]>
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 | ||
bb58eb9e QW |
163 | static int __must_check submit_one_bio(struct bio *bio, int mirror_num, |
164 | unsigned long bio_flags) | |
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 | ||
171 | if (tree->ops) | |
172 | ret = tree->ops->submit_bio_hook(tree->private_data, bio, | |
50489a57 | 173 | mirror_num, bio_flags); |
bb58eb9e QW |
174 | else |
175 | btrfsic_submit_bio(bio); | |
176 | ||
177 | return blk_status_to_errno(ret); | |
178 | } | |
179 | ||
3065976b QW |
180 | /* Cleanup unsubmitted bios */ |
181 | static void end_write_bio(struct extent_page_data *epd, int ret) | |
182 | { | |
183 | if (epd->bio) { | |
184 | epd->bio->bi_status = errno_to_blk_status(ret); | |
185 | bio_endio(epd->bio); | |
186 | epd->bio = NULL; | |
187 | } | |
188 | } | |
189 | ||
f4340622 QW |
190 | /* |
191 | * Submit bio from extent page data via submit_one_bio | |
192 | * | |
193 | * Return 0 if everything is OK. | |
194 | * Return <0 for error. | |
195 | */ | |
196 | static int __must_check flush_write_bio(struct extent_page_data *epd) | |
bb58eb9e | 197 | { |
f4340622 | 198 | int ret = 0; |
bb58eb9e | 199 | |
f4340622 | 200 | if (epd->bio) { |
bb58eb9e | 201 | ret = submit_one_bio(epd->bio, 0, 0); |
f4340622 QW |
202 | /* |
203 | * Clean up of epd->bio is handled by its endio function. | |
204 | * And endio is either triggered by successful bio execution | |
205 | * or the error handler of submit bio hook. | |
206 | * So at this point, no matter what happened, we don't need | |
207 | * to clean up epd->bio. | |
208 | */ | |
bb58eb9e QW |
209 | epd->bio = NULL; |
210 | } | |
f4340622 | 211 | return ret; |
bb58eb9e | 212 | } |
e2932ee0 | 213 | |
6f0d04f8 | 214 | int __init extent_state_cache_init(void) |
d1310b2e | 215 | { |
837e1972 | 216 | extent_state_cache = kmem_cache_create("btrfs_extent_state", |
9601e3f6 | 217 | sizeof(struct extent_state), 0, |
fba4b697 | 218 | SLAB_MEM_SPREAD, NULL); |
d1310b2e CM |
219 | if (!extent_state_cache) |
220 | return -ENOMEM; | |
6f0d04f8 JB |
221 | return 0; |
222 | } | |
d1310b2e | 223 | |
6f0d04f8 JB |
224 | int __init extent_io_init(void) |
225 | { | |
837e1972 | 226 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 | 227 | sizeof(struct extent_buffer), 0, |
fba4b697 | 228 | SLAB_MEM_SPREAD, NULL); |
d1310b2e | 229 | if (!extent_buffer_cache) |
6f0d04f8 | 230 | return -ENOMEM; |
9be3395b | 231 | |
8ac9f7c1 KO |
232 | if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE, |
233 | offsetof(struct btrfs_io_bio, bio), | |
234 | BIOSET_NEED_BVECS)) | |
9be3395b | 235 | goto free_buffer_cache; |
b208c2f7 | 236 | |
8ac9f7c1 | 237 | if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE)) |
b208c2f7 DW |
238 | goto free_bioset; |
239 | ||
d1310b2e CM |
240 | return 0; |
241 | ||
b208c2f7 | 242 | free_bioset: |
8ac9f7c1 | 243 | bioset_exit(&btrfs_bioset); |
b208c2f7 | 244 | |
9be3395b CM |
245 | free_buffer_cache: |
246 | kmem_cache_destroy(extent_buffer_cache); | |
247 | extent_buffer_cache = NULL; | |
6f0d04f8 JB |
248 | return -ENOMEM; |
249 | } | |
9be3395b | 250 | |
6f0d04f8 JB |
251 | void __cold extent_state_cache_exit(void) |
252 | { | |
253 | btrfs_extent_state_leak_debug_check(); | |
d1310b2e | 254 | kmem_cache_destroy(extent_state_cache); |
d1310b2e CM |
255 | } |
256 | ||
e67c718b | 257 | void __cold extent_io_exit(void) |
d1310b2e | 258 | { |
8c0a8537 KS |
259 | /* |
260 | * Make sure all delayed rcu free are flushed before we | |
261 | * destroy caches. | |
262 | */ | |
263 | rcu_barrier(); | |
5598e900 | 264 | kmem_cache_destroy(extent_buffer_cache); |
8ac9f7c1 | 265 | bioset_exit(&btrfs_bioset); |
d1310b2e CM |
266 | } |
267 | ||
41a2ee75 JB |
268 | /* |
269 | * For the file_extent_tree, we want to hold the inode lock when we lookup and | |
270 | * update the disk_i_size, but lockdep will complain because our io_tree we hold | |
271 | * the tree lock and get the inode lock when setting delalloc. These two things | |
272 | * are unrelated, so make a class for the file_extent_tree so we don't get the | |
273 | * two locking patterns mixed up. | |
274 | */ | |
275 | static struct lock_class_key file_extent_tree_class; | |
276 | ||
c258d6e3 | 277 | void extent_io_tree_init(struct btrfs_fs_info *fs_info, |
43eb5f29 QW |
278 | struct extent_io_tree *tree, unsigned int owner, |
279 | void *private_data) | |
d1310b2e | 280 | { |
c258d6e3 | 281 | tree->fs_info = fs_info; |
6bef4d31 | 282 | tree->state = RB_ROOT; |
d1310b2e CM |
283 | tree->ops = NULL; |
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 | |
d1310b2e | 983 | again: |
d0164adc | 984 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
059f791c DS |
985 | /* |
986 | * Don't care for allocation failure here because we might end | |
987 | * up not needing the pre-allocated extent state at all, which | |
988 | * is the case if we only have in the tree extent states that | |
989 | * cover our input range and don't cover too any other range. | |
990 | * If we end up needing a new extent state we allocate it later. | |
991 | */ | |
d1310b2e | 992 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
993 | } |
994 | ||
cad321ad | 995 | spin_lock(&tree->lock); |
9655d298 CM |
996 | if (cached_state && *cached_state) { |
997 | state = *cached_state; | |
df98b6e2 | 998 | if (state->start <= start && state->end > start && |
27a3507d | 999 | extent_state_in_tree(state)) { |
9655d298 CM |
1000 | node = &state->rb_node; |
1001 | goto hit_next; | |
1002 | } | |
1003 | } | |
d1310b2e CM |
1004 | /* |
1005 | * this search will find all the extents that end after | |
1006 | * our range starts. | |
1007 | */ | |
12cfbad9 | 1008 | node = tree_search_for_insert(tree, start, &p, &parent); |
d1310b2e | 1009 | if (!node) { |
8233767a XG |
1010 | prealloc = alloc_extent_state_atomic(prealloc); |
1011 | BUG_ON(!prealloc); | |
12cfbad9 | 1012 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 1013 | &p, &parent, &bits, changeset); |
c2d904e0 JM |
1014 | if (err) |
1015 | extent_io_tree_panic(tree, err); | |
1016 | ||
c42ac0bc | 1017 | cache_state(prealloc, cached_state); |
d1310b2e | 1018 | prealloc = NULL; |
d1310b2e CM |
1019 | goto out; |
1020 | } | |
d1310b2e | 1021 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 1022 | hit_next: |
d1310b2e CM |
1023 | last_start = state->start; |
1024 | last_end = state->end; | |
1025 | ||
1026 | /* | |
1027 | * | ---- desired range ---- | | |
1028 | * | state | | |
1029 | * | |
1030 | * Just lock what we found and keep going | |
1031 | */ | |
1032 | if (state->start == start && state->end <= end) { | |
1edbb734 | 1033 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1034 | *failed_start = state->start; |
1035 | err = -EEXIST; | |
1036 | goto out; | |
1037 | } | |
42daec29 | 1038 | |
d38ed27f | 1039 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 1040 | cache_state(state, cached_state); |
d1310b2e | 1041 | merge_state(tree, state); |
5c939df5 YZ |
1042 | if (last_end == (u64)-1) |
1043 | goto out; | |
1044 | start = last_end + 1; | |
d1ac6e41 LB |
1045 | state = next_state(state); |
1046 | if (start < end && state && state->start == start && | |
1047 | !need_resched()) | |
1048 | goto hit_next; | |
d1310b2e CM |
1049 | goto search_again; |
1050 | } | |
1051 | ||
1052 | /* | |
1053 | * | ---- desired range ---- | | |
1054 | * | state | | |
1055 | * or | |
1056 | * | ------------- state -------------- | | |
1057 | * | |
1058 | * We need to split the extent we found, and may flip bits on | |
1059 | * second half. | |
1060 | * | |
1061 | * If the extent we found extends past our | |
1062 | * range, we just split and search again. It'll get split | |
1063 | * again the next time though. | |
1064 | * | |
1065 | * If the extent we found is inside our range, we set the | |
1066 | * desired bit on it. | |
1067 | */ | |
1068 | if (state->start < start) { | |
1edbb734 | 1069 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1070 | *failed_start = start; |
1071 | err = -EEXIST; | |
1072 | goto out; | |
1073 | } | |
8233767a | 1074 | |
55ffaabe FM |
1075 | /* |
1076 | * If this extent already has all the bits we want set, then | |
1077 | * skip it, not necessary to split it or do anything with it. | |
1078 | */ | |
1079 | if ((state->state & bits) == bits) { | |
1080 | start = state->end + 1; | |
1081 | cache_state(state, cached_state); | |
1082 | goto search_again; | |
1083 | } | |
1084 | ||
8233767a XG |
1085 | prealloc = alloc_extent_state_atomic(prealloc); |
1086 | BUG_ON(!prealloc); | |
d1310b2e | 1087 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1088 | if (err) |
1089 | extent_io_tree_panic(tree, err); | |
1090 | ||
d1310b2e CM |
1091 | prealloc = NULL; |
1092 | if (err) | |
1093 | goto out; | |
1094 | if (state->end <= end) { | |
d38ed27f | 1095 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 1096 | cache_state(state, cached_state); |
d1310b2e | 1097 | merge_state(tree, state); |
5c939df5 YZ |
1098 | if (last_end == (u64)-1) |
1099 | goto out; | |
1100 | start = last_end + 1; | |
d1ac6e41 LB |
1101 | state = next_state(state); |
1102 | if (start < end && state && state->start == start && | |
1103 | !need_resched()) | |
1104 | goto hit_next; | |
d1310b2e CM |
1105 | } |
1106 | goto search_again; | |
1107 | } | |
1108 | /* | |
1109 | * | ---- desired range ---- | | |
1110 | * | state | or | state | | |
1111 | * | |
1112 | * There's a hole, we need to insert something in it and | |
1113 | * ignore the extent we found. | |
1114 | */ | |
1115 | if (state->start > start) { | |
1116 | u64 this_end; | |
1117 | if (end < last_start) | |
1118 | this_end = end; | |
1119 | else | |
d397712b | 1120 | this_end = last_start - 1; |
8233767a XG |
1121 | |
1122 | prealloc = alloc_extent_state_atomic(prealloc); | |
1123 | BUG_ON(!prealloc); | |
c7f895a2 XG |
1124 | |
1125 | /* | |
1126 | * Avoid to free 'prealloc' if it can be merged with | |
1127 | * the later extent. | |
1128 | */ | |
d1310b2e | 1129 | err = insert_state(tree, prealloc, start, this_end, |
d38ed27f | 1130 | NULL, NULL, &bits, changeset); |
c2d904e0 JM |
1131 | if (err) |
1132 | extent_io_tree_panic(tree, err); | |
1133 | ||
9ed74f2d JB |
1134 | cache_state(prealloc, cached_state); |
1135 | prealloc = NULL; | |
d1310b2e CM |
1136 | start = this_end + 1; |
1137 | goto search_again; | |
1138 | } | |
1139 | /* | |
1140 | * | ---- desired range ---- | | |
1141 | * | state | | |
1142 | * We need to split the extent, and set the bit | |
1143 | * on the first half | |
1144 | */ | |
1145 | if (state->start <= end && state->end > end) { | |
1edbb734 | 1146 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1147 | *failed_start = start; |
1148 | err = -EEXIST; | |
1149 | goto out; | |
1150 | } | |
8233767a XG |
1151 | |
1152 | prealloc = alloc_extent_state_atomic(prealloc); | |
1153 | BUG_ON(!prealloc); | |
d1310b2e | 1154 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
1155 | if (err) |
1156 | extent_io_tree_panic(tree, err); | |
d1310b2e | 1157 | |
d38ed27f | 1158 | set_state_bits(tree, prealloc, &bits, changeset); |
2c64c53d | 1159 | cache_state(prealloc, cached_state); |
d1310b2e CM |
1160 | merge_state(tree, prealloc); |
1161 | prealloc = NULL; | |
1162 | goto out; | |
1163 | } | |
1164 | ||
b5a4ba14 DS |
1165 | search_again: |
1166 | if (start > end) | |
1167 | goto out; | |
1168 | spin_unlock(&tree->lock); | |
1169 | if (gfpflags_allow_blocking(mask)) | |
1170 | cond_resched(); | |
1171 | goto again; | |
d1310b2e CM |
1172 | |
1173 | out: | |
cad321ad | 1174 | spin_unlock(&tree->lock); |
d1310b2e CM |
1175 | if (prealloc) |
1176 | free_extent_state(prealloc); | |
1177 | ||
1178 | return err; | |
1179 | ||
d1310b2e | 1180 | } |
d1310b2e | 1181 | |
41074888 | 1182 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
9ee49a04 | 1183 | unsigned bits, u64 * failed_start, |
41074888 | 1184 | struct extent_state **cached_state, gfp_t mask) |
3fbe5c02 JM |
1185 | { |
1186 | return __set_extent_bit(tree, start, end, bits, 0, failed_start, | |
d38ed27f | 1187 | cached_state, mask, NULL); |
3fbe5c02 JM |
1188 | } |
1189 | ||
1190 | ||
462d6fac | 1191 | /** |
10983f2e LB |
1192 | * convert_extent_bit - convert all bits in a given range from one bit to |
1193 | * another | |
462d6fac JB |
1194 | * @tree: the io tree to search |
1195 | * @start: the start offset in bytes | |
1196 | * @end: the end offset in bytes (inclusive) | |
1197 | * @bits: the bits to set in this range | |
1198 | * @clear_bits: the bits to clear in this range | |
e6138876 | 1199 | * @cached_state: state that we're going to cache |
462d6fac JB |
1200 | * |
1201 | * This will go through and set bits for the given range. If any states exist | |
1202 | * already in this range they are set with the given bit and cleared of the | |
1203 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
1204 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
1205 | * boundary bits like LOCK. | |
210aa277 DS |
1206 | * |
1207 | * All allocations are done with GFP_NOFS. | |
462d6fac JB |
1208 | */ |
1209 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1210 | unsigned bits, unsigned clear_bits, |
210aa277 | 1211 | struct extent_state **cached_state) |
462d6fac JB |
1212 | { |
1213 | struct extent_state *state; | |
1214 | struct extent_state *prealloc = NULL; | |
1215 | struct rb_node *node; | |
12cfbad9 FDBM |
1216 | struct rb_node **p; |
1217 | struct rb_node *parent; | |
462d6fac JB |
1218 | int err = 0; |
1219 | u64 last_start; | |
1220 | u64 last_end; | |
c8fd3de7 | 1221 | bool first_iteration = true; |
462d6fac | 1222 | |
a5dee37d | 1223 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 QW |
1224 | trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits, |
1225 | clear_bits); | |
8d599ae1 | 1226 | |
462d6fac | 1227 | again: |
210aa277 | 1228 | if (!prealloc) { |
c8fd3de7 FM |
1229 | /* |
1230 | * Best effort, don't worry if extent state allocation fails | |
1231 | * here for the first iteration. We might have a cached state | |
1232 | * that matches exactly the target range, in which case no | |
1233 | * extent state allocations are needed. We'll only know this | |
1234 | * after locking the tree. | |
1235 | */ | |
210aa277 | 1236 | prealloc = alloc_extent_state(GFP_NOFS); |
c8fd3de7 | 1237 | if (!prealloc && !first_iteration) |
462d6fac JB |
1238 | return -ENOMEM; |
1239 | } | |
1240 | ||
1241 | spin_lock(&tree->lock); | |
e6138876 JB |
1242 | if (cached_state && *cached_state) { |
1243 | state = *cached_state; | |
1244 | if (state->start <= start && state->end > start && | |
27a3507d | 1245 | extent_state_in_tree(state)) { |
e6138876 JB |
1246 | node = &state->rb_node; |
1247 | goto hit_next; | |
1248 | } | |
1249 | } | |
1250 | ||
462d6fac JB |
1251 | /* |
1252 | * this search will find all the extents that end after | |
1253 | * our range starts. | |
1254 | */ | |
12cfbad9 | 1255 | node = tree_search_for_insert(tree, start, &p, &parent); |
462d6fac JB |
1256 | if (!node) { |
1257 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1258 | if (!prealloc) { |
1259 | err = -ENOMEM; | |
1260 | goto out; | |
1261 | } | |
12cfbad9 | 1262 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 1263 | &p, &parent, &bits, NULL); |
c2d904e0 JM |
1264 | if (err) |
1265 | extent_io_tree_panic(tree, err); | |
c42ac0bc FDBM |
1266 | cache_state(prealloc, cached_state); |
1267 | prealloc = NULL; | |
462d6fac JB |
1268 | goto out; |
1269 | } | |
1270 | state = rb_entry(node, struct extent_state, rb_node); | |
1271 | hit_next: | |
1272 | last_start = state->start; | |
1273 | last_end = state->end; | |
1274 | ||
1275 | /* | |
1276 | * | ---- desired range ---- | | |
1277 | * | state | | |
1278 | * | |
1279 | * Just lock what we found and keep going | |
1280 | */ | |
1281 | if (state->start == start && state->end <= end) { | |
d38ed27f | 1282 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1283 | cache_state(state, cached_state); |
fefdc557 | 1284 | state = clear_state_bit(tree, state, &clear_bits, 0, NULL); |
462d6fac JB |
1285 | if (last_end == (u64)-1) |
1286 | goto out; | |
462d6fac | 1287 | start = last_end + 1; |
d1ac6e41 LB |
1288 | if (start < end && state && state->start == start && |
1289 | !need_resched()) | |
1290 | goto hit_next; | |
462d6fac JB |
1291 | goto search_again; |
1292 | } | |
1293 | ||
1294 | /* | |
1295 | * | ---- desired range ---- | | |
1296 | * | state | | |
1297 | * or | |
1298 | * | ------------- state -------------- | | |
1299 | * | |
1300 | * We need to split the extent we found, and may flip bits on | |
1301 | * second half. | |
1302 | * | |
1303 | * If the extent we found extends past our | |
1304 | * range, we just split and search again. It'll get split | |
1305 | * again the next time though. | |
1306 | * | |
1307 | * If the extent we found is inside our range, we set the | |
1308 | * desired bit on it. | |
1309 | */ | |
1310 | if (state->start < start) { | |
1311 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1312 | if (!prealloc) { |
1313 | err = -ENOMEM; | |
1314 | goto out; | |
1315 | } | |
462d6fac | 1316 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1317 | if (err) |
1318 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1319 | prealloc = NULL; |
1320 | if (err) | |
1321 | goto out; | |
1322 | if (state->end <= end) { | |
d38ed27f | 1323 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1324 | cache_state(state, cached_state); |
fefdc557 QW |
1325 | state = clear_state_bit(tree, state, &clear_bits, 0, |
1326 | NULL); | |
462d6fac JB |
1327 | if (last_end == (u64)-1) |
1328 | goto out; | |
1329 | start = last_end + 1; | |
d1ac6e41 LB |
1330 | if (start < end && state && state->start == start && |
1331 | !need_resched()) | |
1332 | goto hit_next; | |
462d6fac JB |
1333 | } |
1334 | goto search_again; | |
1335 | } | |
1336 | /* | |
1337 | * | ---- desired range ---- | | |
1338 | * | state | or | state | | |
1339 | * | |
1340 | * There's a hole, we need to insert something in it and | |
1341 | * ignore the extent we found. | |
1342 | */ | |
1343 | if (state->start > start) { | |
1344 | u64 this_end; | |
1345 | if (end < last_start) | |
1346 | this_end = end; | |
1347 | else | |
1348 | this_end = last_start - 1; | |
1349 | ||
1350 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1351 | if (!prealloc) { |
1352 | err = -ENOMEM; | |
1353 | goto out; | |
1354 | } | |
462d6fac JB |
1355 | |
1356 | /* | |
1357 | * Avoid to free 'prealloc' if it can be merged with | |
1358 | * the later extent. | |
1359 | */ | |
1360 | err = insert_state(tree, prealloc, start, this_end, | |
d38ed27f | 1361 | NULL, NULL, &bits, NULL); |
c2d904e0 JM |
1362 | if (err) |
1363 | extent_io_tree_panic(tree, err); | |
e6138876 | 1364 | cache_state(prealloc, cached_state); |
462d6fac JB |
1365 | prealloc = NULL; |
1366 | start = this_end + 1; | |
1367 | goto search_again; | |
1368 | } | |
1369 | /* | |
1370 | * | ---- desired range ---- | | |
1371 | * | state | | |
1372 | * We need to split the extent, and set the bit | |
1373 | * on the first half | |
1374 | */ | |
1375 | if (state->start <= end && state->end > end) { | |
1376 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1377 | if (!prealloc) { |
1378 | err = -ENOMEM; | |
1379 | goto out; | |
1380 | } | |
462d6fac JB |
1381 | |
1382 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1383 | if (err) |
1384 | extent_io_tree_panic(tree, err); | |
462d6fac | 1385 | |
d38ed27f | 1386 | set_state_bits(tree, prealloc, &bits, NULL); |
e6138876 | 1387 | cache_state(prealloc, cached_state); |
fefdc557 | 1388 | clear_state_bit(tree, prealloc, &clear_bits, 0, NULL); |
462d6fac JB |
1389 | prealloc = NULL; |
1390 | goto out; | |
1391 | } | |
1392 | ||
462d6fac JB |
1393 | search_again: |
1394 | if (start > end) | |
1395 | goto out; | |
1396 | spin_unlock(&tree->lock); | |
210aa277 | 1397 | cond_resched(); |
c8fd3de7 | 1398 | first_iteration = false; |
462d6fac | 1399 | goto again; |
462d6fac JB |
1400 | |
1401 | out: | |
1402 | spin_unlock(&tree->lock); | |
1403 | if (prealloc) | |
1404 | free_extent_state(prealloc); | |
1405 | ||
1406 | return err; | |
462d6fac JB |
1407 | } |
1408 | ||
d1310b2e | 1409 | /* wrappers around set/clear extent bit */ |
d38ed27f | 1410 | int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
2c53b912 | 1411 | unsigned bits, struct extent_changeset *changeset) |
d38ed27f QW |
1412 | { |
1413 | /* | |
1414 | * We don't support EXTENT_LOCKED yet, as current changeset will | |
1415 | * record any bits changed, so for EXTENT_LOCKED case, it will | |
1416 | * either fail with -EEXIST or changeset will record the whole | |
1417 | * range. | |
1418 | */ | |
1419 | BUG_ON(bits & EXTENT_LOCKED); | |
1420 | ||
2c53b912 | 1421 | return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS, |
d38ed27f QW |
1422 | changeset); |
1423 | } | |
1424 | ||
4ca73656 NB |
1425 | int set_extent_bits_nowait(struct extent_io_tree *tree, u64 start, u64 end, |
1426 | unsigned bits) | |
1427 | { | |
1428 | return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, | |
1429 | GFP_NOWAIT, NULL); | |
1430 | } | |
1431 | ||
fefdc557 QW |
1432 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
1433 | unsigned bits, int wake, int delete, | |
ae0f1625 | 1434 | struct extent_state **cached) |
fefdc557 QW |
1435 | { |
1436 | return __clear_extent_bit(tree, start, end, bits, wake, delete, | |
ae0f1625 | 1437 | cached, GFP_NOFS, NULL); |
fefdc557 QW |
1438 | } |
1439 | ||
fefdc557 | 1440 | int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
f734c44a | 1441 | unsigned bits, struct extent_changeset *changeset) |
fefdc557 QW |
1442 | { |
1443 | /* | |
1444 | * Don't support EXTENT_LOCKED case, same reason as | |
1445 | * set_record_extent_bits(). | |
1446 | */ | |
1447 | BUG_ON(bits & EXTENT_LOCKED); | |
1448 | ||
f734c44a | 1449 | return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS, |
fefdc557 QW |
1450 | changeset); |
1451 | } | |
1452 | ||
d352ac68 CM |
1453 | /* |
1454 | * either insert or lock state struct between start and end use mask to tell | |
1455 | * us if waiting is desired. | |
1456 | */ | |
1edbb734 | 1457 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
ff13db41 | 1458 | struct extent_state **cached_state) |
d1310b2e CM |
1459 | { |
1460 | int err; | |
1461 | u64 failed_start; | |
9ee49a04 | 1462 | |
d1310b2e | 1463 | while (1) { |
ff13db41 | 1464 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, |
3fbe5c02 | 1465 | EXTENT_LOCKED, &failed_start, |
d38ed27f | 1466 | cached_state, GFP_NOFS, NULL); |
d0082371 | 1467 | if (err == -EEXIST) { |
d1310b2e CM |
1468 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1469 | start = failed_start; | |
d0082371 | 1470 | } else |
d1310b2e | 1471 | break; |
d1310b2e CM |
1472 | WARN_ON(start > end); |
1473 | } | |
1474 | return err; | |
1475 | } | |
d1310b2e | 1476 | |
d0082371 | 1477 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1478 | { |
1479 | int err; | |
1480 | u64 failed_start; | |
1481 | ||
3fbe5c02 | 1482 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
d38ed27f | 1483 | &failed_start, NULL, GFP_NOFS, NULL); |
6643558d YZ |
1484 | if (err == -EEXIST) { |
1485 | if (failed_start > start) | |
1486 | clear_extent_bit(tree, start, failed_start - 1, | |
ae0f1625 | 1487 | EXTENT_LOCKED, 1, 0, NULL); |
25179201 | 1488 | return 0; |
6643558d | 1489 | } |
25179201 JB |
1490 | return 1; |
1491 | } | |
25179201 | 1492 | |
bd1fa4f0 | 1493 | void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1494 | { |
09cbfeaf KS |
1495 | unsigned long index = start >> PAGE_SHIFT; |
1496 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1497 | struct page *page; |
1498 | ||
1499 | while (index <= end_index) { | |
1500 | page = find_get_page(inode->i_mapping, index); | |
1501 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1502 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1503 | put_page(page); |
4adaa611 CM |
1504 | index++; |
1505 | } | |
4adaa611 CM |
1506 | } |
1507 | ||
f6311572 | 1508 | void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1509 | { |
09cbfeaf KS |
1510 | unsigned long index = start >> PAGE_SHIFT; |
1511 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1512 | struct page *page; |
1513 | ||
1514 | while (index <= end_index) { | |
1515 | page = find_get_page(inode->i_mapping, index); | |
1516 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
4adaa611 | 1517 | __set_page_dirty_nobuffers(page); |
8d38633c | 1518 | account_page_redirty(page); |
09cbfeaf | 1519 | put_page(page); |
4adaa611 CM |
1520 | index++; |
1521 | } | |
4adaa611 CM |
1522 | } |
1523 | ||
d352ac68 CM |
1524 | /* find the first state struct with 'bits' set after 'start', and |
1525 | * return it. tree->lock must be held. NULL will returned if | |
1526 | * nothing was found after 'start' | |
1527 | */ | |
48a3b636 ES |
1528 | static struct extent_state * |
1529 | find_first_extent_bit_state(struct extent_io_tree *tree, | |
9ee49a04 | 1530 | u64 start, unsigned bits) |
d7fc640e CM |
1531 | { |
1532 | struct rb_node *node; | |
1533 | struct extent_state *state; | |
1534 | ||
1535 | /* | |
1536 | * this search will find all the extents that end after | |
1537 | * our range starts. | |
1538 | */ | |
1539 | node = tree_search(tree, start); | |
d397712b | 1540 | if (!node) |
d7fc640e | 1541 | goto out; |
d7fc640e | 1542 | |
d397712b | 1543 | while (1) { |
d7fc640e | 1544 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1545 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1546 | return state; |
d397712b | 1547 | |
d7fc640e CM |
1548 | node = rb_next(node); |
1549 | if (!node) | |
1550 | break; | |
1551 | } | |
1552 | out: | |
1553 | return NULL; | |
1554 | } | |
d7fc640e | 1555 | |
69261c4b XG |
1556 | /* |
1557 | * find the first offset in the io tree with 'bits' set. zero is | |
