]>
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. | |
52042d8e | 495 | * it will optionally wake up anyone 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; | |
0ca1f7ce | 588 | |
2ac55d41 JB |
589 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
590 | clear = 1; | |
d1310b2e | 591 | again: |
d0164adc | 592 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
c7bc6319 FM |
593 | /* |
594 | * Don't care for allocation failure here because we might end | |
595 | * up not needing the pre-allocated extent state at all, which | |
596 | * is the case if we only have in the tree extent states that | |
597 | * cover our input range and don't cover too any other range. | |
598 | * If we end up needing a new extent state we allocate it later. | |
599 | */ | |
d1310b2e | 600 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
601 | } |
602 | ||
cad321ad | 603 | spin_lock(&tree->lock); |
2c64c53d CM |
604 | if (cached_state) { |
605 | cached = *cached_state; | |
2ac55d41 JB |
606 | |
607 | if (clear) { | |
608 | *cached_state = NULL; | |
609 | cached_state = NULL; | |
610 | } | |
611 | ||
27a3507d FM |
612 | if (cached && extent_state_in_tree(cached) && |
613 | cached->start <= start && cached->end > start) { | |
2ac55d41 | 614 | if (clear) |
b7ac31b7 | 615 | refcount_dec(&cached->refs); |
2c64c53d | 616 | state = cached; |
42daec29 | 617 | goto hit_next; |
2c64c53d | 618 | } |
2ac55d41 JB |
619 | if (clear) |
620 | free_extent_state(cached); | |
2c64c53d | 621 | } |
d1310b2e CM |
622 | /* |
623 | * this search will find the extents that end after | |
624 | * our range starts | |
625 | */ | |
80ea96b1 | 626 | node = tree_search(tree, start); |
d1310b2e CM |
627 | if (!node) |
628 | goto out; | |
629 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 630 | hit_next: |
d1310b2e CM |
631 | if (state->start > end) |
632 | goto out; | |
633 | WARN_ON(state->end < start); | |
5c939df5 | 634 | last_end = state->end; |
d1310b2e | 635 | |
0449314a | 636 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
637 | if (!(state->state & bits)) { |
638 | state = next_state(state); | |
0449314a | 639 | goto next; |
cdc6a395 | 640 | } |
0449314a | 641 | |
d1310b2e CM |
642 | /* |
643 | * | ---- desired range ---- | | |
644 | * | state | or | |
645 | * | ------------- state -------------- | | |
646 | * | |
647 | * We need to split the extent we found, and may flip | |
648 | * bits on second half. | |
649 | * | |
650 | * If the extent we found extends past our range, we | |
651 | * just split and search again. It'll get split again | |
652 | * the next time though. | |
653 | * | |
654 | * If the extent we found is inside our range, we clear | |
655 | * the desired bit on it. | |
656 | */ | |
657 | ||
658 | if (state->start < start) { | |
8233767a XG |
659 | prealloc = alloc_extent_state_atomic(prealloc); |
660 | BUG_ON(!prealloc); | |
d1310b2e | 661 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
662 | if (err) |
663 | extent_io_tree_panic(tree, err); | |
664 | ||
d1310b2e CM |
665 | prealloc = NULL; |
666 | if (err) | |
667 | goto out; | |
668 | if (state->end <= end) { | |
fefdc557 QW |
669 | state = clear_state_bit(tree, state, &bits, wake, |
670 | changeset); | |
d1ac6e41 | 671 | goto next; |
d1310b2e CM |
672 | } |
673 | goto search_again; | |
674 | } | |
675 | /* | |
676 | * | ---- desired range ---- | | |
677 | * | state | | |
678 | * We need to split the extent, and clear the bit | |
679 | * on the first half | |
680 | */ | |
681 | if (state->start <= end && state->end > end) { | |
8233767a XG |
682 | prealloc = alloc_extent_state_atomic(prealloc); |
683 | BUG_ON(!prealloc); | |
d1310b2e | 684 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
685 | if (err) |
686 | extent_io_tree_panic(tree, err); | |
687 | ||
d1310b2e CM |
688 | if (wake) |
689 | wake_up(&state->wq); | |
42daec29 | 690 | |
fefdc557 | 691 | clear_state_bit(tree, prealloc, &bits, wake, changeset); |
9ed74f2d | 692 | |
d1310b2e CM |
693 | prealloc = NULL; |
694 | goto out; | |
695 | } | |
42daec29 | 696 | |
fefdc557 | 697 | state = clear_state_bit(tree, state, &bits, wake, changeset); |
0449314a | 698 | next: |
5c939df5 YZ |
699 | if (last_end == (u64)-1) |
700 | goto out; | |
701 | start = last_end + 1; | |
cdc6a395 | 702 | if (start <= end && state && !need_resched()) |
692e5759 | 703 | goto hit_next; |
d1310b2e CM |
704 | |
705 | search_again: | |
706 | if (start > end) | |
707 | goto out; | |
cad321ad | 708 | spin_unlock(&tree->lock); |
d0164adc | 709 | if (gfpflags_allow_blocking(mask)) |
d1310b2e CM |
710 | cond_resched(); |
711 | goto again; | |
7ab5cb2a DS |
712 | |
713 | out: | |
714 | spin_unlock(&tree->lock); | |
715 | if (prealloc) | |
716 | free_extent_state(prealloc); | |
717 | ||
718 | return 0; | |
719 | ||
d1310b2e | 720 | } |
d1310b2e | 721 | |
143bede5 JM |
722 | static void wait_on_state(struct extent_io_tree *tree, |
723 | struct extent_state *state) | |
641f5219 CH |
724 | __releases(tree->lock) |
725 | __acquires(tree->lock) | |
d1310b2e CM |
726 | { |
727 | DEFINE_WAIT(wait); | |
728 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 729 | spin_unlock(&tree->lock); |
d1310b2e | 730 | schedule(); |
cad321ad | 731 | spin_lock(&tree->lock); |
d1310b2e | 732 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
733 | } |
734 | ||
735 | /* | |
736 | * waits for one or more bits to clear on a range in the state tree. | |
737 | * The range [start, end] is inclusive. | |
738 | * The tree lock is taken by this function | |
739 | */ | |
41074888 DS |
740 | static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
741 | unsigned long bits) | |
d1310b2e CM |
742 | { |
743 | struct extent_state *state; | |
744 | struct rb_node *node; | |
745 | ||
a5dee37d | 746 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 747 | |
cad321ad | 748 | spin_lock(&tree->lock); |
d1310b2e CM |
749 | again: |
750 | while (1) { | |
751 | /* | |
752 | * this search will find all the extents that end after | |
753 | * our range starts | |
754 | */ | |
80ea96b1 | 755 | node = tree_search(tree, start); |
c50d3e71 | 756 | process_node: |
d1310b2e CM |
757 | if (!node) |
758 | break; | |
759 | ||
760 | state = rb_entry(node, struct extent_state, rb_node); | |
761 | ||
762 | if (state->start > end) | |
763 | goto out; | |
764 | ||
765 | if (state->state & bits) { | |
766 | start = state->start; | |
b7ac31b7 | 767 | refcount_inc(&state->refs); |
d1310b2e CM |
768 | wait_on_state(tree, state); |
769 | free_extent_state(state); | |
770 | goto again; | |
771 | } | |
772 | start = state->end + 1; | |
773 | ||
774 | if (start > end) | |
775 | break; | |
776 | ||
c50d3e71 FM |
777 | if (!cond_resched_lock(&tree->lock)) { |
778 | node = rb_next(node); | |
779 | goto process_node; | |
780 | } | |
d1310b2e CM |
781 | } |
782 | out: | |
cad321ad | 783 | spin_unlock(&tree->lock); |
d1310b2e | 784 | } |
d1310b2e | 785 | |
1bf85046 | 786 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 787 | struct extent_state *state, |
d38ed27f | 788 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e | 789 | { |
9ee49a04 | 790 | unsigned bits_to_set = *bits & ~EXTENT_CTLBITS; |
57599c7e | 791 | int ret; |
9ed74f2d | 792 | |
e06a1fc9 NB |
793 | if (tree->private_data && is_data_inode(tree->private_data)) |
794 | btrfs_set_delalloc_extent(tree->private_data, state, bits); | |
795 | ||
0ca1f7ce | 796 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
797 | u64 range = state->end - state->start + 1; |
798 | tree->dirty_bytes += range; | |
799 | } | |
57599c7e DS |
800 | ret = add_extent_changeset(state, bits_to_set, changeset, 1); |
801 | BUG_ON(ret < 0); | |
0ca1f7ce | 802 | state->state |= bits_to_set; |
d1310b2e CM |
803 | } |
804 | ||
e38e2ed7 FM |
805 | static void cache_state_if_flags(struct extent_state *state, |
806 | struct extent_state **cached_ptr, | |
9ee49a04 | 807 | unsigned flags) |
2c64c53d CM |
808 | { |
809 | if (cached_ptr && !(*cached_ptr)) { | |
e38e2ed7 | 810 | if (!flags || (state->state & flags)) { |
2c64c53d | 811 | *cached_ptr = state; |
b7ac31b7 | 812 | refcount_inc(&state->refs); |
2c64c53d CM |
813 | } |
814 | } | |
815 | } | |
816 | ||
e38e2ed7 FM |
817 | static void cache_state(struct extent_state *state, |
818 | struct extent_state **cached_ptr) | |
819 | { | |
820 | return cache_state_if_flags(state, cached_ptr, | |
821 | EXTENT_IOBITS | EXTENT_BOUNDARY); | |
822 | } | |
823 | ||
d1310b2e | 824 | /* |
1edbb734 CM |
825 | * set some bits on a range in the tree. This may require allocations or |
826 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 827 | * |
1edbb734 CM |
828 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
829 | * part of the range already has the desired bits set. The start of the | |
830 | * existing range is returned in failed_start in this case. | |
d1310b2e | 831 | * |
1edbb734 | 832 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 833 | */ |
1edbb734 | 834 | |
3fbe5c02 JM |
835 | static int __must_check |
836 | __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 837 | unsigned bits, unsigned exclusive_bits, |
41074888 | 838 | u64 *failed_start, struct extent_state **cached_state, |
d38ed27f | 839 | gfp_t mask, struct extent_changeset *changeset) |
d1310b2e CM |
840 | { |
841 | struct extent_state *state; | |
842 | struct extent_state *prealloc = NULL; | |
843 | struct rb_node *node; | |
12cfbad9 FDBM |
844 | struct rb_node **p; |
845 | struct rb_node *parent; | |
d1310b2e | 846 | int err = 0; |
d1310b2e CM |
847 | u64 last_start; |
848 | u64 last_end; | |
42daec29 | 849 | |
a5dee37d | 850 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 851 | |
d1310b2e | 852 | again: |
d0164adc | 853 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
059f791c DS |
854 | /* |
855 | * Don't care for allocation failure here because we might end | |
856 | * up not needing the pre-allocated extent state at all, which | |
857 | * is the case if we only have in the tree extent states that | |
858 | * cover our input range and don't cover too any other range. | |
859 | * If we end up needing a new extent state we allocate it later. | |
860 | */ | |
d1310b2e | 861 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
862 | } |
863 | ||
cad321ad | 864 | spin_lock(&tree->lock); |
9655d298 CM |
865 | if (cached_state && *cached_state) { |
866 | state = *cached_state; | |
df98b6e2 | 867 | if (state->start <= start && state->end > start && |
27a3507d | 868 | extent_state_in_tree(state)) { |
9655d298 CM |
869 | node = &state->rb_node; |
870 | goto hit_next; | |
871 | } | |
872 | } | |
d1310b2e CM |
873 | /* |
874 | * this search will find all the extents that end after | |
875 | * our range starts. | |
876 | */ | |
12cfbad9 | 877 | node = tree_search_for_insert(tree, start, &p, &parent); |
d1310b2e | 878 | if (!node) { |
8233767a XG |
879 | prealloc = alloc_extent_state_atomic(prealloc); |
880 | BUG_ON(!prealloc); | |
12cfbad9 | 881 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 882 | &p, &parent, &bits, changeset); |
c2d904e0 JM |
883 | if (err) |
884 | extent_io_tree_panic(tree, err); | |
885 | ||
c42ac0bc | 886 | cache_state(prealloc, cached_state); |
d1310b2e | 887 | prealloc = NULL; |
d1310b2e CM |
888 | goto out; |
889 | } | |
d1310b2e | 890 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 891 | hit_next: |
d1310b2e CM |
892 | last_start = state->start; |
893 | last_end = state->end; | |
894 | ||
895 | /* | |
896 | * | ---- desired range ---- | | |
897 | * | state | | |
898 | * | |
899 | * Just lock what we found and keep going | |
900 | */ | |
901 | if (state->start == start && state->end <= end) { | |
1edbb734 | 902 | if (state->state & exclusive_bits) { |
d1310b2e CM |
903 | *failed_start = state->start; |
904 | err = -EEXIST; | |
905 | goto out; | |
906 | } | |
42daec29 | 907 | |
d38ed27f | 908 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 909 | cache_state(state, cached_state); |
d1310b2e | 910 | merge_state(tree, state); |
5c939df5 YZ |
911 | if (last_end == (u64)-1) |
912 | goto out; | |
913 | start = last_end + 1; | |
d1ac6e41 LB |
914 | state = next_state(state); |
915 | if (start < end && state && state->start == start && | |
916 | !need_resched()) | |
917 | goto hit_next; | |
d1310b2e CM |
918 | goto search_again; |
919 | } | |
920 | ||
921 | /* | |
922 | * | ---- desired range ---- | | |
923 | * | state | | |
924 | * or | |
925 | * | ------------- state -------------- | | |
926 | * | |
927 | * We need to split the extent we found, and may flip bits on | |
928 | * second half. | |
929 | * | |
930 | * If the extent we found extends past our | |
931 | * range, we just split and search again. It'll get split | |
932 | * again the next time though. | |
933 | * | |
934 | * If the extent we found is inside our range, we set the | |
935 | * desired bit on it. | |
936 | */ | |
937 | if (state->start < start) { | |
1edbb734 | 938 | if (state->state & exclusive_bits) { |
d1310b2e CM |
939 | *failed_start = start; |
940 | err = -EEXIST; | |
941 | goto out; | |
942 | } | |
8233767a XG |
943 | |
944 | prealloc = alloc_extent_state_atomic(prealloc); | |
945 | BUG_ON(!prealloc); | |
d1310b2e | 946 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
947 | if (err) |
948 | extent_io_tree_panic(tree, err); | |
949 | ||
d1310b2e CM |
950 | prealloc = NULL; |
951 | if (err) | |
952 | goto out; | |
953 | if (state->end <= end) { | |
d38ed27f | 954 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 955 | cache_state(state, cached_state); |
d1310b2e | 956 | merge_state(tree, state); |
5c939df5 YZ |
957 | if (last_end == (u64)-1) |
958 | goto out; | |
959 | start = last_end + 1; | |
d1ac6e41 LB |
960 | state = next_state(state); |
961 | if (start < end && state && state->start == start && | |
962 | !need_resched()) | |
963 | goto hit_next; | |
d1310b2e CM |
964 | } |
965 | goto search_again; | |
966 | } | |
967 | /* | |
968 | * | ---- desired range ---- | | |
969 | * | state | or | state | | |
970 | * | |
971 | * There's a hole, we need to insert something in it and | |
972 | * ignore the extent we found. | |
973 | */ | |
974 | if (state->start > start) { | |
975 | u64 this_end; | |
976 | if (end < last_start) | |
977 | this_end = end; | |
978 | else | |
d397712b | 979 | this_end = last_start - 1; |
8233767a XG |
980 | |
981 | prealloc = alloc_extent_state_atomic(prealloc); | |
982 | BUG_ON(!prealloc); | |
c7f895a2 XG |
983 | |
984 | /* | |
985 | * Avoid to free 'prealloc' if it can be merged with | |
986 | * the later extent. | |
987 | */ | |
d1310b2e | 988 | err = insert_state(tree, prealloc, start, this_end, |
d38ed27f | 989 | NULL, NULL, &bits, changeset); |
c2d904e0 JM |
990 | if (err) |
991 | extent_io_tree_panic(tree, err); | |
992 | ||
9ed74f2d JB |
993 | cache_state(prealloc, cached_state); |
994 | prealloc = NULL; | |
d1310b2e CM |
995 | start = this_end + 1; |
996 | goto search_again; | |
997 | } | |
998 | /* | |
999 | * | ---- desired range ---- | | |
1000 | * | state | | |
1001 | * We need to split the extent, and set the bit | |
1002 | * on the first half | |
1003 | */ | |
1004 | if (state->start <= end && state->end > end) { | |
1edbb734 | 1005 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1006 | *failed_start = start; |
1007 | err = -EEXIST; | |
1008 | goto out; | |
1009 | } | |
8233767a XG |
1010 | |
1011 | prealloc = alloc_extent_state_atomic(prealloc); | |
1012 | BUG_ON(!prealloc); | |
d1310b2e | 1013 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
1014 | if (err) |
1015 | extent_io_tree_panic(tree, err); | |
d1310b2e | 1016 | |
d38ed27f | 1017 | set_state_bits(tree, prealloc, &bits, changeset); |
2c64c53d | 1018 | cache_state(prealloc, cached_state); |
d1310b2e CM |
1019 | merge_state(tree, prealloc); |
1020 | prealloc = NULL; | |
1021 | goto out; | |
1022 | } | |
1023 | ||
b5a4ba14 DS |
1024 | search_again: |
1025 | if (start > end) | |
1026 | goto out; | |
1027 | spin_unlock(&tree->lock); | |
1028 | if (gfpflags_allow_blocking(mask)) | |
1029 | cond_resched(); | |
1030 | goto again; | |
d1310b2e CM |
1031 | |
1032 | out: | |
cad321ad | 1033 | spin_unlock(&tree->lock); |
d1310b2e CM |
1034 | if (prealloc) |
1035 | free_extent_state(prealloc); | |
1036 | ||
1037 | return err; | |
1038 | ||
d1310b2e | 1039 | } |
d1310b2e | 1040 | |
41074888 | 1041 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
9ee49a04 | 1042 | unsigned bits, u64 * failed_start, |
41074888 | 1043 | struct extent_state **cached_state, gfp_t mask) |
3fbe5c02 JM |
1044 | { |
1045 | return __set_extent_bit(tree, start, end, bits, 0, failed_start, | |
d38ed27f | 1046 | cached_state, mask, NULL); |
3fbe5c02 JM |
1047 | } |
1048 | ||
1049 | ||
462d6fac | 1050 | /** |
10983f2e LB |
1051 | * convert_extent_bit - convert all bits in a given range from one bit to |
1052 | * another | |
462d6fac JB |
1053 | * @tree: the io tree to search |
1054 | * @start: the start offset in bytes | |
1055 | * @end: the end offset in bytes (inclusive) | |
1056 | * @bits: the bits to set in this range | |
1057 | * @clear_bits: the bits to clear in this range | |
e6138876 | 1058 | * @cached_state: state that we're going to cache |
462d6fac JB |
1059 | * |
1060 | * This will go through and set bits for the given range. If any states exist | |
1061 | * already in this range they are set with the given bit and cleared of the | |
1062 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
1063 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
1064 | * boundary bits like LOCK. | |
210aa277 DS |
1065 | * |
1066 | * All allocations are done with GFP_NOFS. | |
462d6fac JB |
1067 | */ |
1068 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1069 | unsigned bits, unsigned clear_bits, |
210aa277 | 1070 | struct extent_state **cached_state) |
462d6fac JB |
1071 | { |
1072 | struct extent_state *state; | |
1073 | struct extent_state *prealloc = NULL; | |
1074 | struct rb_node *node; | |
12cfbad9 FDBM |
1075 | struct rb_node **p; |
1076 | struct rb_node *parent; | |
462d6fac JB |
1077 | int err = 0; |
1078 | u64 last_start; | |
1079 | u64 last_end; | |
c8fd3de7 | 1080 | bool first_iteration = true; |
462d6fac | 1081 | |
a5dee37d | 1082 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 1083 | |
462d6fac | 1084 | again: |
210aa277 | 1085 | if (!prealloc) { |
c8fd3de7 FM |
1086 | /* |
1087 | * Best effort, don't worry if extent state allocation fails | |
1088 | * here for the first iteration. We might have a cached state | |
1089 | * that matches exactly the target range, in which case no | |
1090 | * extent state allocations are needed. We'll only know this | |
1091 | * after locking the tree. | |
1092 | */ | |
210aa277 | 1093 | prealloc = alloc_extent_state(GFP_NOFS); |
c8fd3de7 | 1094 | if (!prealloc && !first_iteration) |
462d6fac JB |
1095 | return -ENOMEM; |
1096 | } | |
1097 | ||
1098 | spin_lock(&tree->lock); | |
e6138876 JB |
1099 | if (cached_state && *cached_state) { |
1100 | state = *cached_state; | |
1101 | if (state->start <= start && state->end > start && | |
27a3507d | 1102 | extent_state_in_tree(state)) { |
e6138876 JB |
1103 | node = &state->rb_node; |
1104 | goto hit_next; | |
1105 | } | |
1106 | } | |
1107 | ||
462d6fac JB |
1108 | /* |
1109 | * this search will find all the extents that end after | |
1110 | * our range starts. | |
1111 | */ | |
12cfbad9 | 1112 | node = tree_search_for_insert(tree, start, &p, &parent); |
462d6fac JB |
1113 | if (!node) { |
1114 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1115 | if (!prealloc) { |
1116 | err = -ENOMEM; | |
1117 | goto out; | |
1118 | } | |
12cfbad9 | 1119 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 1120 | &p, &parent, &bits, NULL); |
c2d904e0 JM |
1121 | if (err) |
1122 | extent_io_tree_panic(tree, err); | |
c42ac0bc FDBM |
1123 | cache_state(prealloc, cached_state); |
1124 | prealloc = NULL; | |
462d6fac JB |
1125 | goto out; |
1126 | } | |
1127 | state = rb_entry(node, struct extent_state, rb_node); | |
1128 | hit_next: | |
1129 | last_start = state->start; | |
1130 | last_end = state->end; | |
1131 | ||
1132 | /* | |
1133 | * | ---- desired range ---- | | |
1134 | * | state | | |
1135 | * | |
1136 | * Just lock what we found and keep going | |
1137 | */ | |
1138 | if (state->start == start && state->end <= end) { | |
d38ed27f | 1139 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1140 | cache_state(state, cached_state); |
fefdc557 | 1141 | state = clear_state_bit(tree, state, &clear_bits, 0, NULL); |
462d6fac JB |
1142 | if (last_end == (u64)-1) |
1143 | goto out; | |
462d6fac | 1144 | start = last_end + 1; |
d1ac6e41 LB |
1145 | if (start < end && state && state->start == start && |
1146 | !need_resched()) | |
1147 | goto hit_next; | |
462d6fac JB |
1148 | goto search_again; |
1149 | } | |
1150 | ||
1151 | /* | |
1152 | * | ---- desired range ---- | | |
1153 | * | state | | |
1154 | * or | |
1155 | * | ------------- state -------------- | | |
1156 | * | |
1157 | * We need to split the extent we found, and may flip bits on | |
1158 | * second half. | |
1159 | * | |
1160 | * If the extent we found extends past our | |
1161 | * range, we just split and search again. It'll get split | |
1162 | * again the next time though. | |
1163 | * | |
1164 | * If the extent we found is inside our range, we set the | |
1165 | * desired bit on it. | |
1166 | */ | |
1167 | if (state->start < start) { | |
1168 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1169 | if (!prealloc) { |
1170 | err = -ENOMEM; | |
1171 | goto out; | |
1172 | } | |
462d6fac | 1173 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1174 | if (err) |
1175 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1176 | prealloc = NULL; |
1177 | if (err) | |
1178 | goto out; | |
1179 | if (state->end <= end) { | |
d38ed27f | 1180 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1181 | cache_state(state, cached_state); |
fefdc557 QW |
1182 | state = clear_state_bit(tree, state, &clear_bits, 0, |
1183 | NULL); | |
462d6fac JB |
1184 | if (last_end == (u64)-1) |
1185 | goto out; | |
1186 | start = last_end + 1; | |
d1ac6e41 LB |
1187 | if (start < end && state && state->start == start && |
1188 | !need_resched()) | |
1189 | goto hit_next; | |
462d6fac JB |
1190 | } |
1191 | goto search_again; | |
1192 | } | |
1193 | /* | |
1194 | * | ---- desired range ---- | | |
1195 | * | state | or | state | | |
1196 | * | |
1197 | * There's a hole, we need to insert something in it and | |
1198 | * ignore the extent we found. | |
1199 | */ | |
1200 | if (state->start > start) { | |
1201 | u64 this_end; | |
1202 | if (end < last_start) | |
1203 | this_end = end; | |
1204 | else | |
1205 | this_end = last_start - 1; | |
1206 | ||
1207 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1208 | if (!prealloc) { |
1209 | err = -ENOMEM; | |
1210 | goto out; | |
1211 | } | |
462d6fac JB |
1212 | |
1213 | /* | |
1214 | * Avoid to free 'prealloc' if it can be merged with | |
1215 | * the later extent. | |
1216 | */ | |
1217 | err = insert_state(tree, prealloc, start, this_end, | |
d38ed27f | 1218 | NULL, NULL, &bits, NULL); |
c2d904e0 JM |
1219 | if (err) |
1220 | extent_io_tree_panic(tree, err); | |
e6138876 | 1221 | cache_state(prealloc, cached_state); |
462d6fac JB |
1222 | prealloc = NULL; |
1223 | start = this_end + 1; | |
1224 | goto search_again; | |
1225 | } | |
1226 | /* | |
1227 | * | ---- desired range ---- | | |
1228 | * | state | | |
1229 | * We need to split the extent, and set the bit | |
1230 | * on the first half | |
1231 | */ | |
1232 | if (state->start <= end && state->end > end) { | |
1233 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1234 | if (!prealloc) { |
1235 | err = -ENOMEM; | |
1236 | goto out; | |
1237 | } | |
462d6fac JB |
1238 | |
1239 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1240 | if (err) |
1241 | extent_io_tree_panic(tree, err); | |
462d6fac | 1242 | |
d38ed27f | 1243 | set_state_bits(tree, prealloc, &bits, NULL); |
e6138876 | 1244 | cache_state(prealloc, cached_state); |
fefdc557 | 1245 | clear_state_bit(tree, prealloc, &clear_bits, 0, NULL); |
462d6fac JB |
1246 | prealloc = NULL; |
1247 | goto out; | |
1248 | } | |
1249 | ||
462d6fac JB |
1250 | search_again: |
1251 | if (start > end) | |
1252 | goto out; | |
1253 | spin_unlock(&tree->lock); | |
210aa277 | 1254 | cond_resched(); |
c8fd3de7 | 1255 | first_iteration = false; |
462d6fac | 1256 | goto again; |
462d6fac JB |
1257 | |
1258 | out: | |
1259 | spin_unlock(&tree->lock); | |
1260 | if (prealloc) | |
1261 | free_extent_state(prealloc); | |
1262 | ||
1263 | return err; | |
462d6fac JB |
1264 | } |
1265 | ||
d1310b2e | 1266 | /* wrappers around set/clear extent bit */ |
d38ed27f | 1267 | int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
2c53b912 | 1268 | unsigned bits, struct extent_changeset *changeset) |
d38ed27f QW |
1269 | { |
1270 | /* | |
1271 | * We don't support EXTENT_LOCKED yet, as current changeset will | |
1272 | * record any bits changed, so for EXTENT_LOCKED case, it will | |
1273 | * either fail with -EEXIST or changeset will record the whole | |
1274 | * range. | |
1275 | */ | |
1276 | BUG_ON(bits & EXTENT_LOCKED); | |
