page); | |
4b384318 MF |
3133 | if (!parent_locked) |
3134 | unlock_extent(tree, cur, cur + iosize - 1); | |
70dec807 | 3135 | cur = cur + iosize; |
306e16ce | 3136 | pg_offset += iosize; |
70dec807 CM |
3137 | continue; |
3138 | } | |
d1310b2e | 3139 | |
c8f2f24b | 3140 | pnr -= page->index; |
da2f0f74 | 3141 | ret = submit_extent_page(rw, tree, NULL, page, |
306e16ce | 3142 | sector, disk_io_size, pg_offset, |
89642229 | 3143 | bdev, bio, pnr, |
c8b97818 CM |
3144 | end_bio_extent_readpage, mirror_num, |
3145 | *bio_flags, | |
005efedf FM |
3146 | this_bio_flag, |
3147 | force_bio_submit); | |
c8f2f24b JB |
3148 | if (!ret) { |
3149 | nr++; | |
3150 | *bio_flags = this_bio_flag; | |
3151 | } else { | |
d1310b2e | 3152 | SetPageError(page); |
4b384318 MF |
3153 | if (!parent_locked) |
3154 | unlock_extent(tree, cur, cur + iosize - 1); | |
edd33c99 | 3155 | } |
d1310b2e | 3156 | cur = cur + iosize; |
306e16ce | 3157 | pg_offset += iosize; |
d1310b2e | 3158 | } |
90a887c9 | 3159 | out: |
d1310b2e CM |
3160 | if (!nr) { |
3161 | if (!PageError(page)) | |
3162 | SetPageUptodate(page); | |
3163 | unlock_page(page); | |
3164 | } | |
3165 | return 0; | |
3166 | } | |
3167 | ||
9974090b MX |
3168 | static inline void __do_contiguous_readpages(struct extent_io_tree *tree, |
3169 | struct page *pages[], int nr_pages, | |
3170 | u64 start, u64 end, | |
3171 | get_extent_t *get_extent, | |
125bac01 | 3172 | struct extent_map **em_cached, |
9974090b | 3173 | struct bio **bio, int mirror_num, |
808f80b4 FM |
3174 | unsigned long *bio_flags, int rw, |
3175 | u64 *prev_em_start) | |
9974090b MX |
3176 | { |
3177 | struct inode *inode; | |
3178 | struct btrfs_ordered_extent *ordered; | |
3179 | int index; | |
3180 | ||
3181 | inode = pages[0]->mapping->host; | |
3182 | while (1) { | |
3183 | lock_extent(tree, start, end); | |
3184 | ordered = btrfs_lookup_ordered_range(inode, start, | |
3185 | end - start + 1); | |
3186 | if (!ordered) | |
3187 | break; | |
3188 | unlock_extent(tree, start, end); | |
3189 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3190 | btrfs_put_ordered_extent(ordered); | |
3191 | } | |
3192 | ||
3193 | for (index = 0; index < nr_pages; index++) { | |
125bac01 | 3194 | __do_readpage(tree, pages[index], get_extent, em_cached, bio, |
808f80b4 | 3195 | mirror_num, bio_flags, rw, prev_em_start); |
9974090b MX |
3196 | page_cache_release(pages[index]); |
3197 | } | |
3198 | } | |
3199 | ||
3200 | static void __extent_readpages(struct extent_io_tree *tree, | |
3201 | struct page *pages[], | |
3202 | int nr_pages, get_extent_t *get_extent, | |
125bac01 | 3203 | struct extent_map **em_cached, |
9974090b | 3204 | struct bio **bio, int mirror_num, |
808f80b4 FM |
3205 | unsigned long *bio_flags, int rw, |
3206 | u64 *prev_em_start) | |
9974090b | 3207 | { |
35a3621b | 3208 | u64 start = 0; |
9974090b MX |
3209 | u64 end = 0; |
3210 | u64 page_start; | |
3211 | int index; | |
35a3621b | 3212 | int first_index = 0; |
9974090b MX |
3213 | |
3214 | for (index = 0; index < nr_pages; index++) { | |
3215 | page_start = page_offset(pages[index]); | |
3216 | if (!end) { | |
3217 | start = page_start; | |
3218 | end = start + PAGE_CACHE_SIZE - 1; | |
3219 | first_index = index; | |
3220 | } else if (end + 1 == page_start) { | |
3221 | end += PAGE_CACHE_SIZE; | |
3222 | } else { | |
3223 | __do_contiguous_readpages(tree, &pages[first_index], | |
3224 | index - first_index, start, | |
125bac01 MX |
3225 | end, get_extent, em_cached, |
3226 | bio, mirror_num, bio_flags, | |
808f80b4 | 3227 | rw, prev_em_start); |
9974090b MX |
3228 | start = page_start; |
3229 | end = start + PAGE_CACHE_SIZE - 1; | |
3230 | first_index = index; | |
3231 | } | |
3232 | } | |
3233 | ||
3234 | if (end) | |
3235 | __do_contiguous_readpages(tree, &pages[first_index], | |
3236 | index - first_index, start, | |
125bac01 | 3237 | end, get_extent, em_cached, bio, |
808f80b4 FM |
3238 | mirror_num, bio_flags, rw, |
3239 | prev_em_start); | |
9974090b MX |
3240 | } |
3241 | ||
3242 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
3243 | struct page *page, | |
3244 | get_extent_t *get_extent, | |
3245 | struct bio **bio, int mirror_num, | |
3246 | unsigned long *bio_flags, int rw) | |
3247 | { | |
3248 | struct inode *inode = page->mapping->host; | |
3249 | struct btrfs_ordered_extent *ordered; | |
3250 | u64 start = page_offset(page); | |
3251 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
3252 | int ret; | |
3253 | ||
3254 | while (1) { | |
3255 | lock_extent(tree, start, end); | |
3256 | ordered = btrfs_lookup_ordered_extent(inode, start); | |
3257 | if (!ordered) | |
3258 | break; | |
3259 | unlock_extent(tree, start, end); | |
3260 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3261 | btrfs_put_ordered_extent(ordered); | |
3262 | } | |
3263 | ||
125bac01 | 3264 | ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num, |
005efedf | 3265 | bio_flags, rw, NULL); |
9974090b MX |
3266 | return ret; |
3267 | } | |
3268 | ||
d1310b2e | 3269 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, |
8ddc7d9c | 3270 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3271 | { |
3272 | struct bio *bio = NULL; | |
c8b97818 | 3273 | unsigned long bio_flags = 0; |
d1310b2e CM |
3274 | int ret; |
3275 | ||
8ddc7d9c | 3276 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, |
d4c7ca86 | 3277 | &bio_flags, READ); |
d1310b2e | 3278 | if (bio) |
8ddc7d9c | 3279 | ret = submit_one_bio(READ, bio, mirror_num, bio_flags); |
d1310b2e CM |
3280 | return ret; |
3281 | } | |
d1310b2e | 3282 | |
4b384318 MF |
3283 | int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page, |
3284 | get_extent_t *get_extent, int mirror_num) | |
3285 | { | |
3286 | struct bio *bio = NULL; | |
3287 | unsigned long bio_flags = EXTENT_BIO_PARENT_LOCKED; | |
3288 | int ret; | |
3289 | ||
3290 | ret = __do_readpage(tree, page, get_extent, NULL, &bio, mirror_num, | |
005efedf | 3291 | &bio_flags, READ, NULL); |
4b384318 MF |
3292 | if (bio) |
3293 | ret = submit_one_bio(READ, bio, mirror_num, bio_flags); | |
3294 | return ret; | |
3295 | } | |
3296 | ||
11c8349b CM |
3297 | static noinline void update_nr_written(struct page *page, |
3298 | struct writeback_control *wbc, | |
3299 | unsigned long nr_written) | |
3300 | { | |
3301 | wbc->nr_to_write -= nr_written; | |
3302 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && | |
3303 | wbc->range_start == 0 && wbc->range_end == LLONG_MAX)) | |
3304 | page->mapping->writeback_index = page->index + nr_written; | |
3305 | } | |
3306 | ||
d1310b2e | 3307 | /* |
40f76580 CM |
3308 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
3309 | * | |
3310 | * This returns 1 if our fill_delalloc function did all the work required | |
3311 | * to write the page (copy into inline extent). In this case the IO has | |
3312 | * been started and the page is already unlocked. | |
3313 | * | |
3314 | * This returns 0 if all went well (page still locked) | |
3315 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 3316 | */ |
40f76580 CM |
3317 | static noinline_for_stack int writepage_delalloc(struct inode *inode, |
3318 | struct page *page, struct writeback_control *wbc, | |
3319 | struct extent_page_data *epd, | |
3320 | u64 delalloc_start, | |
3321 | unsigned long *nr_written) | |
3322 | { | |
3323 | struct extent_io_tree *tree = epd->tree; | |
3324 | u64 page_end = delalloc_start + PAGE_CACHE_SIZE - 1; | |
3325 | u64 nr_delalloc; | |
3326 | u64 delalloc_to_write = 0; | |
3327 | u64 delalloc_end = 0; | |
3328 | int ret; | |
3329 | int page_started = 0; | |
3330 | ||
3331 | if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc) | |
3332 | return 0; | |
3333 | ||
3334 | while (delalloc_end < page_end) { | |
3335 | nr_delalloc = find_lock_delalloc_range(inode, tree, | |
3336 | page, | |
3337 | &delalloc_start, | |
3338 | &delalloc_end, | |
dcab6a3b | 3339 | BTRFS_MAX_EXTENT_SIZE); |
40f76580 CM |
3340 | if (nr_delalloc == 0) { |
3341 | delalloc_start = delalloc_end + 1; | |
3342 | continue; | |
3343 | } | |
3344 | ret = tree->ops->fill_delalloc(inode, page, | |
3345 | delalloc_start, | |
3346 | delalloc_end, | |
3347 | &page_started, | |
3348 | nr_written); | |
3349 | /* File system has been set read-only */ | |
3350 | if (ret) { | |
3351 | SetPageError(page); | |
3352 | /* fill_delalloc should be return < 0 for error | |
3353 | * but just in case, we use > 0 here meaning the | |
3354 | * IO is started, so we don't want to return > 0 | |
3355 | * unless things are going well. | |
3356 | */ | |
3357 | ret = ret < 0 ? ret : -EIO; | |
3358 | goto done; | |
3359 | } | |
3360 | /* | |
3361 | * delalloc_end is already one less than the total | |
3362 | * length, so we don't subtract one from | |
3363 | * PAGE_CACHE_SIZE | |
3364 | */ | |
3365 | delalloc_to_write += (delalloc_end - delalloc_start + | |
3366 | PAGE_CACHE_SIZE) >> | |
3367 | PAGE_CACHE_SHIFT; | |
3368 | delalloc_start = delalloc_end + 1; | |
3369 | } | |
3370 | if (wbc->nr_to_write < delalloc_to_write) { | |
3371 | int thresh = 8192; | |
3372 | ||
3373 | if (delalloc_to_write < thresh * 2) | |
3374 | thresh = delalloc_to_write; | |
3375 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3376 | thresh); | |
3377 | } | |
3378 | ||
3379 | /* did the fill delalloc function already unlock and start | |
3380 | * the IO? | |
3381 | */ | |
3382 | if (page_started) { | |
3383 | /* | |
3384 | * we've unlocked the page, so we can't update | |
3385 | * the mapping's writeback index, just update | |
3386 | * nr_to_write. | |
3387 | */ | |
3388 | wbc->nr_to_write -= *nr_written; | |
3389 | return 1; | |
3390 | } | |
3391 | ||
3392 | ret = 0; | |
3393 | ||
3394 | done: | |
3395 | return ret; | |
3396 | } | |
3397 | ||
3398 | /* | |
3399 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3400 | * and does the loop to map the page into extents and bios. | |
3401 | * | |
3402 | * We return 1 if the IO is started and the page is unlocked, | |
3403 | * 0 if all went well (page still locked) | |
3404 | * < 0 if there were errors (page still locked) | |
3405 | */ | |
3406 | static noinline_for_stack int __extent_writepage_io(struct inode *inode, | |
3407 | struct page *page, | |
3408 | struct writeback_control *wbc, | |
3409 | struct extent_page_data *epd, | |
3410 | loff_t i_size, | |
3411 | unsigned long nr_written, | |
3412 | int write_flags, int *nr_ret) | |
d1310b2e | 3413 | { |
d1310b2e | 3414 | struct extent_io_tree *tree = epd->tree; |
4eee4fa4 | 3415 | u64 start = page_offset(page); |
d1310b2e CM |
3416 | u64 page_end = start + PAGE_CACHE_SIZE - 1; |
3417 | u64 end; | |
3418 | u64 cur = start; | |
3419 | u64 extent_offset; | |
d1310b2e CM |
3420 | u64 block_start; |
3421 | u64 iosize; | |
3422 | sector_t sector; | |
2c64c53d | 3423 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
3424 | struct extent_map *em; |
3425 | struct block_device *bdev; | |
7f3c74fb | 3426 | size_t pg_offset = 0; |
d1310b2e | 3427 | size_t blocksize; |
40f76580 CM |
3428 | int ret = 0; |
3429 | int nr = 0; | |
3430 | bool compressed; | |
c8b97818 | 3431 | |
247e743c | 3432 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
3433 | ret = tree->ops->writepage_start_hook(page, start, |
3434 | page_end); | |
87826df0 JM |
3435 | if (ret) { |
3436 | /* Fixup worker will requeue */ | |
3437 | if (ret == -EBUSY) | |
3438 | wbc->pages_skipped++; | |
3439 | else | |
3440 | redirty_page_for_writepage(wbc, page); | |
40f76580 | 3441 | |
11c8349b | 3442 | update_nr_written(page, wbc, nr_written); |
247e743c | 3443 | unlock_page(page); |
40f76580 | 3444 | ret = 1; |
11c8349b | 3445 | goto done_unlocked; |
247e743c CM |
3446 | } |
3447 | } | |
3448 | ||
11c8349b CM |
3449 | /* |
3450 | * we don't want to touch the inode after unlocking the page, | |
3451 | * so we update the mapping writeback index now | |
3452 | */ | |
3453 | update_nr_written(page, wbc, nr_written + 1); | |
771ed689 | 3454 | |
d1310b2e | 3455 | end = page_end; |
40f76580 | 3456 | if (i_size <= start) { |
e6dcd2dc CM |
3457 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3458 | tree->ops->writepage_end_io_hook(page, start, | |