1558 | * returned if we find something, and *start_ret and *end_ret are | |
1559 | * set to reflect the state struct that was found. | |
1560 | * | |
477d7eaf | 1561 | * If nothing was found, 1 is returned. If found something, return 0. |
69261c4b XG |
1562 | */ |
1563 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
9ee49a04 | 1564 | u64 *start_ret, u64 *end_ret, unsigned bits, |
e6138876 | 1565 | struct extent_state **cached_state) |
69261c4b XG |
1566 | { |
1567 | struct extent_state *state; | |
1568 | int ret = 1; | |
1569 | ||
1570 | spin_lock(&tree->lock); | |
e6138876 JB |
1571 | if (cached_state && *cached_state) { |
1572 | state = *cached_state; | |
27a3507d | 1573 | if (state->end == start - 1 && extent_state_in_tree(state)) { |
9688e9a9 | 1574 | while ((state = next_state(state)) != NULL) { |
e6138876 JB |
1575 | if (state->state & bits) |
1576 | goto got_it; | |
e6138876 JB |
1577 | } |
1578 | free_extent_state(*cached_state); | |
1579 | *cached_state = NULL; | |
1580 | goto out; | |
1581 | } | |
1582 | free_extent_state(*cached_state); | |
1583 | *cached_state = NULL; | |
1584 | } | |
1585 | ||
69261c4b | 1586 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1587 | got_it: |
69261c4b | 1588 | if (state) { |
e38e2ed7 | 1589 | cache_state_if_flags(state, cached_state, 0); |
69261c4b XG |
1590 | *start_ret = state->start; |
1591 | *end_ret = state->end; | |
1592 | ret = 0; | |
1593 | } | |
e6138876 | 1594 | out: |
69261c4b XG |
1595 | spin_unlock(&tree->lock); |
1596 | return ret; | |
1597 | } | |
1598 | ||
41a2ee75 JB |
1599 | /** |
1600 | * find_contiguous_extent_bit: find a contiguous area of bits | |
1601 | * @tree - io tree to check | |
1602 | * @start - offset to start the search from | |
1603 | * @start_ret - the first offset we found with the bits set | |
1604 | * @end_ret - the final contiguous range of the bits that were set | |
1605 | * @bits - bits to look for | |
1606 | * | |
1607 | * set_extent_bit and clear_extent_bit can temporarily split contiguous ranges | |
1608 | * to set bits appropriately, and then merge them again. During this time it | |
1609 | * will drop the tree->lock, so use this helper if you want to find the actual | |
1610 | * contiguous area for given bits. We will search to the first bit we find, and | |
1611 | * then walk down the tree until we find a non-contiguous area. The area | |
1612 | * returned will be the full contiguous area with the bits set. | |
1613 | */ | |
1614 | int find_contiguous_extent_bit(struct extent_io_tree *tree, u64 start, | |
1615 | u64 *start_ret, u64 *end_ret, unsigned bits) | |
1616 | { | |
1617 | struct extent_state *state; | |
1618 | int ret = 1; | |
1619 | ||
1620 | spin_lock(&tree->lock); | |
1621 | state = find_first_extent_bit_state(tree, start, bits); | |
1622 | if (state) { | |
1623 | *start_ret = state->start; | |
1624 | *end_ret = state->end; | |
1625 | while ((state = next_state(state)) != NULL) { | |
1626 | if (state->start > (*end_ret + 1)) | |
1627 | break; | |
1628 | *end_ret = state->end; | |
1629 | } | |
1630 | ret = 0; | |
1631 | } | |
1632 | spin_unlock(&tree->lock); | |
1633 | return ret; | |
1634 | } | |
1635 | ||
45bfcfc1 | 1636 | /** |
1eaebb34 NB |
1637 | * find_first_clear_extent_bit - find the first range that has @bits not set. |
1638 | * This range could start before @start. | |
45bfcfc1 NB |
1639 | * |
1640 | * @tree - the tree to search | |
1641 | * @start - the offset at/after which the found extent should start | |
1642 | * @start_ret - records the beginning of the range | |
1643 | * @end_ret - records the end of the range (inclusive) | |
1644 | * @bits - the set of bits which must be unset | |
1645 | * | |
1646 | * Since unallocated range is also considered one which doesn't have the bits | |
1647 | * set it's possible that @end_ret contains -1, this happens in case the range | |
1648 | * spans (last_range_end, end of device]. In this case it's up to the caller to | |
1649 | * trim @end_ret to the appropriate size. | |
1650 | */ | |
1651 | void find_first_clear_extent_bit(struct extent_io_tree *tree, u64 start, | |
1652 | u64 *start_ret, u64 *end_ret, unsigned bits) | |
1653 | { | |
1654 | struct extent_state *state; | |
1655 | struct rb_node *node, *prev = NULL, *next; | |
1656 | ||
1657 | spin_lock(&tree->lock); | |
1658 | ||
1659 | /* Find first extent with bits cleared */ | |
1660 | while (1) { | |
1661 | node = __etree_search(tree, start, &next, &prev, NULL, NULL); | |
5750c375 NB |
1662 | if (!node && !next && !prev) { |
1663 | /* | |
1664 | * Tree is completely empty, send full range and let | |
1665 | * caller deal with it | |
1666 | */ | |
1667 | *start_ret = 0; | |
1668 | *end_ret = -1; | |
1669 | goto out; | |
1670 | } else if (!node && !next) { | |
1671 | /* | |
1672 | * We are past the last allocated chunk, set start at | |
1673 | * the end of the last extent. | |
1674 | */ | |
1675 | state = rb_entry(prev, struct extent_state, rb_node); | |
1676 | *start_ret = state->end + 1; | |
1677 | *end_ret = -1; | |
1678 | goto out; | |
1679 | } else if (!node) { | |
45bfcfc1 | 1680 | node = next; |
45bfcfc1 | 1681 | } |
1eaebb34 NB |
1682 | /* |
1683 | * At this point 'node' either contains 'start' or start is | |
1684 | * before 'node' | |
1685 | */ | |
45bfcfc1 | 1686 | state = rb_entry(node, struct extent_state, rb_node); |
1eaebb34 NB |
1687 | |
1688 | if (in_range(start, state->start, state->end - state->start + 1)) { | |
1689 | if (state->state & bits) { | |
1690 | /* | |
1691 | * |--range with bits sets--| | |
1692 | * | | |
1693 | * start | |
1694 | */ | |
1695 | start = state->end + 1; | |
1696 | } else { | |
1697 | /* | |
1698 | * 'start' falls within a range that doesn't | |
1699 | * have the bits set, so take its start as | |
1700 | * the beginning of the desired range | |
1701 | * | |
1702 | * |--range with bits cleared----| | |
1703 | * | | |
1704 | * start | |
1705 | */ | |
1706 | *start_ret = state->start; | |
1707 | break; | |
1708 | } | |
45bfcfc1 | 1709 | } else { |
1eaebb34 NB |
1710 | /* |
1711 | * |---prev range---|---hole/unset---|---node range---| | |
1712 | * | | |
1713 | * start | |
1714 | * | |
1715 | * or | |
1716 | * | |
1717 | * |---hole/unset--||--first node--| | |
1718 | * 0 | | |
1719 | * start | |
1720 | */ | |
1721 | if (prev) { | |
1722 | state = rb_entry(prev, struct extent_state, | |
1723 | rb_node); | |
1724 | *start_ret = state->end + 1; | |
1725 | } else { | |
1726 | *start_ret = 0; | |
1727 | } | |
45bfcfc1 NB |
1728 | break; |
1729 | } | |
1730 | } | |
1731 | ||
1732 | /* | |
1733 | * Find the longest stretch from start until an entry which has the | |
1734 | * bits set | |
1735 | */ | |
1736 | while (1) { | |
1737 | state = rb_entry(node, struct extent_state, rb_node); | |
1738 | if (state->end >= start && !(state->state & bits)) { | |
1739 | *end_ret = state->end; | |
1740 | } else { | |
1741 | *end_ret = state->start - 1; | |
1742 | break; | |
1743 | } | |
1744 | ||
1745 | node = rb_next(node); | |
1746 | if (!node) | |
1747 | break; | |
1748 | } | |
1749 | out: | |
1750 | spin_unlock(&tree->lock); | |
1751 | } | |
1752 | ||
d352ac68 CM |
1753 | /* |
1754 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1755 | * more than 'max_bytes'. start and end are used to return the range, | |
1756 | * | |
3522e903 | 1757 | * true is returned if we find something, false if nothing was in the tree |
d352ac68 | 1758 | */ |
083e75e7 JB |
1759 | bool btrfs_find_delalloc_range(struct extent_io_tree *tree, u64 *start, |
1760 | u64 *end, u64 max_bytes, | |
1761 | struct extent_state **cached_state) | |
d1310b2e CM |
1762 | { |
1763 | struct rb_node *node; | |
1764 | struct extent_state *state; | |
1765 | u64 cur_start = *start; | |
3522e903 | 1766 | bool found = false; |
d1310b2e CM |
1767 | u64 total_bytes = 0; |
1768 | ||
cad321ad | 1769 | spin_lock(&tree->lock); |
c8b97818 | 1770 | |
d1310b2e CM |
1771 | /* |
1772 | * this search will find all the extents that end after | |
1773 | * our range starts. | |
1774 | */ | |
80ea96b1 | 1775 | node = tree_search(tree, cur_start); |
2b114d1d | 1776 | if (!node) { |
3522e903 | 1777 | *end = (u64)-1; |
d1310b2e CM |
1778 | goto out; |
1779 | } | |
1780 | ||
d397712b | 1781 | while (1) { |
d1310b2e | 1782 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1783 | if (found && (state->start != cur_start || |
1784 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1785 | goto out; |
1786 | } | |
1787 | if (!(state->state & EXTENT_DELALLOC)) { | |
1788 | if (!found) | |
1789 | *end = state->end; | |
1790 | goto out; | |
1791 | } | |
c2a128d2 | 1792 | if (!found) { |
d1310b2e | 1793 | *start = state->start; |
c2a128d2 | 1794 | *cached_state = state; |
b7ac31b7 | 1795 | refcount_inc(&state->refs); |
c2a128d2 | 1796 | } |
3522e903 | 1797 | found = true; |
d1310b2e CM |
1798 | *end = state->end; |
1799 | cur_start = state->end + 1; | |
1800 | node = rb_next(node); | |
d1310b2e | 1801 | total_bytes += state->end - state->start + 1; |
7bf811a5 | 1802 | if (total_bytes >= max_bytes) |
573aecaf | 1803 | break; |
573aecaf | 1804 | if (!node) |
d1310b2e CM |
1805 | break; |
1806 | } | |
1807 | out: | |
cad321ad | 1808 | spin_unlock(&tree->lock); |
d1310b2e CM |
1809 | return found; |
1810 | } | |
1811 | ||
da2c7009 LB |
1812 | static int __process_pages_contig(struct address_space *mapping, |
1813 | struct page *locked_page, | |
1814 | pgoff_t start_index, pgoff_t end_index, | |
1815 | unsigned long page_ops, pgoff_t *index_ret); | |
1816 | ||
143bede5 JM |
1817 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1818 | struct page *locked_page, | |
1819 | u64 start, u64 end) | |
c8b97818 | 1820 | { |
09cbfeaf KS |
1821 | unsigned long index = start >> PAGE_SHIFT; |
1822 | unsigned long end_index = end >> PAGE_SHIFT; | |
c8b97818 | 1823 | |
76c0021d | 1824 | ASSERT(locked_page); |
c8b97818 | 1825 | if (index == locked_page->index && end_index == index) |
143bede5 | 1826 | return; |
c8b97818 | 1827 | |
76c0021d LB |
1828 | __process_pages_contig(inode->i_mapping, locked_page, index, end_index, |
1829 | PAGE_UNLOCK, NULL); | |
c8b97818 CM |
1830 | } |
1831 | ||
1832 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1833 | struct page *locked_page, | |
1834 | u64 delalloc_start, | |
1835 | u64 delalloc_end) | |
1836 | { | |
09cbfeaf | 1837 | unsigned long index = delalloc_start >> PAGE_SHIFT; |
76c0021d | 1838 | unsigned long index_ret = index; |
09cbfeaf | 1839 | unsigned long end_index = delalloc_end >> PAGE_SHIFT; |
c8b97818 | 1840 | int ret; |
c8b97818 | 1841 | |
76c0021d | 1842 | ASSERT(locked_page); |
c8b97818 CM |
1843 | if (index == locked_page->index && index == end_index) |
1844 | return 0; | |
1845 | ||
76c0021d LB |
1846 | ret = __process_pages_contig(inode->i_mapping, locked_page, index, |
1847 | end_index, PAGE_LOCK, &index_ret); | |
1848 | if (ret == -EAGAIN) | |
1849 | __unlock_for_delalloc(inode, locked_page, delalloc_start, | |
1850 | (u64)index_ret << PAGE_SHIFT); | |
c8b97818 CM |
1851 | return ret; |
1852 | } | |
1853 | ||
1854 | /* | |
3522e903 LF |
1855 | * Find and lock a contiguous range of bytes in the file marked as delalloc, no |
1856 | * more than @max_bytes. @Start and @end are used to return the range, | |
c8b97818 | 1857 | * |
3522e903 LF |
1858 | * Return: true if we find something |
1859 | * false if nothing was in the tree | |
c8b97818 | 1860 | */ |
ce9f967f | 1861 | EXPORT_FOR_TESTS |
3522e903 | 1862 | noinline_for_stack bool find_lock_delalloc_range(struct inode *inode, |
294e30fe | 1863 | struct page *locked_page, u64 *start, |
917aacec | 1864 | u64 *end) |
c8b97818 | 1865 | { |
9978059b | 1866 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
917aacec | 1867 | u64 max_bytes = BTRFS_MAX_EXTENT_SIZE; |
c8b97818 CM |
1868 | u64 delalloc_start; |
1869 | u64 delalloc_end; | |
3522e903 | 1870 | bool found; |
9655d298 | 1871 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1872 | int ret; |
1873 | int loops = 0; | |
1874 | ||
1875 | again: | |
1876 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1877 | delalloc_start = *start; | |
1878 | delalloc_end = 0; | |
083e75e7 JB |
1879 | found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end, |
1880 | max_bytes, &cached_state); | |
70b99e69 | 1881 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1882 | *start = delalloc_start; |
1883 | *end = delalloc_end; | |
c2a128d2 | 1884 | free_extent_state(cached_state); |
3522e903 | 1885 | return false; |
c8b97818 CM |
1886 | } |
1887 | ||
70b99e69 CM |
1888 | /* |
1889 | * start comes from the offset of locked_page. We have to lock | |
1890 | * pages in order, so we can't process delalloc bytes before | |
1891 | * locked_page | |
1892 | */ | |
d397712b | 1893 | if (delalloc_start < *start) |
70b99e69 | 1894 | delalloc_start = *start; |
70b99e69 | 1895 | |
c8b97818 CM |
1896 | /* |
1897 | * make sure to limit the number of pages we try to lock down | |
c8b97818 | 1898 | */ |
7bf811a5 JB |
1899 | if (delalloc_end + 1 - delalloc_start > max_bytes) |
1900 | delalloc_end = delalloc_start + max_bytes - 1; | |
d397712b | 1901 | |
c8b97818 CM |
1902 | /* step two, lock all the pages after the page that has start */ |
1903 | ret = lock_delalloc_pages(inode, locked_page, | |
1904 | delalloc_start, delalloc_end); | |
9bfd61d9 | 1905 | ASSERT(!ret || ret == -EAGAIN); |
c8b97818 CM |
1906 | if (ret == -EAGAIN) { |
1907 | /* some of the pages are gone, lets avoid looping by | |
1908 | * shortening the size of the delalloc range we're searching | |
1909 | */ | |
9655d298 | 1910 | free_extent_state(cached_state); |
7d788742 | 1911 | cached_state = NULL; |
c8b97818 | 1912 | if (!loops) { |
09cbfeaf | 1913 | max_bytes = PAGE_SIZE; |
c8b97818 CM |
1914 | loops = 1; |
1915 | goto again; | |
1916 | } else { | |
3522e903 | 1917 | found = false; |
c8b97818 CM |
1918 | goto out_failed; |
1919 | } | |
1920 | } | |
c8b97818 CM |
1921 | |
1922 | /* step three, lock the state bits for the whole range */ | |
ff13db41 | 1923 | lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state); |
c8b97818 CM |
1924 | |
1925 | /* then test to make sure it is all still delalloc */ | |
1926 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1927 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1928 | if (!ret) { |
9655d298 | 1929 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
e43bbe5e | 1930 | &cached_state); |
c8b97818 CM |
1931 | __unlock_for_delalloc(inode, locked_page, |
1932 | delalloc_start, delalloc_end); | |
1933 | cond_resched(); | |
1934 | goto again; | |
1935 | } | |
9655d298 | 1936 | free_extent_state(cached_state); |
c8b97818 CM |
1937 | *start = delalloc_start; |
1938 | *end = delalloc_end; | |
1939 | out_failed: | |
1940 | return found; | |
1941 | } | |
1942 | ||
da2c7009 LB |
1943 | static int __process_pages_contig(struct address_space *mapping, |
1944 | struct page *locked_page, | |
1945 | pgoff_t start_index, pgoff_t end_index, | |
1946 | unsigned long page_ops, pgoff_t *index_ret) | |
c8b97818 | 1947 | { |
873695b3 | 1948 | unsigned long nr_pages = end_index - start_index + 1; |
da2c7009 | 1949 | unsigned long pages_locked = 0; |
873695b3 | 1950 | pgoff_t index = start_index; |
c8b97818 | 1951 | struct page *pages[16]; |
873695b3 | 1952 | unsigned ret; |
da2c7009 | 1953 | int err = 0; |
c8b97818 | 1954 | int i; |
771ed689 | 1955 | |
da2c7009 LB |
1956 | if (page_ops & PAGE_LOCK) { |
1957 | ASSERT(page_ops == PAGE_LOCK); | |
1958 | ASSERT(index_ret && *index_ret == start_index); | |
1959 | } | |
1960 | ||
704de49d | 1961 | if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0) |
873695b3 | 1962 | mapping_set_error(mapping, -EIO); |
704de49d | 1963 | |
d397712b | 1964 | while (nr_pages > 0) { |
873695b3 | 1965 | ret = find_get_pages_contig(mapping, index, |
5b050f04 CM |
1966 | min_t(unsigned long, |
1967 | nr_pages, ARRAY_SIZE(pages)), pages); | |
da2c7009 LB |
1968 | if (ret == 0) { |
1969 | /* | |
1970 | * Only if we're going to lock these pages, | |
1971 | * can we find nothing at @index. | |
1972 | */ | |
1973 | ASSERT(page_ops & PAGE_LOCK); | |
49d4a334 LB |
1974 | err = -EAGAIN; |
1975 | goto out; | |
da2c7009 | 1976 | } |
8b62b72b | 1977 | |
da2c7009 | 1978 | for (i = 0; i < ret; i++) { |
c2790a2e | 1979 | if (page_ops & PAGE_SET_PRIVATE2) |
8b62b72b CM |
1980 | SetPagePrivate2(pages[i]); |
1981 | ||
1d53c9e6 | 1982 | if (locked_page && pages[i] == locked_page) { |
09cbfeaf | 1983 | put_page(pages[i]); |
da2c7009 | 1984 | pages_locked++; |
c8b97818 CM |
1985 | continue; |
1986 | } | |
c2790a2e | 1987 | if (page_ops & PAGE_CLEAR_DIRTY) |
c8b97818 | 1988 | clear_page_dirty_for_io(pages[i]); |
c2790a2e | 1989 | if (page_ops & PAGE_SET_WRITEBACK) |
c8b97818 | 1990 | set_page_writeback(pages[i]); |
704de49d FM |
1991 | if (page_ops & PAGE_SET_ERROR) |
1992 | SetPageError(pages[i]); | |
c2790a2e | 1993 | if (page_ops & PAGE_END_WRITEBACK) |
c8b97818 | 1994 | end_page_writeback(pages[i]); |
c2790a2e | 1995 | if (page_ops & PAGE_UNLOCK) |
771ed689 | 1996 | unlock_page(pages[i]); |
da2c7009 LB |
1997 | if (page_ops & PAGE_LOCK) { |
1998 | lock_page(pages[i]); | |
1999 | if (!PageDirty(pages[i]) || | |
2000 | pages[i]->mapping != mapping) { | |
2001 | unlock_page(pages[i]); | |
2002 | put_page(pages[i]); | |
2003 | err = -EAGAIN; | |
2004 | goto out; | |
2005 | } | |
2006 | } | |
09cbfeaf | 2007 | put_page(pages[i]); |
da2c7009 | 2008 | pages_locked++; |
c8b97818 CM |
2009 | } |
2010 | nr_pages -= ret; | |
2011 | index += ret; | |
2012 | cond_resched(); | |
2013 | } | |
da2c7009 LB |
2014 | out: |
2015 | if (err && index_ret) | |
2016 | *index_ret = start_index + pages_locked - 1; | |
2017 | return err; | |
c8b97818 | 2018 | } |
c8b97818 | 2019 | |
873695b3 | 2020 | void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end, |
74e9194a NB |
2021 | struct page *locked_page, |
2022 | unsigned clear_bits, | |
2023 | unsigned long page_ops) | |
873695b3 LB |
2024 | { |
2025 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0, | |
ae0f1625 | 2026 | NULL); |
873695b3 LB |
2027 | |
2028 | __process_pages_contig(inode->i_mapping, locked_page, | |
2029 | start >> PAGE_SHIFT, end >> PAGE_SHIFT, | |
da2c7009 | 2030 | page_ops, NULL); |
873695b3 LB |
2031 | } |
2032 | ||
d352ac68 CM |
2033 | /* |
2034 | * count the number of bytes in the tree that have a given bit(s) | |
2035 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
2036 | * cached. The total number found is returned. | |
2037 | */ | |
d1310b2e CM |
2038 | u64 count_range_bits(struct extent_io_tree *tree, |
2039 | u64 *start, u64 search_end, u64 max_bytes, | |
9ee49a04 | 2040 | unsigned bits, int contig) |
d1310b2e CM |
2041 | { |
2042 | struct rb_node *node; | |
2043 | struct extent_state *state; | |
2044 | u64 cur_start = *start; | |
2045 | u64 total_bytes = 0; | |
ec29ed5b | 2046 | u64 last = 0; |
d1310b2e CM |
2047 | int found = 0; |
2048 | ||
fae7f21c | 2049 | if (WARN_ON(search_end <= cur_start)) |
d1310b2e | 2050 | return 0; |
d1310b2e | 2051 | |
cad321ad | 2052 | spin_lock(&tree->lock); |
d1310b2e CM |
2053 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
2054 | total_bytes = tree->dirty_bytes; | |
2055 | goto out; | |
2056 | } | |
2057 | /* | |
2058 | * this search will find all the extents that end after | |
2059 | * our range starts. | |
2060 | */ | |
80ea96b1 | 2061 | node = tree_search(tree, cur_start); |
d397712b | 2062 | if (!node) |
d1310b2e | 2063 | goto out; |
d1310b2e | 2064 | |
d397712b | 2065 | while (1) { |
d1310b2e CM |
2066 | state = rb_entry(node, struct extent_state, rb_node); |
2067 | if (state->start > search_end) | |
2068 | break; | |
ec29ed5b CM |
2069 | if (contig && found && state->start > last + 1) |
2070 | break; | |
2071 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
2072 | total_bytes += min(search_end, state->end) + 1 - |
2073 | max(cur_start, state->start); | |
2074 | if (total_bytes >= max_bytes) | |
2075 | break; | |
2076 | if (!found) { | |
af60bed2 | 2077 | *start = max(cur_start, state->start); |
d1310b2e CM |
2078 | found = 1; |
2079 | } | |
ec29ed5b CM |
2080 | last = state->end; |
2081 | } else if (contig && found) { | |
2082 | break; | |
d1310b2e CM |
2083 | } |
2084 | node = rb_next(node); | |
2085 | if (!node) | |
2086 | break; | |
2087 | } | |
2088 | out: | |
cad321ad | 2089 | spin_unlock(&tree->lock); |
d1310b2e CM |
2090 | return total_bytes; |
2091 | } | |
b2950863 | 2092 | |
d352ac68 CM |
2093 | /* |
2094 | * set the private field for a given byte offset in the tree. If there isn't | |
2095 | * an extent_state there already, this does nothing. | |
2096 | */ | |
b3f167aa JB |
2097 | int set_state_failrec(struct extent_io_tree *tree, u64 start, |
2098 | struct io_failure_record *failrec) | |
d1310b2e CM |
2099 | { |
2100 | struct rb_node *node; | |
2101 | struct extent_state *state; | |
2102 | int ret = 0; | |
2103 | ||
cad321ad | 2104 | spin_lock(&tree->lock); |
d1310b2e CM |
2105 | /* |
2106 | * this search will find all the extents that end after | |
2107 | * our range starts. | |
2108 | */ | |
80ea96b1 | 2109 | node = tree_search(tree, start); |
2b114d1d | 2110 | if (!node) { |
d1310b2e CM |
2111 | ret = -ENOENT; |
2112 | goto out; | |
2113 | } | |
2114 | state = rb_entry(node, struct extent_state, rb_node); | |
2115 | if (state->start != start) { | |
2116 | ret = -ENOENT; | |
2117 | goto out; | |
2118 | } | |
47dc196a | 2119 | state->failrec = failrec; |
d1310b2e | 2120 | out: |
cad321ad | 2121 | spin_unlock(&tree->lock); |
d1310b2e CM |
2122 | return ret; |
2123 | } | |
2124 | ||
b3f167aa JB |
2125 | int get_state_failrec(struct extent_io_tree *tree, u64 start, |
2126 | struct io_failure_record **failrec) | |
d1310b2e CM |
2127 | { |
2128 | struct rb_node *node; | |
2129 | struct extent_state *state; | |
2130 | int ret = 0; | |
2131 | ||
cad321ad | 2132 | spin_lock(&tree->lock); |
d1310b2e CM |
2133 | /* |
2134 | * this search will find all the extents that end after | |
2135 | * our range starts. | |
2136 | */ | |
80ea96b1 | 2137 | node = tree_search(tree, start); |
2b114d1d | 2138 | if (!node) { |
d1310b2e CM |
2139 | ret = -ENOENT; |
2140 | goto out; | |
2141 | } | |
2142 | state = rb_entry(node, struct extent_state, rb_node); | |
2143 | if (state->start != start) { | |
2144 | ret = -ENOENT; | |
2145 | goto out; | |
2146 | } | |
47dc196a | 2147 | *failrec = state->failrec; |
d1310b2e | 2148 | out: |
cad321ad | 2149 | spin_unlock(&tree->lock); |
d1310b2e CM |
2150 | return ret; |
2151 | } | |
2152 | ||
2153 | /* | |
2154 | * searches a range in the state tree for a given mask. | |
70dec807 | 2155 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
2156 | * has the bits set. Otherwise, 1 is returned if any bit in the |
2157 | * range is found set. | |
2158 | */ | |
2159 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 2160 | unsigned bits, int filled, struct extent_state *cached) |
d1310b2e CM |
2161 | { |
2162 | struct extent_state *state = NULL; | |
2163 | struct rb_node *node; | |
2164 | int bitset = 0; | |
d1310b2e | 2165 | |
cad321ad | 2166 | spin_lock(&tree->lock); |
27a3507d | 2167 | if (cached && extent_state_in_tree(cached) && cached->start <= start && |
df98b6e2 | 2168 | cached->end > start) |
9655d298 CM |
2169 | node = &cached->rb_node; |
2170 | else | |
2171 | node = tree_search(tree, start); | |
d1310b2e CM |
2172 | while (node && start <= end) { |
2173 | state = rb_entry(node, struct extent_state, rb_node); | |
2174 | ||
2175 | if (filled && state->start > start) { | |
2176 | bitset = 0; | |
2177 | break; | |
2178 | } | |
2179 | ||
2180 | if (state->start > end) | |
2181 | break; | |
2182 | ||
2183 | if (state->state & bits) { | |
2184 | bitset = 1; | |
2185 | if (!filled) | |
2186 | break; | |
2187 | } else if (filled) { | |
2188 | bitset = 0; | |
2189 | break; | |
2190 | } | |
46562cec CM |
2191 | |
2192 | if (state->end == (u64)-1) | |
2193 | break; | |
2194 | ||
d1310b2e CM |
2195 | start = state->end + 1; |
2196 | if (start > end) | |
2197 | break; | |
2198 | node = rb_next(node); | |
2199 | if (!node) { | |
2200 | if (filled) | |
2201 | bitset = 0; | |
2202 | break; | |
2203 | } | |
2204 | } | |
cad321ad | 2205 | spin_unlock(&tree->lock); |
d1310b2e CM |
2206 | return bitset; |
2207 | } | |
d1310b2e CM |
2208 | |
2209 | /* | |
2210 | * helper function to set a given page up to date if all the | |
2211 | * extents in the tree for that page are up to date | |
2212 | */ | |
143bede5 | 2213 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e | 2214 | { |
4eee4fa4 | 2215 | u64 start = page_offset(page); |
09cbfeaf | 2216 | u64 end = start + PAGE_SIZE - 1; |
9655d298 | 2217 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 2218 | SetPageUptodate(page); |
d1310b2e CM |
2219 | } |
2220 | ||
7870d082 JB |
2221 | int free_io_failure(struct extent_io_tree *failure_tree, |
2222 | struct extent_io_tree *io_tree, | |
2223 | struct io_failure_record *rec) | |
4a54c8c1 JS |
2224 | { |
2225 | int ret; | |
2226 | int err = 0; | |
4a54c8c1 | 2227 | |
47dc196a | 2228 | set_state_failrec(failure_tree, rec->start, NULL); |
4a54c8c1 JS |
2229 | ret = clear_extent_bits(failure_tree, rec->start, |
2230 | rec->start + rec->len - 1, | |
91166212 | 2231 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 JS |
2232 | if (ret) |
2233 | err = ret; | |
2234 | ||
7870d082 | 2235 | ret = clear_extent_bits(io_tree, rec->start, |
53b381b3 | 2236 | rec->start + rec->len - 1, |
91166212 | 2237 | EXTENT_DAMAGED); |
53b381b3 DW |
2238 | if (ret && !err) |
2239 | err = ret; | |
4a54c8c1 JS |
2240 | |
2241 | kfree(rec); | |
2242 | return err; | |
2243 | } | |
2244 | ||
4a54c8c1 JS |
2245 | /* |
2246 | * this bypasses the standard btrfs submit functions deliberately, as | |
2247 | * the standard behavior is to write all copies in a raid setup. here we only | |
2248 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
2249 | * submit_bio directly. | |