1277 | ||
2c53b912 | 1278 | return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS, |
d38ed27f QW |
1279 | changeset); |
1280 | } | |
1281 | ||
fefdc557 QW |
1282 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
1283 | unsigned bits, int wake, int delete, | |
ae0f1625 | 1284 | struct extent_state **cached) |
fefdc557 QW |
1285 | { |
1286 | return __clear_extent_bit(tree, start, end, bits, wake, delete, | |
ae0f1625 | 1287 | cached, GFP_NOFS, NULL); |
fefdc557 QW |
1288 | } |
1289 | ||
fefdc557 | 1290 | int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
f734c44a | 1291 | unsigned bits, struct extent_changeset *changeset) |
fefdc557 QW |
1292 | { |
1293 | /* | |
1294 | * Don't support EXTENT_LOCKED case, same reason as | |
1295 | * set_record_extent_bits(). | |
1296 | */ | |
1297 | BUG_ON(bits & EXTENT_LOCKED); | |
1298 | ||
f734c44a | 1299 | return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS, |
fefdc557 QW |
1300 | changeset); |
1301 | } | |
1302 | ||
d352ac68 CM |
1303 | /* |
1304 | * either insert or lock state struct between start and end use mask to tell | |
1305 | * us if waiting is desired. | |
1306 | */ | |
1edbb734 | 1307 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
ff13db41 | 1308 | struct extent_state **cached_state) |
d1310b2e CM |
1309 | { |
1310 | int err; | |
1311 | u64 failed_start; | |
9ee49a04 | 1312 | |
d1310b2e | 1313 | while (1) { |
ff13db41 | 1314 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, |
3fbe5c02 | 1315 | EXTENT_LOCKED, &failed_start, |
d38ed27f | 1316 | cached_state, GFP_NOFS, NULL); |
d0082371 | 1317 | if (err == -EEXIST) { |
d1310b2e CM |
1318 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1319 | start = failed_start; | |
d0082371 | 1320 | } else |
d1310b2e | 1321 | break; |
d1310b2e CM |
1322 | WARN_ON(start > end); |
1323 | } | |
1324 | return err; | |
1325 | } | |
d1310b2e | 1326 | |
d0082371 | 1327 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1328 | { |
1329 | int err; | |
1330 | u64 failed_start; | |
1331 | ||
3fbe5c02 | 1332 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
d38ed27f | 1333 | &failed_start, NULL, GFP_NOFS, NULL); |
6643558d YZ |
1334 | if (err == -EEXIST) { |
1335 | if (failed_start > start) | |
1336 | clear_extent_bit(tree, start, failed_start - 1, | |
ae0f1625 | 1337 | EXTENT_LOCKED, 1, 0, NULL); |
25179201 | 1338 | return 0; |
6643558d | 1339 | } |
25179201 JB |
1340 | return 1; |
1341 | } | |
25179201 | 1342 | |
bd1fa4f0 | 1343 | void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1344 | { |
09cbfeaf KS |
1345 | unsigned long index = start >> PAGE_SHIFT; |
1346 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1347 | struct page *page; |
1348 | ||
1349 | while (index <= end_index) { | |
1350 | page = find_get_page(inode->i_mapping, index); | |
1351 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1352 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1353 | put_page(page); |
4adaa611 CM |
1354 | index++; |
1355 | } | |
4adaa611 CM |
1356 | } |
1357 | ||
f6311572 | 1358 | void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1359 | { |
09cbfeaf KS |
1360 | unsigned long index = start >> PAGE_SHIFT; |
1361 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1362 | struct page *page; |
1363 | ||
1364 | while (index <= end_index) { | |
1365 | page = find_get_page(inode->i_mapping, index); | |
1366 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
4adaa611 | 1367 | __set_page_dirty_nobuffers(page); |
8d38633c | 1368 | account_page_redirty(page); |
09cbfeaf | 1369 | put_page(page); |
4adaa611 CM |
1370 | index++; |
1371 | } | |
4adaa611 CM |
1372 | } |
1373 | ||
d352ac68 CM |
1374 | /* find the first state struct with 'bits' set after 'start', and |
1375 | * return it. tree->lock must be held. NULL will returned if | |
1376 | * nothing was found after 'start' | |
1377 | */ | |
48a3b636 ES |
1378 | static struct extent_state * |
1379 | find_first_extent_bit_state(struct extent_io_tree *tree, | |
9ee49a04 | 1380 | u64 start, unsigned bits) |
d7fc640e CM |
1381 | { |
1382 | struct rb_node *node; | |
1383 | struct extent_state *state; | |
1384 | ||
1385 | /* | |
1386 | * this search will find all the extents that end after | |
1387 | * our range starts. | |
1388 | */ | |
1389 | node = tree_search(tree, start); | |
d397712b | 1390 | if (!node) |
d7fc640e | 1391 | goto out; |
d7fc640e | 1392 | |
d397712b | 1393 | while (1) { |
d7fc640e | 1394 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1395 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1396 | return state; |
d397712b | 1397 | |
d7fc640e CM |
1398 | node = rb_next(node); |
1399 | if (!node) | |
1400 | break; | |
1401 | } | |
1402 | out: | |
1403 | return NULL; | |
1404 | } | |
d7fc640e | 1405 | |
69261c4b XG |
1406 | /* |
1407 | * find the first offset in the io tree with 'bits' set. zero is | |
1408 | * returned if we find something, and *start_ret and *end_ret are | |
1409 | * set to reflect the state struct that was found. | |
1410 | * | |
477d7eaf | 1411 | * If nothing was found, 1 is returned. If found something, return 0. |
69261c4b XG |
1412 | */ |
1413 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
9ee49a04 | 1414 | u64 *start_ret, u64 *end_ret, unsigned bits, |
e6138876 | 1415 | struct extent_state **cached_state) |
69261c4b XG |
1416 | { |
1417 | struct extent_state *state; | |
1418 | int ret = 1; | |
1419 | ||
1420 | spin_lock(&tree->lock); | |
e6138876 JB |
1421 | if (cached_state && *cached_state) { |
1422 | state = *cached_state; | |
27a3507d | 1423 | if (state->end == start - 1 && extent_state_in_tree(state)) { |
9688e9a9 | 1424 | while ((state = next_state(state)) != NULL) { |
e6138876 JB |
1425 | if (state->state & bits) |
1426 | goto got_it; | |
e6138876 JB |
1427 | } |
1428 | free_extent_state(*cached_state); | |
1429 | *cached_state = NULL; | |
1430 | goto out; | |
1431 | } | |
1432 | free_extent_state(*cached_state); | |
1433 | *cached_state = NULL; | |
1434 | } | |
1435 | ||
69261c4b | 1436 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1437 | got_it: |
69261c4b | 1438 | if (state) { |
e38e2ed7 | 1439 | cache_state_if_flags(state, cached_state, 0); |
69261c4b XG |
1440 | *start_ret = state->start; |
1441 | *end_ret = state->end; | |
1442 | ret = 0; | |
1443 | } | |
e6138876 | 1444 | out: |
69261c4b XG |
1445 | spin_unlock(&tree->lock); |
1446 | return ret; | |
1447 | } | |
1448 | ||
d352ac68 CM |
1449 | /* |
1450 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1451 | * more than 'max_bytes'. start and end are used to return the range, | |
1452 | * | |
3522e903 | 1453 | * true is returned if we find something, false if nothing was in the tree |
d352ac68 | 1454 | */ |
3522e903 | 1455 | static noinline bool find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1456 | u64 *start, u64 *end, u64 max_bytes, |
1457 | struct extent_state **cached_state) | |
d1310b2e CM |
1458 | { |
1459 | struct rb_node *node; | |
1460 | struct extent_state *state; | |
1461 | u64 cur_start = *start; | |
3522e903 | 1462 | bool found = false; |
d1310b2e CM |
1463 | u64 total_bytes = 0; |
1464 | ||
cad321ad | 1465 | spin_lock(&tree->lock); |
c8b97818 | 1466 | |
d1310b2e CM |
1467 | /* |
1468 | * this search will find all the extents that end after | |
1469 | * our range starts. | |
1470 | */ | |
80ea96b1 | 1471 | node = tree_search(tree, cur_start); |
2b114d1d | 1472 | if (!node) { |
3522e903 | 1473 | *end = (u64)-1; |
d1310b2e CM |
1474 | goto out; |
1475 | } | |
1476 | ||
d397712b | 1477 | while (1) { |
d1310b2e | 1478 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1479 | if (found && (state->start != cur_start || |
1480 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1481 | goto out; |
1482 | } | |
1483 | if (!(state->state & EXTENT_DELALLOC)) { | |
1484 | if (!found) | |
1485 | *end = state->end; | |
1486 | goto out; | |
1487 | } | |
c2a128d2 | 1488 | if (!found) { |
d1310b2e | 1489 | *start = state->start; |
c2a128d2 | 1490 | *cached_state = state; |
b7ac31b7 | 1491 | refcount_inc(&state->refs); |
c2a128d2 | 1492 | } |
3522e903 | 1493 | found = true; |
d1310b2e CM |
1494 | *end = state->end; |
1495 | cur_start = state->end + 1; | |
1496 | node = rb_next(node); | |
d1310b2e | 1497 | total_bytes += state->end - state->start + 1; |
7bf811a5 | 1498 | if (total_bytes >= max_bytes) |
573aecaf | 1499 | break; |
573aecaf | 1500 | if (!node) |
d1310b2e CM |
1501 | break; |
1502 | } | |
1503 | out: | |
cad321ad | 1504 | spin_unlock(&tree->lock); |
d1310b2e CM |
1505 | return found; |
1506 | } | |
1507 | ||
da2c7009 LB |
1508 | static int __process_pages_contig(struct address_space *mapping, |
1509 | struct page *locked_page, | |
1510 | pgoff_t start_index, pgoff_t end_index, | |
1511 | unsigned long page_ops, pgoff_t *index_ret); | |
1512 | ||
143bede5 JM |
1513 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1514 | struct page *locked_page, | |
1515 | u64 start, u64 end) | |
c8b97818 | 1516 | { |
09cbfeaf KS |
1517 | unsigned long index = start >> PAGE_SHIFT; |
1518 | unsigned long end_index = end >> PAGE_SHIFT; | |
c8b97818 | 1519 | |
76c0021d | 1520 | ASSERT(locked_page); |
c8b97818 | 1521 | if (index == locked_page->index && end_index == index) |
143bede5 | 1522 | return; |
c8b97818 | 1523 | |
76c0021d LB |
1524 | __process_pages_contig(inode->i_mapping, locked_page, index, end_index, |
1525 | PAGE_UNLOCK, NULL); | |
c8b97818 CM |
1526 | } |
1527 | ||
1528 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1529 | struct page *locked_page, | |
1530 | u64 delalloc_start, | |
1531 | u64 delalloc_end) | |
1532 | { | |
09cbfeaf | 1533 | unsigned long index = delalloc_start >> PAGE_SHIFT; |
76c0021d | 1534 | unsigned long index_ret = index; |
09cbfeaf | 1535 | unsigned long end_index = delalloc_end >> PAGE_SHIFT; |
c8b97818 | 1536 | int ret; |
c8b97818 | 1537 | |
76c0021d | 1538 | ASSERT(locked_page); |
c8b97818 CM |
1539 | if (index == locked_page->index && index == end_index) |
1540 | return 0; | |
1541 | ||
76c0021d LB |
1542 | ret = __process_pages_contig(inode->i_mapping, locked_page, index, |
1543 | end_index, PAGE_LOCK, &index_ret); | |
1544 | if (ret == -EAGAIN) | |
1545 | __unlock_for_delalloc(inode, locked_page, delalloc_start, | |
1546 | (u64)index_ret << PAGE_SHIFT); | |
c8b97818 CM |
1547 | return ret; |
1548 | } | |
1549 | ||
1550 | /* | |
3522e903 LF |
1551 | * Find and lock a contiguous range of bytes in the file marked as delalloc, no |
1552 | * more than @max_bytes. @Start and @end are used to return the range, | |
c8b97818 | 1553 | * |
3522e903 LF |
1554 | * Return: true if we find something |
1555 | * false if nothing was in the tree | |
c8b97818 | 1556 | */ |
ce9f967f | 1557 | EXPORT_FOR_TESTS |
3522e903 | 1558 | noinline_for_stack bool find_lock_delalloc_range(struct inode *inode, |
294e30fe JB |
1559 | struct extent_io_tree *tree, |
1560 | struct page *locked_page, u64 *start, | |
917aacec | 1561 | u64 *end) |
c8b97818 | 1562 | { |
917aacec | 1563 | u64 max_bytes = BTRFS_MAX_EXTENT_SIZE; |
c8b97818 CM |
1564 | u64 delalloc_start; |
1565 | u64 delalloc_end; | |
3522e903 | 1566 | bool 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); |
3522e903 | 1581 | return false; |
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); | |
9bfd61d9 | 1601 | ASSERT(!ret || ret == -EAGAIN); |
c8b97818 CM |
1602 | if (ret == -EAGAIN) { |
1603 | /* some of the pages are gone, lets avoid looping by | |
1604 | * shortening the size of the delalloc range we're searching | |
1605 | */ | |
9655d298 | 1606 | free_extent_state(cached_state); |
7d788742 | 1607 | cached_state = NULL; |
c8b97818 | 1608 | if (!loops) { |
09cbfeaf | 1609 | max_bytes = PAGE_SIZE; |
c8b97818 CM |
1610 | loops = 1; |
1611 | goto again; | |
1612 | } else { | |
3522e903 | 1613 | found = false; |
c8b97818 CM |
1614 | goto out_failed; |
1615 | } | |
1616 | } | |
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 | ||
da2c7009 LB |
1639 | static int __process_pages_contig(struct address_space *mapping, |
1640 | struct page *locked_page, | |
1641 | pgoff_t start_index, pgoff_t end_index, | |
1642 | unsigned long page_ops, pgoff_t *index_ret) | |
c8b97818 | 1643 | { |
873695b3 | 1644 | unsigned long nr_pages = end_index - start_index + 1; |
da2c7009 | 1645 | unsigned long pages_locked = 0; |
873695b3 | 1646 | pgoff_t index = start_index; |
c8b97818 | 1647 | struct page *pages[16]; |
873695b3 | 1648 | unsigned ret; |
da2c7009 | 1649 | int err = 0; |
c8b97818 | 1650 | int i; |
771ed689 | 1651 | |
da2c7009 LB |
1652 | if (page_ops & PAGE_LOCK) { |
1653 | ASSERT(page_ops == PAGE_LOCK); | |
1654 | ASSERT(index_ret && *index_ret == start_index); | |
1655 | } | |
1656 | ||
704de49d | 1657 | if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0) |
873695b3 | 1658 | mapping_set_error(mapping, -EIO); |
704de49d | 1659 | |
d397712b | 1660 | while (nr_pages > 0) { |
873695b3 | 1661 | ret = find_get_pages_contig(mapping, index, |
5b050f04 CM |
1662 | min_t(unsigned long, |
1663 | nr_pages, ARRAY_SIZE(pages)), pages); | |
da2c7009 LB |
1664 | if (ret == 0) { |
1665 | /* | |
1666 | * Only if we're going to lock these pages, | |
1667 | * can we find nothing at @index. | |
1668 | */ | |
1669 | ASSERT(page_ops & PAGE_LOCK); | |
49d4a334 LB |
1670 | err = -EAGAIN; |
1671 | goto out; | |
da2c7009 | 1672 | } |
8b62b72b | 1673 | |
da2c7009 | 1674 | for (i = 0; i < ret; i++) { |
c2790a2e | 1675 | if (page_ops & PAGE_SET_PRIVATE2) |
8b62b72b CM |
1676 | SetPagePrivate2(pages[i]); |
1677 | ||
c8b97818 | 1678 | if (pages[i] == locked_page) { |
09cbfeaf | 1679 | put_page(pages[i]); |
da2c7009 | 1680 | pages_locked++; |
c8b97818 CM |
1681 | continue; |
1682 | } | |
c2790a2e | 1683 | if (page_ops & PAGE_CLEAR_DIRTY) |
c8b97818 | 1684 | clear_page_dirty_for_io(pages[i]); |
c2790a2e | 1685 | if (page_ops & PAGE_SET_WRITEBACK) |
c8b97818 | 1686 | set_page_writeback(pages[i]); |
704de49d FM |
1687 | if (page_ops & PAGE_SET_ERROR) |
1688 | SetPageError(pages[i]); | |
c2790a2e | 1689 | if (page_ops & PAGE_END_WRITEBACK) |
c8b97818 | 1690 | end_page_writeback(pages[i]); |
c2790a2e | 1691 | if (page_ops & PAGE_UNLOCK) |
771ed689 | 1692 | unlock_page(pages[i]); |
da2c7009 LB |
1693 | if (page_ops & PAGE_LOCK) { |
1694 | lock_page(pages[i]); | |
1695 | if (!PageDirty(pages[i]) || | |
1696 | pages[i]->mapping != mapping) { | |
1697 | unlock_page(pages[i]); | |
1698 | put_page(pages[i]); | |
1699 | err = -EAGAIN; | |
1700 | goto out; | |
1701 | } | |
1702 | } | |
09cbfeaf | 1703 | put_page(pages[i]); |
da2c7009 | 1704 | pages_locked++; |
c8b97818 CM |
1705 | } |
1706 | nr_pages -= ret; | |
1707 | index += ret; | |
1708 | cond_resched(); | |
1709 | } | |
da2c7009 LB |
1710 | out: |
1711 | if (err && index_ret) | |
1712 | *index_ret = start_index + pages_locked - 1; | |
1713 | return err; | |
c8b97818 | 1714 | } |
c8b97818 | 1715 | |
873695b3 LB |
1716 | void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end, |
1717 | u64 delalloc_end, struct page *locked_page, | |
1718 | unsigned clear_bits, | |
1719 | unsigned long page_ops) | |
1720 | { | |
1721 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0, | |
ae0f1625 | 1722 | NULL); |
873695b3 LB |
1723 | |
1724 | __process_pages_contig(inode->i_mapping, locked_page, | |
1725 | start >> PAGE_SHIFT, end >> PAGE_SHIFT, | |
da2c7009 | 1726 | page_ops, NULL); |
873695b3 LB |
1727 | } |
1728 | ||
d352ac68 CM |
1729 | /* |
1730 | * count the number of bytes in the tree that have a given bit(s) | |
1731 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1732 | * cached. The total number found is returned. | |
1733 | */ | |
d1310b2e CM |
1734 | u64 count_range_bits(struct extent_io_tree *tree, |
1735 | u64 *start, u64 search_end, u64 max_bytes, | |
9ee49a04 | 1736 | unsigned bits, int contig) |
d1310b2e CM |
1737 | { |
1738 | struct rb_node *node; | |
1739 | struct extent_state *state; | |
1740 | u64 cur_start = *start; | |
1741 | u64 total_bytes = 0; | |
ec29ed5b | 1742 | u64 last = 0; |
d1310b2e CM |
1743 | int found = 0; |
1744 | ||
fae7f21c | 1745 | if (WARN_ON(search_end <= cur_start)) |
d1310b2e | 1746 | return 0; |
d1310b2e | 1747 | |
cad321ad | 1748 | spin_lock(&tree->lock); |
d1310b2e CM |
1749 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1750 | total_bytes = tree->dirty_bytes; | |
1751 | goto out; | |
1752 | } | |
1753 | /* | |
1754 | * this search will find all the extents that end after | |
1755 | * our range starts. | |
1756 | */ | |
80ea96b1 | 1757 | node = tree_search(tree, cur_start); |
d397712b | 1758 | if (!node) |
d1310b2e | 1759 | goto out; |
d1310b2e | 1760 | |
d397712b | 1761 | while (1) { |
d1310b2e CM |
1762 | state = rb_entry(node, struct extent_state, rb_node); |
1763 | if (state->start > search_end) | |
1764 | break; | |
ec29ed5b CM |
1765 | if (contig && found && state->start > last + 1) |
1766 | break; | |
1767 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1768 | total_bytes += min(search_end, state->end) + 1 - |
1769 | max(cur_start, state->start); | |
1770 | if (total_bytes >= max_bytes) | |
1771 | break; | |
1772 | if (!found) { | |
af60bed2 | 1773 | *start = max(cur_start, state->start); |
d1310b2e CM |
1774 | found = 1; |
1775 | } | |
ec29ed5b CM |
1776 | last = state->end; |
1777 | } else if (contig && found) { | |
1778 | break; | |
d1310b2e CM |
1779 | } |
1780 | node = rb_next(node); | |
1781 | if (!node) | |
1782 | break; | |
1783 | } | |
1784 | out: | |
cad321ad | 1785 | spin_unlock(&tree->lock); |
d1310b2e CM |
1786 | return total_bytes; |
1787 | } | |
b2950863 | 1788 | |
d352ac68 CM |
1789 | /* |
1790 | * set the private field for a given byte offset in the tree. If there isn't | |
1791 | * an extent_state there already, this does nothing. | |
1792 | */ | |
f827ba9a | 1793 | static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start, |
47dc196a | 1794 | struct io_failure_record *failrec) |
d1310b2e CM |
1795 | { |
1796 | struct rb_node *node; | |
1797 | struct extent_state *state; | |
1798 | int ret = 0; | |
1799 | ||
cad321ad | 1800 | spin_lock(&tree->lock); |
d1310b2e CM |
1801 | /* |
1802 | * this search will find all the extents that end after | |
1803 | * our range starts. | |
1804 | */ | |
80ea96b1 | 1805 | node = tree_search(tree, start); |
2b114d1d | 1806 | if (!node) { |
d1310b2e CM |
1807 | ret = -ENOENT; |
1808 | goto out; | |
1809 | } | |
1810 | state = rb_entry(node, struct extent_state, rb_node); | |
1811 | if (state->start != start) { | |
1812 | ret = -ENOENT; | |
1813 | goto out; | |
1814 | } | |
47dc196a | 1815 | state->failrec = failrec; |
d1310b2e | 1816 | out: |
cad321ad | 1817 | spin_unlock(&tree->lock); |
d1310b2e CM |
1818 | return ret; |
1819 | } | |
1820 | ||
f827ba9a | 1821 | static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start, |
47dc196a | 1822 | struct io_failure_record **failrec) |
d1310b2e CM |
1823 | { |
1824 | struct rb_node *node; | |
1825 | struct extent_state *state; | |
1826 | int ret = 0; | |
1827 | ||
cad321ad | 1828 | spin_lock(&tree->lock); |
d1310b2e CM |
1829 | /* |
1830 | * this search will find all the extents that end after | |
1831 | * our range starts. | |
1832 | */ | |
80ea96b1 | 1833 | node = tree_search(tree, start); |
2b114d1d | 1834 | if (!node) { |
d1310b2e CM |
1835 | ret = -ENOENT; |
1836 | goto out; | |
1837 | } | |
1838 | state = rb_entry(node, struct extent_state, rb_node); | |
1839 | if (state->start != start) { | |
1840 | ret = -ENOENT; | |
1841 | goto out; | |
1842 | } | |
47dc196a | 1843 | *failrec = state->failrec; |
d1310b2e | 1844 | out: |
cad321ad | 1845 | spin_unlock(&tree->lock); |
d1310b2e CM |
1846 | return ret; |
1847 | } | |
1848 | ||
1849 | /* | |
1850 | * searches a range in the state tree for a given mask. | |
70dec807 | 1851 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1852 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1853 | * range is found set. | |
1854 | */ | |
1855 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1856 | unsigned bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1857 | { |
1858 | struct extent_state *state = NULL; | |
1859 | struct rb_node *node; | |
1860 | int bitset = 0; | |
d1310b2e | 1861 | |
cad321ad | 1862 | spin_lock(&tree->lock); |
27a3507d | 1863 | if (cached && extent_state_in_tree(cached) && cached->start <= start && |
df98b6e2 | 1864 | cached->end > start) |
9655d298 CM |
1865 | node = &cached->rb_node; |
1866 | else | |
1867 | node = tree_search(tree, start); | |
d1310b2e CM |
1868 | while (node && start <= end) { |
1869 | state = rb_entry(node, struct extent_state, rb_node); | |
1870 | ||
1871 | if (filled && state->start > start) { | |
1872 | bitset = 0; | |
1873 | break; | |
1874 | } | |
1875 | ||
1876 | if (state->start > end) | |
1877 | break; | |
1878 | ||
1879 | if (state->state & bits) { | |
1880 | bitset = 1; | |
1881 | if (!filled) | |
1882 | break; | |
1883 | } else if (filled) { | |
1884 | bitset = 0; | |
1885 | break; | |
1886 | } | |
46562cec CM |
1887 | |
1888 | if (state->end == (u64)-1) | |
1889 | break; | |
1890 | ||
d1310b2e CM |
1891 | start = state->end + 1; |
1892 | if (start > end) | |
1893 | break; | |
1894 | node = rb_next(node); | |
1895 | if (!node) { | |
1896 | if (filled) | |
1897 | bitset = 0; | |
1898 | break; | |
1899 | } | |
1900 | } | |
cad321ad | 1901 | spin_unlock(&tree->lock); |
d1310b2e CM |
1902 | return bitset; |
1903 | } | |
d1310b2e CM |
1904 | |
1905 | /* | |
1906 | * helper function to set a given page up to date if all the | |
1907 | * extents in the tree for that page are up to date | |
1908 | */ | |
143bede5 | 1909 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e | 1910 | { |
4eee4fa4 | 1911 | u64 start = page_offset(page); |
09cbfeaf | 1912 | u64 end = start + PAGE_SIZE - 1; |
9655d298 | 1913 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 1914 | SetPageUptodate(page); |
d1310b2e CM |
1915 | } |
1916 | ||
7870d082 JB |
1917 | int free_io_failure(struct extent_io_tree *failure_tree, |
1918 | struct extent_io_tree *io_tree, | |
1919 | struct io_failure_record *rec) | |
4a54c8c1 JS |
1920 | { |
1921 | int ret; | |
1922 | int err = 0; | |
4a54c8c1 | 1923 | |
47dc196a | 1924 | set_state_failrec(failure_tree, rec->start, NULL); |
4a54c8c1 JS |
1925 | ret = clear_extent_bits(failure_tree, rec->start, |
1926 | rec->start + rec->len - 1, | |
91166212 | 1927 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 JS |
1928 | if (ret) |
1929 | err = ret; | |
1930 | ||
7870d082 | 1931 | ret = clear_extent_bits(io_tree, rec->start, |
53b381b3 | 1932 | rec->start + rec->len - 1, |
91166212 | 1933 | EXTENT_DAMAGED); |
53b381b3 DW |
1934 | if (ret && !err) |
1935 | err = ret; | |
4a54c8c1 JS |
1936 | |
1937 | kfree(rec); | |
1938 | return err; | |
1939 | } | |
1940 | ||
4a54c8c1 JS |
1941 | /* |
1942 | * this bypasses the standard btrfs submit functions deliberately, as | |
1943 | * the standard behavior is to write all copies in a raid setup. here we only | |
1944 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
1945 | * submit_bio directly. | |
3ec706c8 | 1946 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
1947 | * actually prevents the read that triggered the error from finishing. |
1948 | * currently, there can be no more than two copies of every data bit. thus, | |
1949 | * exactly one rewrite is required. | |
1950 | */ | |
6ec656bc JB |
1951 | int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start, |
1952 | u64 length, u64 logical, struct page *page, | |
1953 | unsigned int pg_offset, int mirror_num) | |
4a54c8c1 JS |
1954 | { |
1955 | struct bio *bio; | |
1956 | struct btrfs_device *dev; | |
4a54c8c1 JS |
1957 | u64 map_length = 0; |
1958 | u64 sector; | |
1959 | struct btrfs_bio *bbio = NULL; | |
1960 | int ret; | |
1961 | ||
1751e8a6 | 1962 | ASSERT(!(fs_info->sb->s_flags & SB_RDONLY)); |
4a54c8c1 JS |
1963 | BUG_ON(!mirror_num); |
1964 | ||
c5e4c3d7 | 1965 | bio = btrfs_io_bio_alloc(1); |
4f024f37 | 1966 | bio->bi_iter.bi_size = 0; |
4a54c8c1 JS |
1967 | map_length = length; |
1968 | ||
b5de8d0d FM |
1969 | /* |
1970 | * Avoid races with device replace and make sure our bbio has devices | |
1971 | * associated to its stripes that don't go away while we are doing the | |
1972 | * read repair operation. | |
1973 | */ | |
1974 | btrfs_bio_counter_inc_blocked(fs_info); | |
e4ff5fb5 | 1975 | if (btrfs_is_parity_mirror(fs_info, logical, length)) { |
c725328c LB |
1976 | /* |
1977 | * Note that we don't use BTRFS_MAP_WRITE because it's supposed | |
1978 | * to update all raid stripes, but here we just want to correct | |
1979 | * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad | |
1980 | * stripe's dev and sector. | |
1981 | */ | |
1982 | ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical, | |
1983 | &map_length, &bbio, 0); | |
1984 | if (ret) { | |
1985 | btrfs_bio_counter_dec(fs_info); | |
1986 | bio_put(bio); | |
1987 | return -EIO; | |
1988 | } | |
1989 | ASSERT(bbio->mirror_num == 1); | |
1990 | } else { | |
1991 | ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, | |
1992 | &map_length, &bbio, mirror_num); | |
1993 | if (ret) { | |
1994 | btrfs_bio_counter_dec(fs_info); | |