3459 | page_end, NULL, 1); | |
d1310b2e CM |
3460 | goto done; |
3461 | } | |
3462 | ||
d1310b2e CM |
3463 | blocksize = inode->i_sb->s_blocksize; |
3464 | ||
3465 | while (cur <= end) { | |
40f76580 CM |
3466 | u64 em_end; |
3467 | if (cur >= i_size) { | |
e6dcd2dc CM |
3468 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3469 | tree->ops->writepage_end_io_hook(page, cur, | |
3470 | page_end, NULL, 1); | |
d1310b2e CM |
3471 | break; |
3472 | } | |
7f3c74fb | 3473 | em = epd->get_extent(inode, page, pg_offset, cur, |
d1310b2e | 3474 | end - cur + 1, 1); |
c704005d | 3475 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3476 | SetPageError(page); |
61391d56 | 3477 | ret = PTR_ERR_OR_ZERO(em); |
d1310b2e CM |
3478 | break; |
3479 | } | |
3480 | ||
3481 | extent_offset = cur - em->start; | |
40f76580 CM |
3482 | em_end = extent_map_end(em); |
3483 | BUG_ON(em_end <= cur); | |
d1310b2e | 3484 | BUG_ON(end < cur); |
40f76580 | 3485 | iosize = min(em_end - cur, end - cur + 1); |
fda2832f | 3486 | iosize = ALIGN(iosize, blocksize); |
d1310b2e CM |
3487 | sector = (em->block_start + extent_offset) >> 9; |
3488 | bdev = em->bdev; | |
3489 | block_start = em->block_start; | |
c8b97818 | 3490 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
3491 | free_extent_map(em); |
3492 | em = NULL; | |
3493 | ||
c8b97818 CM |
3494 | /* |
3495 | * compressed and inline extents are written through other | |
3496 | * paths in the FS | |
3497 | */ | |
3498 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3499 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
3500 | /* |
3501 | * end_io notification does not happen here for | |
3502 | * compressed extents | |
3503 | */ | |
3504 | if (!compressed && tree->ops && | |
3505 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
3506 | tree->ops->writepage_end_io_hook(page, cur, |
3507 | cur + iosize - 1, | |
3508 | NULL, 1); | |
c8b97818 CM |
3509 | else if (compressed) { |
3510 | /* we don't want to end_page_writeback on | |
3511 | * a compressed extent. this happens | |
3512 | * elsewhere | |
3513 | */ | |
3514 | nr++; | |
3515 | } | |
3516 | ||
3517 | cur += iosize; | |
7f3c74fb | 3518 | pg_offset += iosize; |
d1310b2e CM |
3519 | continue; |
3520 | } | |
c8b97818 | 3521 | |
d1310b2e CM |
3522 | if (tree->ops && tree->ops->writepage_io_hook) { |
3523 | ret = tree->ops->writepage_io_hook(page, cur, | |
3524 | cur + iosize - 1); | |
3525 | } else { | |
3526 | ret = 0; | |
3527 | } | |
1259ab75 | 3528 | if (ret) { |
d1310b2e | 3529 | SetPageError(page); |
1259ab75 | 3530 | } else { |
40f76580 | 3531 | unsigned long max_nr = (i_size >> PAGE_CACHE_SHIFT) + 1; |
7f3c74fb | 3532 | |
d1310b2e CM |
3533 | set_range_writeback(tree, cur, cur + iosize - 1); |
3534 | if (!PageWriteback(page)) { | |
efe120a0 FH |
3535 | btrfs_err(BTRFS_I(inode)->root->fs_info, |
3536 | "page %lu not writeback, cur %llu end %llu", | |
c1c9ff7c | 3537 | page->index, cur, end); |
d1310b2e CM |
3538 | } |
3539 | ||
da2f0f74 | 3540 | ret = submit_extent_page(write_flags, tree, wbc, page, |
ffbd517d CM |
3541 | sector, iosize, pg_offset, |
3542 | bdev, &epd->bio, max_nr, | |
c8b97818 | 3543 | end_bio_extent_writepage, |
005efedf | 3544 | 0, 0, 0, false); |
d1310b2e CM |
3545 | if (ret) |
3546 | SetPageError(page); | |
3547 | } | |
3548 | cur = cur + iosize; | |
7f3c74fb | 3549 | pg_offset += iosize; |
d1310b2e CM |
3550 | nr++; |
3551 | } | |
40f76580 CM |
3552 | done: |
3553 | *nr_ret = nr; | |
3554 | ||
3555 | done_unlocked: | |
3556 | ||
3557 | /* drop our reference on any cached states */ | |
3558 | free_extent_state(cached_state); | |
3559 | return ret; | |
3560 | } | |
3561 | ||
3562 | /* | |
3563 | * the writepage semantics are similar to regular writepage. extent | |
3564 | * records are inserted to lock ranges in the tree, and as dirty areas | |
3565 | * are found, they are marked writeback. Then the lock bits are removed | |
3566 | * and the end_io handler clears the writeback ranges | |
3567 | */ | |
3568 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
3569 | void *data) | |
3570 | { | |
3571 | struct inode *inode = page->mapping->host; | |
3572 | struct extent_page_data *epd = data; | |
3573 | u64 start = page_offset(page); | |
3574 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
3575 | int ret; | |
3576 | int nr = 0; | |
3577 | size_t pg_offset = 0; | |
3578 | loff_t i_size = i_size_read(inode); | |
3579 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
3580 | int write_flags; | |
3581 | unsigned long nr_written = 0; | |
3582 | ||
3583 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3584 | write_flags = WRITE_SYNC; | |
3585 | else | |
3586 | write_flags = WRITE; | |
3587 | ||
3588 | trace___extent_writepage(page, inode, wbc); | |
3589 | ||
3590 | WARN_ON(!PageLocked(page)); | |
3591 | ||
3592 | ClearPageError(page); | |
3593 | ||
3594 | pg_offset = i_size & (PAGE_CACHE_SIZE - 1); | |
3595 | if (page->index > end_index || | |
3596 | (page->index == end_index && !pg_offset)) { | |
3597 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE); | |
3598 | unlock_page(page); | |
3599 | return 0; | |
3600 | } | |
3601 | ||
3602 | if (page->index == end_index) { | |
3603 | char *userpage; | |
3604 | ||
3605 | userpage = kmap_atomic(page); | |
3606 | memset(userpage + pg_offset, 0, | |
3607 | PAGE_CACHE_SIZE - pg_offset); | |
3608 | kunmap_atomic(userpage); | |
3609 | flush_dcache_page(page); | |
3610 | } | |
3611 | ||
3612 | pg_offset = 0; | |
3613 | ||
3614 | set_page_extent_mapped(page); | |
3615 | ||
3616 | ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written); | |
3617 | if (ret == 1) | |
3618 | goto done_unlocked; | |
3619 | if (ret) | |
3620 | goto done; | |
3621 | ||
3622 | ret = __extent_writepage_io(inode, page, wbc, epd, | |
3623 | i_size, nr_written, write_flags, &nr); | |
3624 | if (ret == 1) | |
3625 | goto done_unlocked; | |
3626 | ||
d1310b2e CM |
3627 | done: |
3628 | if (nr == 0) { | |
3629 | /* make sure the mapping tag for page dirty gets cleared */ | |
3630 | set_page_writeback(page); | |
3631 | end_page_writeback(page); | |
3632 | } | |
61391d56 FM |
3633 | if (PageError(page)) { |
3634 | ret = ret < 0 ? ret : -EIO; | |
3635 | end_extent_writepage(page, ret, start, page_end); | |
3636 | } | |
d1310b2e | 3637 | unlock_page(page); |
40f76580 | 3638 | return ret; |
771ed689 | 3639 | |
11c8349b | 3640 | done_unlocked: |
d1310b2e CM |
3641 | return 0; |
3642 | } | |
3643 | ||
fd8b2b61 | 3644 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 3645 | { |
74316201 N |
3646 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
3647 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
3648 | } |
3649 | ||
0e378df1 CM |
3650 | static noinline_for_stack int |
3651 | lock_extent_buffer_for_io(struct extent_buffer *eb, | |
3652 | struct btrfs_fs_info *fs_info, | |
3653 | struct extent_page_data *epd) | |
0b32f4bb JB |
3654 | { |
3655 | unsigned long i, num_pages; | |
3656 | int flush = 0; | |
3657 | int ret = 0; | |
3658 | ||
3659 | if (!btrfs_try_tree_write_lock(eb)) { | |
3660 | flush = 1; | |
3661 | flush_write_bio(epd); | |
3662 | btrfs_tree_lock(eb); | |
3663 | } | |
3664 | ||
3665 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3666 | btrfs_tree_unlock(eb); | |
3667 | if (!epd->sync_io) | |
3668 | return 0; | |
3669 | if (!flush) { | |
3670 | flush_write_bio(epd); | |
3671 | flush = 1; | |
3672 | } | |
a098d8e8 CM |
3673 | while (1) { |
3674 | wait_on_extent_buffer_writeback(eb); | |
3675 | btrfs_tree_lock(eb); | |
3676 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3677 | break; | |
0b32f4bb | 3678 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3679 | } |
3680 | } | |
3681 | ||
51561ffe JB |
3682 | /* |
3683 | * We need to do this to prevent races in people who check if the eb is | |
3684 | * under IO since we can end up having no IO bits set for a short period | |
3685 | * of time. | |
3686 | */ | |
3687 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
3688 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3689 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 3690 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 3691 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
e2d84521 MX |
3692 | __percpu_counter_add(&fs_info->dirty_metadata_bytes, |
3693 | -eb->len, | |
3694 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 3695 | ret = 1; |
51561ffe JB |
3696 | } else { |
3697 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
3698 | } |
3699 | ||
3700 | btrfs_tree_unlock(eb); | |
3701 | ||
3702 | if (!ret) | |
3703 | return ret; | |
3704 | ||
3705 | num_pages = num_extent_pages(eb->start, eb->len); | |
3706 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 3707 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3708 | |
3709 | if (!trylock_page(p)) { | |
3710 | if (!flush) { | |
3711 | flush_write_bio(epd); | |
3712 | flush = 1; | |
3713 | } | |
3714 | lock_page(p); | |
3715 | } | |
3716 | } | |
3717 | ||
3718 | return ret; | |
3719 | } | |
3720 | ||
3721 | static void end_extent_buffer_writeback(struct extent_buffer *eb) | |
3722 | { | |
3723 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
4e857c58 | 3724 | smp_mb__after_atomic(); |
0b32f4bb JB |
3725 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); |
3726 | } | |
3727 | ||
656f30db FM |
3728 | static void set_btree_ioerr(struct page *page) |
3729 | { | |
3730 | struct extent_buffer *eb = (struct extent_buffer *)page->private; | |
3731 | struct btrfs_inode *btree_ino = BTRFS_I(eb->fs_info->btree_inode); | |
3732 | ||
3733 | SetPageError(page); | |
3734 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) | |
3735 | return; | |
3736 | ||
3737 | /* | |
3738 | * If writeback for a btree extent that doesn't belong to a log tree | |
3739 | * failed, increment the counter transaction->eb_write_errors. | |
3740 | * We do this because while the transaction is running and before it's | |
3741 | * committing (when we call filemap_fdata[write|wait]_range against | |
3742 | * the btree inode), we might have | |
3743 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
3744 | * returns an error or an error happens during writeback, when we're | |
3745 | * committing the transaction we wouldn't know about it, since the pages | |
3746 | * can be no longer dirty nor marked anymore for writeback (if a | |
3747 | * subsequent modification to the extent buffer didn't happen before the | |
3748 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
3749 | * able to find the pages tagged with SetPageError at transaction | |
3750 | * commit time. So if this happens we must abort the transaction, | |
3751 | * otherwise we commit a super block with btree roots that point to | |
3752 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
3753 | * or the content of some node/leaf from a past generation that got | |
3754 | * cowed or deleted and is no longer valid. | |
3755 | * | |
3756 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
3757 | * not be enough - we need to distinguish between log tree extents vs | |
3758 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
3759 | * will catch and clear such errors in the mapping - and that call might | |
3760 | * be from a log sync and not from a transaction commit. Also, checking | |
3761 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
3762 | * not done and would not be reliable - the eb might have been released | |
3763 | * from memory and reading it back again means that flag would not be | |
3764 | * set (since it's a runtime flag, not persisted on disk). | |
3765 | * | |
3766 | * Using the flags below in the btree inode also makes us achieve the | |
3767 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
3768 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
3769 | * is called, the writeback for all dirty pages had already finished | |
3770 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
3771 | * filemap_fdatawait_range() would return success, as it could not know | |
