]>
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 | ||
bb58eb9e QW |
150 | static int __must_check submit_one_bio(struct bio *bio, int mirror_num, |
151 | unsigned long bio_flags) | |
152 | { | |
153 | blk_status_t ret = 0; | |
154 | struct bio_vec *bvec = bio_last_bvec_all(bio); | |
80201fe1 | 155 | struct bio_vec bv; |
bb58eb9e QW |
156 | struct extent_io_tree *tree = bio->bi_private; |
157 | u64 start; | |
158 | ||
80201fe1 LT |
159 | mp_bvec_last_segment(bvec, &bv); |
160 | start = page_offset(bv.bv_page) + bv.bv_offset; | |
bb58eb9e QW |
161 | |
162 | bio->bi_private = NULL; | |
163 | ||
164 | if (tree->ops) | |
165 | ret = tree->ops->submit_bio_hook(tree->private_data, bio, | |
166 | mirror_num, bio_flags, start); | |
167 | else | |
168 | btrfsic_submit_bio(bio); | |
169 | ||
170 | return blk_status_to_errno(ret); | |
171 | } | |
172 | ||
173 | static void flush_write_bio(struct extent_page_data *epd) | |
174 | { | |
175 | if (epd->bio) { | |
176 | int ret; | |
177 | ||
178 | ret = submit_one_bio(epd->bio, 0, 0); | |
179 | BUG_ON(ret < 0); /* -ENOMEM */ | |
180 | epd->bio = NULL; | |
181 | } | |
182 | } | |
e2932ee0 | 183 | |
d1310b2e CM |
184 | int __init extent_io_init(void) |
185 | { | |
837e1972 | 186 | extent_state_cache = kmem_cache_create("btrfs_extent_state", |
9601e3f6 | 187 | sizeof(struct extent_state), 0, |
fba4b697 | 188 | SLAB_MEM_SPREAD, NULL); |
d1310b2e CM |
189 | if (!extent_state_cache) |
190 | return -ENOMEM; | |
191 | ||
837e1972 | 192 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 | 193 | sizeof(struct extent_buffer), 0, |
fba4b697 | 194 | SLAB_MEM_SPREAD, NULL); |
d1310b2e CM |
195 | if (!extent_buffer_cache) |
196 | goto free_state_cache; | |
9be3395b | 197 | |
8ac9f7c1 KO |
198 | if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE, |
199 | offsetof(struct btrfs_io_bio, bio), | |
200 | BIOSET_NEED_BVECS)) | |
9be3395b | 201 | goto free_buffer_cache; |
b208c2f7 | 202 | |
8ac9f7c1 | 203 | if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE)) |
b208c2f7 DW |
204 | goto free_bioset; |
205 | ||
d1310b2e CM |
206 | return 0; |
207 | ||
b208c2f7 | 208 | free_bioset: |
8ac9f7c1 | 209 | bioset_exit(&btrfs_bioset); |
b208c2f7 | 210 | |
9be3395b CM |
211 | free_buffer_cache: |
212 | kmem_cache_destroy(extent_buffer_cache); | |
213 | extent_buffer_cache = NULL; | |
214 | ||
d1310b2e CM |
215 | free_state_cache: |
216 | kmem_cache_destroy(extent_state_cache); | |
9be3395b | 217 | extent_state_cache = NULL; |
d1310b2e CM |
218 | return -ENOMEM; |
219 | } | |
220 | ||
e67c718b | 221 | void __cold extent_io_exit(void) |
d1310b2e | 222 | { |
6d49ba1b | 223 | btrfs_leak_debug_check(); |
8c0a8537 KS |
224 | |
225 | /* | |
226 | * Make sure all delayed rcu free are flushed before we | |
227 | * destroy caches. | |
228 | */ | |
229 | rcu_barrier(); | |
5598e900 KM |
230 | kmem_cache_destroy(extent_state_cache); |
231 | kmem_cache_destroy(extent_buffer_cache); | |
8ac9f7c1 | 232 | bioset_exit(&btrfs_bioset); |
d1310b2e CM |
233 | } |
234 | ||
c258d6e3 | 235 | void extent_io_tree_init(struct btrfs_fs_info *fs_info, |
43eb5f29 QW |
236 | struct extent_io_tree *tree, unsigned int owner, |
237 | void *private_data) | |
d1310b2e | 238 | { |
c258d6e3 | 239 | tree->fs_info = fs_info; |
6bef4d31 | 240 | tree->state = RB_ROOT; |
d1310b2e CM |
241 | tree->ops = NULL; |
242 | tree->dirty_bytes = 0; | |
70dec807 | 243 | spin_lock_init(&tree->lock); |
c6100a4b | 244 | tree->private_data = private_data; |
43eb5f29 | 245 | tree->owner = owner; |
d1310b2e | 246 | } |
d1310b2e | 247 | |
b2950863 | 248 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
249 | { |
250 | struct extent_state *state; | |
d1310b2e | 251 | |
3ba7ab22 MH |
252 | /* |
253 | * The given mask might be not appropriate for the slab allocator, | |
254 | * drop the unsupported bits | |
255 | */ | |
256 | mask &= ~(__GFP_DMA32|__GFP_HIGHMEM); | |
d1310b2e | 257 | state = kmem_cache_alloc(extent_state_cache, mask); |
2b114d1d | 258 | if (!state) |
d1310b2e CM |
259 | return state; |
260 | state->state = 0; | |
47dc196a | 261 | state->failrec = NULL; |
27a3507d | 262 | RB_CLEAR_NODE(&state->rb_node); |
6d49ba1b | 263 | btrfs_leak_debug_add(&state->leak_list, &states); |
b7ac31b7 | 264 | refcount_set(&state->refs, 1); |
d1310b2e | 265 | init_waitqueue_head(&state->wq); |
143bede5 | 266 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
267 | return state; |
268 | } | |
d1310b2e | 269 | |
4845e44f | 270 | void free_extent_state(struct extent_state *state) |
d1310b2e | 271 | { |
d1310b2e CM |
272 | if (!state) |
273 | return; | |
b7ac31b7 | 274 | if (refcount_dec_and_test(&state->refs)) { |
27a3507d | 275 | WARN_ON(extent_state_in_tree(state)); |
6d49ba1b | 276 | btrfs_leak_debug_del(&state->leak_list); |
143bede5 | 277 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
278 | kmem_cache_free(extent_state_cache, state); |
279 | } | |
280 | } | |
d1310b2e | 281 | |
f2071b21 FM |
282 | static struct rb_node *tree_insert(struct rb_root *root, |
283 | struct rb_node *search_start, | |
284 | u64 offset, | |
12cfbad9 FDBM |
285 | struct rb_node *node, |
286 | struct rb_node ***p_in, | |
287 | struct rb_node **parent_in) | |
d1310b2e | 288 | { |
f2071b21 | 289 | struct rb_node **p; |
d397712b | 290 | struct rb_node *parent = NULL; |
d1310b2e CM |
291 | struct tree_entry *entry; |
292 | ||
12cfbad9 FDBM |
293 | if (p_in && parent_in) { |
294 | p = *p_in; | |
295 | parent = *parent_in; | |
296 | goto do_insert; | |
297 | } | |
298 | ||
f2071b21 | 299 | p = search_start ? &search_start : &root->rb_node; |
d397712b | 300 | while (*p) { |
d1310b2e CM |
301 | parent = *p; |
302 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
303 | ||
304 | if (offset < entry->start) | |
305 | p = &(*p)->rb_left; | |
306 | else if (offset > entry->end) | |
307 | p = &(*p)->rb_right; | |
308 | else | |
309 | return parent; | |
310 | } | |
311 | ||
12cfbad9 | 312 | do_insert: |
d1310b2e CM |
313 | rb_link_node(node, parent, p); |
314 | rb_insert_color(node, root); | |
315 | return NULL; | |
316 | } | |
317 | ||
80ea96b1 | 318 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
12cfbad9 | 319 | struct rb_node **next_ret, |
352646c7 | 320 | struct rb_node **prev_ret, |
12cfbad9 FDBM |
321 | struct rb_node ***p_ret, |
322 | struct rb_node **parent_ret) | |
d1310b2e | 323 | { |
80ea96b1 | 324 | struct rb_root *root = &tree->state; |
12cfbad9 | 325 | struct rb_node **n = &root->rb_node; |
d1310b2e CM |
326 | struct rb_node *prev = NULL; |
327 | struct rb_node *orig_prev = NULL; | |
328 | struct tree_entry *entry; | |
329 | struct tree_entry *prev_entry = NULL; | |
330 | ||
12cfbad9 FDBM |
331 | while (*n) { |
332 | prev = *n; | |
333 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
d1310b2e CM |
334 | prev_entry = entry; |
335 | ||
336 | if (offset < entry->start) | |
12cfbad9 | 337 | n = &(*n)->rb_left; |
d1310b2e | 338 | else if (offset > entry->end) |
12cfbad9 | 339 | n = &(*n)->rb_right; |
d397712b | 340 | else |
12cfbad9 | 341 | return *n; |
d1310b2e CM |
342 | } |
343 | ||
12cfbad9 FDBM |
344 | if (p_ret) |
345 | *p_ret = n; | |
346 | if (parent_ret) | |
347 | *parent_ret = prev; | |
348 | ||
352646c7 | 349 | if (next_ret) { |
d1310b2e | 350 | orig_prev = prev; |
d397712b | 351 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
352 | prev = rb_next(prev); |
353 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
354 | } | |
352646c7 | 355 | *next_ret = prev; |
d1310b2e CM |
356 | prev = orig_prev; |
357 | } | |
358 | ||
352646c7 | 359 | if (prev_ret) { |
d1310b2e | 360 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); |
d397712b | 361 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
362 | prev = rb_prev(prev); |
363 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
364 | } | |
352646c7 | 365 | *prev_ret = prev; |
d1310b2e CM |
366 | } |
367 | return NULL; | |
368 | } | |
369 | ||
12cfbad9 FDBM |
370 | static inline struct rb_node * |
371 | tree_search_for_insert(struct extent_io_tree *tree, | |
372 | u64 offset, | |
373 | struct rb_node ***p_ret, | |
374 | struct rb_node **parent_ret) | |
d1310b2e | 375 | { |
352646c7 | 376 | struct rb_node *next= NULL; |
d1310b2e | 377 | struct rb_node *ret; |
70dec807 | 378 | |
352646c7 | 379 | ret = __etree_search(tree, offset, &next, NULL, p_ret, parent_ret); |
d397712b | 380 | if (!ret) |
352646c7 | 381 | return next; |
d1310b2e CM |
382 | return ret; |
383 | } | |
384 | ||
12cfbad9 FDBM |
385 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
386 | u64 offset) | |
387 | { | |
388 | return tree_search_for_insert(tree, offset, NULL, NULL); | |
389 | } | |
390 | ||
d1310b2e CM |
391 | /* |
392 | * utility function to look for merge candidates inside a given range. | |
393 | * Any extents with matching state are merged together into a single | |
394 | * extent in the tree. Extents with EXTENT_IO in their state field | |
395 | * are not merged because the end_io handlers need to be able to do | |
396 | * operations on them without sleeping (or doing allocations/splits). | |
397 | * | |
398 | * This should be called with the tree lock held. | |
399 | */ | |
1bf85046 JM |
400 | static void merge_state(struct extent_io_tree *tree, |
401 | struct extent_state *state) | |
d1310b2e CM |
402 | { |
403 | struct extent_state *other; | |
404 | struct rb_node *other_node; | |
405 | ||
8882679e | 406 | if (state->state & (EXTENT_LOCKED | EXTENT_BOUNDARY)) |
1bf85046 | 407 | return; |
d1310b2e CM |
408 | |
409 | other_node = rb_prev(&state->rb_node); | |
410 | if (other_node) { | |
411 | other = rb_entry(other_node, struct extent_state, rb_node); | |
412 | if (other->end == state->start - 1 && | |
413 | other->state == state->state) { | |
5c848198 NB |
414 | if (tree->private_data && |
415 | is_data_inode(tree->private_data)) | |
416 | btrfs_merge_delalloc_extent(tree->private_data, | |
417 | state, other); | |
d1310b2e | 418 | state->start = other->start; |
d1310b2e | 419 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 420 | RB_CLEAR_NODE(&other->rb_node); |
d1310b2e CM |
421 | free_extent_state(other); |
422 | } | |
423 | } | |
424 | other_node = rb_next(&state->rb_node); | |
425 | if (other_node) { | |
426 | other = rb_entry(other_node, struct extent_state, rb_node); | |
427 | if (other->start == state->end + 1 && | |
428 | other->state == state->state) { | |
5c848198 NB |
429 | if (tree->private_data && |
430 | is_data_inode(tree->private_data)) | |
431 | btrfs_merge_delalloc_extent(tree->private_data, | |
432 | state, other); | |
df98b6e2 | 433 | state->end = other->end; |
df98b6e2 | 434 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 435 | RB_CLEAR_NODE(&other->rb_node); |
df98b6e2 | 436 | free_extent_state(other); |
d1310b2e CM |
437 | } |
438 | } | |
d1310b2e CM |
439 | } |
440 | ||
3150b699 | 441 | static void set_state_bits(struct extent_io_tree *tree, |
d38ed27f QW |
442 | struct extent_state *state, unsigned *bits, |
443 | struct extent_changeset *changeset); | |
3150b699 | 444 | |
d1310b2e CM |
445 | /* |
446 | * insert an extent_state struct into the tree. 'bits' are set on the | |
447 | * struct before it is inserted. | |
448 | * | |
449 | * This may return -EEXIST if the extent is already there, in which case the | |
450 | * state struct is freed. | |
451 | * | |
452 | * The tree lock is not taken internally. This is a utility function and | |
453 | * probably isn't what you want to call (see set/clear_extent_bit). | |
454 | */ | |
455 | static int insert_state(struct extent_io_tree *tree, | |
456 | struct extent_state *state, u64 start, u64 end, | |
12cfbad9 FDBM |
457 | struct rb_node ***p, |
458 | struct rb_node **parent, | |
d38ed27f | 459 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e CM |
460 | { |
461 | struct rb_node *node; | |
462 | ||
31b1a2bd | 463 | if (end < start) |
efe120a0 | 464 | WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n", |
c1c9ff7c | 465 | end, start); |
d1310b2e CM |
466 | state->start = start; |
467 | state->end = end; | |
9ed74f2d | 468 | |
d38ed27f | 469 | set_state_bits(tree, state, bits, changeset); |
3150b699 | 470 | |
f2071b21 | 471 | node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent); |
d1310b2e CM |
472 | if (node) { |
473 | struct extent_state *found; | |
474 | found = rb_entry(node, struct extent_state, rb_node); | |
62e85577 | 475 | pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n", |
c1c9ff7c | 476 | found->start, found->end, start, end); |
d1310b2e CM |
477 | return -EEXIST; |
478 | } | |
479 | merge_state(tree, state); | |
480 | return 0; | |
481 | } | |
482 | ||
483 | /* | |
484 | * split a given extent state struct in two, inserting the preallocated | |
485 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
486 | * offset inside 'orig' where it should be split. | |
487 | * | |
488 | * Before calling, | |
489 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
490 | * are two extent state structs in the tree: | |
491 | * prealloc: [orig->start, split - 1] | |
492 | * orig: [ split, orig->end ] | |
493 | * | |
494 | * The tree locks are not taken by this function. They need to be held | |
495 | * by the caller. | |
496 | */ | |
497 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
498 | struct extent_state *prealloc, u64 split) | |
499 | { | |
500 | struct rb_node *node; | |
9ed74f2d | 501 | |
abbb55f4 NB |
502 | if (tree->private_data && is_data_inode(tree->private_data)) |
503 | btrfs_split_delalloc_extent(tree->private_data, orig, split); | |
9ed74f2d | 504 | |
d1310b2e CM |
505 | prealloc->start = orig->start; |
506 | prealloc->end = split - 1; | |
507 | prealloc->state = orig->state; | |
508 | orig->start = split; | |
509 | ||
f2071b21 FM |
510 | node = tree_insert(&tree->state, &orig->rb_node, prealloc->end, |
511 | &prealloc->rb_node, NULL, NULL); | |
d1310b2e | 512 | if (node) { |
d1310b2e CM |
513 | free_extent_state(prealloc); |
514 | return -EEXIST; | |
515 | } | |
516 | return 0; | |
517 | } | |
518 | ||
cdc6a395 LZ |
519 | static struct extent_state *next_state(struct extent_state *state) |
520 | { | |
521 | struct rb_node *next = rb_next(&state->rb_node); | |
522 | if (next) | |
523 | return rb_entry(next, struct extent_state, rb_node); | |
524 | else | |
525 | return NULL; | |
526 | } | |
527 | ||
d1310b2e CM |
528 | /* |
529 | * utility function to clear some bits in an extent state struct. | |
52042d8e | 530 | * it will optionally wake up anyone waiting on this state (wake == 1). |
d1310b2e CM |
531 | * |
532 | * If no bits are set on the state struct after clearing things, the | |
533 | * struct is freed and removed from the tree | |
534 | */ | |
cdc6a395 LZ |
535 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
536 | struct extent_state *state, | |
fefdc557 QW |
537 | unsigned *bits, int wake, |
538 | struct extent_changeset *changeset) | |
d1310b2e | 539 | { |
cdc6a395 | 540 | struct extent_state *next; |
9ee49a04 | 541 | unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS; |
57599c7e | 542 | int ret; |
d1310b2e | 543 | |
0ca1f7ce | 544 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
545 | u64 range = state->end - state->start + 1; |
546 | WARN_ON(range > tree->dirty_bytes); | |
547 | tree->dirty_bytes -= range; | |
548 | } | |
a36bb5f9 NB |
549 | |
550 | if (tree->private_data && is_data_inode(tree->private_data)) | |
551 | btrfs_clear_delalloc_extent(tree->private_data, state, bits); | |
552 | ||
57599c7e DS |
553 | ret = add_extent_changeset(state, bits_to_clear, changeset, 0); |
554 | BUG_ON(ret < 0); | |
32c00aff | 555 | state->state &= ~bits_to_clear; |
d1310b2e CM |
556 | if (wake) |
557 | wake_up(&state->wq); | |
0ca1f7ce | 558 | if (state->state == 0) { |
cdc6a395 | 559 | next = next_state(state); |
27a3507d | 560 | if (extent_state_in_tree(state)) { |
d1310b2e | 561 | rb_erase(&state->rb_node, &tree->state); |
27a3507d | 562 | RB_CLEAR_NODE(&state->rb_node); |
d1310b2e CM |
563 | free_extent_state(state); |
564 | } else { | |
565 | WARN_ON(1); | |
566 | } | |
567 | } else { | |
568 | merge_state(tree, state); | |
cdc6a395 | 569 | next = next_state(state); |
d1310b2e | 570 | } |
cdc6a395 | 571 | return next; |
d1310b2e CM |
572 | } |
573 | ||
8233767a XG |
574 | static struct extent_state * |
575 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
576 | { | |
577 | if (!prealloc) | |
578 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
579 | ||
580 | return prealloc; | |
581 | } | |
582 | ||
48a3b636 | 583 | static void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
c2d904e0 | 584 | { |
05912a3c DS |
585 | struct inode *inode = tree->private_data; |
586 | ||
587 | btrfs_panic(btrfs_sb(inode->i_sb), err, | |
588 | "locking error: extent tree was modified by another thread while locked"); | |
c2d904e0 JM |
589 | } |
590 | ||
d1310b2e CM |
591 | /* |
592 | * clear some bits on a range in the tree. This may require splitting | |
593 | * or inserting elements in the tree, so the gfp mask is used to | |
594 | * indicate which allocations or sleeping are allowed. | |
595 | * | |
596 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
597 | * the given range from the tree regardless of state (ie for truncate). | |
598 | * | |
599 | * the range [start, end] is inclusive. | |
600 | * | |
6763af84 | 601 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e | 602 | */ |
66b0c887 | 603 | int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
fefdc557 QW |
604 | unsigned bits, int wake, int delete, |
605 | struct extent_state **cached_state, | |
606 | gfp_t mask, struct extent_changeset *changeset) | |
d1310b2e CM |
607 | { |
608 | struct extent_state *state; | |
2c64c53d | 609 | struct extent_state *cached; |
d1310b2e CM |
610 | struct extent_state *prealloc = NULL; |
611 | struct rb_node *node; | |
5c939df5 | 612 | u64 last_end; |
d1310b2e | 613 | int err; |
2ac55d41 | 614 | int clear = 0; |
d1310b2e | 615 | |
a5dee37d | 616 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 | 617 | trace_btrfs_clear_extent_bit(tree, start, end - start + 1, bits); |
8d599ae1 | 618 | |
7ee9e440 JB |
619 | if (bits & EXTENT_DELALLOC) |
620 | bits |= EXTENT_NORESERVE; | |
621 | ||
0ca1f7ce YZ |
622 | if (delete) |
623 | bits |= ~EXTENT_CTLBITS; | |
0ca1f7ce | 624 | |
8882679e | 625 | if (bits & (EXTENT_LOCKED | EXTENT_BOUNDARY)) |
2ac55d41 | 626 | clear = 1; |
d1310b2e | 627 | again: |
d0164adc | 628 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
c7bc6319 FM |
629 | /* |
630 | * Don't care for allocation failure here because we might end | |
631 | * up not needing the pre-allocated extent state at all, which | |
632 | * is the case if we only have in the tree extent states that | |
633 | * cover our input range and don't cover too any other range. | |
634 | * If we end up needing a new extent state we allocate it later. | |
635 | */ | |
d1310b2e | 636 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
637 | } |
638 | ||
cad321ad | 639 | spin_lock(&tree->lock); |
2c64c53d CM |
640 | if (cached_state) { |
641 | cached = *cached_state; | |
2ac55d41 JB |
642 | |
643 | if (clear) { | |
644 | *cached_state = NULL; | |
645 | cached_state = NULL; | |
646 | } | |
647 | ||
27a3507d FM |
648 | if (cached && extent_state_in_tree(cached) && |
649 | cached->start <= start && cached->end > start) { | |
2ac55d41 | 650 | if (clear) |
b7ac31b7 | 651 | refcount_dec(&cached->refs); |
2c64c53d | 652 | state = cached; |
42daec29 | 653 | goto hit_next; |
2c64c53d | 654 | } |
2ac55d41 JB |
655 | if (clear) |
656 | free_extent_state(cached); | |
2c64c53d | 657 | } |
d1310b2e CM |
658 | /* |
659 | * this search will find the extents that end after | |
660 | * our range starts | |
661 | */ | |
80ea96b1 | 662 | node = tree_search(tree, start); |
d1310b2e CM |
663 | if (!node) |
664 | goto out; | |
665 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 666 | hit_next: |
d1310b2e CM |
667 | if (state->start > end) |
668 | goto out; | |
669 | WARN_ON(state->end < start); | |
5c939df5 | 670 | last_end = state->end; |
d1310b2e | 671 | |
0449314a | 672 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
673 | if (!(state->state & bits)) { |
674 | state = next_state(state); | |
0449314a | 675 | goto next; |
cdc6a395 | 676 | } |
0449314a | 677 | |
d1310b2e CM |
678 | /* |
679 | * | ---- desired range ---- | | |
680 | * | state | or | |
681 | * | ------------- state -------------- | | |
682 | * | |
683 | * We need to split the extent we found, and may flip | |
684 | * bits on second half. | |
685 | * | |
686 | * If the extent we found extends past our range, we | |
687 | * just split and search again. It'll get split again | |
688 | * the next time though. | |
689 | * | |
690 | * If the extent we found is inside our range, we clear | |
691 | * the desired bit on it. | |
692 | */ | |
693 | ||
694 | if (state->start < start) { | |
8233767a XG |
695 | prealloc = alloc_extent_state_atomic(prealloc); |
696 | BUG_ON(!prealloc); | |
d1310b2e | 697 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
698 | if (err) |
699 | extent_io_tree_panic(tree, err); | |
700 | ||
d1310b2e CM |
701 | prealloc = NULL; |
702 | if (err) | |
703 | goto out; | |
704 | if (state->end <= end) { | |
fefdc557 QW |
705 | state = clear_state_bit(tree, state, &bits, wake, |
706 | changeset); | |
d1ac6e41 | 707 | goto next; |
d1310b2e CM |
708 | } |
709 | goto search_again; | |
710 | } | |
711 | /* | |
712 | * | ---- desired range ---- | | |
713 | * | state | | |
714 | * We need to split the extent, and clear the bit | |
715 | * on the first half | |
716 | */ | |
717 | if (state->start <= end && state->end > end) { | |
8233767a XG |
718 | prealloc = alloc_extent_state_atomic(prealloc); |
719 | BUG_ON(!prealloc); | |
d1310b2e | 720 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
721 | if (err) |
722 | extent_io_tree_panic(tree, err); | |
723 | ||
d1310b2e CM |
724 | if (wake) |
725 | wake_up(&state->wq); | |
42daec29 | 726 | |
fefdc557 | 727 | clear_state_bit(tree, prealloc, &bits, wake, changeset); |
9ed74f2d | 728 | |
d1310b2e CM |
729 | prealloc = NULL; |
730 | goto out; | |
731 | } | |
42daec29 | 732 | |
fefdc557 | 733 | state = clear_state_bit(tree, state, &bits, wake, changeset); |
0449314a | 734 | next: |
5c939df5 YZ |
735 | if (last_end == (u64)-1) |
736 | goto out; | |
737 | start = last_end + 1; | |
cdc6a395 | 738 | if (start <= end && state && !need_resched()) |
692e5759 | 739 | goto hit_next; |
d1310b2e CM |
740 | |
741 | search_again: | |
742 | if (start > end) | |
743 | goto out; | |
cad321ad | 744 | spin_unlock(&tree->lock); |
d0164adc | 745 | if (gfpflags_allow_blocking(mask)) |
d1310b2e CM |
746 | cond_resched(); |
747 | goto again; | |
7ab5cb2a DS |
748 | |
749 | out: | |
750 | spin_unlock(&tree->lock); | |
751 | if (prealloc) | |
752 | free_extent_state(prealloc); | |
753 | ||
754 | return 0; | |
755 | ||
d1310b2e | 756 | } |
d1310b2e | 757 | |
143bede5 JM |
758 | static void wait_on_state(struct extent_io_tree *tree, |
759 | struct extent_state *state) | |
641f5219 CH |
760 | __releases(tree->lock) |
761 | __acquires(tree->lock) | |
d1310b2e CM |
762 | { |
763 | DEFINE_WAIT(wait); | |
764 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 765 | spin_unlock(&tree->lock); |
d1310b2e | 766 | schedule(); |
cad321ad | 767 | spin_lock(&tree->lock); |
d1310b2e | 768 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
769 | } |
770 | ||
771 | /* | |
772 | * waits for one or more bits to clear on a range in the state tree. | |
773 | * The range [start, end] is inclusive. | |
774 | * The tree lock is taken by this function | |
775 | */ | |
41074888 DS |
776 | static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
777 | unsigned long bits) | |
d1310b2e CM |
778 | { |
779 | struct extent_state *state; | |
780 | struct rb_node *node; | |
781 | ||
a5dee37d | 782 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 783 | |
cad321ad | 784 | spin_lock(&tree->lock); |
d1310b2e CM |
785 | again: |
786 | while (1) { | |
787 | /* | |
788 | * this search will find all the extents that end after | |
789 | * our range starts | |
790 | */ | |
80ea96b1 | 791 | node = tree_search(tree, start); |
c50d3e71 | 792 | process_node: |
d1310b2e CM |
793 | if (!node) |
794 | break; | |
795 | ||
796 | state = rb_entry(node, struct extent_state, rb_node); | |
797 | ||
798 | if (state->start > end) | |
799 | goto out; | |
800 | ||
801 | if (state->state & bits) { | |
802 | start = state->start; | |
b7ac31b7 | 803 | refcount_inc(&state->refs); |
d1310b2e CM |
804 | wait_on_state(tree, state); |
805 | free_extent_state(state); | |
806 | goto again; | |
807 | } | |
808 | start = state->end + 1; | |
809 | ||
810 | if (start > end) | |
811 | break; | |
812 | ||
c50d3e71 FM |
813 | if (!cond_resched_lock(&tree->lock)) { |
814 | node = rb_next(node); | |
815 | goto process_node; | |
816 | } | |
d1310b2e CM |
817 | } |
818 | out: | |
cad321ad | 819 | spin_unlock(&tree->lock); |
d1310b2e | 820 | } |
d1310b2e | 821 | |
1bf85046 | 822 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 823 | struct extent_state *state, |
d38ed27f | 824 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e | 825 | { |
9ee49a04 | 826 | unsigned bits_to_set = *bits & ~EXTENT_CTLBITS; |
57599c7e | 827 | int ret; |
9ed74f2d | 828 | |
e06a1fc9 NB |
829 | if (tree->private_data && is_data_inode(tree->private_data)) |
830 | btrfs_set_delalloc_extent(tree->private_data, state, bits); | |
831 | ||
0ca1f7ce | 832 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
833 | u64 range = state->end - state->start + 1; |
834 | tree->dirty_bytes += range; | |
835 | } | |
57599c7e DS |
836 | ret = add_extent_changeset(state, bits_to_set, changeset, 1); |
837 | BUG_ON(ret < 0); | |
0ca1f7ce | 838 | state->state |= bits_to_set; |
d1310b2e CM |
839 | } |
840 | ||
e38e2ed7 FM |
841 | static void cache_state_if_flags(struct extent_state *state, |
842 | struct extent_state **cached_ptr, | |
9ee49a04 | 843 | unsigned flags) |
2c64c53d CM |
844 | { |
845 | if (cached_ptr && !(*cached_ptr)) { | |
e38e2ed7 | 846 | if (!flags || (state->state & flags)) { |
2c64c53d | 847 | *cached_ptr = state; |
b7ac31b7 | 848 | refcount_inc(&state->refs); |
2c64c53d CM |
849 | } |
850 | } | |
851 | } | |
852 | ||
e38e2ed7 FM |
853 | static void cache_state(struct extent_state *state, |
854 | struct extent_state **cached_ptr) | |
855 | { | |
856 | return cache_state_if_flags(state, cached_ptr, | |
8882679e | 857 | EXTENT_LOCKED | EXTENT_BOUNDARY); |
e38e2ed7 FM |
858 | } |
859 | ||
d1310b2e | 860 | /* |
1edbb734 CM |
861 | * set some bits on a range in the tree. This may require allocations or |
862 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 863 | * |
1edbb734 CM |
864 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
865 | * part of the range already has the desired bits set. The start of the | |
866 | * existing range is returned in failed_start in this case. | |
d1310b2e | 867 | * |
1edbb734 | 868 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 869 | */ |
1edbb734 | 870 | |
3fbe5c02 JM |
871 | static int __must_check |
872 | __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 873 | unsigned bits, unsigned exclusive_bits, |
41074888 | 874 | u64 *failed_start, struct extent_state **cached_state, |
d38ed27f | 875 | gfp_t mask, struct extent_changeset *changeset) |
d1310b2e CM |
876 | { |
877 | struct extent_state *state; | |
878 | struct extent_state *prealloc = NULL; | |
