]>
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 | 16 | #include "extent_io.h" |
9c7d3a54 | 17 | #include "extent-io-tree.h" |
d1310b2e | 18 | #include "extent_map.h" |
902b22f3 DW |
19 | #include "ctree.h" |
20 | #include "btrfs_inode.h" | |
4a54c8c1 | 21 | #include "volumes.h" |
21adbd5c | 22 | #include "check-integrity.h" |
0b32f4bb | 23 | #include "locking.h" |
606686ee | 24 | #include "rcu-string.h" |
fe09e16c | 25 | #include "backref.h" |
6af49dbd | 26 | #include "disk-io.h" |
d1310b2e | 27 | |
d1310b2e CM |
28 | static struct kmem_cache *extent_state_cache; |
29 | static struct kmem_cache *extent_buffer_cache; | |
8ac9f7c1 | 30 | static struct bio_set btrfs_bioset; |
d1310b2e | 31 | |
27a3507d FM |
32 | static inline bool extent_state_in_tree(const struct extent_state *state) |
33 | { | |
34 | return !RB_EMPTY_NODE(&state->rb_node); | |
35 | } | |
36 | ||
6d49ba1b | 37 | #ifdef CONFIG_BTRFS_DEBUG |
d1310b2e | 38 | static LIST_HEAD(states); |
d397712b | 39 | static DEFINE_SPINLOCK(leak_lock); |
6d49ba1b | 40 | |
3fd63727 JB |
41 | static inline void btrfs_leak_debug_add(spinlock_t *lock, |
42 | struct list_head *new, | |
43 | struct list_head *head) | |
6d49ba1b ES |
44 | { |
45 | unsigned long flags; | |
46 | ||
3fd63727 | 47 | spin_lock_irqsave(lock, flags); |
6d49ba1b | 48 | list_add(new, head); |
3fd63727 | 49 | spin_unlock_irqrestore(lock, flags); |
6d49ba1b ES |
50 | } |
51 | ||
3fd63727 JB |
52 | static inline void btrfs_leak_debug_del(spinlock_t *lock, |
53 | struct list_head *entry) | |
6d49ba1b ES |
54 | { |
55 | unsigned long flags; | |
56 | ||
3fd63727 | 57 | spin_lock_irqsave(lock, flags); |
6d49ba1b | 58 | list_del(entry); |
3fd63727 | 59 | spin_unlock_irqrestore(lock, flags); |
6d49ba1b ES |
60 | } |
61 | ||
3fd63727 | 62 | void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info) |
6d49ba1b | 63 | { |
6d49ba1b | 64 | struct extent_buffer *eb; |
3fd63727 | 65 | unsigned long flags; |
6d49ba1b | 66 | |
8c38938c JB |
67 | /* |
68 | * If we didn't get into open_ctree our allocated_ebs will not be | |
69 | * initialized, so just skip this. | |
70 | */ | |
71 | if (!fs_info->allocated_ebs.next) | |
72 | return; | |
73 | ||
3fd63727 JB |
74 | spin_lock_irqsave(&fs_info->eb_leak_lock, flags); |
75 | while (!list_empty(&fs_info->allocated_ebs)) { | |
76 | eb = list_first_entry(&fs_info->allocated_ebs, | |
77 | struct extent_buffer, leak_list); | |
8c38938c JB |
78 | pr_err( |
79 | "BTRFS: buffer leak start %llu len %lu refs %d bflags %lu owner %llu\n", | |
80 | eb->start, eb->len, atomic_read(&eb->refs), eb->bflags, | |
81 | btrfs_header_owner(eb)); | |
33ca832f JB |
82 | list_del(&eb->leak_list); |
83 | kmem_cache_free(extent_buffer_cache, eb); | |
84 | } | |
3fd63727 | 85 | spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags); |
33ca832f JB |
86 | } |
87 | ||
88 | static inline void btrfs_extent_state_leak_debug_check(void) | |
89 | { | |
90 | struct extent_state *state; | |
91 | ||
6d49ba1b ES |
92 | while (!list_empty(&states)) { |
93 | state = list_entry(states.next, struct extent_state, leak_list); | |
9ee49a04 | 94 | pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n", |
27a3507d FM |
95 | state->start, state->end, state->state, |
96 | extent_state_in_tree(state), | |
b7ac31b7 | 97 | refcount_read(&state->refs)); |
6d49ba1b ES |
98 | list_del(&state->leak_list); |
99 | kmem_cache_free(extent_state_cache, state); | |
100 | } | |
6d49ba1b | 101 | } |
8d599ae1 | 102 | |
a5dee37d JB |
103 | #define btrfs_debug_check_extent_io_range(tree, start, end) \ |
104 | __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end)) | |
8d599ae1 | 105 | static inline void __btrfs_debug_check_extent_io_range(const char *caller, |
a5dee37d | 106 | struct extent_io_tree *tree, u64 start, u64 end) |
8d599ae1 | 107 | { |
65a680f6 NB |
108 | struct inode *inode = tree->private_data; |
109 | u64 isize; | |
110 | ||
111 | if (!inode || !is_data_inode(inode)) | |
112 | return; | |
113 | ||
114 | isize = i_size_read(inode); | |
115 | if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) { | |
116 | btrfs_debug_rl(BTRFS_I(inode)->root->fs_info, | |
117 | "%s: ino %llu isize %llu odd range [%llu,%llu]", | |
118 | caller, btrfs_ino(BTRFS_I(inode)), isize, start, end); | |
119 | } | |
8d599ae1 | 120 | } |
6d49ba1b | 121 | #else |
3fd63727 JB |
122 | #define btrfs_leak_debug_add(lock, new, head) do {} while (0) |
123 | #define btrfs_leak_debug_del(lock, entry) do {} while (0) | |
33ca832f | 124 | #define btrfs_extent_state_leak_debug_check() do {} while (0) |
8d599ae1 | 125 | #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0) |
4bef0848 | 126 | #endif |
d1310b2e | 127 | |
d1310b2e CM |
128 | struct tree_entry { |
129 | u64 start; | |
130 | u64 end; | |
d1310b2e CM |
131 | struct rb_node rb_node; |
132 | }; | |
133 | ||
134 | struct extent_page_data { | |
135 | struct bio *bio; | |
771ed689 CM |
136 | /* tells writepage not to lock the state bits for this range |
137 | * it still does the unlocking | |
138 | */ | |
ffbd517d CM |
139 | unsigned int extent_locked:1; |
140 | ||
70fd7614 | 141 | /* tells the submit_bio code to use REQ_SYNC */ |
ffbd517d | 142 | unsigned int sync_io:1; |
d1310b2e CM |
143 | }; |
144 | ||
57599c7e | 145 | static int add_extent_changeset(struct extent_state *state, unsigned bits, |
d38ed27f QW |
146 | struct extent_changeset *changeset, |
147 | int set) | |
148 | { | |
149 | int ret; | |
150 | ||
151 | if (!changeset) | |
57599c7e | 152 | return 0; |
d38ed27f | 153 | if (set && (state->state & bits) == bits) |
57599c7e | 154 | return 0; |
fefdc557 | 155 | if (!set && (state->state & bits) == 0) |
57599c7e | 156 | return 0; |
d38ed27f | 157 | changeset->bytes_changed += state->end - state->start + 1; |
53d32359 | 158 | ret = ulist_add(&changeset->range_changed, state->start, state->end, |
d38ed27f | 159 | GFP_ATOMIC); |
57599c7e | 160 | return ret; |
d38ed27f QW |
161 | } |
162 | ||
c1be9c1a NB |
163 | int __must_check submit_one_bio(struct bio *bio, int mirror_num, |
164 | unsigned long bio_flags) | |
bb58eb9e QW |
165 | { |
166 | blk_status_t ret = 0; | |
bb58eb9e | 167 | struct extent_io_tree *tree = bio->bi_private; |
bb58eb9e QW |
168 | |
169 | bio->bi_private = NULL; | |
170 | ||
908930f3 NB |
171 | if (is_data_inode(tree->private_data)) |
172 | ret = btrfs_submit_data_bio(tree->private_data, bio, mirror_num, | |
173 | bio_flags); | |
174 | else | |
1b36294a NB |
175 | ret = btrfs_submit_metadata_bio(tree->private_data, bio, |
176 | mirror_num, bio_flags); | |
bb58eb9e QW |
177 | |
178 | return blk_status_to_errno(ret); | |
179 | } | |
180 | ||
3065976b QW |
181 | /* Cleanup unsubmitted bios */ |
182 | static void end_write_bio(struct extent_page_data *epd, int ret) | |
183 | { | |
184 | if (epd->bio) { | |
185 | epd->bio->bi_status = errno_to_blk_status(ret); | |
186 | bio_endio(epd->bio); | |
187 | epd->bio = NULL; | |
188 | } | |
189 | } | |
190 | ||
f4340622 QW |
191 | /* |
192 | * Submit bio from extent page data via submit_one_bio | |
193 | * | |
194 | * Return 0 if everything is OK. | |
195 | * Return <0 for error. | |
196 | */ | |
197 | static int __must_check flush_write_bio(struct extent_page_data *epd) | |
bb58eb9e | 198 | { |
f4340622 | 199 | int ret = 0; |
bb58eb9e | 200 | |
f4340622 | 201 | if (epd->bio) { |
bb58eb9e | 202 | ret = submit_one_bio(epd->bio, 0, 0); |
f4340622 QW |
203 | /* |
204 | * Clean up of epd->bio is handled by its endio function. | |
205 | * And endio is either triggered by successful bio execution | |
206 | * or the error handler of submit bio hook. | |
207 | * So at this point, no matter what happened, we don't need | |
208 | * to clean up epd->bio. | |
209 | */ | |
bb58eb9e QW |
210 | epd->bio = NULL; |
211 | } | |
f4340622 | 212 | return ret; |
bb58eb9e | 213 | } |
e2932ee0 | 214 | |
6f0d04f8 | 215 | int __init extent_state_cache_init(void) |
d1310b2e | 216 | { |
837e1972 | 217 | extent_state_cache = kmem_cache_create("btrfs_extent_state", |
9601e3f6 | 218 | sizeof(struct extent_state), 0, |
fba4b697 | 219 | SLAB_MEM_SPREAD, NULL); |
d1310b2e CM |
220 | if (!extent_state_cache) |
221 | return -ENOMEM; | |
6f0d04f8 JB |
222 | return 0; |
223 | } | |
d1310b2e | 224 | |
6f0d04f8 JB |
225 | int __init extent_io_init(void) |
226 | { | |
837e1972 | 227 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 | 228 | sizeof(struct extent_buffer), 0, |
fba4b697 | 229 | SLAB_MEM_SPREAD, NULL); |
d1310b2e | 230 | if (!extent_buffer_cache) |
6f0d04f8 | 231 | return -ENOMEM; |
9be3395b | 232 | |
8ac9f7c1 KO |
233 | if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE, |
234 | offsetof(struct btrfs_io_bio, bio), | |
235 | BIOSET_NEED_BVECS)) | |
9be3395b | 236 | goto free_buffer_cache; |
b208c2f7 | 237 | |
8ac9f7c1 | 238 | if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE)) |
b208c2f7 DW |
239 | goto free_bioset; |
240 | ||
d1310b2e CM |
241 | return 0; |
242 | ||
b208c2f7 | 243 | free_bioset: |
8ac9f7c1 | 244 | bioset_exit(&btrfs_bioset); |
b208c2f7 | 245 | |
9be3395b CM |
246 | free_buffer_cache: |
247 | kmem_cache_destroy(extent_buffer_cache); | |
248 | extent_buffer_cache = NULL; | |
6f0d04f8 JB |
249 | return -ENOMEM; |
250 | } | |
9be3395b | 251 | |
6f0d04f8 JB |
252 | void __cold extent_state_cache_exit(void) |
253 | { | |
254 | btrfs_extent_state_leak_debug_check(); | |
d1310b2e | 255 | kmem_cache_destroy(extent_state_cache); |
d1310b2e CM |
256 | } |
257 | ||
e67c718b | 258 | void __cold extent_io_exit(void) |
d1310b2e | 259 | { |
8c0a8537 KS |
260 | /* |
261 | * Make sure all delayed rcu free are flushed before we | |
262 | * destroy caches. | |
263 | */ | |
264 | rcu_barrier(); | |
5598e900 | 265 | kmem_cache_destroy(extent_buffer_cache); |
8ac9f7c1 | 266 | bioset_exit(&btrfs_bioset); |
d1310b2e CM |
267 | } |
268 | ||
41a2ee75 JB |
269 | /* |
270 | * For the file_extent_tree, we want to hold the inode lock when we lookup and | |
271 | * update the disk_i_size, but lockdep will complain because our io_tree we hold | |
272 | * the tree lock and get the inode lock when setting delalloc. These two things | |
273 | * are unrelated, so make a class for the file_extent_tree so we don't get the | |
274 | * two locking patterns mixed up. | |
275 | */ | |
276 | static struct lock_class_key file_extent_tree_class; | |
277 | ||
c258d6e3 | 278 | void extent_io_tree_init(struct btrfs_fs_info *fs_info, |
43eb5f29 QW |
279 | struct extent_io_tree *tree, unsigned int owner, |
280 | void *private_data) | |
d1310b2e | 281 | { |
c258d6e3 | 282 | tree->fs_info = fs_info; |
6bef4d31 | 283 | tree->state = RB_ROOT; |
d1310b2e | 284 | tree->dirty_bytes = 0; |
70dec807 | 285 | spin_lock_init(&tree->lock); |
c6100a4b | 286 | tree->private_data = private_data; |
43eb5f29 | 287 | tree->owner = owner; |
41a2ee75 JB |
288 | if (owner == IO_TREE_INODE_FILE_EXTENT) |
289 | lockdep_set_class(&tree->lock, &file_extent_tree_class); | |
d1310b2e | 290 | } |
d1310b2e | 291 | |
41e7acd3 NB |
292 | void extent_io_tree_release(struct extent_io_tree *tree) |
293 | { | |
294 | spin_lock(&tree->lock); | |
295 | /* | |
296 | * Do a single barrier for the waitqueue_active check here, the state | |
297 | * of the waitqueue should not change once extent_io_tree_release is | |
298 | * called. | |
299 | */ | |
300 | smp_mb(); | |
301 | while (!RB_EMPTY_ROOT(&tree->state)) { | |
302 | struct rb_node *node; | |
303 | struct extent_state *state; | |
304 | ||
305 | node = rb_first(&tree->state); | |
306 | state = rb_entry(node, struct extent_state, rb_node); | |
307 | rb_erase(&state->rb_node, &tree->state); | |
308 | RB_CLEAR_NODE(&state->rb_node); | |
309 | /* | |
310 | * btree io trees aren't supposed to have tasks waiting for | |
311 | * changes in the flags of extent states ever. | |
312 | */ | |
313 | ASSERT(!waitqueue_active(&state->wq)); | |
314 | free_extent_state(state); | |
315 | ||
316 | cond_resched_lock(&tree->lock); | |
317 | } | |
318 | spin_unlock(&tree->lock); | |
319 | } | |
320 | ||
b2950863 | 321 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
322 | { |
323 | struct extent_state *state; | |
d1310b2e | 324 | |
3ba7ab22 MH |
325 | /* |
326 | * The given mask might be not appropriate for the slab allocator, | |
327 | * drop the unsupported bits | |
328 | */ | |
329 | mask &= ~(__GFP_DMA32|__GFP_HIGHMEM); | |
d1310b2e | 330 | state = kmem_cache_alloc(extent_state_cache, mask); |
2b114d1d | 331 | if (!state) |
d1310b2e CM |
332 | return state; |
333 | state->state = 0; | |
47dc196a | 334 | state->failrec = NULL; |
27a3507d | 335 | RB_CLEAR_NODE(&state->rb_node); |
3fd63727 | 336 | btrfs_leak_debug_add(&leak_lock, &state->leak_list, &states); |
b7ac31b7 | 337 | refcount_set(&state->refs, 1); |
d1310b2e | 338 | init_waitqueue_head(&state->wq); |
143bede5 | 339 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
340 | return state; |
341 | } | |
d1310b2e | 342 | |
4845e44f | 343 | void free_extent_state(struct extent_state *state) |
d1310b2e | 344 | { |
d1310b2e CM |
345 | if (!state) |
346 | return; | |
b7ac31b7 | 347 | if (refcount_dec_and_test(&state->refs)) { |
27a3507d | 348 | WARN_ON(extent_state_in_tree(state)); |
3fd63727 | 349 | btrfs_leak_debug_del(&leak_lock, &state->leak_list); |
143bede5 | 350 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
351 | kmem_cache_free(extent_state_cache, state); |
352 | } | |
353 | } | |
d1310b2e | 354 | |
f2071b21 FM |
355 | static struct rb_node *tree_insert(struct rb_root *root, |
356 | struct rb_node *search_start, | |
357 | u64 offset, | |
12cfbad9 FDBM |
358 | struct rb_node *node, |
359 | struct rb_node ***p_in, | |
360 | struct rb_node **parent_in) | |
d1310b2e | 361 | { |
f2071b21 | 362 | struct rb_node **p; |
d397712b | 363 | struct rb_node *parent = NULL; |
d1310b2e CM |
364 | struct tree_entry *entry; |
365 | ||
12cfbad9 FDBM |
366 | if (p_in && parent_in) { |
367 | p = *p_in; | |
368 | parent = *parent_in; | |
369 | goto do_insert; | |
370 | } | |
371 | ||
f2071b21 | 372 | p = search_start ? &search_start : &root->rb_node; |
d397712b | 373 | while (*p) { |
d1310b2e CM |
374 | parent = *p; |
375 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
376 | ||
377 | if (offset < entry->start) | |
378 | p = &(*p)->rb_left; | |
379 | else if (offset > entry->end) | |
380 | p = &(*p)->rb_right; | |
381 | else | |
382 | return parent; | |
383 | } | |
384 | ||
12cfbad9 | 385 | do_insert: |
d1310b2e CM |
386 | rb_link_node(node, parent, p); |
387 | rb_insert_color(node, root); | |
388 | return NULL; | |
389 | } | |
390 | ||
8666e638 NB |
391 | /** |
392 | * __etree_search - searche @tree for an entry that contains @offset. Such | |
393 | * entry would have entry->start <= offset && entry->end >= offset. | |
394 | * | |
395 | * @tree - the tree to search | |
396 | * @offset - offset that should fall within an entry in @tree | |
397 | * @next_ret - pointer to the first entry whose range ends after @offset | |
398 | * @prev - pointer to the first entry whose range begins before @offset | |
399 | * @p_ret - pointer where new node should be anchored (used when inserting an | |
400 | * entry in the tree) | |
401 | * @parent_ret - points to entry which would have been the parent of the entry, | |
402 | * containing @offset | |
403 | * | |
404 | * This function returns a pointer to the entry that contains @offset byte | |
405 | * address. If no such entry exists, then NULL is returned and the other | |
406 | * pointer arguments to the function are filled, otherwise the found entry is | |
407 | * returned and other pointers are left untouched. | |
408 | */ | |
80ea96b1 | 409 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
12cfbad9 | 410 | struct rb_node **next_ret, |
352646c7 | 411 | struct rb_node **prev_ret, |
12cfbad9 FDBM |
412 | struct rb_node ***p_ret, |
413 | struct rb_node **parent_ret) | |
d1310b2e | 414 | { |
80ea96b1 | 415 | struct rb_root *root = &tree->state; |
12cfbad9 | 416 | struct rb_node **n = &root->rb_node; |
d1310b2e CM |
417 | struct rb_node *prev = NULL; |
418 | struct rb_node *orig_prev = NULL; | |
419 | struct tree_entry *entry; | |
420 | struct tree_entry *prev_entry = NULL; | |
421 | ||
12cfbad9 FDBM |
422 | while (*n) { |
423 | prev = *n; | |
424 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
d1310b2e CM |
425 | prev_entry = entry; |
426 | ||
427 | if (offset < entry->start) | |
12cfbad9 | 428 | n = &(*n)->rb_left; |
d1310b2e | 429 | else if (offset > entry->end) |
12cfbad9 | 430 | n = &(*n)->rb_right; |
d397712b | 431 | else |
12cfbad9 | 432 | return *n; |
d1310b2e CM |
433 | } |
434 | ||
12cfbad9 FDBM |
435 | if (p_ret) |
436 | *p_ret = n; | |
437 | if (parent_ret) | |
438 | *parent_ret = prev; | |
439 | ||
352646c7 | 440 | if (next_ret) { |
d1310b2e | 441 | orig_prev = prev; |
d397712b | 442 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
443 | prev = rb_next(prev); |
444 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
445 | } | |
352646c7 | 446 | *next_ret = prev; |
d1310b2e CM |
447 | prev = orig_prev; |
448 | } | |
449 | ||
352646c7 | 450 | if (prev_ret) { |
d1310b2e | 451 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); |
d397712b | 452 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
453 | prev = rb_prev(prev); |
454 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
455 | } | |
352646c7 | 456 | *prev_ret = prev; |
d1310b2e CM |
457 | } |
458 | return NULL; | |
459 | } | |
460 | ||
12cfbad9 FDBM |
461 | static inline struct rb_node * |
462 | tree_search_for_insert(struct extent_io_tree *tree, | |
463 | u64 offset, | |
464 | struct rb_node ***p_ret, | |
465 | struct rb_node **parent_ret) | |
d1310b2e | 466 | { |
352646c7 | 467 | struct rb_node *next= NULL; |
d1310b2e | 468 | struct rb_node *ret; |
70dec807 | 469 | |
352646c7 | 470 | ret = __etree_search(tree, offset, &next, NULL, p_ret, parent_ret); |
d397712b | 471 | if (!ret) |
352646c7 | 472 | return next; |
d1310b2e CM |
473 | return ret; |
474 | } | |
475 | ||
12cfbad9 FDBM |
476 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
477 | u64 offset) | |
478 | { | |
479 | return tree_search_for_insert(tree, offset, NULL, NULL); | |
480 | } | |
481 | ||
d1310b2e CM |
482 | /* |
483 | * utility function to look for merge candidates inside a given range. | |
484 | * Any extents with matching state are merged together into a single | |
485 | * extent in the tree. Extents with EXTENT_IO in their state field | |
486 | * are not merged because the end_io handlers need to be able to do | |
487 | * operations on them without sleeping (or doing allocations/splits). | |
488 | * | |
489 | * This should be called with the tree lock held. | |
490 | */ | |
1bf85046 JM |
491 | static void merge_state(struct extent_io_tree *tree, |
492 | struct extent_state *state) | |
d1310b2e CM |
493 | { |
494 | struct extent_state *other; | |
495 | struct rb_node *other_node; | |
496 | ||
8882679e | 497 | if (state->state & (EXTENT_LOCKED | EXTENT_BOUNDARY)) |
1bf85046 | 498 | return; |
d1310b2e CM |
499 | |
500 | other_node = rb_prev(&state->rb_node); | |
501 | if (other_node) { | |
502 | other = rb_entry(other_node, struct extent_state, rb_node); | |
503 | if (other->end == state->start - 1 && | |
504 | other->state == state->state) { | |
5c848198 NB |
505 | if (tree->private_data && |
506 | is_data_inode(tree->private_data)) | |
507 | btrfs_merge_delalloc_extent(tree->private_data, | |
508 | state, other); | |
d1310b2e | 509 | state->start = other->start; |
d1310b2e | 510 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 511 | RB_CLEAR_NODE(&other->rb_node); |
d1310b2e CM |
512 | free_extent_state(other); |
513 | } | |
514 | } | |
515 | other_node = rb_next(&state->rb_node); | |
516 | if (other_node) { | |
517 | other = rb_entry(other_node, struct extent_state, rb_node); | |
518 | if (other->start == state->end + 1 && | |
519 | other->state == state->state) { | |
5c848198 NB |
520 | if (tree->private_data && |
521 | is_data_inode(tree->private_data)) | |
522 | btrfs_merge_delalloc_extent(tree->private_data, | |
523 | state, other); | |
df98b6e2 | 524 | state->end = other->end; |
df98b6e2 | 525 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 526 | RB_CLEAR_NODE(&other->rb_node); |
df98b6e2 | 527 | free_extent_state(other); |
d1310b2e CM |
528 | } |
529 | } | |
d1310b2e CM |
530 | } |
531 | ||
3150b699 | 532 | static void set_state_bits(struct extent_io_tree *tree, |
d38ed27f QW |
533 | struct extent_state *state, unsigned *bits, |
534 | struct extent_changeset *changeset); | |
3150b699 | 535 | |
d1310b2e CM |
536 | /* |
537 | * insert an extent_state struct into the tree. 'bits' are set on the | |
538 | * struct before it is inserted. | |
539 | * | |
540 | * This may return -EEXIST if the extent is already there, in which case the | |
541 | * state struct is freed. | |
542 | * | |
543 | * The tree lock is not taken internally. This is a utility function and | |
544 | * probably isn't what you want to call (see set/clear_extent_bit). | |
545 | */ | |
546 | static int insert_state(struct extent_io_tree *tree, | |
547 | struct extent_state *state, u64 start, u64 end, | |
12cfbad9 FDBM |
548 | struct rb_node ***p, |
549 | struct rb_node **parent, | |
d38ed27f | 550 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e CM |
551 | { |
552 | struct rb_node *node; | |
553 | ||
2792237d DS |
554 | if (end < start) { |
555 | btrfs_err(tree->fs_info, | |
556 | "insert state: end < start %llu %llu", end, start); | |
557 | WARN_ON(1); | |
558 | } | |
d1310b2e CM |
559 | state->start = start; |
560 | state->end = end; | |
9ed74f2d | 561 | |
d38ed27f | 562 | set_state_bits(tree, state, bits, changeset); |
3150b699 | 563 | |
f2071b21 | 564 | node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent); |
d1310b2e CM |
565 | if (node) { |
566 | struct extent_state *found; | |
567 | found = rb_entry(node, struct extent_state, rb_node); | |
2792237d DS |
568 | btrfs_err(tree->fs_info, |
569 | "found node %llu %llu on insert of %llu %llu", | |
c1c9ff7c | 570 | found->start, found->end, start, end); |
d1310b2e CM |
571 | return -EEXIST; |
572 | } | |
573 | merge_state(tree, state); | |
574 | return 0; | |
575 | } | |
576 | ||
577 | /* | |
578 | * split a given extent state struct in two, inserting the preallocated | |
579 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
580 | * offset inside 'orig' where it should be split. | |
581 | * | |
582 | * Before calling, | |
583 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
584 | * are two extent state structs in the tree: | |
585 | * prealloc: [orig->start, split - 1] | |
586 | * orig: [ split, orig->end ] | |
587 | * | |
588 | * The tree locks are not taken by this function. They need to be held | |
589 | * by the caller. | |
590 | */ | |
591 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
592 | struct extent_state *prealloc, u64 split) | |
593 | { | |
594 | struct rb_node *node; | |
9ed74f2d | 595 | |
abbb55f4 NB |
596 | if (tree->private_data && is_data_inode(tree->private_data)) |
597 | btrfs_split_delalloc_extent(tree->private_data, orig, split); | |
9ed74f2d | 598 | |
d1310b2e CM |
599 | prealloc->start = orig->start; |
600 | prealloc->end = split - 1; | |
601 | prealloc->state = orig->state; | |
602 | orig->start = split; | |
603 | ||
f2071b21 FM |
604 | node = tree_insert(&tree->state, &orig->rb_node, prealloc->end, |
605 | &prealloc->rb_node, NULL, NULL); | |
d1310b2e | 606 | if (node) { |
d1310b2e CM |
607 | free_extent_state(prealloc); |
608 | return -EEXIST; | |
609 | } | |
610 | return 0; | |
611 | } | |
612 | ||
cdc6a395 LZ |
613 | static struct extent_state *next_state(struct extent_state *state) |
614 | { | |
615 | struct rb_node *next = rb_next(&state->rb_node); | |
616 | if (next) | |
617 | return rb_entry(next, struct extent_state, rb_node); | |
618 | else | |
619 | return NULL; | |
620 | } | |
621 | ||
d1310b2e CM |
622 | /* |
623 | * utility function to clear some bits in an extent state struct. | |
52042d8e | 624 | * it will optionally wake up anyone waiting on this state (wake == 1). |
d1310b2e CM |
625 | * |
626 | * If no bits are set on the state struct after clearing things, the | |
627 | * struct is freed and removed from the tree | |
628 | */ | |
cdc6a395 LZ |
629 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
630 | struct extent_state *state, | |
fefdc557 QW |
631 | unsigned *bits, int wake, |
632 | struct extent_changeset *changeset) | |
d1310b2e | 633 | { |
cdc6a395 | 634 | struct extent_state *next; |
9ee49a04 | 635 | unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS; |
57599c7e | 636 | int ret; |
d1310b2e | 637 | |
0ca1f7ce | 638 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
639 | u64 range = state->end - state->start + 1; |
640 | WARN_ON(range > tree->dirty_bytes); | |
641 | tree->dirty_bytes -= range; | |
642 | } | |
a36bb5f9 NB |
643 | |
644 | if (tree->private_data && is_data_inode(tree->private_data)) | |
645 | btrfs_clear_delalloc_extent(tree->private_data, state, bits); | |
646 | ||
57599c7e DS |
647 | ret = add_extent_changeset(state, bits_to_clear, changeset, 0); |
648 | BUG_ON(ret < 0); | |
32c00aff | 649 | state->state &= ~bits_to_clear; |
d1310b2e CM |
650 | if (wake) |
651 | wake_up(&state->wq); | |
0ca1f7ce | 652 | if (state->state == 0) { |
cdc6a395 | 653 | next = next_state(state); |
27a3507d | 654 | if (extent_state_in_tree(state)) { |
d1310b2e | 655 | rb_erase(&state->rb_node, &tree->state); |
27a3507d | 656 | RB_CLEAR_NODE(&state->rb_node); |
d1310b2e CM |
657 | free_extent_state(state); |
658 | } else { | |
659 | WARN_ON(1); | |
660 | } | |
661 | } else { | |
662 | merge_state(tree, state); | |
cdc6a395 | 663 | next = next_state(state); |
d1310b2e | 664 | } |
cdc6a395 | 665 | return next; |
d1310b2e CM |
666 | } |
667 | ||
8233767a XG |
668 | static struct extent_state * |
669 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
670 | { | |
671 | if (!prealloc) | |
672 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
673 | ||
674 | return prealloc; | |
675 | } | |
676 | ||
48a3b636 | 677 | static void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
c2d904e0 | 678 | { |
05912a3c DS |
679 | struct inode *inode = tree->private_data; |
680 | ||
681 | btrfs_panic(btrfs_sb(inode->i_sb), err, | |
682 | "locking error: extent tree was modified by another thread while locked"); | |
c2d904e0 JM |
683 | } |
684 | ||
d1310b2e CM |
685 | /* |
686 | * clear some bits on a range in the tree. This may require splitting | |
687 | * or inserting elements in the tree, so the gfp mask is used to | |
688 | * indicate which allocations or sleeping are allowed. | |
689 | * | |
690 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
691 | * the given range from the tree regardless of state (ie for truncate). | |
692 | * | |
693 | * the range [start, end] is inclusive. | |
694 | * | |
6763af84 | 695 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e | 696 | */ |
66b0c887 | 697 | int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
fefdc557 QW |
698 | unsigned bits, int wake, int delete, |
699 | struct extent_state **cached_state, | |
700 | gfp_t mask, struct extent_changeset *changeset) | |
d1310b2e CM |
701 | { |
702 | struct extent_state *state; | |
2c64c53d | 703 | struct extent_state *cached; |
d1310b2e CM |
704 | struct extent_state *prealloc = NULL; |
705 | struct rb_node *node; | |
5c939df5 | 706 | u64 last_end; |
d1310b2e | 707 | int err; |
2ac55d41 | 708 | int clear = 0; |
d1310b2e | 709 | |
a5dee37d | 710 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 | 711 | trace_btrfs_clear_extent_bit(tree, start, end - start + 1, bits); |
8d599ae1 | 712 | |
7ee9e440 JB |
713 | if (bits & EXTENT_DELALLOC) |
714 | bits |= EXTENT_NORESERVE; | |
715 | ||
0ca1f7ce YZ |
716 | if (delete) |
717 | bits |= ~EXTENT_CTLBITS; | |
0ca1f7ce | 718 | |
8882679e | 719 | if (bits & (EXTENT_LOCKED | EXTENT_BOUNDARY)) |
2ac55d41 | 720 | clear = 1; |
d1310b2e | 721 | again: |
d0164adc | 722 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
c7bc6319 FM |
723 | /* |
724 | * Don't care for allocation failure here because we might end | |
725 | * up not needing the pre-allocated extent state at all, which | |
726 | * is the case if we only have in the tree extent states that | |
727 | * cover our input range and don't cover too any other range. | |
728 | * If we end up needing a new extent state we allocate it later. | |
729 | */ | |
d1310b2e | 730 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
731 | } |
732 | ||
cad321ad | 733 | spin_lock(&tree->lock); |
2c64c53d CM |
734 | if (cached_state) { |
735 | cached = *cached_state; | |
2ac55d41 JB |
736 | |
737 | if (clear) { | |
738 | *cached_state = NULL; | |
739 | cached_state = NULL; | |
740 | } | |
741 | ||
27a3507d FM |
742 | if (cached && extent_state_in_tree(cached) && |
743 | cached->start <= start && cached->end > start) { | |
2ac55d41 | 744 | if (clear) |
b7ac31b7 | 745 | refcount_dec(&cached->refs); |
2c64c53d | 746 | state = cached; |
42daec29 | 747 | goto hit_next; |
2c64c53d | 748 | } |
2ac55d41 JB |
749 | if (clear) |
750 | free_extent_state(cached); | |
2c64c53d | 751 | } |
d1310b2e CM |
752 | /* |
753 | * this search will find the extents that end after | |
754 | * our range starts | |
755 | */ | |
80ea96b1 | 756 | node = tree_search(tree, start); |
d1310b2e CM |
757 | if (!node) |
758 | goto out; | |
759 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 760 | hit_next: |
d1310b2e CM |
761 | if (state->start > end) |
762 | goto out; | |
763 | WARN_ON(state->end < start); | |
5c939df5 | 764 | last_end = state->end; |
d1310b2e | 765 | |
0449314a | 766 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
767 | if (!(state->state & bits)) { |
768 | state = next_state(state); | |
0449314a | 769 | goto next; |
cdc6a395 | 770 | } |
0449314a | 771 | |
d1310b2e CM |
772 | /* |
773 | * | ---- desired range ---- | | |
774 | * | state | or | |
775 | * | ------------- state -------------- | | |
776 | * | |
777 | * We need to split the extent we found, and may flip | |
778 | * bits on second half. | |
