]>
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 DS |
98 | if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) { |
99 | printk_ratelimited(KERN_DEBUG | |
efe120a0 | 100 | "BTRFS: %s: ino %llu isize %llu odd range [%llu,%llu]\n", |
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 | ||
0b32f4bb | 134 | static noinline void flush_write_bio(void *data); |
c2d904e0 JM |
135 | static inline struct btrfs_fs_info * |
136 | tree_fs_info(struct extent_io_tree *tree) | |
137 | { | |
a5dee37d JB |
138 | if (!tree->mapping) |
139 | return NULL; | |
c2d904e0 JM |
140 | return btrfs_sb(tree->mapping->host->i_sb); |
141 | } | |
0b32f4bb | 142 | |
d1310b2e CM |
143 | int __init extent_io_init(void) |
144 | { | |
837e1972 | 145 | extent_state_cache = kmem_cache_create("btrfs_extent_state", |
9601e3f6 CH |
146 | sizeof(struct extent_state), 0, |
147 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
148 | if (!extent_state_cache) |
149 | return -ENOMEM; | |
150 | ||
837e1972 | 151 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 CH |
152 | sizeof(struct extent_buffer), 0, |
153 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
154 | if (!extent_buffer_cache) |
155 | goto free_state_cache; | |
9be3395b CM |
156 | |
157 | btrfs_bioset = bioset_create(BIO_POOL_SIZE, | |
158 | offsetof(struct btrfs_io_bio, bio)); | |
159 | if (!btrfs_bioset) | |
160 | goto free_buffer_cache; | |
b208c2f7 DW |
161 | |
162 | if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE)) | |
163 | goto free_bioset; | |
164 | ||
d1310b2e CM |
165 | return 0; |
166 | ||
b208c2f7 DW |
167 | free_bioset: |
168 | bioset_free(btrfs_bioset); | |
169 | btrfs_bioset = NULL; | |
170 | ||
9be3395b CM |
171 | free_buffer_cache: |
172 | kmem_cache_destroy(extent_buffer_cache); | |
173 | extent_buffer_cache = NULL; | |
174 | ||
d1310b2e CM |
175 | free_state_cache: |
176 | kmem_cache_destroy(extent_state_cache); | |
9be3395b | 177 | extent_state_cache = NULL; |
d1310b2e CM |
178 | return -ENOMEM; |
179 | } | |
180 | ||
181 | void extent_io_exit(void) | |
182 | { | |
6d49ba1b | 183 | btrfs_leak_debug_check(); |
8c0a8537 KS |
184 | |
185 | /* | |
186 | * Make sure all delayed rcu free are flushed before we | |
187 | * destroy caches. | |
188 | */ | |
189 | rcu_barrier(); | |
d1310b2e CM |
190 | if (extent_state_cache) |
191 | kmem_cache_destroy(extent_state_cache); | |
192 | if (extent_buffer_cache) | |
193 | kmem_cache_destroy(extent_buffer_cache); | |
9be3395b CM |
194 | if (btrfs_bioset) |
195 | bioset_free(btrfs_bioset); | |
d1310b2e CM |
196 | } |
197 | ||
198 | void extent_io_tree_init(struct extent_io_tree *tree, | |
f993c883 | 199 | struct address_space *mapping) |
d1310b2e | 200 | { |
6bef4d31 | 201 | tree->state = RB_ROOT; |
d1310b2e CM |
202 | tree->ops = NULL; |
203 | tree->dirty_bytes = 0; | |
70dec807 | 204 | spin_lock_init(&tree->lock); |
d1310b2e | 205 | tree->mapping = mapping; |
d1310b2e | 206 | } |
d1310b2e | 207 | |
b2950863 | 208 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
209 | { |
210 | struct extent_state *state; | |
d1310b2e CM |
211 | |
212 | state = kmem_cache_alloc(extent_state_cache, mask); | |
2b114d1d | 213 | if (!state) |
d1310b2e CM |
214 | return state; |
215 | state->state = 0; | |
d1310b2e | 216 | state->private = 0; |
27a3507d | 217 | RB_CLEAR_NODE(&state->rb_node); |
6d49ba1b | 218 | btrfs_leak_debug_add(&state->leak_list, &states); |
d1310b2e CM |
219 | atomic_set(&state->refs, 1); |
220 | init_waitqueue_head(&state->wq); | |
143bede5 | 221 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
222 | return state; |
223 | } | |
d1310b2e | 224 | |
4845e44f | 225 | void free_extent_state(struct extent_state *state) |
d1310b2e | 226 | { |
d1310b2e CM |
227 | if (!state) |
228 | return; | |
229 | if (atomic_dec_and_test(&state->refs)) { | |
27a3507d | 230 | WARN_ON(extent_state_in_tree(state)); |
6d49ba1b | 231 | btrfs_leak_debug_del(&state->leak_list); |
143bede5 | 232 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
233 | kmem_cache_free(extent_state_cache, state); |
234 | } | |
235 | } | |
d1310b2e | 236 | |
f2071b21 FM |
237 | static struct rb_node *tree_insert(struct rb_root *root, |
238 | struct rb_node *search_start, | |
239 | u64 offset, | |
12cfbad9 FDBM |
240 | struct rb_node *node, |
241 | struct rb_node ***p_in, | |
242 | struct rb_node **parent_in) | |
d1310b2e | 243 | { |
f2071b21 | 244 | struct rb_node **p; |
d397712b | 245 | struct rb_node *parent = NULL; |
d1310b2e CM |
246 | struct tree_entry *entry; |
247 | ||
12cfbad9 FDBM |
248 | if (p_in && parent_in) { |
249 | p = *p_in; | |
250 | parent = *parent_in; | |
251 | goto do_insert; | |
252 | } | |
253 | ||
f2071b21 | 254 | p = search_start ? &search_start : &root->rb_node; |
d397712b | 255 | while (*p) { |
d1310b2e CM |
256 | parent = *p; |
257 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
258 | ||
259 | if (offset < entry->start) | |
260 | p = &(*p)->rb_left; | |
261 | else if (offset > entry->end) | |
262 | p = &(*p)->rb_right; | |
263 | else | |
264 | return parent; | |
265 | } | |
266 | ||
12cfbad9 | 267 | do_insert: |
d1310b2e CM |
268 | rb_link_node(node, parent, p); |
269 | rb_insert_color(node, root); | |
270 | return NULL; | |
271 | } | |
272 | ||
80ea96b1 | 273 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
12cfbad9 FDBM |
274 | struct rb_node **prev_ret, |
275 | struct rb_node **next_ret, | |
276 | struct rb_node ***p_ret, | |
277 | struct rb_node **parent_ret) | |
d1310b2e | 278 | { |
80ea96b1 | 279 | struct rb_root *root = &tree->state; |
12cfbad9 | 280 | struct rb_node **n = &root->rb_node; |
d1310b2e CM |
281 | struct rb_node *prev = NULL; |
282 | struct rb_node *orig_prev = NULL; | |
283 | struct tree_entry *entry; | |
284 | struct tree_entry *prev_entry = NULL; | |
285 | ||
12cfbad9 FDBM |
286 | while (*n) { |
287 | prev = *n; | |
288 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
d1310b2e CM |
289 | prev_entry = entry; |
290 | ||
291 | if (offset < entry->start) | |
12cfbad9 | 292 | n = &(*n)->rb_left; |
d1310b2e | 293 | else if (offset > entry->end) |
12cfbad9 | 294 | n = &(*n)->rb_right; |
d397712b | 295 | else |
12cfbad9 | 296 | return *n; |
d1310b2e CM |
297 | } |
298 | ||
12cfbad9 FDBM |
299 | if (p_ret) |
300 | *p_ret = n; | |
301 | if (parent_ret) | |
302 | *parent_ret = prev; | |
303 | ||
d1310b2e CM |
304 | if (prev_ret) { |
305 | orig_prev = prev; | |
d397712b | 306 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
307 | prev = rb_next(prev); |
308 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
309 | } | |
310 | *prev_ret = prev; | |
311 | prev = orig_prev; | |
312 | } | |
313 | ||
314 | if (next_ret) { | |
315 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 316 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
317 | prev = rb_prev(prev); |
318 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
319 | } | |
320 | *next_ret = prev; | |
321 | } | |
322 | return NULL; | |
323 | } | |
324 | ||
12cfbad9 FDBM |
325 | static inline struct rb_node * |
326 | tree_search_for_insert(struct extent_io_tree *tree, | |
327 | u64 offset, | |
328 | struct rb_node ***p_ret, | |
329 | struct rb_node **parent_ret) | |
d1310b2e | 330 | { |
70dec807 | 331 | struct rb_node *prev = NULL; |
d1310b2e | 332 | struct rb_node *ret; |
70dec807 | 333 | |
12cfbad9 | 334 | ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret); |
d397712b | 335 | if (!ret) |
d1310b2e CM |
336 | return prev; |
337 | return ret; | |
338 | } | |
339 | ||
12cfbad9 FDBM |
340 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
341 | u64 offset) | |
342 | { | |
343 | return tree_search_for_insert(tree, offset, NULL, NULL); | |
344 | } | |
345 | ||
9ed74f2d JB |
346 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, |
347 | struct extent_state *other) | |
348 | { | |
349 | if (tree->ops && tree->ops->merge_extent_hook) | |
350 | tree->ops->merge_extent_hook(tree->mapping->host, new, | |
351 | other); | |
352 | } | |
353 | ||
d1310b2e CM |
354 | /* |
355 | * utility function to look for merge candidates inside a given range. | |
356 | * Any extents with matching state are merged together into a single | |
357 | * extent in the tree. Extents with EXTENT_IO in their state field | |
358 | * are not merged because the end_io handlers need to be able to do | |
359 | * operations on them without sleeping (or doing allocations/splits). | |
360 | * | |
361 | * This should be called with the tree lock held. | |
362 | */ | |
1bf85046 JM |
363 | static void merge_state(struct extent_io_tree *tree, |
364 | struct extent_state *state) | |
d1310b2e CM |
365 | { |
366 | struct extent_state *other; | |
367 | struct rb_node *other_node; | |
368 | ||
5b21f2ed | 369 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
1bf85046 | 370 | return; |
d1310b2e CM |
371 | |
372 | other_node = rb_prev(&state->rb_node); | |
373 | if (other_node) { | |
374 | other = rb_entry(other_node, struct extent_state, rb_node); | |
375 | if (other->end == state->start - 1 && | |
376 | other->state == state->state) { | |
9ed74f2d | 377 | merge_cb(tree, state, other); |
d1310b2e | 378 | state->start = other->start; |
d1310b2e | 379 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 380 | RB_CLEAR_NODE(&other->rb_node); |
d1310b2e CM |
381 | free_extent_state(other); |
382 | } | |
383 | } | |
384 | other_node = rb_next(&state->rb_node); | |
385 | if (other_node) { | |
386 | other = rb_entry(other_node, struct extent_state, rb_node); | |
387 | if (other->start == state->end + 1 && | |
388 | other->state == state->state) { | |
9ed74f2d | 389 | merge_cb(tree, state, other); |
df98b6e2 | 390 | state->end = other->end; |
df98b6e2 | 391 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 392 | RB_CLEAR_NODE(&other->rb_node); |
df98b6e2 | 393 | free_extent_state(other); |
d1310b2e CM |
394 | } |
395 | } | |
d1310b2e CM |
396 | } |
397 | ||
1bf85046 | 398 | static void set_state_cb(struct extent_io_tree *tree, |
9ee49a04 | 399 | struct extent_state *state, unsigned *bits) |
291d673e | 400 | { |
1bf85046 JM |
401 | if (tree->ops && tree->ops->set_bit_hook) |
402 | tree->ops->set_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
403 | } |
404 | ||
405 | static void clear_state_cb(struct extent_io_tree *tree, | |
9ee49a04 | 406 | struct extent_state *state, unsigned *bits) |
291d673e | 407 | { |
9ed74f2d JB |
408 | if (tree->ops && tree->ops->clear_bit_hook) |
409 | tree->ops->clear_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
410 | } |
411 | ||
3150b699 | 412 | static void set_state_bits(struct extent_io_tree *tree, |
9ee49a04 | 413 | struct extent_state *state, unsigned *bits); |
3150b699 | 414 | |
d1310b2e CM |
415 | /* |
416 | * insert an extent_state struct into the tree. 'bits' are set on the | |
417 | * struct before it is inserted. | |
418 | * | |
419 | * This may return -EEXIST if the extent is already there, in which case the | |
420 | * state struct is freed. | |
421 | * | |
422 | * The tree lock is not taken internally. This is a utility function and | |
423 | * probably isn't what you want to call (see set/clear_extent_bit). | |
424 | */ | |
425 | static int insert_state(struct extent_io_tree *tree, | |
426 | struct extent_state *state, u64 start, u64 end, | |
12cfbad9 FDBM |
427 | struct rb_node ***p, |
428 | struct rb_node **parent, | |
9ee49a04 | 429 | unsigned *bits) |
d1310b2e CM |
430 | { |
431 | struct rb_node *node; | |
432 | ||
31b1a2bd | 433 | if (end < start) |
efe120a0 | 434 | WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n", |
c1c9ff7c | 435 | end, start); |
d1310b2e CM |
436 | state->start = start; |
437 | state->end = end; | |
9ed74f2d | 438 | |
3150b699 XG |
439 | set_state_bits(tree, state, bits); |
440 | ||
f2071b21 | 441 | node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent); |
d1310b2e CM |
442 | if (node) { |
443 | struct extent_state *found; | |
444 | found = rb_entry(node, struct extent_state, rb_node); | |
efe120a0 | 445 | printk(KERN_ERR "BTRFS: found node %llu %llu on insert of " |
c1c9ff7c GU |
446 | "%llu %llu\n", |
447 | found->start, found->end, start, end); | |
d1310b2e CM |
448 | return -EEXIST; |
449 | } | |
450 | merge_state(tree, state); | |
451 | return 0; | |
452 | } | |
453 | ||
1bf85046 | 454 | static void split_cb(struct extent_io_tree *tree, struct extent_state *orig, |
9ed74f2d JB |
455 | u64 split) |
456 | { | |
457 | if (tree->ops && tree->ops->split_extent_hook) | |
1bf85046 | 458 | tree->ops->split_extent_hook(tree->mapping->host, orig, split); |
9ed74f2d JB |
459 | } |
460 | ||
d1310b2e CM |
461 | /* |
462 | * split a given extent state struct in two, inserting the preallocated | |
463 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
464 | * offset inside 'orig' where it should be split. | |
465 | * | |
466 | * Before calling, | |
467 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
468 | * are two extent state structs in the tree: | |
469 | * prealloc: [orig->start, split - 1] | |
470 | * orig: [ split, orig->end ] | |
471 | * | |
472 | * The tree locks are not taken by this function. They need to be held | |
473 | * by the caller. | |
474 | */ | |
475 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
476 | struct extent_state *prealloc, u64 split) | |
477 | { | |
478 | struct rb_node *node; | |
9ed74f2d JB |
479 | |
480 | split_cb(tree, orig, split); | |
481 | ||
d1310b2e CM |
482 | prealloc->start = orig->start; |
483 | prealloc->end = split - 1; | |
484 | prealloc->state = orig->state; | |
485 | orig->start = split; | |
486 | ||
f2071b21 FM |
487 | node = tree_insert(&tree->state, &orig->rb_node, prealloc->end, |
488 | &prealloc->rb_node, NULL, NULL); | |
d1310b2e | 489 | if (node) { |
d1310b2e CM |
490 | free_extent_state(prealloc); |
491 | return -EEXIST; | |
492 | } | |
493 | return 0; | |
494 | } | |
495 | ||
cdc6a395 LZ |
496 | static struct extent_state *next_state(struct extent_state *state) |
497 | { | |
498 | struct rb_node *next = rb_next(&state->rb_node); | |
499 | if (next) | |
500 | return rb_entry(next, struct extent_state, rb_node); | |
501 | else | |
502 | return NULL; | |
503 | } | |
504 | ||
d1310b2e CM |
505 | /* |
506 | * utility function to clear some bits in an extent state struct. | |
1b303fc0 | 507 | * it will optionally wake up any one waiting on this state (wake == 1). |
d1310b2e CM |
508 | * |
509 | * If no bits are set on the state struct after clearing things, the | |
510 | * struct is freed and removed from the tree | |
511 | */ | |
cdc6a395 LZ |
512 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
513 | struct extent_state *state, | |
9ee49a04 | 514 | unsigned *bits, int wake) |
d1310b2e | 515 | { |
cdc6a395 | 516 | struct extent_state *next; |
9ee49a04 | 517 | unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS; |
d1310b2e | 518 | |
0ca1f7ce | 519 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
520 | u64 range = state->end - state->start + 1; |
521 | WARN_ON(range > tree->dirty_bytes); | |
522 | tree->dirty_bytes -= range; | |
523 | } | |
291d673e | 524 | clear_state_cb(tree, state, bits); |
32c00aff | 525 | state->state &= ~bits_to_clear; |
d1310b2e CM |
526 | if (wake) |
527 | wake_up(&state->wq); | |
0ca1f7ce | 528 | if (state->state == 0) { |
cdc6a395 | 529 | next = next_state(state); |
27a3507d | 530 | if (extent_state_in_tree(state)) { |
d1310b2e | 531 | rb_erase(&state->rb_node, &tree->state); |
27a3507d | 532 | RB_CLEAR_NODE(&state->rb_node); |
d1310b2e CM |
533 | free_extent_state(state); |
534 | } else { | |
535 | WARN_ON(1); | |
536 | } | |
537 | } else { | |
538 | merge_state(tree, state); | |
cdc6a395 | 539 | next = next_state(state); |
d1310b2e | 540 | } |
cdc6a395 | 541 | return next; |
d1310b2e CM |
542 | } |
543 | ||
8233767a XG |
544 | static struct extent_state * |
545 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
546 | { | |
547 | if (!prealloc) | |
548 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
549 | ||
550 | return prealloc; | |
551 | } | |
552 | ||
48a3b636 | 553 | static void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
c2d904e0 JM |
554 | { |
555 | btrfs_panic(tree_fs_info(tree), err, "Locking error: " | |
556 | "Extent tree was modified by another " | |
557 | "thread while locked."); | |
558 | } | |
559 | ||
d1310b2e CM |
560 | /* |
561 | * clear some bits on a range in the tree. This may require splitting | |
562 | * or inserting elements in the tree, so the gfp mask is used to | |
563 | * indicate which allocations or sleeping are allowed. | |
564 | * | |
565 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
566 | * the given range from the tree regardless of state (ie for truncate). | |
567 | * | |
568 | * the range [start, end] is inclusive. | |
569 | * | |
6763af84 | 570 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e CM |
571 | */ |
572 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 573 | unsigned bits, int wake, int delete, |
2c64c53d CM |
574 | struct extent_state **cached_state, |
575 | gfp_t mask) | |
d1310b2e CM |
576 | { |
577 | struct extent_state *state; | |
2c64c53d | 578 | struct extent_state *cached; |
d1310b2e CM |
579 | struct extent_state *prealloc = NULL; |
580 | struct rb_node *node; | |
5c939df5 | 581 | u64 last_end; |
d1310b2e | 582 | int err; |
2ac55d41 | 583 | int clear = 0; |
d1310b2e | 584 | |
a5dee37d | 585 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 586 | |
7ee9e440 JB |
587 | if (bits & EXTENT_DELALLOC) |
588 | bits |= EXTENT_NORESERVE; | |
589 | ||
0ca1f7ce YZ |
590 | if (delete) |
591 | bits |= ~EXTENT_CTLBITS; | |
592 | bits |= EXTENT_FIRST_DELALLOC; | |
593 | ||
2ac55d41 JB |
594 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
595 | clear = 1; | |
d1310b2e CM |
596 | again: |
597 | if (!prealloc && (mask & __GFP_WAIT)) { | |
c7bc6319 FM |
598 | /* |
599 | * Don't care for allocation failure here because we might end | |
600 | * up not needing the pre-allocated extent state at all, which | |
601 | * is the case if we only have in the tree extent states that | |
602 | * cover our input range and don't cover too any other range. | |
603 | * If we end up needing a new extent state we allocate it later. | |
604 | */ | |
d1310b2e | 605 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
606 | } |
607 | ||
cad321ad | 608 | spin_lock(&tree->lock); |
2c64c53d CM |
609 | if (cached_state) { |
610 | cached = *cached_state; | |
2ac55d41 JB |
611 | |
612 | if (clear) { | |
613 | *cached_state = NULL; | |
614 | cached_state = NULL; | |
615 | } | |
616 | ||
27a3507d FM |
617 | if (cached && extent_state_in_tree(cached) && |
618 | cached->start <= start && cached->end > start) { | |
2ac55d41 JB |
619 | if (clear) |
620 | atomic_dec(&cached->refs); | |
2c64c53d | 621 | state = cached; |
42daec29 | 622 | goto hit_next; |
2c64c53d | 623 | } |
2ac55d41 JB |
624 | if (clear) |
625 | free_extent_state(cached); | |
2c64c53d | 626 | } |
d1310b2e CM |
627 | /* |
628 | * this search will find the extents that end after | |
629 | * our range starts | |
630 | */ | |
80ea96b1 | 631 | node = tree_search(tree, start); |
d1310b2e CM |
632 | if (!node) |
633 | goto out; | |
634 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 635 | hit_next: |
d1310b2e CM |
636 | if (state->start > end) |
637 | goto out; | |
638 | WARN_ON(state->end < start); | |
5c939df5 | 639 | last_end = state->end; |
d1310b2e | 640 | |
0449314a | 641 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
642 | if (!(state->state & bits)) { |
643 | state = next_state(state); | |
0449314a | 644 | goto next; |
cdc6a395 | 645 | } |
0449314a | 646 | |
d1310b2e CM |
647 | /* |
648 | * | ---- desired range ---- | | |
649 | * | state | or | |
650 | * | ------------- state -------------- | | |
651 | * | |
652 | * We need to split the extent we found, and may flip | |
653 | * bits on second half. | |
654 | * | |
655 | * If the extent we found extends past our range, we | |
656 | * just split and search again. It'll get split again | |
657 | * the next time though. | |
658 | * | |
659 | * If the extent we found is inside our range, we clear | |
660 | * the desired bit on it. | |
661 | */ | |
662 | ||
663 | if (state->start < start) { | |
8233767a XG |
664 | prealloc = alloc_extent_state_atomic(prealloc); |
665 | BUG_ON(!prealloc); | |
d1310b2e | 666 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
667 | if (err) |
668 | extent_io_tree_panic(tree, err); | |
669 | ||
d1310b2e CM |
670 | prealloc = NULL; |
671 | if (err) | |
672 | goto out; | |
673 | if (state->end <= end) { | |
d1ac6e41 LB |
674 | state = clear_state_bit(tree, state, &bits, wake); |
675 | goto next; | |
d1310b2e CM |
676 | } |
677 | goto search_again; | |
678 | } | |
679 | /* | |
680 | * | ---- desired range ---- | | |
681 | * | state | | |
682 | * We need to split the extent, and clear the bit | |
683 | * on the first half | |
684 | */ | |
685 | if (state->start <= end && state->end > end) { | |
8233767a XG |
686 | prealloc = alloc_extent_state_atomic(prealloc); |
687 | BUG_ON(!prealloc); | |
d1310b2e | 688 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
689 | if (err) |
690 | extent_io_tree_panic(tree, err); | |
691 | ||
d1310b2e CM |
692 | if (wake) |
693 | wake_up(&state->wq); | |
42daec29 | 694 | |
6763af84 | 695 | clear_state_bit(tree, prealloc, &bits, wake); |
9ed74f2d | 696 | |
d1310b2e CM |
697 | prealloc = NULL; |
698 | goto out; | |
699 | } | |
42daec29 | 700 | |
cdc6a395 | 701 | state = clear_state_bit(tree, state, &bits, wake); |
0449314a | 702 | next: |
5c939df5 YZ |
703 | if (last_end == (u64)-1) |
704 | goto out; | |
705 | start = last_end + 1; | |
cdc6a395 | 706 | if (start <= end && state && !need_resched()) |
692e5759 | 707 | goto hit_next; |
d1310b2e CM |
708 | goto search_again; |
709 | ||
710 | out: | |
cad321ad | 711 | spin_unlock(&tree->lock); |
d1310b2e CM |
712 | if (prealloc) |
713 | free_extent_state(prealloc); | |
714 | ||
6763af84 | 715 | return 0; |
d1310b2e CM |
716 | |
717 | search_again: | |
718 | if (start > end) | |
719 | goto out; | |
cad321ad | 720 | spin_unlock(&tree->lock); |
d1310b2e CM |
721 | if (mask & __GFP_WAIT) |
722 | cond_resched(); | |
723 | goto again; | |
724 | } | |
d1310b2e | 725 | |
143bede5 JM |
726 | static void wait_on_state(struct extent_io_tree *tree, |
727 | struct extent_state *state) | |
641f5219 CH |
728 | __releases(tree->lock) |
729 | __acquires(tree->lock) | |
d1310b2e CM |
730 | { |
731 | DEFINE_WAIT(wait); | |
732 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 733 | spin_unlock(&tree->lock); |
d1310b2e | 734 | schedule(); |
cad321ad | 735 | spin_lock(&tree->lock); |
d1310b2e | 736 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
737 | } |
738 | ||
739 | /* | |
740 | * waits for one or more bits to clear on a range in the state tree. | |
741 | * The range [start, end] is inclusive. | |
742 | * The tree lock is taken by this function | |
743 | */ | |
41074888 DS |
744 | static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
745 | unsigned long bits) | |
d1310b2e CM |
746 | { |
747 | struct extent_state *state; | |
748 | struct rb_node *node; | |
749 | ||
a5dee37d | 750 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 751 | |
cad321ad | 752 | spin_lock(&tree->lock); |
d1310b2e CM |
753 | again: |
754 | while (1) { | |
755 | /* | |
756 | * this search will find all the extents that end after | |
757 | * our range starts | |
758 | */ | |
80ea96b1 | 759 | node = tree_search(tree, start); |
c50d3e71 | 760 | process_node: |
d1310b2e CM |
761 | if (!node) |
762 | break; | |
763 | ||
764 | state = rb_entry(node, struct extent_state, rb_node); | |
765 | ||
766 | if (state->start > end) | |
767 | goto out; | |
768 | ||
769 | if (state->state & bits) { | |
770 | start = state->start; | |
771 | atomic_inc(&state->refs); | |
772 | wait_on_state(tree, state); | |
773 | free_extent_state(state); | |
774 | goto again; | |
775 | } | |
776 | start = state->end + 1; | |
777 | ||
778 | if (start > end) | |
779 | break; | |
780 | ||
c50d3e71 FM |
781 | if (!cond_resched_lock(&tree->lock)) { |
782 | node = rb_next(node); | |
783 | goto process_node; | |
784 | } | |
d1310b2e CM |
785 | } |
786 | out: | |
cad321ad | 787 | spin_unlock(&tree->lock); |
d1310b2e | 788 | } |
d1310b2e | 789 | |
1bf85046 | 790 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 791 | struct extent_state *state, |
9ee49a04 | 792 | unsigned *bits) |
d1310b2e | 793 | { |
9ee49a04 | 794 | unsigned bits_to_set = *bits & ~EXTENT_CTLBITS; |
9ed74f2d | 795 | |
1bf85046 | 796 | set_state_cb(tree, state, bits); |
0ca1f7ce | 797 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
798 | u64 range = state->end - state->start + 1; |
799 | tree->dirty_bytes += range; | |
800 | } | |
0ca1f7ce | 801 | state->state |= bits_to_set; |
d1310b2e CM |
802 | } |
803 | ||
e38e2ed7 FM |
804 | static void cache_state_if_flags(struct extent_state *state, |
805 | struct extent_state **cached_ptr, | |
9ee49a04 | 806 | unsigned flags) |
2c64c53d CM |
807 | { |
808 | if (cached_ptr && !(*cached_ptr)) { | |
e38e2ed7 | 809 | if (!flags || (state->state & flags)) { |
2c64c53d CM |
810 | *cached_ptr = state; |
811 | atomic_inc(&state->refs); | |
812 | } | |
813 | } | |
814 | } | |
815 | ||
e38e2ed7 FM |
816 | static void cache_state(struct extent_state *state, |
817 | struct extent_state **cached_ptr) | |
818 | { | |
819 | return cache_state_if_flags(state, cached_ptr, | |
820 | EXTENT_IOBITS | EXTENT_BOUNDARY); | |
821 | } | |
822 | ||
d1310b2e | 823 | /* |
1edbb734 CM |
824 | * set some bits on a range in the tree. This may require allocations or |
825 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 826 | * |
1edbb734 CM |
827 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
828 | * part of the range already has the desired bits set. The start of the | |
829 | * existing range is returned in failed_start in this case. | |
d1310b2e | 830 | * |
1edbb734 | 831 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 832 | */ |
1edbb734 | 833 | |
3fbe5c02 JM |
834 | static int __must_check |