879 | struct rb_node *node; | |
12cfbad9 FDBM |
880 | struct rb_node **p; |
881 | struct rb_node *parent; | |
d1310b2e | 882 | int err = 0; |
d1310b2e CM |
883 | u64 last_start; |
884 | u64 last_end; | |
42daec29 | 885 | |
a5dee37d | 886 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 | 887 | trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits); |
8d599ae1 | 888 | |
d1310b2e | 889 | again: |
d0164adc | 890 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
059f791c DS |
891 | /* |
892 | * Don't care for allocation failure here because we might end | |
893 | * up not needing the pre-allocated extent state at all, which | |
894 | * is the case if we only have in the tree extent states that | |
895 | * cover our input range and don't cover too any other range. | |
896 | * If we end up needing a new extent state we allocate it later. | |
897 | */ | |
d1310b2e | 898 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
899 | } |
900 | ||
cad321ad | 901 | spin_lock(&tree->lock); |
9655d298 CM |
902 | if (cached_state && *cached_state) { |
903 | state = *cached_state; | |
df98b6e2 | 904 | if (state->start <= start && state->end > start && |
27a3507d | 905 | extent_state_in_tree(state)) { |
9655d298 CM |
906 | node = &state->rb_node; |
907 | goto hit_next; | |
908 | } | |
909 | } | |
d1310b2e CM |
910 | /* |
911 | * this search will find all the extents that end after | |
912 | * our range starts. | |
913 | */ | |
12cfbad9 | 914 | node = tree_search_for_insert(tree, start, &p, &parent); |
d1310b2e | 915 | if (!node) { |
8233767a XG |
916 | prealloc = alloc_extent_state_atomic(prealloc); |
917 | BUG_ON(!prealloc); | |
12cfbad9 | 918 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 919 | &p, &parent, &bits, changeset); |
c2d904e0 JM |
920 | if (err) |
921 | extent_io_tree_panic(tree, err); | |
922 | ||
c42ac0bc | 923 | cache_state(prealloc, cached_state); |
d1310b2e | 924 | prealloc = NULL; |
d1310b2e CM |
925 | goto out; |
926 | } | |
d1310b2e | 927 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 928 | hit_next: |
d1310b2e CM |
929 | last_start = state->start; |
930 | last_end = state->end; | |
931 | ||
932 | /* | |
933 | * | ---- desired range ---- | | |
934 | * | state | | |
935 | * | |
936 | * Just lock what we found and keep going | |
937 | */ | |
938 | if (state->start == start && state->end <= end) { | |
1edbb734 | 939 | if (state->state & exclusive_bits) { |
d1310b2e CM |
940 | *failed_start = state->start; |
941 | err = -EEXIST; | |
942 | goto out; | |
943 | } | |
42daec29 | 944 | |
d38ed27f | 945 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 946 | cache_state(state, cached_state); |
d1310b2e | 947 | merge_state(tree, state); |
5c939df5 YZ |
948 | if (last_end == (u64)-1) |
949 | goto out; | |
950 | start = last_end + 1; | |
d1ac6e41 LB |
951 | state = next_state(state); |
952 | if (start < end && state && state->start == start && | |
953 | !need_resched()) | |
954 | goto hit_next; | |
d1310b2e CM |
955 | goto search_again; |
956 | } | |
957 | ||
958 | /* | |
959 | * | ---- desired range ---- | | |
960 | * | state | | |
961 | * or | |
962 | * | ------------- state -------------- | | |
963 | * | |
964 | * We need to split the extent we found, and may flip bits on | |
965 | * second half. | |
966 | * | |
967 | * If the extent we found extends past our | |
968 | * range, we just split and search again. It'll get split | |
969 | * again the next time though. | |
970 | * | |
971 | * If the extent we found is inside our range, we set the | |
972 | * desired bit on it. | |
973 | */ | |
974 | if (state->start < start) { | |
1edbb734 | 975 | if (state->state & exclusive_bits) { |
d1310b2e CM |
976 | *failed_start = start; |
977 | err = -EEXIST; | |
978 | goto out; | |
979 | } | |
8233767a XG |
980 | |
981 | prealloc = alloc_extent_state_atomic(prealloc); | |
982 | BUG_ON(!prealloc); | |
d1310b2e | 983 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
984 | if (err) |
985 | extent_io_tree_panic(tree, err); | |
986 | ||
d1310b2e CM |
987 | prealloc = NULL; |
988 | if (err) | |
989 | goto out; | |
990 | if (state->end <= end) { | |
d38ed27f | 991 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 992 | cache_state(state, cached_state); |
d1310b2e | 993 | merge_state(tree, state); |
5c939df5 YZ |
994 | if (last_end == (u64)-1) |
995 | goto out; | |
996 | start = last_end + 1; | |
d1ac6e41 LB |
997 | state = next_state(state); |
998 | if (start < end && state && state->start == start && | |
999 | !need_resched()) | |
1000 | goto hit_next; | |
d1310b2e CM |
1001 | } |
1002 | goto search_again; | |
1003 | } | |
1004 | /* | |
1005 | * | ---- desired range ---- | | |
1006 | * | state | or | state | | |
1007 | * | |
1008 | * There's a hole, we need to insert something in it and | |
1009 | * ignore the extent we found. | |
1010 | */ | |
1011 | if (state->start > start) { | |
1012 | u64 this_end; | |
1013 | if (end < last_start) | |
1014 | this_end = end; | |
1015 | else | |
d397712b | 1016 | this_end = last_start - 1; |
8233767a XG |
1017 | |
1018 | prealloc = alloc_extent_state_atomic(prealloc); | |
1019 | BUG_ON(!prealloc); | |
c7f895a2 XG |
1020 | |
1021 | /* | |
1022 | * Avoid to free 'prealloc' if it can be merged with | |
1023 | * the later extent. | |
1024 | */ | |
d1310b2e | 1025 | err = insert_state(tree, prealloc, start, this_end, |
d38ed27f | 1026 | NULL, NULL, &bits, changeset); |
c2d904e0 JM |
1027 | if (err) |
1028 | extent_io_tree_panic(tree, err); | |
1029 | ||
9ed74f2d JB |
1030 | cache_state(prealloc, cached_state); |
1031 | prealloc = NULL; | |
d1310b2e CM |
1032 | start = this_end + 1; |
1033 | goto search_again; | |
1034 | } | |
1035 | /* | |
1036 | * | ---- desired range ---- | | |
1037 | * | state | | |
1038 | * We need to split the extent, and set the bit | |
1039 | * on the first half | |
1040 | */ | |
1041 | if (state->start <= end && state->end > end) { | |
1edbb734 | 1042 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1043 | *failed_start = start; |
1044 | err = -EEXIST; | |
1045 | goto out; | |
1046 | } | |
8233767a XG |
1047 | |
1048 | prealloc = alloc_extent_state_atomic(prealloc); | |
1049 | BUG_ON(!prealloc); | |
d1310b2e | 1050 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
1051 | if (err) |
1052 | extent_io_tree_panic(tree, err); | |
d1310b2e | 1053 | |
d38ed27f | 1054 | set_state_bits(tree, prealloc, &bits, changeset); |
2c64c53d | 1055 | cache_state(prealloc, cached_state); |
d1310b2e CM |
1056 | merge_state(tree, prealloc); |
1057 | prealloc = NULL; | |
1058 | goto out; | |
1059 | } | |
1060 | ||
b5a4ba14 DS |
1061 | search_again: |
1062 | if (start > end) | |
1063 | goto out; | |
1064 | spin_unlock(&tree->lock); | |
1065 | if (gfpflags_allow_blocking(mask)) | |
1066 | cond_resched(); | |
1067 | goto again; | |
d1310b2e CM |
1068 | |
1069 | out: | |
cad321ad | 1070 | spin_unlock(&tree->lock); |
d1310b2e CM |
1071 | if (prealloc) |
1072 | free_extent_state(prealloc); | |
1073 | ||
1074 | return err; | |
1075 | ||
d1310b2e | 1076 | } |
d1310b2e | 1077 | |
41074888 | 1078 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
9ee49a04 | 1079 | unsigned bits, u64 * failed_start, |
41074888 | 1080 | struct extent_state **cached_state, gfp_t mask) |
3fbe5c02 JM |
1081 | { |
1082 | return __set_extent_bit(tree, start, end, bits, 0, failed_start, | |
d38ed27f | 1083 | cached_state, mask, NULL); |
3fbe5c02 JM |
1084 | } |
1085 | ||
1086 | ||
462d6fac | 1087 | /** |
10983f2e LB |
1088 | * convert_extent_bit - convert all bits in a given range from one bit to |
1089 | * another | |
462d6fac JB |
1090 | * @tree: the io tree to search |
1091 | * @start: the start offset in bytes | |
1092 | * @end: the end offset in bytes (inclusive) | |
1093 | * @bits: the bits to set in this range | |
1094 | * @clear_bits: the bits to clear in this range | |
e6138876 | 1095 | * @cached_state: state that we're going to cache |
462d6fac JB |
1096 | * |
1097 | * This will go through and set bits for the given range. If any states exist | |
1098 | * already in this range they are set with the given bit and cleared of the | |
1099 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
1100 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
1101 | * boundary bits like LOCK. | |
210aa277 DS |
1102 | * |
1103 | * All allocations are done with GFP_NOFS. | |
462d6fac JB |
1104 | */ |
1105 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1106 | unsigned bits, unsigned clear_bits, |
210aa277 | 1107 | struct extent_state **cached_state) |
462d6fac JB |
1108 | { |
1109 | struct extent_state *state; | |
1110 | struct extent_state *prealloc = NULL; | |
1111 | struct rb_node *node; | |
12cfbad9 FDBM |
1112 | struct rb_node **p; |
1113 | struct rb_node *parent; | |
462d6fac JB |
1114 | int err = 0; |
1115 | u64 last_start; | |
1116 | u64 last_end; | |
c8fd3de7 | 1117 | bool first_iteration = true; |
462d6fac | 1118 | |
a5dee37d | 1119 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 QW |
1120 | trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits, |
1121 | clear_bits); | |
8d599ae1 | 1122 | |
462d6fac | 1123 | again: |
210aa277 | 1124 | if (!prealloc) { |
c8fd3de7 FM |
1125 | /* |
1126 | * Best effort, don't worry if extent state allocation fails | |
1127 | * here for the first iteration. We might have a cached state | |
1128 | * that matches exactly the target range, in which case no | |
1129 | * extent state allocations are needed. We'll only know this | |
1130 | * after locking the tree. | |
1131 | */ | |
210aa277 | 1132 | prealloc = alloc_extent_state(GFP_NOFS); |
c8fd3de7 | 1133 | if (!prealloc && !first_iteration) |
462d6fac JB |
1134 | return -ENOMEM; |
1135 | } | |
1136 | ||
1137 | spin_lock(&tree->lock); | |
e6138876 JB |
1138 | if (cached_state && *cached_state) { |
1139 | state = *cached_state; | |
1140 | if (state->start <= start && state->end > start && | |
27a3507d | 1141 | extent_state_in_tree(state)) { |
e6138876 JB |
1142 | node = &state->rb_node; |
1143 | goto hit_next; | |
1144 | } | |
1145 | } | |
1146 | ||
462d6fac JB |
1147 | /* |
1148 | * this search will find all the extents that end after | |
1149 | * our range starts. | |
1150 | */ | |
12cfbad9 | 1151 | node = tree_search_for_insert(tree, start, &p, &parent); |
462d6fac JB |
1152 | if (!node) { |
1153 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1154 | if (!prealloc) { |
1155 | err = -ENOMEM; | |
1156 | goto out; | |
1157 | } | |
12cfbad9 | 1158 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 1159 | &p, &parent, &bits, NULL); |
c2d904e0 JM |
1160 | if (err) |
1161 | extent_io_tree_panic(tree, err); | |
c42ac0bc FDBM |
1162 | cache_state(prealloc, cached_state); |
1163 | prealloc = NULL; | |
462d6fac JB |
1164 | goto out; |
1165 | } | |
1166 | state = rb_entry(node, struct extent_state, rb_node); | |
1167 | hit_next: | |
1168 | last_start = state->start; | |
1169 | last_end = state->end; | |
1170 | ||
1171 | /* | |
1172 | * | ---- desired range ---- | | |
1173 | * | state | | |
1174 | * | |
1175 | * Just lock what we found and keep going | |
1176 | */ | |
1177 | if (state->start == start && state->end <= end) { | |
d38ed27f | 1178 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1179 | cache_state(state, cached_state); |
fefdc557 | 1180 | state = clear_state_bit(tree, state, &clear_bits, 0, NULL); |
462d6fac JB |
1181 | if (last_end == (u64)-1) |
1182 | goto out; | |
462d6fac | 1183 | start = last_end + 1; |
d1ac6e41 LB |
1184 | if (start < end && state && state->start == start && |
1185 | !need_resched()) | |
1186 | goto hit_next; | |
462d6fac JB |
1187 | goto search_again; |
1188 | } | |
1189 | ||
1190 | /* | |
1191 | * | ---- desired range ---- | | |
1192 | * | state | | |
1193 | * or | |
1194 | * | ------------- state -------------- | | |
1195 | * | |
1196 | * We need to split the extent we found, and may flip bits on | |
1197 | * second half. | |
1198 | * | |
1199 | * If the extent we found extends past our | |
1200 | * range, we just split and search again. It'll get split | |
1201 | * again the next time though. | |
1202 | * | |
1203 | * If the extent we found is inside our range, we set the | |
1204 | * desired bit on it. | |
1205 | */ | |
1206 | if (state->start < start) { | |
1207 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1208 | if (!prealloc) { |
1209 | err = -ENOMEM; | |
1210 | goto out; | |
1211 | } | |
462d6fac | 1212 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1213 | if (err) |
1214 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1215 | prealloc = NULL; |
1216 | if (err) | |
1217 | goto out; | |
1218 | if (state->end <= end) { | |
d38ed27f | 1219 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1220 | cache_state(state, cached_state); |
fefdc557 QW |
1221 | state = clear_state_bit(tree, state, &clear_bits, 0, |
1222 | NULL); | |
462d6fac JB |
1223 | if (last_end == (u64)-1) |
1224 | goto out; | |
1225 | start = last_end + 1; | |
d1ac6e41 LB |
1226 | if (start < end && state && state->start == start && |
1227 | !need_resched()) | |
1228 | goto hit_next; | |
462d6fac JB |
1229 | } |
1230 | goto search_again; | |
1231 | } | |
1232 | /* | |
1233 | * | ---- desired range ---- | | |
1234 | * | state | or | state | | |
1235 | * | |
1236 | * There's a hole, we need to insert something in it and | |
1237 | * ignore the extent we found. | |
1238 | */ | |
1239 | if (state->start > start) { | |
1240 | u64 this_end; | |
1241 | if (end < last_start) | |
1242 | this_end = end; | |
1243 | else | |
1244 | this_end = last_start - 1; | |
1245 | ||
1246 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1247 | if (!prealloc) { |
1248 | err = -ENOMEM; | |
1249 | goto out; | |
1250 | } | |
462d6fac JB |
1251 | |
1252 | /* | |
1253 | * Avoid to free 'prealloc' if it can be merged with | |
1254 | * the later extent. | |
1255 | */ | |
1256 | err = insert_state(tree, prealloc, start, this_end, | |
d38ed27f | 1257 | NULL, NULL, &bits, NULL); |
c2d904e0 JM |
1258 | if (err) |
1259 | extent_io_tree_panic(tree, err); | |
e6138876 | 1260 | cache_state(prealloc, cached_state); |
462d6fac JB |
1261 | prealloc = NULL; |
1262 | start = this_end + 1; | |
1263 | goto search_again; | |
1264 | } | |
1265 | /* | |
1266 | * | ---- desired range ---- | | |
1267 | * | state | | |
1268 | * We need to split the extent, and set the bit | |
1269 | * on the first half | |
1270 | */ | |
1271 | if (state->start <= end && state->end > end) { | |
1272 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1273 | if (!prealloc) { |
1274 | err = -ENOMEM; | |
1275 | goto out; | |
1276 | } | |
462d6fac JB |
1277 | |
1278 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1279 | if (err) |
1280 | extent_io_tree_panic(tree, err); | |
462d6fac | 1281 | |
d38ed27f | 1282 | set_state_bits(tree, prealloc, &bits, NULL); |
e6138876 | 1283 | cache_state(prealloc, cached_state); |
fefdc557 | 1284 | clear_state_bit(tree, prealloc, &clear_bits, 0, NULL); |
462d6fac JB |
1285 | prealloc = NULL; |
1286 | goto out; | |
1287 | } | |
1288 | ||
462d6fac JB |
1289 | search_again: |
1290 | if (start > end) | |
1291 | goto out; | |
1292 | spin_unlock(&tree->lock); | |
210aa277 | 1293 | cond_resched(); |
c8fd3de7 | 1294 | first_iteration = false; |
462d6fac | 1295 | goto again; |
462d6fac JB |
1296 | |
1297 | out: | |
1298 | spin_unlock(&tree->lock); | |
1299 | if (prealloc) | |
1300 | free_extent_state(prealloc); | |
1301 | ||
1302 | return err; | |
462d6fac JB |
1303 | } |
1304 | ||
d1310b2e | 1305 | /* wrappers around set/clear extent bit */ |
d38ed27f | 1306 | int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
2c53b912 | 1307 | unsigned bits, struct extent_changeset *changeset) |
d38ed27f QW |
1308 | { |
1309 | /* | |
1310 | * We don't support EXTENT_LOCKED yet, as current changeset will | |
1311 | * record any bits changed, so for EXTENT_LOCKED case, it will | |
1312 | * either fail with -EEXIST or changeset will record the whole | |
1313 | * range. | |
1314 | */ | |
1315 | BUG_ON(bits & EXTENT_LOCKED); | |
1316 | ||
2c53b912 | 1317 | return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS, |
d38ed27f QW |
1318 | changeset); |
1319 | } | |
1320 | ||
fefdc557 QW |
1321 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
1322 | unsigned bits, int wake, int delete, | |
ae0f1625 | 1323 | struct extent_state **cached) |
fefdc557 QW |
1324 | { |
1325 | return __clear_extent_bit(tree, start, end, bits, wake, delete, | |
ae0f1625 | 1326 | cached, GFP_NOFS, NULL); |
fefdc557 QW |
1327 | } |
1328 | ||
fefdc557 | 1329 | int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
f734c44a | 1330 | unsigned bits, struct extent_changeset *changeset) |
fefdc557 QW |
1331 | { |
1332 | /* | |
1333 | * Don't support EXTENT_LOCKED case, same reason as | |
1334 | * set_record_extent_bits(). | |
1335 | */ | |
1336 | BUG_ON(bits & EXTENT_LOCKED); | |
1337 | ||
f734c44a | 1338 | return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS, |
fefdc557 QW |
1339 | changeset); |
1340 | } | |
1341 | ||
d352ac68 CM |
1342 | /* |
1343 | * either insert or lock state struct between start and end use mask to tell | |
1344 | * us if waiting is desired. | |
1345 | */ | |
1edbb734 | 1346 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
ff13db41 | 1347 | struct extent_state **cached_state) |
d1310b2e CM |
1348 | { |
1349 | int err; | |
1350 | u64 failed_start; | |
9ee49a04 | 1351 | |
d1310b2e | 1352 | while (1) { |
ff13db41 | 1353 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, |
3fbe5c02 | 1354 | EXTENT_LOCKED, &failed_start, |
d38ed27f | 1355 | cached_state, GFP_NOFS, NULL); |
d0082371 | 1356 | if (err == -EEXIST) { |
d1310b2e CM |
1357 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1358 | start = failed_start; | |
d0082371 | 1359 | } else |
d1310b2e | 1360 | break; |
d1310b2e CM |
1361 | WARN_ON(start > end); |
1362 | } | |
1363 | return err; | |
1364 | } | |
d1310b2e | 1365 | |
d0082371 | 1366 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1367 | { |
1368 | int err; | |
1369 | u64 failed_start; | |
1370 | ||
3fbe5c02 | 1371 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
d38ed27f | 1372 | &failed_start, NULL, GFP_NOFS, NULL); |
6643558d YZ |
1373 | if (err == -EEXIST) { |
1374 | if (failed_start > start) | |
1375 | clear_extent_bit(tree, start, failed_start - 1, | |
ae0f1625 | 1376 | EXTENT_LOCKED, 1, 0, NULL); |
25179201 | 1377 | return 0; |
6643558d | 1378 | } |
25179201 JB |
1379 | return 1; |
1380 | } | |
25179201 | 1381 | |
bd1fa4f0 | 1382 | void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1383 | { |
09cbfeaf KS |
1384 | unsigned long index = start >> PAGE_SHIFT; |
1385 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1386 | struct page *page; |
1387 | ||
1388 | while (index <= end_index) { | |
1389 | page = find_get_page(inode->i_mapping, index); | |
1390 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1391 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1392 | put_page(page); |
4adaa611 CM |
1393 | index++; |
1394 | } | |
4adaa611 CM |
1395 | } |
1396 | ||
f6311572 | 1397 | void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1398 | { |
09cbfeaf KS |
1399 | unsigned long index = start >> PAGE_SHIFT; |
1400 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1401 | struct page *page; |
1402 | ||
1403 | while (index <= end_index) { | |
1404 | page = find_get_page(inode->i_mapping, index); | |
1405 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
4adaa611 | 1406 | __set_page_dirty_nobuffers(page); |
8d38633c | 1407 | account_page_redirty(page); |
09cbfeaf | 1408 | put_page(page); |
4adaa611 CM |
1409 | index++; |
1410 | } | |
4adaa611 CM |
1411 | } |
1412 | ||
d352ac68 CM |
1413 | /* find the first state struct with 'bits' set after 'start', and |
1414 | * return it. tree->lock must be held. NULL will returned if | |
1415 | * nothing was found after 'start' | |
1416 | */ | |
48a3b636 ES |
1417 | static struct extent_state * |
1418 | find_first_extent_bit_state(struct extent_io_tree *tree, | |
9ee49a04 | 1419 | u64 start, unsigned bits) |
d7fc640e CM |
1420 | { |
1421 | struct rb_node *node; | |
1422 | struct extent_state *state; | |
1423 | ||
1424 | /* | |
1425 | * this search will find all the extents that end after | |
1426 | * our range starts. | |
1427 | */ | |
1428 | node = tree_search(tree, start); | |
d397712b | 1429 | if (!node) |
d7fc640e | 1430 | goto out; |
d7fc640e | 1431 | |
d397712b | 1432 | while (1) { |
d7fc640e | 1433 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1434 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1435 | return state; |
d397712b | 1436 | |
d7fc640e CM |
1437 | node = rb_next(node); |
1438 | if (!node) | |
1439 | break; | |
1440 | } | |
1441 | out: | |
1442 | return NULL; | |
1443 | } | |
d7fc640e | 1444 | |
69261c4b XG |
1445 | /* |
1446 | * find the first offset in the io tree with 'bits' set. zero is | |
1447 | * returned if we find something, and *start_ret and *end_ret are | |
1448 | * set to reflect the state struct that was found. | |
1449 | * | |
477d7eaf | 1450 | * If nothing was found, 1 is returned. If found something, return 0. |
69261c4b XG |
1451 | */ |
1452 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
9ee49a04 | 1453 | u64 *start_ret, u64 *end_ret, unsigned bits, |
e6138876 | 1454 | struct extent_state **cached_state) |
69261c4b XG |
1455 | { |
1456 | struct extent_state *state; | |
1457 | int ret = 1; | |
1458 | ||
1459 | spin_lock(&tree->lock); | |
e6138876 JB |
1460 | if (cached_state && *cached_state) { |
1461 | state = *cached_state; | |
27a3507d | 1462 | if (state->end == start - 1 && extent_state_in_tree(state)) { |
9688e9a9 | 1463 | while ((state = next_state(state)) != NULL) { |
e6138876 JB |
1464 | if (state->state & bits) |
1465 | goto got_it; | |
e6138876 JB |
1466 | } |
1467 | free_extent_state(*cached_state); | |
1468 | *cached_state = NULL; | |
1469 | goto out; | |
1470 | } | |
1471 | free_extent_state(*cached_state); | |
1472 | *cached_state = NULL; | |
1473 | } | |
1474 | ||
69261c4b | 1475 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1476 | got_it: |
69261c4b | 1477 | if (state) { |
e38e2ed7 | 1478 | cache_state_if_flags(state, cached_state, 0); |
69261c4b XG |
1479 | *start_ret = state->start; |
1480 | *end_ret = state->end; | |
1481 | ret = 0; | |
1482 | } | |
e6138876 | 1483 | out: |
69261c4b XG |
1484 | spin_unlock(&tree->lock); |
1485 | return ret; | |
1486 | } | |
1487 | ||
d352ac68 CM |
1488 | /* |
1489 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1490 | * more than 'max_bytes'. start and end are used to return the range, | |
1491 | * | |
3522e903 | 1492 | * true is returned if we find something, false if nothing was in the tree |
d352ac68 | 1493 | */ |
3522e903 | 1494 | static noinline bool find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1495 | u64 *start, u64 *end, u64 max_bytes, |
1496 | struct extent_state **cached_state) | |
d1310b2e CM |
1497 | { |
1498 | struct rb_node *node; | |
1499 | struct extent_state *state; | |
1500 | u64 cur_start = *start; | |
3522e903 | 1501 | bool found = false; |
d1310b2e CM |
1502 | u64 total_bytes = 0; |
1503 | ||
cad321ad | 1504 | spin_lock(&tree->lock); |
c8b97818 | 1505 | |
d1310b2e CM |
1506 | /* |
1507 | * this search will find all the extents that end after | |
1508 | * our range starts. | |
1509 | */ | |
80ea96b1 | 1510 | node = tree_search(tree, cur_start); |
2b114d1d | 1511 | if (!node) { |
3522e903 | 1512 | *end = (u64)-1; |
d1310b2e CM |
1513 | goto out; |
1514 | } | |
1515 | ||
d397712b | 1516 | while (1) { |
d1310b2e | 1517 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1518 | if (found && (state->start != cur_start || |
1519 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1520 | goto out; |
1521 | } | |
1522 | if (!(state->state & EXTENT_DELALLOC)) { | |
1523 | if (!found) | |
1524 | *end = state->end; | |
1525 | goto out; | |
1526 | } | |
c2a128d2 | 1527 | if (!found) { |
d1310b2e | 1528 | *start = state->start; |
c2a128d2 | 1529 | *cached_state = state; |
b7ac31b7 | 1530 | refcount_inc(&state->refs); |
c2a128d2 | 1531 | } |
3522e903 | 1532 | found = true; |
d1310b2e CM |
1533 | *end = state->end; |
1534 | cur_start = state->end + 1; | |
1535 | node = rb_next(node); | |
d1310b2e | 1536 | total_bytes += state->end - state->start + 1; |
7bf811a5 | 1537 | if (total_bytes >= max_bytes) |
573aecaf | 1538 | break; |
573aecaf | 1539 | if (!node) |
d1310b2e CM |
1540 | break; |
1541 | } | |
1542 | out: | |
cad321ad | 1543 | spin_unlock(&tree->lock); |
d1310b2e CM |
1544 | return found; |
1545 | } | |
1546 | ||
da2c7009 LB |
1547 | static int __process_pages_contig(struct address_space *mapping, |
1548 | struct page *locked_page, | |
1549 | pgoff_t start_index, pgoff_t end_index, | |
1550 | unsigned long page_ops, pgoff_t *index_ret); | |
1551 | ||
143bede5 JM |
1552 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1553 | struct page *locked_page, | |
1554 | u64 start, u64 end) | |
c8b97818 | 1555 | { |
09cbfeaf KS |
1556 | unsigned long index = start >> PAGE_SHIFT; |
1557 | unsigned long end_index = end >> PAGE_SHIFT; | |
c8b97818 | 1558 | |
76c0021d | 1559 | ASSERT(locked_page); |
c8b97818 | 1560 | if (index == locked_page->index && end_index == index) |
143bede5 | 1561 | return; |
c8b97818 | 1562 | |
76c0021d LB |
1563 | __process_pages_contig(inode->i_mapping, locked_page, index, end_index, |
1564 | PAGE_UNLOCK, NULL); | |
c8b97818 CM |
1565 | } |
1566 | ||
1567 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1568 | struct page *locked_page, | |
1569 | u64 delalloc_start, | |
1570 | u64 delalloc_end) | |
1571 | { | |
09cbfeaf | 1572 | unsigned long index = delalloc_start >> PAGE_SHIFT; |
76c0021d | 1573 | unsigned long index_ret = index; |
09cbfeaf | 1574 | unsigned long end_index = delalloc_end >> PAGE_SHIFT; |
c8b97818 | 1575 | int ret; |
c8b97818 | 1576 | |
76c0021d | 1577 | ASSERT(locked_page); |
c8b97818 CM |
1578 | if (index == locked_page->index && index == end_index) |
1579 | return 0; | |
1580 | ||
76c0021d LB |
1581 | ret = __process_pages_contig(inode->i_mapping, locked_page, index, |
1582 | end_index, PAGE_LOCK, &index_ret); | |
1583 | if (ret == -EAGAIN) | |
1584 | __unlock_for_delalloc(inode, locked_page, delalloc_start, | |
1585 | (u64)index_ret << PAGE_SHIFT); | |
c8b97818 CM |
1586 | return ret; |
1587 | } | |
1588 | ||
1589 | /* | |
3522e903 LF |
1590 | * Find and lock a contiguous range of bytes in the file marked as delalloc, no |
1591 | * more than @max_bytes. @Start and @end are used to return the range, | |
c8b97818 | 1592 | * |
3522e903 LF |
1593 | * Return: true if we find something |
1594 | * false if nothing was in the tree | |
c8b97818 | 1595 | */ |
ce9f967f | 1596 | EXPORT_FOR_TESTS |
3522e903 | 1597 | noinline_for_stack bool find_lock_delalloc_range(struct inode *inode, |
294e30fe JB |
1598 | struct extent_io_tree *tree, |
1599 | struct page *locked_page, u64 *start, | |
917aacec | 1600 | u64 *end) |
c8b97818 | 1601 | { |
917aacec | 1602 | u64 max_bytes = BTRFS_MAX_EXTENT_SIZE; |
c8b97818 CM |
1603 | u64 delalloc_start; |
1604 | u64 delalloc_end; | |
3522e903 | 1605 | bool found; |
9655d298 | 1606 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1607 | int ret; |
1608 | int loops = 0; | |
1609 | ||
1610 | again: | |
1611 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1612 | delalloc_start = *start; | |
1613 | delalloc_end = 0; | |