3772 | * that writeback errors happened (the pages were no longer tagged for | |
3773 | * writeback). | |
3774 | */ | |
3775 | switch (eb->log_index) { | |
3776 | case -1: | |
3777 | set_bit(BTRFS_INODE_BTREE_ERR, &btree_ino->runtime_flags); | |
3778 | break; | |
3779 | case 0: | |
3780 | set_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags); | |
3781 | break; | |
3782 | case 1: | |
3783 | set_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags); | |
3784 | break; | |
3785 | default: | |
3786 | BUG(); /* unexpected, logic error */ | |
3787 | } | |
3788 | } | |
3789 | ||
4246a0b6 | 3790 | static void end_bio_extent_buffer_writepage(struct bio *bio) |
0b32f4bb | 3791 | { |
2c30c71b | 3792 | struct bio_vec *bvec; |
0b32f4bb | 3793 | struct extent_buffer *eb; |
2c30c71b | 3794 | int i, done; |
0b32f4bb | 3795 | |
2c30c71b | 3796 | bio_for_each_segment_all(bvec, bio, i) { |
0b32f4bb JB |
3797 | struct page *page = bvec->bv_page; |
3798 | ||
0b32f4bb JB |
3799 | eb = (struct extent_buffer *)page->private; |
3800 | BUG_ON(!eb); | |
3801 | done = atomic_dec_and_test(&eb->io_pages); | |
3802 | ||
4246a0b6 CH |
3803 | if (bio->bi_error || |
3804 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { | |
0b32f4bb | 3805 | ClearPageUptodate(page); |
656f30db | 3806 | set_btree_ioerr(page); |
0b32f4bb JB |
3807 | } |
3808 | ||
3809 | end_page_writeback(page); | |
3810 | ||
3811 | if (!done) | |
3812 | continue; | |
3813 | ||
3814 | end_extent_buffer_writeback(eb); | |
2c30c71b | 3815 | } |
0b32f4bb JB |
3816 | |
3817 | bio_put(bio); | |
0b32f4bb JB |
3818 | } |
3819 | ||
0e378df1 | 3820 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb JB |
3821 | struct btrfs_fs_info *fs_info, |
3822 | struct writeback_control *wbc, | |
3823 | struct extent_page_data *epd) | |
3824 | { | |
3825 | struct block_device *bdev = fs_info->fs_devices->latest_bdev; | |
f28491e0 | 3826 | struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree; |
0b32f4bb JB |
3827 | u64 offset = eb->start; |
3828 | unsigned long i, num_pages; | |
de0022b9 | 3829 | unsigned long bio_flags = 0; |
d4c7ca86 | 3830 | int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META; |
d7dbe9e7 | 3831 | int ret = 0; |
0b32f4bb | 3832 | |
656f30db | 3833 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); |
0b32f4bb JB |
3834 | num_pages = num_extent_pages(eb->start, eb->len); |
3835 | atomic_set(&eb->io_pages, num_pages); | |
de0022b9 JB |
3836 | if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID) |
3837 | bio_flags = EXTENT_BIO_TREE_LOG; | |
3838 | ||
0b32f4bb | 3839 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3840 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3841 | |
3842 | clear_page_dirty_for_io(p); | |
3843 | set_page_writeback(p); | |
da2f0f74 | 3844 | ret = submit_extent_page(rw, tree, wbc, p, offset >> 9, |
0b32f4bb JB |
3845 | PAGE_CACHE_SIZE, 0, bdev, &epd->bio, |
3846 | -1, end_bio_extent_buffer_writepage, | |
005efedf | 3847 | 0, epd->bio_flags, bio_flags, false); |
de0022b9 | 3848 | epd->bio_flags = bio_flags; |
0b32f4bb | 3849 | if (ret) { |
656f30db | 3850 | set_btree_ioerr(p); |
55e3bd2e | 3851 | end_page_writeback(p); |
0b32f4bb JB |
3852 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
3853 | end_extent_buffer_writeback(eb); | |
3854 | ret = -EIO; | |
3855 | break; | |
3856 | } | |
3857 | offset += PAGE_CACHE_SIZE; | |
3858 | update_nr_written(p, wbc, 1); | |
3859 | unlock_page(p); | |
3860 | } | |
3861 | ||
3862 | if (unlikely(ret)) { | |
3863 | for (; i < num_pages; i++) { | |
bbf65cf0 | 3864 | struct page *p = eb->pages[i]; |
81465028 | 3865 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
3866 | unlock_page(p); |
3867 | } | |
3868 | } | |
3869 | ||
3870 | return ret; | |
3871 | } | |
3872 | ||
3873 | int btree_write_cache_pages(struct address_space *mapping, | |
3874 | struct writeback_control *wbc) | |
3875 | { | |
3876 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; | |
3877 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; | |
3878 | struct extent_buffer *eb, *prev_eb = NULL; | |
3879 | struct extent_page_data epd = { | |
3880 | .bio = NULL, | |
3881 | .tree = tree, | |
3882 | .extent_locked = 0, | |
3883 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
de0022b9 | 3884 | .bio_flags = 0, |
0b32f4bb JB |
3885 | }; |
3886 | int ret = 0; | |
3887 | int done = 0; | |
3888 | int nr_to_write_done = 0; | |
3889 | struct pagevec pvec; | |
3890 | int nr_pages; | |
3891 | pgoff_t index; | |
3892 | pgoff_t end; /* Inclusive */ | |
3893 | int scanned = 0; | |
3894 | int tag; | |
3895 | ||
3896 | pagevec_init(&pvec, 0); | |
3897 | if (wbc->range_cyclic) { | |
3898 | index = mapping->writeback_index; /* Start from prev offset */ | |
3899 | end = -1; | |
3900 | } else { | |
3901 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
3902 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
3903 | scanned = 1; | |
3904 | } | |
3905 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3906 | tag = PAGECACHE_TAG_TOWRITE; | |
3907 | else | |
3908 | tag = PAGECACHE_TAG_DIRTY; | |
3909 | retry: | |
3910 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3911 | tag_pages_for_writeback(mapping, index, end); | |
3912 | while (!done && !nr_to_write_done && (index <= end) && | |
3913 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, | |
3914 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
3915 | unsigned i; | |
3916 | ||
3917 | scanned = 1; | |
3918 | for (i = 0; i < nr_pages; i++) { | |
3919 | struct page *page = pvec.pages[i]; | |
3920 | ||
3921 | if (!PagePrivate(page)) | |
3922 | continue; | |
3923 | ||
3924 | if (!wbc->range_cyclic && page->index > end) { | |
3925 | done = 1; | |
3926 | break; | |
3927 | } | |
3928 | ||
b5bae261 JB |
3929 | spin_lock(&mapping->private_lock); |
3930 | if (!PagePrivate(page)) { | |
3931 | spin_unlock(&mapping->private_lock); | |
3932 | continue; | |
3933 | } | |
3934 | ||
0b32f4bb | 3935 | eb = (struct extent_buffer *)page->private; |
b5bae261 JB |
3936 | |
3937 | /* | |
3938 | * Shouldn't happen and normally this would be a BUG_ON | |
3939 | * but no sense in crashing the users box for something | |
3940 | * we can survive anyway. | |
3941 | */ | |
fae7f21c | 3942 | if (WARN_ON(!eb)) { |
b5bae261 | 3943 | spin_unlock(&mapping->private_lock); |
0b32f4bb JB |
3944 | continue; |
3945 | } | |
3946 | ||
b5bae261 JB |
3947 | if (eb == prev_eb) { |
3948 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 3949 | continue; |
b5bae261 | 3950 | } |
0b32f4bb | 3951 | |
b5bae261 JB |
3952 | ret = atomic_inc_not_zero(&eb->refs); |
3953 | spin_unlock(&mapping->private_lock); | |
3954 | if (!ret) | |
0b32f4bb | 3955 | continue; |
0b32f4bb JB |
3956 | |
3957 | prev_eb = eb; | |
3958 | ret = lock_extent_buffer_for_io(eb, fs_info, &epd); | |
3959 | if (!ret) { | |
3960 | free_extent_buffer(eb); | |
3961 | continue; | |
3962 | } | |
3963 | ||
3964 | ret = write_one_eb(eb, fs_info, wbc, &epd); | |
3965 | if (ret) { | |
3966 | done = 1; | |
3967 | free_extent_buffer(eb); | |
3968 | break; | |
3969 | } | |
3970 | free_extent_buffer(eb); | |
3971 | ||
3972 | /* | |
3973 | * the filesystem may choose to bump up nr_to_write. | |
3974 | * We have to make sure to honor the new nr_to_write | |
3975 | * at any time | |
3976 | */ | |
3977 | nr_to_write_done = wbc->nr_to_write <= 0; | |
3978 | } | |
3979 | pagevec_release(&pvec); | |
3980 | cond_resched(); | |
3981 | } | |
3982 | if (!scanned && !done) { | |
3983 | /* | |
3984 | * We hit the last page and there is more work to be done: wrap | |
3985 | * back to the start of the file | |
3986 | */ | |
3987 | scanned = 1; | |
3988 | index = 0; | |
3989 | goto retry; | |
3990 | } | |
3991 | flush_write_bio(&epd); | |
3992 | return ret; | |
3993 | } | |
3994 | ||
d1310b2e | 3995 | /** |
4bef0848 | 3996 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
3997 | * @mapping: address space structure to write |
3998 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
3999 | * @writepage: function called for each page | |
4000 | * @data: data passed to writepage function | |
4001 | * | |
4002 | * If a page is already under I/O, write_cache_pages() skips it, even | |
4003 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
4004 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
4005 | * and msync() need to guarantee that all the data which was dirty at the time | |
4006 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
4007 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
4008 | * existing IO to complete. | |
4009 | */ | |
b2950863 | 4010 | static int extent_write_cache_pages(struct extent_io_tree *tree, |
4bef0848 CM |
4011 | struct address_space *mapping, |
4012 | struct writeback_control *wbc, | |
d2c3f4f6 CM |
4013 | writepage_t writepage, void *data, |
4014 | void (*flush_fn)(void *)) | |
d1310b2e | 4015 | { |
7fd1a3f7 | 4016 | struct inode *inode = mapping->host; |
d1310b2e CM |
4017 | int ret = 0; |
4018 | int done = 0; | |
61391d56 | 4019 | int err = 0; |
f85d7d6c | 4020 | int nr_to_write_done = 0; |
d1310b2e CM |
4021 | struct pagevec pvec; |
4022 | int nr_pages; | |
4023 | pgoff_t index; | |
4024 | pgoff_t end; /* Inclusive */ | |
4025 | int scanned = 0; | |
f7aaa06b | 4026 | int tag; |
d1310b2e | 4027 | |
7fd1a3f7 JB |
4028 | /* |
4029 | * We have to hold onto the inode so that ordered extents can do their | |
4030 | * work when the IO finishes. The alternative to this is failing to add | |
4031 | * an ordered extent if the igrab() fails there and that is a huge pain | |
4032 | * to deal with, so instead just hold onto the inode throughout the | |
4033 | * writepages operation. If it fails here we are freeing up the inode | |
4034 | * anyway and we'd rather not waste our time writing out stuff that is | |
4035 | * going to be truncated anyway. | |
4036 | */ | |
4037 | if (!igrab(inode)) | |
4038 | return 0; | |
4039 | ||
d1310b2e CM |
4040 | pagevec_init(&pvec, 0); |
4041 | if (wbc->range_cyclic) { | |
4042 | index = mapping->writeback_index; /* Start from prev offset */ | |
4043 | end = -1; | |
4044 | } else { | |
4045 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
4046 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
4047 | scanned = 1; |
4048 | } | |
f7aaa06b JB |
4049 | if (wbc->sync_mode == WB_SYNC_ALL) |
4050 | tag = PAGECACHE_TAG_TOWRITE; | |
4051 | else | |
4052 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 4053 | retry: |
f7aaa06b JB |
4054 | if (wbc->sync_mode == WB_SYNC_ALL) |
4055 | tag_pages_for_writeback(mapping, index, end); | |
f85d7d6c | 4056 | while (!done && !nr_to_write_done && (index <= end) && |
f7aaa06b JB |
4057 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
4058 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
d1310b2e CM |
4059 | unsigned i; |
4060 | ||
4061 | scanned = 1; | |
4062 | for (i = 0; i < nr_pages; i++) { | |
4063 | struct page *page = pvec.pages[i]; | |
4064 | ||
4065 | /* | |
4066 | * At this point we hold neither mapping->tree_lock nor | |
4067 | * lock on the page itself: the page may be truncated or | |
4068 | * invalidated (changing page->mapping to NULL), or even | |
4069 | * swizzled back from swapper_space to tmpfs file | |
4070 | * mapping | |
4071 | */ | |
c8f2f24b JB |
4072 | if (!trylock_page(page)) { |
4073 | flush_fn(data); | |
4074 | lock_page(page); | |
01d658f2 | 4075 | } |
d1310b2e CM |
4076 | |
4077 | if (unlikely(page->mapping != mapping)) { | |
4078 | unlock_page(page); | |
4079 | continue; | |
4080 | } | |
4081 | ||
4082 | if (!wbc->range_cyclic && page->index > end) { | |
4083 | done = 1; | |
4084 | unlock_page(page); | |
4085 | continue; | |
4086 | } | |
4087 | ||
d2c3f4f6 | 4088 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 CM |
4089 | if (PageWriteback(page)) |
4090 | flush_fn(data); | |
d1310b2e | 4091 | wait_on_page_writeback(page); |
d2c3f4f6 | 4092 | } |
d1310b2e CM |
4093 | |
4094 | if (PageWriteback(page) || | |
4095 | !clear_page_dirty_for_io(page)) { | |
4096 | unlock_page(page); | |
4097 | continue; | |
4098 | } | |
4099 | ||
4100 | ret = (*writepage)(page, wbc, data); | |