835 | __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 836 | unsigned bits, unsigned exclusive_bits, |
41074888 DS |
837 | u64 *failed_start, struct extent_state **cached_state, |
838 | gfp_t mask) | |
d1310b2e CM |
839 | { |
840 | struct extent_state *state; | |
841 | struct extent_state *prealloc = NULL; | |
842 | struct rb_node *node; | |
12cfbad9 FDBM |
843 | struct rb_node **p; |
844 | struct rb_node *parent; | |
d1310b2e | 845 | int err = 0; |
d1310b2e CM |
846 | u64 last_start; |
847 | u64 last_end; | |
42daec29 | 848 | |
a5dee37d | 849 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 850 | |
0ca1f7ce | 851 | bits |= EXTENT_FIRST_DELALLOC; |
d1310b2e CM |
852 | again: |
853 | if (!prealloc && (mask & __GFP_WAIT)) { | |
854 | prealloc = alloc_extent_state(mask); | |
8233767a | 855 | BUG_ON(!prealloc); |
d1310b2e CM |
856 | } |
857 | ||
cad321ad | 858 | spin_lock(&tree->lock); |
9655d298 CM |
859 | if (cached_state && *cached_state) { |
860 | state = *cached_state; | |
df98b6e2 | 861 | if (state->start <= start && state->end > start && |
27a3507d | 862 | extent_state_in_tree(state)) { |
9655d298 CM |
863 | node = &state->rb_node; |
864 | goto hit_next; | |
865 | } | |
866 | } | |
d1310b2e CM |
867 | /* |
868 | * this search will find all the extents that end after | |
869 | * our range starts. | |
870 | */ | |
12cfbad9 | 871 | node = tree_search_for_insert(tree, start, &p, &parent); |
d1310b2e | 872 | if (!node) { |
8233767a XG |
873 | prealloc = alloc_extent_state_atomic(prealloc); |
874 | BUG_ON(!prealloc); | |
12cfbad9 FDBM |
875 | err = insert_state(tree, prealloc, start, end, |
876 | &p, &parent, &bits); | |
c2d904e0 JM |
877 | if (err) |
878 | extent_io_tree_panic(tree, err); | |
879 | ||
c42ac0bc | 880 | cache_state(prealloc, cached_state); |
d1310b2e | 881 | prealloc = NULL; |
d1310b2e CM |
882 | goto out; |
883 | } | |
d1310b2e | 884 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 885 | hit_next: |
d1310b2e CM |
886 | last_start = state->start; |
887 | last_end = state->end; | |
888 | ||
889 | /* | |
890 | * | ---- desired range ---- | | |
891 | * | state | | |
892 | * | |
893 | * Just lock what we found and keep going | |
894 | */ | |
895 | if (state->start == start && state->end <= end) { | |
1edbb734 | 896 | if (state->state & exclusive_bits) { |
d1310b2e CM |
897 | *failed_start = state->start; |
898 | err = -EEXIST; | |
899 | goto out; | |
900 | } | |
42daec29 | 901 | |
1bf85046 | 902 | set_state_bits(tree, state, &bits); |
2c64c53d | 903 | cache_state(state, cached_state); |
d1310b2e | 904 | merge_state(tree, state); |
5c939df5 YZ |
905 | if (last_end == (u64)-1) |
906 | goto out; | |
907 | start = last_end + 1; | |
d1ac6e41 LB |
908 | state = next_state(state); |
909 | if (start < end && state && state->start == start && | |
910 | !need_resched()) | |
911 | goto hit_next; | |
d1310b2e CM |
912 | goto search_again; |
913 | } | |
914 | ||
915 | /* | |
916 | * | ---- desired range ---- | | |
917 | * | state | | |
918 | * or | |
919 | * | ------------- state -------------- | | |
920 | * | |
921 | * We need to split the extent we found, and may flip bits on | |
922 | * second half. | |
923 | * | |
924 | * If the extent we found extends past our | |
925 | * range, we just split and search again. It'll get split | |
926 | * again the next time though. | |
927 | * | |
928 | * If the extent we found is inside our range, we set the | |
929 | * desired bit on it. | |
930 | */ | |
931 | if (state->start < start) { | |
1edbb734 | 932 | if (state->state & exclusive_bits) { |
d1310b2e CM |
933 | *failed_start = start; |
934 | err = -EEXIST; | |
935 | goto out; | |
936 | } | |
8233767a XG |
937 | |
938 | prealloc = alloc_extent_state_atomic(prealloc); | |
939 | BUG_ON(!prealloc); | |
d1310b2e | 940 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
941 | if (err) |
942 | extent_io_tree_panic(tree, err); | |
943 | ||
d1310b2e CM |
944 | prealloc = NULL; |
945 | if (err) | |
946 | goto out; | |
947 | if (state->end <= end) { | |
1bf85046 | 948 | set_state_bits(tree, state, &bits); |
2c64c53d | 949 | cache_state(state, cached_state); |
d1310b2e | 950 | merge_state(tree, state); |
5c939df5 YZ |
951 | if (last_end == (u64)-1) |
952 | goto out; | |
953 | start = last_end + 1; | |
d1ac6e41 LB |
954 | state = next_state(state); |
955 | if (start < end && state && state->start == start && | |
956 | !need_resched()) | |
957 | goto hit_next; | |
d1310b2e CM |
958 | } |
959 | goto search_again; | |
960 | } | |
961 | /* | |
962 | * | ---- desired range ---- | | |
963 | * | state | or | state | | |
964 | * | |
965 | * There's a hole, we need to insert something in it and | |
966 | * ignore the extent we found. | |
967 | */ | |
968 | if (state->start > start) { | |
969 | u64 this_end; | |
970 | if (end < last_start) | |
971 | this_end = end; | |
972 | else | |
d397712b | 973 | this_end = last_start - 1; |
8233767a XG |
974 | |
975 | prealloc = alloc_extent_state_atomic(prealloc); | |
976 | BUG_ON(!prealloc); | |
c7f895a2 XG |
977 | |
978 | /* | |
979 | * Avoid to free 'prealloc' if it can be merged with | |
980 | * the later extent. | |
981 | */ | |
d1310b2e | 982 | err = insert_state(tree, prealloc, start, this_end, |
12cfbad9 | 983 | NULL, NULL, &bits); |
c2d904e0 JM |
984 | if (err) |
985 | extent_io_tree_panic(tree, err); | |
986 | ||
9ed74f2d JB |
987 | cache_state(prealloc, cached_state); |
988 | prealloc = NULL; | |
d1310b2e CM |
989 | start = this_end + 1; |
990 | goto search_again; | |
991 | } | |
992 | /* | |
993 | * | ---- desired range ---- | | |
994 | * | state | | |
995 | * We need to split the extent, and set the bit | |
996 | * on the first half | |
997 | */ | |
998 | if (state->start <= end && state->end > end) { | |
1edbb734 | 999 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1000 | *failed_start = start; |
1001 | err = -EEXIST; | |
1002 | goto out; | |
1003 | } | |
8233767a XG |
1004 | |
1005 | prealloc = alloc_extent_state_atomic(prealloc); | |
1006 | BUG_ON(!prealloc); | |
d1310b2e | 1007 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
1008 | if (err) |
1009 | extent_io_tree_panic(tree, err); | |
d1310b2e | 1010 | |
1bf85046 | 1011 | set_state_bits(tree, prealloc, &bits); |
2c64c53d | 1012 | cache_state(prealloc, cached_state); |
d1310b2e CM |
1013 | merge_state(tree, prealloc); |
1014 | prealloc = NULL; | |
1015 | goto out; | |
1016 | } | |
1017 | ||
1018 | goto search_again; | |
1019 | ||
1020 | out: | |
cad321ad | 1021 | spin_unlock(&tree->lock); |
d1310b2e CM |
1022 | if (prealloc) |
1023 | free_extent_state(prealloc); | |
1024 | ||
1025 | return err; | |
1026 | ||
1027 | search_again: | |
1028 | if (start > end) | |
1029 | goto out; | |
cad321ad | 1030 | spin_unlock(&tree->lock); |
d1310b2e CM |
1031 | if (mask & __GFP_WAIT) |
1032 | cond_resched(); | |
1033 | goto again; | |
1034 | } | |
d1310b2e | 1035 | |
41074888 | 1036 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
9ee49a04 | 1037 | unsigned bits, u64 * failed_start, |
41074888 | 1038 | struct extent_state **cached_state, gfp_t mask) |
3fbe5c02 JM |
1039 | { |
1040 | return __set_extent_bit(tree, start, end, bits, 0, failed_start, | |
1041 | cached_state, mask); | |
1042 | } | |
1043 | ||
1044 | ||
462d6fac | 1045 | /** |
10983f2e LB |
1046 | * convert_extent_bit - convert all bits in a given range from one bit to |
1047 | * another | |
462d6fac JB |
1048 | * @tree: the io tree to search |
1049 | * @start: the start offset in bytes | |
1050 | * @end: the end offset in bytes (inclusive) | |
1051 | * @bits: the bits to set in this range | |
1052 | * @clear_bits: the bits to clear in this range | |
e6138876 | 1053 | * @cached_state: state that we're going to cache |
462d6fac JB |
1054 | * @mask: the allocation mask |
1055 | * | |
1056 | * This will go through and set bits for the given range. If any states exist | |
1057 | * already in this range they are set with the given bit and cleared of the | |
1058 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
1059 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
1060 | * boundary bits like LOCK. | |
1061 | */ | |
1062 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1063 | unsigned bits, unsigned clear_bits, |
e6138876 | 1064 | struct extent_state **cached_state, gfp_t mask) |
462d6fac JB |
1065 | { |
1066 | struct extent_state *state; | |
1067 | struct extent_state *prealloc = NULL; | |
1068 | struct rb_node *node; | |
12cfbad9 FDBM |
1069 | struct rb_node **p; |
1070 | struct rb_node *parent; | |
462d6fac JB |
1071 | int err = 0; |
1072 | u64 last_start; | |
1073 | u64 last_end; | |
c8fd3de7 | 1074 | bool first_iteration = true; |
462d6fac | 1075 | |
a5dee37d | 1076 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 1077 | |
462d6fac JB |
1078 | again: |
1079 | if (!prealloc && (mask & __GFP_WAIT)) { | |
c8fd3de7 FM |
1080 | /* |
1081 | * Best effort, don't worry if extent state allocation fails | |
1082 | * here for the first iteration. We might have a cached state | |
1083 | * that matches exactly the target range, in which case no | |
1084 | * extent state allocations are needed. We'll only know this | |
1085 | * after locking the tree. | |
1086 | */ | |
462d6fac | 1087 | prealloc = alloc_extent_state(mask); |
c8fd3de7 | 1088 | if (!prealloc && !first_iteration) |
462d6fac JB |
1089 | return -ENOMEM; |
1090 | } | |
1091 | ||
1092 | spin_lock(&tree->lock); | |
e6138876 JB |
1093 | if (cached_state && *cached_state) { |
1094 | state = *cached_state; | |
1095 | if (state->start <= start && state->end > start && | |
27a3507d | 1096 | extent_state_in_tree(state)) { |
e6138876 JB |
1097 | node = &state->rb_node; |
1098 | goto hit_next; | |
1099 | } | |
1100 | } | |
1101 | ||
462d6fac JB |
1102 | /* |
1103 | * this search will find all the extents that end after | |
1104 | * our range starts. | |
1105 | */ | |
12cfbad9 | 1106 | node = tree_search_for_insert(tree, start, &p, &parent); |
462d6fac JB |
1107 | if (!node) { |
1108 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1109 | if (!prealloc) { |
1110 | err = -ENOMEM; | |
1111 | goto out; | |
1112 | } | |
12cfbad9 FDBM |
1113 | err = insert_state(tree, prealloc, start, end, |
1114 | &p, &parent, &bits); | |
c2d904e0 JM |
1115 | if (err) |
1116 | extent_io_tree_panic(tree, err); | |
c42ac0bc FDBM |
1117 | cache_state(prealloc, cached_state); |
1118 | prealloc = NULL; | |
462d6fac JB |
1119 | goto out; |
1120 | } | |
1121 | state = rb_entry(node, struct extent_state, rb_node); | |
1122 | hit_next: | |
1123 | last_start = state->start; | |
1124 | last_end = state->end; | |
1125 | ||
1126 | /* | |
1127 | * | ---- desired range ---- | | |
1128 | * | state | | |
1129 | * | |
1130 | * Just lock what we found and keep going | |
1131 | */ | |
1132 | if (state->start == start && state->end <= end) { | |
462d6fac | 1133 | set_state_bits(tree, state, &bits); |
e6138876 | 1134 | cache_state(state, cached_state); |
d1ac6e41 | 1135 | state = clear_state_bit(tree, state, &clear_bits, 0); |
462d6fac JB |
1136 | if (last_end == (u64)-1) |
1137 | goto out; | |
462d6fac | 1138 | start = last_end + 1; |
d1ac6e41 LB |
1139 | if (start < end && state && state->start == start && |
1140 | !need_resched()) | |
1141 | goto hit_next; | |
462d6fac JB |
1142 | goto search_again; |
1143 | } | |
1144 | ||
1145 | /* | |
1146 | * | ---- desired range ---- | | |
1147 | * | state | | |
1148 | * or | |
1149 | * | ------------- state -------------- | | |
1150 | * | |
1151 | * We need to split the extent we found, and may flip bits on | |
1152 | * second half. | |
1153 | * | |
1154 | * If the extent we found extends past our | |
1155 | * range, we just split and search again. It'll get split | |
1156 | * again the next time though. | |
1157 | * | |
1158 | * If the extent we found is inside our range, we set the | |
1159 | * desired bit on it. | |
1160 | */ | |
1161 | if (state->start < start) { | |
1162 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1163 | if (!prealloc) { |
1164 | err = -ENOMEM; | |
1165 | goto out; | |
1166 | } | |
462d6fac | 1167 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1168 | if (err) |
1169 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1170 | prealloc = NULL; |
1171 | if (err) | |
1172 | goto out; | |
1173 | if (state->end <= end) { | |
1174 | set_state_bits(tree, state, &bits); | |
e6138876 | 1175 | cache_state(state, cached_state); |
d1ac6e41 | 1176 | state = clear_state_bit(tree, state, &clear_bits, 0); |
462d6fac JB |
1177 | if (last_end == (u64)-1) |
1178 | goto out; | |
1179 | start = last_end + 1; | |
d1ac6e41 LB |
1180 | if (start < end && state && state->start == start && |
1181 | !need_resched()) | |
1182 | goto hit_next; | |
462d6fac JB |
1183 | } |
1184 | goto search_again; | |
1185 | } | |
1186 | /* | |
1187 | * | ---- desired range ---- | | |
1188 | * | state | or | state | | |
1189 | * | |
1190 | * There's a hole, we need to insert something in it and | |
1191 | * ignore the extent we found. | |
1192 | */ | |
1193 | if (state->start > start) { | |
1194 | u64 this_end; | |
1195 | if (end < last_start) | |
1196 | this_end = end; | |
1197 | else | |
1198 | this_end = last_start - 1; | |
1199 | ||
1200 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1201 | if (!prealloc) { |
1202 | err = -ENOMEM; | |
1203 | goto out; | |
1204 | } | |
462d6fac JB |
1205 | |
1206 | /* | |
1207 | * Avoid to free 'prealloc' if it can be merged with | |
1208 | * the later extent. | |
1209 | */ | |
1210 | err = insert_state(tree, prealloc, start, this_end, | |
12cfbad9 | 1211 | NULL, NULL, &bits); |
c2d904e0 JM |
1212 | if (err) |
1213 | extent_io_tree_panic(tree, err); | |
e6138876 | 1214 | cache_state(prealloc, cached_state); |
462d6fac JB |
1215 | prealloc = NULL; |
1216 | start = this_end + 1; | |
1217 | goto search_again; | |
1218 | } | |
1219 | /* | |
1220 | * | ---- desired range ---- | | |
1221 | * | state | | |
1222 | * We need to split the extent, and set the bit | |
1223 | * on the first half | |
1224 | */ | |
1225 | if (state->start <= end && state->end > end) { | |
1226 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1227 | if (!prealloc) { |
1228 | err = -ENOMEM; | |
1229 | goto out; | |
1230 | } | |
462d6fac JB |
1231 | |
1232 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1233 | if (err) |
1234 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1235 | |
1236 | set_state_bits(tree, prealloc, &bits); | |
e6138876 | 1237 | cache_state(prealloc, cached_state); |
462d6fac | 1238 | clear_state_bit(tree, prealloc, &clear_bits, 0); |
462d6fac JB |
1239 | prealloc = NULL; |
1240 | goto out; | |
1241 | } | |
1242 | ||
1243 | goto search_again; | |
1244 | ||
1245 | out: | |
1246 | spin_unlock(&tree->lock); | |
1247 | if (prealloc) | |
1248 | free_extent_state(prealloc); | |
1249 | ||
1250 | return err; | |
1251 | ||
1252 | search_again: | |
1253 | if (start > end) | |
1254 | goto out; | |
1255 | spin_unlock(&tree->lock); | |
1256 | if (mask & __GFP_WAIT) | |
1257 | cond_resched(); | |
c8fd3de7 | 1258 | first_iteration = false; |
462d6fac JB |
1259 | goto again; |
1260 | } | |
1261 | ||
d1310b2e CM |
1262 | /* wrappers around set/clear extent bit */ |
1263 | int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
1264 | gfp_t mask) | |
1265 | { | |
3fbe5c02 | 1266 | return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL, |
2c64c53d | 1267 | NULL, mask); |
d1310b2e | 1268 | } |
d1310b2e CM |
1269 | |
1270 | int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1271 | unsigned bits, gfp_t mask) |
d1310b2e | 1272 | { |
3fbe5c02 | 1273 | return set_extent_bit(tree, start, end, bits, NULL, |
2c64c53d | 1274 | NULL, mask); |
d1310b2e | 1275 | } |
d1310b2e CM |
1276 | |
1277 | int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1278 | unsigned bits, gfp_t mask) |
d1310b2e | 1279 | { |
0f31871f FM |
1280 | int wake = 0; |
1281 | ||
1282 | if (bits & EXTENT_LOCKED) | |
1283 | wake = 1; | |
1284 | ||
1285 | return clear_extent_bit(tree, start, end, bits, wake, 0, NULL, mask); | |
d1310b2e | 1286 | } |
d1310b2e CM |
1287 | |
1288 | int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, | |
2ac55d41 | 1289 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
1290 | { |
1291 | return set_extent_bit(tree, start, end, | |
fee187d9 | 1292 | EXTENT_DELALLOC | EXTENT_UPTODATE, |
3fbe5c02 | 1293 | NULL, cached_state, mask); |
d1310b2e | 1294 | } |
d1310b2e | 1295 | |
9e8a4a8b LB |
1296 | int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end, |
1297 | struct extent_state **cached_state, gfp_t mask) | |
1298 | { | |
1299 | return set_extent_bit(tree, start, end, | |
1300 | EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG, | |
1301 | NULL, cached_state, mask); | |
1302 | } | |
1303 | ||
d1310b2e CM |
1304 | int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, |
1305 | gfp_t mask) | |
1306 | { | |
1307 | return clear_extent_bit(tree, start, end, | |
32c00aff | 1308 | EXTENT_DIRTY | EXTENT_DELALLOC | |
0ca1f7ce | 1309 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask); |
d1310b2e | 1310 | } |
d1310b2e CM |
1311 | |
1312 | int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, | |
1313 | gfp_t mask) | |
1314 | { | |
3fbe5c02 | 1315 | return set_extent_bit(tree, start, end, EXTENT_NEW, NULL, |
2c64c53d | 1316 | NULL, mask); |
d1310b2e | 1317 | } |
d1310b2e | 1318 | |
d1310b2e | 1319 | int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
507903b8 | 1320 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e | 1321 | { |
6b67a320 | 1322 | return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL, |
3fbe5c02 | 1323 | cached_state, mask); |
d1310b2e | 1324 | } |
d1310b2e | 1325 | |
5fd02043 JB |
1326 | int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
1327 | struct extent_state **cached_state, gfp_t mask) | |
d1310b2e | 1328 | { |
2c64c53d | 1329 | return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, |
2ac55d41 | 1330 | cached_state, mask); |
d1310b2e | 1331 | } |
d1310b2e | 1332 | |
d352ac68 CM |
1333 | /* |
1334 | * either insert or lock state struct between start and end use mask to tell | |
1335 | * us if waiting is desired. | |
1336 | */ | |
1edbb734 | 1337 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
9ee49a04 | 1338 | unsigned bits, struct extent_state **cached_state) |
d1310b2e CM |
1339 | { |
1340 | int err; | |
1341 | u64 failed_start; | |
9ee49a04 | 1342 | |
d1310b2e | 1343 | while (1) { |
3fbe5c02 JM |
1344 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits, |
1345 | EXTENT_LOCKED, &failed_start, | |
1346 | cached_state, GFP_NOFS); | |
d0082371 | 1347 | if (err == -EEXIST) { |
d1310b2e CM |
1348 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1349 | start = failed_start; | |
d0082371 | 1350 | } else |
d1310b2e | 1351 | break; |
d1310b2e CM |
1352 | WARN_ON(start > end); |
1353 | } | |
1354 | return err; | |
1355 | } | |
d1310b2e | 1356 | |
d0082371 | 1357 | int lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
1edbb734 | 1358 | { |
d0082371 | 1359 | return lock_extent_bits(tree, start, end, 0, NULL); |
1edbb734 CM |
1360 | } |
1361 | ||
d0082371 | 1362 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1363 | { |
1364 | int err; | |
1365 | u64 failed_start; | |
1366 | ||
3fbe5c02 JM |
1367 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
1368 | &failed_start, NULL, GFP_NOFS); | |
6643558d YZ |
1369 | if (err == -EEXIST) { |
1370 | if (failed_start > start) | |
1371 | clear_extent_bit(tree, start, failed_start - 1, | |
d0082371 | 1372 | EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS); |
25179201 | 1373 | return 0; |
6643558d | 1374 | } |
25179201 JB |
1375 | return 1; |
1376 | } | |
25179201 | 1377 | |
2c64c53d CM |
1378 | int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end, |
1379 | struct extent_state **cached, gfp_t mask) | |
1380 | { | |
1381 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, | |
1382 | mask); | |
1383 | } | |
1384 | ||
d0082371 | 1385 | int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e | 1386 | { |
2c64c53d | 1387 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL, |
d0082371 | 1388 | GFP_NOFS); |
d1310b2e | 1389 | } |
d1310b2e | 1390 | |
4adaa611 CM |
1391 | int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
1392 | { | |
1393 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1394 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1395 | struct page *page; | |
1396 | ||
1397 | while (index <= end_index) { | |
1398 | page = find_get_page(inode->i_mapping, index); | |
1399 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1400 | clear_page_dirty_for_io(page); | |
1401 | page_cache_release(page); | |
1402 | index++; | |
1403 | } | |
1404 | return 0; | |
1405 | } | |
1406 | ||
1407 | int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) | |
1408 | { | |
1409 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1410 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1411 | struct page *page; | |
1412 | ||
1413 | while (index <= end_index) { | |
1414 | page = find_get_page(inode->i_mapping, index); | |
1415 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
4adaa611 | 1416 | __set_page_dirty_nobuffers(page); |
8d38633c | 1417 | account_page_redirty(page); |
4adaa611 CM |
1418 | page_cache_release(page); |
1419 | index++; | |
1420 | } | |
1421 | return 0; | |
1422 | } | |
1423 | ||
d1310b2e CM |
1424 | /* |
1425 | * helper function to set both pages and extents in the tree writeback | |
1426 | */ | |
b2950863 | 1427 | static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e CM |
1428 | { |
1429 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1430 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1431 | struct page *page; | |
1432 | ||
1433 | while (index <= end_index) { | |
1434 | page = find_get_page(tree->mapping, index); | |
79787eaa | 1435 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ |
d1310b2e CM |
1436 | set_page_writeback(page); |
1437 | page_cache_release(page); | |
1438 | index++; | |
1439 | } | |
d1310b2e CM |
1440 | return 0; |
1441 | } | |
d1310b2e | 1442 | |
d352ac68 CM |
1443 | /* find the first state struct with 'bits' set after 'start', and |
1444 | * return it. tree->lock must be held. NULL will returned if | |
1445 | * nothing was found after 'start' | |
1446 | */ | |
48a3b636 ES |
1447 | static struct extent_state * |
1448 | find_first_extent_bit_state(struct extent_io_tree *tree, | |
9ee49a04 | 1449 | u64 start, unsigned bits) |
d7fc640e CM |
1450 | { |
1451 | struct rb_node *node; | |
1452 | struct extent_state *state; | |
1453 | ||
1454 | /* | |
1455 | * this search will find all the extents that end after | |
1456 | * our range starts. | |
1457 | */ | |
1458 | node = tree_search(tree, start); | |
d397712b | 1459 | if (!node) |
d7fc640e | 1460 | goto out; |
d7fc640e | 1461 | |
d397712b | 1462 | while (1) { |
d7fc640e | 1463 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1464 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1465 | return state; |
d397712b | 1466 | |
d7fc640e CM |
1467 | node = rb_next(node); |
1468 | if (!node) | |
1469 | break; | |
1470 | } | |
1471 | out: | |
1472 | return NULL; | |
1473 | } | |
d7fc640e | 1474 | |
69261c4b XG |
1475 | /* |
1476 | * find the first offset in the io tree with 'bits' set. zero is | |
1477 | * returned if we find something, and *start_ret and *end_ret are | |
1478 | * set to reflect the state struct that was found. | |
1479 | * | |
477d7eaf | 1480 | * If nothing was found, 1 is returned. If found something, return 0. |
69261c4b XG |
1481 | */ |
1482 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
9ee49a04 | 1483 | u64 *start_ret, u64 *end_ret, unsigned bits, |
e6138876 | 1484 | struct extent_state **cached_state) |
69261c4b XG |
1485 | { |
1486 | struct extent_state *state; | |
e6138876 | 1487 | struct rb_node *n; |
69261c4b XG |
1488 | int ret = 1; |
1489 | ||
1490 | spin_lock(&tree->lock); | |
e6138876 JB |
1491 | if (cached_state && *cached_state) { |
1492 | state = *cached_state; | |
27a3507d | 1493 | if (state->end == start - 1 && extent_state_in_tree(state)) { |
e6138876 JB |
1494 | n = rb_next(&state->rb_node); |
1495 | while (n) { | |
1496 | state = rb_entry(n, struct extent_state, | |
1497 | rb_node); | |
1498 | if (state->state & bits) | |
1499 | goto got_it; | |
1500 | n = rb_next(n); | |
1501 | } | |
1502 | free_extent_state(*cached_state); | |
1503 | *cached_state = NULL; | |
1504 | goto out; | |
1505 | } | |
1506 | free_extent_state(*cached_state); | |
1507 | *cached_state = NULL; | |
1508 | } | |
1509 | ||
69261c4b | 1510 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1511 | got_it: |
69261c4b | 1512 | if (state) { |
e38e2ed7 | 1513 | cache_state_if_flags(state, cached_state, 0); |
69261c4b XG |
1514 | *start_ret = state->start; |
1515 | *end_ret = state->end; | |
1516 | ret = 0; | |
1517 | } | |
e6138876 | 1518 | out: |
69261c4b XG |
1519 | spin_unlock(&tree->lock); |
1520 | return ret; | |
1521 | } | |
1522 | ||
d352ac68 CM |
1523 | /* |
1524 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1525 | * more than 'max_bytes'. start and end are used to return the range, | |
1526 | * | |
1527 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1528 | */ | |
c8b97818 | 1529 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1530 | u64 *start, u64 *end, u64 max_bytes, |
1531 | struct extent_state **cached_state) | |
d1310b2e CM |
1532 | { |
1533 | struct rb_node *node; | |
1534 | struct extent_state *state; | |
1535 | u64 cur_start = *start; | |
1536 | u64 found = 0; | |
1537 | u64 total_bytes = 0; | |
1538 | ||
cad321ad | 1539 | spin_lock(&tree->lock); |
c8b97818 | 1540 | |
d1310b2e CM |
1541 | /* |
1542 | * this search will find all the extents that end after | |
1543 | * our range starts. | |
1544 | */ | |
80ea96b1 | 1545 | node = tree_search(tree, cur_start); |
2b114d1d | 1546 | if (!node) { |
3b951516 CM |
1547 | if (!found) |
1548 | *end = (u64)-1; | |
d1310b2e CM |
1549 | goto out; |
1550 | } | |
1551 | ||
d397712b | 1552 | while (1) { |