1614 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1615 | max_bytes, &cached_state); |
70b99e69 | 1616 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1617 | *start = delalloc_start; |
1618 | *end = delalloc_end; | |
c2a128d2 | 1619 | free_extent_state(cached_state); |
3522e903 | 1620 | return false; |
c8b97818 CM |
1621 | } |
1622 | ||
70b99e69 CM |
1623 | /* |
1624 | * start comes from the offset of locked_page. We have to lock | |
1625 | * pages in order, so we can't process delalloc bytes before | |
1626 | * locked_page | |
1627 | */ | |
d397712b | 1628 | if (delalloc_start < *start) |
70b99e69 | 1629 | delalloc_start = *start; |
70b99e69 | 1630 | |
c8b97818 CM |
1631 | /* |
1632 | * make sure to limit the number of pages we try to lock down | |
c8b97818 | 1633 | */ |
7bf811a5 JB |
1634 | if (delalloc_end + 1 - delalloc_start > max_bytes) |
1635 | delalloc_end = delalloc_start + max_bytes - 1; | |
d397712b | 1636 | |
c8b97818 CM |
1637 | /* step two, lock all the pages after the page that has start */ |
1638 | ret = lock_delalloc_pages(inode, locked_page, | |
1639 | delalloc_start, delalloc_end); | |
9bfd61d9 | 1640 | ASSERT(!ret || ret == -EAGAIN); |
c8b97818 CM |
1641 | if (ret == -EAGAIN) { |
1642 | /* some of the pages are gone, lets avoid looping by | |
1643 | * shortening the size of the delalloc range we're searching | |
1644 | */ | |
9655d298 | 1645 | free_extent_state(cached_state); |
7d788742 | 1646 | cached_state = NULL; |
c8b97818 | 1647 | if (!loops) { |
09cbfeaf | 1648 | max_bytes = PAGE_SIZE; |
c8b97818 CM |
1649 | loops = 1; |
1650 | goto again; | |
1651 | } else { | |
3522e903 | 1652 | found = false; |
c8b97818 CM |
1653 | goto out_failed; |
1654 | } | |
1655 | } | |
c8b97818 CM |
1656 | |
1657 | /* step three, lock the state bits for the whole range */ | |
ff13db41 | 1658 | lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state); |
c8b97818 CM |
1659 | |
1660 | /* then test to make sure it is all still delalloc */ | |
1661 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1662 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1663 | if (!ret) { |
9655d298 | 1664 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
e43bbe5e | 1665 | &cached_state); |
c8b97818 CM |
1666 | __unlock_for_delalloc(inode, locked_page, |
1667 | delalloc_start, delalloc_end); | |
1668 | cond_resched(); | |
1669 | goto again; | |
1670 | } | |
9655d298 | 1671 | free_extent_state(cached_state); |
c8b97818 CM |
1672 | *start = delalloc_start; |
1673 | *end = delalloc_end; | |
1674 | out_failed: | |
1675 | return found; | |
1676 | } | |
1677 | ||
da2c7009 LB |
1678 | static int __process_pages_contig(struct address_space *mapping, |
1679 | struct page *locked_page, | |
1680 | pgoff_t start_index, pgoff_t end_index, | |
1681 | unsigned long page_ops, pgoff_t *index_ret) | |
c8b97818 | 1682 | { |
873695b3 | 1683 | unsigned long nr_pages = end_index - start_index + 1; |
da2c7009 | 1684 | unsigned long pages_locked = 0; |
873695b3 | 1685 | pgoff_t index = start_index; |
c8b97818 | 1686 | struct page *pages[16]; |
873695b3 | 1687 | unsigned ret; |
da2c7009 | 1688 | int err = 0; |
c8b97818 | 1689 | int i; |
771ed689 | 1690 | |
da2c7009 LB |
1691 | if (page_ops & PAGE_LOCK) { |
1692 | ASSERT(page_ops == PAGE_LOCK); | |
1693 | ASSERT(index_ret && *index_ret == start_index); | |
1694 | } | |
1695 | ||
704de49d | 1696 | if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0) |
873695b3 | 1697 | mapping_set_error(mapping, -EIO); |
704de49d | 1698 | |
d397712b | 1699 | while (nr_pages > 0) { |
873695b3 | 1700 | ret = find_get_pages_contig(mapping, index, |
5b050f04 CM |
1701 | min_t(unsigned long, |
1702 | nr_pages, ARRAY_SIZE(pages)), pages); | |
da2c7009 LB |
1703 | if (ret == 0) { |
1704 | /* | |
1705 | * Only if we're going to lock these pages, | |
1706 | * can we find nothing at @index. | |
1707 | */ | |
1708 | ASSERT(page_ops & PAGE_LOCK); | |
49d4a334 LB |
1709 | err = -EAGAIN; |
1710 | goto out; | |
da2c7009 | 1711 | } |
8b62b72b | 1712 | |
da2c7009 | 1713 | for (i = 0; i < ret; i++) { |
c2790a2e | 1714 | if (page_ops & PAGE_SET_PRIVATE2) |
8b62b72b CM |
1715 | SetPagePrivate2(pages[i]); |
1716 | ||
c8b97818 | 1717 | if (pages[i] == locked_page) { |
09cbfeaf | 1718 | put_page(pages[i]); |
da2c7009 | 1719 | pages_locked++; |
c8b97818 CM |
1720 | continue; |
1721 | } | |
c2790a2e | 1722 | if (page_ops & PAGE_CLEAR_DIRTY) |
c8b97818 | 1723 | clear_page_dirty_for_io(pages[i]); |
c2790a2e | 1724 | if (page_ops & PAGE_SET_WRITEBACK) |
c8b97818 | 1725 | set_page_writeback(pages[i]); |
704de49d FM |
1726 | if (page_ops & PAGE_SET_ERROR) |
1727 | SetPageError(pages[i]); | |
c2790a2e | 1728 | if (page_ops & PAGE_END_WRITEBACK) |
c8b97818 | 1729 | end_page_writeback(pages[i]); |
c2790a2e | 1730 | if (page_ops & PAGE_UNLOCK) |
771ed689 | 1731 | unlock_page(pages[i]); |
da2c7009 LB |
1732 | if (page_ops & PAGE_LOCK) { |
1733 | lock_page(pages[i]); | |
1734 | if (!PageDirty(pages[i]) || | |
1735 | pages[i]->mapping != mapping) { | |
1736 | unlock_page(pages[i]); | |
1737 | put_page(pages[i]); | |
1738 | err = -EAGAIN; | |
1739 | goto out; | |
1740 | } | |
1741 | } | |
09cbfeaf | 1742 | put_page(pages[i]); |
da2c7009 | 1743 | pages_locked++; |
c8b97818 CM |
1744 | } |
1745 | nr_pages -= ret; | |
1746 | index += ret; | |
1747 | cond_resched(); | |
1748 | } | |
da2c7009 LB |
1749 | out: |
1750 | if (err && index_ret) | |
1751 | *index_ret = start_index + pages_locked - 1; | |
1752 | return err; | |
c8b97818 | 1753 | } |
c8b97818 | 1754 | |
873695b3 LB |
1755 | void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end, |
1756 | u64 delalloc_end, struct page *locked_page, | |
1757 | unsigned clear_bits, | |
1758 | unsigned long page_ops) | |
1759 | { | |
1760 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0, | |
ae0f1625 | 1761 | NULL); |
873695b3 LB |
1762 | |
1763 | __process_pages_contig(inode->i_mapping, locked_page, | |
1764 | start >> PAGE_SHIFT, end >> PAGE_SHIFT, | |
da2c7009 | 1765 | page_ops, NULL); |
873695b3 LB |
1766 | } |
1767 | ||
d352ac68 CM |
1768 | /* |
1769 | * count the number of bytes in the tree that have a given bit(s) | |
1770 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1771 | * cached. The total number found is returned. | |
1772 | */ | |
d1310b2e CM |
1773 | u64 count_range_bits(struct extent_io_tree *tree, |
1774 | u64 *start, u64 search_end, u64 max_bytes, | |
9ee49a04 | 1775 | unsigned bits, int contig) |
d1310b2e CM |
1776 | { |
1777 | struct rb_node *node; | |
1778 | struct extent_state *state; | |
1779 | u64 cur_start = *start; | |
1780 | u64 total_bytes = 0; | |
ec29ed5b | 1781 | u64 last = 0; |
d1310b2e CM |
1782 | int found = 0; |
1783 | ||
fae7f21c | 1784 | if (WARN_ON(search_end <= cur_start)) |
d1310b2e | 1785 | return 0; |
d1310b2e | 1786 | |
cad321ad | 1787 | spin_lock(&tree->lock); |
d1310b2e CM |
1788 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1789 | total_bytes = tree->dirty_bytes; | |
1790 | goto out; | |
1791 | } | |
1792 | /* | |
1793 | * this search will find all the extents that end after | |
1794 | * our range starts. | |
1795 | */ | |
80ea96b1 | 1796 | node = tree_search(tree, cur_start); |
d397712b | 1797 | if (!node) |
d1310b2e | 1798 | goto out; |
d1310b2e | 1799 | |
d397712b | 1800 | while (1) { |
d1310b2e CM |
1801 | state = rb_entry(node, struct extent_state, rb_node); |
1802 | if (state->start > search_end) | |
1803 | break; | |
ec29ed5b CM |
1804 | if (contig && found && state->start > last + 1) |
1805 | break; | |
1806 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1807 | total_bytes += min(search_end, state->end) + 1 - |
1808 | max(cur_start, state->start); | |
1809 | if (total_bytes >= max_bytes) | |
1810 | break; | |
1811 | if (!found) { | |
af60bed2 | 1812 | *start = max(cur_start, state->start); |
d1310b2e CM |
1813 | found = 1; |
1814 | } | |
ec29ed5b CM |
1815 | last = state->end; |
1816 | } else if (contig && found) { | |
1817 | break; | |
d1310b2e CM |
1818 | } |
1819 | node = rb_next(node); | |
1820 | if (!node) | |
1821 | break; | |
1822 | } | |
1823 | out: | |
cad321ad | 1824 | spin_unlock(&tree->lock); |
d1310b2e CM |
1825 | return total_bytes; |
1826 | } | |
b2950863 | 1827 | |
d352ac68 CM |
1828 | /* |
1829 | * set the private field for a given byte offset in the tree. If there isn't | |
1830 | * an extent_state there already, this does nothing. | |
1831 | */ | |
f827ba9a | 1832 | static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start, |
47dc196a | 1833 | struct io_failure_record *failrec) |
d1310b2e CM |
1834 | { |
1835 | struct rb_node *node; | |
1836 | struct extent_state *state; | |
1837 | int ret = 0; | |
1838 | ||
cad321ad | 1839 | spin_lock(&tree->lock); |
d1310b2e CM |
1840 | /* |
1841 | * this search will find all the extents that end after | |
1842 | * our range starts. | |
1843 | */ | |
80ea96b1 | 1844 | node = tree_search(tree, start); |
2b114d1d | 1845 | if (!node) { |
d1310b2e CM |
1846 | ret = -ENOENT; |
1847 | goto out; | |
1848 | } | |
1849 | state = rb_entry(node, struct extent_state, rb_node); | |
1850 | if (state->start != start) { | |
1851 | ret = -ENOENT; | |
1852 | goto out; | |
1853 | } | |
47dc196a | 1854 | state->failrec = failrec; |
d1310b2e | 1855 | out: |
cad321ad | 1856 | spin_unlock(&tree->lock); |
d1310b2e CM |
1857 | return ret; |
1858 | } | |
1859 | ||
f827ba9a | 1860 | static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start, |
47dc196a | 1861 | struct io_failure_record **failrec) |
d1310b2e CM |
1862 | { |
1863 | struct rb_node *node; | |
1864 | struct extent_state *state; | |
1865 | int ret = 0; | |
1866 | ||
cad321ad | 1867 | spin_lock(&tree->lock); |
d1310b2e CM |
1868 | /* |
1869 | * this search will find all the extents that end after | |
1870 | * our range starts. | |
1871 | */ | |
80ea96b1 | 1872 | node = tree_search(tree, start); |
2b114d1d | 1873 | if (!node) { |
d1310b2e CM |
1874 | ret = -ENOENT; |
1875 | goto out; | |
1876 | } | |
1877 | state = rb_entry(node, struct extent_state, rb_node); | |
1878 | if (state->start != start) { | |
1879 | ret = -ENOENT; | |
1880 | goto out; | |
1881 | } | |
47dc196a | 1882 | *failrec = state->failrec; |
d1310b2e | 1883 | out: |
cad321ad | 1884 | spin_unlock(&tree->lock); |
d1310b2e CM |
1885 | return ret; |
1886 | } | |
1887 | ||
1888 | /* | |
1889 | * searches a range in the state tree for a given mask. | |
70dec807 | 1890 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1891 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1892 | * range is found set. | |
1893 | */ | |
1894 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1895 | unsigned bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1896 | { |
1897 | struct extent_state *state = NULL; | |
1898 | struct rb_node *node; | |
1899 | int bitset = 0; | |
d1310b2e | 1900 | |
cad321ad | 1901 | spin_lock(&tree->lock); |
27a3507d | 1902 | if (cached && extent_state_in_tree(cached) && cached->start <= start && |
df98b6e2 | 1903 | cached->end > start) |
9655d298 CM |
1904 | node = &cached->rb_node; |
1905 | else | |
1906 | node = tree_search(tree, start); | |
d1310b2e CM |
1907 | while (node && start <= end) { |
1908 | state = rb_entry(node, struct extent_state, rb_node); | |
1909 | ||
1910 | if (filled && state->start > start) { | |
1911 | bitset = 0; | |
1912 | break; | |
1913 | } | |
1914 | ||
1915 | if (state->start > end) | |
1916 | break; | |
1917 | ||
1918 | if (state->state & bits) { | |
1919 | bitset = 1; | |
1920 | if (!filled) | |
1921 | break; | |
1922 | } else if (filled) { | |
1923 | bitset = 0; | |
1924 | break; | |
1925 | } | |
46562cec CM |
1926 | |
1927 | if (state->end == (u64)-1) | |
1928 | break; | |
1929 | ||
d1310b2e CM |
1930 | start = state->end + 1; |
1931 | if (start > end) | |
1932 | break; | |
1933 | node = rb_next(node); | |
1934 | if (!node) { | |
1935 | if (filled) | |
1936 | bitset = 0; | |
1937 | break; | |
1938 | } | |
1939 | } | |
cad321ad | 1940 | spin_unlock(&tree->lock); |
d1310b2e CM |
1941 | return bitset; |
1942 | } | |
d1310b2e CM |
1943 | |
1944 | /* | |
1945 | * helper function to set a given page up to date if all the | |
1946 | * extents in the tree for that page are up to date | |
1947 | */ | |
143bede5 | 1948 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e | 1949 | { |
4eee4fa4 | 1950 | u64 start = page_offset(page); |
09cbfeaf | 1951 | u64 end = start + PAGE_SIZE - 1; |
9655d298 | 1952 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 1953 | SetPageUptodate(page); |
d1310b2e CM |
1954 | } |
1955 | ||
7870d082 JB |
1956 | int free_io_failure(struct extent_io_tree *failure_tree, |
1957 | struct extent_io_tree *io_tree, | |
1958 | struct io_failure_record *rec) | |
4a54c8c1 JS |
1959 | { |
1960 | int ret; | |
1961 | int err = 0; | |
4a54c8c1 | 1962 | |
47dc196a | 1963 | set_state_failrec(failure_tree, rec->start, NULL); |
4a54c8c1 JS |
1964 | ret = clear_extent_bits(failure_tree, rec->start, |
1965 | rec->start + rec->len - 1, | |
91166212 | 1966 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 JS |
1967 | if (ret) |
1968 | err = ret; | |
1969 | ||
7870d082 | 1970 | ret = clear_extent_bits(io_tree, rec->start, |
53b381b3 | 1971 | rec->start + rec->len - 1, |
91166212 | 1972 | EXTENT_DAMAGED); |
53b381b3 DW |
1973 | if (ret && !err) |
1974 | err = ret; | |
4a54c8c1 JS |
1975 | |
1976 | kfree(rec); | |
1977 | return err; | |
1978 | } | |
1979 | ||
4a54c8c1 JS |
1980 | /* |
1981 | * this bypasses the standard btrfs submit functions deliberately, as | |
1982 | * the standard behavior is to write all copies in a raid setup. here we only | |
1983 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
1984 | * submit_bio directly. | |
3ec706c8 | 1985 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
1986 | * actually prevents the read that triggered the error from finishing. |
1987 | * currently, there can be no more than two copies of every data bit. thus, | |
1988 | * exactly one rewrite is required. | |
1989 | */ | |
6ec656bc JB |
1990 | int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start, |
1991 | u64 length, u64 logical, struct page *page, | |
1992 | unsigned int pg_offset, int mirror_num) | |
4a54c8c1 JS |
1993 | { |
1994 | struct bio *bio; | |
1995 | struct btrfs_device *dev; | |
4a54c8c1 JS |
1996 | u64 map_length = 0; |
1997 | u64 sector; | |
1998 | struct btrfs_bio *bbio = NULL; | |
1999 | int ret; | |
2000 | ||
1751e8a6 | 2001 | ASSERT(!(fs_info->sb->s_flags & SB_RDONLY)); |
4a54c8c1 JS |
2002 | BUG_ON(!mirror_num); |
2003 | ||
c5e4c3d7 | 2004 | bio = btrfs_io_bio_alloc(1); |
4f024f37 | 2005 | bio->bi_iter.bi_size = 0; |
4a54c8c1 JS |
2006 | map_length = length; |
2007 | ||
b5de8d0d FM |
2008 | /* |
2009 | * Avoid races with device replace and make sure our bbio has devices | |
2010 | * associated to its stripes that don't go away while we are doing the | |
2011 | * read repair operation. | |
2012 | */ | |
2013 | btrfs_bio_counter_inc_blocked(fs_info); | |
e4ff5fb5 | 2014 | if (btrfs_is_parity_mirror(fs_info, logical, length)) { |
c725328c LB |
2015 | /* |
2016 | * Note that we don't use BTRFS_MAP_WRITE because it's supposed | |
2017 | * to update all raid stripes, but here we just want to correct | |
2018 | * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad | |
2019 | * stripe's dev and sector. | |
2020 | */ | |
2021 | ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical, | |
2022 | &map_length, &bbio, 0); | |
2023 | if (ret) { | |
2024 | btrfs_bio_counter_dec(fs_info); | |
2025 | bio_put(bio); | |
2026 | return -EIO; | |
2027 | } | |
2028 | ASSERT(bbio->mirror_num == 1); | |
2029 | } else { | |
2030 | ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, | |
2031 | &map_length, &bbio, mirror_num); | |
2032 | if (ret) { | |
2033 | btrfs_bio_counter_dec(fs_info); | |
2034 | bio_put(bio); | |
2035 | return -EIO; | |
2036 | } | |
2037 | BUG_ON(mirror_num != bbio->mirror_num); | |
4a54c8c1 | 2038 | } |
c725328c LB |
2039 | |
2040 | sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9; | |
4f024f37 | 2041 | bio->bi_iter.bi_sector = sector; |
c725328c | 2042 | dev = bbio->stripes[bbio->mirror_num - 1].dev; |
6e9606d2 | 2043 | btrfs_put_bbio(bbio); |
ebbede42 AJ |
2044 | if (!dev || !dev->bdev || |
2045 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) { | |
b5de8d0d | 2046 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2047 | bio_put(bio); |
2048 | return -EIO; | |
2049 | } | |
74d46992 | 2050 | bio_set_dev(bio, dev->bdev); |
70fd7614 | 2051 | bio->bi_opf = REQ_OP_WRITE | REQ_SYNC; |
ffdd2018 | 2052 | bio_add_page(bio, page, length, pg_offset); |
4a54c8c1 | 2053 | |
4e49ea4a | 2054 | if (btrfsic_submit_bio_wait(bio)) { |
4a54c8c1 | 2055 | /* try to remap that extent elsewhere? */ |
b5de8d0d | 2056 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 | 2057 | bio_put(bio); |
442a4f63 | 2058 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
2059 | return -EIO; |
2060 | } | |
2061 | ||
b14af3b4 DS |
2062 | btrfs_info_rl_in_rcu(fs_info, |
2063 | "read error corrected: ino %llu off %llu (dev %s sector %llu)", | |
6ec656bc | 2064 | ino, start, |
1203b681 | 2065 | rcu_str_deref(dev->name), sector); |
b5de8d0d | 2066 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2067 | bio_put(bio); |
2068 | return 0; | |
2069 | } | |
2070 | ||
2ff7e61e JM |
2071 | int repair_eb_io_failure(struct btrfs_fs_info *fs_info, |
2072 | struct extent_buffer *eb, int mirror_num) | |
ea466794 | 2073 | { |
ea466794 | 2074 | u64 start = eb->start; |
cc5e31a4 | 2075 | int i, num_pages = num_extent_pages(eb); |
d95603b2 | 2076 | int ret = 0; |
ea466794 | 2077 | |
bc98a42c | 2078 | if (sb_rdonly(fs_info->sb)) |
908960c6 ID |
2079 | return -EROFS; |
2080 | ||
ea466794 | 2081 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 2082 | struct page *p = eb->pages[i]; |
1203b681 | 2083 | |
6ec656bc | 2084 | ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p, |
1203b681 | 2085 | start - page_offset(p), mirror_num); |
ea466794 JB |
2086 | if (ret) |
2087 | break; | |
09cbfeaf | 2088 | start += PAGE_SIZE; |
ea466794 JB |
2089 | } |
2090 | ||
2091 | return ret; | |
2092 | } | |
2093 | ||
4a54c8c1 JS |
2094 | /* |
2095 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2096 | * to see if we need to process or clean up an io_failure_record | |
2097 | */ | |
7870d082 JB |
2098 | int clean_io_failure(struct btrfs_fs_info *fs_info, |
2099 | struct extent_io_tree *failure_tree, | |
2100 | struct extent_io_tree *io_tree, u64 start, | |
2101 | struct page *page, u64 ino, unsigned int pg_offset) | |
4a54c8c1 JS |
2102 | { |
2103 | u64 private; | |
4a54c8c1 | 2104 | struct io_failure_record *failrec; |
4a54c8c1 JS |
2105 | struct extent_state *state; |
2106 | int num_copies; | |
4a54c8c1 | 2107 | int ret; |
4a54c8c1 JS |
2108 | |
2109 | private = 0; | |
7870d082 JB |
2110 | ret = count_range_bits(failure_tree, &private, (u64)-1, 1, |
2111 | EXTENT_DIRTY, 0); | |
4a54c8c1 JS |
2112 | if (!ret) |
2113 | return 0; | |
2114 | ||
7870d082 | 2115 | ret = get_state_failrec(failure_tree, start, &failrec); |
4a54c8c1 JS |
2116 | if (ret) |
2117 | return 0; | |
2118 | ||
4a54c8c1 JS |
2119 | BUG_ON(!failrec->this_mirror); |
2120 | ||
2121 | if (failrec->in_validation) { | |
2122 | /* there was no real error, just free the record */ | |
ab8d0fc4 JM |
2123 | btrfs_debug(fs_info, |
2124 | "clean_io_failure: freeing dummy error at %llu", | |
2125 | failrec->start); | |
4a54c8c1 JS |
2126 | goto out; |
2127 | } | |
bc98a42c | 2128 | if (sb_rdonly(fs_info->sb)) |
908960c6 | 2129 | goto out; |
4a54c8c1 | 2130 | |
7870d082 JB |
2131 | spin_lock(&io_tree->lock); |
2132 | state = find_first_extent_bit_state(io_tree, | |
4a54c8c1 JS |
2133 | failrec->start, |
2134 | EXTENT_LOCKED); | |
7870d082 | 2135 | spin_unlock(&io_tree->lock); |
4a54c8c1 | 2136 | |
883d0de4 MX |
2137 | if (state && state->start <= failrec->start && |
2138 | state->end >= failrec->start + failrec->len - 1) { | |
3ec706c8 SB |
2139 | num_copies = btrfs_num_copies(fs_info, failrec->logical, |
2140 | failrec->len); | |
4a54c8c1 | 2141 | if (num_copies > 1) { |
7870d082 JB |
2142 | repair_io_failure(fs_info, ino, start, failrec->len, |
2143 | failrec->logical, page, pg_offset, | |
2144 | failrec->failed_mirror); | |
4a54c8c1 JS |
2145 | } |
2146 | } | |
2147 | ||
2148 | out: | |
7870d082 | 2149 | free_io_failure(failure_tree, io_tree, failrec); |
4a54c8c1 | 2150 | |
454ff3de | 2151 | return 0; |
4a54c8c1 JS |
2152 | } |
2153 | ||
f612496b MX |
2154 | /* |
2155 | * Can be called when | |
2156 | * - hold extent lock | |
2157 | * - under ordered extent | |
2158 | * - the inode is freeing | |
2159 | */ | |
7ab7956e | 2160 | void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end) |
f612496b | 2161 | { |
7ab7956e | 2162 | struct extent_io_tree *failure_tree = &inode->io_failure_tree; |
f612496b MX |
2163 | struct io_failure_record *failrec; |
2164 | struct extent_state *state, *next; | |
2165 | ||
2166 | if (RB_EMPTY_ROOT(&failure_tree->state)) | |
2167 | return; | |
2168 | ||
2169 | spin_lock(&failure_tree->lock); | |
2170 | state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY); | |
2171 | while (state) { | |
2172 | if (state->start > end) | |
2173 | break; | |
2174 | ||
2175 | ASSERT(state->end <= end); | |
2176 | ||
2177 | next = next_state(state); | |
2178 | ||
47dc196a | 2179 | failrec = state->failrec; |
f612496b MX |
2180 | free_extent_state(state); |
2181 | kfree(failrec); | |
2182 | ||
2183 | state = next; | |
2184 | } | |
2185 | spin_unlock(&failure_tree->lock); | |
2186 | } | |
2187 | ||
2fe6303e | 2188 | int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end, |
47dc196a | 2189 | struct io_failure_record **failrec_ret) |
4a54c8c1 | 2190 | { |
ab8d0fc4 | 2191 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e | 2192 | struct io_failure_record *failrec; |
4a54c8c1 | 2193 | struct extent_map *em; |
4a54c8c1 JS |
2194 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2195 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2196 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
4a54c8c1 | 2197 | int ret; |
4a54c8c1 JS |
2198 | u64 logical; |
2199 | ||
47dc196a | 2200 | ret = get_state_failrec(failure_tree, start, &failrec); |
4a54c8c1 JS |
2201 | if (ret) { |
2202 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2203 | if (!failrec) | |
2204 | return -ENOMEM; | |
2fe6303e | 2205 | |
4a54c8c1 JS |
2206 | failrec->start = start; |
2207 | failrec->len = end - start + 1; | |
2208 | failrec->this_mirror = 0; | |
2209 | failrec->bio_flags = 0; | |
2210 | failrec->in_validation = 0; | |
2211 | ||
2212 | read_lock(&em_tree->lock); | |
2213 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2214 | if (!em) { | |
2215 | read_unlock(&em_tree->lock); | |
2216 | kfree(failrec); | |
2217 | return -EIO; | |
2218 | } | |
2219 | ||
68ba990f | 2220 | if (em->start > start || em->start + em->len <= start) { |
4a54c8c1 JS |
2221 | free_extent_map(em); |
2222 | em = NULL; | |
2223 | } | |
2224 | read_unlock(&em_tree->lock); | |
7a2d6a64 | 2225 | if (!em) { |
4a54c8c1 JS |
2226 | kfree(failrec); |
2227 | return -EIO; | |
2228 | } | |
2fe6303e | 2229 | |
4a54c8c1 JS |
2230 | logical = start - em->start; |
2231 | logical = em->block_start + logical; | |
2232 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2233 | logical = em->block_start; | |
2234 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2235 | extent_set_compress_type(&failrec->bio_flags, | |
2236 | em->compress_type); | |
2237 | } | |
2fe6303e | 2238 | |
ab8d0fc4 JM |
2239 | btrfs_debug(fs_info, |
2240 | "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu", | |
2241 | logical, start, failrec->len); | |
2fe6303e | 2242 | |
4a54c8c1 JS |
2243 | failrec->logical = logical; |
2244 | free_extent_map(em); | |
2245 | ||
2246 | /* set the bits in the private failure tree */ | |
2247 | ret = set_extent_bits(failure_tree, start, end, | |
ceeb0ae7 | 2248 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 | 2249 | if (ret >= 0) |
47dc196a | 2250 | ret = set_state_failrec(failure_tree, start, failrec); |
4a54c8c1 JS |
2251 | /* set the bits in the inode's tree */ |
2252 | if (ret >= 0) | |
ceeb0ae7 | 2253 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED); |
4a54c8c1 JS |
2254 | if (ret < 0) { |
2255 | kfree(failrec); | |
2256 | return ret; | |
2257 | } | |
2258 | } else { | |
ab8d0fc4 JM |
2259 | btrfs_debug(fs_info, |
2260 | "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d", | |
2261 | failrec->logical, failrec->start, failrec->len, | |
2262 | failrec->in_validation); | |
4a54c8c1 JS |
2263 | /* |
2264 | * when data can be on disk more than twice, add to failrec here | |
2265 | * (e.g. with a list for failed_mirror) to make | |
2266 | * clean_io_failure() clean all those errors at once. | |
2267 | */ | |
2268 | } | |
2fe6303e MX |
2269 | |
2270 | *failrec_ret = failrec; | |
2271 | ||
2272 | return 0; | |
2273 | } | |
2274 | ||
a0b60d72 | 2275 | bool btrfs_check_repairable(struct inode *inode, unsigned failed_bio_pages, |
2fe6303e MX |
2276 | struct io_failure_record *failrec, int failed_mirror) |
2277 | { | |
ab8d0fc4 | 2278 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e MX |
2279 | int num_copies; |
2280 | ||
ab8d0fc4 | 2281 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
4a54c8c1 JS |
2282 | if (num_copies == 1) { |
2283 | /* | |
2284 | * we only have a single copy of the data, so don't bother with | |
2285 | * all the retry and error correction code that follows. no | |
2286 | * matter what the error is, it is very likely to persist. | |
2287 | */ | |
ab8d0fc4 JM |
2288 | btrfs_debug(fs_info, |
2289 | "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
2290 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2291 | return false; |
4a54c8c1 JS |
2292 | } |
2293 | ||
4a54c8c1 JS |
2294 | /* |
2295 | * there are two premises: | |
2296 | * a) deliver good data to the caller | |
2297 | * b) correct the bad sectors on disk | |
2298 | */ | |
a0b60d72 | 2299 | if (failed_bio_pages > 1) { |
4a54c8c1 JS |
2300 | /* |
2301 | * to fulfill b), we need to know the exact failing sectors, as | |
2302 | * we don't want to rewrite any more than the failed ones. thus, | |
2303 | * we need separate read requests for the failed bio | |
2304 | * | |
2305 | * if the following BUG_ON triggers, our validation request got | |
2306 | * merged. we need separate requests for our algorithm to work. | |
2307 | */ | |
2308 | BUG_ON(failrec->in_validation); | |
2309 | failrec->in_validation = 1; | |
2310 | failrec->this_mirror = failed_mirror; | |
4a54c8c1 JS |
2311 | } else { |
2312 | /* | |
2313 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2314 | * of the failed sector and if we succeed, we have setup | |
2315 | * everything for repair_io_failure to do the rest for us. | |
2316 | */ | |
2317 | if (failrec->in_validation) { | |
2318 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2319 | failrec->in_validation = 0; | |
2320 | failrec->this_mirror = 0; | |
2321 | } | |
2322 | failrec->failed_mirror = failed_mirror; | |
2323 | failrec->this_mirror++; | |
2324 | if (failrec->this_mirror == failed_mirror) | |
2325 | failrec->this_mirror++; | |
4a54c8c1 JS |
2326 | } |
2327 | ||
facc8a22 | 2328 | if (failrec->this_mirror > num_copies) { |
ab8d0fc4 JM |
2329 | btrfs_debug(fs_info, |
2330 | "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
