]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
c1d7c514 | 2 | |
d1310b2e CM |
3 | #include <linux/bitops.h> |
4 | #include <linux/slab.h> | |
5 | #include <linux/bio.h> | |
6 | #include <linux/mm.h> | |
d1310b2e CM |
7 | #include <linux/pagemap.h> |
8 | #include <linux/page-flags.h> | |
d1310b2e CM |
9 | #include <linux/spinlock.h> |
10 | #include <linux/blkdev.h> | |
11 | #include <linux/swap.h> | |
d1310b2e CM |
12 | #include <linux/writeback.h> |
13 | #include <linux/pagevec.h> | |
268bb0ce | 14 | #include <linux/prefetch.h> |
90a887c9 | 15 | #include <linux/cleancache.h> |
d1310b2e CM |
16 | #include "extent_io.h" |
17 | #include "extent_map.h" | |
902b22f3 DW |
18 | #include "ctree.h" |
19 | #include "btrfs_inode.h" | |
4a54c8c1 | 20 | #include "volumes.h" |
21adbd5c | 21 | #include "check-integrity.h" |
0b32f4bb | 22 | #include "locking.h" |
606686ee | 23 | #include "rcu-string.h" |
fe09e16c | 24 | #include "backref.h" |
6af49dbd | 25 | #include "disk-io.h" |
d1310b2e | 26 | |
d1310b2e CM |
27 | static struct kmem_cache *extent_state_cache; |
28 | static struct kmem_cache *extent_buffer_cache; | |
8ac9f7c1 | 29 | static struct bio_set btrfs_bioset; |
d1310b2e | 30 | |
27a3507d FM |
31 | static inline bool extent_state_in_tree(const struct extent_state *state) |
32 | { | |
33 | return !RB_EMPTY_NODE(&state->rb_node); | |
34 | } | |
35 | ||
6d49ba1b | 36 | #ifdef CONFIG_BTRFS_DEBUG |
d1310b2e CM |
37 | static LIST_HEAD(buffers); |
38 | static LIST_HEAD(states); | |
4bef0848 | 39 | |
d397712b | 40 | static DEFINE_SPINLOCK(leak_lock); |
6d49ba1b ES |
41 | |
42 | static inline | |
43 | void btrfs_leak_debug_add(struct list_head *new, struct list_head *head) | |
44 | { | |
45 | unsigned long flags; | |
46 | ||
47 | spin_lock_irqsave(&leak_lock, flags); | |
48 | list_add(new, head); | |
49 | spin_unlock_irqrestore(&leak_lock, flags); | |
50 | } | |
51 | ||
52 | static inline | |
53 | void btrfs_leak_debug_del(struct list_head *entry) | |
54 | { | |
55 | unsigned long flags; | |
56 | ||
57 | spin_lock_irqsave(&leak_lock, flags); | |
58 | list_del(entry); | |
59 | spin_unlock_irqrestore(&leak_lock, flags); | |
60 | } | |
61 | ||
62 | static inline | |
63 | void btrfs_leak_debug_check(void) | |
64 | { | |
65 | struct extent_state *state; | |
66 | struct extent_buffer *eb; | |
67 | ||
68 | while (!list_empty(&states)) { | |
69 | state = list_entry(states.next, struct extent_state, leak_list); | |
9ee49a04 | 70 | pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n", |
27a3507d FM |
71 | state->start, state->end, state->state, |
72 | extent_state_in_tree(state), | |
b7ac31b7 | 73 | refcount_read(&state->refs)); |
6d49ba1b ES |
74 | list_del(&state->leak_list); |
75 | kmem_cache_free(extent_state_cache, state); | |
76 | } | |
77 | ||
78 | while (!list_empty(&buffers)) { | |
79 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
af2679e4 LB |
80 | pr_err("BTRFS: buffer leak start %llu len %lu refs %d bflags %lu\n", |
81 | eb->start, eb->len, atomic_read(&eb->refs), eb->bflags); | |
6d49ba1b ES |
82 | list_del(&eb->leak_list); |
83 | kmem_cache_free(extent_buffer_cache, eb); | |
84 | } | |
85 | } | |
8d599ae1 | 86 | |
a5dee37d JB |
87 | #define btrfs_debug_check_extent_io_range(tree, start, end) \ |
88 | __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end)) | |
8d599ae1 | 89 | static inline void __btrfs_debug_check_extent_io_range(const char *caller, |
a5dee37d | 90 | struct extent_io_tree *tree, u64 start, u64 end) |
8d599ae1 | 91 | { |
65a680f6 NB |
92 | struct inode *inode = tree->private_data; |
93 | u64 isize; | |
94 | ||
95 | if (!inode || !is_data_inode(inode)) | |
96 | return; | |
97 | ||
98 | isize = i_size_read(inode); | |
99 | if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) { | |
100 | btrfs_debug_rl(BTRFS_I(inode)->root->fs_info, | |
101 | "%s: ino %llu isize %llu odd range [%llu,%llu]", | |
102 | caller, btrfs_ino(BTRFS_I(inode)), isize, start, end); | |
103 | } | |
8d599ae1 | 104 | } |
6d49ba1b ES |
105 | #else |
106 | #define btrfs_leak_debug_add(new, head) do {} while (0) | |
107 | #define btrfs_leak_debug_del(entry) do {} while (0) | |
108 | #define btrfs_leak_debug_check() do {} while (0) | |
8d599ae1 | 109 | #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0) |
4bef0848 | 110 | #endif |
d1310b2e | 111 | |
d1310b2e CM |
112 | #define BUFFER_LRU_MAX 64 |
113 | ||
114 | struct tree_entry { | |
115 | u64 start; | |
116 | u64 end; | |
d1310b2e CM |
117 | struct rb_node rb_node; |
118 | }; | |
119 | ||
120 | struct extent_page_data { | |
121 | struct bio *bio; | |
122 | struct extent_io_tree *tree; | |
771ed689 CM |
123 | /* tells writepage not to lock the state bits for this range |
124 | * it still does the unlocking | |
125 | */ | |
ffbd517d CM |
126 | unsigned int extent_locked:1; |
127 | ||
70fd7614 | 128 | /* tells the submit_bio code to use REQ_SYNC */ |
ffbd517d | 129 | unsigned int sync_io:1; |
d1310b2e CM |
130 | }; |
131 | ||
57599c7e | 132 | static int add_extent_changeset(struct extent_state *state, unsigned bits, |
d38ed27f QW |
133 | struct extent_changeset *changeset, |
134 | int set) | |
135 | { | |
136 | int ret; | |
137 | ||
138 | if (!changeset) | |
57599c7e | 139 | return 0; |
d38ed27f | 140 | if (set && (state->state & bits) == bits) |
57599c7e | 141 | return 0; |
fefdc557 | 142 | if (!set && (state->state & bits) == 0) |
57599c7e | 143 | return 0; |
d38ed27f | 144 | changeset->bytes_changed += state->end - state->start + 1; |
53d32359 | 145 | ret = ulist_add(&changeset->range_changed, state->start, state->end, |
d38ed27f | 146 | GFP_ATOMIC); |
57599c7e | 147 | return ret; |
d38ed27f QW |
148 | } |
149 | ||
aab6e9ed | 150 | static void flush_write_bio(struct extent_page_data *epd); |
e2932ee0 | 151 | |
d1310b2e CM |
152 | int __init extent_io_init(void) |
153 | { | |
837e1972 | 154 | extent_state_cache = kmem_cache_create("btrfs_extent_state", |
9601e3f6 | 155 | sizeof(struct extent_state), 0, |
fba4b697 | 156 | SLAB_MEM_SPREAD, NULL); |
d1310b2e CM |
157 | if (!extent_state_cache) |
158 | return -ENOMEM; | |
159 | ||
837e1972 | 160 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 | 161 | sizeof(struct extent_buffer), 0, |
fba4b697 | 162 | SLAB_MEM_SPREAD, NULL); |
d1310b2e CM |
163 | if (!extent_buffer_cache) |
164 | goto free_state_cache; | |
9be3395b | 165 | |
8ac9f7c1 KO |
166 | if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE, |
167 | offsetof(struct btrfs_io_bio, bio), | |
168 | BIOSET_NEED_BVECS)) | |
9be3395b | 169 | goto free_buffer_cache; |
b208c2f7 | 170 | |
8ac9f7c1 | 171 | if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE)) |
b208c2f7 DW |
172 | goto free_bioset; |
173 | ||
d1310b2e CM |
174 | return 0; |
175 | ||
b208c2f7 | 176 | free_bioset: |
8ac9f7c1 | 177 | bioset_exit(&btrfs_bioset); |
b208c2f7 | 178 | |
9be3395b CM |
179 | free_buffer_cache: |
180 | kmem_cache_destroy(extent_buffer_cache); | |
181 | extent_buffer_cache = NULL; | |
182 | ||
d1310b2e CM |
183 | free_state_cache: |
184 | kmem_cache_destroy(extent_state_cache); | |
9be3395b | 185 | extent_state_cache = NULL; |
d1310b2e CM |
186 | return -ENOMEM; |
187 | } | |
188 | ||
e67c718b | 189 | void __cold extent_io_exit(void) |
d1310b2e | 190 | { |
6d49ba1b | 191 | btrfs_leak_debug_check(); |
8c0a8537 KS |
192 | |
193 | /* | |
194 | * Make sure all delayed rcu free are flushed before we | |
195 | * destroy caches. | |
196 | */ | |
197 | rcu_barrier(); | |
5598e900 KM |
198 | kmem_cache_destroy(extent_state_cache); |
199 | kmem_cache_destroy(extent_buffer_cache); | |
8ac9f7c1 | 200 | bioset_exit(&btrfs_bioset); |
d1310b2e CM |
201 | } |
202 | ||
203 | void extent_io_tree_init(struct extent_io_tree *tree, | |
c6100a4b | 204 | void *private_data) |
d1310b2e | 205 | { |
6bef4d31 | 206 | tree->state = RB_ROOT; |
d1310b2e CM |
207 | tree->ops = NULL; |
208 | tree->dirty_bytes = 0; | |
70dec807 | 209 | spin_lock_init(&tree->lock); |
c6100a4b | 210 | tree->private_data = private_data; |
d1310b2e | 211 | } |
d1310b2e | 212 | |
b2950863 | 213 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
214 | { |
215 | struct extent_state *state; | |
d1310b2e | 216 | |
3ba7ab22 MH |
217 | /* |
218 | * The given mask might be not appropriate for the slab allocator, | |
219 | * drop the unsupported bits | |
220 | */ | |
221 | mask &= ~(__GFP_DMA32|__GFP_HIGHMEM); | |
d1310b2e | 222 | state = kmem_cache_alloc(extent_state_cache, mask); |
2b114d1d | 223 | if (!state) |
d1310b2e CM |
224 | return state; |
225 | state->state = 0; | |
47dc196a | 226 | state->failrec = NULL; |
27a3507d | 227 | RB_CLEAR_NODE(&state->rb_node); |
6d49ba1b | 228 | btrfs_leak_debug_add(&state->leak_list, &states); |
b7ac31b7 | 229 | refcount_set(&state->refs, 1); |
d1310b2e | 230 | init_waitqueue_head(&state->wq); |
143bede5 | 231 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
232 | return state; |
233 | } | |
d1310b2e | 234 | |
4845e44f | 235 | void free_extent_state(struct extent_state *state) |
d1310b2e | 236 | { |
d1310b2e CM |
237 | if (!state) |
238 | return; | |
b7ac31b7 | 239 | if (refcount_dec_and_test(&state->refs)) { |
27a3507d | 240 | WARN_ON(extent_state_in_tree(state)); |
6d49ba1b | 241 | btrfs_leak_debug_del(&state->leak_list); |
143bede5 | 242 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
243 | kmem_cache_free(extent_state_cache, state); |
244 | } | |
245 | } | |
d1310b2e | 246 | |
f2071b21 FM |
247 | static struct rb_node *tree_insert(struct rb_root *root, |
248 | struct rb_node *search_start, | |
249 | u64 offset, | |
12cfbad9 FDBM |
250 | struct rb_node *node, |
251 | struct rb_node ***p_in, | |
252 | struct rb_node **parent_in) | |
d1310b2e | 253 | { |
f2071b21 | 254 | struct rb_node **p; |
d397712b | 255 | struct rb_node *parent = NULL; |
d1310b2e CM |
256 | struct tree_entry *entry; |
257 | ||
12cfbad9 FDBM |
258 | if (p_in && parent_in) { |
259 | p = *p_in; | |
260 | parent = *parent_in; | |
261 | goto do_insert; | |
262 | } | |
263 | ||
f2071b21 | 264 | p = search_start ? &search_start : &root->rb_node; |
d397712b | 265 | while (*p) { |
d1310b2e CM |
266 | parent = *p; |
267 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
268 | ||
269 | if (offset < entry->start) | |
270 | p = &(*p)->rb_left; | |
271 | else if (offset > entry->end) | |
272 | p = &(*p)->rb_right; | |
273 | else | |
274 | return parent; | |
275 | } | |
276 | ||
12cfbad9 | 277 | do_insert: |
d1310b2e CM |
278 | rb_link_node(node, parent, p); |
279 | rb_insert_color(node, root); | |
280 | return NULL; | |
281 | } | |
282 | ||
80ea96b1 | 283 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
12cfbad9 FDBM |
284 | struct rb_node **prev_ret, |
285 | struct rb_node **next_ret, | |
286 | struct rb_node ***p_ret, | |
287 | struct rb_node **parent_ret) | |
d1310b2e | 288 | { |
80ea96b1 | 289 | struct rb_root *root = &tree->state; |
12cfbad9 | 290 | struct rb_node **n = &root->rb_node; |
d1310b2e CM |
291 | struct rb_node *prev = NULL; |
292 | struct rb_node *orig_prev = NULL; | |
293 | struct tree_entry *entry; | |
294 | struct tree_entry *prev_entry = NULL; | |
295 | ||
12cfbad9 FDBM |
296 | while (*n) { |
297 | prev = *n; | |
298 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
d1310b2e CM |
299 | prev_entry = entry; |
300 | ||
301 | if (offset < entry->start) | |
12cfbad9 | 302 | n = &(*n)->rb_left; |
d1310b2e | 303 | else if (offset > entry->end) |
12cfbad9 | 304 | n = &(*n)->rb_right; |
d397712b | 305 | else |
12cfbad9 | 306 | return *n; |
d1310b2e CM |
307 | } |
308 | ||
12cfbad9 FDBM |
309 | if (p_ret) |
310 | *p_ret = n; | |
311 | if (parent_ret) | |
312 | *parent_ret = prev; | |
313 | ||
d1310b2e CM |
314 | if (prev_ret) { |
315 | orig_prev = prev; | |
d397712b | 316 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
317 | prev = rb_next(prev); |
318 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
319 | } | |
320 | *prev_ret = prev; | |
321 | prev = orig_prev; | |
322 | } | |
323 | ||
324 | if (next_ret) { | |
325 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 326 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
327 | prev = rb_prev(prev); |
328 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
329 | } | |
330 | *next_ret = prev; | |
331 | } | |
332 | return NULL; | |
333 | } | |
334 | ||
12cfbad9 FDBM |
335 | static inline struct rb_node * |
336 | tree_search_for_insert(struct extent_io_tree *tree, | |
337 | u64 offset, | |
338 | struct rb_node ***p_ret, | |
339 | struct rb_node **parent_ret) | |
d1310b2e | 340 | { |
70dec807 | 341 | struct rb_node *prev = NULL; |
d1310b2e | 342 | struct rb_node *ret; |
70dec807 | 343 | |
12cfbad9 | 344 | ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret); |
d397712b | 345 | if (!ret) |
d1310b2e CM |
346 | return prev; |
347 | return ret; | |
348 | } | |
349 | ||
12cfbad9 FDBM |
350 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
351 | u64 offset) | |
352 | { | |
353 | return tree_search_for_insert(tree, offset, NULL, NULL); | |
354 | } | |
355 | ||
d1310b2e CM |
356 | /* |
357 | * utility function to look for merge candidates inside a given range. | |
358 | * Any extents with matching state are merged together into a single | |
359 | * extent in the tree. Extents with EXTENT_IO in their state field | |
360 | * are not merged because the end_io handlers need to be able to do | |
361 | * operations on them without sleeping (or doing allocations/splits). | |
362 | * | |
363 | * This should be called with the tree lock held. | |
364 | */ | |
1bf85046 JM |
365 | static void merge_state(struct extent_io_tree *tree, |
366 | struct extent_state *state) | |
d1310b2e CM |
367 | { |
368 | struct extent_state *other; | |
369 | struct rb_node *other_node; | |
370 | ||
5b21f2ed | 371 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
1bf85046 | 372 | return; |
d1310b2e CM |
373 | |
374 | other_node = rb_prev(&state->rb_node); | |
375 | if (other_node) { | |
376 | other = rb_entry(other_node, struct extent_state, rb_node); | |
377 | if (other->end == state->start - 1 && | |
378 | other->state == state->state) { | |
5c848198 NB |
379 | if (tree->private_data && |
380 | is_data_inode(tree->private_data)) | |
381 | btrfs_merge_delalloc_extent(tree->private_data, | |
382 | state, other); | |
d1310b2e | 383 | state->start = other->start; |
d1310b2e | 384 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 385 | RB_CLEAR_NODE(&other->rb_node); |
d1310b2e CM |
386 | free_extent_state(other); |
387 | } | |
388 | } | |
389 | other_node = rb_next(&state->rb_node); | |
390 | if (other_node) { | |
391 | other = rb_entry(other_node, struct extent_state, rb_node); | |
392 | if (other->start == state->end + 1 && | |
393 | other->state == state->state) { | |
5c848198 NB |
394 | if (tree->private_data && |
395 | is_data_inode(tree->private_data)) | |
396 | btrfs_merge_delalloc_extent(tree->private_data, | |
397 | state, other); | |
df98b6e2 | 398 | state->end = other->end; |
df98b6e2 | 399 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 400 | RB_CLEAR_NODE(&other->rb_node); |
df98b6e2 | 401 | free_extent_state(other); |
d1310b2e CM |
402 | } |
403 | } | |
d1310b2e CM |
404 | } |
405 | ||
3150b699 | 406 | static void set_state_bits(struct extent_io_tree *tree, |
d38ed27f QW |
407 | struct extent_state *state, unsigned *bits, |
408 | struct extent_changeset *changeset); | |
3150b699 | 409 | |
d1310b2e CM |
410 | /* |
411 | * insert an extent_state struct into the tree. 'bits' are set on the | |
412 | * struct before it is inserted. | |
413 | * | |
414 | * This may return -EEXIST if the extent is already there, in which case the | |
415 | * state struct is freed. | |
416 | * | |
417 | * The tree lock is not taken internally. This is a utility function and | |
418 | * probably isn't what you want to call (see set/clear_extent_bit). | |
419 | */ | |
420 | static int insert_state(struct extent_io_tree *tree, | |
421 | struct extent_state *state, u64 start, u64 end, | |
12cfbad9 FDBM |
422 | struct rb_node ***p, |
423 | struct rb_node **parent, | |
d38ed27f | 424 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e CM |
425 | { |
426 | struct rb_node *node; | |
427 | ||
31b1a2bd | 428 | if (end < start) |
efe120a0 | 429 | WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n", |
c1c9ff7c | 430 | end, start); |
d1310b2e CM |
431 | state->start = start; |
432 | state->end = end; | |
9ed74f2d | 433 | |
d38ed27f | 434 | set_state_bits(tree, state, bits, changeset); |
3150b699 | 435 | |
f2071b21 | 436 | node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent); |
d1310b2e CM |
437 | if (node) { |
438 | struct extent_state *found; | |
439 | found = rb_entry(node, struct extent_state, rb_node); | |
62e85577 | 440 | pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n", |
c1c9ff7c | 441 | found->start, found->end, start, end); |
d1310b2e CM |
442 | return -EEXIST; |
443 | } | |
444 | merge_state(tree, state); | |
445 | return 0; | |
446 | } | |
447 | ||
448 | /* | |
449 | * split a given extent state struct in two, inserting the preallocated | |
450 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
451 | * offset inside 'orig' where it should be split. | |
452 | * | |
453 | * Before calling, | |
454 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
455 | * are two extent state structs in the tree: | |
456 | * prealloc: [orig->start, split - 1] | |
457 | * orig: [ split, orig->end ] | |
458 | * | |
459 | * The tree locks are not taken by this function. They need to be held | |
460 | * by the caller. | |
461 | */ | |
462 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
463 | struct extent_state *prealloc, u64 split) | |
464 | { | |
465 | struct rb_node *node; | |
9ed74f2d | 466 | |
abbb55f4 NB |
467 | if (tree->private_data && is_data_inode(tree->private_data)) |
468 | btrfs_split_delalloc_extent(tree->private_data, orig, split); | |
9ed74f2d | 469 | |
d1310b2e CM |
470 | prealloc->start = orig->start; |
471 | prealloc->end = split - 1; | |
472 | prealloc->state = orig->state; | |
473 | orig->start = split; | |
474 | ||
f2071b21 FM |
475 | node = tree_insert(&tree->state, &orig->rb_node, prealloc->end, |
476 | &prealloc->rb_node, NULL, NULL); | |
d1310b2e | 477 | if (node) { |
d1310b2e CM |
478 | free_extent_state(prealloc); |
479 | return -EEXIST; | |
480 | } | |
481 | return 0; | |
482 | } | |
483 | ||
cdc6a395 LZ |
484 | static struct extent_state *next_state(struct extent_state *state) |
485 | { | |
486 | struct rb_node *next = rb_next(&state->rb_node); | |
487 | if (next) | |
488 | return rb_entry(next, struct extent_state, rb_node); | |
489 | else | |
490 | return NULL; | |
491 | } | |
492 | ||
d1310b2e CM |
493 | /* |
494 | * utility function to clear some bits in an extent state struct. | |
1b303fc0 | 495 | * it will optionally wake up any one waiting on this state (wake == 1). |
d1310b2e CM |
496 | * |
497 | * If no bits are set on the state struct after clearing things, the | |
498 | * struct is freed and removed from the tree | |
499 | */ | |
cdc6a395 LZ |
500 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
501 | struct extent_state *state, | |
fefdc557 QW |
502 | unsigned *bits, int wake, |
503 | struct extent_changeset *changeset) | |
d1310b2e | 504 | { |
cdc6a395 | 505 | struct extent_state *next; |
9ee49a04 | 506 | unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS; |
57599c7e | 507 | int ret; |
d1310b2e | 508 | |
0ca1f7ce | 509 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
510 | u64 range = state->end - state->start + 1; |
511 | WARN_ON(range > tree->dirty_bytes); | |
512 | tree->dirty_bytes -= range; | |
513 | } | |
a36bb5f9 NB |
514 | |
515 | if (tree->private_data && is_data_inode(tree->private_data)) | |
516 | btrfs_clear_delalloc_extent(tree->private_data, state, bits); | |
517 | ||
57599c7e DS |
518 | ret = add_extent_changeset(state, bits_to_clear, changeset, 0); |
519 | BUG_ON(ret < 0); | |
32c00aff | 520 | state->state &= ~bits_to_clear; |
d1310b2e CM |
521 | if (wake) |
522 | wake_up(&state->wq); | |
0ca1f7ce | 523 | if (state->state == 0) { |
cdc6a395 | 524 | next = next_state(state); |
27a3507d | 525 | if (extent_state_in_tree(state)) { |
d1310b2e | 526 | rb_erase(&state->rb_node, &tree->state); |
27a3507d | 527 | RB_CLEAR_NODE(&state->rb_node); |
d1310b2e CM |
528 | free_extent_state(state); |
529 | } else { | |
530 | WARN_ON(1); | |
531 | } | |
532 | } else { | |
533 | merge_state(tree, state); | |
cdc6a395 | 534 | next = next_state(state); |
d1310b2e | 535 | } |
cdc6a395 | 536 | return next; |
d1310b2e CM |
537 | } |
538 | ||
8233767a XG |
539 | static struct extent_state * |
540 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
541 | { | |
542 | if (!prealloc) | |
543 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
544 | ||
545 | return prealloc; | |
546 | } | |
547 | ||
48a3b636 | 548 | static void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
c2d904e0 | 549 | { |
05912a3c DS |
550 | struct inode *inode = tree->private_data; |
551 | ||
552 | btrfs_panic(btrfs_sb(inode->i_sb), err, | |
553 | "locking error: extent tree was modified by another thread while locked"); | |
c2d904e0 JM |
554 | } |
555 | ||
d1310b2e CM |
556 | /* |
557 | * clear some bits on a range in the tree. This may require splitting | |
558 | * or inserting elements in the tree, so the gfp mask is used to | |
559 | * indicate which allocations or sleeping are allowed. | |
560 | * | |
561 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
562 | * the given range from the tree regardless of state (ie for truncate). | |
563 | * | |
564 | * the range [start, end] is inclusive. | |
565 | * | |
6763af84 | 566 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e | 567 | */ |
66b0c887 | 568 | int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
fefdc557 QW |
569 | unsigned bits, int wake, int delete, |
570 | struct extent_state **cached_state, | |
571 | gfp_t mask, struct extent_changeset *changeset) | |
d1310b2e CM |
572 | { |
573 | struct extent_state *state; | |
2c64c53d | 574 | struct extent_state *cached; |
d1310b2e CM |
575 | struct extent_state *prealloc = NULL; |
576 | struct rb_node *node; | |
5c939df5 | 577 | u64 last_end; |
d1310b2e | 578 | int err; |
2ac55d41 | 579 | int clear = 0; |
d1310b2e | 580 | |
a5dee37d | 581 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 582 | |
7ee9e440 JB |
583 | if (bits & EXTENT_DELALLOC) |
584 | bits |= EXTENT_NORESERVE; | |
585 | ||
0ca1f7ce YZ |
586 | if (delete) |
587 | bits |= ~EXTENT_CTLBITS; | |
588 | bits |= EXTENT_FIRST_DELALLOC; | |
589 | ||
2ac55d41 JB |
590 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
591 | clear = 1; | |
d1310b2e | 592 | again: |
d0164adc | 593 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
c7bc6319 FM |
594 | /* |
595 | * Don't care for allocation failure here because we might end | |
596 | * up not needing the pre-allocated extent state at all, which | |
597 | * is the case if we only have in the tree extent states that | |
598 | * cover our input range and don't cover too any other range. | |
599 | * If we end up needing a new extent state we allocate it later. | |
600 | */ | |
d1310b2e | 601 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
602 | } |
603 | ||
cad321ad | 604 | spin_lock(&tree->lock); |
2c64c53d CM |
605 | if (cached_state) { |
606 | cached = *cached_state; | |
2ac55d41 JB |
607 | |
608 | if (clear) { | |
609 | *cached_state = NULL; | |
610 | cached_state = NULL; | |
611 | } | |
612 | ||
27a3507d FM |
613 | if (cached && extent_state_in_tree(cached) && |
614 | cached->start <= start && cached->end > start) { | |
2ac55d41 | 615 | if (clear) |
b7ac31b7 | 616 | refcount_dec(&cached->refs); |
2c64c53d | 617 | state = cached; |
42daec29 | 618 | goto hit_next; |
2c64c53d | 619 | } |
2ac55d41 JB |
620 | if (clear) |
621 | free_extent_state(cached); | |
2c64c53d | 622 | } |
d1310b2e CM |
623 | /* |
624 | * this search will find the extents that end after | |
625 | * our range starts | |
626 | */ | |
80ea96b1 | 627 | node = tree_search(tree, start); |
d1310b2e CM |
628 | if (!node) |
629 | goto out; | |
630 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 631 | hit_next: |
d1310b2e CM |
632 | if (state->start > end) |
633 | goto out; | |
634 | WARN_ON(state->end < start); | |
5c939df5 | 635 | last_end = state->end; |
d1310b2e | 636 | |
0449314a | 637 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
638 | if (!(state->state & bits)) { |
639 | state = next_state(state); | |
0449314a | 640 | goto next; |
cdc6a395 | 641 | } |
0449314a | 642 | |
d1310b2e CM |
643 | /* |
644 | * | ---- desired range ---- | | |
645 | * | state | or | |
646 | * | ------------- state -------------- | | |
647 | * | |
648 | * We need to split the extent we found, and may flip | |
649 | * bits on second half. | |
650 | * | |
651 | * If the extent we found extends past our range, we | |
652 | * just split and search again. It'll get split again | |
653 | * the next time though. | |
654 | * | |
655 | * If the extent we found is inside our range, we clear | |
656 | * the desired bit on it. | |
657 | */ | |
658 | ||
659 | if (state->start < start) { | |
8233767a XG |
660 | prealloc = alloc_extent_state_atomic(prealloc); |
661 | BUG_ON(!prealloc); | |
d1310b2e | 662 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
663 | if (err) |
664 | extent_io_tree_panic(tree, err); | |
665 | ||
d1310b2e CM |
666 | prealloc = NULL; |
667 | if (err) | |
668 | goto out; | |
669 | if (state->end <= end) { | |
fefdc557 QW |
670 | state = clear_state_bit(tree, state, &bits, wake, |
671 | changeset); | |
d1ac6e41 | 672 | goto next; |
d1310b2e CM |
673 | } |
674 | goto search_again; | |
675 | } | |
676 | /* | |
677 | * | ---- desired range ---- | | |
678 | * | state | | |
679 | * We need to split the extent, and clear the bit | |
680 | * on the first half | |
681 | */ | |
682 | if (state->start <= end && state->end > end) { | |
8233767a XG |
683 | prealloc = alloc_extent_state_atomic(prealloc); |
684 | BUG_ON(!prealloc); | |
d1310b2e | 685 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
686 | if (err) |
687 | extent_io_tree_panic(tree, err); | |
688 | ||
d1310b2e CM |
689 | if (wake) |
690 | wake_up(&state->wq); | |
42daec29 | 691 | |
fefdc557 | 692 | clear_state_bit(tree, prealloc, &bits, wake, changeset); |
9ed74f2d | 693 | |
d1310b2e CM |
694 | prealloc = NULL; |
695 | goto out; | |
696 | } | |
42daec29 | 697 | |
fefdc557 | 698 | state = clear_state_bit(tree, state, &bits, wake, changeset); |
0449314a | 699 | next: |
5c939df5 YZ |
700 | if (last_end == (u64)-1) |
701 | goto out; | |
702 | start = last_end + 1; | |
cdc6a395 | 703 | if (start <= end && state && !need_resched()) |
692e5759 | 704 | goto hit_next; |
d1310b2e CM |
705 | |
706 | search_again: | |
707 | if (start > end) | |
708 | goto out; | |
cad321ad | 709 | spin_unlock(&tree->lock); |
d0164adc | 710 | if (gfpflags_allow_blocking(mask)) |
d1310b2e CM |
711 | cond_resched(); |
712 | goto again; | |
7ab5cb2a DS |
713 | |
714 | out: | |
715 | spin_unlock(&tree->lock); | |
716 | if (prealloc) | |
717 | free_extent_state(prealloc); | |
718 | ||
719 | return 0; | |
720 | ||
d1310b2e | 721 | } |
d1310b2e | 722 | |
143bede5 JM |
723 | static void wait_on_state(struct extent_io_tree *tree, |
724 | struct extent_state *state) | |
641f5219 CH |
725 | __releases(tree->lock) |
726 | __acquires(tree->lock) | |
d1310b2e CM |
727 | { |
728 | DEFINE_WAIT(wait); | |
729 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 730 | spin_unlock(&tree->lock); |
d1310b2e | 731 | schedule(); |
cad321ad | 732 | spin_lock(&tree->lock); |
d1310b2e | 733 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
734 | } |
735 | ||
736 | /* | |