4101 | ||
4102 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
4103 | unlock_page(page); | |
4104 | ret = 0; | |
4105 | } | |
61391d56 FM |
4106 | if (!err && ret < 0) |
4107 | err = ret; | |
f85d7d6c CM |
4108 | |
4109 | /* | |
4110 | * the filesystem may choose to bump up nr_to_write. | |
4111 | * We have to make sure to honor the new nr_to_write | |
4112 | * at any time | |
4113 | */ | |
4114 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
4115 | } |
4116 | pagevec_release(&pvec); | |
4117 | cond_resched(); | |
4118 | } | |
61391d56 | 4119 | if (!scanned && !done && !err) { |
d1310b2e CM |
4120 | /* |
4121 | * We hit the last page and there is more work to be done: wrap | |
4122 | * back to the start of the file | |
4123 | */ | |
4124 | scanned = 1; | |
4125 | index = 0; | |
4126 | goto retry; | |
4127 | } | |
7fd1a3f7 | 4128 | btrfs_add_delayed_iput(inode); |
61391d56 | 4129 | return err; |
d1310b2e | 4130 | } |
d1310b2e | 4131 | |
ffbd517d | 4132 | static void flush_epd_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 4133 | { |
d2c3f4f6 | 4134 | if (epd->bio) { |
355808c2 JM |
4135 | int rw = WRITE; |
4136 | int ret; | |
4137 | ||
ffbd517d | 4138 | if (epd->sync_io) |
355808c2 JM |
4139 | rw = WRITE_SYNC; |
4140 | ||
de0022b9 | 4141 | ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags); |
79787eaa | 4142 | BUG_ON(ret < 0); /* -ENOMEM */ |
d2c3f4f6 CM |
4143 | epd->bio = NULL; |
4144 | } | |
4145 | } | |
4146 | ||
ffbd517d CM |
4147 | static noinline void flush_write_bio(void *data) |
4148 | { | |
4149 | struct extent_page_data *epd = data; | |
4150 | flush_epd_write_bio(epd); | |
4151 | } | |
4152 | ||
d1310b2e CM |
4153 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, |
4154 | get_extent_t *get_extent, | |
4155 | struct writeback_control *wbc) | |
4156 | { | |
4157 | int ret; | |
d1310b2e CM |
4158 | struct extent_page_data epd = { |
4159 | .bio = NULL, | |
4160 | .tree = tree, | |
4161 | .get_extent = get_extent, | |
771ed689 | 4162 | .extent_locked = 0, |
ffbd517d | 4163 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
de0022b9 | 4164 | .bio_flags = 0, |
d1310b2e | 4165 | }; |
d1310b2e | 4166 | |
d1310b2e CM |
4167 | ret = __extent_writepage(page, wbc, &epd); |
4168 | ||
ffbd517d | 4169 | flush_epd_write_bio(&epd); |
d1310b2e CM |
4170 | return ret; |
4171 | } | |
d1310b2e | 4172 | |
771ed689 CM |
4173 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, |
4174 | u64 start, u64 end, get_extent_t *get_extent, | |
4175 | int mode) | |
4176 | { | |
4177 | int ret = 0; | |
4178 | struct address_space *mapping = inode->i_mapping; | |
4179 | struct page *page; | |
4180 | unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >> | |
4181 | PAGE_CACHE_SHIFT; | |
4182 | ||
4183 | struct extent_page_data epd = { | |
4184 | .bio = NULL, | |
4185 | .tree = tree, | |
4186 | .get_extent = get_extent, | |
4187 | .extent_locked = 1, | |
ffbd517d | 4188 | .sync_io = mode == WB_SYNC_ALL, |
de0022b9 | 4189 | .bio_flags = 0, |
771ed689 CM |
4190 | }; |
4191 | struct writeback_control wbc_writepages = { | |
771ed689 | 4192 | .sync_mode = mode, |
771ed689 CM |
4193 | .nr_to_write = nr_pages * 2, |
4194 | .range_start = start, | |
4195 | .range_end = end + 1, | |
4196 | }; | |
4197 | ||
d397712b | 4198 | while (start <= end) { |
771ed689 CM |
4199 | page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); |
4200 | if (clear_page_dirty_for_io(page)) | |
4201 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
4202 | else { | |
4203 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
4204 | tree->ops->writepage_end_io_hook(page, start, | |
4205 | start + PAGE_CACHE_SIZE - 1, | |
4206 | NULL, 1); | |
4207 | unlock_page(page); | |
4208 | } | |
4209 | page_cache_release(page); | |
4210 | start += PAGE_CACHE_SIZE; | |
4211 | } | |
4212 | ||
ffbd517d | 4213 | flush_epd_write_bio(&epd); |
771ed689 CM |
4214 | return ret; |
4215 | } | |
d1310b2e CM |
4216 | |
4217 | int extent_writepages(struct extent_io_tree *tree, | |
4218 | struct address_space *mapping, | |
4219 | get_extent_t *get_extent, | |
4220 | struct writeback_control *wbc) | |
4221 | { | |
4222 | int ret = 0; | |
4223 | struct extent_page_data epd = { | |
4224 | .bio = NULL, | |
4225 | .tree = tree, | |
4226 | .get_extent = get_extent, | |
771ed689 | 4227 | .extent_locked = 0, |
ffbd517d | 4228 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
de0022b9 | 4229 | .bio_flags = 0, |
d1310b2e CM |
4230 | }; |
4231 | ||
4bef0848 | 4232 | ret = extent_write_cache_pages(tree, mapping, wbc, |
d2c3f4f6 CM |
4233 | __extent_writepage, &epd, |
4234 | flush_write_bio); | |
ffbd517d | 4235 | flush_epd_write_bio(&epd); |
d1310b2e CM |
4236 | return ret; |
4237 | } | |
d1310b2e CM |
4238 | |
4239 | int extent_readpages(struct extent_io_tree *tree, | |
4240 | struct address_space *mapping, | |
4241 | struct list_head *pages, unsigned nr_pages, | |
4242 | get_extent_t get_extent) | |
4243 | { | |
4244 | struct bio *bio = NULL; | |
4245 | unsigned page_idx; | |
c8b97818 | 4246 | unsigned long bio_flags = 0; |
67c9684f LB |
4247 | struct page *pagepool[16]; |
4248 | struct page *page; | |
125bac01 | 4249 | struct extent_map *em_cached = NULL; |
67c9684f | 4250 | int nr = 0; |
808f80b4 | 4251 | u64 prev_em_start = (u64)-1; |
d1310b2e | 4252 | |
d1310b2e | 4253 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
67c9684f | 4254 | page = list_entry(pages->prev, struct page, lru); |
d1310b2e CM |
4255 | |
4256 | prefetchw(&page->flags); | |
4257 | list_del(&page->lru); | |
67c9684f | 4258 | if (add_to_page_cache_lru(page, mapping, |
43e817a1 | 4259 | page->index, GFP_NOFS)) { |
67c9684f LB |
4260 | page_cache_release(page); |
4261 | continue; | |
d1310b2e | 4262 | } |
67c9684f LB |
4263 | |
4264 | pagepool[nr++] = page; | |
4265 | if (nr < ARRAY_SIZE(pagepool)) | |
4266 | continue; | |
125bac01 | 4267 | __extent_readpages(tree, pagepool, nr, get_extent, &em_cached, |
808f80b4 | 4268 | &bio, 0, &bio_flags, READ, &prev_em_start); |
67c9684f | 4269 | nr = 0; |
d1310b2e | 4270 | } |
9974090b | 4271 | if (nr) |
125bac01 | 4272 | __extent_readpages(tree, pagepool, nr, get_extent, &em_cached, |
808f80b4 | 4273 | &bio, 0, &bio_flags, READ, &prev_em_start); |
67c9684f | 4274 | |
125bac01 MX |
4275 | if (em_cached) |
4276 | free_extent_map(em_cached); | |
4277 | ||
d1310b2e CM |
4278 | BUG_ON(!list_empty(pages)); |
4279 | if (bio) | |
79787eaa | 4280 | return submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
4281 | return 0; |
4282 | } | |
d1310b2e CM |
4283 | |
4284 | /* | |
4285 | * basic invalidatepage code, this waits on any locked or writeback | |
4286 | * ranges corresponding to the page, and then deletes any extent state | |
4287 | * records from the tree | |
4288 | */ | |
4289 | int extent_invalidatepage(struct extent_io_tree *tree, | |
4290 | struct page *page, unsigned long offset) | |
4291 | { | |
2ac55d41 | 4292 | struct extent_state *cached_state = NULL; |
4eee4fa4 | 4293 | u64 start = page_offset(page); |
d1310b2e CM |
4294 | u64 end = start + PAGE_CACHE_SIZE - 1; |
4295 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
4296 | ||
fda2832f | 4297 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
4298 | if (start > end) |
4299 | return 0; | |
4300 | ||
d0082371 | 4301 | lock_extent_bits(tree, start, end, 0, &cached_state); |
1edbb734 | 4302 | wait_on_page_writeback(page); |
d1310b2e | 4303 | clear_extent_bit(tree, start, end, |
32c00aff JB |
4304 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
4305 | EXTENT_DO_ACCOUNTING, | |
2ac55d41 | 4306 | 1, 1, &cached_state, GFP_NOFS); |
d1310b2e CM |
4307 | return 0; |
4308 | } | |
d1310b2e | 4309 | |
7b13b7b1 CM |
4310 | /* |
4311 | * a helper for releasepage, this tests for areas of the page that | |
4312 | * are locked or under IO and drops the related state bits if it is safe | |
4313 | * to drop the page. | |
4314 | */ | |
48a3b636 ES |
4315 | static int try_release_extent_state(struct extent_map_tree *map, |
4316 | struct extent_io_tree *tree, | |
4317 | struct page *page, gfp_t mask) | |
7b13b7b1 | 4318 | { |
4eee4fa4 | 4319 | u64 start = page_offset(page); |
7b13b7b1 CM |
4320 | u64 end = start + PAGE_CACHE_SIZE - 1; |
4321 | int ret = 1; | |
4322 | ||
211f90e6 | 4323 | if (test_range_bit(tree, start, end, |
8b62b72b | 4324 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
4325 | ret = 0; |
4326 | else { | |
4327 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
4328 | mask = GFP_NOFS; | |
11ef160f CM |
4329 | /* |
4330 | * at this point we can safely clear everything except the | |
4331 | * locked bit and the nodatasum bit | |
4332 | */ | |
e3f24cc5 | 4333 | ret = clear_extent_bit(tree, start, end, |
11ef160f CM |
4334 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
4335 | 0, 0, NULL, mask); | |
e3f24cc5 CM |
4336 | |
4337 | /* if clear_extent_bit failed for enomem reasons, | |
4338 | * we can't allow the release to continue. | |
4339 | */ | |
4340 | if (ret < 0) | |
4341 | ret = 0; | |
4342 | else | |
4343 | ret = 1; | |
7b13b7b1 CM |
4344 | } |
4345 | return ret; | |
4346 | } | |
7b13b7b1 | 4347 | |
d1310b2e CM |
4348 | /* |
4349 | * a helper for releasepage. As long as there are no locked extents | |
4350 | * in the range corresponding to the page, both state records and extent | |
4351 | * map records are removed | |
4352 | */ | |
4353 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
4354 | struct extent_io_tree *tree, struct page *page, |
4355 | gfp_t mask) | |
d1310b2e CM |
4356 | { |
4357 | struct extent_map *em; | |
4eee4fa4 | 4358 | u64 start = page_offset(page); |
d1310b2e | 4359 | u64 end = start + PAGE_CACHE_SIZE - 1; |
7b13b7b1 | 4360 | |
70dec807 CM |
4361 | if ((mask & __GFP_WAIT) && |
4362 | page->mapping->host->i_size > 16 * 1024 * 1024) { | |
39b5637f | 4363 | u64 len; |
70dec807 | 4364 | while (start <= end) { |
39b5637f | 4365 | len = end - start + 1; |
890871be | 4366 | write_lock(&map->lock); |
39b5637f | 4367 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 4368 | if (!em) { |
890871be | 4369 | write_unlock(&map->lock); |
70dec807 CM |
4370 | break; |
4371 | } | |
7f3c74fb CM |
4372 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
4373 | em->start != start) { | |
890871be | 4374 | write_unlock(&map->lock); |
70dec807 CM |
4375 | free_extent_map(em); |
4376 | break; | |
4377 | } | |
4378 | if (!test_range_bit(tree, em->start, | |
4379 | extent_map_end(em) - 1, | |
8b62b72b | 4380 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 4381 | 0, NULL)) { |
70dec807 CM |
4382 | remove_extent_mapping(map, em); |
4383 | /* once for the rb tree */ | |
4384 | free_extent_map(em); | |
4385 | } | |
4386 | start = extent_map_end(em); | |
890871be | 4387 | write_unlock(&map->lock); |
70dec807 CM |
4388 | |
4389 | /* once for us */ | |
d1310b2e CM |
4390 | free_extent_map(em); |
4391 | } | |
d1310b2e | 4392 | } |
7b13b7b1 | 4393 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 4394 | } |
d1310b2e | 4395 | |
ec29ed5b CM |
4396 | /* |
4397 | * helper function for fiemap, which doesn't want to see any holes. | |
4398 | * This maps until we find something past 'last' | |
4399 | */ | |
4400 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
4401 | u64 offset, | |
4402 | u64 last, | |
4403 | get_extent_t *get_extent) | |
4404 | { | |
4405 | u64 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
4406 | struct extent_map *em; | |
4407 | u64 len; | |
4408 | ||
4409 | if (offset >= last) | |
4410 | return NULL; | |
4411 | ||
67871254 | 4412 | while (1) { |
ec29ed5b CM |
4413 | len = last - offset; |
4414 | if (len == 0) | |
4415 | break; | |
fda2832f | 4416 | len = ALIGN(len, sectorsize); |
ec29ed5b | 4417 | em = get_extent(inode, NULL, 0, offset, len, 0); |
c704005d | 4418 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
4419 | return em; |
4420 | ||
4421 | /* if this isn't a hole return it */ | |
4422 | if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) && | |
4423 | em->block_start != EXTENT_MAP_HOLE) { | |
4424 | return em; | |
4425 | } | |
4426 | ||