3ec706c8 | 2250 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
2251 | * actually prevents the read that triggered the error from finishing. |
2252 | * currently, there can be no more than two copies of every data bit. thus, | |
2253 | * exactly one rewrite is required. | |
2254 | */ | |
6ec656bc JB |
2255 | int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start, |
2256 | u64 length, u64 logical, struct page *page, | |
2257 | unsigned int pg_offset, int mirror_num) | |
4a54c8c1 JS |
2258 | { |
2259 | struct bio *bio; | |
2260 | struct btrfs_device *dev; | |
4a54c8c1 JS |
2261 | u64 map_length = 0; |
2262 | u64 sector; | |
2263 | struct btrfs_bio *bbio = NULL; | |
2264 | int ret; | |
2265 | ||
1751e8a6 | 2266 | ASSERT(!(fs_info->sb->s_flags & SB_RDONLY)); |
4a54c8c1 JS |
2267 | BUG_ON(!mirror_num); |
2268 | ||
c5e4c3d7 | 2269 | bio = btrfs_io_bio_alloc(1); |
4f024f37 | 2270 | bio->bi_iter.bi_size = 0; |
4a54c8c1 JS |
2271 | map_length = length; |
2272 | ||
b5de8d0d FM |
2273 | /* |
2274 | * Avoid races with device replace and make sure our bbio has devices | |
2275 | * associated to its stripes that don't go away while we are doing the | |
2276 | * read repair operation. | |
2277 | */ | |
2278 | btrfs_bio_counter_inc_blocked(fs_info); | |
e4ff5fb5 | 2279 | if (btrfs_is_parity_mirror(fs_info, logical, length)) { |
c725328c LB |
2280 | /* |
2281 | * Note that we don't use BTRFS_MAP_WRITE because it's supposed | |
2282 | * to update all raid stripes, but here we just want to correct | |
2283 | * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad | |
2284 | * stripe's dev and sector. | |
2285 | */ | |
2286 | ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical, | |
2287 | &map_length, &bbio, 0); | |
2288 | if (ret) { | |
2289 | btrfs_bio_counter_dec(fs_info); | |
2290 | bio_put(bio); | |
2291 | return -EIO; | |
2292 | } | |
2293 | ASSERT(bbio->mirror_num == 1); | |
2294 | } else { | |
2295 | ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, | |
2296 | &map_length, &bbio, mirror_num); | |
2297 | if (ret) { | |
2298 | btrfs_bio_counter_dec(fs_info); | |
2299 | bio_put(bio); | |
2300 | return -EIO; | |
2301 | } | |
2302 | BUG_ON(mirror_num != bbio->mirror_num); | |
4a54c8c1 | 2303 | } |
c725328c LB |
2304 | |
2305 | sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9; | |
4f024f37 | 2306 | bio->bi_iter.bi_sector = sector; |
c725328c | 2307 | dev = bbio->stripes[bbio->mirror_num - 1].dev; |
6e9606d2 | 2308 | btrfs_put_bbio(bbio); |
ebbede42 AJ |
2309 | if (!dev || !dev->bdev || |
2310 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) { | |
b5de8d0d | 2311 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2312 | bio_put(bio); |
2313 | return -EIO; | |
2314 | } | |
74d46992 | 2315 | bio_set_dev(bio, dev->bdev); |
70fd7614 | 2316 | bio->bi_opf = REQ_OP_WRITE | REQ_SYNC; |
ffdd2018 | 2317 | bio_add_page(bio, page, length, pg_offset); |
4a54c8c1 | 2318 | |
4e49ea4a | 2319 | if (btrfsic_submit_bio_wait(bio)) { |
4a54c8c1 | 2320 | /* try to remap that extent elsewhere? */ |
b5de8d0d | 2321 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 | 2322 | bio_put(bio); |
442a4f63 | 2323 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
2324 | return -EIO; |
2325 | } | |
2326 | ||
b14af3b4 DS |
2327 | btrfs_info_rl_in_rcu(fs_info, |
2328 | "read error corrected: ino %llu off %llu (dev %s sector %llu)", | |
6ec656bc | 2329 | ino, start, |
1203b681 | 2330 | rcu_str_deref(dev->name), sector); |
b5de8d0d | 2331 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2332 | bio_put(bio); |
2333 | return 0; | |
2334 | } | |
2335 | ||
20a1fbf9 | 2336 | int btrfs_repair_eb_io_failure(struct extent_buffer *eb, int mirror_num) |
ea466794 | 2337 | { |
20a1fbf9 | 2338 | struct btrfs_fs_info *fs_info = eb->fs_info; |
ea466794 | 2339 | u64 start = eb->start; |
cc5e31a4 | 2340 | int i, num_pages = num_extent_pages(eb); |
d95603b2 | 2341 | int ret = 0; |
ea466794 | 2342 | |
bc98a42c | 2343 | if (sb_rdonly(fs_info->sb)) |
908960c6 ID |
2344 | return -EROFS; |
2345 | ||
ea466794 | 2346 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 2347 | struct page *p = eb->pages[i]; |
1203b681 | 2348 | |
6ec656bc | 2349 | ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p, |
1203b681 | 2350 | start - page_offset(p), mirror_num); |
ea466794 JB |
2351 | if (ret) |
2352 | break; | |
09cbfeaf | 2353 | start += PAGE_SIZE; |
ea466794 JB |
2354 | } |
2355 | ||
2356 | return ret; | |
2357 | } | |
2358 | ||
4a54c8c1 JS |
2359 | /* |
2360 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2361 | * to see if we need to process or clean up an io_failure_record | |
2362 | */ | |
7870d082 JB |
2363 | int clean_io_failure(struct btrfs_fs_info *fs_info, |
2364 | struct extent_io_tree *failure_tree, | |
2365 | struct extent_io_tree *io_tree, u64 start, | |
2366 | struct page *page, u64 ino, unsigned int pg_offset) | |
4a54c8c1 JS |
2367 | { |
2368 | u64 private; | |
4a54c8c1 | 2369 | struct io_failure_record *failrec; |
4a54c8c1 JS |
2370 | struct extent_state *state; |
2371 | int num_copies; | |
4a54c8c1 | 2372 | int ret; |
4a54c8c1 JS |
2373 | |
2374 | private = 0; | |
7870d082 JB |
2375 | ret = count_range_bits(failure_tree, &private, (u64)-1, 1, |
2376 | EXTENT_DIRTY, 0); | |
4a54c8c1 JS |
2377 | if (!ret) |
2378 | return 0; | |
2379 | ||
7870d082 | 2380 | ret = get_state_failrec(failure_tree, start, &failrec); |
4a54c8c1 JS |
2381 | if (ret) |
2382 | return 0; | |
2383 | ||
4a54c8c1 JS |
2384 | BUG_ON(!failrec->this_mirror); |
2385 | ||
2386 | if (failrec->in_validation) { | |
2387 | /* there was no real error, just free the record */ | |
ab8d0fc4 JM |
2388 | btrfs_debug(fs_info, |
2389 | "clean_io_failure: freeing dummy error at %llu", | |
2390 | failrec->start); | |
4a54c8c1 JS |
2391 | goto out; |
2392 | } | |
bc98a42c | 2393 | if (sb_rdonly(fs_info->sb)) |
908960c6 | 2394 | goto out; |
4a54c8c1 | 2395 | |
7870d082 JB |
2396 | spin_lock(&io_tree->lock); |
2397 | state = find_first_extent_bit_state(io_tree, | |
4a54c8c1 JS |
2398 | failrec->start, |
2399 | EXTENT_LOCKED); | |
7870d082 | 2400 | spin_unlock(&io_tree->lock); |
4a54c8c1 | 2401 | |
883d0de4 MX |
2402 | if (state && state->start <= failrec->start && |
2403 | state->end >= failrec->start + failrec->len - 1) { | |
3ec706c8 SB |
2404 | num_copies = btrfs_num_copies(fs_info, failrec->logical, |
2405 | failrec->len); | |
4a54c8c1 | 2406 | if (num_copies > 1) { |
7870d082 JB |
2407 | repair_io_failure(fs_info, ino, start, failrec->len, |
2408 | failrec->logical, page, pg_offset, | |
2409 | failrec->failed_mirror); | |
4a54c8c1 JS |
2410 | } |
2411 | } | |
2412 | ||
2413 | out: | |
7870d082 | 2414 | free_io_failure(failure_tree, io_tree, failrec); |
4a54c8c1 | 2415 | |
454ff3de | 2416 | return 0; |
4a54c8c1 JS |
2417 | } |
2418 | ||
f612496b MX |
2419 | /* |
2420 | * Can be called when | |
2421 | * - hold extent lock | |
2422 | * - under ordered extent | |
2423 | * - the inode is freeing | |
2424 | */ | |
7ab7956e | 2425 | void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end) |
f612496b | 2426 | { |
7ab7956e | 2427 | struct extent_io_tree *failure_tree = &inode->io_failure_tree; |
f612496b MX |
2428 | struct io_failure_record *failrec; |
2429 | struct extent_state *state, *next; | |
2430 | ||
2431 | if (RB_EMPTY_ROOT(&failure_tree->state)) | |
2432 | return; | |
2433 | ||
2434 | spin_lock(&failure_tree->lock); | |
2435 | state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY); | |
2436 | while (state) { | |
2437 | if (state->start > end) | |
2438 | break; | |
2439 | ||
2440 | ASSERT(state->end <= end); | |
2441 | ||
2442 | next = next_state(state); | |
2443 | ||
47dc196a | 2444 | failrec = state->failrec; |
f612496b MX |
2445 | free_extent_state(state); |
2446 | kfree(failrec); | |
2447 | ||
2448 | state = next; | |
2449 | } | |
2450 | spin_unlock(&failure_tree->lock); | |
2451 | } | |
2452 | ||
2fe6303e | 2453 | int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end, |
47dc196a | 2454 | struct io_failure_record **failrec_ret) |
4a54c8c1 | 2455 | { |
ab8d0fc4 | 2456 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e | 2457 | struct io_failure_record *failrec; |
4a54c8c1 | 2458 | struct extent_map *em; |
4a54c8c1 JS |
2459 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2460 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2461 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
4a54c8c1 | 2462 | int ret; |
4a54c8c1 JS |
2463 | u64 logical; |
2464 | ||
47dc196a | 2465 | ret = get_state_failrec(failure_tree, start, &failrec); |
4a54c8c1 JS |
2466 | if (ret) { |
2467 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2468 | if (!failrec) | |
2469 | return -ENOMEM; | |
2fe6303e | 2470 | |
4a54c8c1 JS |
2471 | failrec->start = start; |
2472 | failrec->len = end - start + 1; | |
2473 | failrec->this_mirror = 0; | |
2474 | failrec->bio_flags = 0; | |
2475 | failrec->in_validation = 0; | |
2476 | ||
2477 | read_lock(&em_tree->lock); | |
2478 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2479 | if (!em) { | |
2480 | read_unlock(&em_tree->lock); | |
2481 | kfree(failrec); | |
2482 | return -EIO; | |
2483 | } | |
2484 | ||
68ba990f | 2485 | if (em->start > start || em->start + em->len <= start) { |
4a54c8c1 JS |
2486 | free_extent_map(em); |
2487 | em = NULL; | |
2488 | } | |
2489 | read_unlock(&em_tree->lock); | |
7a2d6a64 | 2490 | if (!em) { |
4a54c8c1 JS |
2491 | kfree(failrec); |
2492 | return -EIO; | |
2493 | } | |
2fe6303e | 2494 | |
4a54c8c1 JS |
2495 | logical = start - em->start; |
2496 | logical = em->block_start + logical; | |
2497 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2498 | logical = em->block_start; | |
2499 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2500 | extent_set_compress_type(&failrec->bio_flags, | |
2501 | em->compress_type); | |
2502 | } | |
2fe6303e | 2503 | |
ab8d0fc4 JM |
2504 | btrfs_debug(fs_info, |
2505 | "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu", | |
2506 | logical, start, failrec->len); | |
2fe6303e | 2507 | |
4a54c8c1 JS |
2508 | failrec->logical = logical; |
2509 | free_extent_map(em); | |
2510 | ||
2511 | /* set the bits in the private failure tree */ | |
2512 | ret = set_extent_bits(failure_tree, start, end, | |
ceeb0ae7 | 2513 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 | 2514 | if (ret >= 0) |
47dc196a | 2515 | ret = set_state_failrec(failure_tree, start, failrec); |
4a54c8c1 JS |
2516 | /* set the bits in the inode's tree */ |
2517 | if (ret >= 0) | |
ceeb0ae7 | 2518 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED); |
4a54c8c1 JS |
2519 | if (ret < 0) { |
2520 | kfree(failrec); | |
2521 | return ret; | |
2522 | } | |
2523 | } else { | |
ab8d0fc4 JM |
2524 | btrfs_debug(fs_info, |
2525 | "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d", | |
2526 | failrec->logical, failrec->start, failrec->len, | |
2527 | failrec->in_validation); | |
4a54c8c1 JS |
2528 | /* |
2529 | * when data can be on disk more than twice, add to failrec here | |
2530 | * (e.g. with a list for failed_mirror) to make | |
2531 | * clean_io_failure() clean all those errors at once. | |
2532 | */ | |
2533 | } | |
2fe6303e MX |
2534 | |
2535 | *failrec_ret = failrec; | |
2536 | ||
2537 | return 0; | |
2538 | } | |
2539 | ||
ce06d3ec OS |
2540 | static bool btrfs_check_repairable(struct inode *inode, bool needs_validation, |
2541 | struct io_failure_record *failrec, | |
2542 | int failed_mirror) | |
2fe6303e | 2543 | { |
ab8d0fc4 | 2544 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e MX |
2545 | int num_copies; |
2546 | ||
ab8d0fc4 | 2547 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
4a54c8c1 JS |
2548 | if (num_copies == 1) { |
2549 | /* | |
2550 | * we only have a single copy of the data, so don't bother with | |
2551 | * all the retry and error correction code that follows. no | |
2552 | * matter what the error is, it is very likely to persist. | |
2553 | */ | |
ab8d0fc4 JM |
2554 | btrfs_debug(fs_info, |
2555 | "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
2556 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2557 | return false; |
4a54c8c1 JS |
2558 | } |
2559 | ||
4a54c8c1 JS |
2560 | /* |
2561 | * there are two premises: | |
2562 | * a) deliver good data to the caller | |
2563 | * b) correct the bad sectors on disk | |
2564 | */ | |
c7333972 | 2565 | if (needs_validation) { |
4a54c8c1 JS |
2566 | /* |
2567 | * to fulfill b), we need to know the exact failing sectors, as | |
2568 | * we don't want to rewrite any more than the failed ones. thus, | |
2569 | * we need separate read requests for the failed bio | |
2570 | * | |
2571 | * if the following BUG_ON triggers, our validation request got | |
2572 | * merged. we need separate requests for our algorithm to work. | |
2573 | */ | |
2574 | BUG_ON(failrec->in_validation); | |
2575 | failrec->in_validation = 1; | |
2576 | failrec->this_mirror = failed_mirror; | |
4a54c8c1 JS |
2577 | } else { |
2578 | /* | |
2579 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2580 | * of the failed sector and if we succeed, we have setup | |
2581 | * everything for repair_io_failure to do the rest for us. | |
2582 | */ | |
2583 | if (failrec->in_validation) { | |
2584 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2585 | failrec->in_validation = 0; | |
2586 | failrec->this_mirror = 0; | |
2587 | } | |
2588 | failrec->failed_mirror = failed_mirror; | |
2589 | failrec->this_mirror++; | |
2590 | if (failrec->this_mirror == failed_mirror) | |
2591 | failrec->this_mirror++; | |
4a54c8c1 JS |
2592 | } |
2593 | ||
facc8a22 | 2594 | if (failrec->this_mirror > num_copies) { |
ab8d0fc4 JM |
2595 | btrfs_debug(fs_info, |
2596 | "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
2597 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2598 | return false; |
4a54c8c1 JS |
2599 | } |
2600 | ||
c3cfb656 | 2601 | return true; |
2fe6303e MX |
2602 | } |
2603 | ||
2604 | ||
2605 | struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio, | |
2606 | struct io_failure_record *failrec, | |
2607 | struct page *page, int pg_offset, int icsum, | |
8b110e39 | 2608 | bio_end_io_t *endio_func, void *data) |
2fe6303e | 2609 | { |
0b246afa | 2610 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e MX |
2611 | struct bio *bio; |
2612 | struct btrfs_io_bio *btrfs_failed_bio; | |
2613 | struct btrfs_io_bio *btrfs_bio; | |
2614 | ||
c5e4c3d7 | 2615 | bio = btrfs_io_bio_alloc(1); |
2fe6303e | 2616 | bio->bi_end_io = endio_func; |
4f024f37 | 2617 | bio->bi_iter.bi_sector = failrec->logical >> 9; |
4f024f37 | 2618 | bio->bi_iter.bi_size = 0; |
8b110e39 | 2619 | bio->bi_private = data; |
4a54c8c1 | 2620 | |
facc8a22 MX |
2621 | btrfs_failed_bio = btrfs_io_bio(failed_bio); |
2622 | if (btrfs_failed_bio->csum) { | |
facc8a22 MX |
2623 | u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); |
2624 | ||
2625 | btrfs_bio = btrfs_io_bio(bio); | |
2626 | btrfs_bio->csum = btrfs_bio->csum_inline; | |
2fe6303e MX |
2627 | icsum *= csum_size; |
2628 | memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum, | |
facc8a22 MX |
2629 | csum_size); |
2630 | } | |
2631 | ||
2fe6303e | 2632 | bio_add_page(bio, page, failrec->len, pg_offset); |
fd9d6670 OS |
2633 | btrfs_io_bio(bio)->logical = failrec->start; |
2634 | btrfs_io_bio(bio)->iter = bio->bi_iter; | |
2fe6303e MX |
2635 | |
2636 | return bio; | |
2637 | } | |
2638 | ||
c7333972 OS |
2639 | static bool btrfs_io_needs_validation(struct inode *inode, struct bio *bio) |
2640 | { | |
2641 | struct bio_vec *bvec; | |
2642 | u64 len = 0; | |
2643 | int i; | |
2644 | ||
f337bd74 OS |
2645 | /* |
2646 | * If bi_status is BLK_STS_OK, then this was a checksum error, not an | |
2647 | * I/O error. In this case, we already know exactly which sector was | |
2648 | * bad, so we don't need to validate. | |
2649 | */ | |
2650 | if (bio->bi_status == BLK_STS_OK) | |
2651 | return false; | |
2652 | ||
c7333972 OS |
2653 | /* |
2654 | * We need to validate each sector individually if the failed I/O was | |
2655 | * for multiple sectors. | |
2656 | */ | |
2657 | bio_for_each_bvec_all(bvec, bio, i) { | |
2658 | len += bvec->bv_len; | |
2659 | if (len > inode->i_sb->s_blocksize) | |
2660 | return true; | |
2661 | } | |
2662 | return false; | |
2663 | } | |
2664 | ||
2fe6303e | 2665 | /* |
78e62c02 NB |
2666 | * This is a generic handler for readpage errors. If other copies exist, read |
2667 | * those and write back good data to the failed position. Does not investigate | |
2668 | * in remapping the failed extent elsewhere, hoping the device will be smart | |
2669 | * enough to do this as needed | |
2fe6303e | 2670 | */ |
2fe6303e MX |
2671 | static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset, |
2672 | struct page *page, u64 start, u64 end, | |
2673 | int failed_mirror) | |
2674 | { | |
2675 | struct io_failure_record *failrec; | |
2676 | struct inode *inode = page->mapping->host; | |
2677 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
7870d082 | 2678 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
c7333972 | 2679 | bool need_validation; |
2fe6303e | 2680 | struct bio *bio; |
70fd7614 | 2681 | int read_mode = 0; |
4e4cbee9 | 2682 | blk_status_t status; |
2fe6303e MX |
2683 | int ret; |
2684 | ||
1f7ad75b | 2685 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
2fe6303e MX |
2686 | |
2687 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
2688 | if (ret) | |
2689 | return ret; | |
2690 | ||
c7333972 OS |
2691 | need_validation = btrfs_io_needs_validation(inode, failed_bio); |
2692 | ||
2693 | if (!btrfs_check_repairable(inode, need_validation, failrec, | |
c3cfb656 | 2694 | failed_mirror)) { |
7870d082 | 2695 | free_io_failure(failure_tree, tree, failrec); |
2fe6303e MX |
2696 | return -EIO; |
2697 | } | |
2698 | ||
c7333972 | 2699 | if (need_validation) |
70fd7614 | 2700 | read_mode |= REQ_FAILFAST_DEV; |
2fe6303e MX |
2701 | |
2702 | phy_offset >>= inode->i_sb->s_blocksize_bits; | |
2703 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2704 | start - page_offset(page), | |
8b110e39 MX |
2705 | (int)phy_offset, failed_bio->bi_end_io, |
2706 | NULL); | |
ebcc3263 | 2707 | bio->bi_opf = REQ_OP_READ | read_mode; |
4a54c8c1 | 2708 | |
ab8d0fc4 JM |
2709 | btrfs_debug(btrfs_sb(inode->i_sb), |
2710 | "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d", | |
2711 | read_mode, failrec->this_mirror, failrec->in_validation); | |
4a54c8c1 | 2712 | |
8c27cb35 | 2713 | status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror, |
50489a57 | 2714 | failrec->bio_flags); |
4e4cbee9 | 2715 | if (status) { |
7870d082 | 2716 | free_io_failure(failure_tree, tree, failrec); |
6c387ab2 | 2717 | bio_put(bio); |
4e4cbee9 | 2718 | ret = blk_status_to_errno(status); |
6c387ab2 MX |
2719 | } |
2720 | ||
013bd4c3 | 2721 | return ret; |
4a54c8c1 JS |
2722 | } |
2723 | ||
d1310b2e CM |
2724 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2725 | ||
b5227c07 | 2726 | void end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
87826df0 JM |
2727 | { |
2728 | int uptodate = (err == 0); | |
3e2426bd | 2729 | int ret = 0; |
87826df0 | 2730 | |
c629732d | 2731 | btrfs_writepage_endio_finish_ordered(page, start, end, uptodate); |
87826df0 | 2732 | |
87826df0 | 2733 | if (!uptodate) { |
87826df0 JM |
2734 | ClearPageUptodate(page); |
2735 | SetPageError(page); | |
bff5baf8 | 2736 | ret = err < 0 ? err : -EIO; |
5dca6eea | 2737 | mapping_set_error(page->mapping, ret); |
87826df0 | 2738 | } |
87826df0 JM |
2739 | } |
2740 | ||
d1310b2e CM |
2741 | /* |
2742 | * after a writepage IO is done, we need to: | |
2743 | * clear the uptodate bits on error | |
2744 | * clear the writeback bits in the extent tree for this IO | |
2745 | * end_page_writeback if the page has no more pending IO | |
2746 | * | |
2747 | * Scheduling is not allowed, so the extent state tree is expected | |
2748 | * to have one and only one object corresponding to this IO. | |
2749 | */ | |
4246a0b6 | 2750 | static void end_bio_extent_writepage(struct bio *bio) |
d1310b2e | 2751 | { |
4e4cbee9 | 2752 | int error = blk_status_to_errno(bio->bi_status); |
2c30c71b | 2753 | struct bio_vec *bvec; |
d1310b2e CM |
2754 | u64 start; |
2755 | u64 end; | |
6dc4f100 | 2756 | struct bvec_iter_all iter_all; |
d1310b2e | 2757 | |
c09abff8 | 2758 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 2759 | bio_for_each_segment_all(bvec, bio, iter_all) { |
d1310b2e | 2760 | struct page *page = bvec->bv_page; |
0b246afa JM |
2761 | struct inode *inode = page->mapping->host; |
2762 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
902b22f3 | 2763 | |
17a5adcc AO |
2764 | /* We always issue full-page reads, but if some block |
2765 | * in a page fails to read, blk_update_request() will | |
2766 | * advance bv_offset and adjust bv_len to compensate. | |
2767 | * Print a warning for nonzero offsets, and an error | |
2768 | * if they don't add up to a full page. */ | |
09cbfeaf KS |
2769 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { |
2770 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
0b246afa | 2771 | btrfs_err(fs_info, |
efe120a0 FH |
2772 | "partial page write in btrfs with offset %u and length %u", |
2773 | bvec->bv_offset, bvec->bv_len); | |
2774 | else | |
0b246afa | 2775 | btrfs_info(fs_info, |
5d163e0e | 2776 | "incomplete page write in btrfs with offset %u and length %u", |
efe120a0 FH |
2777 | bvec->bv_offset, bvec->bv_len); |
2778 | } | |
d1310b2e | 2779 | |
17a5adcc AO |
2780 | start = page_offset(page); |
2781 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
d1310b2e | 2782 | |
4e4cbee9 | 2783 | end_extent_writepage(page, error, start, end); |
17a5adcc | 2784 | end_page_writeback(page); |
2c30c71b | 2785 | } |
2b1f55b0 | 2786 | |
d1310b2e | 2787 | bio_put(bio); |
d1310b2e CM |
2788 | } |
2789 | ||
883d0de4 MX |
2790 | static void |
2791 | endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len, | |
2792 | int uptodate) | |
2793 | { | |
2794 | struct extent_state *cached = NULL; | |
2795 | u64 end = start + len - 1; | |
2796 | ||
2797 | if (uptodate && tree->track_uptodate) | |
2798 | set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC); | |
d810a4be | 2799 | unlock_extent_cached_atomic(tree, start, end, &cached); |
883d0de4 MX |
2800 | } |
2801 | ||
d1310b2e CM |
2802 | /* |
2803 | * after a readpage IO is done, we need to: | |
2804 | * clear the uptodate bits on error | |
2805 | * set the uptodate bits if things worked | |
2806 | * set the page up to date if all extents in the tree are uptodate | |
2807 | * clear the lock bit in the extent tree | |
2808 | * unlock the page if there are no other extents locked for it | |
2809 | * | |
2810 | * Scheduling is not allowed, so the extent state tree is expected | |
2811 | * to have one and only one object corresponding to this IO. | |
2812 | */ | |
4246a0b6 | 2813 | static void end_bio_extent_readpage(struct bio *bio) |
d1310b2e | 2814 | { |
2c30c71b | 2815 | struct bio_vec *bvec; |
4e4cbee9 | 2816 | int uptodate = !bio->bi_status; |
facc8a22 | 2817 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); |
7870d082 | 2818 | struct extent_io_tree *tree, *failure_tree; |
facc8a22 | 2819 | u64 offset = 0; |
d1310b2e CM |
2820 | u64 start; |
2821 | u64 end; | |
facc8a22 | 2822 | u64 len; |
883d0de4 MX |
2823 | u64 extent_start = 0; |
2824 | u64 extent_len = 0; | |
5cf1ab56 | 2825 | int mirror; |
d1310b2e | 2826 | int ret; |
6dc4f100 | 2827 | struct bvec_iter_all iter_all; |
d1310b2e | 2828 | |
c09abff8 | 2829 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 2830 | bio_for_each_segment_all(bvec, bio, iter_all) { |
d1310b2e | 2831 | struct page *page = bvec->bv_page; |
a71754fc | 2832 | struct inode *inode = page->mapping->host; |
ab8d0fc4 | 2833 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
78e62c02 NB |
2834 | bool data_inode = btrfs_ino(BTRFS_I(inode)) |
2835 | != BTRFS_BTREE_INODE_OBJECTID; | |
507903b8 | 2836 | |
ab8d0fc4 JM |
2837 | btrfs_debug(fs_info, |
2838 | "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u", | |
4e4cbee9 | 2839 | (u64)bio->bi_iter.bi_sector, bio->bi_status, |
ab8d0fc4 | 2840 | io_bio->mirror_num); |
a71754fc | 2841 | tree = &BTRFS_I(inode)->io_tree; |
7870d082 | 2842 | failure_tree = &BTRFS_I(inode)->io_failure_tree; |
902b22f3 | 2843 | |
17a5adcc AO |
2844 | /* We always issue full-page reads, but if some block |
2845 | * in a page fails to read, blk_update_request() will | |
2846 | * advance bv_offset and adjust bv_len to compensate. | |
2847 | * Print a warning for nonzero offsets, and an error | |
2848 | * if they don't add up to a full page. */ | |
09cbfeaf KS |
2849 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { |
2850 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
ab8d0fc4 JM |
2851 | btrfs_err(fs_info, |
2852 | "partial page read in btrfs with offset %u and length %u", | |
efe120a0 FH |
2853 | bvec->bv_offset, bvec->bv_len); |
2854 | else | |
ab8d0fc4 JM |
2855 | btrfs_info(fs_info, |
2856 | "incomplete page read in btrfs with offset %u and length %u", | |
efe120a0 FH |
2857 | bvec->bv_offset, bvec->bv_len); |
2858 | } | |
d1310b2e | 2859 | |
17a5adcc AO |
2860 | start = page_offset(page); |
2861 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
facc8a22 | 2862 | len = bvec->bv_len; |
d1310b2e | 2863 | |
9be3395b | 2864 | mirror = io_bio->mirror_num; |
78e62c02 | 2865 | if (likely(uptodate)) { |
facc8a22 MX |
2866 | ret = tree->ops->readpage_end_io_hook(io_bio, offset, |
2867 | page, start, end, | |
2868 | mirror); | |
5ee0844d | 2869 | if (ret) |
d1310b2e | 2870 | uptodate = 0; |
5ee0844d | 2871 | else |
7870d082 JB |
2872 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
2873 | failure_tree, tree, start, | |
2874 | page, | |
2875 | btrfs_ino(BTRFS_I(inode)), 0); | |
d1310b2e | 2876 | } |
ea466794 | 2877 | |
f2a09da9 MX |
2878 | if (likely(uptodate)) |