737 | * waits for one or more bits to clear on a range in the state tree. | |
738 | * The range [start, end] is inclusive. | |
739 | * The tree lock is taken by this function | |
740 | */ | |
41074888 DS |
741 | static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
742 | unsigned long bits) | |
d1310b2e CM |
743 | { |
744 | struct extent_state *state; | |
745 | struct rb_node *node; | |
746 | ||
a5dee37d | 747 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 748 | |
cad321ad | 749 | spin_lock(&tree->lock); |
d1310b2e CM |
750 | again: |
751 | while (1) { | |
752 | /* | |
753 | * this search will find all the extents that end after | |
754 | * our range starts | |
755 | */ | |
80ea96b1 | 756 | node = tree_search(tree, start); |
c50d3e71 | 757 | process_node: |
d1310b2e CM |
758 | if (!node) |
759 | break; | |
760 | ||
761 | state = rb_entry(node, struct extent_state, rb_node); | |
762 | ||
763 | if (state->start > end) | |
764 | goto out; | |
765 | ||
766 | if (state->state & bits) { | |
767 | start = state->start; | |
b7ac31b7 | 768 | refcount_inc(&state->refs); |
d1310b2e CM |
769 | wait_on_state(tree, state); |
770 | free_extent_state(state); | |
771 | goto again; | |
772 | } | |
773 | start = state->end + 1; | |
774 | ||
775 | if (start > end) | |
776 | break; | |
777 | ||
c50d3e71 FM |
778 | if (!cond_resched_lock(&tree->lock)) { |
779 | node = rb_next(node); | |
780 | goto process_node; | |
781 | } | |
d1310b2e CM |
782 | } |
783 | out: | |
cad321ad | 784 | spin_unlock(&tree->lock); |
d1310b2e | 785 | } |
d1310b2e | 786 | |
1bf85046 | 787 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 788 | struct extent_state *state, |
d38ed27f | 789 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e | 790 | { |
9ee49a04 | 791 | unsigned bits_to_set = *bits & ~EXTENT_CTLBITS; |
57599c7e | 792 | int ret; |
9ed74f2d | 793 | |
e06a1fc9 NB |
794 | if (tree->private_data && is_data_inode(tree->private_data)) |
795 | btrfs_set_delalloc_extent(tree->private_data, state, bits); | |
796 | ||
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 | } | |
57599c7e DS |
801 | ret = add_extent_changeset(state, bits_to_set, changeset, 1); |
802 | BUG_ON(ret < 0); | |
0ca1f7ce | 803 | state->state |= bits_to_set; |
d1310b2e CM |
804 | } |
805 | ||
e38e2ed7 FM |
806 | static void cache_state_if_flags(struct extent_state *state, |
807 | struct extent_state **cached_ptr, | |
9ee49a04 | 808 | unsigned flags) |
2c64c53d CM |
809 | { |
810 | if (cached_ptr && !(*cached_ptr)) { | |
e38e2ed7 | 811 | if (!flags || (state->state & flags)) { |
2c64c53d | 812 | *cached_ptr = state; |
b7ac31b7 | 813 | refcount_inc(&state->refs); |
2c64c53d CM |
814 | } |
815 | } | |
816 | } | |
817 | ||
e38e2ed7 FM |
818 | static void cache_state(struct extent_state *state, |
819 | struct extent_state **cached_ptr) | |
820 | { | |
821 | return cache_state_if_flags(state, cached_ptr, | |
822 | EXTENT_IOBITS | EXTENT_BOUNDARY); | |
823 | } | |
824 | ||
d1310b2e | 825 | /* |
1edbb734 CM |
826 | * set some bits on a range in the tree. This may require allocations or |
827 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 828 | * |
1edbb734 CM |
829 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
830 | * part of the range already has the desired bits set. The start of the | |
831 | * existing range is returned in failed_start in this case. | |
d1310b2e | 832 | * |
1edbb734 | 833 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 834 | */ |
1edbb734 | 835 | |
3fbe5c02 JM |
836 | static int __must_check |
837 | __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 838 | unsigned bits, unsigned exclusive_bits, |
41074888 | 839 | u64 *failed_start, struct extent_state **cached_state, |
d38ed27f | 840 | gfp_t mask, struct extent_changeset *changeset) |
d1310b2e CM |
841 | { |
842 | struct extent_state *state; | |
843 | struct extent_state *prealloc = NULL; | |
844 | struct rb_node *node; | |
12cfbad9 FDBM |
845 | struct rb_node **p; |
846 | struct rb_node *parent; | |
d1310b2e | 847 | int err = 0; |
d1310b2e CM |
848 | u64 last_start; |
849 | u64 last_end; | |
42daec29 | 850 | |
a5dee37d | 851 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 852 | |
0ca1f7ce | 853 | bits |= EXTENT_FIRST_DELALLOC; |
d1310b2e | 854 | again: |
d0164adc | 855 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
059f791c DS |
856 | /* |
857 | * Don't care for allocation failure here because we might end | |
858 | * up not needing the pre-allocated extent state at all, which | |
859 | * is the case if we only have in the tree extent states that | |
860 | * cover our input range and don't cover too any other range. | |
861 | * If we end up needing a new extent state we allocate it later. | |
862 | */ | |
d1310b2e | 863 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
864 | } |
865 | ||
cad321ad | 866 | spin_lock(&tree->lock); |
9655d298 CM |
867 | if (cached_state && *cached_state) { |
868 | state = *cached_state; | |
df98b6e2 | 869 | if (state->start <= start && state->end > start && |
27a3507d | 870 | extent_state_in_tree(state)) { |
9655d298 CM |
871 | node = &state->rb_node; |
872 | goto hit_next; | |
873 | } | |
874 | } | |
d1310b2e CM |
875 | /* |
876 | * this search will find all the extents that end after | |
877 | * our range starts. | |
878 | */ | |
12cfbad9 | 879 | node = tree_search_for_insert(tree, start, &p, &parent); |
d1310b2e | 880 | if (!node) { |
8233767a XG |
881 | prealloc = alloc_extent_state_atomic(prealloc); |
882 | BUG_ON(!prealloc); | |
12cfbad9 | 883 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 884 | &p, &parent, &bits, changeset); |
c2d904e0 JM |
885 | if (err) |
886 | extent_io_tree_panic(tree, err); | |
887 | ||
c42ac0bc | 888 | cache_state(prealloc, cached_state); |
d1310b2e | 889 | prealloc = NULL; |
d1310b2e CM |
890 | goto out; |
891 | } | |
d1310b2e | 892 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 893 | hit_next: |
d1310b2e CM |
894 | last_start = state->start; |
895 | last_end = state->end; | |
896 | ||
897 | /* | |
898 | * | ---- desired range ---- | | |
899 | * | state | | |
900 | * | |
901 | * Just lock what we found and keep going | |
902 | */ | |
903 | if (state->start == start && state->end <= end) { | |
1edbb734 | 904 | if (state->state & exclusive_bits) { |
d1310b2e CM |
905 | *failed_start = state->start; |
906 | err = -EEXIST; | |
907 | goto out; | |
908 | } | |
42daec29 | 909 | |
d38ed27f | 910 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 911 | cache_state(state, cached_state); |
d1310b2e | 912 | merge_state(tree, state); |
5c939df5 YZ |
913 | if (last_end == (u64)-1) |
914 | goto out; | |
915 | start = last_end + 1; | |
d1ac6e41 LB |
916 | state = next_state(state); |
917 | if (start < end && state && state->start == start && | |
918 | !need_resched()) | |
919 | goto hit_next; | |
d1310b2e CM |
920 | goto search_again; |
921 | } | |
922 | ||
923 | /* | |
924 | * | ---- desired range ---- | | |
925 | * | state | | |
926 | * or | |
927 | * | ------------- state -------------- | | |
928 | * | |
929 | * We need to split the extent we found, and may flip bits on | |
930 | * second half. | |
931 | * | |
932 | * If the extent we found extends past our | |
933 | * range, we just split and search again. It'll get split | |
934 | * again the next time though. | |
935 | * | |
936 | * If the extent we found is inside our range, we set the | |
937 | * desired bit on it. | |
938 | */ | |
939 | if (state->start < start) { | |
1edbb734 | 940 | if (state->state & exclusive_bits) { |
d1310b2e CM |
941 | *failed_start = start; |
942 | err = -EEXIST; | |
943 | goto out; | |
944 | } | |
8233767a XG |
945 | |
946 | prealloc = alloc_extent_state_atomic(prealloc); | |
947 | BUG_ON(!prealloc); | |
d1310b2e | 948 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
949 | if (err) |
950 | extent_io_tree_panic(tree, err); | |
951 | ||
d1310b2e CM |
952 | prealloc = NULL; |
953 | if (err) | |
954 | goto out; | |
955 | if (state->end <= end) { | |
d38ed27f | 956 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 957 | cache_state(state, cached_state); |
d1310b2e | 958 | merge_state(tree, state); |
5c939df5 YZ |
959 | if (last_end == (u64)-1) |
960 | goto out; | |
961 | start = last_end + 1; | |
d1ac6e41 LB |
962 | state = next_state(state); |
963 | if (start < end && state && state->start == start && | |
964 | !need_resched()) | |
965 | goto hit_next; | |
d1310b2e CM |
966 | } |
967 | goto search_again; | |
968 | } | |
969 | /* | |
970 | * | ---- desired range ---- | | |
971 | * | state | or | state | | |
972 | * | |
973 | * There's a hole, we need to insert something in it and | |
974 | * ignore the extent we found. | |
975 | */ | |
976 | if (state->start > start) { | |
977 | u64 this_end; | |
978 | if (end < last_start) | |
979 | this_end = end; | |
980 | else | |
d397712b | 981 | this_end = last_start - 1; |
8233767a XG |
982 | |
983 | prealloc = alloc_extent_state_atomic(prealloc); | |
984 | BUG_ON(!prealloc); | |
c7f895a2 XG |
985 | |
986 | /* | |
987 | * Avoid to free 'prealloc' if it can be merged with | |
988 | * the later extent. | |
989 | */ | |
d1310b2e | 990 | err = insert_state(tree, prealloc, start, this_end, |
d38ed27f | 991 | NULL, NULL, &bits, changeset); |
c2d904e0 JM |
992 | if (err) |
993 | extent_io_tree_panic(tree, err); | |
994 | ||
9ed74f2d JB |
995 | cache_state(prealloc, cached_state); |
996 | prealloc = NULL; | |
d1310b2e CM |
997 | start = this_end + 1; |
998 | goto search_again; | |
999 | } | |
1000 | /* | |
1001 | * | ---- desired range ---- | | |
1002 | * | state | | |
1003 | * We need to split the extent, and set the bit | |
1004 | * on the first half | |
1005 | */ | |
1006 | if (state->start <= end && state->end > end) { | |
1edbb734 | 1007 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1008 | *failed_start = start; |
1009 | err = -EEXIST; | |
1010 | goto out; | |
1011 | } | |
8233767a XG |
1012 | |
1013 | prealloc = alloc_extent_state_atomic(prealloc); | |
1014 | BUG_ON(!prealloc); | |
d1310b2e | 1015 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
1016 | if (err) |
1017 | extent_io_tree_panic(tree, err); | |
d1310b2e | 1018 | |
d38ed27f | 1019 | set_state_bits(tree, prealloc, &bits, changeset); |
2c64c53d | 1020 | cache_state(prealloc, cached_state); |
d1310b2e CM |
1021 | merge_state(tree, prealloc); |
1022 | prealloc = NULL; | |
1023 | goto out; | |
1024 | } | |
1025 | ||
b5a4ba14 DS |
1026 | search_again: |
1027 | if (start > end) | |
1028 | goto out; | |
1029 | spin_unlock(&tree->lock); | |
1030 | if (gfpflags_allow_blocking(mask)) | |
1031 | cond_resched(); | |
1032 | goto again; | |
d1310b2e CM |
1033 | |
1034 | out: | |
cad321ad | 1035 | spin_unlock(&tree->lock); |
d1310b2e CM |
1036 | if (prealloc) |
1037 | free_extent_state(prealloc); | |
1038 | ||
1039 | return err; | |
1040 | ||
d1310b2e | 1041 | } |
d1310b2e | 1042 | |
41074888 | 1043 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
9ee49a04 | 1044 | unsigned bits, u64 * failed_start, |
41074888 | 1045 | struct extent_state **cached_state, gfp_t mask) |
3fbe5c02 JM |
1046 | { |
1047 | return __set_extent_bit(tree, start, end, bits, 0, failed_start, | |
d38ed27f | 1048 | cached_state, mask, NULL); |
3fbe5c02 JM |
1049 | } |
1050 | ||
1051 | ||
462d6fac | 1052 | /** |
10983f2e LB |
1053 | * convert_extent_bit - convert all bits in a given range from one bit to |
1054 | * another | |
462d6fac JB |
1055 | * @tree: the io tree to search |
1056 | * @start: the start offset in bytes | |
1057 | * @end: the end offset in bytes (inclusive) | |
1058 | * @bits: the bits to set in this range | |
1059 | * @clear_bits: the bits to clear in this range | |
e6138876 | 1060 | * @cached_state: state that we're going to cache |
462d6fac JB |
1061 | * |
1062 | * This will go through and set bits for the given range. If any states exist | |
1063 | * already in this range they are set with the given bit and cleared of the | |
1064 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
1065 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
1066 | * boundary bits like LOCK. | |
210aa277 DS |
1067 | * |
1068 | * All allocations are done with GFP_NOFS. | |
462d6fac JB |
1069 | */ |
1070 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1071 | unsigned bits, unsigned clear_bits, |
210aa277 | 1072 | struct extent_state **cached_state) |
462d6fac JB |
1073 | { |
1074 | struct extent_state *state; | |
1075 | struct extent_state *prealloc = NULL; | |
1076 | struct rb_node *node; | |
12cfbad9 FDBM |
1077 | struct rb_node **p; |
1078 | struct rb_node *parent; | |
462d6fac JB |
1079 | int err = 0; |
1080 | u64 last_start; | |
1081 | u64 last_end; | |
c8fd3de7 | 1082 | bool first_iteration = true; |
462d6fac | 1083 | |
a5dee37d | 1084 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 1085 | |
462d6fac | 1086 | again: |
210aa277 | 1087 | if (!prealloc) { |
c8fd3de7 FM |
1088 | /* |
1089 | * Best effort, don't worry if extent state allocation fails | |
1090 | * here for the first iteration. We might have a cached state | |
1091 | * that matches exactly the target range, in which case no | |
1092 | * extent state allocations are needed. We'll only know this | |
1093 | * after locking the tree. | |
1094 | */ | |
210aa277 | 1095 | prealloc = alloc_extent_state(GFP_NOFS); |
c8fd3de7 | 1096 | if (!prealloc && !first_iteration) |
462d6fac JB |
1097 | return -ENOMEM; |
1098 | } | |
1099 | ||
1100 | spin_lock(&tree->lock); | |
e6138876 JB |
1101 | if (cached_state && *cached_state) { |
1102 | state = *cached_state; | |
1103 | if (state->start <= start && state->end > start && | |
27a3507d | 1104 | extent_state_in_tree(state)) { |
e6138876 JB |
1105 | node = &state->rb_node; |
1106 | goto hit_next; | |
1107 | } | |
1108 | } | |
1109 | ||
462d6fac JB |
1110 | /* |
1111 | * this search will find all the extents that end after | |
1112 | * our range starts. | |
1113 | */ | |
12cfbad9 | 1114 | node = tree_search_for_insert(tree, start, &p, &parent); |
462d6fac JB |
1115 | if (!node) { |
1116 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1117 | if (!prealloc) { |
1118 | err = -ENOMEM; | |
1119 | goto out; | |
1120 | } | |
12cfbad9 | 1121 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 1122 | &p, &parent, &bits, NULL); |
c2d904e0 JM |
1123 | if (err) |
1124 | extent_io_tree_panic(tree, err); | |
c42ac0bc FDBM |
1125 | cache_state(prealloc, cached_state); |
1126 | prealloc = NULL; | |
462d6fac JB |
1127 | goto out; |
1128 | } | |
1129 | state = rb_entry(node, struct extent_state, rb_node); | |
1130 | hit_next: | |
1131 | last_start = state->start; | |
1132 | last_end = state->end; | |
1133 | ||
1134 | /* | |
1135 | * | ---- desired range ---- | | |
1136 | * | state | | |
1137 | * | |
1138 | * Just lock what we found and keep going | |
1139 | */ | |
1140 | if (state->start == start && state->end <= end) { | |
d38ed27f | 1141 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1142 | cache_state(state, cached_state); |
fefdc557 | 1143 | state = clear_state_bit(tree, state, &clear_bits, 0, NULL); |
462d6fac JB |
1144 | if (last_end == (u64)-1) |
1145 | goto out; | |
462d6fac | 1146 | start = last_end + 1; |
d1ac6e41 LB |
1147 | if (start < end && state && state->start == start && |
1148 | !need_resched()) | |
1149 | goto hit_next; | |
462d6fac JB |
1150 | goto search_again; |
1151 | } | |
1152 | ||
1153 | /* | |
1154 | * | ---- desired range ---- | | |
1155 | * | state | | |
1156 | * or | |
1157 | * | ------------- state -------------- | | |
1158 | * | |
1159 | * We need to split the extent we found, and may flip bits on | |
1160 | * second half. | |
1161 | * | |
1162 | * If the extent we found extends past our | |
1163 | * range, we just split and search again. It'll get split | |
1164 | * again the next time though. | |
1165 | * | |
1166 | * If the extent we found is inside our range, we set the | |
1167 | * desired bit on it. | |
1168 | */ | |
1169 | if (state->start < start) { | |
1170 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1171 | if (!prealloc) { |
1172 | err = -ENOMEM; | |
1173 | goto out; | |
1174 | } | |
462d6fac | 1175 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1176 | if (err) |
1177 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1178 | prealloc = NULL; |
1179 | if (err) | |
1180 | goto out; | |
1181 | if (state->end <= end) { | |
d38ed27f | 1182 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1183 | cache_state(state, cached_state); |
fefdc557 QW |
1184 | state = clear_state_bit(tree, state, &clear_bits, 0, |
1185 | NULL); | |
462d6fac JB |
1186 | if (last_end == (u64)-1) |
1187 | goto out; | |
1188 | start = last_end + 1; | |
d1ac6e41 LB |
1189 | if (start < end && state && state->start == start && |
1190 | !need_resched()) | |
1191 | goto hit_next; | |
462d6fac JB |
1192 | } |
1193 | goto search_again; | |
1194 | } | |
1195 | /* | |
1196 | * | ---- desired range ---- | | |
1197 | * | state | or | state | | |
1198 | * | |
1199 | * There's a hole, we need to insert something in it and | |
1200 | * ignore the extent we found. | |
1201 | */ | |
1202 | if (state->start > start) { | |
1203 | u64 this_end; | |
1204 | if (end < last_start) | |
1205 | this_end = end; | |
1206 | else | |
1207 | this_end = last_start - 1; | |
1208 | ||
1209 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1210 | if (!prealloc) { |
1211 | err = -ENOMEM; | |
1212 | goto out; | |
1213 | } | |
462d6fac JB |
1214 | |
1215 | /* | |
1216 | * Avoid to free 'prealloc' if it can be merged with | |
1217 | * the later extent. | |
1218 | */ | |
1219 | err = insert_state(tree, prealloc, start, this_end, | |
d38ed27f | 1220 | NULL, NULL, &bits, NULL); |
c2d904e0 JM |
1221 | if (err) |
1222 | extent_io_tree_panic(tree, err); | |
e6138876 | 1223 | cache_state(prealloc, cached_state); |
462d6fac JB |
1224 | prealloc = NULL; |
1225 | start = this_end + 1; | |
1226 | goto search_again; | |
1227 | } | |
1228 | /* | |
1229 | * | ---- desired range ---- | | |
1230 | * | state | | |
1231 | * We need to split the extent, and set the bit | |
1232 | * on the first half | |
1233 | */ | |
1234 | if (state->start <= end && state->end > end) { | |
1235 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1236 | if (!prealloc) { |
1237 | err = -ENOMEM; | |
1238 | goto out; | |
1239 | } | |
462d6fac JB |
1240 | |
1241 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1242 | if (err) |
1243 | extent_io_tree_panic(tree, err); | |
462d6fac | 1244 | |
d38ed27f | 1245 | set_state_bits(tree, prealloc, &bits, NULL); |
e6138876 | 1246 | cache_state(prealloc, cached_state); |
fefdc557 | 1247 | clear_state_bit(tree, prealloc, &clear_bits, 0, NULL); |
462d6fac JB |
1248 | prealloc = NULL; |
1249 | goto out; | |
1250 | } | |
1251 | ||
462d6fac JB |
1252 | search_again: |
1253 | if (start > end) | |
1254 | goto out; | |
1255 | spin_unlock(&tree->lock); | |
210aa277 | 1256 | cond_resched(); |
c8fd3de7 | 1257 | first_iteration = false; |
462d6fac | 1258 | goto again; |
462d6fac JB |
1259 | |
1260 | out: | |
1261 | spin_unlock(&tree->lock); | |
1262 | if (prealloc) | |
1263 | free_extent_state(prealloc); | |
1264 | ||
1265 | return err; | |
462d6fac JB |
1266 | } |
1267 | ||
d1310b2e | 1268 | /* wrappers around set/clear extent bit */ |
d38ed27f | 1269 | int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
2c53b912 | 1270 | unsigned bits, struct extent_changeset *changeset) |
d38ed27f QW |
1271 | { |
1272 | /* | |
1273 | * We don't support EXTENT_LOCKED yet, as current changeset will | |
1274 | * record any bits changed, so for EXTENT_LOCKED case, it will | |
1275 | * either fail with -EEXIST or changeset will record the whole | |
1276 | * range. | |
1277 | */ | |
1278 | BUG_ON(bits & EXTENT_LOCKED); | |
1279 | ||
2c53b912 | 1280 | return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS, |
d38ed27f QW |
1281 | changeset); |
1282 | } | |
1283 | ||
fefdc557 QW |
1284 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
1285 | unsigned bits, int wake, int delete, | |
ae0f1625 | 1286 | struct extent_state **cached) |
fefdc557 QW |
1287 | { |
1288 | return __clear_extent_bit(tree, start, end, bits, wake, delete, | |
ae0f1625 | 1289 | cached, GFP_NOFS, NULL); |
fefdc557 QW |
1290 | } |
1291 | ||
fefdc557 | 1292 | int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
f734c44a | 1293 | unsigned bits, struct extent_changeset *changeset) |
fefdc557 QW |
1294 | { |
1295 | /* | |
1296 | * Don't support EXTENT_LOCKED case, same reason as | |
1297 | * set_record_extent_bits(). | |
1298 | */ | |
1299 | BUG_ON(bits & EXTENT_LOCKED); | |
1300 | ||
f734c44a | 1301 | return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS, |
fefdc557 QW |
1302 | changeset); |
1303 | } | |
1304 | ||
d352ac68 CM |
1305 | /* |
1306 | * either insert or lock state struct between start and end use mask to tell | |
1307 | * us if waiting is desired. | |
1308 | */ | |
1edbb734 | 1309 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
ff13db41 | 1310 | struct extent_state **cached_state) |
d1310b2e CM |
1311 | { |
1312 | int err; | |
1313 | u64 failed_start; | |
9ee49a04 | 1314 | |
d1310b2e | 1315 | while (1) { |
ff13db41 | 1316 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, |
3fbe5c02 | 1317 | EXTENT_LOCKED, &failed_start, |
d38ed27f | 1318 | cached_state, GFP_NOFS, NULL); |
d0082371 | 1319 | if (err == -EEXIST) { |
d1310b2e CM |
1320 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1321 | start = failed_start; | |
d0082371 | 1322 | } else |
d1310b2e | 1323 | break; |
d1310b2e CM |
1324 | WARN_ON(start > end); |
1325 | } | |
1326 | return err; | |
1327 | } | |
d1310b2e | 1328 | |
d0082371 | 1329 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1330 | { |
1331 | int err; | |
1332 | u64 failed_start; | |
1333 | ||
3fbe5c02 | 1334 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
d38ed27f | 1335 | &failed_start, NULL, GFP_NOFS, NULL); |
6643558d YZ |
1336 | if (err == -EEXIST) { |
1337 | if (failed_start > start) | |
1338 | clear_extent_bit(tree, start, failed_start - 1, | |
ae0f1625 | 1339 | EXTENT_LOCKED, 1, 0, NULL); |
25179201 | 1340 | return 0; |
6643558d | 1341 | } |
25179201 JB |
1342 | return 1; |
1343 | } | |
25179201 | 1344 | |
bd1fa4f0 | 1345 | void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1346 | { |
09cbfeaf KS |
1347 | unsigned long index = start >> PAGE_SHIFT; |
1348 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1349 | struct page *page; |
1350 | ||
1351 | while (index <= end_index) { | |
1352 | page = find_get_page(inode->i_mapping, index); | |
1353 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1354 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1355 | put_page(page); |
4adaa611 CM |
1356 | index++; |
1357 | } | |
4adaa611 CM |
1358 | } |
1359 | ||
f6311572 | 1360 | void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1361 | { |
09cbfeaf KS |
1362 | unsigned long index = start >> PAGE_SHIFT; |
1363 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1364 | struct page *page; |
1365 | ||
1366 | while (index <= end_index) { | |
1367 | page = find_get_page(inode->i_mapping, index); | |
1368 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
4adaa611 | 1369 | __set_page_dirty_nobuffers(page); |
8d38633c | 1370 | account_page_redirty(page); |
09cbfeaf | 1371 | put_page(page); |
4adaa611 CM |
1372 | index++; |
1373 | } | |
4adaa611 CM |
1374 | } |
1375 | ||
d352ac68 CM |
1376 | /* find the first state struct with 'bits' set after 'start', and |
1377 | * return it. tree->lock must be held. NULL will returned if | |
1378 | * nothing was found after 'start' | |
1379 | */ | |
48a3b636 ES |
1380 | static struct extent_state * |
1381 | find_first_extent_bit_state(struct extent_io_tree *tree, | |
9ee49a04 | 1382 | u64 start, unsigned bits) |
d7fc640e CM |
1383 | { |
1384 | struct rb_node *node; | |
1385 | struct extent_state *state; | |
1386 | ||
1387 | /* | |
1388 | * this search will find all the extents that end after | |
1389 | * our range starts. | |
1390 | */ | |
1391 | node = tree_search(tree, start); | |
d397712b | 1392 | if (!node) |
d7fc640e | 1393 | goto out; |
d7fc640e | 1394 | |
d397712b | 1395 | while (1) { |
d7fc640e | 1396 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1397 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1398 | return state; |
d397712b | 1399 | |
d7fc640e CM |
1400 | node = rb_next(node); |
1401 | if (!node) | |
1402 | break; | |
1403 | } | |
1404 | out: | |
1405 | return NULL; | |
1406 | } | |
d7fc640e | 1407 | |
69261c4b XG |
1408 | /* |
1409 | * find the first offset in the io tree with 'bits' set. zero is | |
1410 | * returned if we find something, and *start_ret and *end_ret are | |
1411 | * set to reflect the state struct that was found. | |
1412 | * | |
477d7eaf | 1413 | * If nothing was found, 1 is returned. If found something, return 0. |
69261c4b XG |
1414 | */ |
1415 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
9ee49a04 | 1416 | u64 *start_ret, u64 *end_ret, unsigned bits, |
e6138876 | 1417 | struct extent_state **cached_state) |
69261c4b XG |
1418 | { |
1419 | struct extent_state *state; | |
1420 | int ret = 1; | |
1421 | ||
1422 | spin_lock(&tree->lock); | |
e6138876 JB |
1423 | if (cached_state && *cached_state) { |
1424 | state = *cached_state; | |
27a3507d | 1425 | if (state->end == start - 1 && extent_state_in_tree(state)) { |
9688e9a9 | 1426 | while ((state = next_state(state)) != NULL) { |
e6138876 JB |
1427 | if (state->state & bits) |
1428 | goto got_it; | |
e6138876 JB |
1429 | } |
1430 | free_extent_state(*cached_state); | |
1431 | *cached_state = NULL; | |
1432 | goto out; | |
1433 | } | |
1434 | free_extent_state(*cached_state); | |
1435 | *cached_state = NULL; | |
1436 | } | |
1437 | ||
69261c4b | 1438 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1439 | got_it: |
69261c4b | 1440 | if (state) { |
e38e2ed7 | 1441 | cache_state_if_flags(state, cached_state, 0); |
69261c4b XG |
1442 | *start_ret = state->start; |
1443 | *end_ret = state->end; | |
1444 | ret = 0; | |
1445 | } | |
e6138876 | 1446 | out: |
69261c4b XG |
1447 | spin_unlock(&tree->lock); |
1448 | return ret; | |
1449 | } | |
1450 | ||
d352ac68 CM |
1451 | /* |
1452 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1453 | * more than 'max_bytes'. start and end are used to return the range, | |
1454 | * | |
1455 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1456 | */ | |
c8b97818 | 1457 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1458 | u64 *start, u64 *end, u64 max_bytes, |
1459 | struct extent_state **cached_state) | |
d1310b2e CM |
1460 | { |
1461 | struct rb_node *node; | |
1462 | struct extent_state *state; | |
1463 | u64 cur_start = *start; | |
1464 | u64 found = 0; | |
1465 | u64 total_bytes = 0; | |
1466 | ||
cad321ad | 1467 | spin_lock(&tree->lock); |
c8b97818 | 1468 | |
d1310b2e CM |
1469 | /* |
1470 | * this search will find all the extents that end after | |
1471 | * our range starts. | |
1472 | */ | |
80ea96b1 | 1473 | node = tree_search(tree, cur_start); |
2b114d1d | 1474 | if (!node) { |
3b951516 CM |
1475 | if (!found) |
1476 | *end = (u64)-1; | |
d1310b2e CM |
1477 | goto out; |
1478 | } | |
1479 | ||
d397712b | 1480 | while (1) { |
d1310b2e | 1481 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1482 | if (found && (state->start != cur_start || |
1483 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1484 | goto out; |
1485 | } | |
1486 | if (!(state->state & EXTENT_DELALLOC)) { | |
1487 | if (!found) | |
1488 | *end = state->end; | |
1489 | goto out; | |
1490 | } | |
c2a128d2 | 1491 | if (!found) { |
d1310b2e | 1492 | *start = state->start; |
c2a128d2 | 1493 | *cached_state = state; |