d1310b2e | 1553 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1554 | if (found && (state->start != cur_start || |
1555 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1556 | goto out; |
1557 | } | |
1558 | if (!(state->state & EXTENT_DELALLOC)) { | |
1559 | if (!found) | |
1560 | *end = state->end; | |
1561 | goto out; | |
1562 | } | |
c2a128d2 | 1563 | if (!found) { |
d1310b2e | 1564 | *start = state->start; |
c2a128d2 JB |
1565 | *cached_state = state; |
1566 | atomic_inc(&state->refs); | |
1567 | } | |
d1310b2e CM |
1568 | found++; |
1569 | *end = state->end; | |
1570 | cur_start = state->end + 1; | |
1571 | node = rb_next(node); | |
d1310b2e | 1572 | total_bytes += state->end - state->start + 1; |
7bf811a5 | 1573 | if (total_bytes >= max_bytes) |
573aecaf | 1574 | break; |
573aecaf | 1575 | if (!node) |
d1310b2e CM |
1576 | break; |
1577 | } | |
1578 | out: | |
cad321ad | 1579 | spin_unlock(&tree->lock); |
d1310b2e CM |
1580 | return found; |
1581 | } | |
1582 | ||
143bede5 JM |
1583 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1584 | struct page *locked_page, | |
1585 | u64 start, u64 end) | |
c8b97818 CM |
1586 | { |
1587 | int ret; | |
1588 | struct page *pages[16]; | |
1589 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1590 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1591 | unsigned long nr_pages = end_index - index + 1; | |
1592 | int i; | |
1593 | ||
1594 | if (index == locked_page->index && end_index == index) | |
143bede5 | 1595 | return; |
c8b97818 | 1596 | |
d397712b | 1597 | while (nr_pages > 0) { |
c8b97818 | 1598 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1599 | min_t(unsigned long, nr_pages, |
1600 | ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1601 | for (i = 0; i < ret; i++) { |
1602 | if (pages[i] != locked_page) | |
1603 | unlock_page(pages[i]); | |
1604 | page_cache_release(pages[i]); | |
1605 | } | |
1606 | nr_pages -= ret; | |
1607 | index += ret; | |
1608 | cond_resched(); | |
1609 | } | |
c8b97818 CM |
1610 | } |
1611 | ||
1612 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1613 | struct page *locked_page, | |
1614 | u64 delalloc_start, | |
1615 | u64 delalloc_end) | |
1616 | { | |
1617 | unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT; | |
1618 | unsigned long start_index = index; | |
1619 | unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT; | |
1620 | unsigned long pages_locked = 0; | |
1621 | struct page *pages[16]; | |
1622 | unsigned long nrpages; | |
1623 | int ret; | |
1624 | int i; | |
1625 | ||
1626 | /* the caller is responsible for locking the start index */ | |
1627 | if (index == locked_page->index && index == end_index) | |
1628 | return 0; | |
1629 | ||
1630 | /* skip the page at the start index */ | |
1631 | nrpages = end_index - index + 1; | |
d397712b | 1632 | while (nrpages > 0) { |
c8b97818 | 1633 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1634 | min_t(unsigned long, |
1635 | nrpages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1636 | if (ret == 0) { |
1637 | ret = -EAGAIN; | |
1638 | goto done; | |
1639 | } | |
1640 | /* now we have an array of pages, lock them all */ | |
1641 | for (i = 0; i < ret; i++) { | |
1642 | /* | |
1643 | * the caller is taking responsibility for | |
1644 | * locked_page | |
1645 | */ | |
771ed689 | 1646 | if (pages[i] != locked_page) { |
c8b97818 | 1647 | lock_page(pages[i]); |
f2b1c41c CM |
1648 | if (!PageDirty(pages[i]) || |
1649 | pages[i]->mapping != inode->i_mapping) { | |
771ed689 CM |
1650 | ret = -EAGAIN; |
1651 | unlock_page(pages[i]); | |
1652 | page_cache_release(pages[i]); | |
1653 | goto done; | |
1654 | } | |
1655 | } | |
c8b97818 | 1656 | page_cache_release(pages[i]); |
771ed689 | 1657 | pages_locked++; |
c8b97818 | 1658 | } |
c8b97818 CM |
1659 | nrpages -= ret; |
1660 | index += ret; | |
1661 | cond_resched(); | |
1662 | } | |
1663 | ret = 0; | |
1664 | done: | |
1665 | if (ret && pages_locked) { | |
1666 | __unlock_for_delalloc(inode, locked_page, | |
1667 | delalloc_start, | |
1668 | ((u64)(start_index + pages_locked - 1)) << | |
1669 | PAGE_CACHE_SHIFT); | |
1670 | } | |
1671 | return ret; | |
1672 | } | |
1673 | ||
1674 | /* | |
1675 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1676 | * more than 'max_bytes'. start and end are used to return the range, | |
1677 | * | |
1678 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1679 | */ | |
294e30fe JB |
1680 | STATIC u64 find_lock_delalloc_range(struct inode *inode, |
1681 | struct extent_io_tree *tree, | |
1682 | struct page *locked_page, u64 *start, | |
1683 | u64 *end, u64 max_bytes) | |
c8b97818 CM |
1684 | { |
1685 | u64 delalloc_start; | |
1686 | u64 delalloc_end; | |
1687 | u64 found; | |
9655d298 | 1688 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1689 | int ret; |
1690 | int loops = 0; | |
1691 | ||
1692 | again: | |
1693 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1694 | delalloc_start = *start; | |
1695 | delalloc_end = 0; | |
1696 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1697 | max_bytes, &cached_state); |
70b99e69 | 1698 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1699 | *start = delalloc_start; |
1700 | *end = delalloc_end; | |
c2a128d2 | 1701 | free_extent_state(cached_state); |
385fe0be | 1702 | return 0; |
c8b97818 CM |
1703 | } |
1704 | ||
70b99e69 CM |
1705 | /* |
1706 | * start comes from the offset of locked_page. We have to lock | |
1707 | * pages in order, so we can't process delalloc bytes before | |
1708 | * locked_page | |
1709 | */ | |
d397712b | 1710 | if (delalloc_start < *start) |
70b99e69 | 1711 | delalloc_start = *start; |
70b99e69 | 1712 | |
c8b97818 CM |
1713 | /* |
1714 | * make sure to limit the number of pages we try to lock down | |
c8b97818 | 1715 | */ |
7bf811a5 JB |
1716 | if (delalloc_end + 1 - delalloc_start > max_bytes) |
1717 | delalloc_end = delalloc_start + max_bytes - 1; | |
d397712b | 1718 | |
c8b97818 CM |
1719 | /* step two, lock all the pages after the page that has start */ |
1720 | ret = lock_delalloc_pages(inode, locked_page, | |
1721 | delalloc_start, delalloc_end); | |
1722 | if (ret == -EAGAIN) { | |
1723 | /* some of the pages are gone, lets avoid looping by | |
1724 | * shortening the size of the delalloc range we're searching | |
1725 | */ | |
9655d298 | 1726 | free_extent_state(cached_state); |
7d788742 | 1727 | cached_state = NULL; |
c8b97818 | 1728 | if (!loops) { |
7bf811a5 | 1729 | max_bytes = PAGE_CACHE_SIZE; |
c8b97818 CM |
1730 | loops = 1; |
1731 | goto again; | |
1732 | } else { | |
1733 | found = 0; | |
1734 | goto out_failed; | |
1735 | } | |
1736 | } | |
79787eaa | 1737 | BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */ |
c8b97818 CM |
1738 | |
1739 | /* step three, lock the state bits for the whole range */ | |
d0082371 | 1740 | lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state); |
c8b97818 CM |
1741 | |
1742 | /* then test to make sure it is all still delalloc */ | |
1743 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1744 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1745 | if (!ret) { |
9655d298 CM |
1746 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
1747 | &cached_state, GFP_NOFS); | |
c8b97818 CM |
1748 | __unlock_for_delalloc(inode, locked_page, |
1749 | delalloc_start, delalloc_end); | |
1750 | cond_resched(); | |
1751 | goto again; | |
1752 | } | |
9655d298 | 1753 | free_extent_state(cached_state); |
c8b97818 CM |
1754 | *start = delalloc_start; |
1755 | *end = delalloc_end; | |
1756 | out_failed: | |
1757 | return found; | |
1758 | } | |
1759 | ||
c2790a2e JB |
1760 | int extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end, |
1761 | struct page *locked_page, | |
9ee49a04 | 1762 | unsigned clear_bits, |
c2790a2e | 1763 | unsigned long page_ops) |
c8b97818 | 1764 | { |
c2790a2e | 1765 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
c8b97818 CM |
1766 | int ret; |
1767 | struct page *pages[16]; | |
1768 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1769 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1770 | unsigned long nr_pages = end_index - index + 1; | |
1771 | int i; | |
771ed689 | 1772 | |
2c64c53d | 1773 | clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); |
c2790a2e | 1774 | if (page_ops == 0) |
771ed689 | 1775 | return 0; |
c8b97818 | 1776 | |
704de49d FM |
1777 | if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0) |
1778 | mapping_set_error(inode->i_mapping, -EIO); | |
1779 | ||
d397712b | 1780 | while (nr_pages > 0) { |
c8b97818 | 1781 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1782 | min_t(unsigned long, |
1783 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 | 1784 | for (i = 0; i < ret; i++) { |
8b62b72b | 1785 | |
c2790a2e | 1786 | if (page_ops & PAGE_SET_PRIVATE2) |
8b62b72b CM |
1787 | SetPagePrivate2(pages[i]); |
1788 | ||
c8b97818 CM |
1789 | if (pages[i] == locked_page) { |
1790 | page_cache_release(pages[i]); | |
1791 | continue; | |
1792 | } | |
c2790a2e | 1793 | if (page_ops & PAGE_CLEAR_DIRTY) |
c8b97818 | 1794 | clear_page_dirty_for_io(pages[i]); |
c2790a2e | 1795 | if (page_ops & PAGE_SET_WRITEBACK) |
c8b97818 | 1796 | set_page_writeback(pages[i]); |
704de49d FM |
1797 | if (page_ops & PAGE_SET_ERROR) |
1798 | SetPageError(pages[i]); | |
c2790a2e | 1799 | if (page_ops & PAGE_END_WRITEBACK) |
c8b97818 | 1800 | end_page_writeback(pages[i]); |
c2790a2e | 1801 | if (page_ops & PAGE_UNLOCK) |
771ed689 | 1802 | unlock_page(pages[i]); |
c8b97818 CM |
1803 | page_cache_release(pages[i]); |
1804 | } | |
1805 | nr_pages -= ret; | |
1806 | index += ret; | |
1807 | cond_resched(); | |
1808 | } | |
1809 | return 0; | |
1810 | } | |
c8b97818 | 1811 | |
d352ac68 CM |
1812 | /* |
1813 | * count the number of bytes in the tree that have a given bit(s) | |
1814 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1815 | * cached. The total number found is returned. | |
1816 | */ | |
d1310b2e CM |
1817 | u64 count_range_bits(struct extent_io_tree *tree, |
1818 | u64 *start, u64 search_end, u64 max_bytes, | |
9ee49a04 | 1819 | unsigned bits, int contig) |
d1310b2e CM |
1820 | { |
1821 | struct rb_node *node; | |
1822 | struct extent_state *state; | |
1823 | u64 cur_start = *start; | |
1824 | u64 total_bytes = 0; | |
ec29ed5b | 1825 | u64 last = 0; |
d1310b2e CM |
1826 | int found = 0; |
1827 | ||
fae7f21c | 1828 | if (WARN_ON(search_end <= cur_start)) |
d1310b2e | 1829 | return 0; |
d1310b2e | 1830 | |
cad321ad | 1831 | spin_lock(&tree->lock); |
d1310b2e CM |
1832 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1833 | total_bytes = tree->dirty_bytes; | |
1834 | goto out; | |
1835 | } | |
1836 | /* | |
1837 | * this search will find all the extents that end after | |
1838 | * our range starts. | |
1839 | */ | |
80ea96b1 | 1840 | node = tree_search(tree, cur_start); |
d397712b | 1841 | if (!node) |
d1310b2e | 1842 | goto out; |
d1310b2e | 1843 | |
d397712b | 1844 | while (1) { |
d1310b2e CM |
1845 | state = rb_entry(node, struct extent_state, rb_node); |
1846 | if (state->start > search_end) | |
1847 | break; | |
ec29ed5b CM |
1848 | if (contig && found && state->start > last + 1) |
1849 | break; | |
1850 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1851 | total_bytes += min(search_end, state->end) + 1 - |
1852 | max(cur_start, state->start); | |
1853 | if (total_bytes >= max_bytes) | |
1854 | break; | |
1855 | if (!found) { | |
af60bed2 | 1856 | *start = max(cur_start, state->start); |
d1310b2e CM |
1857 | found = 1; |
1858 | } | |
ec29ed5b CM |
1859 | last = state->end; |
1860 | } else if (contig && found) { | |
1861 | break; | |
d1310b2e CM |
1862 | } |
1863 | node = rb_next(node); | |
1864 | if (!node) | |
1865 | break; | |
1866 | } | |
1867 | out: | |
cad321ad | 1868 | spin_unlock(&tree->lock); |
d1310b2e CM |
1869 | return total_bytes; |
1870 | } | |
b2950863 | 1871 | |
d352ac68 CM |
1872 | /* |
1873 | * set the private field for a given byte offset in the tree. If there isn't | |
1874 | * an extent_state there already, this does nothing. | |
1875 | */ | |
171170c1 | 1876 | static int set_state_private(struct extent_io_tree *tree, u64 start, u64 private) |
d1310b2e CM |
1877 | { |
1878 | struct rb_node *node; | |
1879 | struct extent_state *state; | |
1880 | int ret = 0; | |
1881 | ||
cad321ad | 1882 | spin_lock(&tree->lock); |
d1310b2e CM |
1883 | /* |
1884 | * this search will find all the extents that end after | |
1885 | * our range starts. | |
1886 | */ | |
80ea96b1 | 1887 | node = tree_search(tree, start); |
2b114d1d | 1888 | if (!node) { |
d1310b2e CM |
1889 | ret = -ENOENT; |
1890 | goto out; | |
1891 | } | |
1892 | state = rb_entry(node, struct extent_state, rb_node); | |
1893 | if (state->start != start) { | |
1894 | ret = -ENOENT; | |
1895 | goto out; | |
1896 | } | |
1897 | state->private = private; | |
1898 | out: | |
cad321ad | 1899 | spin_unlock(&tree->lock); |
d1310b2e CM |
1900 | return ret; |
1901 | } | |
1902 | ||
1903 | int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private) | |
1904 | { | |
1905 | struct rb_node *node; | |
1906 | struct extent_state *state; | |
1907 | int ret = 0; | |
1908 | ||
cad321ad | 1909 | spin_lock(&tree->lock); |
d1310b2e CM |
1910 | /* |
1911 | * this search will find all the extents that end after | |
1912 | * our range starts. | |
1913 | */ | |
80ea96b1 | 1914 | node = tree_search(tree, start); |
2b114d1d | 1915 | if (!node) { |
d1310b2e CM |
1916 | ret = -ENOENT; |
1917 | goto out; | |
1918 | } | |
1919 | state = rb_entry(node, struct extent_state, rb_node); | |
1920 | if (state->start != start) { | |
1921 | ret = -ENOENT; | |
1922 | goto out; | |
1923 | } | |
1924 | *private = state->private; | |
1925 | out: | |
cad321ad | 1926 | spin_unlock(&tree->lock); |
d1310b2e CM |
1927 | return ret; |
1928 | } | |
1929 | ||
1930 | /* | |
1931 | * searches a range in the state tree for a given mask. | |
70dec807 | 1932 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1933 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1934 | * range is found set. | |
1935 | */ | |
1936 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1937 | unsigned bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1938 | { |
1939 | struct extent_state *state = NULL; | |
1940 | struct rb_node *node; | |
1941 | int bitset = 0; | |
d1310b2e | 1942 | |
cad321ad | 1943 | spin_lock(&tree->lock); |
27a3507d | 1944 | if (cached && extent_state_in_tree(cached) && cached->start <= start && |
df98b6e2 | 1945 | cached->end > start) |
9655d298 CM |
1946 | node = &cached->rb_node; |
1947 | else | |
1948 | node = tree_search(tree, start); | |
d1310b2e CM |
1949 | while (node && start <= end) { |
1950 | state = rb_entry(node, struct extent_state, rb_node); | |
1951 | ||
1952 | if (filled && state->start > start) { | |
1953 | bitset = 0; | |
1954 | break; | |
1955 | } | |
1956 | ||
1957 | if (state->start > end) | |
1958 | break; | |
1959 | ||
1960 | if (state->state & bits) { | |
1961 | bitset = 1; | |
1962 | if (!filled) | |
1963 | break; | |
1964 | } else if (filled) { | |
1965 | bitset = 0; | |
1966 | break; | |
1967 | } | |
46562cec CM |
1968 | |
1969 | if (state->end == (u64)-1) | |
1970 | break; | |
1971 | ||
d1310b2e CM |
1972 | start = state->end + 1; |
1973 | if (start > end) | |
1974 | break; | |
1975 | node = rb_next(node); | |
1976 | if (!node) { | |
1977 | if (filled) | |
1978 | bitset = 0; | |
1979 | break; | |
1980 | } | |
1981 | } | |
cad321ad | 1982 | spin_unlock(&tree->lock); |
d1310b2e CM |
1983 | return bitset; |
1984 | } | |
d1310b2e CM |
1985 | |
1986 | /* | |
1987 | * helper function to set a given page up to date if all the | |
1988 | * extents in the tree for that page are up to date | |
1989 | */ | |
143bede5 | 1990 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e | 1991 | { |
4eee4fa4 | 1992 | u64 start = page_offset(page); |
d1310b2e | 1993 | u64 end = start + PAGE_CACHE_SIZE - 1; |
9655d298 | 1994 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 1995 | SetPageUptodate(page); |
d1310b2e CM |
1996 | } |
1997 | ||
8b110e39 | 1998 | int free_io_failure(struct inode *inode, struct io_failure_record *rec) |
4a54c8c1 JS |
1999 | { |
2000 | int ret; | |
2001 | int err = 0; | |
2002 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
2003 | ||
2004 | set_state_private(failure_tree, rec->start, 0); | |
2005 | ret = clear_extent_bits(failure_tree, rec->start, | |
2006 | rec->start + rec->len - 1, | |
2007 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
2008 | if (ret) | |
2009 | err = ret; | |
2010 | ||
53b381b3 DW |
2011 | ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start, |
2012 | rec->start + rec->len - 1, | |
2013 | EXTENT_DAMAGED, GFP_NOFS); | |
2014 | if (ret && !err) | |
2015 | err = ret; | |
4a54c8c1 JS |
2016 | |
2017 | kfree(rec); | |
2018 | return err; | |
2019 | } | |
2020 | ||
4a54c8c1 JS |
2021 | /* |
2022 | * this bypasses the standard btrfs submit functions deliberately, as | |
2023 | * the standard behavior is to write all copies in a raid setup. here we only | |
2024 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
2025 | * submit_bio directly. | |
3ec706c8 | 2026 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
2027 | * actually prevents the read that triggered the error from finishing. |
2028 | * currently, there can be no more than two copies of every data bit. thus, | |
2029 | * exactly one rewrite is required. | |
2030 | */ | |
1203b681 MX |
2031 | int repair_io_failure(struct inode *inode, u64 start, u64 length, u64 logical, |
2032 | struct page *page, unsigned int pg_offset, int mirror_num) | |
4a54c8c1 | 2033 | { |
1203b681 | 2034 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; |
4a54c8c1 JS |
2035 | struct bio *bio; |
2036 | struct btrfs_device *dev; | |
4a54c8c1 JS |
2037 | u64 map_length = 0; |
2038 | u64 sector; | |
2039 | struct btrfs_bio *bbio = NULL; | |
53b381b3 | 2040 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
4a54c8c1 JS |
2041 | int ret; |
2042 | ||
908960c6 | 2043 | ASSERT(!(fs_info->sb->s_flags & MS_RDONLY)); |
4a54c8c1 JS |
2044 | BUG_ON(!mirror_num); |
2045 | ||
53b381b3 DW |
2046 | /* we can't repair anything in raid56 yet */ |
2047 | if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num)) | |
2048 | return 0; | |
2049 | ||
9be3395b | 2050 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
4a54c8c1 JS |
2051 | if (!bio) |
2052 | return -EIO; | |
4f024f37 | 2053 | bio->bi_iter.bi_size = 0; |
4a54c8c1 JS |
2054 | map_length = length; |
2055 | ||
3ec706c8 | 2056 | ret = btrfs_map_block(fs_info, WRITE, logical, |
4a54c8c1 JS |
2057 | &map_length, &bbio, mirror_num); |
2058 | if (ret) { | |
2059 | bio_put(bio); | |
2060 | return -EIO; | |
2061 | } | |
2062 | BUG_ON(mirror_num != bbio->mirror_num); | |
2063 | sector = bbio->stripes[mirror_num-1].physical >> 9; | |
4f024f37 | 2064 | bio->bi_iter.bi_sector = sector; |
4a54c8c1 | 2065 | dev = bbio->stripes[mirror_num-1].dev; |
6e9606d2 | 2066 | btrfs_put_bbio(bbio); |
4a54c8c1 JS |
2067 | if (!dev || !dev->bdev || !dev->writeable) { |
2068 | bio_put(bio); | |
2069 | return -EIO; | |
2070 | } | |
2071 | bio->bi_bdev = dev->bdev; | |
ffdd2018 | 2072 | bio_add_page(bio, page, length, pg_offset); |
4a54c8c1 | 2073 | |
33879d45 | 2074 | if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) { |
4a54c8c1 JS |
2075 | /* try to remap that extent elsewhere? */ |
2076 | bio_put(bio); | |
442a4f63 | 2077 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
2078 | return -EIO; |
2079 | } | |
2080 | ||
efe120a0 | 2081 | printk_ratelimited_in_rcu(KERN_INFO |
1203b681 MX |
2082 | "BTRFS: read error corrected: ino %llu off %llu (dev %s sector %llu)\n", |
2083 | btrfs_ino(inode), start, | |
2084 | rcu_str_deref(dev->name), sector); | |
4a54c8c1 JS |
2085 | bio_put(bio); |
2086 | return 0; | |
2087 | } | |
2088 | ||
ea466794 JB |
2089 | int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb, |
2090 | int mirror_num) | |
2091 | { | |
ea466794 JB |
2092 | u64 start = eb->start; |
2093 | unsigned long i, num_pages = num_extent_pages(eb->start, eb->len); | |
d95603b2 | 2094 | int ret = 0; |
ea466794 | 2095 | |
908960c6 ID |
2096 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
2097 | return -EROFS; | |
2098 | ||
ea466794 | 2099 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 2100 | struct page *p = eb->pages[i]; |
1203b681 MX |
2101 | |
2102 | ret = repair_io_failure(root->fs_info->btree_inode, start, | |
2103 | PAGE_CACHE_SIZE, start, p, | |
2104 | start - page_offset(p), mirror_num); | |
ea466794 JB |
2105 | if (ret) |
2106 | break; | |
2107 | start += PAGE_CACHE_SIZE; | |
2108 | } | |
2109 | ||
2110 | return ret; | |
2111 | } | |
2112 | ||
4a54c8c1 JS |
2113 | /* |
2114 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2115 | * to see if we need to process or clean up an io_failure_record | |
2116 | */ | |
8b110e39 MX |
2117 | int clean_io_failure(struct inode *inode, u64 start, struct page *page, |
2118 | unsigned int pg_offset) | |
4a54c8c1 JS |
2119 | { |
2120 | u64 private; | |
2121 | u64 private_failure; | |
2122 | struct io_failure_record *failrec; | |
908960c6 | 2123 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; |
4a54c8c1 JS |
2124 | struct extent_state *state; |
2125 | int num_copies; | |
4a54c8c1 | 2126 | int ret; |
4a54c8c1 JS |
2127 | |
2128 | private = 0; | |
2129 | ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
2130 | (u64)-1, 1, EXTENT_DIRTY, 0); | |
2131 | if (!ret) | |
2132 | return 0; | |
2133 | ||
2134 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start, | |
2135 | &private_failure); | |
2136 | if (ret) | |
2137 | return 0; | |
2138 | ||
2139 | failrec = (struct io_failure_record *)(unsigned long) private_failure; | |
2140 | BUG_ON(!failrec->this_mirror); | |
2141 | ||
2142 | if (failrec->in_validation) { | |
2143 | /* there was no real error, just free the record */ | |
2144 | pr_debug("clean_io_failure: freeing dummy error at %llu\n", | |
2145 | failrec->start); | |
4a54c8c1 JS |
2146 | goto out; |
2147 | } | |
908960c6 ID |
2148 | if (fs_info->sb->s_flags & MS_RDONLY) |
2149 | goto out; | |
4a54c8c1 JS |
2150 | |
2151 | spin_lock(&BTRFS_I(inode)->io_tree.lock); | |
2152 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, | |
2153 | failrec->start, | |
2154 | EXTENT_LOCKED); | |
2155 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); | |
2156 | ||
883d0de4 MX |
2157 | if (state && state->start <= failrec->start && |
2158 | state->end >= failrec->start + failrec->len - 1) { | |
3ec706c8 SB |
2159 | num_copies = btrfs_num_copies(fs_info, failrec->logical, |
2160 | failrec->len); | |
4a54c8c1 | 2161 | if (num_copies > 1) { |
1203b681 | 2162 | repair_io_failure(inode, start, failrec->len, |
454ff3de | 2163 | failrec->logical, page, |
1203b681 | 2164 | pg_offset, failrec->failed_mirror); |
4a54c8c1 JS |
2165 | } |
2166 | } | |
2167 | ||
2168 | out: | |
454ff3de | 2169 | free_io_failure(inode, failrec); |
4a54c8c1 | 2170 | |
454ff3de | 2171 | return 0; |
4a54c8c1 JS |
2172 | } |
2173 | ||
f612496b MX |
2174 | /* |
2175 | * Can be called when | |
2176 | * - hold extent lock | |
2177 | * - under ordered extent | |
2178 | * - the inode is freeing | |
2179 | */ | |
2180 | void btrfs_free_io_failure_record(struct inode *inode, u64 start, u64 end) | |
2181 | { | |
2182 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
2183 | struct io_failure_record *failrec; | |
2184 | struct extent_state *state, *next; | |
2185 | ||
2186 | if (RB_EMPTY_ROOT(&failure_tree->state)) | |
2187 | return; | |
2188 | ||
2189 | spin_lock(&failure_tree->lock); | |
2190 | state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY); | |
2191 | while (state) { | |
2192 | if (state->start > end) | |
2193 | break; | |
2194 | ||
2195 | ASSERT(state->end <= end); | |
2196 | ||
2197 | next = next_state(state); | |
2198 | ||
6e1103a6 | 2199 | failrec = (struct io_failure_record *)(unsigned long)state->private; |
f612496b MX |
2200 | free_extent_state(state); |
2201 | kfree(failrec); | |
2202 | ||
2203 | state = next; | |
2204 | } | |
2205 | spin_unlock(&failure_tree->lock); | |
2206 | } | |
2207 | ||
2fe6303e MX |
2208 | int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end, |
2209 | struct io_failure_record **failrec_ret) | |
4a54c8c1 | 2210 | { |
2fe6303e | 2211 | struct io_failure_record *failrec; |
4a54c8c1 JS |
2212 | u64 private; |
2213 | struct extent_map *em; | |
4a54c8c1 JS |
2214 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2215 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2216 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
4a54c8c1 | 2217 | int ret; |
4a54c8c1 JS |
2218 | u64 logical; |
2219 | ||
4a54c8c1 JS |
2220 | ret = get_state_private(failure_tree, start, &private); |
2221 | if (ret) { | |
2222 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2223 | if (!failrec) | |
2224 | return -ENOMEM; | |
2fe6303e | 2225 | |
4a54c8c1 JS |
2226 | failrec->start = start; |
2227 | failrec->len = end - start + 1; | |
2228 | failrec->this_mirror = 0; | |
2229 | failrec->bio_flags = 0; | |
2230 | failrec->in_validation = 0; | |
2231 | ||
2232 | read_lock(&em_tree->lock); | |
2233 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2234 | if (!em) { | |
2235 | read_unlock(&em_tree->lock); | |
2236 | kfree(failrec); | |
2237 | return -EIO; | |
2238 | } | |
2239 | ||
68ba990f | 2240 | if (em->start > start || em->start + em->len <= start) { |
4a54c8c1 JS |
2241 | free_extent_map(em); |
2242 | em = NULL; | |
2243 | } | |
2244 | read_unlock(&em_tree->lock); | |
7a2d6a64 | 2245 | if (!em) { |
4a54c8c1 JS |
2246 | kfree(failrec); |
2247 | return -EIO; | |
2248 | } | |
2fe6303e | 2249 | |
4a54c8c1 JS |
2250 | logical = start - em->start; |
2251 | logical = em->block_start + logical; | |
2252 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2253 | logical = em->block_start; | |
2254 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2255 | extent_set_compress_type(&failrec->bio_flags, | |
2256 | em->compress_type); | |
2257 | } | |
2fe6303e MX |
2258 | |
2259 | pr_debug("Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu\n", | |