1995 | bio_put(bio); | |
1996 | return -EIO; | |
1997 | } | |
1998 | BUG_ON(mirror_num != bbio->mirror_num); | |
4a54c8c1 | 1999 | } |
c725328c LB |
2000 | |
2001 | sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9; | |
4f024f37 | 2002 | bio->bi_iter.bi_sector = sector; |
c725328c | 2003 | dev = bbio->stripes[bbio->mirror_num - 1].dev; |
6e9606d2 | 2004 | btrfs_put_bbio(bbio); |
ebbede42 AJ |
2005 | if (!dev || !dev->bdev || |
2006 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) { | |
b5de8d0d | 2007 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2008 | bio_put(bio); |
2009 | return -EIO; | |
2010 | } | |
74d46992 | 2011 | bio_set_dev(bio, dev->bdev); |
70fd7614 | 2012 | bio->bi_opf = REQ_OP_WRITE | REQ_SYNC; |
ffdd2018 | 2013 | bio_add_page(bio, page, length, pg_offset); |
4a54c8c1 | 2014 | |
4e49ea4a | 2015 | if (btrfsic_submit_bio_wait(bio)) { |
4a54c8c1 | 2016 | /* try to remap that extent elsewhere? */ |
b5de8d0d | 2017 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 | 2018 | bio_put(bio); |
442a4f63 | 2019 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
2020 | return -EIO; |
2021 | } | |
2022 | ||
b14af3b4 DS |
2023 | btrfs_info_rl_in_rcu(fs_info, |
2024 | "read error corrected: ino %llu off %llu (dev %s sector %llu)", | |
6ec656bc | 2025 | ino, start, |
1203b681 | 2026 | rcu_str_deref(dev->name), sector); |
b5de8d0d | 2027 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2028 | bio_put(bio); |
2029 | return 0; | |
2030 | } | |
2031 | ||
2ff7e61e JM |
2032 | int repair_eb_io_failure(struct btrfs_fs_info *fs_info, |
2033 | struct extent_buffer *eb, int mirror_num) | |
ea466794 | 2034 | { |
ea466794 | 2035 | u64 start = eb->start; |
cc5e31a4 | 2036 | int i, num_pages = num_extent_pages(eb); |
d95603b2 | 2037 | int ret = 0; |
ea466794 | 2038 | |
bc98a42c | 2039 | if (sb_rdonly(fs_info->sb)) |
908960c6 ID |
2040 | return -EROFS; |
2041 | ||
ea466794 | 2042 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 2043 | struct page *p = eb->pages[i]; |
1203b681 | 2044 | |
6ec656bc | 2045 | ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p, |
1203b681 | 2046 | start - page_offset(p), mirror_num); |
ea466794 JB |
2047 | if (ret) |
2048 | break; | |
09cbfeaf | 2049 | start += PAGE_SIZE; |
ea466794 JB |
2050 | } |
2051 | ||
2052 | return ret; | |
2053 | } | |
2054 | ||
4a54c8c1 JS |
2055 | /* |
2056 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2057 | * to see if we need to process or clean up an io_failure_record | |
2058 | */ | |
7870d082 JB |
2059 | int clean_io_failure(struct btrfs_fs_info *fs_info, |
2060 | struct extent_io_tree *failure_tree, | |
2061 | struct extent_io_tree *io_tree, u64 start, | |
2062 | struct page *page, u64 ino, unsigned int pg_offset) | |
4a54c8c1 JS |
2063 | { |
2064 | u64 private; | |
4a54c8c1 | 2065 | struct io_failure_record *failrec; |
4a54c8c1 JS |
2066 | struct extent_state *state; |
2067 | int num_copies; | |
4a54c8c1 | 2068 | int ret; |
4a54c8c1 JS |
2069 | |
2070 | private = 0; | |
7870d082 JB |
2071 | ret = count_range_bits(failure_tree, &private, (u64)-1, 1, |
2072 | EXTENT_DIRTY, 0); | |
4a54c8c1 JS |
2073 | if (!ret) |
2074 | return 0; | |
2075 | ||
7870d082 | 2076 | ret = get_state_failrec(failure_tree, start, &failrec); |
4a54c8c1 JS |
2077 | if (ret) |
2078 | return 0; | |
2079 | ||
4a54c8c1 JS |
2080 | BUG_ON(!failrec->this_mirror); |
2081 | ||
2082 | if (failrec->in_validation) { | |
2083 | /* there was no real error, just free the record */ | |
ab8d0fc4 JM |
2084 | btrfs_debug(fs_info, |
2085 | "clean_io_failure: freeing dummy error at %llu", | |
2086 | failrec->start); | |
4a54c8c1 JS |
2087 | goto out; |
2088 | } | |
bc98a42c | 2089 | if (sb_rdonly(fs_info->sb)) |
908960c6 | 2090 | goto out; |
4a54c8c1 | 2091 | |
7870d082 JB |
2092 | spin_lock(&io_tree->lock); |
2093 | state = find_first_extent_bit_state(io_tree, | |
4a54c8c1 JS |
2094 | failrec->start, |
2095 | EXTENT_LOCKED); | |
7870d082 | 2096 | spin_unlock(&io_tree->lock); |
4a54c8c1 | 2097 | |
883d0de4 MX |
2098 | if (state && state->start <= failrec->start && |
2099 | state->end >= failrec->start + failrec->len - 1) { | |
3ec706c8 SB |
2100 | num_copies = btrfs_num_copies(fs_info, failrec->logical, |
2101 | failrec->len); | |
4a54c8c1 | 2102 | if (num_copies > 1) { |
7870d082 JB |
2103 | repair_io_failure(fs_info, ino, start, failrec->len, |
2104 | failrec->logical, page, pg_offset, | |
2105 | failrec->failed_mirror); | |
4a54c8c1 JS |
2106 | } |
2107 | } | |
2108 | ||
2109 | out: | |
7870d082 | 2110 | free_io_failure(failure_tree, io_tree, failrec); |
4a54c8c1 | 2111 | |
454ff3de | 2112 | return 0; |
4a54c8c1 JS |
2113 | } |
2114 | ||
f612496b MX |
2115 | /* |
2116 | * Can be called when | |
2117 | * - hold extent lock | |
2118 | * - under ordered extent | |
2119 | * - the inode is freeing | |
2120 | */ | |
7ab7956e | 2121 | void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end) |
f612496b | 2122 | { |
7ab7956e | 2123 | struct extent_io_tree *failure_tree = &inode->io_failure_tree; |
f612496b MX |
2124 | struct io_failure_record *failrec; |
2125 | struct extent_state *state, *next; | |
2126 | ||
2127 | if (RB_EMPTY_ROOT(&failure_tree->state)) | |
2128 | return; | |
2129 | ||
2130 | spin_lock(&failure_tree->lock); | |
2131 | state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY); | |
2132 | while (state) { | |
2133 | if (state->start > end) | |
2134 | break; | |
2135 | ||
2136 | ASSERT(state->end <= end); | |
2137 | ||
2138 | next = next_state(state); | |
2139 | ||
47dc196a | 2140 | failrec = state->failrec; |
f612496b MX |
2141 | free_extent_state(state); |
2142 | kfree(failrec); | |
2143 | ||
2144 | state = next; | |
2145 | } | |
2146 | spin_unlock(&failure_tree->lock); | |
2147 | } | |
2148 | ||
2fe6303e | 2149 | int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end, |
47dc196a | 2150 | struct io_failure_record **failrec_ret) |
4a54c8c1 | 2151 | { |
ab8d0fc4 | 2152 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e | 2153 | struct io_failure_record *failrec; |
4a54c8c1 | 2154 | struct extent_map *em; |
4a54c8c1 JS |
2155 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2156 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2157 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
4a54c8c1 | 2158 | int ret; |
4a54c8c1 JS |
2159 | u64 logical; |
2160 | ||
47dc196a | 2161 | ret = get_state_failrec(failure_tree, start, &failrec); |
4a54c8c1 JS |
2162 | if (ret) { |
2163 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2164 | if (!failrec) | |
2165 | return -ENOMEM; | |
2fe6303e | 2166 | |
4a54c8c1 JS |
2167 | failrec->start = start; |
2168 | failrec->len = end - start + 1; | |
2169 | failrec->this_mirror = 0; | |
2170 | failrec->bio_flags = 0; | |
2171 | failrec->in_validation = 0; | |
2172 | ||
2173 | read_lock(&em_tree->lock); | |
2174 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2175 | if (!em) { | |
2176 | read_unlock(&em_tree->lock); | |
2177 | kfree(failrec); | |
2178 | return -EIO; | |
2179 | } | |
2180 | ||
68ba990f | 2181 | if (em->start > start || em->start + em->len <= start) { |
4a54c8c1 JS |
2182 | free_extent_map(em); |
2183 | em = NULL; | |
2184 | } | |
2185 | read_unlock(&em_tree->lock); | |
7a2d6a64 | 2186 | if (!em) { |
4a54c8c1 JS |
2187 | kfree(failrec); |
2188 | return -EIO; | |
2189 | } | |
2fe6303e | 2190 | |
4a54c8c1 JS |
2191 | logical = start - em->start; |
2192 | logical = em->block_start + logical; | |
2193 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2194 | logical = em->block_start; | |
2195 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2196 | extent_set_compress_type(&failrec->bio_flags, | |
2197 | em->compress_type); | |
2198 | } | |
2fe6303e | 2199 | |
ab8d0fc4 JM |
2200 | btrfs_debug(fs_info, |
2201 | "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu", | |
2202 | logical, start, failrec->len); | |
2fe6303e | 2203 | |
4a54c8c1 JS |
2204 | failrec->logical = logical; |
2205 | free_extent_map(em); | |
2206 | ||
2207 | /* set the bits in the private failure tree */ | |
2208 | ret = set_extent_bits(failure_tree, start, end, | |
ceeb0ae7 | 2209 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 | 2210 | if (ret >= 0) |
47dc196a | 2211 | ret = set_state_failrec(failure_tree, start, failrec); |
4a54c8c1 JS |
2212 | /* set the bits in the inode's tree */ |
2213 | if (ret >= 0) | |
ceeb0ae7 | 2214 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED); |
4a54c8c1 JS |
2215 | if (ret < 0) { |
2216 | kfree(failrec); | |
2217 | return ret; | |
2218 | } | |
2219 | } else { | |
ab8d0fc4 JM |
2220 | btrfs_debug(fs_info, |
2221 | "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d", | |
2222 | failrec->logical, failrec->start, failrec->len, | |
2223 | failrec->in_validation); | |
4a54c8c1 JS |
2224 | /* |
2225 | * when data can be on disk more than twice, add to failrec here | |
2226 | * (e.g. with a list for failed_mirror) to make | |
2227 | * clean_io_failure() clean all those errors at once. | |
2228 | */ | |
2229 | } | |
2fe6303e MX |
2230 | |
2231 | *failrec_ret = failrec; | |
2232 | ||
2233 | return 0; | |
2234 | } | |
2235 | ||
a0b60d72 | 2236 | bool btrfs_check_repairable(struct inode *inode, unsigned failed_bio_pages, |
2fe6303e MX |
2237 | struct io_failure_record *failrec, int failed_mirror) |
2238 | { | |
ab8d0fc4 | 2239 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e MX |
2240 | int num_copies; |
2241 | ||
ab8d0fc4 | 2242 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
4a54c8c1 JS |
2243 | if (num_copies == 1) { |
2244 | /* | |
2245 | * we only have a single copy of the data, so don't bother with | |
2246 | * all the retry and error correction code that follows. no | |
2247 | * matter what the error is, it is very likely to persist. | |
2248 | */ | |
ab8d0fc4 JM |
2249 | btrfs_debug(fs_info, |
2250 | "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
2251 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2252 | return false; |
4a54c8c1 JS |
2253 | } |
2254 | ||
4a54c8c1 JS |
2255 | /* |
2256 | * there are two premises: | |
2257 | * a) deliver good data to the caller | |
2258 | * b) correct the bad sectors on disk | |
2259 | */ | |
a0b60d72 | 2260 | if (failed_bio_pages > 1) { |
4a54c8c1 JS |
2261 | /* |
2262 | * to fulfill b), we need to know the exact failing sectors, as | |
2263 | * we don't want to rewrite any more than the failed ones. thus, | |
2264 | * we need separate read requests for the failed bio | |
2265 | * | |
2266 | * if the following BUG_ON triggers, our validation request got | |
2267 | * merged. we need separate requests for our algorithm to work. | |
2268 | */ | |
2269 | BUG_ON(failrec->in_validation); | |
2270 | failrec->in_validation = 1; | |
2271 | failrec->this_mirror = failed_mirror; | |
4a54c8c1 JS |
2272 | } else { |
2273 | /* | |
2274 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2275 | * of the failed sector and if we succeed, we have setup | |
2276 | * everything for repair_io_failure to do the rest for us. | |
2277 | */ | |
2278 | if (failrec->in_validation) { | |
2279 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2280 | failrec->in_validation = 0; | |
2281 | failrec->this_mirror = 0; | |
2282 | } | |
2283 | failrec->failed_mirror = failed_mirror; | |
2284 | failrec->this_mirror++; | |
2285 | if (failrec->this_mirror == failed_mirror) | |
2286 | failrec->this_mirror++; | |
4a54c8c1 JS |
2287 | } |
2288 | ||
facc8a22 | 2289 | if (failrec->this_mirror > num_copies) { |
ab8d0fc4 JM |
2290 | btrfs_debug(fs_info, |
2291 | "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
2292 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2293 | return false; |
4a54c8c1 JS |
2294 | } |
2295 | ||
c3cfb656 | 2296 | return true; |
2fe6303e MX |
2297 | } |
2298 | ||
2299 | ||
2300 | struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio, | |
2301 | struct io_failure_record *failrec, | |
2302 | struct page *page, int pg_offset, int icsum, | |
8b110e39 | 2303 | bio_end_io_t *endio_func, void *data) |
2fe6303e | 2304 | { |
0b246afa | 2305 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e MX |
2306 | struct bio *bio; |
2307 | struct btrfs_io_bio *btrfs_failed_bio; | |
2308 | struct btrfs_io_bio *btrfs_bio; | |
2309 | ||
c5e4c3d7 | 2310 | bio = btrfs_io_bio_alloc(1); |
2fe6303e | 2311 | bio->bi_end_io = endio_func; |
4f024f37 | 2312 | bio->bi_iter.bi_sector = failrec->logical >> 9; |
74d46992 | 2313 | bio_set_dev(bio, fs_info->fs_devices->latest_bdev); |
4f024f37 | 2314 | bio->bi_iter.bi_size = 0; |
8b110e39 | 2315 | bio->bi_private = data; |
4a54c8c1 | 2316 | |
facc8a22 MX |
2317 | btrfs_failed_bio = btrfs_io_bio(failed_bio); |
2318 | if (btrfs_failed_bio->csum) { | |
facc8a22 MX |
2319 | u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); |
2320 | ||
2321 | btrfs_bio = btrfs_io_bio(bio); | |
2322 | btrfs_bio->csum = btrfs_bio->csum_inline; | |
2fe6303e MX |
2323 | icsum *= csum_size; |
2324 | memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum, | |
facc8a22 MX |
2325 | csum_size); |
2326 | } | |
2327 | ||
2fe6303e MX |
2328 | bio_add_page(bio, page, failrec->len, pg_offset); |
2329 | ||
2330 | return bio; | |
2331 | } | |
2332 | ||
2333 | /* | |
78e62c02 NB |
2334 | * This is a generic handler for readpage errors. If other copies exist, read |
2335 | * those and write back good data to the failed position. Does not investigate | |
2336 | * in remapping the failed extent elsewhere, hoping the device will be smart | |
2337 | * enough to do this as needed | |
2fe6303e | 2338 | */ |
2fe6303e MX |
2339 | static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset, |
2340 | struct page *page, u64 start, u64 end, | |
2341 | int failed_mirror) | |
2342 | { | |
2343 | struct io_failure_record *failrec; | |
2344 | struct inode *inode = page->mapping->host; | |
2345 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
7870d082 | 2346 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2fe6303e | 2347 | struct bio *bio; |
70fd7614 | 2348 | int read_mode = 0; |
4e4cbee9 | 2349 | blk_status_t status; |
2fe6303e | 2350 | int ret; |
a0b60d72 | 2351 | unsigned failed_bio_pages = bio_pages_all(failed_bio); |
2fe6303e | 2352 | |
1f7ad75b | 2353 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
2fe6303e MX |
2354 | |
2355 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
2356 | if (ret) | |
2357 | return ret; | |
2358 | ||
a0b60d72 | 2359 | if (!btrfs_check_repairable(inode, failed_bio_pages, failrec, |
c3cfb656 | 2360 | failed_mirror)) { |
7870d082 | 2361 | free_io_failure(failure_tree, tree, failrec); |
2fe6303e MX |
2362 | return -EIO; |
2363 | } | |
2364 | ||
a0b60d72 | 2365 | if (failed_bio_pages > 1) |
70fd7614 | 2366 | read_mode |= REQ_FAILFAST_DEV; |
2fe6303e MX |
2367 | |
2368 | phy_offset >>= inode->i_sb->s_blocksize_bits; | |
2369 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2370 | start - page_offset(page), | |
8b110e39 MX |
2371 | (int)phy_offset, failed_bio->bi_end_io, |
2372 | NULL); | |
ebcc3263 | 2373 | bio->bi_opf = REQ_OP_READ | read_mode; |
4a54c8c1 | 2374 | |
ab8d0fc4 JM |
2375 | btrfs_debug(btrfs_sb(inode->i_sb), |
2376 | "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d", | |
2377 | read_mode, failrec->this_mirror, failrec->in_validation); | |
4a54c8c1 | 2378 | |
8c27cb35 | 2379 | status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror, |
013bd4c3 | 2380 | failrec->bio_flags, 0); |
4e4cbee9 | 2381 | if (status) { |
7870d082 | 2382 | free_io_failure(failure_tree, tree, failrec); |
6c387ab2 | 2383 | bio_put(bio); |
4e4cbee9 | 2384 | ret = blk_status_to_errno(status); |
6c387ab2 MX |
2385 | } |
2386 | ||
013bd4c3 | 2387 | return ret; |
4a54c8c1 JS |
2388 | } |
2389 | ||
d1310b2e CM |
2390 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2391 | ||
b5227c07 | 2392 | void end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
87826df0 JM |
2393 | { |
2394 | int uptodate = (err == 0); | |
3e2426bd | 2395 | int ret = 0; |
87826df0 | 2396 | |
c629732d | 2397 | btrfs_writepage_endio_finish_ordered(page, start, end, uptodate); |
87826df0 | 2398 | |
87826df0 | 2399 | if (!uptodate) { |
87826df0 JM |
2400 | ClearPageUptodate(page); |
2401 | SetPageError(page); | |
bff5baf8 | 2402 | ret = err < 0 ? err : -EIO; |
5dca6eea | 2403 | mapping_set_error(page->mapping, ret); |
87826df0 | 2404 | } |
87826df0 JM |
2405 | } |
2406 | ||
d1310b2e CM |
2407 | /* |
2408 | * after a writepage IO is done, we need to: | |
2409 | * clear the uptodate bits on error | |
2410 | * clear the writeback bits in the extent tree for this IO | |
2411 | * end_page_writeback if the page has no more pending IO | |
2412 | * | |
2413 | * Scheduling is not allowed, so the extent state tree is expected | |
2414 | * to have one and only one object corresponding to this IO. | |
2415 | */ | |
4246a0b6 | 2416 | static void end_bio_extent_writepage(struct bio *bio) |
d1310b2e | 2417 | { |
4e4cbee9 | 2418 | int error = blk_status_to_errno(bio->bi_status); |
2c30c71b | 2419 | struct bio_vec *bvec; |
d1310b2e CM |
2420 | u64 start; |
2421 | u64 end; | |
2c30c71b | 2422 | int i; |
d1310b2e | 2423 | |
c09abff8 | 2424 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2c30c71b | 2425 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2426 | struct page *page = bvec->bv_page; |
0b246afa JM |
2427 | struct inode *inode = page->mapping->host; |
2428 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
902b22f3 | 2429 | |
17a5adcc AO |
2430 | /* We always issue full-page reads, but if some block |
2431 | * in a page fails to read, blk_update_request() will | |
2432 | * advance bv_offset and adjust bv_len to compensate. | |
2433 | * Print a warning for nonzero offsets, and an error | |
2434 | * if they don't add up to a full page. */ | |
09cbfeaf KS |
2435 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { |
2436 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
0b246afa | 2437 | btrfs_err(fs_info, |
efe120a0 FH |
2438 | "partial page write in btrfs with offset %u and length %u", |
2439 | bvec->bv_offset, bvec->bv_len); | |
2440 | else | |
0b246afa | 2441 | btrfs_info(fs_info, |
5d163e0e | 2442 | "incomplete page write in btrfs with offset %u and length %u", |
efe120a0 FH |
2443 | bvec->bv_offset, bvec->bv_len); |
2444 | } | |
d1310b2e | 2445 | |
17a5adcc AO |
2446 | start = page_offset(page); |
2447 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
d1310b2e | 2448 | |
4e4cbee9 | 2449 | end_extent_writepage(page, error, start, end); |
17a5adcc | 2450 | end_page_writeback(page); |
2c30c71b | 2451 | } |
2b1f55b0 | 2452 | |
d1310b2e | 2453 | bio_put(bio); |
d1310b2e CM |
2454 | } |
2455 | ||
883d0de4 MX |
2456 | static void |
2457 | endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len, | |
2458 | int uptodate) | |
2459 | { | |
2460 | struct extent_state *cached = NULL; | |
2461 | u64 end = start + len - 1; | |
2462 | ||
2463 | if (uptodate && tree->track_uptodate) | |
2464 | set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC); | |
d810a4be | 2465 | unlock_extent_cached_atomic(tree, start, end, &cached); |
883d0de4 MX |
2466 | } |
2467 | ||
d1310b2e CM |
2468 | /* |
2469 | * after a readpage IO is done, we need to: | |
2470 | * clear the uptodate bits on error | |
2471 | * set the uptodate bits if things worked | |
2472 | * set the page up to date if all extents in the tree are uptodate | |
2473 | * clear the lock bit in the extent tree | |
2474 | * unlock the page if there are no other extents locked for it | |
2475 | * | |
2476 | * Scheduling is not allowed, so the extent state tree is expected | |
2477 | * to have one and only one object corresponding to this IO. | |
2478 | */ | |
4246a0b6 | 2479 | static void end_bio_extent_readpage(struct bio *bio) |
d1310b2e | 2480 | { |
2c30c71b | 2481 | struct bio_vec *bvec; |
4e4cbee9 | 2482 | int uptodate = !bio->bi_status; |
facc8a22 | 2483 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); |
7870d082 | 2484 | struct extent_io_tree *tree, *failure_tree; |
facc8a22 | 2485 | u64 offset = 0; |
d1310b2e CM |
2486 | u64 start; |
2487 | u64 end; | |
facc8a22 | 2488 | u64 len; |
883d0de4 MX |
2489 | u64 extent_start = 0; |
2490 | u64 extent_len = 0; | |
5cf1ab56 | 2491 | int mirror; |
d1310b2e | 2492 | int ret; |
2c30c71b | 2493 | int i; |
d1310b2e | 2494 | |
c09abff8 | 2495 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2c30c71b | 2496 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2497 | struct page *page = bvec->bv_page; |
a71754fc | 2498 | struct inode *inode = page->mapping->host; |
ab8d0fc4 | 2499 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
78e62c02 NB |
2500 | bool data_inode = btrfs_ino(BTRFS_I(inode)) |
2501 | != BTRFS_BTREE_INODE_OBJECTID; | |
507903b8 | 2502 | |
ab8d0fc4 JM |
2503 | btrfs_debug(fs_info, |
2504 | "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u", | |
4e4cbee9 | 2505 | (u64)bio->bi_iter.bi_sector, bio->bi_status, |
ab8d0fc4 | 2506 | io_bio->mirror_num); |
a71754fc | 2507 | tree = &BTRFS_I(inode)->io_tree; |
7870d082 | 2508 | failure_tree = &BTRFS_I(inode)->io_failure_tree; |
902b22f3 | 2509 | |
17a5adcc AO |
2510 | /* We always issue full-page reads, but if some block |
2511 | * in a page fails to read, blk_update_request() will | |
2512 | * advance bv_offset and adjust bv_len to compensate. | |
2513 | * Print a warning for nonzero offsets, and an error | |
2514 | * if they don't add up to a full page. */ | |
09cbfeaf KS |
2515 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { |
2516 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
ab8d0fc4 JM |
2517 | btrfs_err(fs_info, |
2518 | "partial page read in btrfs with offset %u and length %u", | |
efe120a0 FH |
2519 | bvec->bv_offset, bvec->bv_len); |
2520 | else | |
ab8d0fc4 JM |
2521 | btrfs_info(fs_info, |
2522 | "incomplete page read in btrfs with offset %u and length %u", | |
efe120a0 FH |
2523 | bvec->bv_offset, bvec->bv_len); |
2524 | } | |
d1310b2e | 2525 | |
17a5adcc AO |
2526 | start = page_offset(page); |
2527 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
facc8a22 | 2528 | len = bvec->bv_len; |
d1310b2e | 2529 | |
9be3395b | 2530 | mirror = io_bio->mirror_num; |
78e62c02 | 2531 | if (likely(uptodate)) { |
facc8a22 MX |
2532 | ret = tree->ops->readpage_end_io_hook(io_bio, offset, |
2533 | page, start, end, | |
2534 | mirror); | |
5ee0844d | 2535 | if (ret) |
d1310b2e | 2536 | uptodate = 0; |
5ee0844d | 2537 | else |
7870d082 JB |
2538 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
2539 | failure_tree, tree, start, | |
2540 | page, | |
2541 | btrfs_ino(BTRFS_I(inode)), 0); | |
d1310b2e | 2542 | } |
ea466794 | 2543 | |
f2a09da9 MX |
2544 | if (likely(uptodate)) |
2545 | goto readpage_ok; | |
2546 | ||
78e62c02 | 2547 | if (data_inode) { |
9d0d1c8b | 2548 | |
f4a8e656 | 2549 | /* |
78e62c02 NB |
2550 | * The generic bio_readpage_error handles errors the |
2551 | * following way: If possible, new read requests are | |
2552 | * created and submitted and will end up in | |
2553 | * end_bio_extent_readpage as well (if we're lucky, | |
2554 | * not in the !uptodate case). In that case it returns | |
2555 | * 0 and we just go on with the next page in our bio. | |
2556 | * If it can't handle the error it will return -EIO and | |
2557 | * we remain responsible for that page. | |
f4a8e656 | 2558 | */ |
78e62c02 NB |
2559 | ret = bio_readpage_error(bio, offset, page, start, end, |
2560 | mirror); | |
2561 | if (ret == 0) { | |
2562 | uptodate = !bio->bi_status; | |
2563 | offset += len; | |
2564 | continue; | |
2565 | } | |
2566 | } else { | |
2567 | struct extent_buffer *eb; | |
2568 | ||
2569 | eb = (struct extent_buffer *)page->private; | |
2570 | set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); | |
2571 | eb->read_mirror = mirror; | |
2572 | atomic_dec(&eb->io_pages); | |
2573 | if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, | |
2574 | &eb->bflags)) | |
2575 | btree_readahead_hook(eb, -EIO); | |
2576 | ||
2577 | ret = -EIO; | |
7e38326f | 2578 | } |
f2a09da9 | 2579 | readpage_ok: |
883d0de4 | 2580 | if (likely(uptodate)) { |
a71754fc | 2581 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 2582 | pgoff_t end_index = i_size >> PAGE_SHIFT; |
a583c026 | 2583 | unsigned off; |
a71754fc JB |
2584 | |
2585 | /* Zero out the end if this page straddles i_size */ | |
7073017a | 2586 | off = offset_in_page(i_size); |
a583c026 | 2587 | if (page->index == end_index && off) |
09cbfeaf | 2588 | zero_user_segment(page, off, PAGE_SIZE); |
17a5adcc | 2589 | SetPageUptodate(page); |
70dec807 | 2590 | } else { |
17a5adcc AO |
2591 | ClearPageUptodate(page); |
2592 | SetPageError(page); | |
70dec807 | 2593 | } |
17a5adcc | 2594 | unlock_page(page); |
facc8a22 | 2595 | offset += len; |
883d0de4 MX |
2596 | |
2597 | if (unlikely(!uptodate)) { | |
2598 | if (extent_len) { | |
2599 | endio_readpage_release_extent(tree, | |
2600 | extent_start, | |
2601 | extent_len, 1); | |
2602 | extent_start = 0; | |
2603 | extent_len = 0; | |
2604 | } | |
2605 | endio_readpage_release_extent(tree, start, | |
2606 | end - start + 1, 0); | |
2607 | } else if (!extent_len) { | |
2608 | extent_start = start; | |