b7ac31b7 | 1494 | refcount_inc(&state->refs); |
c2a128d2 | 1495 | } |
d1310b2e CM |
1496 | found++; |
1497 | *end = state->end; | |
1498 | cur_start = state->end + 1; | |
1499 | node = rb_next(node); | |
d1310b2e | 1500 | total_bytes += state->end - state->start + 1; |
7bf811a5 | 1501 | if (total_bytes >= max_bytes) |
573aecaf | 1502 | break; |
573aecaf | 1503 | if (!node) |
d1310b2e CM |
1504 | break; |
1505 | } | |
1506 | out: | |
cad321ad | 1507 | spin_unlock(&tree->lock); |
d1310b2e CM |
1508 | return found; |
1509 | } | |
1510 | ||
da2c7009 LB |
1511 | static int __process_pages_contig(struct address_space *mapping, |
1512 | struct page *locked_page, | |
1513 | pgoff_t start_index, pgoff_t end_index, | |
1514 | unsigned long page_ops, pgoff_t *index_ret); | |
1515 | ||
143bede5 JM |
1516 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1517 | struct page *locked_page, | |
1518 | u64 start, u64 end) | |
c8b97818 | 1519 | { |
09cbfeaf KS |
1520 | unsigned long index = start >> PAGE_SHIFT; |
1521 | unsigned long end_index = end >> PAGE_SHIFT; | |
c8b97818 | 1522 | |
76c0021d | 1523 | ASSERT(locked_page); |
c8b97818 | 1524 | if (index == locked_page->index && end_index == index) |
143bede5 | 1525 | return; |
c8b97818 | 1526 | |
76c0021d LB |
1527 | __process_pages_contig(inode->i_mapping, locked_page, index, end_index, |
1528 | PAGE_UNLOCK, NULL); | |
c8b97818 CM |
1529 | } |
1530 | ||
1531 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1532 | struct page *locked_page, | |
1533 | u64 delalloc_start, | |
1534 | u64 delalloc_end) | |
1535 | { | |
09cbfeaf | 1536 | unsigned long index = delalloc_start >> PAGE_SHIFT; |
76c0021d | 1537 | unsigned long index_ret = index; |
09cbfeaf | 1538 | unsigned long end_index = delalloc_end >> PAGE_SHIFT; |
c8b97818 | 1539 | int ret; |
c8b97818 | 1540 | |
76c0021d | 1541 | ASSERT(locked_page); |
c8b97818 CM |
1542 | if (index == locked_page->index && index == end_index) |
1543 | return 0; | |
1544 | ||
76c0021d LB |
1545 | ret = __process_pages_contig(inode->i_mapping, locked_page, index, |
1546 | end_index, PAGE_LOCK, &index_ret); | |
1547 | if (ret == -EAGAIN) | |
1548 | __unlock_for_delalloc(inode, locked_page, delalloc_start, | |
1549 | (u64)index_ret << PAGE_SHIFT); | |
c8b97818 CM |
1550 | return ret; |
1551 | } | |
1552 | ||
1553 | /* | |
1554 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1555 | * more than 'max_bytes'. start and end are used to return the range, | |
1556 | * | |
1557 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1558 | */ | |
9c36396c | 1559 | static noinline_for_stack u64 find_lock_delalloc_range(struct inode *inode, |
294e30fe JB |
1560 | struct extent_io_tree *tree, |
1561 | struct page *locked_page, u64 *start, | |
1562 | u64 *end, u64 max_bytes) | |
c8b97818 CM |
1563 | { |
1564 | u64 delalloc_start; | |
1565 | u64 delalloc_end; | |
1566 | u64 found; | |
9655d298 | 1567 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1568 | int ret; |
1569 | int loops = 0; | |
1570 | ||
1571 | again: | |
1572 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1573 | delalloc_start = *start; | |
1574 | delalloc_end = 0; | |
1575 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1576 | max_bytes, &cached_state); |
70b99e69 | 1577 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1578 | *start = delalloc_start; |
1579 | *end = delalloc_end; | |
c2a128d2 | 1580 | free_extent_state(cached_state); |
385fe0be | 1581 | return 0; |
c8b97818 CM |
1582 | } |
1583 | ||
70b99e69 CM |
1584 | /* |
1585 | * start comes from the offset of locked_page. We have to lock | |
1586 | * pages in order, so we can't process delalloc bytes before | |
1587 | * locked_page | |
1588 | */ | |
d397712b | 1589 | if (delalloc_start < *start) |
70b99e69 | 1590 | delalloc_start = *start; |
70b99e69 | 1591 | |
c8b97818 CM |
1592 | /* |
1593 | * make sure to limit the number of pages we try to lock down | |
c8b97818 | 1594 | */ |
7bf811a5 JB |
1595 | if (delalloc_end + 1 - delalloc_start > max_bytes) |
1596 | delalloc_end = delalloc_start + max_bytes - 1; | |
d397712b | 1597 | |
c8b97818 CM |
1598 | /* step two, lock all the pages after the page that has start */ |
1599 | ret = lock_delalloc_pages(inode, locked_page, | |
1600 | delalloc_start, delalloc_end); | |
1601 | if (ret == -EAGAIN) { | |
1602 | /* some of the pages are gone, lets avoid looping by | |
1603 | * shortening the size of the delalloc range we're searching | |
1604 | */ | |
9655d298 | 1605 | free_extent_state(cached_state); |
7d788742 | 1606 | cached_state = NULL; |
c8b97818 | 1607 | if (!loops) { |
09cbfeaf | 1608 | max_bytes = PAGE_SIZE; |
c8b97818 CM |
1609 | loops = 1; |
1610 | goto again; | |
1611 | } else { | |
1612 | found = 0; | |
1613 | goto out_failed; | |
1614 | } | |
1615 | } | |
79787eaa | 1616 | BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */ |
c8b97818 CM |
1617 | |
1618 | /* step three, lock the state bits for the whole range */ | |
ff13db41 | 1619 | lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state); |
c8b97818 CM |
1620 | |
1621 | /* then test to make sure it is all still delalloc */ | |
1622 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1623 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1624 | if (!ret) { |
9655d298 | 1625 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
e43bbe5e | 1626 | &cached_state); |
c8b97818 CM |
1627 | __unlock_for_delalloc(inode, locked_page, |
1628 | delalloc_start, delalloc_end); | |
1629 | cond_resched(); | |
1630 | goto again; | |
1631 | } | |
9655d298 | 1632 | free_extent_state(cached_state); |
c8b97818 CM |
1633 | *start = delalloc_start; |
1634 | *end = delalloc_end; | |
1635 | out_failed: | |
1636 | return found; | |
1637 | } | |
1638 | ||
9c36396c DS |
1639 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
1640 | u64 btrfs_find_lock_delalloc_range(struct inode *inode, | |
1641 | struct extent_io_tree *tree, | |
1642 | struct page *locked_page, u64 *start, | |
1643 | u64 *end, u64 max_bytes) | |
1644 | { | |
1645 | return find_lock_delalloc_range(inode, tree, locked_page, start, end, | |
1646 | max_bytes); | |
1647 | } | |
1648 | #endif | |
1649 | ||
da2c7009 LB |
1650 | static int __process_pages_contig(struct address_space *mapping, |
1651 | struct page *locked_page, | |
1652 | pgoff_t start_index, pgoff_t end_index, | |
1653 | unsigned long page_ops, pgoff_t *index_ret) | |
c8b97818 | 1654 | { |
873695b3 | 1655 | unsigned long nr_pages = end_index - start_index + 1; |
da2c7009 | 1656 | unsigned long pages_locked = 0; |
873695b3 | 1657 | pgoff_t index = start_index; |
c8b97818 | 1658 | struct page *pages[16]; |
873695b3 | 1659 | unsigned ret; |
da2c7009 | 1660 | int err = 0; |
c8b97818 | 1661 | int i; |
771ed689 | 1662 | |
da2c7009 LB |
1663 | if (page_ops & PAGE_LOCK) { |
1664 | ASSERT(page_ops == PAGE_LOCK); | |
1665 | ASSERT(index_ret && *index_ret == start_index); | |
1666 | } | |
1667 | ||
704de49d | 1668 | if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0) |
873695b3 | 1669 | mapping_set_error(mapping, -EIO); |
704de49d | 1670 | |
d397712b | 1671 | while (nr_pages > 0) { |
873695b3 | 1672 | ret = find_get_pages_contig(mapping, index, |
5b050f04 CM |
1673 | min_t(unsigned long, |
1674 | nr_pages, ARRAY_SIZE(pages)), pages); | |
da2c7009 LB |
1675 | if (ret == 0) { |
1676 | /* | |
1677 | * Only if we're going to lock these pages, | |
1678 | * can we find nothing at @index. | |
1679 | */ | |
1680 | ASSERT(page_ops & PAGE_LOCK); | |
49d4a334 LB |
1681 | err = -EAGAIN; |
1682 | goto out; | |
da2c7009 | 1683 | } |
8b62b72b | 1684 | |
da2c7009 | 1685 | for (i = 0; i < ret; i++) { |
c2790a2e | 1686 | if (page_ops & PAGE_SET_PRIVATE2) |
8b62b72b CM |
1687 | SetPagePrivate2(pages[i]); |
1688 | ||
c8b97818 | 1689 | if (pages[i] == locked_page) { |
09cbfeaf | 1690 | put_page(pages[i]); |
da2c7009 | 1691 | pages_locked++; |
c8b97818 CM |
1692 | continue; |
1693 | } | |
c2790a2e | 1694 | if (page_ops & PAGE_CLEAR_DIRTY) |
c8b97818 | 1695 | clear_page_dirty_for_io(pages[i]); |
c2790a2e | 1696 | if (page_ops & PAGE_SET_WRITEBACK) |
c8b97818 | 1697 | set_page_writeback(pages[i]); |
704de49d FM |
1698 | if (page_ops & PAGE_SET_ERROR) |
1699 | SetPageError(pages[i]); | |
c2790a2e | 1700 | if (page_ops & PAGE_END_WRITEBACK) |
c8b97818 | 1701 | end_page_writeback(pages[i]); |
c2790a2e | 1702 | if (page_ops & PAGE_UNLOCK) |
771ed689 | 1703 | unlock_page(pages[i]); |
da2c7009 LB |
1704 | if (page_ops & PAGE_LOCK) { |
1705 | lock_page(pages[i]); | |
1706 | if (!PageDirty(pages[i]) || | |
1707 | pages[i]->mapping != mapping) { | |
1708 | unlock_page(pages[i]); | |
1709 | put_page(pages[i]); | |
1710 | err = -EAGAIN; | |
1711 | goto out; | |
1712 | } | |
1713 | } | |
09cbfeaf | 1714 | put_page(pages[i]); |
da2c7009 | 1715 | pages_locked++; |
c8b97818 CM |
1716 | } |
1717 | nr_pages -= ret; | |
1718 | index += ret; | |
1719 | cond_resched(); | |
1720 | } | |
da2c7009 LB |
1721 | out: |
1722 | if (err && index_ret) | |
1723 | *index_ret = start_index + pages_locked - 1; | |
1724 | return err; | |
c8b97818 | 1725 | } |
c8b97818 | 1726 | |
873695b3 LB |
1727 | void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end, |
1728 | u64 delalloc_end, struct page *locked_page, | |
1729 | unsigned clear_bits, | |
1730 | unsigned long page_ops) | |
1731 | { | |
1732 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0, | |
ae0f1625 | 1733 | NULL); |
873695b3 LB |
1734 | |
1735 | __process_pages_contig(inode->i_mapping, locked_page, | |
1736 | start >> PAGE_SHIFT, end >> PAGE_SHIFT, | |
da2c7009 | 1737 | page_ops, NULL); |
873695b3 LB |
1738 | } |
1739 | ||
d352ac68 CM |
1740 | /* |
1741 | * count the number of bytes in the tree that have a given bit(s) | |
1742 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1743 | * cached. The total number found is returned. | |
1744 | */ | |
d1310b2e CM |
1745 | u64 count_range_bits(struct extent_io_tree *tree, |
1746 | u64 *start, u64 search_end, u64 max_bytes, | |
9ee49a04 | 1747 | unsigned bits, int contig) |
d1310b2e CM |
1748 | { |
1749 | struct rb_node *node; | |
1750 | struct extent_state *state; | |
1751 | u64 cur_start = *start; | |
1752 | u64 total_bytes = 0; | |
ec29ed5b | 1753 | u64 last = 0; |
d1310b2e CM |
1754 | int found = 0; |
1755 | ||
fae7f21c | 1756 | if (WARN_ON(search_end <= cur_start)) |
d1310b2e | 1757 | return 0; |
d1310b2e | 1758 | |
cad321ad | 1759 | spin_lock(&tree->lock); |
d1310b2e CM |
1760 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1761 | total_bytes = tree->dirty_bytes; | |
1762 | goto out; | |
1763 | } | |
1764 | /* | |
1765 | * this search will find all the extents that end after | |
1766 | * our range starts. | |
1767 | */ | |
80ea96b1 | 1768 | node = tree_search(tree, cur_start); |
d397712b | 1769 | if (!node) |
d1310b2e | 1770 | goto out; |
d1310b2e | 1771 | |
d397712b | 1772 | while (1) { |
d1310b2e CM |
1773 | state = rb_entry(node, struct extent_state, rb_node); |
1774 | if (state->start > search_end) | |
1775 | break; | |
ec29ed5b CM |
1776 | if (contig && found && state->start > last + 1) |
1777 | break; | |
1778 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1779 | total_bytes += min(search_end, state->end) + 1 - |
1780 | max(cur_start, state->start); | |
1781 | if (total_bytes >= max_bytes) | |
1782 | break; | |
1783 | if (!found) { | |
af60bed2 | 1784 | *start = max(cur_start, state->start); |
d1310b2e CM |
1785 | found = 1; |
1786 | } | |
ec29ed5b CM |
1787 | last = state->end; |
1788 | } else if (contig && found) { | |
1789 | break; | |
d1310b2e CM |
1790 | } |
1791 | node = rb_next(node); | |
1792 | if (!node) | |
1793 | break; | |
1794 | } | |
1795 | out: | |
cad321ad | 1796 | spin_unlock(&tree->lock); |
d1310b2e CM |
1797 | return total_bytes; |
1798 | } | |
b2950863 | 1799 | |
d352ac68 CM |
1800 | /* |
1801 | * set the private field for a given byte offset in the tree. If there isn't | |
1802 | * an extent_state there already, this does nothing. | |
1803 | */ | |
f827ba9a | 1804 | static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start, |
47dc196a | 1805 | struct io_failure_record *failrec) |
d1310b2e CM |
1806 | { |
1807 | struct rb_node *node; | |
1808 | struct extent_state *state; | |
1809 | int ret = 0; | |
1810 | ||
cad321ad | 1811 | spin_lock(&tree->lock); |
d1310b2e CM |
1812 | /* |
1813 | * this search will find all the extents that end after | |
1814 | * our range starts. | |
1815 | */ | |
80ea96b1 | 1816 | node = tree_search(tree, start); |
2b114d1d | 1817 | if (!node) { |
d1310b2e CM |
1818 | ret = -ENOENT; |
1819 | goto out; | |
1820 | } | |
1821 | state = rb_entry(node, struct extent_state, rb_node); | |
1822 | if (state->start != start) { | |
1823 | ret = -ENOENT; | |
1824 | goto out; | |
1825 | } | |
47dc196a | 1826 | state->failrec = failrec; |
d1310b2e | 1827 | out: |
cad321ad | 1828 | spin_unlock(&tree->lock); |
d1310b2e CM |
1829 | return ret; |
1830 | } | |
1831 | ||
f827ba9a | 1832 | static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start, |
47dc196a | 1833 | struct io_failure_record **failrec) |
d1310b2e CM |
1834 | { |
1835 | struct rb_node *node; | |
1836 | struct extent_state *state; | |
1837 | int ret = 0; | |
1838 | ||
cad321ad | 1839 | spin_lock(&tree->lock); |
d1310b2e CM |
1840 | /* |
1841 | * this search will find all the extents that end after | |
1842 | * our range starts. | |
1843 | */ | |
80ea96b1 | 1844 | node = tree_search(tree, start); |
2b114d1d | 1845 | if (!node) { |
d1310b2e CM |
1846 | ret = -ENOENT; |
1847 | goto out; | |
1848 | } | |
1849 | state = rb_entry(node, struct extent_state, rb_node); | |
1850 | if (state->start != start) { | |
1851 | ret = -ENOENT; | |
1852 | goto out; | |
1853 | } | |
47dc196a | 1854 | *failrec = state->failrec; |
d1310b2e | 1855 | out: |
cad321ad | 1856 | spin_unlock(&tree->lock); |
d1310b2e CM |
1857 | return ret; |
1858 | } | |
1859 | ||
1860 | /* | |
1861 | * searches a range in the state tree for a given mask. | |
70dec807 | 1862 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1863 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1864 | * range is found set. | |
1865 | */ | |
1866 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1867 | unsigned bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1868 | { |
1869 | struct extent_state *state = NULL; | |
1870 | struct rb_node *node; | |
1871 | int bitset = 0; | |
d1310b2e | 1872 | |
cad321ad | 1873 | spin_lock(&tree->lock); |
27a3507d | 1874 | if (cached && extent_state_in_tree(cached) && cached->start <= start && |
df98b6e2 | 1875 | cached->end > start) |
9655d298 CM |
1876 | node = &cached->rb_node; |
1877 | else | |
1878 | node = tree_search(tree, start); | |
d1310b2e CM |
1879 | while (node && start <= end) { |
1880 | state = rb_entry(node, struct extent_state, rb_node); | |
1881 | ||
1882 | if (filled && state->start > start) { | |
1883 | bitset = 0; | |
1884 | break; | |
1885 | } | |
1886 | ||
1887 | if (state->start > end) | |
1888 | break; | |
1889 | ||
1890 | if (state->state & bits) { | |
1891 | bitset = 1; | |
1892 | if (!filled) | |
1893 | break; | |
1894 | } else if (filled) { | |
1895 | bitset = 0; | |
1896 | break; | |
1897 | } | |
46562cec CM |
1898 | |
1899 | if (state->end == (u64)-1) | |
1900 | break; | |
1901 | ||
d1310b2e CM |
1902 | start = state->end + 1; |
1903 | if (start > end) | |
1904 | break; | |
1905 | node = rb_next(node); | |
1906 | if (!node) { | |
1907 | if (filled) | |
1908 | bitset = 0; | |
1909 | break; | |
1910 | } | |
1911 | } | |
cad321ad | 1912 | spin_unlock(&tree->lock); |
d1310b2e CM |
1913 | return bitset; |
1914 | } | |
d1310b2e CM |
1915 | |
1916 | /* | |
1917 | * helper function to set a given page up to date if all the | |
1918 | * extents in the tree for that page are up to date | |
1919 | */ | |
143bede5 | 1920 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e | 1921 | { |
4eee4fa4 | 1922 | u64 start = page_offset(page); |
09cbfeaf | 1923 | u64 end = start + PAGE_SIZE - 1; |
9655d298 | 1924 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 1925 | SetPageUptodate(page); |
d1310b2e CM |
1926 | } |
1927 | ||
7870d082 JB |
1928 | int free_io_failure(struct extent_io_tree *failure_tree, |
1929 | struct extent_io_tree *io_tree, | |
1930 | struct io_failure_record *rec) | |
4a54c8c1 JS |
1931 | { |
1932 | int ret; | |
1933 | int err = 0; | |
4a54c8c1 | 1934 | |
47dc196a | 1935 | set_state_failrec(failure_tree, rec->start, NULL); |
4a54c8c1 JS |
1936 | ret = clear_extent_bits(failure_tree, rec->start, |
1937 | rec->start + rec->len - 1, | |
91166212 | 1938 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 JS |
1939 | if (ret) |
1940 | err = ret; | |
1941 | ||
7870d082 | 1942 | ret = clear_extent_bits(io_tree, rec->start, |
53b381b3 | 1943 | rec->start + rec->len - 1, |
91166212 | 1944 | EXTENT_DAMAGED); |
53b381b3 DW |
1945 | if (ret && !err) |
1946 | err = ret; | |
4a54c8c1 JS |
1947 | |
1948 | kfree(rec); | |
1949 | return err; | |
1950 | } | |
1951 | ||
4a54c8c1 JS |
1952 | /* |
1953 | * this bypasses the standard btrfs submit functions deliberately, as | |
1954 | * the standard behavior is to write all copies in a raid setup. here we only | |
1955 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
1956 | * submit_bio directly. | |
3ec706c8 | 1957 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
1958 | * actually prevents the read that triggered the error from finishing. |
1959 | * currently, there can be no more than two copies of every data bit. thus, | |
1960 | * exactly one rewrite is required. | |
1961 | */ | |
6ec656bc JB |
1962 | int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start, |
1963 | u64 length, u64 logical, struct page *page, | |
1964 | unsigned int pg_offset, int mirror_num) | |
4a54c8c1 JS |
1965 | { |
1966 | struct bio *bio; | |
1967 | struct btrfs_device *dev; | |
4a54c8c1 JS |
1968 | u64 map_length = 0; |
1969 | u64 sector; | |
1970 | struct btrfs_bio *bbio = NULL; | |
1971 | int ret; | |
1972 | ||
1751e8a6 | 1973 | ASSERT(!(fs_info->sb->s_flags & SB_RDONLY)); |
4a54c8c1 JS |
1974 | BUG_ON(!mirror_num); |
1975 | ||
c5e4c3d7 | 1976 | bio = btrfs_io_bio_alloc(1); |
4f024f37 | 1977 | bio->bi_iter.bi_size = 0; |
4a54c8c1 JS |
1978 | map_length = length; |
1979 | ||
b5de8d0d FM |
1980 | /* |
1981 | * Avoid races with device replace and make sure our bbio has devices | |
1982 | * associated to its stripes that don't go away while we are doing the | |
1983 | * read repair operation. | |
1984 | */ | |
1985 | btrfs_bio_counter_inc_blocked(fs_info); | |
e4ff5fb5 | 1986 | if (btrfs_is_parity_mirror(fs_info, logical, length)) { |
c725328c LB |
1987 | /* |
1988 | * Note that we don't use BTRFS_MAP_WRITE because it's supposed | |
1989 | * to update all raid stripes, but here we just want to correct | |
1990 | * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad | |
1991 | * stripe's dev and sector. | |
1992 | */ | |
1993 | ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical, | |
1994 | &map_length, &bbio, 0); | |
1995 | if (ret) { | |
1996 | btrfs_bio_counter_dec(fs_info); | |
1997 | bio_put(bio); | |
1998 | return -EIO; | |
1999 | } | |
2000 | ASSERT(bbio->mirror_num == 1); | |
2001 | } else { | |
2002 | ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, | |
2003 | &map_length, &bbio, mirror_num); | |
2004 | if (ret) { | |
2005 | btrfs_bio_counter_dec(fs_info); | |
2006 | bio_put(bio); | |
2007 | return -EIO; | |
2008 | } | |
2009 | BUG_ON(mirror_num != bbio->mirror_num); | |
4a54c8c1 | 2010 | } |
c725328c LB |
2011 | |
2012 | sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9; | |
4f024f37 | 2013 | bio->bi_iter.bi_sector = sector; |
c725328c | 2014 | dev = bbio->stripes[bbio->mirror_num - 1].dev; |
6e9606d2 | 2015 | btrfs_put_bbio(bbio); |
ebbede42 AJ |
2016 | if (!dev || !dev->bdev || |
2017 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) { | |
b5de8d0d | 2018 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2019 | bio_put(bio); |
2020 | return -EIO; | |
2021 | } | |
74d46992 | 2022 | bio_set_dev(bio, dev->bdev); |
70fd7614 | 2023 | bio->bi_opf = REQ_OP_WRITE | REQ_SYNC; |
ffdd2018 | 2024 | bio_add_page(bio, page, length, pg_offset); |
4a54c8c1 | 2025 | |
4e49ea4a | 2026 | if (btrfsic_submit_bio_wait(bio)) { |
4a54c8c1 | 2027 | /* try to remap that extent elsewhere? */ |
b5de8d0d | 2028 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 | 2029 | bio_put(bio); |
442a4f63 | 2030 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
2031 | return -EIO; |
2032 | } | |
2033 | ||
b14af3b4 DS |
2034 | btrfs_info_rl_in_rcu(fs_info, |
2035 | "read error corrected: ino %llu off %llu (dev %s sector %llu)", | |
6ec656bc | 2036 | ino, start, |
1203b681 | 2037 | rcu_str_deref(dev->name), sector); |
b5de8d0d | 2038 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2039 | bio_put(bio); |
2040 | return 0; | |
2041 | } | |
2042 | ||
2ff7e61e JM |
2043 | int repair_eb_io_failure(struct btrfs_fs_info *fs_info, |
2044 | struct extent_buffer *eb, int mirror_num) | |
ea466794 | 2045 | { |
ea466794 | 2046 | u64 start = eb->start; |
cc5e31a4 | 2047 | int i, num_pages = num_extent_pages(eb); |
d95603b2 | 2048 | int ret = 0; |
ea466794 | 2049 | |
bc98a42c | 2050 | if (sb_rdonly(fs_info->sb)) |
908960c6 ID |
2051 | return -EROFS; |
2052 | ||
ea466794 | 2053 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 2054 | struct page *p = eb->pages[i]; |
1203b681 | 2055 | |
6ec656bc | 2056 | ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p, |
1203b681 | 2057 | start - page_offset(p), mirror_num); |
ea466794 JB |
2058 | if (ret) |
2059 | break; | |
09cbfeaf | 2060 | start += PAGE_SIZE; |
ea466794 JB |
2061 | } |
2062 | ||
2063 | return ret; | |
2064 | } | |
2065 | ||
4a54c8c1 JS |
2066 | /* |
2067 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2068 | * to see if we need to process or clean up an io_failure_record | |
2069 | */ | |
7870d082 JB |
2070 | int clean_io_failure(struct btrfs_fs_info *fs_info, |
2071 | struct extent_io_tree *failure_tree, | |
2072 | struct extent_io_tree *io_tree, u64 start, | |
2073 | struct page *page, u64 ino, unsigned int pg_offset) | |
4a54c8c1 JS |
2074 | { |
2075 | u64 private; | |
4a54c8c1 | 2076 | struct io_failure_record *failrec; |
4a54c8c1 JS |
2077 | struct extent_state *state; |
2078 | int num_copies; | |
4a54c8c1 | 2079 | int ret; |
4a54c8c1 JS |
2080 | |
2081 | private = 0; | |
7870d082 JB |
2082 | ret = count_range_bits(failure_tree, &private, (u64)-1, 1, |
2083 | EXTENT_DIRTY, 0); | |
4a54c8c1 JS |
2084 | if (!ret) |
2085 | return 0; | |
2086 | ||
7870d082 | 2087 | ret = get_state_failrec(failure_tree, start, &failrec); |
4a54c8c1 JS |
2088 | if (ret) |
2089 | return 0; | |
2090 | ||
4a54c8c1 JS |
2091 | BUG_ON(!failrec->this_mirror); |
2092 | ||
2093 | if (failrec->in_validation) { | |
2094 | /* there was no real error, just free the record */ | |
ab8d0fc4 JM |
2095 | btrfs_debug(fs_info, |
2096 | "clean_io_failure: freeing dummy error at %llu", | |
2097 | failrec->start); | |
4a54c8c1 JS |
2098 | goto out; |
2099 | } | |
bc98a42c | 2100 | if (sb_rdonly(fs_info->sb)) |
908960c6 | 2101 | goto out; |
4a54c8c1 | 2102 | |
7870d082 JB |
2103 | spin_lock(&io_tree->lock); |
2104 | state = find_first_extent_bit_state(io_tree, | |
4a54c8c1 JS |
2105 | failrec->start, |
2106 | EXTENT_LOCKED); | |
7870d082 | 2107 | spin_unlock(&io_tree->lock); |
4a54c8c1 | 2108 | |
883d0de4 MX |
2109 | if (state && state->start <= failrec->start && |
2110 | state->end >= failrec->start + failrec->len - 1) { | |
3ec706c8 SB |
2111 | num_copies = btrfs_num_copies(fs_info, failrec->logical, |
2112 | failrec->len); | |
4a54c8c1 | 2113 | if (num_copies > 1) { |
7870d082 JB |
2114 | repair_io_failure(fs_info, ino, start, failrec->len, |
2115 | failrec->logical, page, pg_offset, | |
2116 | failrec->failed_mirror); | |
4a54c8c1 JS |
2117 | } |
2118 | } | |
2119 | ||
2120 | out: | |
7870d082 | 2121 | free_io_failure(failure_tree, io_tree, failrec); |
4a54c8c1 | 2122 | |
454ff3de | 2123 | return 0; |
4a54c8c1 JS |
2124 | } |
2125 | ||
f612496b MX |
2126 | /* |
2127 | * Can be called when | |
2128 | * - hold extent lock | |
2129 | * - under ordered extent | |
2130 | * - the inode is freeing | |
2131 | */ | |
7ab7956e | 2132 | void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end) |
f612496b | 2133 | { |
7ab7956e | 2134 | struct extent_io_tree *failure_tree = &inode->io_failure_tree; |
f612496b MX |
2135 | struct io_failure_record *failrec; |
2136 | struct extent_state *state, *next; | |
2137 | ||
2138 | if (RB_EMPTY_ROOT(&failure_tree->state)) | |
2139 | return; | |
2140 | ||
2141 | spin_lock(&failure_tree->lock); | |
2142 | state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY); | |
2143 | while (state) { | |
2144 | if (state->start > end) | |
2145 | break; | |
2146 | ||
2147 | ASSERT(state->end <= end); | |
2148 | ||
2149 | next = next_state(state); | |
2150 | ||
47dc196a | 2151 | failrec = state->failrec; |
f612496b MX |
2152 | free_extent_state(state); |
2153 | kfree(failrec); | |
2154 | ||
2155 | state = next; | |
2156 | } | |
2157 | spin_unlock(&failure_tree->lock); | |
2158 | } | |
2159 | ||
2fe6303e | 2160 | int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end, |
47dc196a | 2161 | struct io_failure_record **failrec_ret) |
4a54c8c1 | 2162 | { |
ab8d0fc4 | 2163 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e | 2164 | struct io_failure_record *failrec; |
4a54c8c1 | 2165 | struct extent_map *em; |
4a54c8c1 JS |
2166 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2167 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2168 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
4a54c8c1 | 2169 | int ret; |
4a54c8c1 JS |
2170 | u64 logical; |
2171 | ||
47dc196a | 2172 | ret = get_state_failrec(failure_tree, start, &failrec); |
4a54c8c1 JS |
2173 | if (ret) { |
2174 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2175 | if (!failrec) | |
2176 | return -ENOMEM; | |
2fe6303e | 2177 | |
4a54c8c1 JS |
2178 | failrec->start = start; |
2179 | failrec->len = end - start + 1; | |
2180 | failrec->this_mirror = 0; | |
2181 | failrec->bio_flags = 0; | |
2182 | failrec->in_validation = 0; | |
2183 | ||
2184 | read_lock(&em_tree->lock); | |
2185 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2186 | if (!em) { | |
2187 | read_unlock(&em_tree->lock); | |
2188 | kfree(failrec); | |
2189 | return -EIO; | |
2190 | } | |
2191 | ||
68ba990f | 2192 | if (em->start > start || em->start + em->len <= start) { |
4a54c8c1 JS |
2193 | free_extent_map(em); |
2194 | em = NULL; | |
2195 | } | |
2196 | read_unlock(&em_tree->lock); | |
7a2d6a64 | 2197 | if (!em) { |
4a54c8c1 JS |
2198 | kfree(failrec); |
2199 | return -EIO; | |
2200 | } | |
2fe6303e | 2201 | |
4a54c8c1 JS |
2202 | logical = start - em->start; |
2203 | logical = em->block_start + logical; | |
2204 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2205 | logical = em->block_start; | |
2206 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2207 | extent_set_compress_type(&failrec->bio_flags, | |
2208 | em->compress_type); | |
2209 | } | |
2fe6303e | 2210 | |
ab8d0fc4 JM |
2211 | btrfs_debug(fs_info, |
2212 | "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu", | |
2213 | logical, start, failrec->len); | |
2fe6303e | 2214 | |
4a54c8c1 JS |
2215 | failrec->logical = logical; |
2216 | free_extent_map(em); | |
2217 | ||
2218 | /* set the bits in the private failure tree */ | |
2219 | ret = set_extent_bits(failure_tree, start, end, | |
ceeb0ae7 | 2220 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 | 2221 | if (ret >= 0) |
47dc196a | 2222 | ret = set_state_failrec(failure_tree, start, failrec); |
4a54c8c1 JS |
2223 | /* set the bits in the inode's tree */ |