2260 | logical, start, failrec->len); | |
2261 | ||
4a54c8c1 JS |
2262 | failrec->logical = logical; |
2263 | free_extent_map(em); | |
2264 | ||
2265 | /* set the bits in the private failure tree */ | |
2266 | ret = set_extent_bits(failure_tree, start, end, | |
2267 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
2268 | if (ret >= 0) | |
2269 | ret = set_state_private(failure_tree, start, | |
2270 | (u64)(unsigned long)failrec); | |
2271 | /* set the bits in the inode's tree */ | |
2272 | if (ret >= 0) | |
2273 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED, | |
2274 | GFP_NOFS); | |
2275 | if (ret < 0) { | |
2276 | kfree(failrec); | |
2277 | return ret; | |
2278 | } | |
2279 | } else { | |
2280 | failrec = (struct io_failure_record *)(unsigned long)private; | |
2fe6303e | 2281 | pr_debug("Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d\n", |
4a54c8c1 JS |
2282 | failrec->logical, failrec->start, failrec->len, |
2283 | failrec->in_validation); | |
2284 | /* | |
2285 | * when data can be on disk more than twice, add to failrec here | |
2286 | * (e.g. with a list for failed_mirror) to make | |
2287 | * clean_io_failure() clean all those errors at once. | |
2288 | */ | |
2289 | } | |
2fe6303e MX |
2290 | |
2291 | *failrec_ret = failrec; | |
2292 | ||
2293 | return 0; | |
2294 | } | |
2295 | ||
2296 | int btrfs_check_repairable(struct inode *inode, struct bio *failed_bio, | |
2297 | struct io_failure_record *failrec, int failed_mirror) | |
2298 | { | |
2299 | int num_copies; | |
2300 | ||
5d964051 SB |
2301 | num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info, |
2302 | failrec->logical, failrec->len); | |
4a54c8c1 JS |
2303 | if (num_copies == 1) { |
2304 | /* | |
2305 | * we only have a single copy of the data, so don't bother with | |
2306 | * all the retry and error correction code that follows. no | |
2307 | * matter what the error is, it is very likely to persist. | |
2308 | */ | |
2fe6303e | 2309 | pr_debug("Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n", |
09a7f7a2 | 2310 | num_copies, failrec->this_mirror, failed_mirror); |
2fe6303e | 2311 | return 0; |
4a54c8c1 JS |
2312 | } |
2313 | ||
4a54c8c1 JS |
2314 | /* |
2315 | * there are two premises: | |
2316 | * a) deliver good data to the caller | |
2317 | * b) correct the bad sectors on disk | |
2318 | */ | |
2319 | if (failed_bio->bi_vcnt > 1) { | |
2320 | /* | |
2321 | * to fulfill b), we need to know the exact failing sectors, as | |
2322 | * we don't want to rewrite any more than the failed ones. thus, | |
2323 | * we need separate read requests for the failed bio | |
2324 | * | |
2325 | * if the following BUG_ON triggers, our validation request got | |
2326 | * merged. we need separate requests for our algorithm to work. | |
2327 | */ | |
2328 | BUG_ON(failrec->in_validation); | |
2329 | failrec->in_validation = 1; | |
2330 | failrec->this_mirror = failed_mirror; | |
4a54c8c1 JS |
2331 | } else { |
2332 | /* | |
2333 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2334 | * of the failed sector and if we succeed, we have setup | |
2335 | * everything for repair_io_failure to do the rest for us. | |
2336 | */ | |
2337 | if (failrec->in_validation) { | |
2338 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2339 | failrec->in_validation = 0; | |
2340 | failrec->this_mirror = 0; | |
2341 | } | |
2342 | failrec->failed_mirror = failed_mirror; | |
2343 | failrec->this_mirror++; | |
2344 | if (failrec->this_mirror == failed_mirror) | |
2345 | failrec->this_mirror++; | |
4a54c8c1 JS |
2346 | } |
2347 | ||
facc8a22 | 2348 | if (failrec->this_mirror > num_copies) { |
2fe6303e | 2349 | pr_debug("Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n", |
4a54c8c1 | 2350 | num_copies, failrec->this_mirror, failed_mirror); |
2fe6303e | 2351 | return 0; |
4a54c8c1 JS |
2352 | } |
2353 | ||
2fe6303e MX |
2354 | return 1; |
2355 | } | |
2356 | ||
2357 | ||
2358 | struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio, | |
2359 | struct io_failure_record *failrec, | |
2360 | struct page *page, int pg_offset, int icsum, | |
8b110e39 | 2361 | bio_end_io_t *endio_func, void *data) |
2fe6303e MX |
2362 | { |
2363 | struct bio *bio; | |
2364 | struct btrfs_io_bio *btrfs_failed_bio; | |
2365 | struct btrfs_io_bio *btrfs_bio; | |
2366 | ||
9be3395b | 2367 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
2fe6303e MX |
2368 | if (!bio) |
2369 | return NULL; | |
2370 | ||
2371 | bio->bi_end_io = endio_func; | |
4f024f37 | 2372 | bio->bi_iter.bi_sector = failrec->logical >> 9; |
4a54c8c1 | 2373 | bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; |
4f024f37 | 2374 | bio->bi_iter.bi_size = 0; |
8b110e39 | 2375 | bio->bi_private = data; |
4a54c8c1 | 2376 | |
facc8a22 MX |
2377 | btrfs_failed_bio = btrfs_io_bio(failed_bio); |
2378 | if (btrfs_failed_bio->csum) { | |
2379 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2380 | u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); | |
2381 | ||
2382 | btrfs_bio = btrfs_io_bio(bio); | |
2383 | btrfs_bio->csum = btrfs_bio->csum_inline; | |
2fe6303e MX |
2384 | icsum *= csum_size; |
2385 | memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum, | |
facc8a22 MX |
2386 | csum_size); |
2387 | } | |
2388 | ||
2fe6303e MX |
2389 | bio_add_page(bio, page, failrec->len, pg_offset); |
2390 | ||
2391 | return bio; | |
2392 | } | |
2393 | ||
2394 | /* | |
2395 | * this is a generic handler for readpage errors (default | |
2396 | * readpage_io_failed_hook). if other copies exist, read those and write back | |
2397 | * good data to the failed position. does not investigate in remapping the | |
2398 | * failed extent elsewhere, hoping the device will be smart enough to do this as | |
2399 | * needed | |
2400 | */ | |
2401 | ||
2402 | static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset, | |
2403 | struct page *page, u64 start, u64 end, | |
2404 | int failed_mirror) | |
2405 | { | |
2406 | struct io_failure_record *failrec; | |
2407 | struct inode *inode = page->mapping->host; | |
2408 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2409 | struct bio *bio; | |
2410 | int read_mode; | |
2411 | int ret; | |
2412 | ||
2413 | BUG_ON(failed_bio->bi_rw & REQ_WRITE); | |
2414 | ||
2415 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
2416 | if (ret) | |
2417 | return ret; | |
2418 | ||
2419 | ret = btrfs_check_repairable(inode, failed_bio, failrec, failed_mirror); | |
2420 | if (!ret) { | |
2421 | free_io_failure(inode, failrec); | |
2422 | return -EIO; | |
2423 | } | |
2424 | ||
2425 | if (failed_bio->bi_vcnt > 1) | |
2426 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; | |
2427 | else | |
2428 | read_mode = READ_SYNC; | |
2429 | ||
2430 | phy_offset >>= inode->i_sb->s_blocksize_bits; | |
2431 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2432 | start - page_offset(page), | |
8b110e39 MX |
2433 | (int)phy_offset, failed_bio->bi_end_io, |
2434 | NULL); | |
2fe6303e MX |
2435 | if (!bio) { |
2436 | free_io_failure(inode, failrec); | |
2437 | return -EIO; | |
2438 | } | |
4a54c8c1 | 2439 | |
2fe6303e MX |
2440 | pr_debug("Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d\n", |
2441 | read_mode, failrec->this_mirror, failrec->in_validation); | |
4a54c8c1 | 2442 | |
013bd4c3 TI |
2443 | ret = tree->ops->submit_bio_hook(inode, read_mode, bio, |
2444 | failrec->this_mirror, | |
2445 | failrec->bio_flags, 0); | |
6c387ab2 | 2446 | if (ret) { |
454ff3de | 2447 | free_io_failure(inode, failrec); |
6c387ab2 MX |
2448 | bio_put(bio); |
2449 | } | |
2450 | ||
013bd4c3 | 2451 | return ret; |
4a54c8c1 JS |
2452 | } |
2453 | ||
d1310b2e CM |
2454 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2455 | ||
87826df0 JM |
2456 | int end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
2457 | { | |
2458 | int uptodate = (err == 0); | |
2459 | struct extent_io_tree *tree; | |
3e2426bd | 2460 | int ret = 0; |
87826df0 JM |
2461 | |
2462 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
2463 | ||
2464 | if (tree->ops && tree->ops->writepage_end_io_hook) { | |
2465 | ret = tree->ops->writepage_end_io_hook(page, start, | |
2466 | end, NULL, uptodate); | |
2467 | if (ret) | |
2468 | uptodate = 0; | |
2469 | } | |
2470 | ||
87826df0 | 2471 | if (!uptodate) { |
87826df0 JM |
2472 | ClearPageUptodate(page); |
2473 | SetPageError(page); | |
5dca6eea LB |
2474 | ret = ret < 0 ? ret : -EIO; |
2475 | mapping_set_error(page->mapping, ret); | |
87826df0 JM |
2476 | } |
2477 | return 0; | |
2478 | } | |
2479 | ||
d1310b2e CM |
2480 | /* |
2481 | * after a writepage IO is done, we need to: | |
2482 | * clear the uptodate bits on error | |
2483 | * clear the writeback bits in the extent tree for this IO | |
2484 | * end_page_writeback if the page has no more pending IO | |
2485 | * | |
2486 | * Scheduling is not allowed, so the extent state tree is expected | |
2487 | * to have one and only one object corresponding to this IO. | |
2488 | */ | |
4246a0b6 | 2489 | static void end_bio_extent_writepage(struct bio *bio) |
d1310b2e | 2490 | { |
2c30c71b | 2491 | struct bio_vec *bvec; |
d1310b2e CM |
2492 | u64 start; |
2493 | u64 end; | |
2c30c71b | 2494 | int i; |
d1310b2e | 2495 | |
2c30c71b | 2496 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2497 | struct page *page = bvec->bv_page; |
902b22f3 | 2498 | |
17a5adcc AO |
2499 | /* We always issue full-page reads, but if some block |
2500 | * in a page fails to read, blk_update_request() will | |
2501 | * advance bv_offset and adjust bv_len to compensate. | |
2502 | * Print a warning for nonzero offsets, and an error | |
2503 | * if they don't add up to a full page. */ | |
efe120a0 FH |
2504 | if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) { |
2505 | if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE) | |
2506 | btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info, | |
2507 | "partial page write in btrfs with offset %u and length %u", | |
2508 | bvec->bv_offset, bvec->bv_len); | |
2509 | else | |
2510 | btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info, | |
2511 | "incomplete page write in btrfs with offset %u and " | |
2512 | "length %u", | |
2513 | bvec->bv_offset, bvec->bv_len); | |
2514 | } | |
d1310b2e | 2515 | |
17a5adcc AO |
2516 | start = page_offset(page); |
2517 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
d1310b2e | 2518 | |
4246a0b6 | 2519 | if (end_extent_writepage(page, bio->bi_error, start, end)) |
87826df0 | 2520 | continue; |
70dec807 | 2521 | |
17a5adcc | 2522 | end_page_writeback(page); |
2c30c71b | 2523 | } |
2b1f55b0 | 2524 | |
d1310b2e | 2525 | bio_put(bio); |
d1310b2e CM |
2526 | } |
2527 | ||
883d0de4 MX |
2528 | static void |
2529 | endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len, | |
2530 | int uptodate) | |
2531 | { | |
2532 | struct extent_state *cached = NULL; | |
2533 | u64 end = start + len - 1; | |
2534 | ||
2535 | if (uptodate && tree->track_uptodate) | |
2536 | set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC); | |
2537 | unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC); | |
2538 | } | |
2539 | ||
d1310b2e CM |
2540 | /* |
2541 | * after a readpage IO is done, we need to: | |
2542 | * clear the uptodate bits on error | |
2543 | * set the uptodate bits if things worked | |
2544 | * set the page up to date if all extents in the tree are uptodate | |
2545 | * clear the lock bit in the extent tree | |
2546 | * unlock the page if there are no other extents locked for it | |
2547 | * | |
2548 | * Scheduling is not allowed, so the extent state tree is expected | |
2549 | * to have one and only one object corresponding to this IO. | |
2550 | */ | |
4246a0b6 | 2551 | static void end_bio_extent_readpage(struct bio *bio) |
d1310b2e | 2552 | { |
2c30c71b | 2553 | struct bio_vec *bvec; |
4246a0b6 | 2554 | int uptodate = !bio->bi_error; |
facc8a22 | 2555 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); |
902b22f3 | 2556 | struct extent_io_tree *tree; |
facc8a22 | 2557 | u64 offset = 0; |
d1310b2e CM |
2558 | u64 start; |
2559 | u64 end; | |
facc8a22 | 2560 | u64 len; |
883d0de4 MX |
2561 | u64 extent_start = 0; |
2562 | u64 extent_len = 0; | |
5cf1ab56 | 2563 | int mirror; |
d1310b2e | 2564 | int ret; |
2c30c71b | 2565 | int i; |
d1310b2e | 2566 | |
2c30c71b | 2567 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2568 | struct page *page = bvec->bv_page; |
a71754fc | 2569 | struct inode *inode = page->mapping->host; |
507903b8 | 2570 | |
be3940c0 | 2571 | pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, " |
4246a0b6 CH |
2572 | "mirror=%u\n", (u64)bio->bi_iter.bi_sector, |
2573 | bio->bi_error, io_bio->mirror_num); | |
a71754fc | 2574 | tree = &BTRFS_I(inode)->io_tree; |
902b22f3 | 2575 | |
17a5adcc AO |
2576 | /* We always issue full-page reads, but if some block |
2577 | * in a page fails to read, blk_update_request() will | |
2578 | * advance bv_offset and adjust bv_len to compensate. | |
2579 | * Print a warning for nonzero offsets, and an error | |
2580 | * if they don't add up to a full page. */ | |
efe120a0 FH |
2581 | if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) { |
2582 | if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE) | |
2583 | btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info, | |
2584 | "partial page read in btrfs with offset %u and length %u", | |
2585 | bvec->bv_offset, bvec->bv_len); | |
2586 | else | |
2587 | btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info, | |
2588 | "incomplete page read in btrfs with offset %u and " | |
2589 | "length %u", | |
2590 | bvec->bv_offset, bvec->bv_len); | |
2591 | } | |
d1310b2e | 2592 | |
17a5adcc AO |
2593 | start = page_offset(page); |
2594 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
facc8a22 | 2595 | len = bvec->bv_len; |
d1310b2e | 2596 | |
9be3395b | 2597 | mirror = io_bio->mirror_num; |
f2a09da9 MX |
2598 | if (likely(uptodate && tree->ops && |
2599 | tree->ops->readpage_end_io_hook)) { | |
facc8a22 MX |
2600 | ret = tree->ops->readpage_end_io_hook(io_bio, offset, |
2601 | page, start, end, | |
2602 | mirror); | |
5ee0844d | 2603 | if (ret) |
d1310b2e | 2604 | uptodate = 0; |
5ee0844d | 2605 | else |
1203b681 | 2606 | clean_io_failure(inode, start, page, 0); |
d1310b2e | 2607 | } |
ea466794 | 2608 | |
f2a09da9 MX |
2609 | if (likely(uptodate)) |
2610 | goto readpage_ok; | |
2611 | ||
2612 | if (tree->ops && tree->ops->readpage_io_failed_hook) { | |
5cf1ab56 | 2613 | ret = tree->ops->readpage_io_failed_hook(page, mirror); |
4246a0b6 | 2614 | if (!ret && !bio->bi_error) |
ea466794 | 2615 | uptodate = 1; |
f2a09da9 | 2616 | } else { |
f4a8e656 JS |
2617 | /* |
2618 | * The generic bio_readpage_error handles errors the | |
2619 | * following way: If possible, new read requests are | |
2620 | * created and submitted and will end up in | |
2621 | * end_bio_extent_readpage as well (if we're lucky, not | |
2622 | * in the !uptodate case). In that case it returns 0 and | |
2623 | * we just go on with the next page in our bio. If it | |
2624 | * can't handle the error it will return -EIO and we | |
2625 | * remain responsible for that page. | |
2626 | */ | |
facc8a22 MX |
2627 | ret = bio_readpage_error(bio, offset, page, start, end, |
2628 | mirror); | |
7e38326f | 2629 | if (ret == 0) { |
4246a0b6 | 2630 | uptodate = !bio->bi_error; |
38c1c2e4 | 2631 | offset += len; |
7e38326f CM |
2632 | continue; |
2633 | } | |
2634 | } | |
f2a09da9 | 2635 | readpage_ok: |
883d0de4 | 2636 | if (likely(uptodate)) { |
a71754fc JB |
2637 | loff_t i_size = i_size_read(inode); |
2638 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
a583c026 | 2639 | unsigned off; |
a71754fc JB |
2640 | |
2641 | /* Zero out the end if this page straddles i_size */ | |
a583c026 LB |
2642 | off = i_size & (PAGE_CACHE_SIZE-1); |
2643 | if (page->index == end_index && off) | |
2644 | zero_user_segment(page, off, PAGE_CACHE_SIZE); | |
17a5adcc | 2645 | SetPageUptodate(page); |
70dec807 | 2646 | } else { |
17a5adcc AO |
2647 | ClearPageUptodate(page); |
2648 | SetPageError(page); | |
70dec807 | 2649 | } |
17a5adcc | 2650 | unlock_page(page); |
facc8a22 | 2651 | offset += len; |
883d0de4 MX |
2652 | |
2653 | if (unlikely(!uptodate)) { | |
2654 | if (extent_len) { | |
2655 | endio_readpage_release_extent(tree, | |
2656 | extent_start, | |
2657 | extent_len, 1); | |
2658 | extent_start = 0; | |
2659 | extent_len = 0; | |
2660 | } | |
2661 | endio_readpage_release_extent(tree, start, | |
2662 | end - start + 1, 0); | |
2663 | } else if (!extent_len) { | |
2664 | extent_start = start; | |
2665 | extent_len = end + 1 - start; | |
2666 | } else if (extent_start + extent_len == start) { | |
2667 | extent_len += end + 1 - start; | |
2668 | } else { | |
2669 | endio_readpage_release_extent(tree, extent_start, | |
2670 | extent_len, uptodate); | |
2671 | extent_start = start; | |
2672 | extent_len = end + 1 - start; | |
2673 | } | |
2c30c71b | 2674 | } |
d1310b2e | 2675 | |
883d0de4 MX |
2676 | if (extent_len) |
2677 | endio_readpage_release_extent(tree, extent_start, extent_len, | |
2678 | uptodate); | |
facc8a22 | 2679 | if (io_bio->end_io) |
4246a0b6 | 2680 | io_bio->end_io(io_bio, bio->bi_error); |
d1310b2e | 2681 | bio_put(bio); |
d1310b2e CM |
2682 | } |
2683 | ||
9be3395b CM |
2684 | /* |
2685 | * this allocates from the btrfs_bioset. We're returning a bio right now | |
2686 | * but you can call btrfs_io_bio for the appropriate container_of magic | |
2687 | */ | |
88f794ed MX |
2688 | struct bio * |
2689 | btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
2690 | gfp_t gfp_flags) | |
d1310b2e | 2691 | { |
facc8a22 | 2692 | struct btrfs_io_bio *btrfs_bio; |
d1310b2e CM |
2693 | struct bio *bio; |
2694 | ||
9be3395b | 2695 | bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset); |
d1310b2e CM |
2696 | |
2697 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
9be3395b CM |
2698 | while (!bio && (nr_vecs /= 2)) { |
2699 | bio = bio_alloc_bioset(gfp_flags, | |
2700 | nr_vecs, btrfs_bioset); | |
2701 | } | |
d1310b2e CM |
2702 | } |
2703 | ||
2704 | if (bio) { | |
2705 | bio->bi_bdev = bdev; | |
4f024f37 | 2706 | bio->bi_iter.bi_sector = first_sector; |
facc8a22 MX |
2707 | btrfs_bio = btrfs_io_bio(bio); |
2708 | btrfs_bio->csum = NULL; | |
2709 | btrfs_bio->csum_allocated = NULL; | |
2710 | btrfs_bio->end_io = NULL; | |
d1310b2e CM |
2711 | } |
2712 | return bio; | |
2713 | } | |
2714 | ||
9be3395b CM |
2715 | struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask) |
2716 | { | |
23ea8e5a MX |
2717 | struct btrfs_io_bio *btrfs_bio; |
2718 | struct bio *new; | |
9be3395b | 2719 | |
23ea8e5a MX |
2720 | new = bio_clone_bioset(bio, gfp_mask, btrfs_bioset); |
2721 | if (new) { | |
2722 | btrfs_bio = btrfs_io_bio(new); | |
2723 | btrfs_bio->csum = NULL; | |
2724 | btrfs_bio->csum_allocated = NULL; | |
2725 | btrfs_bio->end_io = NULL; | |
3a9508b0 CM |
2726 | |
2727 | #ifdef CONFIG_BLK_CGROUP | |
da2f0f74 CM |
2728 | /* FIXME, put this into bio_clone_bioset */ |
2729 | if (bio->bi_css) | |
2730 | bio_associate_blkcg(new, bio->bi_css); | |
3a9508b0 | 2731 | #endif |
23ea8e5a MX |
2732 | } |
2733 | return new; | |
2734 | } | |
9be3395b CM |
2735 | |
2736 | /* this also allocates from the btrfs_bioset */ | |
2737 | struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs) | |
2738 | { | |
facc8a22 MX |
2739 | struct btrfs_io_bio *btrfs_bio; |
2740 | struct bio *bio; | |
2741 | ||
2742 | bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset); | |
2743 | if (bio) { | |
2744 | btrfs_bio = btrfs_io_bio(bio); | |
2745 | btrfs_bio->csum = NULL; | |
2746 | btrfs_bio->csum_allocated = NULL; | |
2747 | btrfs_bio->end_io = NULL; | |
2748 | } | |
2749 | return bio; | |
9be3395b CM |
2750 | } |
2751 | ||
2752 | ||
355808c2 JM |
2753 | static int __must_check submit_one_bio(int rw, struct bio *bio, |
2754 | int mirror_num, unsigned long bio_flags) | |
d1310b2e | 2755 | { |
d1310b2e | 2756 | int ret = 0; |
70dec807 CM |
2757 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
2758 | struct page *page = bvec->bv_page; | |
2759 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 2760 | u64 start; |
70dec807 | 2761 | |
4eee4fa4 | 2762 | start = page_offset(page) + bvec->bv_offset; |
70dec807 | 2763 | |
902b22f3 | 2764 | bio->bi_private = NULL; |
d1310b2e CM |
2765 | |
2766 | bio_get(bio); | |
2767 | ||
065631f6 | 2768 | if (tree->ops && tree->ops->submit_bio_hook) |
6b82ce8d | 2769 | ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio, |
eaf25d93 | 2770 | mirror_num, bio_flags, start); |
0b86a832 | 2771 | else |
21adbd5c | 2772 | btrfsic_submit_bio(rw, bio); |
4a54c8c1 | 2773 | |
d1310b2e CM |
2774 | bio_put(bio); |
2775 | return ret; | |
2776 | } | |
2777 | ||
64a16701 | 2778 | static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page, |
3444a972 JM |
2779 | unsigned long offset, size_t size, struct bio *bio, |
2780 | unsigned long bio_flags) | |
2781 | { | |
2782 | int ret = 0; | |
2783 | if (tree->ops && tree->ops->merge_bio_hook) | |
64a16701 | 2784 | ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio, |
3444a972 JM |
2785 | bio_flags); |
2786 | BUG_ON(ret < 0); | |
2787 | return ret; | |
2788 | ||
2789 | } | |
2790 | ||
d1310b2e | 2791 | static int submit_extent_page(int rw, struct extent_io_tree *tree, |
da2f0f74 | 2792 | struct writeback_control *wbc, |
d1310b2e CM |
2793 | struct page *page, sector_t sector, |
2794 | size_t size, unsigned long offset, | |
2795 | struct block_device *bdev, | |
2796 | struct bio **bio_ret, | |
2797 | unsigned long max_pages, | |
f188591e | 2798 | bio_end_io_t end_io_func, |
c8b97818 CM |
2799 | int mirror_num, |
2800 | unsigned long prev_bio_flags, | |
005efedf FM |
2801 | unsigned long bio_flags, |
2802 | bool force_bio_submit) | |
d1310b2e CM |
2803 | { |
2804 | int ret = 0; | |
2805 | struct bio *bio; | |
c8b97818 | 2806 | int contig = 0; |
c8b97818 | 2807 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; |
5b050f04 | 2808 | size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE); |
d1310b2e CM |
2809 | |
2810 | if (bio_ret && *bio_ret) { | |
2811 | bio = *bio_ret; | |
c8b97818 | 2812 | if (old_compressed) |
4f024f37 | 2813 | contig = bio->bi_iter.bi_sector == sector; |
c8b97818 | 2814 | else |
f73a1c7d | 2815 | contig = bio_end_sector(bio) == sector; |
c8b97818 CM |
2816 | |
2817 | if (prev_bio_flags != bio_flags || !contig || | |
005efedf | 2818 | force_bio_submit || |
64a16701 | 2819 | merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) || |
c8b97818 CM |
2820 | bio_add_page(bio, page, page_size, offset) < page_size) { |
2821 | ret = submit_one_bio(rw, bio, mirror_num, | |
2822 | prev_bio_flags); | |
289454ad NA |
2823 | if (ret < 0) { |
2824 | *bio_ret = NULL; | |
79787eaa | 2825 | return ret; |
289454ad | 2826 | } |
d1310b2e CM |
2827 | bio = NULL; |
2828 | } else { | |
da2f0f74 CM |
2829 | if (wbc) |
2830 | wbc_account_io(wbc, page, page_size); | |
d1310b2e CM |
2831 | return 0; |
2832 | } | |
2833 | } | |
c8b97818 | 2834 | |
b54ffb73 KO |
2835 | bio = btrfs_bio_alloc(bdev, sector, BIO_MAX_PAGES, |
2836 | GFP_NOFS | __GFP_HIGH); | |
5df67083 TI |
2837 | if (!bio) |
2838 | return -ENOMEM; | |
70dec807 | 2839 | |
c8b97818 | 2840 | bio_add_page(bio, page, page_size, offset); |
d1310b2e CM |
2841 | bio->bi_end_io = end_io_func; |
2842 | bio->bi_private = tree; | |
da2f0f74 CM |
2843 | if (wbc) { |
2844 | wbc_init_bio(wbc, bio); | |
2845 | wbc_account_io(wbc, page, page_size); | |
2846 | } | |
70dec807 | 2847 | |
d397712b | 2848 | if (bio_ret) |
d1310b2e | 2849 | *bio_ret = bio; |
d397712b | 2850 | else |
c8b97818 | 2851 | ret = submit_one_bio(rw, bio, mirror_num, bio_flags); |
d1310b2e CM |
2852 | |
2853 | return ret; | |
2854 | } | |
2855 | ||
48a3b636 ES |
2856 | static void attach_extent_buffer_page(struct extent_buffer *eb, |
2857 | struct page *page) | |
d1310b2e CM |
2858 | { |
2859 | if (!PagePrivate(page)) { | |
2860 | SetPagePrivate(page); | |
d1310b2e | 2861 | page_cache_get(page); |
4f2de97a JB |
2862 | set_page_private(page, (unsigned long)eb); |
2863 | } else { | |
2864 | WARN_ON(page->private != (unsigned long)eb); | |
d1310b2e CM |
2865 | } |
2866 | } | |
2867 | ||
4f2de97a | 2868 | void set_page_extent_mapped(struct page *page) |
d1310b2e | 2869 | { |
4f2de97a JB |
2870 | if (!PagePrivate(page)) { |
2871 | SetPagePrivate(page); | |
2872 | page_cache_get(page); | |
2873 | set_page_private(page, EXTENT_PAGE_PRIVATE); | |
2874 | } | |
d1310b2e CM |
2875 | } |
2876 | ||
125bac01 MX |
2877 | static struct extent_map * |
2878 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
2879 | u64 start, u64 len, get_extent_t *get_extent, | |
2880 | struct extent_map **em_cached) | |
2881 | { | |
2882 | struct extent_map *em; | |
2883 | ||
2884 | if (em_cached && *em_cached) { | |
2885 | em = *em_cached; | |
cbc0e928 | 2886 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 MX |
2887 | start < extent_map_end(em)) { |
2888 | atomic_inc(&em->refs); | |
2889 | return em; | |
2890 | } | |
2891 | ||
2892 | free_extent_map(em); | |
2893 | *em_cached = NULL; | |
2894 | } | |
2895 | ||
2896 | em = get_extent(inode, page, pg_offset, start, len, 0); | |
2897 | if (em_cached && !IS_ERR_OR_NULL(em)) { | |
2898 | BUG_ON(*em_cached); | |
2899 | atomic_inc(&em->refs); | |
2900 | *em_cached = em; | |
2901 | } | |
2902 | return em; | |
2903 | } | |
d1310b2e CM |
2904 | /* |
2905 | * basic readpage implementation. Locked extent state structs are inserted | |
2906 | * into the tree that are removed when the IO is done (by the end_io | |
2907 | * handlers) | |
79787eaa | 2908 | * XXX JDM: This needs looking at to ensure proper page locking |
d1310b2e | 2909 | */ |
9974090b MX |
2910 | static int __do_readpage(struct extent_io_tree *tree, |
2911 | struct page *page, | |
2912 | get_extent_t *get_extent, | |
125bac01 | 2913 | struct extent_map **em_cached, |
9974090b | 2914 | struct bio **bio, int mirror_num, |
005efedf FM |
2915 | unsigned long *bio_flags, int rw, |
2916 | u64 *prev_em_start) | |
d1310b2e CM |
2917 | { |
2918 | struct inode *inode = page->mapping->host; | |