2331 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2332 | return false; |
4a54c8c1 JS |
2333 | } |
2334 | ||
c3cfb656 | 2335 | return true; |
2fe6303e MX |
2336 | } |
2337 | ||
2338 | ||
2339 | struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio, | |
2340 | struct io_failure_record *failrec, | |
2341 | struct page *page, int pg_offset, int icsum, | |
8b110e39 | 2342 | bio_end_io_t *endio_func, void *data) |
2fe6303e | 2343 | { |
0b246afa | 2344 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e MX |
2345 | struct bio *bio; |
2346 | struct btrfs_io_bio *btrfs_failed_bio; | |
2347 | struct btrfs_io_bio *btrfs_bio; | |
2348 | ||
c5e4c3d7 | 2349 | bio = btrfs_io_bio_alloc(1); |
2fe6303e | 2350 | bio->bi_end_io = endio_func; |
4f024f37 | 2351 | bio->bi_iter.bi_sector = failrec->logical >> 9; |
74d46992 | 2352 | bio_set_dev(bio, fs_info->fs_devices->latest_bdev); |
4f024f37 | 2353 | bio->bi_iter.bi_size = 0; |
8b110e39 | 2354 | bio->bi_private = data; |
4a54c8c1 | 2355 | |
facc8a22 MX |
2356 | btrfs_failed_bio = btrfs_io_bio(failed_bio); |
2357 | if (btrfs_failed_bio->csum) { | |
facc8a22 MX |
2358 | u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); |
2359 | ||
2360 | btrfs_bio = btrfs_io_bio(bio); | |
2361 | btrfs_bio->csum = btrfs_bio->csum_inline; | |
2fe6303e MX |
2362 | icsum *= csum_size; |
2363 | memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum, | |
facc8a22 MX |
2364 | csum_size); |
2365 | } | |
2366 | ||
2fe6303e MX |
2367 | bio_add_page(bio, page, failrec->len, pg_offset); |
2368 | ||
2369 | return bio; | |
2370 | } | |
2371 | ||
2372 | /* | |
78e62c02 NB |
2373 | * This is a generic handler for readpage errors. If other copies exist, read |
2374 | * those and write back good data to the failed position. Does not investigate | |
2375 | * in remapping the failed extent elsewhere, hoping the device will be smart | |
2376 | * enough to do this as needed | |
2fe6303e | 2377 | */ |
2fe6303e MX |
2378 | static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset, |
2379 | struct page *page, u64 start, u64 end, | |
2380 | int failed_mirror) | |
2381 | { | |
2382 | struct io_failure_record *failrec; | |
2383 | struct inode *inode = page->mapping->host; | |
2384 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
7870d082 | 2385 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2fe6303e | 2386 | struct bio *bio; |
70fd7614 | 2387 | int read_mode = 0; |
4e4cbee9 | 2388 | blk_status_t status; |
2fe6303e | 2389 | int ret; |
8a2ee44a | 2390 | unsigned failed_bio_pages = failed_bio->bi_iter.bi_size >> PAGE_SHIFT; |
2fe6303e | 2391 | |
1f7ad75b | 2392 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
2fe6303e MX |
2393 | |
2394 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
2395 | if (ret) | |
2396 | return ret; | |
2397 | ||
a0b60d72 | 2398 | if (!btrfs_check_repairable(inode, failed_bio_pages, failrec, |
c3cfb656 | 2399 | failed_mirror)) { |
7870d082 | 2400 | free_io_failure(failure_tree, tree, failrec); |
2fe6303e MX |
2401 | return -EIO; |
2402 | } | |
2403 | ||
a0b60d72 | 2404 | if (failed_bio_pages > 1) |
70fd7614 | 2405 | read_mode |= REQ_FAILFAST_DEV; |
2fe6303e MX |
2406 | |
2407 | phy_offset >>= inode->i_sb->s_blocksize_bits; | |
2408 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2409 | start - page_offset(page), | |
8b110e39 MX |
2410 | (int)phy_offset, failed_bio->bi_end_io, |
2411 | NULL); | |
ebcc3263 | 2412 | bio->bi_opf = REQ_OP_READ | read_mode; |
4a54c8c1 | 2413 | |
ab8d0fc4 JM |
2414 | btrfs_debug(btrfs_sb(inode->i_sb), |
2415 | "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d", | |
2416 | read_mode, failrec->this_mirror, failrec->in_validation); | |
4a54c8c1 | 2417 | |
8c27cb35 | 2418 | status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror, |
013bd4c3 | 2419 | failrec->bio_flags, 0); |
4e4cbee9 | 2420 | if (status) { |
7870d082 | 2421 | free_io_failure(failure_tree, tree, failrec); |
6c387ab2 | 2422 | bio_put(bio); |
4e4cbee9 | 2423 | ret = blk_status_to_errno(status); |
6c387ab2 MX |
2424 | } |
2425 | ||
013bd4c3 | 2426 | return ret; |
4a54c8c1 JS |
2427 | } |
2428 | ||
d1310b2e CM |
2429 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2430 | ||
b5227c07 | 2431 | void end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
87826df0 JM |
2432 | { |
2433 | int uptodate = (err == 0); | |
3e2426bd | 2434 | int ret = 0; |
87826df0 | 2435 | |
c629732d | 2436 | btrfs_writepage_endio_finish_ordered(page, start, end, uptodate); |
87826df0 | 2437 | |
87826df0 | 2438 | if (!uptodate) { |
87826df0 JM |
2439 | ClearPageUptodate(page); |
2440 | SetPageError(page); | |
bff5baf8 | 2441 | ret = err < 0 ? err : -EIO; |
5dca6eea | 2442 | mapping_set_error(page->mapping, ret); |
87826df0 | 2443 | } |
87826df0 JM |
2444 | } |
2445 | ||
d1310b2e CM |
2446 | /* |
2447 | * after a writepage IO is done, we need to: | |
2448 | * clear the uptodate bits on error | |
2449 | * clear the writeback bits in the extent tree for this IO | |
2450 | * end_page_writeback if the page has no more pending IO | |
2451 | * | |
2452 | * Scheduling is not allowed, so the extent state tree is expected | |
2453 | * to have one and only one object corresponding to this IO. | |
2454 | */ | |
4246a0b6 | 2455 | static void end_bio_extent_writepage(struct bio *bio) |
d1310b2e | 2456 | { |
4e4cbee9 | 2457 | int error = blk_status_to_errno(bio->bi_status); |
2c30c71b | 2458 | struct bio_vec *bvec; |
d1310b2e CM |
2459 | u64 start; |
2460 | u64 end; | |
2c30c71b | 2461 | int i; |
6dc4f100 | 2462 | struct bvec_iter_all iter_all; |
d1310b2e | 2463 | |
c09abff8 | 2464 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
6dc4f100 | 2465 | bio_for_each_segment_all(bvec, bio, i, iter_all) { |
d1310b2e | 2466 | struct page *page = bvec->bv_page; |
0b246afa JM |
2467 | struct inode *inode = page->mapping->host; |
2468 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
902b22f3 | 2469 | |
17a5adcc AO |
2470 | /* We always issue full-page reads, but if some block |
2471 | * in a page fails to read, blk_update_request() will | |
2472 | * advance bv_offset and adjust bv_len to compensate. | |
2473 | * Print a warning for nonzero offsets, and an error | |
2474 | * if they don't add up to a full page. */ | |
09cbfeaf KS |
2475 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { |
2476 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
0b246afa | 2477 | btrfs_err(fs_info, |
efe120a0 FH |
2478 | "partial page write in btrfs with offset %u and length %u", |
2479 | bvec->bv_offset, bvec->bv_len); | |
2480 | else | |
0b246afa | 2481 | btrfs_info(fs_info, |
5d163e0e | 2482 | "incomplete page write in btrfs with offset %u and length %u", |
efe120a0 FH |
2483 | bvec->bv_offset, bvec->bv_len); |
2484 | } | |
d1310b2e | 2485 | |
17a5adcc AO |
2486 | start = page_offset(page); |
2487 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
d1310b2e | 2488 | |
4e4cbee9 | 2489 | end_extent_writepage(page, error, start, end); |
17a5adcc | 2490 | end_page_writeback(page); |
2c30c71b | 2491 | } |
2b1f55b0 | 2492 | |
d1310b2e | 2493 | bio_put(bio); |
d1310b2e CM |
2494 | } |
2495 | ||
883d0de4 MX |
2496 | static void |
2497 | endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len, | |
2498 | int uptodate) | |
2499 | { | |
2500 | struct extent_state *cached = NULL; | |
2501 | u64 end = start + len - 1; | |
2502 | ||
2503 | if (uptodate && tree->track_uptodate) | |
2504 | set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC); | |
d810a4be | 2505 | unlock_extent_cached_atomic(tree, start, end, &cached); |
883d0de4 MX |
2506 | } |
2507 | ||
d1310b2e CM |
2508 | /* |
2509 | * after a readpage IO is done, we need to: | |
2510 | * clear the uptodate bits on error | |
2511 | * set the uptodate bits if things worked | |
2512 | * set the page up to date if all extents in the tree are uptodate | |
2513 | * clear the lock bit in the extent tree | |
2514 | * unlock the page if there are no other extents locked for it | |
2515 | * | |
2516 | * Scheduling is not allowed, so the extent state tree is expected | |
2517 | * to have one and only one object corresponding to this IO. | |
2518 | */ | |
4246a0b6 | 2519 | static void end_bio_extent_readpage(struct bio *bio) |
d1310b2e | 2520 | { |
2c30c71b | 2521 | struct bio_vec *bvec; |
4e4cbee9 | 2522 | int uptodate = !bio->bi_status; |
facc8a22 | 2523 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); |
7870d082 | 2524 | struct extent_io_tree *tree, *failure_tree; |
facc8a22 | 2525 | u64 offset = 0; |
d1310b2e CM |
2526 | u64 start; |
2527 | u64 end; | |
facc8a22 | 2528 | u64 len; |
883d0de4 MX |
2529 | u64 extent_start = 0; |
2530 | u64 extent_len = 0; | |
5cf1ab56 | 2531 | int mirror; |
d1310b2e | 2532 | int ret; |
2c30c71b | 2533 | int i; |
6dc4f100 | 2534 | struct bvec_iter_all iter_all; |
d1310b2e | 2535 | |
c09abff8 | 2536 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
6dc4f100 | 2537 | bio_for_each_segment_all(bvec, bio, i, iter_all) { |
d1310b2e | 2538 | struct page *page = bvec->bv_page; |
a71754fc | 2539 | struct inode *inode = page->mapping->host; |
ab8d0fc4 | 2540 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
78e62c02 NB |
2541 | bool data_inode = btrfs_ino(BTRFS_I(inode)) |
2542 | != BTRFS_BTREE_INODE_OBJECTID; | |
507903b8 | 2543 | |
ab8d0fc4 JM |
2544 | btrfs_debug(fs_info, |
2545 | "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u", | |
4e4cbee9 | 2546 | (u64)bio->bi_iter.bi_sector, bio->bi_status, |
ab8d0fc4 | 2547 | io_bio->mirror_num); |
a71754fc | 2548 | tree = &BTRFS_I(inode)->io_tree; |
7870d082 | 2549 | failure_tree = &BTRFS_I(inode)->io_failure_tree; |
902b22f3 | 2550 | |
17a5adcc AO |
2551 | /* We always issue full-page reads, but if some block |
2552 | * in a page fails to read, blk_update_request() will | |
2553 | * advance bv_offset and adjust bv_len to compensate. | |
2554 | * Print a warning for nonzero offsets, and an error | |
2555 | * if they don't add up to a full page. */ | |
09cbfeaf KS |
2556 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { |
2557 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
ab8d0fc4 JM |
2558 | btrfs_err(fs_info, |
2559 | "partial page read in btrfs with offset %u and length %u", | |
efe120a0 FH |
2560 | bvec->bv_offset, bvec->bv_len); |
2561 | else | |
ab8d0fc4 JM |
2562 | btrfs_info(fs_info, |
2563 | "incomplete page read in btrfs with offset %u and length %u", | |
efe120a0 FH |
2564 | bvec->bv_offset, bvec->bv_len); |
2565 | } | |
d1310b2e | 2566 | |
17a5adcc AO |
2567 | start = page_offset(page); |
2568 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
facc8a22 | 2569 | len = bvec->bv_len; |
d1310b2e | 2570 | |
9be3395b | 2571 | mirror = io_bio->mirror_num; |
78e62c02 | 2572 | if (likely(uptodate)) { |
facc8a22 MX |
2573 | ret = tree->ops->readpage_end_io_hook(io_bio, offset, |
2574 | page, start, end, | |
2575 | mirror); | |
5ee0844d | 2576 | if (ret) |
d1310b2e | 2577 | uptodate = 0; |
5ee0844d | 2578 | else |
7870d082 JB |
2579 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
2580 | failure_tree, tree, start, | |
2581 | page, | |
2582 | btrfs_ino(BTRFS_I(inode)), 0); | |
d1310b2e | 2583 | } |
ea466794 | 2584 | |
f2a09da9 MX |
2585 | if (likely(uptodate)) |
2586 | goto readpage_ok; | |
2587 | ||
78e62c02 | 2588 | if (data_inode) { |
9d0d1c8b | 2589 | |
f4a8e656 | 2590 | /* |
78e62c02 NB |
2591 | * The generic bio_readpage_error handles errors the |
2592 | * following way: If possible, new read requests are | |
2593 | * created and submitted and will end up in | |
2594 | * end_bio_extent_readpage as well (if we're lucky, | |
2595 | * not in the !uptodate case). In that case it returns | |
2596 | * 0 and we just go on with the next page in our bio. | |
2597 | * If it can't handle the error it will return -EIO and | |
2598 | * we remain responsible for that page. | |
f4a8e656 | 2599 | */ |
78e62c02 NB |
2600 | ret = bio_readpage_error(bio, offset, page, start, end, |
2601 | mirror); | |
2602 | if (ret == 0) { | |
2603 | uptodate = !bio->bi_status; | |
2604 | offset += len; | |
2605 | continue; | |
2606 | } | |
2607 | } else { | |
2608 | struct extent_buffer *eb; | |
2609 | ||
2610 | eb = (struct extent_buffer *)page->private; | |
2611 | set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); | |
2612 | eb->read_mirror = mirror; | |
2613 | atomic_dec(&eb->io_pages); | |
2614 | if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, | |
2615 | &eb->bflags)) | |
2616 | btree_readahead_hook(eb, -EIO); | |
7e38326f | 2617 | } |
f2a09da9 | 2618 | readpage_ok: |
883d0de4 | 2619 | if (likely(uptodate)) { |
a71754fc | 2620 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 2621 | pgoff_t end_index = i_size >> PAGE_SHIFT; |
a583c026 | 2622 | unsigned off; |
a71754fc JB |
2623 | |
2624 | /* Zero out the end if this page straddles i_size */ | |
7073017a | 2625 | off = offset_in_page(i_size); |
a583c026 | 2626 | if (page->index == end_index && off) |
09cbfeaf | 2627 | zero_user_segment(page, off, PAGE_SIZE); |
17a5adcc | 2628 | SetPageUptodate(page); |
70dec807 | 2629 | } else { |
17a5adcc AO |
2630 | ClearPageUptodate(page); |
2631 | SetPageError(page); | |
70dec807 | 2632 | } |
17a5adcc | 2633 | unlock_page(page); |
facc8a22 | 2634 | offset += len; |
883d0de4 MX |
2635 | |
2636 | if (unlikely(!uptodate)) { | |
2637 | if (extent_len) { | |
2638 | endio_readpage_release_extent(tree, | |
2639 | extent_start, | |
2640 | extent_len, 1); | |
2641 | extent_start = 0; | |
2642 | extent_len = 0; | |
2643 | } | |
2644 | endio_readpage_release_extent(tree, start, | |
2645 | end - start + 1, 0); | |
2646 | } else if (!extent_len) { | |
2647 | extent_start = start; | |
2648 | extent_len = end + 1 - start; | |
2649 | } else if (extent_start + extent_len == start) { | |
2650 | extent_len += end + 1 - start; | |
2651 | } else { | |
2652 | endio_readpage_release_extent(tree, extent_start, | |
2653 | extent_len, uptodate); | |
2654 | extent_start = start; | |
2655 | extent_len = end + 1 - start; | |
2656 | } | |
2c30c71b | 2657 | } |
d1310b2e | 2658 | |
883d0de4 MX |
2659 | if (extent_len) |
2660 | endio_readpage_release_extent(tree, extent_start, extent_len, | |
2661 | uptodate); | |
b3a0dd50 | 2662 | btrfs_io_bio_free_csum(io_bio); |
d1310b2e | 2663 | bio_put(bio); |
d1310b2e CM |
2664 | } |
2665 | ||
9be3395b | 2666 | /* |
184f999e DS |
2667 | * Initialize the members up to but not including 'bio'. Use after allocating a |
2668 | * new bio by bio_alloc_bioset as it does not initialize the bytes outside of | |
2669 | * 'bio' because use of __GFP_ZERO is not supported. | |
9be3395b | 2670 | */ |
184f999e | 2671 | static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio) |
d1310b2e | 2672 | { |
184f999e DS |
2673 | memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio)); |
2674 | } | |
d1310b2e | 2675 | |
9be3395b | 2676 | /* |
6e707bcd DS |
2677 | * The following helpers allocate a bio. As it's backed by a bioset, it'll |
2678 | * never fail. We're returning a bio right now but you can call btrfs_io_bio | |
2679 | * for the appropriate container_of magic | |
9be3395b | 2680 | */ |
c821e7f3 | 2681 | struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte) |
d1310b2e CM |
2682 | { |
2683 | struct bio *bio; | |
d1310b2e | 2684 | |
8ac9f7c1 | 2685 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &btrfs_bioset); |
74d46992 | 2686 | bio_set_dev(bio, bdev); |
c821e7f3 | 2687 | bio->bi_iter.bi_sector = first_byte >> 9; |
184f999e | 2688 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
d1310b2e CM |
2689 | return bio; |
2690 | } | |
2691 | ||
8b6c1d56 | 2692 | struct bio *btrfs_bio_clone(struct bio *bio) |
9be3395b | 2693 | { |
23ea8e5a MX |
2694 | struct btrfs_io_bio *btrfs_bio; |
2695 | struct bio *new; | |
9be3395b | 2696 | |
6e707bcd | 2697 | /* Bio allocation backed by a bioset does not fail */ |
8ac9f7c1 | 2698 | new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset); |
6e707bcd | 2699 | btrfs_bio = btrfs_io_bio(new); |
184f999e | 2700 | btrfs_io_bio_init(btrfs_bio); |
6e707bcd | 2701 | btrfs_bio->iter = bio->bi_iter; |
23ea8e5a MX |
2702 | return new; |
2703 | } | |
9be3395b | 2704 | |
c5e4c3d7 | 2705 | struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs) |
9be3395b | 2706 | { |
facc8a22 MX |
2707 | struct bio *bio; |
2708 | ||
6e707bcd | 2709 | /* Bio allocation backed by a bioset does not fail */ |
8ac9f7c1 | 2710 | bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset); |
184f999e | 2711 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
facc8a22 | 2712 | return bio; |
9be3395b CM |
2713 | } |
2714 | ||
e477094f | 2715 | struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size) |
2f8e9140 LB |
2716 | { |
2717 | struct bio *bio; | |
2718 | struct btrfs_io_bio *btrfs_bio; | |
2719 | ||
2720 | /* this will never fail when it's backed by a bioset */ | |
8ac9f7c1 | 2721 | bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset); |
2f8e9140 LB |
2722 | ASSERT(bio); |
2723 | ||
2724 | btrfs_bio = btrfs_io_bio(bio); | |
184f999e | 2725 | btrfs_io_bio_init(btrfs_bio); |
2f8e9140 LB |
2726 | |
2727 | bio_trim(bio, offset >> 9, size >> 9); | |
17347cec | 2728 | btrfs_bio->iter = bio->bi_iter; |
2f8e9140 LB |
2729 | return bio; |
2730 | } | |
9be3395b | 2731 | |
4b81ba48 DS |
2732 | /* |
2733 | * @opf: bio REQ_OP_* and REQ_* flags as one value | |
b8b3d625 DS |
2734 | * @tree: tree so we can call our merge_bio hook |
2735 | * @wbc: optional writeback control for io accounting | |
2736 | * @page: page to add to the bio | |
2737 | * @pg_offset: offset of the new bio or to check whether we are adding | |
2738 | * a contiguous page to the previous one | |
2739 | * @size: portion of page that we want to write | |
2740 | * @offset: starting offset in the page | |
2741 | * @bdev: attach newly created bios to this bdev | |
5c2b1fd7 | 2742 | * @bio_ret: must be valid pointer, newly allocated bio will be stored there |
b8b3d625 DS |
2743 | * @end_io_func: end_io callback for new bio |
2744 | * @mirror_num: desired mirror to read/write | |
2745 | * @prev_bio_flags: flags of previous bio to see if we can merge the current one | |
2746 | * @bio_flags: flags of the current bio to see if we can merge them | |
4b81ba48 DS |
2747 | */ |
2748 | static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree, | |
da2f0f74 | 2749 | struct writeback_control *wbc, |
6273b7f8 | 2750 | struct page *page, u64 offset, |
6c5a4e2c | 2751 | size_t size, unsigned long pg_offset, |
d1310b2e CM |
2752 | struct block_device *bdev, |
2753 | struct bio **bio_ret, | |
f188591e | 2754 | bio_end_io_t end_io_func, |
c8b97818 CM |
2755 | int mirror_num, |
2756 | unsigned long prev_bio_flags, | |
005efedf FM |
2757 | unsigned long bio_flags, |
2758 | bool force_bio_submit) | |
d1310b2e CM |
2759 | { |
2760 | int ret = 0; | |
2761 | struct bio *bio; | |
09cbfeaf | 2762 | size_t page_size = min_t(size_t, size, PAGE_SIZE); |
6273b7f8 | 2763 | sector_t sector = offset >> 9; |
d1310b2e | 2764 | |
5c2b1fd7 DS |
2765 | ASSERT(bio_ret); |
2766 | ||
2767 | if (*bio_ret) { | |
0c8508a6 DS |
2768 | bool contig; |
2769 | bool can_merge = true; | |
2770 | ||
d1310b2e | 2771 | bio = *bio_ret; |
0c8508a6 | 2772 | if (prev_bio_flags & EXTENT_BIO_COMPRESSED) |
4f024f37 | 2773 | contig = bio->bi_iter.bi_sector == sector; |
c8b97818 | 2774 | else |
f73a1c7d | 2775 | contig = bio_end_sector(bio) == sector; |
c8b97818 | 2776 | |
da12fe54 NB |
2777 | ASSERT(tree->ops); |
2778 | if (btrfs_bio_fits_in_stripe(page, page_size, bio, bio_flags)) | |
0c8508a6 DS |
2779 | can_merge = false; |
2780 | ||
2781 | if (prev_bio_flags != bio_flags || !contig || !can_merge || | |
005efedf | 2782 | force_bio_submit || |
6c5a4e2c | 2783 | bio_add_page(bio, page, page_size, pg_offset) < page_size) { |
1f7ad75b | 2784 | ret = submit_one_bio(bio, mirror_num, prev_bio_flags); |
289454ad NA |
2785 | if (ret < 0) { |
2786 | *bio_ret = NULL; | |
79787eaa | 2787 | return ret; |
289454ad | 2788 | } |
d1310b2e CM |
2789 | bio = NULL; |
2790 | } else { | |
da2f0f74 CM |
2791 | if (wbc) |
2792 | wbc_account_io(wbc, page, page_size); | |
d1310b2e CM |
2793 | return 0; |
2794 | } | |
2795 | } | |
c8b97818 | 2796 | |
6273b7f8 | 2797 | bio = btrfs_bio_alloc(bdev, offset); |
6c5a4e2c | 2798 | bio_add_page(bio, page, page_size, pg_offset); |
d1310b2e CM |
2799 | bio->bi_end_io = end_io_func; |
2800 | bio->bi_private = tree; | |
e6959b93 | 2801 | bio->bi_write_hint = page->mapping->host->i_write_hint; |
4b81ba48 | 2802 | bio->bi_opf = opf; |
da2f0f74 CM |
2803 | if (wbc) { |
2804 | wbc_init_bio(wbc, bio); | |
2805 | wbc_account_io(wbc, page, page_size); | |
2806 | } | |
70dec807 | 2807 | |
5c2b1fd7 | 2808 | *bio_ret = bio; |
d1310b2e CM |
2809 | |
2810 | return ret; | |
2811 | } | |
2812 | ||
48a3b636 ES |
2813 | static void attach_extent_buffer_page(struct extent_buffer *eb, |
2814 | struct page *page) | |
d1310b2e CM |
2815 | { |
2816 | if (!PagePrivate(page)) { | |
2817 | SetPagePrivate(page); | |
09cbfeaf | 2818 | get_page(page); |
4f2de97a JB |
2819 | set_page_private(page, (unsigned long)eb); |
2820 | } else { | |
2821 | WARN_ON(page->private != (unsigned long)eb); | |
d1310b2e CM |
2822 | } |
2823 | } | |
2824 | ||
4f2de97a | 2825 | void set_page_extent_mapped(struct page *page) |
d1310b2e | 2826 | { |
4f2de97a JB |
2827 | if (!PagePrivate(page)) { |
2828 | SetPagePrivate(page); | |
09cbfeaf | 2829 | get_page(page); |
4f2de97a JB |
2830 | set_page_private(page, EXTENT_PAGE_PRIVATE); |
2831 | } | |
d1310b2e CM |
2832 | } |
2833 | ||
125bac01 MX |
2834 | static struct extent_map * |
2835 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
2836 | u64 start, u64 len, get_extent_t *get_extent, | |
2837 | struct extent_map **em_cached) | |
2838 | { | |
2839 | struct extent_map *em; | |
2840 | ||
2841 | if (em_cached && *em_cached) { | |
2842 | em = *em_cached; | |
cbc0e928 | 2843 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 | 2844 | start < extent_map_end(em)) { |
490b54d6 | 2845 | refcount_inc(&em->refs); |
125bac01 MX |
2846 | return em; |
2847 | } | |
2848 | ||
2849 | free_extent_map(em); | |
2850 | *em_cached = NULL; | |
2851 | } | |
2852 | ||
fc4f21b1 | 2853 | em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0); |
125bac01 MX |
2854 | if (em_cached && !IS_ERR_OR_NULL(em)) { |
2855 | BUG_ON(*em_cached); | |
490b54d6 | 2856 | refcount_inc(&em->refs); |
125bac01 MX |
2857 | *em_cached = em; |
2858 | } | |
2859 | return em; | |
2860 | } | |
d1310b2e CM |
2861 | /* |
2862 | * basic readpage implementation. Locked extent state structs are inserted | |
2863 | * into the tree that are removed when the IO is done (by the end_io | |
2864 | * handlers) | |
79787eaa | 2865 | * XXX JDM: This needs looking at to ensure proper page locking |
baf863b9 | 2866 | * return 0 on success, otherwise return error |
d1310b2e | 2867 | */ |
9974090b MX |
2868 | static int __do_readpage(struct extent_io_tree *tree, |
2869 | struct page *page, | |
2870 | get_extent_t *get_extent, | |
125bac01 | 2871 | struct extent_map **em_cached, |
9974090b | 2872 | struct bio **bio, int mirror_num, |
f1c77c55 | 2873 | unsigned long *bio_flags, unsigned int read_flags, |
005efedf | 2874 | u64 *prev_em_start) |
d1310b2e CM |
2875 | { |
2876 | struct inode *inode = page->mapping->host; | |
4eee4fa4 | 2877 | u64 start = page_offset(page); |
8eec8296 | 2878 | const u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
2879 | u64 cur = start; |
2880 | u64 extent_offset; | |
2881 | u64 last_byte = i_size_read(inode); | |
2882 | u64 block_start; | |
2883 | u64 cur_end; | |
d1310b2e CM |
2884 | struct extent_map *em; |
2885 | struct block_device *bdev; | |
baf863b9 | 2886 | int ret = 0; |
d1310b2e | 2887 | int nr = 0; |
306e16ce | 2888 | size_t pg_offset = 0; |
d1310b2e | 2889 | size_t iosize; |
c8b97818 | 2890 | size_t disk_io_size; |
d1310b2e | 2891 | size_t blocksize = inode->i_sb->s_blocksize; |
7f042a83 | 2892 | unsigned long this_bio_flag = 0; |
d1310b2e CM |
2893 | |
2894 | set_page_extent_mapped(page); | |
2895 | ||
90a887c9 DM |
2896 | if (!PageUptodate(page)) { |
2897 | if (cleancache_get_page(page) == 0) { | |
2898 | BUG_ON(blocksize != PAGE_SIZE); | |
9974090b | 2899 | unlock_extent(tree, start, end); |
90a887c9 DM |
2900 | goto out; |
2901 | } | |
2902 | } | |
2903 | ||
09cbfeaf | 2904 | if (page->index == last_byte >> PAGE_SHIFT) { |
c8b97818 | 2905 | char *userpage; |
7073017a | 2906 | size_t zero_offset = offset_in_page(last_byte); |
c8b97818 CM |
2907 | |
2908 | if (zero_offset) { | |
09cbfeaf | 2909 | iosize = PAGE_SIZE - zero_offset; |
7ac687d9 | 2910 | userpage = kmap_atomic(page); |
c8b97818 CM |
2911 | memset(userpage + zero_offset, 0, iosize); |
2912 | flush_dcache_page(page); | |
7ac687d9 | 2913 | kunmap_atomic(userpage); |
c8b97818 CM |
2914 | } |
2915 | } | |
d1310b2e | 2916 | while (cur <= end) { |
005efedf | 2917 | bool force_bio_submit = false; |
6273b7f8 | 2918 | u64 offset; |
c8f2f24b | 2919 | |
d1310b2e CM |
2920 | if (cur >= last_byte) { |
2921 | char *userpage; | |
507903b8 AJ |
2922 | struct extent_state *cached = NULL; |
2923 | ||
09cbfeaf | 2924 | iosize = PAGE_SIZE - pg_offset; |
7ac687d9 | 2925 | userpage = kmap_atomic(page); |
306e16ce | 2926 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2927 | flush_dcache_page(page); |
7ac687d9 | 2928 | kunmap_atomic(userpage); |
d1310b2e | 2929 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 2930 | &cached, GFP_NOFS); |
7f042a83 | 2931 | unlock_extent_cached(tree, cur, |
e43bbe5e | 2932 | cur + iosize - 1, &cached); |
d1310b2e CM |
2933 | break; |
2934 | } | |
125bac01 MX |
2935 | em = __get_extent_map(inode, page, pg_offset, cur, |
2936 | end - cur + 1, get_extent, em_cached); | |
c704005d | 2937 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 2938 | SetPageError(page); |
7f042a83 | 2939 | unlock_extent(tree, cur, end); |
d1310b2e CM |
2940 | break; |
2941 | } | |
d1310b2e CM |
2942 | extent_offset = cur - em->start; |
2943 | BUG_ON(extent_map_end(em) <= cur); | |
2944 | BUG_ON(end < cur); | |
2945 | ||
261507a0 | 2946 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
4b384318 | 2947 | this_bio_flag |= EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
2948 | extent_set_compress_type(&this_bio_flag, |
2949 | em->compress_type); | |
2950 | } | |
c8b97818 | 2951 | |
d1310b2e CM |
2952 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
2953 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 2954 | iosize = ALIGN(iosize, blocksize); |
c8b97818 CM |
2955 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
2956 | disk_io_size = em->block_len; | |
6273b7f8 | 2957 | offset = em->block_start; |
c8b97818 | 2958 | } else { |
6273b7f8 | 2959 | offset = em->block_start + extent_offset; |
c8b97818 CM |
2960 | disk_io_size = iosize; |
2961 | } | |
d1310b2e CM |
2962 | bdev = em->bdev; |
2963 | block_start = em->block_start; | |
d899e052 YZ |
2964 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2965 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
2966 | |
2967 | /* | |
2968 | * If we have a file range that points to a compressed extent | |
2969 | * and it's followed by a consecutive file range that points to | |
2970 | * to the same compressed extent (possibly with a different | |
2971 | * offset and/or length, so it either points to the whole extent | |
2972 | * or only part of it), we must make sure we do not submit a | |