2224 | if (ret >= 0) | |
ceeb0ae7 | 2225 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED); |
4a54c8c1 JS |
2226 | if (ret < 0) { |
2227 | kfree(failrec); | |
2228 | return ret; | |
2229 | } | |
2230 | } else { | |
ab8d0fc4 JM |
2231 | btrfs_debug(fs_info, |
2232 | "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d", | |
2233 | failrec->logical, failrec->start, failrec->len, | |
2234 | failrec->in_validation); | |
4a54c8c1 JS |
2235 | /* |
2236 | * when data can be on disk more than twice, add to failrec here | |
2237 | * (e.g. with a list for failed_mirror) to make | |
2238 | * clean_io_failure() clean all those errors at once. | |
2239 | */ | |
2240 | } | |
2fe6303e MX |
2241 | |
2242 | *failrec_ret = failrec; | |
2243 | ||
2244 | return 0; | |
2245 | } | |
2246 | ||
a0b60d72 | 2247 | bool btrfs_check_repairable(struct inode *inode, unsigned failed_bio_pages, |
2fe6303e MX |
2248 | struct io_failure_record *failrec, int failed_mirror) |
2249 | { | |
ab8d0fc4 | 2250 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e MX |
2251 | int num_copies; |
2252 | ||
ab8d0fc4 | 2253 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
4a54c8c1 JS |
2254 | if (num_copies == 1) { |
2255 | /* | |
2256 | * we only have a single copy of the data, so don't bother with | |
2257 | * all the retry and error correction code that follows. no | |
2258 | * matter what the error is, it is very likely to persist. | |
2259 | */ | |
ab8d0fc4 JM |
2260 | btrfs_debug(fs_info, |
2261 | "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
2262 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2263 | return false; |
4a54c8c1 JS |
2264 | } |
2265 | ||
4a54c8c1 JS |
2266 | /* |
2267 | * there are two premises: | |
2268 | * a) deliver good data to the caller | |
2269 | * b) correct the bad sectors on disk | |
2270 | */ | |
a0b60d72 | 2271 | if (failed_bio_pages > 1) { |
4a54c8c1 JS |
2272 | /* |
2273 | * to fulfill b), we need to know the exact failing sectors, as | |
2274 | * we don't want to rewrite any more than the failed ones. thus, | |
2275 | * we need separate read requests for the failed bio | |
2276 | * | |
2277 | * if the following BUG_ON triggers, our validation request got | |
2278 | * merged. we need separate requests for our algorithm to work. | |
2279 | */ | |
2280 | BUG_ON(failrec->in_validation); | |
2281 | failrec->in_validation = 1; | |
2282 | failrec->this_mirror = failed_mirror; | |
4a54c8c1 JS |
2283 | } else { |
2284 | /* | |
2285 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2286 | * of the failed sector and if we succeed, we have setup | |
2287 | * everything for repair_io_failure to do the rest for us. | |
2288 | */ | |
2289 | if (failrec->in_validation) { | |
2290 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2291 | failrec->in_validation = 0; | |
2292 | failrec->this_mirror = 0; | |
2293 | } | |
2294 | failrec->failed_mirror = failed_mirror; | |
2295 | failrec->this_mirror++; | |
2296 | if (failrec->this_mirror == failed_mirror) | |
2297 | failrec->this_mirror++; | |
4a54c8c1 JS |
2298 | } |
2299 | ||
facc8a22 | 2300 | if (failrec->this_mirror > num_copies) { |
ab8d0fc4 JM |
2301 | btrfs_debug(fs_info, |
2302 | "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
2303 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2304 | return false; |
4a54c8c1 JS |
2305 | } |
2306 | ||
c3cfb656 | 2307 | return true; |
2fe6303e MX |
2308 | } |
2309 | ||
2310 | ||
2311 | struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio, | |
2312 | struct io_failure_record *failrec, | |
2313 | struct page *page, int pg_offset, int icsum, | |
8b110e39 | 2314 | bio_end_io_t *endio_func, void *data) |
2fe6303e | 2315 | { |
0b246afa | 2316 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e MX |
2317 | struct bio *bio; |
2318 | struct btrfs_io_bio *btrfs_failed_bio; | |
2319 | struct btrfs_io_bio *btrfs_bio; | |
2320 | ||
c5e4c3d7 | 2321 | bio = btrfs_io_bio_alloc(1); |
2fe6303e | 2322 | bio->bi_end_io = endio_func; |
4f024f37 | 2323 | bio->bi_iter.bi_sector = failrec->logical >> 9; |
74d46992 | 2324 | bio_set_dev(bio, fs_info->fs_devices->latest_bdev); |
4f024f37 | 2325 | bio->bi_iter.bi_size = 0; |
8b110e39 | 2326 | bio->bi_private = data; |
4a54c8c1 | 2327 | |
facc8a22 MX |
2328 | btrfs_failed_bio = btrfs_io_bio(failed_bio); |
2329 | if (btrfs_failed_bio->csum) { | |
facc8a22 MX |
2330 | u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); |
2331 | ||
2332 | btrfs_bio = btrfs_io_bio(bio); | |
2333 | btrfs_bio->csum = btrfs_bio->csum_inline; | |
2fe6303e MX |
2334 | icsum *= csum_size; |
2335 | memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum, | |
facc8a22 MX |
2336 | csum_size); |
2337 | } | |
2338 | ||
2fe6303e MX |
2339 | bio_add_page(bio, page, failrec->len, pg_offset); |
2340 | ||
2341 | return bio; | |
2342 | } | |
2343 | ||
2344 | /* | |
2345 | * this is a generic handler for readpage errors (default | |
2346 | * readpage_io_failed_hook). if other copies exist, read those and write back | |
2347 | * good data to the failed position. does not investigate in remapping the | |
2348 | * failed extent elsewhere, hoping the device will be smart enough to do this as | |
2349 | * needed | |
2350 | */ | |
2351 | ||
2352 | static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset, | |
2353 | struct page *page, u64 start, u64 end, | |
2354 | int failed_mirror) | |
2355 | { | |
2356 | struct io_failure_record *failrec; | |
2357 | struct inode *inode = page->mapping->host; | |
2358 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
7870d082 | 2359 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2fe6303e | 2360 | struct bio *bio; |
70fd7614 | 2361 | int read_mode = 0; |
4e4cbee9 | 2362 | blk_status_t status; |
2fe6303e | 2363 | int ret; |
a0b60d72 | 2364 | unsigned failed_bio_pages = bio_pages_all(failed_bio); |
2fe6303e | 2365 | |
1f7ad75b | 2366 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
2fe6303e MX |
2367 | |
2368 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
2369 | if (ret) | |
2370 | return ret; | |
2371 | ||
a0b60d72 | 2372 | if (!btrfs_check_repairable(inode, failed_bio_pages, failrec, |
c3cfb656 | 2373 | failed_mirror)) { |
7870d082 | 2374 | free_io_failure(failure_tree, tree, failrec); |
2fe6303e MX |
2375 | return -EIO; |
2376 | } | |
2377 | ||
a0b60d72 | 2378 | if (failed_bio_pages > 1) |
70fd7614 | 2379 | read_mode |= REQ_FAILFAST_DEV; |
2fe6303e MX |
2380 | |
2381 | phy_offset >>= inode->i_sb->s_blocksize_bits; | |
2382 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2383 | start - page_offset(page), | |
8b110e39 MX |
2384 | (int)phy_offset, failed_bio->bi_end_io, |
2385 | NULL); | |
ebcc3263 | 2386 | bio->bi_opf = REQ_OP_READ | read_mode; |
4a54c8c1 | 2387 | |
ab8d0fc4 JM |
2388 | btrfs_debug(btrfs_sb(inode->i_sb), |
2389 | "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d", | |
2390 | read_mode, failrec->this_mirror, failrec->in_validation); | |
4a54c8c1 | 2391 | |
8c27cb35 | 2392 | status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror, |
013bd4c3 | 2393 | failrec->bio_flags, 0); |
4e4cbee9 | 2394 | if (status) { |
7870d082 | 2395 | free_io_failure(failure_tree, tree, failrec); |
6c387ab2 | 2396 | bio_put(bio); |
4e4cbee9 | 2397 | ret = blk_status_to_errno(status); |
6c387ab2 MX |
2398 | } |
2399 | ||
013bd4c3 | 2400 | return ret; |
4a54c8c1 JS |
2401 | } |
2402 | ||
d1310b2e CM |
2403 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2404 | ||
b5227c07 | 2405 | void end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
87826df0 JM |
2406 | { |
2407 | int uptodate = (err == 0); | |
3e2426bd | 2408 | int ret = 0; |
87826df0 | 2409 | |
c629732d | 2410 | btrfs_writepage_endio_finish_ordered(page, start, end, uptodate); |
87826df0 | 2411 | |
87826df0 | 2412 | if (!uptodate) { |
87826df0 JM |
2413 | ClearPageUptodate(page); |
2414 | SetPageError(page); | |
bff5baf8 | 2415 | ret = err < 0 ? err : -EIO; |
5dca6eea | 2416 | mapping_set_error(page->mapping, ret); |
87826df0 | 2417 | } |
87826df0 JM |
2418 | } |
2419 | ||
d1310b2e CM |
2420 | /* |
2421 | * after a writepage IO is done, we need to: | |
2422 | * clear the uptodate bits on error | |
2423 | * clear the writeback bits in the extent tree for this IO | |
2424 | * end_page_writeback if the page has no more pending IO | |
2425 | * | |
2426 | * Scheduling is not allowed, so the extent state tree is expected | |
2427 | * to have one and only one object corresponding to this IO. | |
2428 | */ | |
4246a0b6 | 2429 | static void end_bio_extent_writepage(struct bio *bio) |
d1310b2e | 2430 | { |
4e4cbee9 | 2431 | int error = blk_status_to_errno(bio->bi_status); |
2c30c71b | 2432 | struct bio_vec *bvec; |
d1310b2e CM |
2433 | u64 start; |
2434 | u64 end; | |
2c30c71b | 2435 | int i; |
d1310b2e | 2436 | |
c09abff8 | 2437 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2c30c71b | 2438 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2439 | struct page *page = bvec->bv_page; |
0b246afa JM |
2440 | struct inode *inode = page->mapping->host; |
2441 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
902b22f3 | 2442 | |
17a5adcc AO |
2443 | /* We always issue full-page reads, but if some block |
2444 | * in a page fails to read, blk_update_request() will | |
2445 | * advance bv_offset and adjust bv_len to compensate. | |
2446 | * Print a warning for nonzero offsets, and an error | |
2447 | * if they don't add up to a full page. */ | |
09cbfeaf KS |
2448 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { |
2449 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
0b246afa | 2450 | btrfs_err(fs_info, |
efe120a0 FH |
2451 | "partial page write in btrfs with offset %u and length %u", |
2452 | bvec->bv_offset, bvec->bv_len); | |
2453 | else | |
0b246afa | 2454 | btrfs_info(fs_info, |
5d163e0e | 2455 | "incomplete page write in btrfs with offset %u and length %u", |
efe120a0 FH |
2456 | bvec->bv_offset, bvec->bv_len); |
2457 | } | |
d1310b2e | 2458 | |
17a5adcc AO |
2459 | start = page_offset(page); |
2460 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
d1310b2e | 2461 | |
4e4cbee9 | 2462 | end_extent_writepage(page, error, start, end); |
17a5adcc | 2463 | end_page_writeback(page); |
2c30c71b | 2464 | } |
2b1f55b0 | 2465 | |
d1310b2e | 2466 | bio_put(bio); |
d1310b2e CM |
2467 | } |
2468 | ||
883d0de4 MX |
2469 | static void |
2470 | endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len, | |
2471 | int uptodate) | |
2472 | { | |
2473 | struct extent_state *cached = NULL; | |
2474 | u64 end = start + len - 1; | |
2475 | ||
2476 | if (uptodate && tree->track_uptodate) | |
2477 | set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC); | |
d810a4be | 2478 | unlock_extent_cached_atomic(tree, start, end, &cached); |
883d0de4 MX |
2479 | } |
2480 | ||
d1310b2e CM |
2481 | /* |
2482 | * after a readpage IO is done, we need to: | |
2483 | * clear the uptodate bits on error | |
2484 | * set the uptodate bits if things worked | |
2485 | * set the page up to date if all extents in the tree are uptodate | |
2486 | * clear the lock bit in the extent tree | |
2487 | * unlock the page if there are no other extents locked for it | |
2488 | * | |
2489 | * Scheduling is not allowed, so the extent state tree is expected | |
2490 | * to have one and only one object corresponding to this IO. | |
2491 | */ | |
4246a0b6 | 2492 | static void end_bio_extent_readpage(struct bio *bio) |
d1310b2e | 2493 | { |
2c30c71b | 2494 | struct bio_vec *bvec; |
4e4cbee9 | 2495 | int uptodate = !bio->bi_status; |
facc8a22 | 2496 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); |
7870d082 | 2497 | struct extent_io_tree *tree, *failure_tree; |
facc8a22 | 2498 | u64 offset = 0; |
d1310b2e CM |
2499 | u64 start; |
2500 | u64 end; | |
facc8a22 | 2501 | u64 len; |
883d0de4 MX |
2502 | u64 extent_start = 0; |
2503 | u64 extent_len = 0; | |
5cf1ab56 | 2504 | int mirror; |
d1310b2e | 2505 | int ret; |
2c30c71b | 2506 | int i; |
d1310b2e | 2507 | |
c09abff8 | 2508 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2c30c71b | 2509 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2510 | struct page *page = bvec->bv_page; |
a71754fc | 2511 | struct inode *inode = page->mapping->host; |
ab8d0fc4 | 2512 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
507903b8 | 2513 | |
ab8d0fc4 JM |
2514 | btrfs_debug(fs_info, |
2515 | "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u", | |
4e4cbee9 | 2516 | (u64)bio->bi_iter.bi_sector, bio->bi_status, |
ab8d0fc4 | 2517 | io_bio->mirror_num); |
a71754fc | 2518 | tree = &BTRFS_I(inode)->io_tree; |
7870d082 | 2519 | failure_tree = &BTRFS_I(inode)->io_failure_tree; |
902b22f3 | 2520 | |
17a5adcc AO |
2521 | /* We always issue full-page reads, but if some block |
2522 | * in a page fails to read, blk_update_request() will | |
2523 | * advance bv_offset and adjust bv_len to compensate. | |
2524 | * Print a warning for nonzero offsets, and an error | |
2525 | * if they don't add up to a full page. */ | |
09cbfeaf KS |
2526 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { |
2527 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
ab8d0fc4 JM |
2528 | btrfs_err(fs_info, |
2529 | "partial page read in btrfs with offset %u and length %u", | |
efe120a0 FH |
2530 | bvec->bv_offset, bvec->bv_len); |
2531 | else | |
ab8d0fc4 JM |
2532 | btrfs_info(fs_info, |
2533 | "incomplete page read in btrfs with offset %u and length %u", | |
efe120a0 FH |
2534 | bvec->bv_offset, bvec->bv_len); |
2535 | } | |
d1310b2e | 2536 | |
17a5adcc AO |
2537 | start = page_offset(page); |
2538 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
facc8a22 | 2539 | len = bvec->bv_len; |
d1310b2e | 2540 | |
9be3395b | 2541 | mirror = io_bio->mirror_num; |
20c9801d | 2542 | if (likely(uptodate && tree->ops)) { |
facc8a22 MX |
2543 | ret = tree->ops->readpage_end_io_hook(io_bio, offset, |
2544 | page, start, end, | |
2545 | mirror); | |
5ee0844d | 2546 | if (ret) |
d1310b2e | 2547 | uptodate = 0; |
5ee0844d | 2548 | else |
7870d082 JB |
2549 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
2550 | failure_tree, tree, start, | |
2551 | page, | |
2552 | btrfs_ino(BTRFS_I(inode)), 0); | |
d1310b2e | 2553 | } |
ea466794 | 2554 | |
f2a09da9 MX |
2555 | if (likely(uptodate)) |
2556 | goto readpage_ok; | |
2557 | ||
20a7db8a | 2558 | if (tree->ops) { |
5cf1ab56 | 2559 | ret = tree->ops->readpage_io_failed_hook(page, mirror); |
9d0d1c8b LB |
2560 | if (ret == -EAGAIN) { |
2561 | /* | |
2562 | * Data inode's readpage_io_failed_hook() always | |
2563 | * returns -EAGAIN. | |
2564 | * | |
2565 | * The generic bio_readpage_error handles errors | |
2566 | * the following way: If possible, new read | |
2567 | * requests are created and submitted and will | |
2568 | * end up in end_bio_extent_readpage as well (if | |
2569 | * we're lucky, not in the !uptodate case). In | |
2570 | * that case it returns 0 and we just go on with | |
2571 | * the next page in our bio. If it can't handle | |
2572 | * the error it will return -EIO and we remain | |
2573 | * responsible for that page. | |
2574 | */ | |
2575 | ret = bio_readpage_error(bio, offset, page, | |
2576 | start, end, mirror); | |
2577 | if (ret == 0) { | |
4e4cbee9 | 2578 | uptodate = !bio->bi_status; |
9d0d1c8b LB |
2579 | offset += len; |
2580 | continue; | |
2581 | } | |
2582 | } | |
2583 | ||
f4a8e656 | 2584 | /* |
9d0d1c8b LB |
2585 | * metadata's readpage_io_failed_hook() always returns |
2586 | * -EIO and fixes nothing. -EIO is also returned if | |
2587 | * data inode error could not be fixed. | |
f4a8e656 | 2588 | */ |
9d0d1c8b | 2589 | ASSERT(ret == -EIO); |
7e38326f | 2590 | } |
f2a09da9 | 2591 | readpage_ok: |
883d0de4 | 2592 | if (likely(uptodate)) { |
a71754fc | 2593 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 2594 | pgoff_t end_index = i_size >> PAGE_SHIFT; |
a583c026 | 2595 | unsigned off; |
a71754fc JB |
2596 | |
2597 | /* Zero out the end if this page straddles i_size */ | |
09cbfeaf | 2598 | off = i_size & (PAGE_SIZE-1); |
a583c026 | 2599 | if (page->index == end_index && off) |
09cbfeaf | 2600 | zero_user_segment(page, off, PAGE_SIZE); |
17a5adcc | 2601 | SetPageUptodate(page); |
70dec807 | 2602 | } else { |
17a5adcc AO |
2603 | ClearPageUptodate(page); |
2604 | SetPageError(page); | |
70dec807 | 2605 | } |
17a5adcc | 2606 | unlock_page(page); |
facc8a22 | 2607 | offset += len; |
883d0de4 MX |
2608 | |
2609 | if (unlikely(!uptodate)) { | |
2610 | if (extent_len) { | |
2611 | endio_readpage_release_extent(tree, | |
2612 | extent_start, | |
2613 | extent_len, 1); | |
2614 | extent_start = 0; | |
2615 | extent_len = 0; | |
2616 | } | |
2617 | endio_readpage_release_extent(tree, start, | |
2618 | end - start + 1, 0); | |
2619 | } else if (!extent_len) { | |
2620 | extent_start = start; | |
2621 | extent_len = end + 1 - start; | |
2622 | } else if (extent_start + extent_len == start) { | |
2623 | extent_len += end + 1 - start; | |
2624 | } else { | |
2625 | endio_readpage_release_extent(tree, extent_start, | |
2626 | extent_len, uptodate); | |
2627 | extent_start = start; | |
2628 | extent_len = end + 1 - start; | |
2629 | } | |
2c30c71b | 2630 | } |
d1310b2e | 2631 | |
883d0de4 MX |
2632 | if (extent_len) |
2633 | endio_readpage_release_extent(tree, extent_start, extent_len, | |
2634 | uptodate); | |
facc8a22 | 2635 | if (io_bio->end_io) |
4e4cbee9 | 2636 | io_bio->end_io(io_bio, blk_status_to_errno(bio->bi_status)); |
d1310b2e | 2637 | bio_put(bio); |
d1310b2e CM |
2638 | } |
2639 | ||
9be3395b | 2640 | /* |
184f999e DS |
2641 | * Initialize the members up to but not including 'bio'. Use after allocating a |
2642 | * new bio by bio_alloc_bioset as it does not initialize the bytes outside of | |
2643 | * 'bio' because use of __GFP_ZERO is not supported. | |
9be3395b | 2644 | */ |
184f999e | 2645 | static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio) |
d1310b2e | 2646 | { |
184f999e DS |
2647 | memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio)); |
2648 | } | |
d1310b2e | 2649 | |
9be3395b | 2650 | /* |
6e707bcd DS |
2651 | * The following helpers allocate a bio. As it's backed by a bioset, it'll |
2652 | * never fail. We're returning a bio right now but you can call btrfs_io_bio | |
2653 | * for the appropriate container_of magic | |
9be3395b | 2654 | */ |
c821e7f3 | 2655 | struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte) |
d1310b2e CM |
2656 | { |
2657 | struct bio *bio; | |
d1310b2e | 2658 | |
8ac9f7c1 | 2659 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &btrfs_bioset); |
74d46992 | 2660 | bio_set_dev(bio, bdev); |
c821e7f3 | 2661 | bio->bi_iter.bi_sector = first_byte >> 9; |
184f999e | 2662 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
d1310b2e CM |
2663 | return bio; |
2664 | } | |
2665 | ||
8b6c1d56 | 2666 | struct bio *btrfs_bio_clone(struct bio *bio) |
9be3395b | 2667 | { |
23ea8e5a MX |
2668 | struct btrfs_io_bio *btrfs_bio; |
2669 | struct bio *new; | |
9be3395b | 2670 | |
6e707bcd | 2671 | /* Bio allocation backed by a bioset does not fail */ |
8ac9f7c1 | 2672 | new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset); |
6e707bcd | 2673 | btrfs_bio = btrfs_io_bio(new); |
184f999e | 2674 | btrfs_io_bio_init(btrfs_bio); |
6e707bcd | 2675 | btrfs_bio->iter = bio->bi_iter; |
23ea8e5a MX |
2676 | return new; |
2677 | } | |
9be3395b | 2678 | |
c5e4c3d7 | 2679 | struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs) |
9be3395b | 2680 | { |
facc8a22 MX |
2681 | struct bio *bio; |
2682 | ||
6e707bcd | 2683 | /* Bio allocation backed by a bioset does not fail */ |
8ac9f7c1 | 2684 | bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset); |
184f999e | 2685 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
facc8a22 | 2686 | return bio; |
9be3395b CM |
2687 | } |
2688 | ||
e477094f | 2689 | struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size) |
2f8e9140 LB |
2690 | { |
2691 | struct bio *bio; | |
2692 | struct btrfs_io_bio *btrfs_bio; | |
2693 | ||
2694 | /* this will never fail when it's backed by a bioset */ | |
8ac9f7c1 | 2695 | bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset); |
2f8e9140 LB |
2696 | ASSERT(bio); |
2697 | ||
2698 | btrfs_bio = btrfs_io_bio(bio); | |
184f999e | 2699 | btrfs_io_bio_init(btrfs_bio); |
2f8e9140 LB |
2700 | |
2701 | bio_trim(bio, offset >> 9, size >> 9); | |
17347cec | 2702 | btrfs_bio->iter = bio->bi_iter; |
2f8e9140 LB |
2703 | return bio; |
2704 | } | |
9be3395b | 2705 | |
1f7ad75b MC |
2706 | static int __must_check submit_one_bio(struct bio *bio, int mirror_num, |
2707 | unsigned long bio_flags) | |
d1310b2e | 2708 | { |
4e4cbee9 | 2709 | blk_status_t ret = 0; |
c45a8f2d | 2710 | struct bio_vec *bvec = bio_last_bvec_all(bio); |
70dec807 CM |
2711 | struct page *page = bvec->bv_page; |
2712 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 2713 | u64 start; |
70dec807 | 2714 | |
4eee4fa4 | 2715 | start = page_offset(page) + bvec->bv_offset; |
70dec807 | 2716 | |
902b22f3 | 2717 | bio->bi_private = NULL; |
d1310b2e | 2718 | |
20c9801d | 2719 | if (tree->ops) |
c6100a4b | 2720 | ret = tree->ops->submit_bio_hook(tree->private_data, bio, |
eaf25d93 | 2721 | mirror_num, bio_flags, start); |
0b86a832 | 2722 | else |
4e49ea4a | 2723 | btrfsic_submit_bio(bio); |
4a54c8c1 | 2724 | |
4e4cbee9 | 2725 | return blk_status_to_errno(ret); |
d1310b2e CM |
2726 | } |
2727 | ||
4b81ba48 DS |
2728 | /* |
2729 | * @opf: bio REQ_OP_* and REQ_* flags as one value | |
b8b3d625 DS |
2730 | * @tree: tree so we can call our merge_bio hook |
2731 | * @wbc: optional writeback control for io accounting | |
2732 | * @page: page to add to the bio | |
2733 | * @pg_offset: offset of the new bio or to check whether we are adding | |
2734 | * a contiguous page to the previous one | |
2735 | * @size: portion of page that we want to write | |
2736 | * @offset: starting offset in the page | |
2737 | * @bdev: attach newly created bios to this bdev | |
5c2b1fd7 | 2738 | * @bio_ret: must be valid pointer, newly allocated bio will be stored there |
b8b3d625 DS |
2739 | * @end_io_func: end_io callback for new bio |
2740 | * @mirror_num: desired mirror to read/write | |
2741 | * @prev_bio_flags: flags of previous bio to see if we can merge the current one | |
2742 | * @bio_flags: flags of the current bio to see if we can merge them | |
4b81ba48 DS |
2743 | */ |
2744 | static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree, | |
da2f0f74 | 2745 | struct writeback_control *wbc, |
6273b7f8 | 2746 | struct page *page, u64 offset, |
6c5a4e2c | 2747 | size_t size, unsigned long pg_offset, |
d1310b2e CM |
2748 | struct block_device *bdev, |
2749 | struct bio **bio_ret, | |
f188591e | 2750 | bio_end_io_t end_io_func, |
c8b97818 CM |
2751 | int mirror_num, |
2752 | unsigned long prev_bio_flags, | |
005efedf FM |
2753 | unsigned long bio_flags, |
2754 | bool force_bio_submit) | |
d1310b2e CM |
2755 | { |
2756 | int ret = 0; | |
2757 | struct bio *bio; | |
09cbfeaf | 2758 | size_t page_size = min_t(size_t, size, PAGE_SIZE); |
6273b7f8 | 2759 | sector_t sector = offset >> 9; |
d1310b2e | 2760 | |
5c2b1fd7 DS |
2761 | ASSERT(bio_ret); |
2762 | ||
2763 | if (*bio_ret) { | |
0c8508a6 DS |
2764 | bool contig; |
2765 | bool can_merge = true; | |
2766 | ||
d1310b2e | 2767 | bio = *bio_ret; |
0c8508a6 | 2768 | if (prev_bio_flags & EXTENT_BIO_COMPRESSED) |
4f024f37 | 2769 | contig = bio->bi_iter.bi_sector == sector; |
c8b97818 | 2770 | else |
f73a1c7d | 2771 | contig = bio_end_sector(bio) == sector; |
c8b97818 | 2772 | |
00032d38 DS |
2773 | if (tree->ops && btrfs_merge_bio_hook(page, offset, page_size, |
2774 | bio, bio_flags)) | |
0c8508a6 DS |
2775 | can_merge = false; |
2776 | ||
2777 | if (prev_bio_flags != bio_flags || !contig || !can_merge || | |
005efedf | 2778 | force_bio_submit || |
6c5a4e2c | 2779 | bio_add_page(bio, page, page_size, pg_offset) < page_size) { |
1f7ad75b | 2780 | ret = submit_one_bio(bio, mirror_num, prev_bio_flags); |
289454ad NA |
2781 | if (ret < 0) { |
2782 | *bio_ret = NULL; | |
79787eaa | 2783 | return ret; |
289454ad | 2784 | } |
d1310b2e CM |
2785 | bio = NULL; |
2786 | } else { | |
da2f0f74 CM |
2787 | if (wbc) |
2788 | wbc_account_io(wbc, page, page_size); | |
d1310b2e CM |
2789 | return 0; |
2790 | } | |
2791 | } | |
c8b97818 | 2792 | |
6273b7f8 | 2793 | bio = btrfs_bio_alloc(bdev, offset); |
6c5a4e2c | 2794 | bio_add_page(bio, page, page_size, pg_offset); |
d1310b2e CM |
2795 | bio->bi_end_io = end_io_func; |
2796 | bio->bi_private = tree; | |
e6959b93 | 2797 | bio->bi_write_hint = page->mapping->host->i_write_hint; |
4b81ba48 | 2798 | bio->bi_opf = opf; |
da2f0f74 CM |
2799 | if (wbc) { |
2800 | wbc_init_bio(wbc, bio); | |
2801 | wbc_account_io(wbc, page, page_size); | |
2802 | } | |
70dec807 | 2803 | |
5c2b1fd7 | 2804 | *bio_ret = bio; |
d1310b2e CM |
2805 | |
2806 | return ret; | |
2807 | } | |
2808 | ||
48a3b636 ES |
2809 | static void attach_extent_buffer_page(struct extent_buffer *eb, |
2810 | struct page *page) | |
d1310b2e CM |
2811 | { |
2812 | if (!PagePrivate(page)) { | |
2813 | SetPagePrivate(page); | |
09cbfeaf | 2814 | get_page(page); |
4f2de97a JB |
2815 | set_page_private(page, (unsigned long)eb); |
2816 | } else { | |
2817 | WARN_ON(page->private != (unsigned long)eb); | |
d1310b2e CM |
2818 | } |
2819 | } | |
2820 | ||
4f2de97a | 2821 | void set_page_extent_mapped(struct page *page) |
d1310b2e | 2822 | { |
4f2de97a JB |
2823 | if (!PagePrivate(page)) { |
2824 | SetPagePrivate(page); | |
09cbfeaf | 2825 | get_page(page); |
4f2de97a JB |
2826 | set_page_private(page, EXTENT_PAGE_PRIVATE); |
2827 | } | |
d1310b2e CM |
2828 | } |
2829 | ||
125bac01 MX |
2830 | static struct extent_map * |
2831 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
2832 | u64 start, u64 len, get_extent_t *get_extent, | |
2833 | struct extent_map **em_cached) | |
2834 | { | |
2835 | struct extent_map *em; | |
2836 | ||
2837 | if (em_cached && *em_cached) { | |
2838 | em = *em_cached; | |
cbc0e928 | 2839 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 | 2840 | start < extent_map_end(em)) { |
490b54d6 | 2841 | refcount_inc(&em->refs); |
125bac01 MX |
2842 | return em; |
2843 | } | |
2844 | ||
2845 | free_extent_map(em); | |
2846 | *em_cached = NULL; | |
2847 | } | |
2848 | ||
fc4f21b1 | 2849 | em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0); |
125bac01 MX |
2850 | if (em_cached && !IS_ERR_OR_NULL(em)) { |
2851 | BUG_ON(*em_cached); | |
490b54d6 | 2852 | refcount_inc(&em->refs); |
125bac01 MX |
2853 | *em_cached = em; |
2854 | } | |
2855 | return em; | |
2856 | } | |
d1310b2e CM |
2857 | /* |
2858 | * basic readpage implementation. Locked extent state structs are inserted | |
2859 | * into the tree that are removed when the IO is done (by the end_io | |
2860 | * handlers) | |
79787eaa | 2861 | * XXX JDM: This needs looking at to ensure proper page locking |
baf863b9 | 2862 | * return 0 on success, otherwise return error |
d1310b2e | 2863 | */ |
9974090b MX |
2864 | static int __do_readpage(struct extent_io_tree *tree, |
2865 | struct page *page, | |
2866 | get_extent_t *get_extent, | |
125bac01 | 2867 | struct extent_map **em_cached, |
9974090b | 2868 | struct bio **bio, int mirror_num, |
f1c77c55 | 2869 | unsigned long *bio_flags, unsigned int read_flags, |
005efedf | 2870 | u64 *prev_em_start) |
d1310b2e CM |
2871 | { |
2872 | struct inode *inode = page->mapping->host; | |
4eee4fa4 | 2873 | u64 start = page_offset(page); |
8eec8296 | 2874 | const u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
2875 | u64 cur = start; |
2876 | u64 extent_offset; | |
2877 | u64 last_byte = i_size_read(inode); | |
2878 | u64 block_start; | |
2879 | u64 cur_end; | |
d1310b2e CM |