4eee4fa4 | 2919 | u64 start = page_offset(page); |
d1310b2e CM |
2920 | u64 page_end = start + PAGE_CACHE_SIZE - 1; |
2921 | u64 end; | |
2922 | u64 cur = start; | |
2923 | u64 extent_offset; | |
2924 | u64 last_byte = i_size_read(inode); | |
2925 | u64 block_start; | |
2926 | u64 cur_end; | |
2927 | sector_t sector; | |
2928 | struct extent_map *em; | |
2929 | struct block_device *bdev; | |
2930 | int ret; | |
2931 | int nr = 0; | |
4b384318 | 2932 | int parent_locked = *bio_flags & EXTENT_BIO_PARENT_LOCKED; |
306e16ce | 2933 | size_t pg_offset = 0; |
d1310b2e | 2934 | size_t iosize; |
c8b97818 | 2935 | size_t disk_io_size; |
d1310b2e | 2936 | size_t blocksize = inode->i_sb->s_blocksize; |
4b384318 | 2937 | unsigned long this_bio_flag = *bio_flags & EXTENT_BIO_PARENT_LOCKED; |
d1310b2e CM |
2938 | |
2939 | set_page_extent_mapped(page); | |
2940 | ||
9974090b | 2941 | end = page_end; |
90a887c9 DM |
2942 | if (!PageUptodate(page)) { |
2943 | if (cleancache_get_page(page) == 0) { | |
2944 | BUG_ON(blocksize != PAGE_SIZE); | |
9974090b | 2945 | unlock_extent(tree, start, end); |
90a887c9 DM |
2946 | goto out; |
2947 | } | |
2948 | } | |
2949 | ||
c8b97818 CM |
2950 | if (page->index == last_byte >> PAGE_CACHE_SHIFT) { |
2951 | char *userpage; | |
2952 | size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1); | |
2953 | ||
2954 | if (zero_offset) { | |
2955 | iosize = PAGE_CACHE_SIZE - zero_offset; | |
7ac687d9 | 2956 | userpage = kmap_atomic(page); |
c8b97818 CM |
2957 | memset(userpage + zero_offset, 0, iosize); |
2958 | flush_dcache_page(page); | |
7ac687d9 | 2959 | kunmap_atomic(userpage); |
c8b97818 CM |
2960 | } |
2961 | } | |
d1310b2e | 2962 | while (cur <= end) { |
c8f2f24b | 2963 | unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1; |
005efedf | 2964 | bool force_bio_submit = false; |
c8f2f24b | 2965 | |
d1310b2e CM |
2966 | if (cur >= last_byte) { |
2967 | char *userpage; | |
507903b8 AJ |
2968 | struct extent_state *cached = NULL; |
2969 | ||
306e16ce | 2970 | iosize = PAGE_CACHE_SIZE - pg_offset; |
7ac687d9 | 2971 | userpage = kmap_atomic(page); |
306e16ce | 2972 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2973 | flush_dcache_page(page); |
7ac687d9 | 2974 | kunmap_atomic(userpage); |
d1310b2e | 2975 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 2976 | &cached, GFP_NOFS); |
4b384318 MF |
2977 | if (!parent_locked) |
2978 | unlock_extent_cached(tree, cur, | |
2979 | cur + iosize - 1, | |
2980 | &cached, GFP_NOFS); | |
d1310b2e CM |
2981 | break; |
2982 | } | |
125bac01 MX |
2983 | em = __get_extent_map(inode, page, pg_offset, cur, |
2984 | end - cur + 1, get_extent, em_cached); | |
c704005d | 2985 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 2986 | SetPageError(page); |
4b384318 MF |
2987 | if (!parent_locked) |
2988 | unlock_extent(tree, cur, end); | |
d1310b2e CM |
2989 | break; |
2990 | } | |
d1310b2e CM |
2991 | extent_offset = cur - em->start; |
2992 | BUG_ON(extent_map_end(em) <= cur); | |
2993 | BUG_ON(end < cur); | |
2994 | ||
261507a0 | 2995 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
4b384318 | 2996 | this_bio_flag |= EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
2997 | extent_set_compress_type(&this_bio_flag, |
2998 | em->compress_type); | |
2999 | } | |
c8b97818 | 3000 | |
d1310b2e CM |
3001 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
3002 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 3003 | iosize = ALIGN(iosize, blocksize); |
c8b97818 CM |
3004 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
3005 | disk_io_size = em->block_len; | |
3006 | sector = em->block_start >> 9; | |
3007 | } else { | |
3008 | sector = (em->block_start + extent_offset) >> 9; | |
3009 | disk_io_size = iosize; | |
3010 | } | |
d1310b2e CM |
3011 | bdev = em->bdev; |
3012 | block_start = em->block_start; | |
d899e052 YZ |
3013 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
3014 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
3015 | |
3016 | /* | |
3017 | * If we have a file range that points to a compressed extent | |
3018 | * and it's followed by a consecutive file range that points to | |
3019 | * to the same compressed extent (possibly with a different | |
3020 | * offset and/or length, so it either points to the whole extent | |
3021 | * or only part of it), we must make sure we do not submit a | |
3022 | * single bio to populate the pages for the 2 ranges because | |
3023 | * this makes the compressed extent read zero out the pages | |
3024 | * belonging to the 2nd range. Imagine the following scenario: | |
3025 | * | |
3026 | * File layout | |
3027 | * [0 - 8K] [8K - 24K] | |
3028 | * | | | |
3029 | * | | | |
3030 | * points to extent X, points to extent X, | |
3031 | * offset 4K, length of 8K offset 0, length 16K | |
3032 | * | |
3033 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
3034 | * | |
3035 | * If the bio to read the compressed extent covers both ranges, | |
3036 | * it will decompress extent X into the pages belonging to the | |
3037 | * first range and then it will stop, zeroing out the remaining | |
3038 | * pages that belong to the other range that points to extent X. | |
3039 | * So here we make sure we submit 2 bios, one for the first | |
3040 | * range and another one for the third range. Both will target | |
3041 | * the same physical extent from disk, but we can't currently | |
3042 | * make the compressed bio endio callback populate the pages | |
3043 | * for both ranges because each compressed bio is tightly | |
3044 | * coupled with a single extent map, and each range can have | |
3045 | * an extent map with a different offset value relative to the | |
3046 | * uncompressed data of our extent and different lengths. This | |
3047 | * is a corner case so we prioritize correctness over | |
3048 | * non-optimal behavior (submitting 2 bios for the same extent). | |
3049 | */ | |
3050 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
3051 | prev_em_start && *prev_em_start != (u64)-1 && | |
3052 | *prev_em_start != em->orig_start) | |
3053 | force_bio_submit = true; | |
3054 | ||
3055 | if (prev_em_start) | |
3056 | *prev_em_start = em->orig_start; | |
3057 | ||
d1310b2e CM |
3058 | free_extent_map(em); |
3059 | em = NULL; | |
3060 | ||
3061 | /* we've found a hole, just zero and go on */ | |
3062 | if (block_start == EXTENT_MAP_HOLE) { | |
3063 | char *userpage; | |
507903b8 AJ |
3064 | struct extent_state *cached = NULL; |
3065 | ||
7ac687d9 | 3066 | userpage = kmap_atomic(page); |
306e16ce | 3067 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3068 | flush_dcache_page(page); |
7ac687d9 | 3069 | kunmap_atomic(userpage); |
d1310b2e CM |
3070 | |
3071 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 AJ |
3072 | &cached, GFP_NOFS); |
3073 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
3074 | &cached, GFP_NOFS); | |
d1310b2e | 3075 | cur = cur + iosize; |
306e16ce | 3076 | pg_offset += iosize; |
d1310b2e CM |
3077 | continue; |
3078 | } | |
3079 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
3080 | if (test_range_bit(tree, cur, cur_end, |
3081 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 3082 | check_page_uptodate(tree, page); |
4b384318 MF |
3083 | if (!parent_locked) |
3084 | unlock_extent(tree, cur, cur + iosize - 1); | |
d1310b2e | 3085 | cur = cur + iosize; |
306e16ce | 3086 | pg_offset += iosize; |
d1310b2e CM |
3087 | continue; |
3088 | } | |
70dec807 CM |
3089 | /* we have an inline extent but it didn't get marked up |
3090 | * to date. Error out | |
3091 | */ | |
3092 | if (block_start == EXTENT_MAP_INLINE) { | |
3093 | SetPageError(page); | |
4b384318 MF |
3094 | if (!parent_locked) |
3095 | unlock_extent(tree, cur, cur + iosize - 1); | |
70dec807 | 3096 | cur = cur + iosize; |
306e16ce | 3097 | pg_offset += iosize; |
70dec807 CM |
3098 | continue; |
3099 | } | |
d1310b2e | 3100 | |
c8f2f24b | 3101 | pnr -= page->index; |
da2f0f74 | 3102 | ret = submit_extent_page(rw, tree, NULL, page, |
306e16ce | 3103 | sector, disk_io_size, pg_offset, |
89642229 | 3104 | bdev, bio, pnr, |
c8b97818 CM |
3105 | end_bio_extent_readpage, mirror_num, |
3106 | *bio_flags, | |
005efedf FM |
3107 | this_bio_flag, |
3108 | force_bio_submit); | |
c8f2f24b JB |
3109 | if (!ret) { |
3110 | nr++; | |
3111 | *bio_flags = this_bio_flag; | |
3112 | } else { | |
d1310b2e | 3113 | SetPageError(page); |
4b384318 MF |
3114 | if (!parent_locked) |
3115 | unlock_extent(tree, cur, cur + iosize - 1); | |
edd33c99 | 3116 | } |
d1310b2e | 3117 | cur = cur + iosize; |
306e16ce | 3118 | pg_offset += iosize; |
d1310b2e | 3119 | } |
90a887c9 | 3120 | out: |
d1310b2e CM |
3121 | if (!nr) { |
3122 | if (!PageError(page)) | |
3123 | SetPageUptodate(page); | |
3124 | unlock_page(page); | |
3125 | } | |
3126 | return 0; | |
3127 | } | |
3128 | ||
9974090b MX |
3129 | static inline void __do_contiguous_readpages(struct extent_io_tree *tree, |
3130 | struct page *pages[], int nr_pages, | |
3131 | u64 start, u64 end, | |
3132 | get_extent_t *get_extent, | |
125bac01 | 3133 | struct extent_map **em_cached, |
9974090b | 3134 | struct bio **bio, int mirror_num, |
808f80b4 FM |
3135 | unsigned long *bio_flags, int rw, |
3136 | u64 *prev_em_start) | |
9974090b MX |
3137 | { |
3138 | struct inode *inode; | |
3139 | struct btrfs_ordered_extent *ordered; | |
3140 | int index; | |
3141 | ||
3142 | inode = pages[0]->mapping->host; | |
3143 | while (1) { | |
3144 | lock_extent(tree, start, end); | |
3145 | ordered = btrfs_lookup_ordered_range(inode, start, | |
3146 | end - start + 1); | |
3147 | if (!ordered) | |
3148 | break; | |
3149 | unlock_extent(tree, start, end); | |
3150 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3151 | btrfs_put_ordered_extent(ordered); | |
3152 | } | |
3153 | ||
3154 | for (index = 0; index < nr_pages; index++) { | |
125bac01 | 3155 | __do_readpage(tree, pages[index], get_extent, em_cached, bio, |
808f80b4 | 3156 | mirror_num, bio_flags, rw, prev_em_start); |
9974090b MX |
3157 | page_cache_release(pages[index]); |
3158 | } | |
3159 | } | |
3160 | ||
3161 | static void __extent_readpages(struct extent_io_tree *tree, | |
3162 | struct page *pages[], | |
3163 | int nr_pages, get_extent_t *get_extent, | |
125bac01 | 3164 | struct extent_map **em_cached, |
9974090b | 3165 | struct bio **bio, int mirror_num, |
808f80b4 FM |
3166 | unsigned long *bio_flags, int rw, |
3167 | u64 *prev_em_start) | |
9974090b | 3168 | { |
35a3621b | 3169 | u64 start = 0; |
9974090b MX |
3170 | u64 end = 0; |
3171 | u64 page_start; | |
3172 | int index; | |
35a3621b | 3173 | int first_index = 0; |
9974090b MX |
3174 | |
3175 | for (index = 0; index < nr_pages; index++) { | |
3176 | page_start = page_offset(pages[index]); | |
3177 | if (!end) { | |
3178 | start = page_start; | |
3179 | end = start + PAGE_CACHE_SIZE - 1; | |
3180 | first_index = index; | |
3181 | } else if (end + 1 == page_start) { | |
3182 | end += PAGE_CACHE_SIZE; | |
3183 | } else { | |
3184 | __do_contiguous_readpages(tree, &pages[first_index], | |
3185 | index - first_index, start, | |
125bac01 MX |
3186 | end, get_extent, em_cached, |
3187 | bio, mirror_num, bio_flags, | |
808f80b4 | 3188 | rw, prev_em_start); |
9974090b MX |
3189 | start = page_start; |
3190 | end = start + PAGE_CACHE_SIZE - 1; | |
3191 | first_index = index; | |
3192 | } | |
3193 | } | |
3194 | ||
3195 | if (end) | |
3196 | __do_contiguous_readpages(tree, &pages[first_index], | |
3197 | index - first_index, start, | |
125bac01 | 3198 | end, get_extent, em_cached, bio, |
808f80b4 FM |
3199 | mirror_num, bio_flags, rw, |
3200 | prev_em_start); | |
9974090b MX |
3201 | } |
3202 | ||
3203 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
3204 | struct page *page, | |
3205 | get_extent_t *get_extent, | |
3206 | struct bio **bio, int mirror_num, | |
3207 | unsigned long *bio_flags, int rw) | |
3208 | { | |
3209 | struct inode *inode = page->mapping->host; | |
3210 | struct btrfs_ordered_extent *ordered; | |
3211 | u64 start = page_offset(page); | |
3212 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
3213 | int ret; | |
3214 | ||
3215 | while (1) { | |
3216 | lock_extent(tree, start, end); | |
3217 | ordered = btrfs_lookup_ordered_extent(inode, start); | |
3218 | if (!ordered) | |
3219 | break; | |
3220 | unlock_extent(tree, start, end); | |
3221 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3222 | btrfs_put_ordered_extent(ordered); | |
3223 | } | |
3224 | ||
125bac01 | 3225 | ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num, |
005efedf | 3226 | bio_flags, rw, NULL); |
9974090b MX |
3227 | return ret; |
3228 | } | |
3229 | ||
d1310b2e | 3230 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, |
8ddc7d9c | 3231 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3232 | { |
3233 | struct bio *bio = NULL; | |
c8b97818 | 3234 | unsigned long bio_flags = 0; |
d1310b2e CM |
3235 | int ret; |
3236 | ||
8ddc7d9c | 3237 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, |
d4c7ca86 | 3238 | &bio_flags, READ); |
d1310b2e | 3239 | if (bio) |
8ddc7d9c | 3240 | ret = submit_one_bio(READ, bio, mirror_num, bio_flags); |
d1310b2e CM |
3241 | return ret; |
3242 | } | |
d1310b2e | 3243 | |
4b384318 MF |
3244 | int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page, |
3245 | get_extent_t *get_extent, int mirror_num) | |
3246 | { | |
3247 | struct bio *bio = NULL; | |
3248 | unsigned long bio_flags = EXTENT_BIO_PARENT_LOCKED; | |
3249 | int ret; | |
3250 | ||
3251 | ret = __do_readpage(tree, page, get_extent, NULL, &bio, mirror_num, | |
005efedf | 3252 | &bio_flags, READ, NULL); |
4b384318 MF |
3253 | if (bio) |
3254 | ret = submit_one_bio(READ, bio, mirror_num, bio_flags); | |
3255 | return ret; | |
3256 | } | |
3257 | ||
11c8349b CM |
3258 | static noinline void update_nr_written(struct page *page, |
3259 | struct writeback_control *wbc, | |
3260 | unsigned long nr_written) | |
3261 | { | |
3262 | wbc->nr_to_write -= nr_written; | |
3263 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && | |
3264 | wbc->range_start == 0 && wbc->range_end == LLONG_MAX)) | |
3265 | page->mapping->writeback_index = page->index + nr_written; | |
3266 | } | |
3267 | ||
d1310b2e | 3268 | /* |
40f76580 CM |
3269 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
3270 | * | |
3271 | * This returns 1 if our fill_delalloc function did all the work required | |
3272 | * to write the page (copy into inline extent). In this case the IO has | |
3273 | * been started and the page is already unlocked. | |
3274 | * | |
3275 | * This returns 0 if all went well (page still locked) | |
3276 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 3277 | */ |
40f76580 CM |
3278 | static noinline_for_stack int writepage_delalloc(struct inode *inode, |
3279 | struct page *page, struct writeback_control *wbc, | |
3280 | struct extent_page_data *epd, | |
3281 | u64 delalloc_start, | |
3282 | unsigned long *nr_written) | |
3283 | { | |
3284 | struct extent_io_tree *tree = epd->tree; | |
3285 | u64 page_end = delalloc_start + PAGE_CACHE_SIZE - 1; | |
3286 | u64 nr_delalloc; | |
3287 | u64 delalloc_to_write = 0; | |
3288 | u64 delalloc_end = 0; | |
3289 | int ret; | |
3290 | int page_started = 0; | |
3291 | ||
3292 | if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc) | |
3293 | return 0; | |
3294 | ||
3295 | while (delalloc_end < page_end) { | |
3296 | nr_delalloc = find_lock_delalloc_range(inode, tree, | |
3297 | page, | |
3298 | &delalloc_start, | |
3299 | &delalloc_end, | |
dcab6a3b | 3300 | BTRFS_MAX_EXTENT_SIZE); |
40f76580 CM |
3301 | if (nr_delalloc == 0) { |
3302 | delalloc_start = delalloc_end + 1; | |
3303 | continue; | |
3304 | } | |
3305 | ret = tree->ops->fill_delalloc(inode, page, | |
3306 | delalloc_start, | |
3307 | delalloc_end, | |
3308 | &page_started, | |
3309 | nr_written); | |
3310 | /* File system has been set read-only */ | |
3311 | if (ret) { | |
3312 | SetPageError(page); | |
3313 | /* fill_delalloc should be return < 0 for error | |
3314 | * but just in case, we use > 0 here meaning the | |
3315 | * IO is started, so we don't want to return > 0 | |
3316 | * unless things are going well. | |
3317 | */ | |
3318 | ret = ret < 0 ? ret : -EIO; | |
3319 | goto done; | |
3320 | } | |
3321 | /* | |
3322 | * delalloc_end is already one less than the total | |
3323 | * length, so we don't subtract one from | |
3324 | * PAGE_CACHE_SIZE | |
3325 | */ | |
3326 | delalloc_to_write += (delalloc_end - delalloc_start + | |
3327 | PAGE_CACHE_SIZE) >> | |
3328 | PAGE_CACHE_SHIFT; | |
3329 | delalloc_start = delalloc_end + 1; | |
3330 | } | |
3331 | if (wbc->nr_to_write < delalloc_to_write) { | |
3332 | int thresh = 8192; | |
3333 | ||
3334 | if (delalloc_to_write < thresh * 2) | |
3335 | thresh = delalloc_to_write; | |
3336 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3337 | thresh); | |
3338 | } | |
3339 | ||
3340 | /* did the fill delalloc function already unlock and start | |
3341 | * the IO? | |
3342 | */ | |
3343 | if (page_started) { | |
3344 | /* | |
3345 | * we've unlocked the page, so we can't update | |
3346 | * the mapping's writeback index, just update | |
3347 | * nr_to_write. | |
3348 | */ | |
3349 | wbc->nr_to_write -= *nr_written; | |
3350 | return 1; | |
3351 | } | |
3352 | ||
3353 | ret = 0; | |
3354 | ||
3355 | done: | |
3356 | return ret; | |
3357 | } | |
3358 | ||
3359 | /* | |
3360 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3361 | * and does the loop to map the page into extents and bios. | |
3362 | * | |
3363 | * We return 1 if the IO is started and the page is unlocked, | |
3364 | * 0 if all went well (page still locked) | |
3365 | * < 0 if there were errors (page still locked) | |
3366 | */ | |
3367 | static noinline_for_stack int __extent_writepage_io(struct inode *inode, | |
3368 | struct page *page, | |
3369 | struct writeback_control *wbc, | |
3370 | struct extent_page_data *epd, | |
3371 | loff_t i_size, | |
3372 | unsigned long nr_written, | |
3373 | int write_flags, int *nr_ret) | |
d1310b2e | 3374 | { |
d1310b2e | 3375 | struct extent_io_tree *tree = epd->tree; |
4eee4fa4 | 3376 | u64 start = page_offset(page); |
d1310b2e CM |
3377 | u64 page_end = start + PAGE_CACHE_SIZE - 1; |
3378 | u64 end; | |
3379 | u64 cur = start; | |
3380 | u64 extent_offset; | |
d1310b2e CM |
3381 | u64 block_start; |
3382 | u64 iosize; | |
3383 | sector_t sector; | |
2c64c53d | 3384 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
3385 | struct extent_map *em; |
3386 | struct block_device *bdev; | |
7f3c74fb | 3387 | size_t pg_offset = 0; |
d1310b2e | 3388 | size_t blocksize; |
40f76580 CM |
3389 | int ret = 0; |
3390 | int nr = 0; | |
3391 | bool compressed; | |
c8b97818 | 3392 | |
247e743c | 3393 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
3394 | ret = tree->ops->writepage_start_hook(page, start, |
3395 | page_end); | |
87826df0 JM |
3396 | if (ret) { |
3397 | /* Fixup worker will requeue */ | |
3398 | if (ret == -EBUSY) | |
3399 | wbc->pages_skipped++; | |
3400 | else | |
3401 | redirty_page_for_writepage(wbc, page); | |
40f76580 | 3402 | |
11c8349b | 3403 | update_nr_written(page, wbc, nr_written); |
247e743c | 3404 | unlock_page(page); |
40f76580 | 3405 | ret = 1; |
11c8349b | 3406 | goto done_unlocked; |
247e743c CM |
3407 | } |
3408 | } | |
3409 | ||
11c8349b CM |
3410 | /* |
3411 | * we don't want to touch the inode after unlocking the page, | |
3412 | * so we update the mapping writeback index now | |
3413 | */ | |
3414 | update_nr_written(page, wbc, nr_written + 1); | |
771ed689 | 3415 | |
d1310b2e | 3416 | end = page_end; |
40f76580 | 3417 | if (i_size <= start) { |
e6dcd2dc CM |
3418 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3419 | tree->ops->writepage_end_io_hook(page, start, | |
3420 | page_end, NULL, 1); | |
d1310b2e CM |
3421 | goto done; |
3422 | } | |
3423 | ||
d1310b2e CM |
3424 | blocksize = inode->i_sb->s_blocksize; |
3425 | ||
3426 | while (cur <= end) { | |
40f76580 CM |
3427 | u64 em_end; |
3428 | if (cur >= i_size) { | |
e6dcd2dc CM |
3429 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3430 | tree->ops->writepage_end_io_hook(page, cur, | |
3431 | page_end, NULL, 1); | |
d1310b2e CM |
3432 | break; |
3433 | } | |
7f3c74fb | 3434 | em = epd->get_extent(inode, page, pg_offset, cur, |
d1310b2e | 3435 | end - cur + 1, 1); |
c704005d | 3436 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3437 | SetPageError(page); |
61391d56 | 3438 | ret = PTR_ERR_OR_ZERO(em); |
d1310b2e CM |
3439 | break; |
3440 | } | |
3441 | ||
3442 | extent_offset = cur - em->start; | |
40f76580 CM |
3443 | em_end = extent_map_end(em); |
3444 | BUG_ON(em_end <= cur); | |
d1310b2e | 3445 | BUG_ON(end < cur); |
40f76580 | 3446 | iosize = min(em_end - cur, end - cur + 1); |
fda2832f | 3447 | iosize = ALIGN(iosize, blocksize); |
d1310b2e CM |
3448 | sector = (em->block_start + extent_offset) >> 9; |
3449 | bdev = em->bdev; | |
3450 | block_start = em->block_start; | |
c8b97818 | 3451 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
3452 | free_extent_map(em); |
3453 | em = NULL; | |
3454 | ||
c8b97818 CM |
3455 | /* |
3456 | * compressed and inline extents are written through other | |
3457 | * paths in the FS | |
3458 | */ | |
3459 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3460 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
3461 | /* |
3462 | * end_io notification does not happen here for | |
3463 | * compressed extents | |
3464 | */ | |
3465 | if (!compressed && tree->ops && | |
3466 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
3467 | tree->ops->writepage_end_io_hook(page, cur, |
3468 | cur + iosize - 1, | |
3469 | NULL, 1); | |
c8b97818 CM |
3470 | else if (compressed) { |
3471 | /* we don't want to end_page_writeback on | |
3472 | * a compressed extent. this happens | |
3473 | * elsewhere | |
3474 | */ | |
3475 | nr++; | |
3476 | } | |
3477 | ||
3478 | cur += iosize; | |
7f3c74fb | 3479 | pg_offset += iosize; |
d1310b2e CM |
3480 | continue; |
3481 | } | |
c8b97818 | 3482 | |
d1310b2e CM |
3483 | if (tree->ops && tree->ops->writepage_io_hook) { |
3484 | ret = tree->ops->writepage_io_hook(page, cur, | |
3485 | cur + iosize - 1); | |
3486 | } else { | |
3487 | ret = 0; | |
3488 | } | |
1259ab75 | 3489 | if (ret) { |
d1310b2e | 3490 | SetPageError(page); |
1259ab75 | 3491 | } else { |
40f76580 | 3492 | unsigned long max_nr = (i_size >> PAGE_CACHE_SHIFT) + 1; |
7f3c74fb | 3493 | |
d1310b2e CM |
3494 | set_range_writeback(tree, cur, cur + iosize - 1); |
3495 | if (!PageWriteback(page)) { | |
efe120a0 FH |
3496 | btrfs_err(BTRFS_I(inode)->root->fs_info, |
3497 | "page %lu not writeback, cur %llu end %llu", | |
c1c9ff7c | 3498 | page->index, cur, end); |
d1310b2e CM |
3499 | } |
3500 | ||
da2f0f74 | 3501 | ret = submit_extent_page(write_flags, tree, wbc, page, |
ffbd517d CM |
3502 | sector, iosize, pg_offset, |
3503 | bdev, &epd->bio, max_nr, | |
c8b97818 | 3504 | end_bio_extent_writepage, |
005efedf | 3505 | 0, 0, 0, false); |
d1310b2e CM |
3506 | if (ret) |
3507 | SetPageError(page); | |
3508 | } | |
3509 | cur = cur + iosize; | |
7f3c74fb | 3510 | pg_offset += iosize; |
d1310b2e CM |
3511 | nr++; |
3512 | } | |
40f76580 CM |
3513 | done: |
3514 | *nr_ret = nr; | |
3515 | ||
3516 | done_unlocked: | |
3517 | ||
3518 | /* drop our reference on any cached states */ | |
3519 | free_extent_state(cached_state); | |
3520 | return ret; | |
3521 | } | |
3522 | ||
3523 | /* | |
3524 | * the writepage semantics are similar to regular writepage. extent | |
3525 | * records are inserted to lock ranges in the tree, and as dirty areas | |
3526 | * are found, they are marked writeback. Then the lock bits are removed | |
3527 | * and the end_io handler clears the writeback ranges | |
3528 | */ | |
3529 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
3530 | void *data) | |
3531 | { | |
3532 | struct inode *inode = page->mapping->host; | |
3533 | struct extent_page_data *epd = data; | |
3534 | u64 start = page_offset(page); | |
3535 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
3536 | int ret; | |
3537 | int nr = 0; | |
3538 | size_t pg_offset = 0; | |
3539 | loff_t i_size = i_size_read(inode); | |
3540 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
3541 | int write_flags; | |
3542 | unsigned long nr_written = 0; | |
3543 | ||
3544 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3545 | write_flags = WRITE_SYNC; | |
3546 | else | |
3547 | write_flags = WRITE; | |
3548 | ||
3549 | trace___extent_writepage(page, inode, wbc); | |
3550 | ||
3551 | WARN_ON(!PageLocked(page)); | |
3552 | ||
3553 | ClearPageError(page); | |
3554 | ||
3555 | pg_offset = i_size & (PAGE_CACHE_SIZE - 1); | |
3556 | if (page->index > end_index || | |
3557 | (page->index == end_index && !pg_offset)) { | |
3558 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE); | |
3559 | unlock_page(page); | |
3560 | return 0; | |
3561 | } | |
3562 | ||
3563 | if (page->index == end_index) { | |
3564 | char *userpage; | |
3565 | ||
3566 | userpage = kmap_atomic(page); | |
3567 | memset(userpage + pg_offset, 0, | |
3568 | PAGE_CACHE_SIZE - pg_offset); | |
3569 | kunmap_atomic(userpage); | |
3570 | flush_dcache_page(page); | |
3571 | } | |
3572 | ||
3573 | pg_offset = 0; | |
3574 | ||
3575 | set_page_extent_mapped(page); | |
3576 | ||
3577 | ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written); | |
3578 | if (ret == 1) | |
3579 | goto done_unlocked; | |
3580 | if (ret) | |
3581 | goto done; | |
3582 | ||
3583 | ret = __extent_writepage_io(inode, page, wbc, epd, | |
3584 | i_size, nr_written, write_flags, &nr); | |
3585 | if (ret == 1) | |
3586 | goto done_unlocked; | |
3587 | ||
d1310b2e CM |
3588 | done: |
3589 | if (nr == 0) { | |
3590 | /* make sure the mapping tag for page dirty gets cleared */ | |
3591 | set_page_writeback(page); | |
3592 | end_page_writeback(page); | |
3593 | } | |
61391d56 FM |
3594 | if (PageError(page)) { |
3595 | ret = ret < 0 ? ret : -EIO; | |
3596 | end_extent_writepage(page, ret, start, page_end); | |
3597 | } | |
d1310b2e | 3598 | unlock_page(page); |