2973 | * single bio to populate the pages for the 2 ranges because | |
2974 | * this makes the compressed extent read zero out the pages | |
2975 | * belonging to the 2nd range. Imagine the following scenario: | |
2976 | * | |
2977 | * File layout | |
2978 | * [0 - 8K] [8K - 24K] | |
2979 | * | | | |
2980 | * | | | |
2981 | * points to extent X, points to extent X, | |
2982 | * offset 4K, length of 8K offset 0, length 16K | |
2983 | * | |
2984 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
2985 | * | |
2986 | * If the bio to read the compressed extent covers both ranges, | |
2987 | * it will decompress extent X into the pages belonging to the | |
2988 | * first range and then it will stop, zeroing out the remaining | |
2989 | * pages that belong to the other range that points to extent X. | |
2990 | * So here we make sure we submit 2 bios, one for the first | |
2991 | * range and another one for the third range. Both will target | |
2992 | * the same physical extent from disk, but we can't currently | |
2993 | * make the compressed bio endio callback populate the pages | |
2994 | * for both ranges because each compressed bio is tightly | |
2995 | * coupled with a single extent map, and each range can have | |
2996 | * an extent map with a different offset value relative to the | |
2997 | * uncompressed data of our extent and different lengths. This | |
2998 | * is a corner case so we prioritize correctness over | |
2999 | * non-optimal behavior (submitting 2 bios for the same extent). | |
3000 | */ | |
3001 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
3002 | prev_em_start && *prev_em_start != (u64)-1 && | |
8e928218 | 3003 | *prev_em_start != em->start) |
005efedf FM |
3004 | force_bio_submit = true; |
3005 | ||
3006 | if (prev_em_start) | |
8e928218 | 3007 | *prev_em_start = em->start; |
005efedf | 3008 | |
d1310b2e CM |
3009 | free_extent_map(em); |
3010 | em = NULL; | |
3011 | ||
3012 | /* we've found a hole, just zero and go on */ | |
3013 | if (block_start == EXTENT_MAP_HOLE) { | |
3014 | char *userpage; | |
507903b8 AJ |
3015 | struct extent_state *cached = NULL; |
3016 | ||
7ac687d9 | 3017 | userpage = kmap_atomic(page); |
306e16ce | 3018 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3019 | flush_dcache_page(page); |
7ac687d9 | 3020 | kunmap_atomic(userpage); |
d1310b2e CM |
3021 | |
3022 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 | 3023 | &cached, GFP_NOFS); |
7f042a83 | 3024 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3025 | cur + iosize - 1, &cached); |
d1310b2e | 3026 | cur = cur + iosize; |
306e16ce | 3027 | pg_offset += iosize; |
d1310b2e CM |
3028 | continue; |
3029 | } | |
3030 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
3031 | if (test_range_bit(tree, cur, cur_end, |
3032 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 3033 | check_page_uptodate(tree, page); |
7f042a83 | 3034 | unlock_extent(tree, cur, cur + iosize - 1); |
d1310b2e | 3035 | cur = cur + iosize; |
306e16ce | 3036 | pg_offset += iosize; |
d1310b2e CM |
3037 | continue; |
3038 | } | |
70dec807 CM |
3039 | /* we have an inline extent but it didn't get marked up |
3040 | * to date. Error out | |
3041 | */ | |
3042 | if (block_start == EXTENT_MAP_INLINE) { | |
3043 | SetPageError(page); | |
7f042a83 | 3044 | unlock_extent(tree, cur, cur + iosize - 1); |
70dec807 | 3045 | cur = cur + iosize; |
306e16ce | 3046 | pg_offset += iosize; |
70dec807 CM |
3047 | continue; |
3048 | } | |
d1310b2e | 3049 | |
4b81ba48 | 3050 | ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL, |
6273b7f8 DS |
3051 | page, offset, disk_io_size, |
3052 | pg_offset, bdev, bio, | |
c8b97818 CM |
3053 | end_bio_extent_readpage, mirror_num, |
3054 | *bio_flags, | |
005efedf FM |
3055 | this_bio_flag, |
3056 | force_bio_submit); | |
c8f2f24b JB |
3057 | if (!ret) { |
3058 | nr++; | |
3059 | *bio_flags = this_bio_flag; | |
3060 | } else { | |
d1310b2e | 3061 | SetPageError(page); |
7f042a83 | 3062 | unlock_extent(tree, cur, cur + iosize - 1); |
baf863b9 | 3063 | goto out; |
edd33c99 | 3064 | } |
d1310b2e | 3065 | cur = cur + iosize; |
306e16ce | 3066 | pg_offset += iosize; |
d1310b2e | 3067 | } |
90a887c9 | 3068 | out: |
d1310b2e CM |
3069 | if (!nr) { |
3070 | if (!PageError(page)) | |
3071 | SetPageUptodate(page); | |
3072 | unlock_page(page); | |
3073 | } | |
baf863b9 | 3074 | return ret; |
d1310b2e CM |
3075 | } |
3076 | ||
e65ef21e | 3077 | static inline void contiguous_readpages(struct extent_io_tree *tree, |
9974090b MX |
3078 | struct page *pages[], int nr_pages, |
3079 | u64 start, u64 end, | |
125bac01 | 3080 | struct extent_map **em_cached, |
d3fac6ba | 3081 | struct bio **bio, |
1f7ad75b | 3082 | unsigned long *bio_flags, |
808f80b4 | 3083 | u64 *prev_em_start) |
9974090b MX |
3084 | { |
3085 | struct inode *inode; | |
3086 | struct btrfs_ordered_extent *ordered; | |
3087 | int index; | |
3088 | ||
3089 | inode = pages[0]->mapping->host; | |
3090 | while (1) { | |
3091 | lock_extent(tree, start, end); | |
a776c6fa | 3092 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
9974090b MX |
3093 | end - start + 1); |
3094 | if (!ordered) | |
3095 | break; | |
3096 | unlock_extent(tree, start, end); | |
3097 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3098 | btrfs_put_ordered_extent(ordered); | |
3099 | } | |
3100 | ||
3101 | for (index = 0; index < nr_pages; index++) { | |
4ef77695 | 3102 | __do_readpage(tree, pages[index], btrfs_get_extent, em_cached, |
5e9d3982 | 3103 | bio, 0, bio_flags, REQ_RAHEAD, prev_em_start); |
09cbfeaf | 3104 | put_page(pages[index]); |
9974090b MX |
3105 | } |
3106 | } | |
3107 | ||
9974090b MX |
3108 | static int __extent_read_full_page(struct extent_io_tree *tree, |
3109 | struct page *page, | |
3110 | get_extent_t *get_extent, | |
3111 | struct bio **bio, int mirror_num, | |
f1c77c55 DS |
3112 | unsigned long *bio_flags, |
3113 | unsigned int read_flags) | |
9974090b MX |
3114 | { |
3115 | struct inode *inode = page->mapping->host; | |
3116 | struct btrfs_ordered_extent *ordered; | |
3117 | u64 start = page_offset(page); | |
09cbfeaf | 3118 | u64 end = start + PAGE_SIZE - 1; |
9974090b MX |
3119 | int ret; |
3120 | ||
3121 | while (1) { | |
3122 | lock_extent(tree, start, end); | |
a776c6fa | 3123 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
09cbfeaf | 3124 | PAGE_SIZE); |
9974090b MX |
3125 | if (!ordered) |
3126 | break; | |
3127 | unlock_extent(tree, start, end); | |
3128 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3129 | btrfs_put_ordered_extent(ordered); | |
3130 | } | |
3131 | ||
125bac01 | 3132 | ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num, |
1f7ad75b | 3133 | bio_flags, read_flags, NULL); |
9974090b MX |
3134 | return ret; |
3135 | } | |
3136 | ||
d1310b2e | 3137 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, |
8ddc7d9c | 3138 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3139 | { |
3140 | struct bio *bio = NULL; | |
c8b97818 | 3141 | unsigned long bio_flags = 0; |
d1310b2e CM |
3142 | int ret; |
3143 | ||
8ddc7d9c | 3144 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, |
1f7ad75b | 3145 | &bio_flags, 0); |
d1310b2e | 3146 | if (bio) |
1f7ad75b | 3147 | ret = submit_one_bio(bio, mirror_num, bio_flags); |
d1310b2e CM |
3148 | return ret; |
3149 | } | |
d1310b2e | 3150 | |
3d4b9496 | 3151 | static void update_nr_written(struct writeback_control *wbc, |
a9132667 | 3152 | unsigned long nr_written) |
11c8349b CM |
3153 | { |
3154 | wbc->nr_to_write -= nr_written; | |
11c8349b CM |
3155 | } |
3156 | ||
d1310b2e | 3157 | /* |
40f76580 CM |
3158 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
3159 | * | |
5eaad97a | 3160 | * This returns 1 if btrfs_run_delalloc_range function did all the work required |
40f76580 CM |
3161 | * to write the page (copy into inline extent). In this case the IO has |
3162 | * been started and the page is already unlocked. | |
3163 | * | |
3164 | * This returns 0 if all went well (page still locked) | |
3165 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 3166 | */ |
40f76580 | 3167 | static noinline_for_stack int writepage_delalloc(struct inode *inode, |
8cc0237a NB |
3168 | struct page *page, struct writeback_control *wbc, |
3169 | u64 delalloc_start, unsigned long *nr_written) | |
40f76580 | 3170 | { |
8cc0237a | 3171 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
09cbfeaf | 3172 | u64 page_end = delalloc_start + PAGE_SIZE - 1; |
3522e903 | 3173 | bool found; |
40f76580 CM |
3174 | u64 delalloc_to_write = 0; |
3175 | u64 delalloc_end = 0; | |
3176 | int ret; | |
3177 | int page_started = 0; | |
3178 | ||
40f76580 CM |
3179 | |
3180 | while (delalloc_end < page_end) { | |
3522e903 | 3181 | found = find_lock_delalloc_range(inode, tree, |
40f76580 CM |
3182 | page, |
3183 | &delalloc_start, | |
917aacec | 3184 | &delalloc_end); |
3522e903 | 3185 | if (!found) { |
40f76580 CM |
3186 | delalloc_start = delalloc_end + 1; |
3187 | continue; | |
3188 | } | |
5eaad97a NB |
3189 | ret = btrfs_run_delalloc_range(inode, page, delalloc_start, |
3190 | delalloc_end, &page_started, nr_written, wbc); | |
40f76580 CM |
3191 | /* File system has been set read-only */ |
3192 | if (ret) { | |
3193 | SetPageError(page); | |
5eaad97a NB |
3194 | /* |
3195 | * btrfs_run_delalloc_range should return < 0 for error | |
3196 | * but just in case, we use > 0 here meaning the IO is | |
3197 | * started, so we don't want to return > 0 unless | |
3198 | * things are going well. | |
40f76580 CM |
3199 | */ |
3200 | ret = ret < 0 ? ret : -EIO; | |
3201 | goto done; | |
3202 | } | |
3203 | /* | |
ea1754a0 KS |
3204 | * delalloc_end is already one less than the total length, so |
3205 | * we don't subtract one from PAGE_SIZE | |
40f76580 CM |
3206 | */ |
3207 | delalloc_to_write += (delalloc_end - delalloc_start + | |
ea1754a0 | 3208 | PAGE_SIZE) >> PAGE_SHIFT; |
40f76580 CM |
3209 | delalloc_start = delalloc_end + 1; |
3210 | } | |
3211 | if (wbc->nr_to_write < delalloc_to_write) { | |
3212 | int thresh = 8192; | |
3213 | ||
3214 | if (delalloc_to_write < thresh * 2) | |
3215 | thresh = delalloc_to_write; | |
3216 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3217 | thresh); | |
3218 | } | |
3219 | ||
3220 | /* did the fill delalloc function already unlock and start | |
3221 | * the IO? | |
3222 | */ | |
3223 | if (page_started) { | |
3224 | /* | |
3225 | * we've unlocked the page, so we can't update | |
3226 | * the mapping's writeback index, just update | |
3227 | * nr_to_write. | |
3228 | */ | |
3229 | wbc->nr_to_write -= *nr_written; | |
3230 | return 1; | |
3231 | } | |
3232 | ||
3233 | ret = 0; | |
3234 | ||
3235 | done: | |
3236 | return ret; | |
3237 | } | |
3238 | ||
3239 | /* | |
3240 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3241 | * and does the loop to map the page into extents and bios. | |
3242 | * | |
3243 | * We return 1 if the IO is started and the page is unlocked, | |
3244 | * 0 if all went well (page still locked) | |
3245 | * < 0 if there were errors (page still locked) | |
3246 | */ | |
3247 | static noinline_for_stack int __extent_writepage_io(struct inode *inode, | |
3248 | struct page *page, | |
3249 | struct writeback_control *wbc, | |
3250 | struct extent_page_data *epd, | |
3251 | loff_t i_size, | |
3252 | unsigned long nr_written, | |
f1c77c55 | 3253 | unsigned int write_flags, int *nr_ret) |
d1310b2e | 3254 | { |
d1310b2e | 3255 | struct extent_io_tree *tree = epd->tree; |
4eee4fa4 | 3256 | u64 start = page_offset(page); |
09cbfeaf | 3257 | u64 page_end = start + PAGE_SIZE - 1; |
d1310b2e CM |
3258 | u64 end; |
3259 | u64 cur = start; | |
3260 | u64 extent_offset; | |
d1310b2e CM |
3261 | u64 block_start; |
3262 | u64 iosize; | |
d1310b2e CM |
3263 | struct extent_map *em; |
3264 | struct block_device *bdev; | |
7f3c74fb | 3265 | size_t pg_offset = 0; |
d1310b2e | 3266 | size_t blocksize; |
40f76580 CM |
3267 | int ret = 0; |
3268 | int nr = 0; | |
3269 | bool compressed; | |
c8b97818 | 3270 | |
d75855b4 NB |
3271 | ret = btrfs_writepage_cow_fixup(page, start, page_end); |
3272 | if (ret) { | |
3273 | /* Fixup worker will requeue */ | |
3274 | if (ret == -EBUSY) | |
3275 | wbc->pages_skipped++; | |
3276 | else | |
3277 | redirty_page_for_writepage(wbc, page); | |
40f76580 | 3278 | |
d75855b4 NB |
3279 | update_nr_written(wbc, nr_written); |
3280 | unlock_page(page); | |
3281 | return 1; | |
247e743c CM |
3282 | } |
3283 | ||
11c8349b CM |
3284 | /* |
3285 | * we don't want to touch the inode after unlocking the page, | |
3286 | * so we update the mapping writeback index now | |
3287 | */ | |
3d4b9496 | 3288 | update_nr_written(wbc, nr_written + 1); |
771ed689 | 3289 | |
d1310b2e | 3290 | end = page_end; |
40f76580 | 3291 | if (i_size <= start) { |
c629732d | 3292 | btrfs_writepage_endio_finish_ordered(page, start, page_end, 1); |
d1310b2e CM |
3293 | goto done; |
3294 | } | |
3295 | ||
d1310b2e CM |
3296 | blocksize = inode->i_sb->s_blocksize; |
3297 | ||
3298 | while (cur <= end) { | |
40f76580 | 3299 | u64 em_end; |
6273b7f8 | 3300 | u64 offset; |
58409edd | 3301 | |
40f76580 | 3302 | if (cur >= i_size) { |
7087a9d8 | 3303 | btrfs_writepage_endio_finish_ordered(page, cur, |
c629732d | 3304 | page_end, 1); |
d1310b2e CM |
3305 | break; |
3306 | } | |
3c98c62f | 3307 | em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, cur, |
d1310b2e | 3308 | end - cur + 1, 1); |
c704005d | 3309 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3310 | SetPageError(page); |
61391d56 | 3311 | ret = PTR_ERR_OR_ZERO(em); |
d1310b2e CM |
3312 | break; |
3313 | } | |
3314 | ||
3315 | extent_offset = cur - em->start; | |
40f76580 CM |
3316 | em_end = extent_map_end(em); |
3317 | BUG_ON(em_end <= cur); | |
d1310b2e | 3318 | BUG_ON(end < cur); |
40f76580 | 3319 | iosize = min(em_end - cur, end - cur + 1); |
fda2832f | 3320 | iosize = ALIGN(iosize, blocksize); |
6273b7f8 | 3321 | offset = em->block_start + extent_offset; |
d1310b2e CM |
3322 | bdev = em->bdev; |
3323 | block_start = em->block_start; | |
c8b97818 | 3324 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
3325 | free_extent_map(em); |
3326 | em = NULL; | |
3327 | ||
c8b97818 CM |
3328 | /* |
3329 | * compressed and inline extents are written through other | |
3330 | * paths in the FS | |
3331 | */ | |
3332 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3333 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
3334 | /* |
3335 | * end_io notification does not happen here for | |
3336 | * compressed extents | |
3337 | */ | |
7087a9d8 NB |
3338 | if (!compressed) |
3339 | btrfs_writepage_endio_finish_ordered(page, cur, | |
3340 | cur + iosize - 1, | |
c629732d | 3341 | 1); |
c8b97818 CM |
3342 | else if (compressed) { |
3343 | /* we don't want to end_page_writeback on | |
3344 | * a compressed extent. this happens | |
3345 | * elsewhere | |
3346 | */ | |
3347 | nr++; | |
3348 | } | |
3349 | ||
3350 | cur += iosize; | |
7f3c74fb | 3351 | pg_offset += iosize; |
d1310b2e CM |
3352 | continue; |
3353 | } | |
c8b97818 | 3354 | |
5cdc84bf | 3355 | btrfs_set_range_writeback(tree, cur, cur + iosize - 1); |
58409edd DS |
3356 | if (!PageWriteback(page)) { |
3357 | btrfs_err(BTRFS_I(inode)->root->fs_info, | |
3358 | "page %lu not writeback, cur %llu end %llu", | |
3359 | page->index, cur, end); | |
d1310b2e | 3360 | } |
7f3c74fb | 3361 | |
4b81ba48 | 3362 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc, |
6273b7f8 | 3363 | page, offset, iosize, pg_offset, |
c2df8bb4 | 3364 | bdev, &epd->bio, |
58409edd DS |
3365 | end_bio_extent_writepage, |
3366 | 0, 0, 0, false); | |
fe01aa65 | 3367 | if (ret) { |
58409edd | 3368 | SetPageError(page); |
fe01aa65 TK |
3369 | if (PageWriteback(page)) |
3370 | end_page_writeback(page); | |
3371 | } | |
d1310b2e | 3372 | |
d1310b2e | 3373 | cur = cur + iosize; |
7f3c74fb | 3374 | pg_offset += iosize; |
d1310b2e CM |
3375 | nr++; |
3376 | } | |
40f76580 CM |
3377 | done: |
3378 | *nr_ret = nr; | |
40f76580 CM |
3379 | return ret; |
3380 | } | |
3381 | ||
3382 | /* | |
3383 | * the writepage semantics are similar to regular writepage. extent | |
3384 | * records are inserted to lock ranges in the tree, and as dirty areas | |
3385 | * are found, they are marked writeback. Then the lock bits are removed | |
3386 | * and the end_io handler clears the writeback ranges | |
3387 | */ | |
3388 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
aab6e9ed | 3389 | struct extent_page_data *epd) |
40f76580 CM |
3390 | { |
3391 | struct inode *inode = page->mapping->host; | |
40f76580 | 3392 | u64 start = page_offset(page); |
09cbfeaf | 3393 | u64 page_end = start + PAGE_SIZE - 1; |
40f76580 CM |
3394 | int ret; |
3395 | int nr = 0; | |
3396 | size_t pg_offset = 0; | |
3397 | loff_t i_size = i_size_read(inode); | |
09cbfeaf | 3398 | unsigned long end_index = i_size >> PAGE_SHIFT; |
f1c77c55 | 3399 | unsigned int write_flags = 0; |
40f76580 CM |
3400 | unsigned long nr_written = 0; |
3401 | ||
ff40adf7 | 3402 | write_flags = wbc_to_write_flags(wbc); |
40f76580 CM |
3403 | |
3404 | trace___extent_writepage(page, inode, wbc); | |
3405 | ||
3406 | WARN_ON(!PageLocked(page)); | |
3407 | ||
3408 | ClearPageError(page); | |
3409 | ||
7073017a | 3410 | pg_offset = offset_in_page(i_size); |
40f76580 CM |
3411 | if (page->index > end_index || |
3412 | (page->index == end_index && !pg_offset)) { | |
09cbfeaf | 3413 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); |
40f76580 CM |
3414 | unlock_page(page); |
3415 | return 0; | |
3416 | } | |
3417 | ||
3418 | if (page->index == end_index) { | |
3419 | char *userpage; | |
3420 | ||
3421 | userpage = kmap_atomic(page); | |
3422 | memset(userpage + pg_offset, 0, | |
09cbfeaf | 3423 | PAGE_SIZE - pg_offset); |
40f76580 CM |
3424 | kunmap_atomic(userpage); |
3425 | flush_dcache_page(page); | |
3426 | } | |
3427 | ||
3428 | pg_offset = 0; | |
3429 | ||
3430 | set_page_extent_mapped(page); | |
3431 | ||
7789a55a | 3432 | if (!epd->extent_locked) { |
8cc0237a | 3433 | ret = writepage_delalloc(inode, page, wbc, start, &nr_written); |
7789a55a NB |
3434 | if (ret == 1) |
3435 | goto done_unlocked; | |
3436 | if (ret) | |
3437 | goto done; | |
3438 | } | |
40f76580 CM |
3439 | |
3440 | ret = __extent_writepage_io(inode, page, wbc, epd, | |
3441 | i_size, nr_written, write_flags, &nr); | |
3442 | if (ret == 1) | |
3443 | goto done_unlocked; | |
3444 | ||
d1310b2e CM |
3445 | done: |
3446 | if (nr == 0) { | |
3447 | /* make sure the mapping tag for page dirty gets cleared */ | |
3448 | set_page_writeback(page); | |
3449 | end_page_writeback(page); | |
3450 | } | |
61391d56 FM |
3451 | if (PageError(page)) { |
3452 | ret = ret < 0 ? ret : -EIO; | |
3453 | end_extent_writepage(page, ret, start, page_end); | |
3454 | } | |
d1310b2e | 3455 | unlock_page(page); |
40f76580 | 3456 | return ret; |
771ed689 | 3457 | |
11c8349b | 3458 | done_unlocked: |
d1310b2e CM |
3459 | return 0; |
3460 | } | |
3461 | ||
fd8b2b61 | 3462 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 3463 | { |
74316201 N |
3464 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
3465 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
3466 | } |
3467 | ||
0e378df1 CM |
3468 | static noinline_for_stack int |
3469 | lock_extent_buffer_for_io(struct extent_buffer *eb, | |
3470 | struct btrfs_fs_info *fs_info, | |
3471 | struct extent_page_data *epd) | |
0b32f4bb | 3472 | { |
cc5e31a4 | 3473 | int i, num_pages; |
0b32f4bb JB |
3474 | int flush = 0; |
3475 | int ret = 0; | |
3476 | ||
3477 | if (!btrfs_try_tree_write_lock(eb)) { | |
3478 | flush = 1; | |
3479 | flush_write_bio(epd); | |
3480 | btrfs_tree_lock(eb); | |
3481 | } | |
3482 | ||
3483 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3484 | btrfs_tree_unlock(eb); | |
3485 | if (!epd->sync_io) | |
3486 | return 0; | |
3487 | if (!flush) { | |
3488 | flush_write_bio(epd); | |
3489 | flush = 1; | |
3490 | } | |
a098d8e8 CM |
3491 | while (1) { |
3492 | wait_on_extent_buffer_writeback(eb); | |
3493 | btrfs_tree_lock(eb); | |
3494 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3495 | break; | |
0b32f4bb | 3496 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3497 | } |
3498 | } | |
3499 | ||
51561ffe JB |
3500 | /* |
3501 | * We need to do this to prevent races in people who check if the eb is | |
3502 | * under IO since we can end up having no IO bits set for a short period | |
3503 | * of time. | |
3504 | */ | |
3505 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
3506 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3507 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 3508 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 3509 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
104b4e51 NB |
3510 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
3511 | -eb->len, | |
3512 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 3513 | ret = 1; |
51561ffe JB |
3514 | } else { |
3515 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
3516 | } |
3517 | ||
3518 | btrfs_tree_unlock(eb); | |
3519 | ||
3520 | if (!ret) | |
3521 | return ret; | |
3522 | ||
65ad0104 | 3523 | num_pages = num_extent_pages(eb); |
0b32f4bb | 3524 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3525 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3526 | |
3527 | if (!trylock_page(p)) { | |
3528 | if (!flush) { | |
3529 | flush_write_bio(epd); | |
3530 | flush = 1; | |
3531 | } | |
3532 | lock_page(p); | |
3533 | } | |
3534 | } | |
3535 | ||
3536 | return ret; | |
3537 | } | |
3538 | ||
3539 | static void end_extent_buffer_writeback(struct extent_buffer *eb) | |
3540 | { | |
3541 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
4e857c58 | 3542 | smp_mb__after_atomic(); |
0b32f4bb JB |
3543 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); |
3544 | } | |
3545 | ||
656f30db FM |
3546 | static void set_btree_ioerr(struct page *page) |
3547 | { | |
3548 | struct extent_buffer *eb = (struct extent_buffer *)page->private; | |
656f30db FM |
3549 | |
3550 | SetPageError(page); | |
3551 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) | |
3552 | return; | |
3553 | ||
3554 | /* | |
3555 | * If writeback for a btree extent that doesn't belong to a log tree | |
3556 | * failed, increment the counter transaction->eb_write_errors. | |
3557 | * We do this because while the transaction is running and before it's | |
3558 | * committing (when we call filemap_fdata[write|wait]_range against | |
3559 | * the btree inode), we might have | |
3560 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
3561 | * returns an error or an error happens during writeback, when we're | |
3562 | * committing the transaction we wouldn't know about it, since the pages | |
3563 | * can be no longer dirty nor marked anymore for writeback (if a | |
3564 | * subsequent modification to the extent buffer didn't happen before the | |
3565 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
3566 | * able to find the pages tagged with SetPageError at transaction | |
3567 | * commit time. So if this happens we must abort the transaction, | |
3568 | * otherwise we commit a super block with btree roots that point to | |
3569 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
3570 | * or the content of some node/leaf from a past generation that got | |
3571 | * cowed or deleted and is no longer valid. | |
3572 | * | |
3573 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
3574 | * not be enough - we need to distinguish between log tree extents vs | |
3575 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
3576 | * will catch and clear such errors in the mapping - and that call might | |
3577 | * be from a log sync and not from a transaction commit. Also, checking | |
3578 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
3579 | * not done and would not be reliable - the eb might have been released | |
3580 | * from memory and reading it back again means that flag would not be | |
3581 | * set (since it's a runtime flag, not persisted on disk). | |
3582 | * | |
3583 | * Using the flags below in the btree inode also makes us achieve the | |
3584 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
3585 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
3586 | * is called, the writeback for all dirty pages had already finished | |
3587 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
3588 | * filemap_fdatawait_range() would return success, as it could not know | |
3589 | * that writeback errors happened (the pages were no longer tagged for | |
3590 | * writeback). | |
3591 | */ | |
3592 | switch (eb->log_index) { | |
3593 | case -1: | |
afcdd129 | 3594 | set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags); |
656f30db FM |
3595 | break; |
3596 | case 0: | |
afcdd129 | 3597 | set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags); |
656f30db FM |
3598 | break; |
3599 | case 1: | |
afcdd129 | 3600 | set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags); |
656f30db FM |
3601 | break; |
3602 | default: | |
3603 | BUG(); /* unexpected, logic error */ | |
3604 | } | |
3605 | } | |
3606 | ||
4246a0b6 | 3607 | static void end_bio_extent_buffer_writepage(struct bio *bio) |
0b32f4bb | 3608 | { |
2c30c71b | 3609 | struct bio_vec *bvec; |
0b32f4bb | 3610 | struct extent_buffer *eb; |
2c30c71b | 3611 | int i, done; |
6dc4f100 | 3612 | struct bvec_iter_all iter_all; |
0b32f4bb | 3613 | |
c09abff8 | 3614 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
6dc4f100 | 3615 | bio_for_each_segment_all(bvec, bio, i, iter_all) { |
0b32f4bb JB |
3616 | struct page *page = bvec->bv_page; |
3617 | ||
0b32f4bb JB |
3618 | eb = (struct extent_buffer *)page->private; |
3619 | BUG_ON(!eb); | |
3620 | done = atomic_dec_and_test(&eb->io_pages); | |
3621 | ||
4e4cbee9 | 3622 | if (bio->bi_status || |
4246a0b6 | 3623 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { |
0b32f4bb | 3624 | ClearPageUptodate(page); |
656f30db | 3625 | set_btree_ioerr(page); |
0b32f4bb JB |
3626 | } |
3627 | ||
3628 | end_page_writeback(page); | |
3629 | ||
3630 | if (!done) | |
3631 | continue; | |
3632 | ||
3633 | end_extent_buffer_writeback(eb); | |
2c30c71b | 3634 | } |
0b32f4bb JB |
3635 | |
3636 | bio_put(bio); | |
0b32f4bb JB |
3637 | } |
3638 | ||
0e378df1 | 3639 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb JB |
3640 | struct btrfs_fs_info *fs_info, |
3641 | struct writeback_control *wbc, | |
3642 | struct extent_page_data *epd) | |
3643 | { | |
3644 | struct block_device *bdev = fs_info->fs_devices->latest_bdev; | |
f28491e0 | 3645 | struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree; |
0b32f4bb | 3646 | u64 offset = eb->start; |
851cd173 | 3647 | u32 nritems; |
cc5e31a4 | 3648 | int i, num_pages; |
851cd173 | 3649 | unsigned long start, end; |
ff40adf7 | 3650 | unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META; |
d7dbe9e7 | 3651 | int ret = 0; |
0b32f4bb | 3652 | |
656f30db | 3653 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); |
65ad0104 | 3654 | num_pages = num_extent_pages(eb); |
0b32f4bb | 3655 | atomic_set(&eb->io_pages, num_pages); |
de0022b9 | 3656 | |
851cd173 LB |
3657 | /* set btree blocks beyond nritems with 0 to avoid stale content. */ |
3658 | nritems = btrfs_header_nritems(eb); | |
3eb548ee | 3659 | if (btrfs_header_level(eb) > 0) { |
3eb548ee LB |
3660 | end = btrfs_node_key_ptr_offset(nritems); |
3661 | ||
b159fa28 | 3662 | memzero_extent_buffer(eb, end, eb->len - end); |
851cd173 LB |
3663 | } else { |