4427 | /* this is a hole, advance to the next extent */ | |
4428 | offset = extent_map_end(em); | |
4429 | free_extent_map(em); | |
4430 | if (offset >= last) | |
4431 | break; | |
4432 | } | |
4433 | return NULL; | |
4434 | } | |
4435 | ||
1506fcc8 YS |
4436 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4437 | __u64 start, __u64 len, get_extent_t *get_extent) | |
4438 | { | |
975f84fe | 4439 | int ret = 0; |
1506fcc8 YS |
4440 | u64 off = start; |
4441 | u64 max = start + len; | |
4442 | u32 flags = 0; | |
975f84fe JB |
4443 | u32 found_type; |
4444 | u64 last; | |
ec29ed5b | 4445 | u64 last_for_get_extent = 0; |
1506fcc8 | 4446 | u64 disko = 0; |
ec29ed5b | 4447 | u64 isize = i_size_read(inode); |
975f84fe | 4448 | struct btrfs_key found_key; |
1506fcc8 | 4449 | struct extent_map *em = NULL; |
2ac55d41 | 4450 | struct extent_state *cached_state = NULL; |
975f84fe | 4451 | struct btrfs_path *path; |
dc046b10 | 4452 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1506fcc8 | 4453 | int end = 0; |
ec29ed5b CM |
4454 | u64 em_start = 0; |
4455 | u64 em_len = 0; | |
4456 | u64 em_end = 0; | |
1506fcc8 YS |
4457 | |
4458 | if (len == 0) | |
4459 | return -EINVAL; | |
4460 | ||
975f84fe JB |
4461 | path = btrfs_alloc_path(); |
4462 | if (!path) | |
4463 | return -ENOMEM; | |
4464 | path->leave_spinning = 1; | |
4465 | ||
2c91943b QW |
4466 | start = round_down(start, BTRFS_I(inode)->root->sectorsize); |
4467 | len = round_up(max, BTRFS_I(inode)->root->sectorsize) - start; | |
4d479cf0 | 4468 | |
ec29ed5b CM |
4469 | /* |
4470 | * lookup the last file extent. We're not using i_size here | |
4471 | * because there might be preallocation past i_size | |
4472 | */ | |
dc046b10 JB |
4473 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1, |
4474 | 0); | |
975f84fe JB |
4475 | if (ret < 0) { |
4476 | btrfs_free_path(path); | |
4477 | return ret; | |
4478 | } | |
4479 | WARN_ON(!ret); | |
4480 | path->slots[0]--; | |
975f84fe | 4481 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
962a298f | 4482 | found_type = found_key.type; |
975f84fe | 4483 | |
ec29ed5b | 4484 | /* No extents, but there might be delalloc bits */ |
33345d01 | 4485 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 4486 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
4487 | /* have to trust i_size as the end */ |
4488 | last = (u64)-1; | |
4489 | last_for_get_extent = isize; | |
4490 | } else { | |
4491 | /* | |
4492 | * remember the start of the last extent. There are a | |
4493 | * bunch of different factors that go into the length of the | |
4494 | * extent, so its much less complex to remember where it started | |
4495 | */ | |
4496 | last = found_key.offset; | |
4497 | last_for_get_extent = last + 1; | |
975f84fe | 4498 | } |
fe09e16c | 4499 | btrfs_release_path(path); |
975f84fe | 4500 | |
ec29ed5b CM |
4501 | /* |
4502 | * we might have some extents allocated but more delalloc past those | |
4503 | * extents. so, we trust isize unless the start of the last extent is | |
4504 | * beyond isize | |
4505 | */ | |
4506 | if (last < isize) { | |
4507 | last = (u64)-1; | |
4508 | last_for_get_extent = isize; | |
4509 | } | |
4510 | ||
a52f4cd2 | 4511 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0, |
d0082371 | 4512 | &cached_state); |
ec29ed5b | 4513 | |
4d479cf0 | 4514 | em = get_extent_skip_holes(inode, start, last_for_get_extent, |
ec29ed5b | 4515 | get_extent); |
1506fcc8 YS |
4516 | if (!em) |
4517 | goto out; | |
4518 | if (IS_ERR(em)) { | |
4519 | ret = PTR_ERR(em); | |
4520 | goto out; | |
4521 | } | |
975f84fe | 4522 | |
1506fcc8 | 4523 | while (!end) { |
b76bb701 | 4524 | u64 offset_in_extent = 0; |
ea8efc74 CM |
4525 | |
4526 | /* break if the extent we found is outside the range */ | |
4527 | if (em->start >= max || extent_map_end(em) < off) | |
4528 | break; | |
4529 | ||
4530 | /* | |
4531 | * get_extent may return an extent that starts before our | |
4532 | * requested range. We have to make sure the ranges | |
4533 | * we return to fiemap always move forward and don't | |
4534 | * overlap, so adjust the offsets here | |
4535 | */ | |
4536 | em_start = max(em->start, off); | |
1506fcc8 | 4537 | |
ea8efc74 CM |
4538 | /* |
4539 | * record the offset from the start of the extent | |
b76bb701 JB |
4540 | * for adjusting the disk offset below. Only do this if the |
4541 | * extent isn't compressed since our in ram offset may be past | |
4542 | * what we have actually allocated on disk. | |
ea8efc74 | 4543 | */ |
b76bb701 JB |
4544 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
4545 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 4546 | em_end = extent_map_end(em); |
ea8efc74 | 4547 | em_len = em_end - em_start; |
1506fcc8 YS |
4548 | disko = 0; |
4549 | flags = 0; | |
4550 | ||
ea8efc74 CM |
4551 | /* |
4552 | * bump off for our next call to get_extent | |
4553 | */ | |
4554 | off = extent_map_end(em); | |
4555 | if (off >= max) | |
4556 | end = 1; | |
4557 | ||
93dbfad7 | 4558 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
4559 | end = 1; |
4560 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 4561 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
4562 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
4563 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 4564 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
4565 | flags |= (FIEMAP_EXTENT_DELALLOC | |
4566 | FIEMAP_EXTENT_UNKNOWN); | |
dc046b10 JB |
4567 | } else if (fieinfo->fi_extents_max) { |
4568 | u64 bytenr = em->block_start - | |
4569 | (em->start - em->orig_start); | |
fe09e16c | 4570 | |
ea8efc74 | 4571 | disko = em->block_start + offset_in_extent; |
fe09e16c LB |
4572 | |
4573 | /* | |
4574 | * As btrfs supports shared space, this information | |
4575 | * can be exported to userspace tools via | |
dc046b10 JB |
4576 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 |
4577 | * then we're just getting a count and we can skip the | |
4578 | * lookup stuff. | |
fe09e16c | 4579 | */ |
dc046b10 JB |
4580 | ret = btrfs_check_shared(NULL, root->fs_info, |
4581 | root->objectid, | |
4582 | btrfs_ino(inode), bytenr); | |
4583 | if (ret < 0) | |
fe09e16c | 4584 | goto out_free; |
dc046b10 | 4585 | if (ret) |
fe09e16c | 4586 | flags |= FIEMAP_EXTENT_SHARED; |
dc046b10 | 4587 | ret = 0; |
1506fcc8 YS |
4588 | } |
4589 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
4590 | flags |= FIEMAP_EXTENT_ENCODED; | |
0d2b2372 JB |
4591 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
4592 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
1506fcc8 | 4593 | |
1506fcc8 YS |
4594 | free_extent_map(em); |
4595 | em = NULL; | |
ec29ed5b CM |
4596 | if ((em_start >= last) || em_len == (u64)-1 || |
4597 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
4598 | flags |= FIEMAP_EXTENT_LAST; |
4599 | end = 1; | |
4600 | } | |
4601 | ||
ec29ed5b CM |
4602 | /* now scan forward to see if this is really the last extent. */ |
4603 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
4604 | get_extent); | |
4605 | if (IS_ERR(em)) { | |
4606 | ret = PTR_ERR(em); | |
4607 | goto out; | |
4608 | } | |
4609 | if (!em) { | |
975f84fe JB |
4610 | flags |= FIEMAP_EXTENT_LAST; |
4611 | end = 1; | |
4612 | } | |
ec29ed5b CM |
4613 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, |
4614 | em_len, flags); | |
26e726af CS |
4615 | if (ret) { |
4616 | if (ret == 1) | |
4617 | ret = 0; | |
ec29ed5b | 4618 | goto out_free; |
26e726af | 4619 | } |
1506fcc8 YS |
4620 | } |
4621 | out_free: | |
4622 | free_extent_map(em); | |
4623 | out: | |
fe09e16c | 4624 | btrfs_free_path(path); |
a52f4cd2 | 4625 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
2ac55d41 | 4626 | &cached_state, GFP_NOFS); |
1506fcc8 YS |
4627 | return ret; |
4628 | } | |
4629 | ||
727011e0 CM |
4630 | static void __free_extent_buffer(struct extent_buffer *eb) |
4631 | { | |
6d49ba1b | 4632 | btrfs_leak_debug_del(&eb->leak_list); |
727011e0 CM |
4633 | kmem_cache_free(extent_buffer_cache, eb); |
4634 | } | |
4635 | ||
a26e8c9f | 4636 | int extent_buffer_under_io(struct extent_buffer *eb) |
db7f3436 JB |
4637 | { |
4638 | return (atomic_read(&eb->io_pages) || | |
4639 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4640 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4641 | } | |
4642 | ||
4643 | /* | |
4644 | * Helper for releasing extent buffer page. | |
4645 | */ | |
a50924e3 | 4646 | static void btrfs_release_extent_buffer_page(struct extent_buffer *eb) |
db7f3436 JB |
4647 | { |
4648 | unsigned long index; | |
db7f3436 JB |
4649 | struct page *page; |
4650 | int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4651 | ||
4652 | BUG_ON(extent_buffer_under_io(eb)); | |
4653 | ||
a50924e3 DS |
4654 | index = num_extent_pages(eb->start, eb->len); |
4655 | if (index == 0) | |
db7f3436 JB |
4656 | return; |
4657 | ||
4658 | do { | |
4659 | index--; | |
fb85fc9a | 4660 | page = eb->pages[index]; |
5d2361db FL |
4661 | if (!page) |
4662 | continue; | |
4663 | if (mapped) | |
db7f3436 | 4664 | spin_lock(&page->mapping->private_lock); |
5d2361db FL |
4665 | /* |
4666 | * We do this since we'll remove the pages after we've | |
4667 | * removed the eb from the radix tree, so we could race | |
4668 | * and have this page now attached to the new eb. So | |
4669 | * only clear page_private if it's still connected to | |
4670 | * this eb. | |
4671 | */ | |
4672 | if (PagePrivate(page) && | |
4673 | page->private == (unsigned long)eb) { | |
4674 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4675 | BUG_ON(PageDirty(page)); | |
4676 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 4677 | /* |
5d2361db FL |
4678 | * We need to make sure we haven't be attached |
4679 | * to a new eb. | |
db7f3436 | 4680 | */ |
5d2361db FL |
4681 | ClearPagePrivate(page); |
4682 | set_page_private(page, 0); | |
4683 | /* One for the page private */ | |
db7f3436 JB |
4684 | page_cache_release(page); |
4685 | } | |
5d2361db FL |
4686 | |
4687 | if (mapped) | |
4688 | spin_unlock(&page->mapping->private_lock); | |
4689 | ||
4690 | /* One for when we alloced the page */ | |
4691 | page_cache_release(page); | |
a50924e3 | 4692 | } while (index != 0); |
db7f3436 JB |
4693 | } |
4694 | ||
4695 | /* | |
4696 | * Helper for releasing the extent buffer. | |
4697 | */ | |
4698 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4699 | { | |
a50924e3 | 4700 | btrfs_release_extent_buffer_page(eb); |
db7f3436 JB |
4701 | __free_extent_buffer(eb); |
4702 | } | |
4703 | ||
f28491e0 JB |
4704 | static struct extent_buffer * |
4705 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 4706 | unsigned long len) |
d1310b2e CM |
4707 | { |
4708 | struct extent_buffer *eb = NULL; | |
4709 | ||
d1b5c567 | 4710 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
4711 | eb->start = start; |
4712 | eb->len = len; | |
f28491e0 | 4713 | eb->fs_info = fs_info; |
815a51c7 | 4714 | eb->bflags = 0; |
bd681513 CM |
4715 | rwlock_init(&eb->lock); |
4716 | atomic_set(&eb->write_locks, 0); | |
4717 | atomic_set(&eb->read_locks, 0); | |
4718 | atomic_set(&eb->blocking_readers, 0); | |
4719 | atomic_set(&eb->blocking_writers, 0); | |
4720 | atomic_set(&eb->spinning_readers, 0); | |
4721 | atomic_set(&eb->spinning_writers, 0); | |
5b25f70f | 4722 | eb->lock_nested = 0; |
bd681513 CM |
4723 | init_waitqueue_head(&eb->write_lock_wq); |
4724 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 4725 | |
6d49ba1b ES |
4726 | btrfs_leak_debug_add(&eb->leak_list, &buffers); |
4727 | ||
3083ee2e | 4728 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 4729 | atomic_set(&eb->refs, 1); |
0b32f4bb | 4730 | atomic_set(&eb->io_pages, 0); |
727011e0 | 4731 | |
b8dae313 DS |
4732 | /* |
4733 | * Sanity checks, currently the maximum is 64k covered by 16x 4k pages | |
4734 | */ | |
4735 | BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE | |
4736 | > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
4737 | BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
d1310b2e CM |
4738 | |
4739 | return eb; | |
4740 | } | |
4741 | ||
815a51c7 JS |