40f76580 | 3599 | return ret; |
771ed689 | 3600 | |
11c8349b | 3601 | done_unlocked: |
d1310b2e CM |
3602 | return 0; |
3603 | } | |
3604 | ||
fd8b2b61 | 3605 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 3606 | { |
74316201 N |
3607 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
3608 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
3609 | } |
3610 | ||
0e378df1 CM |
3611 | static noinline_for_stack int |
3612 | lock_extent_buffer_for_io(struct extent_buffer *eb, | |
3613 | struct btrfs_fs_info *fs_info, | |
3614 | struct extent_page_data *epd) | |
0b32f4bb JB |
3615 | { |
3616 | unsigned long i, num_pages; | |
3617 | int flush = 0; | |
3618 | int ret = 0; | |
3619 | ||
3620 | if (!btrfs_try_tree_write_lock(eb)) { | |
3621 | flush = 1; | |
3622 | flush_write_bio(epd); | |
3623 | btrfs_tree_lock(eb); | |
3624 | } | |
3625 | ||
3626 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3627 | btrfs_tree_unlock(eb); | |
3628 | if (!epd->sync_io) | |
3629 | return 0; | |
3630 | if (!flush) { | |
3631 | flush_write_bio(epd); | |
3632 | flush = 1; | |
3633 | } | |
a098d8e8 CM |
3634 | while (1) { |
3635 | wait_on_extent_buffer_writeback(eb); | |
3636 | btrfs_tree_lock(eb); | |
3637 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3638 | break; | |
0b32f4bb | 3639 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3640 | } |
3641 | } | |
3642 | ||
51561ffe JB |
3643 | /* |
3644 | * We need to do this to prevent races in people who check if the eb is | |
3645 | * under IO since we can end up having no IO bits set for a short period | |
3646 | * of time. | |
3647 | */ | |
3648 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
3649 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3650 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 3651 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 3652 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
e2d84521 MX |
3653 | __percpu_counter_add(&fs_info->dirty_metadata_bytes, |
3654 | -eb->len, | |
3655 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 3656 | ret = 1; |
51561ffe JB |
3657 | } else { |
3658 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
3659 | } |
3660 | ||
3661 | btrfs_tree_unlock(eb); | |
3662 | ||
3663 | if (!ret) | |
3664 | return ret; | |
3665 | ||
3666 | num_pages = num_extent_pages(eb->start, eb->len); | |
3667 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 3668 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3669 | |
3670 | if (!trylock_page(p)) { | |
3671 | if (!flush) { | |
3672 | flush_write_bio(epd); | |
3673 | flush = 1; | |
3674 | } | |
3675 | lock_page(p); | |
3676 | } | |
3677 | } | |
3678 | ||
3679 | return ret; | |
3680 | } | |
3681 | ||
3682 | static void end_extent_buffer_writeback(struct extent_buffer *eb) | |
3683 | { | |
3684 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
4e857c58 | 3685 | smp_mb__after_atomic(); |
0b32f4bb JB |
3686 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); |
3687 | } | |
3688 | ||
656f30db FM |
3689 | static void set_btree_ioerr(struct page *page) |
3690 | { | |
3691 | struct extent_buffer *eb = (struct extent_buffer *)page->private; | |
3692 | struct btrfs_inode *btree_ino = BTRFS_I(eb->fs_info->btree_inode); | |
3693 | ||
3694 | SetPageError(page); | |
3695 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) | |
3696 | return; | |
3697 | ||
3698 | /* | |
3699 | * If writeback for a btree extent that doesn't belong to a log tree | |
3700 | * failed, increment the counter transaction->eb_write_errors. | |
3701 | * We do this because while the transaction is running and before it's | |
3702 | * committing (when we call filemap_fdata[write|wait]_range against | |
3703 | * the btree inode), we might have | |
3704 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
3705 | * returns an error or an error happens during writeback, when we're | |
3706 | * committing the transaction we wouldn't know about it, since the pages | |
3707 | * can be no longer dirty nor marked anymore for writeback (if a | |
3708 | * subsequent modification to the extent buffer didn't happen before the | |
3709 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
3710 | * able to find the pages tagged with SetPageError at transaction | |
3711 | * commit time. So if this happens we must abort the transaction, | |
3712 | * otherwise we commit a super block with btree roots that point to | |
3713 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
3714 | * or the content of some node/leaf from a past generation that got | |
3715 | * cowed or deleted and is no longer valid. | |
3716 | * | |
3717 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
3718 | * not be enough - we need to distinguish between log tree extents vs | |
3719 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
3720 | * will catch and clear such errors in the mapping - and that call might | |
3721 | * be from a log sync and not from a transaction commit. Also, checking | |
3722 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
3723 | * not done and would not be reliable - the eb might have been released | |
3724 | * from memory and reading it back again means that flag would not be | |
3725 | * set (since it's a runtime flag, not persisted on disk). | |
3726 | * | |
3727 | * Using the flags below in the btree inode also makes us achieve the | |
3728 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
3729 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
3730 | * is called, the writeback for all dirty pages had already finished | |
3731 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
3732 | * filemap_fdatawait_range() would return success, as it could not know | |
3733 | * that writeback errors happened (the pages were no longer tagged for | |
3734 | * writeback). | |
3735 | */ | |
3736 | switch (eb->log_index) { | |
3737 | case -1: | |
3738 | set_bit(BTRFS_INODE_BTREE_ERR, &btree_ino->runtime_flags); | |
3739 | break; | |
3740 | case 0: | |
3741 | set_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags); | |
3742 | break; | |
3743 | case 1: | |
3744 | set_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags); | |
3745 | break; | |
3746 | default: | |
3747 | BUG(); /* unexpected, logic error */ | |
3748 | } | |
3749 | } | |
3750 | ||
4246a0b6 | 3751 | static void end_bio_extent_buffer_writepage(struct bio *bio) |
0b32f4bb | 3752 | { |
2c30c71b | 3753 | struct bio_vec *bvec; |
0b32f4bb | 3754 | struct extent_buffer *eb; |
2c30c71b | 3755 | int i, done; |
0b32f4bb | 3756 | |
2c30c71b | 3757 | bio_for_each_segment_all(bvec, bio, i) { |
0b32f4bb JB |
3758 | struct page *page = bvec->bv_page; |
3759 | ||
0b32f4bb JB |
3760 | eb = (struct extent_buffer *)page->private; |
3761 | BUG_ON(!eb); | |
3762 | done = atomic_dec_and_test(&eb->io_pages); | |
3763 | ||
4246a0b6 CH |
3764 | if (bio->bi_error || |
3765 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { | |
0b32f4bb | 3766 | ClearPageUptodate(page); |
656f30db | 3767 | set_btree_ioerr(page); |
0b32f4bb JB |
3768 | } |
3769 | ||
3770 | end_page_writeback(page); | |
3771 | ||
3772 | if (!done) | |
3773 | continue; | |
3774 | ||
3775 | end_extent_buffer_writeback(eb); | |
2c30c71b | 3776 | } |
0b32f4bb JB |
3777 | |
3778 | bio_put(bio); | |
0b32f4bb JB |
3779 | } |
3780 | ||
0e378df1 | 3781 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb JB |
3782 | struct btrfs_fs_info *fs_info, |
3783 | struct writeback_control *wbc, | |
3784 | struct extent_page_data *epd) | |
3785 | { | |
3786 | struct block_device *bdev = fs_info->fs_devices->latest_bdev; | |
f28491e0 | 3787 | struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree; |
0b32f4bb JB |
3788 | u64 offset = eb->start; |
3789 | unsigned long i, num_pages; | |
de0022b9 | 3790 | unsigned long bio_flags = 0; |
d4c7ca86 | 3791 | int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META; |
d7dbe9e7 | 3792 | int ret = 0; |
0b32f4bb | 3793 | |
656f30db | 3794 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); |
0b32f4bb JB |
3795 | num_pages = num_extent_pages(eb->start, eb->len); |
3796 | atomic_set(&eb->io_pages, num_pages); | |
de0022b9 JB |
3797 | if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID) |
3798 | bio_flags = EXTENT_BIO_TREE_LOG; | |
3799 | ||
0b32f4bb | 3800 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3801 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3802 | |
3803 | clear_page_dirty_for_io(p); | |
3804 | set_page_writeback(p); | |
da2f0f74 | 3805 | ret = submit_extent_page(rw, tree, wbc, p, offset >> 9, |
0b32f4bb JB |
3806 | PAGE_CACHE_SIZE, 0, bdev, &epd->bio, |
3807 | -1, end_bio_extent_buffer_writepage, | |
005efedf | 3808 | 0, epd->bio_flags, bio_flags, false); |
de0022b9 | 3809 | epd->bio_flags = bio_flags; |
0b32f4bb | 3810 | if (ret) { |
656f30db | 3811 | set_btree_ioerr(p); |
55e3bd2e | 3812 | end_page_writeback(p); |
0b32f4bb JB |
3813 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
3814 | end_extent_buffer_writeback(eb); | |
3815 | ret = -EIO; | |
3816 | break; | |
3817 | } | |
3818 | offset += PAGE_CACHE_SIZE; | |
3819 | update_nr_written(p, wbc, 1); | |
3820 | unlock_page(p); | |
3821 | } | |
3822 | ||
3823 | if (unlikely(ret)) { | |
3824 | for (; i < num_pages; i++) { | |
bbf65cf0 | 3825 | struct page *p = eb->pages[i]; |
81465028 | 3826 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
3827 | unlock_page(p); |
3828 | } | |
3829 | } | |
3830 | ||
3831 | return ret; | |
3832 | } | |
3833 | ||
3834 | int btree_write_cache_pages(struct address_space *mapping, | |
3835 | struct writeback_control *wbc) | |
3836 | { | |
3837 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; | |
3838 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; | |
3839 | struct extent_buffer *eb, *prev_eb = NULL; | |
3840 | struct extent_page_data epd = { | |
3841 | .bio = NULL, | |
3842 | .tree = tree, | |
3843 | .extent_locked = 0, | |
3844 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
de0022b9 | 3845 | .bio_flags = 0, |
0b32f4bb JB |
3846 | }; |
3847 | int ret = 0; | |
3848 | int done = 0; | |
3849 | int nr_to_write_done = 0; | |
3850 | struct pagevec pvec; | |
3851 | int nr_pages; | |
3852 | pgoff_t index; | |
3853 | pgoff_t end; /* Inclusive */ | |
3854 | int scanned = 0; | |
3855 | int tag; | |
3856 | ||
3857 | pagevec_init(&pvec, 0); | |
3858 | if (wbc->range_cyclic) { | |
3859 | index = mapping->writeback_index; /* Start from prev offset */ | |
3860 | end = -1; | |
3861 | } else { | |
3862 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
3863 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
3864 | scanned = 1; | |
3865 | } | |
3866 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3867 | tag = PAGECACHE_TAG_TOWRITE; | |
3868 | else | |
3869 | tag = PAGECACHE_TAG_DIRTY; | |
3870 | retry: | |
3871 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3872 | tag_pages_for_writeback(mapping, index, end); | |
3873 | while (!done && !nr_to_write_done && (index <= end) && | |
3874 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, | |
3875 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
3876 | unsigned i; | |
3877 | ||
3878 | scanned = 1; | |
3879 | for (i = 0; i < nr_pages; i++) { | |
3880 | struct page *page = pvec.pages[i]; | |
3881 | ||
3882 | if (!PagePrivate(page)) | |
3883 | continue; | |
3884 | ||
3885 | if (!wbc->range_cyclic && page->index > end) { | |
3886 | done = 1; | |
3887 | break; | |
3888 | } | |
3889 | ||
b5bae261 JB |
3890 | spin_lock(&mapping->private_lock); |
3891 | if (!PagePrivate(page)) { | |
3892 | spin_unlock(&mapping->private_lock); | |
3893 | continue; | |
3894 | } | |
3895 | ||
0b32f4bb | 3896 | eb = (struct extent_buffer *)page->private; |
b5bae261 JB |
3897 | |
3898 | /* | |
3899 | * Shouldn't happen and normally this would be a BUG_ON | |
3900 | * but no sense in crashing the users box for something | |
3901 | * we can survive anyway. | |
3902 | */ | |
fae7f21c | 3903 | if (WARN_ON(!eb)) { |
b5bae261 | 3904 | spin_unlock(&mapping->private_lock); |
0b32f4bb JB |
3905 | continue; |
3906 | } | |
3907 | ||
b5bae261 JB |
3908 | if (eb == prev_eb) { |
3909 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 3910 | continue; |
b5bae261 | 3911 | } |
0b32f4bb | 3912 | |
b5bae261 JB |
3913 | ret = atomic_inc_not_zero(&eb->refs); |
3914 | spin_unlock(&mapping->private_lock); | |
3915 | if (!ret) | |
0b32f4bb | 3916 | continue; |
0b32f4bb JB |
3917 | |
3918 | prev_eb = eb; | |
3919 | ret = lock_extent_buffer_for_io(eb, fs_info, &epd); | |
3920 | if (!ret) { | |
3921 | free_extent_buffer(eb); | |
3922 | continue; | |
3923 | } | |
3924 | ||
3925 | ret = write_one_eb(eb, fs_info, wbc, &epd); | |
3926 | if (ret) { | |
3927 | done = 1; | |
3928 | free_extent_buffer(eb); | |
3929 | break; | |
3930 | } | |
3931 | free_extent_buffer(eb); | |
3932 | ||
3933 | /* | |
3934 | * the filesystem may choose to bump up nr_to_write. | |
3935 | * We have to make sure to honor the new nr_to_write | |
3936 | * at any time | |
3937 | */ | |
3938 | nr_to_write_done = wbc->nr_to_write <= 0; | |
3939 | } | |
3940 | pagevec_release(&pvec); | |
3941 | cond_resched(); | |
3942 | } | |
3943 | if (!scanned && !done) { | |
3944 | /* | |
3945 | * We hit the last page and there is more work to be done: wrap | |
3946 | * back to the start of the file | |
3947 | */ | |
3948 | scanned = 1; | |
3949 | index = 0; | |
3950 | goto retry; | |
3951 | } | |
3952 | flush_write_bio(&epd); | |
3953 | return ret; | |
3954 | } | |
3955 | ||
d1310b2e | 3956 | /** |
4bef0848 | 3957 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
3958 | * @mapping: address space structure to write |
3959 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
3960 | * @writepage: function called for each page | |
3961 | * @data: data passed to writepage function | |
3962 | * | |
3963 | * If a page is already under I/O, write_cache_pages() skips it, even | |
3964 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
3965 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
3966 | * and msync() need to guarantee that all the data which was dirty at the time | |
3967 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
3968 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
3969 | * existing IO to complete. | |
3970 | */ | |
b2950863 | 3971 | static int extent_write_cache_pages(struct extent_io_tree *tree, |
4bef0848 CM |
3972 | struct address_space *mapping, |
3973 | struct writeback_control *wbc, | |
d2c3f4f6 CM |
3974 | writepage_t writepage, void *data, |
3975 | void (*flush_fn)(void *)) | |
d1310b2e | 3976 | { |
7fd1a3f7 | 3977 | struct inode *inode = mapping->host; |
d1310b2e CM |
3978 | int ret = 0; |
3979 | int done = 0; | |
61391d56 | 3980 | int err = 0; |
f85d7d6c | 3981 | int nr_to_write_done = 0; |
d1310b2e CM |
3982 | struct pagevec pvec; |
3983 | int nr_pages; | |
3984 | pgoff_t index; | |
3985 | pgoff_t end; /* Inclusive */ | |
3986 | int scanned = 0; | |
f7aaa06b | 3987 | int tag; |
d1310b2e | 3988 | |
7fd1a3f7 JB |
3989 | /* |
3990 | * We have to hold onto the inode so that ordered extents can do their | |
3991 | * work when the IO finishes. The alternative to this is failing to add | |
3992 | * an ordered extent if the igrab() fails there and that is a huge pain | |
3993 | * to deal with, so instead just hold onto the inode throughout the | |
3994 | * writepages operation. If it fails here we are freeing up the inode | |
3995 | * anyway and we'd rather not waste our time writing out stuff that is | |
3996 | * going to be truncated anyway. | |
3997 | */ | |
3998 | if (!igrab(inode)) | |
3999 | return 0; | |
4000 | ||
d1310b2e CM |
4001 | pagevec_init(&pvec, 0); |
4002 | if (wbc->range_cyclic) { | |
4003 | index = mapping->writeback_index; /* Start from prev offset */ | |
4004 | end = -1; | |
4005 | } else { | |
4006 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
4007 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
4008 | scanned = 1; |
4009 | } | |
f7aaa06b JB |
4010 | if (wbc->sync_mode == WB_SYNC_ALL) |
4011 | tag = PAGECACHE_TAG_TOWRITE; | |
4012 | else | |
4013 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 4014 | retry: |
f7aaa06b JB |
4015 | if (wbc->sync_mode == WB_SYNC_ALL) |
4016 | tag_pages_for_writeback(mapping, index, end); | |
f85d7d6c | 4017 | while (!done && !nr_to_write_done && (index <= end) && |
f7aaa06b JB |
4018 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
4019 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
d1310b2e CM |
4020 | unsigned i; |
4021 | ||
4022 | scanned = 1; | |
4023 | for (i = 0; i < nr_pages; i++) { | |
4024 | struct page *page = pvec.pages[i]; | |
4025 | ||
4026 | /* | |
4027 | * At this point we hold neither mapping->tree_lock nor | |
4028 | * lock on the page itself: the page may be truncated or | |
4029 | * invalidated (changing page->mapping to NULL), or even | |
4030 | * swizzled back from swapper_space to tmpfs file | |
4031 | * mapping | |
4032 | */ | |
c8f2f24b JB |
4033 | if (!trylock_page(page)) { |
4034 | flush_fn(data); | |
4035 | lock_page(page); | |
01d658f2 | 4036 | } |
d1310b2e CM |
4037 | |
4038 | if (unlikely(page->mapping != mapping)) { | |
4039 | unlock_page(page); | |
4040 | continue; | |
4041 | } | |
4042 | ||
4043 | if (!wbc->range_cyclic && page->index > end) { | |
4044 | done = 1; | |
4045 | unlock_page(page); | |
4046 | continue; | |
4047 | } | |
4048 | ||
d2c3f4f6 | 4049 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 CM |
4050 | if (PageWriteback(page)) |
4051 | flush_fn(data); | |
d1310b2e | 4052 | wait_on_page_writeback(page); |
d2c3f4f6 | 4053 | } |
d1310b2e CM |
4054 | |
4055 | if (PageWriteback(page) || | |
4056 | !clear_page_dirty_for_io(page)) { | |
4057 | unlock_page(page); | |
4058 | continue; | |
4059 | } | |
4060 | ||
4061 | ret = (*writepage)(page, wbc, data); | |
4062 | ||
4063 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
4064 | unlock_page(page); | |
4065 | ret = 0; | |
4066 | } | |
61391d56 FM |
4067 | if (!err && ret < 0) |
4068 | err = ret; | |
f85d7d6c CM |
4069 | |
4070 | /* | |
4071 | * the filesystem may choose to bump up nr_to_write. | |
4072 | * We have to make sure to honor the new nr_to_write | |
4073 | * at any time | |
4074 | */ | |
4075 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
4076 | } |
4077 | pagevec_release(&pvec); | |
4078 | cond_resched(); | |
4079 | } | |
61391d56 | 4080 | if (!scanned && !done && !err) { |
d1310b2e CM |
4081 | /* |
4082 | * We hit the last page and there is more work to be done: wrap | |
4083 | * back to the start of the file | |
4084 | */ | |
4085 | scanned = 1; | |
4086 | index = 0; | |
4087 | goto retry; | |
4088 | } | |
7fd1a3f7 | 4089 | btrfs_add_delayed_iput(inode); |
61391d56 | 4090 | return err; |
d1310b2e | 4091 | } |
d1310b2e | 4092 | |
ffbd517d | 4093 | static void flush_epd_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 4094 | { |
d2c3f4f6 | 4095 | if (epd->bio) { |
355808c2 JM |
4096 | int rw = WRITE; |
4097 | int ret; | |
4098 | ||
ffbd517d | 4099 | if (epd->sync_io) |
355808c2 JM |
4100 | rw = WRITE_SYNC; |
4101 | ||
de0022b9 | 4102 | ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags); |
79787eaa | 4103 | BUG_ON(ret < 0); /* -ENOMEM */ |
d2c3f4f6 CM |
4104 | epd->bio = NULL; |
4105 | } | |
4106 | } | |
4107 | ||
ffbd517d CM |
4108 | static noinline void flush_write_bio(void *data) |
4109 | { | |
4110 | struct extent_page_data *epd = data; | |
4111 | flush_epd_write_bio(epd); | |
4112 | } | |
4113 | ||
d1310b2e CM |
4114 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, |
4115 | get_extent_t *get_extent, | |
4116 | struct writeback_control *wbc) | |
4117 | { | |
4118 | int ret; | |
d1310b2e CM |
4119 | struct extent_page_data epd = { |
4120 | .bio = NULL, | |
4121 | .tree = tree, | |
4122 | .get_extent = get_extent, | |
771ed689 | 4123 | .extent_locked = 0, |
ffbd517d | 4124 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
de0022b9 | 4125 | .bio_flags = 0, |
d1310b2e | 4126 | }; |
d1310b2e | 4127 | |
d1310b2e CM |
4128 | ret = __extent_writepage(page, wbc, &epd); |
4129 | ||
ffbd517d | 4130 | flush_epd_write_bio(&epd); |
d1310b2e CM |
4131 | return ret; |
4132 | } | |
d1310b2e | 4133 | |
771ed689 CM |
4134 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, |
4135 | u64 start, u64 end, get_extent_t *get_extent, | |
4136 | int mode) | |
4137 | { | |
4138 | int ret = 0; | |
4139 | struct address_space *mapping = inode->i_mapping; | |
4140 | struct page *page; | |
4141 | unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >> | |
4142 | PAGE_CACHE_SHIFT; | |
4143 | ||
4144 | struct extent_page_data epd = { | |
4145 | .bio = NULL, | |
4146 | .tree = tree, | |
4147 | .get_extent = get_extent, | |
4148 | .extent_locked = 1, | |
ffbd517d | 4149 | .sync_io = mode == WB_SYNC_ALL, |
de0022b9 | 4150 | .bio_flags = 0, |
771ed689 CM |
4151 | }; |
4152 | struct writeback_control wbc_writepages = { | |
771ed689 | 4153 | .sync_mode = mode, |
771ed689 CM |
4154 | .nr_to_write = nr_pages * 2, |
4155 | .range_start = start, | |
4156 | .range_end = end + 1, | |
4157 | }; | |
4158 | ||
d397712b | 4159 | while (start <= end) { |
771ed689 CM |
4160 | page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); |
4161 | if (clear_page_dirty_for_io(page)) | |
4162 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
4163 | else { | |
4164 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
4165 | tree->ops->writepage_end_io_hook(page, start, | |
4166 | start + PAGE_CACHE_SIZE - 1, | |
4167 | NULL, 1); | |
4168 | unlock_page(page); | |
4169 | } | |
4170 | page_cache_release(page); | |
4171 | start += PAGE_CACHE_SIZE; | |
4172 | } | |
4173 | ||
ffbd517d | 4174 | flush_epd_write_bio(&epd); |
771ed689 CM |
4175 | return ret; |
4176 | } | |
d1310b2e CM |
4177 | |
4178 | int extent_writepages(struct extent_io_tree *tree, | |
4179 | struct address_space *mapping, | |
4180 | get_extent_t *get_extent, | |
4181 | struct writeback_control *wbc) | |
4182 | { | |
4183 | int ret = 0; | |
4184 | struct extent_page_data epd = { | |
4185 | .bio = NULL, | |
4186 | .tree = tree, | |
4187 | .get_extent = get_extent, | |
771ed689 | 4188 | .extent_locked = 0, |
ffbd517d | 4189 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
de0022b9 | 4190 | .bio_flags = 0, |
d1310b2e CM |
4191 | }; |
4192 | ||
4bef0848 | 4193 | ret = extent_write_cache_pages(tree, mapping, wbc, |
d2c3f4f6 CM |
4194 | __extent_writepage, &epd, |
4195 | flush_write_bio); | |
ffbd517d | 4196 | flush_epd_write_bio(&epd); |
d1310b2e CM |
4197 | return ret; |
4198 | } | |
d1310b2e CM |
4199 | |
4200 | int extent_readpages(struct extent_io_tree *tree, | |
4201 | struct address_space *mapping, | |
4202 | struct list_head *pages, unsigned nr_pages, | |
4203 | get_extent_t get_extent) | |
4204 | { | |
4205 | struct bio *bio = NULL; | |
4206 | unsigned page_idx; | |
c8b97818 | 4207 | unsigned long bio_flags = 0; |
67c9684f LB |
4208 | struct page *pagepool[16]; |
4209 | struct page *page; | |
125bac01 | 4210 | struct extent_map *em_cached = NULL; |
67c9684f | 4211 | int nr = 0; |
808f80b4 | 4212 | u64 prev_em_start = (u64)-1; |
d1310b2e | 4213 | |
d1310b2e | 4214 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
67c9684f | 4215 | page = list_entry(pages->prev, struct page, lru); |
d1310b2e CM |
4216 | |
4217 | prefetchw(&page->flags); | |
4218 | list_del(&page->lru); | |
67c9684f | 4219 | if (add_to_page_cache_lru(page, mapping, |
43e817a1 | 4220 | page->index, GFP_NOFS)) { |
67c9684f LB |
4221 | page_cache_release(page); |
4222 | continue; | |
d1310b2e | 4223 | } |
67c9684f LB |
4224 | |
4225 | pagepool[nr++] = page; | |
4226 | if (nr < ARRAY_SIZE(pagepool)) | |
4227 | continue; | |
125bac01 | 4228 | __extent_readpages(tree, pagepool, nr, get_extent, &em_cached, |
808f80b4 | 4229 | &bio, 0, &bio_flags, READ, &prev_em_start); |
67c9684f | 4230 | nr = 0; |
d1310b2e | 4231 | } |
9974090b | 4232 | if (nr) |
125bac01 | 4233 | __extent_readpages(tree, pagepool, nr, get_extent, &em_cached, |
808f80b4 | 4234 | &bio, 0, &bio_flags, READ, &prev_em_start); |
67c9684f | 4235 | |
125bac01 MX |
4236 | if (em_cached) |
4237 | free_extent_map(em_cached); | |
4238 | ||
d1310b2e CM |
4239 | BUG_ON(!list_empty(pages)); |
4240 | if (bio) | |
79787eaa | 4241 | return submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
4242 | return 0; |
4243 | } | |
d1310b2e CM |
4244 | |
4245 | /* | |
4246 | * basic invalidatepage code, this waits on any locked or writeback | |
4247 | * ranges corresponding to the page, and then deletes any extent state | |
4248 | * records from the tree | |
4249 | */ | |
4250 | int extent_invalidatepage(struct extent_io_tree *tree, | |
4251 | struct page *page, unsigned long offset) | |
4252 | { | |
2ac55d41 | 4253 | struct extent_state *cached_state = NULL; |
4eee4fa4 | 4254 | u64 start = page_offset(page); |
d1310b2e CM |
4255 | u64 end = start + PAGE_CACHE_SIZE - 1; |
4256 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
4257 | ||
fda2832f | 4258 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
4259 | if (start > end) |
4260 | return 0; | |
4261 | ||
d0082371 | 4262 | lock_extent_bits(tree, start, end, 0, &cached_state); |
1edbb734 | 4263 | wait_on_page_writeback(page); |
d1310b2e | 4264 | clear_extent_bit(tree, start, end, |
32c00aff JB |
4265 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
4266 | EXTENT_DO_ACCOUNTING, | |
2ac55d41 | 4267 | 1, 1, &cached_state, GFP_NOFS); |
d1310b2e CM |
4268 | return 0; |
4269 | } | |
d1310b2e | 4270 | |
7b13b7b1 CM |
4271 | /* |
4272 | * a helper for releasepage, this tests for areas of the page that | |
4273 | * are locked or under IO and drops the related state bits if it is safe | |
4274 | * to drop the page. | |
4275 | */ | |
48a3b636 ES |
4276 | static int try_release_extent_state(struct extent_map_tree *map, |
4277 | struct extent_io_tree *tree, | |
4278 | struct page *page, gfp_t mask) | |
7b13b7b1 | 4279 | { |
4eee4fa4 | 4280 | u64 start = page_offset(page); |
7b13b7b1 CM |
4281 | u64 end = start + PAGE_CACHE_SIZE - 1; |
4282 | int ret = 1; | |
4283 | ||