2880 | struct extent_map *em; |
2881 | struct block_device *bdev; | |
baf863b9 | 2882 | int ret = 0; |
d1310b2e | 2883 | int nr = 0; |
306e16ce | 2884 | size_t pg_offset = 0; |
d1310b2e | 2885 | size_t iosize; |
c8b97818 | 2886 | size_t disk_io_size; |
d1310b2e | 2887 | size_t blocksize = inode->i_sb->s_blocksize; |
7f042a83 | 2888 | unsigned long this_bio_flag = 0; |
d1310b2e CM |
2889 | |
2890 | set_page_extent_mapped(page); | |
2891 | ||
90a887c9 DM |
2892 | if (!PageUptodate(page)) { |
2893 | if (cleancache_get_page(page) == 0) { | |
2894 | BUG_ON(blocksize != PAGE_SIZE); | |
9974090b | 2895 | unlock_extent(tree, start, end); |
90a887c9 DM |
2896 | goto out; |
2897 | } | |
2898 | } | |
2899 | ||
09cbfeaf | 2900 | if (page->index == last_byte >> PAGE_SHIFT) { |
c8b97818 | 2901 | char *userpage; |
09cbfeaf | 2902 | size_t zero_offset = last_byte & (PAGE_SIZE - 1); |
c8b97818 CM |
2903 | |
2904 | if (zero_offset) { | |
09cbfeaf | 2905 | iosize = PAGE_SIZE - zero_offset; |
7ac687d9 | 2906 | userpage = kmap_atomic(page); |
c8b97818 CM |
2907 | memset(userpage + zero_offset, 0, iosize); |
2908 | flush_dcache_page(page); | |
7ac687d9 | 2909 | kunmap_atomic(userpage); |
c8b97818 CM |
2910 | } |
2911 | } | |
d1310b2e | 2912 | while (cur <= end) { |
005efedf | 2913 | bool force_bio_submit = false; |
6273b7f8 | 2914 | u64 offset; |
c8f2f24b | 2915 | |
d1310b2e CM |
2916 | if (cur >= last_byte) { |
2917 | char *userpage; | |
507903b8 AJ |
2918 | struct extent_state *cached = NULL; |
2919 | ||
09cbfeaf | 2920 | iosize = PAGE_SIZE - pg_offset; |
7ac687d9 | 2921 | userpage = kmap_atomic(page); |
306e16ce | 2922 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2923 | flush_dcache_page(page); |
7ac687d9 | 2924 | kunmap_atomic(userpage); |
d1310b2e | 2925 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 2926 | &cached, GFP_NOFS); |
7f042a83 | 2927 | unlock_extent_cached(tree, cur, |
e43bbe5e | 2928 | cur + iosize - 1, &cached); |
d1310b2e CM |
2929 | break; |
2930 | } | |
125bac01 MX |
2931 | em = __get_extent_map(inode, page, pg_offset, cur, |
2932 | end - cur + 1, get_extent, em_cached); | |
c704005d | 2933 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 2934 | SetPageError(page); |
7f042a83 | 2935 | unlock_extent(tree, cur, end); |
d1310b2e CM |
2936 | break; |
2937 | } | |
d1310b2e CM |
2938 | extent_offset = cur - em->start; |
2939 | BUG_ON(extent_map_end(em) <= cur); | |
2940 | BUG_ON(end < cur); | |
2941 | ||
261507a0 | 2942 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
4b384318 | 2943 | this_bio_flag |= EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
2944 | extent_set_compress_type(&this_bio_flag, |
2945 | em->compress_type); | |
2946 | } | |
c8b97818 | 2947 | |
d1310b2e CM |
2948 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
2949 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 2950 | iosize = ALIGN(iosize, blocksize); |
c8b97818 CM |
2951 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
2952 | disk_io_size = em->block_len; | |
6273b7f8 | 2953 | offset = em->block_start; |
c8b97818 | 2954 | } else { |
6273b7f8 | 2955 | offset = em->block_start + extent_offset; |
c8b97818 CM |
2956 | disk_io_size = iosize; |
2957 | } | |
d1310b2e CM |
2958 | bdev = em->bdev; |
2959 | block_start = em->block_start; | |
d899e052 YZ |
2960 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2961 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
2962 | |
2963 | /* | |
2964 | * If we have a file range that points to a compressed extent | |
2965 | * and it's followed by a consecutive file range that points to | |
2966 | * to the same compressed extent (possibly with a different | |
2967 | * offset and/or length, so it either points to the whole extent | |
2968 | * or only part of it), we must make sure we do not submit a | |
2969 | * single bio to populate the pages for the 2 ranges because | |
2970 | * this makes the compressed extent read zero out the pages | |
2971 | * belonging to the 2nd range. Imagine the following scenario: | |
2972 | * | |
2973 | * File layout | |
2974 | * [0 - 8K] [8K - 24K] | |
2975 | * | | | |
2976 | * | | | |
2977 | * points to extent X, points to extent X, | |
2978 | * offset 4K, length of 8K offset 0, length 16K | |
2979 | * | |
2980 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
2981 | * | |
2982 | * If the bio to read the compressed extent covers both ranges, | |
2983 | * it will decompress extent X into the pages belonging to the | |
2984 | * first range and then it will stop, zeroing out the remaining | |
2985 | * pages that belong to the other range that points to extent X. | |
2986 | * So here we make sure we submit 2 bios, one for the first | |
2987 | * range and another one for the third range. Both will target | |
2988 | * the same physical extent from disk, but we can't currently | |
2989 | * make the compressed bio endio callback populate the pages | |
2990 | * for both ranges because each compressed bio is tightly | |
2991 | * coupled with a single extent map, and each range can have | |
2992 | * an extent map with a different offset value relative to the | |
2993 | * uncompressed data of our extent and different lengths. This | |
2994 | * is a corner case so we prioritize correctness over | |
2995 | * non-optimal behavior (submitting 2 bios for the same extent). | |
2996 | */ | |
2997 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
2998 | prev_em_start && *prev_em_start != (u64)-1 && | |
2999 | *prev_em_start != em->orig_start) | |
3000 | force_bio_submit = true; | |
3001 | ||
3002 | if (prev_em_start) | |
3003 | *prev_em_start = em->orig_start; | |
3004 | ||
d1310b2e CM |
3005 | free_extent_map(em); |
3006 | em = NULL; | |
3007 | ||
3008 | /* we've found a hole, just zero and go on */ | |
3009 | if (block_start == EXTENT_MAP_HOLE) { | |
3010 | char *userpage; | |
507903b8 AJ |
3011 | struct extent_state *cached = NULL; |
3012 | ||
7ac687d9 | 3013 | userpage = kmap_atomic(page); |
306e16ce | 3014 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3015 | flush_dcache_page(page); |
7ac687d9 | 3016 | kunmap_atomic(userpage); |
d1310b2e CM |
3017 | |
3018 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 | 3019 | &cached, GFP_NOFS); |
7f042a83 | 3020 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3021 | cur + iosize - 1, &cached); |
d1310b2e | 3022 | cur = cur + iosize; |
306e16ce | 3023 | pg_offset += iosize; |
d1310b2e CM |
3024 | continue; |
3025 | } | |
3026 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
3027 | if (test_range_bit(tree, cur, cur_end, |
3028 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 3029 | check_page_uptodate(tree, page); |
7f042a83 | 3030 | unlock_extent(tree, cur, cur + iosize - 1); |
d1310b2e | 3031 | cur = cur + iosize; |
306e16ce | 3032 | pg_offset += iosize; |
d1310b2e CM |
3033 | continue; |
3034 | } | |
70dec807 CM |
3035 | /* we have an inline extent but it didn't get marked up |
3036 | * to date. Error out | |
3037 | */ | |
3038 | if (block_start == EXTENT_MAP_INLINE) { | |
3039 | SetPageError(page); | |
7f042a83 | 3040 | unlock_extent(tree, cur, cur + iosize - 1); |
70dec807 | 3041 | cur = cur + iosize; |
306e16ce | 3042 | pg_offset += iosize; |
70dec807 CM |
3043 | continue; |
3044 | } | |
d1310b2e | 3045 | |
4b81ba48 | 3046 | ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL, |
6273b7f8 DS |
3047 | page, offset, disk_io_size, |
3048 | pg_offset, bdev, bio, | |
c8b97818 CM |
3049 | end_bio_extent_readpage, mirror_num, |
3050 | *bio_flags, | |
005efedf FM |
3051 | this_bio_flag, |
3052 | force_bio_submit); | |
c8f2f24b JB |
3053 | if (!ret) { |
3054 | nr++; | |
3055 | *bio_flags = this_bio_flag; | |
3056 | } else { | |
d1310b2e | 3057 | SetPageError(page); |
7f042a83 | 3058 | unlock_extent(tree, cur, cur + iosize - 1); |
baf863b9 | 3059 | goto out; |
edd33c99 | 3060 | } |
d1310b2e | 3061 | cur = cur + iosize; |
306e16ce | 3062 | pg_offset += iosize; |
d1310b2e | 3063 | } |
90a887c9 | 3064 | out: |
d1310b2e CM |
3065 | if (!nr) { |
3066 | if (!PageError(page)) | |
3067 | SetPageUptodate(page); | |
3068 | unlock_page(page); | |
3069 | } | |
baf863b9 | 3070 | return ret; |
d1310b2e CM |
3071 | } |
3072 | ||
9974090b MX |
3073 | static inline void __do_contiguous_readpages(struct extent_io_tree *tree, |
3074 | struct page *pages[], int nr_pages, | |
3075 | u64 start, u64 end, | |
125bac01 | 3076 | struct extent_map **em_cached, |
d3fac6ba | 3077 | struct bio **bio, |
1f7ad75b | 3078 | unsigned long *bio_flags, |
808f80b4 | 3079 | u64 *prev_em_start) |
9974090b MX |
3080 | { |
3081 | struct inode *inode; | |
3082 | struct btrfs_ordered_extent *ordered; | |
3083 | int index; | |
3084 | ||
3085 | inode = pages[0]->mapping->host; | |
3086 | while (1) { | |
3087 | lock_extent(tree, start, end); | |
a776c6fa | 3088 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
9974090b MX |
3089 | end - start + 1); |
3090 | if (!ordered) | |
3091 | break; | |
3092 | unlock_extent(tree, start, end); | |
3093 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3094 | btrfs_put_ordered_extent(ordered); | |
3095 | } | |
3096 | ||
3097 | for (index = 0; index < nr_pages; index++) { | |
4ef77695 | 3098 | __do_readpage(tree, pages[index], btrfs_get_extent, em_cached, |
5e9d3982 | 3099 | bio, 0, bio_flags, REQ_RAHEAD, prev_em_start); |
09cbfeaf | 3100 | put_page(pages[index]); |
9974090b MX |
3101 | } |
3102 | } | |
3103 | ||
3104 | static void __extent_readpages(struct extent_io_tree *tree, | |
3105 | struct page *pages[], | |
e4d17ef5 | 3106 | int nr_pages, |
125bac01 | 3107 | struct extent_map **em_cached, |
d3fac6ba | 3108 | struct bio **bio, unsigned long *bio_flags, |
808f80b4 | 3109 | u64 *prev_em_start) |
9974090b | 3110 | { |
35a3621b | 3111 | u64 start = 0; |
9974090b MX |
3112 | u64 end = 0; |
3113 | u64 page_start; | |
3114 | int index; | |
35a3621b | 3115 | int first_index = 0; |
9974090b MX |
3116 | |
3117 | for (index = 0; index < nr_pages; index++) { | |
3118 | page_start = page_offset(pages[index]); | |
3119 | if (!end) { | |
3120 | start = page_start; | |
09cbfeaf | 3121 | end = start + PAGE_SIZE - 1; |
9974090b MX |
3122 | first_index = index; |
3123 | } else if (end + 1 == page_start) { | |
09cbfeaf | 3124 | end += PAGE_SIZE; |
9974090b MX |
3125 | } else { |
3126 | __do_contiguous_readpages(tree, &pages[first_index], | |
3127 | index - first_index, start, | |
4ef77695 | 3128 | end, em_cached, |
d3fac6ba | 3129 | bio, bio_flags, |
1f7ad75b | 3130 | prev_em_start); |
9974090b | 3131 | start = page_start; |
09cbfeaf | 3132 | end = start + PAGE_SIZE - 1; |
9974090b MX |
3133 | first_index = index; |
3134 | } | |
3135 | } | |
3136 | ||
3137 | if (end) | |
3138 | __do_contiguous_readpages(tree, &pages[first_index], | |
3139 | index - first_index, start, | |
4ef77695 | 3140 | end, em_cached, bio, |
d3fac6ba | 3141 | bio_flags, prev_em_start); |
9974090b MX |
3142 | } |
3143 | ||
3144 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
3145 | struct page *page, | |
3146 | get_extent_t *get_extent, | |
3147 | struct bio **bio, int mirror_num, | |
f1c77c55 DS |
3148 | unsigned long *bio_flags, |
3149 | unsigned int read_flags) | |
9974090b MX |
3150 | { |
3151 | struct inode *inode = page->mapping->host; | |
3152 | struct btrfs_ordered_extent *ordered; | |
3153 | u64 start = page_offset(page); | |
09cbfeaf | 3154 | u64 end = start + PAGE_SIZE - 1; |
9974090b MX |
3155 | int ret; |
3156 | ||
3157 | while (1) { | |
3158 | lock_extent(tree, start, end); | |
a776c6fa | 3159 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
09cbfeaf | 3160 | PAGE_SIZE); |
9974090b MX |
3161 | if (!ordered) |
3162 | break; | |
3163 | unlock_extent(tree, start, end); | |
3164 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3165 | btrfs_put_ordered_extent(ordered); | |
3166 | } | |
3167 | ||
125bac01 | 3168 | ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num, |
1f7ad75b | 3169 | bio_flags, read_flags, NULL); |
9974090b MX |
3170 | return ret; |
3171 | } | |
3172 | ||
d1310b2e | 3173 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, |
8ddc7d9c | 3174 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3175 | { |
3176 | struct bio *bio = NULL; | |
c8b97818 | 3177 | unsigned long bio_flags = 0; |
d1310b2e CM |
3178 | int ret; |
3179 | ||
8ddc7d9c | 3180 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, |
1f7ad75b | 3181 | &bio_flags, 0); |
d1310b2e | 3182 | if (bio) |
1f7ad75b | 3183 | ret = submit_one_bio(bio, mirror_num, bio_flags); |
d1310b2e CM |
3184 | return ret; |
3185 | } | |
d1310b2e | 3186 | |
3d4b9496 | 3187 | static void update_nr_written(struct writeback_control *wbc, |
a9132667 | 3188 | unsigned long nr_written) |
11c8349b CM |
3189 | { |
3190 | wbc->nr_to_write -= nr_written; | |
11c8349b CM |
3191 | } |
3192 | ||
d1310b2e | 3193 | /* |
40f76580 CM |
3194 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
3195 | * | |
5eaad97a | 3196 | * This returns 1 if btrfs_run_delalloc_range function did all the work required |
40f76580 CM |
3197 | * to write the page (copy into inline extent). In this case the IO has |
3198 | * been started and the page is already unlocked. | |
3199 | * | |
3200 | * This returns 0 if all went well (page still locked) | |
3201 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 3202 | */ |
40f76580 | 3203 | static noinline_for_stack int writepage_delalloc(struct inode *inode, |
8cc0237a NB |
3204 | struct page *page, struct writeback_control *wbc, |
3205 | u64 delalloc_start, unsigned long *nr_written) | |
40f76580 | 3206 | { |
8cc0237a | 3207 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
09cbfeaf | 3208 | u64 page_end = delalloc_start + PAGE_SIZE - 1; |
40f76580 CM |
3209 | u64 nr_delalloc; |
3210 | u64 delalloc_to_write = 0; | |
3211 | u64 delalloc_end = 0; | |
3212 | int ret; | |
3213 | int page_started = 0; | |
3214 | ||
40f76580 CM |
3215 | |
3216 | while (delalloc_end < page_end) { | |
3217 | nr_delalloc = find_lock_delalloc_range(inode, tree, | |
3218 | page, | |
3219 | &delalloc_start, | |
3220 | &delalloc_end, | |
dcab6a3b | 3221 | BTRFS_MAX_EXTENT_SIZE); |
40f76580 CM |
3222 | if (nr_delalloc == 0) { |
3223 | delalloc_start = delalloc_end + 1; | |
3224 | continue; | |
3225 | } | |
5eaad97a NB |
3226 | ret = btrfs_run_delalloc_range(inode, page, delalloc_start, |
3227 | delalloc_end, &page_started, nr_written, wbc); | |
40f76580 CM |
3228 | /* File system has been set read-only */ |
3229 | if (ret) { | |
3230 | SetPageError(page); | |
5eaad97a NB |
3231 | /* |
3232 | * btrfs_run_delalloc_range should return < 0 for error | |
3233 | * but just in case, we use > 0 here meaning the IO is | |
3234 | * started, so we don't want to return > 0 unless | |
3235 | * things are going well. | |
40f76580 CM |
3236 | */ |
3237 | ret = ret < 0 ? ret : -EIO; | |
3238 | goto done; | |
3239 | } | |
3240 | /* | |
ea1754a0 KS |
3241 | * delalloc_end is already one less than the total length, so |
3242 | * we don't subtract one from PAGE_SIZE | |
40f76580 CM |
3243 | */ |
3244 | delalloc_to_write += (delalloc_end - delalloc_start + | |
ea1754a0 | 3245 | PAGE_SIZE) >> PAGE_SHIFT; |
40f76580 CM |
3246 | delalloc_start = delalloc_end + 1; |
3247 | } | |
3248 | if (wbc->nr_to_write < delalloc_to_write) { | |
3249 | int thresh = 8192; | |
3250 | ||
3251 | if (delalloc_to_write < thresh * 2) | |
3252 | thresh = delalloc_to_write; | |
3253 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3254 | thresh); | |
3255 | } | |
3256 | ||
3257 | /* did the fill delalloc function already unlock and start | |
3258 | * the IO? | |
3259 | */ | |
3260 | if (page_started) { | |
3261 | /* | |
3262 | * we've unlocked the page, so we can't update | |
3263 | * the mapping's writeback index, just update | |
3264 | * nr_to_write. | |
3265 | */ | |
3266 | wbc->nr_to_write -= *nr_written; | |
3267 | return 1; | |
3268 | } | |
3269 | ||
3270 | ret = 0; | |
3271 | ||
3272 | done: | |
3273 | return ret; | |
3274 | } | |
3275 | ||
3276 | /* | |
3277 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3278 | * and does the loop to map the page into extents and bios. | |
3279 | * | |
3280 | * We return 1 if the IO is started and the page is unlocked, | |
3281 | * 0 if all went well (page still locked) | |
3282 | * < 0 if there were errors (page still locked) | |
3283 | */ | |
3284 | static noinline_for_stack int __extent_writepage_io(struct inode *inode, | |
3285 | struct page *page, | |
3286 | struct writeback_control *wbc, | |
3287 | struct extent_page_data *epd, | |
3288 | loff_t i_size, | |
3289 | unsigned long nr_written, | |
f1c77c55 | 3290 | unsigned int write_flags, int *nr_ret) |
d1310b2e | 3291 | { |
d1310b2e | 3292 | struct extent_io_tree *tree = epd->tree; |
4eee4fa4 | 3293 | u64 start = page_offset(page); |
09cbfeaf | 3294 | u64 page_end = start + PAGE_SIZE - 1; |
d1310b2e CM |
3295 | u64 end; |
3296 | u64 cur = start; | |
3297 | u64 extent_offset; | |
d1310b2e CM |
3298 | u64 block_start; |
3299 | u64 iosize; | |
d1310b2e CM |
3300 | struct extent_map *em; |
3301 | struct block_device *bdev; | |
7f3c74fb | 3302 | size_t pg_offset = 0; |
d1310b2e | 3303 | size_t blocksize; |
40f76580 CM |
3304 | int ret = 0; |
3305 | int nr = 0; | |
3306 | bool compressed; | |
c8b97818 | 3307 | |
d75855b4 NB |
3308 | ret = btrfs_writepage_cow_fixup(page, start, page_end); |
3309 | if (ret) { | |
3310 | /* Fixup worker will requeue */ | |
3311 | if (ret == -EBUSY) | |
3312 | wbc->pages_skipped++; | |
3313 | else | |
3314 | redirty_page_for_writepage(wbc, page); | |
40f76580 | 3315 | |
d75855b4 NB |
3316 | update_nr_written(wbc, nr_written); |
3317 | unlock_page(page); | |
3318 | return 1; | |
247e743c CM |
3319 | } |
3320 | ||
11c8349b CM |
3321 | /* |
3322 | * we don't want to touch the inode after unlocking the page, | |
3323 | * so we update the mapping writeback index now | |
3324 | */ | |
3d4b9496 | 3325 | update_nr_written(wbc, nr_written + 1); |
771ed689 | 3326 | |
d1310b2e | 3327 | end = page_end; |
40f76580 | 3328 | if (i_size <= start) { |
c629732d | 3329 | btrfs_writepage_endio_finish_ordered(page, start, page_end, 1); |
d1310b2e CM |
3330 | goto done; |
3331 | } | |
3332 | ||
d1310b2e CM |
3333 | blocksize = inode->i_sb->s_blocksize; |
3334 | ||
3335 | while (cur <= end) { | |
40f76580 | 3336 | u64 em_end; |
6273b7f8 | 3337 | u64 offset; |
58409edd | 3338 | |
40f76580 | 3339 | if (cur >= i_size) { |
7087a9d8 | 3340 | btrfs_writepage_endio_finish_ordered(page, cur, |
c629732d | 3341 | page_end, 1); |
d1310b2e CM |
3342 | break; |
3343 | } | |
3c98c62f | 3344 | em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, cur, |
d1310b2e | 3345 | end - cur + 1, 1); |
c704005d | 3346 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3347 | SetPageError(page); |
61391d56 | 3348 | ret = PTR_ERR_OR_ZERO(em); |
d1310b2e CM |
3349 | break; |
3350 | } | |
3351 | ||
3352 | extent_offset = cur - em->start; | |
40f76580 CM |
3353 | em_end = extent_map_end(em); |
3354 | BUG_ON(em_end <= cur); | |
d1310b2e | 3355 | BUG_ON(end < cur); |
40f76580 | 3356 | iosize = min(em_end - cur, end - cur + 1); |
fda2832f | 3357 | iosize = ALIGN(iosize, blocksize); |
6273b7f8 | 3358 | offset = em->block_start + extent_offset; |
d1310b2e CM |
3359 | bdev = em->bdev; |
3360 | block_start = em->block_start; | |
c8b97818 | 3361 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
3362 | free_extent_map(em); |
3363 | em = NULL; | |
3364 | ||
c8b97818 CM |
3365 | /* |
3366 | * compressed and inline extents are written through other | |
3367 | * paths in the FS | |
3368 | */ | |
3369 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3370 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
3371 | /* |
3372 | * end_io notification does not happen here for | |
3373 | * compressed extents | |
3374 | */ | |
7087a9d8 NB |
3375 | if (!compressed) |
3376 | btrfs_writepage_endio_finish_ordered(page, cur, | |
3377 | cur + iosize - 1, | |
c629732d | 3378 | 1); |
c8b97818 CM |
3379 | else if (compressed) { |
3380 | /* we don't want to end_page_writeback on | |
3381 | * a compressed extent. this happens | |
3382 | * elsewhere | |
3383 | */ | |
3384 | nr++; | |
3385 | } | |
3386 | ||
3387 | cur += iosize; | |
7f3c74fb | 3388 | pg_offset += iosize; |
d1310b2e CM |
3389 | continue; |
3390 | } | |
c8b97818 | 3391 | |
5cdc84bf | 3392 | btrfs_set_range_writeback(tree, cur, cur + iosize - 1); |
58409edd DS |
3393 | if (!PageWriteback(page)) { |
3394 | btrfs_err(BTRFS_I(inode)->root->fs_info, | |
3395 | "page %lu not writeback, cur %llu end %llu", | |
3396 | page->index, cur, end); | |
d1310b2e | 3397 | } |
7f3c74fb | 3398 | |
4b81ba48 | 3399 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc, |
6273b7f8 | 3400 | page, offset, iosize, pg_offset, |
c2df8bb4 | 3401 | bdev, &epd->bio, |
58409edd DS |
3402 | end_bio_extent_writepage, |
3403 | 0, 0, 0, false); | |
fe01aa65 | 3404 | if (ret) { |
58409edd | 3405 | SetPageError(page); |
fe01aa65 TK |
3406 | if (PageWriteback(page)) |
3407 | end_page_writeback(page); | |
3408 | } | |
d1310b2e | 3409 | |
d1310b2e | 3410 | cur = cur + iosize; |
7f3c74fb | 3411 | pg_offset += iosize; |
d1310b2e CM |
3412 | nr++; |
3413 | } | |
40f76580 CM |
3414 | done: |
3415 | *nr_ret = nr; | |
40f76580 CM |
3416 | return ret; |
3417 | } | |
3418 | ||
3419 | /* | |
3420 | * the writepage semantics are similar to regular writepage. extent | |
3421 | * records are inserted to lock ranges in the tree, and as dirty areas | |
3422 | * are found, they are marked writeback. Then the lock bits are removed | |
3423 | * and the end_io handler clears the writeback ranges | |
3424 | */ | |
3425 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
aab6e9ed | 3426 | struct extent_page_data *epd) |
40f76580 CM |
3427 | { |
3428 | struct inode *inode = page->mapping->host; | |
40f76580 | 3429 | u64 start = page_offset(page); |
09cbfeaf | 3430 | u64 page_end = start + PAGE_SIZE - 1; |
40f76580 CM |
3431 | int ret; |
3432 | int nr = 0; | |
3433 | size_t pg_offset = 0; | |
3434 | loff_t i_size = i_size_read(inode); | |
09cbfeaf | 3435 | unsigned long end_index = i_size >> PAGE_SHIFT; |
f1c77c55 | 3436 | unsigned int write_flags = 0; |
40f76580 CM |
3437 | unsigned long nr_written = 0; |
3438 | ||
ff40adf7 | 3439 | write_flags = wbc_to_write_flags(wbc); |
40f76580 CM |
3440 | |
3441 | trace___extent_writepage(page, inode, wbc); | |
3442 | ||
3443 | WARN_ON(!PageLocked(page)); | |
3444 | ||
3445 | ClearPageError(page); | |
3446 | ||
09cbfeaf | 3447 | pg_offset = i_size & (PAGE_SIZE - 1); |
40f76580 CM |
3448 | if (page->index > end_index || |
3449 | (page->index == end_index && !pg_offset)) { | |
09cbfeaf | 3450 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); |
40f76580 CM |
3451 | unlock_page(page); |
3452 | return 0; | |
3453 | } | |
3454 | ||
3455 | if (page->index == end_index) { | |
3456 | char *userpage; | |
3457 | ||
3458 | userpage = kmap_atomic(page); | |
3459 | memset(userpage + pg_offset, 0, | |
09cbfeaf | 3460 | PAGE_SIZE - pg_offset); |
40f76580 CM |
3461 | kunmap_atomic(userpage); |
3462 | flush_dcache_page(page); | |
3463 | } | |
3464 | ||
3465 | pg_offset = 0; | |
3466 | ||
3467 | set_page_extent_mapped(page); | |
3468 | ||
7789a55a | 3469 | if (!epd->extent_locked) { |
8cc0237a | 3470 | ret = writepage_delalloc(inode, page, wbc, start, &nr_written); |
7789a55a NB |
3471 | if (ret == 1) |
3472 | goto done_unlocked; | |
3473 | if (ret) | |
3474 | goto done; | |
3475 | } | |
40f76580 CM |
3476 | |
3477 | ret = __extent_writepage_io(inode, page, wbc, epd, | |
3478 | i_size, nr_written, write_flags, &nr); | |
3479 | if (ret == 1) | |
3480 | goto done_unlocked; | |
3481 | ||
d1310b2e CM |
3482 | done: |
3483 | if (nr == 0) { | |
3484 | /* make sure the mapping tag for page dirty gets cleared */ | |
3485 | set_page_writeback(page); | |
3486 | end_page_writeback(page); | |
3487 | } | |
61391d56 FM |
3488 | if (PageError(page)) { |
3489 | ret = ret < 0 ? ret : -EIO; | |
3490 | end_extent_writepage(page, ret, start, page_end); | |
3491 | } | |
d1310b2e | 3492 | unlock_page(page); |
40f76580 | 3493 | return ret; |
771ed689 | 3494 | |
11c8349b | 3495 | done_unlocked: |
d1310b2e CM |
3496 | return 0; |
3497 | } | |
3498 | ||
fd8b2b61 | 3499 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 3500 | { |
74316201 N |
3501 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
3502 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
3503 | } |
3504 | ||
0e378df1 CM |
3505 | static noinline_for_stack int |
3506 | lock_extent_buffer_for_io(struct extent_buffer *eb, | |
3507 | struct btrfs_fs_info *fs_info, | |
3508 | struct extent_page_data *epd) | |
0b32f4bb | 3509 | { |
cc5e31a4 | 3510 | int i, num_pages; |
0b32f4bb JB |
3511 | int flush = 0; |
3512 | int ret = 0; | |
3513 | ||
3514 | if (!btrfs_try_tree_write_lock(eb)) { | |
3515 | flush = 1; | |
3516 | flush_write_bio(epd); | |
3517 | btrfs_tree_lock(eb); | |
3518 | } | |
3519 | ||
3520 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3521 | btrfs_tree_unlock(eb); | |
3522 | if (!epd->sync_io) | |
3523 | return 0; | |
3524 | if (!flush) { | |
3525 | flush_write_bio(epd); | |
3526 | flush = 1; | |
3527 | } | |
a098d8e8 CM |
3528 | while (1) { |
3529 | wait_on_extent_buffer_writeback(eb); | |
3530 | btrfs_tree_lock(eb); | |
3531 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3532 | break; | |
0b32f4bb | 3533 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3534 | } |
3535 | } | |
3536 | ||
51561ffe JB |
3537 | /* |
3538 | * We need to do this to prevent races in people who check if the eb is | |
3539 | * under IO since we can end up having no IO bits set for a short period | |
3540 | * of time. | |
3541 | */ | |
3542 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
3543 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3544 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 3545 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 3546 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
104b4e51 NB |
3547 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
3548 | -eb->len, | |
3549 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 3550 | ret = 1; |
51561ffe JB |
3551 | } else { |
3552 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
3553 | } |
3554 | ||
3555 | btrfs_tree_unlock(eb); | |
3556 | ||
3557 | if (!ret) | |
3558 | return ret; | |
3559 | ||
65ad0104 | 3560 | num_pages = num_extent_pages(eb); |
0b32f4bb | 3561 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3562 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3563 | |
3564 | if (!trylock_page(p)) { | |
3565 | if (!flush) { | |
3566 | flush_write_bio(epd); | |
3567 | flush = 1; | |
3568 | } | |
3569 | lock_page(p); | |
3570 | } | |
3571 | } | |
3572 | ||
3573 | return ret; | |
3574 | } | |
3575 | ||
3576 | static void end_extent_buffer_writeback(struct extent_buffer *eb) | |
3577 | { | |
3578 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
4e857c58 | 3579 | smp_mb__after_atomic(); |
0b32f4bb JB |
3580 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); |
3581 | } | |
3582 | ||
656f30db FM |
3583 | static void set_btree_ioerr(struct page *page) |
3584 | { | |
3585 | struct extent_buffer *eb = (struct extent_buffer *)page->private; | |
656f30db FM |
3586 | |
3587 | SetPageError(page); | |
3588 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) | |
3589 | return; | |
3590 | ||
3591 | /* | |
3592 | * If writeback for a btree extent that doesn't belong to a log tree | |
3593 | * failed, increment the counter transaction->eb_write_errors. | |
3594 | * We do this because while the transaction is running and before it's | |
3595 | * committing (when we call filemap_fdata[write|wait]_range against | |
3596 | * the btree inode), we might have | |