2879 | goto readpage_ok; | |
2880 | ||
78e62c02 | 2881 | if (data_inode) { |
9d0d1c8b | 2882 | |
f4a8e656 | 2883 | /* |
78e62c02 NB |
2884 | * The generic bio_readpage_error handles errors the |
2885 | * following way: If possible, new read requests are | |
2886 | * created and submitted and will end up in | |
2887 | * end_bio_extent_readpage as well (if we're lucky, | |
2888 | * not in the !uptodate case). In that case it returns | |
2889 | * 0 and we just go on with the next page in our bio. | |
2890 | * If it can't handle the error it will return -EIO and | |
2891 | * we remain responsible for that page. | |
f4a8e656 | 2892 | */ |
78e62c02 NB |
2893 | ret = bio_readpage_error(bio, offset, page, start, end, |
2894 | mirror); | |
2895 | if (ret == 0) { | |
2896 | uptodate = !bio->bi_status; | |
2897 | offset += len; | |
2898 | continue; | |
2899 | } | |
2900 | } else { | |
2901 | struct extent_buffer *eb; | |
2902 | ||
2903 | eb = (struct extent_buffer *)page->private; | |
2904 | set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); | |
2905 | eb->read_mirror = mirror; | |
2906 | atomic_dec(&eb->io_pages); | |
2907 | if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, | |
2908 | &eb->bflags)) | |
2909 | btree_readahead_hook(eb, -EIO); | |
7e38326f | 2910 | } |
f2a09da9 | 2911 | readpage_ok: |
883d0de4 | 2912 | if (likely(uptodate)) { |
a71754fc | 2913 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 2914 | pgoff_t end_index = i_size >> PAGE_SHIFT; |
a583c026 | 2915 | unsigned off; |
a71754fc JB |
2916 | |
2917 | /* Zero out the end if this page straddles i_size */ | |
7073017a | 2918 | off = offset_in_page(i_size); |
a583c026 | 2919 | if (page->index == end_index && off) |
09cbfeaf | 2920 | zero_user_segment(page, off, PAGE_SIZE); |
17a5adcc | 2921 | SetPageUptodate(page); |
70dec807 | 2922 | } else { |
17a5adcc AO |
2923 | ClearPageUptodate(page); |
2924 | SetPageError(page); | |
70dec807 | 2925 | } |
17a5adcc | 2926 | unlock_page(page); |
facc8a22 | 2927 | offset += len; |
883d0de4 MX |
2928 | |
2929 | if (unlikely(!uptodate)) { | |
2930 | if (extent_len) { | |
2931 | endio_readpage_release_extent(tree, | |
2932 | extent_start, | |
2933 | extent_len, 1); | |
2934 | extent_start = 0; | |
2935 | extent_len = 0; | |
2936 | } | |
2937 | endio_readpage_release_extent(tree, start, | |
2938 | end - start + 1, 0); | |
2939 | } else if (!extent_len) { | |
2940 | extent_start = start; | |
2941 | extent_len = end + 1 - start; | |
2942 | } else if (extent_start + extent_len == start) { | |
2943 | extent_len += end + 1 - start; | |
2944 | } else { | |
2945 | endio_readpage_release_extent(tree, extent_start, | |
2946 | extent_len, uptodate); | |
2947 | extent_start = start; | |
2948 | extent_len = end + 1 - start; | |
2949 | } | |
2c30c71b | 2950 | } |
d1310b2e | 2951 | |
883d0de4 MX |
2952 | if (extent_len) |
2953 | endio_readpage_release_extent(tree, extent_start, extent_len, | |
2954 | uptodate); | |
b3a0dd50 | 2955 | btrfs_io_bio_free_csum(io_bio); |
d1310b2e | 2956 | bio_put(bio); |
d1310b2e CM |
2957 | } |
2958 | ||
9be3395b | 2959 | /* |
184f999e DS |
2960 | * Initialize the members up to but not including 'bio'. Use after allocating a |
2961 | * new bio by bio_alloc_bioset as it does not initialize the bytes outside of | |
2962 | * 'bio' because use of __GFP_ZERO is not supported. | |
9be3395b | 2963 | */ |
184f999e | 2964 | static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio) |
d1310b2e | 2965 | { |
184f999e DS |
2966 | memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio)); |
2967 | } | |
d1310b2e | 2968 | |
9be3395b | 2969 | /* |
6e707bcd DS |
2970 | * The following helpers allocate a bio. As it's backed by a bioset, it'll |
2971 | * never fail. We're returning a bio right now but you can call btrfs_io_bio | |
2972 | * for the appropriate container_of magic | |
9be3395b | 2973 | */ |
e749af44 | 2974 | struct bio *btrfs_bio_alloc(u64 first_byte) |
d1310b2e CM |
2975 | { |
2976 | struct bio *bio; | |
d1310b2e | 2977 | |
8ac9f7c1 | 2978 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &btrfs_bioset); |
c821e7f3 | 2979 | bio->bi_iter.bi_sector = first_byte >> 9; |
184f999e | 2980 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
d1310b2e CM |
2981 | return bio; |
2982 | } | |
2983 | ||
8b6c1d56 | 2984 | struct bio *btrfs_bio_clone(struct bio *bio) |
9be3395b | 2985 | { |
23ea8e5a MX |
2986 | struct btrfs_io_bio *btrfs_bio; |
2987 | struct bio *new; | |
9be3395b | 2988 | |
6e707bcd | 2989 | /* Bio allocation backed by a bioset does not fail */ |
8ac9f7c1 | 2990 | new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset); |
6e707bcd | 2991 | btrfs_bio = btrfs_io_bio(new); |
184f999e | 2992 | btrfs_io_bio_init(btrfs_bio); |
6e707bcd | 2993 | btrfs_bio->iter = bio->bi_iter; |
23ea8e5a MX |
2994 | return new; |
2995 | } | |
9be3395b | 2996 | |
c5e4c3d7 | 2997 | struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs) |
9be3395b | 2998 | { |
facc8a22 MX |
2999 | struct bio *bio; |
3000 | ||
6e707bcd | 3001 | /* Bio allocation backed by a bioset does not fail */ |
8ac9f7c1 | 3002 | bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset); |
184f999e | 3003 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
facc8a22 | 3004 | return bio; |
9be3395b CM |
3005 | } |
3006 | ||
e477094f | 3007 | struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size) |
2f8e9140 LB |
3008 | { |
3009 | struct bio *bio; | |
3010 | struct btrfs_io_bio *btrfs_bio; | |
3011 | ||
3012 | /* this will never fail when it's backed by a bioset */ | |
8ac9f7c1 | 3013 | bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset); |
2f8e9140 LB |
3014 | ASSERT(bio); |
3015 | ||
3016 | btrfs_bio = btrfs_io_bio(bio); | |
184f999e | 3017 | btrfs_io_bio_init(btrfs_bio); |
2f8e9140 LB |
3018 | |
3019 | bio_trim(bio, offset >> 9, size >> 9); | |
17347cec | 3020 | btrfs_bio->iter = bio->bi_iter; |
2f8e9140 LB |
3021 | return bio; |
3022 | } | |
9be3395b | 3023 | |
4b81ba48 DS |
3024 | /* |
3025 | * @opf: bio REQ_OP_* and REQ_* flags as one value | |
b8b3d625 DS |
3026 | * @wbc: optional writeback control for io accounting |
3027 | * @page: page to add to the bio | |
3028 | * @pg_offset: offset of the new bio or to check whether we are adding | |
3029 | * a contiguous page to the previous one | |
3030 | * @size: portion of page that we want to write | |
3031 | * @offset: starting offset in the page | |
5c2b1fd7 | 3032 | * @bio_ret: must be valid pointer, newly allocated bio will be stored there |
b8b3d625 DS |
3033 | * @end_io_func: end_io callback for new bio |
3034 | * @mirror_num: desired mirror to read/write | |
3035 | * @prev_bio_flags: flags of previous bio to see if we can merge the current one | |
3036 | * @bio_flags: flags of the current bio to see if we can merge them | |
4b81ba48 | 3037 | */ |
0ceb34bf | 3038 | static int submit_extent_page(unsigned int opf, |
da2f0f74 | 3039 | struct writeback_control *wbc, |
6273b7f8 | 3040 | struct page *page, u64 offset, |
6c5a4e2c | 3041 | size_t size, unsigned long pg_offset, |
d1310b2e | 3042 | struct bio **bio_ret, |
f188591e | 3043 | bio_end_io_t end_io_func, |
c8b97818 CM |
3044 | int mirror_num, |
3045 | unsigned long prev_bio_flags, | |
005efedf FM |
3046 | unsigned long bio_flags, |
3047 | bool force_bio_submit) | |
d1310b2e CM |
3048 | { |
3049 | int ret = 0; | |
3050 | struct bio *bio; | |
09cbfeaf | 3051 | size_t page_size = min_t(size_t, size, PAGE_SIZE); |
6273b7f8 | 3052 | sector_t sector = offset >> 9; |
0ceb34bf | 3053 | struct extent_io_tree *tree = &BTRFS_I(page->mapping->host)->io_tree; |
d1310b2e | 3054 | |
5c2b1fd7 DS |
3055 | ASSERT(bio_ret); |
3056 | ||
3057 | if (*bio_ret) { | |
0c8508a6 DS |
3058 | bool contig; |
3059 | bool can_merge = true; | |
3060 | ||
d1310b2e | 3061 | bio = *bio_ret; |
0c8508a6 | 3062 | if (prev_bio_flags & EXTENT_BIO_COMPRESSED) |
4f024f37 | 3063 | contig = bio->bi_iter.bi_sector == sector; |
c8b97818 | 3064 | else |
f73a1c7d | 3065 | contig = bio_end_sector(bio) == sector; |
c8b97818 | 3066 | |
da12fe54 NB |
3067 | ASSERT(tree->ops); |
3068 | if (btrfs_bio_fits_in_stripe(page, page_size, bio, bio_flags)) | |
0c8508a6 DS |
3069 | can_merge = false; |
3070 | ||
3071 | if (prev_bio_flags != bio_flags || !contig || !can_merge || | |
005efedf | 3072 | force_bio_submit || |
6c5a4e2c | 3073 | bio_add_page(bio, page, page_size, pg_offset) < page_size) { |
1f7ad75b | 3074 | ret = submit_one_bio(bio, mirror_num, prev_bio_flags); |
289454ad NA |
3075 | if (ret < 0) { |
3076 | *bio_ret = NULL; | |
79787eaa | 3077 | return ret; |
289454ad | 3078 | } |
d1310b2e CM |
3079 | bio = NULL; |
3080 | } else { | |
da2f0f74 | 3081 | if (wbc) |
34e51a5e | 3082 | wbc_account_cgroup_owner(wbc, page, page_size); |
d1310b2e CM |
3083 | return 0; |
3084 | } | |
3085 | } | |
c8b97818 | 3086 | |
e749af44 | 3087 | bio = btrfs_bio_alloc(offset); |
6c5a4e2c | 3088 | bio_add_page(bio, page, page_size, pg_offset); |
d1310b2e CM |
3089 | bio->bi_end_io = end_io_func; |
3090 | bio->bi_private = tree; | |
e6959b93 | 3091 | bio->bi_write_hint = page->mapping->host->i_write_hint; |
4b81ba48 | 3092 | bio->bi_opf = opf; |
da2f0f74 | 3093 | if (wbc) { |
429aebc0 DS |
3094 | struct block_device *bdev; |
3095 | ||
3096 | bdev = BTRFS_I(page->mapping->host)->root->fs_info->fs_devices->latest_bdev; | |
3097 | bio_set_dev(bio, bdev); | |
da2f0f74 | 3098 | wbc_init_bio(wbc, bio); |
34e51a5e | 3099 | wbc_account_cgroup_owner(wbc, page, page_size); |
da2f0f74 | 3100 | } |
70dec807 | 3101 | |
5c2b1fd7 | 3102 | *bio_ret = bio; |
d1310b2e CM |
3103 | |
3104 | return ret; | |
3105 | } | |
3106 | ||
48a3b636 ES |
3107 | static void attach_extent_buffer_page(struct extent_buffer *eb, |
3108 | struct page *page) | |
d1310b2e CM |
3109 | { |
3110 | if (!PagePrivate(page)) { | |
3111 | SetPagePrivate(page); | |
09cbfeaf | 3112 | get_page(page); |
4f2de97a JB |
3113 | set_page_private(page, (unsigned long)eb); |
3114 | } else { | |
3115 | WARN_ON(page->private != (unsigned long)eb); | |
d1310b2e CM |
3116 | } |
3117 | } | |
3118 | ||
4f2de97a | 3119 | void set_page_extent_mapped(struct page *page) |
d1310b2e | 3120 | { |
4f2de97a JB |
3121 | if (!PagePrivate(page)) { |
3122 | SetPagePrivate(page); | |
09cbfeaf | 3123 | get_page(page); |
4f2de97a JB |
3124 | set_page_private(page, EXTENT_PAGE_PRIVATE); |
3125 | } | |
d1310b2e CM |
3126 | } |
3127 | ||
125bac01 MX |
3128 | static struct extent_map * |
3129 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
3130 | u64 start, u64 len, get_extent_t *get_extent, | |
3131 | struct extent_map **em_cached) | |
3132 | { | |
3133 | struct extent_map *em; | |
3134 | ||
3135 | if (em_cached && *em_cached) { | |
3136 | em = *em_cached; | |
cbc0e928 | 3137 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 | 3138 | start < extent_map_end(em)) { |
490b54d6 | 3139 | refcount_inc(&em->refs); |
125bac01 MX |
3140 | return em; |
3141 | } | |
3142 | ||
3143 | free_extent_map(em); | |
3144 | *em_cached = NULL; | |
3145 | } | |
3146 | ||
39b07b5d | 3147 | em = get_extent(BTRFS_I(inode), page, pg_offset, start, len); |
125bac01 MX |
3148 | if (em_cached && !IS_ERR_OR_NULL(em)) { |
3149 | BUG_ON(*em_cached); | |
490b54d6 | 3150 | refcount_inc(&em->refs); |
125bac01 MX |
3151 | *em_cached = em; |
3152 | } | |
3153 | return em; | |
3154 | } | |
d1310b2e CM |
3155 | /* |
3156 | * basic readpage implementation. Locked extent state structs are inserted | |
3157 | * into the tree that are removed when the IO is done (by the end_io | |
3158 | * handlers) | |
79787eaa | 3159 | * XXX JDM: This needs looking at to ensure proper page locking |
baf863b9 | 3160 | * return 0 on success, otherwise return error |
d1310b2e | 3161 | */ |
f657a31c | 3162 | static int __do_readpage(struct page *page, |
9974090b | 3163 | get_extent_t *get_extent, |
125bac01 | 3164 | struct extent_map **em_cached, |
9974090b | 3165 | struct bio **bio, int mirror_num, |
f1c77c55 | 3166 | unsigned long *bio_flags, unsigned int read_flags, |
005efedf | 3167 | u64 *prev_em_start) |
d1310b2e CM |
3168 | { |
3169 | struct inode *inode = page->mapping->host; | |
4eee4fa4 | 3170 | u64 start = page_offset(page); |
8eec8296 | 3171 | const u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
3172 | u64 cur = start; |
3173 | u64 extent_offset; | |
3174 | u64 last_byte = i_size_read(inode); | |
3175 | u64 block_start; | |
3176 | u64 cur_end; | |
d1310b2e | 3177 | struct extent_map *em; |
baf863b9 | 3178 | int ret = 0; |
d1310b2e | 3179 | int nr = 0; |
306e16ce | 3180 | size_t pg_offset = 0; |
d1310b2e | 3181 | size_t iosize; |
c8b97818 | 3182 | size_t disk_io_size; |
d1310b2e | 3183 | size_t blocksize = inode->i_sb->s_blocksize; |
7f042a83 | 3184 | unsigned long this_bio_flag = 0; |
f657a31c | 3185 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
ae6957eb | 3186 | |
d1310b2e CM |
3187 | set_page_extent_mapped(page); |
3188 | ||
90a887c9 DM |
3189 | if (!PageUptodate(page)) { |
3190 | if (cleancache_get_page(page) == 0) { | |
3191 | BUG_ON(blocksize != PAGE_SIZE); | |
9974090b | 3192 | unlock_extent(tree, start, end); |
90a887c9 DM |
3193 | goto out; |
3194 | } | |
3195 | } | |
3196 | ||
09cbfeaf | 3197 | if (page->index == last_byte >> PAGE_SHIFT) { |
c8b97818 | 3198 | char *userpage; |
7073017a | 3199 | size_t zero_offset = offset_in_page(last_byte); |
c8b97818 CM |
3200 | |
3201 | if (zero_offset) { | |
09cbfeaf | 3202 | iosize = PAGE_SIZE - zero_offset; |
7ac687d9 | 3203 | userpage = kmap_atomic(page); |
c8b97818 CM |
3204 | memset(userpage + zero_offset, 0, iosize); |
3205 | flush_dcache_page(page); | |
7ac687d9 | 3206 | kunmap_atomic(userpage); |
c8b97818 CM |
3207 | } |
3208 | } | |
d1310b2e | 3209 | while (cur <= end) { |
005efedf | 3210 | bool force_bio_submit = false; |
6273b7f8 | 3211 | u64 offset; |
c8f2f24b | 3212 | |
d1310b2e CM |
3213 | if (cur >= last_byte) { |
3214 | char *userpage; | |
507903b8 AJ |
3215 | struct extent_state *cached = NULL; |
3216 | ||
09cbfeaf | 3217 | iosize = PAGE_SIZE - pg_offset; |
7ac687d9 | 3218 | userpage = kmap_atomic(page); |
306e16ce | 3219 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3220 | flush_dcache_page(page); |
7ac687d9 | 3221 | kunmap_atomic(userpage); |
d1310b2e | 3222 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 3223 | &cached, GFP_NOFS); |
7f042a83 | 3224 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3225 | cur + iosize - 1, &cached); |
d1310b2e CM |
3226 | break; |
3227 | } | |
125bac01 MX |
3228 | em = __get_extent_map(inode, page, pg_offset, cur, |
3229 | end - cur + 1, get_extent, em_cached); | |
c704005d | 3230 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3231 | SetPageError(page); |
7f042a83 | 3232 | unlock_extent(tree, cur, end); |
d1310b2e CM |
3233 | break; |
3234 | } | |
d1310b2e CM |
3235 | extent_offset = cur - em->start; |
3236 | BUG_ON(extent_map_end(em) <= cur); | |
3237 | BUG_ON(end < cur); | |
3238 | ||
261507a0 | 3239 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
4b384318 | 3240 | this_bio_flag |= EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
3241 | extent_set_compress_type(&this_bio_flag, |
3242 | em->compress_type); | |
3243 | } | |
c8b97818 | 3244 | |
d1310b2e CM |
3245 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
3246 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 3247 | iosize = ALIGN(iosize, blocksize); |
c8b97818 CM |
3248 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
3249 | disk_io_size = em->block_len; | |
6273b7f8 | 3250 | offset = em->block_start; |
c8b97818 | 3251 | } else { |
6273b7f8 | 3252 | offset = em->block_start + extent_offset; |
c8b97818 CM |
3253 | disk_io_size = iosize; |
3254 | } | |
d1310b2e | 3255 | block_start = em->block_start; |
d899e052 YZ |
3256 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
3257 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
3258 | |
3259 | /* | |
3260 | * If we have a file range that points to a compressed extent | |
3261 | * and it's followed by a consecutive file range that points to | |
3262 | * to the same compressed extent (possibly with a different | |
3263 | * offset and/or length, so it either points to the whole extent | |
3264 | * or only part of it), we must make sure we do not submit a | |
3265 | * single bio to populate the pages for the 2 ranges because | |
3266 | * this makes the compressed extent read zero out the pages | |
3267 | * belonging to the 2nd range. Imagine the following scenario: | |
3268 | * | |
3269 | * File layout | |
3270 | * [0 - 8K] [8K - 24K] | |
3271 | * | | | |
3272 | * | | | |
3273 | * points to extent X, points to extent X, | |
3274 | * offset 4K, length of 8K offset 0, length 16K | |
3275 | * | |
3276 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
3277 | * | |
3278 | * If the bio to read the compressed extent covers both ranges, | |
3279 | * it will decompress extent X into the pages belonging to the | |
3280 | * first range and then it will stop, zeroing out the remaining | |
3281 | * pages that belong to the other range that points to extent X. | |
3282 | * So here we make sure we submit 2 bios, one for the first | |
3283 | * range and another one for the third range. Both will target | |
3284 | * the same physical extent from disk, but we can't currently | |
3285 | * make the compressed bio endio callback populate the pages | |
3286 | * for both ranges because each compressed bio is tightly | |
3287 | * coupled with a single extent map, and each range can have | |
3288 | * an extent map with a different offset value relative to the | |
3289 | * uncompressed data of our extent and different lengths. This | |
3290 | * is a corner case so we prioritize correctness over | |
3291 | * non-optimal behavior (submitting 2 bios for the same extent). | |
3292 | */ | |
3293 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
3294 | prev_em_start && *prev_em_start != (u64)-1 && | |
8e928218 | 3295 | *prev_em_start != em->start) |
005efedf FM |
3296 | force_bio_submit = true; |
3297 | ||
3298 | if (prev_em_start) | |
8e928218 | 3299 | *prev_em_start = em->start; |
005efedf | 3300 | |
d1310b2e CM |
3301 | free_extent_map(em); |
3302 | em = NULL; | |
3303 | ||
3304 | /* we've found a hole, just zero and go on */ | |
3305 | if (block_start == EXTENT_MAP_HOLE) { | |
3306 | char *userpage; | |
507903b8 AJ |
3307 | struct extent_state *cached = NULL; |
3308 | ||
7ac687d9 | 3309 | userpage = kmap_atomic(page); |
306e16ce | 3310 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3311 | flush_dcache_page(page); |
7ac687d9 | 3312 | kunmap_atomic(userpage); |
d1310b2e CM |
3313 | |
3314 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 | 3315 | &cached, GFP_NOFS); |
7f042a83 | 3316 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3317 | cur + iosize - 1, &cached); |
d1310b2e | 3318 | cur = cur + iosize; |
306e16ce | 3319 | pg_offset += iosize; |
d1310b2e CM |
3320 | continue; |
3321 | } | |
3322 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
3323 | if (test_range_bit(tree, cur, cur_end, |
3324 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 3325 | check_page_uptodate(tree, page); |
7f042a83 | 3326 | unlock_extent(tree, cur, cur + iosize - 1); |
d1310b2e | 3327 | cur = cur + iosize; |
306e16ce | 3328 | pg_offset += iosize; |
d1310b2e CM |
3329 | continue; |
3330 | } | |
70dec807 CM |
3331 | /* we have an inline extent but it didn't get marked up |
3332 | * to date. Error out | |
3333 | */ | |
3334 | if (block_start == EXTENT_MAP_INLINE) { | |
3335 | SetPageError(page); | |
7f042a83 | 3336 | unlock_extent(tree, cur, cur + iosize - 1); |
70dec807 | 3337 | cur = cur + iosize; |
306e16ce | 3338 | pg_offset += iosize; |
70dec807 CM |
3339 | continue; |
3340 | } | |
d1310b2e | 3341 | |
0ceb34bf | 3342 | ret = submit_extent_page(REQ_OP_READ | read_flags, NULL, |
6273b7f8 | 3343 | page, offset, disk_io_size, |
fa17ed06 | 3344 | pg_offset, bio, |
c8b97818 CM |
3345 | end_bio_extent_readpage, mirror_num, |
3346 | *bio_flags, | |
005efedf FM |
3347 | this_bio_flag, |
3348 | force_bio_submit); | |
c8f2f24b JB |
3349 | if (!ret) { |
3350 | nr++; | |
3351 | *bio_flags = this_bio_flag; | |
3352 | } else { | |
d1310b2e | 3353 | SetPageError(page); |
7f042a83 | 3354 | unlock_extent(tree, cur, cur + iosize - 1); |
baf863b9 | 3355 | goto out; |
edd33c99 | 3356 | } |
d1310b2e | 3357 | cur = cur + iosize; |
306e16ce | 3358 | pg_offset += iosize; |
d1310b2e | 3359 | } |
90a887c9 | 3360 | out: |
d1310b2e CM |
3361 | if (!nr) { |
3362 | if (!PageError(page)) | |
3363 | SetPageUptodate(page); | |
3364 | unlock_page(page); | |
3365 | } | |
baf863b9 | 3366 | return ret; |
d1310b2e CM |
3367 | } |
3368 | ||
b6660e80 | 3369 | static inline void contiguous_readpages(struct page *pages[], int nr_pages, |
9974090b | 3370 | u64 start, u64 end, |
125bac01 | 3371 | struct extent_map **em_cached, |
d3fac6ba | 3372 | struct bio **bio, |
1f7ad75b | 3373 | unsigned long *bio_flags, |
808f80b4 | 3374 | u64 *prev_em_start) |
9974090b | 3375 | { |
23d31bd4 | 3376 | struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host); |
9974090b MX |
3377 | int index; |
3378 | ||
b272ae22 | 3379 | btrfs_lock_and_flush_ordered_range(inode, start, end, NULL); |
9974090b MX |
3380 | |
3381 | for (index = 0; index < nr_pages; index++) { | |
f657a31c | 3382 | __do_readpage(pages[index], btrfs_get_extent, em_cached, |
5e9d3982 | 3383 | bio, 0, bio_flags, REQ_RAHEAD, prev_em_start); |
09cbfeaf | 3384 | put_page(pages[index]); |
9974090b MX |
3385 | } |
3386 | } | |
3387 | ||
0d44fea7 | 3388 | static int __extent_read_full_page(struct page *page, |
9974090b MX |
3389 | get_extent_t *get_extent, |
3390 | struct bio **bio, int mirror_num, | |
f1c77c55 DS |
3391 | unsigned long *bio_flags, |
3392 | unsigned int read_flags) | |
9974090b | 3393 | { |
23d31bd4 | 3394 | struct btrfs_inode *inode = BTRFS_I(page->mapping->host); |
9974090b | 3395 | u64 start = page_offset(page); |
09cbfeaf | 3396 | u64 end = start + PAGE_SIZE - 1; |
9974090b MX |
3397 | int ret; |
3398 | ||
b272ae22 | 3399 | btrfs_lock_and_flush_ordered_range(inode, start, end, NULL); |
9974090b | 3400 | |
f657a31c | 3401 | ret = __do_readpage(page, get_extent, NULL, bio, mirror_num, |
1f7ad75b | 3402 | bio_flags, read_flags, NULL); |
9974090b MX |
3403 | return ret; |
3404 | } | |
3405 | ||
71ad38b4 DS |
3406 | int extent_read_full_page(struct page *page, get_extent_t *get_extent, |
3407 | int mirror_num) | |
d1310b2e CM |
3408 | { |
3409 | struct bio *bio = NULL; | |
c8b97818 | 3410 | unsigned long bio_flags = 0; |
d1310b2e CM |
3411 | int ret; |
3412 | ||
0d44fea7 | 3413 | ret = __extent_read_full_page(page, get_extent, &bio, mirror_num, |
1f7ad75b | 3414 | &bio_flags, 0); |
d1310b2e | 3415 | if (bio) |
1f7ad75b | 3416 | ret = submit_one_bio(bio, mirror_num, bio_flags); |
d1310b2e CM |
3417 | return ret; |
3418 | } | |
d1310b2e | 3419 | |
3d4b9496 | 3420 | static void update_nr_written(struct writeback_control *wbc, |
a9132667 | 3421 | unsigned long nr_written) |
11c8349b CM |
3422 | { |
3423 | wbc->nr_to_write -= nr_written; | |
11c8349b CM |
3424 | } |
3425 | ||
d1310b2e | 3426 | /* |
40f76580 CM |
3427 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
3428 | * | |
5eaad97a | 3429 | * This returns 1 if btrfs_run_delalloc_range function did all the work required |
40f76580 CM |
3430 | * to write the page (copy into inline extent). In this case the IO has |
3431 | * been started and the page is already unlocked. | |
3432 | * | |
3433 | * This returns 0 if all went well (page still locked) | |
3434 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 3435 | */ |
40f76580 | 3436 | static noinline_for_stack int writepage_delalloc(struct inode *inode, |
8cc0237a NB |
3437 | struct page *page, struct writeback_control *wbc, |
3438 | u64 delalloc_start, unsigned long *nr_written) | |
40f76580 | 3439 | { |
09cbfeaf | 3440 | u64 page_end = delalloc_start + PAGE_SIZE - 1; |
3522e903 | 3441 | bool found; |
40f76580 CM |
3442 | u64 delalloc_to_write = 0; |
3443 | u64 delalloc_end = 0; | |
3444 | int ret; | |
3445 | int page_started = 0; | |
3446 | ||
40f76580 CM |
3447 | |
3448 | while (delalloc_end < page_end) { | |
9978059b | 3449 | found = find_lock_delalloc_range(inode, page, |
40f76580 | 3450 | &delalloc_start, |
917aacec | 3451 | &delalloc_end); |
3522e903 | 3452 | if (!found) { |
40f76580 CM |
3453 | delalloc_start = delalloc_end + 1; |
3454 | continue; | |
3455 | } | |
5eaad97a NB |
3456 | ret = btrfs_run_delalloc_range(inode, page, delalloc_start, |
3457 | delalloc_end, &page_started, nr_written, wbc); | |
40f76580 CM |
3458 | if (ret) { |
3459 | SetPageError(page); | |
5eaad97a NB |
3460 | /* |
3461 | * btrfs_run_delalloc_range should return < 0 for error | |
3462 | * but just in case, we use > 0 here meaning the IO is | |
3463 | * started, so we don't want to return > 0 unless | |
3464 | * things are going well. | |
40f76580 CM |
3465 | */ |
3466 | ret = ret < 0 ? ret : -EIO; | |
3467 | goto done; | |
3468 | } | |
3469 | /* | |
ea1754a0 KS |
3470 | * delalloc_end is already one less than the total length, so |
3471 | * we don't subtract one from PAGE_SIZE | |
40f76580 CM |
3472 | */ |
3473 | delalloc_to_write += (delalloc_end - delalloc_start + | |
ea1754a0 | 3474 | PAGE_SIZE) >> PAGE_SHIFT; |
40f76580 CM |
3475 | delalloc_start = delalloc_end + 1; |
3476 | } | |
3477 | if (wbc->nr_to_write < delalloc_to_write) { | |
3478 | int thresh = 8192; | |
3479 | ||
3480 | if (delalloc_to_write < thresh * 2) | |
3481 | thresh = delalloc_to_write; | |
3482 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3483 | thresh); | |
3484 | } | |
3485 | ||
3486 | /* did the fill delalloc function already unlock and start | |
3487 | * the IO? | |
3488 | */ | |
3489 | if (page_started) { | |
3490 | /* | |
3491 | * we've unlocked the page, so we can't update | |