3664 | /* | |
3665 | * leaf: | |
3666 | * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0 | |
3667 | */ | |
3668 | start = btrfs_item_nr_offset(nritems); | |
3d9ec8c4 | 3669 | end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, eb); |
b159fa28 | 3670 | memzero_extent_buffer(eb, start, end - start); |
3eb548ee LB |
3671 | } |
3672 | ||
0b32f4bb | 3673 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3674 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3675 | |
3676 | clear_page_dirty_for_io(p); | |
3677 | set_page_writeback(p); | |
4b81ba48 | 3678 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc, |
6273b7f8 | 3679 | p, offset, PAGE_SIZE, 0, bdev, |
c2df8bb4 | 3680 | &epd->bio, |
1f7ad75b | 3681 | end_bio_extent_buffer_writepage, |
18fdc679 | 3682 | 0, 0, 0, false); |
0b32f4bb | 3683 | if (ret) { |
656f30db | 3684 | set_btree_ioerr(p); |
fe01aa65 TK |
3685 | if (PageWriteback(p)) |
3686 | end_page_writeback(p); | |
0b32f4bb JB |
3687 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
3688 | end_extent_buffer_writeback(eb); | |
3689 | ret = -EIO; | |
3690 | break; | |
3691 | } | |
09cbfeaf | 3692 | offset += PAGE_SIZE; |
3d4b9496 | 3693 | update_nr_written(wbc, 1); |
0b32f4bb JB |
3694 | unlock_page(p); |
3695 | } | |
3696 | ||
3697 | if (unlikely(ret)) { | |
3698 | for (; i < num_pages; i++) { | |
bbf65cf0 | 3699 | struct page *p = eb->pages[i]; |
81465028 | 3700 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
3701 | unlock_page(p); |
3702 | } | |
3703 | } | |
3704 | ||
3705 | return ret; | |
3706 | } | |
3707 | ||
3708 | int btree_write_cache_pages(struct address_space *mapping, | |
3709 | struct writeback_control *wbc) | |
3710 | { | |
3711 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; | |
3712 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; | |
3713 | struct extent_buffer *eb, *prev_eb = NULL; | |
3714 | struct extent_page_data epd = { | |
3715 | .bio = NULL, | |
3716 | .tree = tree, | |
3717 | .extent_locked = 0, | |
3718 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
3719 | }; | |
3720 | int ret = 0; | |
3721 | int done = 0; | |
3722 | int nr_to_write_done = 0; | |
3723 | struct pagevec pvec; | |
3724 | int nr_pages; | |
3725 | pgoff_t index; | |
3726 | pgoff_t end; /* Inclusive */ | |
3727 | int scanned = 0; | |
10bbd235 | 3728 | xa_mark_t tag; |
0b32f4bb | 3729 | |
86679820 | 3730 | pagevec_init(&pvec); |
0b32f4bb JB |
3731 | if (wbc->range_cyclic) { |
3732 | index = mapping->writeback_index; /* Start from prev offset */ | |
3733 | end = -1; | |
3734 | } else { | |
09cbfeaf KS |
3735 | index = wbc->range_start >> PAGE_SHIFT; |
3736 | end = wbc->range_end >> PAGE_SHIFT; | |
0b32f4bb JB |
3737 | scanned = 1; |
3738 | } | |
3739 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3740 | tag = PAGECACHE_TAG_TOWRITE; | |
3741 | else | |
3742 | tag = PAGECACHE_TAG_DIRTY; | |
3743 | retry: | |
3744 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3745 | tag_pages_for_writeback(mapping, index, end); | |
3746 | while (!done && !nr_to_write_done && (index <= end) && | |
4006f437 | 3747 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 3748 | tag))) { |
0b32f4bb JB |
3749 | unsigned i; |
3750 | ||
3751 | scanned = 1; | |
3752 | for (i = 0; i < nr_pages; i++) { | |
3753 | struct page *page = pvec.pages[i]; | |
3754 | ||
3755 | if (!PagePrivate(page)) | |
3756 | continue; | |
3757 | ||
b5bae261 JB |
3758 | spin_lock(&mapping->private_lock); |
3759 | if (!PagePrivate(page)) { | |
3760 | spin_unlock(&mapping->private_lock); | |
3761 | continue; | |
3762 | } | |
3763 | ||
0b32f4bb | 3764 | eb = (struct extent_buffer *)page->private; |
b5bae261 JB |
3765 | |
3766 | /* | |
3767 | * Shouldn't happen and normally this would be a BUG_ON | |
3768 | * but no sense in crashing the users box for something | |
3769 | * we can survive anyway. | |
3770 | */ | |
fae7f21c | 3771 | if (WARN_ON(!eb)) { |
b5bae261 | 3772 | spin_unlock(&mapping->private_lock); |
0b32f4bb JB |
3773 | continue; |
3774 | } | |
3775 | ||
b5bae261 JB |
3776 | if (eb == prev_eb) { |
3777 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 3778 | continue; |
b5bae261 | 3779 | } |
0b32f4bb | 3780 | |
b5bae261 JB |
3781 | ret = atomic_inc_not_zero(&eb->refs); |
3782 | spin_unlock(&mapping->private_lock); | |
3783 | if (!ret) | |
0b32f4bb | 3784 | continue; |
0b32f4bb JB |
3785 | |
3786 | prev_eb = eb; | |
3787 | ret = lock_extent_buffer_for_io(eb, fs_info, &epd); | |
3788 | if (!ret) { | |
3789 | free_extent_buffer(eb); | |
3790 | continue; | |
3791 | } | |
3792 | ||
3793 | ret = write_one_eb(eb, fs_info, wbc, &epd); | |
3794 | if (ret) { | |
3795 | done = 1; | |
3796 | free_extent_buffer(eb); | |
3797 | break; | |
3798 | } | |
3799 | free_extent_buffer(eb); | |
3800 | ||
3801 | /* | |
3802 | * the filesystem may choose to bump up nr_to_write. | |
3803 | * We have to make sure to honor the new nr_to_write | |
3804 | * at any time | |
3805 | */ | |
3806 | nr_to_write_done = wbc->nr_to_write <= 0; | |
3807 | } | |
3808 | pagevec_release(&pvec); | |
3809 | cond_resched(); | |
3810 | } | |
3811 | if (!scanned && !done) { | |
3812 | /* | |
3813 | * We hit the last page and there is more work to be done: wrap | |
3814 | * back to the start of the file | |
3815 | */ | |
3816 | scanned = 1; | |
3817 | index = 0; | |
3818 | goto retry; | |
3819 | } | |
3820 | flush_write_bio(&epd); | |
3821 | return ret; | |
3822 | } | |
3823 | ||
d1310b2e | 3824 | /** |
4bef0848 | 3825 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
3826 | * @mapping: address space structure to write |
3827 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
935db853 | 3828 | * @data: data passed to __extent_writepage function |
d1310b2e CM |
3829 | * |
3830 | * If a page is already under I/O, write_cache_pages() skips it, even | |
3831 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
3832 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
3833 | * and msync() need to guarantee that all the data which was dirty at the time | |
3834 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
3835 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
3836 | * existing IO to complete. | |
3837 | */ | |
4242b64a | 3838 | static int extent_write_cache_pages(struct address_space *mapping, |
4bef0848 | 3839 | struct writeback_control *wbc, |
aab6e9ed | 3840 | struct extent_page_data *epd) |
d1310b2e | 3841 | { |
7fd1a3f7 | 3842 | struct inode *inode = mapping->host; |
d1310b2e CM |
3843 | int ret = 0; |
3844 | int done = 0; | |
f85d7d6c | 3845 | int nr_to_write_done = 0; |
d1310b2e CM |
3846 | struct pagevec pvec; |
3847 | int nr_pages; | |
3848 | pgoff_t index; | |
3849 | pgoff_t end; /* Inclusive */ | |
a9132667 LB |
3850 | pgoff_t done_index; |
3851 | int range_whole = 0; | |
d1310b2e | 3852 | int scanned = 0; |
10bbd235 | 3853 | xa_mark_t tag; |
d1310b2e | 3854 | |
7fd1a3f7 JB |
3855 | /* |
3856 | * We have to hold onto the inode so that ordered extents can do their | |
3857 | * work when the IO finishes. The alternative to this is failing to add | |
3858 | * an ordered extent if the igrab() fails there and that is a huge pain | |
3859 | * to deal with, so instead just hold onto the inode throughout the | |
3860 | * writepages operation. If it fails here we are freeing up the inode | |
3861 | * anyway and we'd rather not waste our time writing out stuff that is | |
3862 | * going to be truncated anyway. | |
3863 | */ | |
3864 | if (!igrab(inode)) | |
3865 | return 0; | |
3866 | ||
86679820 | 3867 | pagevec_init(&pvec); |
d1310b2e CM |
3868 | if (wbc->range_cyclic) { |
3869 | index = mapping->writeback_index; /* Start from prev offset */ | |
3870 | end = -1; | |
3871 | } else { | |
09cbfeaf KS |
3872 | index = wbc->range_start >> PAGE_SHIFT; |
3873 | end = wbc->range_end >> PAGE_SHIFT; | |
a9132667 LB |
3874 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
3875 | range_whole = 1; | |
d1310b2e CM |
3876 | scanned = 1; |
3877 | } | |
3cd24c69 EL |
3878 | |
3879 | /* | |
3880 | * We do the tagged writepage as long as the snapshot flush bit is set | |
3881 | * and we are the first one who do the filemap_flush() on this inode. | |
3882 | * | |
3883 | * The nr_to_write == LONG_MAX is needed to make sure other flushers do | |
3884 | * not race in and drop the bit. | |
3885 | */ | |
3886 | if (range_whole && wbc->nr_to_write == LONG_MAX && | |
3887 | test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
3888 | &BTRFS_I(inode)->runtime_flags)) | |
3889 | wbc->tagged_writepages = 1; | |
3890 | ||
3891 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) | |
f7aaa06b JB |
3892 | tag = PAGECACHE_TAG_TOWRITE; |
3893 | else | |
3894 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 3895 | retry: |
3cd24c69 | 3896 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
f7aaa06b | 3897 | tag_pages_for_writeback(mapping, index, end); |
a9132667 | 3898 | done_index = index; |
f85d7d6c | 3899 | while (!done && !nr_to_write_done && (index <= end) && |
67fd707f JK |
3900 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, |
3901 | &index, end, tag))) { | |
d1310b2e CM |
3902 | unsigned i; |
3903 | ||
3904 | scanned = 1; | |
3905 | for (i = 0; i < nr_pages; i++) { | |
3906 | struct page *page = pvec.pages[i]; | |
3907 | ||
a9132667 | 3908 | done_index = page->index; |
d1310b2e | 3909 | /* |
b93b0163 MW |
3910 | * At this point we hold neither the i_pages lock nor |
3911 | * the page lock: the page may be truncated or | |
3912 | * invalidated (changing page->mapping to NULL), | |
3913 | * or even swizzled back from swapper_space to | |
3914 | * tmpfs file mapping | |
d1310b2e | 3915 | */ |
c8f2f24b | 3916 | if (!trylock_page(page)) { |
aab6e9ed | 3917 | flush_write_bio(epd); |
c8f2f24b | 3918 | lock_page(page); |
01d658f2 | 3919 | } |
d1310b2e CM |
3920 | |
3921 | if (unlikely(page->mapping != mapping)) { | |
3922 | unlock_page(page); | |
3923 | continue; | |
3924 | } | |
3925 | ||
d2c3f4f6 | 3926 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 | 3927 | if (PageWriteback(page)) |
aab6e9ed | 3928 | flush_write_bio(epd); |
d1310b2e | 3929 | wait_on_page_writeback(page); |
d2c3f4f6 | 3930 | } |
d1310b2e CM |
3931 | |
3932 | if (PageWriteback(page) || | |
3933 | !clear_page_dirty_for_io(page)) { | |
3934 | unlock_page(page); | |
3935 | continue; | |
3936 | } | |
3937 | ||
aab6e9ed | 3938 | ret = __extent_writepage(page, wbc, epd); |
d1310b2e CM |
3939 | |
3940 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
3941 | unlock_page(page); | |
3942 | ret = 0; | |
3943 | } | |
a9132667 LB |
3944 | if (ret < 0) { |
3945 | /* | |
3946 | * done_index is set past this page, | |
3947 | * so media errors will not choke | |
3948 | * background writeout for the entire | |
3949 | * file. This has consequences for | |
3950 | * range_cyclic semantics (ie. it may | |
3951 | * not be suitable for data integrity | |
3952 | * writeout). | |
3953 | */ | |
3954 | done_index = page->index + 1; | |
3955 | done = 1; | |
3956 | break; | |
3957 | } | |
f85d7d6c CM |
3958 | |
3959 | /* | |
3960 | * the filesystem may choose to bump up nr_to_write. | |
3961 | * We have to make sure to honor the new nr_to_write | |
3962 | * at any time | |
3963 | */ | |
3964 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
3965 | } |
3966 | pagevec_release(&pvec); | |
3967 | cond_resched(); | |
3968 | } | |
894b36e3 | 3969 | if (!scanned && !done) { |
d1310b2e CM |
3970 | /* |
3971 | * We hit the last page and there is more work to be done: wrap | |
3972 | * back to the start of the file | |
3973 | */ | |
3974 | scanned = 1; | |
3975 | index = 0; | |
3976 | goto retry; | |
3977 | } | |
a9132667 LB |
3978 | |
3979 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole)) | |
3980 | mapping->writeback_index = done_index; | |
3981 | ||
7fd1a3f7 | 3982 | btrfs_add_delayed_iput(inode); |
894b36e3 | 3983 | return ret; |
d1310b2e | 3984 | } |
d1310b2e | 3985 | |
0a9b0e53 | 3986 | int extent_write_full_page(struct page *page, struct writeback_control *wbc) |
d1310b2e CM |
3987 | { |
3988 | int ret; | |
d1310b2e CM |
3989 | struct extent_page_data epd = { |
3990 | .bio = NULL, | |
0a9b0e53 | 3991 | .tree = &BTRFS_I(page->mapping->host)->io_tree, |
771ed689 | 3992 | .extent_locked = 0, |
ffbd517d | 3993 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e | 3994 | }; |
d1310b2e | 3995 | |
d1310b2e CM |
3996 | ret = __extent_writepage(page, wbc, &epd); |
3997 | ||
e2932ee0 | 3998 | flush_write_bio(&epd); |
d1310b2e CM |
3999 | return ret; |
4000 | } | |
d1310b2e | 4001 | |
5e3ee236 | 4002 | int extent_write_locked_range(struct inode *inode, u64 start, u64 end, |
771ed689 CM |
4003 | int mode) |
4004 | { | |
4005 | int ret = 0; | |
4006 | struct address_space *mapping = inode->i_mapping; | |
5e3ee236 | 4007 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
771ed689 | 4008 | struct page *page; |
09cbfeaf KS |
4009 | unsigned long nr_pages = (end - start + PAGE_SIZE) >> |
4010 | PAGE_SHIFT; | |
771ed689 CM |
4011 | |
4012 | struct extent_page_data epd = { | |
4013 | .bio = NULL, | |
4014 | .tree = tree, | |
771ed689 | 4015 | .extent_locked = 1, |
ffbd517d | 4016 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
4017 | }; |
4018 | struct writeback_control wbc_writepages = { | |
771ed689 | 4019 | .sync_mode = mode, |
771ed689 CM |
4020 | .nr_to_write = nr_pages * 2, |
4021 | .range_start = start, | |
4022 | .range_end = end + 1, | |
4023 | }; | |
4024 | ||
d397712b | 4025 | while (start <= end) { |
09cbfeaf | 4026 | page = find_get_page(mapping, start >> PAGE_SHIFT); |
771ed689 CM |
4027 | if (clear_page_dirty_for_io(page)) |
4028 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
4029 | else { | |
7087a9d8 | 4030 | btrfs_writepage_endio_finish_ordered(page, start, |
c629732d | 4031 | start + PAGE_SIZE - 1, 1); |
771ed689 CM |
4032 | unlock_page(page); |
4033 | } | |
09cbfeaf KS |
4034 | put_page(page); |
4035 | start += PAGE_SIZE; | |
771ed689 CM |
4036 | } |
4037 | ||
e2932ee0 | 4038 | flush_write_bio(&epd); |
771ed689 CM |
4039 | return ret; |
4040 | } | |
d1310b2e | 4041 | |
8ae225a8 | 4042 | int extent_writepages(struct address_space *mapping, |
d1310b2e CM |
4043 | struct writeback_control *wbc) |
4044 | { | |
4045 | int ret = 0; | |
4046 | struct extent_page_data epd = { | |
4047 | .bio = NULL, | |
8ae225a8 | 4048 | .tree = &BTRFS_I(mapping->host)->io_tree, |
771ed689 | 4049 | .extent_locked = 0, |
ffbd517d | 4050 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
4051 | }; |
4052 | ||
935db853 | 4053 | ret = extent_write_cache_pages(mapping, wbc, &epd); |
e2932ee0 | 4054 | flush_write_bio(&epd); |
d1310b2e CM |
4055 | return ret; |
4056 | } | |
d1310b2e | 4057 | |
2a3ff0ad NB |
4058 | int extent_readpages(struct address_space *mapping, struct list_head *pages, |
4059 | unsigned nr_pages) | |
d1310b2e CM |
4060 | { |
4061 | struct bio *bio = NULL; | |
c8b97818 | 4062 | unsigned long bio_flags = 0; |
67c9684f | 4063 | struct page *pagepool[16]; |
125bac01 | 4064 | struct extent_map *em_cached = NULL; |
2a3ff0ad | 4065 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; |
67c9684f | 4066 | int nr = 0; |
808f80b4 | 4067 | u64 prev_em_start = (u64)-1; |
d1310b2e | 4068 | |
61ed3a14 | 4069 | while (!list_empty(pages)) { |
e65ef21e NB |
4070 | u64 contig_end = 0; |
4071 | ||
61ed3a14 | 4072 | for (nr = 0; nr < ARRAY_SIZE(pagepool) && !list_empty(pages);) { |
f86196ea | 4073 | struct page *page = lru_to_page(pages); |
d1310b2e | 4074 | |
61ed3a14 NB |
4075 | prefetchw(&page->flags); |
4076 | list_del(&page->lru); | |
4077 | if (add_to_page_cache_lru(page, mapping, page->index, | |
4078 | readahead_gfp_mask(mapping))) { | |
4079 | put_page(page); | |
e65ef21e | 4080 | break; |
61ed3a14 NB |
4081 | } |
4082 | ||
4083 | pagepool[nr++] = page; | |
e65ef21e | 4084 | contig_end = page_offset(page) + PAGE_SIZE - 1; |
d1310b2e | 4085 | } |
67c9684f | 4086 | |
e65ef21e NB |
4087 | if (nr) { |
4088 | u64 contig_start = page_offset(pagepool[0]); | |
4089 | ||
4090 | ASSERT(contig_start + nr * PAGE_SIZE - 1 == contig_end); | |
4091 | ||
4092 | contiguous_readpages(tree, pagepool, nr, contig_start, | |
4093 | contig_end, &em_cached, &bio, &bio_flags, | |
4094 | &prev_em_start); | |
4095 | } | |
d1310b2e | 4096 | } |
67c9684f | 4097 | |
125bac01 MX |
4098 | if (em_cached) |
4099 | free_extent_map(em_cached); | |
4100 | ||
d1310b2e | 4101 | if (bio) |
1f7ad75b | 4102 | return submit_one_bio(bio, 0, bio_flags); |
d1310b2e CM |
4103 | return 0; |
4104 | } | |
d1310b2e CM |
4105 | |
4106 | /* | |
4107 | * basic invalidatepage code, this waits on any locked or writeback | |
4108 | * ranges corresponding to the page, and then deletes any extent state | |
4109 | * records from the tree | |
4110 | */ | |
4111 | int extent_invalidatepage(struct extent_io_tree *tree, | |
4112 | struct page *page, unsigned long offset) | |
4113 | { | |
2ac55d41 | 4114 | struct extent_state *cached_state = NULL; |
4eee4fa4 | 4115 | u64 start = page_offset(page); |
09cbfeaf | 4116 | u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
4117 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; |
4118 | ||
fda2832f | 4119 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
4120 | if (start > end) |
4121 | return 0; | |
4122 | ||
ff13db41 | 4123 | lock_extent_bits(tree, start, end, &cached_state); |
1edbb734 | 4124 | wait_on_page_writeback(page); |
d1310b2e | 4125 | clear_extent_bit(tree, start, end, |
32c00aff JB |
4126 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
4127 | EXTENT_DO_ACCOUNTING, | |
ae0f1625 | 4128 | 1, 1, &cached_state); |
d1310b2e CM |
4129 | return 0; |
4130 | } | |
d1310b2e | 4131 | |
7b13b7b1 CM |
4132 | /* |
4133 | * a helper for releasepage, this tests for areas of the page that | |
4134 | * are locked or under IO and drops the related state bits if it is safe | |
4135 | * to drop the page. | |
4136 | */ | |
29c68b2d | 4137 | static int try_release_extent_state(struct extent_io_tree *tree, |
48a3b636 | 4138 | struct page *page, gfp_t mask) |
7b13b7b1 | 4139 | { |
4eee4fa4 | 4140 | u64 start = page_offset(page); |
09cbfeaf | 4141 | u64 end = start + PAGE_SIZE - 1; |
7b13b7b1 CM |
4142 | int ret = 1; |
4143 | ||
8882679e | 4144 | if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) { |
7b13b7b1 | 4145 | ret = 0; |
8882679e | 4146 | } else { |
11ef160f CM |
4147 | /* |
4148 | * at this point we can safely clear everything except the | |
4149 | * locked bit and the nodatasum bit | |
4150 | */ | |
66b0c887 | 4151 | ret = __clear_extent_bit(tree, start, end, |
11ef160f | 4152 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
66b0c887 | 4153 | 0, 0, NULL, mask, NULL); |
e3f24cc5 CM |
4154 | |
4155 | /* if clear_extent_bit failed for enomem reasons, | |
4156 | * we can't allow the release to continue. | |
4157 | */ | |
4158 | if (ret < 0) | |
4159 | ret = 0; | |
4160 | else | |
4161 | ret = 1; | |
7b13b7b1 CM |
4162 | } |
4163 | return ret; | |
4164 | } | |
7b13b7b1 | 4165 | |
d1310b2e CM |
4166 | /* |
4167 | * a helper for releasepage. As long as there are no locked extents | |
4168 | * in the range corresponding to the page, both state records and extent | |
4169 | * map records are removed | |
4170 | */ | |
477a30ba | 4171 | int try_release_extent_mapping(struct page *page, gfp_t mask) |
d1310b2e CM |
4172 | { |
4173 | struct extent_map *em; | |
4eee4fa4 | 4174 | u64 start = page_offset(page); |
09cbfeaf | 4175 | u64 end = start + PAGE_SIZE - 1; |
bd3599a0 FM |
4176 | struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host); |
4177 | struct extent_io_tree *tree = &btrfs_inode->io_tree; | |
4178 | struct extent_map_tree *map = &btrfs_inode->extent_tree; | |
7b13b7b1 | 4179 | |
d0164adc | 4180 | if (gfpflags_allow_blocking(mask) && |
ee22184b | 4181 | page->mapping->host->i_size > SZ_16M) { |
39b5637f | 4182 | u64 len; |
70dec807 | 4183 | while (start <= end) { |
39b5637f | 4184 | len = end - start + 1; |
890871be | 4185 | write_lock(&map->lock); |
39b5637f | 4186 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 4187 | if (!em) { |
890871be | 4188 | write_unlock(&map->lock); |
70dec807 CM |
4189 | break; |
4190 | } | |
7f3c74fb CM |
4191 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
4192 | em->start != start) { | |
890871be | 4193 | write_unlock(&map->lock); |
70dec807 CM |
4194 | free_extent_map(em); |
4195 | break; | |
4196 | } | |
4197 | if (!test_range_bit(tree, em->start, | |
4198 | extent_map_end(em) - 1, | |
4e586ca3 | 4199 | EXTENT_LOCKED, 0, NULL)) { |
bd3599a0 FM |
4200 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
4201 | &btrfs_inode->runtime_flags); | |
70dec807 CM |
4202 | remove_extent_mapping(map, em); |
4203 | /* once for the rb tree */ | |
4204 | free_extent_map(em); | |
4205 | } | |
4206 | start = extent_map_end(em); | |
890871be | 4207 | write_unlock(&map->lock); |
70dec807 CM |
4208 | |
4209 | /* once for us */ | |
d1310b2e CM |
4210 | free_extent_map(em); |
4211 | } | |
d1310b2e | 4212 | } |
29c68b2d | 4213 | return try_release_extent_state(tree, page, mask); |
d1310b2e | 4214 | } |
d1310b2e | 4215 | |
ec29ed5b CM |
4216 | /* |
4217 | * helper function for fiemap, which doesn't want to see any holes. | |
4218 | * This maps until we find something past 'last' | |
4219 | */ | |
4220 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
e3350e16 | 4221 | u64 offset, u64 last) |
ec29ed5b | 4222 | { |
da17066c | 4223 | u64 sectorsize = btrfs_inode_sectorsize(inode); |
ec29ed5b CM |
4224 | struct extent_map *em; |
4225 | u64 len; | |
4226 | ||
4227 | if (offset >= last) | |
4228 | return NULL; | |
4229 | ||
67871254 | 4230 | while (1) { |
ec29ed5b CM |
4231 | len = last - offset; |
4232 | if (len == 0) | |
4233 | break; | |
fda2832f | 4234 | len = ALIGN(len, sectorsize); |
4ab47a8d | 4235 | em = btrfs_get_extent_fiemap(BTRFS_I(inode), offset, len); |
c704005d | 4236 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
4237 | return em; |
4238 | ||
4239 | /* if this isn't a hole return it */ | |
4a2d25cd | 4240 | if (em->block_start != EXTENT_MAP_HOLE) |
ec29ed5b | 4241 | return em; |
ec29ed5b CM |
4242 | |
4243 | /* this is a hole, advance to the next extent */ | |
4244 | offset = extent_map_end(em); | |
4245 | free_extent_map(em); | |
4246 | if (offset >= last) | |
4247 | break; | |
4248 | } | |
4249 | return NULL; | |
4250 | } | |
4251 | ||
4751832d QW |
4252 | /* |
4253 | * To cache previous fiemap extent | |
4254 | * | |
4255 | * Will be used for merging fiemap extent | |
4256 | */ | |
4257 | struct fiemap_cache { | |
4258 | u64 offset; | |
4259 | u64 phys; | |
4260 | u64 len; | |
4261 | u32 flags; | |
4262 | bool cached; | |
4263 | }; | |
4264 | ||
4265 | /* | |
4266 | * Helper to submit fiemap extent. | |
4267 | * | |
4268 | * Will try to merge current fiemap extent specified by @offset, @phys, | |
4269 | * @len and @flags with cached one. | |
4270 | * And only when we fails to merge, cached one will be submitted as | |
4271 | * fiemap extent. | |
4272 | * | |
4273 | * Return value is the same as fiemap_fill_next_extent(). | |
4274 | */ | |
4275 | static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, | |
4276 | struct fiemap_cache *cache, | |
4277 | u64 offset, u64 phys, u64 len, u32 flags) | |
4278 | { | |
4279 | int ret = 0; | |
4280 | ||
4281 | if (!cache->cached) | |
4282 | goto assign; | |
4283 | ||
4284 | /* | |
4285 | * Sanity check, extent_fiemap() should have ensured that new | |
52042d8e | 4286 | * fiemap extent won't overlap with cached one. |
4751832d QW |
4287 | * Not recoverable. |
4288 | * | |
4289 | * NOTE: Physical address can overlap, due to compression | |
4290 | */ | |
4291 | if (cache->offset + cache->len > offset) { | |
4292 | WARN_ON(1); | |
4293 | return -EINVAL; | |
4294 | } | |
4295 | ||
4296 | /* | |
4297 | * Only merges fiemap extents if | |
4298 | * 1) Their logical addresses are continuous | |
4299 | * | |
4300 | * 2) Their physical addresses are continuous | |
4301 | * So truly compressed (physical size smaller than logical size) | |
4302 | * extents won't get merged with each other | |
4303 | * | |
4304 | * 3) Share same flags except FIEMAP_EXTENT_LAST | |
4305 | * So regular extent won't get merged with prealloc extent | |
4306 | */ | |
4307 | if (cache->offset + cache->len == offset && | |
4308 | cache->phys + cache->len == phys && | |
4309 | (cache->flags & ~FIEMAP_EXTENT_LAST) == | |
4310 | (flags & ~FIEMAP_EXTENT_LAST)) { | |
4311 | cache->len += len; | |
4312 | cache->flags |= flags; | |
4313 | goto try_submit_last; | |
4314 | } | |
4315 | ||
4316 | /* Not mergeable, need to submit cached one */ | |
4317 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, | |
4318 | cache->len, cache->flags); | |
4319 | cache->cached = false; | |
4320 | if (ret) | |
4321 | return ret; | |
4322 | assign: | |
4323 | cache->cached = true; | |
4324 | cache->offset = offset; | |
4325 | cache->phys = phys; | |
4326 | cache->len = len; | |
4327 | cache->flags = flags; | |
4328 | try_submit_last: | |
4329 | if (cache->flags & FIEMAP_EXTENT_LAST) { | |
4330 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, | |
4331 | cache->phys, cache->len, cache->flags); | |
4332 | cache->cached = false; | |
4333 | } | |
4334 | return ret; | |
4335 | } | |
4336 | ||
4337 | /* | |
848c23b7 | 4338 | * Emit last fiemap cache |
4751832d | 4339 | * |
848c23b7 QW |
4340 | * The last fiemap cache may still be cached in the following case: |
4341 | * 0 4k 8k | |
4342 | * |<- Fiemap range ->| | |
4343 | * |<------------ First extent ----------->| | |
4344 | * | |
4345 | * In this case, the first extent range will be cached but not emitted. | |
4346 | * So we must emit it before ending extent_fiemap(). | |
4751832d | 4347 | */ |
848c23b7 QW |
4348 | static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info, |
4349 | struct fiemap_extent_info *fieinfo, | |
4350 | struct fiemap_cache *cache) | |
4751832d QW |
4351 | { |
4352 | int ret; | |
4353 | ||
4354 | if (!cache->cached) | |
4355 | return 0; | |
4356 | ||
4751832d QW |
4357 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, |
4358 | cache->len, cache->flags); | |
4359 | cache->cached = false; | |
4360 | if (ret > 0) | |
4361 | ret = 0; | |
4362 | return ret; | |
4363 | } | |
4364 | ||
1506fcc8 | 4365 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
2135fb9b | 4366 | __u64 start, __u64 len) |
1506fcc8 | 4367 | { |
975f84fe | 4368 | int ret = 0; |
1506fcc8 YS |
4369 | u64 off = start; |
4370 | u64 max = start + len; | |
4371 | u32 flags = 0; | |
975f84fe JB |
4372 | u32 found_type; |
4373 | u64 last; | |
ec29ed5b | 4374 | u64 last_for_get_extent = 0; |
1506fcc8 | 4375 | u64 disko = 0; |
ec29ed5b | 4376 | u64 isize = i_size_read(inode); |
975f84fe | 4377 | struct btrfs_key found_key; |
1506fcc8 | 4378 | struct extent_map *em = NULL; |
2ac55d41 | 4379 | struct extent_state *cached_state = NULL; |
975f84fe | 4380 | struct btrfs_path *path; |
dc046b10 | 4381 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4751832d | 4382 | struct fiemap_cache cache = { 0 }; |
1506fcc8 | 4383 | int end = 0; |
ec29ed5b CM |
4384 | u64 em_start = 0; |
4385 | u64 em_len = 0; | |
4386 | u64 em_end = 0; | |
1506fcc8 YS |
4387 | |
4388 | if (len == 0) | |
4389 | return -EINVAL; | |
4390 | ||
975f84fe JB |
4391 | path = btrfs_alloc_path(); |
4392 | if (!path) | |
4393 | return -ENOMEM; | |
4394 | path->leave_spinning = 1; | |
4395 | ||
da17066c JM |
4396 | start = round_down(start, btrfs_inode_sectorsize(inode)); |