3597 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
3598 | * returns an error or an error happens during writeback, when we're | |
3599 | * committing the transaction we wouldn't know about it, since the pages | |
3600 | * can be no longer dirty nor marked anymore for writeback (if a | |
3601 | * subsequent modification to the extent buffer didn't happen before the | |
3602 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
3603 | * able to find the pages tagged with SetPageError at transaction | |
3604 | * commit time. So if this happens we must abort the transaction, | |
3605 | * otherwise we commit a super block with btree roots that point to | |
3606 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
3607 | * or the content of some node/leaf from a past generation that got | |
3608 | * cowed or deleted and is no longer valid. | |
3609 | * | |
3610 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
3611 | * not be enough - we need to distinguish between log tree extents vs | |
3612 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
3613 | * will catch and clear such errors in the mapping - and that call might | |
3614 | * be from a log sync and not from a transaction commit. Also, checking | |
3615 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
3616 | * not done and would not be reliable - the eb might have been released | |
3617 | * from memory and reading it back again means that flag would not be | |
3618 | * set (since it's a runtime flag, not persisted on disk). | |
3619 | * | |
3620 | * Using the flags below in the btree inode also makes us achieve the | |
3621 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
3622 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
3623 | * is called, the writeback for all dirty pages had already finished | |
3624 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
3625 | * filemap_fdatawait_range() would return success, as it could not know | |
3626 | * that writeback errors happened (the pages were no longer tagged for | |
3627 | * writeback). | |
3628 | */ | |
3629 | switch (eb->log_index) { | |
3630 | case -1: | |
afcdd129 | 3631 | set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags); |
656f30db FM |
3632 | break; |
3633 | case 0: | |
afcdd129 | 3634 | set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags); |
656f30db FM |
3635 | break; |
3636 | case 1: | |
afcdd129 | 3637 | set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags); |
656f30db FM |
3638 | break; |
3639 | default: | |
3640 | BUG(); /* unexpected, logic error */ | |
3641 | } | |
3642 | } | |
3643 | ||
4246a0b6 | 3644 | static void end_bio_extent_buffer_writepage(struct bio *bio) |
0b32f4bb | 3645 | { |
2c30c71b | 3646 | struct bio_vec *bvec; |
0b32f4bb | 3647 | struct extent_buffer *eb; |
2c30c71b | 3648 | int i, done; |
0b32f4bb | 3649 | |
c09abff8 | 3650 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2c30c71b | 3651 | bio_for_each_segment_all(bvec, bio, i) { |
0b32f4bb JB |
3652 | struct page *page = bvec->bv_page; |
3653 | ||
0b32f4bb JB |
3654 | eb = (struct extent_buffer *)page->private; |
3655 | BUG_ON(!eb); | |
3656 | done = atomic_dec_and_test(&eb->io_pages); | |
3657 | ||
4e4cbee9 | 3658 | if (bio->bi_status || |
4246a0b6 | 3659 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { |
0b32f4bb | 3660 | ClearPageUptodate(page); |
656f30db | 3661 | set_btree_ioerr(page); |
0b32f4bb JB |
3662 | } |
3663 | ||
3664 | end_page_writeback(page); | |
3665 | ||
3666 | if (!done) | |
3667 | continue; | |
3668 | ||
3669 | end_extent_buffer_writeback(eb); | |
2c30c71b | 3670 | } |
0b32f4bb JB |
3671 | |
3672 | bio_put(bio); | |
0b32f4bb JB |
3673 | } |
3674 | ||
0e378df1 | 3675 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb JB |
3676 | struct btrfs_fs_info *fs_info, |
3677 | struct writeback_control *wbc, | |
3678 | struct extent_page_data *epd) | |
3679 | { | |
3680 | struct block_device *bdev = fs_info->fs_devices->latest_bdev; | |
f28491e0 | 3681 | struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree; |
0b32f4bb | 3682 | u64 offset = eb->start; |
851cd173 | 3683 | u32 nritems; |
cc5e31a4 | 3684 | int i, num_pages; |
851cd173 | 3685 | unsigned long start, end; |
ff40adf7 | 3686 | unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META; |
d7dbe9e7 | 3687 | int ret = 0; |
0b32f4bb | 3688 | |
656f30db | 3689 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); |
65ad0104 | 3690 | num_pages = num_extent_pages(eb); |
0b32f4bb | 3691 | atomic_set(&eb->io_pages, num_pages); |
de0022b9 | 3692 | |
851cd173 LB |
3693 | /* set btree blocks beyond nritems with 0 to avoid stale content. */ |
3694 | nritems = btrfs_header_nritems(eb); | |
3eb548ee | 3695 | if (btrfs_header_level(eb) > 0) { |
3eb548ee LB |
3696 | end = btrfs_node_key_ptr_offset(nritems); |
3697 | ||
b159fa28 | 3698 | memzero_extent_buffer(eb, end, eb->len - end); |
851cd173 LB |
3699 | } else { |
3700 | /* | |
3701 | * leaf: | |
3702 | * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0 | |
3703 | */ | |
3704 | start = btrfs_item_nr_offset(nritems); | |
3d9ec8c4 | 3705 | end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, eb); |
b159fa28 | 3706 | memzero_extent_buffer(eb, start, end - start); |
3eb548ee LB |
3707 | } |
3708 | ||
0b32f4bb | 3709 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3710 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3711 | |
3712 | clear_page_dirty_for_io(p); | |
3713 | set_page_writeback(p); | |
4b81ba48 | 3714 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc, |
6273b7f8 | 3715 | p, offset, PAGE_SIZE, 0, bdev, |
c2df8bb4 | 3716 | &epd->bio, |
1f7ad75b | 3717 | end_bio_extent_buffer_writepage, |
18fdc679 | 3718 | 0, 0, 0, false); |
0b32f4bb | 3719 | if (ret) { |
656f30db | 3720 | set_btree_ioerr(p); |
fe01aa65 TK |
3721 | if (PageWriteback(p)) |
3722 | end_page_writeback(p); | |
0b32f4bb JB |
3723 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
3724 | end_extent_buffer_writeback(eb); | |
3725 | ret = -EIO; | |
3726 | break; | |
3727 | } | |
09cbfeaf | 3728 | offset += PAGE_SIZE; |
3d4b9496 | 3729 | update_nr_written(wbc, 1); |
0b32f4bb JB |
3730 | unlock_page(p); |
3731 | } | |
3732 | ||
3733 | if (unlikely(ret)) { | |
3734 | for (; i < num_pages; i++) { | |
bbf65cf0 | 3735 | struct page *p = eb->pages[i]; |
81465028 | 3736 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
3737 | unlock_page(p); |
3738 | } | |
3739 | } | |
3740 | ||
3741 | return ret; | |
3742 | } | |
3743 | ||
3744 | int btree_write_cache_pages(struct address_space *mapping, | |
3745 | struct writeback_control *wbc) | |
3746 | { | |
3747 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; | |
3748 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; | |
3749 | struct extent_buffer *eb, *prev_eb = NULL; | |
3750 | struct extent_page_data epd = { | |
3751 | .bio = NULL, | |
3752 | .tree = tree, | |
3753 | .extent_locked = 0, | |
3754 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
3755 | }; | |
3756 | int ret = 0; | |
3757 | int done = 0; | |
3758 | int nr_to_write_done = 0; | |
3759 | struct pagevec pvec; | |
3760 | int nr_pages; | |
3761 | pgoff_t index; | |
3762 | pgoff_t end; /* Inclusive */ | |
3763 | int scanned = 0; | |
10bbd235 | 3764 | xa_mark_t tag; |
0b32f4bb | 3765 | |
86679820 | 3766 | pagevec_init(&pvec); |
0b32f4bb JB |
3767 | if (wbc->range_cyclic) { |
3768 | index = mapping->writeback_index; /* Start from prev offset */ | |
3769 | end = -1; | |
3770 | } else { | |
09cbfeaf KS |
3771 | index = wbc->range_start >> PAGE_SHIFT; |
3772 | end = wbc->range_end >> PAGE_SHIFT; | |
0b32f4bb JB |
3773 | scanned = 1; |
3774 | } | |
3775 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3776 | tag = PAGECACHE_TAG_TOWRITE; | |
3777 | else | |
3778 | tag = PAGECACHE_TAG_DIRTY; | |
3779 | retry: | |
3780 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3781 | tag_pages_for_writeback(mapping, index, end); | |
3782 | while (!done && !nr_to_write_done && (index <= end) && | |
4006f437 | 3783 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 3784 | tag))) { |
0b32f4bb JB |
3785 | unsigned i; |
3786 | ||
3787 | scanned = 1; | |
3788 | for (i = 0; i < nr_pages; i++) { | |
3789 | struct page *page = pvec.pages[i]; | |
3790 | ||
3791 | if (!PagePrivate(page)) | |
3792 | continue; | |
3793 | ||
b5bae261 JB |
3794 | spin_lock(&mapping->private_lock); |
3795 | if (!PagePrivate(page)) { | |
3796 | spin_unlock(&mapping->private_lock); | |
3797 | continue; | |
3798 | } | |
3799 | ||
0b32f4bb | 3800 | eb = (struct extent_buffer *)page->private; |
b5bae261 JB |
3801 | |
3802 | /* | |
3803 | * Shouldn't happen and normally this would be a BUG_ON | |
3804 | * but no sense in crashing the users box for something | |
3805 | * we can survive anyway. | |
3806 | */ | |
fae7f21c | 3807 | if (WARN_ON(!eb)) { |
b5bae261 | 3808 | spin_unlock(&mapping->private_lock); |
0b32f4bb JB |
3809 | continue; |
3810 | } | |
3811 | ||
b5bae261 JB |
3812 | if (eb == prev_eb) { |
3813 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 3814 | continue; |
b5bae261 | 3815 | } |
0b32f4bb | 3816 | |
b5bae261 JB |
3817 | ret = atomic_inc_not_zero(&eb->refs); |
3818 | spin_unlock(&mapping->private_lock); | |
3819 | if (!ret) | |
0b32f4bb | 3820 | continue; |
0b32f4bb JB |
3821 | |
3822 | prev_eb = eb; | |
3823 | ret = lock_extent_buffer_for_io(eb, fs_info, &epd); | |
3824 | if (!ret) { | |
3825 | free_extent_buffer(eb); | |
3826 | continue; | |
3827 | } | |
3828 | ||
3829 | ret = write_one_eb(eb, fs_info, wbc, &epd); | |
3830 | if (ret) { | |
3831 | done = 1; | |
3832 | free_extent_buffer(eb); | |
3833 | break; | |
3834 | } | |
3835 | free_extent_buffer(eb); | |
3836 | ||
3837 | /* | |
3838 | * the filesystem may choose to bump up nr_to_write. | |
3839 | * We have to make sure to honor the new nr_to_write | |
3840 | * at any time | |
3841 | */ | |
3842 | nr_to_write_done = wbc->nr_to_write <= 0; | |
3843 | } | |
3844 | pagevec_release(&pvec); | |
3845 | cond_resched(); | |
3846 | } | |
3847 | if (!scanned && !done) { | |
3848 | /* | |
3849 | * We hit the last page and there is more work to be done: wrap | |
3850 | * back to the start of the file | |
3851 | */ | |
3852 | scanned = 1; | |
3853 | index = 0; | |
3854 | goto retry; | |
3855 | } | |
3856 | flush_write_bio(&epd); | |
3857 | return ret; | |
3858 | } | |
3859 | ||
d1310b2e | 3860 | /** |
4bef0848 | 3861 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
3862 | * @mapping: address space structure to write |
3863 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
935db853 | 3864 | * @data: data passed to __extent_writepage function |
d1310b2e CM |
3865 | * |
3866 | * If a page is already under I/O, write_cache_pages() skips it, even | |
3867 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
3868 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
3869 | * and msync() need to guarantee that all the data which was dirty at the time | |
3870 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
3871 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
3872 | * existing IO to complete. | |
3873 | */ | |
4242b64a | 3874 | static int extent_write_cache_pages(struct address_space *mapping, |
4bef0848 | 3875 | struct writeback_control *wbc, |
aab6e9ed | 3876 | struct extent_page_data *epd) |
d1310b2e | 3877 | { |
7fd1a3f7 | 3878 | struct inode *inode = mapping->host; |
d1310b2e CM |
3879 | int ret = 0; |
3880 | int done = 0; | |
f85d7d6c | 3881 | int nr_to_write_done = 0; |
d1310b2e CM |
3882 | struct pagevec pvec; |
3883 | int nr_pages; | |
3884 | pgoff_t index; | |
3885 | pgoff_t end; /* Inclusive */ | |
a9132667 LB |
3886 | pgoff_t done_index; |
3887 | int range_whole = 0; | |
d1310b2e | 3888 | int scanned = 0; |
10bbd235 | 3889 | xa_mark_t tag; |
d1310b2e | 3890 | |
7fd1a3f7 JB |
3891 | /* |
3892 | * We have to hold onto the inode so that ordered extents can do their | |
3893 | * work when the IO finishes. The alternative to this is failing to add | |
3894 | * an ordered extent if the igrab() fails there and that is a huge pain | |
3895 | * to deal with, so instead just hold onto the inode throughout the | |
3896 | * writepages operation. If it fails here we are freeing up the inode | |
3897 | * anyway and we'd rather not waste our time writing out stuff that is | |
3898 | * going to be truncated anyway. | |
3899 | */ | |
3900 | if (!igrab(inode)) | |
3901 | return 0; | |
3902 | ||
86679820 | 3903 | pagevec_init(&pvec); |
d1310b2e CM |
3904 | if (wbc->range_cyclic) { |
3905 | index = mapping->writeback_index; /* Start from prev offset */ | |
3906 | end = -1; | |
3907 | } else { | |
09cbfeaf KS |
3908 | index = wbc->range_start >> PAGE_SHIFT; |
3909 | end = wbc->range_end >> PAGE_SHIFT; | |
a9132667 LB |
3910 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
3911 | range_whole = 1; | |
d1310b2e CM |
3912 | scanned = 1; |
3913 | } | |
f7aaa06b JB |
3914 | if (wbc->sync_mode == WB_SYNC_ALL) |
3915 | tag = PAGECACHE_TAG_TOWRITE; | |
3916 | else | |
3917 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 3918 | retry: |
f7aaa06b JB |
3919 | if (wbc->sync_mode == WB_SYNC_ALL) |
3920 | tag_pages_for_writeback(mapping, index, end); | |
a9132667 | 3921 | done_index = index; |
f85d7d6c | 3922 | while (!done && !nr_to_write_done && (index <= end) && |
67fd707f JK |
3923 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, |
3924 | &index, end, tag))) { | |
d1310b2e CM |
3925 | unsigned i; |
3926 | ||
3927 | scanned = 1; | |
3928 | for (i = 0; i < nr_pages; i++) { | |
3929 | struct page *page = pvec.pages[i]; | |
3930 | ||
a9132667 | 3931 | done_index = page->index; |
d1310b2e | 3932 | /* |
b93b0163 MW |
3933 | * At this point we hold neither the i_pages lock nor |
3934 | * the page lock: the page may be truncated or | |
3935 | * invalidated (changing page->mapping to NULL), | |
3936 | * or even swizzled back from swapper_space to | |
3937 | * tmpfs file mapping | |
d1310b2e | 3938 | */ |
c8f2f24b | 3939 | if (!trylock_page(page)) { |
aab6e9ed | 3940 | flush_write_bio(epd); |
c8f2f24b | 3941 | lock_page(page); |
01d658f2 | 3942 | } |
d1310b2e CM |
3943 | |
3944 | if (unlikely(page->mapping != mapping)) { | |
3945 | unlock_page(page); | |
3946 | continue; | |
3947 | } | |
3948 | ||
d2c3f4f6 | 3949 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 | 3950 | if (PageWriteback(page)) |
aab6e9ed | 3951 | flush_write_bio(epd); |
d1310b2e | 3952 | wait_on_page_writeback(page); |
d2c3f4f6 | 3953 | } |
d1310b2e CM |
3954 | |
3955 | if (PageWriteback(page) || | |
3956 | !clear_page_dirty_for_io(page)) { | |
3957 | unlock_page(page); | |
3958 | continue; | |
3959 | } | |
3960 | ||
aab6e9ed | 3961 | ret = __extent_writepage(page, wbc, epd); |
d1310b2e CM |
3962 | |
3963 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
3964 | unlock_page(page); | |
3965 | ret = 0; | |
3966 | } | |
a9132667 LB |
3967 | if (ret < 0) { |
3968 | /* | |
3969 | * done_index is set past this page, | |
3970 | * so media errors will not choke | |
3971 | * background writeout for the entire | |
3972 | * file. This has consequences for | |
3973 | * range_cyclic semantics (ie. it may | |
3974 | * not be suitable for data integrity | |
3975 | * writeout). | |
3976 | */ | |
3977 | done_index = page->index + 1; | |
3978 | done = 1; | |
3979 | break; | |
3980 | } | |
f85d7d6c CM |
3981 | |
3982 | /* | |
3983 | * the filesystem may choose to bump up nr_to_write. | |
3984 | * We have to make sure to honor the new nr_to_write | |
3985 | * at any time | |
3986 | */ | |
3987 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
3988 | } |
3989 | pagevec_release(&pvec); | |
3990 | cond_resched(); | |
3991 | } | |
894b36e3 | 3992 | if (!scanned && !done) { |
d1310b2e CM |
3993 | /* |
3994 | * We hit the last page and there is more work to be done: wrap | |
3995 | * back to the start of the file | |
3996 | */ | |
3997 | scanned = 1; | |
3998 | index = 0; | |
3999 | goto retry; | |
4000 | } | |
a9132667 LB |
4001 | |
4002 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole)) | |
4003 | mapping->writeback_index = done_index; | |
4004 | ||
7fd1a3f7 | 4005 | btrfs_add_delayed_iput(inode); |
894b36e3 | 4006 | return ret; |
d1310b2e | 4007 | } |
d1310b2e | 4008 | |
aab6e9ed | 4009 | static void flush_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 4010 | { |
d2c3f4f6 | 4011 | if (epd->bio) { |
355808c2 JM |
4012 | int ret; |
4013 | ||
18fdc679 | 4014 | ret = submit_one_bio(epd->bio, 0, 0); |
79787eaa | 4015 | BUG_ON(ret < 0); /* -ENOMEM */ |
d2c3f4f6 CM |
4016 | epd->bio = NULL; |
4017 | } | |
4018 | } | |
4019 | ||
0a9b0e53 | 4020 | int extent_write_full_page(struct page *page, struct writeback_control *wbc) |
d1310b2e CM |
4021 | { |
4022 | int ret; | |
d1310b2e CM |
4023 | struct extent_page_data epd = { |
4024 | .bio = NULL, | |
0a9b0e53 | 4025 | .tree = &BTRFS_I(page->mapping->host)->io_tree, |
771ed689 | 4026 | .extent_locked = 0, |
ffbd517d | 4027 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e | 4028 | }; |
d1310b2e | 4029 | |
d1310b2e CM |
4030 | ret = __extent_writepage(page, wbc, &epd); |
4031 | ||
e2932ee0 | 4032 | flush_write_bio(&epd); |
d1310b2e CM |
4033 | return ret; |
4034 | } | |
d1310b2e | 4035 | |
5e3ee236 | 4036 | int extent_write_locked_range(struct inode *inode, u64 start, u64 end, |
771ed689 CM |
4037 | int mode) |
4038 | { | |
4039 | int ret = 0; | |
4040 | struct address_space *mapping = inode->i_mapping; | |
5e3ee236 | 4041 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
771ed689 | 4042 | struct page *page; |
09cbfeaf KS |
4043 | unsigned long nr_pages = (end - start + PAGE_SIZE) >> |
4044 | PAGE_SHIFT; | |
771ed689 CM |
4045 | |
4046 | struct extent_page_data epd = { | |
4047 | .bio = NULL, | |
4048 | .tree = tree, | |
771ed689 | 4049 | .extent_locked = 1, |
ffbd517d | 4050 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
4051 | }; |
4052 | struct writeback_control wbc_writepages = { | |
771ed689 | 4053 | .sync_mode = mode, |
771ed689 CM |
4054 | .nr_to_write = nr_pages * 2, |
4055 | .range_start = start, | |
4056 | .range_end = end + 1, | |
4057 | }; | |
4058 | ||
d397712b | 4059 | while (start <= end) { |
09cbfeaf | 4060 | page = find_get_page(mapping, start >> PAGE_SHIFT); |
771ed689 CM |
4061 | if (clear_page_dirty_for_io(page)) |
4062 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
4063 | else { | |
7087a9d8 | 4064 | btrfs_writepage_endio_finish_ordered(page, start, |
c629732d | 4065 | start + PAGE_SIZE - 1, 1); |
771ed689 CM |
4066 | unlock_page(page); |
4067 | } | |
09cbfeaf KS |
4068 | put_page(page); |
4069 | start += PAGE_SIZE; | |
771ed689 CM |
4070 | } |
4071 | ||
e2932ee0 | 4072 | flush_write_bio(&epd); |
771ed689 CM |
4073 | return ret; |
4074 | } | |
d1310b2e | 4075 | |
8ae225a8 | 4076 | int extent_writepages(struct address_space *mapping, |
d1310b2e CM |
4077 | struct writeback_control *wbc) |
4078 | { | |
4079 | int ret = 0; | |
4080 | struct extent_page_data epd = { | |
4081 | .bio = NULL, | |
8ae225a8 | 4082 | .tree = &BTRFS_I(mapping->host)->io_tree, |
771ed689 | 4083 | .extent_locked = 0, |
ffbd517d | 4084 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
4085 | }; |
4086 | ||
935db853 | 4087 | ret = extent_write_cache_pages(mapping, wbc, &epd); |
e2932ee0 | 4088 | flush_write_bio(&epd); |
d1310b2e CM |
4089 | return ret; |
4090 | } | |
d1310b2e | 4091 | |
2a3ff0ad NB |
4092 | int extent_readpages(struct address_space *mapping, struct list_head *pages, |
4093 | unsigned nr_pages) | |
d1310b2e CM |
4094 | { |
4095 | struct bio *bio = NULL; | |
4096 | unsigned page_idx; | |
c8b97818 | 4097 | unsigned long bio_flags = 0; |
67c9684f LB |
4098 | struct page *pagepool[16]; |
4099 | struct page *page; | |
125bac01 | 4100 | struct extent_map *em_cached = NULL; |
2a3ff0ad | 4101 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; |
67c9684f | 4102 | int nr = 0; |
808f80b4 | 4103 | u64 prev_em_start = (u64)-1; |
d1310b2e | 4104 | |
d1310b2e | 4105 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
67c9684f | 4106 | page = list_entry(pages->prev, struct page, lru); |
d1310b2e CM |
4107 | |
4108 | prefetchw(&page->flags); | |
4109 | list_del(&page->lru); | |
67c9684f | 4110 | if (add_to_page_cache_lru(page, mapping, |
8a5c743e MH |
4111 | page->index, |
4112 | readahead_gfp_mask(mapping))) { | |
09cbfeaf | 4113 | put_page(page); |
67c9684f | 4114 | continue; |
d1310b2e | 4115 | } |
67c9684f LB |
4116 | |
4117 | pagepool[nr++] = page; | |
4118 | if (nr < ARRAY_SIZE(pagepool)) | |
4119 | continue; | |
e4d17ef5 DS |
4120 | __extent_readpages(tree, pagepool, nr, &em_cached, &bio, |
4121 | &bio_flags, &prev_em_start); | |
67c9684f | 4122 | nr = 0; |
d1310b2e | 4123 | } |
9974090b | 4124 | if (nr) |
e4d17ef5 DS |
4125 | __extent_readpages(tree, pagepool, nr, &em_cached, &bio, |
4126 | &bio_flags, &prev_em_start); | |
67c9684f | 4127 | |
125bac01 MX |
4128 | if (em_cached) |
4129 | free_extent_map(em_cached); | |
4130 | ||
d1310b2e CM |
4131 | BUG_ON(!list_empty(pages)); |
4132 | if (bio) | |
1f7ad75b | 4133 | return submit_one_bio(bio, 0, bio_flags); |
d1310b2e CM |
4134 | return 0; |
4135 | } | |
d1310b2e CM |
4136 | |
4137 | /* | |
4138 | * basic invalidatepage code, this waits on any locked or writeback | |
4139 | * ranges corresponding to the page, and then deletes any extent state | |
4140 | * records from the tree | |
4141 | */ | |
4142 | int extent_invalidatepage(struct extent_io_tree *tree, | |
4143 | struct page *page, unsigned long offset) | |
4144 | { | |
2ac55d41 | 4145 | struct extent_state *cached_state = NULL; |
4eee4fa4 | 4146 | u64 start = page_offset(page); |
09cbfeaf | 4147 | u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
4148 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; |
4149 | ||
fda2832f | 4150 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
4151 | if (start > end) |
4152 | return 0; | |
4153 | ||
ff13db41 | 4154 | lock_extent_bits(tree, start, end, &cached_state); |
1edbb734 | 4155 | wait_on_page_writeback(page); |
d1310b2e | 4156 | clear_extent_bit(tree, start, end, |
32c00aff JB |
4157 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
4158 | EXTENT_DO_ACCOUNTING, | |
ae0f1625 | 4159 | 1, 1, &cached_state); |
d1310b2e CM |
4160 | return 0; |
4161 | } | |
d1310b2e | 4162 | |
7b13b7b1 CM |
4163 | /* |
4164 | * a helper for releasepage, this tests for areas of the page that | |
4165 | * are locked or under IO and drops the related state bits if it is safe | |
4166 | * to drop the page. | |
4167 | */ | |
29c68b2d | 4168 | static int try_release_extent_state(struct extent_io_tree *tree, |
48a3b636 | 4169 | struct page *page, gfp_t mask) |
7b13b7b1 | 4170 | { |
4eee4fa4 | 4171 | u64 start = page_offset(page); |
09cbfeaf | 4172 | u64 end = start + PAGE_SIZE - 1; |
7b13b7b1 CM |
4173 | int ret = 1; |
4174 | ||
211f90e6 | 4175 | if (test_range_bit(tree, start, end, |
8b62b72b | 4176 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
4177 | ret = 0; |
4178 | else { | |
11ef160f CM |
4179 | /* |
4180 | * at this point we can safely clear everything except the | |
4181 | * locked bit and the nodatasum bit | |
4182 | */ | |
66b0c887 | 4183 | ret = __clear_extent_bit(tree, start, end, |
11ef160f | 4184 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
66b0c887 | 4185 | 0, 0, NULL, mask, NULL); |
e3f24cc5 CM |
4186 | |
4187 | /* if clear_extent_bit failed for enomem reasons, | |
4188 | * we can't allow the release to continue. | |
4189 | */ | |
4190 | if (ret < 0) | |
4191 | ret = 0; | |
4192 | else | |
4193 | ret = 1; | |
7b13b7b1 CM |
4194 | } |
4195 | return ret; | |
4196 | } | |
7b13b7b1 | 4197 | |
d1310b2e CM |
4198 | /* |
4199 | * a helper for releasepage. As long as there are no locked extents | |
4200 | * in the range corresponding to the page, both state records and extent | |
4201 | * map records are removed | |
4202 | */ | |
477a30ba | 4203 | int try_release_extent_mapping(struct page *page, gfp_t mask) |
d1310b2e CM |
4204 | { |
4205 | struct extent_map *em; | |
4eee4fa4 | 4206 | u64 start = page_offset(page); |
09cbfeaf | 4207 | u64 end = start + PAGE_SIZE - 1; |
bd3599a0 FM |
4208 | struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host); |
4209 | struct extent_io_tree *tree = &btrfs_inode->io_tree; | |
4210 | struct extent_map_tree *map = &btrfs_inode->extent_tree; | |
7b13b7b1 | 4211 | |
d0164adc | 4212 | if (gfpflags_allow_blocking(mask) && |
ee22184b | 4213 | page->mapping->host->i_size > SZ_16M) { |
39b5637f | 4214 | u64 len; |
70dec807 | 4215 | while (start <= end) { |
39b5637f | 4216 | len = end - start + 1; |
890871be | 4217 | write_lock(&map->lock); |
39b5637f | 4218 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 4219 | if (!em) { |
890871be | 4220 | write_unlock(&map->lock); |
70dec807 CM |
4221 | break; |
4222 | } | |
7f3c74fb CM |
4223 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
4224 | em->start != start) { | |
890871be | 4225 | write_unlock(&map->lock); |
70dec807 CM |
4226 | free_extent_map(em); |
4227 | break; | |
4228 | } | |
4229 | if (!test_range_bit(tree, em->start, | |
4230 | extent_map_end(em) - 1, | |
8b62b72b | 4231 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 4232 | 0, NULL)) { |
bd3599a0 FM |
4233 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
4234 | &btrfs_inode->runtime_flags); | |
70dec807 CM |
4235 | remove_extent_mapping(map, em); |
4236 | /* once for the rb tree */ | |
4237 | free_extent_map(em); | |
4238 | } | |
4239 | start = extent_map_end(em); | |
890871be | 4240 | write_unlock(&map->lock); |
70dec807 CM |
4241 | |
4242 | /* once for us */ | |
d1310b2e CM |
4243 | free_extent_map(em); |
4244 | } | |
d1310b2e | 4245 | } |
29c68b2d | 4246 | return try_release_extent_state(tree, page, mask); |
d1310b2e | 4247 | } |
d1310b2e | 4248 | |
ec29ed5b CM |
4249 | /* |
4250 | * helper function for fiemap, which doesn't want to see any holes. | |
4251 | * This maps until we find something past 'last' | |
4252 | */ | |
4253 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
e3350e16 | 4254 | u64 offset, u64 last) |
ec29ed5b | 4255 | { |
da17066c | 4256 | u64 sectorsize = btrfs_inode_sectorsize(inode); |
ec29ed5b CM |
4257 | struct extent_map *em; |
4258 | u64 len; | |
4259 | ||
4260 | if (offset >= last) | |
4261 | return NULL; | |
4262 | ||
67871254 | 4263 | while (1) { |
ec29ed5b CM |
4264 | len = last - offset; |
4265 | if (len == 0) | |
4266 | break; | |
fda2832f | 4267 | len = ALIGN(len, sectorsize); |
e3350e16 DS |
4268 | em = btrfs_get_extent_fiemap(BTRFS_I(inode), NULL, 0, offset, |
4269 | len, 0); | |
c704005d | 4270 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
4271 | return em; |
4272 | ||
4273 | /* if this isn't a hole return it */ | |
4a2d25cd | 4274 | if (em->block_start != EXTENT_MAP_HOLE) |
ec29ed5b | 4275 | return em; |
ec29ed5b CM |
4276 | |
4277 | /* this is a hole, advance to the next extent */ | |
4278 | offset = extent_map_end(em); | |
4279 | free_extent_map(em); | |
4280 | if (offset >= last) | |
4281 | break; | |
4282 | } | |
4283 | return NULL; | |
4284 | } | |
4285 | ||
4751832d QW |
4286 | /* |
4287 | * To cache previous fiemap extent | |
4288 | * | |
4289 | * Will be used for merging fiemap extent | |
4290 | */ | |
4291 | struct fiemap_cache { | |
4292 | u64 offset; | |
4293 | u64 phys; | |
4294 | u64 len; | |
4295 | u32 flags; | |
4296 | bool cached; | |
4297 | }; | |
4298 | ||
4299 | /* | |
4300 | * Helper to submit fiemap extent. | |
4301 | * | |
4302 | * Will try to merge current fiemap extent specified by @offset, @phys, | |
4303 | * @len and @flags with cached one. | |
4304 | * And only when we fails to merge, cached one will be submitted as | |
4305 | * fiemap extent. | |