3492 | * the mapping's writeback index, just update | |
3493 | * nr_to_write. | |
3494 | */ | |
3495 | wbc->nr_to_write -= *nr_written; | |
3496 | return 1; | |
3497 | } | |
3498 | ||
3499 | ret = 0; | |
3500 | ||
3501 | done: | |
3502 | return ret; | |
3503 | } | |
3504 | ||
3505 | /* | |
3506 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3507 | * and does the loop to map the page into extents and bios. | |
3508 | * | |
3509 | * We return 1 if the IO is started and the page is unlocked, | |
3510 | * 0 if all went well (page still locked) | |
3511 | * < 0 if there were errors (page still locked) | |
3512 | */ | |
3513 | static noinline_for_stack int __extent_writepage_io(struct inode *inode, | |
3514 | struct page *page, | |
3515 | struct writeback_control *wbc, | |
3516 | struct extent_page_data *epd, | |
3517 | loff_t i_size, | |
3518 | unsigned long nr_written, | |
57e5ffeb | 3519 | int *nr_ret) |
d1310b2e | 3520 | { |
45b08405 | 3521 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
4eee4fa4 | 3522 | u64 start = page_offset(page); |
09cbfeaf | 3523 | u64 page_end = start + PAGE_SIZE - 1; |
d1310b2e CM |
3524 | u64 end; |
3525 | u64 cur = start; | |
3526 | u64 extent_offset; | |
d1310b2e CM |
3527 | u64 block_start; |
3528 | u64 iosize; | |
d1310b2e | 3529 | struct extent_map *em; |
7f3c74fb | 3530 | size_t pg_offset = 0; |
d1310b2e | 3531 | size_t blocksize; |
40f76580 CM |
3532 | int ret = 0; |
3533 | int nr = 0; | |
57e5ffeb | 3534 | const unsigned int write_flags = wbc_to_write_flags(wbc); |
40f76580 | 3535 | bool compressed; |
c8b97818 | 3536 | |
d75855b4 NB |
3537 | ret = btrfs_writepage_cow_fixup(page, start, page_end); |
3538 | if (ret) { | |
3539 | /* Fixup worker will requeue */ | |
5ab58055 | 3540 | redirty_page_for_writepage(wbc, page); |
d75855b4 NB |
3541 | update_nr_written(wbc, nr_written); |
3542 | unlock_page(page); | |
3543 | return 1; | |
247e743c CM |
3544 | } |
3545 | ||
11c8349b CM |
3546 | /* |
3547 | * we don't want to touch the inode after unlocking the page, | |
3548 | * so we update the mapping writeback index now | |
3549 | */ | |
3d4b9496 | 3550 | update_nr_written(wbc, nr_written + 1); |
771ed689 | 3551 | |
d1310b2e | 3552 | end = page_end; |
d1310b2e CM |
3553 | blocksize = inode->i_sb->s_blocksize; |
3554 | ||
3555 | while (cur <= end) { | |
40f76580 | 3556 | u64 em_end; |
6273b7f8 | 3557 | u64 offset; |
58409edd | 3558 | |
40f76580 | 3559 | if (cur >= i_size) { |
7087a9d8 | 3560 | btrfs_writepage_endio_finish_ordered(page, cur, |
c629732d | 3561 | page_end, 1); |
d1310b2e CM |
3562 | break; |
3563 | } | |
39b07b5d OS |
3564 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur, |
3565 | end - cur + 1); | |
c704005d | 3566 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3567 | SetPageError(page); |
61391d56 | 3568 | ret = PTR_ERR_OR_ZERO(em); |
d1310b2e CM |
3569 | break; |
3570 | } | |
3571 | ||
3572 | extent_offset = cur - em->start; | |
40f76580 CM |
3573 | em_end = extent_map_end(em); |
3574 | BUG_ON(em_end <= cur); | |
d1310b2e | 3575 | BUG_ON(end < cur); |
40f76580 | 3576 | iosize = min(em_end - cur, end - cur + 1); |
fda2832f | 3577 | iosize = ALIGN(iosize, blocksize); |
6273b7f8 | 3578 | offset = em->block_start + extent_offset; |
d1310b2e | 3579 | block_start = em->block_start; |
c8b97818 | 3580 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
3581 | free_extent_map(em); |
3582 | em = NULL; | |
3583 | ||
c8b97818 CM |
3584 | /* |
3585 | * compressed and inline extents are written through other | |
3586 | * paths in the FS | |
3587 | */ | |
3588 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3589 | block_start == EXTENT_MAP_INLINE) { |
c8b04030 | 3590 | if (compressed) |
c8b97818 | 3591 | nr++; |
c8b04030 OS |
3592 | else |
3593 | btrfs_writepage_endio_finish_ordered(page, cur, | |
3594 | cur + iosize - 1, 1); | |
c8b97818 | 3595 | cur += iosize; |
7f3c74fb | 3596 | pg_offset += iosize; |
d1310b2e CM |
3597 | continue; |
3598 | } | |
c8b97818 | 3599 | |
5cdc84bf | 3600 | btrfs_set_range_writeback(tree, cur, cur + iosize - 1); |
58409edd DS |
3601 | if (!PageWriteback(page)) { |
3602 | btrfs_err(BTRFS_I(inode)->root->fs_info, | |
3603 | "page %lu not writeback, cur %llu end %llu", | |
3604 | page->index, cur, end); | |
d1310b2e | 3605 | } |
7f3c74fb | 3606 | |
0ceb34bf | 3607 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc, |
6273b7f8 | 3608 | page, offset, iosize, pg_offset, |
fa17ed06 | 3609 | &epd->bio, |
58409edd DS |
3610 | end_bio_extent_writepage, |
3611 | 0, 0, 0, false); | |
fe01aa65 | 3612 | if (ret) { |
58409edd | 3613 | SetPageError(page); |
fe01aa65 TK |
3614 | if (PageWriteback(page)) |
3615 | end_page_writeback(page); | |
3616 | } | |
d1310b2e | 3617 | |
d1310b2e | 3618 | cur = cur + iosize; |
7f3c74fb | 3619 | pg_offset += iosize; |
d1310b2e CM |
3620 | nr++; |
3621 | } | |
40f76580 | 3622 | *nr_ret = nr; |
40f76580 CM |
3623 | return ret; |
3624 | } | |
3625 | ||
3626 | /* | |
3627 | * the writepage semantics are similar to regular writepage. extent | |
3628 | * records are inserted to lock ranges in the tree, and as dirty areas | |
3629 | * are found, they are marked writeback. Then the lock bits are removed | |
3630 | * and the end_io handler clears the writeback ranges | |
3065976b QW |
3631 | * |
3632 | * Return 0 if everything goes well. | |
3633 | * Return <0 for error. | |
40f76580 CM |
3634 | */ |
3635 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
aab6e9ed | 3636 | struct extent_page_data *epd) |
40f76580 CM |
3637 | { |
3638 | struct inode *inode = page->mapping->host; | |
40f76580 | 3639 | u64 start = page_offset(page); |
09cbfeaf | 3640 | u64 page_end = start + PAGE_SIZE - 1; |
40f76580 CM |
3641 | int ret; |
3642 | int nr = 0; | |
eb70d222 | 3643 | size_t pg_offset; |
40f76580 | 3644 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 3645 | unsigned long end_index = i_size >> PAGE_SHIFT; |
40f76580 CM |
3646 | unsigned long nr_written = 0; |
3647 | ||
40f76580 CM |
3648 | trace___extent_writepage(page, inode, wbc); |
3649 | ||
3650 | WARN_ON(!PageLocked(page)); | |
3651 | ||
3652 | ClearPageError(page); | |
3653 | ||
7073017a | 3654 | pg_offset = offset_in_page(i_size); |
40f76580 CM |
3655 | if (page->index > end_index || |
3656 | (page->index == end_index && !pg_offset)) { | |
09cbfeaf | 3657 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); |
40f76580 CM |
3658 | unlock_page(page); |
3659 | return 0; | |
3660 | } | |
3661 | ||
3662 | if (page->index == end_index) { | |
3663 | char *userpage; | |
3664 | ||
3665 | userpage = kmap_atomic(page); | |
3666 | memset(userpage + pg_offset, 0, | |
09cbfeaf | 3667 | PAGE_SIZE - pg_offset); |
40f76580 CM |
3668 | kunmap_atomic(userpage); |
3669 | flush_dcache_page(page); | |
3670 | } | |
3671 | ||
40f76580 CM |
3672 | set_page_extent_mapped(page); |
3673 | ||
7789a55a | 3674 | if (!epd->extent_locked) { |
8cc0237a | 3675 | ret = writepage_delalloc(inode, page, wbc, start, &nr_written); |
7789a55a | 3676 | if (ret == 1) |
169d2c87 | 3677 | return 0; |
7789a55a NB |
3678 | if (ret) |
3679 | goto done; | |
3680 | } | |
40f76580 CM |
3681 | |
3682 | ret = __extent_writepage_io(inode, page, wbc, epd, | |
57e5ffeb | 3683 | i_size, nr_written, &nr); |
40f76580 | 3684 | if (ret == 1) |
169d2c87 | 3685 | return 0; |
40f76580 | 3686 | |
d1310b2e CM |
3687 | done: |
3688 | if (nr == 0) { | |
3689 | /* make sure the mapping tag for page dirty gets cleared */ | |
3690 | set_page_writeback(page); | |
3691 | end_page_writeback(page); | |
3692 | } | |
61391d56 FM |
3693 | if (PageError(page)) { |
3694 | ret = ret < 0 ? ret : -EIO; | |
3695 | end_extent_writepage(page, ret, start, page_end); | |
3696 | } | |
d1310b2e | 3697 | unlock_page(page); |
3065976b | 3698 | ASSERT(ret <= 0); |
40f76580 | 3699 | return ret; |
d1310b2e CM |
3700 | } |
3701 | ||
fd8b2b61 | 3702 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 3703 | { |
74316201 N |
3704 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
3705 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
3706 | } |
3707 | ||
18dfa711 FM |
3708 | static void end_extent_buffer_writeback(struct extent_buffer *eb) |
3709 | { | |
3710 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
3711 | smp_mb__after_atomic(); | |
3712 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); | |
3713 | } | |
3714 | ||
2e3c2513 QW |
3715 | /* |
3716 | * Lock eb pages and flush the bio if we can't the locks | |
3717 | * | |
3718 | * Return 0 if nothing went wrong | |
3719 | * Return >0 is same as 0, except bio is not submitted | |
3720 | * Return <0 if something went wrong, no page is locked | |
3721 | */ | |
9df76fb5 | 3722 | static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb, |
0e378df1 | 3723 | struct extent_page_data *epd) |
0b32f4bb | 3724 | { |
9df76fb5 | 3725 | struct btrfs_fs_info *fs_info = eb->fs_info; |
2e3c2513 | 3726 | int i, num_pages, failed_page_nr; |
0b32f4bb JB |
3727 | int flush = 0; |
3728 | int ret = 0; | |
3729 | ||
3730 | if (!btrfs_try_tree_write_lock(eb)) { | |
f4340622 | 3731 | ret = flush_write_bio(epd); |
2e3c2513 QW |
3732 | if (ret < 0) |
3733 | return ret; | |
3734 | flush = 1; | |
0b32f4bb JB |
3735 | btrfs_tree_lock(eb); |
3736 | } | |
3737 | ||
3738 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3739 | btrfs_tree_unlock(eb); | |
3740 | if (!epd->sync_io) | |
3741 | return 0; | |
3742 | if (!flush) { | |
f4340622 | 3743 | ret = flush_write_bio(epd); |
2e3c2513 QW |
3744 | if (ret < 0) |
3745 | return ret; | |
0b32f4bb JB |
3746 | flush = 1; |
3747 | } | |
a098d8e8 CM |
3748 | while (1) { |
3749 | wait_on_extent_buffer_writeback(eb); | |
3750 | btrfs_tree_lock(eb); | |
3751 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3752 | break; | |
0b32f4bb | 3753 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3754 | } |
3755 | } | |
3756 | ||
51561ffe JB |
3757 | /* |
3758 | * We need to do this to prevent races in people who check if the eb is | |
3759 | * under IO since we can end up having no IO bits set for a short period | |
3760 | * of time. | |
3761 | */ | |
3762 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
3763 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3764 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 3765 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 3766 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
104b4e51 NB |
3767 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
3768 | -eb->len, | |
3769 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 3770 | ret = 1; |
51561ffe JB |
3771 | } else { |
3772 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
3773 | } |
3774 | ||
3775 | btrfs_tree_unlock(eb); | |
3776 | ||
3777 | if (!ret) | |
3778 | return ret; | |
3779 | ||
65ad0104 | 3780 | num_pages = num_extent_pages(eb); |
0b32f4bb | 3781 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3782 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3783 | |
3784 | if (!trylock_page(p)) { | |
3785 | if (!flush) { | |
18dfa711 FM |
3786 | int err; |
3787 | ||
3788 | err = flush_write_bio(epd); | |
3789 | if (err < 0) { | |
3790 | ret = err; | |
2e3c2513 QW |
3791 | failed_page_nr = i; |
3792 | goto err_unlock; | |
3793 | } | |
0b32f4bb JB |
3794 | flush = 1; |
3795 | } | |
3796 | lock_page(p); | |
3797 | } | |
3798 | } | |
3799 | ||
3800 | return ret; | |
2e3c2513 QW |
3801 | err_unlock: |
3802 | /* Unlock already locked pages */ | |
3803 | for (i = 0; i < failed_page_nr; i++) | |
3804 | unlock_page(eb->pages[i]); | |
18dfa711 FM |
3805 | /* |
3806 | * Clear EXTENT_BUFFER_WRITEBACK and wake up anyone waiting on it. | |
3807 | * Also set back EXTENT_BUFFER_DIRTY so future attempts to this eb can | |
3808 | * be made and undo everything done before. | |
3809 | */ | |
3810 | btrfs_tree_lock(eb); | |
3811 | spin_lock(&eb->refs_lock); | |
3812 | set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); | |
3813 | end_extent_buffer_writeback(eb); | |
3814 | spin_unlock(&eb->refs_lock); | |
3815 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, eb->len, | |
3816 | fs_info->dirty_metadata_batch); | |
3817 | btrfs_clear_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); | |
3818 | btrfs_tree_unlock(eb); | |
2e3c2513 | 3819 | return ret; |
0b32f4bb JB |
3820 | } |
3821 | ||
656f30db FM |
3822 | static void set_btree_ioerr(struct page *page) |
3823 | { | |
3824 | struct extent_buffer *eb = (struct extent_buffer *)page->private; | |
eb5b64f1 | 3825 | struct btrfs_fs_info *fs_info; |
656f30db FM |
3826 | |
3827 | SetPageError(page); | |
3828 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) | |
3829 | return; | |
3830 | ||
eb5b64f1 DZ |
3831 | /* |
3832 | * If we error out, we should add back the dirty_metadata_bytes | |
3833 | * to make it consistent. | |
3834 | */ | |
3835 | fs_info = eb->fs_info; | |
3836 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, | |
3837 | eb->len, fs_info->dirty_metadata_batch); | |
3838 | ||
656f30db FM |
3839 | /* |
3840 | * If writeback for a btree extent that doesn't belong to a log tree | |
3841 | * failed, increment the counter transaction->eb_write_errors. | |
3842 | * We do this because while the transaction is running and before it's | |
3843 | * committing (when we call filemap_fdata[write|wait]_range against | |
3844 | * the btree inode), we might have | |
3845 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
3846 | * returns an error or an error happens during writeback, when we're | |
3847 | * committing the transaction we wouldn't know about it, since the pages | |
3848 | * can be no longer dirty nor marked anymore for writeback (if a | |
3849 | * subsequent modification to the extent buffer didn't happen before the | |
3850 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
3851 | * able to find the pages tagged with SetPageError at transaction | |
3852 | * commit time. So if this happens we must abort the transaction, | |
3853 | * otherwise we commit a super block with btree roots that point to | |
3854 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
3855 | * or the content of some node/leaf from a past generation that got | |
3856 | * cowed or deleted and is no longer valid. | |
3857 | * | |
3858 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
3859 | * not be enough - we need to distinguish between log tree extents vs | |
3860 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
3861 | * will catch and clear such errors in the mapping - and that call might | |
3862 | * be from a log sync and not from a transaction commit. Also, checking | |
3863 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
3864 | * not done and would not be reliable - the eb might have been released | |
3865 | * from memory and reading it back again means that flag would not be | |
3866 | * set (since it's a runtime flag, not persisted on disk). | |
3867 | * | |
3868 | * Using the flags below in the btree inode also makes us achieve the | |
3869 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
3870 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
3871 | * is called, the writeback for all dirty pages had already finished | |
3872 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
3873 | * filemap_fdatawait_range() would return success, as it could not know | |
3874 | * that writeback errors happened (the pages were no longer tagged for | |
3875 | * writeback). | |
3876 | */ | |
3877 | switch (eb->log_index) { | |
3878 | case -1: | |
afcdd129 | 3879 | set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags); |
656f30db FM |
3880 | break; |
3881 | case 0: | |
afcdd129 | 3882 | set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags); |
656f30db FM |
3883 | break; |
3884 | case 1: | |
afcdd129 | 3885 | set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags); |
656f30db FM |
3886 | break; |
3887 | default: | |
3888 | BUG(); /* unexpected, logic error */ | |
3889 | } | |
3890 | } | |
3891 | ||
4246a0b6 | 3892 | static void end_bio_extent_buffer_writepage(struct bio *bio) |
0b32f4bb | 3893 | { |
2c30c71b | 3894 | struct bio_vec *bvec; |
0b32f4bb | 3895 | struct extent_buffer *eb; |
2b070cfe | 3896 | int done; |
6dc4f100 | 3897 | struct bvec_iter_all iter_all; |
0b32f4bb | 3898 | |
c09abff8 | 3899 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 3900 | bio_for_each_segment_all(bvec, bio, iter_all) { |
0b32f4bb JB |
3901 | struct page *page = bvec->bv_page; |
3902 | ||
0b32f4bb JB |
3903 | eb = (struct extent_buffer *)page->private; |
3904 | BUG_ON(!eb); | |
3905 | done = atomic_dec_and_test(&eb->io_pages); | |
3906 | ||
4e4cbee9 | 3907 | if (bio->bi_status || |
4246a0b6 | 3908 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { |
0b32f4bb | 3909 | ClearPageUptodate(page); |
656f30db | 3910 | set_btree_ioerr(page); |
0b32f4bb JB |
3911 | } |
3912 | ||
3913 | end_page_writeback(page); | |
3914 | ||
3915 | if (!done) | |
3916 | continue; | |
3917 | ||
3918 | end_extent_buffer_writeback(eb); | |
2c30c71b | 3919 | } |
0b32f4bb JB |
3920 | |
3921 | bio_put(bio); | |
0b32f4bb JB |
3922 | } |
3923 | ||
0e378df1 | 3924 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb JB |
3925 | struct writeback_control *wbc, |
3926 | struct extent_page_data *epd) | |
3927 | { | |
0b32f4bb | 3928 | u64 offset = eb->start; |
851cd173 | 3929 | u32 nritems; |
cc5e31a4 | 3930 | int i, num_pages; |
851cd173 | 3931 | unsigned long start, end; |
ff40adf7 | 3932 | unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META; |
d7dbe9e7 | 3933 | int ret = 0; |
0b32f4bb | 3934 | |
656f30db | 3935 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); |
65ad0104 | 3936 | num_pages = num_extent_pages(eb); |
0b32f4bb | 3937 | atomic_set(&eb->io_pages, num_pages); |
de0022b9 | 3938 | |
851cd173 LB |
3939 | /* set btree blocks beyond nritems with 0 to avoid stale content. */ |
3940 | nritems = btrfs_header_nritems(eb); | |
3eb548ee | 3941 | if (btrfs_header_level(eb) > 0) { |
3eb548ee LB |
3942 | end = btrfs_node_key_ptr_offset(nritems); |
3943 | ||
b159fa28 | 3944 | memzero_extent_buffer(eb, end, eb->len - end); |
851cd173 LB |
3945 | } else { |
3946 | /* | |
3947 | * leaf: | |
3948 | * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0 | |
3949 | */ | |
3950 | start = btrfs_item_nr_offset(nritems); | |
8f881e8c | 3951 | end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(eb); |
b159fa28 | 3952 | memzero_extent_buffer(eb, start, end - start); |
3eb548ee LB |
3953 | } |
3954 | ||
0b32f4bb | 3955 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3956 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3957 | |
3958 | clear_page_dirty_for_io(p); | |
3959 | set_page_writeback(p); | |
0ceb34bf | 3960 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc, |
fa17ed06 | 3961 | p, offset, PAGE_SIZE, 0, |
c2df8bb4 | 3962 | &epd->bio, |
1f7ad75b | 3963 | end_bio_extent_buffer_writepage, |
18fdc679 | 3964 | 0, 0, 0, false); |
0b32f4bb | 3965 | if (ret) { |
656f30db | 3966 | set_btree_ioerr(p); |
fe01aa65 TK |
3967 | if (PageWriteback(p)) |
3968 | end_page_writeback(p); | |
0b32f4bb JB |
3969 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
3970 | end_extent_buffer_writeback(eb); | |
3971 | ret = -EIO; | |
3972 | break; | |
3973 | } | |
09cbfeaf | 3974 | offset += PAGE_SIZE; |
3d4b9496 | 3975 | update_nr_written(wbc, 1); |
0b32f4bb JB |
3976 | unlock_page(p); |
3977 | } | |
3978 | ||
3979 | if (unlikely(ret)) { | |
3980 | for (; i < num_pages; i++) { | |
bbf65cf0 | 3981 | struct page *p = eb->pages[i]; |
81465028 | 3982 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
3983 | unlock_page(p); |
3984 | } | |
3985 | } | |
3986 | ||
3987 | return ret; | |
3988 | } | |
3989 | ||
3990 | int btree_write_cache_pages(struct address_space *mapping, | |
3991 | struct writeback_control *wbc) | |
3992 | { | |
0b32f4bb JB |
3993 | struct extent_buffer *eb, *prev_eb = NULL; |
3994 | struct extent_page_data epd = { | |
3995 | .bio = NULL, | |
0b32f4bb JB |
3996 | .extent_locked = 0, |
3997 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
3998 | }; | |
b3ff8f1d | 3999 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; |
0b32f4bb JB |
4000 | int ret = 0; |
4001 | int done = 0; | |
4002 | int nr_to_write_done = 0; | |
4003 | struct pagevec pvec; | |
4004 | int nr_pages; | |
4005 | pgoff_t index; | |
4006 | pgoff_t end; /* Inclusive */ | |
4007 | int scanned = 0; | |
10bbd235 | 4008 | xa_mark_t tag; |
0b32f4bb | 4009 | |
86679820 | 4010 | pagevec_init(&pvec); |
0b32f4bb JB |
4011 | if (wbc->range_cyclic) { |
4012 | index = mapping->writeback_index; /* Start from prev offset */ | |
4013 | end = -1; | |
556755a8 JB |
4014 | /* |
4015 | * Start from the beginning does not need to cycle over the | |
4016 | * range, mark it as scanned. | |
4017 | */ | |
4018 | scanned = (index == 0); | |
0b32f4bb | 4019 | } else { |
09cbfeaf KS |
4020 | index = wbc->range_start >> PAGE_SHIFT; |
4021 | end = wbc->range_end >> PAGE_SHIFT; | |
0b32f4bb JB |
4022 | scanned = 1; |
4023 | } | |
4024 | if (wbc->sync_mode == WB_SYNC_ALL) | |
4025 | tag = PAGECACHE_TAG_TOWRITE; | |
4026 | else | |
4027 | tag = PAGECACHE_TAG_DIRTY; | |
4028 | retry: | |
4029 | if (wbc->sync_mode == WB_SYNC_ALL) | |
4030 | tag_pages_for_writeback(mapping, index, end); | |
4031 | while (!done && !nr_to_write_done && (index <= end) && | |
4006f437 | 4032 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 4033 | tag))) { |
0b32f4bb JB |
4034 | unsigned i; |
4035 | ||
0b32f4bb JB |
4036 | for (i = 0; i < nr_pages; i++) { |
4037 | struct page *page = pvec.pages[i]; | |
4038 | ||
4039 | if (!PagePrivate(page)) | |
4040 | continue; | |
4041 | ||
b5bae261 JB |
4042 | spin_lock(&mapping->private_lock); |
4043 | if (!PagePrivate(page)) { | |
4044 | spin_unlock(&mapping->private_lock); | |
4045 | continue; | |
4046 | } | |
4047 | ||
0b32f4bb | 4048 | eb = (struct extent_buffer *)page->private; |
b5bae261 JB |
4049 | |
4050 | /* | |
4051 | * Shouldn't happen and normally this would be a BUG_ON | |
4052 | * but no sense in crashing the users box for something | |
4053 | * we can survive anyway. | |
4054 | */ | |
fae7f21c | 4055 | if (WARN_ON(!eb)) { |
b5bae261 | 4056 | spin_unlock(&mapping->private_lock); |
0b32f4bb JB |
4057 | continue; |
4058 | } | |
4059 | ||
b5bae261 JB |
4060 | if (eb == prev_eb) { |
4061 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 4062 | continue; |
b5bae261 | 4063 | } |
0b32f4bb | 4064 | |
b5bae261 JB |
4065 | ret = atomic_inc_not_zero(&eb->refs); |
4066 | spin_unlock(&mapping->private_lock); | |
4067 | if (!ret) | |
0b32f4bb | 4068 | continue; |
0b32f4bb JB |
4069 | |
4070 | prev_eb = eb; | |
9df76fb5 | 4071 | ret = lock_extent_buffer_for_io(eb, &epd); |
0b32f4bb JB |
4072 | if (!ret) { |
4073 | free_extent_buffer(eb); | |
4074 | continue; | |
0607eb1d FM |
4075 | } else if (ret < 0) { |
4076 | done = 1; | |
4077 | free_extent_buffer(eb); | |
4078 | break; | |
0b32f4bb JB |
4079 | } |
4080 | ||
0ab02063 | 4081 | ret = write_one_eb(eb, wbc, &epd); |
0b32f4bb JB |
4082 | if (ret) { |
4083 | done = 1; | |
4084 | free_extent_buffer(eb); | |
4085 | break; | |
4086 | } | |
4087 | free_extent_buffer(eb); | |
4088 | ||
4089 | /* | |
4090 | * the filesystem may choose to bump up nr_to_write. | |
4091 | * We have to make sure to honor the new nr_to_write | |
4092 | * at any time | |
4093 | */ | |
4094 | nr_to_write_done = wbc->nr_to_write <= 0; | |
4095 | } | |
4096 | pagevec_release(&pvec); | |
4097 | cond_resched(); | |
4098 | } | |
4099 | if (!scanned && !done) { | |
4100 | /* | |
4101 | * We hit the last page and there is more work to be done: wrap | |
4102 | * back to the start of the file | |
4103 | */ | |
4104 | scanned = 1; | |
4105 | index = 0; | |
4106 | goto retry; | |
4107 | } | |
2b952eea QW |
4108 | ASSERT(ret <= 0); |
4109 | if (ret < 0) { | |
4110 | end_write_bio(&epd, ret); | |
4111 | return ret; | |
4112 | } | |
b3ff8f1d QW |
4113 | /* |
4114 | * If something went wrong, don't allow any metadata write bio to be | |
4115 | * submitted. | |
4116 | * | |
4117 | * This would prevent use-after-free if we had dirty pages not | |
4118 | * cleaned up, which can still happen by fuzzed images. | |
4119 | * | |
4120 | * - Bad extent tree | |
4121 | * Allowing existing tree block to be allocated for other trees. | |
4122 | * | |
4123 | * - Log tree operations | |
4124 | * Exiting tree blocks get allocated to log tree, bumps its | |
4125 | * generation, then get cleaned in tree re-balance. | |
4126 | * Such tree block will not be written back, since it's clean, | |
4127 | * thus no WRITTEN flag set. | |
4128 | * And after log writes back, this tree block is not traced by | |
4129 | * any dirty extent_io_tree. | |
4130 | * | |
4131 | * - Offending tree block gets re-dirtied from its original owner | |
4132 | * Since it has bumped generation, no WRITTEN flag, it can be | |
4133 | * reused without COWing. This tree block will not be traced | |
4134 | * by btrfs_transaction::dirty_pages. | |
4135 | * | |
4136 | * Now such dirty tree block will not be cleaned by any dirty | |
4137 | * extent io tree. Thus we don't want to submit such wild eb | |
4138 | * if the fs already has error. | |
4139 | */ | |
4140 | if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { | |
4141 | ret = flush_write_bio(&epd); | |
4142 | } else { | |
4143 | ret = -EUCLEAN; | |
4144 | end_write_bio(&epd, ret); | |
4145 | } | |
0b32f4bb JB |
4146 | return ret; |
4147 | } | |
4148 | ||
d1310b2e | 4149 | /** |
4bef0848 | 4150 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
4151 | * @mapping: address space structure to write |
4152 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
935db853 | 4153 | * @data: data passed to __extent_writepage function |
d1310b2e CM |
4154 | * |
4155 | * If a page is already under I/O, write_cache_pages() skips it, even | |
4156 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