211f90e6 | 4284 | if (test_range_bit(tree, start, end, |
8b62b72b | 4285 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
4286 | ret = 0; |
4287 | else { | |
4288 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
4289 | mask = GFP_NOFS; | |
11ef160f CM |
4290 | /* |
4291 | * at this point we can safely clear everything except the | |
4292 | * locked bit and the nodatasum bit | |
4293 | */ | |
e3f24cc5 | 4294 | ret = clear_extent_bit(tree, start, end, |
11ef160f CM |
4295 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
4296 | 0, 0, NULL, mask); | |
e3f24cc5 CM |
4297 | |
4298 | /* if clear_extent_bit failed for enomem reasons, | |
4299 | * we can't allow the release to continue. | |
4300 | */ | |
4301 | if (ret < 0) | |
4302 | ret = 0; | |
4303 | else | |
4304 | ret = 1; | |
7b13b7b1 CM |
4305 | } |
4306 | return ret; | |
4307 | } | |
7b13b7b1 | 4308 | |
d1310b2e CM |
4309 | /* |
4310 | * a helper for releasepage. As long as there are no locked extents | |
4311 | * in the range corresponding to the page, both state records and extent | |
4312 | * map records are removed | |
4313 | */ | |
4314 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
4315 | struct extent_io_tree *tree, struct page *page, |
4316 | gfp_t mask) | |
d1310b2e CM |
4317 | { |
4318 | struct extent_map *em; | |
4eee4fa4 | 4319 | u64 start = page_offset(page); |
d1310b2e | 4320 | u64 end = start + PAGE_CACHE_SIZE - 1; |
7b13b7b1 | 4321 | |
70dec807 CM |
4322 | if ((mask & __GFP_WAIT) && |
4323 | page->mapping->host->i_size > 16 * 1024 * 1024) { | |
39b5637f | 4324 | u64 len; |
70dec807 | 4325 | while (start <= end) { |
39b5637f | 4326 | len = end - start + 1; |
890871be | 4327 | write_lock(&map->lock); |
39b5637f | 4328 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 4329 | if (!em) { |
890871be | 4330 | write_unlock(&map->lock); |
70dec807 CM |
4331 | break; |
4332 | } | |
7f3c74fb CM |
4333 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
4334 | em->start != start) { | |
890871be | 4335 | write_unlock(&map->lock); |
70dec807 CM |
4336 | free_extent_map(em); |
4337 | break; | |
4338 | } | |
4339 | if (!test_range_bit(tree, em->start, | |
4340 | extent_map_end(em) - 1, | |
8b62b72b | 4341 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 4342 | 0, NULL)) { |
70dec807 CM |
4343 | remove_extent_mapping(map, em); |
4344 | /* once for the rb tree */ | |
4345 | free_extent_map(em); | |
4346 | } | |
4347 | start = extent_map_end(em); | |
890871be | 4348 | write_unlock(&map->lock); |
70dec807 CM |
4349 | |
4350 | /* once for us */ | |
d1310b2e CM |
4351 | free_extent_map(em); |
4352 | } | |
d1310b2e | 4353 | } |
7b13b7b1 | 4354 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 4355 | } |
d1310b2e | 4356 | |
ec29ed5b CM |
4357 | /* |
4358 | * helper function for fiemap, which doesn't want to see any holes. | |
4359 | * This maps until we find something past 'last' | |
4360 | */ | |
4361 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
4362 | u64 offset, | |
4363 | u64 last, | |
4364 | get_extent_t *get_extent) | |
4365 | { | |
4366 | u64 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
4367 | struct extent_map *em; | |
4368 | u64 len; | |
4369 | ||
4370 | if (offset >= last) | |
4371 | return NULL; | |
4372 | ||
67871254 | 4373 | while (1) { |
ec29ed5b CM |
4374 | len = last - offset; |
4375 | if (len == 0) | |
4376 | break; | |
fda2832f | 4377 | len = ALIGN(len, sectorsize); |
ec29ed5b | 4378 | em = get_extent(inode, NULL, 0, offset, len, 0); |
c704005d | 4379 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
4380 | return em; |
4381 | ||
4382 | /* if this isn't a hole return it */ | |
4383 | if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) && | |
4384 | em->block_start != EXTENT_MAP_HOLE) { | |
4385 | return em; | |
4386 | } | |
4387 | ||
4388 | /* this is a hole, advance to the next extent */ | |
4389 | offset = extent_map_end(em); | |
4390 | free_extent_map(em); | |
4391 | if (offset >= last) | |
4392 | break; | |
4393 | } | |
4394 | return NULL; | |
4395 | } | |
4396 | ||
1506fcc8 YS |
4397 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4398 | __u64 start, __u64 len, get_extent_t *get_extent) | |
4399 | { | |
975f84fe | 4400 | int ret = 0; |
1506fcc8 YS |
4401 | u64 off = start; |
4402 | u64 max = start + len; | |
4403 | u32 flags = 0; | |
975f84fe JB |
4404 | u32 found_type; |
4405 | u64 last; | |
ec29ed5b | 4406 | u64 last_for_get_extent = 0; |
1506fcc8 | 4407 | u64 disko = 0; |
ec29ed5b | 4408 | u64 isize = i_size_read(inode); |
975f84fe | 4409 | struct btrfs_key found_key; |
1506fcc8 | 4410 | struct extent_map *em = NULL; |
2ac55d41 | 4411 | struct extent_state *cached_state = NULL; |
975f84fe | 4412 | struct btrfs_path *path; |
dc046b10 | 4413 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1506fcc8 | 4414 | int end = 0; |
ec29ed5b CM |
4415 | u64 em_start = 0; |
4416 | u64 em_len = 0; | |
4417 | u64 em_end = 0; | |
1506fcc8 YS |
4418 | |
4419 | if (len == 0) | |
4420 | return -EINVAL; | |
4421 | ||
975f84fe JB |
4422 | path = btrfs_alloc_path(); |
4423 | if (!path) | |
4424 | return -ENOMEM; | |
4425 | path->leave_spinning = 1; | |
4426 | ||
2c91943b QW |
4427 | start = round_down(start, BTRFS_I(inode)->root->sectorsize); |
4428 | len = round_up(max, BTRFS_I(inode)->root->sectorsize) - start; | |
4d479cf0 | 4429 | |
ec29ed5b CM |
4430 | /* |
4431 | * lookup the last file extent. We're not using i_size here | |
4432 | * because there might be preallocation past i_size | |
4433 | */ | |
dc046b10 JB |
4434 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1, |
4435 | 0); | |
975f84fe JB |
4436 | if (ret < 0) { |
4437 | btrfs_free_path(path); | |
4438 | return ret; | |
4439 | } | |
4440 | WARN_ON(!ret); | |
4441 | path->slots[0]--; | |
975f84fe | 4442 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
962a298f | 4443 | found_type = found_key.type; |
975f84fe | 4444 | |
ec29ed5b | 4445 | /* No extents, but there might be delalloc bits */ |
33345d01 | 4446 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 4447 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
4448 | /* have to trust i_size as the end */ |
4449 | last = (u64)-1; | |
4450 | last_for_get_extent = isize; | |
4451 | } else { | |
4452 | /* | |
4453 | * remember the start of the last extent. There are a | |
4454 | * bunch of different factors that go into the length of the | |
4455 | * extent, so its much less complex to remember where it started | |
4456 | */ | |
4457 | last = found_key.offset; | |
4458 | last_for_get_extent = last + 1; | |
975f84fe | 4459 | } |
fe09e16c | 4460 | btrfs_release_path(path); |
975f84fe | 4461 | |
ec29ed5b CM |
4462 | /* |
4463 | * we might have some extents allocated but more delalloc past those | |
4464 | * extents. so, we trust isize unless the start of the last extent is | |
4465 | * beyond isize | |
4466 | */ | |
4467 | if (last < isize) { | |
4468 | last = (u64)-1; | |
4469 | last_for_get_extent = isize; | |
4470 | } | |
4471 | ||
a52f4cd2 | 4472 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0, |
d0082371 | 4473 | &cached_state); |
ec29ed5b | 4474 | |
4d479cf0 | 4475 | em = get_extent_skip_holes(inode, start, last_for_get_extent, |
ec29ed5b | 4476 | get_extent); |
1506fcc8 YS |
4477 | if (!em) |
4478 | goto out; | |
4479 | if (IS_ERR(em)) { | |
4480 | ret = PTR_ERR(em); | |
4481 | goto out; | |
4482 | } | |
975f84fe | 4483 | |
1506fcc8 | 4484 | while (!end) { |
b76bb701 | 4485 | u64 offset_in_extent = 0; |
ea8efc74 CM |
4486 | |
4487 | /* break if the extent we found is outside the range */ | |
4488 | if (em->start >= max || extent_map_end(em) < off) | |
4489 | break; | |
4490 | ||
4491 | /* | |
4492 | * get_extent may return an extent that starts before our | |
4493 | * requested range. We have to make sure the ranges | |
4494 | * we return to fiemap always move forward and don't | |
4495 | * overlap, so adjust the offsets here | |
4496 | */ | |
4497 | em_start = max(em->start, off); | |
1506fcc8 | 4498 | |
ea8efc74 CM |
4499 | /* |
4500 | * record the offset from the start of the extent | |
b76bb701 JB |
4501 | * for adjusting the disk offset below. Only do this if the |
4502 | * extent isn't compressed since our in ram offset may be past | |
4503 | * what we have actually allocated on disk. | |
ea8efc74 | 4504 | */ |
b76bb701 JB |
4505 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
4506 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 4507 | em_end = extent_map_end(em); |
ea8efc74 | 4508 | em_len = em_end - em_start; |
1506fcc8 YS |
4509 | disko = 0; |
4510 | flags = 0; | |
4511 | ||
ea8efc74 CM |
4512 | /* |
4513 | * bump off for our next call to get_extent | |
4514 | */ | |
4515 | off = extent_map_end(em); | |
4516 | if (off >= max) | |
4517 | end = 1; | |
4518 | ||
93dbfad7 | 4519 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
4520 | end = 1; |
4521 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 4522 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
4523 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
4524 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 4525 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
4526 | flags |= (FIEMAP_EXTENT_DELALLOC | |
4527 | FIEMAP_EXTENT_UNKNOWN); | |
dc046b10 JB |
4528 | } else if (fieinfo->fi_extents_max) { |
4529 | u64 bytenr = em->block_start - | |
4530 | (em->start - em->orig_start); | |
fe09e16c | 4531 | |
ea8efc74 | 4532 | disko = em->block_start + offset_in_extent; |
fe09e16c LB |
4533 | |
4534 | /* | |
4535 | * As btrfs supports shared space, this information | |
4536 | * can be exported to userspace tools via | |
dc046b10 JB |
4537 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 |
4538 | * then we're just getting a count and we can skip the | |
4539 | * lookup stuff. | |
fe09e16c | 4540 | */ |
dc046b10 JB |
4541 | ret = btrfs_check_shared(NULL, root->fs_info, |
4542 | root->objectid, | |
4543 | btrfs_ino(inode), bytenr); | |
4544 | if (ret < 0) | |
fe09e16c | 4545 | goto out_free; |
dc046b10 | 4546 | if (ret) |
fe09e16c | 4547 | flags |= FIEMAP_EXTENT_SHARED; |
dc046b10 | 4548 | ret = 0; |
1506fcc8 YS |
4549 | } |
4550 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
4551 | flags |= FIEMAP_EXTENT_ENCODED; | |
0d2b2372 JB |
4552 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
4553 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
1506fcc8 | 4554 | |
1506fcc8 YS |
4555 | free_extent_map(em); |
4556 | em = NULL; | |
ec29ed5b CM |
4557 | if ((em_start >= last) || em_len == (u64)-1 || |
4558 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
4559 | flags |= FIEMAP_EXTENT_LAST; |
4560 | end = 1; | |
4561 | } | |
4562 | ||
ec29ed5b CM |
4563 | /* now scan forward to see if this is really the last extent. */ |
4564 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
4565 | get_extent); | |
4566 | if (IS_ERR(em)) { | |
4567 | ret = PTR_ERR(em); | |
4568 | goto out; | |
4569 | } | |
4570 | if (!em) { | |
975f84fe JB |
4571 | flags |= FIEMAP_EXTENT_LAST; |
4572 | end = 1; | |
4573 | } | |
ec29ed5b CM |
4574 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, |
4575 | em_len, flags); | |
26e726af CS |
4576 | if (ret) { |
4577 | if (ret == 1) | |
4578 | ret = 0; | |
ec29ed5b | 4579 | goto out_free; |
26e726af | 4580 | } |
1506fcc8 YS |
4581 | } |
4582 | out_free: | |
4583 | free_extent_map(em); | |
4584 | out: | |
fe09e16c | 4585 | btrfs_free_path(path); |
a52f4cd2 | 4586 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
2ac55d41 | 4587 | &cached_state, GFP_NOFS); |
1506fcc8 YS |
4588 | return ret; |
4589 | } | |
4590 | ||
727011e0 CM |
4591 | static void __free_extent_buffer(struct extent_buffer *eb) |
4592 | { | |
6d49ba1b | 4593 | btrfs_leak_debug_del(&eb->leak_list); |
727011e0 CM |
4594 | kmem_cache_free(extent_buffer_cache, eb); |
4595 | } | |
4596 | ||
a26e8c9f | 4597 | int extent_buffer_under_io(struct extent_buffer *eb) |
db7f3436 JB |
4598 | { |
4599 | return (atomic_read(&eb->io_pages) || | |
4600 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4601 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4602 | } | |
4603 | ||
4604 | /* | |
4605 | * Helper for releasing extent buffer page. | |
4606 | */ | |
a50924e3 | 4607 | static void btrfs_release_extent_buffer_page(struct extent_buffer *eb) |
db7f3436 JB |
4608 | { |
4609 | unsigned long index; | |
db7f3436 JB |
4610 | struct page *page; |
4611 | int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4612 | ||
4613 | BUG_ON(extent_buffer_under_io(eb)); | |
4614 | ||
a50924e3 DS |
4615 | index = num_extent_pages(eb->start, eb->len); |
4616 | if (index == 0) | |
db7f3436 JB |
4617 | return; |
4618 | ||
4619 | do { | |
4620 | index--; | |
fb85fc9a | 4621 | page = eb->pages[index]; |
5d2361db FL |
4622 | if (!page) |
4623 | continue; | |
4624 | if (mapped) | |
db7f3436 | 4625 | spin_lock(&page->mapping->private_lock); |
5d2361db FL |
4626 | /* |
4627 | * We do this since we'll remove the pages after we've | |
4628 | * removed the eb from the radix tree, so we could race | |
4629 | * and have this page now attached to the new eb. So | |
4630 | * only clear page_private if it's still connected to | |
4631 | * this eb. | |
4632 | */ | |
4633 | if (PagePrivate(page) && | |
4634 | page->private == (unsigned long)eb) { | |
4635 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4636 | BUG_ON(PageDirty(page)); | |
4637 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 4638 | /* |
5d2361db FL |
4639 | * We need to make sure we haven't be attached |
4640 | * to a new eb. | |
db7f3436 | 4641 | */ |
5d2361db FL |
4642 | ClearPagePrivate(page); |
4643 | set_page_private(page, 0); | |
4644 | /* One for the page private */ | |
db7f3436 JB |
4645 | page_cache_release(page); |
4646 | } | |
5d2361db FL |
4647 | |
4648 | if (mapped) | |
4649 | spin_unlock(&page->mapping->private_lock); | |
4650 | ||
4651 | /* One for when we alloced the page */ | |
4652 | page_cache_release(page); | |
a50924e3 | 4653 | } while (index != 0); |
db7f3436 JB |
4654 | } |
4655 | ||
4656 | /* | |
4657 | * Helper for releasing the extent buffer. | |
4658 | */ | |
4659 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4660 | { | |
a50924e3 | 4661 | btrfs_release_extent_buffer_page(eb); |
db7f3436 JB |
4662 | __free_extent_buffer(eb); |
4663 | } | |
4664 | ||
f28491e0 JB |
4665 | static struct extent_buffer * |
4666 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 4667 | unsigned long len) |
d1310b2e CM |
4668 | { |
4669 | struct extent_buffer *eb = NULL; | |
4670 | ||
d1b5c567 | 4671 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
4672 | eb->start = start; |
4673 | eb->len = len; | |
f28491e0 | 4674 | eb->fs_info = fs_info; |
815a51c7 | 4675 | eb->bflags = 0; |
bd681513 CM |
4676 | rwlock_init(&eb->lock); |
4677 | atomic_set(&eb->write_locks, 0); | |
4678 | atomic_set(&eb->read_locks, 0); | |
4679 | atomic_set(&eb->blocking_readers, 0); | |
4680 | atomic_set(&eb->blocking_writers, 0); | |
4681 | atomic_set(&eb->spinning_readers, 0); | |
4682 | atomic_set(&eb->spinning_writers, 0); | |
5b25f70f | 4683 | eb->lock_nested = 0; |
bd681513 CM |
4684 | init_waitqueue_head(&eb->write_lock_wq); |
4685 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 4686 | |
6d49ba1b ES |
4687 | btrfs_leak_debug_add(&eb->leak_list, &buffers); |
4688 | ||
3083ee2e | 4689 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 4690 | atomic_set(&eb->refs, 1); |
0b32f4bb | 4691 | atomic_set(&eb->io_pages, 0); |
727011e0 | 4692 | |
b8dae313 DS |
4693 | /* |
4694 | * Sanity checks, currently the maximum is 64k covered by 16x 4k pages | |
4695 | */ | |
4696 | BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE | |
4697 | > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
4698 | BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
d1310b2e CM |
4699 | |
4700 | return eb; | |
4701 | } | |
4702 | ||
815a51c7 JS |
4703 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) |
4704 | { | |
4705 | unsigned long i; | |
4706 | struct page *p; | |
4707 | struct extent_buffer *new; | |
4708 | unsigned long num_pages = num_extent_pages(src->start, src->len); | |
4709 | ||
3f556f78 | 4710 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
4711 | if (new == NULL) |
4712 | return NULL; | |
4713 | ||
4714 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4715 | p = alloc_page(GFP_NOFS); |
db7f3436 JB |
4716 | if (!p) { |
4717 | btrfs_release_extent_buffer(new); | |
4718 | return NULL; | |
4719 | } | |
815a51c7 JS |
4720 | attach_extent_buffer_page(new, p); |
4721 | WARN_ON(PageDirty(p)); | |
4722 | SetPageUptodate(p); | |
4723 | new->pages[i] = p; | |
4724 | } | |
4725 | ||
4726 | copy_extent_buffer(new, src, 0, 0, src->len); | |
4727 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); | |
4728 | set_bit(EXTENT_BUFFER_DUMMY, &new->bflags); | |
4729 | ||
4730 | return new; | |
4731 | } | |
4732 | ||
3f556f78 DS |
4733 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
4734 | u64 start) | |
815a51c7 JS |
4735 | { |
4736 | struct extent_buffer *eb; | |
3f556f78 DS |
4737 | unsigned long len; |
4738 | unsigned long num_pages; | |
815a51c7 JS |
4739 | unsigned long i; |
4740 | ||
3f556f78 DS |
4741 | if (!fs_info) { |
4742 | /* | |
4743 | * Called only from tests that don't always have a fs_info | |
4744 | * available, but we know that nodesize is 4096 | |
4745 | */ | |
4746 | len = 4096; | |
4747 | } else { | |
4748 | len = fs_info->tree_root->nodesize; | |
4749 | } | |
4750 | num_pages = num_extent_pages(0, len); | |
4751 | ||
4752 | eb = __alloc_extent_buffer(fs_info, start, len); | |
815a51c7 JS |
4753 | if (!eb) |
4754 | return NULL; | |
4755 | ||
4756 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4757 | eb->pages[i] = alloc_page(GFP_NOFS); |
815a51c7 JS |
4758 | if (!eb->pages[i]) |
4759 | goto err; | |
4760 | } | |
4761 | set_extent_buffer_uptodate(eb); | |
4762 | btrfs_set_header_nritems(eb, 0); | |
4763 | set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4764 | ||
4765 | return eb; | |
4766 | err: | |
84167d19 SB |
4767 | for (; i > 0; i--) |
4768 | __free_page(eb->pages[i - 1]); | |
815a51c7 JS |
4769 | __free_extent_buffer(eb); |
4770 | return NULL; | |
4771 | } | |
4772 | ||
0b32f4bb JB |
4773 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
4774 | { | |
242e18c7 | 4775 | int refs; |
0b32f4bb JB |
4776 | /* the ref bit is tricky. We have to make sure it is set |
4777 | * if we have the buffer dirty. Otherwise the | |
4778 | * code to free a buffer can end up dropping a dirty | |
4779 | * page | |
4780 | * | |
4781 | * Once the ref bit is set, it won't go away while the | |
4782 | * buffer is dirty or in writeback, and it also won't | |
4783 | * go away while we have the reference count on the | |
4784 | * eb bumped. | |
4785 | * | |
4786 | * We can't just set the ref bit without bumping the | |
4787 | * ref on the eb because free_extent_buffer might | |
4788 | * see the ref bit and try to clear it. If this happens | |
4789 | * free_extent_buffer might end up dropping our original | |
4790 | * ref by mistake and freeing the page before we are able | |
4791 | * to add one more ref. | |
4792 | * | |
4793 | * So bump the ref count first, then set the bit. If someone | |
4794 | * beat us to it, drop the ref we added. | |
4795 | */ | |
242e18c7 CM |
4796 | refs = atomic_read(&eb->refs); |
4797 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
4798 | return; | |
4799 | ||
594831c4 JB |
4800 | spin_lock(&eb->refs_lock); |
4801 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 4802 | atomic_inc(&eb->refs); |
594831c4 | 4803 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
4804 | } |
4805 | ||
2457aec6 MG |
4806 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
4807 | struct page *accessed) | |
5df4235e JB |
4808 | { |
4809 | unsigned long num_pages, i; | |
4810 | ||
0b32f4bb JB |
4811 | check_buffer_tree_ref(eb); |
4812 | ||
5df4235e JB |
4813 | num_pages = num_extent_pages(eb->start, eb->len); |
4814 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a DS |
4815 | struct page *p = eb->pages[i]; |
4816 | ||
2457aec6 MG |
4817 | if (p != accessed) |
4818 | mark_page_accessed(p); | |
5df4235e JB |
4819 | } |
4820 | } | |
4821 | ||
f28491e0 JB |
4822 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
4823 | u64 start) | |
452c75c3 CS |
4824 | { |
4825 | struct extent_buffer *eb; | |
4826 | ||
4827 | rcu_read_lock(); | |
f28491e0 JB |
4828 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
4829 | start >> PAGE_CACHE_SHIFT); | |
452c75c3 CS |
4830 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
4831 | rcu_read_unlock(); | |
062c19e9 FM |
4832 | /* |
4833 | * Lock our eb's refs_lock to avoid races with | |
4834 | * free_extent_buffer. When we get our eb it might be flagged | |
4835 | * with EXTENT_BUFFER_STALE and another task running | |
4836 | * free_extent_buffer might have seen that flag set, | |
4837 | * eb->refs == 2, that the buffer isn't under IO (dirty and | |
4838 | * writeback flags not set) and it's still in the tree (flag | |
4839 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process | |
4840 | * of decrementing the extent buffer's reference count twice. | |
4841 | * So here we could race and increment the eb's reference count, | |
4842 | * clear its stale flag, mark it as dirty and drop our reference | |
4843 | * before the other task finishes executing free_extent_buffer, | |
4844 | * which would later result in an attempt to free an extent | |
4845 | * buffer that is dirty. | |
4846 | */ | |
4847 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
4848 | spin_lock(&eb->refs_lock); | |
4849 | spin_unlock(&eb->refs_lock); | |
4850 | } | |
2457aec6 | 4851 | mark_extent_buffer_accessed(eb, NULL); |
452c75c3 CS |
4852 | return eb; |
4853 | } | |
4854 | rcu_read_unlock(); | |
4855 | ||
4856 | return NULL; | |
4857 | } | |
4858 | ||
faa2dbf0 JB |
4859 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
4860 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
ce3e6984 | 4861 | u64 start) |
faa2dbf0 JB |
4862 | { |
4863 | struct extent_buffer *eb, *exists = NULL; | |
4864 | int ret; | |
4865 | ||
4866 | eb = find_extent_buffer(fs_info, start); | |
4867 | if (eb) | |
4868 | return eb; | |
3f556f78 | 4869 | eb = alloc_dummy_extent_buffer(fs_info, start); |
faa2dbf0 JB |
4870 | if (!eb) |
4871 | return NULL; | |
4872 | eb->fs_info = fs_info; | |
4873 | again: | |
4874 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); | |
4875 | if (ret) | |
4876 | goto free_eb; | |
4877 | spin_lock(&fs_info->buffer_lock); | |
4878 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
4879 | start >> PAGE_CACHE_SHIFT, eb); | |
4880 | spin_unlock(&fs_info->buffer_lock); | |
4881 | radix_tree_preload_end(); | |
4882 | if (ret == -EEXIST) { | |
4883 | exists = find_extent_buffer(fs_info, start); | |
4884 | if (exists) | |
4885 | goto free_eb; | |
4886 | else | |
4887 | goto again; | |
4888 | } | |
4889 | check_buffer_tree_ref(eb); | |
4890 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
4891 | ||
4892 | /* | |
4893 | * We will free dummy extent buffer's if they come into | |
4894 | * free_extent_buffer with a ref count of 2, but if we are using this we | |
4895 | * want the buffers to stay in memory until we're done with them, so | |
4896 | * bump the ref count again. | |
4897 | */ | |
4898 | atomic_inc(&eb->refs); | |
4899 | return eb; | |
4900 | free_eb: | |
4901 | btrfs_release_extent_buffer(eb); | |
4902 | return exists; | |
4903 | } | |
4904 | #endif | |
4905 | ||
f28491e0 | 4906 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
ce3e6984 | 4907 | u64 start) |
d1310b2e | 4908 | { |
ce3e6984 | 4909 | unsigned long len = fs_info->tree_root->nodesize; |
d1310b2e CM |
4910 | unsigned long num_pages = num_extent_pages(start, len); |
4911 | unsigned long i; | |
4912 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
4913 | struct extent_buffer *eb; | |
6af118ce | 4914 | struct extent_buffer *exists = NULL; |
d1310b2e | 4915 | struct page *p; |
f28491e0 | 4916 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
d1310b2e | 4917 | int uptodate = 1; |
19fe0a8b | 4918 | int ret; |
d1310b2e | 4919 | |
f28491e0 | 4920 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 4921 | if (eb) |
6af118ce | 4922 | return eb; |
6af118ce | 4923 | |
23d79d81 | 4924 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 4925 | if (!eb) |
d1310b2e CM |
4926 | return NULL; |
4927 | ||
727011e0 | 4928 | for (i = 0; i < num_pages; i++, index++) { |
d1b5c567 | 4929 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
4804b382 | 4930 | if (!p) |
6af118ce | 4931 | goto free_eb; |
4f2de97a JB |
4932 | |
4933 | spin_lock(&mapping->private_lock); | |
4934 | if (PagePrivate(p)) { | |
4935 | /* | |
4936 | * We could have already allocated an eb for this page | |
4937 | * and attached one so lets see if we can get a ref on | |
4938 | * the existing eb, and if we can we know it's good and | |
4939 | * we can just return that one, else we know we can just | |
4940 | * overwrite page->private. | |
4941 | */ | |
4942 | exists = (struct extent_buffer *)p->private; | |
4943 | if (atomic_inc_not_zero(&exists->refs)) { | |
4944 | spin_unlock(&mapping->private_lock); | |
4945 | unlock_page(p); | |
17de39ac | 4946 | page_cache_release(p); |
2457aec6 | 4947 | mark_extent_buffer_accessed(exists, p); |
4f2de97a JB |
4948 | goto free_eb; |
4949 | } | |
5ca64f45 | 4950 | exists = NULL; |
4f2de97a | 4951 | |
0b32f4bb | 4952 | /* |
4f2de97a JB |
4953 | * Do this so attach doesn't complain and we need to |
4954 | * drop the ref the old guy had. | |
4955 | */ | |
4956 | ClearPagePrivate(p); | |
0b32f4bb | 4957 | WARN_ON(PageDirty(p)); |
4f2de97a | 4958 | page_cache_release(p); |
d1310b2e | 4959 | } |
4f2de97a JB |
4960 | attach_extent_buffer_page(eb, p); |
4961 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 4962 | WARN_ON(PageDirty(p)); |
727011e0 | 4963 | eb->pages[i] = p; |
d1310b2e CM |
4964 | if (!PageUptodate(p)) |
4965 | uptodate = 0; | |
eb14ab8e CM |
4966 | |
4967 | /* | |