4306 | * | |
4307 | * Return value is the same as fiemap_fill_next_extent(). | |
4308 | */ | |
4309 | static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, | |
4310 | struct fiemap_cache *cache, | |
4311 | u64 offset, u64 phys, u64 len, u32 flags) | |
4312 | { | |
4313 | int ret = 0; | |
4314 | ||
4315 | if (!cache->cached) | |
4316 | goto assign; | |
4317 | ||
4318 | /* | |
4319 | * Sanity check, extent_fiemap() should have ensured that new | |
4320 | * fiemap extent won't overlap with cahced one. | |
4321 | * Not recoverable. | |
4322 | * | |
4323 | * NOTE: Physical address can overlap, due to compression | |
4324 | */ | |
4325 | if (cache->offset + cache->len > offset) { | |
4326 | WARN_ON(1); | |
4327 | return -EINVAL; | |
4328 | } | |
4329 | ||
4330 | /* | |
4331 | * Only merges fiemap extents if | |
4332 | * 1) Their logical addresses are continuous | |
4333 | * | |
4334 | * 2) Their physical addresses are continuous | |
4335 | * So truly compressed (physical size smaller than logical size) | |
4336 | * extents won't get merged with each other | |
4337 | * | |
4338 | * 3) Share same flags except FIEMAP_EXTENT_LAST | |
4339 | * So regular extent won't get merged with prealloc extent | |
4340 | */ | |
4341 | if (cache->offset + cache->len == offset && | |
4342 | cache->phys + cache->len == phys && | |
4343 | (cache->flags & ~FIEMAP_EXTENT_LAST) == | |
4344 | (flags & ~FIEMAP_EXTENT_LAST)) { | |
4345 | cache->len += len; | |
4346 | cache->flags |= flags; | |
4347 | goto try_submit_last; | |
4348 | } | |
4349 | ||
4350 | /* Not mergeable, need to submit cached one */ | |
4351 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, | |
4352 | cache->len, cache->flags); | |
4353 | cache->cached = false; | |
4354 | if (ret) | |
4355 | return ret; | |
4356 | assign: | |
4357 | cache->cached = true; | |
4358 | cache->offset = offset; | |
4359 | cache->phys = phys; | |
4360 | cache->len = len; | |
4361 | cache->flags = flags; | |
4362 | try_submit_last: | |
4363 | if (cache->flags & FIEMAP_EXTENT_LAST) { | |
4364 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, | |
4365 | cache->phys, cache->len, cache->flags); | |
4366 | cache->cached = false; | |
4367 | } | |
4368 | return ret; | |
4369 | } | |
4370 | ||
4371 | /* | |
848c23b7 | 4372 | * Emit last fiemap cache |
4751832d | 4373 | * |
848c23b7 QW |
4374 | * The last fiemap cache may still be cached in the following case: |
4375 | * 0 4k 8k | |
4376 | * |<- Fiemap range ->| | |
4377 | * |<------------ First extent ----------->| | |
4378 | * | |
4379 | * In this case, the first extent range will be cached but not emitted. | |
4380 | * So we must emit it before ending extent_fiemap(). | |
4751832d | 4381 | */ |
848c23b7 QW |
4382 | static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info, |
4383 | struct fiemap_extent_info *fieinfo, | |
4384 | struct fiemap_cache *cache) | |
4751832d QW |
4385 | { |
4386 | int ret; | |
4387 | ||
4388 | if (!cache->cached) | |
4389 | return 0; | |
4390 | ||
4751832d QW |
4391 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, |
4392 | cache->len, cache->flags); | |
4393 | cache->cached = false; | |
4394 | if (ret > 0) | |
4395 | ret = 0; | |
4396 | return ret; | |
4397 | } | |
4398 | ||
1506fcc8 | 4399 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
2135fb9b | 4400 | __u64 start, __u64 len) |
1506fcc8 | 4401 | { |
975f84fe | 4402 | int ret = 0; |
1506fcc8 YS |
4403 | u64 off = start; |
4404 | u64 max = start + len; | |
4405 | u32 flags = 0; | |
975f84fe JB |
4406 | u32 found_type; |
4407 | u64 last; | |
ec29ed5b | 4408 | u64 last_for_get_extent = 0; |
1506fcc8 | 4409 | u64 disko = 0; |
ec29ed5b | 4410 | u64 isize = i_size_read(inode); |
975f84fe | 4411 | struct btrfs_key found_key; |
1506fcc8 | 4412 | struct extent_map *em = NULL; |
2ac55d41 | 4413 | struct extent_state *cached_state = NULL; |
975f84fe | 4414 | struct btrfs_path *path; |
dc046b10 | 4415 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4751832d | 4416 | struct fiemap_cache cache = { 0 }; |
1506fcc8 | 4417 | int end = 0; |
ec29ed5b CM |
4418 | u64 em_start = 0; |
4419 | u64 em_len = 0; | |
4420 | u64 em_end = 0; | |
1506fcc8 YS |
4421 | |
4422 | if (len == 0) | |
4423 | return -EINVAL; | |
4424 | ||
975f84fe JB |
4425 | path = btrfs_alloc_path(); |
4426 | if (!path) | |
4427 | return -ENOMEM; | |
4428 | path->leave_spinning = 1; | |
4429 | ||
da17066c JM |
4430 | start = round_down(start, btrfs_inode_sectorsize(inode)); |
4431 | len = round_up(max, btrfs_inode_sectorsize(inode)) - start; | |
4d479cf0 | 4432 | |
ec29ed5b CM |
4433 | /* |
4434 | * lookup the last file extent. We're not using i_size here | |
4435 | * because there might be preallocation past i_size | |
4436 | */ | |
f85b7379 DS |
4437 | ret = btrfs_lookup_file_extent(NULL, root, path, |
4438 | btrfs_ino(BTRFS_I(inode)), -1, 0); | |
975f84fe JB |
4439 | if (ret < 0) { |
4440 | btrfs_free_path(path); | |
4441 | return ret; | |
2d324f59 LB |
4442 | } else { |
4443 | WARN_ON(!ret); | |
4444 | if (ret == 1) | |
4445 | ret = 0; | |
975f84fe | 4446 | } |
2d324f59 | 4447 | |
975f84fe | 4448 | path->slots[0]--; |
975f84fe | 4449 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
962a298f | 4450 | found_type = found_key.type; |
975f84fe | 4451 | |
ec29ed5b | 4452 | /* No extents, but there might be delalloc bits */ |
4a0cc7ca | 4453 | if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) || |
975f84fe | 4454 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
4455 | /* have to trust i_size as the end */ |
4456 | last = (u64)-1; | |
4457 | last_for_get_extent = isize; | |
4458 | } else { | |
4459 | /* | |
4460 | * remember the start of the last extent. There are a | |
4461 | * bunch of different factors that go into the length of the | |
4462 | * extent, so its much less complex to remember where it started | |
4463 | */ | |
4464 | last = found_key.offset; | |
4465 | last_for_get_extent = last + 1; | |
975f84fe | 4466 | } |
fe09e16c | 4467 | btrfs_release_path(path); |
975f84fe | 4468 | |
ec29ed5b CM |
4469 | /* |
4470 | * we might have some extents allocated but more delalloc past those | |
4471 | * extents. so, we trust isize unless the start of the last extent is | |
4472 | * beyond isize | |
4473 | */ | |
4474 | if (last < isize) { | |
4475 | last = (u64)-1; | |
4476 | last_for_get_extent = isize; | |
4477 | } | |
4478 | ||
ff13db41 | 4479 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
d0082371 | 4480 | &cached_state); |
ec29ed5b | 4481 | |
e3350e16 | 4482 | em = get_extent_skip_holes(inode, start, last_for_get_extent); |
1506fcc8 YS |
4483 | if (!em) |
4484 | goto out; | |
4485 | if (IS_ERR(em)) { | |
4486 | ret = PTR_ERR(em); | |
4487 | goto out; | |
4488 | } | |
975f84fe | 4489 | |
1506fcc8 | 4490 | while (!end) { |
b76bb701 | 4491 | u64 offset_in_extent = 0; |
ea8efc74 CM |
4492 | |
4493 | /* break if the extent we found is outside the range */ | |
4494 | if (em->start >= max || extent_map_end(em) < off) | |
4495 | break; | |
4496 | ||
4497 | /* | |
4498 | * get_extent may return an extent that starts before our | |
4499 | * requested range. We have to make sure the ranges | |
4500 | * we return to fiemap always move forward and don't | |
4501 | * overlap, so adjust the offsets here | |
4502 | */ | |
4503 | em_start = max(em->start, off); | |
1506fcc8 | 4504 | |
ea8efc74 CM |
4505 | /* |
4506 | * record the offset from the start of the extent | |
b76bb701 JB |
4507 | * for adjusting the disk offset below. Only do this if the |
4508 | * extent isn't compressed since our in ram offset may be past | |
4509 | * what we have actually allocated on disk. | |
ea8efc74 | 4510 | */ |
b76bb701 JB |
4511 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
4512 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 4513 | em_end = extent_map_end(em); |
ea8efc74 | 4514 | em_len = em_end - em_start; |
1506fcc8 | 4515 | flags = 0; |
f0986318 FM |
4516 | if (em->block_start < EXTENT_MAP_LAST_BYTE) |
4517 | disko = em->block_start + offset_in_extent; | |
4518 | else | |
4519 | disko = 0; | |
1506fcc8 | 4520 | |
ea8efc74 CM |
4521 | /* |
4522 | * bump off for our next call to get_extent | |
4523 | */ | |
4524 | off = extent_map_end(em); | |
4525 | if (off >= max) | |
4526 | end = 1; | |
4527 | ||
93dbfad7 | 4528 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
4529 | end = 1; |
4530 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 4531 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
4532 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
4533 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 4534 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
4535 | flags |= (FIEMAP_EXTENT_DELALLOC | |
4536 | FIEMAP_EXTENT_UNKNOWN); | |
dc046b10 JB |
4537 | } else if (fieinfo->fi_extents_max) { |
4538 | u64 bytenr = em->block_start - | |
4539 | (em->start - em->orig_start); | |
fe09e16c | 4540 | |
fe09e16c LB |
4541 | /* |
4542 | * As btrfs supports shared space, this information | |
4543 | * can be exported to userspace tools via | |
dc046b10 JB |
4544 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 |
4545 | * then we're just getting a count and we can skip the | |
4546 | * lookup stuff. | |
fe09e16c | 4547 | */ |
bb739cf0 EN |
4548 | ret = btrfs_check_shared(root, |
4549 | btrfs_ino(BTRFS_I(inode)), | |
4550 | bytenr); | |
dc046b10 | 4551 | if (ret < 0) |
fe09e16c | 4552 | goto out_free; |
dc046b10 | 4553 | if (ret) |
fe09e16c | 4554 | flags |= FIEMAP_EXTENT_SHARED; |
dc046b10 | 4555 | ret = 0; |
1506fcc8 YS |
4556 | } |
4557 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
4558 | flags |= FIEMAP_EXTENT_ENCODED; | |
0d2b2372 JB |
4559 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
4560 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
1506fcc8 | 4561 | |
1506fcc8 YS |
4562 | free_extent_map(em); |
4563 | em = NULL; | |
ec29ed5b CM |
4564 | if ((em_start >= last) || em_len == (u64)-1 || |
4565 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
4566 | flags |= FIEMAP_EXTENT_LAST; |
4567 | end = 1; | |
4568 | } | |
4569 | ||
ec29ed5b | 4570 | /* now scan forward to see if this is really the last extent. */ |
e3350e16 | 4571 | em = get_extent_skip_holes(inode, off, last_for_get_extent); |
ec29ed5b CM |
4572 | if (IS_ERR(em)) { |
4573 | ret = PTR_ERR(em); | |
4574 | goto out; | |
4575 | } | |
4576 | if (!em) { | |
975f84fe JB |
4577 | flags |= FIEMAP_EXTENT_LAST; |
4578 | end = 1; | |
4579 | } | |
4751832d QW |
4580 | ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko, |
4581 | em_len, flags); | |
26e726af CS |
4582 | if (ret) { |
4583 | if (ret == 1) | |
4584 | ret = 0; | |
ec29ed5b | 4585 | goto out_free; |
26e726af | 4586 | } |
1506fcc8 YS |
4587 | } |
4588 | out_free: | |
4751832d | 4589 | if (!ret) |
848c23b7 | 4590 | ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache); |
1506fcc8 YS |
4591 | free_extent_map(em); |
4592 | out: | |
fe09e16c | 4593 | btrfs_free_path(path); |
a52f4cd2 | 4594 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
e43bbe5e | 4595 | &cached_state); |
1506fcc8 YS |
4596 | return ret; |
4597 | } | |
4598 | ||
727011e0 CM |
4599 | static void __free_extent_buffer(struct extent_buffer *eb) |
4600 | { | |
6d49ba1b | 4601 | btrfs_leak_debug_del(&eb->leak_list); |
727011e0 CM |
4602 | kmem_cache_free(extent_buffer_cache, eb); |
4603 | } | |
4604 | ||
a26e8c9f | 4605 | int extent_buffer_under_io(struct extent_buffer *eb) |
db7f3436 JB |
4606 | { |
4607 | return (atomic_read(&eb->io_pages) || | |
4608 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4609 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4610 | } | |
4611 | ||
4612 | /* | |
55ac0139 | 4613 | * Release all pages attached to the extent buffer. |
db7f3436 | 4614 | */ |
55ac0139 | 4615 | static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb) |
db7f3436 | 4616 | { |
d64766fd NB |
4617 | int i; |
4618 | int num_pages; | |
b0132a3b | 4619 | int mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
db7f3436 JB |
4620 | |
4621 | BUG_ON(extent_buffer_under_io(eb)); | |
4622 | ||
d64766fd NB |
4623 | num_pages = num_extent_pages(eb); |
4624 | for (i = 0; i < num_pages; i++) { | |
4625 | struct page *page = eb->pages[i]; | |
db7f3436 | 4626 | |
5d2361db FL |
4627 | if (!page) |
4628 | continue; | |
4629 | if (mapped) | |
db7f3436 | 4630 | spin_lock(&page->mapping->private_lock); |
5d2361db FL |
4631 | /* |
4632 | * We do this since we'll remove the pages after we've | |
4633 | * removed the eb from the radix tree, so we could race | |
4634 | * and have this page now attached to the new eb. So | |
4635 | * only clear page_private if it's still connected to | |
4636 | * this eb. | |
4637 | */ | |
4638 | if (PagePrivate(page) && | |
4639 | page->private == (unsigned long)eb) { | |
4640 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4641 | BUG_ON(PageDirty(page)); | |
4642 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 4643 | /* |
5d2361db FL |
4644 | * We need to make sure we haven't be attached |
4645 | * to a new eb. | |
db7f3436 | 4646 | */ |
5d2361db FL |
4647 | ClearPagePrivate(page); |
4648 | set_page_private(page, 0); | |
4649 | /* One for the page private */ | |
09cbfeaf | 4650 | put_page(page); |
db7f3436 | 4651 | } |
5d2361db FL |
4652 | |
4653 | if (mapped) | |
4654 | spin_unlock(&page->mapping->private_lock); | |
4655 | ||
01327610 | 4656 | /* One for when we allocated the page */ |
09cbfeaf | 4657 | put_page(page); |
d64766fd | 4658 | } |
db7f3436 JB |
4659 | } |
4660 | ||
4661 | /* | |
4662 | * Helper for releasing the extent buffer. | |
4663 | */ | |
4664 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4665 | { | |
55ac0139 | 4666 | btrfs_release_extent_buffer_pages(eb); |
db7f3436 JB |
4667 | __free_extent_buffer(eb); |
4668 | } | |
4669 | ||
f28491e0 JB |
4670 | static struct extent_buffer * |
4671 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 4672 | unsigned long len) |
d1310b2e CM |
4673 | { |
4674 | struct extent_buffer *eb = NULL; | |
4675 | ||
d1b5c567 | 4676 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
4677 | eb->start = start; |
4678 | eb->len = len; | |
f28491e0 | 4679 | eb->fs_info = fs_info; |
815a51c7 | 4680 | eb->bflags = 0; |
bd681513 CM |
4681 | rwlock_init(&eb->lock); |
4682 | atomic_set(&eb->write_locks, 0); | |
4683 | atomic_set(&eb->read_locks, 0); | |
4684 | atomic_set(&eb->blocking_readers, 0); | |
4685 | atomic_set(&eb->blocking_writers, 0); | |
4686 | atomic_set(&eb->spinning_readers, 0); | |
4687 | atomic_set(&eb->spinning_writers, 0); | |
5b25f70f | 4688 | eb->lock_nested = 0; |
bd681513 CM |
4689 | init_waitqueue_head(&eb->write_lock_wq); |
4690 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 4691 | |
6d49ba1b ES |
4692 | btrfs_leak_debug_add(&eb->leak_list, &buffers); |
4693 | ||
3083ee2e | 4694 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 4695 | atomic_set(&eb->refs, 1); |
0b32f4bb | 4696 | atomic_set(&eb->io_pages, 0); |
727011e0 | 4697 | |
b8dae313 DS |
4698 | /* |
4699 | * Sanity checks, currently the maximum is 64k covered by 16x 4k pages | |
4700 | */ | |
4701 | BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE | |
4702 | > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
4703 | BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
d1310b2e CM |
4704 | |
4705 | return eb; | |
4706 | } | |
4707 | ||
815a51c7 JS |
4708 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) |
4709 | { | |
cc5e31a4 | 4710 | int i; |
815a51c7 JS |
4711 | struct page *p; |
4712 | struct extent_buffer *new; | |
cc5e31a4 | 4713 | int num_pages = num_extent_pages(src); |
815a51c7 | 4714 | |
3f556f78 | 4715 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
4716 | if (new == NULL) |
4717 | return NULL; | |
4718 | ||
4719 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4720 | p = alloc_page(GFP_NOFS); |
db7f3436 JB |
4721 | if (!p) { |
4722 | btrfs_release_extent_buffer(new); | |
4723 | return NULL; | |
4724 | } | |
815a51c7 JS |
4725 | attach_extent_buffer_page(new, p); |
4726 | WARN_ON(PageDirty(p)); | |
4727 | SetPageUptodate(p); | |
4728 | new->pages[i] = p; | |
fba1acf9 | 4729 | copy_page(page_address(p), page_address(src->pages[i])); |
815a51c7 JS |
4730 | } |
4731 | ||
815a51c7 | 4732 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); |
b0132a3b | 4733 | set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags); |
815a51c7 JS |
4734 | |
4735 | return new; | |
4736 | } | |
4737 | ||
0f331229 OS |
4738 | struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
4739 | u64 start, unsigned long len) | |
815a51c7 JS |
4740 | { |
4741 | struct extent_buffer *eb; | |
cc5e31a4 DS |
4742 | int num_pages; |
4743 | int i; | |
815a51c7 | 4744 | |
3f556f78 | 4745 | eb = __alloc_extent_buffer(fs_info, start, len); |
815a51c7 JS |
4746 | if (!eb) |
4747 | return NULL; | |
4748 | ||
65ad0104 | 4749 | num_pages = num_extent_pages(eb); |
815a51c7 | 4750 | for (i = 0; i < num_pages; i++) { |
9ec72677 | 4751 | eb->pages[i] = alloc_page(GFP_NOFS); |
815a51c7 JS |
4752 | if (!eb->pages[i]) |
4753 | goto err; | |
4754 | } | |
4755 | set_extent_buffer_uptodate(eb); | |
4756 | btrfs_set_header_nritems(eb, 0); | |
b0132a3b | 4757 | set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
815a51c7 JS |
4758 | |
4759 | return eb; | |
4760 | err: | |
84167d19 SB |
4761 | for (; i > 0; i--) |
4762 | __free_page(eb->pages[i - 1]); | |
815a51c7 JS |
4763 | __free_extent_buffer(eb); |
4764 | return NULL; | |
4765 | } | |
4766 | ||
0f331229 | 4767 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
da17066c | 4768 | u64 start) |
0f331229 | 4769 | { |
da17066c | 4770 | return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize); |
0f331229 OS |
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 | 4808 | { |
cc5e31a4 | 4809 | int num_pages, i; |
5df4235e | 4810 | |
0b32f4bb JB |
4811 | check_buffer_tree_ref(eb); |
4812 | ||
65ad0104 | 4813 | num_pages = num_extent_pages(eb); |
5df4235e | 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 | 4828 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
09cbfeaf | 4829 | start >> PAGE_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, | |
da17066c | 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; | |
da17066c | 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: | |
e1860a77 | 4874 | ret = radix_tree_preload(GFP_NOFS); |
faa2dbf0 JB |
4875 | if (ret) |
4876 | goto free_eb; | |
4877 | spin_lock(&fs_info->buffer_lock); | |
4878 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
09cbfeaf | 4879 | start >> PAGE_SHIFT, eb); |
faa2dbf0 JB |
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 | ||
faa2dbf0 JB |
4892 | return eb; |
4893 | free_eb: | |
4894 | btrfs_release_extent_buffer(eb); | |
4895 | return exists; | |
4896 | } | |
4897 | #endif | |
4898 | ||
f28491e0 | 4899 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
ce3e6984 | 4900 | u64 start) |
d1310b2e | 4901 | { |
da17066c | 4902 | unsigned long len = fs_info->nodesize; |
cc5e31a4 DS |
4903 | int num_pages; |
4904 | int i; | |
09cbfeaf | 4905 | unsigned long index = start >> PAGE_SHIFT; |
d1310b2e | 4906 | struct extent_buffer *eb; |
6af118ce | 4907 | struct extent_buffer *exists = NULL; |
d1310b2e | 4908 | struct page *p; |
f28491e0 | 4909 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
d1310b2e | 4910 | int uptodate = 1; |
19fe0a8b | 4911 | int ret; |
d1310b2e | 4912 | |
da17066c | 4913 | if (!IS_ALIGNED(start, fs_info->sectorsize)) { |
c871b0f2 LB |
4914 | btrfs_err(fs_info, "bad tree block start %llu", start); |
4915 | return ERR_PTR(-EINVAL); | |
4916 | } | |
4917 | ||
f28491e0 | 4918 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 4919 | if (eb) |
6af118ce | 4920 | return eb; |
6af118ce | 4921 | |
23d79d81 | 4922 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 4923 | if (!eb) |
c871b0f2 | 4924 | return ERR_PTR(-ENOMEM); |
d1310b2e | 4925 | |
65ad0104 | 4926 | num_pages = num_extent_pages(eb); |
727011e0 | 4927 | for (i = 0; i < num_pages; i++, index++) { |
d1b5c567 | 4928 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
c871b0f2 LB |
4929 | if (!p) { |
4930 | exists = ERR_PTR(-ENOMEM); | |
6af118ce | 4931 | goto free_eb; |
c871b0f2 | 4932 | } |
4f2de97a JB |
4933 | |
4934 | spin_lock(&mapping->private_lock); | |
4935 | if (PagePrivate(p)) { | |
4936 | /* | |
4937 | * We could have already allocated an eb for this page | |
4938 | * and attached one so lets see if we can get a ref on | |
4939 | * the existing eb, and if we can we know it's good and | |
4940 | * we can just return that one, else we know we can just | |
4941 | * overwrite page->private. | |
4942 | */ | |
4943 | exists = (struct extent_buffer *)p->private; | |
4944 | if (atomic_inc_not_zero(&exists->refs)) { | |
4945 | spin_unlock(&mapping->private_lock); | |
4946 | unlock_page(p); | |
09cbfeaf | 4947 | put_page(p); |
2457aec6 | 4948 | mark_extent_buffer_accessed(exists, p); |
4f2de97a JB |
4949 | goto free_eb; |
4950 | } | |
5ca64f45 | 4951 | exists = NULL; |
4f2de97a | 4952 | |
0b32f4bb | 4953 | /* |
4f2de97a JB |
4954 | * Do this so attach doesn't complain and we need to |
4955 | * drop the ref the old guy had. | |
4956 | */ | |
4957 | ClearPagePrivate(p); | |
0b32f4bb | 4958 | WARN_ON(PageDirty(p)); |
09cbfeaf | 4959 | put_page(p); |
d1310b2e | 4960 | } |
4f2de97a JB |
4961 | attach_extent_buffer_page(eb, p); |
4962 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 4963 | WARN_ON(PageDirty(p)); |
727011e0 | 4964 | eb->pages[i] = p; |
d1310b2e CM |
4965 | if (!PageUptodate(p)) |
4966 | uptodate = 0; | |
eb14ab8e CM |
4967 | |
4968 | /* | |
b16d011e NB |
4969 | * We can't unlock the pages just yet since the extent buffer |
4970 | * hasn't been properly inserted in the radix tree, this | |
4971 | * opens a race with btree_releasepage which can free a page | |
4972 | * while we are still filling in all pages for the buffer and | |
4973 | * we could crash. | |
eb14ab8e | 4974 | */ |
d1310b2e CM |
4975 | } |
4976 | if (uptodate) | |
b4ce94de | 4977 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 4978 | again: |
e1860a77 | 4979 | ret = radix_tree_preload(GFP_NOFS); |
c871b0f2 LB |
4980 | if (ret) { |
4981 | exists = ERR_PTR(ret); | |
19fe0a8b | 4982 | goto free_eb; |
c871b0f2 | 4983 | } |
19fe0a8b | 4984 | |
f28491e0 JB |
4985 | spin_lock(&fs_info->buffer_lock); |
4986 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
09cbfeaf | 4987 | start >> PAGE_SHIFT, eb); |
f28491e0 | 4988 | spin_unlock(&fs_info->buffer_lock); |
452c75c3 | 4989 | radix_tree_preload_end(); |
19fe0a8b | 4990 | if (ret == -EEXIST) { |
f28491e0 | 4991 | exists = find_extent_buffer(fs_info, start); |
452c75c3 CS |
4992 | if (exists) |
4993 | goto free_eb; | |
4994 | else | |
115391d2 | 4995 | goto again; |
6af118ce | 4996 | } |
6af118ce | 4997 | /* add one reference for the tree */ |
0b32f4bb | 4998 | check_buffer_tree_ref(eb); |
34b41ace | 4999 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
5000 | |
5001 | /* | |
b16d011e NB |
5002 | * Now it's safe to unlock the pages because any calls to |
5003 | * btree_releasepage will correctly detect that a page belongs to a | |
5004 | * live buffer and won't free them prematurely. | |
eb14ab8e | 5005 | */ |
28187ae5 NB |
5006 | for (i = 0; i < num_pages; i++) |
5007 | unlock_page(eb->pages[i]); | |
d1310b2e CM |
5008 | return eb; |
5009 | ||
6af118ce | 5010 | free_eb: |
5ca64f45 | 5011 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
5012 | for (i = 0; i < num_pages; i++) { |
5013 | if (eb->pages[i]) | |
5014 | unlock_page(eb->pages[i]); | |
5015 | } | |
eb14ab8e | 5016 | |
897ca6e9 | 5017 | btrfs_release_extent_buffer(eb); |
6af118ce | 5018 | return exists; |
d1310b2e | 5019 | } |
d1310b2e | 5020 | |
3083ee2e JB |
5021 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
5022 | { | |
5023 | struct extent_buffer *eb = | |
5024 | container_of(head, struct extent_buffer, rcu_head); | |
5025 | ||
5026 | __free_extent_buffer(eb); | |
5027 | } | |
5028 | ||
f7a52a40 | 5029 | static int release_extent_buffer(struct extent_buffer *eb) |
3083ee2e | 5030 | { |
07e21c4d NB |
5031 | lockdep_assert_held(&eb->refs_lock); |
5032 | ||
3083ee2e JB |
5033 | WARN_ON(atomic_read(&eb->refs) == 0); |
5034 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 5035 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 5036 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 5037 | |
815a51c7 | 5038 | spin_unlock(&eb->refs_lock); |
3083ee2e | 5039 | |
f28491e0 JB |
5040 | spin_lock(&fs_info->buffer_lock); |
5041 | radix_tree_delete(&fs_info->buffer_radix, | |
09cbfeaf | 5042 | eb->start >> PAGE_SHIFT); |
f28491e0 | 5043 | spin_unlock(&fs_info->buffer_lock); |
34b41ace JB |
5044 | } else { |
5045 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 5046 | } |
3083ee2e JB |
5047 | |
5048 | /* Should be safe to release our pages at this point */ | |
55ac0139 | 5049 | btrfs_release_extent_buffer_pages(eb); |
bcb7e449 | 5050 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
b0132a3b | 5051 | if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) { |
bcb7e449 JB |
5052 | __free_extent_buffer(eb); |
5053 | return 1; | |
5054 | } | |
5055 | #endif | |
3083ee2e | 5056 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 5057 | return 1; |
3083ee2e JB |
5058 | } |
5059 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
5060 | |
5061 | return 0; | |
3083ee2e JB |
5062 | } |
5063 | ||
d1310b2e CM |
5064 | void free_extent_buffer(struct extent_buffer *eb) |
5065 | { | |
242e18c7 CM |
5066 | int refs; |
5067 | int old; | |
d1310b2e CM |
5068 | if (!eb) |
5069 | return; | |
5070 | ||
242e18c7 CM |
5071 | while (1) { |
5072 | refs = atomic_read(&eb->refs); | |
46cc775e NB |
5073 | if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3) |
5074 | || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && | |
5075 | refs == 1)) | |
242e18c7 CM |
5076 | break; |
5077 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
5078 | if (old == refs) | |
5079 | return; | |
5080 | } | |
5081 | ||
3083ee2e JB |
5082 | spin_lock(&eb->refs_lock); |
5083 | if (atomic_read(&eb->refs) == 2 && | |
5084 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 5085 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
5086 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5087 | atomic_dec(&eb->refs); | |
5088 | ||
5089 | /* | |
5090 | * I know this is terrible, but it's temporary until we stop tracking | |
5091 | * the uptodate bits and such for the extent buffers. | |
5092 | */ | |
f7a52a40 | 5093 | release_extent_buffer(eb); |
3083ee2e JB |
5094 | } |
5095 | ||
5096 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
5097 | { | |
5098 | if (!eb) | |
d1310b2e CM |
5099 | return; |
5100 | ||
3083ee2e JB |
5101 | spin_lock(&eb->refs_lock); |
5102 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
5103 | ||
0b32f4bb | 5104 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
5105 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5106 | atomic_dec(&eb->refs); | |
f7a52a40 | 5107 | release_extent_buffer(eb); |
d1310b2e | 5108 | } |
d1310b2e | 5109 | |
1d4284bd | 5110 | void clear_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5111 | { |
cc5e31a4 DS |
5112 | int i; |
5113 | int num_pages; | |
d1310b2e CM |
5114 | struct page *page; |
5115 | ||
65ad0104 | 5116 | num_pages = num_extent_pages(eb); |
d1310b2e CM |
5117 | |
5118 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 5119 | page = eb->pages[i]; |
b9473439 | 5120 | if (!PageDirty(page)) |
d2c3f4f6 CM |
5121 | continue; |
5122 | ||
a61e6f29 | 5123 | lock_page(page); |
eb14ab8e CM |
5124 | WARN_ON(!PagePrivate(page)); |
5125 | ||