779 | * | |
780 | * If the extent we found extends past our range, we | |
781 | * just split and search again. It'll get split again | |
782 | * the next time though. | |
783 | * | |
784 | * If the extent we found is inside our range, we clear | |
785 | * the desired bit on it. | |
786 | */ | |
787 | ||
788 | if (state->start < start) { | |
8233767a XG |
789 | prealloc = alloc_extent_state_atomic(prealloc); |
790 | BUG_ON(!prealloc); | |
d1310b2e | 791 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
792 | if (err) |
793 | extent_io_tree_panic(tree, err); | |
794 | ||
d1310b2e CM |
795 | prealloc = NULL; |
796 | if (err) | |
797 | goto out; | |
798 | if (state->end <= end) { | |
fefdc557 QW |
799 | state = clear_state_bit(tree, state, &bits, wake, |
800 | changeset); | |
d1ac6e41 | 801 | goto next; |
d1310b2e CM |
802 | } |
803 | goto search_again; | |
804 | } | |
805 | /* | |
806 | * | ---- desired range ---- | | |
807 | * | state | | |
808 | * We need to split the extent, and clear the bit | |
809 | * on the first half | |
810 | */ | |
811 | if (state->start <= end && state->end > end) { | |
8233767a XG |
812 | prealloc = alloc_extent_state_atomic(prealloc); |
813 | BUG_ON(!prealloc); | |
d1310b2e | 814 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
815 | if (err) |
816 | extent_io_tree_panic(tree, err); | |
817 | ||
d1310b2e CM |
818 | if (wake) |
819 | wake_up(&state->wq); | |
42daec29 | 820 | |
fefdc557 | 821 | clear_state_bit(tree, prealloc, &bits, wake, changeset); |
9ed74f2d | 822 | |
d1310b2e CM |
823 | prealloc = NULL; |
824 | goto out; | |
825 | } | |
42daec29 | 826 | |
fefdc557 | 827 | state = clear_state_bit(tree, state, &bits, wake, changeset); |
0449314a | 828 | next: |
5c939df5 YZ |
829 | if (last_end == (u64)-1) |
830 | goto out; | |
831 | start = last_end + 1; | |
cdc6a395 | 832 | if (start <= end && state && !need_resched()) |
692e5759 | 833 | goto hit_next; |
d1310b2e CM |
834 | |
835 | search_again: | |
836 | if (start > end) | |
837 | goto out; | |
cad321ad | 838 | spin_unlock(&tree->lock); |
d0164adc | 839 | if (gfpflags_allow_blocking(mask)) |
d1310b2e CM |
840 | cond_resched(); |
841 | goto again; | |
7ab5cb2a DS |
842 | |
843 | out: | |
844 | spin_unlock(&tree->lock); | |
845 | if (prealloc) | |
846 | free_extent_state(prealloc); | |
847 | ||
848 | return 0; | |
849 | ||
d1310b2e | 850 | } |
d1310b2e | 851 | |
143bede5 JM |
852 | static void wait_on_state(struct extent_io_tree *tree, |
853 | struct extent_state *state) | |
641f5219 CH |
854 | __releases(tree->lock) |
855 | __acquires(tree->lock) | |
d1310b2e CM |
856 | { |
857 | DEFINE_WAIT(wait); | |
858 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 859 | spin_unlock(&tree->lock); |
d1310b2e | 860 | schedule(); |
cad321ad | 861 | spin_lock(&tree->lock); |
d1310b2e | 862 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
863 | } |
864 | ||
865 | /* | |
866 | * waits for one or more bits to clear on a range in the state tree. | |
867 | * The range [start, end] is inclusive. | |
868 | * The tree lock is taken by this function | |
869 | */ | |
41074888 DS |
870 | static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
871 | unsigned long bits) | |
d1310b2e CM |
872 | { |
873 | struct extent_state *state; | |
874 | struct rb_node *node; | |
875 | ||
a5dee37d | 876 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 877 | |
cad321ad | 878 | spin_lock(&tree->lock); |
d1310b2e CM |
879 | again: |
880 | while (1) { | |
881 | /* | |
882 | * this search will find all the extents that end after | |
883 | * our range starts | |
884 | */ | |
80ea96b1 | 885 | node = tree_search(tree, start); |
c50d3e71 | 886 | process_node: |
d1310b2e CM |
887 | if (!node) |
888 | break; | |
889 | ||
890 | state = rb_entry(node, struct extent_state, rb_node); | |
891 | ||
892 | if (state->start > end) | |
893 | goto out; | |
894 | ||
895 | if (state->state & bits) { | |
896 | start = state->start; | |
b7ac31b7 | 897 | refcount_inc(&state->refs); |
d1310b2e CM |
898 | wait_on_state(tree, state); |
899 | free_extent_state(state); | |
900 | goto again; | |
901 | } | |
902 | start = state->end + 1; | |
903 | ||
904 | if (start > end) | |
905 | break; | |
906 | ||
c50d3e71 FM |
907 | if (!cond_resched_lock(&tree->lock)) { |
908 | node = rb_next(node); | |
909 | goto process_node; | |
910 | } | |
d1310b2e CM |
911 | } |
912 | out: | |
cad321ad | 913 | spin_unlock(&tree->lock); |
d1310b2e | 914 | } |
d1310b2e | 915 | |
1bf85046 | 916 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 917 | struct extent_state *state, |
d38ed27f | 918 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e | 919 | { |
9ee49a04 | 920 | unsigned bits_to_set = *bits & ~EXTENT_CTLBITS; |
57599c7e | 921 | int ret; |
9ed74f2d | 922 | |
e06a1fc9 NB |
923 | if (tree->private_data && is_data_inode(tree->private_data)) |
924 | btrfs_set_delalloc_extent(tree->private_data, state, bits); | |
925 | ||
0ca1f7ce | 926 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
927 | u64 range = state->end - state->start + 1; |
928 | tree->dirty_bytes += range; | |
929 | } | |
57599c7e DS |
930 | ret = add_extent_changeset(state, bits_to_set, changeset, 1); |
931 | BUG_ON(ret < 0); | |
0ca1f7ce | 932 | state->state |= bits_to_set; |
d1310b2e CM |
933 | } |
934 | ||
e38e2ed7 FM |
935 | static void cache_state_if_flags(struct extent_state *state, |
936 | struct extent_state **cached_ptr, | |
9ee49a04 | 937 | unsigned flags) |
2c64c53d CM |
938 | { |
939 | if (cached_ptr && !(*cached_ptr)) { | |
e38e2ed7 | 940 | if (!flags || (state->state & flags)) { |
2c64c53d | 941 | *cached_ptr = state; |
b7ac31b7 | 942 | refcount_inc(&state->refs); |
2c64c53d CM |
943 | } |
944 | } | |
945 | } | |
946 | ||
e38e2ed7 FM |
947 | static void cache_state(struct extent_state *state, |
948 | struct extent_state **cached_ptr) | |
949 | { | |
950 | return cache_state_if_flags(state, cached_ptr, | |
8882679e | 951 | EXTENT_LOCKED | EXTENT_BOUNDARY); |
e38e2ed7 FM |
952 | } |
953 | ||
d1310b2e | 954 | /* |
1edbb734 CM |
955 | * set some bits on a range in the tree. This may require allocations or |
956 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 957 | * |
1edbb734 CM |
958 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
959 | * part of the range already has the desired bits set. The start of the | |
960 | * existing range is returned in failed_start in this case. | |
d1310b2e | 961 | * |
1edbb734 | 962 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 963 | */ |
1edbb734 | 964 | |
3fbe5c02 JM |
965 | static int __must_check |
966 | __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 967 | unsigned bits, unsigned exclusive_bits, |
41074888 | 968 | u64 *failed_start, struct extent_state **cached_state, |
d38ed27f | 969 | gfp_t mask, struct extent_changeset *changeset) |
d1310b2e CM |
970 | { |
971 | struct extent_state *state; | |
972 | struct extent_state *prealloc = NULL; | |
973 | struct rb_node *node; | |
12cfbad9 FDBM |
974 | struct rb_node **p; |
975 | struct rb_node *parent; | |
d1310b2e | 976 | int err = 0; |
d1310b2e CM |
977 | u64 last_start; |
978 | u64 last_end; | |
42daec29 | 979 | |
a5dee37d | 980 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 | 981 | trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits); |
8d599ae1 | 982 | |
3f6bb4ae QW |
983 | if (exclusive_bits) |
984 | ASSERT(failed_start); | |
985 | else | |
986 | ASSERT(failed_start == NULL); | |
d1310b2e | 987 | again: |
d0164adc | 988 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
059f791c DS |
989 | /* |
990 | * Don't care for allocation failure here because we might end | |
991 | * up not needing the pre-allocated extent state at all, which | |
992 | * is the case if we only have in the tree extent states that | |
993 | * cover our input range and don't cover too any other range. | |
994 | * If we end up needing a new extent state we allocate it later. | |
995 | */ | |
d1310b2e | 996 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
997 | } |
998 | ||
cad321ad | 999 | spin_lock(&tree->lock); |
9655d298 CM |
1000 | if (cached_state && *cached_state) { |
1001 | state = *cached_state; | |
df98b6e2 | 1002 | if (state->start <= start && state->end > start && |
27a3507d | 1003 | extent_state_in_tree(state)) { |
9655d298 CM |
1004 | node = &state->rb_node; |
1005 | goto hit_next; | |
1006 | } | |
1007 | } | |
d1310b2e CM |
1008 | /* |
1009 | * this search will find all the extents that end after | |
1010 | * our range starts. | |
1011 | */ | |
12cfbad9 | 1012 | node = tree_search_for_insert(tree, start, &p, &parent); |
d1310b2e | 1013 | if (!node) { |
8233767a XG |
1014 | prealloc = alloc_extent_state_atomic(prealloc); |
1015 | BUG_ON(!prealloc); | |
12cfbad9 | 1016 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 1017 | &p, &parent, &bits, changeset); |
c2d904e0 JM |
1018 | if (err) |
1019 | extent_io_tree_panic(tree, err); | |
1020 | ||
c42ac0bc | 1021 | cache_state(prealloc, cached_state); |
d1310b2e | 1022 | prealloc = NULL; |
d1310b2e CM |
1023 | goto out; |
1024 | } | |
d1310b2e | 1025 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 1026 | hit_next: |
d1310b2e CM |
1027 | last_start = state->start; |
1028 | last_end = state->end; | |
1029 | ||
1030 | /* | |
1031 | * | ---- desired range ---- | | |
1032 | * | state | | |
1033 | * | |
1034 | * Just lock what we found and keep going | |
1035 | */ | |
1036 | if (state->start == start && state->end <= end) { | |
1edbb734 | 1037 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1038 | *failed_start = state->start; |
1039 | err = -EEXIST; | |
1040 | goto out; | |
1041 | } | |
42daec29 | 1042 | |
d38ed27f | 1043 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 1044 | cache_state(state, cached_state); |
d1310b2e | 1045 | merge_state(tree, state); |
5c939df5 YZ |
1046 | if (last_end == (u64)-1) |
1047 | goto out; | |
1048 | start = last_end + 1; | |
d1ac6e41 LB |
1049 | state = next_state(state); |
1050 | if (start < end && state && state->start == start && | |
1051 | !need_resched()) | |
1052 | goto hit_next; | |
d1310b2e CM |
1053 | goto search_again; |
1054 | } | |
1055 | ||
1056 | /* | |
1057 | * | ---- desired range ---- | | |
1058 | * | state | | |
1059 | * or | |
1060 | * | ------------- state -------------- | | |
1061 | * | |
1062 | * We need to split the extent we found, and may flip bits on | |
1063 | * second half. | |
1064 | * | |
1065 | * If the extent we found extends past our | |
1066 | * range, we just split and search again. It'll get split | |
1067 | * again the next time though. | |
1068 | * | |
1069 | * If the extent we found is inside our range, we set the | |
1070 | * desired bit on it. | |
1071 | */ | |
1072 | if (state->start < start) { | |
1edbb734 | 1073 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1074 | *failed_start = start; |
1075 | err = -EEXIST; | |
1076 | goto out; | |
1077 | } | |
8233767a | 1078 | |
55ffaabe FM |
1079 | /* |
1080 | * If this extent already has all the bits we want set, then | |
1081 | * skip it, not necessary to split it or do anything with it. | |
1082 | */ | |
1083 | if ((state->state & bits) == bits) { | |
1084 | start = state->end + 1; | |
1085 | cache_state(state, cached_state); | |
1086 | goto search_again; | |
1087 | } | |
1088 | ||
8233767a XG |
1089 | prealloc = alloc_extent_state_atomic(prealloc); |
1090 | BUG_ON(!prealloc); | |
d1310b2e | 1091 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1092 | if (err) |
1093 | extent_io_tree_panic(tree, err); | |
1094 | ||
d1310b2e CM |
1095 | prealloc = NULL; |
1096 | if (err) | |
1097 | goto out; | |
1098 | if (state->end <= end) { | |
d38ed27f | 1099 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 1100 | cache_state(state, cached_state); |
d1310b2e | 1101 | merge_state(tree, state); |
5c939df5 YZ |
1102 | if (last_end == (u64)-1) |
1103 | goto out; | |
1104 | start = last_end + 1; | |
d1ac6e41 LB |
1105 | state = next_state(state); |
1106 | if (start < end && state && state->start == start && | |
1107 | !need_resched()) | |
1108 | goto hit_next; | |
d1310b2e CM |
1109 | } |
1110 | goto search_again; | |
1111 | } | |
1112 | /* | |
1113 | * | ---- desired range ---- | | |
1114 | * | state | or | state | | |
1115 | * | |
1116 | * There's a hole, we need to insert something in it and | |
1117 | * ignore the extent we found. | |
1118 | */ | |
1119 | if (state->start > start) { | |
1120 | u64 this_end; | |
1121 | if (end < last_start) | |
1122 | this_end = end; | |
1123 | else | |
d397712b | 1124 | this_end = last_start - 1; |
8233767a XG |
1125 | |
1126 | prealloc = alloc_extent_state_atomic(prealloc); | |
1127 | BUG_ON(!prealloc); | |
c7f895a2 XG |
1128 | |
1129 | /* | |
1130 | * Avoid to free 'prealloc' if it can be merged with | |
1131 | * the later extent. | |
1132 | */ | |
d1310b2e | 1133 | err = insert_state(tree, prealloc, start, this_end, |
d38ed27f | 1134 | NULL, NULL, &bits, changeset); |
c2d904e0 JM |
1135 | if (err) |
1136 | extent_io_tree_panic(tree, err); | |
1137 | ||
9ed74f2d JB |
1138 | cache_state(prealloc, cached_state); |
1139 | prealloc = NULL; | |
d1310b2e CM |
1140 | start = this_end + 1; |
1141 | goto search_again; | |
1142 | } | |
1143 | /* | |
1144 | * | ---- desired range ---- | | |
1145 | * | state | | |
1146 | * We need to split the extent, and set the bit | |
1147 | * on the first half | |
1148 | */ | |
1149 | if (state->start <= end && state->end > end) { | |
1edbb734 | 1150 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1151 | *failed_start = start; |
1152 | err = -EEXIST; | |
1153 | goto out; | |
1154 | } | |
8233767a XG |
1155 | |
1156 | prealloc = alloc_extent_state_atomic(prealloc); | |
1157 | BUG_ON(!prealloc); | |
d1310b2e | 1158 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
1159 | if (err) |
1160 | extent_io_tree_panic(tree, err); | |
d1310b2e | 1161 | |
d38ed27f | 1162 | set_state_bits(tree, prealloc, &bits, changeset); |
2c64c53d | 1163 | cache_state(prealloc, cached_state); |
d1310b2e CM |
1164 | merge_state(tree, prealloc); |
1165 | prealloc = NULL; | |
1166 | goto out; | |
1167 | } | |
1168 | ||
b5a4ba14 DS |
1169 | search_again: |
1170 | if (start > end) | |
1171 | goto out; | |
1172 | spin_unlock(&tree->lock); | |
1173 | if (gfpflags_allow_blocking(mask)) | |
1174 | cond_resched(); | |
1175 | goto again; | |
d1310b2e CM |
1176 | |
1177 | out: | |
cad321ad | 1178 | spin_unlock(&tree->lock); |
d1310b2e CM |
1179 | if (prealloc) |
1180 | free_extent_state(prealloc); | |
1181 | ||
1182 | return err; | |
1183 | ||
d1310b2e | 1184 | } |
d1310b2e | 1185 | |
41074888 | 1186 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
3f6bb4ae | 1187 | unsigned bits, struct extent_state **cached_state, gfp_t mask) |
3fbe5c02 | 1188 | { |
3f6bb4ae QW |
1189 | return __set_extent_bit(tree, start, end, bits, 0, NULL, cached_state, |
1190 | mask, NULL); | |
3fbe5c02 JM |
1191 | } |
1192 | ||
1193 | ||
462d6fac | 1194 | /** |
10983f2e LB |
1195 | * convert_extent_bit - convert all bits in a given range from one bit to |
1196 | * another | |
462d6fac JB |
1197 | * @tree: the io tree to search |
1198 | * @start: the start offset in bytes | |
1199 | * @end: the end offset in bytes (inclusive) | |
1200 | * @bits: the bits to set in this range | |
1201 | * @clear_bits: the bits to clear in this range | |
e6138876 | 1202 | * @cached_state: state that we're going to cache |
462d6fac JB |
1203 | * |
1204 | * This will go through and set bits for the given range. If any states exist | |
1205 | * already in this range they are set with the given bit and cleared of the | |
1206 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
1207 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
1208 | * boundary bits like LOCK. | |
210aa277 DS |
1209 | * |
1210 | * All allocations are done with GFP_NOFS. | |
462d6fac JB |
1211 | */ |
1212 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1213 | unsigned bits, unsigned clear_bits, |
210aa277 | 1214 | struct extent_state **cached_state) |
462d6fac JB |
1215 | { |
1216 | struct extent_state *state; | |
1217 | struct extent_state *prealloc = NULL; | |
1218 | struct rb_node *node; | |
12cfbad9 FDBM |
1219 | struct rb_node **p; |
1220 | struct rb_node *parent; | |
462d6fac JB |
1221 | int err = 0; |
1222 | u64 last_start; | |
1223 | u64 last_end; | |
c8fd3de7 | 1224 | bool first_iteration = true; |
462d6fac | 1225 | |
a5dee37d | 1226 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 QW |
1227 | trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits, |
1228 | clear_bits); | |
8d599ae1 | 1229 | |
462d6fac | 1230 | again: |
210aa277 | 1231 | if (!prealloc) { |
c8fd3de7 FM |
1232 | /* |
1233 | * Best effort, don't worry if extent state allocation fails | |
1234 | * here for the first iteration. We might have a cached state | |
1235 | * that matches exactly the target range, in which case no | |
1236 | * extent state allocations are needed. We'll only know this | |
1237 | * after locking the tree. | |
1238 | */ | |
210aa277 | 1239 | prealloc = alloc_extent_state(GFP_NOFS); |
c8fd3de7 | 1240 | if (!prealloc && !first_iteration) |
462d6fac JB |
1241 | return -ENOMEM; |
1242 | } | |
1243 | ||
1244 | spin_lock(&tree->lock); | |
e6138876 JB |
1245 | if (cached_state && *cached_state) { |
1246 | state = *cached_state; | |
1247 | if (state->start <= start && state->end > start && | |
27a3507d | 1248 | extent_state_in_tree(state)) { |
e6138876 JB |
1249 | node = &state->rb_node; |
1250 | goto hit_next; | |
1251 | } | |
1252 | } | |
1253 | ||
462d6fac JB |
1254 | /* |
1255 | * this search will find all the extents that end after | |
1256 | * our range starts. | |
1257 | */ | |
12cfbad9 | 1258 | node = tree_search_for_insert(tree, start, &p, &parent); |
462d6fac JB |
1259 | if (!node) { |
1260 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1261 | if (!prealloc) { |
1262 | err = -ENOMEM; | |
1263 | goto out; | |
1264 | } | |
12cfbad9 | 1265 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 1266 | &p, &parent, &bits, NULL); |
c2d904e0 JM |
1267 | if (err) |
1268 | extent_io_tree_panic(tree, err); | |
c42ac0bc FDBM |
1269 | cache_state(prealloc, cached_state); |
1270 | prealloc = NULL; | |
462d6fac JB |
1271 | goto out; |
1272 | } | |
1273 | state = rb_entry(node, struct extent_state, rb_node); | |
1274 | hit_next: | |
1275 | last_start = state->start; | |
1276 | last_end = state->end; | |
1277 | ||
1278 | /* | |
1279 | * | ---- desired range ---- | | |
1280 | * | state | | |
1281 | * | |
1282 | * Just lock what we found and keep going | |
1283 | */ | |
1284 | if (state->start == start && state->end <= end) { | |
d38ed27f | 1285 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1286 | cache_state(state, cached_state); |
fefdc557 | 1287 | state = clear_state_bit(tree, state, &clear_bits, 0, NULL); |
462d6fac JB |
1288 | if (last_end == (u64)-1) |
1289 | goto out; | |
462d6fac | 1290 | start = last_end + 1; |
d1ac6e41 LB |
1291 | if (start < end && state && state->start == start && |
1292 | !need_resched()) | |
1293 | goto hit_next; | |
462d6fac JB |
1294 | goto search_again; |
1295 | } | |
1296 | ||
1297 | /* | |
1298 | * | ---- desired range ---- | | |
1299 | * | state | | |
1300 | * or | |
1301 | * | ------------- state -------------- | | |
1302 | * | |
1303 | * We need to split the extent we found, and may flip bits on | |
1304 | * second half. | |
1305 | * | |
1306 | * If the extent we found extends past our | |
1307 | * range, we just split and search again. It'll get split | |
1308 | * again the next time though. | |
1309 | * | |
1310 | * If the extent we found is inside our range, we set the | |
1311 | * desired bit on it. | |
1312 | */ | |
1313 | if (state->start < start) { | |
1314 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1315 | if (!prealloc) { |
1316 | err = -ENOMEM; | |
1317 | goto out; | |
1318 | } | |
462d6fac | 1319 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1320 | if (err) |
1321 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1322 | prealloc = NULL; |
1323 | if (err) | |
1324 | goto out; | |
1325 | if (state->end <= end) { | |
d38ed27f | 1326 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1327 | cache_state(state, cached_state); |
fefdc557 QW |
1328 | state = clear_state_bit(tree, state, &clear_bits, 0, |
1329 | NULL); | |
462d6fac JB |
1330 | if (last_end == (u64)-1) |
1331 | goto out; | |
1332 | start = last_end + 1; | |
d1ac6e41 LB |
1333 | if (start < end && state && state->start == start && |
1334 | !need_resched()) | |
1335 | goto hit_next; | |
462d6fac JB |
1336 | } |
1337 | goto search_again; | |
1338 | } | |
1339 | /* | |
1340 | * | ---- desired range ---- | | |
1341 | * | state | or | state | | |
1342 | * | |
1343 | * There's a hole, we need to insert something in it and | |
1344 | * ignore the extent we found. | |
1345 | */ | |
1346 | if (state->start > start) { | |
1347 | u64 this_end; | |
1348 | if (end < last_start) | |
1349 | this_end = end; | |
1350 | else | |
1351 | this_end = last_start - 1; | |
1352 | ||
1353 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1354 | if (!prealloc) { |
1355 | err = -ENOMEM; | |
1356 | goto out; | |
1357 | } | |
462d6fac JB |
1358 | |
1359 | /* | |
1360 | * Avoid to free 'prealloc' if it can be merged with | |
1361 | * the later extent. | |
1362 | */ | |
1363 | err = insert_state(tree, prealloc, start, this_end, | |
d38ed27f | 1364 | NULL, NULL, &bits, NULL); |
c2d904e0 JM |
1365 | if (err) |
1366 | extent_io_tree_panic(tree, err); | |
e6138876 | 1367 | cache_state(prealloc, cached_state); |
462d6fac JB |
1368 | prealloc = NULL; |
1369 | start = this_end + 1; | |
1370 | goto search_again; | |
1371 | } | |
1372 | /* | |
1373 | * | ---- desired range ---- | | |
1374 | * | state | | |
1375 | * We need to split the extent, and set the bit | |
1376 | * on the first half | |
1377 | */ | |
1378 | if (state->start <= end && state->end > end) { | |
1379 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1380 | if (!prealloc) { |
1381 | err = -ENOMEM; | |
1382 | goto out; | |
1383 | } | |
462d6fac JB |
1384 | |
1385 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1386 | if (err) |
1387 | extent_io_tree_panic(tree, err); | |
462d6fac | 1388 | |
d38ed27f | 1389 | set_state_bits(tree, prealloc, &bits, NULL); |
e6138876 | 1390 | cache_state(prealloc, cached_state); |
fefdc557 | 1391 | clear_state_bit(tree, prealloc, &clear_bits, 0, NULL); |
462d6fac JB |
1392 | prealloc = NULL; |
1393 | goto out; | |
1394 | } | |
1395 | ||
462d6fac JB |
1396 | search_again: |
1397 | if (start > end) | |
1398 | goto out; | |
1399 | spin_unlock(&tree->lock); | |
210aa277 | 1400 | cond_resched(); |
c8fd3de7 | 1401 | first_iteration = false; |
462d6fac | 1402 | goto again; |
462d6fac JB |
1403 | |
1404 | out: | |
1405 | spin_unlock(&tree->lock); | |
1406 | if (prealloc) | |
1407 | free_extent_state(prealloc); | |
1408 | ||
1409 | return err; | |
462d6fac JB |
1410 | } |
1411 | ||
d1310b2e | 1412 | /* wrappers around set/clear extent bit */ |
d38ed27f | 1413 | int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
2c53b912 | 1414 | unsigned bits, struct extent_changeset *changeset) |
d38ed27f QW |
1415 | { |
1416 | /* | |
1417 | * We don't support EXTENT_LOCKED yet, as current changeset will | |
1418 | * record any bits changed, so for EXTENT_LOCKED case, it will | |
1419 | * either fail with -EEXIST or changeset will record the whole | |
1420 | * range. | |
1421 | */ | |
1422 | BUG_ON(bits & EXTENT_LOCKED); | |
1423 | ||
2c53b912 | 1424 | return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS, |
d38ed27f QW |
1425 | changeset); |
1426 | } | |
1427 | ||
4ca73656 NB |
1428 | int set_extent_bits_nowait(struct extent_io_tree *tree, u64 start, u64 end, |
1429 | unsigned bits) | |
1430 | { | |
1431 | return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, | |
1432 | GFP_NOWAIT, NULL); | |
1433 | } | |
1434 | ||
fefdc557 QW |
1435 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
1436 | unsigned bits, int wake, int delete, | |
ae0f1625 | 1437 | struct extent_state **cached) |
fefdc557 QW |
1438 | { |
1439 | return __clear_extent_bit(tree, start, end, bits, wake, delete, | |
ae0f1625 | 1440 | cached, GFP_NOFS, NULL); |
fefdc557 QW |
1441 | } |
1442 | ||
fefdc557 | 1443 | int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
f734c44a | 1444 | unsigned bits, struct extent_changeset *changeset) |
fefdc557 QW |
1445 | { |
1446 | /* | |
1447 | * Don't support EXTENT_LOCKED case, same reason as | |
1448 | * set_record_extent_bits(). | |
1449 | */ | |
1450 | BUG_ON(bits & EXTENT_LOCKED); | |
1451 | ||
f734c44a | 1452 | return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS, |
fefdc557 QW |
1453 | changeset); |
1454 | } | |
1455 | ||
d352ac68 CM |
1456 | /* |
1457 | * either insert or lock state struct between start and end use mask to tell | |
1458 | * us if waiting is desired. | |
1459 | */ | |
1edbb734 | 1460 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
ff13db41 | 1461 | struct extent_state **cached_state) |
d1310b2e CM |
1462 | { |
1463 | int err; | |
1464 | u64 failed_start; | |
9ee49a04 | 1465 | |
d1310b2e | 1466 | while (1) { |
ff13db41 | 1467 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, |
3fbe5c02 | 1468 | EXTENT_LOCKED, &failed_start, |
d38ed27f | 1469 | cached_state, GFP_NOFS, NULL); |
d0082371 | 1470 | if (err == -EEXIST) { |
d1310b2e CM |
1471 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1472 | start = failed_start; | |
d0082371 | 1473 | } else |
d1310b2e | 1474 | break; |
d1310b2e CM |
1475 | WARN_ON(start > end); |
1476 | } | |
1477 | return err; | |
1478 | } | |
d1310b2e | 1479 | |
d0082371 | 1480 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1481 | { |
1482 | int err; | |
1483 | u64 failed_start; | |
1484 | ||
3fbe5c02 | 1485 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
d38ed27f | 1486 | &failed_start, NULL, GFP_NOFS, NULL); |
6643558d YZ |
1487 | if (err == -EEXIST) { |
1488 | if (failed_start > start) | |
1489 | clear_extent_bit(tree, start, failed_start - 1, | |
ae0f1625 | 1490 | EXTENT_LOCKED, 1, 0, NULL); |
25179201 | 1491 | return 0; |
6643558d | 1492 | } |
25179201 JB |
1493 | return 1; |
1494 | } | |
25179201 | 1495 | |
bd1fa4f0 | 1496 | void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1497 | { |
09cbfeaf KS |
1498 | unsigned long index = start >> PAGE_SHIFT; |
1499 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1500 | struct page *page; |
1501 | ||
1502 | while (index <= end_index) { | |
1503 | page = find_get_page(inode->i_mapping, index); | |
1504 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1505 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1506 | put_page(page); |
4adaa611 CM |
1507 | index++; |
1508 | } | |
4adaa611 CM |
1509 | } |
1510 | ||
f6311572 | 1511 | void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1512 | { |
09cbfeaf KS |
1513 | unsigned long index = start >> PAGE_SHIFT; |
1514 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1515 | struct page *page; |
1516 | ||
1517 | while (index <= end_index) { | |
1518 | page = find_get_page(inode->i_mapping, index); | |
1519 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
4adaa611 | 1520 | __set_page_dirty_nobuffers(page); |
8d38633c | 1521 | account_page_redirty(page); |
09cbfeaf | 1522 | put_page(page); |
4adaa611 CM |
1523 | index++; |
1524 | } | |
4adaa611 CM |
1525 | } |
1526 | ||
d352ac68 CM |
1527 | /* find the first state struct with 'bits' set after 'start', and |
1528 | * return it. tree->lock must be held. NULL will returned if | |
1529 | * nothing was found after 'start' | |
1530 | */ | |
48a3b636 ES |
1531 | static struct extent_state * |
1532 | find_first_extent_bit_state(struct extent_io_tree *tree, | |
9ee49a04 | 1533 | u64 start, unsigned bits) |
d7fc640e CM |
1534 | { |
1535 | struct rb_node *node; | |
1536 | struct extent_state *state; | |
1537 | ||
1538 | /* | |
1539 | * this search will find all the extents that end after | |
1540 | * our range starts. | |
1541 | */ | |
1542 | node = tree_search(tree, start); | |
d397712b | 1543 | if (!node) |
d7fc640e | 1544 | goto out; |
d7fc640e | 1545 | |
d397712b | 1546 | while (1) { |
d7fc640e | 1547 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1548 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1549 | return state; |
d397712b | 1550 | |
d7fc640e CM |
1551 | node = rb_next(node); |
1552 | if (!node) | |
1553 | break; | |
1554 | } | |
1555 | out: | |
1556 | return NULL; | |
1557 | } | |
d7fc640e | 1558 | |
69261c4b | 1559 | /* |
03509b78 | 1560 | * Find the first offset in the io tree with one or more @bits set. |
69261c4b | 1561 | * |
03509b78 QW |
1562 | * Note: If there are multiple bits set in @bits, any of them will match. |
1563 | * | |
1564 | * Return 0 if we find something, and update @start_ret and @end_ret. | |
1565 | * Return 1 if we found nothing. | |
69261c4b XG |
1566 | */ |
1567 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
9ee49a04 | 1568 | u64 *start_ret, u64 *end_ret, unsigned bits, |
e6138876 | 1569 | struct extent_state **cached_state) |
69261c4b XG |
1570 | { |
1571 | struct extent_state *state; | |
1572 | int ret = 1; | |
1573 | ||
1574 | spin_lock(&tree->lock); | |
e6138876 JB |
1575 | if (cached_state && *cached_state) { |
1576 | state = *cached_state; | |
27a3507d | 1577 | if (state->end == start - 1 && extent_state_in_tree(state)) { |
9688e9a9 | 1578 | while ((state = next_state(state)) != NULL) { |
e6138876 JB |
1579 | if (state->state & bits) |
1580 | goto got_it; | |
e6138876 JB |
1581 | } |
1582 | free_extent_state(*cached_state); | |
1583 | *cached_state = NULL; | |
1584 | goto out; | |
1585 | } | |
1586 | free_extent_state(*cached_state); | |
1587 | *cached_state = NULL; | |
1588 | } | |
1589 | ||
69261c4b | 1590 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1591 | got_it: |
69261c4b | 1592 | if (state) { |
e38e2ed7 | 1593 | cache_state_if_flags(state, cached_state, 0); |
69261c4b XG |
1594 | *start_ret = state->start; |