4397 | len = round_up(max, btrfs_inode_sectorsize(inode)) - start; | |
4d479cf0 | 4398 | |
ec29ed5b CM |
4399 | /* |
4400 | * lookup the last file extent. We're not using i_size here | |
4401 | * because there might be preallocation past i_size | |
4402 | */ | |
f85b7379 DS |
4403 | ret = btrfs_lookup_file_extent(NULL, root, path, |
4404 | btrfs_ino(BTRFS_I(inode)), -1, 0); | |
975f84fe JB |
4405 | if (ret < 0) { |
4406 | btrfs_free_path(path); | |
4407 | return ret; | |
2d324f59 LB |
4408 | } else { |
4409 | WARN_ON(!ret); | |
4410 | if (ret == 1) | |
4411 | ret = 0; | |
975f84fe | 4412 | } |
2d324f59 | 4413 | |
975f84fe | 4414 | path->slots[0]--; |
975f84fe | 4415 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
962a298f | 4416 | found_type = found_key.type; |
975f84fe | 4417 | |
ec29ed5b | 4418 | /* No extents, but there might be delalloc bits */ |
4a0cc7ca | 4419 | if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) || |
975f84fe | 4420 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
4421 | /* have to trust i_size as the end */ |
4422 | last = (u64)-1; | |
4423 | last_for_get_extent = isize; | |
4424 | } else { | |
4425 | /* | |
4426 | * remember the start of the last extent. There are a | |
4427 | * bunch of different factors that go into the length of the | |
4428 | * extent, so its much less complex to remember where it started | |
4429 | */ | |
4430 | last = found_key.offset; | |
4431 | last_for_get_extent = last + 1; | |
975f84fe | 4432 | } |
fe09e16c | 4433 | btrfs_release_path(path); |
975f84fe | 4434 | |
ec29ed5b CM |
4435 | /* |
4436 | * we might have some extents allocated but more delalloc past those | |
4437 | * extents. so, we trust isize unless the start of the last extent is | |
4438 | * beyond isize | |
4439 | */ | |
4440 | if (last < isize) { | |
4441 | last = (u64)-1; | |
4442 | last_for_get_extent = isize; | |
4443 | } | |
4444 | ||
ff13db41 | 4445 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
d0082371 | 4446 | &cached_state); |
ec29ed5b | 4447 | |
e3350e16 | 4448 | em = get_extent_skip_holes(inode, start, last_for_get_extent); |
1506fcc8 YS |
4449 | if (!em) |
4450 | goto out; | |
4451 | if (IS_ERR(em)) { | |
4452 | ret = PTR_ERR(em); | |
4453 | goto out; | |
4454 | } | |
975f84fe | 4455 | |
1506fcc8 | 4456 | while (!end) { |
b76bb701 | 4457 | u64 offset_in_extent = 0; |
ea8efc74 CM |
4458 | |
4459 | /* break if the extent we found is outside the range */ | |
4460 | if (em->start >= max || extent_map_end(em) < off) | |
4461 | break; | |
4462 | ||
4463 | /* | |
4464 | * get_extent may return an extent that starts before our | |
4465 | * requested range. We have to make sure the ranges | |
4466 | * we return to fiemap always move forward and don't | |
4467 | * overlap, so adjust the offsets here | |
4468 | */ | |
4469 | em_start = max(em->start, off); | |
1506fcc8 | 4470 | |
ea8efc74 CM |
4471 | /* |
4472 | * record the offset from the start of the extent | |
b76bb701 JB |
4473 | * for adjusting the disk offset below. Only do this if the |
4474 | * extent isn't compressed since our in ram offset may be past | |
4475 | * what we have actually allocated on disk. | |
ea8efc74 | 4476 | */ |
b76bb701 JB |
4477 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
4478 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 4479 | em_end = extent_map_end(em); |
ea8efc74 | 4480 | em_len = em_end - em_start; |
1506fcc8 | 4481 | flags = 0; |
f0986318 FM |
4482 | if (em->block_start < EXTENT_MAP_LAST_BYTE) |
4483 | disko = em->block_start + offset_in_extent; | |
4484 | else | |
4485 | disko = 0; | |
1506fcc8 | 4486 | |
ea8efc74 CM |
4487 | /* |
4488 | * bump off for our next call to get_extent | |
4489 | */ | |
4490 | off = extent_map_end(em); | |
4491 | if (off >= max) | |
4492 | end = 1; | |
4493 | ||
93dbfad7 | 4494 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
4495 | end = 1; |
4496 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 4497 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
4498 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
4499 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 4500 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
4501 | flags |= (FIEMAP_EXTENT_DELALLOC | |
4502 | FIEMAP_EXTENT_UNKNOWN); | |
dc046b10 JB |
4503 | } else if (fieinfo->fi_extents_max) { |
4504 | u64 bytenr = em->block_start - | |
4505 | (em->start - em->orig_start); | |
fe09e16c | 4506 | |
fe09e16c LB |
4507 | /* |
4508 | * As btrfs supports shared space, this information | |
4509 | * can be exported to userspace tools via | |
dc046b10 JB |
4510 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 |
4511 | * then we're just getting a count and we can skip the | |
4512 | * lookup stuff. | |
fe09e16c | 4513 | */ |
bb739cf0 EN |
4514 | ret = btrfs_check_shared(root, |
4515 | btrfs_ino(BTRFS_I(inode)), | |
4516 | bytenr); | |
dc046b10 | 4517 | if (ret < 0) |
fe09e16c | 4518 | goto out_free; |
dc046b10 | 4519 | if (ret) |
fe09e16c | 4520 | flags |= FIEMAP_EXTENT_SHARED; |
dc046b10 | 4521 | ret = 0; |
1506fcc8 YS |
4522 | } |
4523 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
4524 | flags |= FIEMAP_EXTENT_ENCODED; | |
0d2b2372 JB |
4525 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
4526 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
1506fcc8 | 4527 | |
1506fcc8 YS |
4528 | free_extent_map(em); |
4529 | em = NULL; | |
ec29ed5b CM |
4530 | if ((em_start >= last) || em_len == (u64)-1 || |
4531 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
4532 | flags |= FIEMAP_EXTENT_LAST; |
4533 | end = 1; | |
4534 | } | |
4535 | ||
ec29ed5b | 4536 | /* now scan forward to see if this is really the last extent. */ |
e3350e16 | 4537 | em = get_extent_skip_holes(inode, off, last_for_get_extent); |
ec29ed5b CM |
4538 | if (IS_ERR(em)) { |
4539 | ret = PTR_ERR(em); | |
4540 | goto out; | |
4541 | } | |
4542 | if (!em) { | |
975f84fe JB |
4543 | flags |= FIEMAP_EXTENT_LAST; |
4544 | end = 1; | |
4545 | } | |
4751832d QW |
4546 | ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko, |
4547 | em_len, flags); | |
26e726af CS |
4548 | if (ret) { |
4549 | if (ret == 1) | |
4550 | ret = 0; | |
ec29ed5b | 4551 | goto out_free; |
26e726af | 4552 | } |
1506fcc8 YS |
4553 | } |
4554 | out_free: | |
4751832d | 4555 | if (!ret) |
848c23b7 | 4556 | ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache); |
1506fcc8 YS |
4557 | free_extent_map(em); |
4558 | out: | |
fe09e16c | 4559 | btrfs_free_path(path); |
a52f4cd2 | 4560 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
e43bbe5e | 4561 | &cached_state); |
1506fcc8 YS |
4562 | return ret; |
4563 | } | |
4564 | ||
727011e0 CM |
4565 | static void __free_extent_buffer(struct extent_buffer *eb) |
4566 | { | |
6d49ba1b | 4567 | btrfs_leak_debug_del(&eb->leak_list); |
727011e0 CM |
4568 | kmem_cache_free(extent_buffer_cache, eb); |
4569 | } | |
4570 | ||
a26e8c9f | 4571 | int extent_buffer_under_io(struct extent_buffer *eb) |
db7f3436 JB |
4572 | { |
4573 | return (atomic_read(&eb->io_pages) || | |
4574 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4575 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4576 | } | |
4577 | ||
4578 | /* | |
55ac0139 | 4579 | * Release all pages attached to the extent buffer. |
db7f3436 | 4580 | */ |
55ac0139 | 4581 | static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb) |
db7f3436 | 4582 | { |
d64766fd NB |
4583 | int i; |
4584 | int num_pages; | |
b0132a3b | 4585 | int mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
db7f3436 JB |
4586 | |
4587 | BUG_ON(extent_buffer_under_io(eb)); | |
4588 | ||
d64766fd NB |
4589 | num_pages = num_extent_pages(eb); |
4590 | for (i = 0; i < num_pages; i++) { | |
4591 | struct page *page = eb->pages[i]; | |
db7f3436 | 4592 | |
5d2361db FL |
4593 | if (!page) |
4594 | continue; | |
4595 | if (mapped) | |
db7f3436 | 4596 | spin_lock(&page->mapping->private_lock); |
5d2361db FL |
4597 | /* |
4598 | * We do this since we'll remove the pages after we've | |
4599 | * removed the eb from the radix tree, so we could race | |
4600 | * and have this page now attached to the new eb. So | |
4601 | * only clear page_private if it's still connected to | |
4602 | * this eb. | |
4603 | */ | |
4604 | if (PagePrivate(page) && | |
4605 | page->private == (unsigned long)eb) { | |
4606 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4607 | BUG_ON(PageDirty(page)); | |
4608 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 4609 | /* |
5d2361db FL |
4610 | * We need to make sure we haven't be attached |
4611 | * to a new eb. | |
db7f3436 | 4612 | */ |
5d2361db FL |
4613 | ClearPagePrivate(page); |
4614 | set_page_private(page, 0); | |
4615 | /* One for the page private */ | |
09cbfeaf | 4616 | put_page(page); |
db7f3436 | 4617 | } |
5d2361db FL |
4618 | |
4619 | if (mapped) | |
4620 | spin_unlock(&page->mapping->private_lock); | |
4621 | ||
01327610 | 4622 | /* One for when we allocated the page */ |
09cbfeaf | 4623 | put_page(page); |
d64766fd | 4624 | } |
db7f3436 JB |
4625 | } |
4626 | ||
4627 | /* | |
4628 | * Helper for releasing the extent buffer. | |
4629 | */ | |
4630 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4631 | { | |
55ac0139 | 4632 | btrfs_release_extent_buffer_pages(eb); |
db7f3436 JB |
4633 | __free_extent_buffer(eb); |
4634 | } | |
4635 | ||
f28491e0 JB |
4636 | static struct extent_buffer * |
4637 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 4638 | unsigned long len) |
d1310b2e CM |
4639 | { |
4640 | struct extent_buffer *eb = NULL; | |
4641 | ||
d1b5c567 | 4642 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
4643 | eb->start = start; |
4644 | eb->len = len; | |
f28491e0 | 4645 | eb->fs_info = fs_info; |
815a51c7 | 4646 | eb->bflags = 0; |
bd681513 | 4647 | rwlock_init(&eb->lock); |
bd681513 CM |
4648 | atomic_set(&eb->blocking_readers, 0); |
4649 | atomic_set(&eb->blocking_writers, 0); | |
ed1b4ed7 | 4650 | eb->lock_nested = false; |
bd681513 CM |
4651 | init_waitqueue_head(&eb->write_lock_wq); |
4652 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 4653 | |
6d49ba1b ES |
4654 | btrfs_leak_debug_add(&eb->leak_list, &buffers); |
4655 | ||
3083ee2e | 4656 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 4657 | atomic_set(&eb->refs, 1); |
0b32f4bb | 4658 | atomic_set(&eb->io_pages, 0); |
727011e0 | 4659 | |
b8dae313 DS |
4660 | /* |
4661 | * Sanity checks, currently the maximum is 64k covered by 16x 4k pages | |
4662 | */ | |
4663 | BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE | |
4664 | > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
4665 | BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
d1310b2e | 4666 | |
843ccf9f DS |
4667 | #ifdef CONFIG_BTRFS_DEBUG |
4668 | atomic_set(&eb->spinning_writers, 0); | |
afd495a8 | 4669 | atomic_set(&eb->spinning_readers, 0); |
5c9c799a | 4670 | atomic_set(&eb->read_locks, 0); |
c79adfc0 | 4671 | atomic_set(&eb->write_locks, 0); |
843ccf9f DS |
4672 | #endif |
4673 | ||
d1310b2e CM |
4674 | return eb; |
4675 | } | |
4676 | ||
815a51c7 JS |
4677 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) |
4678 | { | |
cc5e31a4 | 4679 | int i; |
815a51c7 JS |
4680 | struct page *p; |
4681 | struct extent_buffer *new; | |
cc5e31a4 | 4682 | int num_pages = num_extent_pages(src); |
815a51c7 | 4683 | |
3f556f78 | 4684 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
4685 | if (new == NULL) |
4686 | return NULL; | |
4687 | ||
4688 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4689 | p = alloc_page(GFP_NOFS); |
db7f3436 JB |
4690 | if (!p) { |
4691 | btrfs_release_extent_buffer(new); | |
4692 | return NULL; | |
4693 | } | |
815a51c7 JS |
4694 | attach_extent_buffer_page(new, p); |
4695 | WARN_ON(PageDirty(p)); | |
4696 | SetPageUptodate(p); | |
4697 | new->pages[i] = p; | |
fba1acf9 | 4698 | copy_page(page_address(p), page_address(src->pages[i])); |
815a51c7 JS |
4699 | } |
4700 | ||
815a51c7 | 4701 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); |
b0132a3b | 4702 | set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags); |
815a51c7 JS |
4703 | |
4704 | return new; | |
4705 | } | |
4706 | ||
0f331229 OS |
4707 | struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
4708 | u64 start, unsigned long len) | |
815a51c7 JS |
4709 | { |
4710 | struct extent_buffer *eb; | |
cc5e31a4 DS |
4711 | int num_pages; |
4712 | int i; | |
815a51c7 | 4713 | |
3f556f78 | 4714 | eb = __alloc_extent_buffer(fs_info, start, len); |
815a51c7 JS |
4715 | if (!eb) |
4716 | return NULL; | |
4717 | ||
65ad0104 | 4718 | num_pages = num_extent_pages(eb); |
815a51c7 | 4719 | for (i = 0; i < num_pages; i++) { |
9ec72677 | 4720 | eb->pages[i] = alloc_page(GFP_NOFS); |
815a51c7 JS |
4721 | if (!eb->pages[i]) |
4722 | goto err; | |
4723 | } | |
4724 | set_extent_buffer_uptodate(eb); | |
4725 | btrfs_set_header_nritems(eb, 0); | |
b0132a3b | 4726 | set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
815a51c7 JS |
4727 | |
4728 | return eb; | |
4729 | err: | |
84167d19 SB |
4730 | for (; i > 0; i--) |
4731 | __free_page(eb->pages[i - 1]); | |
815a51c7 JS |
4732 | __free_extent_buffer(eb); |
4733 | return NULL; | |
4734 | } | |
4735 | ||
0f331229 | 4736 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
da17066c | 4737 | u64 start) |
0f331229 | 4738 | { |
da17066c | 4739 | return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize); |
0f331229 OS |
4740 | } |
4741 | ||
0b32f4bb JB |
4742 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
4743 | { | |
242e18c7 | 4744 | int refs; |
0b32f4bb JB |
4745 | /* the ref bit is tricky. We have to make sure it is set |
4746 | * if we have the buffer dirty. Otherwise the | |
4747 | * code to free a buffer can end up dropping a dirty | |
4748 | * page | |
4749 | * | |
4750 | * Once the ref bit is set, it won't go away while the | |
4751 | * buffer is dirty or in writeback, and it also won't | |
4752 | * go away while we have the reference count on the | |
4753 | * eb bumped. | |
4754 | * | |
4755 | * We can't just set the ref bit without bumping the | |
4756 | * ref on the eb because free_extent_buffer might | |
4757 | * see the ref bit and try to clear it. If this happens | |
4758 | * free_extent_buffer might end up dropping our original | |
4759 | * ref by mistake and freeing the page before we are able | |
4760 | * to add one more ref. | |
4761 | * | |
4762 | * So bump the ref count first, then set the bit. If someone | |
4763 | * beat us to it, drop the ref we added. | |
4764 | */ | |
242e18c7 CM |
4765 | refs = atomic_read(&eb->refs); |
4766 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
4767 | return; | |
4768 | ||
594831c4 JB |
4769 | spin_lock(&eb->refs_lock); |
4770 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 4771 | atomic_inc(&eb->refs); |
594831c4 | 4772 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
4773 | } |
4774 | ||
2457aec6 MG |
4775 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
4776 | struct page *accessed) | |
5df4235e | 4777 | { |
cc5e31a4 | 4778 | int num_pages, i; |
5df4235e | 4779 | |
0b32f4bb JB |
4780 | check_buffer_tree_ref(eb); |
4781 | ||
65ad0104 | 4782 | num_pages = num_extent_pages(eb); |
5df4235e | 4783 | for (i = 0; i < num_pages; i++) { |
fb85fc9a DS |
4784 | struct page *p = eb->pages[i]; |
4785 | ||
2457aec6 MG |
4786 | if (p != accessed) |
4787 | mark_page_accessed(p); | |
5df4235e JB |
4788 | } |
4789 | } | |
4790 | ||
f28491e0 JB |
4791 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
4792 | u64 start) | |
452c75c3 CS |
4793 | { |
4794 | struct extent_buffer *eb; | |
4795 | ||
4796 | rcu_read_lock(); | |
f28491e0 | 4797 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
09cbfeaf | 4798 | start >> PAGE_SHIFT); |
452c75c3 CS |
4799 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
4800 | rcu_read_unlock(); | |
062c19e9 FM |
4801 | /* |
4802 | * Lock our eb's refs_lock to avoid races with | |
4803 | * free_extent_buffer. When we get our eb it might be flagged | |
4804 | * with EXTENT_BUFFER_STALE and another task running | |
4805 | * free_extent_buffer might have seen that flag set, | |
4806 | * eb->refs == 2, that the buffer isn't under IO (dirty and | |
4807 | * writeback flags not set) and it's still in the tree (flag | |
4808 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process | |
4809 | * of decrementing the extent buffer's reference count twice. | |
4810 | * So here we could race and increment the eb's reference count, | |
4811 | * clear its stale flag, mark it as dirty and drop our reference | |
4812 | * before the other task finishes executing free_extent_buffer, | |
4813 | * which would later result in an attempt to free an extent | |
4814 | * buffer that is dirty. | |
4815 | */ | |
4816 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
4817 | spin_lock(&eb->refs_lock); | |
4818 | spin_unlock(&eb->refs_lock); | |
4819 | } | |
2457aec6 | 4820 | mark_extent_buffer_accessed(eb, NULL); |
452c75c3 CS |
4821 | return eb; |
4822 | } | |
4823 | rcu_read_unlock(); | |
4824 | ||
4825 | return NULL; | |
4826 | } | |
4827 | ||
faa2dbf0 JB |
4828 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
4829 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
da17066c | 4830 | u64 start) |
faa2dbf0 JB |
4831 | { |
4832 | struct extent_buffer *eb, *exists = NULL; | |
4833 | int ret; | |
4834 | ||
4835 | eb = find_extent_buffer(fs_info, start); | |
4836 | if (eb) | |
4837 | return eb; | |
da17066c | 4838 | eb = alloc_dummy_extent_buffer(fs_info, start); |
faa2dbf0 JB |
4839 | if (!eb) |
4840 | return NULL; | |
4841 | eb->fs_info = fs_info; | |
4842 | again: | |
e1860a77 | 4843 | ret = radix_tree_preload(GFP_NOFS); |
faa2dbf0 JB |
4844 | if (ret) |
4845 | goto free_eb; | |
4846 | spin_lock(&fs_info->buffer_lock); | |
4847 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
09cbfeaf | 4848 | start >> PAGE_SHIFT, eb); |
faa2dbf0 JB |
4849 | spin_unlock(&fs_info->buffer_lock); |
4850 | radix_tree_preload_end(); | |
4851 | if (ret == -EEXIST) { | |
4852 | exists = find_extent_buffer(fs_info, start); | |
4853 | if (exists) | |
4854 | goto free_eb; | |
4855 | else | |
4856 | goto again; | |
4857 | } | |
4858 | check_buffer_tree_ref(eb); | |
4859 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
4860 | ||
faa2dbf0 JB |
4861 | return eb; |
4862 | free_eb: | |
4863 | btrfs_release_extent_buffer(eb); | |
4864 | return exists; | |
4865 | } | |
4866 | #endif | |
4867 | ||
f28491e0 | 4868 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
ce3e6984 | 4869 | u64 start) |
d1310b2e | 4870 | { |
da17066c | 4871 | unsigned long len = fs_info->nodesize; |
cc5e31a4 DS |
4872 | int num_pages; |
4873 | int i; | |
09cbfeaf | 4874 | unsigned long index = start >> PAGE_SHIFT; |
d1310b2e | 4875 | struct extent_buffer *eb; |
6af118ce | 4876 | struct extent_buffer *exists = NULL; |
d1310b2e | 4877 | struct page *p; |
f28491e0 | 4878 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
d1310b2e | 4879 | int uptodate = 1; |
19fe0a8b | 4880 | int ret; |
d1310b2e | 4881 | |
da17066c | 4882 | if (!IS_ALIGNED(start, fs_info->sectorsize)) { |
c871b0f2 LB |
4883 | btrfs_err(fs_info, "bad tree block start %llu", start); |
4884 | return ERR_PTR(-EINVAL); | |
4885 | } | |
4886 | ||
f28491e0 | 4887 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 4888 | if (eb) |
6af118ce | 4889 | return eb; |
6af118ce | 4890 | |
23d79d81 | 4891 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 4892 | if (!eb) |
c871b0f2 | 4893 | return ERR_PTR(-ENOMEM); |
d1310b2e | 4894 | |
65ad0104 | 4895 | num_pages = num_extent_pages(eb); |
727011e0 | 4896 | for (i = 0; i < num_pages; i++, index++) { |
d1b5c567 | 4897 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
c871b0f2 LB |
4898 | if (!p) { |
4899 | exists = ERR_PTR(-ENOMEM); | |
6af118ce | 4900 | goto free_eb; |
c871b0f2 | 4901 | } |
4f2de97a JB |
4902 | |
4903 | spin_lock(&mapping->private_lock); | |
4904 | if (PagePrivate(p)) { | |
4905 | /* | |
4906 | * We could have already allocated an eb for this page | |
4907 | * and attached one so lets see if we can get a ref on | |
4908 | * the existing eb, and if we can we know it's good and | |
4909 | * we can just return that one, else we know we can just | |
4910 | * overwrite page->private. | |
4911 | */ | |
4912 | exists = (struct extent_buffer *)p->private; | |
4913 | if (atomic_inc_not_zero(&exists->refs)) { | |
4914 | spin_unlock(&mapping->private_lock); | |
4915 | unlock_page(p); | |
09cbfeaf | 4916 | put_page(p); |
2457aec6 | 4917 | mark_extent_buffer_accessed(exists, p); |
4f2de97a JB |
4918 | goto free_eb; |
4919 | } | |
5ca64f45 | 4920 | exists = NULL; |
4f2de97a | 4921 | |
0b32f4bb | 4922 | /* |
4f2de97a JB |
4923 | * Do this so attach doesn't complain and we need to |
4924 | * drop the ref the old guy had. | |
4925 | */ | |
4926 | ClearPagePrivate(p); | |
0b32f4bb | 4927 | WARN_ON(PageDirty(p)); |
09cbfeaf | 4928 | put_page(p); |
d1310b2e | 4929 | } |
4f2de97a JB |
4930 | attach_extent_buffer_page(eb, p); |
4931 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 4932 | WARN_ON(PageDirty(p)); |
727011e0 | 4933 | eb->pages[i] = p; |
d1310b2e CM |
4934 | if (!PageUptodate(p)) |
4935 | uptodate = 0; | |
eb14ab8e CM |
4936 | |
4937 | /* | |
b16d011e NB |
4938 | * We can't unlock the pages just yet since the extent buffer |
4939 | * hasn't been properly inserted in the radix tree, this | |
4940 | * opens a race with btree_releasepage which can free a page | |
4941 | * while we are still filling in all pages for the buffer and | |
4942 | * we could crash. | |
eb14ab8e | 4943 | */ |
d1310b2e CM |
4944 | } |
4945 | if (uptodate) | |
b4ce94de | 4946 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 4947 | again: |
e1860a77 | 4948 | ret = radix_tree_preload(GFP_NOFS); |
c871b0f2 LB |
4949 | if (ret) { |
4950 | exists = ERR_PTR(ret); | |
19fe0a8b | 4951 | goto free_eb; |
c871b0f2 | 4952 | } |
19fe0a8b | 4953 | |
f28491e0 JB |
4954 | spin_lock(&fs_info->buffer_lock); |
4955 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
09cbfeaf | 4956 | start >> PAGE_SHIFT, eb); |
f28491e0 | 4957 | spin_unlock(&fs_info->buffer_lock); |
452c75c3 | 4958 | radix_tree_preload_end(); |
19fe0a8b | 4959 | if (ret == -EEXIST) { |
f28491e0 | 4960 | exists = find_extent_buffer(fs_info, start); |
452c75c3 CS |
4961 | if (exists) |
4962 | goto free_eb; | |
4963 | else | |
115391d2 | 4964 | goto again; |
6af118ce | 4965 | } |
6af118ce | 4966 | /* add one reference for the tree */ |
0b32f4bb | 4967 | check_buffer_tree_ref(eb); |
34b41ace | 4968 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
4969 | |
4970 | /* | |
b16d011e NB |
4971 | * Now it's safe to unlock the pages because any calls to |
4972 | * btree_releasepage will correctly detect that a page belongs to a | |
4973 | * live buffer and won't free them prematurely. | |
eb14ab8e | 4974 | */ |
28187ae5 NB |
4975 | for (i = 0; i < num_pages; i++) |
4976 | unlock_page(eb->pages[i]); | |
d1310b2e CM |
4977 | return eb; |
4978 | ||
6af118ce | 4979 | free_eb: |
5ca64f45 | 4980 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
4981 | for (i = 0; i < num_pages; i++) { |
4982 | if (eb->pages[i]) | |
4983 | unlock_page(eb->pages[i]); | |
4984 | } | |
eb14ab8e | 4985 | |
897ca6e9 | 4986 | btrfs_release_extent_buffer(eb); |
6af118ce | 4987 | return exists; |
d1310b2e | 4988 | } |
d1310b2e | 4989 | |
3083ee2e JB |
4990 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
4991 | { | |
4992 | struct extent_buffer *eb = | |
4993 | container_of(head, struct extent_buffer, rcu_head); | |
4994 | ||
4995 | __free_extent_buffer(eb); | |
4996 | } | |
4997 | ||
f7a52a40 | 4998 | static int release_extent_buffer(struct extent_buffer *eb) |
3083ee2e | 4999 | { |
07e21c4d NB |
5000 | lockdep_assert_held(&eb->refs_lock); |
5001 | ||
3083ee2e JB |
5002 | WARN_ON(atomic_read(&eb->refs) == 0); |
5003 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 5004 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 5005 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 5006 | |
815a51c7 | 5007 | spin_unlock(&eb->refs_lock); |
3083ee2e | 5008 | |
f28491e0 JB |
5009 | spin_lock(&fs_info->buffer_lock); |
5010 | radix_tree_delete(&fs_info->buffer_radix, | |
09cbfeaf | 5011 | eb->start >> PAGE_SHIFT); |
f28491e0 | 5012 | spin_unlock(&fs_info->buffer_lock); |
34b41ace JB |
5013 | } else { |
5014 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 5015 | } |
3083ee2e JB |
5016 | |
5017 | /* Should be safe to release our pages at this point */ | |
55ac0139 | 5018 | btrfs_release_extent_buffer_pages(eb); |
bcb7e449 | 5019 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
b0132a3b | 5020 | if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) { |
bcb7e449 JB |
5021 | __free_extent_buffer(eb); |
5022 | return 1; | |
5023 | } | |
5024 | #endif | |
3083ee2e | 5025 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 5026 | return 1; |
3083ee2e JB |
5027 | } |
5028 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
5029 | |
5030 | return 0; | |
3083ee2e JB |
5031 | } |
5032 | ||
d1310b2e CM |
5033 | void free_extent_buffer(struct extent_buffer *eb) |
5034 | { | |
242e18c7 CM |
5035 | int refs; |
5036 | int old; | |
d1310b2e CM |
5037 | if (!eb) |
5038 | return; | |
5039 | ||
242e18c7 CM |
5040 | while (1) { |
5041 | refs = atomic_read(&eb->refs); | |
46cc775e NB |
5042 | if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3) |
5043 | || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && | |
5044 | refs == 1)) | |
242e18c7 CM |
5045 | break; |
5046 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
5047 | if (old == refs) | |
5048 | return; | |
5049 | } | |
5050 | ||
3083ee2e JB |
5051 | spin_lock(&eb->refs_lock); |
5052 | if (atomic_read(&eb->refs) == 2 && | |
5053 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 5054 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
5055 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5056 | atomic_dec(&eb->refs); | |
5057 | ||
5058 | /* | |
5059 | * I know this is terrible, but it's temporary until we stop tracking | |
5060 | * the uptodate bits and such for the extent buffers. | |
5061 | */ | |
f7a52a40 | 5062 | release_extent_buffer(eb); |
3083ee2e JB |
5063 | } |
5064 | ||
5065 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
5066 | { | |
5067 | if (!eb) | |
d1310b2e CM |
5068 | return; |
5069 | ||
3083ee2e JB |
5070 | spin_lock(&eb->refs_lock); |
5071 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
5072 | ||
0b32f4bb | 5073 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
5074 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5075 | atomic_dec(&eb->refs); | |
f7a52a40 | 5076 | release_extent_buffer(eb); |
d1310b2e | 5077 | } |
d1310b2e | 5078 | |
1d4284bd | 5079 | void clear_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5080 | { |
cc5e31a4 DS |
5081 | int i; |
5082 | int num_pages; | |
d1310b2e CM |
5083 | struct page *page; |
5084 | ||
65ad0104 | 5085 | num_pages = num_extent_pages(eb); |
d1310b2e CM |
5086 | |
5087 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 5088 | page = eb->pages[i]; |
b9473439 | 5089 | if (!PageDirty(page)) |
d2c3f4f6 CM |
5090 | continue; |
5091 | ||
a61e6f29 | 5092 | lock_page(page); |
eb14ab8e CM |
5093 | WARN_ON(!PagePrivate(page)); |
5094 | ||
d1310b2e | 5095 | clear_page_dirty_for_io(page); |
b93b0163 | 5096 | xa_lock_irq(&page->mapping->i_pages); |
0a943c65 MW |