d1310b2e | 5126 | clear_page_dirty_for_io(page); |
b93b0163 | 5127 | xa_lock_irq(&page->mapping->i_pages); |
0a943c65 MW |
5128 | if (!PageDirty(page)) |
5129 | __xa_clear_mark(&page->mapping->i_pages, | |
5130 | page_index(page), PAGECACHE_TAG_DIRTY); | |
b93b0163 | 5131 | xa_unlock_irq(&page->mapping->i_pages); |
bf0da8c1 | 5132 | ClearPageError(page); |
a61e6f29 | 5133 | unlock_page(page); |
d1310b2e | 5134 | } |
0b32f4bb | 5135 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 5136 | } |
d1310b2e | 5137 | |
abb57ef3 | 5138 | bool set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5139 | { |
cc5e31a4 DS |
5140 | int i; |
5141 | int num_pages; | |
abb57ef3 | 5142 | bool was_dirty; |
d1310b2e | 5143 | |
0b32f4bb JB |
5144 | check_buffer_tree_ref(eb); |
5145 | ||
b9473439 | 5146 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 5147 | |
65ad0104 | 5148 | num_pages = num_extent_pages(eb); |
3083ee2e | 5149 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
5150 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
5151 | ||
abb57ef3 LB |
5152 | if (!was_dirty) |
5153 | for (i = 0; i < num_pages; i++) | |
5154 | set_page_dirty(eb->pages[i]); | |
51995c39 LB |
5155 | |
5156 | #ifdef CONFIG_BTRFS_DEBUG | |
5157 | for (i = 0; i < num_pages; i++) | |
5158 | ASSERT(PageDirty(eb->pages[i])); | |
5159 | #endif | |
5160 | ||
b9473439 | 5161 | return was_dirty; |
d1310b2e | 5162 | } |
d1310b2e | 5163 | |
69ba3927 | 5164 | void clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 | 5165 | { |
cc5e31a4 | 5166 | int i; |
1259ab75 | 5167 | struct page *page; |
cc5e31a4 | 5168 | int num_pages; |
1259ab75 | 5169 | |
b4ce94de | 5170 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 5171 | num_pages = num_extent_pages(eb); |
1259ab75 | 5172 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5173 | page = eb->pages[i]; |
33958dc6 CM |
5174 | if (page) |
5175 | ClearPageUptodate(page); | |
1259ab75 | 5176 | } |
1259ab75 CM |
5177 | } |
5178 | ||
09c25a8c | 5179 | void set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 5180 | { |
cc5e31a4 | 5181 | int i; |
d1310b2e | 5182 | struct page *page; |
cc5e31a4 | 5183 | int num_pages; |
d1310b2e | 5184 | |
0b32f4bb | 5185 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 5186 | num_pages = num_extent_pages(eb); |
d1310b2e | 5187 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5188 | page = eb->pages[i]; |
d1310b2e CM |
5189 | SetPageUptodate(page); |
5190 | } | |
d1310b2e | 5191 | } |
d1310b2e | 5192 | |
d1310b2e | 5193 | int read_extent_buffer_pages(struct extent_io_tree *tree, |
6af49dbd | 5194 | struct extent_buffer *eb, int wait, int mirror_num) |
d1310b2e | 5195 | { |
cc5e31a4 | 5196 | int i; |
d1310b2e CM |
5197 | struct page *page; |
5198 | int err; | |
5199 | int ret = 0; | |
ce9adaa5 CM |
5200 | int locked_pages = 0; |
5201 | int all_uptodate = 1; | |
cc5e31a4 | 5202 | int num_pages; |
727011e0 | 5203 | unsigned long num_reads = 0; |
a86c12c7 | 5204 | struct bio *bio = NULL; |
c8b97818 | 5205 | unsigned long bio_flags = 0; |
a86c12c7 | 5206 | |
b4ce94de | 5207 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
5208 | return 0; |
5209 | ||
65ad0104 | 5210 | num_pages = num_extent_pages(eb); |
8436ea91 | 5211 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5212 | page = eb->pages[i]; |
bb82ab88 | 5213 | if (wait == WAIT_NONE) { |
2db04966 | 5214 | if (!trylock_page(page)) |
ce9adaa5 | 5215 | goto unlock_exit; |
d1310b2e CM |
5216 | } else { |
5217 | lock_page(page); | |
5218 | } | |
ce9adaa5 | 5219 | locked_pages++; |
2571e739 LB |
5220 | } |
5221 | /* | |
5222 | * We need to firstly lock all pages to make sure that | |
5223 | * the uptodate bit of our pages won't be affected by | |
5224 | * clear_extent_buffer_uptodate(). | |
5225 | */ | |
8436ea91 | 5226 | for (i = 0; i < num_pages; i++) { |
2571e739 | 5227 | page = eb->pages[i]; |
727011e0 CM |
5228 | if (!PageUptodate(page)) { |
5229 | num_reads++; | |
ce9adaa5 | 5230 | all_uptodate = 0; |
727011e0 | 5231 | } |
ce9adaa5 | 5232 | } |
2571e739 | 5233 | |
ce9adaa5 | 5234 | if (all_uptodate) { |
8436ea91 | 5235 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
5236 | goto unlock_exit; |
5237 | } | |
5238 | ||
656f30db | 5239 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 5240 | eb->read_mirror = 0; |
0b32f4bb | 5241 | atomic_set(&eb->io_pages, num_reads); |
8436ea91 | 5242 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5243 | page = eb->pages[i]; |
baf863b9 | 5244 | |
ce9adaa5 | 5245 | if (!PageUptodate(page)) { |
baf863b9 LB |
5246 | if (ret) { |
5247 | atomic_dec(&eb->io_pages); | |
5248 | unlock_page(page); | |
5249 | continue; | |
5250 | } | |
5251 | ||
f188591e | 5252 | ClearPageError(page); |
a86c12c7 | 5253 | err = __extent_read_full_page(tree, page, |
6af49dbd | 5254 | btree_get_extent, &bio, |
d4c7ca86 | 5255 | mirror_num, &bio_flags, |
1f7ad75b | 5256 | REQ_META); |
baf863b9 | 5257 | if (err) { |
d1310b2e | 5258 | ret = err; |
baf863b9 LB |
5259 | /* |
5260 | * We use &bio in above __extent_read_full_page, | |
5261 | * so we ensure that if it returns error, the | |
5262 | * current page fails to add itself to bio and | |
5263 | * it's been unlocked. | |
5264 | * | |
5265 | * We must dec io_pages by ourselves. | |
5266 | */ | |
5267 | atomic_dec(&eb->io_pages); | |
5268 | } | |
d1310b2e CM |
5269 | } else { |
5270 | unlock_page(page); | |
5271 | } | |
5272 | } | |
5273 | ||
355808c2 | 5274 | if (bio) { |
1f7ad75b | 5275 | err = submit_one_bio(bio, mirror_num, bio_flags); |
79787eaa JM |
5276 | if (err) |
5277 | return err; | |
355808c2 | 5278 | } |
a86c12c7 | 5279 | |
bb82ab88 | 5280 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 5281 | return ret; |
d397712b | 5282 | |
8436ea91 | 5283 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5284 | page = eb->pages[i]; |
d1310b2e | 5285 | wait_on_page_locked(page); |
d397712b | 5286 | if (!PageUptodate(page)) |
d1310b2e | 5287 | ret = -EIO; |
d1310b2e | 5288 | } |
d397712b | 5289 | |
d1310b2e | 5290 | return ret; |
ce9adaa5 CM |
5291 | |
5292 | unlock_exit: | |
d397712b | 5293 | while (locked_pages > 0) { |
ce9adaa5 | 5294 | locked_pages--; |
8436ea91 JB |
5295 | page = eb->pages[locked_pages]; |
5296 | unlock_page(page); | |
ce9adaa5 CM |
5297 | } |
5298 | return ret; | |
d1310b2e | 5299 | } |
d1310b2e | 5300 | |
1cbb1f45 JM |
5301 | void read_extent_buffer(const struct extent_buffer *eb, void *dstv, |
5302 | unsigned long start, unsigned long len) | |
d1310b2e CM |
5303 | { |
5304 | size_t cur; | |
5305 | size_t offset; | |
5306 | struct page *page; | |
5307 | char *kaddr; | |
5308 | char *dst = (char *)dstv; | |
09cbfeaf KS |
5309 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5310 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e | 5311 | |
f716abd5 LB |
5312 | if (start + len > eb->len) { |
5313 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n", | |
5314 | eb->start, eb->len, start, len); | |
5315 | memset(dst, 0, len); | |
5316 | return; | |
5317 | } | |
d1310b2e | 5318 | |
09cbfeaf | 5319 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
d1310b2e | 5320 | |
d397712b | 5321 | while (len > 0) { |
fb85fc9a | 5322 | page = eb->pages[i]; |
d1310b2e | 5323 | |
09cbfeaf | 5324 | cur = min(len, (PAGE_SIZE - offset)); |
a6591715 | 5325 | kaddr = page_address(page); |
d1310b2e | 5326 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
5327 | |
5328 | dst += cur; | |
5329 | len -= cur; | |
5330 | offset = 0; | |
5331 | i++; | |
5332 | } | |
5333 | } | |
d1310b2e | 5334 | |
1cbb1f45 JM |
5335 | int read_extent_buffer_to_user(const struct extent_buffer *eb, |
5336 | void __user *dstv, | |
5337 | unsigned long start, unsigned long len) | |
550ac1d8 GH |
5338 | { |
5339 | size_t cur; | |
5340 | size_t offset; | |
5341 | struct page *page; | |
5342 | char *kaddr; | |
5343 | char __user *dst = (char __user *)dstv; | |
09cbfeaf KS |
5344 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5345 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
550ac1d8 GH |
5346 | int ret = 0; |
5347 | ||
5348 | WARN_ON(start > eb->len); | |
5349 | WARN_ON(start + len > eb->start + eb->len); | |
5350 | ||
09cbfeaf | 5351 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
550ac1d8 GH |
5352 | |
5353 | while (len > 0) { | |
fb85fc9a | 5354 | page = eb->pages[i]; |
550ac1d8 | 5355 | |
09cbfeaf | 5356 | cur = min(len, (PAGE_SIZE - offset)); |
550ac1d8 GH |
5357 | kaddr = page_address(page); |
5358 | if (copy_to_user(dst, kaddr + offset, cur)) { | |
5359 | ret = -EFAULT; | |
5360 | break; | |
5361 | } | |
5362 | ||
5363 | dst += cur; | |
5364 | len -= cur; | |
5365 | offset = 0; | |
5366 | i++; | |
5367 | } | |
5368 | ||
5369 | return ret; | |
5370 | } | |
5371 | ||
415b35a5 LB |
5372 | /* |
5373 | * return 0 if the item is found within a page. | |
5374 | * return 1 if the item spans two pages. | |
5375 | * return -EINVAL otherwise. | |
5376 | */ | |
1cbb1f45 JM |
5377 | int map_private_extent_buffer(const struct extent_buffer *eb, |
5378 | unsigned long start, unsigned long min_len, | |
5379 | char **map, unsigned long *map_start, | |
5380 | unsigned long *map_len) | |
d1310b2e | 5381 | { |
09cbfeaf | 5382 | size_t offset = start & (PAGE_SIZE - 1); |
d1310b2e CM |
5383 | char *kaddr; |
5384 | struct page *p; | |
09cbfeaf KS |
5385 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5386 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e | 5387 | unsigned long end_i = (start_offset + start + min_len - 1) >> |
09cbfeaf | 5388 | PAGE_SHIFT; |
d1310b2e | 5389 | |
f716abd5 LB |
5390 | if (start + min_len > eb->len) { |
5391 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n", | |
5392 | eb->start, eb->len, start, min_len); | |
5393 | return -EINVAL; | |
5394 | } | |
5395 | ||
d1310b2e | 5396 | if (i != end_i) |
415b35a5 | 5397 | return 1; |
d1310b2e CM |
5398 | |
5399 | if (i == 0) { | |
5400 | offset = start_offset; | |
5401 | *map_start = 0; | |
5402 | } else { | |
5403 | offset = 0; | |
09cbfeaf | 5404 | *map_start = ((u64)i << PAGE_SHIFT) - start_offset; |
d1310b2e | 5405 | } |
d397712b | 5406 | |
fb85fc9a | 5407 | p = eb->pages[i]; |
a6591715 | 5408 | kaddr = page_address(p); |
d1310b2e | 5409 | *map = kaddr + offset; |
09cbfeaf | 5410 | *map_len = PAGE_SIZE - offset; |
d1310b2e CM |
5411 | return 0; |
5412 | } | |
d1310b2e | 5413 | |
1cbb1f45 JM |
5414 | int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv, |
5415 | unsigned long start, unsigned long len) | |
d1310b2e CM |
5416 | { |
5417 | size_t cur; | |
5418 | size_t offset; | |
5419 | struct page *page; | |
5420 | char *kaddr; | |
5421 | char *ptr = (char *)ptrv; | |
09cbfeaf KS |
5422 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5423 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e CM |
5424 | int ret = 0; |
5425 | ||
5426 | WARN_ON(start > eb->len); | |
5427 | WARN_ON(start + len > eb->start + eb->len); | |
5428 | ||
09cbfeaf | 5429 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
d1310b2e | 5430 | |
d397712b | 5431 | while (len > 0) { |
fb85fc9a | 5432 | page = eb->pages[i]; |
d1310b2e | 5433 | |
09cbfeaf | 5434 | cur = min(len, (PAGE_SIZE - offset)); |
d1310b2e | 5435 | |
a6591715 | 5436 | kaddr = page_address(page); |
d1310b2e | 5437 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
5438 | if (ret) |
5439 | break; | |
5440 | ||
5441 | ptr += cur; | |
5442 | len -= cur; | |
5443 | offset = 0; | |
5444 | i++; | |
5445 | } | |
5446 | return ret; | |
5447 | } | |
d1310b2e | 5448 | |
f157bf76 DS |
5449 | void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb, |
5450 | const void *srcv) | |
5451 | { | |
5452 | char *kaddr; | |
5453 | ||
5454 | WARN_ON(!PageUptodate(eb->pages[0])); | |
5455 | kaddr = page_address(eb->pages[0]); | |
5456 | memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv, | |
5457 | BTRFS_FSID_SIZE); | |
5458 | } | |
5459 | ||
5460 | void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv) | |
5461 | { | |
5462 | char *kaddr; | |
5463 | ||
5464 | WARN_ON(!PageUptodate(eb->pages[0])); | |
5465 | kaddr = page_address(eb->pages[0]); | |
5466 | memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv, | |
5467 | BTRFS_FSID_SIZE); | |
5468 | } | |
5469 | ||
d1310b2e CM |
5470 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, |
5471 | unsigned long start, unsigned long len) | |
5472 | { | |
5473 | size_t cur; | |
5474 | size_t offset; | |
5475 | struct page *page; | |
5476 | char *kaddr; | |
5477 | char *src = (char *)srcv; | |
09cbfeaf KS |
5478 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5479 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e CM |
5480 | |
5481 | WARN_ON(start > eb->len); | |
5482 | WARN_ON(start + len > eb->start + eb->len); | |
5483 | ||
09cbfeaf | 5484 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
d1310b2e | 5485 | |
d397712b | 5486 | while (len > 0) { |
fb85fc9a | 5487 | page = eb->pages[i]; |
d1310b2e CM |
5488 | WARN_ON(!PageUptodate(page)); |
5489 | ||
09cbfeaf | 5490 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 5491 | kaddr = page_address(page); |
d1310b2e | 5492 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
5493 | |
5494 | src += cur; | |
5495 | len -= cur; | |
5496 | offset = 0; | |
5497 | i++; | |
5498 | } | |
5499 | } | |
d1310b2e | 5500 | |
b159fa28 DS |
5501 | void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start, |
5502 | unsigned long len) | |
d1310b2e CM |
5503 | { |
5504 | size_t cur; | |
5505 | size_t offset; | |
5506 | struct page *page; | |
5507 | char *kaddr; | |
09cbfeaf KS |
5508 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5509 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e CM |
5510 | |
5511 | WARN_ON(start > eb->len); | |
5512 | WARN_ON(start + len > eb->start + eb->len); | |
5513 | ||
09cbfeaf | 5514 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
d1310b2e | 5515 | |
d397712b | 5516 | while (len > 0) { |
fb85fc9a | 5517 | page = eb->pages[i]; |
d1310b2e CM |
5518 | WARN_ON(!PageUptodate(page)); |
5519 | ||
09cbfeaf | 5520 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 5521 | kaddr = page_address(page); |
b159fa28 | 5522 | memset(kaddr + offset, 0, cur); |
d1310b2e CM |
5523 | |
5524 | len -= cur; | |
5525 | offset = 0; | |
5526 | i++; | |
5527 | } | |
5528 | } | |
d1310b2e | 5529 | |
58e8012c DS |
5530 | void copy_extent_buffer_full(struct extent_buffer *dst, |
5531 | struct extent_buffer *src) | |
5532 | { | |
5533 | int i; | |
cc5e31a4 | 5534 | int num_pages; |
58e8012c DS |
5535 | |
5536 | ASSERT(dst->len == src->len); | |
5537 | ||
65ad0104 | 5538 | num_pages = num_extent_pages(dst); |
58e8012c DS |
5539 | for (i = 0; i < num_pages; i++) |
5540 | copy_page(page_address(dst->pages[i]), | |
5541 | page_address(src->pages[i])); | |
5542 | } | |
5543 | ||
d1310b2e CM |
5544 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, |
5545 | unsigned long dst_offset, unsigned long src_offset, | |
5546 | unsigned long len) | |
5547 | { | |
5548 | u64 dst_len = dst->len; | |
5549 | size_t cur; | |
5550 | size_t offset; | |
5551 | struct page *page; | |
5552 | char *kaddr; | |
09cbfeaf KS |
5553 | size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1); |
5554 | unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT; | |
d1310b2e CM |
5555 | |
5556 | WARN_ON(src->len != dst_len); | |
5557 | ||
5558 | offset = (start_offset + dst_offset) & | |
09cbfeaf | 5559 | (PAGE_SIZE - 1); |
d1310b2e | 5560 | |
d397712b | 5561 | while (len > 0) { |
fb85fc9a | 5562 | page = dst->pages[i]; |
d1310b2e CM |
5563 | WARN_ON(!PageUptodate(page)); |
5564 | ||
09cbfeaf | 5565 | cur = min(len, (unsigned long)(PAGE_SIZE - offset)); |
d1310b2e | 5566 | |
a6591715 | 5567 | kaddr = page_address(page); |
d1310b2e | 5568 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
5569 | |
5570 | src_offset += cur; | |
5571 | len -= cur; | |
5572 | offset = 0; | |
5573 | i++; | |
5574 | } | |
5575 | } | |
d1310b2e | 5576 | |
3e1e8bb7 OS |
5577 | /* |
5578 | * eb_bitmap_offset() - calculate the page and offset of the byte containing the | |
5579 | * given bit number | |
5580 | * @eb: the extent buffer | |
5581 | * @start: offset of the bitmap item in the extent buffer | |
5582 | * @nr: bit number | |
5583 | * @page_index: return index of the page in the extent buffer that contains the | |
5584 | * given bit number | |
5585 | * @page_offset: return offset into the page given by page_index | |
5586 | * | |
5587 | * This helper hides the ugliness of finding the byte in an extent buffer which | |
5588 | * contains a given bit. | |
5589 | */ | |
5590 | static inline void eb_bitmap_offset(struct extent_buffer *eb, | |
5591 | unsigned long start, unsigned long nr, | |
5592 | unsigned long *page_index, | |
5593 | size_t *page_offset) | |
5594 | { | |
09cbfeaf | 5595 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
3e1e8bb7 OS |
5596 | size_t byte_offset = BIT_BYTE(nr); |
5597 | size_t offset; | |
5598 | ||
5599 | /* | |
5600 | * The byte we want is the offset of the extent buffer + the offset of | |
5601 | * the bitmap item in the extent buffer + the offset of the byte in the | |
5602 | * bitmap item. | |
5603 | */ | |
5604 | offset = start_offset + start + byte_offset; | |
5605 | ||
09cbfeaf KS |
5606 | *page_index = offset >> PAGE_SHIFT; |
5607 | *page_offset = offset & (PAGE_SIZE - 1); | |
3e1e8bb7 OS |
5608 | } |
5609 | ||
5610 | /** | |
5611 | * extent_buffer_test_bit - determine whether a bit in a bitmap item is set | |
5612 | * @eb: the extent buffer | |
5613 | * @start: offset of the bitmap item in the extent buffer | |
5614 | * @nr: bit number to test | |
5615 | */ | |
5616 | int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start, | |
5617 | unsigned long nr) | |
5618 | { | |
2fe1d551 | 5619 | u8 *kaddr; |
3e1e8bb7 OS |
5620 | struct page *page; |
5621 | unsigned long i; | |
5622 | size_t offset; | |
5623 | ||
5624 | eb_bitmap_offset(eb, start, nr, &i, &offset); | |
5625 | page = eb->pages[i]; | |
5626 | WARN_ON(!PageUptodate(page)); | |
5627 | kaddr = page_address(page); | |
5628 | return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1))); | |
5629 | } | |
5630 | ||
5631 | /** | |
5632 | * extent_buffer_bitmap_set - set an area of a bitmap | |
5633 | * @eb: the extent buffer | |
5634 | * @start: offset of the bitmap item in the extent buffer | |
5635 | * @pos: bit number of the first bit | |
5636 | * @len: number of bits to set | |
5637 | */ | |
5638 | void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start, | |
5639 | unsigned long pos, unsigned long len) | |
5640 | { | |
2fe1d551 | 5641 | u8 *kaddr; |
3e1e8bb7 OS |
5642 | struct page *page; |
5643 | unsigned long i; | |
5644 | size_t offset; | |
5645 | const unsigned int size = pos + len; | |
5646 | int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 5647 | u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
5648 | |
5649 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5650 | page = eb->pages[i]; | |
5651 | WARN_ON(!PageUptodate(page)); | |
5652 | kaddr = page_address(page); | |
5653 | ||
5654 | while (len >= bits_to_set) { | |
5655 | kaddr[offset] |= mask_to_set; | |
5656 | len -= bits_to_set; | |
5657 | bits_to_set = BITS_PER_BYTE; | |
9c894696 | 5658 | mask_to_set = ~0; |
09cbfeaf | 5659 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5660 | offset = 0; |
5661 | page = eb->pages[++i]; | |
5662 | WARN_ON(!PageUptodate(page)); | |
5663 | kaddr = page_address(page); | |
5664 | } | |
5665 | } | |
5666 | if (len) { | |
5667 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
5668 | kaddr[offset] |= mask_to_set; | |
5669 | } | |
5670 | } | |
5671 | ||
5672 | ||
5673 | /** | |
5674 | * extent_buffer_bitmap_clear - clear an area of a bitmap | |
5675 | * @eb: the extent buffer | |
5676 | * @start: offset of the bitmap item in the extent buffer | |
5677 | * @pos: bit number of the first bit | |
5678 | * @len: number of bits to clear | |
5679 | */ | |
5680 | void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start, | |
5681 | unsigned long pos, unsigned long len) | |
5682 | { | |
2fe1d551 | 5683 | u8 *kaddr; |
3e1e8bb7 OS |
5684 | struct page *page; |
5685 | unsigned long i; | |
5686 | size_t offset; | |
5687 | const unsigned int size = pos + len; | |
5688 | int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 5689 | u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
5690 | |
5691 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5692 | page = eb->pages[i]; | |
5693 | WARN_ON(!PageUptodate(page)); | |
5694 | kaddr = page_address(page); | |
5695 | ||
5696 | while (len >= bits_to_clear) { | |
5697 | kaddr[offset] &= ~mask_to_clear; | |
5698 | len -= bits_to_clear; | |
5699 | bits_to_clear = BITS_PER_BYTE; | |
9c894696 | 5700 | mask_to_clear = ~0; |
09cbfeaf | 5701 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5702 | offset = 0; |
5703 | page = eb->pages[++i]; | |
5704 | WARN_ON(!PageUptodate(page)); | |
5705 | kaddr = page_address(page); | |
5706 | } | |
5707 | } | |
5708 | if (len) { | |
5709 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
5710 | kaddr[offset] &= ~mask_to_clear; | |
5711 | } | |
5712 | } | |
5713 | ||
3387206f ST |
5714 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
5715 | { | |
5716 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
5717 | return distance < len; | |
5718 | } | |
5719 | ||
d1310b2e CM |
5720 | static void copy_pages(struct page *dst_page, struct page *src_page, |
5721 | unsigned long dst_off, unsigned long src_off, | |
5722 | unsigned long len) | |
5723 | { | |
a6591715 | 5724 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 5725 | char *src_kaddr; |
727011e0 | 5726 | int must_memmove = 0; |
d1310b2e | 5727 | |
3387206f | 5728 | if (dst_page != src_page) { |
a6591715 | 5729 | src_kaddr = page_address(src_page); |
3387206f | 5730 | } else { |
d1310b2e | 5731 | src_kaddr = dst_kaddr; |
727011e0 CM |
5732 | if (areas_overlap(src_off, dst_off, len)) |
5733 | must_memmove = 1; | |
3387206f | 5734 | } |
d1310b2e | 5735 | |
727011e0 CM |
5736 | if (must_memmove) |
5737 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
5738 | else | |
5739 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
5740 | } |
5741 | ||
5742 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5743 | unsigned long src_offset, unsigned long len) | |
5744 | { | |
0b246afa | 5745 | struct btrfs_fs_info *fs_info = dst->fs_info; |
d1310b2e CM |
5746 | size_t cur; |
5747 | size_t dst_off_in_page; | |
5748 | size_t src_off_in_page; | |
09cbfeaf | 5749 | size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1); |
d1310b2e CM |
5750 | unsigned long dst_i; |
5751 | unsigned long src_i; | |
5752 | ||
5753 | if (src_offset + len > dst->len) { | |
0b246afa | 5754 | btrfs_err(fs_info, |
5d163e0e JM |
5755 | "memmove bogus src_offset %lu move len %lu dst len %lu", |
5756 | src_offset, len, dst->len); | |
d1310b2e CM |
5757 | BUG_ON(1); |
5758 | } | |
5759 | if (dst_offset + len > dst->len) { | |
0b246afa | 5760 | btrfs_err(fs_info, |
5d163e0e JM |
5761 | "memmove bogus dst_offset %lu move len %lu dst len %lu", |
5762 | dst_offset, len, dst->len); | |
d1310b2e CM |
5763 | BUG_ON(1); |
5764 | } | |
5765 | ||
d397712b | 5766 | while (len > 0) { |
d1310b2e | 5767 | dst_off_in_page = (start_offset + dst_offset) & |
09cbfeaf | 5768 | (PAGE_SIZE - 1); |
d1310b2e | 5769 | src_off_in_page = (start_offset + src_offset) & |
09cbfeaf | 5770 | (PAGE_SIZE - 1); |
d1310b2e | 5771 | |
09cbfeaf KS |
5772 | dst_i = (start_offset + dst_offset) >> PAGE_SHIFT; |
5773 | src_i = (start_offset + src_offset) >> PAGE_SHIFT; | |
d1310b2e | 5774 | |
09cbfeaf | 5775 | cur = min(len, (unsigned long)(PAGE_SIZE - |
d1310b2e CM |
5776 | src_off_in_page)); |
5777 | cur = min_t(unsigned long, cur, | |
09cbfeaf | 5778 | (unsigned long)(PAGE_SIZE - dst_off_in_page)); |
d1310b2e | 5779 | |
fb85fc9a | 5780 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5781 | dst_off_in_page, src_off_in_page, cur); |
5782 | ||
5783 | src_offset += cur; | |
5784 | dst_offset += cur; | |
5785 | len -= cur; | |
5786 | } | |
5787 | } | |
d1310b2e CM |
5788 | |
5789 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5790 | unsigned long src_offset, unsigned long len) | |
5791 | { | |
0b246afa | 5792 | struct btrfs_fs_info *fs_info = dst->fs_info; |
d1310b2e CM |
5793 | size_t cur; |
5794 | size_t dst_off_in_page; | |
5795 | size_t src_off_in_page; | |
5796 | unsigned long dst_end = dst_offset + len - 1; | |
5797 | unsigned long src_end = src_offset + len - 1; | |
09cbfeaf | 5798 | size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1); |
d1310b2e CM |
5799 | unsigned long dst_i; |
5800 | unsigned long src_i; | |
5801 | ||
5802 | if (src_offset + len > dst->len) { | |
0b246afa | 5803 | btrfs_err(fs_info, |
5d163e0e JM |
5804 | "memmove bogus src_offset %lu move len %lu len %lu", |
5805 | src_offset, len, dst->len); | |
d1310b2e CM |
5806 | BUG_ON(1); |
5807 | } | |
5808 | if (dst_offset + len > dst->len) { | |
0b246afa | 5809 | btrfs_err(fs_info, |
5d163e0e JM |
5810 | "memmove bogus dst_offset %lu move len %lu len %lu", |
5811 | dst_offset, len, dst->len); | |
d1310b2e CM |
5812 | BUG_ON(1); |
5813 | } | |
727011e0 | 5814 | if (dst_offset < src_offset) { |
d1310b2e CM |
5815 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
5816 | return; | |
5817 | } | |
d397712b | 5818 | while (len > 0) { |
09cbfeaf KS |
5819 | dst_i = (start_offset + dst_end) >> PAGE_SHIFT; |
5820 | src_i = (start_offset + src_end) >> PAGE_SHIFT; | |
d1310b2e CM |
5821 | |
5822 | dst_off_in_page = (start_offset + dst_end) & | |
09cbfeaf | 5823 | (PAGE_SIZE - 1); |
d1310b2e | 5824 | src_off_in_page = (start_offset + src_end) & |
09cbfeaf | 5825 | (PAGE_SIZE - 1); |
d1310b2e CM |
5826 | |
5827 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
5828 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 5829 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5830 | dst_off_in_page - cur + 1, |
5831 | src_off_in_page - cur + 1, cur); | |
5832 | ||
5833 | dst_end -= cur; | |
5834 | src_end -= cur; | |
5835 | len -= cur; | |
5836 | } | |
5837 | } | |
6af118ce | 5838 | |
f7a52a40 | 5839 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 5840 | { |
6af118ce | 5841 | struct extent_buffer *eb; |
6af118ce | 5842 | |
3083ee2e | 5843 | /* |
01327610 | 5844 | * We need to make sure nobody is attaching this page to an eb right |
3083ee2e JB |
5845 | * now. |
5846 | */ | |
5847 | spin_lock(&page->mapping->private_lock); | |
5848 | if (!PagePrivate(page)) { | |
5849 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 5850 | return 1; |
45f49bce | 5851 | } |
6af118ce | 5852 | |
3083ee2e JB |
5853 | eb = (struct extent_buffer *)page->private; |
5854 | BUG_ON(!eb); | |
19fe0a8b MX |
5855 | |
5856 | /* | |
3083ee2e JB |
5857 | * This is a little awful but should be ok, we need to make sure that |
5858 | * the eb doesn't disappear out from under us while we're looking at | |
5859 | * this page. | |
19fe0a8b | 5860 | */ |
3083ee2e | 5861 | spin_lock(&eb->refs_lock); |
0b32f4bb | 5862 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
5863 | spin_unlock(&eb->refs_lock); |
5864 | spin_unlock(&page->mapping->private_lock); | |
5865 | return 0; | |
b9473439 | 5866 | } |
3083ee2e | 5867 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 5868 | |
19fe0a8b | 5869 | /* |
3083ee2e JB |
5870 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
5871 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 5872 | */ |
3083ee2e JB |
5873 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
5874 | spin_unlock(&eb->refs_lock); | |
5875 | return 0; | |
b9473439 | 5876 | } |
19fe0a8b | 5877 | |
f7a52a40 | 5878 | return release_extent_buffer(eb); |
6af118ce | 5879 | } |