1595 | *end_ret = state->end; | |
1596 | ret = 0; | |
1597 | } | |
e6138876 | 1598 | out: |
69261c4b XG |
1599 | spin_unlock(&tree->lock); |
1600 | return ret; | |
1601 | } | |
1602 | ||
41a2ee75 JB |
1603 | /** |
1604 | * find_contiguous_extent_bit: find a contiguous area of bits | |
1605 | * @tree - io tree to check | |
1606 | * @start - offset to start the search from | |
1607 | * @start_ret - the first offset we found with the bits set | |
1608 | * @end_ret - the final contiguous range of the bits that were set | |
1609 | * @bits - bits to look for | |
1610 | * | |
1611 | * set_extent_bit and clear_extent_bit can temporarily split contiguous ranges | |
1612 | * to set bits appropriately, and then merge them again. During this time it | |
1613 | * will drop the tree->lock, so use this helper if you want to find the actual | |
1614 | * contiguous area for given bits. We will search to the first bit we find, and | |
1615 | * then walk down the tree until we find a non-contiguous area. The area | |
1616 | * returned will be the full contiguous area with the bits set. | |
1617 | */ | |
1618 | int find_contiguous_extent_bit(struct extent_io_tree *tree, u64 start, | |
1619 | u64 *start_ret, u64 *end_ret, unsigned bits) | |
1620 | { | |
1621 | struct extent_state *state; | |
1622 | int ret = 1; | |
1623 | ||
1624 | spin_lock(&tree->lock); | |
1625 | state = find_first_extent_bit_state(tree, start, bits); | |
1626 | if (state) { | |
1627 | *start_ret = state->start; | |
1628 | *end_ret = state->end; | |
1629 | while ((state = next_state(state)) != NULL) { | |
1630 | if (state->start > (*end_ret + 1)) | |
1631 | break; | |
1632 | *end_ret = state->end; | |
1633 | } | |
1634 | ret = 0; | |
1635 | } | |
1636 | spin_unlock(&tree->lock); | |
1637 | return ret; | |
1638 | } | |
1639 | ||
45bfcfc1 | 1640 | /** |
1eaebb34 NB |
1641 | * find_first_clear_extent_bit - find the first range that has @bits not set. |
1642 | * This range could start before @start. | |
45bfcfc1 NB |
1643 | * |
1644 | * @tree - the tree to search | |
1645 | * @start - the offset at/after which the found extent should start | |
1646 | * @start_ret - records the beginning of the range | |
1647 | * @end_ret - records the end of the range (inclusive) | |
1648 | * @bits - the set of bits which must be unset | |
1649 | * | |
1650 | * Since unallocated range is also considered one which doesn't have the bits | |
1651 | * set it's possible that @end_ret contains -1, this happens in case the range | |
1652 | * spans (last_range_end, end of device]. In this case it's up to the caller to | |
1653 | * trim @end_ret to the appropriate size. | |
1654 | */ | |
1655 | void find_first_clear_extent_bit(struct extent_io_tree *tree, u64 start, | |
1656 | u64 *start_ret, u64 *end_ret, unsigned bits) | |
1657 | { | |
1658 | struct extent_state *state; | |
1659 | struct rb_node *node, *prev = NULL, *next; | |
1660 | ||
1661 | spin_lock(&tree->lock); | |
1662 | ||
1663 | /* Find first extent with bits cleared */ | |
1664 | while (1) { | |
1665 | node = __etree_search(tree, start, &next, &prev, NULL, NULL); | |
5750c375 NB |
1666 | if (!node && !next && !prev) { |
1667 | /* | |
1668 | * Tree is completely empty, send full range and let | |
1669 | * caller deal with it | |
1670 | */ | |
1671 | *start_ret = 0; | |
1672 | *end_ret = -1; | |
1673 | goto out; | |
1674 | } else if (!node && !next) { | |
1675 | /* | |
1676 | * We are past the last allocated chunk, set start at | |
1677 | * the end of the last extent. | |
1678 | */ | |
1679 | state = rb_entry(prev, struct extent_state, rb_node); | |
1680 | *start_ret = state->end + 1; | |
1681 | *end_ret = -1; | |
1682 | goto out; | |
1683 | } else if (!node) { | |
45bfcfc1 | 1684 | node = next; |
45bfcfc1 | 1685 | } |
1eaebb34 NB |
1686 | /* |
1687 | * At this point 'node' either contains 'start' or start is | |
1688 | * before 'node' | |
1689 | */ | |
45bfcfc1 | 1690 | state = rb_entry(node, struct extent_state, rb_node); |
1eaebb34 NB |
1691 | |
1692 | if (in_range(start, state->start, state->end - state->start + 1)) { | |
1693 | if (state->state & bits) { | |
1694 | /* | |
1695 | * |--range with bits sets--| | |
1696 | * | | |
1697 | * start | |
1698 | */ | |
1699 | start = state->end + 1; | |
1700 | } else { | |
1701 | /* | |
1702 | * 'start' falls within a range that doesn't | |
1703 | * have the bits set, so take its start as | |
1704 | * the beginning of the desired range | |
1705 | * | |
1706 | * |--range with bits cleared----| | |
1707 | * | | |
1708 | * start | |
1709 | */ | |
1710 | *start_ret = state->start; | |
1711 | break; | |
1712 | } | |
45bfcfc1 | 1713 | } else { |
1eaebb34 NB |
1714 | /* |
1715 | * |---prev range---|---hole/unset---|---node range---| | |
1716 | * | | |
1717 | * start | |
1718 | * | |
1719 | * or | |
1720 | * | |
1721 | * |---hole/unset--||--first node--| | |
1722 | * 0 | | |
1723 | * start | |
1724 | */ | |
1725 | if (prev) { | |
1726 | state = rb_entry(prev, struct extent_state, | |
1727 | rb_node); | |
1728 | *start_ret = state->end + 1; | |
1729 | } else { | |
1730 | *start_ret = 0; | |
1731 | } | |
45bfcfc1 NB |
1732 | break; |
1733 | } | |
1734 | } | |
1735 | ||
1736 | /* | |
1737 | * Find the longest stretch from start until an entry which has the | |
1738 | * bits set | |
1739 | */ | |
1740 | while (1) { | |
1741 | state = rb_entry(node, struct extent_state, rb_node); | |
1742 | if (state->end >= start && !(state->state & bits)) { | |
1743 | *end_ret = state->end; | |
1744 | } else { | |
1745 | *end_ret = state->start - 1; | |
1746 | break; | |
1747 | } | |
1748 | ||
1749 | node = rb_next(node); | |
1750 | if (!node) | |
1751 | break; | |
1752 | } | |
1753 | out: | |
1754 | spin_unlock(&tree->lock); | |
1755 | } | |
1756 | ||
d352ac68 CM |
1757 | /* |
1758 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1759 | * more than 'max_bytes'. start and end are used to return the range, | |
1760 | * | |
3522e903 | 1761 | * true is returned if we find something, false if nothing was in the tree |
d352ac68 | 1762 | */ |
083e75e7 JB |
1763 | bool btrfs_find_delalloc_range(struct extent_io_tree *tree, u64 *start, |
1764 | u64 *end, u64 max_bytes, | |
1765 | struct extent_state **cached_state) | |
d1310b2e CM |
1766 | { |
1767 | struct rb_node *node; | |
1768 | struct extent_state *state; | |
1769 | u64 cur_start = *start; | |
3522e903 | 1770 | bool found = false; |
d1310b2e CM |
1771 | u64 total_bytes = 0; |
1772 | ||
cad321ad | 1773 | spin_lock(&tree->lock); |
c8b97818 | 1774 | |
d1310b2e CM |
1775 | /* |
1776 | * this search will find all the extents that end after | |
1777 | * our range starts. | |
1778 | */ | |
80ea96b1 | 1779 | node = tree_search(tree, cur_start); |
2b114d1d | 1780 | if (!node) { |
3522e903 | 1781 | *end = (u64)-1; |
d1310b2e CM |
1782 | goto out; |
1783 | } | |
1784 | ||
d397712b | 1785 | while (1) { |
d1310b2e | 1786 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1787 | if (found && (state->start != cur_start || |
1788 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1789 | goto out; |
1790 | } | |
1791 | if (!(state->state & EXTENT_DELALLOC)) { | |
1792 | if (!found) | |
1793 | *end = state->end; | |
1794 | goto out; | |
1795 | } | |
c2a128d2 | 1796 | if (!found) { |
d1310b2e | 1797 | *start = state->start; |
c2a128d2 | 1798 | *cached_state = state; |
b7ac31b7 | 1799 | refcount_inc(&state->refs); |
c2a128d2 | 1800 | } |
3522e903 | 1801 | found = true; |
d1310b2e CM |
1802 | *end = state->end; |
1803 | cur_start = state->end + 1; | |
1804 | node = rb_next(node); | |
d1310b2e | 1805 | total_bytes += state->end - state->start + 1; |
7bf811a5 | 1806 | if (total_bytes >= max_bytes) |
573aecaf | 1807 | break; |
573aecaf | 1808 | if (!node) |
d1310b2e CM |
1809 | break; |
1810 | } | |
1811 | out: | |
cad321ad | 1812 | spin_unlock(&tree->lock); |
d1310b2e CM |
1813 | return found; |
1814 | } | |
1815 | ||
da2c7009 LB |
1816 | static int __process_pages_contig(struct address_space *mapping, |
1817 | struct page *locked_page, | |
1818 | pgoff_t start_index, pgoff_t end_index, | |
1819 | unsigned long page_ops, pgoff_t *index_ret); | |
1820 | ||
143bede5 JM |
1821 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1822 | struct page *locked_page, | |
1823 | u64 start, u64 end) | |
c8b97818 | 1824 | { |
09cbfeaf KS |
1825 | unsigned long index = start >> PAGE_SHIFT; |
1826 | unsigned long end_index = end >> PAGE_SHIFT; | |
c8b97818 | 1827 | |
76c0021d | 1828 | ASSERT(locked_page); |
c8b97818 | 1829 | if (index == locked_page->index && end_index == index) |
143bede5 | 1830 | return; |
c8b97818 | 1831 | |
76c0021d LB |
1832 | __process_pages_contig(inode->i_mapping, locked_page, index, end_index, |
1833 | PAGE_UNLOCK, NULL); | |
c8b97818 CM |
1834 | } |
1835 | ||
1836 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1837 | struct page *locked_page, | |
1838 | u64 delalloc_start, | |
1839 | u64 delalloc_end) | |
1840 | { | |
09cbfeaf | 1841 | unsigned long index = delalloc_start >> PAGE_SHIFT; |
76c0021d | 1842 | unsigned long index_ret = index; |
09cbfeaf | 1843 | unsigned long end_index = delalloc_end >> PAGE_SHIFT; |
c8b97818 | 1844 | int ret; |
c8b97818 | 1845 | |
76c0021d | 1846 | ASSERT(locked_page); |
c8b97818 CM |
1847 | if (index == locked_page->index && index == end_index) |
1848 | return 0; | |
1849 | ||
76c0021d LB |
1850 | ret = __process_pages_contig(inode->i_mapping, locked_page, index, |
1851 | end_index, PAGE_LOCK, &index_ret); | |
1852 | if (ret == -EAGAIN) | |
1853 | __unlock_for_delalloc(inode, locked_page, delalloc_start, | |
1854 | (u64)index_ret << PAGE_SHIFT); | |
c8b97818 CM |
1855 | return ret; |
1856 | } | |
1857 | ||
1858 | /* | |
3522e903 LF |
1859 | * Find and lock a contiguous range of bytes in the file marked as delalloc, no |
1860 | * more than @max_bytes. @Start and @end are used to return the range, | |
c8b97818 | 1861 | * |
3522e903 LF |
1862 | * Return: true if we find something |
1863 | * false if nothing was in the tree | |
c8b97818 | 1864 | */ |
ce9f967f | 1865 | EXPORT_FOR_TESTS |
3522e903 | 1866 | noinline_for_stack bool find_lock_delalloc_range(struct inode *inode, |
294e30fe | 1867 | struct page *locked_page, u64 *start, |
917aacec | 1868 | u64 *end) |
c8b97818 | 1869 | { |
9978059b | 1870 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
917aacec | 1871 | u64 max_bytes = BTRFS_MAX_EXTENT_SIZE; |
c8b97818 CM |
1872 | u64 delalloc_start; |
1873 | u64 delalloc_end; | |
3522e903 | 1874 | bool found; |
9655d298 | 1875 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1876 | int ret; |
1877 | int loops = 0; | |
1878 | ||
1879 | again: | |
1880 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1881 | delalloc_start = *start; | |
1882 | delalloc_end = 0; | |
083e75e7 JB |
1883 | found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end, |
1884 | max_bytes, &cached_state); | |
70b99e69 | 1885 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1886 | *start = delalloc_start; |
1887 | *end = delalloc_end; | |
c2a128d2 | 1888 | free_extent_state(cached_state); |
3522e903 | 1889 | return false; |
c8b97818 CM |
1890 | } |
1891 | ||
70b99e69 CM |
1892 | /* |
1893 | * start comes from the offset of locked_page. We have to lock | |
1894 | * pages in order, so we can't process delalloc bytes before | |
1895 | * locked_page | |
1896 | */ | |
d397712b | 1897 | if (delalloc_start < *start) |
70b99e69 | 1898 | delalloc_start = *start; |
70b99e69 | 1899 | |
c8b97818 CM |
1900 | /* |
1901 | * make sure to limit the number of pages we try to lock down | |
c8b97818 | 1902 | */ |
7bf811a5 JB |
1903 | if (delalloc_end + 1 - delalloc_start > max_bytes) |
1904 | delalloc_end = delalloc_start + max_bytes - 1; | |
d397712b | 1905 | |
c8b97818 CM |
1906 | /* step two, lock all the pages after the page that has start */ |
1907 | ret = lock_delalloc_pages(inode, locked_page, | |
1908 | delalloc_start, delalloc_end); | |
9bfd61d9 | 1909 | ASSERT(!ret || ret == -EAGAIN); |
c8b97818 CM |
1910 | if (ret == -EAGAIN) { |
1911 | /* some of the pages are gone, lets avoid looping by | |
1912 | * shortening the size of the delalloc range we're searching | |
1913 | */ | |
9655d298 | 1914 | free_extent_state(cached_state); |
7d788742 | 1915 | cached_state = NULL; |
c8b97818 | 1916 | if (!loops) { |
09cbfeaf | 1917 | max_bytes = PAGE_SIZE; |
c8b97818 CM |
1918 | loops = 1; |
1919 | goto again; | |
1920 | } else { | |
3522e903 | 1921 | found = false; |
c8b97818 CM |
1922 | goto out_failed; |
1923 | } | |
1924 | } | |
c8b97818 CM |
1925 | |
1926 | /* step three, lock the state bits for the whole range */ | |
ff13db41 | 1927 | lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state); |
c8b97818 CM |
1928 | |
1929 | /* then test to make sure it is all still delalloc */ | |
1930 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1931 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1932 | if (!ret) { |
9655d298 | 1933 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
e43bbe5e | 1934 | &cached_state); |
c8b97818 CM |
1935 | __unlock_for_delalloc(inode, locked_page, |
1936 | delalloc_start, delalloc_end); | |
1937 | cond_resched(); | |
1938 | goto again; | |
1939 | } | |
9655d298 | 1940 | free_extent_state(cached_state); |
c8b97818 CM |
1941 | *start = delalloc_start; |
1942 | *end = delalloc_end; | |
1943 | out_failed: | |
1944 | return found; | |
1945 | } | |
1946 | ||
da2c7009 LB |
1947 | static int __process_pages_contig(struct address_space *mapping, |
1948 | struct page *locked_page, | |
1949 | pgoff_t start_index, pgoff_t end_index, | |
1950 | unsigned long page_ops, pgoff_t *index_ret) | |
c8b97818 | 1951 | { |
873695b3 | 1952 | unsigned long nr_pages = end_index - start_index + 1; |
12e3360f | 1953 | unsigned long pages_processed = 0; |
873695b3 | 1954 | pgoff_t index = start_index; |
c8b97818 | 1955 | struct page *pages[16]; |
873695b3 | 1956 | unsigned ret; |
da2c7009 | 1957 | int err = 0; |
c8b97818 | 1958 | int i; |
771ed689 | 1959 | |
da2c7009 LB |
1960 | if (page_ops & PAGE_LOCK) { |
1961 | ASSERT(page_ops == PAGE_LOCK); | |
1962 | ASSERT(index_ret && *index_ret == start_index); | |
1963 | } | |
1964 | ||
704de49d | 1965 | if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0) |
873695b3 | 1966 | mapping_set_error(mapping, -EIO); |
704de49d | 1967 | |
d397712b | 1968 | while (nr_pages > 0) { |
873695b3 | 1969 | ret = find_get_pages_contig(mapping, index, |
5b050f04 CM |
1970 | min_t(unsigned long, |
1971 | nr_pages, ARRAY_SIZE(pages)), pages); | |
da2c7009 LB |
1972 | if (ret == 0) { |
1973 | /* | |
1974 | * Only if we're going to lock these pages, | |
1975 | * can we find nothing at @index. | |
1976 | */ | |
1977 | ASSERT(page_ops & PAGE_LOCK); | |
49d4a334 LB |
1978 | err = -EAGAIN; |
1979 | goto out; | |
da2c7009 | 1980 | } |
8b62b72b | 1981 | |
da2c7009 | 1982 | for (i = 0; i < ret; i++) { |
c2790a2e | 1983 | if (page_ops & PAGE_SET_PRIVATE2) |
8b62b72b CM |
1984 | SetPagePrivate2(pages[i]); |
1985 | ||
1d53c9e6 | 1986 | if (locked_page && pages[i] == locked_page) { |
09cbfeaf | 1987 | put_page(pages[i]); |
12e3360f | 1988 | pages_processed++; |
c8b97818 CM |
1989 | continue; |
1990 | } | |
c2790a2e | 1991 | if (page_ops & PAGE_CLEAR_DIRTY) |
c8b97818 | 1992 | clear_page_dirty_for_io(pages[i]); |
c2790a2e | 1993 | if (page_ops & PAGE_SET_WRITEBACK) |
c8b97818 | 1994 | set_page_writeback(pages[i]); |
704de49d FM |
1995 | if (page_ops & PAGE_SET_ERROR) |
1996 | SetPageError(pages[i]); | |
c2790a2e | 1997 | if (page_ops & PAGE_END_WRITEBACK) |
c8b97818 | 1998 | end_page_writeback(pages[i]); |
c2790a2e | 1999 | if (page_ops & PAGE_UNLOCK) |
771ed689 | 2000 | unlock_page(pages[i]); |
da2c7009 LB |
2001 | if (page_ops & PAGE_LOCK) { |
2002 | lock_page(pages[i]); | |
2003 | if (!PageDirty(pages[i]) || | |
2004 | pages[i]->mapping != mapping) { | |
2005 | unlock_page(pages[i]); | |
5909ca11 RK |
2006 | for (; i < ret; i++) |
2007 | put_page(pages[i]); | |
da2c7009 LB |
2008 | err = -EAGAIN; |
2009 | goto out; | |
2010 | } | |
2011 | } | |
09cbfeaf | 2012 | put_page(pages[i]); |
12e3360f | 2013 | pages_processed++; |
c8b97818 CM |
2014 | } |
2015 | nr_pages -= ret; | |
2016 | index += ret; | |
2017 | cond_resched(); | |
2018 | } | |
da2c7009 LB |
2019 | out: |
2020 | if (err && index_ret) | |
12e3360f | 2021 | *index_ret = start_index + pages_processed - 1; |
da2c7009 | 2022 | return err; |
c8b97818 | 2023 | } |
c8b97818 | 2024 | |
ad7ff17b | 2025 | void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end, |
74e9194a NB |
2026 | struct page *locked_page, |
2027 | unsigned clear_bits, | |
2028 | unsigned long page_ops) | |
873695b3 | 2029 | { |
ad7ff17b | 2030 | clear_extent_bit(&inode->io_tree, start, end, clear_bits, 1, 0, NULL); |
873695b3 | 2031 | |
ad7ff17b | 2032 | __process_pages_contig(inode->vfs_inode.i_mapping, locked_page, |
873695b3 | 2033 | start >> PAGE_SHIFT, end >> PAGE_SHIFT, |
da2c7009 | 2034 | page_ops, NULL); |
873695b3 LB |
2035 | } |
2036 | ||
d352ac68 CM |
2037 | /* |
2038 | * count the number of bytes in the tree that have a given bit(s) | |
2039 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
2040 | * cached. The total number found is returned. | |
2041 | */ | |
d1310b2e CM |
2042 | u64 count_range_bits(struct extent_io_tree *tree, |
2043 | u64 *start, u64 search_end, u64 max_bytes, | |
9ee49a04 | 2044 | unsigned bits, int contig) |
d1310b2e CM |
2045 | { |
2046 | struct rb_node *node; | |
2047 | struct extent_state *state; | |
2048 | u64 cur_start = *start; | |
2049 | u64 total_bytes = 0; | |
ec29ed5b | 2050 | u64 last = 0; |
d1310b2e CM |
2051 | int found = 0; |
2052 | ||
fae7f21c | 2053 | if (WARN_ON(search_end <= cur_start)) |
d1310b2e | 2054 | return 0; |
d1310b2e | 2055 | |
cad321ad | 2056 | spin_lock(&tree->lock); |
d1310b2e CM |
2057 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
2058 | total_bytes = tree->dirty_bytes; | |
2059 | goto out; | |
2060 | } | |
2061 | /* | |
2062 | * this search will find all the extents that end after | |
2063 | * our range starts. | |
2064 | */ | |
80ea96b1 | 2065 | node = tree_search(tree, cur_start); |
d397712b | 2066 | if (!node) |
d1310b2e | 2067 | goto out; |
d1310b2e | 2068 | |
d397712b | 2069 | while (1) { |
d1310b2e CM |
2070 | state = rb_entry(node, struct extent_state, rb_node); |
2071 | if (state->start > search_end) | |
2072 | break; | |
ec29ed5b CM |
2073 | if (contig && found && state->start > last + 1) |
2074 | break; | |
2075 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
2076 | total_bytes += min(search_end, state->end) + 1 - |
2077 | max(cur_start, state->start); | |
2078 | if (total_bytes >= max_bytes) | |
2079 | break; | |
2080 | if (!found) { | |
af60bed2 | 2081 | *start = max(cur_start, state->start); |
d1310b2e CM |
2082 | found = 1; |
2083 | } | |
ec29ed5b CM |
2084 | last = state->end; |
2085 | } else if (contig && found) { | |
2086 | break; | |
d1310b2e CM |
2087 | } |
2088 | node = rb_next(node); | |
2089 | if (!node) | |
2090 | break; | |
2091 | } | |
2092 | out: | |
cad321ad | 2093 | spin_unlock(&tree->lock); |
d1310b2e CM |
2094 | return total_bytes; |
2095 | } | |
b2950863 | 2096 | |
d352ac68 CM |
2097 | /* |
2098 | * set the private field for a given byte offset in the tree. If there isn't | |
2099 | * an extent_state there already, this does nothing. | |
2100 | */ | |
b3f167aa JB |
2101 | int set_state_failrec(struct extent_io_tree *tree, u64 start, |
2102 | struct io_failure_record *failrec) | |
d1310b2e CM |
2103 | { |
2104 | struct rb_node *node; | |
2105 | struct extent_state *state; | |
2106 | int ret = 0; | |
2107 | ||
cad321ad | 2108 | spin_lock(&tree->lock); |
d1310b2e CM |
2109 | /* |
2110 | * this search will find all the extents that end after | |
2111 | * our range starts. | |
2112 | */ | |
80ea96b1 | 2113 | node = tree_search(tree, start); |
2b114d1d | 2114 | if (!node) { |
d1310b2e CM |
2115 | ret = -ENOENT; |
2116 | goto out; | |
2117 | } | |
2118 | state = rb_entry(node, struct extent_state, rb_node); | |
2119 | if (state->start != start) { | |
2120 | ret = -ENOENT; | |
2121 | goto out; | |
2122 | } | |
47dc196a | 2123 | state->failrec = failrec; |
d1310b2e | 2124 | out: |
cad321ad | 2125 | spin_unlock(&tree->lock); |
d1310b2e CM |
2126 | return ret; |
2127 | } | |
2128 | ||
2279a270 | 2129 | struct io_failure_record *get_state_failrec(struct extent_io_tree *tree, u64 start) |
d1310b2e CM |
2130 | { |
2131 | struct rb_node *node; | |
2132 | struct extent_state *state; | |
2279a270 | 2133 | struct io_failure_record *failrec; |
d1310b2e | 2134 | |
cad321ad | 2135 | spin_lock(&tree->lock); |
d1310b2e CM |
2136 | /* |
2137 | * this search will find all the extents that end after | |
2138 | * our range starts. | |
2139 | */ | |
80ea96b1 | 2140 | node = tree_search(tree, start); |
2b114d1d | 2141 | if (!node) { |
2279a270 | 2142 | failrec = ERR_PTR(-ENOENT); |
d1310b2e CM |
2143 | goto out; |
2144 | } | |
2145 | state = rb_entry(node, struct extent_state, rb_node); | |
2146 | if (state->start != start) { | |
2279a270 | 2147 | failrec = ERR_PTR(-ENOENT); |
d1310b2e CM |
2148 | goto out; |
2149 | } | |
2279a270 NB |
2150 | |
2151 | failrec = state->failrec; | |
d1310b2e | 2152 | out: |
cad321ad | 2153 | spin_unlock(&tree->lock); |
2279a270 | 2154 | return failrec; |
d1310b2e CM |
2155 | } |
2156 | ||
2157 | /* | |
2158 | * searches a range in the state tree for a given mask. | |
70dec807 | 2159 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
2160 | * has the bits set. Otherwise, 1 is returned if any bit in the |
2161 | * range is found set. | |
2162 | */ | |
2163 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 2164 | unsigned bits, int filled, struct extent_state *cached) |
d1310b2e CM |
2165 | { |
2166 | struct extent_state *state = NULL; | |
2167 | struct rb_node *node; | |
2168 | int bitset = 0; | |
d1310b2e | 2169 | |
cad321ad | 2170 | spin_lock(&tree->lock); |
27a3507d | 2171 | if (cached && extent_state_in_tree(cached) && cached->start <= start && |
df98b6e2 | 2172 | cached->end > start) |
9655d298 CM |
2173 | node = &cached->rb_node; |
2174 | else | |
2175 | node = tree_search(tree, start); | |
d1310b2e CM |
2176 | while (node && start <= end) { |
2177 | state = rb_entry(node, struct extent_state, rb_node); | |
2178 | ||
2179 | if (filled && state->start > start) { | |
2180 | bitset = 0; | |
2181 | break; | |
2182 | } | |
2183 | ||
2184 | if (state->start > end) | |
2185 | break; | |
2186 | ||
2187 | if (state->state & bits) { | |
2188 | bitset = 1; | |
2189 | if (!filled) | |
2190 | break; | |
2191 | } else if (filled) { | |
2192 | bitset = 0; | |
2193 | break; | |
2194 | } | |
46562cec CM |
2195 | |
2196 | if (state->end == (u64)-1) | |
2197 | break; | |
2198 | ||
d1310b2e CM |
2199 | start = state->end + 1; |
2200 | if (start > end) | |
2201 | break; | |
2202 | node = rb_next(node); | |
2203 | if (!node) { | |
2204 | if (filled) | |
2205 | bitset = 0; | |
2206 | break; | |
2207 | } | |
2208 | } | |
cad321ad | 2209 | spin_unlock(&tree->lock); |
d1310b2e CM |
2210 | return bitset; |
2211 | } | |
d1310b2e CM |
2212 | |
2213 | /* | |
2214 | * helper function to set a given page up to date if all the | |
2215 | * extents in the tree for that page are up to date | |
2216 | */ | |
143bede5 | 2217 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e | 2218 | { |
4eee4fa4 | 2219 | u64 start = page_offset(page); |
09cbfeaf | 2220 | u64 end = start + PAGE_SIZE - 1; |
9655d298 | 2221 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 2222 | SetPageUptodate(page); |
d1310b2e CM |
2223 | } |
2224 | ||
7870d082 JB |
2225 | int free_io_failure(struct extent_io_tree *failure_tree, |
2226 | struct extent_io_tree *io_tree, | |
2227 | struct io_failure_record *rec) | |
4a54c8c1 JS |
2228 | { |
2229 | int ret; | |
2230 | int err = 0; | |
4a54c8c1 | 2231 | |
47dc196a | 2232 | set_state_failrec(failure_tree, rec->start, NULL); |
4a54c8c1 JS |
2233 | ret = clear_extent_bits(failure_tree, rec->start, |
2234 | rec->start + rec->len - 1, | |
91166212 | 2235 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 JS |
2236 | if (ret) |
2237 | err = ret; | |
2238 | ||
7870d082 | 2239 | ret = clear_extent_bits(io_tree, rec->start, |
53b381b3 | 2240 | rec->start + rec->len - 1, |
91166212 | 2241 | EXTENT_DAMAGED); |
53b381b3 DW |
2242 | if (ret && !err) |
2243 | err = ret; | |
4a54c8c1 JS |
2244 | |
2245 | kfree(rec); | |
2246 | return err; | |
2247 | } | |
2248 | ||
4a54c8c1 JS |
2249 | /* |
2250 | * this bypasses the standard btrfs submit functions deliberately, as | |
2251 | * the standard behavior is to write all copies in a raid setup. here we only | |
2252 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
2253 | * submit_bio directly. | |
3ec706c8 | 2254 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
2255 | * actually prevents the read that triggered the error from finishing. |
2256 | * currently, there can be no more than two copies of every data bit. thus, | |
2257 | * exactly one rewrite is required. | |
2258 | */ | |
6ec656bc JB |
2259 | int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start, |
2260 | u64 length, u64 logical, struct page *page, | |
2261 | unsigned int pg_offset, int mirror_num) | |
4a54c8c1 JS |
2262 | { |
2263 | struct bio *bio; | |
2264 | struct btrfs_device *dev; | |
4a54c8c1 JS |
2265 | u64 map_length = 0; |
2266 | u64 sector; | |
2267 | struct btrfs_bio *bbio = NULL; | |
2268 | int ret; | |
2269 | ||
1751e8a6 | 2270 | ASSERT(!(fs_info->sb->s_flags & SB_RDONLY)); |
4a54c8c1 JS |
2271 | BUG_ON(!mirror_num); |
2272 | ||
c5e4c3d7 | 2273 | bio = btrfs_io_bio_alloc(1); |
4f024f37 | 2274 | bio->bi_iter.bi_size = 0; |
4a54c8c1 JS |
2275 | map_length = length; |
2276 | ||
b5de8d0d FM |
2277 | /* |
2278 | * Avoid races with device replace and make sure our bbio has devices | |
2279 | * associated to its stripes that don't go away while we are doing the | |
2280 | * read repair operation. | |
2281 | */ | |
2282 | btrfs_bio_counter_inc_blocked(fs_info); | |
e4ff5fb5 | 2283 | if (btrfs_is_parity_mirror(fs_info, logical, length)) { |
c725328c LB |
2284 | /* |
2285 | * Note that we don't use BTRFS_MAP_WRITE because it's supposed | |
2286 | * to update all raid stripes, but here we just want to correct | |
2287 | * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad | |
2288 | * stripe's dev and sector. | |
2289 | */ | |
2290 | ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical, | |
2291 | &map_length, &bbio, 0); | |
2292 | if (ret) { | |
2293 | btrfs_bio_counter_dec(fs_info); | |
2294 | bio_put(bio); | |
2295 | return -EIO; | |
2296 | } | |
2297 | ASSERT(bbio->mirror_num == 1); | |
2298 | } else { | |
2299 | ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, | |
2300 | &map_length, &bbio, mirror_num); | |
2301 | if (ret) { | |
2302 | btrfs_bio_counter_dec(fs_info); | |
2303 | bio_put(bio); | |
2304 | return -EIO; | |
2305 | } | |
2306 | BUG_ON(mirror_num != bbio->mirror_num); | |
4a54c8c1 | 2307 | } |
c725328c LB |
2308 | |
2309 | sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9; | |
4f024f37 | 2310 | bio->bi_iter.bi_sector = sector; |
c725328c | 2311 | dev = bbio->stripes[bbio->mirror_num - 1].dev; |
6e9606d2 | 2312 | btrfs_put_bbio(bbio); |
ebbede42 AJ |
2313 | if (!dev || !dev->bdev || |
2314 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) { | |
b5de8d0d | 2315 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2316 | bio_put(bio); |
2317 | return -EIO; | |
2318 | } | |
74d46992 | 2319 | bio_set_dev(bio, dev->bdev); |
70fd7614 | 2320 | bio->bi_opf = REQ_OP_WRITE | REQ_SYNC; |
ffdd2018 | 2321 | bio_add_page(bio, page, length, pg_offset); |
4a54c8c1 | 2322 | |
4e49ea4a | 2323 | if (btrfsic_submit_bio_wait(bio)) { |
4a54c8c1 | 2324 | /* try to remap that extent elsewhere? */ |
b5de8d0d | 2325 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 | 2326 | bio_put(bio); |
442a4f63 | 2327 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
2328 | return -EIO; |
2329 | } | |
2330 | ||
b14af3b4 DS |
2331 | btrfs_info_rl_in_rcu(fs_info, |
2332 | "read error corrected: ino %llu off %llu (dev %s sector %llu)", | |
6ec656bc | 2333 | ino, start, |
1203b681 | 2334 | rcu_str_deref(dev->name), sector); |
b5de8d0d | 2335 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2336 | bio_put(bio); |
2337 | return 0; | |
2338 | } | |
2339 | ||
2b48966a | 2340 | int btrfs_repair_eb_io_failure(const struct extent_buffer *eb, int mirror_num) |
ea466794 | 2341 | { |
20a1fbf9 | 2342 | struct btrfs_fs_info *fs_info = eb->fs_info; |
ea466794 | 2343 | u64 start = eb->start; |
cc5e31a4 | 2344 | int i, num_pages = num_extent_pages(eb); |
d95603b2 | 2345 | int ret = 0; |
ea466794 | 2346 | |
bc98a42c | 2347 | if (sb_rdonly(fs_info->sb)) |
908960c6 ID |
2348 | return -EROFS; |
2349 | ||
ea466794 | 2350 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 2351 | struct page *p = eb->pages[i]; |
1203b681 | 2352 | |
6ec656bc | 2353 | ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p, |
1203b681 | 2354 | start - page_offset(p), mirror_num); |
ea466794 JB |
2355 | if (ret) |
2356 | break; | |
09cbfeaf | 2357 | start += PAGE_SIZE; |
ea466794 JB |
2358 | } |
2359 | ||
2360 | return ret; | |
2361 | } | |
2362 | ||
4a54c8c1 JS |
2363 | /* |
2364 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2365 | * to see if we need to process or clean up an io_failure_record | |
2366 | */ | |
7870d082 JB |
2367 | int clean_io_failure(struct btrfs_fs_info *fs_info, |
2368 | struct extent_io_tree *failure_tree, | |
2369 | struct extent_io_tree *io_tree, u64 start, | |
2370 | struct page *page, u64 ino, unsigned int pg_offset) | |
4a54c8c1 JS |
2371 | { |
2372 | u64 private; | |
4a54c8c1 | 2373 | struct io_failure_record *failrec; |
4a54c8c1 JS |
2374 | struct extent_state *state; |
2375 | int num_copies; | |
4a54c8c1 | 2376 | int ret; |
4a54c8c1 JS |
2377 | |
2378 | private = 0; | |
7870d082 JB |
2379 | ret = count_range_bits(failure_tree, &private, (u64)-1, 1, |
2380 | EXTENT_DIRTY, 0); | |
4a54c8c1 JS |
2381 | if (!ret) |
2382 | return 0; | |