4157 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
4158 | * and msync() need to guarantee that all the data which was dirty at the time | |
4159 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
4160 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
4161 | * existing IO to complete. | |
4162 | */ | |
4242b64a | 4163 | static int extent_write_cache_pages(struct address_space *mapping, |
4bef0848 | 4164 | struct writeback_control *wbc, |
aab6e9ed | 4165 | struct extent_page_data *epd) |
d1310b2e | 4166 | { |
7fd1a3f7 | 4167 | struct inode *inode = mapping->host; |
d1310b2e CM |
4168 | int ret = 0; |
4169 | int done = 0; | |
f85d7d6c | 4170 | int nr_to_write_done = 0; |
d1310b2e CM |
4171 | struct pagevec pvec; |
4172 | int nr_pages; | |
4173 | pgoff_t index; | |
4174 | pgoff_t end; /* Inclusive */ | |
a9132667 LB |
4175 | pgoff_t done_index; |
4176 | int range_whole = 0; | |
d1310b2e | 4177 | int scanned = 0; |
10bbd235 | 4178 | xa_mark_t tag; |
d1310b2e | 4179 | |
7fd1a3f7 JB |
4180 | /* |
4181 | * We have to hold onto the inode so that ordered extents can do their | |
4182 | * work when the IO finishes. The alternative to this is failing to add | |
4183 | * an ordered extent if the igrab() fails there and that is a huge pain | |
4184 | * to deal with, so instead just hold onto the inode throughout the | |
4185 | * writepages operation. If it fails here we are freeing up the inode | |
4186 | * anyway and we'd rather not waste our time writing out stuff that is | |
4187 | * going to be truncated anyway. | |
4188 | */ | |
4189 | if (!igrab(inode)) | |
4190 | return 0; | |
4191 | ||
86679820 | 4192 | pagevec_init(&pvec); |
d1310b2e CM |
4193 | if (wbc->range_cyclic) { |
4194 | index = mapping->writeback_index; /* Start from prev offset */ | |
4195 | end = -1; | |
556755a8 JB |
4196 | /* |
4197 | * Start from the beginning does not need to cycle over the | |
4198 | * range, mark it as scanned. | |
4199 | */ | |
4200 | scanned = (index == 0); | |
d1310b2e | 4201 | } else { |
09cbfeaf KS |
4202 | index = wbc->range_start >> PAGE_SHIFT; |
4203 | end = wbc->range_end >> PAGE_SHIFT; | |
a9132667 LB |
4204 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
4205 | range_whole = 1; | |
d1310b2e CM |
4206 | scanned = 1; |
4207 | } | |
3cd24c69 EL |
4208 | |
4209 | /* | |
4210 | * We do the tagged writepage as long as the snapshot flush bit is set | |
4211 | * and we are the first one who do the filemap_flush() on this inode. | |
4212 | * | |
4213 | * The nr_to_write == LONG_MAX is needed to make sure other flushers do | |
4214 | * not race in and drop the bit. | |
4215 | */ | |
4216 | if (range_whole && wbc->nr_to_write == LONG_MAX && | |
4217 | test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
4218 | &BTRFS_I(inode)->runtime_flags)) | |
4219 | wbc->tagged_writepages = 1; | |
4220 | ||
4221 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) | |
f7aaa06b JB |
4222 | tag = PAGECACHE_TAG_TOWRITE; |
4223 | else | |
4224 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 4225 | retry: |
3cd24c69 | 4226 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
f7aaa06b | 4227 | tag_pages_for_writeback(mapping, index, end); |
a9132667 | 4228 | done_index = index; |
f85d7d6c | 4229 | while (!done && !nr_to_write_done && (index <= end) && |
67fd707f JK |
4230 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, |
4231 | &index, end, tag))) { | |
d1310b2e CM |
4232 | unsigned i; |
4233 | ||
d1310b2e CM |
4234 | for (i = 0; i < nr_pages; i++) { |
4235 | struct page *page = pvec.pages[i]; | |
4236 | ||
f7bddf1e | 4237 | done_index = page->index + 1; |
d1310b2e | 4238 | /* |
b93b0163 MW |
4239 | * At this point we hold neither the i_pages lock nor |
4240 | * the page lock: the page may be truncated or | |
4241 | * invalidated (changing page->mapping to NULL), | |
4242 | * or even swizzled back from swapper_space to | |
4243 | * tmpfs file mapping | |
d1310b2e | 4244 | */ |
c8f2f24b | 4245 | if (!trylock_page(page)) { |
f4340622 QW |
4246 | ret = flush_write_bio(epd); |
4247 | BUG_ON(ret < 0); | |
c8f2f24b | 4248 | lock_page(page); |
01d658f2 | 4249 | } |
d1310b2e CM |
4250 | |
4251 | if (unlikely(page->mapping != mapping)) { | |
4252 | unlock_page(page); | |
4253 | continue; | |
4254 | } | |
4255 | ||
d2c3f4f6 | 4256 | if (wbc->sync_mode != WB_SYNC_NONE) { |
f4340622 QW |
4257 | if (PageWriteback(page)) { |
4258 | ret = flush_write_bio(epd); | |
4259 | BUG_ON(ret < 0); | |
4260 | } | |
d1310b2e | 4261 | wait_on_page_writeback(page); |
d2c3f4f6 | 4262 | } |
d1310b2e CM |
4263 | |
4264 | if (PageWriteback(page) || | |
4265 | !clear_page_dirty_for_io(page)) { | |
4266 | unlock_page(page); | |
4267 | continue; | |
4268 | } | |
4269 | ||
aab6e9ed | 4270 | ret = __extent_writepage(page, wbc, epd); |
a9132667 | 4271 | if (ret < 0) { |
a9132667 LB |
4272 | done = 1; |
4273 | break; | |
4274 | } | |
f85d7d6c CM |
4275 | |
4276 | /* | |
4277 | * the filesystem may choose to bump up nr_to_write. | |
4278 | * We have to make sure to honor the new nr_to_write | |
4279 | * at any time | |
4280 | */ | |
4281 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
4282 | } |
4283 | pagevec_release(&pvec); | |
4284 | cond_resched(); | |
4285 | } | |
894b36e3 | 4286 | if (!scanned && !done) { |
d1310b2e CM |
4287 | /* |
4288 | * We hit the last page and there is more work to be done: wrap | |
4289 | * back to the start of the file | |
4290 | */ | |
4291 | scanned = 1; | |
4292 | index = 0; | |
42ffb0bf JB |
4293 | |
4294 | /* | |
4295 | * If we're looping we could run into a page that is locked by a | |
4296 | * writer and that writer could be waiting on writeback for a | |
4297 | * page in our current bio, and thus deadlock, so flush the | |
4298 | * write bio here. | |
4299 | */ | |
4300 | ret = flush_write_bio(epd); | |
4301 | if (!ret) | |
4302 | goto retry; | |
d1310b2e | 4303 | } |
a9132667 LB |
4304 | |
4305 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole)) | |
4306 | mapping->writeback_index = done_index; | |
4307 | ||
7fd1a3f7 | 4308 | btrfs_add_delayed_iput(inode); |
894b36e3 | 4309 | return ret; |
d1310b2e | 4310 | } |
d1310b2e | 4311 | |
0a9b0e53 | 4312 | int extent_write_full_page(struct page *page, struct writeback_control *wbc) |
d1310b2e CM |
4313 | { |
4314 | int ret; | |
d1310b2e CM |
4315 | struct extent_page_data epd = { |
4316 | .bio = NULL, | |
771ed689 | 4317 | .extent_locked = 0, |
ffbd517d | 4318 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e | 4319 | }; |
d1310b2e | 4320 | |
d1310b2e | 4321 | ret = __extent_writepage(page, wbc, &epd); |
3065976b QW |
4322 | ASSERT(ret <= 0); |
4323 | if (ret < 0) { | |
4324 | end_write_bio(&epd, ret); | |
4325 | return ret; | |
4326 | } | |
d1310b2e | 4327 | |
3065976b QW |
4328 | ret = flush_write_bio(&epd); |
4329 | ASSERT(ret <= 0); | |
d1310b2e CM |
4330 | return ret; |
4331 | } | |
d1310b2e | 4332 | |
5e3ee236 | 4333 | int extent_write_locked_range(struct inode *inode, u64 start, u64 end, |
771ed689 CM |
4334 | int mode) |
4335 | { | |
4336 | int ret = 0; | |
4337 | struct address_space *mapping = inode->i_mapping; | |
4338 | struct page *page; | |
09cbfeaf KS |
4339 | unsigned long nr_pages = (end - start + PAGE_SIZE) >> |
4340 | PAGE_SHIFT; | |
771ed689 CM |
4341 | |
4342 | struct extent_page_data epd = { | |
4343 | .bio = NULL, | |
771ed689 | 4344 | .extent_locked = 1, |
ffbd517d | 4345 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
4346 | }; |
4347 | struct writeback_control wbc_writepages = { | |
771ed689 | 4348 | .sync_mode = mode, |
771ed689 CM |
4349 | .nr_to_write = nr_pages * 2, |
4350 | .range_start = start, | |
4351 | .range_end = end + 1, | |
ec39f769 CM |
4352 | /* We're called from an async helper function */ |
4353 | .punt_to_cgroup = 1, | |
4354 | .no_cgroup_owner = 1, | |
771ed689 CM |
4355 | }; |
4356 | ||
dbb70bec | 4357 | wbc_attach_fdatawrite_inode(&wbc_writepages, inode); |
d397712b | 4358 | while (start <= end) { |
09cbfeaf | 4359 | page = find_get_page(mapping, start >> PAGE_SHIFT); |
771ed689 CM |
4360 | if (clear_page_dirty_for_io(page)) |
4361 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
4362 | else { | |
7087a9d8 | 4363 | btrfs_writepage_endio_finish_ordered(page, start, |
c629732d | 4364 | start + PAGE_SIZE - 1, 1); |
771ed689 CM |
4365 | unlock_page(page); |
4366 | } | |
09cbfeaf KS |
4367 | put_page(page); |
4368 | start += PAGE_SIZE; | |
771ed689 CM |
4369 | } |
4370 | ||
02c6db4f | 4371 | ASSERT(ret <= 0); |
dbb70bec CM |
4372 | if (ret == 0) |
4373 | ret = flush_write_bio(&epd); | |
4374 | else | |
02c6db4f | 4375 | end_write_bio(&epd, ret); |
dbb70bec CM |
4376 | |
4377 | wbc_detach_inode(&wbc_writepages); | |
771ed689 CM |
4378 | return ret; |
4379 | } | |
d1310b2e | 4380 | |
8ae225a8 | 4381 | int extent_writepages(struct address_space *mapping, |
d1310b2e CM |
4382 | struct writeback_control *wbc) |
4383 | { | |
4384 | int ret = 0; | |
4385 | struct extent_page_data epd = { | |
4386 | .bio = NULL, | |
771ed689 | 4387 | .extent_locked = 0, |
ffbd517d | 4388 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
4389 | }; |
4390 | ||
935db853 | 4391 | ret = extent_write_cache_pages(mapping, wbc, &epd); |
a2a72fbd QW |
4392 | ASSERT(ret <= 0); |
4393 | if (ret < 0) { | |
4394 | end_write_bio(&epd, ret); | |
4395 | return ret; | |
4396 | } | |
4397 | ret = flush_write_bio(&epd); | |
d1310b2e CM |
4398 | return ret; |
4399 | } | |
d1310b2e | 4400 | |
2a3ff0ad NB |
4401 | int extent_readpages(struct address_space *mapping, struct list_head *pages, |
4402 | unsigned nr_pages) | |
d1310b2e CM |
4403 | { |
4404 | struct bio *bio = NULL; | |
c8b97818 | 4405 | unsigned long bio_flags = 0; |
67c9684f | 4406 | struct page *pagepool[16]; |
125bac01 | 4407 | struct extent_map *em_cached = NULL; |
67c9684f | 4408 | int nr = 0; |
808f80b4 | 4409 | u64 prev_em_start = (u64)-1; |
d1310b2e | 4410 | |
61ed3a14 | 4411 | while (!list_empty(pages)) { |
e65ef21e NB |
4412 | u64 contig_end = 0; |
4413 | ||
61ed3a14 | 4414 | for (nr = 0; nr < ARRAY_SIZE(pagepool) && !list_empty(pages);) { |
f86196ea | 4415 | struct page *page = lru_to_page(pages); |
d1310b2e | 4416 | |
61ed3a14 NB |
4417 | prefetchw(&page->flags); |
4418 | list_del(&page->lru); | |
4419 | if (add_to_page_cache_lru(page, mapping, page->index, | |
4420 | readahead_gfp_mask(mapping))) { | |
4421 | put_page(page); | |
e65ef21e | 4422 | break; |
61ed3a14 NB |
4423 | } |
4424 | ||
4425 | pagepool[nr++] = page; | |
e65ef21e | 4426 | contig_end = page_offset(page) + PAGE_SIZE - 1; |
d1310b2e | 4427 | } |
67c9684f | 4428 | |
e65ef21e NB |
4429 | if (nr) { |
4430 | u64 contig_start = page_offset(pagepool[0]); | |
4431 | ||
4432 | ASSERT(contig_start + nr * PAGE_SIZE - 1 == contig_end); | |
4433 | ||
b6660e80 | 4434 | contiguous_readpages(pagepool, nr, contig_start, |
e65ef21e NB |
4435 | contig_end, &em_cached, &bio, &bio_flags, |
4436 | &prev_em_start); | |
4437 | } | |
d1310b2e | 4438 | } |
67c9684f | 4439 | |
125bac01 MX |
4440 | if (em_cached) |
4441 | free_extent_map(em_cached); | |
4442 | ||
d1310b2e | 4443 | if (bio) |
1f7ad75b | 4444 | return submit_one_bio(bio, 0, bio_flags); |
d1310b2e CM |
4445 | return 0; |
4446 | } | |
d1310b2e CM |
4447 | |
4448 | /* | |
4449 | * basic invalidatepage code, this waits on any locked or writeback | |
4450 | * ranges corresponding to the page, and then deletes any extent state | |
4451 | * records from the tree | |
4452 | */ | |
4453 | int extent_invalidatepage(struct extent_io_tree *tree, | |
4454 | struct page *page, unsigned long offset) | |
4455 | { | |
2ac55d41 | 4456 | struct extent_state *cached_state = NULL; |
4eee4fa4 | 4457 | u64 start = page_offset(page); |
09cbfeaf | 4458 | u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
4459 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; |
4460 | ||
fda2832f | 4461 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
4462 | if (start > end) |
4463 | return 0; | |
4464 | ||
ff13db41 | 4465 | lock_extent_bits(tree, start, end, &cached_state); |
1edbb734 | 4466 | wait_on_page_writeback(page); |
e182163d OS |
4467 | clear_extent_bit(tree, start, end, EXTENT_LOCKED | EXTENT_DELALLOC | |
4468 | EXTENT_DO_ACCOUNTING, 1, 1, &cached_state); | |
d1310b2e CM |
4469 | return 0; |
4470 | } | |
d1310b2e | 4471 | |
7b13b7b1 CM |
4472 | /* |
4473 | * a helper for releasepage, this tests for areas of the page that | |
4474 | * are locked or under IO and drops the related state bits if it is safe | |
4475 | * to drop the page. | |
4476 | */ | |
29c68b2d | 4477 | static int try_release_extent_state(struct extent_io_tree *tree, |
48a3b636 | 4478 | struct page *page, gfp_t mask) |
7b13b7b1 | 4479 | { |
4eee4fa4 | 4480 | u64 start = page_offset(page); |
09cbfeaf | 4481 | u64 end = start + PAGE_SIZE - 1; |
7b13b7b1 CM |
4482 | int ret = 1; |
4483 | ||
8882679e | 4484 | if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) { |
7b13b7b1 | 4485 | ret = 0; |
8882679e | 4486 | } else { |
11ef160f CM |
4487 | /* |
4488 | * at this point we can safely clear everything except the | |
4489 | * locked bit and the nodatasum bit | |
4490 | */ | |
66b0c887 | 4491 | ret = __clear_extent_bit(tree, start, end, |
11ef160f | 4492 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
66b0c887 | 4493 | 0, 0, NULL, mask, NULL); |
e3f24cc5 CM |
4494 | |
4495 | /* if clear_extent_bit failed for enomem reasons, | |
4496 | * we can't allow the release to continue. | |
4497 | */ | |
4498 | if (ret < 0) | |
4499 | ret = 0; | |
4500 | else | |
4501 | ret = 1; | |
7b13b7b1 CM |
4502 | } |
4503 | return ret; | |
4504 | } | |
7b13b7b1 | 4505 | |
d1310b2e CM |
4506 | /* |
4507 | * a helper for releasepage. As long as there are no locked extents | |
4508 | * in the range corresponding to the page, both state records and extent | |
4509 | * map records are removed | |
4510 | */ | |
477a30ba | 4511 | int try_release_extent_mapping(struct page *page, gfp_t mask) |
d1310b2e CM |
4512 | { |
4513 | struct extent_map *em; | |
4eee4fa4 | 4514 | u64 start = page_offset(page); |
09cbfeaf | 4515 | u64 end = start + PAGE_SIZE - 1; |
bd3599a0 FM |
4516 | struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host); |
4517 | struct extent_io_tree *tree = &btrfs_inode->io_tree; | |
4518 | struct extent_map_tree *map = &btrfs_inode->extent_tree; | |
7b13b7b1 | 4519 | |
d0164adc | 4520 | if (gfpflags_allow_blocking(mask) && |
ee22184b | 4521 | page->mapping->host->i_size > SZ_16M) { |
39b5637f | 4522 | u64 len; |
70dec807 | 4523 | while (start <= end) { |
39b5637f | 4524 | len = end - start + 1; |
890871be | 4525 | write_lock(&map->lock); |
39b5637f | 4526 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 4527 | if (!em) { |
890871be | 4528 | write_unlock(&map->lock); |
70dec807 CM |
4529 | break; |
4530 | } | |
7f3c74fb CM |
4531 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
4532 | em->start != start) { | |
890871be | 4533 | write_unlock(&map->lock); |
70dec807 CM |
4534 | free_extent_map(em); |
4535 | break; | |
4536 | } | |
4537 | if (!test_range_bit(tree, em->start, | |
4538 | extent_map_end(em) - 1, | |
4e586ca3 | 4539 | EXTENT_LOCKED, 0, NULL)) { |
bd3599a0 FM |
4540 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
4541 | &btrfs_inode->runtime_flags); | |
70dec807 CM |
4542 | remove_extent_mapping(map, em); |
4543 | /* once for the rb tree */ | |
4544 | free_extent_map(em); | |
4545 | } | |
4546 | start = extent_map_end(em); | |
890871be | 4547 | write_unlock(&map->lock); |
70dec807 CM |
4548 | |
4549 | /* once for us */ | |
d1310b2e CM |
4550 | free_extent_map(em); |
4551 | } | |
d1310b2e | 4552 | } |
29c68b2d | 4553 | return try_release_extent_state(tree, page, mask); |
d1310b2e | 4554 | } |
d1310b2e | 4555 | |
ec29ed5b CM |
4556 | /* |
4557 | * helper function for fiemap, which doesn't want to see any holes. | |
4558 | * This maps until we find something past 'last' | |
4559 | */ | |
4560 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
e3350e16 | 4561 | u64 offset, u64 last) |
ec29ed5b | 4562 | { |
da17066c | 4563 | u64 sectorsize = btrfs_inode_sectorsize(inode); |
ec29ed5b CM |
4564 | struct extent_map *em; |
4565 | u64 len; | |
4566 | ||
4567 | if (offset >= last) | |
4568 | return NULL; | |
4569 | ||
67871254 | 4570 | while (1) { |
ec29ed5b CM |
4571 | len = last - offset; |
4572 | if (len == 0) | |
4573 | break; | |
fda2832f | 4574 | len = ALIGN(len, sectorsize); |
4ab47a8d | 4575 | em = btrfs_get_extent_fiemap(BTRFS_I(inode), offset, len); |
c704005d | 4576 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
4577 | return em; |
4578 | ||
4579 | /* if this isn't a hole return it */ | |
4a2d25cd | 4580 | if (em->block_start != EXTENT_MAP_HOLE) |
ec29ed5b | 4581 | return em; |
ec29ed5b CM |
4582 | |
4583 | /* this is a hole, advance to the next extent */ | |
4584 | offset = extent_map_end(em); | |
4585 | free_extent_map(em); | |
4586 | if (offset >= last) | |
4587 | break; | |
4588 | } | |
4589 | return NULL; | |
4590 | } | |
4591 | ||
4751832d QW |
4592 | /* |
4593 | * To cache previous fiemap extent | |
4594 | * | |
4595 | * Will be used for merging fiemap extent | |
4596 | */ | |
4597 | struct fiemap_cache { | |
4598 | u64 offset; | |
4599 | u64 phys; | |
4600 | u64 len; | |
4601 | u32 flags; | |
4602 | bool cached; | |
4603 | }; | |
4604 | ||
4605 | /* | |
4606 | * Helper to submit fiemap extent. | |
4607 | * | |
4608 | * Will try to merge current fiemap extent specified by @offset, @phys, | |
4609 | * @len and @flags with cached one. | |
4610 | * And only when we fails to merge, cached one will be submitted as | |
4611 | * fiemap extent. | |
4612 | * | |
4613 | * Return value is the same as fiemap_fill_next_extent(). | |
4614 | */ | |
4615 | static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, | |
4616 | struct fiemap_cache *cache, | |
4617 | u64 offset, u64 phys, u64 len, u32 flags) | |
4618 | { | |
4619 | int ret = 0; | |
4620 | ||
4621 | if (!cache->cached) | |
4622 | goto assign; | |
4623 | ||
4624 | /* | |
4625 | * Sanity check, extent_fiemap() should have ensured that new | |
52042d8e | 4626 | * fiemap extent won't overlap with cached one. |
4751832d QW |
4627 | * Not recoverable. |
4628 | * | |
4629 | * NOTE: Physical address can overlap, due to compression | |
4630 | */ | |
4631 | if (cache->offset + cache->len > offset) { | |
4632 | WARN_ON(1); | |
4633 | return -EINVAL; | |
4634 | } | |
4635 | ||
4636 | /* | |
4637 | * Only merges fiemap extents if | |
4638 | * 1) Their logical addresses are continuous | |
4639 | * | |
4640 | * 2) Their physical addresses are continuous | |
4641 | * So truly compressed (physical size smaller than logical size) | |
4642 | * extents won't get merged with each other | |
4643 | * | |
4644 | * 3) Share same flags except FIEMAP_EXTENT_LAST | |
4645 | * So regular extent won't get merged with prealloc extent | |
4646 | */ | |
4647 | if (cache->offset + cache->len == offset && | |
4648 | cache->phys + cache->len == phys && | |
4649 | (cache->flags & ~FIEMAP_EXTENT_LAST) == | |
4650 | (flags & ~FIEMAP_EXTENT_LAST)) { | |
4651 | cache->len += len; | |
4652 | cache->flags |= flags; | |
4653 | goto try_submit_last; | |
4654 | } | |
4655 | ||
4656 | /* Not mergeable, need to submit cached one */ | |
4657 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, | |
4658 | cache->len, cache->flags); | |
4659 | cache->cached = false; | |
4660 | if (ret) | |
4661 | return ret; | |
4662 | assign: | |
4663 | cache->cached = true; | |
4664 | cache->offset = offset; | |
4665 | cache->phys = phys; | |
4666 | cache->len = len; | |
4667 | cache->flags = flags; | |
4668 | try_submit_last: | |
4669 | if (cache->flags & FIEMAP_EXTENT_LAST) { | |
4670 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, | |
4671 | cache->phys, cache->len, cache->flags); | |
4672 | cache->cached = false; | |
4673 | } | |
4674 | return ret; | |
4675 | } | |
4676 | ||
4677 | /* | |
848c23b7 | 4678 | * Emit last fiemap cache |
4751832d | 4679 | * |
848c23b7 QW |
4680 | * The last fiemap cache may still be cached in the following case: |
4681 | * 0 4k 8k | |
4682 | * |<- Fiemap range ->| | |
4683 | * |<------------ First extent ----------->| | |
4684 | * | |
4685 | * In this case, the first extent range will be cached but not emitted. | |
4686 | * So we must emit it before ending extent_fiemap(). | |
4751832d | 4687 | */ |
5c5aff98 | 4688 | static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo, |
848c23b7 | 4689 | struct fiemap_cache *cache) |
4751832d QW |
4690 | { |
4691 | int ret; | |
4692 | ||
4693 | if (!cache->cached) | |
4694 | return 0; | |
4695 | ||
4751832d QW |
4696 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, |
4697 | cache->len, cache->flags); | |
4698 | cache->cached = false; | |
4699 | if (ret > 0) | |
4700 | ret = 0; | |
4701 | return ret; | |
4702 | } | |
4703 | ||
1506fcc8 | 4704 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
2135fb9b | 4705 | __u64 start, __u64 len) |
1506fcc8 | 4706 | { |
975f84fe | 4707 | int ret = 0; |
1506fcc8 YS |
4708 | u64 off = start; |
4709 | u64 max = start + len; | |
4710 | u32 flags = 0; | |
975f84fe JB |
4711 | u32 found_type; |
4712 | u64 last; | |
ec29ed5b | 4713 | u64 last_for_get_extent = 0; |
1506fcc8 | 4714 | u64 disko = 0; |
ec29ed5b | 4715 | u64 isize = i_size_read(inode); |
975f84fe | 4716 | struct btrfs_key found_key; |
1506fcc8 | 4717 | struct extent_map *em = NULL; |
2ac55d41 | 4718 | struct extent_state *cached_state = NULL; |
975f84fe | 4719 | struct btrfs_path *path; |
dc046b10 | 4720 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4751832d | 4721 | struct fiemap_cache cache = { 0 }; |
5911c8fe DS |
4722 | struct ulist *roots; |
4723 | struct ulist *tmp_ulist; | |
1506fcc8 | 4724 | int end = 0; |
ec29ed5b CM |
4725 | u64 em_start = 0; |
4726 | u64 em_len = 0; | |
4727 | u64 em_end = 0; | |
1506fcc8 YS |
4728 | |
4729 | if (len == 0) | |
4730 | return -EINVAL; | |
4731 | ||
975f84fe JB |
4732 | path = btrfs_alloc_path(); |
4733 | if (!path) | |
4734 | return -ENOMEM; | |
4735 | path->leave_spinning = 1; | |
4736 | ||
5911c8fe DS |
4737 | roots = ulist_alloc(GFP_KERNEL); |
4738 | tmp_ulist = ulist_alloc(GFP_KERNEL); | |
4739 | if (!roots || !tmp_ulist) { | |
4740 | ret = -ENOMEM; | |
4741 | goto out_free_ulist; | |
4742 | } | |
4743 | ||
da17066c JM |
4744 | start = round_down(start, btrfs_inode_sectorsize(inode)); |
4745 | len = round_up(max, btrfs_inode_sectorsize(inode)) - start; | |
4d479cf0 | 4746 | |
ec29ed5b CM |
4747 | /* |
4748 | * lookup the last file extent. We're not using i_size here | |
4749 | * because there might be preallocation past i_size | |
4750 | */ | |
f85b7379 DS |
4751 | ret = btrfs_lookup_file_extent(NULL, root, path, |
4752 | btrfs_ino(BTRFS_I(inode)), -1, 0); | |
975f84fe | 4753 | if (ret < 0) { |
5911c8fe | 4754 | goto out_free_ulist; |
2d324f59 LB |
4755 | } else { |
4756 | WARN_ON(!ret); | |
4757 | if (ret == 1) | |
4758 | ret = 0; | |
975f84fe | 4759 | } |
2d324f59 | 4760 | |
975f84fe | 4761 | path->slots[0]--; |
975f84fe | 4762 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
962a298f | 4763 | found_type = found_key.type; |
975f84fe | 4764 | |
ec29ed5b | 4765 | /* No extents, but there might be delalloc bits */ |
4a0cc7ca | 4766 | if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) || |
975f84fe | 4767 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
4768 | /* have to trust i_size as the end */ |
4769 | last = (u64)-1; | |
4770 | last_for_get_extent = isize; | |
4771 | } else { | |
4772 | /* | |
4773 | * remember the start of the last extent. There are a | |
4774 | * bunch of different factors that go into the length of the | |
4775 | * extent, so its much less complex to remember where it started | |
4776 | */ | |
4777 | last = found_key.offset; | |
4778 | last_for_get_extent = last + 1; | |
975f84fe | 4779 | } |
fe09e16c | 4780 | btrfs_release_path(path); |
975f84fe | 4781 | |
ec29ed5b CM |
4782 | /* |
4783 | * we might have some extents allocated but more delalloc past those | |
4784 | * extents. so, we trust isize unless the start of the last extent is | |
4785 | * beyond isize | |
4786 | */ | |
4787 | if (last < isize) { | |
4788 | last = (u64)-1; | |
4789 | last_for_get_extent = isize; | |
4790 | } | |
4791 | ||
ff13db41 | 4792 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
d0082371 | 4793 | &cached_state); |
ec29ed5b | 4794 | |
e3350e16 | 4795 | em = get_extent_skip_holes(inode, start, last_for_get_extent); |
1506fcc8 YS |
4796 | if (!em) |
4797 | goto out; | |
4798 | if (IS_ERR(em)) { | |
4799 | ret = PTR_ERR(em); | |
4800 | goto out; | |
4801 | } | |
975f84fe | 4802 | |
1506fcc8 | 4803 | while (!end) { |
b76bb701 | 4804 | u64 offset_in_extent = 0; |
ea8efc74 CM |
4805 | |
4806 | /* break if the extent we found is outside the range */ | |
4807 | if (em->start >= max || extent_map_end(em) < off) | |
4808 | break; | |
4809 | ||
4810 | /* | |
4811 | * get_extent may return an extent that starts before our | |
4812 | * requested range. We have to make sure the ranges | |
4813 | * we return to fiemap always move forward and don't | |
4814 | * overlap, so adjust the offsets here | |
4815 | */ | |
4816 | em_start = max(em->start, off); | |
1506fcc8 | 4817 | |
ea8efc74 CM |
4818 | /* |
4819 | * record the offset from the start of the extent | |
b76bb701 JB |
4820 | * for adjusting the disk offset below. Only do this if the |
4821 | * extent isn't compressed since our in ram offset may be past | |
4822 | * what we have actually allocated on disk. | |
ea8efc74 | 4823 | */ |
b76bb701 JB |
4824 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
4825 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 4826 | em_end = extent_map_end(em); |
ea8efc74 | 4827 | em_len = em_end - em_start; |
1506fcc8 | 4828 | flags = 0; |
f0986318 FM |
4829 | if (em->block_start < EXTENT_MAP_LAST_BYTE) |