4742 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) |
4743 | { | |
4744 | unsigned long i; | |
4745 | struct page *p; | |
4746 | struct extent_buffer *new; | |
4747 | unsigned long num_pages = num_extent_pages(src->start, src->len); | |
4748 | ||
3f556f78 | 4749 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
4750 | if (new == NULL) |
4751 | return NULL; | |
4752 | ||
4753 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4754 | p = alloc_page(GFP_NOFS); |
db7f3436 JB |
4755 | if (!p) { |
4756 | btrfs_release_extent_buffer(new); | |
4757 | return NULL; | |
4758 | } | |
815a51c7 JS |
4759 | attach_extent_buffer_page(new, p); |
4760 | WARN_ON(PageDirty(p)); | |
4761 | SetPageUptodate(p); | |
4762 | new->pages[i] = p; | |
4763 | } | |
4764 | ||
4765 | copy_extent_buffer(new, src, 0, 0, src->len); | |
4766 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); | |
4767 | set_bit(EXTENT_BUFFER_DUMMY, &new->bflags); | |
4768 | ||
4769 | return new; | |
4770 | } | |
4771 | ||
3f556f78 DS |
4772 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
4773 | u64 start) | |
815a51c7 JS |
4774 | { |
4775 | struct extent_buffer *eb; | |
3f556f78 DS |
4776 | unsigned long len; |
4777 | unsigned long num_pages; | |
815a51c7 JS |
4778 | unsigned long i; |
4779 | ||
3f556f78 DS |
4780 | if (!fs_info) { |
4781 | /* | |
4782 | * Called only from tests that don't always have a fs_info | |
4783 | * available, but we know that nodesize is 4096 | |
4784 | */ | |
4785 | len = 4096; | |
4786 | } else { | |
4787 | len = fs_info->tree_root->nodesize; | |
4788 | } | |
4789 | num_pages = num_extent_pages(0, len); | |
4790 | ||
4791 | eb = __alloc_extent_buffer(fs_info, start, len); | |
815a51c7 JS |
4792 | if (!eb) |
4793 | return NULL; | |
4794 | ||
4795 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4796 | eb->pages[i] = alloc_page(GFP_NOFS); |
815a51c7 JS |
4797 | if (!eb->pages[i]) |
4798 | goto err; | |
4799 | } | |
4800 | set_extent_buffer_uptodate(eb); | |
4801 | btrfs_set_header_nritems(eb, 0); | |
4802 | set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4803 | ||
4804 | return eb; | |
4805 | err: | |
84167d19 SB |
4806 | for (; i > 0; i--) |
4807 | __free_page(eb->pages[i - 1]); | |
815a51c7 JS |
4808 | __free_extent_buffer(eb); |
4809 | return NULL; | |
4810 | } | |
4811 | ||
0b32f4bb JB |
4812 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
4813 | { | |
242e18c7 | 4814 | int refs; |
0b32f4bb JB |
4815 | /* the ref bit is tricky. We have to make sure it is set |
4816 | * if we have the buffer dirty. Otherwise the | |
4817 | * code to free a buffer can end up dropping a dirty | |
4818 | * page | |
4819 | * | |
4820 | * Once the ref bit is set, it won't go away while the | |
4821 | * buffer is dirty or in writeback, and it also won't | |
4822 | * go away while we have the reference count on the | |
4823 | * eb bumped. | |
4824 | * | |
4825 | * We can't just set the ref bit without bumping the | |
4826 | * ref on the eb because free_extent_buffer might | |
4827 | * see the ref bit and try to clear it. If this happens | |
4828 | * free_extent_buffer might end up dropping our original | |
4829 | * ref by mistake and freeing the page before we are able | |
4830 | * to add one more ref. | |
4831 | * | |
4832 | * So bump the ref count first, then set the bit. If someone | |
4833 | * beat us to it, drop the ref we added. | |
4834 | */ | |
242e18c7 CM |
4835 | refs = atomic_read(&eb->refs); |
4836 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
4837 | return; | |
4838 | ||
594831c4 JB |
4839 | spin_lock(&eb->refs_lock); |
4840 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 4841 | atomic_inc(&eb->refs); |
594831c4 | 4842 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
4843 | } |
4844 | ||
2457aec6 MG |
4845 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
4846 | struct page *accessed) | |
5df4235e JB |
4847 | { |
4848 | unsigned long num_pages, i; | |
4849 | ||
0b32f4bb JB |
4850 | check_buffer_tree_ref(eb); |
4851 | ||
5df4235e JB |
4852 | num_pages = num_extent_pages(eb->start, eb->len); |
4853 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a DS |
4854 | struct page *p = eb->pages[i]; |
4855 | ||
2457aec6 MG |
4856 | if (p != accessed) |
4857 | mark_page_accessed(p); | |
5df4235e JB |
4858 | } |
4859 | } | |
4860 | ||
f28491e0 JB |
4861 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
4862 | u64 start) | |
452c75c3 CS |
4863 | { |
4864 | struct extent_buffer *eb; | |
4865 | ||
4866 | rcu_read_lock(); | |
f28491e0 JB |
4867 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
4868 | start >> PAGE_CACHE_SHIFT); | |
452c75c3 CS |
4869 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
4870 | rcu_read_unlock(); | |
062c19e9 FM |
4871 | /* |
4872 | * Lock our eb's refs_lock to avoid races with | |
4873 | * free_extent_buffer. When we get our eb it might be flagged | |
4874 | * with EXTENT_BUFFER_STALE and another task running | |
4875 | * free_extent_buffer might have seen that flag set, | |
4876 | * eb->refs == 2, that the buffer isn't under IO (dirty and | |
4877 | * writeback flags not set) and it's still in the tree (flag | |
4878 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process | |
4879 | * of decrementing the extent buffer's reference count twice. | |
4880 | * So here we could race and increment the eb's reference count, | |
4881 | * clear its stale flag, mark it as dirty and drop our reference | |
4882 | * before the other task finishes executing free_extent_buffer, | |
4883 | * which would later result in an attempt to free an extent | |
4884 | * buffer that is dirty. | |
4885 | */ | |
4886 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
4887 | spin_lock(&eb->refs_lock); | |
4888 | spin_unlock(&eb->refs_lock); | |
4889 | } | |
2457aec6 | 4890 | mark_extent_buffer_accessed(eb, NULL); |
452c75c3 CS |
4891 | return eb; |
4892 | } | |
4893 | rcu_read_unlock(); | |
4894 | ||
4895 | return NULL; | |
4896 | } | |
4897 | ||
faa2dbf0 JB |
4898 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
4899 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
ce3e6984 | 4900 | u64 start) |
faa2dbf0 JB |
4901 | { |
4902 | struct extent_buffer *eb, *exists = NULL; | |
4903 | int ret; | |
4904 | ||
4905 | eb = find_extent_buffer(fs_info, start); | |
4906 | if (eb) | |
4907 | return eb; | |
3f556f78 | 4908 | eb = alloc_dummy_extent_buffer(fs_info, start); |
faa2dbf0 JB |
4909 | if (!eb) |
4910 | return NULL; | |
4911 | eb->fs_info = fs_info; | |
4912 | again: | |
4913 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); | |
4914 | if (ret) | |
4915 | goto free_eb; | |
4916 | spin_lock(&fs_info->buffer_lock); | |
4917 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
4918 | start >> PAGE_CACHE_SHIFT, eb); | |
4919 | spin_unlock(&fs_info->buffer_lock); | |
4920 | radix_tree_preload_end(); | |
4921 | if (ret == -EEXIST) { | |
4922 | exists = find_extent_buffer(fs_info, start); | |
4923 | if (exists) | |
4924 | goto free_eb; | |
4925 | else | |
4926 | goto again; | |
4927 | } | |
4928 | check_buffer_tree_ref(eb); | |
4929 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
4930 | ||
4931 | /* | |
4932 | * We will free dummy extent buffer's if they come into | |
4933 | * free_extent_buffer with a ref count of 2, but if we are using this we | |
4934 | * want the buffers to stay in memory until we're done with them, so | |
4935 | * bump the ref count again. | |
4936 | */ | |
4937 | atomic_inc(&eb->refs); | |
4938 | return eb; | |
4939 | free_eb: | |
4940 | btrfs_release_extent_buffer(eb); | |
4941 | return exists; | |
4942 | } | |
4943 | #endif | |
4944 | ||
f28491e0 | 4945 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
ce3e6984 | 4946 | u64 start) |
d1310b2e | 4947 | { |
ce3e6984 | 4948 | unsigned long len = fs_info->tree_root->nodesize; |
d1310b2e CM |
4949 | unsigned long num_pages = num_extent_pages(start, len); |
4950 | unsigned long i; | |
4951 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
4952 | struct extent_buffer *eb; | |
6af118ce | 4953 | struct extent_buffer *exists = NULL; |
d1310b2e | 4954 | struct page *p; |
f28491e0 | 4955 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
d1310b2e | 4956 | int uptodate = 1; |
19fe0a8b | 4957 | int ret; |
d1310b2e | 4958 | |
f28491e0 | 4959 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 4960 | if (eb) |
6af118ce | 4961 | return eb; |
6af118ce | 4962 | |
23d79d81 | 4963 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 4964 | if (!eb) |
d1310b2e CM |
4965 | return NULL; |
4966 | ||
727011e0 | 4967 | for (i = 0; i < num_pages; i++, index++) { |
d1b5c567 | 4968 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
4804b382 | 4969 | if (!p) |
6af118ce | 4970 | goto free_eb; |
4f2de97a JB |
4971 | |
4972 | spin_lock(&mapping->private_lock); | |
4973 | if (PagePrivate(p)) { | |
4974 | /* | |
4975 | * We could have already allocated an eb for this page | |
4976 | * and attached one so lets see if we can get a ref on | |
4977 | * the existing eb, and if we can we know it's good and | |
4978 | * we can just return that one, else we know we can just | |
4979 | * overwrite page->private. | |
4980 | */ | |
4981 | exists = (struct extent_buffer *)p->private; | |
4982 | if (atomic_inc_not_zero(&exists->refs)) { | |
4983 | spin_unlock(&mapping->private_lock); | |
4984 | unlock_page(p); | |
17de39ac | 4985 | page_cache_release(p); |
2457aec6 | 4986 | mark_extent_buffer_accessed(exists, p); |
4f2de97a JB |
4987 | goto free_eb; |
4988 | } | |
5ca64f45 | 4989 | exists = NULL; |
4f2de97a | 4990 | |
0b32f4bb | 4991 | /* |
4f2de97a JB |
4992 | * Do this so attach doesn't complain and we need to |
4993 | * drop the ref the old guy had. | |
4994 | */ | |
4995 | ClearPagePrivate(p); | |
0b32f4bb | 4996 | WARN_ON(PageDirty(p)); |
4f2de97a | 4997 | page_cache_release(p); |
d1310b2e | 4998 | } |
4f2de97a JB |
4999 | attach_extent_buffer_page(eb, p); |
5000 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 5001 | WARN_ON(PageDirty(p)); |
727011e0 | 5002 | eb->pages[i] = p; |
d1310b2e CM |
5003 | if (!PageUptodate(p)) |
5004 | uptodate = 0; | |
eb14ab8e CM |
5005 | |
5006 | /* | |
5007 | * see below about how we avoid a nasty race with release page | |
5008 | * and why we unlock later | |
5009 | */ | |
d1310b2e CM |
5010 | } |
5011 | if (uptodate) | |
b4ce94de | 5012 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 5013 | again: |
19fe0a8b MX |
5014 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); |
5015 | if (ret) | |
5016 | goto free_eb; | |
5017 | ||
f28491e0 JB |
5018 | spin_lock(&fs_info->buffer_lock); |
5019 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
5020 | start >> PAGE_CACHE_SHIFT, eb); | |
5021 | spin_unlock(&fs_info->buffer_lock); | |
452c75c3 | 5022 | radix_tree_preload_end(); |
19fe0a8b | 5023 | if (ret == -EEXIST) { |
f28491e0 | 5024 | exists = find_extent_buffer(fs_info, start); |
452c75c3 CS |
5025 | if (exists) |
5026 | goto free_eb; | |
5027 | else | |
115391d2 | 5028 | goto again; |
6af118ce | 5029 | } |
6af118ce | 5030 | /* add one reference for the tree */ |
0b32f4bb | 5031 | check_buffer_tree_ref(eb); |
34b41ace | 5032 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
5033 | |
5034 | /* | |
5035 | * there is a race where release page may have | |
5036 | * tried to find this extent buffer in the radix | |
5037 | * but failed. It will tell the VM it is safe to | |
5038 | * reclaim the, and it will clear the page private bit. | |
5039 | * We must make sure to set the page private bit properly | |
5040 | * after the extent buffer is in the radix tree so | |
5041 | * it doesn't get lost | |
5042 | */ | |
727011e0 CM |
5043 | SetPageChecked(eb->pages[0]); |
5044 | for (i = 1; i < num_pages; i++) { | |
fb85fc9a | 5045 | p = eb->pages[i]; |
727011e0 CM |
5046 | ClearPageChecked(p); |
5047 | unlock_page(p); | |
5048 | } | |
5049 | unlock_page(eb->pages[0]); | |
d1310b2e CM |
5050 | return eb; |
5051 | ||
6af118ce | 5052 | free_eb: |
5ca64f45 | 5053 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