2383 | ||
2279a270 NB |
2384 | failrec = get_state_failrec(failure_tree, start); |
2385 | if (IS_ERR(failrec)) | |
4a54c8c1 JS |
2386 | return 0; |
2387 | ||
4a54c8c1 JS |
2388 | BUG_ON(!failrec->this_mirror); |
2389 | ||
2390 | if (failrec->in_validation) { | |
2391 | /* there was no real error, just free the record */ | |
ab8d0fc4 JM |
2392 | btrfs_debug(fs_info, |
2393 | "clean_io_failure: freeing dummy error at %llu", | |
2394 | failrec->start); | |
4a54c8c1 JS |
2395 | goto out; |
2396 | } | |
bc98a42c | 2397 | if (sb_rdonly(fs_info->sb)) |
908960c6 | 2398 | goto out; |
4a54c8c1 | 2399 | |
7870d082 JB |
2400 | spin_lock(&io_tree->lock); |
2401 | state = find_first_extent_bit_state(io_tree, | |
4a54c8c1 JS |
2402 | failrec->start, |
2403 | EXTENT_LOCKED); | |
7870d082 | 2404 | spin_unlock(&io_tree->lock); |
4a54c8c1 | 2405 | |
883d0de4 MX |
2406 | if (state && state->start <= failrec->start && |
2407 | state->end >= failrec->start + failrec->len - 1) { | |
3ec706c8 SB |
2408 | num_copies = btrfs_num_copies(fs_info, failrec->logical, |
2409 | failrec->len); | |
4a54c8c1 | 2410 | if (num_copies > 1) { |
7870d082 JB |
2411 | repair_io_failure(fs_info, ino, start, failrec->len, |
2412 | failrec->logical, page, pg_offset, | |
2413 | failrec->failed_mirror); | |
4a54c8c1 JS |
2414 | } |
2415 | } | |
2416 | ||
2417 | out: | |
7870d082 | 2418 | free_io_failure(failure_tree, io_tree, failrec); |
4a54c8c1 | 2419 | |
454ff3de | 2420 | return 0; |
4a54c8c1 JS |
2421 | } |
2422 | ||
f612496b MX |
2423 | /* |
2424 | * Can be called when | |
2425 | * - hold extent lock | |
2426 | * - under ordered extent | |
2427 | * - the inode is freeing | |
2428 | */ | |
7ab7956e | 2429 | void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end) |
f612496b | 2430 | { |
7ab7956e | 2431 | struct extent_io_tree *failure_tree = &inode->io_failure_tree; |
f612496b MX |
2432 | struct io_failure_record *failrec; |
2433 | struct extent_state *state, *next; | |
2434 | ||
2435 | if (RB_EMPTY_ROOT(&failure_tree->state)) | |
2436 | return; | |
2437 | ||
2438 | spin_lock(&failure_tree->lock); | |
2439 | state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY); | |
2440 | while (state) { | |
2441 | if (state->start > end) | |
2442 | break; | |
2443 | ||
2444 | ASSERT(state->end <= end); | |
2445 | ||
2446 | next = next_state(state); | |
2447 | ||
47dc196a | 2448 | failrec = state->failrec; |
f612496b MX |
2449 | free_extent_state(state); |
2450 | kfree(failrec); | |
2451 | ||
2452 | state = next; | |
2453 | } | |
2454 | spin_unlock(&failure_tree->lock); | |
2455 | } | |
2456 | ||
3526302f NB |
2457 | static struct io_failure_record *btrfs_get_io_failure_record(struct inode *inode, |
2458 | u64 start, u64 end) | |
4a54c8c1 | 2459 | { |
ab8d0fc4 | 2460 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e | 2461 | struct io_failure_record *failrec; |
4a54c8c1 | 2462 | struct extent_map *em; |
4a54c8c1 JS |
2463 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2464 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2465 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
4a54c8c1 | 2466 | int ret; |
4a54c8c1 JS |
2467 | u64 logical; |
2468 | ||
2279a270 | 2469 | failrec = get_state_failrec(failure_tree, start); |
3526302f | 2470 | if (!IS_ERR(failrec)) { |
ab8d0fc4 JM |
2471 | btrfs_debug(fs_info, |
2472 | "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d", | |
2473 | failrec->logical, failrec->start, failrec->len, | |
2474 | failrec->in_validation); | |
4a54c8c1 JS |
2475 | /* |
2476 | * when data can be on disk more than twice, add to failrec here | |
2477 | * (e.g. with a list for failed_mirror) to make | |
2478 | * clean_io_failure() clean all those errors at once. | |
2479 | */ | |
3526302f NB |
2480 | |
2481 | return failrec; | |
4a54c8c1 | 2482 | } |
2fe6303e | 2483 | |
3526302f NB |
2484 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); |
2485 | if (!failrec) | |
2486 | return ERR_PTR(-ENOMEM); | |
2fe6303e | 2487 | |
3526302f NB |
2488 | failrec->start = start; |
2489 | failrec->len = end - start + 1; | |
2490 | failrec->this_mirror = 0; | |
2491 | failrec->bio_flags = 0; | |
2492 | failrec->in_validation = 0; | |
2493 | ||
2494 | read_lock(&em_tree->lock); | |
2495 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2496 | if (!em) { | |
2497 | read_unlock(&em_tree->lock); | |
2498 | kfree(failrec); | |
2499 | return ERR_PTR(-EIO); | |
2500 | } | |
2501 | ||
2502 | if (em->start > start || em->start + em->len <= start) { | |
2503 | free_extent_map(em); | |
2504 | em = NULL; | |
2505 | } | |
2506 | read_unlock(&em_tree->lock); | |
2507 | if (!em) { | |
2508 | kfree(failrec); | |
2509 | return ERR_PTR(-EIO); | |
2510 | } | |
2511 | ||
2512 | logical = start - em->start; | |
2513 | logical = em->block_start + logical; | |
2514 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2515 | logical = em->block_start; | |
2516 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2517 | extent_set_compress_type(&failrec->bio_flags, em->compress_type); | |
2518 | } | |
2519 | ||
2520 | btrfs_debug(fs_info, | |
2521 | "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu", | |
2522 | logical, start, failrec->len); | |
2523 | ||
2524 | failrec->logical = logical; | |
2525 | free_extent_map(em); | |
2526 | ||
2527 | /* Set the bits in the private failure tree */ | |
2528 | ret = set_extent_bits(failure_tree, start, end, | |
2529 | EXTENT_LOCKED | EXTENT_DIRTY); | |
2530 | if (ret >= 0) { | |
2531 | ret = set_state_failrec(failure_tree, start, failrec); | |
2532 | /* Set the bits in the inode's tree */ | |
2533 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED); | |
2534 | } else if (ret < 0) { | |
2535 | kfree(failrec); | |
2536 | return ERR_PTR(ret); | |
2537 | } | |
2538 | ||
2539 | return failrec; | |
2fe6303e MX |
2540 | } |
2541 | ||
ce06d3ec OS |
2542 | static bool btrfs_check_repairable(struct inode *inode, bool needs_validation, |
2543 | struct io_failure_record *failrec, | |
2544 | int failed_mirror) | |
2fe6303e | 2545 | { |
ab8d0fc4 | 2546 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e MX |
2547 | int num_copies; |
2548 | ||
ab8d0fc4 | 2549 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
4a54c8c1 JS |
2550 | if (num_copies == 1) { |
2551 | /* | |
2552 | * we only have a single copy of the data, so don't bother with | |
2553 | * all the retry and error correction code that follows. no | |
2554 | * matter what the error is, it is very likely to persist. | |
2555 | */ | |
ab8d0fc4 JM |
2556 | btrfs_debug(fs_info, |
2557 | "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
2558 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2559 | return false; |
4a54c8c1 JS |
2560 | } |
2561 | ||
4a54c8c1 JS |
2562 | /* |
2563 | * there are two premises: | |
2564 | * a) deliver good data to the caller | |
2565 | * b) correct the bad sectors on disk | |
2566 | */ | |
c7333972 | 2567 | if (needs_validation) { |
4a54c8c1 JS |
2568 | /* |
2569 | * to fulfill b), we need to know the exact failing sectors, as | |
2570 | * we don't want to rewrite any more than the failed ones. thus, | |
2571 | * we need separate read requests for the failed bio | |
2572 | * | |
2573 | * if the following BUG_ON triggers, our validation request got | |
2574 | * merged. we need separate requests for our algorithm to work. | |
2575 | */ | |
2576 | BUG_ON(failrec->in_validation); | |
2577 | failrec->in_validation = 1; | |
2578 | failrec->this_mirror = failed_mirror; | |
4a54c8c1 JS |
2579 | } else { |
2580 | /* | |
2581 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2582 | * of the failed sector and if we succeed, we have setup | |
2583 | * everything for repair_io_failure to do the rest for us. | |
2584 | */ | |
2585 | if (failrec->in_validation) { | |
2586 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2587 | failrec->in_validation = 0; | |
2588 | failrec->this_mirror = 0; | |
2589 | } | |
2590 | failrec->failed_mirror = failed_mirror; | |
2591 | failrec->this_mirror++; | |
2592 | if (failrec->this_mirror == failed_mirror) | |
2593 | failrec->this_mirror++; | |
4a54c8c1 JS |
2594 | } |
2595 | ||
facc8a22 | 2596 | if (failrec->this_mirror > num_copies) { |
ab8d0fc4 JM |
2597 | btrfs_debug(fs_info, |
2598 | "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
2599 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2600 | return false; |
4a54c8c1 JS |
2601 | } |
2602 | ||
c3cfb656 | 2603 | return true; |
2fe6303e MX |
2604 | } |
2605 | ||
c7333972 | 2606 | static bool btrfs_io_needs_validation(struct inode *inode, struct bio *bio) |
2fe6303e | 2607 | { |
c7333972 | 2608 | u64 len = 0; |
77d5d689 | 2609 | const u32 blocksize = inode->i_sb->s_blocksize; |
2fe6303e | 2610 | |
f337bd74 OS |
2611 | /* |
2612 | * If bi_status is BLK_STS_OK, then this was a checksum error, not an | |
2613 | * I/O error. In this case, we already know exactly which sector was | |
2614 | * bad, so we don't need to validate. | |
2615 | */ | |
2616 | if (bio->bi_status == BLK_STS_OK) | |
2617 | return false; | |
4a54c8c1 | 2618 | |
c7333972 OS |
2619 | /* |
2620 | * We need to validate each sector individually if the failed I/O was | |
2621 | * for multiple sectors. | |
77d5d689 OS |
2622 | * |
2623 | * There are a few possible bios that can end up here: | |
2624 | * 1. A buffered read bio, which is not cloned. | |
2625 | * 2. A direct I/O read bio, which is cloned. | |
2626 | * 3. A (buffered or direct) repair bio, which is not cloned. | |
2627 | * | |
2628 | * For cloned bios (case 2), we can get the size from | |
2629 | * btrfs_io_bio->iter; for non-cloned bios (cases 1 and 3), we can get | |
2630 | * it from the bvecs. | |
c7333972 | 2631 | */ |
77d5d689 OS |
2632 | if (bio_flagged(bio, BIO_CLONED)) { |
2633 | if (btrfs_io_bio(bio)->iter.bi_size > blocksize) | |
c7333972 | 2634 | return true; |
77d5d689 OS |
2635 | } else { |
2636 | struct bio_vec *bvec; | |
2637 | int i; | |
facc8a22 | 2638 | |
77d5d689 OS |
2639 | bio_for_each_bvec_all(bvec, bio, i) { |
2640 | len += bvec->bv_len; | |
2641 | if (len > blocksize) | |
2642 | return true; | |
2643 | } | |
facc8a22 | 2644 | } |
c7333972 | 2645 | return false; |
2fe6303e MX |
2646 | } |
2647 | ||
77d5d689 OS |
2648 | blk_status_t btrfs_submit_read_repair(struct inode *inode, |
2649 | struct bio *failed_bio, u64 phy_offset, | |
2650 | struct page *page, unsigned int pgoff, | |
2651 | u64 start, u64 end, int failed_mirror, | |
2652 | submit_bio_hook_t *submit_bio_hook) | |
2fe6303e MX |
2653 | { |
2654 | struct io_failure_record *failrec; | |
77d5d689 | 2655 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e | 2656 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
7870d082 | 2657 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
77d5d689 | 2658 | struct btrfs_io_bio *failed_io_bio = btrfs_io_bio(failed_bio); |
265fdfa6 | 2659 | const int icsum = phy_offset >> fs_info->sectorsize_bits; |
c7333972 | 2660 | bool need_validation; |
77d5d689 OS |
2661 | struct bio *repair_bio; |
2662 | struct btrfs_io_bio *repair_io_bio; | |
4e4cbee9 | 2663 | blk_status_t status; |
2fe6303e | 2664 | |
77d5d689 OS |
2665 | btrfs_debug(fs_info, |
2666 | "repair read error: read error at %llu", start); | |
2fe6303e | 2667 | |
1f7ad75b | 2668 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
2fe6303e | 2669 | |
3526302f NB |
2670 | failrec = btrfs_get_io_failure_record(inode, start, end); |
2671 | if (IS_ERR(failrec)) | |
2672 | return errno_to_blk_status(PTR_ERR(failrec)); | |
2fe6303e | 2673 | |
c7333972 | 2674 | need_validation = btrfs_io_needs_validation(inode, failed_bio); |
2fe6303e | 2675 | |
c7333972 | 2676 | if (!btrfs_check_repairable(inode, need_validation, failrec, |
c3cfb656 | 2677 | failed_mirror)) { |
7870d082 | 2678 | free_io_failure(failure_tree, tree, failrec); |
77d5d689 | 2679 | return BLK_STS_IOERR; |
2fe6303e MX |
2680 | } |
2681 | ||
77d5d689 OS |
2682 | repair_bio = btrfs_io_bio_alloc(1); |
2683 | repair_io_bio = btrfs_io_bio(repair_bio); | |
2684 | repair_bio->bi_opf = REQ_OP_READ; | |
c7333972 | 2685 | if (need_validation) |
77d5d689 OS |
2686 | repair_bio->bi_opf |= REQ_FAILFAST_DEV; |
2687 | repair_bio->bi_end_io = failed_bio->bi_end_io; | |
2688 | repair_bio->bi_iter.bi_sector = failrec->logical >> 9; | |
2689 | repair_bio->bi_private = failed_bio->bi_private; | |
2fe6303e | 2690 | |
77d5d689 | 2691 | if (failed_io_bio->csum) { |
223486c2 | 2692 | const u32 csum_size = fs_info->csum_size; |
77d5d689 OS |
2693 | |
2694 | repair_io_bio->csum = repair_io_bio->csum_inline; | |
2695 | memcpy(repair_io_bio->csum, | |
2696 | failed_io_bio->csum + csum_size * icsum, csum_size); | |
2697 | } | |
2fe6303e | 2698 | |
77d5d689 OS |
2699 | bio_add_page(repair_bio, page, failrec->len, pgoff); |
2700 | repair_io_bio->logical = failrec->start; | |
2701 | repair_io_bio->iter = repair_bio->bi_iter; | |
4a54c8c1 | 2702 | |
ab8d0fc4 | 2703 | btrfs_debug(btrfs_sb(inode->i_sb), |
77d5d689 OS |
2704 | "repair read error: submitting new read to mirror %d, in_validation=%d", |
2705 | failrec->this_mirror, failrec->in_validation); | |
4a54c8c1 | 2706 | |
77d5d689 OS |
2707 | status = submit_bio_hook(inode, repair_bio, failrec->this_mirror, |
2708 | failrec->bio_flags); | |
4e4cbee9 | 2709 | if (status) { |
7870d082 | 2710 | free_io_failure(failure_tree, tree, failrec); |
77d5d689 | 2711 | bio_put(repair_bio); |
6c387ab2 | 2712 | } |
77d5d689 | 2713 | return status; |
4a54c8c1 JS |
2714 | } |
2715 | ||
d1310b2e CM |
2716 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2717 | ||
b5227c07 | 2718 | void end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
87826df0 JM |
2719 | { |
2720 | int uptodate = (err == 0); | |
3e2426bd | 2721 | int ret = 0; |
87826df0 | 2722 | |
c629732d | 2723 | btrfs_writepage_endio_finish_ordered(page, start, end, uptodate); |
87826df0 | 2724 | |
87826df0 | 2725 | if (!uptodate) { |
87826df0 JM |
2726 | ClearPageUptodate(page); |
2727 | SetPageError(page); | |
bff5baf8 | 2728 | ret = err < 0 ? err : -EIO; |
5dca6eea | 2729 | mapping_set_error(page->mapping, ret); |
87826df0 | 2730 | } |
87826df0 JM |
2731 | } |
2732 | ||
d1310b2e CM |
2733 | /* |
2734 | * after a writepage IO is done, we need to: | |
2735 | * clear the uptodate bits on error | |
2736 | * clear the writeback bits in the extent tree for this IO | |
2737 | * end_page_writeback if the page has no more pending IO | |
2738 | * | |
2739 | * Scheduling is not allowed, so the extent state tree is expected | |
2740 | * to have one and only one object corresponding to this IO. | |
2741 | */ | |
4246a0b6 | 2742 | static void end_bio_extent_writepage(struct bio *bio) |
d1310b2e | 2743 | { |
4e4cbee9 | 2744 | int error = blk_status_to_errno(bio->bi_status); |
2c30c71b | 2745 | struct bio_vec *bvec; |
d1310b2e CM |
2746 | u64 start; |
2747 | u64 end; | |
6dc4f100 | 2748 | struct bvec_iter_all iter_all; |
d1310b2e | 2749 | |
c09abff8 | 2750 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 2751 | bio_for_each_segment_all(bvec, bio, iter_all) { |
d1310b2e | 2752 | struct page *page = bvec->bv_page; |
0b246afa JM |
2753 | struct inode *inode = page->mapping->host; |
2754 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
902b22f3 | 2755 | |
17a5adcc AO |
2756 | /* We always issue full-page reads, but if some block |
2757 | * in a page fails to read, blk_update_request() will | |
2758 | * advance bv_offset and adjust bv_len to compensate. | |
2759 | * Print a warning for nonzero offsets, and an error | |
2760 | * if they don't add up to a full page. */ | |
09cbfeaf KS |
2761 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { |
2762 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
0b246afa | 2763 | btrfs_err(fs_info, |
efe120a0 FH |
2764 | "partial page write in btrfs with offset %u and length %u", |
2765 | bvec->bv_offset, bvec->bv_len); | |
2766 | else | |
0b246afa | 2767 | btrfs_info(fs_info, |
5d163e0e | 2768 | "incomplete page write in btrfs with offset %u and length %u", |
efe120a0 FH |
2769 | bvec->bv_offset, bvec->bv_len); |
2770 | } | |
d1310b2e | 2771 | |
17a5adcc AO |
2772 | start = page_offset(page); |
2773 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
d1310b2e | 2774 | |
4e4cbee9 | 2775 | end_extent_writepage(page, error, start, end); |
17a5adcc | 2776 | end_page_writeback(page); |
2c30c71b | 2777 | } |
2b1f55b0 | 2778 | |
d1310b2e | 2779 | bio_put(bio); |
d1310b2e CM |
2780 | } |
2781 | ||
883d0de4 MX |
2782 | static void |
2783 | endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len, | |
2784 | int uptodate) | |
2785 | { | |
2786 | struct extent_state *cached = NULL; | |
2787 | u64 end = start + len - 1; | |
2788 | ||
2789 | if (uptodate && tree->track_uptodate) | |
2790 | set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC); | |
d810a4be | 2791 | unlock_extent_cached_atomic(tree, start, end, &cached); |
883d0de4 MX |
2792 | } |
2793 | ||
d1310b2e CM |
2794 | /* |
2795 | * after a readpage IO is done, we need to: | |
2796 | * clear the uptodate bits on error | |
2797 | * set the uptodate bits if things worked | |
2798 | * set the page up to date if all extents in the tree are uptodate | |
2799 | * clear the lock bit in the extent tree | |
2800 | * unlock the page if there are no other extents locked for it | |
2801 | * | |
2802 | * Scheduling is not allowed, so the extent state tree is expected | |
2803 | * to have one and only one object corresponding to this IO. | |
2804 | */ | |
4246a0b6 | 2805 | static void end_bio_extent_readpage(struct bio *bio) |
d1310b2e | 2806 | { |
2c30c71b | 2807 | struct bio_vec *bvec; |
4e4cbee9 | 2808 | int uptodate = !bio->bi_status; |
facc8a22 | 2809 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); |
7870d082 | 2810 | struct extent_io_tree *tree, *failure_tree; |
facc8a22 | 2811 | u64 offset = 0; |
d1310b2e CM |
2812 | u64 start; |
2813 | u64 end; | |
facc8a22 | 2814 | u64 len; |
883d0de4 MX |
2815 | u64 extent_start = 0; |
2816 | u64 extent_len = 0; | |
5cf1ab56 | 2817 | int mirror; |
d1310b2e | 2818 | int ret; |
6dc4f100 | 2819 | struct bvec_iter_all iter_all; |
d1310b2e | 2820 | |
c09abff8 | 2821 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 2822 | bio_for_each_segment_all(bvec, bio, iter_all) { |
d1310b2e | 2823 | struct page *page = bvec->bv_page; |
a71754fc | 2824 | struct inode *inode = page->mapping->host; |
ab8d0fc4 | 2825 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8b8bbd46 | 2826 | u32 sectorsize = fs_info->sectorsize; |
507903b8 | 2827 | |
ab8d0fc4 JM |
2828 | btrfs_debug(fs_info, |
2829 | "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u", | |
4e4cbee9 | 2830 | (u64)bio->bi_iter.bi_sector, bio->bi_status, |
ab8d0fc4 | 2831 | io_bio->mirror_num); |
a71754fc | 2832 | tree = &BTRFS_I(inode)->io_tree; |
7870d082 | 2833 | failure_tree = &BTRFS_I(inode)->io_failure_tree; |
902b22f3 | 2834 | |
8b8bbd46 QW |
2835 | /* |
2836 | * We always issue full-sector reads, but if some block in a | |
2837 | * page fails to read, blk_update_request() will advance | |
2838 | * bv_offset and adjust bv_len to compensate. Print a warning | |
2839 | * for unaligned offsets, and an error if they don't add up to | |
2840 | * a full sector. | |
2841 | */ | |
2842 | if (!IS_ALIGNED(bvec->bv_offset, sectorsize)) | |
2843 | btrfs_err(fs_info, | |
2844 | "partial page read in btrfs with offset %u and length %u", | |
2845 | bvec->bv_offset, bvec->bv_len); | |
2846 | else if (!IS_ALIGNED(bvec->bv_offset + bvec->bv_len, | |
2847 | sectorsize)) | |
2848 | btrfs_info(fs_info, | |
2849 | "incomplete page read with offset %u and length %u", | |
2850 | bvec->bv_offset, bvec->bv_len); | |
2851 | ||
2852 | start = page_offset(page) + bvec->bv_offset; | |
2853 | end = start + bvec->bv_len - 1; | |
facc8a22 | 2854 | len = bvec->bv_len; |
d1310b2e | 2855 | |
9be3395b | 2856 | mirror = io_bio->mirror_num; |
78e62c02 | 2857 | if (likely(uptodate)) { |
be17b3af | 2858 | if (is_data_inode(inode)) |
9a446d6a NB |
2859 | ret = btrfs_verify_data_csum(io_bio, offset, page, |
2860 | start, end, mirror); | |
2861 | else | |
2862 | ret = btrfs_validate_metadata_buffer(io_bio, | |
2863 | offset, page, start, end, mirror); | |
5ee0844d | 2864 | if (ret) |
d1310b2e | 2865 | uptodate = 0; |
5ee0844d | 2866 | else |
7870d082 JB |
2867 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
2868 | failure_tree, tree, start, | |
2869 | page, | |
2870 | btrfs_ino(BTRFS_I(inode)), 0); | |
d1310b2e | 2871 | } |
ea466794 | 2872 | |
f2a09da9 MX |
2873 | if (likely(uptodate)) |
2874 | goto readpage_ok; | |
2875 | ||
be17b3af | 2876 | if (is_data_inode(inode)) { |
9d0d1c8b | 2877 | |
f4a8e656 | 2878 | /* |
78e62c02 NB |
2879 | * The generic bio_readpage_error handles errors the |
2880 | * following way: If possible, new read requests are | |
2881 | * created and submitted and will end up in | |
2882 | * end_bio_extent_readpage as well (if we're lucky, | |
2883 | * not in the !uptodate case). In that case it returns | |
2884 | * 0 and we just go on with the next page in our bio. | |
2885 | * If it can't handle the error it will return -EIO and | |
2886 | * we remain responsible for that page. | |
f4a8e656 | 2887 | */ |
77d5d689 OS |
2888 | if (!btrfs_submit_read_repair(inode, bio, offset, page, |
2889 | start - page_offset(page), | |
2890 | start, end, mirror, | |
908930f3 | 2891 | btrfs_submit_data_bio)) { |
78e62c02 NB |
2892 | uptodate = !bio->bi_status; |
2893 | offset += len; | |
2894 | continue; | |
2895 | } | |
2896 | } else { | |
2897 | struct extent_buffer *eb; | |
2898 | ||
2899 | eb = (struct extent_buffer *)page->private; | |
2900 | set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); | |
2901 | eb->read_mirror = mirror; | |
2902 | atomic_dec(&eb->io_pages); | |
2903 | if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, | |
2904 | &eb->bflags)) | |
2905 | btree_readahead_hook(eb, -EIO); | |
7e38326f | 2906 | } |
f2a09da9 | 2907 | readpage_ok: |
883d0de4 | 2908 | if (likely(uptodate)) { |
a71754fc | 2909 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 2910 | pgoff_t end_index = i_size >> PAGE_SHIFT; |
a583c026 | 2911 | unsigned off; |
a71754fc JB |
2912 | |
2913 | /* Zero out the end if this page straddles i_size */ | |
7073017a | 2914 | off = offset_in_page(i_size); |
a583c026 | 2915 | if (page->index == end_index && off) |
09cbfeaf | 2916 | zero_user_segment(page, off, PAGE_SIZE); |
17a5adcc | 2917 | SetPageUptodate(page); |
70dec807 | 2918 | } else { |
17a5adcc AO |
2919 | ClearPageUptodate(page); |
2920 | SetPageError(page); | |
70dec807 | 2921 | } |
17a5adcc | 2922 | unlock_page(page); |
facc8a22 | 2923 | offset += len; |
883d0de4 MX |
2924 | |
2925 | if (unlikely(!uptodate)) { | |
2926 | if (extent_len) { | |
2927 | endio_readpage_release_extent(tree, | |
2928 | extent_start, | |
2929 | extent_len, 1); | |
2930 | extent_start = 0; | |
2931 | extent_len = 0; | |
2932 | } | |
2933 | endio_readpage_release_extent(tree, start, | |
2934 | end - start + 1, 0); | |
2935 | } else if (!extent_len) { | |
2936 | extent_start = start; | |
2937 | extent_len = end + 1 - start; | |
2938 | } else if (extent_start + extent_len == start) { | |
2939 | extent_len += end + 1 - start; | |
2940 | } else { | |
2941 | endio_readpage_release_extent(tree, extent_start, | |
2942 | extent_len, uptodate); | |
2943 | extent_start = start; | |
2944 | extent_len = end + 1 - start; | |
2945 | } | |
2c30c71b | 2946 | } |
d1310b2e | 2947 | |
883d0de4 MX |
2948 | if (extent_len) |
2949 | endio_readpage_release_extent(tree, extent_start, extent_len, | |
2950 | uptodate); | |
b3a0dd50 | 2951 | btrfs_io_bio_free_csum(io_bio); |
d1310b2e | 2952 | bio_put(bio); |
d1310b2e CM |
2953 | } |
2954 | ||
9be3395b | 2955 | /* |
184f999e DS |
2956 | * Initialize the members up to but not including 'bio'. Use after allocating a |
2957 | * new bio by bio_alloc_bioset as it does not initialize the bytes outside of | |
2958 | * 'bio' because use of __GFP_ZERO is not supported. | |
9be3395b | 2959 | */ |
184f999e | 2960 | static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio) |
d1310b2e | 2961 | { |
184f999e DS |
2962 | memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio)); |
2963 | } | |
d1310b2e | 2964 | |
9be3395b | 2965 | /* |
6e707bcd DS |
2966 | * The following helpers allocate a bio. As it's backed by a bioset, it'll |
2967 | * never fail. We're returning a bio right now but you can call btrfs_io_bio | |
2968 | * for the appropriate container_of magic | |
9be3395b | 2969 | */ |
e749af44 | 2970 | struct bio *btrfs_bio_alloc(u64 first_byte) |
d1310b2e CM |
2971 | { |
2972 | struct bio *bio; | |
d1310b2e | 2973 | |
8ac9f7c1 | 2974 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &btrfs_bioset); |
c821e7f3 | 2975 | bio->bi_iter.bi_sector = first_byte >> 9; |
184f999e | 2976 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
d1310b2e CM |
2977 | return bio; |
2978 | } | |
2979 | ||
8b6c1d56 | 2980 | struct bio *btrfs_bio_clone(struct bio *bio) |
9be3395b | 2981 | { |
23ea8e5a MX |
2982 | struct btrfs_io_bio *btrfs_bio; |
2983 | struct bio *new; | |
9be3395b | 2984 | |
6e707bcd | 2985 | /* Bio allocation backed by a bioset does not fail */ |
8ac9f7c1 | 2986 | new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset); |
6e707bcd | 2987 | btrfs_bio = btrfs_io_bio(new); |
184f999e | 2988 | btrfs_io_bio_init(btrfs_bio); |
6e707bcd | 2989 | btrfs_bio->iter = bio->bi_iter; |
23ea8e5a MX |
2990 | return new; |
2991 | } | |
9be3395b | 2992 | |
c5e4c3d7 | 2993 | struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs) |
9be3395b | 2994 | { |
facc8a22 MX |
2995 | struct bio *bio; |
2996 | ||
6e707bcd | 2997 | /* Bio allocation backed by a bioset does not fail */ |
8ac9f7c1 | 2998 | bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset); |
184f999e | 2999 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
facc8a22 | 3000 | return bio; |
9be3395b CM |
3001 | } |
3002 | ||
e477094f | 3003 | struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size) |
2f8e9140 LB |
3004 | { |
3005 | struct bio *bio; | |
3006 | struct btrfs_io_bio *btrfs_bio; | |
3007 | ||
3008 | /* this will never fail when it's backed by a bioset */ | |
8ac9f7c1 | 3009 | bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset); |
2f8e9140 LB |
3010 | ASSERT(bio); |
3011 | ||
3012 | btrfs_bio = btrfs_io_bio(bio); | |
184f999e | 3013 | btrfs_io_bio_init(btrfs_bio); |
2f8e9140 LB |
3014 | |
3015 | bio_trim(bio, offset >> 9, size >> 9); | |
17347cec | 3016 | btrfs_bio->iter = bio->bi_iter; |
2f8e9140 LB |
3017 | return bio; |
3018 | } | |
9be3395b | 3019 | |
4b81ba48 DS |
3020 | /* |
3021 | * @opf: bio REQ_OP_* and REQ_* flags as one value | |
b8b3d625 DS |
3022 | * @wbc: optional writeback control for io accounting |
3023 | * @page: page to add to the bio | |
3024 | * @pg_offset: offset of the new bio or to check whether we are adding | |
3025 | * a contiguous page to the previous one | |
3026 | * @size: portion of page that we want to write | |
3027 | * @offset: starting offset in the page | |
5c2b1fd7 | 3028 | * @bio_ret: must be valid pointer, newly allocated bio will be stored there |
b8b3d625 DS |
3029 | * @end_io_func: end_io callback for new bio |
3030 | * @mirror_num: desired mirror to read/write | |
3031 | * @prev_bio_flags: flags of previous bio to see if we can merge the current one | |
3032 | * @bio_flags: flags of the current bio to see if we can merge them | |
4b81ba48 | 3033 | */ |
0ceb34bf | 3034 | static int submit_extent_page(unsigned int opf, |
da2f0f74 | 3035 | struct writeback_control *wbc, |
6273b7f8 | 3036 | struct page *page, u64 offset, |
6c5a4e2c | 3037 | size_t size, unsigned long pg_offset, |
d1310b2e | 3038 | struct bio **bio_ret, |
f188591e | 3039 | bio_end_io_t end_io_func, |
c8b97818 CM |
3040 | int mirror_num, |
3041 | unsigned long prev_bio_flags, | |
005efedf FM |
3042 | unsigned long bio_flags, |
3043 | bool force_bio_submit) | |
d1310b2e CM |
3044 | { |
3045 | int ret = 0; | |
3046 | struct bio *bio; | |
e940e9a7 | 3047 | size_t io_size = min_t(size_t, size, PAGE_SIZE); |
6273b7f8 | 3048 | sector_t sector = offset >> 9; |
0ceb34bf | 3049 | struct extent_io_tree *tree = &BTRFS_I(page->mapping->host)->io_tree; |
d1310b2e | 3050 | |
5c2b1fd7 DS |
3051 | ASSERT(bio_ret); |
3052 | ||
3053 | if (*bio_ret) { | |
0c8508a6 DS |
3054 | bool contig; |
3055 | bool can_merge = true; | |
3056 | ||
d1310b2e | 3057 | bio = *bio_ret; |
0c8508a6 | 3058 | if (prev_bio_flags & EXTENT_BIO_COMPRESSED) |
4f024f37 | 3059 | contig = bio->bi_iter.bi_sector == sector; |
c8b97818 | 3060 | else |
f73a1c7d | 3061 | contig = bio_end_sector(bio) == sector; |
c8b97818 | 3062 | |
e940e9a7 | 3063 | if (btrfs_bio_fits_in_stripe(page, io_size, bio, bio_flags)) |
0c8508a6 DS |
3064 | can_merge = false; |
3065 | ||
3066 | if (prev_bio_flags != bio_flags || !contig || !can_merge || | |
005efedf | 3067 | force_bio_submit || |
e940e9a7 | 3068 | bio_add_page(bio, page, io_size, pg_offset) < io_size) { |
1f7ad75b | 3069 | ret = submit_one_bio(bio, mirror_num, prev_bio_flags); |
289454ad NA |
3070 | if (ret < 0) { |
3071 | *bio_ret = NULL; | |
79787eaa | 3072 | return ret; |
289454ad | 3073 | } |
d1310b2e CM |
3074 | bio = NULL; |
3075 | } else { | |
da2f0f74 | 3076 | if (wbc) |
e940e9a7 | 3077 | wbc_account_cgroup_owner(wbc, page, io_size); |
d1310b2e CM |
3078 | return 0; |
3079 | } | |
3080 | } | |
c8b97818 | 3081 | |
e749af44 | 3082 | bio = btrfs_bio_alloc(offset); |
e940e9a7 | 3083 | bio_add_page(bio, page, io_size, pg_offset); |
d1310b2e CM |
3084 | bio->bi_end_io = end_io_func; |
3085 | bio->bi_private = tree; | |
e6959b93 | 3086 | bio->bi_write_hint = page->mapping->host->i_write_hint; |
4b81ba48 | 3087 | bio->bi_opf = opf; |
da2f0f74 | 3088 | if (wbc) { |
429aebc0 DS |
3089 | struct block_device *bdev; |
3090 | ||
3091 | bdev = BTRFS_I(page->mapping->host)->root->fs_info->fs_devices->latest_bdev; | |
3092 | bio_set_dev(bio, bdev); | |
da2f0f74 | 3093 | wbc_init_bio(wbc, bio); |
e940e9a7 | 3094 | wbc_account_cgroup_owner(wbc, page, io_size); |