4968 | * see below about how we avoid a nasty race with release page | |
4969 | * and why we unlock later | |
4970 | */ | |
d1310b2e CM |
4971 | } |
4972 | if (uptodate) | |
b4ce94de | 4973 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 4974 | again: |
19fe0a8b MX |
4975 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); |
4976 | if (ret) | |
4977 | goto free_eb; | |
4978 | ||
f28491e0 JB |
4979 | spin_lock(&fs_info->buffer_lock); |
4980 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
4981 | start >> PAGE_CACHE_SHIFT, eb); | |
4982 | spin_unlock(&fs_info->buffer_lock); | |
452c75c3 | 4983 | radix_tree_preload_end(); |
19fe0a8b | 4984 | if (ret == -EEXIST) { |
f28491e0 | 4985 | exists = find_extent_buffer(fs_info, start); |
452c75c3 CS |
4986 | if (exists) |
4987 | goto free_eb; | |
4988 | else | |
115391d2 | 4989 | goto again; |
6af118ce | 4990 | } |
6af118ce | 4991 | /* add one reference for the tree */ |
0b32f4bb | 4992 | check_buffer_tree_ref(eb); |
34b41ace | 4993 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
4994 | |
4995 | /* | |
4996 | * there is a race where release page may have | |
4997 | * tried to find this extent buffer in the radix | |
4998 | * but failed. It will tell the VM it is safe to | |
4999 | * reclaim the, and it will clear the page private bit. | |
5000 | * We must make sure to set the page private bit properly | |
5001 | * after the extent buffer is in the radix tree so | |
5002 | * it doesn't get lost | |
5003 | */ | |
727011e0 CM |
5004 | SetPageChecked(eb->pages[0]); |
5005 | for (i = 1; i < num_pages; i++) { | |
fb85fc9a | 5006 | p = eb->pages[i]; |
727011e0 CM |
5007 | ClearPageChecked(p); |
5008 | unlock_page(p); | |
5009 | } | |
5010 | unlock_page(eb->pages[0]); | |
d1310b2e CM |
5011 | return eb; |
5012 | ||
6af118ce | 5013 | free_eb: |
5ca64f45 | 5014 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
5015 | for (i = 0; i < num_pages; i++) { |
5016 | if (eb->pages[i]) | |
5017 | unlock_page(eb->pages[i]); | |
5018 | } | |
eb14ab8e | 5019 | |
897ca6e9 | 5020 | btrfs_release_extent_buffer(eb); |
6af118ce | 5021 | return exists; |
d1310b2e | 5022 | } |
d1310b2e | 5023 | |
3083ee2e JB |
5024 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
5025 | { | |
5026 | struct extent_buffer *eb = | |
5027 | container_of(head, struct extent_buffer, rcu_head); | |
5028 | ||
5029 | __free_extent_buffer(eb); | |
5030 | } | |
5031 | ||
3083ee2e | 5032 | /* Expects to have eb->eb_lock already held */ |
f7a52a40 | 5033 | static int release_extent_buffer(struct extent_buffer *eb) |
3083ee2e JB |
5034 | { |
5035 | WARN_ON(atomic_read(&eb->refs) == 0); | |
5036 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 5037 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 5038 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 5039 | |
815a51c7 | 5040 | spin_unlock(&eb->refs_lock); |
3083ee2e | 5041 | |
f28491e0 JB |
5042 | spin_lock(&fs_info->buffer_lock); |
5043 | radix_tree_delete(&fs_info->buffer_radix, | |
815a51c7 | 5044 | eb->start >> PAGE_CACHE_SHIFT); |
f28491e0 | 5045 | spin_unlock(&fs_info->buffer_lock); |
34b41ace JB |
5046 | } else { |
5047 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 5048 | } |
3083ee2e JB |
5049 | |
5050 | /* Should be safe to release our pages at this point */ | |
a50924e3 | 5051 | btrfs_release_extent_buffer_page(eb); |
bcb7e449 JB |
5052 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
5053 | if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) { | |
5054 | __free_extent_buffer(eb); | |
5055 | return 1; | |
5056 | } | |
5057 | #endif | |
3083ee2e | 5058 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 5059 | return 1; |
3083ee2e JB |
5060 | } |
5061 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
5062 | |
5063 | return 0; | |
3083ee2e JB |
5064 | } |
5065 | ||
d1310b2e CM |
5066 | void free_extent_buffer(struct extent_buffer *eb) |
5067 | { | |
242e18c7 CM |
5068 | int refs; |
5069 | int old; | |
d1310b2e CM |
5070 | if (!eb) |
5071 | return; | |
5072 | ||
242e18c7 CM |
5073 | while (1) { |
5074 | refs = atomic_read(&eb->refs); | |
5075 | if (refs <= 3) | |
5076 | break; | |
5077 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
5078 | if (old == refs) | |
5079 | return; | |
5080 | } | |
5081 | ||
3083ee2e | 5082 | spin_lock(&eb->refs_lock); |
815a51c7 JS |
5083 | if (atomic_read(&eb->refs) == 2 && |
5084 | test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) | |
5085 | atomic_dec(&eb->refs); | |
5086 | ||
3083ee2e JB |
5087 | if (atomic_read(&eb->refs) == 2 && |
5088 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 5089 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
5090 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5091 | atomic_dec(&eb->refs); | |
5092 | ||
5093 | /* | |
5094 | * I know this is terrible, but it's temporary until we stop tracking | |
5095 | * the uptodate bits and such for the extent buffers. | |
5096 | */ | |
f7a52a40 | 5097 | release_extent_buffer(eb); |
3083ee2e JB |
5098 | } |
5099 | ||
5100 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
5101 | { | |
5102 | if (!eb) | |
d1310b2e CM |
5103 | return; |
5104 | ||
3083ee2e JB |
5105 | spin_lock(&eb->refs_lock); |
5106 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
5107 | ||
0b32f4bb | 5108 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
5109 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5110 | atomic_dec(&eb->refs); | |
f7a52a40 | 5111 | release_extent_buffer(eb); |
d1310b2e | 5112 | } |
d1310b2e | 5113 | |
1d4284bd | 5114 | void clear_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5115 | { |
d1310b2e CM |
5116 | unsigned long i; |
5117 | unsigned long num_pages; | |
5118 | struct page *page; | |
5119 | ||
d1310b2e CM |
5120 | num_pages = num_extent_pages(eb->start, eb->len); |
5121 | ||
5122 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 5123 | page = eb->pages[i]; |
b9473439 | 5124 | if (!PageDirty(page)) |
d2c3f4f6 CM |
5125 | continue; |
5126 | ||
a61e6f29 | 5127 | lock_page(page); |
eb14ab8e CM |
5128 | WARN_ON(!PagePrivate(page)); |
5129 | ||
d1310b2e | 5130 | clear_page_dirty_for_io(page); |
0ee0fda0 | 5131 | spin_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
5132 | if (!PageDirty(page)) { |
5133 | radix_tree_tag_clear(&page->mapping->page_tree, | |
5134 | page_index(page), | |
5135 | PAGECACHE_TAG_DIRTY); | |
5136 | } | |
0ee0fda0 | 5137 | spin_unlock_irq(&page->mapping->tree_lock); |
bf0da8c1 | 5138 | ClearPageError(page); |
a61e6f29 | 5139 | unlock_page(page); |
d1310b2e | 5140 | } |
0b32f4bb | 5141 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 5142 | } |
d1310b2e | 5143 | |
0b32f4bb | 5144 | int set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e CM |
5145 | { |
5146 | unsigned long i; | |
5147 | unsigned long num_pages; | |
b9473439 | 5148 | int was_dirty = 0; |
d1310b2e | 5149 | |
0b32f4bb JB |
5150 | check_buffer_tree_ref(eb); |
5151 | ||
b9473439 | 5152 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 5153 | |
d1310b2e | 5154 | num_pages = num_extent_pages(eb->start, eb->len); |
3083ee2e | 5155 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
5156 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
5157 | ||
b9473439 | 5158 | for (i = 0; i < num_pages; i++) |
fb85fc9a | 5159 | set_page_dirty(eb->pages[i]); |
b9473439 | 5160 | return was_dirty; |
d1310b2e | 5161 | } |
d1310b2e | 5162 | |
0b32f4bb | 5163 | int clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 CM |
5164 | { |
5165 | unsigned long i; | |
5166 | struct page *page; | |
5167 | unsigned long num_pages; | |
5168 | ||
b4ce94de | 5169 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
0b32f4bb | 5170 | num_pages = num_extent_pages(eb->start, eb->len); |
1259ab75 | 5171 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5172 | page = eb->pages[i]; |
33958dc6 CM |
5173 | if (page) |
5174 | ClearPageUptodate(page); | |
1259ab75 CM |
5175 | } |
5176 | return 0; | |
5177 | } | |
5178 | ||
0b32f4bb | 5179 | int set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e CM |
5180 | { |
5181 | unsigned long i; | |
5182 | struct page *page; | |
5183 | unsigned long num_pages; | |
5184 | ||
0b32f4bb | 5185 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 5186 | num_pages = num_extent_pages(eb->start, eb->len); |
d1310b2e | 5187 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5188 | page = eb->pages[i]; |
d1310b2e CM |
5189 | SetPageUptodate(page); |
5190 | } | |
5191 | return 0; | |
5192 | } | |
d1310b2e | 5193 | |
0b32f4bb | 5194 | int extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 5195 | { |
0b32f4bb | 5196 | return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 5197 | } |
d1310b2e CM |
5198 | |
5199 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
bb82ab88 | 5200 | struct extent_buffer *eb, u64 start, int wait, |
f188591e | 5201 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
5202 | { |
5203 | unsigned long i; | |
5204 | unsigned long start_i; | |
5205 | struct page *page; | |
5206 | int err; | |
5207 | int ret = 0; | |
ce9adaa5 CM |
5208 | int locked_pages = 0; |
5209 | int all_uptodate = 1; | |
d1310b2e | 5210 | unsigned long num_pages; |
727011e0 | 5211 | unsigned long num_reads = 0; |
a86c12c7 | 5212 | struct bio *bio = NULL; |
c8b97818 | 5213 | unsigned long bio_flags = 0; |
a86c12c7 | 5214 | |
b4ce94de | 5215 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
5216 | return 0; |
5217 | ||
d1310b2e CM |
5218 | if (start) { |
5219 | WARN_ON(start < eb->start); | |
5220 | start_i = (start >> PAGE_CACHE_SHIFT) - | |
5221 | (eb->start >> PAGE_CACHE_SHIFT); | |
5222 | } else { | |
5223 | start_i = 0; | |
5224 | } | |
5225 | ||
5226 | num_pages = num_extent_pages(eb->start, eb->len); | |
5227 | for (i = start_i; i < num_pages; i++) { | |
fb85fc9a | 5228 | page = eb->pages[i]; |
bb82ab88 | 5229 | if (wait == WAIT_NONE) { |
2db04966 | 5230 | if (!trylock_page(page)) |
ce9adaa5 | 5231 | goto unlock_exit; |
d1310b2e CM |
5232 | } else { |
5233 | lock_page(page); | |
5234 | } | |
ce9adaa5 | 5235 | locked_pages++; |
727011e0 CM |
5236 | if (!PageUptodate(page)) { |
5237 | num_reads++; | |
ce9adaa5 | 5238 | all_uptodate = 0; |
727011e0 | 5239 | } |
ce9adaa5 CM |
5240 | } |
5241 | if (all_uptodate) { | |
5242 | if (start_i == 0) | |
b4ce94de | 5243 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
5244 | goto unlock_exit; |
5245 | } | |
5246 | ||
656f30db | 5247 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 5248 | eb->read_mirror = 0; |
0b32f4bb | 5249 | atomic_set(&eb->io_pages, num_reads); |
ce9adaa5 | 5250 | for (i = start_i; i < num_pages; i++) { |
fb85fc9a | 5251 | page = eb->pages[i]; |
ce9adaa5 | 5252 | if (!PageUptodate(page)) { |
f188591e | 5253 | ClearPageError(page); |
a86c12c7 | 5254 | err = __extent_read_full_page(tree, page, |
f188591e | 5255 | get_extent, &bio, |
d4c7ca86 JB |
5256 | mirror_num, &bio_flags, |
5257 | READ | REQ_META); | |
d397712b | 5258 | if (err) |
d1310b2e | 5259 | ret = err; |
d1310b2e CM |
5260 | } else { |
5261 | unlock_page(page); | |
5262 | } | |
5263 | } | |
5264 | ||
355808c2 | 5265 | if (bio) { |
d4c7ca86 JB |
5266 | err = submit_one_bio(READ | REQ_META, bio, mirror_num, |
5267 | bio_flags); | |
79787eaa JM |
5268 | if (err) |
5269 | return err; | |
355808c2 | 5270 | } |
a86c12c7 | 5271 | |
bb82ab88 | 5272 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 5273 | return ret; |
d397712b | 5274 | |
d1310b2e | 5275 | for (i = start_i; i < num_pages; i++) { |
fb85fc9a | 5276 | page = eb->pages[i]; |
d1310b2e | 5277 | wait_on_page_locked(page); |
d397712b | 5278 | if (!PageUptodate(page)) |
d1310b2e | 5279 | ret = -EIO; |
d1310b2e | 5280 | } |
d397712b | 5281 | |
d1310b2e | 5282 | return ret; |
ce9adaa5 CM |
5283 | |
5284 | unlock_exit: | |
5285 | i = start_i; | |
d397712b | 5286 | while (locked_pages > 0) { |
fb85fc9a | 5287 | page = eb->pages[i]; |
ce9adaa5 CM |
5288 | i++; |
5289 | unlock_page(page); | |
5290 | locked_pages--; | |
5291 | } | |
5292 | return ret; | |
d1310b2e | 5293 | } |
d1310b2e CM |
5294 | |
5295 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
5296 | unsigned long start, | |
5297 | unsigned long len) | |
5298 | { | |
5299 | size_t cur; | |
5300 | size_t offset; | |
5301 | struct page *page; | |
5302 | char *kaddr; | |
5303 | char *dst = (char *)dstv; | |
5304 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5305 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
5306 | |
5307 | WARN_ON(start > eb->len); | |
5308 | WARN_ON(start + len > eb->start + eb->len); | |
5309 | ||
778746b5 | 5310 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5311 | |
d397712b | 5312 | while (len > 0) { |
fb85fc9a | 5313 | page = eb->pages[i]; |
d1310b2e CM |
5314 | |
5315 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
a6591715 | 5316 | kaddr = page_address(page); |
d1310b2e | 5317 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
5318 | |
5319 | dst += cur; | |
5320 | len -= cur; | |
5321 | offset = 0; | |
5322 | i++; | |
5323 | } | |
5324 | } | |
d1310b2e | 5325 | |
550ac1d8 GH |
5326 | int read_extent_buffer_to_user(struct extent_buffer *eb, void __user *dstv, |
5327 | unsigned long start, | |
5328 | unsigned long len) | |
5329 | { | |
5330 | size_t cur; | |
5331 | size_t offset; | |
5332 | struct page *page; | |
5333 | char *kaddr; | |
5334 | char __user *dst = (char __user *)dstv; | |
5335 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5336 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5337 | int ret = 0; | |
5338 | ||
5339 | WARN_ON(start > eb->len); | |
5340 | WARN_ON(start + len > eb->start + eb->len); | |
5341 | ||
5342 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); | |
5343 | ||
5344 | while (len > 0) { | |
fb85fc9a | 5345 | page = eb->pages[i]; |
550ac1d8 GH |
5346 | |
5347 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
5348 | kaddr = page_address(page); | |
5349 | if (copy_to_user(dst, kaddr + offset, cur)) { | |
5350 | ret = -EFAULT; | |
5351 | break; | |
5352 | } | |
5353 | ||
5354 | dst += cur; | |
5355 | len -= cur; | |
5356 | offset = 0; | |
5357 | i++; | |
5358 | } | |
5359 | ||
5360 | return ret; | |
5361 | } | |
5362 | ||
d1310b2e | 5363 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, |
a6591715 | 5364 | unsigned long min_len, char **map, |
d1310b2e | 5365 | unsigned long *map_start, |
a6591715 | 5366 | unsigned long *map_len) |
d1310b2e CM |
5367 | { |
5368 | size_t offset = start & (PAGE_CACHE_SIZE - 1); | |
5369 | char *kaddr; | |
5370 | struct page *p; | |
5371 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5372 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5373 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
5374 | PAGE_CACHE_SHIFT; | |
5375 | ||
5376 | if (i != end_i) | |
5377 | return -EINVAL; | |
5378 | ||
5379 | if (i == 0) { | |
5380 | offset = start_offset; | |
5381 | *map_start = 0; | |
5382 | } else { | |
5383 | offset = 0; | |
5384 | *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; | |
5385 | } | |
d397712b | 5386 | |
d1310b2e | 5387 | if (start + min_len > eb->len) { |
31b1a2bd | 5388 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, " |
c1c9ff7c GU |
5389 | "wanted %lu %lu\n", |
5390 | eb->start, eb->len, start, min_len); | |
85026533 | 5391 | return -EINVAL; |
d1310b2e CM |
5392 | } |
5393 | ||
fb85fc9a | 5394 | p = eb->pages[i]; |
a6591715 | 5395 | kaddr = page_address(p); |
d1310b2e CM |
5396 | *map = kaddr + offset; |
5397 | *map_len = PAGE_CACHE_SIZE - offset; | |
5398 | return 0; | |
5399 | } | |
d1310b2e | 5400 | |
d1310b2e CM |
5401 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, |
5402 | unsigned long start, | |
5403 | unsigned long len) | |
5404 | { | |
5405 | size_t cur; | |
5406 | size_t offset; | |
5407 | struct page *page; | |
5408 | char *kaddr; | |
5409 | char *ptr = (char *)ptrv; | |
5410 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5411 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5412 | int ret = 0; | |
5413 | ||
5414 | WARN_ON(start > eb->len); | |
5415 | WARN_ON(start + len > eb->start + eb->len); | |
5416 | ||
778746b5 | 5417 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5418 | |
d397712b | 5419 | while (len > 0) { |
fb85fc9a | 5420 | page = eb->pages[i]; |
d1310b2e CM |
5421 | |
5422 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
5423 | ||
a6591715 | 5424 | kaddr = page_address(page); |
d1310b2e | 5425 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
5426 | if (ret) |
5427 | break; | |
5428 | ||
5429 | ptr += cur; | |
5430 | len -= cur; | |
5431 | offset = 0; | |
5432 | i++; | |
5433 | } | |
5434 | return ret; | |
5435 | } | |
d1310b2e CM |
5436 | |
5437 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
5438 | unsigned long start, unsigned long len) | |
5439 | { | |
5440 | size_t cur; | |
5441 | size_t offset; | |
5442 | struct page *page; | |
5443 | char *kaddr; | |
5444 | char *src = (char *)srcv; | |
5445 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5446 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5447 | ||
5448 | WARN_ON(start > eb->len); | |
5449 | WARN_ON(start + len > eb->start + eb->len); | |
5450 | ||
778746b5 | 5451 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5452 | |
d397712b | 5453 | while (len > 0) { |
fb85fc9a | 5454 | page = eb->pages[i]; |
d1310b2e CM |
5455 | WARN_ON(!PageUptodate(page)); |
5456 | ||
5457 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 5458 | kaddr = page_address(page); |
d1310b2e | 5459 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
5460 | |
5461 | src += cur; | |
5462 | len -= cur; | |
5463 | offset = 0; | |
5464 | i++; | |
5465 | } | |
5466 | } | |
d1310b2e CM |
5467 | |
5468 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
5469 | unsigned long start, unsigned long len) | |
5470 | { | |
5471 | size_t cur; | |
5472 | size_t offset; | |
5473 | struct page *page; | |
5474 | char *kaddr; | |
5475 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5476 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5477 | ||
5478 | WARN_ON(start > eb->len); | |
5479 | WARN_ON(start + len > eb->start + eb->len); | |
5480 | ||
778746b5 | 5481 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5482 | |
d397712b | 5483 | while (len > 0) { |
fb85fc9a | 5484 | page = eb->pages[i]; |
d1310b2e CM |
5485 | WARN_ON(!PageUptodate(page)); |
5486 | ||
5487 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 5488 | kaddr = page_address(page); |
d1310b2e | 5489 | memset(kaddr + offset, c, cur); |
d1310b2e CM |
5490 | |
5491 | len -= cur; | |
5492 | offset = 0; | |
5493 | i++; | |
5494 | } | |
5495 | } | |
d1310b2e CM |
5496 | |
5497 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
5498 | unsigned long dst_offset, unsigned long src_offset, | |
5499 | unsigned long len) | |
5500 | { | |
5501 | u64 dst_len = dst->len; | |
5502 | size_t cur; | |
5503 | size_t offset; | |
5504 | struct page *page; | |
5505 | char *kaddr; | |
5506 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5507 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
5508 | ||
5509 | WARN_ON(src->len != dst_len); | |
5510 | ||
5511 | offset = (start_offset + dst_offset) & | |
778746b5 | 5512 | (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5513 | |
d397712b | 5514 | while (len > 0) { |
fb85fc9a | 5515 | page = dst->pages[i]; |
d1310b2e CM |
5516 | WARN_ON(!PageUptodate(page)); |
5517 | ||
5518 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
5519 | ||
a6591715 | 5520 | kaddr = page_address(page); |
d1310b2e | 5521 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
5522 | |
5523 | src_offset += cur; | |
5524 | len -= cur; | |
5525 | offset = 0; | |
5526 | i++; | |
5527 | } | |
5528 | } | |
d1310b2e | 5529 | |
3387206f ST |
5530 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
5531 | { | |
5532 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
5533 | return distance < len; | |
5534 | } | |
5535 | ||
d1310b2e CM |
5536 | static void copy_pages(struct page *dst_page, struct page *src_page, |
5537 | unsigned long dst_off, unsigned long src_off, | |
5538 | unsigned long len) | |
5539 | { | |
a6591715 | 5540 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 5541 | char *src_kaddr; |
727011e0 | 5542 | int must_memmove = 0; |
d1310b2e | 5543 | |
3387206f | 5544 | if (dst_page != src_page) { |
a6591715 | 5545 | src_kaddr = page_address(src_page); |
3387206f | 5546 | } else { |
d1310b2e | 5547 | src_kaddr = dst_kaddr; |
727011e0 CM |
5548 | if (areas_overlap(src_off, dst_off, len)) |
5549 | must_memmove = 1; | |
3387206f | 5550 | } |
d1310b2e | 5551 | |
727011e0 CM |
5552 | if (must_memmove) |
5553 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
5554 | else | |
5555 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
5556 | } |
5557 | ||
5558 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5559 | unsigned long src_offset, unsigned long len) | |
5560 | { | |
5561 | size_t cur; | |
5562 | size_t dst_off_in_page; | |
5563 | size_t src_off_in_page; | |
5564 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5565 | unsigned long dst_i; | |
5566 | unsigned long src_i; | |
5567 | ||
5568 | if (src_offset + len > dst->len) { | |
efe120a0 | 5569 | printk(KERN_ERR "BTRFS: memmove bogus src_offset %lu move " |
d397712b | 5570 | "len %lu dst len %lu\n", src_offset, len, dst->len); |
d1310b2e CM |
5571 | BUG_ON(1); |
5572 | } | |
5573 | if (dst_offset + len > dst->len) { | |
efe120a0 | 5574 | printk(KERN_ERR "BTRFS: memmove bogus dst_offset %lu move " |
d397712b | 5575 | "len %lu dst len %lu\n", dst_offset, len, dst->len); |
d1310b2e CM |
5576 | BUG_ON(1); |
5577 | } | |
5578 | ||
d397712b | 5579 | while (len > 0) { |
d1310b2e | 5580 | dst_off_in_page = (start_offset + dst_offset) & |
778746b5 | 5581 | (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5582 | src_off_in_page = (start_offset + src_offset) & |
778746b5 | 5583 | (PAGE_CACHE_SIZE - 1); |
d1310b2e CM |
5584 | |
5585 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
5586 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
5587 | ||
5588 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
5589 | src_off_in_page)); | |
5590 | cur = min_t(unsigned long, cur, | |
5591 | (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); | |
5592 | ||
fb85fc9a | 5593 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5594 | dst_off_in_page, src_off_in_page, cur); |
5595 | ||
5596 | src_offset += cur; | |
5597 | dst_offset += cur; | |
5598 | len -= cur; | |
5599 | } | |
5600 | } | |
d1310b2e CM |
5601 | |
5602 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5603 | unsigned long src_offset, unsigned long len) | |
5604 | { | |
5605 | size_t cur; | |
5606 | size_t dst_off_in_page; | |
5607 | size_t src_off_in_page; | |
5608 | unsigned long dst_end = dst_offset + len - 1; | |
5609 | unsigned long src_end = src_offset + len - 1; | |
5610 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5611 | unsigned long dst_i; | |
5612 | unsigned long src_i; | |
5613 | ||
5614 | if (src_offset + len > dst->len) { | |
efe120a0 | 5615 | printk(KERN_ERR "BTRFS: memmove bogus src_offset %lu move " |
d397712b | 5616 | "len %lu len %lu\n", src_offset, len, dst->len); |
d1310b2e CM |
5617 | BUG_ON(1); |
5618 | } | |
5619 | if (dst_offset + len > dst->len) { | |
efe120a0 | 5620 | printk(KERN_ERR "BTRFS: memmove bogus dst_offset %lu move " |
d397712b | 5621 | "len %lu len %lu\n", dst_offset, len, dst->len); |
d1310b2e CM |
5622 | BUG_ON(1); |
5623 | } | |
727011e0 | 5624 | if (dst_offset < src_offset) { |
d1310b2e CM |
5625 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
5626 | return; | |
5627 | } | |
d397712b | 5628 | while (len > 0) { |
d1310b2e CM |
5629 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; |
5630 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
5631 | ||
5632 | dst_off_in_page = (start_offset + dst_end) & | |
778746b5 | 5633 | (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5634 | src_off_in_page = (start_offset + src_end) & |
778746b5 | 5635 | (PAGE_CACHE_SIZE - 1); |
d1310b2e CM |
5636 | |
5637 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
5638 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 5639 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5640 | dst_off_in_page - cur + 1, |
5641 | src_off_in_page - cur + 1, cur); | |
5642 | ||
5643 | dst_end -= cur; | |
5644 | src_end -= cur; | |
5645 | len -= cur; | |
5646 | } | |
5647 | } | |
6af118ce | 5648 | |
f7a52a40 | 5649 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 5650 | { |
6af118ce | 5651 | struct extent_buffer *eb; |
6af118ce | 5652 | |
3083ee2e JB |
5653 | /* |
5654 | * We need to make sure noboody is attaching this page to an eb right | |
5655 | * now. | |
5656 | */ | |
5657 | spin_lock(&page->mapping->private_lock); | |
5658 | if (!PagePrivate(page)) { | |
5659 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 5660 | return 1; |
45f49bce | 5661 | } |
6af118ce | 5662 | |
3083ee2e JB |
5663 | eb = (struct extent_buffer *)page->private; |
5664 | BUG_ON(!eb); | |
19fe0a8b MX |
5665 | |
5666 | /* | |
3083ee2e JB |
5667 | * This is a little awful but should be ok, we need to make sure that |
5668 | * the eb doesn't disappear out from under us while we're looking at | |
5669 | * this page. | |
19fe0a8b | 5670 | */ |
3083ee2e | 5671 | spin_lock(&eb->refs_lock); |
0b32f4bb | 5672 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
5673 | spin_unlock(&eb->refs_lock); |
5674 | spin_unlock(&page->mapping->private_lock); | |
5675 | return 0; | |
b9473439 | 5676 | } |
3083ee2e | 5677 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 5678 | |
19fe0a8b | 5679 | /* |
3083ee2e JB |
5680 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
5681 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 5682 | */ |
3083ee2e JB |
5683 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
5684 | spin_unlock(&eb->refs_lock); | |
5685 | return 0; | |
b9473439 | 5686 | } |
19fe0a8b | 5687 | |
f7a52a40 | 5688 | return release_extent_buffer(eb); |
6af118ce | 5689 | } |