2609 | extent_len = end + 1 - start; | |
2610 | } else if (extent_start + extent_len == start) { | |
2611 | extent_len += end + 1 - start; | |
2612 | } else { | |
2613 | endio_readpage_release_extent(tree, extent_start, | |
2614 | extent_len, uptodate); | |
2615 | extent_start = start; | |
2616 | extent_len = end + 1 - start; | |
2617 | } | |
2c30c71b | 2618 | } |
d1310b2e | 2619 | |
883d0de4 MX |
2620 | if (extent_len) |
2621 | endio_readpage_release_extent(tree, extent_start, extent_len, | |
2622 | uptodate); | |
b3a0dd50 | 2623 | btrfs_io_bio_free_csum(io_bio); |
d1310b2e | 2624 | bio_put(bio); |
d1310b2e CM |
2625 | } |
2626 | ||
9be3395b | 2627 | /* |
184f999e DS |
2628 | * Initialize the members up to but not including 'bio'. Use after allocating a |
2629 | * new bio by bio_alloc_bioset as it does not initialize the bytes outside of | |
2630 | * 'bio' because use of __GFP_ZERO is not supported. | |
9be3395b | 2631 | */ |
184f999e | 2632 | static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio) |
d1310b2e | 2633 | { |
184f999e DS |
2634 | memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio)); |
2635 | } | |
d1310b2e | 2636 | |
9be3395b | 2637 | /* |
6e707bcd DS |
2638 | * The following helpers allocate a bio. As it's backed by a bioset, it'll |
2639 | * never fail. We're returning a bio right now but you can call btrfs_io_bio | |
2640 | * for the appropriate container_of magic | |
9be3395b | 2641 | */ |
c821e7f3 | 2642 | struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte) |
d1310b2e CM |
2643 | { |
2644 | struct bio *bio; | |
d1310b2e | 2645 | |
8ac9f7c1 | 2646 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &btrfs_bioset); |
74d46992 | 2647 | bio_set_dev(bio, bdev); |
c821e7f3 | 2648 | bio->bi_iter.bi_sector = first_byte >> 9; |
184f999e | 2649 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
d1310b2e CM |
2650 | return bio; |
2651 | } | |
2652 | ||
8b6c1d56 | 2653 | struct bio *btrfs_bio_clone(struct bio *bio) |
9be3395b | 2654 | { |
23ea8e5a MX |
2655 | struct btrfs_io_bio *btrfs_bio; |
2656 | struct bio *new; | |
9be3395b | 2657 | |
6e707bcd | 2658 | /* Bio allocation backed by a bioset does not fail */ |
8ac9f7c1 | 2659 | new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset); |
6e707bcd | 2660 | btrfs_bio = btrfs_io_bio(new); |
184f999e | 2661 | btrfs_io_bio_init(btrfs_bio); |
6e707bcd | 2662 | btrfs_bio->iter = bio->bi_iter; |
23ea8e5a MX |
2663 | return new; |
2664 | } | |
9be3395b | 2665 | |
c5e4c3d7 | 2666 | struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs) |
9be3395b | 2667 | { |
facc8a22 MX |
2668 | struct bio *bio; |
2669 | ||
6e707bcd | 2670 | /* Bio allocation backed by a bioset does not fail */ |
8ac9f7c1 | 2671 | bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset); |
184f999e | 2672 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
facc8a22 | 2673 | return bio; |
9be3395b CM |
2674 | } |
2675 | ||
e477094f | 2676 | struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size) |
2f8e9140 LB |
2677 | { |
2678 | struct bio *bio; | |
2679 | struct btrfs_io_bio *btrfs_bio; | |
2680 | ||
2681 | /* this will never fail when it's backed by a bioset */ | |
8ac9f7c1 | 2682 | bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset); |
2f8e9140 LB |
2683 | ASSERT(bio); |
2684 | ||
2685 | btrfs_bio = btrfs_io_bio(bio); | |
184f999e | 2686 | btrfs_io_bio_init(btrfs_bio); |
2f8e9140 LB |
2687 | |
2688 | bio_trim(bio, offset >> 9, size >> 9); | |
17347cec | 2689 | btrfs_bio->iter = bio->bi_iter; |
2f8e9140 LB |
2690 | return bio; |
2691 | } | |
9be3395b | 2692 | |
1f7ad75b MC |
2693 | static int __must_check submit_one_bio(struct bio *bio, int mirror_num, |
2694 | unsigned long bio_flags) | |
d1310b2e | 2695 | { |
4e4cbee9 | 2696 | blk_status_t ret = 0; |
c45a8f2d | 2697 | struct bio_vec *bvec = bio_last_bvec_all(bio); |
70dec807 CM |
2698 | struct page *page = bvec->bv_page; |
2699 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 2700 | u64 start; |
70dec807 | 2701 | |
4eee4fa4 | 2702 | start = page_offset(page) + bvec->bv_offset; |
70dec807 | 2703 | |
902b22f3 | 2704 | bio->bi_private = NULL; |
d1310b2e | 2705 | |
20c9801d | 2706 | if (tree->ops) |
c6100a4b | 2707 | ret = tree->ops->submit_bio_hook(tree->private_data, bio, |
eaf25d93 | 2708 | mirror_num, bio_flags, start); |
0b86a832 | 2709 | else |
4e49ea4a | 2710 | btrfsic_submit_bio(bio); |
4a54c8c1 | 2711 | |
4e4cbee9 | 2712 | return blk_status_to_errno(ret); |
d1310b2e CM |
2713 | } |
2714 | ||
4b81ba48 DS |
2715 | /* |
2716 | * @opf: bio REQ_OP_* and REQ_* flags as one value | |
b8b3d625 DS |
2717 | * @tree: tree so we can call our merge_bio hook |
2718 | * @wbc: optional writeback control for io accounting | |
2719 | * @page: page to add to the bio | |
2720 | * @pg_offset: offset of the new bio or to check whether we are adding | |
2721 | * a contiguous page to the previous one | |
2722 | * @size: portion of page that we want to write | |
2723 | * @offset: starting offset in the page | |
2724 | * @bdev: attach newly created bios to this bdev | |
5c2b1fd7 | 2725 | * @bio_ret: must be valid pointer, newly allocated bio will be stored there |
b8b3d625 DS |
2726 | * @end_io_func: end_io callback for new bio |
2727 | * @mirror_num: desired mirror to read/write | |
2728 | * @prev_bio_flags: flags of previous bio to see if we can merge the current one | |
2729 | * @bio_flags: flags of the current bio to see if we can merge them | |
4b81ba48 DS |
2730 | */ |
2731 | static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree, | |
da2f0f74 | 2732 | struct writeback_control *wbc, |
6273b7f8 | 2733 | struct page *page, u64 offset, |
6c5a4e2c | 2734 | size_t size, unsigned long pg_offset, |
d1310b2e CM |
2735 | struct block_device *bdev, |
2736 | struct bio **bio_ret, | |
f188591e | 2737 | bio_end_io_t end_io_func, |
c8b97818 CM |
2738 | int mirror_num, |
2739 | unsigned long prev_bio_flags, | |
005efedf FM |
2740 | unsigned long bio_flags, |
2741 | bool force_bio_submit) | |
d1310b2e CM |
2742 | { |
2743 | int ret = 0; | |
2744 | struct bio *bio; | |
09cbfeaf | 2745 | size_t page_size = min_t(size_t, size, PAGE_SIZE); |
6273b7f8 | 2746 | sector_t sector = offset >> 9; |
d1310b2e | 2747 | |
5c2b1fd7 DS |
2748 | ASSERT(bio_ret); |
2749 | ||
2750 | if (*bio_ret) { | |
0c8508a6 DS |
2751 | bool contig; |
2752 | bool can_merge = true; | |
2753 | ||
d1310b2e | 2754 | bio = *bio_ret; |
0c8508a6 | 2755 | if (prev_bio_flags & EXTENT_BIO_COMPRESSED) |
4f024f37 | 2756 | contig = bio->bi_iter.bi_sector == sector; |
c8b97818 | 2757 | else |
f73a1c7d | 2758 | contig = bio_end_sector(bio) == sector; |
c8b97818 | 2759 | |
da12fe54 NB |
2760 | ASSERT(tree->ops); |
2761 | if (btrfs_bio_fits_in_stripe(page, page_size, bio, bio_flags)) | |
0c8508a6 DS |
2762 | can_merge = false; |
2763 | ||
2764 | if (prev_bio_flags != bio_flags || !contig || !can_merge || | |
005efedf | 2765 | force_bio_submit || |
6c5a4e2c | 2766 | bio_add_page(bio, page, page_size, pg_offset) < page_size) { |
1f7ad75b | 2767 | ret = submit_one_bio(bio, mirror_num, prev_bio_flags); |
289454ad NA |
2768 | if (ret < 0) { |
2769 | *bio_ret = NULL; | |
79787eaa | 2770 | return ret; |
289454ad | 2771 | } |
d1310b2e CM |
2772 | bio = NULL; |
2773 | } else { | |
da2f0f74 CM |
2774 | if (wbc) |
2775 | wbc_account_io(wbc, page, page_size); | |
d1310b2e CM |
2776 | return 0; |
2777 | } | |
2778 | } | |
c8b97818 | 2779 | |
6273b7f8 | 2780 | bio = btrfs_bio_alloc(bdev, offset); |
6c5a4e2c | 2781 | bio_add_page(bio, page, page_size, pg_offset); |
d1310b2e CM |
2782 | bio->bi_end_io = end_io_func; |
2783 | bio->bi_private = tree; | |
e6959b93 | 2784 | bio->bi_write_hint = page->mapping->host->i_write_hint; |
4b81ba48 | 2785 | bio->bi_opf = opf; |
da2f0f74 CM |
2786 | if (wbc) { |
2787 | wbc_init_bio(wbc, bio); | |
2788 | wbc_account_io(wbc, page, page_size); | |
2789 | } | |
70dec807 | 2790 | |
5c2b1fd7 | 2791 | *bio_ret = bio; |
d1310b2e CM |
2792 | |
2793 | return ret; | |
2794 | } | |
2795 | ||
48a3b636 ES |
2796 | static void attach_extent_buffer_page(struct extent_buffer *eb, |
2797 | struct page *page) | |
d1310b2e CM |
2798 | { |
2799 | if (!PagePrivate(page)) { | |
2800 | SetPagePrivate(page); | |
09cbfeaf | 2801 | get_page(page); |
4f2de97a JB |
2802 | set_page_private(page, (unsigned long)eb); |
2803 | } else { | |
2804 | WARN_ON(page->private != (unsigned long)eb); | |
d1310b2e CM |
2805 | } |
2806 | } | |
2807 | ||
4f2de97a | 2808 | void set_page_extent_mapped(struct page *page) |
d1310b2e | 2809 | { |
4f2de97a JB |
2810 | if (!PagePrivate(page)) { |
2811 | SetPagePrivate(page); | |
09cbfeaf | 2812 | get_page(page); |
4f2de97a JB |
2813 | set_page_private(page, EXTENT_PAGE_PRIVATE); |
2814 | } | |
d1310b2e CM |
2815 | } |
2816 | ||
125bac01 MX |
2817 | static struct extent_map * |
2818 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
2819 | u64 start, u64 len, get_extent_t *get_extent, | |
2820 | struct extent_map **em_cached) | |
2821 | { | |
2822 | struct extent_map *em; | |
2823 | ||
2824 | if (em_cached && *em_cached) { | |
2825 | em = *em_cached; | |
cbc0e928 | 2826 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 | 2827 | start < extent_map_end(em)) { |
490b54d6 | 2828 | refcount_inc(&em->refs); |
125bac01 MX |
2829 | return em; |
2830 | } | |
2831 | ||
2832 | free_extent_map(em); | |
2833 | *em_cached = NULL; | |
2834 | } | |
2835 | ||
fc4f21b1 | 2836 | em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0); |
125bac01 MX |
2837 | if (em_cached && !IS_ERR_OR_NULL(em)) { |
2838 | BUG_ON(*em_cached); | |
490b54d6 | 2839 | refcount_inc(&em->refs); |
125bac01 MX |
2840 | *em_cached = em; |
2841 | } | |
2842 | return em; | |
2843 | } | |
d1310b2e CM |
2844 | /* |
2845 | * basic readpage implementation. Locked extent state structs are inserted | |
2846 | * into the tree that are removed when the IO is done (by the end_io | |
2847 | * handlers) | |
79787eaa | 2848 | * XXX JDM: This needs looking at to ensure proper page locking |
baf863b9 | 2849 | * return 0 on success, otherwise return error |
d1310b2e | 2850 | */ |
9974090b MX |
2851 | static int __do_readpage(struct extent_io_tree *tree, |
2852 | struct page *page, | |
2853 | get_extent_t *get_extent, | |
125bac01 | 2854 | struct extent_map **em_cached, |
9974090b | 2855 | struct bio **bio, int mirror_num, |
f1c77c55 | 2856 | unsigned long *bio_flags, unsigned int read_flags, |
005efedf | 2857 | u64 *prev_em_start) |
d1310b2e CM |
2858 | { |
2859 | struct inode *inode = page->mapping->host; | |
4eee4fa4 | 2860 | u64 start = page_offset(page); |
8eec8296 | 2861 | const u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
2862 | u64 cur = start; |
2863 | u64 extent_offset; | |
2864 | u64 last_byte = i_size_read(inode); | |
2865 | u64 block_start; | |
2866 | u64 cur_end; | |
d1310b2e CM |
2867 | struct extent_map *em; |
2868 | struct block_device *bdev; | |
baf863b9 | 2869 | int ret = 0; |
d1310b2e | 2870 | int nr = 0; |
306e16ce | 2871 | size_t pg_offset = 0; |
d1310b2e | 2872 | size_t iosize; |
c8b97818 | 2873 | size_t disk_io_size; |
d1310b2e | 2874 | size_t blocksize = inode->i_sb->s_blocksize; |
7f042a83 | 2875 | unsigned long this_bio_flag = 0; |
d1310b2e CM |
2876 | |
2877 | set_page_extent_mapped(page); | |
2878 | ||
90a887c9 DM |
2879 | if (!PageUptodate(page)) { |
2880 | if (cleancache_get_page(page) == 0) { | |
2881 | BUG_ON(blocksize != PAGE_SIZE); | |
9974090b | 2882 | unlock_extent(tree, start, end); |
90a887c9 DM |
2883 | goto out; |
2884 | } | |
2885 | } | |
2886 | ||
09cbfeaf | 2887 | if (page->index == last_byte >> PAGE_SHIFT) { |
c8b97818 | 2888 | char *userpage; |
7073017a | 2889 | size_t zero_offset = offset_in_page(last_byte); |
c8b97818 CM |
2890 | |
2891 | if (zero_offset) { | |
09cbfeaf | 2892 | iosize = PAGE_SIZE - zero_offset; |
7ac687d9 | 2893 | userpage = kmap_atomic(page); |
c8b97818 CM |
2894 | memset(userpage + zero_offset, 0, iosize); |
2895 | flush_dcache_page(page); | |
7ac687d9 | 2896 | kunmap_atomic(userpage); |
c8b97818 CM |
2897 | } |
2898 | } | |
d1310b2e | 2899 | while (cur <= end) { |
005efedf | 2900 | bool force_bio_submit = false; |
6273b7f8 | 2901 | u64 offset; |
c8f2f24b | 2902 | |
d1310b2e CM |
2903 | if (cur >= last_byte) { |
2904 | char *userpage; | |
507903b8 AJ |
2905 | struct extent_state *cached = NULL; |
2906 | ||
09cbfeaf | 2907 | iosize = PAGE_SIZE - pg_offset; |
7ac687d9 | 2908 | userpage = kmap_atomic(page); |
306e16ce | 2909 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2910 | flush_dcache_page(page); |
7ac687d9 | 2911 | kunmap_atomic(userpage); |
d1310b2e | 2912 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 2913 | &cached, GFP_NOFS); |
7f042a83 | 2914 | unlock_extent_cached(tree, cur, |
e43bbe5e | 2915 | cur + iosize - 1, &cached); |
d1310b2e CM |
2916 | break; |
2917 | } | |
125bac01 MX |
2918 | em = __get_extent_map(inode, page, pg_offset, cur, |
2919 | end - cur + 1, get_extent, em_cached); | |
c704005d | 2920 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 2921 | SetPageError(page); |
7f042a83 | 2922 | unlock_extent(tree, cur, end); |
d1310b2e CM |
2923 | break; |
2924 | } | |
d1310b2e CM |
2925 | extent_offset = cur - em->start; |
2926 | BUG_ON(extent_map_end(em) <= cur); | |
2927 | BUG_ON(end < cur); | |
2928 | ||
261507a0 | 2929 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
4b384318 | 2930 | this_bio_flag |= EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
2931 | extent_set_compress_type(&this_bio_flag, |
2932 | em->compress_type); | |
2933 | } | |
c8b97818 | 2934 | |
d1310b2e CM |
2935 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
2936 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 2937 | iosize = ALIGN(iosize, blocksize); |
c8b97818 CM |
2938 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
2939 | disk_io_size = em->block_len; | |
6273b7f8 | 2940 | offset = em->block_start; |
c8b97818 | 2941 | } else { |
6273b7f8 | 2942 | offset = em->block_start + extent_offset; |
c8b97818 CM |
2943 | disk_io_size = iosize; |
2944 | } | |
d1310b2e CM |
2945 | bdev = em->bdev; |
2946 | block_start = em->block_start; | |
d899e052 YZ |
2947 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2948 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
2949 | |
2950 | /* | |
2951 | * If we have a file range that points to a compressed extent | |
2952 | * and it's followed by a consecutive file range that points to | |
2953 | * to the same compressed extent (possibly with a different | |
2954 | * offset and/or length, so it either points to the whole extent | |
2955 | * or only part of it), we must make sure we do not submit a | |
2956 | * single bio to populate the pages for the 2 ranges because | |
2957 | * this makes the compressed extent read zero out the pages | |
2958 | * belonging to the 2nd range. Imagine the following scenario: | |
2959 | * | |
2960 | * File layout | |
2961 | * [0 - 8K] [8K - 24K] | |
2962 | * | | | |
2963 | * | | | |
2964 | * points to extent X, points to extent X, | |
2965 | * offset 4K, length of 8K offset 0, length 16K | |
2966 | * | |
2967 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
2968 | * | |
2969 | * If the bio to read the compressed extent covers both ranges, | |
2970 | * it will decompress extent X into the pages belonging to the | |
2971 | * first range and then it will stop, zeroing out the remaining | |
2972 | * pages that belong to the other range that points to extent X. | |
2973 | * So here we make sure we submit 2 bios, one for the first | |
2974 | * range and another one for the third range. Both will target | |
2975 | * the same physical extent from disk, but we can't currently | |
2976 | * make the compressed bio endio callback populate the pages | |
2977 | * for both ranges because each compressed bio is tightly | |
2978 | * coupled with a single extent map, and each range can have | |
2979 | * an extent map with a different offset value relative to the | |
2980 | * uncompressed data of our extent and different lengths. This | |
2981 | * is a corner case so we prioritize correctness over | |
2982 | * non-optimal behavior (submitting 2 bios for the same extent). | |
2983 | */ | |
2984 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
2985 | prev_em_start && *prev_em_start != (u64)-1 && | |
2986 | *prev_em_start != em->orig_start) | |
2987 | force_bio_submit = true; | |
2988 | ||
2989 | if (prev_em_start) | |
2990 | *prev_em_start = em->orig_start; | |
2991 | ||
d1310b2e CM |
2992 | free_extent_map(em); |
2993 | em = NULL; | |
2994 | ||
2995 | /* we've found a hole, just zero and go on */ | |
2996 | if (block_start == EXTENT_MAP_HOLE) { | |
2997 | char *userpage; | |
507903b8 AJ |
2998 | struct extent_state *cached = NULL; |
2999 | ||
7ac687d9 | 3000 | userpage = kmap_atomic(page); |
306e16ce | 3001 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3002 | flush_dcache_page(page); |
7ac687d9 | 3003 | kunmap_atomic(userpage); |
d1310b2e CM |
3004 | |
3005 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 | 3006 | &cached, GFP_NOFS); |
7f042a83 | 3007 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3008 | cur + iosize - 1, &cached); |
d1310b2e | 3009 | cur = cur + iosize; |
306e16ce | 3010 | pg_offset += iosize; |
d1310b2e CM |
3011 | continue; |
3012 | } | |
3013 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
3014 | if (test_range_bit(tree, cur, cur_end, |
3015 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 3016 | check_page_uptodate(tree, page); |
7f042a83 | 3017 | unlock_extent(tree, cur, cur + iosize - 1); |
d1310b2e | 3018 | cur = cur + iosize; |
306e16ce | 3019 | pg_offset += iosize; |
d1310b2e CM |
3020 | continue; |
3021 | } | |
70dec807 CM |
3022 | /* we have an inline extent but it didn't get marked up |
3023 | * to date. Error out | |
3024 | */ | |
3025 | if (block_start == EXTENT_MAP_INLINE) { | |
3026 | SetPageError(page); | |
7f042a83 | 3027 | unlock_extent(tree, cur, cur + iosize - 1); |
70dec807 | 3028 | cur = cur + iosize; |
306e16ce | 3029 | pg_offset += iosize; |
70dec807 CM |
3030 | continue; |
3031 | } | |
d1310b2e | 3032 | |
4b81ba48 | 3033 | ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL, |
6273b7f8 DS |
3034 | page, offset, disk_io_size, |
3035 | pg_offset, bdev, bio, | |
c8b97818 CM |
3036 | end_bio_extent_readpage, mirror_num, |
3037 | *bio_flags, | |
005efedf FM |
3038 | this_bio_flag, |
3039 | force_bio_submit); | |
c8f2f24b JB |
3040 | if (!ret) { |
3041 | nr++; | |
3042 | *bio_flags = this_bio_flag; | |
3043 | } else { | |
d1310b2e | 3044 | SetPageError(page); |
7f042a83 | 3045 | unlock_extent(tree, cur, cur + iosize - 1); |
baf863b9 | 3046 | goto out; |
edd33c99 | 3047 | } |
d1310b2e | 3048 | cur = cur + iosize; |
306e16ce | 3049 | pg_offset += iosize; |
d1310b2e | 3050 | } |
90a887c9 | 3051 | out: |
d1310b2e CM |
3052 | if (!nr) { |
3053 | if (!PageError(page)) | |
3054 | SetPageUptodate(page); | |
3055 | unlock_page(page); | |
3056 | } | |
baf863b9 | 3057 | return ret; |
d1310b2e CM |
3058 | } |
3059 | ||
9974090b MX |
3060 | static inline void __do_contiguous_readpages(struct extent_io_tree *tree, |
3061 | struct page *pages[], int nr_pages, | |
3062 | u64 start, u64 end, | |
125bac01 | 3063 | struct extent_map **em_cached, |
d3fac6ba | 3064 | struct bio **bio, |
1f7ad75b | 3065 | unsigned long *bio_flags, |
808f80b4 | 3066 | u64 *prev_em_start) |
9974090b MX |
3067 | { |
3068 | struct inode *inode; | |
3069 | struct btrfs_ordered_extent *ordered; | |
3070 | int index; | |
3071 | ||
3072 | inode = pages[0]->mapping->host; | |
3073 | while (1) { | |
3074 | lock_extent(tree, start, end); | |
a776c6fa | 3075 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
9974090b MX |
3076 | end - start + 1); |
3077 | if (!ordered) | |
3078 | break; | |
3079 | unlock_extent(tree, start, end); | |
3080 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3081 | btrfs_put_ordered_extent(ordered); | |
3082 | } | |
3083 | ||
3084 | for (index = 0; index < nr_pages; index++) { | |
4ef77695 | 3085 | __do_readpage(tree, pages[index], btrfs_get_extent, em_cached, |
5e9d3982 | 3086 | bio, 0, bio_flags, REQ_RAHEAD, prev_em_start); |
09cbfeaf | 3087 | put_page(pages[index]); |
9974090b MX |
3088 | } |
3089 | } | |
3090 | ||
3091 | static void __extent_readpages(struct extent_io_tree *tree, | |
3092 | struct page *pages[], | |
e4d17ef5 | 3093 | int nr_pages, |
125bac01 | 3094 | struct extent_map **em_cached, |
d3fac6ba | 3095 | struct bio **bio, unsigned long *bio_flags, |
808f80b4 | 3096 | u64 *prev_em_start) |
9974090b | 3097 | { |
35a3621b | 3098 | u64 start = 0; |
9974090b MX |
3099 | u64 end = 0; |
3100 | u64 page_start; | |
3101 | int index; | |
35a3621b | 3102 | int first_index = 0; |
9974090b MX |
3103 | |
3104 | for (index = 0; index < nr_pages; index++) { | |
3105 | page_start = page_offset(pages[index]); | |
3106 | if (!end) { | |
3107 | start = page_start; | |
09cbfeaf | 3108 | end = start + PAGE_SIZE - 1; |
9974090b MX |
3109 | first_index = index; |
3110 | } else if (end + 1 == page_start) { | |
09cbfeaf | 3111 | end += PAGE_SIZE; |
9974090b MX |
3112 | } else { |
3113 | __do_contiguous_readpages(tree, &pages[first_index], | |
3114 | index - first_index, start, | |
4ef77695 | 3115 | end, em_cached, |
d3fac6ba | 3116 | bio, bio_flags, |
1f7ad75b | 3117 | prev_em_start); |
9974090b | 3118 | start = page_start; |
09cbfeaf | 3119 | end = start + PAGE_SIZE - 1; |
9974090b MX |
3120 | first_index = index; |
3121 | } | |
3122 | } | |
3123 | ||
3124 | if (end) | |
3125 | __do_contiguous_readpages(tree, &pages[first_index], | |
3126 | index - first_index, start, | |
4ef77695 | 3127 | end, em_cached, bio, |
d3fac6ba | 3128 | bio_flags, prev_em_start); |
9974090b MX |
3129 | } |
3130 | ||
3131 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
3132 | struct page *page, | |
3133 | get_extent_t *get_extent, | |
3134 | struct bio **bio, int mirror_num, | |
f1c77c55 DS |
3135 | unsigned long *bio_flags, |
3136 | unsigned int read_flags) | |
9974090b MX |
3137 | { |
3138 | struct inode *inode = page->mapping->host; | |
3139 | struct btrfs_ordered_extent *ordered; | |
3140 | u64 start = page_offset(page); | |
09cbfeaf | 3141 | u64 end = start + PAGE_SIZE - 1; |
9974090b MX |
3142 | int ret; |
3143 | ||
3144 | while (1) { | |
3145 | lock_extent(tree, start, end); | |
a776c6fa | 3146 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
09cbfeaf | 3147 | PAGE_SIZE); |
9974090b MX |
3148 | if (!ordered) |
3149 | break; | |
3150 | unlock_extent(tree, start, end); | |
3151 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3152 | btrfs_put_ordered_extent(ordered); | |
3153 | } | |
3154 | ||
125bac01 | 3155 | ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num, |
1f7ad75b | 3156 | bio_flags, read_flags, NULL); |
9974090b MX |
3157 | return ret; |
3158 | } | |
3159 | ||
d1310b2e | 3160 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, |
8ddc7d9c | 3161 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3162 | { |
3163 | struct bio *bio = NULL; | |
c8b97818 | 3164 | unsigned long bio_flags = 0; |
d1310b2e CM |
3165 | int ret; |
3166 | ||
8ddc7d9c | 3167 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, |
1f7ad75b | 3168 | &bio_flags, 0); |
d1310b2e | 3169 | if (bio) |
1f7ad75b | 3170 | ret = submit_one_bio(bio, mirror_num, bio_flags); |
d1310b2e CM |
3171 | return ret; |
3172 | } | |
d1310b2e | 3173 | |
3d4b9496 | 3174 | static void update_nr_written(struct writeback_control *wbc, |
a9132667 | 3175 | unsigned long nr_written) |
11c8349b CM |
3176 | { |
3177 | wbc->nr_to_write -= nr_written; | |
11c8349b CM |
3178 | } |
3179 | ||
d1310b2e | 3180 | /* |
40f76580 CM |
3181 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
3182 | * | |
5eaad97a | 3183 | * This returns 1 if btrfs_run_delalloc_range function did all the work required |
40f76580 CM |
3184 | * to write the page (copy into inline extent). In this case the IO has |
3185 | * been started and the page is already unlocked. | |
3186 | * | |
3187 | * This returns 0 if all went well (page still locked) | |
3188 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 3189 | */ |
40f76580 | 3190 | static noinline_for_stack int writepage_delalloc(struct inode *inode, |
8cc0237a NB |
3191 | struct page *page, struct writeback_control *wbc, |
3192 | u64 delalloc_start, unsigned long *nr_written) | |
40f76580 | 3193 | { |
8cc0237a | 3194 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
09cbfeaf | 3195 | u64 page_end = delalloc_start + PAGE_SIZE - 1; |
3522e903 | 3196 | bool found; |
40f76580 CM |
3197 | u64 delalloc_to_write = 0; |
3198 | u64 delalloc_end = 0; | |
3199 | int ret; | |
3200 | int page_started = 0; | |
3201 | ||
40f76580 CM |
3202 | |
3203 | while (delalloc_end < page_end) { | |
3522e903 | 3204 | found = find_lock_delalloc_range(inode, tree, |
40f76580 CM |
3205 | page, |
3206 | &delalloc_start, | |
917aacec | 3207 | &delalloc_end); |
3522e903 | 3208 | if (!found) { |
40f76580 CM |
3209 | delalloc_start = delalloc_end + 1; |
3210 | continue; | |
3211 | } | |
5eaad97a NB |
3212 | ret = btrfs_run_delalloc_range(inode, page, delalloc_start, |