da2f0f74 | 3095 | } |
70dec807 | 3096 | |
5c2b1fd7 | 3097 | *bio_ret = bio; |
d1310b2e CM |
3098 | |
3099 | return ret; | |
3100 | } | |
3101 | ||
48a3b636 ES |
3102 | static void attach_extent_buffer_page(struct extent_buffer *eb, |
3103 | struct page *page) | |
d1310b2e | 3104 | { |
0d01e247 QW |
3105 | /* |
3106 | * If the page is mapped to btree inode, we should hold the private | |
3107 | * lock to prevent race. | |
3108 | * For cloned or dummy extent buffers, their pages are not mapped and | |
3109 | * will not race with any other ebs. | |
3110 | */ | |
3111 | if (page->mapping) | |
3112 | lockdep_assert_held(&page->mapping->private_lock); | |
3113 | ||
d1b89bc0 GJ |
3114 | if (!PagePrivate(page)) |
3115 | attach_page_private(page, eb); | |
3116 | else | |
4f2de97a | 3117 | WARN_ON(page->private != (unsigned long)eb); |
d1310b2e CM |
3118 | } |
3119 | ||
4f2de97a | 3120 | void set_page_extent_mapped(struct page *page) |
d1310b2e | 3121 | { |
d1b89bc0 GJ |
3122 | if (!PagePrivate(page)) |
3123 | attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE); | |
d1310b2e CM |
3124 | } |
3125 | ||
125bac01 MX |
3126 | static struct extent_map * |
3127 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
1a5ee1e6 | 3128 | u64 start, u64 len, struct extent_map **em_cached) |
125bac01 MX |
3129 | { |
3130 | struct extent_map *em; | |
3131 | ||
3132 | if (em_cached && *em_cached) { | |
3133 | em = *em_cached; | |
cbc0e928 | 3134 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 | 3135 | start < extent_map_end(em)) { |
490b54d6 | 3136 | refcount_inc(&em->refs); |
125bac01 MX |
3137 | return em; |
3138 | } | |
3139 | ||
3140 | free_extent_map(em); | |
3141 | *em_cached = NULL; | |
3142 | } | |
3143 | ||
1a5ee1e6 | 3144 | em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len); |
125bac01 MX |
3145 | if (em_cached && !IS_ERR_OR_NULL(em)) { |
3146 | BUG_ON(*em_cached); | |
490b54d6 | 3147 | refcount_inc(&em->refs); |
125bac01 MX |
3148 | *em_cached = em; |
3149 | } | |
3150 | return em; | |
3151 | } | |
d1310b2e CM |
3152 | /* |
3153 | * basic readpage implementation. Locked extent state structs are inserted | |
3154 | * into the tree that are removed when the IO is done (by the end_io | |
3155 | * handlers) | |
79787eaa | 3156 | * XXX JDM: This needs looking at to ensure proper page locking |
baf863b9 | 3157 | * return 0 on success, otherwise return error |
d1310b2e | 3158 | */ |
0f208812 NB |
3159 | int btrfs_do_readpage(struct page *page, struct extent_map **em_cached, |
3160 | struct bio **bio, unsigned long *bio_flags, | |
3161 | unsigned int read_flags, u64 *prev_em_start) | |
d1310b2e CM |
3162 | { |
3163 | struct inode *inode = page->mapping->host; | |
4eee4fa4 | 3164 | u64 start = page_offset(page); |
8eec8296 | 3165 | const u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
3166 | u64 cur = start; |
3167 | u64 extent_offset; | |
3168 | u64 last_byte = i_size_read(inode); | |
3169 | u64 block_start; | |
3170 | u64 cur_end; | |
d1310b2e | 3171 | struct extent_map *em; |
baf863b9 | 3172 | int ret = 0; |
d1310b2e | 3173 | int nr = 0; |
306e16ce | 3174 | size_t pg_offset = 0; |
d1310b2e CM |
3175 | size_t iosize; |
3176 | size_t blocksize = inode->i_sb->s_blocksize; | |
7f042a83 | 3177 | unsigned long this_bio_flag = 0; |
f657a31c | 3178 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
ae6957eb | 3179 | |
d1310b2e CM |
3180 | set_page_extent_mapped(page); |
3181 | ||
90a887c9 DM |
3182 | if (!PageUptodate(page)) { |
3183 | if (cleancache_get_page(page) == 0) { | |
3184 | BUG_ON(blocksize != PAGE_SIZE); | |
9974090b | 3185 | unlock_extent(tree, start, end); |
90a887c9 DM |
3186 | goto out; |
3187 | } | |
3188 | } | |
3189 | ||
09cbfeaf | 3190 | if (page->index == last_byte >> PAGE_SHIFT) { |
c8b97818 | 3191 | char *userpage; |
7073017a | 3192 | size_t zero_offset = offset_in_page(last_byte); |
c8b97818 CM |
3193 | |
3194 | if (zero_offset) { | |
09cbfeaf | 3195 | iosize = PAGE_SIZE - zero_offset; |
7ac687d9 | 3196 | userpage = kmap_atomic(page); |
c8b97818 CM |
3197 | memset(userpage + zero_offset, 0, iosize); |
3198 | flush_dcache_page(page); | |
7ac687d9 | 3199 | kunmap_atomic(userpage); |
c8b97818 CM |
3200 | } |
3201 | } | |
d1310b2e | 3202 | while (cur <= end) { |
005efedf | 3203 | bool force_bio_submit = false; |
6273b7f8 | 3204 | u64 offset; |
c8f2f24b | 3205 | |
d1310b2e CM |
3206 | if (cur >= last_byte) { |
3207 | char *userpage; | |
507903b8 AJ |
3208 | struct extent_state *cached = NULL; |
3209 | ||
09cbfeaf | 3210 | iosize = PAGE_SIZE - pg_offset; |
7ac687d9 | 3211 | userpage = kmap_atomic(page); |
306e16ce | 3212 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3213 | flush_dcache_page(page); |
7ac687d9 | 3214 | kunmap_atomic(userpage); |
d1310b2e | 3215 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 3216 | &cached, GFP_NOFS); |
7f042a83 | 3217 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3218 | cur + iosize - 1, &cached); |
d1310b2e CM |
3219 | break; |
3220 | } | |
125bac01 | 3221 | em = __get_extent_map(inode, page, pg_offset, cur, |
1a5ee1e6 | 3222 | end - cur + 1, em_cached); |
c704005d | 3223 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3224 | SetPageError(page); |
7f042a83 | 3225 | unlock_extent(tree, cur, end); |
d1310b2e CM |
3226 | break; |
3227 | } | |
d1310b2e CM |
3228 | extent_offset = cur - em->start; |
3229 | BUG_ON(extent_map_end(em) <= cur); | |
3230 | BUG_ON(end < cur); | |
3231 | ||
261507a0 | 3232 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
4b384318 | 3233 | this_bio_flag |= EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
3234 | extent_set_compress_type(&this_bio_flag, |
3235 | em->compress_type); | |
3236 | } | |
c8b97818 | 3237 | |
d1310b2e CM |
3238 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
3239 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 3240 | iosize = ALIGN(iosize, blocksize); |
949b3273 | 3241 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) |
6273b7f8 | 3242 | offset = em->block_start; |
949b3273 | 3243 | else |
6273b7f8 | 3244 | offset = em->block_start + extent_offset; |
d1310b2e | 3245 | block_start = em->block_start; |
d899e052 YZ |
3246 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
3247 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
3248 | |
3249 | /* | |
3250 | * If we have a file range that points to a compressed extent | |
260db43c | 3251 | * and it's followed by a consecutive file range that points |
005efedf FM |
3252 | * to the same compressed extent (possibly with a different |
3253 | * offset and/or length, so it either points to the whole extent | |
3254 | * or only part of it), we must make sure we do not submit a | |
3255 | * single bio to populate the pages for the 2 ranges because | |
3256 | * this makes the compressed extent read zero out the pages | |
3257 | * belonging to the 2nd range. Imagine the following scenario: | |
3258 | * | |
3259 | * File layout | |
3260 | * [0 - 8K] [8K - 24K] | |
3261 | * | | | |
3262 | * | | | |
3263 | * points to extent X, points to extent X, | |
3264 | * offset 4K, length of 8K offset 0, length 16K | |
3265 | * | |
3266 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
3267 | * | |
3268 | * If the bio to read the compressed extent covers both ranges, | |
3269 | * it will decompress extent X into the pages belonging to the | |
3270 | * first range and then it will stop, zeroing out the remaining | |
3271 | * pages that belong to the other range that points to extent X. | |
3272 | * So here we make sure we submit 2 bios, one for the first | |
3273 | * range and another one for the third range. Both will target | |
3274 | * the same physical extent from disk, but we can't currently | |
3275 | * make the compressed bio endio callback populate the pages | |
3276 | * for both ranges because each compressed bio is tightly | |
3277 | * coupled with a single extent map, and each range can have | |
3278 | * an extent map with a different offset value relative to the | |
3279 | * uncompressed data of our extent and different lengths. This | |
3280 | * is a corner case so we prioritize correctness over | |
3281 | * non-optimal behavior (submitting 2 bios for the same extent). | |
3282 | */ | |
3283 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
3284 | prev_em_start && *prev_em_start != (u64)-1 && | |
8e928218 | 3285 | *prev_em_start != em->start) |
005efedf FM |
3286 | force_bio_submit = true; |
3287 | ||
3288 | if (prev_em_start) | |
8e928218 | 3289 | *prev_em_start = em->start; |
005efedf | 3290 | |
d1310b2e CM |
3291 | free_extent_map(em); |
3292 | em = NULL; | |
3293 | ||
3294 | /* we've found a hole, just zero and go on */ | |
3295 | if (block_start == EXTENT_MAP_HOLE) { | |
3296 | char *userpage; | |
507903b8 AJ |
3297 | struct extent_state *cached = NULL; |
3298 | ||
7ac687d9 | 3299 | userpage = kmap_atomic(page); |
306e16ce | 3300 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3301 | flush_dcache_page(page); |
7ac687d9 | 3302 | kunmap_atomic(userpage); |
d1310b2e CM |
3303 | |
3304 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 | 3305 | &cached, GFP_NOFS); |
7f042a83 | 3306 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3307 | cur + iosize - 1, &cached); |
d1310b2e | 3308 | cur = cur + iosize; |
306e16ce | 3309 | pg_offset += iosize; |
d1310b2e CM |
3310 | continue; |
3311 | } | |
3312 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
3313 | if (test_range_bit(tree, cur, cur_end, |
3314 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 3315 | check_page_uptodate(tree, page); |
7f042a83 | 3316 | unlock_extent(tree, cur, cur + iosize - 1); |
d1310b2e | 3317 | cur = cur + iosize; |
306e16ce | 3318 | pg_offset += iosize; |
d1310b2e CM |
3319 | continue; |
3320 | } | |
70dec807 CM |
3321 | /* we have an inline extent but it didn't get marked up |
3322 | * to date. Error out | |
3323 | */ | |
3324 | if (block_start == EXTENT_MAP_INLINE) { | |
3325 | SetPageError(page); | |
7f042a83 | 3326 | unlock_extent(tree, cur, cur + iosize - 1); |
70dec807 | 3327 | cur = cur + iosize; |
306e16ce | 3328 | pg_offset += iosize; |
70dec807 CM |
3329 | continue; |
3330 | } | |
d1310b2e | 3331 | |
0ceb34bf | 3332 | ret = submit_extent_page(REQ_OP_READ | read_flags, NULL, |
949b3273 | 3333 | page, offset, iosize, |
fa17ed06 | 3334 | pg_offset, bio, |
fd513000 | 3335 | end_bio_extent_readpage, 0, |
c8b97818 | 3336 | *bio_flags, |
005efedf FM |
3337 | this_bio_flag, |
3338 | force_bio_submit); | |
c8f2f24b JB |
3339 | if (!ret) { |
3340 | nr++; | |
3341 | *bio_flags = this_bio_flag; | |
3342 | } else { | |
d1310b2e | 3343 | SetPageError(page); |
7f042a83 | 3344 | unlock_extent(tree, cur, cur + iosize - 1); |
baf863b9 | 3345 | goto out; |
edd33c99 | 3346 | } |
d1310b2e | 3347 | cur = cur + iosize; |
306e16ce | 3348 | pg_offset += iosize; |
d1310b2e | 3349 | } |
90a887c9 | 3350 | out: |
d1310b2e CM |
3351 | if (!nr) { |
3352 | if (!PageError(page)) | |
3353 | SetPageUptodate(page); | |
3354 | unlock_page(page); | |
3355 | } | |
baf863b9 | 3356 | return ret; |
d1310b2e CM |
3357 | } |
3358 | ||
b6660e80 | 3359 | static inline void contiguous_readpages(struct page *pages[], int nr_pages, |
9974090b | 3360 | u64 start, u64 end, |
125bac01 | 3361 | struct extent_map **em_cached, |
d3fac6ba | 3362 | struct bio **bio, |
1f7ad75b | 3363 | unsigned long *bio_flags, |
808f80b4 | 3364 | u64 *prev_em_start) |
9974090b | 3365 | { |
23d31bd4 | 3366 | struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host); |
9974090b MX |
3367 | int index; |
3368 | ||
b272ae22 | 3369 | btrfs_lock_and_flush_ordered_range(inode, start, end, NULL); |
9974090b MX |
3370 | |
3371 | for (index = 0; index < nr_pages; index++) { | |
0f208812 NB |
3372 | btrfs_do_readpage(pages[index], em_cached, bio, bio_flags, |
3373 | REQ_RAHEAD, prev_em_start); | |
09cbfeaf | 3374 | put_page(pages[index]); |
9974090b MX |
3375 | } |
3376 | } | |
3377 | ||
3d4b9496 | 3378 | static void update_nr_written(struct writeback_control *wbc, |
a9132667 | 3379 | unsigned long nr_written) |
11c8349b CM |
3380 | { |
3381 | wbc->nr_to_write -= nr_written; | |
11c8349b CM |
3382 | } |
3383 | ||
d1310b2e | 3384 | /* |
40f76580 CM |
3385 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
3386 | * | |
5eaad97a | 3387 | * This returns 1 if btrfs_run_delalloc_range function did all the work required |
40f76580 CM |
3388 | * to write the page (copy into inline extent). In this case the IO has |
3389 | * been started and the page is already unlocked. | |
3390 | * | |
3391 | * This returns 0 if all went well (page still locked) | |
3392 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 3393 | */ |
cd4c0bf9 | 3394 | static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode, |
8cc0237a NB |
3395 | struct page *page, struct writeback_control *wbc, |
3396 | u64 delalloc_start, unsigned long *nr_written) | |
40f76580 | 3397 | { |
09cbfeaf | 3398 | u64 page_end = delalloc_start + PAGE_SIZE - 1; |
3522e903 | 3399 | bool found; |
40f76580 CM |
3400 | u64 delalloc_to_write = 0; |
3401 | u64 delalloc_end = 0; | |
3402 | int ret; | |
3403 | int page_started = 0; | |
3404 | ||
40f76580 CM |
3405 | |
3406 | while (delalloc_end < page_end) { | |
cd4c0bf9 | 3407 | found = find_lock_delalloc_range(&inode->vfs_inode, page, |
40f76580 | 3408 | &delalloc_start, |
917aacec | 3409 | &delalloc_end); |
3522e903 | 3410 | if (!found) { |
40f76580 CM |
3411 | delalloc_start = delalloc_end + 1; |
3412 | continue; | |
3413 | } | |
cd4c0bf9 | 3414 | ret = btrfs_run_delalloc_range(inode, page, delalloc_start, |
5eaad97a | 3415 | delalloc_end, &page_started, nr_written, wbc); |
40f76580 CM |
3416 | if (ret) { |
3417 | SetPageError(page); | |
5eaad97a NB |
3418 | /* |
3419 | * btrfs_run_delalloc_range should return < 0 for error | |
3420 | * but just in case, we use > 0 here meaning the IO is | |
3421 | * started, so we don't want to return > 0 unless | |
3422 | * things are going well. | |
40f76580 | 3423 | */ |
b69d1ee9 | 3424 | return ret < 0 ? ret : -EIO; |
40f76580 CM |
3425 | } |
3426 | /* | |
ea1754a0 KS |
3427 | * delalloc_end is already one less than the total length, so |
3428 | * we don't subtract one from PAGE_SIZE | |
40f76580 CM |
3429 | */ |
3430 | delalloc_to_write += (delalloc_end - delalloc_start + | |
ea1754a0 | 3431 | PAGE_SIZE) >> PAGE_SHIFT; |
40f76580 CM |
3432 | delalloc_start = delalloc_end + 1; |
3433 | } | |
3434 | if (wbc->nr_to_write < delalloc_to_write) { | |
3435 | int thresh = 8192; | |
3436 | ||
3437 | if (delalloc_to_write < thresh * 2) | |
3438 | thresh = delalloc_to_write; | |
3439 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3440 | thresh); | |
3441 | } | |
3442 | ||
3443 | /* did the fill delalloc function already unlock and start | |
3444 | * the IO? | |
3445 | */ | |
3446 | if (page_started) { | |
3447 | /* | |
3448 | * we've unlocked the page, so we can't update | |
3449 | * the mapping's writeback index, just update | |
3450 | * nr_to_write. | |
3451 | */ | |
3452 | wbc->nr_to_write -= *nr_written; | |
3453 | return 1; | |
3454 | } | |
3455 | ||
b69d1ee9 | 3456 | return 0; |
40f76580 CM |
3457 | } |
3458 | ||
3459 | /* | |
3460 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3461 | * and does the loop to map the page into extents and bios. | |
3462 | * | |
3463 | * We return 1 if the IO is started and the page is unlocked, | |
3464 | * 0 if all went well (page still locked) | |
3465 | * < 0 if there were errors (page still locked) | |
3466 | */ | |
d4580fe2 | 3467 | static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode, |
40f76580 CM |
3468 | struct page *page, |
3469 | struct writeback_control *wbc, | |
3470 | struct extent_page_data *epd, | |
3471 | loff_t i_size, | |
3472 | unsigned long nr_written, | |
57e5ffeb | 3473 | int *nr_ret) |
d1310b2e | 3474 | { |
d4580fe2 | 3475 | struct extent_io_tree *tree = &inode->io_tree; |
4eee4fa4 | 3476 | u64 start = page_offset(page); |
09cbfeaf | 3477 | u64 page_end = start + PAGE_SIZE - 1; |
d1310b2e CM |
3478 | u64 end; |
3479 | u64 cur = start; | |
3480 | u64 extent_offset; | |
d1310b2e CM |
3481 | u64 block_start; |
3482 | u64 iosize; | |
d1310b2e | 3483 | struct extent_map *em; |
7f3c74fb | 3484 | size_t pg_offset = 0; |
d1310b2e | 3485 | size_t blocksize; |
40f76580 CM |
3486 | int ret = 0; |
3487 | int nr = 0; | |
57e5ffeb | 3488 | const unsigned int write_flags = wbc_to_write_flags(wbc); |
40f76580 | 3489 | bool compressed; |
c8b97818 | 3490 | |
d75855b4 NB |
3491 | ret = btrfs_writepage_cow_fixup(page, start, page_end); |
3492 | if (ret) { | |
3493 | /* Fixup worker will requeue */ | |
5ab58055 | 3494 | redirty_page_for_writepage(wbc, page); |
d75855b4 NB |
3495 | update_nr_written(wbc, nr_written); |
3496 | unlock_page(page); | |
3497 | return 1; | |
247e743c CM |
3498 | } |
3499 | ||
11c8349b CM |
3500 | /* |
3501 | * we don't want to touch the inode after unlocking the page, | |
3502 | * so we update the mapping writeback index now | |
3503 | */ | |
3d4b9496 | 3504 | update_nr_written(wbc, nr_written + 1); |
771ed689 | 3505 | |
d1310b2e | 3506 | end = page_end; |
d4580fe2 | 3507 | blocksize = inode->vfs_inode.i_sb->s_blocksize; |
d1310b2e CM |
3508 | |
3509 | while (cur <= end) { | |
40f76580 | 3510 | u64 em_end; |
6273b7f8 | 3511 | u64 offset; |
58409edd | 3512 | |
40f76580 | 3513 | if (cur >= i_size) { |
7087a9d8 | 3514 | btrfs_writepage_endio_finish_ordered(page, cur, |
c629732d | 3515 | page_end, 1); |
d1310b2e CM |
3516 | break; |
3517 | } | |
d4580fe2 | 3518 | em = btrfs_get_extent(inode, NULL, 0, cur, end - cur + 1); |
c704005d | 3519 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3520 | SetPageError(page); |
61391d56 | 3521 | ret = PTR_ERR_OR_ZERO(em); |
d1310b2e CM |
3522 | break; |
3523 | } | |
3524 | ||
3525 | extent_offset = cur - em->start; | |
40f76580 CM |
3526 | em_end = extent_map_end(em); |
3527 | BUG_ON(em_end <= cur); | |
d1310b2e | 3528 | BUG_ON(end < cur); |
40f76580 | 3529 | iosize = min(em_end - cur, end - cur + 1); |
fda2832f | 3530 | iosize = ALIGN(iosize, blocksize); |
6273b7f8 | 3531 | offset = em->block_start + extent_offset; |
d1310b2e | 3532 | block_start = em->block_start; |
c8b97818 | 3533 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
3534 | free_extent_map(em); |
3535 | em = NULL; | |
3536 | ||
c8b97818 CM |
3537 | /* |
3538 | * compressed and inline extents are written through other | |
3539 | * paths in the FS | |
3540 | */ | |
3541 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3542 | block_start == EXTENT_MAP_INLINE) { |
c8b04030 | 3543 | if (compressed) |
c8b97818 | 3544 | nr++; |
c8b04030 OS |
3545 | else |
3546 | btrfs_writepage_endio_finish_ordered(page, cur, | |
3547 | cur + iosize - 1, 1); | |
c8b97818 | 3548 | cur += iosize; |
7f3c74fb | 3549 | pg_offset += iosize; |
d1310b2e CM |
3550 | continue; |
3551 | } | |
c8b97818 | 3552 | |
5cdc84bf | 3553 | btrfs_set_range_writeback(tree, cur, cur + iosize - 1); |
58409edd | 3554 | if (!PageWriteback(page)) { |
d4580fe2 | 3555 | btrfs_err(inode->root->fs_info, |
58409edd DS |
3556 | "page %lu not writeback, cur %llu end %llu", |
3557 | page->index, cur, end); | |
d1310b2e | 3558 | } |
7f3c74fb | 3559 | |
0ceb34bf | 3560 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc, |
6273b7f8 | 3561 | page, offset, iosize, pg_offset, |
fa17ed06 | 3562 | &epd->bio, |
58409edd DS |
3563 | end_bio_extent_writepage, |
3564 | 0, 0, 0, false); | |
fe01aa65 | 3565 | if (ret) { |
58409edd | 3566 | SetPageError(page); |
fe01aa65 TK |
3567 | if (PageWriteback(page)) |
3568 | end_page_writeback(page); | |
3569 | } | |
d1310b2e | 3570 | |
d1310b2e | 3571 | cur = cur + iosize; |
7f3c74fb | 3572 | pg_offset += iosize; |
d1310b2e CM |
3573 | nr++; |
3574 | } | |
40f76580 | 3575 | *nr_ret = nr; |
40f76580 CM |
3576 | return ret; |
3577 | } | |
3578 | ||
3579 | /* | |
3580 | * the writepage semantics are similar to regular writepage. extent | |
3581 | * records are inserted to lock ranges in the tree, and as dirty areas | |
3582 | * are found, they are marked writeback. Then the lock bits are removed | |
3583 | * and the end_io handler clears the writeback ranges | |
3065976b QW |
3584 | * |
3585 | * Return 0 if everything goes well. | |
3586 | * Return <0 for error. | |
40f76580 CM |
3587 | */ |
3588 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
aab6e9ed | 3589 | struct extent_page_data *epd) |
40f76580 CM |
3590 | { |
3591 | struct inode *inode = page->mapping->host; | |
40f76580 | 3592 | u64 start = page_offset(page); |
09cbfeaf | 3593 | u64 page_end = start + PAGE_SIZE - 1; |
40f76580 CM |
3594 | int ret; |
3595 | int nr = 0; | |
eb70d222 | 3596 | size_t pg_offset; |
40f76580 | 3597 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 3598 | unsigned long end_index = i_size >> PAGE_SHIFT; |
40f76580 CM |
3599 | unsigned long nr_written = 0; |
3600 | ||
40f76580 CM |
3601 | trace___extent_writepage(page, inode, wbc); |
3602 | ||
3603 | WARN_ON(!PageLocked(page)); | |
3604 | ||
3605 | ClearPageError(page); | |
3606 | ||
7073017a | 3607 | pg_offset = offset_in_page(i_size); |
40f76580 CM |
3608 | if (page->index > end_index || |
3609 | (page->index == end_index && !pg_offset)) { | |
09cbfeaf | 3610 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); |
40f76580 CM |
3611 | unlock_page(page); |
3612 | return 0; | |
3613 | } | |
3614 | ||
3615 | if (page->index == end_index) { | |
3616 | char *userpage; | |
3617 | ||
3618 | userpage = kmap_atomic(page); | |
3619 | memset(userpage + pg_offset, 0, | |
09cbfeaf | 3620 | PAGE_SIZE - pg_offset); |
40f76580 CM |
3621 | kunmap_atomic(userpage); |
3622 | flush_dcache_page(page); | |
3623 | } | |
3624 | ||
40f76580 CM |
3625 | set_page_extent_mapped(page); |
3626 | ||
7789a55a | 3627 | if (!epd->extent_locked) { |
cd4c0bf9 NB |
3628 | ret = writepage_delalloc(BTRFS_I(inode), page, wbc, start, |
3629 | &nr_written); | |
7789a55a | 3630 | if (ret == 1) |
169d2c87 | 3631 | return 0; |
7789a55a NB |
3632 | if (ret) |
3633 | goto done; | |
3634 | } | |
40f76580 | 3635 | |
d4580fe2 NB |
3636 | ret = __extent_writepage_io(BTRFS_I(inode), page, wbc, epd, i_size, |
3637 | nr_written, &nr); | |
40f76580 | 3638 | if (ret == 1) |
169d2c87 | 3639 | return 0; |
40f76580 | 3640 | |
d1310b2e CM |
3641 | done: |
3642 | if (nr == 0) { | |
3643 | /* make sure the mapping tag for page dirty gets cleared */ | |
3644 | set_page_writeback(page); | |
3645 | end_page_writeback(page); | |
3646 | } | |
61391d56 FM |
3647 | if (PageError(page)) { |
3648 | ret = ret < 0 ? ret : -EIO; | |
3649 | end_extent_writepage(page, ret, start, page_end); | |
3650 | } | |
d1310b2e | 3651 | unlock_page(page); |
3065976b | 3652 | ASSERT(ret <= 0); |
40f76580 | 3653 | return ret; |
d1310b2e CM |
3654 | } |
3655 | ||
fd8b2b61 | 3656 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 3657 | { |
74316201 N |
3658 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
3659 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
3660 | } |
3661 | ||
18dfa711 FM |
3662 | static void end_extent_buffer_writeback(struct extent_buffer *eb) |
3663 | { | |
3664 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
3665 | smp_mb__after_atomic(); | |
3666 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); | |
3667 | } | |
3668 | ||
2e3c2513 | 3669 | /* |
a3efb2f0 | 3670 | * Lock extent buffer status and pages for writeback. |
2e3c2513 | 3671 | * |
a3efb2f0 QW |
3672 | * May try to flush write bio if we can't get the lock. |
3673 | * | |
3674 | * Return 0 if the extent buffer doesn't need to be submitted. | |
3675 | * (E.g. the extent buffer is not dirty) | |
3676 | * Return >0 is the extent buffer is submitted to bio. | |
3677 | * Return <0 if something went wrong, no page is locked. | |
2e3c2513 | 3678 | */ |
9df76fb5 | 3679 | static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb, |
0e378df1 | 3680 | struct extent_page_data *epd) |
0b32f4bb | 3681 | { |
9df76fb5 | 3682 | struct btrfs_fs_info *fs_info = eb->fs_info; |
2e3c2513 | 3683 | int i, num_pages, failed_page_nr; |
0b32f4bb JB |
3684 | int flush = 0; |
3685 | int ret = 0; | |
3686 | ||
3687 | if (!btrfs_try_tree_write_lock(eb)) { | |
f4340622 | 3688 | ret = flush_write_bio(epd); |
2e3c2513 QW |
3689 | if (ret < 0) |
3690 | return ret; | |
3691 | flush = 1; | |
0b32f4bb JB |
3692 | btrfs_tree_lock(eb); |
3693 | } | |
3694 | ||
3695 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3696 | btrfs_tree_unlock(eb); | |
3697 | if (!epd->sync_io) | |
3698 | return 0; | |
3699 | if (!flush) { | |
f4340622 | 3700 | ret = flush_write_bio(epd); |
2e3c2513 QW |
3701 | if (ret < 0) |
3702 | return ret; | |
0b32f4bb JB |
3703 | flush = 1; |
3704 | } | |
a098d8e8 CM |
3705 | while (1) { |
3706 | wait_on_extent_buffer_writeback(eb); | |
3707 | btrfs_tree_lock(eb); | |
3708 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3709 | break; | |
0b32f4bb | 3710 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3711 | } |
3712 | } | |
3713 | ||
51561ffe JB |
3714 | /* |
3715 | * We need to do this to prevent races in people who check if the eb is | |
3716 | * under IO since we can end up having no IO bits set for a short period | |
3717 | * of time. | |
3718 | */ | |
3719 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
3720 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3721 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 3722 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 3723 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
104b4e51 NB |
3724 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
3725 | -eb->len, | |
3726 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 3727 | ret = 1; |
51561ffe JB |
3728 | } else { |
3729 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
3730 | } |
3731 | ||
3732 | btrfs_tree_unlock(eb); | |
3733 | ||
3734 | if (!ret) | |
3735 | return ret; | |
3736 | ||
65ad0104 | 3737 | num_pages = num_extent_pages(eb); |
0b32f4bb | 3738 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3739 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3740 | |
3741 | if (!trylock_page(p)) { | |
3742 | if (!flush) { | |
18dfa711 FM |
3743 | int err; |
3744 | ||
3745 | err = flush_write_bio(epd); | |
3746 | if (err < 0) { | |
3747 | ret = err; | |
2e3c2513 QW |
3748 | failed_page_nr = i; |
3749 | goto err_unlock; | |
3750 | } | |
0b32f4bb JB |
3751 | flush = 1; |
3752 | } | |
3753 | lock_page(p); | |
3754 | } | |
3755 | } | |
3756 | ||
3757 | return ret; | |
2e3c2513 QW |
3758 | err_unlock: |
3759 | /* Unlock already locked pages */ | |
3760 | for (i = 0; i < failed_page_nr; i++) | |
3761 | unlock_page(eb->pages[i]); | |
18dfa711 FM |
3762 | /* |
3763 | * Clear EXTENT_BUFFER_WRITEBACK and wake up anyone waiting on it. | |
3764 | * Also set back EXTENT_BUFFER_DIRTY so future attempts to this eb can | |
3765 | * be made and undo everything done before. | |
3766 | */ | |
3767 | btrfs_tree_lock(eb); | |
3768 | spin_lock(&eb->refs_lock); | |
3769 | set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); | |
3770 | end_extent_buffer_writeback(eb); | |
3771 | spin_unlock(&eb->refs_lock); | |
3772 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, eb->len, | |
3773 | fs_info->dirty_metadata_batch); | |
3774 | btrfs_clear_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); | |
3775 | btrfs_tree_unlock(eb); | |
2e3c2513 | 3776 | return ret; |
0b32f4bb JB |
3777 | } |
3778 | ||
656f30db FM |
3779 | static void set_btree_ioerr(struct page *page) |
3780 | { | |
3781 | struct extent_buffer *eb = (struct extent_buffer *)page->private; | |
eb5b64f1 | 3782 | struct btrfs_fs_info *fs_info; |
656f30db FM |
3783 | |
3784 | SetPageError(page); | |
3785 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) | |
3786 | return; | |
3787 | ||
eb5b64f1 DZ |
3788 | /* |
3789 | * If we error out, we should add back the dirty_metadata_bytes | |
3790 | * to make it consistent. | |
3791 | */ | |
3792 | fs_info = eb->fs_info; | |
3793 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, | |
3794 | eb->len, fs_info->dirty_metadata_batch); | |
3795 | ||
656f30db FM |
3796 | /* |
3797 | * If writeback for a btree extent that doesn't belong to a log tree | |
3798 | * failed, increment the counter transaction->eb_write_errors. | |
3799 | * We do this because while the transaction is running and before it's | |
3800 | * committing (when we call filemap_fdata[write|wait]_range against | |
3801 | * the btree inode), we might have | |
3802 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
3803 | * returns an error or an error happens during writeback, when we're | |
3804 | * committing the transaction we wouldn't know about it, since the pages | |
3805 | * can be no longer dirty nor marked anymore for writeback (if a | |
3806 | * subsequent modification to the extent buffer didn't happen before the | |
3807 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
3808 | * able to find the pages tagged with SetPageError at transaction | |
3809 | * commit time. So if this happens we must abort the transaction, | |
3810 | * otherwise we commit a super block with btree roots that point to | |
3811 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
3812 | * or the content of some node/leaf from a past generation that got | |
3813 | * cowed or deleted and is no longer valid. | |
3814 | * | |
3815 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
3816 | * not be enough - we need to distinguish between log tree extents vs | |
3817 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
3818 | * will catch and clear such errors in the mapping - and that call might | |
3819 | * be from a log sync and not from a transaction commit. Also, checking | |
3820 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
3821 | * not done and would not be reliable - the eb might have been released | |
3822 | * from memory and reading it back again means that flag would not be | |
3823 | * set (since it's a runtime flag, not persisted on disk). | |
3824 | * | |