4830 | disko = em->block_start + offset_in_extent; | |
4831 | else | |
4832 | disko = 0; | |
1506fcc8 | 4833 | |
ea8efc74 CM |
4834 | /* |
4835 | * bump off for our next call to get_extent | |
4836 | */ | |
4837 | off = extent_map_end(em); | |
4838 | if (off >= max) | |
4839 | end = 1; | |
4840 | ||
93dbfad7 | 4841 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
4842 | end = 1; |
4843 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 4844 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
4845 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
4846 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 4847 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
4848 | flags |= (FIEMAP_EXTENT_DELALLOC | |
4849 | FIEMAP_EXTENT_UNKNOWN); | |
dc046b10 JB |
4850 | } else if (fieinfo->fi_extents_max) { |
4851 | u64 bytenr = em->block_start - | |
4852 | (em->start - em->orig_start); | |
fe09e16c | 4853 | |
fe09e16c LB |
4854 | /* |
4855 | * As btrfs supports shared space, this information | |
4856 | * can be exported to userspace tools via | |
dc046b10 JB |
4857 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 |
4858 | * then we're just getting a count and we can skip the | |
4859 | * lookup stuff. | |
fe09e16c | 4860 | */ |
bb739cf0 EN |
4861 | ret = btrfs_check_shared(root, |
4862 | btrfs_ino(BTRFS_I(inode)), | |
5911c8fe | 4863 | bytenr, roots, tmp_ulist); |
dc046b10 | 4864 | if (ret < 0) |
fe09e16c | 4865 | goto out_free; |
dc046b10 | 4866 | if (ret) |
fe09e16c | 4867 | flags |= FIEMAP_EXTENT_SHARED; |
dc046b10 | 4868 | ret = 0; |
1506fcc8 YS |
4869 | } |
4870 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
4871 | flags |= FIEMAP_EXTENT_ENCODED; | |
0d2b2372 JB |
4872 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
4873 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
1506fcc8 | 4874 | |
1506fcc8 YS |
4875 | free_extent_map(em); |
4876 | em = NULL; | |
ec29ed5b CM |
4877 | if ((em_start >= last) || em_len == (u64)-1 || |
4878 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
4879 | flags |= FIEMAP_EXTENT_LAST; |
4880 | end = 1; | |
4881 | } | |
4882 | ||
ec29ed5b | 4883 | /* now scan forward to see if this is really the last extent. */ |
e3350e16 | 4884 | em = get_extent_skip_holes(inode, off, last_for_get_extent); |
ec29ed5b CM |
4885 | if (IS_ERR(em)) { |
4886 | ret = PTR_ERR(em); | |
4887 | goto out; | |
4888 | } | |
4889 | if (!em) { | |
975f84fe JB |
4890 | flags |= FIEMAP_EXTENT_LAST; |
4891 | end = 1; | |
4892 | } | |
4751832d QW |
4893 | ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko, |
4894 | em_len, flags); | |
26e726af CS |
4895 | if (ret) { |
4896 | if (ret == 1) | |
4897 | ret = 0; | |
ec29ed5b | 4898 | goto out_free; |
26e726af | 4899 | } |
1506fcc8 YS |
4900 | } |
4901 | out_free: | |
4751832d | 4902 | if (!ret) |
5c5aff98 | 4903 | ret = emit_last_fiemap_cache(fieinfo, &cache); |
1506fcc8 YS |
4904 | free_extent_map(em); |
4905 | out: | |
a52f4cd2 | 4906 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
e43bbe5e | 4907 | &cached_state); |
5911c8fe DS |
4908 | |
4909 | out_free_ulist: | |
e02d48ea | 4910 | btrfs_free_path(path); |
5911c8fe DS |
4911 | ulist_free(roots); |
4912 | ulist_free(tmp_ulist); | |
1506fcc8 YS |
4913 | return ret; |
4914 | } | |
4915 | ||
727011e0 CM |
4916 | static void __free_extent_buffer(struct extent_buffer *eb) |
4917 | { | |
727011e0 CM |
4918 | kmem_cache_free(extent_buffer_cache, eb); |
4919 | } | |
4920 | ||
a26e8c9f | 4921 | int extent_buffer_under_io(struct extent_buffer *eb) |
db7f3436 JB |
4922 | { |
4923 | return (atomic_read(&eb->io_pages) || | |
4924 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4925 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4926 | } | |
4927 | ||
4928 | /* | |
55ac0139 | 4929 | * Release all pages attached to the extent buffer. |
db7f3436 | 4930 | */ |
55ac0139 | 4931 | static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb) |
db7f3436 | 4932 | { |
d64766fd NB |
4933 | int i; |
4934 | int num_pages; | |
b0132a3b | 4935 | int mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
db7f3436 JB |
4936 | |
4937 | BUG_ON(extent_buffer_under_io(eb)); | |
4938 | ||
d64766fd NB |
4939 | num_pages = num_extent_pages(eb); |
4940 | for (i = 0; i < num_pages; i++) { | |
4941 | struct page *page = eb->pages[i]; | |
db7f3436 | 4942 | |
5d2361db FL |
4943 | if (!page) |
4944 | continue; | |
4945 | if (mapped) | |
db7f3436 | 4946 | spin_lock(&page->mapping->private_lock); |
5d2361db FL |
4947 | /* |
4948 | * We do this since we'll remove the pages after we've | |
4949 | * removed the eb from the radix tree, so we could race | |
4950 | * and have this page now attached to the new eb. So | |
4951 | * only clear page_private if it's still connected to | |
4952 | * this eb. | |
4953 | */ | |
4954 | if (PagePrivate(page) && | |
4955 | page->private == (unsigned long)eb) { | |
4956 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4957 | BUG_ON(PageDirty(page)); | |
4958 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 4959 | /* |
5d2361db FL |
4960 | * We need to make sure we haven't be attached |
4961 | * to a new eb. | |
db7f3436 | 4962 | */ |
5d2361db FL |
4963 | ClearPagePrivate(page); |
4964 | set_page_private(page, 0); | |
4965 | /* One for the page private */ | |
09cbfeaf | 4966 | put_page(page); |
db7f3436 | 4967 | } |
5d2361db FL |
4968 | |
4969 | if (mapped) | |
4970 | spin_unlock(&page->mapping->private_lock); | |
4971 | ||
01327610 | 4972 | /* One for when we allocated the page */ |
09cbfeaf | 4973 | put_page(page); |
d64766fd | 4974 | } |
db7f3436 JB |
4975 | } |
4976 | ||
4977 | /* | |
4978 | * Helper for releasing the extent buffer. | |
4979 | */ | |
4980 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4981 | { | |
55ac0139 | 4982 | btrfs_release_extent_buffer_pages(eb); |
8c38938c | 4983 | btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list); |
db7f3436 JB |
4984 | __free_extent_buffer(eb); |
4985 | } | |
4986 | ||
f28491e0 JB |
4987 | static struct extent_buffer * |
4988 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 4989 | unsigned long len) |
d1310b2e CM |
4990 | { |
4991 | struct extent_buffer *eb = NULL; | |
4992 | ||
d1b5c567 | 4993 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
4994 | eb->start = start; |
4995 | eb->len = len; | |
f28491e0 | 4996 | eb->fs_info = fs_info; |
815a51c7 | 4997 | eb->bflags = 0; |
bd681513 | 4998 | rwlock_init(&eb->lock); |
bd681513 | 4999 | atomic_set(&eb->blocking_readers, 0); |
06297d8c | 5000 | eb->blocking_writers = 0; |
ed1b4ed7 | 5001 | eb->lock_nested = false; |
bd681513 CM |
5002 | init_waitqueue_head(&eb->write_lock_wq); |
5003 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 5004 | |
3fd63727 JB |
5005 | btrfs_leak_debug_add(&fs_info->eb_leak_lock, &eb->leak_list, |
5006 | &fs_info->allocated_ebs); | |
6d49ba1b | 5007 | |
3083ee2e | 5008 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 5009 | atomic_set(&eb->refs, 1); |
0b32f4bb | 5010 | atomic_set(&eb->io_pages, 0); |
727011e0 | 5011 | |
b8dae313 DS |
5012 | /* |
5013 | * Sanity checks, currently the maximum is 64k covered by 16x 4k pages | |
5014 | */ | |
5015 | BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE | |
5016 | > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
5017 | BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
d1310b2e | 5018 | |
843ccf9f | 5019 | #ifdef CONFIG_BTRFS_DEBUG |
f3dc24c5 | 5020 | eb->spinning_writers = 0; |
afd495a8 | 5021 | atomic_set(&eb->spinning_readers, 0); |
5c9c799a | 5022 | atomic_set(&eb->read_locks, 0); |
00801ae4 | 5023 | eb->write_locks = 0; |
843ccf9f DS |
5024 | #endif |
5025 | ||
d1310b2e CM |
5026 | return eb; |
5027 | } | |
5028 | ||
815a51c7 JS |
5029 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) |
5030 | { | |
cc5e31a4 | 5031 | int i; |
815a51c7 JS |
5032 | struct page *p; |
5033 | struct extent_buffer *new; | |
cc5e31a4 | 5034 | int num_pages = num_extent_pages(src); |
815a51c7 | 5035 | |
3f556f78 | 5036 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
5037 | if (new == NULL) |
5038 | return NULL; | |
5039 | ||
5040 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 5041 | p = alloc_page(GFP_NOFS); |
db7f3436 JB |
5042 | if (!p) { |
5043 | btrfs_release_extent_buffer(new); | |
5044 | return NULL; | |
5045 | } | |
815a51c7 JS |
5046 | attach_extent_buffer_page(new, p); |
5047 | WARN_ON(PageDirty(p)); | |
5048 | SetPageUptodate(p); | |
5049 | new->pages[i] = p; | |
fba1acf9 | 5050 | copy_page(page_address(p), page_address(src->pages[i])); |
815a51c7 JS |
5051 | } |
5052 | ||
815a51c7 | 5053 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); |
b0132a3b | 5054 | set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags); |
815a51c7 JS |
5055 | |
5056 | return new; | |
5057 | } | |
5058 | ||
0f331229 OS |
5059 | struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
5060 | u64 start, unsigned long len) | |
815a51c7 JS |
5061 | { |
5062 | struct extent_buffer *eb; | |
cc5e31a4 DS |
5063 | int num_pages; |
5064 | int i; | |
815a51c7 | 5065 | |
3f556f78 | 5066 | eb = __alloc_extent_buffer(fs_info, start, len); |
815a51c7 JS |
5067 | if (!eb) |
5068 | return NULL; | |
5069 | ||
65ad0104 | 5070 | num_pages = num_extent_pages(eb); |
815a51c7 | 5071 | for (i = 0; i < num_pages; i++) { |
9ec72677 | 5072 | eb->pages[i] = alloc_page(GFP_NOFS); |
815a51c7 JS |
5073 | if (!eb->pages[i]) |
5074 | goto err; | |
5075 | } | |
5076 | set_extent_buffer_uptodate(eb); | |
5077 | btrfs_set_header_nritems(eb, 0); | |
b0132a3b | 5078 | set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
815a51c7 JS |
5079 | |
5080 | return eb; | |
5081 | err: | |
84167d19 SB |
5082 | for (; i > 0; i--) |
5083 | __free_page(eb->pages[i - 1]); | |
815a51c7 JS |
5084 | __free_extent_buffer(eb); |
5085 | return NULL; | |
5086 | } | |
5087 | ||
0f331229 | 5088 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
da17066c | 5089 | u64 start) |
0f331229 | 5090 | { |
da17066c | 5091 | return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize); |
0f331229 OS |
5092 | } |
5093 | ||
0b32f4bb JB |
5094 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
5095 | { | |
242e18c7 | 5096 | int refs; |
0b32f4bb JB |
5097 | /* the ref bit is tricky. We have to make sure it is set |
5098 | * if we have the buffer dirty. Otherwise the | |
5099 | * code to free a buffer can end up dropping a dirty | |
5100 | * page | |
5101 | * | |
5102 | * Once the ref bit is set, it won't go away while the | |
5103 | * buffer is dirty or in writeback, and it also won't | |
5104 | * go away while we have the reference count on the | |
5105 | * eb bumped. | |
5106 | * | |
5107 | * We can't just set the ref bit without bumping the | |
5108 | * ref on the eb because free_extent_buffer might | |
5109 | * see the ref bit and try to clear it. If this happens | |
5110 | * free_extent_buffer might end up dropping our original | |
5111 | * ref by mistake and freeing the page before we are able | |
5112 | * to add one more ref. | |
5113 | * | |
5114 | * So bump the ref count first, then set the bit. If someone | |
5115 | * beat us to it, drop the ref we added. | |
5116 | */ | |
242e18c7 CM |
5117 | refs = atomic_read(&eb->refs); |
5118 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
5119 | return; | |
5120 | ||
594831c4 JB |
5121 | spin_lock(&eb->refs_lock); |
5122 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 5123 | atomic_inc(&eb->refs); |
594831c4 | 5124 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
5125 | } |
5126 | ||
2457aec6 MG |
5127 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
5128 | struct page *accessed) | |
5df4235e | 5129 | { |
cc5e31a4 | 5130 | int num_pages, i; |
5df4235e | 5131 | |
0b32f4bb JB |
5132 | check_buffer_tree_ref(eb); |
5133 | ||
65ad0104 | 5134 | num_pages = num_extent_pages(eb); |
5df4235e | 5135 | for (i = 0; i < num_pages; i++) { |
fb85fc9a DS |
5136 | struct page *p = eb->pages[i]; |
5137 | ||
2457aec6 MG |
5138 | if (p != accessed) |
5139 | mark_page_accessed(p); | |
5df4235e JB |
5140 | } |
5141 | } | |
5142 | ||
f28491e0 JB |
5143 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
5144 | u64 start) | |
452c75c3 CS |
5145 | { |
5146 | struct extent_buffer *eb; | |
5147 | ||
5148 | rcu_read_lock(); | |
f28491e0 | 5149 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
09cbfeaf | 5150 | start >> PAGE_SHIFT); |
452c75c3 CS |
5151 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
5152 | rcu_read_unlock(); | |
062c19e9 FM |
5153 | /* |
5154 | * Lock our eb's refs_lock to avoid races with | |
5155 | * free_extent_buffer. When we get our eb it might be flagged | |
5156 | * with EXTENT_BUFFER_STALE and another task running | |
5157 | * free_extent_buffer might have seen that flag set, | |
5158 | * eb->refs == 2, that the buffer isn't under IO (dirty and | |
5159 | * writeback flags not set) and it's still in the tree (flag | |
5160 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process | |
5161 | * of decrementing the extent buffer's reference count twice. | |
5162 | * So here we could race and increment the eb's reference count, | |
5163 | * clear its stale flag, mark it as dirty and drop our reference | |
5164 | * before the other task finishes executing free_extent_buffer, | |
5165 | * which would later result in an attempt to free an extent | |
5166 | * buffer that is dirty. | |
5167 | */ | |
5168 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
5169 | spin_lock(&eb->refs_lock); | |
5170 | spin_unlock(&eb->refs_lock); | |
5171 | } | |
2457aec6 | 5172 | mark_extent_buffer_accessed(eb, NULL); |
452c75c3 CS |
5173 | return eb; |
5174 | } | |
5175 | rcu_read_unlock(); | |
5176 | ||
5177 | return NULL; | |
5178 | } | |
5179 | ||
faa2dbf0 JB |
5180 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
5181 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
da17066c | 5182 | u64 start) |
faa2dbf0 JB |
5183 | { |
5184 | struct extent_buffer *eb, *exists = NULL; | |
5185 | int ret; | |
5186 | ||
5187 | eb = find_extent_buffer(fs_info, start); | |
5188 | if (eb) | |
5189 | return eb; | |
da17066c | 5190 | eb = alloc_dummy_extent_buffer(fs_info, start); |
faa2dbf0 | 5191 | if (!eb) |
b6293c82 | 5192 | return ERR_PTR(-ENOMEM); |
faa2dbf0 JB |
5193 | eb->fs_info = fs_info; |
5194 | again: | |
e1860a77 | 5195 | ret = radix_tree_preload(GFP_NOFS); |
b6293c82 DC |
5196 | if (ret) { |
5197 | exists = ERR_PTR(ret); | |
faa2dbf0 | 5198 | goto free_eb; |
b6293c82 | 5199 | } |
faa2dbf0 JB |
5200 | spin_lock(&fs_info->buffer_lock); |
5201 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
09cbfeaf | 5202 | start >> PAGE_SHIFT, eb); |
faa2dbf0 JB |
5203 | spin_unlock(&fs_info->buffer_lock); |
5204 | radix_tree_preload_end(); | |
5205 | if (ret == -EEXIST) { | |
5206 | exists = find_extent_buffer(fs_info, start); | |
5207 | if (exists) | |
5208 | goto free_eb; | |
5209 | else | |
5210 | goto again; | |
5211 | } | |
5212 | check_buffer_tree_ref(eb); | |
5213 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
5214 | ||
faa2dbf0 JB |
5215 | return eb; |
5216 | free_eb: | |
5217 | btrfs_release_extent_buffer(eb); | |
5218 | return exists; | |
5219 | } | |
5220 | #endif | |
5221 | ||
f28491e0 | 5222 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
ce3e6984 | 5223 | u64 start) |
d1310b2e | 5224 | { |
da17066c | 5225 | unsigned long len = fs_info->nodesize; |
cc5e31a4 DS |
5226 | int num_pages; |
5227 | int i; | |
09cbfeaf | 5228 | unsigned long index = start >> PAGE_SHIFT; |
d1310b2e | 5229 | struct extent_buffer *eb; |
6af118ce | 5230 | struct extent_buffer *exists = NULL; |
d1310b2e | 5231 | struct page *p; |
f28491e0 | 5232 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
d1310b2e | 5233 | int uptodate = 1; |
19fe0a8b | 5234 | int ret; |
d1310b2e | 5235 | |
da17066c | 5236 | if (!IS_ALIGNED(start, fs_info->sectorsize)) { |
c871b0f2 LB |
5237 | btrfs_err(fs_info, "bad tree block start %llu", start); |
5238 | return ERR_PTR(-EINVAL); | |
5239 | } | |
5240 | ||
f28491e0 | 5241 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 5242 | if (eb) |
6af118ce | 5243 | return eb; |
6af118ce | 5244 | |
23d79d81 | 5245 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 5246 | if (!eb) |
c871b0f2 | 5247 | return ERR_PTR(-ENOMEM); |
d1310b2e | 5248 | |
65ad0104 | 5249 | num_pages = num_extent_pages(eb); |
727011e0 | 5250 | for (i = 0; i < num_pages; i++, index++) { |
d1b5c567 | 5251 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
c871b0f2 LB |
5252 | if (!p) { |
5253 | exists = ERR_PTR(-ENOMEM); | |
6af118ce | 5254 | goto free_eb; |
c871b0f2 | 5255 | } |
4f2de97a JB |
5256 | |
5257 | spin_lock(&mapping->private_lock); | |
5258 | if (PagePrivate(p)) { | |
5259 | /* | |
5260 | * We could have already allocated an eb for this page | |
5261 | * and attached one so lets see if we can get a ref on | |
5262 | * the existing eb, and if we can we know it's good and | |
5263 | * we can just return that one, else we know we can just | |
5264 | * overwrite page->private. | |
5265 | */ | |
5266 | exists = (struct extent_buffer *)p->private; | |
5267 | if (atomic_inc_not_zero(&exists->refs)) { | |
5268 | spin_unlock(&mapping->private_lock); | |
5269 | unlock_page(p); | |
09cbfeaf | 5270 | put_page(p); |
2457aec6 | 5271 | mark_extent_buffer_accessed(exists, p); |
4f2de97a JB |
5272 | goto free_eb; |
5273 | } | |
5ca64f45 | 5274 | exists = NULL; |
4f2de97a | 5275 | |
0b32f4bb | 5276 | /* |
4f2de97a JB |
5277 | * Do this so attach doesn't complain and we need to |
5278 | * drop the ref the old guy had. | |
5279 | */ | |
5280 | ClearPagePrivate(p); | |
0b32f4bb | 5281 | WARN_ON(PageDirty(p)); |
09cbfeaf | 5282 | put_page(p); |
d1310b2e | 5283 | } |
4f2de97a JB |
5284 | attach_extent_buffer_page(eb, p); |
5285 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 5286 | WARN_ON(PageDirty(p)); |
727011e0 | 5287 | eb->pages[i] = p; |
d1310b2e CM |
5288 | if (!PageUptodate(p)) |
5289 | uptodate = 0; | |
eb14ab8e CM |
5290 | |
5291 | /* | |
b16d011e NB |
5292 | * We can't unlock the pages just yet since the extent buffer |
5293 | * hasn't been properly inserted in the radix tree, this | |
5294 | * opens a race with btree_releasepage which can free a page | |
5295 | * while we are still filling in all pages for the buffer and | |
5296 | * we could crash. | |
eb14ab8e | 5297 | */ |
d1310b2e CM |
5298 | } |
5299 | if (uptodate) | |
b4ce94de | 5300 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 5301 | again: |
e1860a77 | 5302 | ret = radix_tree_preload(GFP_NOFS); |
c871b0f2 LB |
5303 | if (ret) { |
5304 | exists = ERR_PTR(ret); | |
19fe0a8b | 5305 | goto free_eb; |
c871b0f2 | 5306 | } |
19fe0a8b | 5307 | |
f28491e0 JB |
5308 | spin_lock(&fs_info->buffer_lock); |
5309 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
09cbfeaf | 5310 | start >> PAGE_SHIFT, eb); |
f28491e0 | 5311 | spin_unlock(&fs_info->buffer_lock); |
452c75c3 | 5312 | radix_tree_preload_end(); |
19fe0a8b | 5313 | if (ret == -EEXIST) { |
f28491e0 | 5314 | exists = find_extent_buffer(fs_info, start); |
452c75c3 CS |
5315 | if (exists) |
5316 | goto free_eb; | |
5317 | else | |
115391d2 | 5318 | goto again; |
6af118ce | 5319 | } |
6af118ce | 5320 | /* add one reference for the tree */ |
0b32f4bb | 5321 | check_buffer_tree_ref(eb); |
34b41ace | 5322 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
5323 | |
5324 | /* | |
b16d011e NB |
5325 | * Now it's safe to unlock the pages because any calls to |
5326 | * btree_releasepage will correctly detect that a page belongs to a | |
5327 | * live buffer and won't free them prematurely. | |
eb14ab8e | 5328 | */ |
28187ae5 NB |
5329 | for (i = 0; i < num_pages; i++) |
5330 | unlock_page(eb->pages[i]); | |
d1310b2e CM |
5331 | return eb; |
5332 | ||
6af118ce | 5333 | free_eb: |
5ca64f45 | 5334 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
5335 | for (i = 0; i < num_pages; i++) { |
5336 | if (eb->pages[i]) | |
5337 | unlock_page(eb->pages[i]); | |
5338 | } | |
eb14ab8e | 5339 | |
897ca6e9 | 5340 | btrfs_release_extent_buffer(eb); |
6af118ce | 5341 | return exists; |
d1310b2e | 5342 | } |
d1310b2e | 5343 | |
3083ee2e JB |
5344 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
5345 | { | |
5346 | struct extent_buffer *eb = | |
5347 | container_of(head, struct extent_buffer, rcu_head); | |
5348 | ||
5349 | __free_extent_buffer(eb); | |
5350 | } | |
5351 | ||
f7a52a40 | 5352 | static int release_extent_buffer(struct extent_buffer *eb) |
5ce48d0f | 5353 | __releases(&eb->refs_lock) |
3083ee2e | 5354 | { |
07e21c4d NB |
5355 | lockdep_assert_held(&eb->refs_lock); |
5356 | ||
3083ee2e JB |
5357 | WARN_ON(atomic_read(&eb->refs) == 0); |
5358 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 5359 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 5360 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 5361 | |
815a51c7 | 5362 | spin_unlock(&eb->refs_lock); |
3083ee2e | 5363 | |
f28491e0 JB |
5364 | spin_lock(&fs_info->buffer_lock); |
5365 | radix_tree_delete(&fs_info->buffer_radix, | |
09cbfeaf | 5366 | eb->start >> PAGE_SHIFT); |
f28491e0 | 5367 | spin_unlock(&fs_info->buffer_lock); |
34b41ace JB |
5368 | } else { |
5369 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 5370 | } |
3083ee2e | 5371 | |
8c38938c | 5372 | btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list); |
3083ee2e | 5373 | /* Should be safe to release our pages at this point */ |
55ac0139 | 5374 | btrfs_release_extent_buffer_pages(eb); |
bcb7e449 | 5375 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
b0132a3b | 5376 | if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) { |
bcb7e449 JB |
5377 | __free_extent_buffer(eb); |
5378 | return 1; | |
5379 | } | |
5380 | #endif | |
3083ee2e | 5381 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 5382 | return 1; |
3083ee2e JB |
5383 | } |
5384 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
5385 | |
5386 | return 0; | |
3083ee2e JB |
5387 | } |
5388 | ||
d1310b2e CM |
5389 | void free_extent_buffer(struct extent_buffer *eb) |
5390 | { | |
242e18c7 CM |
5391 | int refs; |
5392 | int old; | |
d1310b2e CM |
5393 | if (!eb) |
5394 | return; | |
5395 | ||
242e18c7 CM |
5396 | while (1) { |
5397 | refs = atomic_read(&eb->refs); | |
46cc775e NB |
5398 | if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3) |
5399 | || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && | |
5400 | refs == 1)) | |
242e18c7 CM |
5401 | break; |
5402 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
5403 | if (old == refs) | |
5404 | return; | |
5405 | } | |
5406 | ||
3083ee2e JB |
5407 | spin_lock(&eb->refs_lock); |
5408 | if (atomic_read(&eb->refs) == 2 && | |
5409 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 5410 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
5411 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5412 | atomic_dec(&eb->refs); | |
5413 | ||
5414 | /* | |
5415 | * I know this is terrible, but it's temporary until we stop tracking | |
5416 | * the uptodate bits and such for the extent buffers. | |
5417 | */ | |
f7a52a40 | 5418 | release_extent_buffer(eb); |
3083ee2e JB |
5419 | } |
5420 | ||
5421 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
5422 | { | |
5423 | if (!eb) | |
d1310b2e CM |
5424 | return; |
5425 | ||
3083ee2e JB |
5426 | spin_lock(&eb->refs_lock); |
5427 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
5428 | ||
0b32f4bb | 5429 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
5430 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5431 | atomic_dec(&eb->refs); | |
f7a52a40 | 5432 | release_extent_buffer(eb); |
d1310b2e | 5433 | } |
d1310b2e | 5434 | |
1d4284bd | 5435 | void clear_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5436 | { |
cc5e31a4 DS |
5437 | int i; |
5438 | int num_pages; | |
d1310b2e CM |
5439 | struct page *page; |
5440 | ||
65ad0104 | 5441 | num_pages = num_extent_pages(eb); |
d1310b2e CM |
5442 | |
5443 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 5444 | page = eb->pages[i]; |
b9473439 | 5445 | if (!PageDirty(page)) |
d2c3f4f6 CM |
5446 | continue; |
5447 | ||
a61e6f29 | 5448 | lock_page(page); |
eb14ab8e CM |
5449 | WARN_ON(!PagePrivate(page)); |
5450 | ||
d1310b2e | 5451 | clear_page_dirty_for_io(page); |
b93b0163 | 5452 | xa_lock_irq(&page->mapping->i_pages); |
0a943c65 MW |
5453 | if (!PageDirty(page)) |
5454 | __xa_clear_mark(&page->mapping->i_pages, | |
5455 | page_index(page), PAGECACHE_TAG_DIRTY); | |
b93b0163 | 5456 | xa_unlock_irq(&page->mapping->i_pages); |
bf0da8c1 | 5457 | ClearPageError(page); |
a61e6f29 | 5458 | unlock_page(page); |
d1310b2e | 5459 | } |
0b32f4bb | 5460 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 5461 | } |
d1310b2e | 5462 | |
abb57ef3 | 5463 | bool set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5464 | { |
cc5e31a4 DS |
5465 | int i; |
5466 | int num_pages; | |
abb57ef3 | 5467 | bool was_dirty; |
d1310b2e | 5468 | |
0b32f4bb JB |
5469 | check_buffer_tree_ref(eb); |
5470 | ||
b9473439 | 5471 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 5472 | |
65ad0104 | 5473 | num_pages = num_extent_pages(eb); |
3083ee2e | 5474 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
5475 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
5476 | ||
abb57ef3 LB |
5477 | if (!was_dirty) |
5478 | for (i = 0; i < num_pages; i++) | |
5479 | set_page_dirty(eb->pages[i]); | |
51995c39 LB |
5480 | |
5481 | #ifdef CONFIG_BTRFS_DEBUG | |
5482 | for (i = 0; i < num_pages; i++) | |
5483 | ASSERT(PageDirty(eb->pages[i])); | |
5484 | #endif | |
5485 | ||
b9473439 | 5486 | return was_dirty; |
d1310b2e | 5487 | } |
d1310b2e | 5488 | |
69ba3927 | 5489 | void clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 | 5490 | { |
cc5e31a4 | 5491 | int i; |
1259ab75 | 5492 | struct page *page; |
cc5e31a4 | 5493 | int num_pages; |
1259ab75 | 5494 | |