3213 | delalloc_end, &page_started, nr_written, wbc); | |
40f76580 CM |
3214 | /* File system has been set read-only */ |
3215 | if (ret) { | |
3216 | SetPageError(page); | |
5eaad97a NB |
3217 | /* |
3218 | * btrfs_run_delalloc_range should return < 0 for error | |
3219 | * but just in case, we use > 0 here meaning the IO is | |
3220 | * started, so we don't want to return > 0 unless | |
3221 | * things are going well. | |
40f76580 CM |
3222 | */ |
3223 | ret = ret < 0 ? ret : -EIO; | |
3224 | goto done; | |
3225 | } | |
3226 | /* | |
ea1754a0 KS |
3227 | * delalloc_end is already one less than the total length, so |
3228 | * we don't subtract one from PAGE_SIZE | |
40f76580 CM |
3229 | */ |
3230 | delalloc_to_write += (delalloc_end - delalloc_start + | |
ea1754a0 | 3231 | PAGE_SIZE) >> PAGE_SHIFT; |
40f76580 CM |
3232 | delalloc_start = delalloc_end + 1; |
3233 | } | |
3234 | if (wbc->nr_to_write < delalloc_to_write) { | |
3235 | int thresh = 8192; | |
3236 | ||
3237 | if (delalloc_to_write < thresh * 2) | |
3238 | thresh = delalloc_to_write; | |
3239 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3240 | thresh); | |
3241 | } | |
3242 | ||
3243 | /* did the fill delalloc function already unlock and start | |
3244 | * the IO? | |
3245 | */ | |
3246 | if (page_started) { | |
3247 | /* | |
3248 | * we've unlocked the page, so we can't update | |
3249 | * the mapping's writeback index, just update | |
3250 | * nr_to_write. | |
3251 | */ | |
3252 | wbc->nr_to_write -= *nr_written; | |
3253 | return 1; | |
3254 | } | |
3255 | ||
3256 | ret = 0; | |
3257 | ||
3258 | done: | |
3259 | return ret; | |
3260 | } | |
3261 | ||
3262 | /* | |
3263 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3264 | * and does the loop to map the page into extents and bios. | |
3265 | * | |
3266 | * We return 1 if the IO is started and the page is unlocked, | |
3267 | * 0 if all went well (page still locked) | |
3268 | * < 0 if there were errors (page still locked) | |
3269 | */ | |
3270 | static noinline_for_stack int __extent_writepage_io(struct inode *inode, | |
3271 | struct page *page, | |
3272 | struct writeback_control *wbc, | |
3273 | struct extent_page_data *epd, | |
3274 | loff_t i_size, | |
3275 | unsigned long nr_written, | |
f1c77c55 | 3276 | unsigned int write_flags, int *nr_ret) |
d1310b2e | 3277 | { |
d1310b2e | 3278 | struct extent_io_tree *tree = epd->tree; |
4eee4fa4 | 3279 | u64 start = page_offset(page); |
09cbfeaf | 3280 | u64 page_end = start + PAGE_SIZE - 1; |
d1310b2e CM |
3281 | u64 end; |
3282 | u64 cur = start; | |
3283 | u64 extent_offset; | |
d1310b2e CM |
3284 | u64 block_start; |
3285 | u64 iosize; | |
d1310b2e CM |
3286 | struct extent_map *em; |
3287 | struct block_device *bdev; | |
7f3c74fb | 3288 | size_t pg_offset = 0; |
d1310b2e | 3289 | size_t blocksize; |
40f76580 CM |
3290 | int ret = 0; |
3291 | int nr = 0; | |
3292 | bool compressed; | |
c8b97818 | 3293 | |
d75855b4 NB |
3294 | ret = btrfs_writepage_cow_fixup(page, start, page_end); |
3295 | if (ret) { | |
3296 | /* Fixup worker will requeue */ | |
3297 | if (ret == -EBUSY) | |
3298 | wbc->pages_skipped++; | |
3299 | else | |
3300 | redirty_page_for_writepage(wbc, page); | |
40f76580 | 3301 | |
d75855b4 NB |
3302 | update_nr_written(wbc, nr_written); |
3303 | unlock_page(page); | |
3304 | return 1; | |
247e743c CM |
3305 | } |
3306 | ||
11c8349b CM |
3307 | /* |
3308 | * we don't want to touch the inode after unlocking the page, | |
3309 | * so we update the mapping writeback index now | |
3310 | */ | |
3d4b9496 | 3311 | update_nr_written(wbc, nr_written + 1); |
771ed689 | 3312 | |
d1310b2e | 3313 | end = page_end; |
40f76580 | 3314 | if (i_size <= start) { |
c629732d | 3315 | btrfs_writepage_endio_finish_ordered(page, start, page_end, 1); |
d1310b2e CM |
3316 | goto done; |
3317 | } | |
3318 | ||
d1310b2e CM |
3319 | blocksize = inode->i_sb->s_blocksize; |
3320 | ||
3321 | while (cur <= end) { | |
40f76580 | 3322 | u64 em_end; |
6273b7f8 | 3323 | u64 offset; |
58409edd | 3324 | |
40f76580 | 3325 | if (cur >= i_size) { |
7087a9d8 | 3326 | btrfs_writepage_endio_finish_ordered(page, cur, |
c629732d | 3327 | page_end, 1); |
d1310b2e CM |
3328 | break; |
3329 | } | |
3c98c62f | 3330 | em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, cur, |
d1310b2e | 3331 | end - cur + 1, 1); |
c704005d | 3332 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3333 | SetPageError(page); |
61391d56 | 3334 | ret = PTR_ERR_OR_ZERO(em); |
d1310b2e CM |
3335 | break; |
3336 | } | |
3337 | ||
3338 | extent_offset = cur - em->start; | |
40f76580 CM |
3339 | em_end = extent_map_end(em); |
3340 | BUG_ON(em_end <= cur); | |
d1310b2e | 3341 | BUG_ON(end < cur); |
40f76580 | 3342 | iosize = min(em_end - cur, end - cur + 1); |
fda2832f | 3343 | iosize = ALIGN(iosize, blocksize); |
6273b7f8 | 3344 | offset = em->block_start + extent_offset; |
d1310b2e CM |
3345 | bdev = em->bdev; |
3346 | block_start = em->block_start; | |
c8b97818 | 3347 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
3348 | free_extent_map(em); |
3349 | em = NULL; | |
3350 | ||
c8b97818 CM |
3351 | /* |
3352 | * compressed and inline extents are written through other | |
3353 | * paths in the FS | |
3354 | */ | |
3355 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3356 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
3357 | /* |
3358 | * end_io notification does not happen here for | |
3359 | * compressed extents | |
3360 | */ | |
7087a9d8 NB |
3361 | if (!compressed) |
3362 | btrfs_writepage_endio_finish_ordered(page, cur, | |
3363 | cur + iosize - 1, | |
c629732d | 3364 | 1); |
c8b97818 CM |
3365 | else if (compressed) { |
3366 | /* we don't want to end_page_writeback on | |
3367 | * a compressed extent. this happens | |
3368 | * elsewhere | |
3369 | */ | |
3370 | nr++; | |
3371 | } | |
3372 | ||
3373 | cur += iosize; | |
7f3c74fb | 3374 | pg_offset += iosize; |
d1310b2e CM |
3375 | continue; |
3376 | } | |
c8b97818 | 3377 | |
5cdc84bf | 3378 | btrfs_set_range_writeback(tree, cur, cur + iosize - 1); |
58409edd DS |
3379 | if (!PageWriteback(page)) { |
3380 | btrfs_err(BTRFS_I(inode)->root->fs_info, | |
3381 | "page %lu not writeback, cur %llu end %llu", | |
3382 | page->index, cur, end); | |
d1310b2e | 3383 | } |
7f3c74fb | 3384 | |
4b81ba48 | 3385 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc, |
6273b7f8 | 3386 | page, offset, iosize, pg_offset, |
c2df8bb4 | 3387 | bdev, &epd->bio, |
58409edd DS |
3388 | end_bio_extent_writepage, |
3389 | 0, 0, 0, false); | |
fe01aa65 | 3390 | if (ret) { |
58409edd | 3391 | SetPageError(page); |
fe01aa65 TK |
3392 | if (PageWriteback(page)) |
3393 | end_page_writeback(page); | |
3394 | } | |
d1310b2e | 3395 | |
d1310b2e | 3396 | cur = cur + iosize; |
7f3c74fb | 3397 | pg_offset += iosize; |
d1310b2e CM |
3398 | nr++; |
3399 | } | |
40f76580 CM |
3400 | done: |
3401 | *nr_ret = nr; | |
40f76580 CM |
3402 | return ret; |
3403 | } | |
3404 | ||
3405 | /* | |
3406 | * the writepage semantics are similar to regular writepage. extent | |
3407 | * records are inserted to lock ranges in the tree, and as dirty areas | |
3408 | * are found, they are marked writeback. Then the lock bits are removed | |
3409 | * and the end_io handler clears the writeback ranges | |
3410 | */ | |
3411 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
aab6e9ed | 3412 | struct extent_page_data *epd) |
40f76580 CM |
3413 | { |
3414 | struct inode *inode = page->mapping->host; | |
40f76580 | 3415 | u64 start = page_offset(page); |
09cbfeaf | 3416 | u64 page_end = start + PAGE_SIZE - 1; |
40f76580 CM |
3417 | int ret; |
3418 | int nr = 0; | |
3419 | size_t pg_offset = 0; | |
3420 | loff_t i_size = i_size_read(inode); | |
09cbfeaf | 3421 | unsigned long end_index = i_size >> PAGE_SHIFT; |
f1c77c55 | 3422 | unsigned int write_flags = 0; |
40f76580 CM |
3423 | unsigned long nr_written = 0; |
3424 | ||
ff40adf7 | 3425 | write_flags = wbc_to_write_flags(wbc); |
40f76580 CM |
3426 | |
3427 | trace___extent_writepage(page, inode, wbc); | |
3428 | ||
3429 | WARN_ON(!PageLocked(page)); | |
3430 | ||
3431 | ClearPageError(page); | |
3432 | ||
7073017a | 3433 | pg_offset = offset_in_page(i_size); |
40f76580 CM |
3434 | if (page->index > end_index || |
3435 | (page->index == end_index && !pg_offset)) { | |
09cbfeaf | 3436 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); |
40f76580 CM |
3437 | unlock_page(page); |
3438 | return 0; | |
3439 | } | |
3440 | ||
3441 | if (page->index == end_index) { | |
3442 | char *userpage; | |
3443 | ||
3444 | userpage = kmap_atomic(page); | |
3445 | memset(userpage + pg_offset, 0, | |
09cbfeaf | 3446 | PAGE_SIZE - pg_offset); |
40f76580 CM |
3447 | kunmap_atomic(userpage); |
3448 | flush_dcache_page(page); | |
3449 | } | |
3450 | ||
3451 | pg_offset = 0; | |
3452 | ||
3453 | set_page_extent_mapped(page); | |
3454 | ||
7789a55a | 3455 | if (!epd->extent_locked) { |
8cc0237a | 3456 | ret = writepage_delalloc(inode, page, wbc, start, &nr_written); |
7789a55a NB |
3457 | if (ret == 1) |
3458 | goto done_unlocked; | |
3459 | if (ret) | |
3460 | goto done; | |
3461 | } | |
40f76580 CM |
3462 | |
3463 | ret = __extent_writepage_io(inode, page, wbc, epd, | |
3464 | i_size, nr_written, write_flags, &nr); | |
3465 | if (ret == 1) | |
3466 | goto done_unlocked; | |
3467 | ||
d1310b2e CM |
3468 | done: |
3469 | if (nr == 0) { | |
3470 | /* make sure the mapping tag for page dirty gets cleared */ | |
3471 | set_page_writeback(page); | |
3472 | end_page_writeback(page); | |
3473 | } | |
61391d56 FM |
3474 | if (PageError(page)) { |
3475 | ret = ret < 0 ? ret : -EIO; | |
3476 | end_extent_writepage(page, ret, start, page_end); | |
3477 | } | |
d1310b2e | 3478 | unlock_page(page); |
40f76580 | 3479 | return ret; |
771ed689 | 3480 | |
11c8349b | 3481 | done_unlocked: |
d1310b2e CM |
3482 | return 0; |
3483 | } | |
3484 | ||
fd8b2b61 | 3485 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 3486 | { |
74316201 N |
3487 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
3488 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
3489 | } |
3490 | ||
0e378df1 CM |
3491 | static noinline_for_stack int |
3492 | lock_extent_buffer_for_io(struct extent_buffer *eb, | |
3493 | struct btrfs_fs_info *fs_info, | |
3494 | struct extent_page_data *epd) | |
0b32f4bb | 3495 | { |
cc5e31a4 | 3496 | int i, num_pages; |
0b32f4bb JB |
3497 | int flush = 0; |
3498 | int ret = 0; | |
3499 | ||
3500 | if (!btrfs_try_tree_write_lock(eb)) { | |
3501 | flush = 1; | |
3502 | flush_write_bio(epd); | |
3503 | btrfs_tree_lock(eb); | |
3504 | } | |
3505 | ||
3506 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3507 | btrfs_tree_unlock(eb); | |
3508 | if (!epd->sync_io) | |
3509 | return 0; | |
3510 | if (!flush) { | |
3511 | flush_write_bio(epd); | |
3512 | flush = 1; | |
3513 | } | |
a098d8e8 CM |
3514 | while (1) { |
3515 | wait_on_extent_buffer_writeback(eb); | |
3516 | btrfs_tree_lock(eb); | |
3517 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3518 | break; | |
0b32f4bb | 3519 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3520 | } |
3521 | } | |
3522 | ||
51561ffe JB |
3523 | /* |
3524 | * We need to do this to prevent races in people who check if the eb is | |
3525 | * under IO since we can end up having no IO bits set for a short period | |
3526 | * of time. | |
3527 | */ | |
3528 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
3529 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3530 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 3531 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 3532 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
104b4e51 NB |
3533 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
3534 | -eb->len, | |
3535 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 3536 | ret = 1; |
51561ffe JB |
3537 | } else { |
3538 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
3539 | } |
3540 | ||
3541 | btrfs_tree_unlock(eb); | |
3542 | ||
3543 | if (!ret) | |
3544 | return ret; | |
3545 | ||
65ad0104 | 3546 | num_pages = num_extent_pages(eb); |
0b32f4bb | 3547 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3548 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3549 | |
3550 | if (!trylock_page(p)) { | |
3551 | if (!flush) { | |
3552 | flush_write_bio(epd); | |
3553 | flush = 1; | |
3554 | } | |
3555 | lock_page(p); | |
3556 | } | |
3557 | } | |
3558 | ||
3559 | return ret; | |
3560 | } | |
3561 | ||
3562 | static void end_extent_buffer_writeback(struct extent_buffer *eb) | |
3563 | { | |
3564 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
4e857c58 | 3565 | smp_mb__after_atomic(); |
0b32f4bb JB |
3566 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); |
3567 | } | |
3568 | ||
656f30db FM |
3569 | static void set_btree_ioerr(struct page *page) |
3570 | { | |
3571 | struct extent_buffer *eb = (struct extent_buffer *)page->private; | |
656f30db FM |
3572 | |
3573 | SetPageError(page); | |
3574 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) | |
3575 | return; | |
3576 | ||
3577 | /* | |
3578 | * If writeback for a btree extent that doesn't belong to a log tree | |
3579 | * failed, increment the counter transaction->eb_write_errors. | |
3580 | * We do this because while the transaction is running and before it's | |
3581 | * committing (when we call filemap_fdata[write|wait]_range against | |
3582 | * the btree inode), we might have | |
3583 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
3584 | * returns an error or an error happens during writeback, when we're | |
3585 | * committing the transaction we wouldn't know about it, since the pages | |
3586 | * can be no longer dirty nor marked anymore for writeback (if a | |
3587 | * subsequent modification to the extent buffer didn't happen before the | |
3588 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
3589 | * able to find the pages tagged with SetPageError at transaction | |
3590 | * commit time. So if this happens we must abort the transaction, | |
3591 | * otherwise we commit a super block with btree roots that point to | |
3592 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
3593 | * or the content of some node/leaf from a past generation that got | |
3594 | * cowed or deleted and is no longer valid. | |
3595 | * | |
3596 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
3597 | * not be enough - we need to distinguish between log tree extents vs | |
3598 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
3599 | * will catch and clear such errors in the mapping - and that call might | |
3600 | * be from a log sync and not from a transaction commit. Also, checking | |
3601 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
3602 | * not done and would not be reliable - the eb might have been released | |
3603 | * from memory and reading it back again means that flag would not be | |
3604 | * set (since it's a runtime flag, not persisted on disk). | |
3605 | * | |
3606 | * Using the flags below in the btree inode also makes us achieve the | |
3607 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
3608 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
3609 | * is called, the writeback for all dirty pages had already finished | |
3610 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
3611 | * filemap_fdatawait_range() would return success, as it could not know | |
3612 | * that writeback errors happened (the pages were no longer tagged for | |
3613 | * writeback). | |
3614 | */ | |
3615 | switch (eb->log_index) { | |
3616 | case -1: | |
afcdd129 | 3617 | set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags); |
656f30db FM |
3618 | break; |
3619 | case 0: | |
afcdd129 | 3620 | set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags); |
656f30db FM |
3621 | break; |
3622 | case 1: | |
afcdd129 | 3623 | set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags); |
656f30db FM |
3624 | break; |
3625 | default: | |
3626 | BUG(); /* unexpected, logic error */ | |
3627 | } | |
3628 | } | |
3629 | ||
4246a0b6 | 3630 | static void end_bio_extent_buffer_writepage(struct bio *bio) |
0b32f4bb | 3631 | { |
2c30c71b | 3632 | struct bio_vec *bvec; |
0b32f4bb | 3633 | struct extent_buffer *eb; |
2c30c71b | 3634 | int i, done; |
0b32f4bb | 3635 | |
c09abff8 | 3636 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2c30c71b | 3637 | bio_for_each_segment_all(bvec, bio, i) { |
0b32f4bb JB |
3638 | struct page *page = bvec->bv_page; |
3639 | ||
0b32f4bb JB |
3640 | eb = (struct extent_buffer *)page->private; |
3641 | BUG_ON(!eb); | |
3642 | done = atomic_dec_and_test(&eb->io_pages); | |
3643 | ||
4e4cbee9 | 3644 | if (bio->bi_status || |
4246a0b6 | 3645 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { |
0b32f4bb | 3646 | ClearPageUptodate(page); |
656f30db | 3647 | set_btree_ioerr(page); |
0b32f4bb JB |
3648 | } |
3649 | ||
3650 | end_page_writeback(page); | |
3651 | ||
3652 | if (!done) | |
3653 | continue; | |
3654 | ||
3655 | end_extent_buffer_writeback(eb); | |
2c30c71b | 3656 | } |
0b32f4bb JB |
3657 | |
3658 | bio_put(bio); | |
0b32f4bb JB |
3659 | } |
3660 | ||
0e378df1 | 3661 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb JB |
3662 | struct btrfs_fs_info *fs_info, |
3663 | struct writeback_control *wbc, | |
3664 | struct extent_page_data *epd) | |
3665 | { | |
3666 | struct block_device *bdev = fs_info->fs_devices->latest_bdev; | |
f28491e0 | 3667 | struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree; |
0b32f4bb | 3668 | u64 offset = eb->start; |
851cd173 | 3669 | u32 nritems; |
cc5e31a4 | 3670 | int i, num_pages; |
851cd173 | 3671 | unsigned long start, end; |
ff40adf7 | 3672 | unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META; |
d7dbe9e7 | 3673 | int ret = 0; |
0b32f4bb | 3674 | |
656f30db | 3675 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); |
65ad0104 | 3676 | num_pages = num_extent_pages(eb); |
0b32f4bb | 3677 | atomic_set(&eb->io_pages, num_pages); |
de0022b9 | 3678 | |
851cd173 LB |
3679 | /* set btree blocks beyond nritems with 0 to avoid stale content. */ |
3680 | nritems = btrfs_header_nritems(eb); | |
3eb548ee | 3681 | if (btrfs_header_level(eb) > 0) { |
3eb548ee LB |
3682 | end = btrfs_node_key_ptr_offset(nritems); |
3683 | ||
b159fa28 | 3684 | memzero_extent_buffer(eb, end, eb->len - end); |
851cd173 LB |
3685 | } else { |
3686 | /* | |
3687 | * leaf: | |
3688 | * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0 | |
3689 | */ | |
3690 | start = btrfs_item_nr_offset(nritems); | |
3d9ec8c4 | 3691 | end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, eb); |
b159fa28 | 3692 | memzero_extent_buffer(eb, start, end - start); |
3eb548ee LB |
3693 | } |
3694 | ||
0b32f4bb | 3695 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3696 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3697 | |
3698 | clear_page_dirty_for_io(p); | |
3699 | set_page_writeback(p); | |
4b81ba48 | 3700 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc, |
6273b7f8 | 3701 | p, offset, PAGE_SIZE, 0, bdev, |
c2df8bb4 | 3702 | &epd->bio, |
1f7ad75b | 3703 | end_bio_extent_buffer_writepage, |
18fdc679 | 3704 | 0, 0, 0, false); |
0b32f4bb | 3705 | if (ret) { |
656f30db | 3706 | set_btree_ioerr(p); |
fe01aa65 TK |
3707 | if (PageWriteback(p)) |
3708 | end_page_writeback(p); | |
0b32f4bb JB |
3709 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
3710 | end_extent_buffer_writeback(eb); | |
3711 | ret = -EIO; | |
3712 | break; | |
3713 | } | |
09cbfeaf | 3714 | offset += PAGE_SIZE; |
3d4b9496 | 3715 | update_nr_written(wbc, 1); |
0b32f4bb JB |
3716 | unlock_page(p); |
3717 | } | |
3718 | ||
3719 | if (unlikely(ret)) { | |
3720 | for (; i < num_pages; i++) { | |
bbf65cf0 | 3721 | struct page *p = eb->pages[i]; |
81465028 | 3722 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
3723 | unlock_page(p); |
3724 | } | |
3725 | } | |
3726 | ||
3727 | return ret; | |
3728 | } | |
3729 | ||
3730 | int btree_write_cache_pages(struct address_space *mapping, | |
3731 | struct writeback_control *wbc) | |
3732 | { | |
3733 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; | |
3734 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; | |
3735 | struct extent_buffer *eb, *prev_eb = NULL; | |
3736 | struct extent_page_data epd = { | |
3737 | .bio = NULL, | |
3738 | .tree = tree, | |
3739 | .extent_locked = 0, | |
3740 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
3741 | }; | |
3742 | int ret = 0; | |
3743 | int done = 0; | |
3744 | int nr_to_write_done = 0; | |
3745 | struct pagevec pvec; | |
3746 | int nr_pages; | |
3747 | pgoff_t index; | |
3748 | pgoff_t end; /* Inclusive */ | |
3749 | int scanned = 0; | |
10bbd235 | 3750 | xa_mark_t tag; |
0b32f4bb | 3751 | |
86679820 | 3752 | pagevec_init(&pvec); |
0b32f4bb JB |
3753 | if (wbc->range_cyclic) { |
3754 | index = mapping->writeback_index; /* Start from prev offset */ | |
3755 | end = -1; | |
3756 | } else { | |
09cbfeaf KS |
3757 | index = wbc->range_start >> PAGE_SHIFT; |
3758 | end = wbc->range_end >> PAGE_SHIFT; | |
0b32f4bb JB |
3759 | scanned = 1; |
3760 | } | |
3761 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3762 | tag = PAGECACHE_TAG_TOWRITE; | |
3763 | else | |
3764 | tag = PAGECACHE_TAG_DIRTY; | |
3765 | retry: | |
3766 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3767 | tag_pages_for_writeback(mapping, index, end); | |
3768 | while (!done && !nr_to_write_done && (index <= end) && | |
4006f437 | 3769 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 3770 | tag))) { |
0b32f4bb JB |
3771 | unsigned i; |
3772 | ||
3773 | scanned = 1; | |
3774 | for (i = 0; i < nr_pages; i++) { | |
3775 | struct page *page = pvec.pages[i]; | |
3776 | ||
3777 | if (!PagePrivate(page)) | |
3778 | continue; | |
3779 | ||
b5bae261 JB |
3780 | spin_lock(&mapping->private_lock); |
3781 | if (!PagePrivate(page)) { | |
3782 | spin_unlock(&mapping->private_lock); | |
3783 | continue; | |
3784 | } | |
3785 | ||
0b32f4bb | 3786 | eb = (struct extent_buffer *)page->private; |
b5bae261 JB |
3787 | |
3788 | /* | |
3789 | * Shouldn't happen and normally this would be a BUG_ON | |
3790 | * but no sense in crashing the users box for something | |
3791 | * we can survive anyway. | |
3792 | */ | |
fae7f21c | 3793 | if (WARN_ON(!eb)) { |
b5bae261 | 3794 | spin_unlock(&mapping->private_lock); |
0b32f4bb JB |
3795 | continue; |
3796 | } | |
3797 | ||
b5bae261 JB |
3798 | if (eb == prev_eb) { |
3799 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 3800 | continue; |
b5bae261 | 3801 | } |
0b32f4bb | 3802 | |
b5bae261 JB |
3803 | ret = atomic_inc_not_zero(&eb->refs); |
3804 | spin_unlock(&mapping->private_lock); | |
3805 | if (!ret) | |
0b32f4bb | 3806 | continue; |
0b32f4bb JB |
3807 | |
3808 | prev_eb = eb; | |
3809 | ret = lock_extent_buffer_for_io(eb, fs_info, &epd); | |
3810 | if (!ret) { | |
3811 | free_extent_buffer(eb); | |
3812 | continue; | |
3813 | } | |
3814 | ||
3815 | ret = write_one_eb(eb, fs_info, wbc, &epd); | |
3816 | if (ret) { | |
3817 | done = 1; | |
3818 | free_extent_buffer(eb); | |
3819 | break; | |
3820 | } | |
3821 | free_extent_buffer(eb); | |
3822 | ||
3823 | /* | |
3824 | * the filesystem may choose to bump up nr_to_write. | |
3825 | * We have to make sure to honor the new nr_to_write | |
3826 | * at any time | |
3827 | */ | |
3828 | nr_to_write_done = wbc->nr_to_write <= 0; | |
3829 | } | |
3830 | pagevec_release(&pvec); | |
3831 | cond_resched(); | |
3832 | } | |
3833 | if (!scanned && !done) { | |
3834 | /* | |
3835 | * We hit the last page and there is more work to be done: wrap | |
3836 | * back to the start of the file | |
3837 | */ | |
3838 | scanned = 1; | |
3839 | index = 0; | |
3840 | goto retry; | |
3841 | } | |
3842 | flush_write_bio(&epd); | |
3843 | return ret; | |
3844 | } | |
3845 | ||
d1310b2e | 3846 | /** |
4bef0848 | 3847 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
3848 | * @mapping: address space structure to write |
3849 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
935db853 | 3850 | * @data: data passed to __extent_writepage function |
d1310b2e CM |
3851 | * |
3852 | * If a page is already under I/O, write_cache_pages() skips it, even | |
3853 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
3854 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
3855 | * and msync() need to guarantee that all the data which was dirty at the time | |
3856 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
3857 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
3858 | * existing IO to complete. | |
3859 | */ | |
4242b64a | 3860 | static int extent_write_cache_pages(struct address_space *mapping, |
4bef0848 | 3861 | struct writeback_control *wbc, |
aab6e9ed | 3862 | struct extent_page_data *epd) |
d1310b2e | 3863 | { |
7fd1a3f7 | 3864 | struct inode *inode = mapping->host; |
d1310b2e CM |
3865 | int ret = 0; |
3866 | int done = 0; | |
f85d7d6c | 3867 | int nr_to_write_done = 0; |
d1310b2e CM |
3868 | struct pagevec pvec; |
3869 | int nr_pages; | |
3870 | pgoff_t index; | |
3871 | pgoff_t end; /* Inclusive */ | |
a9132667 LB |
3872 | pgoff_t done_index; |
3873 | int range_whole = 0; | |
d1310b2e | 3874 | int scanned = 0; |