3825 | * Using the flags below in the btree inode also makes us achieve the | |
3826 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
3827 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
3828 | * is called, the writeback for all dirty pages had already finished | |
3829 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
3830 | * filemap_fdatawait_range() would return success, as it could not know | |
3831 | * that writeback errors happened (the pages were no longer tagged for | |
3832 | * writeback). | |
3833 | */ | |
3834 | switch (eb->log_index) { | |
3835 | case -1: | |
afcdd129 | 3836 | set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags); |
656f30db FM |
3837 | break; |
3838 | case 0: | |
afcdd129 | 3839 | set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags); |
656f30db FM |
3840 | break; |
3841 | case 1: | |
afcdd129 | 3842 | set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags); |
656f30db FM |
3843 | break; |
3844 | default: | |
3845 | BUG(); /* unexpected, logic error */ | |
3846 | } | |
3847 | } | |
3848 | ||
4246a0b6 | 3849 | static void end_bio_extent_buffer_writepage(struct bio *bio) |
0b32f4bb | 3850 | { |
2c30c71b | 3851 | struct bio_vec *bvec; |
0b32f4bb | 3852 | struct extent_buffer *eb; |
2b070cfe | 3853 | int done; |
6dc4f100 | 3854 | struct bvec_iter_all iter_all; |
0b32f4bb | 3855 | |
c09abff8 | 3856 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 3857 | bio_for_each_segment_all(bvec, bio, iter_all) { |
0b32f4bb JB |
3858 | struct page *page = bvec->bv_page; |
3859 | ||
0b32f4bb JB |
3860 | eb = (struct extent_buffer *)page->private; |
3861 | BUG_ON(!eb); | |
3862 | done = atomic_dec_and_test(&eb->io_pages); | |
3863 | ||
4e4cbee9 | 3864 | if (bio->bi_status || |
4246a0b6 | 3865 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { |
0b32f4bb | 3866 | ClearPageUptodate(page); |
656f30db | 3867 | set_btree_ioerr(page); |
0b32f4bb JB |
3868 | } |
3869 | ||
3870 | end_page_writeback(page); | |
3871 | ||
3872 | if (!done) | |
3873 | continue; | |
3874 | ||
3875 | end_extent_buffer_writeback(eb); | |
2c30c71b | 3876 | } |
0b32f4bb JB |
3877 | |
3878 | bio_put(bio); | |
0b32f4bb JB |
3879 | } |
3880 | ||
0e378df1 | 3881 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb JB |
3882 | struct writeback_control *wbc, |
3883 | struct extent_page_data *epd) | |
3884 | { | |
0b32f4bb | 3885 | u64 offset = eb->start; |
851cd173 | 3886 | u32 nritems; |
cc5e31a4 | 3887 | int i, num_pages; |
851cd173 | 3888 | unsigned long start, end; |
ff40adf7 | 3889 | unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META; |
d7dbe9e7 | 3890 | int ret = 0; |
0b32f4bb | 3891 | |
656f30db | 3892 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); |
65ad0104 | 3893 | num_pages = num_extent_pages(eb); |
0b32f4bb | 3894 | atomic_set(&eb->io_pages, num_pages); |
de0022b9 | 3895 | |
851cd173 LB |
3896 | /* set btree blocks beyond nritems with 0 to avoid stale content. */ |
3897 | nritems = btrfs_header_nritems(eb); | |
3eb548ee | 3898 | if (btrfs_header_level(eb) > 0) { |
3eb548ee LB |
3899 | end = btrfs_node_key_ptr_offset(nritems); |
3900 | ||
b159fa28 | 3901 | memzero_extent_buffer(eb, end, eb->len - end); |
851cd173 LB |
3902 | } else { |
3903 | /* | |
3904 | * leaf: | |
3905 | * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0 | |
3906 | */ | |
3907 | start = btrfs_item_nr_offset(nritems); | |
8f881e8c | 3908 | end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(eb); |
b159fa28 | 3909 | memzero_extent_buffer(eb, start, end - start); |
3eb548ee LB |
3910 | } |
3911 | ||
0b32f4bb | 3912 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3913 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3914 | |
3915 | clear_page_dirty_for_io(p); | |
3916 | set_page_writeback(p); | |
0ceb34bf | 3917 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc, |
fa17ed06 | 3918 | p, offset, PAGE_SIZE, 0, |
c2df8bb4 | 3919 | &epd->bio, |
1f7ad75b | 3920 | end_bio_extent_buffer_writepage, |
18fdc679 | 3921 | 0, 0, 0, false); |
0b32f4bb | 3922 | if (ret) { |
656f30db | 3923 | set_btree_ioerr(p); |
fe01aa65 TK |
3924 | if (PageWriteback(p)) |
3925 | end_page_writeback(p); | |
0b32f4bb JB |
3926 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
3927 | end_extent_buffer_writeback(eb); | |
3928 | ret = -EIO; | |
3929 | break; | |
3930 | } | |
09cbfeaf | 3931 | offset += PAGE_SIZE; |
3d4b9496 | 3932 | update_nr_written(wbc, 1); |
0b32f4bb JB |
3933 | unlock_page(p); |
3934 | } | |
3935 | ||
3936 | if (unlikely(ret)) { | |
3937 | for (; i < num_pages; i++) { | |
bbf65cf0 | 3938 | struct page *p = eb->pages[i]; |
81465028 | 3939 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
3940 | unlock_page(p); |
3941 | } | |
3942 | } | |
3943 | ||
3944 | return ret; | |
3945 | } | |
3946 | ||
3947 | int btree_write_cache_pages(struct address_space *mapping, | |
3948 | struct writeback_control *wbc) | |
3949 | { | |
0b32f4bb JB |
3950 | struct extent_buffer *eb, *prev_eb = NULL; |
3951 | struct extent_page_data epd = { | |
3952 | .bio = NULL, | |
0b32f4bb JB |
3953 | .extent_locked = 0, |
3954 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
3955 | }; | |
b3ff8f1d | 3956 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; |
0b32f4bb JB |
3957 | int ret = 0; |
3958 | int done = 0; | |
3959 | int nr_to_write_done = 0; | |
3960 | struct pagevec pvec; | |
3961 | int nr_pages; | |
3962 | pgoff_t index; | |
3963 | pgoff_t end; /* Inclusive */ | |
3964 | int scanned = 0; | |
10bbd235 | 3965 | xa_mark_t tag; |
0b32f4bb | 3966 | |
86679820 | 3967 | pagevec_init(&pvec); |
0b32f4bb JB |
3968 | if (wbc->range_cyclic) { |
3969 | index = mapping->writeback_index; /* Start from prev offset */ | |
3970 | end = -1; | |
556755a8 JB |
3971 | /* |
3972 | * Start from the beginning does not need to cycle over the | |
3973 | * range, mark it as scanned. | |
3974 | */ | |
3975 | scanned = (index == 0); | |
0b32f4bb | 3976 | } else { |
09cbfeaf KS |
3977 | index = wbc->range_start >> PAGE_SHIFT; |
3978 | end = wbc->range_end >> PAGE_SHIFT; | |
0b32f4bb JB |
3979 | scanned = 1; |
3980 | } | |
3981 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3982 | tag = PAGECACHE_TAG_TOWRITE; | |
3983 | else | |
3984 | tag = PAGECACHE_TAG_DIRTY; | |
3985 | retry: | |
3986 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3987 | tag_pages_for_writeback(mapping, index, end); | |
3988 | while (!done && !nr_to_write_done && (index <= end) && | |
4006f437 | 3989 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 3990 | tag))) { |
0b32f4bb JB |
3991 | unsigned i; |
3992 | ||
0b32f4bb JB |
3993 | for (i = 0; i < nr_pages; i++) { |
3994 | struct page *page = pvec.pages[i]; | |
3995 | ||
3996 | if (!PagePrivate(page)) | |
3997 | continue; | |
3998 | ||
b5bae261 JB |
3999 | spin_lock(&mapping->private_lock); |
4000 | if (!PagePrivate(page)) { | |
4001 | spin_unlock(&mapping->private_lock); | |
4002 | continue; | |
4003 | } | |
4004 | ||
0b32f4bb | 4005 | eb = (struct extent_buffer *)page->private; |
b5bae261 JB |
4006 | |
4007 | /* | |
4008 | * Shouldn't happen and normally this would be a BUG_ON | |
4009 | * but no sense in crashing the users box for something | |
4010 | * we can survive anyway. | |
4011 | */ | |
fae7f21c | 4012 | if (WARN_ON(!eb)) { |
b5bae261 | 4013 | spin_unlock(&mapping->private_lock); |
0b32f4bb JB |
4014 | continue; |
4015 | } | |
4016 | ||
b5bae261 JB |
4017 | if (eb == prev_eb) { |
4018 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 4019 | continue; |
b5bae261 | 4020 | } |
0b32f4bb | 4021 | |
b5bae261 JB |
4022 | ret = atomic_inc_not_zero(&eb->refs); |
4023 | spin_unlock(&mapping->private_lock); | |
4024 | if (!ret) | |
0b32f4bb | 4025 | continue; |
0b32f4bb JB |
4026 | |
4027 | prev_eb = eb; | |
9df76fb5 | 4028 | ret = lock_extent_buffer_for_io(eb, &epd); |
0b32f4bb JB |
4029 | if (!ret) { |
4030 | free_extent_buffer(eb); | |
4031 | continue; | |
0607eb1d FM |
4032 | } else if (ret < 0) { |
4033 | done = 1; | |
4034 | free_extent_buffer(eb); | |
4035 | break; | |
0b32f4bb JB |
4036 | } |
4037 | ||
0ab02063 | 4038 | ret = write_one_eb(eb, wbc, &epd); |
0b32f4bb JB |
4039 | if (ret) { |
4040 | done = 1; | |
4041 | free_extent_buffer(eb); | |
4042 | break; | |
4043 | } | |
4044 | free_extent_buffer(eb); | |
4045 | ||
4046 | /* | |
4047 | * the filesystem may choose to bump up nr_to_write. | |
4048 | * We have to make sure to honor the new nr_to_write | |
4049 | * at any time | |
4050 | */ | |
4051 | nr_to_write_done = wbc->nr_to_write <= 0; | |
4052 | } | |
4053 | pagevec_release(&pvec); | |
4054 | cond_resched(); | |
4055 | } | |
4056 | if (!scanned && !done) { | |
4057 | /* | |
4058 | * We hit the last page and there is more work to be done: wrap | |
4059 | * back to the start of the file | |
4060 | */ | |
4061 | scanned = 1; | |
4062 | index = 0; | |
4063 | goto retry; | |
4064 | } | |
2b952eea QW |
4065 | ASSERT(ret <= 0); |
4066 | if (ret < 0) { | |
4067 | end_write_bio(&epd, ret); | |
4068 | return ret; | |
4069 | } | |
b3ff8f1d QW |
4070 | /* |
4071 | * If something went wrong, don't allow any metadata write bio to be | |
4072 | * submitted. | |
4073 | * | |
4074 | * This would prevent use-after-free if we had dirty pages not | |
4075 | * cleaned up, which can still happen by fuzzed images. | |
4076 | * | |
4077 | * - Bad extent tree | |
4078 | * Allowing existing tree block to be allocated for other trees. | |
4079 | * | |
4080 | * - Log tree operations | |
4081 | * Exiting tree blocks get allocated to log tree, bumps its | |
4082 | * generation, then get cleaned in tree re-balance. | |
4083 | * Such tree block will not be written back, since it's clean, | |
4084 | * thus no WRITTEN flag set. | |
4085 | * And after log writes back, this tree block is not traced by | |
4086 | * any dirty extent_io_tree. | |
4087 | * | |
4088 | * - Offending tree block gets re-dirtied from its original owner | |
4089 | * Since it has bumped generation, no WRITTEN flag, it can be | |
4090 | * reused without COWing. This tree block will not be traced | |
4091 | * by btrfs_transaction::dirty_pages. | |
4092 | * | |
4093 | * Now such dirty tree block will not be cleaned by any dirty | |
4094 | * extent io tree. Thus we don't want to submit such wild eb | |
4095 | * if the fs already has error. | |
4096 | */ | |
4097 | if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { | |
4098 | ret = flush_write_bio(&epd); | |
4099 | } else { | |
fbabd4a3 | 4100 | ret = -EROFS; |
b3ff8f1d QW |
4101 | end_write_bio(&epd, ret); |
4102 | } | |
0b32f4bb JB |
4103 | return ret; |
4104 | } | |
4105 | ||
d1310b2e | 4106 | /** |
4bef0848 | 4107 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
4108 | * @mapping: address space structure to write |
4109 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
935db853 | 4110 | * @data: data passed to __extent_writepage function |
d1310b2e CM |
4111 | * |
4112 | * If a page is already under I/O, write_cache_pages() skips it, even | |
4113 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
4114 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
4115 | * and msync() need to guarantee that all the data which was dirty at the time | |
4116 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
4117 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
4118 | * existing IO to complete. | |
4119 | */ | |
4242b64a | 4120 | static int extent_write_cache_pages(struct address_space *mapping, |
4bef0848 | 4121 | struct writeback_control *wbc, |
aab6e9ed | 4122 | struct extent_page_data *epd) |
d1310b2e | 4123 | { |
7fd1a3f7 | 4124 | struct inode *inode = mapping->host; |
d1310b2e CM |
4125 | int ret = 0; |
4126 | int done = 0; | |
f85d7d6c | 4127 | int nr_to_write_done = 0; |
d1310b2e CM |
4128 | struct pagevec pvec; |
4129 | int nr_pages; | |
4130 | pgoff_t index; | |
4131 | pgoff_t end; /* Inclusive */ | |
a9132667 LB |
4132 | pgoff_t done_index; |
4133 | int range_whole = 0; | |
d1310b2e | 4134 | int scanned = 0; |
10bbd235 | 4135 | xa_mark_t tag; |
d1310b2e | 4136 | |
7fd1a3f7 JB |
4137 | /* |
4138 | * We have to hold onto the inode so that ordered extents can do their | |
4139 | * work when the IO finishes. The alternative to this is failing to add | |
4140 | * an ordered extent if the igrab() fails there and that is a huge pain | |
4141 | * to deal with, so instead just hold onto the inode throughout the | |
4142 | * writepages operation. If it fails here we are freeing up the inode | |
4143 | * anyway and we'd rather not waste our time writing out stuff that is | |
4144 | * going to be truncated anyway. | |
4145 | */ | |
4146 | if (!igrab(inode)) | |
4147 | return 0; | |
4148 | ||
86679820 | 4149 | pagevec_init(&pvec); |
d1310b2e CM |
4150 | if (wbc->range_cyclic) { |
4151 | index = mapping->writeback_index; /* Start from prev offset */ | |
4152 | end = -1; | |
556755a8 JB |
4153 | /* |
4154 | * Start from the beginning does not need to cycle over the | |
4155 | * range, mark it as scanned. | |
4156 | */ | |
4157 | scanned = (index == 0); | |
d1310b2e | 4158 | } else { |
09cbfeaf KS |
4159 | index = wbc->range_start >> PAGE_SHIFT; |
4160 | end = wbc->range_end >> PAGE_SHIFT; | |
a9132667 LB |
4161 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
4162 | range_whole = 1; | |
d1310b2e CM |
4163 | scanned = 1; |
4164 | } | |
3cd24c69 EL |
4165 | |
4166 | /* | |
4167 | * We do the tagged writepage as long as the snapshot flush bit is set | |
4168 | * and we are the first one who do the filemap_flush() on this inode. | |
4169 | * | |
4170 | * The nr_to_write == LONG_MAX is needed to make sure other flushers do | |
4171 | * not race in and drop the bit. | |
4172 | */ | |
4173 | if (range_whole && wbc->nr_to_write == LONG_MAX && | |
4174 | test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
4175 | &BTRFS_I(inode)->runtime_flags)) | |
4176 | wbc->tagged_writepages = 1; | |
4177 | ||
4178 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) | |
f7aaa06b JB |
4179 | tag = PAGECACHE_TAG_TOWRITE; |
4180 | else | |
4181 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 4182 | retry: |
3cd24c69 | 4183 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
f7aaa06b | 4184 | tag_pages_for_writeback(mapping, index, end); |
a9132667 | 4185 | done_index = index; |
f85d7d6c | 4186 | while (!done && !nr_to_write_done && (index <= end) && |
67fd707f JK |
4187 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, |
4188 | &index, end, tag))) { | |
d1310b2e CM |
4189 | unsigned i; |
4190 | ||
d1310b2e CM |
4191 | for (i = 0; i < nr_pages; i++) { |
4192 | struct page *page = pvec.pages[i]; | |
4193 | ||
f7bddf1e | 4194 | done_index = page->index + 1; |
d1310b2e | 4195 | /* |
b93b0163 MW |
4196 | * At this point we hold neither the i_pages lock nor |
4197 | * the page lock: the page may be truncated or | |
4198 | * invalidated (changing page->mapping to NULL), | |
4199 | * or even swizzled back from swapper_space to | |
4200 | * tmpfs file mapping | |
d1310b2e | 4201 | */ |
c8f2f24b | 4202 | if (!trylock_page(page)) { |
f4340622 QW |
4203 | ret = flush_write_bio(epd); |
4204 | BUG_ON(ret < 0); | |
c8f2f24b | 4205 | lock_page(page); |
01d658f2 | 4206 | } |
d1310b2e CM |
4207 | |
4208 | if (unlikely(page->mapping != mapping)) { | |
4209 | unlock_page(page); | |
4210 | continue; | |
4211 | } | |
4212 | ||
d2c3f4f6 | 4213 | if (wbc->sync_mode != WB_SYNC_NONE) { |
f4340622 QW |
4214 | if (PageWriteback(page)) { |
4215 | ret = flush_write_bio(epd); | |
4216 | BUG_ON(ret < 0); | |
4217 | } | |
d1310b2e | 4218 | wait_on_page_writeback(page); |
d2c3f4f6 | 4219 | } |
d1310b2e CM |
4220 | |
4221 | if (PageWriteback(page) || | |
4222 | !clear_page_dirty_for_io(page)) { | |
4223 | unlock_page(page); | |
4224 | continue; | |
4225 | } | |
4226 | ||
aab6e9ed | 4227 | ret = __extent_writepage(page, wbc, epd); |
a9132667 | 4228 | if (ret < 0) { |
a9132667 LB |
4229 | done = 1; |
4230 | break; | |
4231 | } | |
f85d7d6c CM |
4232 | |
4233 | /* | |
4234 | * the filesystem may choose to bump up nr_to_write. | |
4235 | * We have to make sure to honor the new nr_to_write | |
4236 | * at any time | |
4237 | */ | |
4238 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
4239 | } |
4240 | pagevec_release(&pvec); | |
4241 | cond_resched(); | |
4242 | } | |
894b36e3 | 4243 | if (!scanned && !done) { |
d1310b2e CM |
4244 | /* |
4245 | * We hit the last page and there is more work to be done: wrap | |
4246 | * back to the start of the file | |
4247 | */ | |
4248 | scanned = 1; | |
4249 | index = 0; | |
42ffb0bf JB |
4250 | |
4251 | /* | |
4252 | * If we're looping we could run into a page that is locked by a | |
4253 | * writer and that writer could be waiting on writeback for a | |
4254 | * page in our current bio, and thus deadlock, so flush the | |
4255 | * write bio here. | |
4256 | */ | |
4257 | ret = flush_write_bio(epd); | |
4258 | if (!ret) | |
4259 | goto retry; | |
d1310b2e | 4260 | } |
a9132667 LB |
4261 | |
4262 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole)) | |
4263 | mapping->writeback_index = done_index; | |
4264 | ||
7fd1a3f7 | 4265 | btrfs_add_delayed_iput(inode); |
894b36e3 | 4266 | return ret; |
d1310b2e | 4267 | } |
d1310b2e | 4268 | |
0a9b0e53 | 4269 | int extent_write_full_page(struct page *page, struct writeback_control *wbc) |
d1310b2e CM |
4270 | { |
4271 | int ret; | |
d1310b2e CM |
4272 | struct extent_page_data epd = { |
4273 | .bio = NULL, | |
771ed689 | 4274 | .extent_locked = 0, |
ffbd517d | 4275 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e | 4276 | }; |
d1310b2e | 4277 | |
d1310b2e | 4278 | ret = __extent_writepage(page, wbc, &epd); |
3065976b QW |
4279 | ASSERT(ret <= 0); |
4280 | if (ret < 0) { | |
4281 | end_write_bio(&epd, ret); | |
4282 | return ret; | |
4283 | } | |
d1310b2e | 4284 | |
3065976b QW |
4285 | ret = flush_write_bio(&epd); |
4286 | ASSERT(ret <= 0); | |
d1310b2e CM |
4287 | return ret; |
4288 | } | |
d1310b2e | 4289 | |
5e3ee236 | 4290 | int extent_write_locked_range(struct inode *inode, u64 start, u64 end, |
771ed689 CM |
4291 | int mode) |
4292 | { | |
4293 | int ret = 0; | |
4294 | struct address_space *mapping = inode->i_mapping; | |
4295 | struct page *page; | |
09cbfeaf KS |
4296 | unsigned long nr_pages = (end - start + PAGE_SIZE) >> |
4297 | PAGE_SHIFT; | |
771ed689 CM |
4298 | |
4299 | struct extent_page_data epd = { | |
4300 | .bio = NULL, | |
771ed689 | 4301 | .extent_locked = 1, |
ffbd517d | 4302 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
4303 | }; |
4304 | struct writeback_control wbc_writepages = { | |
771ed689 | 4305 | .sync_mode = mode, |
771ed689 CM |
4306 | .nr_to_write = nr_pages * 2, |
4307 | .range_start = start, | |
4308 | .range_end = end + 1, | |
ec39f769 CM |
4309 | /* We're called from an async helper function */ |
4310 | .punt_to_cgroup = 1, | |
4311 | .no_cgroup_owner = 1, | |
771ed689 CM |
4312 | }; |
4313 | ||
dbb70bec | 4314 | wbc_attach_fdatawrite_inode(&wbc_writepages, inode); |
d397712b | 4315 | while (start <= end) { |
09cbfeaf | 4316 | page = find_get_page(mapping, start >> PAGE_SHIFT); |
771ed689 CM |
4317 | if (clear_page_dirty_for_io(page)) |
4318 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
4319 | else { | |
7087a9d8 | 4320 | btrfs_writepage_endio_finish_ordered(page, start, |
c629732d | 4321 | start + PAGE_SIZE - 1, 1); |
771ed689 CM |
4322 | unlock_page(page); |
4323 | } | |
09cbfeaf KS |
4324 | put_page(page); |
4325 | start += PAGE_SIZE; | |
771ed689 CM |
4326 | } |
4327 | ||
02c6db4f | 4328 | ASSERT(ret <= 0); |
dbb70bec CM |
4329 | if (ret == 0) |
4330 | ret = flush_write_bio(&epd); | |
4331 | else | |
02c6db4f | 4332 | end_write_bio(&epd, ret); |
dbb70bec CM |
4333 | |
4334 | wbc_detach_inode(&wbc_writepages); | |
771ed689 CM |
4335 | return ret; |
4336 | } | |
d1310b2e | 4337 | |
8ae225a8 | 4338 | int extent_writepages(struct address_space *mapping, |
d1310b2e CM |
4339 | struct writeback_control *wbc) |
4340 | { | |
4341 | int ret = 0; | |
4342 | struct extent_page_data epd = { | |
4343 | .bio = NULL, | |
771ed689 | 4344 | .extent_locked = 0, |
ffbd517d | 4345 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
4346 | }; |
4347 | ||
935db853 | 4348 | ret = extent_write_cache_pages(mapping, wbc, &epd); |
a2a72fbd QW |
4349 | ASSERT(ret <= 0); |
4350 | if (ret < 0) { | |
4351 | end_write_bio(&epd, ret); | |
4352 | return ret; | |
4353 | } | |
4354 | ret = flush_write_bio(&epd); | |
d1310b2e CM |
4355 | return ret; |
4356 | } | |
d1310b2e | 4357 | |
ba206a02 | 4358 | void extent_readahead(struct readahead_control *rac) |
d1310b2e CM |
4359 | { |
4360 | struct bio *bio = NULL; | |
c8b97818 | 4361 | unsigned long bio_flags = 0; |
67c9684f | 4362 | struct page *pagepool[16]; |
125bac01 | 4363 | struct extent_map *em_cached = NULL; |
808f80b4 | 4364 | u64 prev_em_start = (u64)-1; |
ba206a02 | 4365 | int nr; |
d1310b2e | 4366 | |
ba206a02 MWO |
4367 | while ((nr = readahead_page_batch(rac, pagepool))) { |
4368 | u64 contig_start = page_offset(pagepool[0]); | |
4369 | u64 contig_end = page_offset(pagepool[nr - 1]) + PAGE_SIZE - 1; | |
e65ef21e | 4370 | |
ba206a02 | 4371 | ASSERT(contig_start + nr * PAGE_SIZE - 1 == contig_end); |
e65ef21e | 4372 | |
ba206a02 MWO |
4373 | contiguous_readpages(pagepool, nr, contig_start, contig_end, |
4374 | &em_cached, &bio, &bio_flags, &prev_em_start); | |
d1310b2e | 4375 | } |
67c9684f | 4376 | |
125bac01 MX |
4377 | if (em_cached) |
4378 | free_extent_map(em_cached); | |
4379 | ||
ba206a02 MWO |
4380 | if (bio) { |
4381 | if (submit_one_bio(bio, 0, bio_flags)) | |
4382 | return; | |
4383 | } | |
d1310b2e | 4384 | } |
d1310b2e CM |
4385 | |
4386 | /* | |
4387 | * basic invalidatepage code, this waits on any locked or writeback | |
4388 | * ranges corresponding to the page, and then deletes any extent state | |
4389 | * records from the tree | |
4390 | */ | |
4391 | int extent_invalidatepage(struct extent_io_tree *tree, | |
4392 | struct page *page, unsigned long offset) | |
4393 | { | |
2ac55d41 | 4394 | struct extent_state *cached_state = NULL; |
4eee4fa4 | 4395 | u64 start = page_offset(page); |
09cbfeaf | 4396 | u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
4397 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; |
4398 | ||
fda2832f | 4399 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
4400 | if (start > end) |
4401 | return 0; | |
4402 | ||
ff13db41 | 4403 | lock_extent_bits(tree, start, end, &cached_state); |
1edbb734 | 4404 | wait_on_page_writeback(page); |
e182163d OS |
4405 | clear_extent_bit(tree, start, end, EXTENT_LOCKED | EXTENT_DELALLOC | |
4406 | EXTENT_DO_ACCOUNTING, 1, 1, &cached_state); | |
d1310b2e CM |
4407 | return 0; |
4408 | } | |
d1310b2e | 4409 | |
7b13b7b1 CM |
4410 | /* |
4411 | * a helper for releasepage, this tests for areas of the page that | |
4412 | * are locked or under IO and drops the related state bits if it is safe | |
4413 | * to drop the page. | |
4414 | */ | |
29c68b2d | 4415 | static int try_release_extent_state(struct extent_io_tree *tree, |
48a3b636 | 4416 | struct page *page, gfp_t mask) |
7b13b7b1 | 4417 | { |
4eee4fa4 | 4418 | u64 start = page_offset(page); |
09cbfeaf | 4419 | u64 end = start + PAGE_SIZE - 1; |
7b13b7b1 CM |
4420 | int ret = 1; |
4421 | ||
8882679e | 4422 | if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) { |
7b13b7b1 | 4423 | ret = 0; |
8882679e | 4424 | } else { |
11ef160f CM |
4425 | /* |
4426 | * at this point we can safely clear everything except the | |
4427 | * locked bit and the nodatasum bit | |
4428 | */ | |
66b0c887 | 4429 | ret = __clear_extent_bit(tree, start, end, |
11ef160f | 4430 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
66b0c887 | 4431 | 0, 0, NULL, mask, NULL); |
e3f24cc5 CM |
4432 | |
4433 | /* if clear_extent_bit failed for enomem reasons, | |
4434 | * we can't allow the release to continue. | |
4435 | */ | |
4436 | if (ret < 0) | |
4437 | ret = 0; | |
4438 | else | |
4439 | ret = 1; | |
7b13b7b1 CM |
4440 | } |
4441 | return ret; | |
4442 | } | |
7b13b7b1 | 4443 | |
d1310b2e CM |
4444 | /* |
4445 | * a helper for releasepage. As long as there are no locked extents | |
4446 | * in the range corresponding to the page, both state records and extent | |
4447 | * map records are removed | |
4448 | */ | |
477a30ba | 4449 | int try_release_extent_mapping(struct page *page, gfp_t mask) |
d1310b2e CM |
4450 | { |
4451 | struct extent_map *em; | |
4eee4fa4 | 4452 | u64 start = page_offset(page); |
09cbfeaf | 4453 | u64 end = start + PAGE_SIZE - 1; |
bd3599a0 FM |
4454 | struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host); |
4455 | struct extent_io_tree *tree = &btrfs_inode->io_tree; | |
4456 | struct extent_map_tree *map = &btrfs_inode->extent_tree; | |
7b13b7b1 | 4457 | |
d0164adc | 4458 | if (gfpflags_allow_blocking(mask) && |
ee22184b | 4459 | page->mapping->host->i_size > SZ_16M) { |
39b5637f | 4460 | u64 len; |
70dec807 | 4461 | while (start <= end) { |
fbc2bd7e FM |
4462 | struct btrfs_fs_info *fs_info; |
4463 | u64 cur_gen; | |
4464 | ||
39b5637f | 4465 | len = end - start + 1; |
890871be | 4466 | write_lock(&map->lock); |
39b5637f | 4467 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 4468 | if (!em) { |
890871be | 4469 | write_unlock(&map->lock); |
70dec807 CM |
4470 | break; |
4471 | } | |
7f3c74fb CM |
4472 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
4473 | em->start != start) { | |
890871be | 4474 | write_unlock(&map->lock); |
70dec807 CM |
4475 | free_extent_map(em); |
4476 | break; | |
4477 | } | |
3d6448e6 FM |
4478 | if (test_range_bit(tree, em->start, |
4479 | extent_map_end(em) - 1, | |
4480 | EXTENT_LOCKED, 0, NULL)) | |
4481 | goto next; | |
4482 | /* | |
4483 | * If it's not in the list of modified extents, used | |
4484 | * by a fast fsync, we can remove it. If it's being | |
4485 | * logged we can safely remove it since fsync took an | |
4486 | * extra reference on the em. | |
4487 | */ | |
4488 | if (list_empty(&em->list) || | |
fbc2bd7e FM |
4489 | test_bit(EXTENT_FLAG_LOGGING, &em->flags)) |
4490 | goto remove_em; | |
4491 | /* | |
4492 | * If it's in the list of modified extents, remove it | |
4493 | * only if its generation is older then the current one, | |
4494 | * in which case we don't need it for a fast fsync. | |
4495 | * Otherwise don't remove it, we could be racing with an | |
4496 | * ongoing fast fsync that could miss the new extent. | |
4497 | */ | |
4498 | fs_info = btrfs_inode->root->fs_info; | |
4499 | spin_lock(&fs_info->trans_lock); | |
4500 | cur_gen = fs_info->generation; | |
4501 | spin_unlock(&fs_info->trans_lock); | |
4502 | if (em->generation >= cur_gen) | |
4503 | goto next; | |
4504 | remove_em: | |
5e548b32 FM |
4505 | /* |
4506 | * We only remove extent maps that are not in the list of | |
4507 | * modified extents or that are in the list but with a | |
4508 | * generation lower then the current generation, so there | |
4509 | * is no need to set the full fsync flag on the inode (it | |
4510 | * hurts the fsync performance for workloads with a data | |
4511 | * size that exceeds or is close to the system's memory). | |
4512 | */ | |
fbc2bd7e FM |
4513 | remove_extent_mapping(map, em); |
4514 | /* once for the rb tree */ | |
4515 | free_extent_map(em); | |
3d6448e6 | 4516 | next: |
70dec807 | 4517 | start = extent_map_end(em); |
890871be | 4518 | write_unlock(&map->lock); |
70dec807 CM |
4519 | |
4520 | /* once for us */ | |
d1310b2e | 4521 | free_extent_map(em); |
9f47eb54 PM |
4522 | |
4523 | cond_resched(); /* Allow large-extent preemption. */ | |
d1310b2e | 4524 | } |
d1310b2e | 4525 | } |
29c68b2d | 4526 | return try_release_extent_state(tree, page, mask); |
d1310b2e | 4527 | } |
d1310b2e | 4528 | |
ec29ed5b CM |
4529 | /* |
4530 | * helper function for fiemap, which doesn't want to see any holes. | |
4531 | * This maps until we find something past 'last' | |
4532 | */ | |
f1bbde8d | 4533 | static struct extent_map *get_extent_skip_holes(struct btrfs_inode *inode, |
e3350e16 | 4534 | u64 offset, u64 last) |
ec29ed5b | 4535 | { |
f1bbde8d | 4536 | u64 sectorsize = btrfs_inode_sectorsize(inode); |
ec29ed5b CM |
4537 | struct extent_map *em; |
4538 | u64 len; | |
4539 | ||
4540 | if (offset >= last) | |
4541 | return NULL; | |
4542 | ||
67871254 | 4543 | while (1) { |
ec29ed5b CM |
4544 | len = last - offset; |
4545 | if (len == 0) | |
4546 | break; | |
fda2832f | 4547 | len = ALIGN(len, sectorsize); |
f1bbde8d | 4548 | em = btrfs_get_extent_fiemap(inode, offset, len); |
c704005d | 4549 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
4550 | return em; |
4551 | ||
4552 | /* if this isn't a hole return it */ | |
4a2d25cd | 4553 | if (em->block_start != EXTENT_MAP_HOLE) |
ec29ed5b | 4554 | return em; |
ec29ed5b CM |
4555 | |
4556 | /* this is a hole, advance to the next extent */ | |
4557 | offset = extent_map_end(em); | |
4558 | free_extent_map(em); | |
4559 | if (offset >= last) | |
4560 | break; | |
4561 | } | |
4562 | return NULL; | |
4563 | } | |
4564 | ||
4751832d QW |
4565 | /* |
4566 | * To cache previous fiemap extent | |
4567 | * | |
4568 | * Will be used for merging fiemap extent | |
4569 | */ | |
4570 | struct fiemap_cache { | |
4571 | u64 offset; | |
4572 | u64 phys; | |
4573 | u64 len; | |
4574 | u32 flags; | |
4575 | bool cached; | |
4576 | }; | |
4577 | ||
4578 | /* | |
4579 | * Helper to submit fiemap extent. | |
4580 | * | |
4581 | * Will try to merge current fiemap extent specified by @offset, @phys, | |
4582 | * @len and @flags with cached one. | |
4583 | * And only when we fails to merge, cached one will be submitted as | |
4584 | * fiemap extent. | |
4585 | * | |
4586 | * Return value is the same as fiemap_fill_next_extent(). | |
4587 | */ | |
4588 | static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, | |
4589 | struct fiemap_cache *cache, | |
4590 | u64 offset, u64 phys, u64 len, u32 flags) | |
4591 | { | |
4592 | int ret = 0; | |
4593 | ||
4594 | if (!cache->cached) | |
4595 | goto assign; | |
4596 | ||
4597 | /* | |
4598 | * Sanity check, extent_fiemap() should have ensured that new | |