5054 | for (i = 0; i < num_pages; i++) { |
5055 | if (eb->pages[i]) | |
5056 | unlock_page(eb->pages[i]); | |
5057 | } | |
eb14ab8e | 5058 | |
897ca6e9 | 5059 | btrfs_release_extent_buffer(eb); |
6af118ce | 5060 | return exists; |
d1310b2e | 5061 | } |
d1310b2e | 5062 | |
3083ee2e JB |
5063 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
5064 | { | |
5065 | struct extent_buffer *eb = | |
5066 | container_of(head, struct extent_buffer, rcu_head); | |
5067 | ||
5068 | __free_extent_buffer(eb); | |
5069 | } | |
5070 | ||
3083ee2e | 5071 | /* Expects to have eb->eb_lock already held */ |
f7a52a40 | 5072 | static int release_extent_buffer(struct extent_buffer *eb) |
3083ee2e JB |
5073 | { |
5074 | WARN_ON(atomic_read(&eb->refs) == 0); | |
5075 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 5076 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 5077 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 5078 | |
815a51c7 | 5079 | spin_unlock(&eb->refs_lock); |
3083ee2e | 5080 | |
f28491e0 JB |
5081 | spin_lock(&fs_info->buffer_lock); |
5082 | radix_tree_delete(&fs_info->buffer_radix, | |
815a51c7 | 5083 | eb->start >> PAGE_CACHE_SHIFT); |
f28491e0 | 5084 | spin_unlock(&fs_info->buffer_lock); |
34b41ace JB |
5085 | } else { |
5086 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 5087 | } |
3083ee2e JB |
5088 | |
5089 | /* Should be safe to release our pages at this point */ | |
a50924e3 | 5090 | btrfs_release_extent_buffer_page(eb); |
bcb7e449 JB |
5091 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
5092 | if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) { | |
5093 | __free_extent_buffer(eb); | |
5094 | return 1; | |
5095 | } | |
5096 | #endif | |
3083ee2e | 5097 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 5098 | return 1; |
3083ee2e JB |
5099 | } |
5100 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
5101 | |
5102 | return 0; | |
3083ee2e JB |
5103 | } |
5104 | ||
d1310b2e CM |
5105 | void free_extent_buffer(struct extent_buffer *eb) |
5106 | { | |
242e18c7 CM |
5107 | int refs; |
5108 | int old; | |
d1310b2e CM |
5109 | if (!eb) |
5110 | return; | |
5111 | ||
242e18c7 CM |
5112 | while (1) { |
5113 | refs = atomic_read(&eb->refs); | |
5114 | if (refs <= 3) | |
5115 | break; | |
5116 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
5117 | if (old == refs) | |
5118 | return; | |
5119 | } | |
5120 | ||
3083ee2e | 5121 | spin_lock(&eb->refs_lock); |
815a51c7 JS |
5122 | if (atomic_read(&eb->refs) == 2 && |
5123 | test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) | |
5124 | atomic_dec(&eb->refs); | |
5125 | ||
3083ee2e JB |
5126 | if (atomic_read(&eb->refs) == 2 && |
5127 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 5128 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
5129 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5130 | atomic_dec(&eb->refs); | |
5131 | ||
5132 | /* | |
5133 | * I know this is terrible, but it's temporary until we stop tracking | |
5134 | * the uptodate bits and such for the extent buffers. | |
5135 | */ | |
f7a52a40 | 5136 | release_extent_buffer(eb); |
3083ee2e JB |
5137 | } |
5138 | ||
5139 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
5140 | { | |
5141 | if (!eb) | |
d1310b2e CM |
5142 | return; |
5143 | ||
3083ee2e JB |
5144 | spin_lock(&eb->refs_lock); |
5145 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
5146 | ||
0b32f4bb | 5147 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
5148 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5149 | atomic_dec(&eb->refs); | |
f7a52a40 | 5150 | release_extent_buffer(eb); |
d1310b2e | 5151 | } |
d1310b2e | 5152 | |
1d4284bd | 5153 | void clear_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5154 | { |
d1310b2e CM |
5155 | unsigned long i; |
5156 | unsigned long num_pages; | |
5157 | struct page *page; | |
5158 | ||
d1310b2e CM |
5159 | num_pages = num_extent_pages(eb->start, eb->len); |
5160 | ||
5161 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 5162 | page = eb->pages[i]; |
b9473439 | 5163 | if (!PageDirty(page)) |
d2c3f4f6 CM |
5164 | continue; |
5165 | ||
a61e6f29 | 5166 | lock_page(page); |
eb14ab8e CM |
5167 | WARN_ON(!PagePrivate(page)); |
5168 | ||
d1310b2e | 5169 | clear_page_dirty_for_io(page); |
0ee0fda0 | 5170 | spin_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
5171 | if (!PageDirty(page)) { |
5172 | radix_tree_tag_clear(&page->mapping->page_tree, | |
5173 | page_index(page), | |
5174 | PAGECACHE_TAG_DIRTY); | |
5175 | } | |
0ee0fda0 | 5176 | spin_unlock_irq(&page->mapping->tree_lock); |
bf0da8c1 | 5177 | ClearPageError(page); |
a61e6f29 | 5178 | unlock_page(page); |
d1310b2e | 5179 | } |
0b32f4bb | 5180 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 5181 | } |
d1310b2e | 5182 | |
0b32f4bb | 5183 | int set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e CM |
5184 | { |
5185 | unsigned long i; | |
5186 | unsigned long num_pages; | |
b9473439 | 5187 | int was_dirty = 0; |
d1310b2e | 5188 | |
0b32f4bb JB |
5189 | check_buffer_tree_ref(eb); |
5190 | ||
b9473439 | 5191 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 5192 | |
d1310b2e | 5193 | num_pages = num_extent_pages(eb->start, eb->len); |
3083ee2e | 5194 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
5195 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
5196 | ||
b9473439 | 5197 | for (i = 0; i < num_pages; i++) |
fb85fc9a | 5198 | set_page_dirty(eb->pages[i]); |
b9473439 | 5199 | return was_dirty; |
d1310b2e | 5200 | } |
d1310b2e | 5201 | |
0b32f4bb | 5202 | int clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 CM |
5203 | { |
5204 | unsigned long i; | |
5205 | struct page *page; | |
5206 | unsigned long num_pages; | |
5207 | ||
b4ce94de | 5208 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
0b32f4bb | 5209 | num_pages = num_extent_pages(eb->start, eb->len); |
1259ab75 | 5210 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5211 | page = eb->pages[i]; |
33958dc6 CM |
5212 | if (page) |
5213 | ClearPageUptodate(page); | |
1259ab75 CM |
5214 | } |
5215 | return 0; | |
5216 | } | |
5217 | ||
0b32f4bb | 5218 | int set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e CM |
5219 | { |
5220 | unsigned long i; | |
5221 | struct page *page; | |
5222 | unsigned long num_pages; | |
5223 | ||
0b32f4bb | 5224 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 5225 | num_pages = num_extent_pages(eb->start, eb->len); |
d1310b2e | 5226 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5227 | page = eb->pages[i]; |
d1310b2e CM |
5228 | SetPageUptodate(page); |
5229 | } | |
5230 | return 0; | |
5231 | } | |
d1310b2e | 5232 | |
0b32f4bb | 5233 | int extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 5234 | { |
0b32f4bb | 5235 | return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 5236 | } |
d1310b2e CM |
5237 | |
5238 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
bb82ab88 | 5239 | struct extent_buffer *eb, u64 start, int wait, |
f188591e | 5240 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
5241 | { |
5242 | unsigned long i; | |
5243 | unsigned long start_i; | |
5244 | struct page *page; | |
5245 | int err; | |
5246 | int ret = 0; | |
ce9adaa5 CM |
5247 | int locked_pages = 0; |
5248 | int all_uptodate = 1; | |
d1310b2e | 5249 | unsigned long num_pages; |
727011e0 | 5250 | unsigned long num_reads = 0; |
a86c12c7 | 5251 | struct bio *bio = NULL; |
c8b97818 | 5252 | unsigned long bio_flags = 0; |
a86c12c7 | 5253 | |
b4ce94de | 5254 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
5255 | return 0; |
5256 | ||
d1310b2e CM |
5257 | if (start) { |
5258 | WARN_ON(start < eb->start); | |
5259 | start_i = (start >> PAGE_CACHE_SHIFT) - | |
5260 | (eb->start >> PAGE_CACHE_SHIFT); | |
5261 | } else { | |
5262 | start_i = 0; | |
5263 | } | |
5264 | ||
5265 | num_pages = num_extent_pages(eb->start, eb->len); | |
5266 | for (i = start_i; i < num_pages; i++) { | |
fb85fc9a | 5267 | page = eb->pages[i]; |
bb82ab88 | 5268 | if (wait == WAIT_NONE) { |
2db04966 | 5269 | if (!trylock_page(page)) |
ce9adaa5 | 5270 | goto unlock_exit; |
d1310b2e CM |
5271 | } else { |
5272 | lock_page(page); | |
5273 | } | |
ce9adaa5 | 5274 | locked_pages++; |
727011e0 CM |
5275 | if (!PageUptodate(page)) { |
5276 | num_reads++; | |
ce9adaa5 | 5277 | all_uptodate = 0; |
727011e0 | 5278 | } |
ce9adaa5 CM |
5279 | } |
5280 | if (all_uptodate) { | |
5281 | if (start_i == 0) | |
b4ce94de | 5282 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
5283 | goto unlock_exit; |
5284 | } | |
5285 | ||
656f30db | 5286 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 5287 | eb->read_mirror = 0; |
0b32f4bb | 5288 | atomic_set(&eb->io_pages, num_reads); |
ce9adaa5 | 5289 | for (i = start_i; i < num_pages; i++) { |
fb85fc9a | 5290 | page = eb->pages[i]; |
ce9adaa5 | 5291 | if (!PageUptodate(page)) { |
f188591e | 5292 | ClearPageError(page); |
a86c12c7 | 5293 | err = __extent_read_full_page(tree, page, |
f188591e | 5294 | get_extent, &bio, |
d4c7ca86 JB |
5295 | mirror_num, &bio_flags, |
5296 | READ | REQ_META); | |
d397712b | 5297 | if (err) |
d1310b2e | 5298 | ret = err; |
d1310b2e CM |
5299 | } else { |
5300 | unlock_page(page); | |
5301 | } | |
5302 | } | |
5303 | ||
355808c2 | 5304 | if (bio) { |
d4c7ca86 JB |
5305 | err = submit_one_bio(READ | REQ_META, bio, mirror_num, |
5306 | bio_flags); | |
79787eaa JM |
5307 | if (err) |
5308 | return err; | |
355808c2 | 5309 | } |
a86c12c7 | 5310 | |
bb82ab88 | 5311 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 5312 | return ret; |
d397712b | 5313 | |
d1310b2e | 5314 | for (i = start_i; i < num_pages; i++) { |
fb85fc9a | 5315 | page = eb->pages[i]; |
d1310b2e | 5316 | wait_on_page_locked(page); |
d397712b | 5317 | if (!PageUptodate(page)) |
d1310b2e | 5318 | ret = -EIO; |
d1310b2e | 5319 | } |
d397712b | 5320 | |
d1310b2e | 5321 | return ret; |
ce9adaa5 CM |
5322 | |
5323 | unlock_exit: | |
5324 | i = start_i; | |
d397712b | 5325 | while (locked_pages > 0) { |
fb85fc9a | 5326 | page = eb->pages[i]; |
ce9adaa5 CM |
5327 | i++; |
5328 | unlock_page(page); | |
5329 | locked_pages--; | |
5330 | } | |
5331 | return ret; | |
d1310b2e | 5332 | } |
d1310b2e CM |
5333 | |
5334 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
5335 | unsigned long start, | |
5336 | unsigned long len) | |
5337 | { | |
5338 | size_t cur; | |
5339 | size_t offset; | |
5340 | struct page *page; | |
5341 | char *kaddr; | |
5342 | char *dst = (char *)dstv; | |
5343 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5344 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
5345 | |
5346 | WARN_ON(start > eb->len); | |
5347 | WARN_ON(start + len > eb->start + eb->len); | |
5348 | ||
778746b5 | 5349 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5350 | |
d397712b | 5351 | while (len > 0) { |
fb85fc9a | 5352 | page = eb->pages[i]; |
d1310b2e CM |
5353 | |
5354 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
a6591715 | 5355 | kaddr = page_address(page); |
d1310b2e | 5356 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
5357 | |
5358 | dst += cur; | |
5359 | len -= cur; | |
5360 | offset = 0; | |
5361 | i++; | |
5362 | } | |
5363 | } | |
d1310b2e | 5364 | |
550ac1d8 GH |
5365 | int read_extent_buffer_to_user(struct extent_buffer *eb, void __user *dstv, |
5366 | unsigned long start, | |
5367 | unsigned long len) | |
5368 | { | |
5369 | size_t cur; | |
5370 | size_t offset; | |
5371 | struct page *page; | |
5372 | char *kaddr; | |
5373 | char __user *dst = (char __user *)dstv; | |
5374 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5375 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5376 | int ret = 0; | |
5377 | ||
5378 | WARN_ON(start > eb->len); | |
5379 | WARN_ON(start + len > eb->start + eb->len); | |
5380 | ||
5381 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); | |
5382 | ||
5383 | while (len > 0) { | |
fb85fc9a | 5384 | page = eb->pages[i]; |
550ac1d8 GH |
5385 | |
5386 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
5387 | kaddr = page_address(page); | |
5388 | if (copy_to_user(dst, kaddr + offset, cur)) { | |
5389 | ret = -EFAULT; | |