10bbd235 | 3875 | xa_mark_t tag; |
d1310b2e | 3876 | |
7fd1a3f7 JB |
3877 | /* |
3878 | * We have to hold onto the inode so that ordered extents can do their | |
3879 | * work when the IO finishes. The alternative to this is failing to add | |
3880 | * an ordered extent if the igrab() fails there and that is a huge pain | |
3881 | * to deal with, so instead just hold onto the inode throughout the | |
3882 | * writepages operation. If it fails here we are freeing up the inode | |
3883 | * anyway and we'd rather not waste our time writing out stuff that is | |
3884 | * going to be truncated anyway. | |
3885 | */ | |
3886 | if (!igrab(inode)) | |
3887 | return 0; | |
3888 | ||
86679820 | 3889 | pagevec_init(&pvec); |
d1310b2e CM |
3890 | if (wbc->range_cyclic) { |
3891 | index = mapping->writeback_index; /* Start from prev offset */ | |
3892 | end = -1; | |
3893 | } else { | |
09cbfeaf KS |
3894 | index = wbc->range_start >> PAGE_SHIFT; |
3895 | end = wbc->range_end >> PAGE_SHIFT; | |
a9132667 LB |
3896 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
3897 | range_whole = 1; | |
d1310b2e CM |
3898 | scanned = 1; |
3899 | } | |
3cd24c69 EL |
3900 | |
3901 | /* | |
3902 | * We do the tagged writepage as long as the snapshot flush bit is set | |
3903 | * and we are the first one who do the filemap_flush() on this inode. | |
3904 | * | |
3905 | * The nr_to_write == LONG_MAX is needed to make sure other flushers do | |
3906 | * not race in and drop the bit. | |
3907 | */ | |
3908 | if (range_whole && wbc->nr_to_write == LONG_MAX && | |
3909 | test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
3910 | &BTRFS_I(inode)->runtime_flags)) | |
3911 | wbc->tagged_writepages = 1; | |
3912 | ||
3913 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) | |
f7aaa06b JB |
3914 | tag = PAGECACHE_TAG_TOWRITE; |
3915 | else | |
3916 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 3917 | retry: |
3cd24c69 | 3918 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
f7aaa06b | 3919 | tag_pages_for_writeback(mapping, index, end); |
a9132667 | 3920 | done_index = index; |
f85d7d6c | 3921 | while (!done && !nr_to_write_done && (index <= end) && |
67fd707f JK |
3922 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, |
3923 | &index, end, tag))) { | |
d1310b2e CM |
3924 | unsigned i; |
3925 | ||
3926 | scanned = 1; | |
3927 | for (i = 0; i < nr_pages; i++) { | |
3928 | struct page *page = pvec.pages[i]; | |
3929 | ||
a9132667 | 3930 | done_index = page->index; |
d1310b2e | 3931 | /* |
b93b0163 MW |
3932 | * At this point we hold neither the i_pages lock nor |
3933 | * the page lock: the page may be truncated or | |
3934 | * invalidated (changing page->mapping to NULL), | |
3935 | * or even swizzled back from swapper_space to | |
3936 | * tmpfs file mapping | |
d1310b2e | 3937 | */ |
c8f2f24b | 3938 | if (!trylock_page(page)) { |
aab6e9ed | 3939 | flush_write_bio(epd); |
c8f2f24b | 3940 | lock_page(page); |
01d658f2 | 3941 | } |
d1310b2e CM |
3942 | |
3943 | if (unlikely(page->mapping != mapping)) { | |
3944 | unlock_page(page); | |
3945 | continue; | |
3946 | } | |
3947 | ||
d2c3f4f6 | 3948 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 | 3949 | if (PageWriteback(page)) |
aab6e9ed | 3950 | flush_write_bio(epd); |
d1310b2e | 3951 | wait_on_page_writeback(page); |
d2c3f4f6 | 3952 | } |
d1310b2e CM |
3953 | |
3954 | if (PageWriteback(page) || | |
3955 | !clear_page_dirty_for_io(page)) { | |
3956 | unlock_page(page); | |
3957 | continue; | |
3958 | } | |
3959 | ||
aab6e9ed | 3960 | ret = __extent_writepage(page, wbc, epd); |
d1310b2e CM |
3961 | |
3962 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
3963 | unlock_page(page); | |
3964 | ret = 0; | |
3965 | } | |
a9132667 LB |
3966 | if (ret < 0) { |
3967 | /* | |
3968 | * done_index is set past this page, | |
3969 | * so media errors will not choke | |
3970 | * background writeout for the entire | |
3971 | * file. This has consequences for | |
3972 | * range_cyclic semantics (ie. it may | |
3973 | * not be suitable for data integrity | |
3974 | * writeout). | |
3975 | */ | |
3976 | done_index = page->index + 1; | |
3977 | done = 1; | |
3978 | break; | |
3979 | } | |
f85d7d6c CM |
3980 | |
3981 | /* | |
3982 | * the filesystem may choose to bump up nr_to_write. | |
3983 | * We have to make sure to honor the new nr_to_write | |
3984 | * at any time | |
3985 | */ | |
3986 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
3987 | } |
3988 | pagevec_release(&pvec); | |
3989 | cond_resched(); | |
3990 | } | |
894b36e3 | 3991 | if (!scanned && !done) { |
d1310b2e CM |
3992 | /* |
3993 | * We hit the last page and there is more work to be done: wrap | |
3994 | * back to the start of the file | |
3995 | */ | |
3996 | scanned = 1; | |
3997 | index = 0; | |
3998 | goto retry; | |
3999 | } | |
a9132667 LB |
4000 | |
4001 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole)) | |
4002 | mapping->writeback_index = done_index; | |
4003 | ||
7fd1a3f7 | 4004 | btrfs_add_delayed_iput(inode); |
894b36e3 | 4005 | return ret; |
d1310b2e | 4006 | } |
d1310b2e | 4007 | |
aab6e9ed | 4008 | static void flush_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 4009 | { |
d2c3f4f6 | 4010 | if (epd->bio) { |
355808c2 JM |
4011 | int ret; |
4012 | ||
18fdc679 | 4013 | ret = submit_one_bio(epd->bio, 0, 0); |
79787eaa | 4014 | BUG_ON(ret < 0); /* -ENOMEM */ |
d2c3f4f6 CM |
4015 | epd->bio = NULL; |
4016 | } | |
4017 | } | |
4018 | ||
0a9b0e53 | 4019 | int extent_write_full_page(struct page *page, struct writeback_control *wbc) |
d1310b2e CM |
4020 | { |
4021 | int ret; | |
d1310b2e CM |
4022 | struct extent_page_data epd = { |
4023 | .bio = NULL, | |
0a9b0e53 | 4024 | .tree = &BTRFS_I(page->mapping->host)->io_tree, |
771ed689 | 4025 | .extent_locked = 0, |
ffbd517d | 4026 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e | 4027 | }; |
d1310b2e | 4028 | |
d1310b2e CM |
4029 | ret = __extent_writepage(page, wbc, &epd); |
4030 | ||
e2932ee0 | 4031 | flush_write_bio(&epd); |
d1310b2e CM |
4032 | return ret; |
4033 | } | |
d1310b2e | 4034 | |
5e3ee236 | 4035 | int extent_write_locked_range(struct inode *inode, u64 start, u64 end, |
771ed689 CM |
4036 | int mode) |
4037 | { | |
4038 | int ret = 0; | |
4039 | struct address_space *mapping = inode->i_mapping; | |
5e3ee236 | 4040 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
771ed689 | 4041 | struct page *page; |
09cbfeaf KS |
4042 | unsigned long nr_pages = (end - start + PAGE_SIZE) >> |
4043 | PAGE_SHIFT; | |
771ed689 CM |
4044 | |
4045 | struct extent_page_data epd = { | |
4046 | .bio = NULL, | |
4047 | .tree = tree, | |
771ed689 | 4048 | .extent_locked = 1, |
ffbd517d | 4049 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
4050 | }; |
4051 | struct writeback_control wbc_writepages = { | |
771ed689 | 4052 | .sync_mode = mode, |
771ed689 CM |
4053 | .nr_to_write = nr_pages * 2, |
4054 | .range_start = start, | |
4055 | .range_end = end + 1, | |
4056 | }; | |
4057 | ||
d397712b | 4058 | while (start <= end) { |
09cbfeaf | 4059 | page = find_get_page(mapping, start >> PAGE_SHIFT); |
771ed689 CM |
4060 | if (clear_page_dirty_for_io(page)) |
4061 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
4062 | else { | |
7087a9d8 | 4063 | btrfs_writepage_endio_finish_ordered(page, start, |
c629732d | 4064 | start + PAGE_SIZE - 1, 1); |
771ed689 CM |
4065 | unlock_page(page); |
4066 | } | |
09cbfeaf KS |
4067 | put_page(page); |
4068 | start += PAGE_SIZE; | |
771ed689 CM |
4069 | } |
4070 | ||
e2932ee0 | 4071 | flush_write_bio(&epd); |
771ed689 CM |
4072 | return ret; |
4073 | } | |
d1310b2e | 4074 | |
8ae225a8 | 4075 | int extent_writepages(struct address_space *mapping, |
d1310b2e CM |
4076 | struct writeback_control *wbc) |
4077 | { | |
4078 | int ret = 0; | |
4079 | struct extent_page_data epd = { | |
4080 | .bio = NULL, | |
8ae225a8 | 4081 | .tree = &BTRFS_I(mapping->host)->io_tree, |
771ed689 | 4082 | .extent_locked = 0, |
ffbd517d | 4083 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
4084 | }; |
4085 | ||
935db853 | 4086 | ret = extent_write_cache_pages(mapping, wbc, &epd); |
e2932ee0 | 4087 | flush_write_bio(&epd); |
d1310b2e CM |
4088 | return ret; |
4089 | } | |
d1310b2e | 4090 | |
2a3ff0ad NB |
4091 | int extent_readpages(struct address_space *mapping, struct list_head *pages, |
4092 | unsigned nr_pages) | |
d1310b2e CM |
4093 | { |
4094 | struct bio *bio = NULL; | |
c8b97818 | 4095 | unsigned long bio_flags = 0; |
67c9684f | 4096 | struct page *pagepool[16]; |
125bac01 | 4097 | struct extent_map *em_cached = NULL; |
2a3ff0ad | 4098 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; |
67c9684f | 4099 | int nr = 0; |
808f80b4 | 4100 | u64 prev_em_start = (u64)-1; |
d1310b2e | 4101 | |
61ed3a14 NB |
4102 | while (!list_empty(pages)) { |
4103 | for (nr = 0; nr < ARRAY_SIZE(pagepool) && !list_empty(pages);) { | |
f86196ea | 4104 | struct page *page = lru_to_page(pages); |
d1310b2e | 4105 | |
61ed3a14 NB |
4106 | prefetchw(&page->flags); |
4107 | list_del(&page->lru); | |
4108 | if (add_to_page_cache_lru(page, mapping, page->index, | |
4109 | readahead_gfp_mask(mapping))) { | |
4110 | put_page(page); | |
4111 | continue; | |
4112 | } | |
4113 | ||
4114 | pagepool[nr++] = page; | |
d1310b2e | 4115 | } |
67c9684f | 4116 | |
e4d17ef5 | 4117 | __extent_readpages(tree, pagepool, nr, &em_cached, &bio, |
61ed3a14 | 4118 | &bio_flags, &prev_em_start); |
d1310b2e | 4119 | } |
67c9684f | 4120 | |
125bac01 MX |
4121 | if (em_cached) |
4122 | free_extent_map(em_cached); | |
4123 | ||
d1310b2e | 4124 | if (bio) |
1f7ad75b | 4125 | return submit_one_bio(bio, 0, bio_flags); |
d1310b2e CM |
4126 | return 0; |
4127 | } | |
d1310b2e CM |
4128 | |
4129 | /* | |
4130 | * basic invalidatepage code, this waits on any locked or writeback | |
4131 | * ranges corresponding to the page, and then deletes any extent state | |
4132 | * records from the tree | |
4133 | */ | |
4134 | int extent_invalidatepage(struct extent_io_tree *tree, | |
4135 | struct page *page, unsigned long offset) | |
4136 | { | |
2ac55d41 | 4137 | struct extent_state *cached_state = NULL; |
4eee4fa4 | 4138 | u64 start = page_offset(page); |
09cbfeaf | 4139 | u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
4140 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; |
4141 | ||
fda2832f | 4142 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
4143 | if (start > end) |
4144 | return 0; | |
4145 | ||
ff13db41 | 4146 | lock_extent_bits(tree, start, end, &cached_state); |
1edbb734 | 4147 | wait_on_page_writeback(page); |
d1310b2e | 4148 | clear_extent_bit(tree, start, end, |
32c00aff JB |
4149 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
4150 | EXTENT_DO_ACCOUNTING, | |
ae0f1625 | 4151 | 1, 1, &cached_state); |
d1310b2e CM |
4152 | return 0; |
4153 | } | |
d1310b2e | 4154 | |
7b13b7b1 CM |
4155 | /* |
4156 | * a helper for releasepage, this tests for areas of the page that | |
4157 | * are locked or under IO and drops the related state bits if it is safe | |
4158 | * to drop the page. | |
4159 | */ | |
29c68b2d | 4160 | static int try_release_extent_state(struct extent_io_tree *tree, |
48a3b636 | 4161 | struct page *page, gfp_t mask) |
7b13b7b1 | 4162 | { |
4eee4fa4 | 4163 | u64 start = page_offset(page); |
09cbfeaf | 4164 | u64 end = start + PAGE_SIZE - 1; |
7b13b7b1 CM |
4165 | int ret = 1; |
4166 | ||
211f90e6 | 4167 | if (test_range_bit(tree, start, end, |
8b62b72b | 4168 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
4169 | ret = 0; |
4170 | else { | |
11ef160f CM |
4171 | /* |
4172 | * at this point we can safely clear everything except the | |
4173 | * locked bit and the nodatasum bit | |
4174 | */ | |
66b0c887 | 4175 | ret = __clear_extent_bit(tree, start, end, |
11ef160f | 4176 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
66b0c887 | 4177 | 0, 0, NULL, mask, NULL); |
e3f24cc5 CM |
4178 | |
4179 | /* if clear_extent_bit failed for enomem reasons, | |
4180 | * we can't allow the release to continue. | |
4181 | */ | |
4182 | if (ret < 0) | |
4183 | ret = 0; | |
4184 | else | |
4185 | ret = 1; | |
7b13b7b1 CM |
4186 | } |
4187 | return ret; | |
4188 | } | |
7b13b7b1 | 4189 | |
d1310b2e CM |
4190 | /* |
4191 | * a helper for releasepage. As long as there are no locked extents | |
4192 | * in the range corresponding to the page, both state records and extent | |
4193 | * map records are removed | |
4194 | */ | |
477a30ba | 4195 | int try_release_extent_mapping(struct page *page, gfp_t mask) |
d1310b2e CM |
4196 | { |
4197 | struct extent_map *em; | |
4eee4fa4 | 4198 | u64 start = page_offset(page); |
09cbfeaf | 4199 | u64 end = start + PAGE_SIZE - 1; |
bd3599a0 FM |
4200 | struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host); |
4201 | struct extent_io_tree *tree = &btrfs_inode->io_tree; | |
4202 | struct extent_map_tree *map = &btrfs_inode->extent_tree; | |
7b13b7b1 | 4203 | |
d0164adc | 4204 | if (gfpflags_allow_blocking(mask) && |
ee22184b | 4205 | page->mapping->host->i_size > SZ_16M) { |
39b5637f | 4206 | u64 len; |
70dec807 | 4207 | while (start <= end) { |
39b5637f | 4208 | len = end - start + 1; |
890871be | 4209 | write_lock(&map->lock); |
39b5637f | 4210 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 4211 | if (!em) { |
890871be | 4212 | write_unlock(&map->lock); |
70dec807 CM |
4213 | break; |
4214 | } | |
7f3c74fb CM |
4215 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
4216 | em->start != start) { | |
890871be | 4217 | write_unlock(&map->lock); |
70dec807 CM |
4218 | free_extent_map(em); |
4219 | break; | |
4220 | } | |
4221 | if (!test_range_bit(tree, em->start, | |
4222 | extent_map_end(em) - 1, | |
8b62b72b | 4223 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 4224 | 0, NULL)) { |
bd3599a0 FM |
4225 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
4226 | &btrfs_inode->runtime_flags); | |
70dec807 CM |
4227 | remove_extent_mapping(map, em); |
4228 | /* once for the rb tree */ | |
4229 | free_extent_map(em); | |
4230 | } | |
4231 | start = extent_map_end(em); | |
890871be | 4232 | write_unlock(&map->lock); |
70dec807 CM |
4233 | |
4234 | /* once for us */ | |
d1310b2e CM |
4235 | free_extent_map(em); |
4236 | } | |
d1310b2e | 4237 | } |
29c68b2d | 4238 | return try_release_extent_state(tree, page, mask); |
d1310b2e | 4239 | } |
d1310b2e | 4240 | |
ec29ed5b CM |
4241 | /* |
4242 | * helper function for fiemap, which doesn't want to see any holes. | |
4243 | * This maps until we find something past 'last' | |
4244 | */ | |
4245 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
e3350e16 | 4246 | u64 offset, u64 last) |
ec29ed5b | 4247 | { |
da17066c | 4248 | u64 sectorsize = btrfs_inode_sectorsize(inode); |
ec29ed5b CM |
4249 | struct extent_map *em; |
4250 | u64 len; | |
4251 | ||
4252 | if (offset >= last) | |
4253 | return NULL; | |
4254 | ||
67871254 | 4255 | while (1) { |
ec29ed5b CM |
4256 | len = last - offset; |
4257 | if (len == 0) | |
4258 | break; | |
fda2832f | 4259 | len = ALIGN(len, sectorsize); |
4ab47a8d | 4260 | em = btrfs_get_extent_fiemap(BTRFS_I(inode), offset, len); |
c704005d | 4261 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
4262 | return em; |
4263 | ||
4264 | /* if this isn't a hole return it */ | |
4a2d25cd | 4265 | if (em->block_start != EXTENT_MAP_HOLE) |
ec29ed5b | 4266 | return em; |
ec29ed5b CM |
4267 | |
4268 | /* this is a hole, advance to the next extent */ | |
4269 | offset = extent_map_end(em); | |
4270 | free_extent_map(em); | |
4271 | if (offset >= last) | |
4272 | break; | |
4273 | } | |
4274 | return NULL; | |
4275 | } | |
4276 | ||
4751832d QW |
4277 | /* |
4278 | * To cache previous fiemap extent | |
4279 | * | |
4280 | * Will be used for merging fiemap extent | |
4281 | */ | |
4282 | struct fiemap_cache { | |
4283 | u64 offset; | |
4284 | u64 phys; | |
4285 | u64 len; | |
4286 | u32 flags; | |
4287 | bool cached; | |
4288 | }; | |
4289 | ||
4290 | /* | |
4291 | * Helper to submit fiemap extent. | |
4292 | * | |
4293 | * Will try to merge current fiemap extent specified by @offset, @phys, | |
4294 | * @len and @flags with cached one. | |
4295 | * And only when we fails to merge, cached one will be submitted as | |
4296 | * fiemap extent. | |
4297 | * | |
4298 | * Return value is the same as fiemap_fill_next_extent(). | |
4299 | */ | |
4300 | static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, | |
4301 | struct fiemap_cache *cache, | |
4302 | u64 offset, u64 phys, u64 len, u32 flags) | |
4303 | { | |
4304 | int ret = 0; | |
4305 | ||
4306 | if (!cache->cached) | |
4307 | goto assign; | |
4308 | ||
4309 | /* | |
4310 | * Sanity check, extent_fiemap() should have ensured that new | |
52042d8e | 4311 | * fiemap extent won't overlap with cached one. |
4751832d QW |
4312 | * Not recoverable. |
4313 | * | |
4314 | * NOTE: Physical address can overlap, due to compression | |
4315 | */ | |
4316 | if (cache->offset + cache->len > offset) { | |
4317 | WARN_ON(1); | |
4318 | return -EINVAL; | |
4319 | } | |
4320 | ||
4321 | /* | |
4322 | * Only merges fiemap extents if | |
4323 | * 1) Their logical addresses are continuous | |
4324 | * | |
4325 | * 2) Their physical addresses are continuous | |
4326 | * So truly compressed (physical size smaller than logical size) | |
4327 | * extents won't get merged with each other | |
4328 | * | |
4329 | * 3) Share same flags except FIEMAP_EXTENT_LAST | |
4330 | * So regular extent won't get merged with prealloc extent | |
4331 | */ | |
4332 | if (cache->offset + cache->len == offset && | |
4333 | cache->phys + cache->len == phys && | |
4334 | (cache->flags & ~FIEMAP_EXTENT_LAST) == | |
4335 | (flags & ~FIEMAP_EXTENT_LAST)) { | |
4336 | cache->len += len; | |
4337 | cache->flags |= flags; | |
4338 | goto try_submit_last; | |
4339 | } | |
4340 | ||
4341 | /* Not mergeable, need to submit cached one */ | |
4342 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, | |
4343 | cache->len, cache->flags); | |
4344 | cache->cached = false; | |
4345 | if (ret) | |
4346 | return ret; | |
4347 | assign: | |
4348 | cache->cached = true; | |
4349 | cache->offset = offset; | |
4350 | cache->phys = phys; | |
4351 | cache->len = len; | |
4352 | cache->flags = flags; | |
4353 | try_submit_last: | |
4354 | if (cache->flags & FIEMAP_EXTENT_LAST) { | |
4355 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, | |
4356 | cache->phys, cache->len, cache->flags); | |
4357 | cache->cached = false; | |
4358 | } | |
4359 | return ret; | |
4360 | } | |
4361 | ||
4362 | /* | |
848c23b7 | 4363 | * Emit last fiemap cache |
4751832d | 4364 | * |
848c23b7 QW |
4365 | * The last fiemap cache may still be cached in the following case: |
4366 | * 0 4k 8k | |
4367 | * |<- Fiemap range ->| | |
4368 | * |<------------ First extent ----------->| | |
4369 | * | |
4370 | * In this case, the first extent range will be cached but not emitted. | |
4371 | * So we must emit it before ending extent_fiemap(). | |
4751832d | 4372 | */ |
848c23b7 QW |
4373 | static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info, |
4374 | struct fiemap_extent_info *fieinfo, | |
4375 | struct fiemap_cache *cache) | |
4751832d QW |
4376 | { |
4377 | int ret; | |
4378 | ||
4379 | if (!cache->cached) | |
4380 | return 0; | |
4381 | ||
4751832d QW |
4382 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, |
4383 | cache->len, cache->flags); | |
4384 | cache->cached = false; | |
4385 | if (ret > 0) | |
4386 | ret = 0; | |
4387 | return ret; | |
4388 | } | |
4389 | ||
1506fcc8 | 4390 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
2135fb9b | 4391 | __u64 start, __u64 len) |
1506fcc8 | 4392 | { |
975f84fe | 4393 | int ret = 0; |
1506fcc8 YS |
4394 | u64 off = start; |
4395 | u64 max = start + len; | |
4396 | u32 flags = 0; | |
975f84fe JB |
4397 | u32 found_type; |
4398 | u64 last; | |
ec29ed5b | 4399 | u64 last_for_get_extent = 0; |
1506fcc8 | 4400 | u64 disko = 0; |
ec29ed5b | 4401 | u64 isize = i_size_read(inode); |
975f84fe | 4402 | struct btrfs_key found_key; |
1506fcc8 | 4403 | struct extent_map *em = NULL; |
2ac55d41 | 4404 | struct extent_state *cached_state = NULL; |
975f84fe | 4405 | struct btrfs_path *path; |
dc046b10 | 4406 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4751832d | 4407 | struct fiemap_cache cache = { 0 }; |
1506fcc8 | 4408 | int end = 0; |
ec29ed5b CM |
4409 | u64 em_start = 0; |
4410 | u64 em_len = 0; | |
4411 | u64 em_end = 0; | |
1506fcc8 YS |
4412 | |
4413 | if (len == 0) | |
4414 | return -EINVAL; | |
4415 | ||
975f84fe JB |
4416 | path = btrfs_alloc_path(); |
4417 | if (!path) | |
4418 | return -ENOMEM; | |
4419 | path->leave_spinning = 1; | |
4420 | ||
da17066c JM |
4421 | start = round_down(start, btrfs_inode_sectorsize(inode)); |
4422 | len = round_up(max, btrfs_inode_sectorsize(inode)) - start; | |
4d479cf0 | 4423 | |
ec29ed5b CM |
4424 | /* |
4425 | * lookup the last file extent. We're not using i_size here | |
4426 | * because there might be preallocation past i_size | |
4427 | */ | |
f85b7379 DS |
4428 | ret = btrfs_lookup_file_extent(NULL, root, path, |
4429 | btrfs_ino(BTRFS_I(inode)), -1, 0); | |
975f84fe JB |
4430 | if (ret < 0) { |
4431 | btrfs_free_path(path); | |
4432 | return ret; | |
2d324f59 LB |
4433 | } else { |
4434 | WARN_ON(!ret); | |
4435 | if (ret == 1) | |
4436 | ret = 0; | |
975f84fe | 4437 | } |
2d324f59 | 4438 | |
975f84fe | 4439 | path->slots[0]--; |
975f84fe | 4440 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
962a298f | 4441 | found_type = found_key.type; |
975f84fe | 4442 | |
ec29ed5b | 4443 | /* No extents, but there might be delalloc bits */ |
4a0cc7ca | 4444 | if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) || |
975f84fe | 4445 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
4446 | /* have to trust i_size as the end */ |
4447 | last = (u64)-1; | |
4448 | last_for_get_extent = isize; | |
4449 | } else { | |
4450 | /* | |
4451 | * remember the start of the last extent. There are a | |
4452 | * bunch of different factors that go into the length of the | |
4453 | * extent, so its much less complex to remember where it started | |
4454 | */ | |
4455 | last = found_key.offset; | |
4456 | last_for_get_extent = last + 1; | |
975f84fe | 4457 | } |
fe09e16c | 4458 | btrfs_release_path(path); |
975f84fe | 4459 | |
ec29ed5b CM |
4460 | /* |
4461 | * we might have some extents allocated but more delalloc past those | |
4462 | * extents. so, we trust isize unless the start of the last extent is | |
4463 | * beyond isize | |
4464 | */ | |
4465 | if (last < isize) { | |
4466 | last = (u64)-1; | |
4467 | last_for_get_extent = isize; | |
4468 | } | |
4469 | ||
ff13db41 | 4470 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
d0082371 | 4471 | &cached_state); |
ec29ed5b | 4472 | |
e3350e16 | 4473 | em = get_extent_skip_holes(inode, start, last_for_get_extent); |
1506fcc8 YS |
4474 | if (!em) |
4475 | goto out; | |
4476 | if (IS_ERR(em)) { | |
4477 | ret = PTR_ERR(em); | |
4478 | goto out; | |
4479 | } | |
975f84fe | 4480 | |
1506fcc8 | 4481 | while (!end) { |
b76bb701 | 4482 | u64 offset_in_extent = 0; |
ea8efc74 CM |
4483 | |
4484 | /* break if the extent we found is outside the range */ | |
4485 | if (em->start >= max || extent_map_end(em) < off) | |
4486 | break; | |
4487 | ||
4488 | /* | |
4489 | * get_extent may return an extent that starts before our | |
4490 | * requested range. We have to make sure the ranges | |
4491 | * we return to fiemap always move forward and don't | |
4492 | * overlap, so adjust the offsets here | |
4493 | */ | |
4494 | em_start = max(em->start, off); | |
1506fcc8 | 4495 | |
ea8efc74 CM |
4496 | /* |
4497 | * record the offset from the start of the extent | |
b76bb701 JB |
4498 | * for adjusting the disk offset below. Only do this if the |
4499 | * extent isn't compressed since our in ram offset may be past | |
4500 | * what we have actually allocated on disk. | |
ea8efc74 | 4501 | */ |
b76bb701 JB |
4502 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
4503 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 4504 | em_end = extent_map_end(em); |
ea8efc74 | 4505 | em_len = em_end - em_start; |
1506fcc8 | 4506 | flags = 0; |
f0986318 FM |
4507 | if (em->block_start < EXTENT_MAP_LAST_BYTE) |
4508 | disko = em->block_start + offset_in_extent; | |
4509 | else | |
4510 | disko = 0; | |
1506fcc8 | 4511 | |
ea8efc74 CM |
4512 | /* |
4513 | * bump off for our next call to get_extent | |
4514 | */ | |
4515 | off = extent_map_end(em); | |
4516 | if (off >= max) | |
4517 | end = 1; | |
4518 | ||
93dbfad7 | 4519 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
4520 | end = 1; |
4521 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 4522 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
4523 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
4524 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 4525 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
4526 | flags |= (FIEMAP_EXTENT_DELALLOC | |
4527 | FIEMAP_EXTENT_UNKNOWN); | |
dc046b10 JB |
4528 | } else if (fieinfo->fi_extents_max) { |
4529 | u64 bytenr = em->block_start - | |
4530 | (em->start - em->orig_start); | |