b4ce94de | 5495 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 5496 | num_pages = num_extent_pages(eb); |
1259ab75 | 5497 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5498 | page = eb->pages[i]; |
33958dc6 CM |
5499 | if (page) |
5500 | ClearPageUptodate(page); | |
1259ab75 | 5501 | } |
1259ab75 CM |
5502 | } |
5503 | ||
09c25a8c | 5504 | void set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 5505 | { |
cc5e31a4 | 5506 | int i; |
d1310b2e | 5507 | struct page *page; |
cc5e31a4 | 5508 | int num_pages; |
d1310b2e | 5509 | |
0b32f4bb | 5510 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 5511 | num_pages = num_extent_pages(eb); |
d1310b2e | 5512 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5513 | page = eb->pages[i]; |
d1310b2e CM |
5514 | SetPageUptodate(page); |
5515 | } | |
d1310b2e | 5516 | } |
d1310b2e | 5517 | |
c2ccfbc6 | 5518 | int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num) |
d1310b2e | 5519 | { |
cc5e31a4 | 5520 | int i; |
d1310b2e CM |
5521 | struct page *page; |
5522 | int err; | |
5523 | int ret = 0; | |
ce9adaa5 CM |
5524 | int locked_pages = 0; |
5525 | int all_uptodate = 1; | |
cc5e31a4 | 5526 | int num_pages; |
727011e0 | 5527 | unsigned long num_reads = 0; |
a86c12c7 | 5528 | struct bio *bio = NULL; |
c8b97818 | 5529 | unsigned long bio_flags = 0; |
a86c12c7 | 5530 | |
b4ce94de | 5531 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
5532 | return 0; |
5533 | ||
65ad0104 | 5534 | num_pages = num_extent_pages(eb); |
8436ea91 | 5535 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5536 | page = eb->pages[i]; |
bb82ab88 | 5537 | if (wait == WAIT_NONE) { |
2db04966 | 5538 | if (!trylock_page(page)) |
ce9adaa5 | 5539 | goto unlock_exit; |
d1310b2e CM |
5540 | } else { |
5541 | lock_page(page); | |
5542 | } | |
ce9adaa5 | 5543 | locked_pages++; |
2571e739 LB |
5544 | } |
5545 | /* | |
5546 | * We need to firstly lock all pages to make sure that | |
5547 | * the uptodate bit of our pages won't be affected by | |
5548 | * clear_extent_buffer_uptodate(). | |
5549 | */ | |
8436ea91 | 5550 | for (i = 0; i < num_pages; i++) { |
2571e739 | 5551 | page = eb->pages[i]; |
727011e0 CM |
5552 | if (!PageUptodate(page)) { |
5553 | num_reads++; | |
ce9adaa5 | 5554 | all_uptodate = 0; |
727011e0 | 5555 | } |
ce9adaa5 | 5556 | } |
2571e739 | 5557 | |
ce9adaa5 | 5558 | if (all_uptodate) { |
8436ea91 | 5559 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
5560 | goto unlock_exit; |
5561 | } | |
5562 | ||
656f30db | 5563 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 5564 | eb->read_mirror = 0; |
0b32f4bb | 5565 | atomic_set(&eb->io_pages, num_reads); |
8436ea91 | 5566 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5567 | page = eb->pages[i]; |
baf863b9 | 5568 | |
ce9adaa5 | 5569 | if (!PageUptodate(page)) { |
baf863b9 LB |
5570 | if (ret) { |
5571 | atomic_dec(&eb->io_pages); | |
5572 | unlock_page(page); | |
5573 | continue; | |
5574 | } | |
5575 | ||
f188591e | 5576 | ClearPageError(page); |
0d44fea7 | 5577 | err = __extent_read_full_page(page, |
6af49dbd | 5578 | btree_get_extent, &bio, |
d4c7ca86 | 5579 | mirror_num, &bio_flags, |
1f7ad75b | 5580 | REQ_META); |
baf863b9 | 5581 | if (err) { |
d1310b2e | 5582 | ret = err; |
baf863b9 LB |
5583 | /* |
5584 | * We use &bio in above __extent_read_full_page, | |
5585 | * so we ensure that if it returns error, the | |
5586 | * current page fails to add itself to bio and | |
5587 | * it's been unlocked. | |
5588 | * | |
5589 | * We must dec io_pages by ourselves. | |
5590 | */ | |
5591 | atomic_dec(&eb->io_pages); | |
5592 | } | |
d1310b2e CM |
5593 | } else { |
5594 | unlock_page(page); | |
5595 | } | |
5596 | } | |
5597 | ||
355808c2 | 5598 | if (bio) { |
1f7ad75b | 5599 | err = submit_one_bio(bio, mirror_num, bio_flags); |
79787eaa JM |
5600 | if (err) |
5601 | return err; | |
355808c2 | 5602 | } |
a86c12c7 | 5603 | |
bb82ab88 | 5604 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 5605 | return ret; |
d397712b | 5606 | |
8436ea91 | 5607 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5608 | page = eb->pages[i]; |
d1310b2e | 5609 | wait_on_page_locked(page); |
d397712b | 5610 | if (!PageUptodate(page)) |
d1310b2e | 5611 | ret = -EIO; |
d1310b2e | 5612 | } |
d397712b | 5613 | |
d1310b2e | 5614 | return ret; |
ce9adaa5 CM |
5615 | |
5616 | unlock_exit: | |
d397712b | 5617 | while (locked_pages > 0) { |
ce9adaa5 | 5618 | locked_pages--; |
8436ea91 JB |
5619 | page = eb->pages[locked_pages]; |
5620 | unlock_page(page); | |
ce9adaa5 CM |
5621 | } |
5622 | return ret; | |
d1310b2e | 5623 | } |
d1310b2e | 5624 | |
1cbb1f45 JM |
5625 | void read_extent_buffer(const struct extent_buffer *eb, void *dstv, |
5626 | unsigned long start, unsigned long len) | |
d1310b2e CM |
5627 | { |
5628 | size_t cur; | |
5629 | size_t offset; | |
5630 | struct page *page; | |
5631 | char *kaddr; | |
5632 | char *dst = (char *)dstv; | |
7073017a | 5633 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5634 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
d1310b2e | 5635 | |
f716abd5 LB |
5636 | if (start + len > eb->len) { |
5637 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n", | |
5638 | eb->start, eb->len, start, len); | |
5639 | memset(dst, 0, len); | |
5640 | return; | |
5641 | } | |
d1310b2e | 5642 | |
7073017a | 5643 | offset = offset_in_page(start_offset + start); |
d1310b2e | 5644 | |
d397712b | 5645 | while (len > 0) { |
fb85fc9a | 5646 | page = eb->pages[i]; |
d1310b2e | 5647 | |
09cbfeaf | 5648 | cur = min(len, (PAGE_SIZE - offset)); |
a6591715 | 5649 | kaddr = page_address(page); |
d1310b2e | 5650 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
5651 | |
5652 | dst += cur; | |
5653 | len -= cur; | |
5654 | offset = 0; | |
5655 | i++; | |
5656 | } | |
5657 | } | |
d1310b2e | 5658 | |
1cbb1f45 JM |
5659 | int read_extent_buffer_to_user(const struct extent_buffer *eb, |
5660 | void __user *dstv, | |
5661 | unsigned long start, unsigned long len) | |
550ac1d8 GH |
5662 | { |
5663 | size_t cur; | |
5664 | size_t offset; | |
5665 | struct page *page; | |
5666 | char *kaddr; | |
5667 | char __user *dst = (char __user *)dstv; | |
7073017a | 5668 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5669 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
550ac1d8 GH |
5670 | int ret = 0; |
5671 | ||
5672 | WARN_ON(start > eb->len); | |
5673 | WARN_ON(start + len > eb->start + eb->len); | |
5674 | ||
7073017a | 5675 | offset = offset_in_page(start_offset + start); |
550ac1d8 GH |
5676 | |
5677 | while (len > 0) { | |
fb85fc9a | 5678 | page = eb->pages[i]; |
550ac1d8 | 5679 | |
09cbfeaf | 5680 | cur = min(len, (PAGE_SIZE - offset)); |
550ac1d8 GH |
5681 | kaddr = page_address(page); |
5682 | if (copy_to_user(dst, kaddr + offset, cur)) { | |
5683 | ret = -EFAULT; | |
5684 | break; | |
5685 | } | |
5686 | ||
5687 | dst += cur; | |
5688 | len -= cur; | |
5689 | offset = 0; | |
5690 | i++; | |
5691 | } | |
5692 | ||
5693 | return ret; | |
5694 | } | |
5695 | ||
415b35a5 LB |
5696 | /* |
5697 | * return 0 if the item is found within a page. | |
5698 | * return 1 if the item spans two pages. | |
5699 | * return -EINVAL otherwise. | |
5700 | */ | |
1cbb1f45 JM |
5701 | int map_private_extent_buffer(const struct extent_buffer *eb, |
5702 | unsigned long start, unsigned long min_len, | |
5703 | char **map, unsigned long *map_start, | |
5704 | unsigned long *map_len) | |
d1310b2e | 5705 | { |
cc2c39d6 | 5706 | size_t offset; |
d1310b2e CM |
5707 | char *kaddr; |
5708 | struct page *p; | |
7073017a | 5709 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5710 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
d1310b2e | 5711 | unsigned long end_i = (start_offset + start + min_len - 1) >> |
09cbfeaf | 5712 | PAGE_SHIFT; |
d1310b2e | 5713 | |
f716abd5 LB |
5714 | if (start + min_len > eb->len) { |
5715 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n", | |
5716 | eb->start, eb->len, start, min_len); | |
5717 | return -EINVAL; | |
5718 | } | |
5719 | ||
d1310b2e | 5720 | if (i != end_i) |
415b35a5 | 5721 | return 1; |
d1310b2e CM |
5722 | |
5723 | if (i == 0) { | |
5724 | offset = start_offset; | |
5725 | *map_start = 0; | |
5726 | } else { | |
5727 | offset = 0; | |
09cbfeaf | 5728 | *map_start = ((u64)i << PAGE_SHIFT) - start_offset; |
d1310b2e | 5729 | } |
d397712b | 5730 | |
fb85fc9a | 5731 | p = eb->pages[i]; |
a6591715 | 5732 | kaddr = page_address(p); |
d1310b2e | 5733 | *map = kaddr + offset; |
09cbfeaf | 5734 | *map_len = PAGE_SIZE - offset; |
d1310b2e CM |
5735 | return 0; |
5736 | } | |
d1310b2e | 5737 | |
1cbb1f45 JM |
5738 | int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv, |
5739 | unsigned long start, unsigned long len) | |
d1310b2e CM |
5740 | { |
5741 | size_t cur; | |
5742 | size_t offset; | |
5743 | struct page *page; | |
5744 | char *kaddr; | |
5745 | char *ptr = (char *)ptrv; | |
7073017a | 5746 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5747 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
d1310b2e CM |
5748 | int ret = 0; |
5749 | ||
5750 | WARN_ON(start > eb->len); | |
5751 | WARN_ON(start + len > eb->start + eb->len); | |
5752 | ||
7073017a | 5753 | offset = offset_in_page(start_offset + start); |
d1310b2e | 5754 | |
d397712b | 5755 | while (len > 0) { |
fb85fc9a | 5756 | page = eb->pages[i]; |
d1310b2e | 5757 | |
09cbfeaf | 5758 | cur = min(len, (PAGE_SIZE - offset)); |
d1310b2e | 5759 | |
a6591715 | 5760 | kaddr = page_address(page); |
d1310b2e | 5761 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
5762 | if (ret) |
5763 | break; | |
5764 | ||
5765 | ptr += cur; | |
5766 | len -= cur; | |
5767 | offset = 0; | |
5768 | i++; | |
5769 | } | |
5770 | return ret; | |
5771 | } | |
d1310b2e | 5772 | |
f157bf76 DS |
5773 | void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb, |
5774 | const void *srcv) | |
5775 | { | |
5776 | char *kaddr; | |
5777 | ||
5778 | WARN_ON(!PageUptodate(eb->pages[0])); | |
5779 | kaddr = page_address(eb->pages[0]); | |
5780 | memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv, | |
5781 | BTRFS_FSID_SIZE); | |
5782 | } | |
5783 | ||
5784 | void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv) | |
5785 | { | |
5786 | char *kaddr; | |
5787 | ||
5788 | WARN_ON(!PageUptodate(eb->pages[0])); | |
5789 | kaddr = page_address(eb->pages[0]); | |
5790 | memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv, | |
5791 | BTRFS_FSID_SIZE); | |
5792 | } | |
5793 | ||
d1310b2e CM |
5794 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, |
5795 | unsigned long start, unsigned long len) | |
5796 | { | |
5797 | size_t cur; | |
5798 | size_t offset; | |
5799 | struct page *page; | |
5800 | char *kaddr; | |
5801 | char *src = (char *)srcv; | |
7073017a | 5802 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5803 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
d1310b2e CM |
5804 | |
5805 | WARN_ON(start > eb->len); | |
5806 | WARN_ON(start + len > eb->start + eb->len); | |
5807 | ||
7073017a | 5808 | offset = offset_in_page(start_offset + start); |
d1310b2e | 5809 | |
d397712b | 5810 | while (len > 0) { |
fb85fc9a | 5811 | page = eb->pages[i]; |
d1310b2e CM |
5812 | WARN_ON(!PageUptodate(page)); |
5813 | ||
09cbfeaf | 5814 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 5815 | kaddr = page_address(page); |
d1310b2e | 5816 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
5817 | |
5818 | src += cur; | |
5819 | len -= cur; | |
5820 | offset = 0; | |
5821 | i++; | |
5822 | } | |
5823 | } | |
d1310b2e | 5824 | |
b159fa28 DS |
5825 | void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start, |
5826 | unsigned long len) | |
d1310b2e CM |
5827 | { |
5828 | size_t cur; | |
5829 | size_t offset; | |
5830 | struct page *page; | |
5831 | char *kaddr; | |
7073017a | 5832 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5833 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
d1310b2e CM |
5834 | |
5835 | WARN_ON(start > eb->len); | |
5836 | WARN_ON(start + len > eb->start + eb->len); | |
5837 | ||
7073017a | 5838 | offset = offset_in_page(start_offset + start); |
d1310b2e | 5839 | |
d397712b | 5840 | while (len > 0) { |
fb85fc9a | 5841 | page = eb->pages[i]; |
d1310b2e CM |
5842 | WARN_ON(!PageUptodate(page)); |
5843 | ||
09cbfeaf | 5844 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 5845 | kaddr = page_address(page); |
b159fa28 | 5846 | memset(kaddr + offset, 0, cur); |
d1310b2e CM |
5847 | |
5848 | len -= cur; | |
5849 | offset = 0; | |
5850 | i++; | |
5851 | } | |
5852 | } | |
d1310b2e | 5853 | |
58e8012c DS |
5854 | void copy_extent_buffer_full(struct extent_buffer *dst, |
5855 | struct extent_buffer *src) | |
5856 | { | |
5857 | int i; | |
cc5e31a4 | 5858 | int num_pages; |
58e8012c DS |
5859 | |
5860 | ASSERT(dst->len == src->len); | |
5861 | ||
65ad0104 | 5862 | num_pages = num_extent_pages(dst); |
58e8012c DS |
5863 | for (i = 0; i < num_pages; i++) |
5864 | copy_page(page_address(dst->pages[i]), | |
5865 | page_address(src->pages[i])); | |
5866 | } | |
5867 | ||
d1310b2e CM |
5868 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, |
5869 | unsigned long dst_offset, unsigned long src_offset, | |
5870 | unsigned long len) | |
5871 | { | |
5872 | u64 dst_len = dst->len; | |
5873 | size_t cur; | |
5874 | size_t offset; | |
5875 | struct page *page; | |
5876 | char *kaddr; | |
7073017a | 5877 | size_t start_offset = offset_in_page(dst->start); |
09cbfeaf | 5878 | unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT; |
d1310b2e CM |
5879 | |
5880 | WARN_ON(src->len != dst_len); | |
5881 | ||
7073017a | 5882 | offset = offset_in_page(start_offset + dst_offset); |
d1310b2e | 5883 | |
d397712b | 5884 | while (len > 0) { |
fb85fc9a | 5885 | page = dst->pages[i]; |
d1310b2e CM |
5886 | WARN_ON(!PageUptodate(page)); |
5887 | ||
09cbfeaf | 5888 | cur = min(len, (unsigned long)(PAGE_SIZE - offset)); |
d1310b2e | 5889 | |
a6591715 | 5890 | kaddr = page_address(page); |
d1310b2e | 5891 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
5892 | |
5893 | src_offset += cur; | |
5894 | len -= cur; | |
5895 | offset = 0; | |
5896 | i++; | |
5897 | } | |
5898 | } | |
d1310b2e | 5899 | |
3e1e8bb7 OS |
5900 | /* |
5901 | * eb_bitmap_offset() - calculate the page and offset of the byte containing the | |
5902 | * given bit number | |
5903 | * @eb: the extent buffer | |
5904 | * @start: offset of the bitmap item in the extent buffer | |
5905 | * @nr: bit number | |
5906 | * @page_index: return index of the page in the extent buffer that contains the | |
5907 | * given bit number | |
5908 | * @page_offset: return offset into the page given by page_index | |
5909 | * | |
5910 | * This helper hides the ugliness of finding the byte in an extent buffer which | |
5911 | * contains a given bit. | |
5912 | */ | |
5913 | static inline void eb_bitmap_offset(struct extent_buffer *eb, | |
5914 | unsigned long start, unsigned long nr, | |
5915 | unsigned long *page_index, | |
5916 | size_t *page_offset) | |
5917 | { | |
7073017a | 5918 | size_t start_offset = offset_in_page(eb->start); |
3e1e8bb7 OS |
5919 | size_t byte_offset = BIT_BYTE(nr); |
5920 | size_t offset; | |
5921 | ||
5922 | /* | |
5923 | * The byte we want is the offset of the extent buffer + the offset of | |
5924 | * the bitmap item in the extent buffer + the offset of the byte in the | |
5925 | * bitmap item. | |
5926 | */ | |
5927 | offset = start_offset + start + byte_offset; | |
5928 | ||
09cbfeaf | 5929 | *page_index = offset >> PAGE_SHIFT; |
7073017a | 5930 | *page_offset = offset_in_page(offset); |
3e1e8bb7 OS |
5931 | } |
5932 | ||
5933 | /** | |
5934 | * extent_buffer_test_bit - determine whether a bit in a bitmap item is set | |
5935 | * @eb: the extent buffer | |
5936 | * @start: offset of the bitmap item in the extent buffer | |
5937 | * @nr: bit number to test | |
5938 | */ | |
5939 | int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start, | |
5940 | unsigned long nr) | |
5941 | { | |
2fe1d551 | 5942 | u8 *kaddr; |
3e1e8bb7 OS |
5943 | struct page *page; |
5944 | unsigned long i; | |
5945 | size_t offset; | |
5946 | ||
5947 | eb_bitmap_offset(eb, start, nr, &i, &offset); | |
5948 | page = eb->pages[i]; | |
5949 | WARN_ON(!PageUptodate(page)); | |
5950 | kaddr = page_address(page); | |
5951 | return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1))); | |
5952 | } | |
5953 | ||
5954 | /** | |
5955 | * extent_buffer_bitmap_set - set an area of a bitmap | |
5956 | * @eb: the extent buffer | |
5957 | * @start: offset of the bitmap item in the extent buffer | |
5958 | * @pos: bit number of the first bit | |
5959 | * @len: number of bits to set | |
5960 | */ | |
5961 | void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start, | |
5962 | unsigned long pos, unsigned long len) | |
5963 | { | |
2fe1d551 | 5964 | u8 *kaddr; |
3e1e8bb7 OS |
5965 | struct page *page; |
5966 | unsigned long i; | |
5967 | size_t offset; | |
5968 | const unsigned int size = pos + len; | |
5969 | int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 5970 | u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
5971 | |
5972 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5973 | page = eb->pages[i]; | |
5974 | WARN_ON(!PageUptodate(page)); | |
5975 | kaddr = page_address(page); | |
5976 | ||
5977 | while (len >= bits_to_set) { | |
5978 | kaddr[offset] |= mask_to_set; | |
5979 | len -= bits_to_set; | |
5980 | bits_to_set = BITS_PER_BYTE; | |
9c894696 | 5981 | mask_to_set = ~0; |
09cbfeaf | 5982 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5983 | offset = 0; |
5984 | page = eb->pages[++i]; | |
5985 | WARN_ON(!PageUptodate(page)); | |
5986 | kaddr = page_address(page); | |
5987 | } | |
5988 | } | |
5989 | if (len) { | |
5990 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
5991 | kaddr[offset] |= mask_to_set; | |
5992 | } | |
5993 | } | |
5994 | ||
5995 | ||
5996 | /** | |
5997 | * extent_buffer_bitmap_clear - clear an area of a bitmap | |
5998 | * @eb: the extent buffer | |
5999 | * @start: offset of the bitmap item in the extent buffer | |
6000 | * @pos: bit number of the first bit | |
6001 | * @len: number of bits to clear | |
6002 | */ | |
6003 | void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start, | |
6004 | unsigned long pos, unsigned long len) | |
6005 | { | |
2fe1d551 | 6006 | u8 *kaddr; |
3e1e8bb7 OS |
6007 | struct page *page; |
6008 | unsigned long i; | |
6009 | size_t offset; | |
6010 | const unsigned int size = pos + len; | |
6011 | int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 6012 | u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
6013 | |
6014 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
6015 | page = eb->pages[i]; | |
6016 | WARN_ON(!PageUptodate(page)); | |
6017 | kaddr = page_address(page); | |
6018 | ||
6019 | while (len >= bits_to_clear) { | |
6020 | kaddr[offset] &= ~mask_to_clear; | |
6021 | len -= bits_to_clear; | |
6022 | bits_to_clear = BITS_PER_BYTE; | |
9c894696 | 6023 | mask_to_clear = ~0; |
09cbfeaf | 6024 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
6025 | offset = 0; |
6026 | page = eb->pages[++i]; | |
6027 | WARN_ON(!PageUptodate(page)); | |
6028 | kaddr = page_address(page); | |
6029 | } | |
6030 | } | |
6031 | if (len) { | |
6032 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
6033 | kaddr[offset] &= ~mask_to_clear; | |
6034 | } | |
6035 | } | |
6036 | ||
3387206f ST |
6037 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
6038 | { | |
6039 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
6040 | return distance < len; | |
6041 | } | |
6042 | ||
d1310b2e CM |
6043 | static void copy_pages(struct page *dst_page, struct page *src_page, |
6044 | unsigned long dst_off, unsigned long src_off, | |
6045 | unsigned long len) | |
6046 | { | |
a6591715 | 6047 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 6048 | char *src_kaddr; |
727011e0 | 6049 | int must_memmove = 0; |
d1310b2e | 6050 | |
3387206f | 6051 | if (dst_page != src_page) { |
a6591715 | 6052 | src_kaddr = page_address(src_page); |
3387206f | 6053 | } else { |
d1310b2e | 6054 | src_kaddr = dst_kaddr; |
727011e0 CM |
6055 | if (areas_overlap(src_off, dst_off, len)) |
6056 | must_memmove = 1; | |
3387206f | 6057 | } |
d1310b2e | 6058 | |
727011e0 CM |
6059 | if (must_memmove) |
6060 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
6061 | else | |
6062 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
6063 | } |
6064 | ||
6065 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
6066 | unsigned long src_offset, unsigned long len) | |
6067 | { | |
0b246afa | 6068 | struct btrfs_fs_info *fs_info = dst->fs_info; |
d1310b2e CM |
6069 | size_t cur; |
6070 | size_t dst_off_in_page; | |
6071 | size_t src_off_in_page; | |
7073017a | 6072 | size_t start_offset = offset_in_page(dst->start); |
d1310b2e CM |
6073 | unsigned long dst_i; |
6074 | unsigned long src_i; | |
6075 | ||
6076 | if (src_offset + len > dst->len) { | |
0b246afa | 6077 | btrfs_err(fs_info, |
5d163e0e JM |
6078 | "memmove bogus src_offset %lu move len %lu dst len %lu", |
6079 | src_offset, len, dst->len); | |
290342f6 | 6080 | BUG(); |
d1310b2e CM |
6081 | } |
6082 | if (dst_offset + len > dst->len) { | |
0b246afa | 6083 | btrfs_err(fs_info, |
5d163e0e JM |
6084 | "memmove bogus dst_offset %lu move len %lu dst len %lu", |
6085 | dst_offset, len, dst->len); | |
290342f6 | 6086 | BUG(); |
d1310b2e CM |
6087 | } |
6088 | ||
d397712b | 6089 | while (len > 0) { |
7073017a JT |
6090 | dst_off_in_page = offset_in_page(start_offset + dst_offset); |
6091 | src_off_in_page = offset_in_page(start_offset + src_offset); | |
d1310b2e | 6092 | |
09cbfeaf KS |
6093 | dst_i = (start_offset + dst_offset) >> PAGE_SHIFT; |
6094 | src_i = (start_offset + src_offset) >> PAGE_SHIFT; | |
d1310b2e | 6095 | |
09cbfeaf | 6096 | cur = min(len, (unsigned long)(PAGE_SIZE - |
d1310b2e CM |
6097 | src_off_in_page)); |
6098 | cur = min_t(unsigned long, cur, | |
09cbfeaf | 6099 | (unsigned long)(PAGE_SIZE - dst_off_in_page)); |
d1310b2e | 6100 | |
fb85fc9a | 6101 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
6102 | dst_off_in_page, src_off_in_page, cur); |
6103 | ||
6104 | src_offset += cur; | |
6105 | dst_offset += cur; | |
6106 | len -= cur; | |
6107 | } | |
6108 | } | |
d1310b2e CM |
6109 | |
6110 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
6111 | unsigned long src_offset, unsigned long len) | |
6112 | { | |
0b246afa | 6113 | struct btrfs_fs_info *fs_info = dst->fs_info; |
d1310b2e CM |
6114 | size_t cur; |
6115 | size_t dst_off_in_page; | |
6116 | size_t src_off_in_page; | |
6117 | unsigned long dst_end = dst_offset + len - 1; | |
6118 | unsigned long src_end = src_offset + len - 1; | |
7073017a | 6119 | size_t start_offset = offset_in_page(dst->start); |
d1310b2e CM |
6120 | unsigned long dst_i; |
6121 | unsigned long src_i; | |
6122 | ||
6123 | if (src_offset + len > dst->len) { | |
0b246afa | 6124 | btrfs_err(fs_info, |
5d163e0e JM |
6125 | "memmove bogus src_offset %lu move len %lu len %lu", |
6126 | src_offset, len, dst->len); | |
290342f6 | 6127 | BUG(); |
d1310b2e CM |
6128 | } |
6129 | if (dst_offset + len > dst->len) { | |
0b246afa | 6130 | btrfs_err(fs_info, |
5d163e0e JM |
6131 | "memmove bogus dst_offset %lu move len %lu len %lu", |
6132 | dst_offset, len, dst->len); | |
290342f6 | 6133 | BUG(); |
d1310b2e | 6134 | } |
727011e0 | 6135 | if (dst_offset < src_offset) { |
d1310b2e CM |
6136 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
6137 | return; | |
6138 | } | |
d397712b | 6139 | while (len > 0) { |
09cbfeaf KS |
6140 | dst_i = (start_offset + dst_end) >> PAGE_SHIFT; |
6141 | src_i = (start_offset + src_end) >> PAGE_SHIFT; | |
d1310b2e | 6142 | |
7073017a JT |
6143 | dst_off_in_page = offset_in_page(start_offset + dst_end); |
6144 | src_off_in_page = offset_in_page(start_offset + src_end); | |
d1310b2e CM |
6145 | |
6146 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
6147 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 6148 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
6149 | dst_off_in_page - cur + 1, |
6150 | src_off_in_page - cur + 1, cur); | |
6151 | ||
6152 | dst_end -= cur; | |
6153 | src_end -= cur; | |
6154 | len -= cur; | |
6155 | } | |
6156 | } | |
6af118ce | 6157 | |
f7a52a40 | 6158 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 6159 | { |
6af118ce | 6160 | struct extent_buffer *eb; |
6af118ce | 6161 | |
3083ee2e | 6162 | /* |
01327610 | 6163 | * We need to make sure nobody is attaching this page to an eb right |
3083ee2e JB |
6164 | * now. |
6165 | */ | |
6166 | spin_lock(&page->mapping->private_lock); | |
6167 | if (!PagePrivate(page)) { | |
6168 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 6169 | return 1; |
45f49bce | 6170 | } |
6af118ce | 6171 | |
3083ee2e JB |
6172 | eb = (struct extent_buffer *)page->private; |
6173 | BUG_ON(!eb); | |
19fe0a8b MX |
6174 | |
6175 | /* | |
3083ee2e JB |
6176 | * This is a little awful but should be ok, we need to make sure that |
6177 | * the eb doesn't disappear out from under us while we're looking at | |
6178 | * this page. | |
19fe0a8b | 6179 | */ |
3083ee2e | 6180 | spin_lock(&eb->refs_lock); |
0b32f4bb | 6181 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
6182 | spin_unlock(&eb->refs_lock); |
6183 | spin_unlock(&page->mapping->private_lock); | |
6184 | return 0; | |
b9473439 | 6185 | } |
3083ee2e | 6186 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 6187 | |
19fe0a8b | 6188 | /* |
3083ee2e JB |
6189 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
6190 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 6191 | */ |
3083ee2e JB |
6192 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
6193 | spin_unlock(&eb->refs_lock); | |
6194 | return 0; | |
b9473439 | 6195 | } |
19fe0a8b | 6196 | |
f7a52a40 | 6197 | return release_extent_buffer(eb); |
6af118ce | 6198 | } |