52042d8e | 4599 | * fiemap extent won't overlap with cached one. |
4751832d QW |
4600 | * Not recoverable. |
4601 | * | |
4602 | * NOTE: Physical address can overlap, due to compression | |
4603 | */ | |
4604 | if (cache->offset + cache->len > offset) { | |
4605 | WARN_ON(1); | |
4606 | return -EINVAL; | |
4607 | } | |
4608 | ||
4609 | /* | |
4610 | * Only merges fiemap extents if | |
4611 | * 1) Their logical addresses are continuous | |
4612 | * | |
4613 | * 2) Their physical addresses are continuous | |
4614 | * So truly compressed (physical size smaller than logical size) | |
4615 | * extents won't get merged with each other | |
4616 | * | |
4617 | * 3) Share same flags except FIEMAP_EXTENT_LAST | |
4618 | * So regular extent won't get merged with prealloc extent | |
4619 | */ | |
4620 | if (cache->offset + cache->len == offset && | |
4621 | cache->phys + cache->len == phys && | |
4622 | (cache->flags & ~FIEMAP_EXTENT_LAST) == | |
4623 | (flags & ~FIEMAP_EXTENT_LAST)) { | |
4624 | cache->len += len; | |
4625 | cache->flags |= flags; | |
4626 | goto try_submit_last; | |
4627 | } | |
4628 | ||
4629 | /* Not mergeable, need to submit cached one */ | |
4630 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, | |
4631 | cache->len, cache->flags); | |
4632 | cache->cached = false; | |
4633 | if (ret) | |
4634 | return ret; | |
4635 | assign: | |
4636 | cache->cached = true; | |
4637 | cache->offset = offset; | |
4638 | cache->phys = phys; | |
4639 | cache->len = len; | |
4640 | cache->flags = flags; | |
4641 | try_submit_last: | |
4642 | if (cache->flags & FIEMAP_EXTENT_LAST) { | |
4643 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, | |
4644 | cache->phys, cache->len, cache->flags); | |
4645 | cache->cached = false; | |
4646 | } | |
4647 | return ret; | |
4648 | } | |
4649 | ||
4650 | /* | |
848c23b7 | 4651 | * Emit last fiemap cache |
4751832d | 4652 | * |
848c23b7 QW |
4653 | * The last fiemap cache may still be cached in the following case: |
4654 | * 0 4k 8k | |
4655 | * |<- Fiemap range ->| | |
4656 | * |<------------ First extent ----------->| | |
4657 | * | |
4658 | * In this case, the first extent range will be cached but not emitted. | |
4659 | * So we must emit it before ending extent_fiemap(). | |
4751832d | 4660 | */ |
5c5aff98 | 4661 | static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo, |
848c23b7 | 4662 | struct fiemap_cache *cache) |
4751832d QW |
4663 | { |
4664 | int ret; | |
4665 | ||
4666 | if (!cache->cached) | |
4667 | return 0; | |
4668 | ||
4751832d QW |
4669 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, |
4670 | cache->len, cache->flags); | |
4671 | cache->cached = false; | |
4672 | if (ret > 0) | |
4673 | ret = 0; | |
4674 | return ret; | |
4675 | } | |
4676 | ||
facee0a0 | 4677 | int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo, |
bab16e21 | 4678 | u64 start, u64 len) |
1506fcc8 | 4679 | { |
975f84fe | 4680 | int ret = 0; |
1506fcc8 YS |
4681 | u64 off = start; |
4682 | u64 max = start + len; | |
4683 | u32 flags = 0; | |
975f84fe JB |
4684 | u32 found_type; |
4685 | u64 last; | |
ec29ed5b | 4686 | u64 last_for_get_extent = 0; |
1506fcc8 | 4687 | u64 disko = 0; |
facee0a0 | 4688 | u64 isize = i_size_read(&inode->vfs_inode); |
975f84fe | 4689 | struct btrfs_key found_key; |
1506fcc8 | 4690 | struct extent_map *em = NULL; |
2ac55d41 | 4691 | struct extent_state *cached_state = NULL; |
975f84fe | 4692 | struct btrfs_path *path; |
facee0a0 | 4693 | struct btrfs_root *root = inode->root; |
4751832d | 4694 | struct fiemap_cache cache = { 0 }; |
5911c8fe DS |
4695 | struct ulist *roots; |
4696 | struct ulist *tmp_ulist; | |
1506fcc8 | 4697 | int end = 0; |
ec29ed5b CM |
4698 | u64 em_start = 0; |
4699 | u64 em_len = 0; | |
4700 | u64 em_end = 0; | |
1506fcc8 YS |
4701 | |
4702 | if (len == 0) | |
4703 | return -EINVAL; | |
4704 | ||
975f84fe JB |
4705 | path = btrfs_alloc_path(); |
4706 | if (!path) | |
4707 | return -ENOMEM; | |
975f84fe | 4708 | |
5911c8fe DS |
4709 | roots = ulist_alloc(GFP_KERNEL); |
4710 | tmp_ulist = ulist_alloc(GFP_KERNEL); | |
4711 | if (!roots || !tmp_ulist) { | |
4712 | ret = -ENOMEM; | |
4713 | goto out_free_ulist; | |
4714 | } | |
4715 | ||
facee0a0 NB |
4716 | start = round_down(start, btrfs_inode_sectorsize(inode)); |
4717 | len = round_up(max, btrfs_inode_sectorsize(inode)) - start; | |
4d479cf0 | 4718 | |
ec29ed5b CM |
4719 | /* |
4720 | * lookup the last file extent. We're not using i_size here | |
4721 | * because there might be preallocation past i_size | |
4722 | */ | |
facee0a0 NB |
4723 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1, |
4724 | 0); | |
975f84fe | 4725 | if (ret < 0) { |
5911c8fe | 4726 | goto out_free_ulist; |
2d324f59 LB |
4727 | } else { |
4728 | WARN_ON(!ret); | |
4729 | if (ret == 1) | |
4730 | ret = 0; | |
975f84fe | 4731 | } |
2d324f59 | 4732 | |
975f84fe | 4733 | path->slots[0]--; |
975f84fe | 4734 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
962a298f | 4735 | found_type = found_key.type; |
975f84fe | 4736 | |
ec29ed5b | 4737 | /* No extents, but there might be delalloc bits */ |
facee0a0 | 4738 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 4739 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
4740 | /* have to trust i_size as the end */ |
4741 | last = (u64)-1; | |
4742 | last_for_get_extent = isize; | |
4743 | } else { | |
4744 | /* | |
4745 | * remember the start of the last extent. There are a | |
4746 | * bunch of different factors that go into the length of the | |
4747 | * extent, so its much less complex to remember where it started | |
4748 | */ | |
4749 | last = found_key.offset; | |
4750 | last_for_get_extent = last + 1; | |
975f84fe | 4751 | } |
fe09e16c | 4752 | btrfs_release_path(path); |
975f84fe | 4753 | |
ec29ed5b CM |
4754 | /* |
4755 | * we might have some extents allocated but more delalloc past those | |
4756 | * extents. so, we trust isize unless the start of the last extent is | |
4757 | * beyond isize | |
4758 | */ | |
4759 | if (last < isize) { | |
4760 | last = (u64)-1; | |
4761 | last_for_get_extent = isize; | |
4762 | } | |
4763 | ||
facee0a0 | 4764 | lock_extent_bits(&inode->io_tree, start, start + len - 1, |
d0082371 | 4765 | &cached_state); |
ec29ed5b | 4766 | |
facee0a0 | 4767 | em = get_extent_skip_holes(inode, start, last_for_get_extent); |
1506fcc8 YS |
4768 | if (!em) |
4769 | goto out; | |
4770 | if (IS_ERR(em)) { | |
4771 | ret = PTR_ERR(em); | |
4772 | goto out; | |
4773 | } | |
975f84fe | 4774 | |
1506fcc8 | 4775 | while (!end) { |
b76bb701 | 4776 | u64 offset_in_extent = 0; |
ea8efc74 CM |
4777 | |
4778 | /* break if the extent we found is outside the range */ | |
4779 | if (em->start >= max || extent_map_end(em) < off) | |
4780 | break; | |
4781 | ||
4782 | /* | |
4783 | * get_extent may return an extent that starts before our | |
4784 | * requested range. We have to make sure the ranges | |
4785 | * we return to fiemap always move forward and don't | |
4786 | * overlap, so adjust the offsets here | |
4787 | */ | |
4788 | em_start = max(em->start, off); | |
1506fcc8 | 4789 | |
ea8efc74 CM |
4790 | /* |
4791 | * record the offset from the start of the extent | |
b76bb701 JB |
4792 | * for adjusting the disk offset below. Only do this if the |
4793 | * extent isn't compressed since our in ram offset may be past | |
4794 | * what we have actually allocated on disk. | |
ea8efc74 | 4795 | */ |
b76bb701 JB |
4796 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
4797 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 4798 | em_end = extent_map_end(em); |
ea8efc74 | 4799 | em_len = em_end - em_start; |
1506fcc8 | 4800 | flags = 0; |
f0986318 FM |
4801 | if (em->block_start < EXTENT_MAP_LAST_BYTE) |
4802 | disko = em->block_start + offset_in_extent; | |
4803 | else | |
4804 | disko = 0; | |
1506fcc8 | 4805 | |
ea8efc74 CM |
4806 | /* |
4807 | * bump off for our next call to get_extent | |
4808 | */ | |
4809 | off = extent_map_end(em); | |
4810 | if (off >= max) | |
4811 | end = 1; | |
4812 | ||
93dbfad7 | 4813 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
4814 | end = 1; |
4815 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 4816 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
4817 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
4818 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 4819 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
4820 | flags |= (FIEMAP_EXTENT_DELALLOC | |
4821 | FIEMAP_EXTENT_UNKNOWN); | |
dc046b10 JB |
4822 | } else if (fieinfo->fi_extents_max) { |
4823 | u64 bytenr = em->block_start - | |
4824 | (em->start - em->orig_start); | |
fe09e16c | 4825 | |
fe09e16c LB |
4826 | /* |
4827 | * As btrfs supports shared space, this information | |
4828 | * can be exported to userspace tools via | |
dc046b10 JB |
4829 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 |
4830 | * then we're just getting a count and we can skip the | |
4831 | * lookup stuff. | |
fe09e16c | 4832 | */ |
facee0a0 | 4833 | ret = btrfs_check_shared(root, btrfs_ino(inode), |
5911c8fe | 4834 | bytenr, roots, tmp_ulist); |
dc046b10 | 4835 | if (ret < 0) |
fe09e16c | 4836 | goto out_free; |
dc046b10 | 4837 | if (ret) |
fe09e16c | 4838 | flags |= FIEMAP_EXTENT_SHARED; |
dc046b10 | 4839 | ret = 0; |
1506fcc8 YS |
4840 | } |
4841 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
4842 | flags |= FIEMAP_EXTENT_ENCODED; | |
0d2b2372 JB |
4843 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
4844 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
1506fcc8 | 4845 | |
1506fcc8 YS |
4846 | free_extent_map(em); |
4847 | em = NULL; | |
ec29ed5b CM |
4848 | if ((em_start >= last) || em_len == (u64)-1 || |
4849 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
4850 | flags |= FIEMAP_EXTENT_LAST; |
4851 | end = 1; | |
4852 | } | |
4853 | ||
ec29ed5b | 4854 | /* now scan forward to see if this is really the last extent. */ |
facee0a0 | 4855 | em = get_extent_skip_holes(inode, off, last_for_get_extent); |
ec29ed5b CM |
4856 | if (IS_ERR(em)) { |
4857 | ret = PTR_ERR(em); | |
4858 | goto out; | |
4859 | } | |
4860 | if (!em) { | |
975f84fe JB |
4861 | flags |= FIEMAP_EXTENT_LAST; |
4862 | end = 1; | |
4863 | } | |
4751832d QW |
4864 | ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko, |
4865 | em_len, flags); | |
26e726af CS |
4866 | if (ret) { |
4867 | if (ret == 1) | |
4868 | ret = 0; | |
ec29ed5b | 4869 | goto out_free; |
26e726af | 4870 | } |
1506fcc8 YS |
4871 | } |
4872 | out_free: | |
4751832d | 4873 | if (!ret) |
5c5aff98 | 4874 | ret = emit_last_fiemap_cache(fieinfo, &cache); |
1506fcc8 YS |
4875 | free_extent_map(em); |
4876 | out: | |
facee0a0 | 4877 | unlock_extent_cached(&inode->io_tree, start, start + len - 1, |
e43bbe5e | 4878 | &cached_state); |
5911c8fe DS |
4879 | |
4880 | out_free_ulist: | |
e02d48ea | 4881 | btrfs_free_path(path); |
5911c8fe DS |
4882 | ulist_free(roots); |
4883 | ulist_free(tmp_ulist); | |
1506fcc8 YS |
4884 | return ret; |
4885 | } | |
4886 | ||
727011e0 CM |
4887 | static void __free_extent_buffer(struct extent_buffer *eb) |
4888 | { | |
727011e0 CM |
4889 | kmem_cache_free(extent_buffer_cache, eb); |
4890 | } | |
4891 | ||
2b48966a | 4892 | int extent_buffer_under_io(const struct extent_buffer *eb) |
db7f3436 JB |
4893 | { |
4894 | return (atomic_read(&eb->io_pages) || | |
4895 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4896 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4897 | } | |
4898 | ||
4899 | /* | |
55ac0139 | 4900 | * Release all pages attached to the extent buffer. |
db7f3436 | 4901 | */ |
55ac0139 | 4902 | static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb) |
db7f3436 | 4903 | { |
d64766fd NB |
4904 | int i; |
4905 | int num_pages; | |
b0132a3b | 4906 | int mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
db7f3436 JB |
4907 | |
4908 | BUG_ON(extent_buffer_under_io(eb)); | |
4909 | ||
d64766fd NB |
4910 | num_pages = num_extent_pages(eb); |
4911 | for (i = 0; i < num_pages; i++) { | |
4912 | struct page *page = eb->pages[i]; | |
db7f3436 | 4913 | |
5d2361db FL |
4914 | if (!page) |
4915 | continue; | |
4916 | if (mapped) | |
db7f3436 | 4917 | spin_lock(&page->mapping->private_lock); |
5d2361db FL |
4918 | /* |
4919 | * We do this since we'll remove the pages after we've | |
4920 | * removed the eb from the radix tree, so we could race | |
4921 | * and have this page now attached to the new eb. So | |
4922 | * only clear page_private if it's still connected to | |
4923 | * this eb. | |
4924 | */ | |
4925 | if (PagePrivate(page) && | |
4926 | page->private == (unsigned long)eb) { | |
4927 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4928 | BUG_ON(PageDirty(page)); | |
4929 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 4930 | /* |
5d2361db FL |
4931 | * We need to make sure we haven't be attached |
4932 | * to a new eb. | |
db7f3436 | 4933 | */ |
d1b89bc0 | 4934 | detach_page_private(page); |
db7f3436 | 4935 | } |
5d2361db FL |
4936 | |
4937 | if (mapped) | |
4938 | spin_unlock(&page->mapping->private_lock); | |
4939 | ||
01327610 | 4940 | /* One for when we allocated the page */ |
09cbfeaf | 4941 | put_page(page); |
d64766fd | 4942 | } |
db7f3436 JB |
4943 | } |
4944 | ||
4945 | /* | |
4946 | * Helper for releasing the extent buffer. | |
4947 | */ | |
4948 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4949 | { | |
55ac0139 | 4950 | btrfs_release_extent_buffer_pages(eb); |
8c38938c | 4951 | btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list); |
db7f3436 JB |
4952 | __free_extent_buffer(eb); |
4953 | } | |
4954 | ||
f28491e0 JB |
4955 | static struct extent_buffer * |
4956 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 4957 | unsigned long len) |
d1310b2e CM |
4958 | { |
4959 | struct extent_buffer *eb = NULL; | |
4960 | ||
d1b5c567 | 4961 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
4962 | eb->start = start; |
4963 | eb->len = len; | |
f28491e0 | 4964 | eb->fs_info = fs_info; |
815a51c7 | 4965 | eb->bflags = 0; |
196d59ab | 4966 | init_rwsem(&eb->lock); |
329ced79 | 4967 | eb->lock_recursed = false; |
b4ce94de | 4968 | |
3fd63727 JB |
4969 | btrfs_leak_debug_add(&fs_info->eb_leak_lock, &eb->leak_list, |
4970 | &fs_info->allocated_ebs); | |
6d49ba1b | 4971 | |
3083ee2e | 4972 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 4973 | atomic_set(&eb->refs, 1); |
0b32f4bb | 4974 | atomic_set(&eb->io_pages, 0); |
727011e0 | 4975 | |
b8dae313 DS |
4976 | /* |
4977 | * Sanity checks, currently the maximum is 64k covered by 16x 4k pages | |
4978 | */ | |
4979 | BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE | |
4980 | > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
4981 | BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
d1310b2e CM |
4982 | |
4983 | return eb; | |
4984 | } | |
4985 | ||
2b48966a | 4986 | struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src) |
815a51c7 | 4987 | { |
cc5e31a4 | 4988 | int i; |
815a51c7 JS |
4989 | struct page *p; |
4990 | struct extent_buffer *new; | |
cc5e31a4 | 4991 | int num_pages = num_extent_pages(src); |
815a51c7 | 4992 | |
3f556f78 | 4993 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
4994 | if (new == NULL) |
4995 | return NULL; | |
4996 | ||
4997 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4998 | p = alloc_page(GFP_NOFS); |
db7f3436 JB |
4999 | if (!p) { |
5000 | btrfs_release_extent_buffer(new); | |
5001 | return NULL; | |
5002 | } | |
815a51c7 JS |
5003 | attach_extent_buffer_page(new, p); |
5004 | WARN_ON(PageDirty(p)); | |
5005 | SetPageUptodate(p); | |
5006 | new->pages[i] = p; | |
fba1acf9 | 5007 | copy_page(page_address(p), page_address(src->pages[i])); |
815a51c7 JS |
5008 | } |
5009 | ||
815a51c7 | 5010 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); |
b0132a3b | 5011 | set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags); |
815a51c7 JS |
5012 | |
5013 | return new; | |
5014 | } | |
5015 | ||
0f331229 OS |
5016 | struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
5017 | u64 start, unsigned long len) | |
815a51c7 JS |
5018 | { |
5019 | struct extent_buffer *eb; | |
cc5e31a4 DS |
5020 | int num_pages; |
5021 | int i; | |
815a51c7 | 5022 | |
3f556f78 | 5023 | eb = __alloc_extent_buffer(fs_info, start, len); |
815a51c7 JS |
5024 | if (!eb) |
5025 | return NULL; | |
5026 | ||
65ad0104 | 5027 | num_pages = num_extent_pages(eb); |
815a51c7 | 5028 | for (i = 0; i < num_pages; i++) { |
9ec72677 | 5029 | eb->pages[i] = alloc_page(GFP_NOFS); |
815a51c7 JS |
5030 | if (!eb->pages[i]) |
5031 | goto err; | |
5032 | } | |
5033 | set_extent_buffer_uptodate(eb); | |
5034 | btrfs_set_header_nritems(eb, 0); | |
b0132a3b | 5035 | set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
815a51c7 JS |
5036 | |
5037 | return eb; | |
5038 | err: | |
84167d19 SB |
5039 | for (; i > 0; i--) |
5040 | __free_page(eb->pages[i - 1]); | |
815a51c7 JS |
5041 | __free_extent_buffer(eb); |
5042 | return NULL; | |
5043 | } | |
5044 | ||
0f331229 | 5045 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
da17066c | 5046 | u64 start) |
0f331229 | 5047 | { |
da17066c | 5048 | return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize); |
0f331229 OS |
5049 | } |
5050 | ||
0b32f4bb JB |
5051 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
5052 | { | |
242e18c7 | 5053 | int refs; |
6bf9cd2e BB |
5054 | /* |
5055 | * The TREE_REF bit is first set when the extent_buffer is added | |
5056 | * to the radix tree. It is also reset, if unset, when a new reference | |
5057 | * is created by find_extent_buffer. | |
0b32f4bb | 5058 | * |
6bf9cd2e BB |
5059 | * It is only cleared in two cases: freeing the last non-tree |
5060 | * reference to the extent_buffer when its STALE bit is set or | |
5061 | * calling releasepage when the tree reference is the only reference. | |
0b32f4bb | 5062 | * |
6bf9cd2e BB |
5063 | * In both cases, care is taken to ensure that the extent_buffer's |
5064 | * pages are not under io. However, releasepage can be concurrently | |
5065 | * called with creating new references, which is prone to race | |
5066 | * conditions between the calls to check_buffer_tree_ref in those | |
5067 | * codepaths and clearing TREE_REF in try_release_extent_buffer. | |
0b32f4bb | 5068 | * |
6bf9cd2e BB |
5069 | * The actual lifetime of the extent_buffer in the radix tree is |
5070 | * adequately protected by the refcount, but the TREE_REF bit and | |
5071 | * its corresponding reference are not. To protect against this | |
5072 | * class of races, we call check_buffer_tree_ref from the codepaths | |
5073 | * which trigger io after they set eb->io_pages. Note that once io is | |
5074 | * initiated, TREE_REF can no longer be cleared, so that is the | |
5075 | * moment at which any such race is best fixed. | |
0b32f4bb | 5076 | */ |
242e18c7 CM |
5077 | refs = atomic_read(&eb->refs); |
5078 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
5079 | return; | |
5080 | ||
594831c4 JB |
5081 | spin_lock(&eb->refs_lock); |
5082 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 5083 | atomic_inc(&eb->refs); |
594831c4 | 5084 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
5085 | } |
5086 | ||
2457aec6 MG |
5087 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
5088 | struct page *accessed) | |
5df4235e | 5089 | { |
cc5e31a4 | 5090 | int num_pages, i; |
5df4235e | 5091 | |
0b32f4bb JB |
5092 | check_buffer_tree_ref(eb); |
5093 | ||
65ad0104 | 5094 | num_pages = num_extent_pages(eb); |
5df4235e | 5095 | for (i = 0; i < num_pages; i++) { |
fb85fc9a DS |
5096 | struct page *p = eb->pages[i]; |
5097 | ||
2457aec6 MG |
5098 | if (p != accessed) |
5099 | mark_page_accessed(p); | |
5df4235e JB |
5100 | } |
5101 | } | |
5102 | ||
f28491e0 JB |
5103 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
5104 | u64 start) | |
452c75c3 CS |
5105 | { |
5106 | struct extent_buffer *eb; | |
5107 | ||
5108 | rcu_read_lock(); | |
f28491e0 | 5109 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
478ef886 | 5110 | start >> fs_info->sectorsize_bits); |
452c75c3 CS |
5111 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
5112 | rcu_read_unlock(); | |
062c19e9 FM |
5113 | /* |
5114 | * Lock our eb's refs_lock to avoid races with | |
5115 | * free_extent_buffer. When we get our eb it might be flagged | |
5116 | * with EXTENT_BUFFER_STALE and another task running | |
5117 | * free_extent_buffer might have seen that flag set, | |
5118 | * eb->refs == 2, that the buffer isn't under IO (dirty and | |
5119 | * writeback flags not set) and it's still in the tree (flag | |
5120 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process | |
5121 | * of decrementing the extent buffer's reference count twice. | |
5122 | * So here we could race and increment the eb's reference count, | |
5123 | * clear its stale flag, mark it as dirty and drop our reference | |
5124 | * before the other task finishes executing free_extent_buffer, | |
5125 | * which would later result in an attempt to free an extent | |
5126 | * buffer that is dirty. | |
5127 | */ | |
5128 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
5129 | spin_lock(&eb->refs_lock); | |
5130 | spin_unlock(&eb->refs_lock); | |
5131 | } | |
2457aec6 | 5132 | mark_extent_buffer_accessed(eb, NULL); |
452c75c3 CS |
5133 | return eb; |
5134 | } | |
5135 | rcu_read_unlock(); | |
5136 | ||
5137 | return NULL; | |
5138 | } | |
5139 | ||
faa2dbf0 JB |
5140 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
5141 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
da17066c | 5142 | u64 start) |
faa2dbf0 JB |
5143 | { |
5144 | struct extent_buffer *eb, *exists = NULL; | |
5145 | int ret; | |
5146 | ||
5147 | eb = find_extent_buffer(fs_info, start); | |
5148 | if (eb) | |
5149 | return eb; | |
da17066c | 5150 | eb = alloc_dummy_extent_buffer(fs_info, start); |
faa2dbf0 | 5151 | if (!eb) |
b6293c82 | 5152 | return ERR_PTR(-ENOMEM); |
faa2dbf0 JB |
5153 | eb->fs_info = fs_info; |
5154 | again: | |
e1860a77 | 5155 | ret = radix_tree_preload(GFP_NOFS); |
b6293c82 DC |
5156 | if (ret) { |
5157 | exists = ERR_PTR(ret); | |
faa2dbf0 | 5158 | goto free_eb; |
b6293c82 | 5159 | } |
faa2dbf0 JB |
5160 | spin_lock(&fs_info->buffer_lock); |
5161 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
478ef886 | 5162 | start >> fs_info->sectorsize_bits, eb); |
faa2dbf0 JB |
5163 | spin_unlock(&fs_info->buffer_lock); |
5164 | radix_tree_preload_end(); | |
5165 | if (ret == -EEXIST) { | |
5166 | exists = find_extent_buffer(fs_info, start); | |
5167 | if (exists) | |
5168 | goto free_eb; | |
5169 | else | |
5170 | goto again; | |
5171 | } | |
5172 | check_buffer_tree_ref(eb); | |
5173 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
5174 | ||
faa2dbf0 JB |
5175 | return eb; |
5176 | free_eb: | |
5177 | btrfs_release_extent_buffer(eb); | |
5178 | return exists; | |
5179 | } | |
5180 | #endif | |
5181 | ||
f28491e0 | 5182 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
3fbaf258 | 5183 | u64 start, u64 owner_root, int level) |
d1310b2e | 5184 | { |
da17066c | 5185 | unsigned long len = fs_info->nodesize; |
cc5e31a4 DS |
5186 | int num_pages; |
5187 | int i; | |
09cbfeaf | 5188 | unsigned long index = start >> PAGE_SHIFT; |
d1310b2e | 5189 | struct extent_buffer *eb; |
6af118ce | 5190 | struct extent_buffer *exists = NULL; |
d1310b2e | 5191 | struct page *p; |
f28491e0 | 5192 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
d1310b2e | 5193 | int uptodate = 1; |
19fe0a8b | 5194 | int ret; |
d1310b2e | 5195 | |
da17066c | 5196 | if (!IS_ALIGNED(start, fs_info->sectorsize)) { |
c871b0f2 LB |
5197 | btrfs_err(fs_info, "bad tree block start %llu", start); |
5198 | return ERR_PTR(-EINVAL); | |
5199 | } | |
5200 | ||
f28491e0 | 5201 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 5202 | if (eb) |
6af118ce | 5203 | return eb; |
6af118ce | 5204 | |
23d79d81 | 5205 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 5206 | if (!eb) |
c871b0f2 | 5207 | return ERR_PTR(-ENOMEM); |
d1310b2e | 5208 | |
65ad0104 | 5209 | num_pages = num_extent_pages(eb); |
727011e0 | 5210 | for (i = 0; i < num_pages; i++, index++) { |
d1b5c567 | 5211 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
c871b0f2 LB |
5212 | if (!p) { |
5213 | exists = ERR_PTR(-ENOMEM); | |
6af118ce | 5214 | goto free_eb; |
c871b0f2 | 5215 | } |
4f2de97a JB |
5216 | |
5217 | spin_lock(&mapping->private_lock); | |
5218 | if (PagePrivate(p)) { | |
5219 | /* | |
5220 | * We could have already allocated an eb for this page | |
5221 | * and attached one so lets see if we can get a ref on | |
5222 | * the existing eb, and if we can we know it's good and | |
5223 | * we can just return that one, else we know we can just | |
5224 | * overwrite page->private. | |
5225 | */ | |
5226 | exists = (struct extent_buffer *)p->private; | |
5227 | if (atomic_inc_not_zero(&exists->refs)) { | |
5228 | spin_unlock(&mapping->private_lock); | |
5229 | unlock_page(p); | |
09cbfeaf | 5230 | put_page(p); |
2457aec6 | 5231 | mark_extent_buffer_accessed(exists, p); |
4f2de97a JB |
5232 | goto free_eb; |
5233 | } | |
5ca64f45 | 5234 | exists = NULL; |
4f2de97a | 5235 | |
0b32f4bb | 5236 | /* |
4f2de97a JB |
5237 | * Do this so attach doesn't complain and we need to |
5238 | * drop the ref the old guy had. | |
5239 | */ | |
5240 | ClearPagePrivate(p); | |
0b32f4bb | 5241 | WARN_ON(PageDirty(p)); |
09cbfeaf | 5242 | put_page(p); |
d1310b2e | 5243 | } |
4f2de97a JB |
5244 | attach_extent_buffer_page(eb, p); |
5245 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 5246 | WARN_ON(PageDirty(p)); |
727011e0 | 5247 | eb->pages[i] = p; |
d1310b2e CM |
5248 | if (!PageUptodate(p)) |
5249 | uptodate = 0; | |
eb14ab8e CM |
5250 | |
5251 | /* | |
b16d011e NB |
5252 | * We can't unlock the pages just yet since the extent buffer |
5253 | * hasn't been properly inserted in the radix tree, this | |
5254 | * opens a race with btree_releasepage which can free a page | |
5255 | * while we are still filling in all pages for the buffer and | |
5256 | * we could crash. | |
eb14ab8e | 5257 | */ |
d1310b2e CM |
5258 | } |
5259 | if (uptodate) | |
b4ce94de | 5260 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 5261 | again: |
e1860a77 | 5262 | ret = radix_tree_preload(GFP_NOFS); |
c871b0f2 LB |
5263 | if (ret) { |
5264 | exists = ERR_PTR(ret); | |
19fe0a8b | 5265 | goto free_eb; |
c871b0f2 | 5266 | } |
19fe0a8b | 5267 | |
f28491e0 JB |
5268 | spin_lock(&fs_info->buffer_lock); |
5269 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
478ef886 | 5270 | start >> fs_info->sectorsize_bits, eb); |
f28491e0 | 5271 | spin_unlock(&fs_info->buffer_lock); |
452c75c3 | 5272 | radix_tree_preload_end(); |
19fe0a8b | 5273 | if (ret == -EEXIST) { |
f28491e0 | 5274 | exists = find_extent_buffer(fs_info, start); |
452c75c3 CS |
5275 | if (exists) |
5276 | goto free_eb; | |
5277 | else | |
115391d2 | 5278 | goto again; |
6af118ce | 5279 | } |
6af118ce | 5280 | /* add one reference for the tree */ |
0b32f4bb | 5281 | check_buffer_tree_ref(eb); |
34b41ace | 5282 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
5283 | |
5284 | /* | |
b16d011e NB |
5285 | * Now it's safe to unlock the pages because any calls to |
5286 | * btree_releasepage will correctly detect that a page belongs to a | |
5287 | * live buffer and won't free them prematurely. | |
eb14ab8e | 5288 | */ |
28187ae5 NB |
5289 | for (i = 0; i < num_pages; i++) |
5290 | unlock_page(eb->pages[i]); | |
d1310b2e CM |
5291 | return eb; |
5292 | ||
6af118ce | 5293 | free_eb: |
5ca64f45 | 5294 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
5295 | for (i = 0; i < num_pages; i++) { |
5296 | if (eb->pages[i]) | |
5297 | unlock_page(eb->pages[i]); | |
5298 | } | |
eb14ab8e | 5299 | |
897ca6e9 | 5300 | btrfs_release_extent_buffer(eb); |
6af118ce | 5301 | return exists; |
d1310b2e | 5302 | } |
d1310b2e | 5303 | |
3083ee2e JB |
5304 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
5305 | { | |
5306 | struct extent_buffer *eb = | |
5307 | container_of(head, struct extent_buffer, rcu_head); | |
5308 | ||
5309 | __free_extent_buffer(eb); | |
5310 | } | |
5311 | ||
f7a52a40 | 5312 | static int release_extent_buffer(struct extent_buffer *eb) |
5ce48d0f | 5313 | __releases(&eb->refs_lock) |
3083ee2e | 5314 | { |
07e21c4d NB |
5315 | lockdep_assert_held(&eb->refs_lock); |
5316 | ||
3083ee2e JB |
5317 | WARN_ON(atomic_read(&eb->refs) == 0); |
5318 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 5319 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 5320 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 5321 | |
815a51c7 | 5322 | spin_unlock(&eb->refs_lock); |
3083ee2e | 5323 | |
f28491e0 JB |
5324 | spin_lock(&fs_info->buffer_lock); |
5325 | radix_tree_delete(&fs_info->buffer_radix, | |
478ef886 | 5326 | eb->start >> fs_info->sectorsize_bits); |
f28491e0 | 5327 | spin_unlock(&fs_info->buffer_lock); |
34b41ace JB |
5328 | } else { |
5329 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 5330 | } |
3083ee2e | 5331 | |
8c38938c | 5332 | btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list); |
3083ee2e | 5333 | /* Should be safe to release our pages at this point */ |
55ac0139 | 5334 | btrfs_release_extent_buffer_pages(eb); |
bcb7e449 | 5335 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
b0132a3b | 5336 | if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) { |
bcb7e449 JB |
5337 | __free_extent_buffer(eb); |
5338 | return 1; | |
5339 | } | |
5340 | #endif | |
3083ee2e | 5341 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 5342 | return 1; |
3083ee2e JB |
5343 | } |
5344 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
5345 | |
5346 | return 0; | |
3083ee2e JB |
5347 | } |
5348 | ||
d1310b2e CM |
5349 | void free_extent_buffer(struct extent_buffer *eb) |
5350 | { | |
242e18c7 CM |
5351 | int refs; |
5352 | int old; | |
d1310b2e CM |
5353 | if (!eb) |
5354 | return; | |
5355 | ||
242e18c7 CM |
5356 | while (1) { |
5357 | refs = atomic_read(&eb->refs); | |
46cc775e NB |
5358 | if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3) |