5390 | break; | |
5391 | } | |
5392 | ||
5393 | dst += cur; | |
5394 | len -= cur; | |
5395 | offset = 0; | |
5396 | i++; | |
5397 | } | |
5398 | ||
5399 | return ret; | |
5400 | } | |
5401 | ||
d1310b2e | 5402 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, |
a6591715 | 5403 | unsigned long min_len, char **map, |
d1310b2e | 5404 | unsigned long *map_start, |
a6591715 | 5405 | unsigned long *map_len) |
d1310b2e CM |
5406 | { |
5407 | size_t offset = start & (PAGE_CACHE_SIZE - 1); | |
5408 | char *kaddr; | |
5409 | struct page *p; | |
5410 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5411 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5412 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
5413 | PAGE_CACHE_SHIFT; | |
5414 | ||
5415 | if (i != end_i) | |
5416 | return -EINVAL; | |
5417 | ||
5418 | if (i == 0) { | |
5419 | offset = start_offset; | |
5420 | *map_start = 0; | |
5421 | } else { | |
5422 | offset = 0; | |
5423 | *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; | |
5424 | } | |
d397712b | 5425 | |
d1310b2e | 5426 | if (start + min_len > eb->len) { |
31b1a2bd | 5427 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, " |
c1c9ff7c GU |
5428 | "wanted %lu %lu\n", |
5429 | eb->start, eb->len, start, min_len); | |
85026533 | 5430 | return -EINVAL; |
d1310b2e CM |
5431 | } |
5432 | ||
fb85fc9a | 5433 | p = eb->pages[i]; |
a6591715 | 5434 | kaddr = page_address(p); |
d1310b2e CM |
5435 | *map = kaddr + offset; |
5436 | *map_len = PAGE_CACHE_SIZE - offset; | |
5437 | return 0; | |
5438 | } | |
d1310b2e | 5439 | |
d1310b2e CM |
5440 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, |
5441 | unsigned long start, | |
5442 | unsigned long len) | |
5443 | { | |
5444 | size_t cur; | |
5445 | size_t offset; | |
5446 | struct page *page; | |
5447 | char *kaddr; | |
5448 | char *ptr = (char *)ptrv; | |
5449 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5450 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5451 | int ret = 0; | |
5452 | ||
5453 | WARN_ON(start > eb->len); | |
5454 | WARN_ON(start + len > eb->start + eb->len); | |
5455 | ||
778746b5 | 5456 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5457 | |
d397712b | 5458 | while (len > 0) { |
fb85fc9a | 5459 | page = eb->pages[i]; |
d1310b2e CM |
5460 | |
5461 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
5462 | ||
a6591715 | 5463 | kaddr = page_address(page); |
d1310b2e | 5464 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
5465 | if (ret) |
5466 | break; | |
5467 | ||
5468 | ptr += cur; | |
5469 | len -= cur; | |
5470 | offset = 0; | |
5471 | i++; | |
5472 | } | |
5473 | return ret; | |
5474 | } | |
d1310b2e CM |
5475 | |
5476 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
5477 | unsigned long start, unsigned long len) | |
5478 | { | |
5479 | size_t cur; | |
5480 | size_t offset; | |
5481 | struct page *page; | |
5482 | char *kaddr; | |
5483 | char *src = (char *)srcv; | |
5484 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5485 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5486 | ||
5487 | WARN_ON(start > eb->len); | |
5488 | WARN_ON(start + len > eb->start + eb->len); | |
5489 | ||
778746b5 | 5490 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5491 | |
d397712b | 5492 | while (len > 0) { |
fb85fc9a | 5493 | page = eb->pages[i]; |
d1310b2e CM |
5494 | WARN_ON(!PageUptodate(page)); |
5495 | ||
5496 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 5497 | kaddr = page_address(page); |
d1310b2e | 5498 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
5499 | |
5500 | src += cur; | |
5501 | len -= cur; | |
5502 | offset = 0; | |
5503 | i++; | |
5504 | } | |
5505 | } | |
d1310b2e CM |
5506 | |
5507 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
5508 | unsigned long start, unsigned long len) | |
5509 | { | |
5510 | size_t cur; | |
5511 | size_t offset; | |
5512 | struct page *page; | |
5513 | char *kaddr; | |
5514 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5515 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5516 | ||
5517 | WARN_ON(start > eb->len); | |
5518 | WARN_ON(start + len > eb->start + eb->len); | |
5519 | ||
778746b5 | 5520 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5521 | |
d397712b | 5522 | while (len > 0) { |
fb85fc9a | 5523 | page = eb->pages[i]; |
d1310b2e CM |
5524 | WARN_ON(!PageUptodate(page)); |
5525 | ||
5526 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 5527 | kaddr = page_address(page); |
d1310b2e | 5528 | memset(kaddr + offset, c, cur); |
d1310b2e CM |
5529 | |
5530 | len -= cur; | |
5531 | offset = 0; | |
5532 | i++; | |
5533 | } | |
5534 | } | |
d1310b2e CM |
5535 | |
5536 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
5537 | unsigned long dst_offset, unsigned long src_offset, | |
5538 | unsigned long len) | |
5539 | { | |
5540 | u64 dst_len = dst->len; | |
5541 | size_t cur; | |
5542 | size_t offset; | |
5543 | struct page *page; | |
5544 | char *kaddr; | |
5545 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5546 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
5547 | ||
5548 | WARN_ON(src->len != dst_len); | |
5549 | ||
5550 | offset = (start_offset + dst_offset) & | |
778746b5 | 5551 | (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5552 | |
d397712b | 5553 | while (len > 0) { |
fb85fc9a | 5554 | page = dst->pages[i]; |
d1310b2e CM |
5555 | WARN_ON(!PageUptodate(page)); |
5556 | ||
5557 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
5558 | ||
a6591715 | 5559 | kaddr = page_address(page); |
d1310b2e | 5560 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
5561 | |
5562 | src_offset += cur; | |
5563 | len -= cur; | |
5564 | offset = 0; | |
5565 | i++; | |
5566 | } | |
5567 | } | |
d1310b2e | 5568 | |
3387206f ST |
5569 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
5570 | { | |
5571 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
5572 | return distance < len; | |
5573 | } | |
5574 | ||
d1310b2e CM |
5575 | static void copy_pages(struct page *dst_page, struct page *src_page, |
5576 | unsigned long dst_off, unsigned long src_off, | |
5577 | unsigned long len) | |
5578 | { | |
a6591715 | 5579 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 5580 | char *src_kaddr; |
727011e0 | 5581 | int must_memmove = 0; |
d1310b2e | 5582 | |
3387206f | 5583 | if (dst_page != src_page) { |
a6591715 | 5584 | src_kaddr = page_address(src_page); |
3387206f | 5585 | } else { |
d1310b2e | 5586 | src_kaddr = dst_kaddr; |
727011e0 CM |
5587 | if (areas_overlap(src_off, dst_off, len)) |
5588 | must_memmove = 1; | |
3387206f | 5589 | } |
d1310b2e | 5590 | |
727011e0 CM |
5591 | if (must_memmove) |
5592 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
5593 | else | |
5594 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
5595 | } |
5596 | ||
5597 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5598 | unsigned long src_offset, unsigned long len) | |
5599 | { | |
5600 | size_t cur; | |
5601 | size_t dst_off_in_page; | |
5602 | size_t src_off_in_page; | |
5603 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5604 | unsigned long dst_i; | |
5605 | unsigned long src_i; | |
5606 | ||
5607 | if (src_offset + len > dst->len) { | |
f14d104d DS |
5608 | btrfs_err(dst->fs_info, |
5609 | "memmove bogus src_offset %lu move " | |
5610 | "len %lu dst len %lu", src_offset, len, dst->len); | |
d1310b2e CM |
5611 | BUG_ON(1); |
5612 | } | |
5613 | if (dst_offset + len > dst->len) { | |
f14d104d DS |
5614 | btrfs_err(dst->fs_info, |
5615 | "memmove bogus dst_offset %lu move " | |
5616 | "len %lu dst len %lu", dst_offset, len, dst->len); | |
d1310b2e CM |
5617 | BUG_ON(1); |
5618 | } | |
5619 | ||
d397712b | 5620 | while (len > 0) { |
d1310b2e | 5621 | dst_off_in_page = (start_offset + dst_offset) & |
778746b5 | 5622 | (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5623 | src_off_in_page = (start_offset + src_offset) & |
778746b5 | 5624 | (PAGE_CACHE_SIZE - 1); |
d1310b2e CM |
5625 | |
5626 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
5627 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
5628 | ||
5629 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
5630 | src_off_in_page)); | |
5631 | cur = min_t(unsigned long, cur, | |
5632 | (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); | |
5633 | ||
fb85fc9a | 5634 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5635 | dst_off_in_page, src_off_in_page, cur); |
5636 | ||
5637 | src_offset += cur; | |
5638 | dst_offset += cur; | |
5639 | len -= cur; | |
5640 | } | |
5641 | } | |
d1310b2e CM |
5642 | |
5643 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5644 | unsigned long src_offset, unsigned long len) | |
5645 | { | |
5646 | size_t cur; | |
5647 | size_t dst_off_in_page; | |
5648 | size_t src_off_in_page; | |
5649 | unsigned long dst_end = dst_offset + len - 1; | |
5650 | unsigned long src_end = src_offset + len - 1; | |
5651 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5652 | unsigned long dst_i; | |
5653 | unsigned long src_i; | |
5654 | ||
5655 | if (src_offset + len > dst->len) { | |
f14d104d DS |
5656 | btrfs_err(dst->fs_info, "memmove bogus src_offset %lu move " |
5657 | "len %lu len %lu", src_offset, len, dst->len); | |
d1310b2e CM |
5658 | BUG_ON(1); |
5659 | } | |
5660 | if (dst_offset + len > dst->len) { | |
f14d104d DS |
5661 | btrfs_err(dst->fs_info, "memmove bogus dst_offset %lu move " |
5662 | "len %lu len %lu", dst_offset, len, dst->len); | |
d1310b2e CM |
5663 | BUG_ON(1); |
5664 | } | |
727011e0 | 5665 | if (dst_offset < src_offset) { |
d1310b2e CM |
5666 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
5667 | return; | |
5668 | } | |
d397712b | 5669 | while (len > 0) { |
d1310b2e CM |
5670 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; |
5671 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
5672 | ||
5673 | dst_off_in_page = (start_offset + dst_end) & | |
778746b5 | 5674 | (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5675 | src_off_in_page = (start_offset + src_end) & |
778746b5 | 5676 | (PAGE_CACHE_SIZE - 1); |
d1310b2e CM |
5677 | |
5678 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
5679 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 5680 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5681 | dst_off_in_page - cur + 1, |
5682 | src_off_in_page - cur + 1, cur); | |
5683 | ||
5684 | dst_end -= cur; | |
5685 | src_end -= cur; | |
5686 | len -= cur; | |
5687 | } | |
5688 | } | |
6af118ce | 5689 | |
f7a52a40 | 5690 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 5691 | { |
6af118ce | 5692 | struct extent_buffer *eb; |
6af118ce | 5693 | |
3083ee2e JB |
5694 | /* |
5695 | * We need to make sure noboody is attaching this page to an eb right | |
5696 | * now. | |
5697 | */ | |
5698 | spin_lock(&page->mapping->private_lock); | |
5699 | if (!PagePrivate(page)) { | |
5700 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 5701 | return 1; |
45f49bce | 5702 | } |
6af118ce | 5703 | |
3083ee2e JB |
5704 | eb = (struct extent_buffer *)page->private; |
5705 | BUG_ON(!eb); | |
19fe0a8b MX |
5706 | |
5707 | /* | |
3083ee2e JB |
5708 | * This is a little awful but should be ok, we need to make sure that |
5709 | * the eb doesn't disappear out from under us while we're looking at | |
5710 | * this page. | |
19fe0a8b | 5711 | */ |
3083ee2e | 5712 | spin_lock(&eb->refs_lock); |
0b32f4bb | 5713 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
5714 | spin_unlock(&eb->refs_lock); |
5715 | spin_unlock(&page->mapping->private_lock); | |
5716 | return 0; | |
b9473439 | 5717 | } |
3083ee2e | 5718 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 5719 | |
19fe0a8b | 5720 | /* |
3083ee2e JB |
5721 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
5722 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 5723 | */ |
3083ee2e JB |
5724 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
5725 | spin_unlock(&eb->refs_lock); | |
5726 | return 0; | |
b9473439 | 5727 | } |
19fe0a8b | 5728 | |
f7a52a40 | 5729 | return release_extent_buffer(eb); |
6af118ce | 5730 | } |