fe09e16c | 4531 | |
fe09e16c LB |
4532 | /* |
4533 | * As btrfs supports shared space, this information | |
4534 | * can be exported to userspace tools via | |
dc046b10 JB |
4535 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 |
4536 | * then we're just getting a count and we can skip the | |
4537 | * lookup stuff. | |
fe09e16c | 4538 | */ |
bb739cf0 EN |
4539 | ret = btrfs_check_shared(root, |
4540 | btrfs_ino(BTRFS_I(inode)), | |
4541 | bytenr); | |
dc046b10 | 4542 | if (ret < 0) |
fe09e16c | 4543 | goto out_free; |
dc046b10 | 4544 | if (ret) |
fe09e16c | 4545 | flags |= FIEMAP_EXTENT_SHARED; |
dc046b10 | 4546 | ret = 0; |
1506fcc8 YS |
4547 | } |
4548 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
4549 | flags |= FIEMAP_EXTENT_ENCODED; | |
0d2b2372 JB |
4550 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
4551 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
1506fcc8 | 4552 | |
1506fcc8 YS |
4553 | free_extent_map(em); |
4554 | em = NULL; | |
ec29ed5b CM |
4555 | if ((em_start >= last) || em_len == (u64)-1 || |
4556 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
4557 | flags |= FIEMAP_EXTENT_LAST; |
4558 | end = 1; | |
4559 | } | |
4560 | ||
ec29ed5b | 4561 | /* now scan forward to see if this is really the last extent. */ |
e3350e16 | 4562 | em = get_extent_skip_holes(inode, off, last_for_get_extent); |
ec29ed5b CM |
4563 | if (IS_ERR(em)) { |
4564 | ret = PTR_ERR(em); | |
4565 | goto out; | |
4566 | } | |
4567 | if (!em) { | |
975f84fe JB |
4568 | flags |= FIEMAP_EXTENT_LAST; |
4569 | end = 1; | |
4570 | } | |
4751832d QW |
4571 | ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko, |
4572 | em_len, flags); | |
26e726af CS |
4573 | if (ret) { |
4574 | if (ret == 1) | |
4575 | ret = 0; | |
ec29ed5b | 4576 | goto out_free; |
26e726af | 4577 | } |
1506fcc8 YS |
4578 | } |
4579 | out_free: | |
4751832d | 4580 | if (!ret) |
848c23b7 | 4581 | ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache); |
1506fcc8 YS |
4582 | free_extent_map(em); |
4583 | out: | |
fe09e16c | 4584 | btrfs_free_path(path); |
a52f4cd2 | 4585 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
e43bbe5e | 4586 | &cached_state); |
1506fcc8 YS |
4587 | return ret; |
4588 | } | |
4589 | ||
727011e0 CM |
4590 | static void __free_extent_buffer(struct extent_buffer *eb) |
4591 | { | |
6d49ba1b | 4592 | btrfs_leak_debug_del(&eb->leak_list); |
727011e0 CM |
4593 | kmem_cache_free(extent_buffer_cache, eb); |
4594 | } | |
4595 | ||
a26e8c9f | 4596 | int extent_buffer_under_io(struct extent_buffer *eb) |
db7f3436 JB |
4597 | { |
4598 | return (atomic_read(&eb->io_pages) || | |
4599 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4600 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4601 | } | |
4602 | ||
4603 | /* | |
55ac0139 | 4604 | * Release all pages attached to the extent buffer. |
db7f3436 | 4605 | */ |
55ac0139 | 4606 | static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb) |
db7f3436 | 4607 | { |
d64766fd NB |
4608 | int i; |
4609 | int num_pages; | |
b0132a3b | 4610 | int mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
db7f3436 JB |
4611 | |
4612 | BUG_ON(extent_buffer_under_io(eb)); | |
4613 | ||
d64766fd NB |
4614 | num_pages = num_extent_pages(eb); |
4615 | for (i = 0; i < num_pages; i++) { | |
4616 | struct page *page = eb->pages[i]; | |
db7f3436 | 4617 | |
5d2361db FL |
4618 | if (!page) |
4619 | continue; | |
4620 | if (mapped) | |
db7f3436 | 4621 | spin_lock(&page->mapping->private_lock); |
5d2361db FL |
4622 | /* |
4623 | * We do this since we'll remove the pages after we've | |
4624 | * removed the eb from the radix tree, so we could race | |
4625 | * and have this page now attached to the new eb. So | |
4626 | * only clear page_private if it's still connected to | |
4627 | * this eb. | |
4628 | */ | |
4629 | if (PagePrivate(page) && | |
4630 | page->private == (unsigned long)eb) { | |
4631 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4632 | BUG_ON(PageDirty(page)); | |
4633 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 4634 | /* |
5d2361db FL |
4635 | * We need to make sure we haven't be attached |
4636 | * to a new eb. | |
db7f3436 | 4637 | */ |
5d2361db FL |
4638 | ClearPagePrivate(page); |
4639 | set_page_private(page, 0); | |
4640 | /* One for the page private */ | |
09cbfeaf | 4641 | put_page(page); |
db7f3436 | 4642 | } |
5d2361db FL |
4643 | |
4644 | if (mapped) | |
4645 | spin_unlock(&page->mapping->private_lock); | |
4646 | ||
01327610 | 4647 | /* One for when we allocated the page */ |
09cbfeaf | 4648 | put_page(page); |
d64766fd | 4649 | } |
db7f3436 JB |
4650 | } |
4651 | ||
4652 | /* | |
4653 | * Helper for releasing the extent buffer. | |
4654 | */ | |
4655 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4656 | { | |
55ac0139 | 4657 | btrfs_release_extent_buffer_pages(eb); |
db7f3436 JB |
4658 | __free_extent_buffer(eb); |
4659 | } | |
4660 | ||
f28491e0 JB |
4661 | static struct extent_buffer * |
4662 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 4663 | unsigned long len) |
d1310b2e CM |
4664 | { |
4665 | struct extent_buffer *eb = NULL; | |
4666 | ||
d1b5c567 | 4667 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
4668 | eb->start = start; |
4669 | eb->len = len; | |
f28491e0 | 4670 | eb->fs_info = fs_info; |
815a51c7 | 4671 | eb->bflags = 0; |
bd681513 CM |
4672 | rwlock_init(&eb->lock); |
4673 | atomic_set(&eb->write_locks, 0); | |
4674 | atomic_set(&eb->read_locks, 0); | |
4675 | atomic_set(&eb->blocking_readers, 0); | |
4676 | atomic_set(&eb->blocking_writers, 0); | |
4677 | atomic_set(&eb->spinning_readers, 0); | |
4678 | atomic_set(&eb->spinning_writers, 0); | |
5b25f70f | 4679 | eb->lock_nested = 0; |
bd681513 CM |
4680 | init_waitqueue_head(&eb->write_lock_wq); |
4681 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 4682 | |
6d49ba1b ES |
4683 | btrfs_leak_debug_add(&eb->leak_list, &buffers); |
4684 | ||
3083ee2e | 4685 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 4686 | atomic_set(&eb->refs, 1); |
0b32f4bb | 4687 | atomic_set(&eb->io_pages, 0); |
727011e0 | 4688 | |
b8dae313 DS |
4689 | /* |
4690 | * Sanity checks, currently the maximum is 64k covered by 16x 4k pages | |
4691 | */ | |
4692 | BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE | |
4693 | > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
4694 | BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
d1310b2e CM |
4695 | |
4696 | return eb; | |
4697 | } | |
4698 | ||
815a51c7 JS |
4699 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) |
4700 | { | |
cc5e31a4 | 4701 | int i; |
815a51c7 JS |
4702 | struct page *p; |
4703 | struct extent_buffer *new; | |
cc5e31a4 | 4704 | int num_pages = num_extent_pages(src); |
815a51c7 | 4705 | |
3f556f78 | 4706 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
4707 | if (new == NULL) |
4708 | return NULL; | |
4709 | ||
4710 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4711 | p = alloc_page(GFP_NOFS); |
db7f3436 JB |
4712 | if (!p) { |
4713 | btrfs_release_extent_buffer(new); | |
4714 | return NULL; | |
4715 | } | |
815a51c7 JS |
4716 | attach_extent_buffer_page(new, p); |
4717 | WARN_ON(PageDirty(p)); | |
4718 | SetPageUptodate(p); | |
4719 | new->pages[i] = p; | |
fba1acf9 | 4720 | copy_page(page_address(p), page_address(src->pages[i])); |
815a51c7 JS |
4721 | } |
4722 | ||
815a51c7 | 4723 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); |
b0132a3b | 4724 | set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags); |
815a51c7 JS |
4725 | |
4726 | return new; | |
4727 | } | |
4728 | ||
0f331229 OS |
4729 | struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
4730 | u64 start, unsigned long len) | |
815a51c7 JS |
4731 | { |
4732 | struct extent_buffer *eb; | |
cc5e31a4 DS |
4733 | int num_pages; |
4734 | int i; | |
815a51c7 | 4735 | |
3f556f78 | 4736 | eb = __alloc_extent_buffer(fs_info, start, len); |
815a51c7 JS |
4737 | if (!eb) |
4738 | return NULL; | |
4739 | ||
65ad0104 | 4740 | num_pages = num_extent_pages(eb); |
815a51c7 | 4741 | for (i = 0; i < num_pages; i++) { |
9ec72677 | 4742 | eb->pages[i] = alloc_page(GFP_NOFS); |
815a51c7 JS |
4743 | if (!eb->pages[i]) |
4744 | goto err; | |
4745 | } | |
4746 | set_extent_buffer_uptodate(eb); | |
4747 | btrfs_set_header_nritems(eb, 0); | |
b0132a3b | 4748 | set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
815a51c7 JS |
4749 | |
4750 | return eb; | |
4751 | err: | |
84167d19 SB |
4752 | for (; i > 0; i--) |
4753 | __free_page(eb->pages[i - 1]); | |
815a51c7 JS |
4754 | __free_extent_buffer(eb); |
4755 | return NULL; | |
4756 | } | |
4757 | ||
0f331229 | 4758 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
da17066c | 4759 | u64 start) |
0f331229 | 4760 | { |
da17066c | 4761 | return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize); |
0f331229 OS |
4762 | } |
4763 | ||
0b32f4bb JB |
4764 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
4765 | { | |
242e18c7 | 4766 | int refs; |
0b32f4bb JB |
4767 | /* the ref bit is tricky. We have to make sure it is set |
4768 | * if we have the buffer dirty. Otherwise the | |
4769 | * code to free a buffer can end up dropping a dirty | |
4770 | * page | |
4771 | * | |
4772 | * Once the ref bit is set, it won't go away while the | |
4773 | * buffer is dirty or in writeback, and it also won't | |
4774 | * go away while we have the reference count on the | |
4775 | * eb bumped. | |
4776 | * | |
4777 | * We can't just set the ref bit without bumping the | |
4778 | * ref on the eb because free_extent_buffer might | |
4779 | * see the ref bit and try to clear it. If this happens | |
4780 | * free_extent_buffer might end up dropping our original | |
4781 | * ref by mistake and freeing the page before we are able | |
4782 | * to add one more ref. | |
4783 | * | |
4784 | * So bump the ref count first, then set the bit. If someone | |
4785 | * beat us to it, drop the ref we added. | |
4786 | */ | |
242e18c7 CM |
4787 | refs = atomic_read(&eb->refs); |
4788 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
4789 | return; | |
4790 | ||
594831c4 JB |
4791 | spin_lock(&eb->refs_lock); |
4792 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 4793 | atomic_inc(&eb->refs); |
594831c4 | 4794 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
4795 | } |
4796 | ||
2457aec6 MG |
4797 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
4798 | struct page *accessed) | |
5df4235e | 4799 | { |
cc5e31a4 | 4800 | int num_pages, i; |
5df4235e | 4801 | |
0b32f4bb JB |
4802 | check_buffer_tree_ref(eb); |
4803 | ||
65ad0104 | 4804 | num_pages = num_extent_pages(eb); |
5df4235e | 4805 | for (i = 0; i < num_pages; i++) { |
fb85fc9a DS |
4806 | struct page *p = eb->pages[i]; |
4807 | ||
2457aec6 MG |
4808 | if (p != accessed) |
4809 | mark_page_accessed(p); | |
5df4235e JB |
4810 | } |
4811 | } | |
4812 | ||
f28491e0 JB |
4813 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
4814 | u64 start) | |
452c75c3 CS |
4815 | { |
4816 | struct extent_buffer *eb; | |
4817 | ||
4818 | rcu_read_lock(); | |
f28491e0 | 4819 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
09cbfeaf | 4820 | start >> PAGE_SHIFT); |
452c75c3 CS |
4821 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
4822 | rcu_read_unlock(); | |
062c19e9 FM |
4823 | /* |
4824 | * Lock our eb's refs_lock to avoid races with | |
4825 | * free_extent_buffer. When we get our eb it might be flagged | |
4826 | * with EXTENT_BUFFER_STALE and another task running | |
4827 | * free_extent_buffer might have seen that flag set, | |
4828 | * eb->refs == 2, that the buffer isn't under IO (dirty and | |
4829 | * writeback flags not set) and it's still in the tree (flag | |
4830 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process | |
4831 | * of decrementing the extent buffer's reference count twice. | |
4832 | * So here we could race and increment the eb's reference count, | |
4833 | * clear its stale flag, mark it as dirty and drop our reference | |
4834 | * before the other task finishes executing free_extent_buffer, | |
4835 | * which would later result in an attempt to free an extent | |
4836 | * buffer that is dirty. | |
4837 | */ | |
4838 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
4839 | spin_lock(&eb->refs_lock); | |
4840 | spin_unlock(&eb->refs_lock); | |
4841 | } | |
2457aec6 | 4842 | mark_extent_buffer_accessed(eb, NULL); |
452c75c3 CS |
4843 | return eb; |
4844 | } | |
4845 | rcu_read_unlock(); | |
4846 | ||
4847 | return NULL; | |
4848 | } | |
4849 | ||
faa2dbf0 JB |
4850 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
4851 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
da17066c | 4852 | u64 start) |
faa2dbf0 JB |
4853 | { |
4854 | struct extent_buffer *eb, *exists = NULL; | |
4855 | int ret; | |
4856 | ||
4857 | eb = find_extent_buffer(fs_info, start); | |
4858 | if (eb) | |
4859 | return eb; | |
da17066c | 4860 | eb = alloc_dummy_extent_buffer(fs_info, start); |
faa2dbf0 JB |
4861 | if (!eb) |
4862 | return NULL; | |
4863 | eb->fs_info = fs_info; | |
4864 | again: | |
e1860a77 | 4865 | ret = radix_tree_preload(GFP_NOFS); |
faa2dbf0 JB |
4866 | if (ret) |
4867 | goto free_eb; | |
4868 | spin_lock(&fs_info->buffer_lock); | |
4869 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
09cbfeaf | 4870 | start >> PAGE_SHIFT, eb); |
faa2dbf0 JB |
4871 | spin_unlock(&fs_info->buffer_lock); |
4872 | radix_tree_preload_end(); | |
4873 | if (ret == -EEXIST) { | |
4874 | exists = find_extent_buffer(fs_info, start); | |
4875 | if (exists) | |
4876 | goto free_eb; | |
4877 | else | |
4878 | goto again; | |
4879 | } | |
4880 | check_buffer_tree_ref(eb); | |
4881 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
4882 | ||
faa2dbf0 JB |
4883 | return eb; |
4884 | free_eb: | |
4885 | btrfs_release_extent_buffer(eb); | |
4886 | return exists; | |
4887 | } | |
4888 | #endif | |
4889 | ||
f28491e0 | 4890 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
ce3e6984 | 4891 | u64 start) |
d1310b2e | 4892 | { |
da17066c | 4893 | unsigned long len = fs_info->nodesize; |
cc5e31a4 DS |
4894 | int num_pages; |
4895 | int i; | |
09cbfeaf | 4896 | unsigned long index = start >> PAGE_SHIFT; |
d1310b2e | 4897 | struct extent_buffer *eb; |
6af118ce | 4898 | struct extent_buffer *exists = NULL; |
d1310b2e | 4899 | struct page *p; |
f28491e0 | 4900 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
d1310b2e | 4901 | int uptodate = 1; |
19fe0a8b | 4902 | int ret; |
d1310b2e | 4903 | |
da17066c | 4904 | if (!IS_ALIGNED(start, fs_info->sectorsize)) { |
c871b0f2 LB |
4905 | btrfs_err(fs_info, "bad tree block start %llu", start); |
4906 | return ERR_PTR(-EINVAL); | |
4907 | } | |
4908 | ||
f28491e0 | 4909 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 4910 | if (eb) |
6af118ce | 4911 | return eb; |
6af118ce | 4912 | |
23d79d81 | 4913 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 4914 | if (!eb) |
c871b0f2 | 4915 | return ERR_PTR(-ENOMEM); |
d1310b2e | 4916 | |
65ad0104 | 4917 | num_pages = num_extent_pages(eb); |
727011e0 | 4918 | for (i = 0; i < num_pages; i++, index++) { |
d1b5c567 | 4919 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
c871b0f2 LB |
4920 | if (!p) { |
4921 | exists = ERR_PTR(-ENOMEM); | |
6af118ce | 4922 | goto free_eb; |
c871b0f2 | 4923 | } |
4f2de97a JB |
4924 | |
4925 | spin_lock(&mapping->private_lock); | |
4926 | if (PagePrivate(p)) { | |
4927 | /* | |
4928 | * We could have already allocated an eb for this page | |
4929 | * and attached one so lets see if we can get a ref on | |
4930 | * the existing eb, and if we can we know it's good and | |
4931 | * we can just return that one, else we know we can just | |
4932 | * overwrite page->private. | |
4933 | */ | |
4934 | exists = (struct extent_buffer *)p->private; | |
4935 | if (atomic_inc_not_zero(&exists->refs)) { | |
4936 | spin_unlock(&mapping->private_lock); | |
4937 | unlock_page(p); | |
09cbfeaf | 4938 | put_page(p); |
2457aec6 | 4939 | mark_extent_buffer_accessed(exists, p); |
4f2de97a JB |
4940 | goto free_eb; |
4941 | } | |
5ca64f45 | 4942 | exists = NULL; |
4f2de97a | 4943 | |
0b32f4bb | 4944 | /* |
4f2de97a JB |
4945 | * Do this so attach doesn't complain and we need to |
4946 | * drop the ref the old guy had. | |
4947 | */ | |
4948 | ClearPagePrivate(p); | |
0b32f4bb | 4949 | WARN_ON(PageDirty(p)); |
09cbfeaf | 4950 | put_page(p); |
d1310b2e | 4951 | } |
4f2de97a JB |
4952 | attach_extent_buffer_page(eb, p); |
4953 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 4954 | WARN_ON(PageDirty(p)); |
727011e0 | 4955 | eb->pages[i] = p; |
d1310b2e CM |
4956 | if (!PageUptodate(p)) |
4957 | uptodate = 0; | |
eb14ab8e CM |
4958 | |
4959 | /* | |
b16d011e NB |
4960 | * We can't unlock the pages just yet since the extent buffer |
4961 | * hasn't been properly inserted in the radix tree, this | |
4962 | * opens a race with btree_releasepage which can free a page | |
4963 | * while we are still filling in all pages for the buffer and | |
4964 | * we could crash. | |
eb14ab8e | 4965 | */ |
d1310b2e CM |
4966 | } |
4967 | if (uptodate) | |
b4ce94de | 4968 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 4969 | again: |
e1860a77 | 4970 | ret = radix_tree_preload(GFP_NOFS); |
c871b0f2 LB |
4971 | if (ret) { |
4972 | exists = ERR_PTR(ret); | |
19fe0a8b | 4973 | goto free_eb; |
c871b0f2 | 4974 | } |
19fe0a8b | 4975 | |
f28491e0 JB |
4976 | spin_lock(&fs_info->buffer_lock); |
4977 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
09cbfeaf | 4978 | start >> PAGE_SHIFT, eb); |
f28491e0 | 4979 | spin_unlock(&fs_info->buffer_lock); |
452c75c3 | 4980 | radix_tree_preload_end(); |
19fe0a8b | 4981 | if (ret == -EEXIST) { |
f28491e0 | 4982 | exists = find_extent_buffer(fs_info, start); |
452c75c3 CS |
4983 | if (exists) |
4984 | goto free_eb; | |
4985 | else | |
115391d2 | 4986 | goto again; |
6af118ce | 4987 | } |
6af118ce | 4988 | /* add one reference for the tree */ |
0b32f4bb | 4989 | check_buffer_tree_ref(eb); |
34b41ace | 4990 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
4991 | |
4992 | /* | |
b16d011e NB |
4993 | * Now it's safe to unlock the pages because any calls to |
4994 | * btree_releasepage will correctly detect that a page belongs to a | |
4995 | * live buffer and won't free them prematurely. | |
eb14ab8e | 4996 | */ |
28187ae5 NB |
4997 | for (i = 0; i < num_pages; i++) |
4998 | unlock_page(eb->pages[i]); | |
d1310b2e CM |
4999 | return eb; |
5000 | ||
6af118ce | 5001 | free_eb: |
5ca64f45 | 5002 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
5003 | for (i = 0; i < num_pages; i++) { |
5004 | if (eb->pages[i]) | |
5005 | unlock_page(eb->pages[i]); | |
5006 | } | |
eb14ab8e | 5007 | |
897ca6e9 | 5008 | btrfs_release_extent_buffer(eb); |
6af118ce | 5009 | return exists; |
d1310b2e | 5010 | } |
d1310b2e | 5011 | |
3083ee2e JB |
5012 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
5013 | { | |
5014 | struct extent_buffer *eb = | |
5015 | container_of(head, struct extent_buffer, rcu_head); | |
5016 | ||
5017 | __free_extent_buffer(eb); | |
5018 | } | |
5019 | ||
f7a52a40 | 5020 | static int release_extent_buffer(struct extent_buffer *eb) |
3083ee2e | 5021 | { |
07e21c4d NB |
5022 | lockdep_assert_held(&eb->refs_lock); |
5023 | ||
3083ee2e JB |
5024 | WARN_ON(atomic_read(&eb->refs) == 0); |
5025 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 5026 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 5027 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 5028 | |
815a51c7 | 5029 | spin_unlock(&eb->refs_lock); |
3083ee2e | 5030 | |
f28491e0 JB |
5031 | spin_lock(&fs_info->buffer_lock); |
5032 | radix_tree_delete(&fs_info->buffer_radix, | |
09cbfeaf | 5033 | eb->start >> PAGE_SHIFT); |
f28491e0 | 5034 | spin_unlock(&fs_info->buffer_lock); |
34b41ace JB |
5035 | } else { |
5036 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 5037 | } |
3083ee2e JB |
5038 | |
5039 | /* Should be safe to release our pages at this point */ | |
55ac0139 | 5040 | btrfs_release_extent_buffer_pages(eb); |
bcb7e449 | 5041 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
b0132a3b | 5042 | if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) { |
bcb7e449 JB |
5043 | __free_extent_buffer(eb); |
5044 | return 1; | |
5045 | } | |
5046 | #endif | |
3083ee2e | 5047 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 5048 | return 1; |
3083ee2e JB |
5049 | } |
5050 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
5051 | |
5052 | return 0; | |
3083ee2e JB |
5053 | } |
5054 | ||
d1310b2e CM |
5055 | void free_extent_buffer(struct extent_buffer *eb) |
5056 | { | |
242e18c7 CM |
5057 | int refs; |
5058 | int old; | |
d1310b2e CM |
5059 | if (!eb) |
5060 | return; | |
5061 | ||
242e18c7 CM |
5062 | while (1) { |
5063 | refs = atomic_read(&eb->refs); | |
46cc775e NB |
5064 | if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3) |
5065 | || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && | |
5066 | refs == 1)) | |
242e18c7 CM |
5067 | break; |
5068 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
5069 | if (old == refs) | |
5070 | return; | |
5071 | } | |
5072 | ||
3083ee2e JB |
5073 | spin_lock(&eb->refs_lock); |
5074 | if (atomic_read(&eb->refs) == 2 && | |
5075 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 5076 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
5077 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5078 | atomic_dec(&eb->refs); | |
5079 | ||
5080 | /* | |
5081 | * I know this is terrible, but it's temporary until we stop tracking | |
5082 | * the uptodate bits and such for the extent buffers. | |
5083 | */ | |
f7a52a40 | 5084 | release_extent_buffer(eb); |
3083ee2e JB |
5085 | } |
5086 | ||
5087 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
5088 | { | |
5089 | if (!eb) | |
d1310b2e CM |
5090 | return; |
5091 | ||
3083ee2e JB |
5092 | spin_lock(&eb->refs_lock); |
5093 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
5094 | ||
0b32f4bb | 5095 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
5096 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5097 | atomic_dec(&eb->refs); | |
f7a52a40 | 5098 | release_extent_buffer(eb); |
d1310b2e | 5099 | } |
d1310b2e | 5100 | |
1d4284bd | 5101 | void clear_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5102 | { |
cc5e31a4 DS |
5103 | int i; |
5104 | int num_pages; | |
d1310b2e CM |
5105 | struct page *page; |
5106 | ||
65ad0104 | 5107 | num_pages = num_extent_pages(eb); |
d1310b2e CM |
5108 | |
5109 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 5110 | page = eb->pages[i]; |
b9473439 | 5111 | if (!PageDirty(page)) |
d2c3f4f6 CM |
5112 | continue; |
5113 | ||
a61e6f29 | 5114 | lock_page(page); |
eb14ab8e CM |
5115 | WARN_ON(!PagePrivate(page)); |
5116 | ||
d1310b2e | 5117 | clear_page_dirty_for_io(page); |
b93b0163 | 5118 | xa_lock_irq(&page->mapping->i_pages); |
0a943c65 MW |
5119 | if (!PageDirty(page)) |
5120 | __xa_clear_mark(&page->mapping->i_pages, | |
5121 | page_index(page), PAGECACHE_TAG_DIRTY); | |
b93b0163 | 5122 | xa_unlock_irq(&page->mapping->i_pages); |
bf0da8c1 | 5123 | ClearPageError(page); |
a61e6f29 | 5124 | unlock_page(page); |
d1310b2e | 5125 | } |
0b32f4bb | 5126 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 5127 | } |
d1310b2e | 5128 | |
abb57ef3 | 5129 | bool set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5130 | { |
cc5e31a4 DS |
5131 | int i; |
5132 | int num_pages; | |
abb57ef3 | 5133 | bool was_dirty; |
d1310b2e | 5134 | |
0b32f4bb JB |
5135 | check_buffer_tree_ref(eb); |
5136 | ||
b9473439 | 5137 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 5138 | |
65ad0104 | 5139 | num_pages = num_extent_pages(eb); |
3083ee2e | 5140 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
5141 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
5142 | ||
abb57ef3 LB |
5143 | if (!was_dirty) |
5144 | for (i = 0; i < num_pages; i++) | |
5145 | set_page_dirty(eb->pages[i]); | |
51995c39 LB |
5146 | |
5147 | #ifdef CONFIG_BTRFS_DEBUG | |
5148 | for (i = 0; i < num_pages; i++) | |
5149 | ASSERT(PageDirty(eb->pages[i])); | |
5150 | #endif | |
5151 | ||
b9473439 | 5152 | return was_dirty; |
d1310b2e | 5153 | } |
d1310b2e | 5154 | |
69ba3927 | 5155 | void clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 | 5156 | { |
cc5e31a4 | 5157 | int i; |
1259ab75 | 5158 | struct page *page; |
cc5e31a4 | 5159 | int num_pages; |
1259ab75 | 5160 | |
b4ce94de | 5161 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 5162 | num_pages = num_extent_pages(eb); |
1259ab75 | 5163 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5164 | page = eb->pages[i]; |
33958dc6 CM |
5165 | if (page) |
5166 | ClearPageUptodate(page); | |
1259ab75 | 5167 | } |
1259ab75 CM |
5168 | } |
5169 | ||
09c25a8c | 5170 | void set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 5171 | { |
cc5e31a4 | 5172 | int i; |
d1310b2e | 5173 | struct page *page; |
cc5e31a4 | 5174 | int num_pages; |
d1310b2e | 5175 | |
0b32f4bb | 5176 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 5177 | num_pages = num_extent_pages(eb); |
d1310b2e | 5178 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5179 | page = eb->pages[i]; |
d1310b2e CM |