5097 | if (!PageDirty(page)) |
5098 | __xa_clear_mark(&page->mapping->i_pages, | |
5099 | page_index(page), PAGECACHE_TAG_DIRTY); | |
b93b0163 | 5100 | xa_unlock_irq(&page->mapping->i_pages); |
bf0da8c1 | 5101 | ClearPageError(page); |
a61e6f29 | 5102 | unlock_page(page); |
d1310b2e | 5103 | } |
0b32f4bb | 5104 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 5105 | } |
d1310b2e | 5106 | |
abb57ef3 | 5107 | bool set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5108 | { |
cc5e31a4 DS |
5109 | int i; |
5110 | int num_pages; | |
abb57ef3 | 5111 | bool was_dirty; |
d1310b2e | 5112 | |
0b32f4bb JB |
5113 | check_buffer_tree_ref(eb); |
5114 | ||
b9473439 | 5115 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 5116 | |
65ad0104 | 5117 | num_pages = num_extent_pages(eb); |
3083ee2e | 5118 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
5119 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
5120 | ||
abb57ef3 LB |
5121 | if (!was_dirty) |
5122 | for (i = 0; i < num_pages; i++) | |
5123 | set_page_dirty(eb->pages[i]); | |
51995c39 LB |
5124 | |
5125 | #ifdef CONFIG_BTRFS_DEBUG | |
5126 | for (i = 0; i < num_pages; i++) | |
5127 | ASSERT(PageDirty(eb->pages[i])); | |
5128 | #endif | |
5129 | ||
b9473439 | 5130 | return was_dirty; |
d1310b2e | 5131 | } |
d1310b2e | 5132 | |
69ba3927 | 5133 | void clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 | 5134 | { |
cc5e31a4 | 5135 | int i; |
1259ab75 | 5136 | struct page *page; |
cc5e31a4 | 5137 | int num_pages; |
1259ab75 | 5138 | |
b4ce94de | 5139 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 5140 | num_pages = num_extent_pages(eb); |
1259ab75 | 5141 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5142 | page = eb->pages[i]; |
33958dc6 CM |
5143 | if (page) |
5144 | ClearPageUptodate(page); | |
1259ab75 | 5145 | } |
1259ab75 CM |
5146 | } |
5147 | ||
09c25a8c | 5148 | void set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 5149 | { |
cc5e31a4 | 5150 | int i; |
d1310b2e | 5151 | struct page *page; |
cc5e31a4 | 5152 | int num_pages; |
d1310b2e | 5153 | |
0b32f4bb | 5154 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 5155 | num_pages = num_extent_pages(eb); |
d1310b2e | 5156 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5157 | page = eb->pages[i]; |
d1310b2e CM |
5158 | SetPageUptodate(page); |
5159 | } | |
d1310b2e | 5160 | } |
d1310b2e | 5161 | |
d1310b2e | 5162 | int read_extent_buffer_pages(struct extent_io_tree *tree, |
6af49dbd | 5163 | struct extent_buffer *eb, int wait, int mirror_num) |
d1310b2e | 5164 | { |
cc5e31a4 | 5165 | int i; |
d1310b2e CM |
5166 | struct page *page; |
5167 | int err; | |
5168 | int ret = 0; | |
ce9adaa5 CM |
5169 | int locked_pages = 0; |
5170 | int all_uptodate = 1; | |
cc5e31a4 | 5171 | int num_pages; |
727011e0 | 5172 | unsigned long num_reads = 0; |
a86c12c7 | 5173 | struct bio *bio = NULL; |
c8b97818 | 5174 | unsigned long bio_flags = 0; |
a86c12c7 | 5175 | |
b4ce94de | 5176 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
5177 | return 0; |
5178 | ||
65ad0104 | 5179 | num_pages = num_extent_pages(eb); |
8436ea91 | 5180 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5181 | page = eb->pages[i]; |
bb82ab88 | 5182 | if (wait == WAIT_NONE) { |
2db04966 | 5183 | if (!trylock_page(page)) |
ce9adaa5 | 5184 | goto unlock_exit; |
d1310b2e CM |
5185 | } else { |
5186 | lock_page(page); | |
5187 | } | |
ce9adaa5 | 5188 | locked_pages++; |
2571e739 LB |
5189 | } |
5190 | /* | |
5191 | * We need to firstly lock all pages to make sure that | |
5192 | * the uptodate bit of our pages won't be affected by | |
5193 | * clear_extent_buffer_uptodate(). | |
5194 | */ | |
8436ea91 | 5195 | for (i = 0; i < num_pages; i++) { |
2571e739 | 5196 | page = eb->pages[i]; |
727011e0 CM |
5197 | if (!PageUptodate(page)) { |
5198 | num_reads++; | |
ce9adaa5 | 5199 | all_uptodate = 0; |
727011e0 | 5200 | } |
ce9adaa5 | 5201 | } |
2571e739 | 5202 | |
ce9adaa5 | 5203 | if (all_uptodate) { |
8436ea91 | 5204 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
5205 | goto unlock_exit; |
5206 | } | |
5207 | ||
656f30db | 5208 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 5209 | eb->read_mirror = 0; |
0b32f4bb | 5210 | atomic_set(&eb->io_pages, num_reads); |
8436ea91 | 5211 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5212 | page = eb->pages[i]; |
baf863b9 | 5213 | |
ce9adaa5 | 5214 | if (!PageUptodate(page)) { |
baf863b9 LB |
5215 | if (ret) { |
5216 | atomic_dec(&eb->io_pages); | |
5217 | unlock_page(page); | |
5218 | continue; | |
5219 | } | |
5220 | ||
f188591e | 5221 | ClearPageError(page); |
a86c12c7 | 5222 | err = __extent_read_full_page(tree, page, |
6af49dbd | 5223 | btree_get_extent, &bio, |
d4c7ca86 | 5224 | mirror_num, &bio_flags, |
1f7ad75b | 5225 | REQ_META); |
baf863b9 | 5226 | if (err) { |
d1310b2e | 5227 | ret = err; |
baf863b9 LB |
5228 | /* |
5229 | * We use &bio in above __extent_read_full_page, | |
5230 | * so we ensure that if it returns error, the | |
5231 | * current page fails to add itself to bio and | |
5232 | * it's been unlocked. | |
5233 | * | |
5234 | * We must dec io_pages by ourselves. | |
5235 | */ | |
5236 | atomic_dec(&eb->io_pages); | |
5237 | } | |
d1310b2e CM |
5238 | } else { |
5239 | unlock_page(page); | |
5240 | } | |
5241 | } | |
5242 | ||
355808c2 | 5243 | if (bio) { |
1f7ad75b | 5244 | err = submit_one_bio(bio, mirror_num, bio_flags); |
79787eaa JM |
5245 | if (err) |
5246 | return err; | |
355808c2 | 5247 | } |
a86c12c7 | 5248 | |
bb82ab88 | 5249 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 5250 | return ret; |
d397712b | 5251 | |
8436ea91 | 5252 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5253 | page = eb->pages[i]; |
d1310b2e | 5254 | wait_on_page_locked(page); |
d397712b | 5255 | if (!PageUptodate(page)) |
d1310b2e | 5256 | ret = -EIO; |
d1310b2e | 5257 | } |
d397712b | 5258 | |
d1310b2e | 5259 | return ret; |
ce9adaa5 CM |
5260 | |
5261 | unlock_exit: | |
d397712b | 5262 | while (locked_pages > 0) { |
ce9adaa5 | 5263 | locked_pages--; |
8436ea91 JB |
5264 | page = eb->pages[locked_pages]; |
5265 | unlock_page(page); | |
ce9adaa5 CM |
5266 | } |
5267 | return ret; | |
d1310b2e | 5268 | } |
d1310b2e | 5269 | |
1cbb1f45 JM |
5270 | void read_extent_buffer(const struct extent_buffer *eb, void *dstv, |
5271 | unsigned long start, unsigned long len) | |
d1310b2e CM |
5272 | { |
5273 | size_t cur; | |
5274 | size_t offset; | |
5275 | struct page *page; | |
5276 | char *kaddr; | |
5277 | char *dst = (char *)dstv; | |
7073017a | 5278 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5279 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
d1310b2e | 5280 | |
f716abd5 LB |
5281 | if (start + len > eb->len) { |
5282 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n", | |
5283 | eb->start, eb->len, start, len); | |
5284 | memset(dst, 0, len); | |
5285 | return; | |
5286 | } | |
d1310b2e | 5287 | |
7073017a | 5288 | offset = offset_in_page(start_offset + start); |
d1310b2e | 5289 | |
d397712b | 5290 | while (len > 0) { |
fb85fc9a | 5291 | page = eb->pages[i]; |
d1310b2e | 5292 | |
09cbfeaf | 5293 | cur = min(len, (PAGE_SIZE - offset)); |
a6591715 | 5294 | kaddr = page_address(page); |
d1310b2e | 5295 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
5296 | |
5297 | dst += cur; | |
5298 | len -= cur; | |
5299 | offset = 0; | |
5300 | i++; | |
5301 | } | |
5302 | } | |
d1310b2e | 5303 | |
1cbb1f45 JM |
5304 | int read_extent_buffer_to_user(const struct extent_buffer *eb, |
5305 | void __user *dstv, | |
5306 | unsigned long start, unsigned long len) | |
550ac1d8 GH |
5307 | { |
5308 | size_t cur; | |
5309 | size_t offset; | |
5310 | struct page *page; | |
5311 | char *kaddr; | |
5312 | char __user *dst = (char __user *)dstv; | |
7073017a | 5313 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5314 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
550ac1d8 GH |
5315 | int ret = 0; |
5316 | ||
5317 | WARN_ON(start > eb->len); | |
5318 | WARN_ON(start + len > eb->start + eb->len); | |
5319 | ||
7073017a | 5320 | offset = offset_in_page(start_offset + start); |
550ac1d8 GH |
5321 | |
5322 | while (len > 0) { | |
fb85fc9a | 5323 | page = eb->pages[i]; |
550ac1d8 | 5324 | |
09cbfeaf | 5325 | cur = min(len, (PAGE_SIZE - offset)); |
550ac1d8 GH |
5326 | kaddr = page_address(page); |
5327 | if (copy_to_user(dst, kaddr + offset, cur)) { | |
5328 | ret = -EFAULT; | |
5329 | break; | |
5330 | } | |
5331 | ||
5332 | dst += cur; | |
5333 | len -= cur; | |
5334 | offset = 0; | |
5335 | i++; | |
5336 | } | |
5337 | ||
5338 | return ret; | |
5339 | } | |
5340 | ||
415b35a5 LB |
5341 | /* |
5342 | * return 0 if the item is found within a page. | |
5343 | * return 1 if the item spans two pages. | |
5344 | * return -EINVAL otherwise. | |
5345 | */ | |
1cbb1f45 JM |
5346 | int map_private_extent_buffer(const struct extent_buffer *eb, |
5347 | unsigned long start, unsigned long min_len, | |
5348 | char **map, unsigned long *map_start, | |
5349 | unsigned long *map_len) | |
d1310b2e | 5350 | { |
cc2c39d6 | 5351 | size_t offset; |
d1310b2e CM |
5352 | char *kaddr; |
5353 | struct page *p; | |
7073017a | 5354 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5355 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
d1310b2e | 5356 | unsigned long end_i = (start_offset + start + min_len - 1) >> |
09cbfeaf | 5357 | PAGE_SHIFT; |
d1310b2e | 5358 | |
f716abd5 LB |
5359 | if (start + min_len > eb->len) { |
5360 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n", | |
5361 | eb->start, eb->len, start, min_len); | |
5362 | return -EINVAL; | |
5363 | } | |
5364 | ||
d1310b2e | 5365 | if (i != end_i) |
415b35a5 | 5366 | return 1; |
d1310b2e CM |
5367 | |
5368 | if (i == 0) { | |
5369 | offset = start_offset; | |
5370 | *map_start = 0; | |
5371 | } else { | |
5372 | offset = 0; | |
09cbfeaf | 5373 | *map_start = ((u64)i << PAGE_SHIFT) - start_offset; |
d1310b2e | 5374 | } |
d397712b | 5375 | |
fb85fc9a | 5376 | p = eb->pages[i]; |
a6591715 | 5377 | kaddr = page_address(p); |
d1310b2e | 5378 | *map = kaddr + offset; |
09cbfeaf | 5379 | *map_len = PAGE_SIZE - offset; |
d1310b2e CM |
5380 | return 0; |
5381 | } | |
d1310b2e | 5382 | |
1cbb1f45 JM |
5383 | int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv, |
5384 | unsigned long start, unsigned long len) | |
d1310b2e CM |
5385 | { |
5386 | size_t cur; | |
5387 | size_t offset; | |
5388 | struct page *page; | |
5389 | char *kaddr; | |
5390 | char *ptr = (char *)ptrv; | |
7073017a | 5391 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5392 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
d1310b2e CM |
5393 | int ret = 0; |
5394 | ||
5395 | WARN_ON(start > eb->len); | |
5396 | WARN_ON(start + len > eb->start + eb->len); | |
5397 | ||
7073017a | 5398 | offset = offset_in_page(start_offset + start); |
d1310b2e | 5399 | |
d397712b | 5400 | while (len > 0) { |
fb85fc9a | 5401 | page = eb->pages[i]; |
d1310b2e | 5402 | |
09cbfeaf | 5403 | cur = min(len, (PAGE_SIZE - offset)); |
d1310b2e | 5404 | |
a6591715 | 5405 | kaddr = page_address(page); |
d1310b2e | 5406 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
5407 | if (ret) |
5408 | break; | |
5409 | ||
5410 | ptr += cur; | |
5411 | len -= cur; | |
5412 | offset = 0; | |
5413 | i++; | |
5414 | } | |
5415 | return ret; | |
5416 | } | |
d1310b2e | 5417 | |
f157bf76 DS |
5418 | void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb, |
5419 | const void *srcv) | |
5420 | { | |
5421 | char *kaddr; | |
5422 | ||
5423 | WARN_ON(!PageUptodate(eb->pages[0])); | |
5424 | kaddr = page_address(eb->pages[0]); | |
5425 | memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv, | |
5426 | BTRFS_FSID_SIZE); | |
5427 | } | |
5428 | ||
5429 | void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv) | |
5430 | { | |
5431 | char *kaddr; | |
5432 | ||
5433 | WARN_ON(!PageUptodate(eb->pages[0])); | |
5434 | kaddr = page_address(eb->pages[0]); | |
5435 | memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv, | |
5436 | BTRFS_FSID_SIZE); | |
5437 | } | |
5438 | ||
d1310b2e CM |
5439 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, |
5440 | unsigned long start, unsigned long len) | |
5441 | { | |
5442 | size_t cur; | |
5443 | size_t offset; | |
5444 | struct page *page; | |
5445 | char *kaddr; | |
5446 | char *src = (char *)srcv; | |
7073017a | 5447 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5448 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
d1310b2e CM |
5449 | |
5450 | WARN_ON(start > eb->len); | |
5451 | WARN_ON(start + len > eb->start + eb->len); | |
5452 | ||
7073017a | 5453 | offset = offset_in_page(start_offset + start); |
d1310b2e | 5454 | |
d397712b | 5455 | while (len > 0) { |
fb85fc9a | 5456 | page = eb->pages[i]; |
d1310b2e CM |
5457 | WARN_ON(!PageUptodate(page)); |
5458 | ||
09cbfeaf | 5459 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 5460 | kaddr = page_address(page); |
d1310b2e | 5461 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
5462 | |
5463 | src += cur; | |
5464 | len -= cur; | |
5465 | offset = 0; | |
5466 | i++; | |
5467 | } | |
5468 | } | |
d1310b2e | 5469 | |
b159fa28 DS |
5470 | void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start, |
5471 | unsigned long len) | |
d1310b2e CM |
5472 | { |
5473 | size_t cur; | |
5474 | size_t offset; | |
5475 | struct page *page; | |
5476 | char *kaddr; | |
7073017a | 5477 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5478 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
d1310b2e CM |
5479 | |
5480 | WARN_ON(start > eb->len); | |
5481 | WARN_ON(start + len > eb->start + eb->len); | |
5482 | ||
7073017a | 5483 | offset = offset_in_page(start_offset + start); |
d1310b2e | 5484 | |
d397712b | 5485 | while (len > 0) { |
fb85fc9a | 5486 | page = eb->pages[i]; |
d1310b2e CM |
5487 | WARN_ON(!PageUptodate(page)); |
5488 | ||
09cbfeaf | 5489 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 5490 | kaddr = page_address(page); |
b159fa28 | 5491 | memset(kaddr + offset, 0, cur); |
d1310b2e CM |
5492 | |
5493 | len -= cur; | |
5494 | offset = 0; | |
5495 | i++; | |
5496 | } | |
5497 | } | |
d1310b2e | 5498 | |
58e8012c DS |
5499 | void copy_extent_buffer_full(struct extent_buffer *dst, |
5500 | struct extent_buffer *src) | |
5501 | { | |
5502 | int i; | |
cc5e31a4 | 5503 | int num_pages; |
58e8012c DS |
5504 | |
5505 | ASSERT(dst->len == src->len); | |
5506 | ||
65ad0104 | 5507 | num_pages = num_extent_pages(dst); |
58e8012c DS |
5508 | for (i = 0; i < num_pages; i++) |
5509 | copy_page(page_address(dst->pages[i]), | |
5510 | page_address(src->pages[i])); | |
5511 | } | |
5512 | ||
d1310b2e CM |
5513 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, |
5514 | unsigned long dst_offset, unsigned long src_offset, | |
5515 | unsigned long len) | |
5516 | { | |
5517 | u64 dst_len = dst->len; | |
5518 | size_t cur; | |
5519 | size_t offset; | |
5520 | struct page *page; | |
5521 | char *kaddr; | |
7073017a | 5522 | size_t start_offset = offset_in_page(dst->start); |
09cbfeaf | 5523 | unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT; |
d1310b2e CM |
5524 | |
5525 | WARN_ON(src->len != dst_len); | |
5526 | ||
7073017a | 5527 | offset = offset_in_page(start_offset + dst_offset); |
d1310b2e | 5528 | |
d397712b | 5529 | while (len > 0) { |
fb85fc9a | 5530 | page = dst->pages[i]; |
d1310b2e CM |
5531 | WARN_ON(!PageUptodate(page)); |
5532 | ||
09cbfeaf | 5533 | cur = min(len, (unsigned long)(PAGE_SIZE - offset)); |
d1310b2e | 5534 | |
a6591715 | 5535 | kaddr = page_address(page); |
d1310b2e | 5536 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
5537 | |
5538 | src_offset += cur; | |
5539 | len -= cur; | |
5540 | offset = 0; | |
5541 | i++; | |
5542 | } | |
5543 | } | |
d1310b2e | 5544 | |
3e1e8bb7 OS |
5545 | /* |
5546 | * eb_bitmap_offset() - calculate the page and offset of the byte containing the | |
5547 | * given bit number | |
5548 | * @eb: the extent buffer | |
5549 | * @start: offset of the bitmap item in the extent buffer | |
5550 | * @nr: bit number | |
5551 | * @page_index: return index of the page in the extent buffer that contains the | |
5552 | * given bit number | |
5553 | * @page_offset: return offset into the page given by page_index | |
5554 | * | |
5555 | * This helper hides the ugliness of finding the byte in an extent buffer which | |
5556 | * contains a given bit. | |
5557 | */ | |
5558 | static inline void eb_bitmap_offset(struct extent_buffer *eb, | |
5559 | unsigned long start, unsigned long nr, | |
5560 | unsigned long *page_index, | |
5561 | size_t *page_offset) | |
5562 | { | |
7073017a | 5563 | size_t start_offset = offset_in_page(eb->start); |
3e1e8bb7 OS |
5564 | size_t byte_offset = BIT_BYTE(nr); |
5565 | size_t offset; | |
5566 | ||
5567 | /* | |
5568 | * The byte we want is the offset of the extent buffer + the offset of | |
5569 | * the bitmap item in the extent buffer + the offset of the byte in the | |
5570 | * bitmap item. | |
5571 | */ | |
5572 | offset = start_offset + start + byte_offset; | |
5573 | ||
09cbfeaf | 5574 | *page_index = offset >> PAGE_SHIFT; |
7073017a | 5575 | *page_offset = offset_in_page(offset); |
3e1e8bb7 OS |
5576 | } |
5577 | ||
5578 | /** | |
5579 | * extent_buffer_test_bit - determine whether a bit in a bitmap item is set | |
5580 | * @eb: the extent buffer | |
5581 | * @start: offset of the bitmap item in the extent buffer | |
5582 | * @nr: bit number to test | |
5583 | */ | |
5584 | int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start, | |
5585 | unsigned long nr) | |
5586 | { | |
2fe1d551 | 5587 | u8 *kaddr; |
3e1e8bb7 OS |
5588 | struct page *page; |
5589 | unsigned long i; | |
5590 | size_t offset; | |
5591 | ||
5592 | eb_bitmap_offset(eb, start, nr, &i, &offset); | |
5593 | page = eb->pages[i]; | |
5594 | WARN_ON(!PageUptodate(page)); | |
5595 | kaddr = page_address(page); | |
5596 | return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1))); | |
5597 | } | |
5598 | ||
5599 | /** | |
5600 | * extent_buffer_bitmap_set - set an area of a bitmap | |
5601 | * @eb: the extent buffer | |
5602 | * @start: offset of the bitmap item in the extent buffer | |
5603 | * @pos: bit number of the first bit | |
5604 | * @len: number of bits to set | |
5605 | */ | |
5606 | void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start, | |
5607 | unsigned long pos, unsigned long len) | |
5608 | { | |
2fe1d551 | 5609 | u8 *kaddr; |
3e1e8bb7 OS |
5610 | struct page *page; |
5611 | unsigned long i; | |
5612 | size_t offset; | |
5613 | const unsigned int size = pos + len; | |
5614 | int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 5615 | u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
5616 | |
5617 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5618 | page = eb->pages[i]; | |
5619 | WARN_ON(!PageUptodate(page)); | |
5620 | kaddr = page_address(page); | |
5621 | ||
5622 | while (len >= bits_to_set) { | |
5623 | kaddr[offset] |= mask_to_set; | |
5624 | len -= bits_to_set; | |
5625 | bits_to_set = BITS_PER_BYTE; | |
9c894696 | 5626 | mask_to_set = ~0; |
09cbfeaf | 5627 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5628 | offset = 0; |
5629 | page = eb->pages[++i]; | |
5630 | WARN_ON(!PageUptodate(page)); | |
5631 | kaddr = page_address(page); | |
5632 | } | |
5633 | } | |
5634 | if (len) { | |
5635 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
5636 | kaddr[offset] |= mask_to_set; | |
5637 | } | |
5638 | } | |
5639 | ||
5640 | ||
5641 | /** | |
5642 | * extent_buffer_bitmap_clear - clear an area of a bitmap | |
5643 | * @eb: the extent buffer | |
5644 | * @start: offset of the bitmap item in the extent buffer | |
5645 | * @pos: bit number of the first bit | |
5646 | * @len: number of bits to clear | |
5647 | */ | |
5648 | void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start, | |
5649 | unsigned long pos, unsigned long len) | |
5650 | { | |
2fe1d551 | 5651 | u8 *kaddr; |
3e1e8bb7 OS |
5652 | struct page *page; |
5653 | unsigned long i; | |
5654 | size_t offset; | |
5655 | const unsigned int size = pos + len; | |
5656 | int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 5657 | u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
5658 | |
5659 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5660 | page = eb->pages[i]; | |
5661 | WARN_ON(!PageUptodate(page)); | |
5662 | kaddr = page_address(page); | |
5663 | ||
5664 | while (len >= bits_to_clear) { | |
5665 | kaddr[offset] &= ~mask_to_clear; | |
5666 | len -= bits_to_clear; | |
5667 | bits_to_clear = BITS_PER_BYTE; | |
9c894696 | 5668 | mask_to_clear = ~0; |
09cbfeaf | 5669 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5670 | offset = 0; |
5671 | page = eb->pages[++i]; | |
5672 | WARN_ON(!PageUptodate(page)); | |
5673 | kaddr = page_address(page); | |
5674 | } | |
5675 | } | |
5676 | if (len) { | |
5677 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
5678 | kaddr[offset] &= ~mask_to_clear; | |
5679 | } | |
5680 | } | |
5681 | ||
3387206f ST |
5682 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
5683 | { | |
5684 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
5685 | return distance < len; | |
5686 | } | |
5687 | ||
d1310b2e CM |
5688 | static void copy_pages(struct page *dst_page, struct page *src_page, |
5689 | unsigned long dst_off, unsigned long src_off, | |
5690 | unsigned long len) | |
5691 | { | |
a6591715 | 5692 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 5693 | char *src_kaddr; |
727011e0 | 5694 | int must_memmove = 0; |
d1310b2e | 5695 | |
3387206f | 5696 | if (dst_page != src_page) { |
a6591715 | 5697 | src_kaddr = page_address(src_page); |
3387206f | 5698 | } else { |
d1310b2e | 5699 | src_kaddr = dst_kaddr; |
727011e0 CM |
5700 | if (areas_overlap(src_off, dst_off, len)) |
5701 | must_memmove = 1; | |
3387206f | 5702 | } |
d1310b2e | 5703 | |
727011e0 CM |
5704 | if (must_memmove) |
5705 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
5706 | else | |
5707 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
5708 | } |
5709 | ||
5710 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5711 | unsigned long src_offset, unsigned long len) | |
5712 | { | |
0b246afa | 5713 | struct btrfs_fs_info *fs_info = dst->fs_info; |
d1310b2e CM |
5714 | size_t cur; |
5715 | size_t dst_off_in_page; | |
5716 | size_t src_off_in_page; | |
7073017a | 5717 | size_t start_offset = offset_in_page(dst->start); |
d1310b2e CM |
5718 | unsigned long dst_i; |
5719 | unsigned long src_i; | |
5720 | ||
5721 | if (src_offset + len > dst->len) { | |
0b246afa | 5722 | btrfs_err(fs_info, |
5d163e0e JM |
5723 | "memmove bogus src_offset %lu move len %lu dst len %lu", |
5724 | src_offset, len, dst->len); | |
d1310b2e CM |
5725 | BUG_ON(1); |
5726 | } | |
5727 | if (dst_offset + len > dst->len) { | |
0b246afa | 5728 | btrfs_err(fs_info, |
5d163e0e JM |
5729 | "memmove bogus dst_offset %lu move len %lu dst len %lu", |
5730 | dst_offset, len, dst->len); | |
d1310b2e CM |
5731 | BUG_ON(1); |
5732 | } | |
5733 | ||
d397712b | 5734 | while (len > 0) { |
7073017a JT |
5735 | dst_off_in_page = offset_in_page(start_offset + dst_offset); |
5736 | src_off_in_page = offset_in_page(start_offset + src_offset); | |
d1310b2e | 5737 | |
09cbfeaf KS |
5738 | dst_i = (start_offset + dst_offset) >> PAGE_SHIFT; |
5739 | src_i = (start_offset + src_offset) >> PAGE_SHIFT; | |
d1310b2e | 5740 | |
09cbfeaf | 5741 | cur = min(len, (unsigned long)(PAGE_SIZE - |
d1310b2e CM |
5742 | src_off_in_page)); |
5743 | cur = min_t(unsigned long, cur, | |
09cbfeaf | 5744 | (unsigned long)(PAGE_SIZE - dst_off_in_page)); |
d1310b2e | 5745 | |
fb85fc9a | 5746 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5747 | dst_off_in_page, src_off_in_page, cur); |
5748 | ||
5749 | src_offset += cur; | |
5750 | dst_offset += cur; | |
5751 | len -= cur; | |
5752 | } | |
5753 | } | |
d1310b2e CM |
5754 | |
5755 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5756 | unsigned long src_offset, unsigned long len) | |
5757 | { | |
0b246afa | 5758 | struct btrfs_fs_info *fs_info = dst->fs_info; |
d1310b2e CM |
5759 | size_t cur; |
5760 | size_t dst_off_in_page; | |
5761 | size_t src_off_in_page; | |
5762 | unsigned long dst_end = dst_offset + len - 1; | |
5763 | unsigned long src_end = src_offset + len - 1; | |
7073017a | 5764 | size_t start_offset = offset_in_page(dst->start); |
d1310b2e CM |
5765 | unsigned long dst_i; |
5766 | unsigned long src_i; | |
5767 | ||
5768 | if (src_offset + len > dst->len) { | |
0b246afa | 5769 | btrfs_err(fs_info, |
5d163e0e JM |
5770 | "memmove bogus src_offset %lu move len %lu len %lu", |
5771 | src_offset, len, dst->len); | |
d1310b2e CM |
5772 | BUG_ON(1); |
5773 | } | |
5774 | if (dst_offset + len > dst->len) { | |
0b246afa | 5775 | btrfs_err(fs_info, |
5d163e0e JM |
5776 | "memmove bogus dst_offset %lu move len %lu len %lu", |
5777 | dst_offset, len, dst->len); | |
d1310b2e CM |
5778 | BUG_ON(1); |
5779 | } | |
727011e0 | 5780 | if (dst_offset < src_offset) { |
d1310b2e CM |
5781 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
5782 | return; | |
5783 | } | |
d397712b | 5784 | while (len > 0) { |
09cbfeaf KS |
5785 | dst_i = (start_offset + dst_end) >> PAGE_SHIFT; |
5786 | src_i = (start_offset + src_end) >> PAGE_SHIFT; | |
d1310b2e | 5787 | |
7073017a JT |
5788 | dst_off_in_page = offset_in_page(start_offset + dst_end); |
5789 | src_off_in_page = offset_in_page(start_offset + src_end); | |
d1310b2e CM |
5790 | |
5791 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
5792 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 5793 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5794 | dst_off_in_page - cur + 1, |
5795 | src_off_in_page - cur + 1, cur); | |
5796 | ||
5797 | dst_end -= cur; | |
5798 | src_end -= cur; | |
5799 | len -= cur; | |
5800 | } | |
5801 | } | |
6af118ce | 5802 | |
f7a52a40 | 5803 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 5804 | { |
6af118ce | 5805 | struct extent_buffer *eb; |
6af118ce | 5806 | |
3083ee2e | 5807 | /* |
01327610 | 5808 | * We need to make sure nobody is attaching this page to an eb right |
3083ee2e JB |
5809 | * now. |
5810 | */ | |
5811 | spin_lock(&page->mapping->private_lock); | |
5812 | if (!PagePrivate(page)) { | |
5813 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 5814 | return 1; |
45f49bce | 5815 | } |
6af118ce | 5816 | |
3083ee2e JB |
5817 | eb = (struct extent_buffer *)page->private; |
5818 | BUG_ON(!eb); | |
19fe0a8b MX |
5819 | |
5820 | /* | |
3083ee2e JB |
5821 | * This is a little awful but should be ok, we need to make sure that |
5822 | * the eb doesn't disappear out from under us while we're looking at | |
5823 | * this page. | |
19fe0a8b | 5824 | */ |
3083ee2e | 5825 | spin_lock(&eb->refs_lock); |
0b32f4bb | 5826 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
5827 | spin_unlock(&eb->refs_lock); |
5828 | spin_unlock(&page->mapping->private_lock); | |
5829 | return 0; | |
b9473439 | 5830 | } |
3083ee2e | 5831 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 5832 | |
19fe0a8b | 5833 | /* |
3083ee2e JB |
5834 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
5835 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 5836 | */ |
3083ee2e JB |
5837 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
5838 | spin_unlock(&eb->refs_lock); | |
5839 | return 0; | |
b9473439 | 5840 | } |
19fe0a8b | 5841 | |
f7a52a40 | 5842 | return release_extent_buffer(eb); |
6af118ce | 5843 | } |