5359 | || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && | |
5360 | refs == 1)) | |
242e18c7 CM |
5361 | break; |
5362 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
5363 | if (old == refs) | |
5364 | return; | |
5365 | } | |
5366 | ||
3083ee2e JB |
5367 | spin_lock(&eb->refs_lock); |
5368 | if (atomic_read(&eb->refs) == 2 && | |
5369 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 5370 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
5371 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5372 | atomic_dec(&eb->refs); | |
5373 | ||
5374 | /* | |
5375 | * I know this is terrible, but it's temporary until we stop tracking | |
5376 | * the uptodate bits and such for the extent buffers. | |
5377 | */ | |
f7a52a40 | 5378 | release_extent_buffer(eb); |
3083ee2e JB |
5379 | } |
5380 | ||
5381 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
5382 | { | |
5383 | if (!eb) | |
d1310b2e CM |
5384 | return; |
5385 | ||
3083ee2e JB |
5386 | spin_lock(&eb->refs_lock); |
5387 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
5388 | ||
0b32f4bb | 5389 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
5390 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5391 | atomic_dec(&eb->refs); | |
f7a52a40 | 5392 | release_extent_buffer(eb); |
d1310b2e | 5393 | } |
d1310b2e | 5394 | |
2b48966a | 5395 | void clear_extent_buffer_dirty(const struct extent_buffer *eb) |
d1310b2e | 5396 | { |
cc5e31a4 DS |
5397 | int i; |
5398 | int num_pages; | |
d1310b2e CM |
5399 | struct page *page; |
5400 | ||
65ad0104 | 5401 | num_pages = num_extent_pages(eb); |
d1310b2e CM |
5402 | |
5403 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 5404 | page = eb->pages[i]; |
b9473439 | 5405 | if (!PageDirty(page)) |
d2c3f4f6 CM |
5406 | continue; |
5407 | ||
a61e6f29 | 5408 | lock_page(page); |
eb14ab8e CM |
5409 | WARN_ON(!PagePrivate(page)); |
5410 | ||
d1310b2e | 5411 | clear_page_dirty_for_io(page); |
b93b0163 | 5412 | xa_lock_irq(&page->mapping->i_pages); |
0a943c65 MW |
5413 | if (!PageDirty(page)) |
5414 | __xa_clear_mark(&page->mapping->i_pages, | |
5415 | page_index(page), PAGECACHE_TAG_DIRTY); | |
b93b0163 | 5416 | xa_unlock_irq(&page->mapping->i_pages); |
bf0da8c1 | 5417 | ClearPageError(page); |
a61e6f29 | 5418 | unlock_page(page); |
d1310b2e | 5419 | } |
0b32f4bb | 5420 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 5421 | } |
d1310b2e | 5422 | |
abb57ef3 | 5423 | bool set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5424 | { |
cc5e31a4 DS |
5425 | int i; |
5426 | int num_pages; | |
abb57ef3 | 5427 | bool was_dirty; |
d1310b2e | 5428 | |
0b32f4bb JB |
5429 | check_buffer_tree_ref(eb); |
5430 | ||
b9473439 | 5431 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 5432 | |
65ad0104 | 5433 | num_pages = num_extent_pages(eb); |
3083ee2e | 5434 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
5435 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
5436 | ||
abb57ef3 LB |
5437 | if (!was_dirty) |
5438 | for (i = 0; i < num_pages; i++) | |
5439 | set_page_dirty(eb->pages[i]); | |
51995c39 LB |
5440 | |
5441 | #ifdef CONFIG_BTRFS_DEBUG | |
5442 | for (i = 0; i < num_pages; i++) | |
5443 | ASSERT(PageDirty(eb->pages[i])); | |
5444 | #endif | |
5445 | ||
b9473439 | 5446 | return was_dirty; |
d1310b2e | 5447 | } |
d1310b2e | 5448 | |
69ba3927 | 5449 | void clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 | 5450 | { |
cc5e31a4 | 5451 | int i; |
1259ab75 | 5452 | struct page *page; |
cc5e31a4 | 5453 | int num_pages; |
1259ab75 | 5454 | |
b4ce94de | 5455 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 5456 | num_pages = num_extent_pages(eb); |
1259ab75 | 5457 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5458 | page = eb->pages[i]; |
33958dc6 CM |
5459 | if (page) |
5460 | ClearPageUptodate(page); | |
1259ab75 | 5461 | } |
1259ab75 CM |
5462 | } |
5463 | ||
09c25a8c | 5464 | void set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 5465 | { |
cc5e31a4 | 5466 | int i; |
d1310b2e | 5467 | struct page *page; |
cc5e31a4 | 5468 | int num_pages; |
d1310b2e | 5469 | |
0b32f4bb | 5470 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 5471 | num_pages = num_extent_pages(eb); |
d1310b2e | 5472 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5473 | page = eb->pages[i]; |
d1310b2e CM |
5474 | SetPageUptodate(page); |
5475 | } | |
d1310b2e | 5476 | } |
d1310b2e | 5477 | |
c2ccfbc6 | 5478 | int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num) |
d1310b2e | 5479 | { |
cc5e31a4 | 5480 | int i; |
d1310b2e CM |
5481 | struct page *page; |
5482 | int err; | |
5483 | int ret = 0; | |
ce9adaa5 CM |
5484 | int locked_pages = 0; |
5485 | int all_uptodate = 1; | |
cc5e31a4 | 5486 | int num_pages; |
727011e0 | 5487 | unsigned long num_reads = 0; |
a86c12c7 | 5488 | struct bio *bio = NULL; |
c8b97818 | 5489 | unsigned long bio_flags = 0; |
a86c12c7 | 5490 | |
b4ce94de | 5491 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
5492 | return 0; |
5493 | ||
65ad0104 | 5494 | num_pages = num_extent_pages(eb); |
8436ea91 | 5495 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5496 | page = eb->pages[i]; |
bb82ab88 | 5497 | if (wait == WAIT_NONE) { |
2db04966 | 5498 | if (!trylock_page(page)) |
ce9adaa5 | 5499 | goto unlock_exit; |
d1310b2e CM |
5500 | } else { |
5501 | lock_page(page); | |
5502 | } | |
ce9adaa5 | 5503 | locked_pages++; |
2571e739 LB |
5504 | } |
5505 | /* | |
5506 | * We need to firstly lock all pages to make sure that | |
5507 | * the uptodate bit of our pages won't be affected by | |
5508 | * clear_extent_buffer_uptodate(). | |
5509 | */ | |
8436ea91 | 5510 | for (i = 0; i < num_pages; i++) { |
2571e739 | 5511 | page = eb->pages[i]; |
727011e0 CM |
5512 | if (!PageUptodate(page)) { |
5513 | num_reads++; | |
ce9adaa5 | 5514 | all_uptodate = 0; |
727011e0 | 5515 | } |
ce9adaa5 | 5516 | } |
2571e739 | 5517 | |
ce9adaa5 | 5518 | if (all_uptodate) { |
8436ea91 | 5519 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
5520 | goto unlock_exit; |
5521 | } | |
5522 | ||
656f30db | 5523 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 5524 | eb->read_mirror = 0; |
0b32f4bb | 5525 | atomic_set(&eb->io_pages, num_reads); |
6bf9cd2e BB |
5526 | /* |
5527 | * It is possible for releasepage to clear the TREE_REF bit before we | |
5528 | * set io_pages. See check_buffer_tree_ref for a more detailed comment. | |
5529 | */ | |
5530 | check_buffer_tree_ref(eb); | |
8436ea91 | 5531 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5532 | page = eb->pages[i]; |
baf863b9 | 5533 | |
ce9adaa5 | 5534 | if (!PageUptodate(page)) { |
baf863b9 LB |
5535 | if (ret) { |
5536 | atomic_dec(&eb->io_pages); | |
5537 | unlock_page(page); | |
5538 | continue; | |
5539 | } | |
5540 | ||
f188591e | 5541 | ClearPageError(page); |
0420177c NB |
5542 | err = submit_extent_page(REQ_OP_READ | REQ_META, NULL, |
5543 | page, page_offset(page), PAGE_SIZE, 0, | |
5544 | &bio, end_bio_extent_readpage, | |
5545 | mirror_num, 0, 0, false); | |
baf863b9 | 5546 | if (err) { |
baf863b9 | 5547 | /* |
0420177c NB |
5548 | * We failed to submit the bio so it's the |
5549 | * caller's responsibility to perform cleanup | |
5550 | * i.e unlock page/set error bit. | |
baf863b9 | 5551 | */ |
0420177c NB |
5552 | ret = err; |
5553 | SetPageError(page); | |
5554 | unlock_page(page); | |
baf863b9 LB |
5555 | atomic_dec(&eb->io_pages); |
5556 | } | |
d1310b2e CM |
5557 | } else { |
5558 | unlock_page(page); | |
5559 | } | |
5560 | } | |
5561 | ||
355808c2 | 5562 | if (bio) { |
1f7ad75b | 5563 | err = submit_one_bio(bio, mirror_num, bio_flags); |
79787eaa JM |
5564 | if (err) |
5565 | return err; | |
355808c2 | 5566 | } |
a86c12c7 | 5567 | |
bb82ab88 | 5568 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 5569 | return ret; |
d397712b | 5570 | |
8436ea91 | 5571 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5572 | page = eb->pages[i]; |
d1310b2e | 5573 | wait_on_page_locked(page); |
d397712b | 5574 | if (!PageUptodate(page)) |
d1310b2e | 5575 | ret = -EIO; |
d1310b2e | 5576 | } |
d397712b | 5577 | |
d1310b2e | 5578 | return ret; |
ce9adaa5 CM |
5579 | |
5580 | unlock_exit: | |
d397712b | 5581 | while (locked_pages > 0) { |
ce9adaa5 | 5582 | locked_pages--; |
8436ea91 JB |
5583 | page = eb->pages[locked_pages]; |
5584 | unlock_page(page); | |
ce9adaa5 CM |
5585 | } |
5586 | return ret; | |
d1310b2e | 5587 | } |
d1310b2e | 5588 | |
f98b6215 QW |
5589 | static bool report_eb_range(const struct extent_buffer *eb, unsigned long start, |
5590 | unsigned long len) | |
5591 | { | |
5592 | btrfs_warn(eb->fs_info, | |
5593 | "access to eb bytenr %llu len %lu out of range start %lu len %lu", | |
5594 | eb->start, eb->len, start, len); | |
5595 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
5596 | ||
5597 | return true; | |
5598 | } | |
5599 | ||
5600 | /* | |
5601 | * Check if the [start, start + len) range is valid before reading/writing | |
5602 | * the eb. | |
5603 | * NOTE: @start and @len are offset inside the eb, not logical address. | |
5604 | * | |
5605 | * Caller should not touch the dst/src memory if this function returns error. | |
5606 | */ | |
5607 | static inline int check_eb_range(const struct extent_buffer *eb, | |
5608 | unsigned long start, unsigned long len) | |
5609 | { | |
5610 | unsigned long offset; | |
5611 | ||
5612 | /* start, start + len should not go beyond eb->len nor overflow */ | |
5613 | if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len)) | |
5614 | return report_eb_range(eb, start, len); | |
5615 | ||
5616 | return false; | |
5617 | } | |
5618 | ||
1cbb1f45 JM |
5619 | void read_extent_buffer(const struct extent_buffer *eb, void *dstv, |
5620 | unsigned long start, unsigned long len) | |
d1310b2e CM |
5621 | { |
5622 | size_t cur; | |
5623 | size_t offset; | |
5624 | struct page *page; | |
5625 | char *kaddr; | |
5626 | char *dst = (char *)dstv; | |
c60ac0ff | 5627 | unsigned long i = start >> PAGE_SHIFT; |
d1310b2e | 5628 | |
f98b6215 | 5629 | if (check_eb_range(eb, start, len)) |
f716abd5 | 5630 | return; |
d1310b2e | 5631 | |
c60ac0ff | 5632 | offset = offset_in_page(start); |
d1310b2e | 5633 | |
d397712b | 5634 | while (len > 0) { |
fb85fc9a | 5635 | page = eb->pages[i]; |
d1310b2e | 5636 | |
09cbfeaf | 5637 | cur = min(len, (PAGE_SIZE - offset)); |
a6591715 | 5638 | kaddr = page_address(page); |
d1310b2e | 5639 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
5640 | |
5641 | dst += cur; | |
5642 | len -= cur; | |
5643 | offset = 0; | |
5644 | i++; | |
5645 | } | |
5646 | } | |
d1310b2e | 5647 | |
a48b73ec JB |
5648 | int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb, |
5649 | void __user *dstv, | |
5650 | unsigned long start, unsigned long len) | |
550ac1d8 GH |
5651 | { |
5652 | size_t cur; | |
5653 | size_t offset; | |
5654 | struct page *page; | |
5655 | char *kaddr; | |
5656 | char __user *dst = (char __user *)dstv; | |
c60ac0ff | 5657 | unsigned long i = start >> PAGE_SHIFT; |
550ac1d8 GH |
5658 | int ret = 0; |
5659 | ||
5660 | WARN_ON(start > eb->len); | |
5661 | WARN_ON(start + len > eb->start + eb->len); | |
5662 | ||
c60ac0ff | 5663 | offset = offset_in_page(start); |
550ac1d8 GH |
5664 | |
5665 | while (len > 0) { | |
fb85fc9a | 5666 | page = eb->pages[i]; |
550ac1d8 | 5667 | |
09cbfeaf | 5668 | cur = min(len, (PAGE_SIZE - offset)); |
550ac1d8 | 5669 | kaddr = page_address(page); |
a48b73ec | 5670 | if (copy_to_user_nofault(dst, kaddr + offset, cur)) { |
550ac1d8 GH |
5671 | ret = -EFAULT; |
5672 | break; | |
5673 | } | |
5674 | ||
5675 | dst += cur; | |
5676 | len -= cur; | |
5677 | offset = 0; | |
5678 | i++; | |
5679 | } | |
5680 | ||
5681 | return ret; | |
5682 | } | |
5683 | ||
1cbb1f45 JM |
5684 | int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv, |
5685 | unsigned long start, unsigned long len) | |
d1310b2e CM |
5686 | { |
5687 | size_t cur; | |
5688 | size_t offset; | |
5689 | struct page *page; | |
5690 | char *kaddr; | |
5691 | char *ptr = (char *)ptrv; | |
c60ac0ff | 5692 | unsigned long i = start >> PAGE_SHIFT; |
d1310b2e CM |
5693 | int ret = 0; |
5694 | ||
f98b6215 QW |
5695 | if (check_eb_range(eb, start, len)) |
5696 | return -EINVAL; | |
d1310b2e | 5697 | |
c60ac0ff | 5698 | offset = offset_in_page(start); |
d1310b2e | 5699 | |
d397712b | 5700 | while (len > 0) { |
fb85fc9a | 5701 | page = eb->pages[i]; |
d1310b2e | 5702 | |
09cbfeaf | 5703 | cur = min(len, (PAGE_SIZE - offset)); |
d1310b2e | 5704 | |
a6591715 | 5705 | kaddr = page_address(page); |
d1310b2e | 5706 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
5707 | if (ret) |
5708 | break; | |
5709 | ||
5710 | ptr += cur; | |
5711 | len -= cur; | |
5712 | offset = 0; | |
5713 | i++; | |
5714 | } | |
5715 | return ret; | |
5716 | } | |
d1310b2e | 5717 | |
2b48966a | 5718 | void write_extent_buffer_chunk_tree_uuid(const struct extent_buffer *eb, |
f157bf76 DS |
5719 | const void *srcv) |
5720 | { | |
5721 | char *kaddr; | |
5722 | ||
5723 | WARN_ON(!PageUptodate(eb->pages[0])); | |
5724 | kaddr = page_address(eb->pages[0]); | |
5725 | memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv, | |
5726 | BTRFS_FSID_SIZE); | |
5727 | } | |
5728 | ||
2b48966a | 5729 | void write_extent_buffer_fsid(const struct extent_buffer *eb, const void *srcv) |
f157bf76 DS |
5730 | { |
5731 | char *kaddr; | |
5732 | ||
5733 | WARN_ON(!PageUptodate(eb->pages[0])); | |
5734 | kaddr = page_address(eb->pages[0]); | |
5735 | memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv, | |
5736 | BTRFS_FSID_SIZE); | |
5737 | } | |
5738 | ||
2b48966a | 5739 | void write_extent_buffer(const struct extent_buffer *eb, const void *srcv, |
d1310b2e CM |
5740 | unsigned long start, unsigned long len) |
5741 | { | |
5742 | size_t cur; | |
5743 | size_t offset; | |
5744 | struct page *page; | |
5745 | char *kaddr; | |
5746 | char *src = (char *)srcv; | |
c60ac0ff | 5747 | unsigned long i = start >> PAGE_SHIFT; |
d1310b2e | 5748 | |
f98b6215 QW |
5749 | if (check_eb_range(eb, start, len)) |
5750 | return; | |
d1310b2e | 5751 | |
c60ac0ff | 5752 | offset = offset_in_page(start); |
d1310b2e | 5753 | |
d397712b | 5754 | while (len > 0) { |
fb85fc9a | 5755 | page = eb->pages[i]; |
d1310b2e CM |
5756 | WARN_ON(!PageUptodate(page)); |
5757 | ||
09cbfeaf | 5758 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 5759 | kaddr = page_address(page); |
d1310b2e | 5760 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
5761 | |
5762 | src += cur; | |
5763 | len -= cur; | |
5764 | offset = 0; | |
5765 | i++; | |
5766 | } | |
5767 | } | |
d1310b2e | 5768 | |
2b48966a | 5769 | void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start, |
b159fa28 | 5770 | unsigned long len) |
d1310b2e CM |
5771 | { |
5772 | size_t cur; | |
5773 | size_t offset; | |
5774 | struct page *page; | |
5775 | char *kaddr; | |
c60ac0ff | 5776 | unsigned long i = start >> PAGE_SHIFT; |
d1310b2e | 5777 | |
f98b6215 QW |
5778 | if (check_eb_range(eb, start, len)) |
5779 | return; | |
d1310b2e | 5780 | |
c60ac0ff | 5781 | offset = offset_in_page(start); |
d1310b2e | 5782 | |
d397712b | 5783 | while (len > 0) { |
fb85fc9a | 5784 | page = eb->pages[i]; |
d1310b2e CM |
5785 | WARN_ON(!PageUptodate(page)); |
5786 | ||
09cbfeaf | 5787 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 5788 | kaddr = page_address(page); |
b159fa28 | 5789 | memset(kaddr + offset, 0, cur); |
d1310b2e CM |
5790 | |
5791 | len -= cur; | |
5792 | offset = 0; | |
5793 | i++; | |
5794 | } | |
5795 | } | |
d1310b2e | 5796 | |
2b48966a DS |
5797 | void copy_extent_buffer_full(const struct extent_buffer *dst, |
5798 | const struct extent_buffer *src) | |
58e8012c DS |
5799 | { |
5800 | int i; | |
cc5e31a4 | 5801 | int num_pages; |
58e8012c DS |
5802 | |
5803 | ASSERT(dst->len == src->len); | |
5804 | ||
65ad0104 | 5805 | num_pages = num_extent_pages(dst); |
58e8012c DS |
5806 | for (i = 0; i < num_pages; i++) |
5807 | copy_page(page_address(dst->pages[i]), | |
5808 | page_address(src->pages[i])); | |
5809 | } | |
5810 | ||
2b48966a DS |
5811 | void copy_extent_buffer(const struct extent_buffer *dst, |
5812 | const struct extent_buffer *src, | |
d1310b2e CM |
5813 | unsigned long dst_offset, unsigned long src_offset, |
5814 | unsigned long len) | |
5815 | { | |
5816 | u64 dst_len = dst->len; | |
5817 | size_t cur; | |
5818 | size_t offset; | |
5819 | struct page *page; | |
5820 | char *kaddr; | |
c60ac0ff | 5821 | unsigned long i = dst_offset >> PAGE_SHIFT; |
d1310b2e | 5822 | |
f98b6215 QW |
5823 | if (check_eb_range(dst, dst_offset, len) || |
5824 | check_eb_range(src, src_offset, len)) | |
5825 | return; | |
5826 | ||
d1310b2e CM |
5827 | WARN_ON(src->len != dst_len); |
5828 | ||
c60ac0ff | 5829 | offset = offset_in_page(dst_offset); |
d1310b2e | 5830 | |
d397712b | 5831 | while (len > 0) { |
fb85fc9a | 5832 | page = dst->pages[i]; |
d1310b2e CM |
5833 | WARN_ON(!PageUptodate(page)); |
5834 | ||
09cbfeaf | 5835 | cur = min(len, (unsigned long)(PAGE_SIZE - offset)); |
d1310b2e | 5836 | |
a6591715 | 5837 | kaddr = page_address(page); |
d1310b2e | 5838 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
5839 | |
5840 | src_offset += cur; | |
5841 | len -= cur; | |
5842 | offset = 0; | |
5843 | i++; | |
5844 | } | |
5845 | } | |
d1310b2e | 5846 | |
3e1e8bb7 OS |
5847 | /* |
5848 | * eb_bitmap_offset() - calculate the page and offset of the byte containing the | |
5849 | * given bit number | |
5850 | * @eb: the extent buffer | |
5851 | * @start: offset of the bitmap item in the extent buffer | |
5852 | * @nr: bit number | |
5853 | * @page_index: return index of the page in the extent buffer that contains the | |
5854 | * given bit number | |
5855 | * @page_offset: return offset into the page given by page_index | |
5856 | * | |
5857 | * This helper hides the ugliness of finding the byte in an extent buffer which | |
5858 | * contains a given bit. | |
5859 | */ | |
2b48966a | 5860 | static inline void eb_bitmap_offset(const struct extent_buffer *eb, |
3e1e8bb7 OS |
5861 | unsigned long start, unsigned long nr, |
5862 | unsigned long *page_index, | |
5863 | size_t *page_offset) | |
5864 | { | |
3e1e8bb7 OS |
5865 | size_t byte_offset = BIT_BYTE(nr); |
5866 | size_t offset; | |
5867 | ||
5868 | /* | |
5869 | * The byte we want is the offset of the extent buffer + the offset of | |
5870 | * the bitmap item in the extent buffer + the offset of the byte in the | |
5871 | * bitmap item. | |
5872 | */ | |
c60ac0ff | 5873 | offset = start + byte_offset; |
3e1e8bb7 | 5874 | |
09cbfeaf | 5875 | *page_index = offset >> PAGE_SHIFT; |
7073017a | 5876 | *page_offset = offset_in_page(offset); |
3e1e8bb7 OS |
5877 | } |
5878 | ||
5879 | /** | |
5880 | * extent_buffer_test_bit - determine whether a bit in a bitmap item is set | |
5881 | * @eb: the extent buffer | |
5882 | * @start: offset of the bitmap item in the extent buffer | |
5883 | * @nr: bit number to test | |
5884 | */ | |
2b48966a | 5885 | int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start, |
3e1e8bb7 OS |
5886 | unsigned long nr) |
5887 | { | |
2fe1d551 | 5888 | u8 *kaddr; |
3e1e8bb7 OS |
5889 | struct page *page; |
5890 | unsigned long i; | |
5891 | size_t offset; | |
5892 | ||
5893 | eb_bitmap_offset(eb, start, nr, &i, &offset); | |
5894 | page = eb->pages[i]; | |
5895 | WARN_ON(!PageUptodate(page)); | |
5896 | kaddr = page_address(page); | |
5897 | return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1))); | |
5898 | } | |
5899 | ||
5900 | /** | |
5901 | * extent_buffer_bitmap_set - set an area of a bitmap | |
5902 | * @eb: the extent buffer | |
5903 | * @start: offset of the bitmap item in the extent buffer | |
5904 | * @pos: bit number of the first bit | |
5905 | * @len: number of bits to set | |
5906 | */ | |
2b48966a | 5907 | void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start, |
3e1e8bb7 OS |
5908 | unsigned long pos, unsigned long len) |
5909 | { | |
2fe1d551 | 5910 | u8 *kaddr; |
3e1e8bb7 OS |
5911 | struct page *page; |
5912 | unsigned long i; | |
5913 | size_t offset; | |
5914 | const unsigned int size = pos + len; | |
5915 | int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 5916 | u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
5917 | |
5918 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5919 | page = eb->pages[i]; | |
5920 | WARN_ON(!PageUptodate(page)); | |
5921 | kaddr = page_address(page); | |
5922 | ||
5923 | while (len >= bits_to_set) { | |
5924 | kaddr[offset] |= mask_to_set; | |
5925 | len -= bits_to_set; | |
5926 | bits_to_set = BITS_PER_BYTE; | |
9c894696 | 5927 | mask_to_set = ~0; |
09cbfeaf | 5928 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5929 | offset = 0; |
5930 | page = eb->pages[++i]; | |
5931 | WARN_ON(!PageUptodate(page)); | |
5932 | kaddr = page_address(page); | |
5933 | } | |
5934 | } | |
5935 | if (len) { | |
5936 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
5937 | kaddr[offset] |= mask_to_set; | |
5938 | } | |
5939 | } | |
5940 | ||
5941 | ||
5942 | /** | |
5943 | * extent_buffer_bitmap_clear - clear an area of a bitmap | |
5944 | * @eb: the extent buffer | |
5945 | * @start: offset of the bitmap item in the extent buffer | |
5946 | * @pos: bit number of the first bit | |
5947 | * @len: number of bits to clear | |
5948 | */ | |
2b48966a DS |
5949 | void extent_buffer_bitmap_clear(const struct extent_buffer *eb, |
5950 | unsigned long start, unsigned long pos, | |
5951 | unsigned long len) | |
3e1e8bb7 | 5952 | { |
2fe1d551 | 5953 | u8 *kaddr; |
3e1e8bb7 OS |
5954 | struct page *page; |
5955 | unsigned long i; | |
5956 | size_t offset; | |
5957 | const unsigned int size = pos + len; | |
5958 | int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 5959 | u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
5960 | |
5961 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5962 | page = eb->pages[i]; | |
5963 | WARN_ON(!PageUptodate(page)); | |
5964 | kaddr = page_address(page); | |
5965 | ||
5966 | while (len >= bits_to_clear) { | |
5967 | kaddr[offset] &= ~mask_to_clear; | |
5968 | len -= bits_to_clear; | |
5969 | bits_to_clear = BITS_PER_BYTE; | |
9c894696 | 5970 | mask_to_clear = ~0; |
09cbfeaf | 5971 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5972 | offset = 0; |
5973 | page = eb->pages[++i]; | |
5974 | WARN_ON(!PageUptodate(page)); | |
5975 | kaddr = page_address(page); | |
5976 | } | |
5977 | } | |
5978 | if (len) { | |
5979 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
5980 | kaddr[offset] &= ~mask_to_clear; | |
5981 | } | |
5982 | } | |
5983 | ||
3387206f ST |
5984 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
5985 | { | |
5986 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
5987 | return distance < len; | |
5988 | } | |
5989 | ||
d1310b2e CM |
5990 | static void copy_pages(struct page *dst_page, struct page *src_page, |
5991 | unsigned long dst_off, unsigned long src_off, | |
5992 | unsigned long len) | |
5993 | { | |
a6591715 | 5994 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 5995 | char *src_kaddr; |
727011e0 | 5996 | int must_memmove = 0; |
d1310b2e | 5997 | |
3387206f | 5998 | if (dst_page != src_page) { |
a6591715 | 5999 | src_kaddr = page_address(src_page); |
3387206f | 6000 | } else { |
d1310b2e | 6001 | src_kaddr = dst_kaddr; |
727011e0 CM |
6002 | if (areas_overlap(src_off, dst_off, len)) |
6003 | must_memmove = 1; | |
3387206f | 6004 | } |
d1310b2e | 6005 | |
727011e0 CM |
6006 | if (must_memmove) |
6007 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
6008 | else | |
6009 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
6010 | } |
6011 | ||
2b48966a DS |
6012 | void memcpy_extent_buffer(const struct extent_buffer *dst, |
6013 | unsigned long dst_offset, unsigned long src_offset, | |
6014 | unsigned long len) | |
d1310b2e CM |
6015 | { |
6016 | size_t cur; | |
6017 | size_t dst_off_in_page; | |
6018 | size_t src_off_in_page; | |
d1310b2e CM |
6019 | unsigned long dst_i; |
6020 | unsigned long src_i; | |
6021 | ||
f98b6215 QW |
6022 | if (check_eb_range(dst, dst_offset, len) || |
6023 | check_eb_range(dst, src_offset, len)) | |
6024 | return; | |
d1310b2e | 6025 | |
d397712b | 6026 | while (len > 0) { |
c60ac0ff DS |
6027 | dst_off_in_page = offset_in_page(dst_offset); |
6028 | src_off_in_page = offset_in_page(src_offset); | |
d1310b2e | 6029 | |
c60ac0ff DS |
6030 | dst_i = dst_offset >> PAGE_SHIFT; |
6031 | src_i = src_offset >> PAGE_SHIFT; | |
d1310b2e | 6032 | |
09cbfeaf | 6033 | cur = min(len, (unsigned long)(PAGE_SIZE - |
d1310b2e CM |
6034 | src_off_in_page)); |
6035 | cur = min_t(unsigned long, cur, | |
09cbfeaf | 6036 | (unsigned long)(PAGE_SIZE - dst_off_in_page)); |
d1310b2e | 6037 | |
fb85fc9a | 6038 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
6039 | dst_off_in_page, src_off_in_page, cur); |
6040 | ||
6041 | src_offset += cur; | |
6042 | dst_offset += cur; | |
6043 | len -= cur; | |
6044 | } | |
6045 | } | |
d1310b2e | 6046 | |
2b48966a DS |
6047 | void memmove_extent_buffer(const struct extent_buffer *dst, |
6048 | unsigned long dst_offset, unsigned long src_offset, | |
6049 | unsigned long len) | |
d1310b2e CM |
6050 | { |
6051 | size_t cur; | |
6052 | size_t dst_off_in_page; | |
6053 | size_t src_off_in_page; | |
6054 | unsigned long dst_end = dst_offset + len - 1; | |
6055 | unsigned long src_end = src_offset + len - 1; | |
d1310b2e CM |
6056 | unsigned long dst_i; |
6057 | unsigned long src_i; | |
6058 | ||
f98b6215 QW |
6059 | if (check_eb_range(dst, dst_offset, len) || |
6060 | check_eb_range(dst, src_offset, len)) | |
6061 | return; | |
727011e0 | 6062 | if (dst_offset < src_offset) { |
d1310b2e CM |
6063 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
6064 | return; | |
6065 | } | |
d397712b | 6066 | while (len > 0) { |
c60ac0ff DS |
6067 | dst_i = dst_end >> PAGE_SHIFT; |
6068 | src_i = src_end >> PAGE_SHIFT; | |
d1310b2e | 6069 | |
c60ac0ff DS |
6070 | dst_off_in_page = offset_in_page(dst_end); |
6071 | src_off_in_page = offset_in_page(src_end); | |
d1310b2e CM |
6072 | |
6073 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
6074 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 6075 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
6076 | dst_off_in_page - cur + 1, |
6077 | src_off_in_page - cur + 1, cur); | |
6078 | ||
6079 | dst_end -= cur; | |
6080 | src_end -= cur; | |
6081 | len -= cur; | |
6082 | } | |
6083 | } | |
6af118ce | 6084 | |
f7a52a40 | 6085 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 6086 | { |
6af118ce | 6087 | struct extent_buffer *eb; |
6af118ce | 6088 | |
3083ee2e | 6089 | /* |
01327610 | 6090 | * We need to make sure nobody is attaching this page to an eb right |
3083ee2e JB |
6091 | * now. |
6092 | */ | |
6093 | spin_lock(&page->mapping->private_lock); | |
6094 | if (!PagePrivate(page)) { | |
6095 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 6096 | return 1; |
45f49bce | 6097 | } |
6af118ce | 6098 | |
3083ee2e JB |
6099 | eb = (struct extent_buffer *)page->private; |
6100 | BUG_ON(!eb); | |
19fe0a8b MX |
6101 | |
6102 | /* | |
3083ee2e JB |
6103 | * This is a little awful but should be ok, we need to make sure that |
6104 | * the eb doesn't disappear out from under us while we're looking at | |
6105 | * this page. | |
19fe0a8b | 6106 | */ |
3083ee2e | 6107 | spin_lock(&eb->refs_lock); |
0b32f4bb | 6108 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
6109 | spin_unlock(&eb->refs_lock); |
6110 | spin_unlock(&page->mapping->private_lock); | |
6111 | return 0; | |
b9473439 | 6112 | } |
3083ee2e | 6113 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 6114 | |
19fe0a8b | 6115 | /* |
3083ee2e JB |
6116 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
6117 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 6118 | */ |
3083ee2e JB |
6119 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
6120 | spin_unlock(&eb->refs_lock); | |
6121 | return 0; | |
b9473439 | 6122 | } |
19fe0a8b | 6123 | |
f7a52a40 | 6124 | return release_extent_buffer(eb); |
6af118ce | 6125 | } |
bfb484d9 JB |
6126 | |
6127 | /* | |
6128 | * btrfs_readahead_tree_block - attempt to readahead a child block | |
6129 | * @fs_info: the fs_info | |
6130 | * @bytenr: bytenr to read | |
3fbaf258 | 6131 | * @owner_root: objectid of the root that owns this eb |
bfb484d9 | 6132 | * @gen: generation for the uptodate check, can be 0 |
3fbaf258 | 6133 | * @level: level for the eb |
bfb484d9 JB |
6134 | * |
6135 | * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a | |
6136 | * normal uptodate check of the eb, without checking the generation. If we have | |
6137 | * to read the block we will not block on anything. | |
6138 | */ | |
6139 | void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info, | |
3fbaf258 | 6140 | u64 bytenr, u64 owner_root, u64 gen, int level) |
bfb484d9 JB |
6141 | { |
6142 | struct extent_buffer *eb; | |
6143 | int ret; | |
6144 | ||
3fbaf258 | 6145 | eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level); |
bfb484d9 JB |
6146 | if (IS_ERR(eb)) |
6147 | return; | |
6148 | ||
6149 | if (btrfs_buffer_uptodate(eb, gen, 1)) { | |
6150 | free_extent_buffer(eb); | |
6151 | return; | |
6152 | } | |
6153 | ||
6154 | ret = read_extent_buffer_pages(eb, WAIT_NONE, 0); | |
6155 | if (ret < 0) | |
6156 | free_extent_buffer_stale(eb); | |
6157 | else | |
6158 | free_extent_buffer(eb); | |
6159 | } | |
6160 | ||
6161 | /* | |
6162 | * btrfs_readahead_node_child - readahead a node's child block | |
6163 | * @node: parent node we're reading from | |
6164 | * @slot: slot in the parent node for the child we want to read | |
6165 | * | |
6166 | * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at | |
6167 | * the slot in the node provided. | |
6168 | */ | |
6169 | void btrfs_readahead_node_child(struct extent_buffer *node, int slot) | |
6170 | { | |
6171 | btrfs_readahead_tree_block(node->fs_info, | |
6172 | btrfs_node_blockptr(node, slot), | |
3fbaf258 JB |
6173 | btrfs_header_owner(node), |
6174 | btrfs_node_ptr_generation(node, slot), | |
6175 | btrfs_header_level(node) - 1); | |
bfb484d9 | 6176 | } |