5180 | SetPageUptodate(page); |
5181 | } | |
d1310b2e | 5182 | } |
d1310b2e | 5183 | |
d1310b2e | 5184 | int read_extent_buffer_pages(struct extent_io_tree *tree, |
6af49dbd | 5185 | struct extent_buffer *eb, int wait, int mirror_num) |
d1310b2e | 5186 | { |
cc5e31a4 | 5187 | int i; |
d1310b2e CM |
5188 | struct page *page; |
5189 | int err; | |
5190 | int ret = 0; | |
ce9adaa5 CM |
5191 | int locked_pages = 0; |
5192 | int all_uptodate = 1; | |
cc5e31a4 | 5193 | int num_pages; |
727011e0 | 5194 | unsigned long num_reads = 0; |
a86c12c7 | 5195 | struct bio *bio = NULL; |
c8b97818 | 5196 | unsigned long bio_flags = 0; |
a86c12c7 | 5197 | |
b4ce94de | 5198 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
5199 | return 0; |
5200 | ||
65ad0104 | 5201 | num_pages = num_extent_pages(eb); |
8436ea91 | 5202 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5203 | page = eb->pages[i]; |
bb82ab88 | 5204 | if (wait == WAIT_NONE) { |
2db04966 | 5205 | if (!trylock_page(page)) |
ce9adaa5 | 5206 | goto unlock_exit; |
d1310b2e CM |
5207 | } else { |
5208 | lock_page(page); | |
5209 | } | |
ce9adaa5 | 5210 | locked_pages++; |
2571e739 LB |
5211 | } |
5212 | /* | |
5213 | * We need to firstly lock all pages to make sure that | |
5214 | * the uptodate bit of our pages won't be affected by | |
5215 | * clear_extent_buffer_uptodate(). | |
5216 | */ | |
8436ea91 | 5217 | for (i = 0; i < num_pages; i++) { |
2571e739 | 5218 | page = eb->pages[i]; |
727011e0 CM |
5219 | if (!PageUptodate(page)) { |
5220 | num_reads++; | |
ce9adaa5 | 5221 | all_uptodate = 0; |
727011e0 | 5222 | } |
ce9adaa5 | 5223 | } |
2571e739 | 5224 | |
ce9adaa5 | 5225 | if (all_uptodate) { |
8436ea91 | 5226 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
5227 | goto unlock_exit; |
5228 | } | |
5229 | ||
656f30db | 5230 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 5231 | eb->read_mirror = 0; |
0b32f4bb | 5232 | atomic_set(&eb->io_pages, num_reads); |
8436ea91 | 5233 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5234 | page = eb->pages[i]; |
baf863b9 | 5235 | |
ce9adaa5 | 5236 | if (!PageUptodate(page)) { |
baf863b9 LB |
5237 | if (ret) { |
5238 | atomic_dec(&eb->io_pages); | |
5239 | unlock_page(page); | |
5240 | continue; | |
5241 | } | |
5242 | ||
f188591e | 5243 | ClearPageError(page); |
a86c12c7 | 5244 | err = __extent_read_full_page(tree, page, |
6af49dbd | 5245 | btree_get_extent, &bio, |
d4c7ca86 | 5246 | mirror_num, &bio_flags, |
1f7ad75b | 5247 | REQ_META); |
baf863b9 | 5248 | if (err) { |
d1310b2e | 5249 | ret = err; |
baf863b9 LB |
5250 | /* |
5251 | * We use &bio in above __extent_read_full_page, | |
5252 | * so we ensure that if it returns error, the | |
5253 | * current page fails to add itself to bio and | |
5254 | * it's been unlocked. | |
5255 | * | |
5256 | * We must dec io_pages by ourselves. | |
5257 | */ | |
5258 | atomic_dec(&eb->io_pages); | |
5259 | } | |
d1310b2e CM |
5260 | } else { |
5261 | unlock_page(page); | |
5262 | } | |
5263 | } | |
5264 | ||
355808c2 | 5265 | if (bio) { |
1f7ad75b | 5266 | err = submit_one_bio(bio, mirror_num, bio_flags); |
79787eaa JM |
5267 | if (err) |
5268 | return err; | |
355808c2 | 5269 | } |
a86c12c7 | 5270 | |
bb82ab88 | 5271 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 5272 | return ret; |
d397712b | 5273 | |
8436ea91 | 5274 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5275 | page = eb->pages[i]; |
d1310b2e | 5276 | wait_on_page_locked(page); |
d397712b | 5277 | if (!PageUptodate(page)) |
d1310b2e | 5278 | ret = -EIO; |
d1310b2e | 5279 | } |
d397712b | 5280 | |
d1310b2e | 5281 | return ret; |
ce9adaa5 CM |
5282 | |
5283 | unlock_exit: | |
d397712b | 5284 | while (locked_pages > 0) { |
ce9adaa5 | 5285 | locked_pages--; |
8436ea91 JB |
5286 | page = eb->pages[locked_pages]; |
5287 | unlock_page(page); | |
ce9adaa5 CM |
5288 | } |
5289 | return ret; | |
d1310b2e | 5290 | } |
d1310b2e | 5291 | |
1cbb1f45 JM |
5292 | void read_extent_buffer(const struct extent_buffer *eb, void *dstv, |
5293 | unsigned long start, unsigned long len) | |
d1310b2e CM |
5294 | { |
5295 | size_t cur; | |
5296 | size_t offset; | |
5297 | struct page *page; | |
5298 | char *kaddr; | |
5299 | char *dst = (char *)dstv; | |
7073017a | 5300 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5301 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
d1310b2e | 5302 | |
f716abd5 LB |
5303 | if (start + len > eb->len) { |
5304 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n", | |
5305 | eb->start, eb->len, start, len); | |
5306 | memset(dst, 0, len); | |
5307 | return; | |
5308 | } | |
d1310b2e | 5309 | |
7073017a | 5310 | offset = offset_in_page(start_offset + start); |
d1310b2e | 5311 | |
d397712b | 5312 | while (len > 0) { |
fb85fc9a | 5313 | page = eb->pages[i]; |
d1310b2e | 5314 | |
09cbfeaf | 5315 | cur = min(len, (PAGE_SIZE - offset)); |
a6591715 | 5316 | kaddr = page_address(page); |
d1310b2e | 5317 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
5318 | |
5319 | dst += cur; | |
5320 | len -= cur; | |
5321 | offset = 0; | |
5322 | i++; | |
5323 | } | |
5324 | } | |
d1310b2e | 5325 | |
1cbb1f45 JM |
5326 | int read_extent_buffer_to_user(const struct extent_buffer *eb, |
5327 | void __user *dstv, | |
5328 | unsigned long start, unsigned long len) | |
550ac1d8 GH |
5329 | { |
5330 | size_t cur; | |
5331 | size_t offset; | |
5332 | struct page *page; | |
5333 | char *kaddr; | |
5334 | char __user *dst = (char __user *)dstv; | |
7073017a | 5335 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5336 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
550ac1d8 GH |
5337 | int ret = 0; |
5338 | ||
5339 | WARN_ON(start > eb->len); | |
5340 | WARN_ON(start + len > eb->start + eb->len); | |
5341 | ||
7073017a | 5342 | offset = offset_in_page(start_offset + start); |
550ac1d8 GH |
5343 | |
5344 | while (len > 0) { | |
fb85fc9a | 5345 | page = eb->pages[i]; |
550ac1d8 | 5346 | |
09cbfeaf | 5347 | cur = min(len, (PAGE_SIZE - offset)); |
550ac1d8 GH |
5348 | kaddr = page_address(page); |
5349 | if (copy_to_user(dst, kaddr + offset, cur)) { | |
5350 | ret = -EFAULT; | |
5351 | break; | |
5352 | } | |
5353 | ||
5354 | dst += cur; | |
5355 | len -= cur; | |
5356 | offset = 0; | |
5357 | i++; | |
5358 | } | |
5359 | ||
5360 | return ret; | |
5361 | } | |
5362 | ||
415b35a5 LB |
5363 | /* |
5364 | * return 0 if the item is found within a page. | |
5365 | * return 1 if the item spans two pages. | |
5366 | * return -EINVAL otherwise. | |
5367 | */ | |
1cbb1f45 JM |
5368 | int map_private_extent_buffer(const struct extent_buffer *eb, |
5369 | unsigned long start, unsigned long min_len, | |
5370 | char **map, unsigned long *map_start, | |
5371 | unsigned long *map_len) | |
d1310b2e | 5372 | { |
cc2c39d6 | 5373 | size_t offset; |
d1310b2e CM |
5374 | char *kaddr; |
5375 | struct page *p; | |
7073017a | 5376 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5377 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
d1310b2e | 5378 | unsigned long end_i = (start_offset + start + min_len - 1) >> |
09cbfeaf | 5379 | PAGE_SHIFT; |
d1310b2e | 5380 | |
f716abd5 LB |
5381 | if (start + min_len > eb->len) { |
5382 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n", | |
5383 | eb->start, eb->len, start, min_len); | |
5384 | return -EINVAL; | |
5385 | } | |
5386 | ||
d1310b2e | 5387 | if (i != end_i) |
415b35a5 | 5388 | return 1; |
d1310b2e CM |
5389 | |
5390 | if (i == 0) { | |
5391 | offset = start_offset; | |
5392 | *map_start = 0; | |
5393 | } else { | |
5394 | offset = 0; | |
09cbfeaf | 5395 | *map_start = ((u64)i << PAGE_SHIFT) - start_offset; |
d1310b2e | 5396 | } |
d397712b | 5397 | |
fb85fc9a | 5398 | p = eb->pages[i]; |
a6591715 | 5399 | kaddr = page_address(p); |
d1310b2e | 5400 | *map = kaddr + offset; |
09cbfeaf | 5401 | *map_len = PAGE_SIZE - offset; |
d1310b2e CM |
5402 | return 0; |
5403 | } | |
d1310b2e | 5404 | |
1cbb1f45 JM |
5405 | int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv, |
5406 | unsigned long start, unsigned long len) | |
d1310b2e CM |
5407 | { |
5408 | size_t cur; | |
5409 | size_t offset; | |
5410 | struct page *page; | |
5411 | char *kaddr; | |
5412 | char *ptr = (char *)ptrv; | |
7073017a | 5413 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5414 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
d1310b2e CM |
5415 | int ret = 0; |
5416 | ||
5417 | WARN_ON(start > eb->len); | |
5418 | WARN_ON(start + len > eb->start + eb->len); | |
5419 | ||
7073017a | 5420 | offset = offset_in_page(start_offset + start); |
d1310b2e | 5421 | |
d397712b | 5422 | while (len > 0) { |
fb85fc9a | 5423 | page = eb->pages[i]; |
d1310b2e | 5424 | |
09cbfeaf | 5425 | cur = min(len, (PAGE_SIZE - offset)); |
d1310b2e | 5426 | |
a6591715 | 5427 | kaddr = page_address(page); |
d1310b2e | 5428 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
5429 | if (ret) |
5430 | break; | |
5431 | ||
5432 | ptr += cur; | |
5433 | len -= cur; | |
5434 | offset = 0; | |
5435 | i++; | |
5436 | } | |
5437 | return ret; | |
5438 | } | |
d1310b2e | 5439 | |
f157bf76 DS |
5440 | void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb, |
5441 | const void *srcv) | |
5442 | { | |
5443 | char *kaddr; | |
5444 | ||
5445 | WARN_ON(!PageUptodate(eb->pages[0])); | |
5446 | kaddr = page_address(eb->pages[0]); | |
5447 | memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv, | |
5448 | BTRFS_FSID_SIZE); | |
5449 | } | |
5450 | ||
5451 | void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv) | |
5452 | { | |
5453 | char *kaddr; | |
5454 | ||
5455 | WARN_ON(!PageUptodate(eb->pages[0])); | |
5456 | kaddr = page_address(eb->pages[0]); | |
5457 | memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv, | |
5458 | BTRFS_FSID_SIZE); | |
5459 | } | |
5460 | ||
d1310b2e CM |
5461 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, |
5462 | unsigned long start, unsigned long len) | |
5463 | { | |
5464 | size_t cur; | |
5465 | size_t offset; | |
5466 | struct page *page; | |
5467 | char *kaddr; | |
5468 | char *src = (char *)srcv; | |
7073017a | 5469 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5470 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
d1310b2e CM |
5471 | |
5472 | WARN_ON(start > eb->len); | |
5473 | WARN_ON(start + len > eb->start + eb->len); | |
5474 | ||
7073017a | 5475 | offset = offset_in_page(start_offset + start); |
d1310b2e | 5476 | |
d397712b | 5477 | while (len > 0) { |
fb85fc9a | 5478 | page = eb->pages[i]; |
d1310b2e CM |
5479 | WARN_ON(!PageUptodate(page)); |
5480 | ||
09cbfeaf | 5481 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 5482 | kaddr = page_address(page); |
d1310b2e | 5483 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
5484 | |
5485 | src += cur; | |
5486 | len -= cur; | |
5487 | offset = 0; | |
5488 | i++; | |
5489 | } | |
5490 | } | |
d1310b2e | 5491 | |
b159fa28 DS |
5492 | void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start, |
5493 | unsigned long len) | |
d1310b2e CM |
5494 | { |
5495 | size_t cur; | |
5496 | size_t offset; | |
5497 | struct page *page; | |
5498 | char *kaddr; | |
7073017a | 5499 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5500 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
d1310b2e CM |
5501 | |
5502 | WARN_ON(start > eb->len); | |
5503 | WARN_ON(start + len > eb->start + eb->len); | |
5504 | ||
7073017a | 5505 | offset = offset_in_page(start_offset + start); |
d1310b2e | 5506 | |
d397712b | 5507 | while (len > 0) { |
fb85fc9a | 5508 | page = eb->pages[i]; |
d1310b2e CM |
5509 | WARN_ON(!PageUptodate(page)); |
5510 | ||
09cbfeaf | 5511 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 5512 | kaddr = page_address(page); |
b159fa28 | 5513 | memset(kaddr + offset, 0, cur); |
d1310b2e CM |
5514 | |
5515 | len -= cur; | |
5516 | offset = 0; | |
5517 | i++; | |
5518 | } | |
5519 | } | |
d1310b2e | 5520 | |
58e8012c DS |
5521 | void copy_extent_buffer_full(struct extent_buffer *dst, |
5522 | struct extent_buffer *src) | |
5523 | { | |
5524 | int i; | |
cc5e31a4 | 5525 | int num_pages; |
58e8012c DS |
5526 | |
5527 | ASSERT(dst->len == src->len); | |
5528 | ||
65ad0104 | 5529 | num_pages = num_extent_pages(dst); |
58e8012c DS |
5530 | for (i = 0; i < num_pages; i++) |
5531 | copy_page(page_address(dst->pages[i]), | |
5532 | page_address(src->pages[i])); | |
5533 | } | |
5534 | ||
d1310b2e CM |
5535 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, |
5536 | unsigned long dst_offset, unsigned long src_offset, | |
5537 | unsigned long len) | |
5538 | { | |
5539 | u64 dst_len = dst->len; | |
5540 | size_t cur; | |
5541 | size_t offset; | |
5542 | struct page *page; | |
5543 | char *kaddr; | |
7073017a | 5544 | size_t start_offset = offset_in_page(dst->start); |
09cbfeaf | 5545 | unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT; |
d1310b2e CM |
5546 | |
5547 | WARN_ON(src->len != dst_len); | |
5548 | ||
7073017a | 5549 | offset = offset_in_page(start_offset + dst_offset); |
d1310b2e | 5550 | |
d397712b | 5551 | while (len > 0) { |
fb85fc9a | 5552 | page = dst->pages[i]; |
d1310b2e CM |
5553 | WARN_ON(!PageUptodate(page)); |
5554 | ||
09cbfeaf | 5555 | cur = min(len, (unsigned long)(PAGE_SIZE - offset)); |
d1310b2e | 5556 | |
a6591715 | 5557 | kaddr = page_address(page); |
d1310b2e | 5558 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
5559 | |
5560 | src_offset += cur; | |
5561 | len -= cur; | |
5562 | offset = 0; | |
5563 | i++; | |
5564 | } | |
5565 | } | |
d1310b2e | 5566 | |
3e1e8bb7 OS |
5567 | /* |
5568 | * eb_bitmap_offset() - calculate the page and offset of the byte containing the | |
5569 | * given bit number | |
5570 | * @eb: the extent buffer | |
5571 | * @start: offset of the bitmap item in the extent buffer | |
5572 | * @nr: bit number | |
5573 | * @page_index: return index of the page in the extent buffer that contains the | |
5574 | * given bit number | |
5575 | * @page_offset: return offset into the page given by page_index | |
5576 | * | |
5577 | * This helper hides the ugliness of finding the byte in an extent buffer which | |
5578 | * contains a given bit. | |
5579 | */ | |
5580 | static inline void eb_bitmap_offset(struct extent_buffer *eb, | |
5581 | unsigned long start, unsigned long nr, | |
5582 | unsigned long *page_index, | |
5583 | size_t *page_offset) | |
5584 | { | |
7073017a | 5585 | size_t start_offset = offset_in_page(eb->start); |
3e1e8bb7 OS |
5586 | size_t byte_offset = BIT_BYTE(nr); |
5587 | size_t offset; | |
5588 | ||
5589 | /* | |
5590 | * The byte we want is the offset of the extent buffer + the offset of | |
5591 | * the bitmap item in the extent buffer + the offset of the byte in the | |
5592 | * bitmap item. | |
5593 | */ | |
5594 | offset = start_offset + start + byte_offset; | |
5595 | ||
09cbfeaf | 5596 | *page_index = offset >> PAGE_SHIFT; |
7073017a | 5597 | *page_offset = offset_in_page(offset); |
3e1e8bb7 OS |
5598 | } |
5599 | ||
5600 | /** | |
5601 | * extent_buffer_test_bit - determine whether a bit in a bitmap item is set | |
5602 | * @eb: the extent buffer | |
5603 | * @start: offset of the bitmap item in the extent buffer | |
5604 | * @nr: bit number to test | |
5605 | */ | |
5606 | int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start, | |
5607 | unsigned long nr) | |
5608 | { | |
2fe1d551 | 5609 | u8 *kaddr; |
3e1e8bb7 OS |
5610 | struct page *page; |
5611 | unsigned long i; | |
5612 | size_t offset; | |
5613 | ||
5614 | eb_bitmap_offset(eb, start, nr, &i, &offset); | |
5615 | page = eb->pages[i]; | |
5616 | WARN_ON(!PageUptodate(page)); | |
5617 | kaddr = page_address(page); | |
5618 | return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1))); | |
5619 | } | |
5620 | ||
5621 | /** | |
5622 | * extent_buffer_bitmap_set - set an area of a bitmap | |
5623 | * @eb: the extent buffer | |
5624 | * @start: offset of the bitmap item in the extent buffer | |
5625 | * @pos: bit number of the first bit | |
5626 | * @len: number of bits to set | |
5627 | */ | |
5628 | void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start, | |
5629 | unsigned long pos, unsigned long len) | |
5630 | { | |
2fe1d551 | 5631 | u8 *kaddr; |
3e1e8bb7 OS |
5632 | struct page *page; |
5633 | unsigned long i; | |
5634 | size_t offset; | |
5635 | const unsigned int size = pos + len; | |
5636 | int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 5637 | u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
5638 | |
5639 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5640 | page = eb->pages[i]; | |
5641 | WARN_ON(!PageUptodate(page)); | |
5642 | kaddr = page_address(page); | |
5643 | ||
5644 | while (len >= bits_to_set) { | |
5645 | kaddr[offset] |= mask_to_set; | |
5646 | len -= bits_to_set; | |
5647 | bits_to_set = BITS_PER_BYTE; | |
9c894696 | 5648 | mask_to_set = ~0; |
09cbfeaf | 5649 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5650 | offset = 0; |
5651 | page = eb->pages[++i]; | |
5652 | WARN_ON(!PageUptodate(page)); | |
5653 | kaddr = page_address(page); | |
5654 | } | |
5655 | } | |
5656 | if (len) { | |
5657 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
5658 | kaddr[offset] |= mask_to_set; | |
5659 | } | |
5660 | } | |
5661 | ||
5662 | ||
5663 | /** | |
5664 | * extent_buffer_bitmap_clear - clear an area of a bitmap | |
5665 | * @eb: the extent buffer | |
5666 | * @start: offset of the bitmap item in the extent buffer | |
5667 | * @pos: bit number of the first bit | |
5668 | * @len: number of bits to clear | |
5669 | */ | |
5670 | void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start, | |
5671 | unsigned long pos, unsigned long len) | |
5672 | { | |
2fe1d551 | 5673 | u8 *kaddr; |
3e1e8bb7 OS |
5674 | struct page *page; |
5675 | unsigned long i; | |
5676 | size_t offset; | |
5677 | const unsigned int size = pos + len; | |
5678 | int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 5679 | u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
5680 | |
5681 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5682 | page = eb->pages[i]; | |
5683 | WARN_ON(!PageUptodate(page)); | |
5684 | kaddr = page_address(page); | |
5685 | ||
5686 | while (len >= bits_to_clear) { | |
5687 | kaddr[offset] &= ~mask_to_clear; | |
5688 | len -= bits_to_clear; | |
5689 | bits_to_clear = BITS_PER_BYTE; | |
9c894696 | 5690 | mask_to_clear = ~0; |
09cbfeaf | 5691 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5692 | offset = 0; |
5693 | page = eb->pages[++i]; | |
5694 | WARN_ON(!PageUptodate(page)); | |
5695 | kaddr = page_address(page); | |
5696 | } | |
5697 | } | |
5698 | if (len) { | |
5699 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
5700 | kaddr[offset] &= ~mask_to_clear; | |
5701 | } | |
5702 | } | |
5703 | ||
3387206f ST |
5704 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
5705 | { | |
5706 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
5707 | return distance < len; | |
5708 | } | |
5709 | ||
d1310b2e CM |
5710 | static void copy_pages(struct page *dst_page, struct page *src_page, |
5711 | unsigned long dst_off, unsigned long src_off, | |
5712 | unsigned long len) | |
5713 | { | |
a6591715 | 5714 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 5715 | char *src_kaddr; |
727011e0 | 5716 | int must_memmove = 0; |
d1310b2e | 5717 | |
3387206f | 5718 | if (dst_page != src_page) { |
a6591715 | 5719 | src_kaddr = page_address(src_page); |
3387206f | 5720 | } else { |
d1310b2e | 5721 | src_kaddr = dst_kaddr; |
727011e0 CM |
5722 | if (areas_overlap(src_off, dst_off, len)) |
5723 | must_memmove = 1; | |
3387206f | 5724 | } |
d1310b2e | 5725 | |
727011e0 CM |
5726 | if (must_memmove) |
5727 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
5728 | else | |
5729 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
5730 | } |
5731 | ||
5732 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5733 | unsigned long src_offset, unsigned long len) | |
5734 | { | |
0b246afa | 5735 | struct btrfs_fs_info *fs_info = dst->fs_info; |
d1310b2e CM |
5736 | size_t cur; |
5737 | size_t dst_off_in_page; | |
5738 | size_t src_off_in_page; | |
7073017a | 5739 | size_t start_offset = offset_in_page(dst->start); |
d1310b2e CM |
5740 | unsigned long dst_i; |
5741 | unsigned long src_i; | |
5742 | ||
5743 | if (src_offset + len > dst->len) { | |
0b246afa | 5744 | btrfs_err(fs_info, |
5d163e0e JM |
5745 | "memmove bogus src_offset %lu move len %lu dst len %lu", |
5746 | src_offset, len, dst->len); | |
d1310b2e CM |
5747 | BUG_ON(1); |
5748 | } | |
5749 | if (dst_offset + len > dst->len) { | |
0b246afa | 5750 | btrfs_err(fs_info, |
5d163e0e JM |
5751 | "memmove bogus dst_offset %lu move len %lu dst len %lu", |
5752 | dst_offset, len, dst->len); | |
d1310b2e CM |
5753 | BUG_ON(1); |
5754 | } | |
5755 | ||
d397712b | 5756 | while (len > 0) { |
7073017a JT |
5757 | dst_off_in_page = offset_in_page(start_offset + dst_offset); |
5758 | src_off_in_page = offset_in_page(start_offset + src_offset); | |
d1310b2e | 5759 | |
09cbfeaf KS |
5760 | dst_i = (start_offset + dst_offset) >> PAGE_SHIFT; |
5761 | src_i = (start_offset + src_offset) >> PAGE_SHIFT; | |
d1310b2e | 5762 | |
09cbfeaf | 5763 | cur = min(len, (unsigned long)(PAGE_SIZE - |
d1310b2e CM |
5764 | src_off_in_page)); |
5765 | cur = min_t(unsigned long, cur, | |
09cbfeaf | 5766 | (unsigned long)(PAGE_SIZE - dst_off_in_page)); |
d1310b2e | 5767 | |
fb85fc9a | 5768 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5769 | dst_off_in_page, src_off_in_page, cur); |
5770 | ||
5771 | src_offset += cur; | |
5772 | dst_offset += cur; | |
5773 | len -= cur; | |
5774 | } | |
5775 | } | |
d1310b2e CM |
5776 | |
5777 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5778 | unsigned long src_offset, unsigned long len) | |
5779 | { | |
0b246afa | 5780 | struct btrfs_fs_info *fs_info = dst->fs_info; |
d1310b2e CM |
5781 | size_t cur; |
5782 | size_t dst_off_in_page; | |
5783 | size_t src_off_in_page; | |
5784 | unsigned long dst_end = dst_offset + len - 1; | |
5785 | unsigned long src_end = src_offset + len - 1; | |
7073017a | 5786 | size_t start_offset = offset_in_page(dst->start); |
d1310b2e CM |
5787 | unsigned long dst_i; |
5788 | unsigned long src_i; | |
5789 | ||
5790 | if (src_offset + len > dst->len) { | |
0b246afa | 5791 | btrfs_err(fs_info, |
5d163e0e JM |
5792 | "memmove bogus src_offset %lu move len %lu len %lu", |
5793 | src_offset, len, dst->len); | |
d1310b2e CM |
5794 | BUG_ON(1); |
5795 | } | |
5796 | if (dst_offset + len > dst->len) { | |
0b246afa | 5797 | btrfs_err(fs_info, |
5d163e0e JM |
5798 | "memmove bogus dst_offset %lu move len %lu len %lu", |
5799 | dst_offset, len, dst->len); | |
d1310b2e CM |
5800 | BUG_ON(1); |
5801 | } | |
727011e0 | 5802 | if (dst_offset < src_offset) { |
d1310b2e CM |
5803 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
5804 | return; | |
5805 | } | |
d397712b | 5806 | while (len > 0) { |
09cbfeaf KS |
5807 | dst_i = (start_offset + dst_end) >> PAGE_SHIFT; |
5808 | src_i = (start_offset + src_end) >> PAGE_SHIFT; | |
d1310b2e | 5809 | |
7073017a JT |
5810 | dst_off_in_page = offset_in_page(start_offset + dst_end); |
5811 | src_off_in_page = offset_in_page(start_offset + src_end); | |
d1310b2e CM |
5812 | |
5813 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
5814 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 5815 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5816 | dst_off_in_page - cur + 1, |
5817 | src_off_in_page - cur + 1, cur); | |
5818 | ||
5819 | dst_end -= cur; | |
5820 | src_end -= cur; | |
5821 | len -= cur; | |
5822 | } | |
5823 | } | |
6af118ce | 5824 | |
f7a52a40 | 5825 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 5826 | { |
6af118ce | 5827 | struct extent_buffer *eb; |
6af118ce | 5828 | |
3083ee2e | 5829 | /* |
01327610 | 5830 | * We need to make sure nobody is attaching this page to an eb right |
3083ee2e JB |
5831 | * now. |
5832 | */ | |
5833 | spin_lock(&page->mapping->private_lock); | |
5834 | if (!PagePrivate(page)) { | |
5835 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 5836 | return 1; |
45f49bce | 5837 | } |
6af118ce | 5838 | |
3083ee2e JB |
5839 | eb = (struct extent_buffer *)page->private; |
5840 | BUG_ON(!eb); | |
19fe0a8b MX |
5841 | |
5842 | /* | |
3083ee2e JB |
5843 | * This is a little awful but should be ok, we need to make sure that |
5844 | * the eb doesn't disappear out from under us while we're looking at | |
5845 | * this page. | |
19fe0a8b | 5846 | */ |
3083ee2e | 5847 | spin_lock(&eb->refs_lock); |
0b32f4bb | 5848 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
5849 | spin_unlock(&eb->refs_lock); |
5850 | spin_unlock(&page->mapping->private_lock); | |
5851 | return 0; | |
b9473439 | 5852 | } |
3083ee2e | 5853 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 5854 | |
19fe0a8b | 5855 | /* |
3083ee2e JB |
5856 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
5857 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 5858 | */ |
3083ee2e JB |
5859 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
5860 | spin_unlock(&eb->refs_lock); | |
5861 | return 0; | |
b9473439 | 5862 | } |
19fe0a8b | 5863 | |
f7a52a40 | 5864 | return release_extent_buffer(eb); |
6af118ce | 5865 | } |