]>
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" |
760f991f | 27 | #include "subpage.h" |
d3575156 | 28 | #include "zoned.h" |
d1310b2e | 29 | |
d1310b2e CM |
30 | static struct kmem_cache *extent_state_cache; |
31 | static struct kmem_cache *extent_buffer_cache; | |
8ac9f7c1 | 32 | static struct bio_set btrfs_bioset; |
d1310b2e | 33 | |
27a3507d FM |
34 | static inline bool extent_state_in_tree(const struct extent_state *state) |
35 | { | |
36 | return !RB_EMPTY_NODE(&state->rb_node); | |
37 | } | |
38 | ||
6d49ba1b | 39 | #ifdef CONFIG_BTRFS_DEBUG |
d1310b2e | 40 | static LIST_HEAD(states); |
d397712b | 41 | static DEFINE_SPINLOCK(leak_lock); |
6d49ba1b | 42 | |
3fd63727 JB |
43 | static inline void btrfs_leak_debug_add(spinlock_t *lock, |
44 | struct list_head *new, | |
45 | struct list_head *head) | |
6d49ba1b ES |
46 | { |
47 | unsigned long flags; | |
48 | ||
3fd63727 | 49 | spin_lock_irqsave(lock, flags); |
6d49ba1b | 50 | list_add(new, head); |
3fd63727 | 51 | spin_unlock_irqrestore(lock, flags); |
6d49ba1b ES |
52 | } |
53 | ||
3fd63727 JB |
54 | static inline void btrfs_leak_debug_del(spinlock_t *lock, |
55 | struct list_head *entry) | |
6d49ba1b ES |
56 | { |
57 | unsigned long flags; | |
58 | ||
3fd63727 | 59 | spin_lock_irqsave(lock, flags); |
6d49ba1b | 60 | list_del(entry); |
3fd63727 | 61 | spin_unlock_irqrestore(lock, flags); |
6d49ba1b ES |
62 | } |
63 | ||
3fd63727 | 64 | void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info) |
6d49ba1b | 65 | { |
6d49ba1b | 66 | struct extent_buffer *eb; |
3fd63727 | 67 | unsigned long flags; |
6d49ba1b | 68 | |
8c38938c JB |
69 | /* |
70 | * If we didn't get into open_ctree our allocated_ebs will not be | |
71 | * initialized, so just skip this. | |
72 | */ | |
73 | if (!fs_info->allocated_ebs.next) | |
74 | return; | |
75 | ||
3fd63727 JB |
76 | spin_lock_irqsave(&fs_info->eb_leak_lock, flags); |
77 | while (!list_empty(&fs_info->allocated_ebs)) { | |
78 | eb = list_first_entry(&fs_info->allocated_ebs, | |
79 | struct extent_buffer, leak_list); | |
8c38938c JB |
80 | pr_err( |
81 | "BTRFS: buffer leak start %llu len %lu refs %d bflags %lu owner %llu\n", | |
82 | eb->start, eb->len, atomic_read(&eb->refs), eb->bflags, | |
83 | btrfs_header_owner(eb)); | |
33ca832f JB |
84 | list_del(&eb->leak_list); |
85 | kmem_cache_free(extent_buffer_cache, eb); | |
86 | } | |
3fd63727 | 87 | spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags); |
33ca832f JB |
88 | } |
89 | ||
90 | static inline void btrfs_extent_state_leak_debug_check(void) | |
91 | { | |
92 | struct extent_state *state; | |
93 | ||
6d49ba1b ES |
94 | while (!list_empty(&states)) { |
95 | state = list_entry(states.next, struct extent_state, leak_list); | |
9ee49a04 | 96 | pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n", |
27a3507d FM |
97 | state->start, state->end, state->state, |
98 | extent_state_in_tree(state), | |
b7ac31b7 | 99 | refcount_read(&state->refs)); |
6d49ba1b ES |
100 | list_del(&state->leak_list); |
101 | kmem_cache_free(extent_state_cache, state); | |
102 | } | |
6d49ba1b | 103 | } |
8d599ae1 | 104 | |
a5dee37d JB |
105 | #define btrfs_debug_check_extent_io_range(tree, start, end) \ |
106 | __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end)) | |
8d599ae1 | 107 | static inline void __btrfs_debug_check_extent_io_range(const char *caller, |
a5dee37d | 108 | struct extent_io_tree *tree, u64 start, u64 end) |
8d599ae1 | 109 | { |
65a680f6 NB |
110 | struct inode *inode = tree->private_data; |
111 | u64 isize; | |
112 | ||
113 | if (!inode || !is_data_inode(inode)) | |
114 | return; | |
115 | ||
116 | isize = i_size_read(inode); | |
117 | if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) { | |
118 | btrfs_debug_rl(BTRFS_I(inode)->root->fs_info, | |
119 | "%s: ino %llu isize %llu odd range [%llu,%llu]", | |
120 | caller, btrfs_ino(BTRFS_I(inode)), isize, start, end); | |
121 | } | |
8d599ae1 | 122 | } |
6d49ba1b | 123 | #else |
3fd63727 JB |
124 | #define btrfs_leak_debug_add(lock, new, head) do {} while (0) |
125 | #define btrfs_leak_debug_del(lock, entry) do {} while (0) | |
33ca832f | 126 | #define btrfs_extent_state_leak_debug_check() do {} while (0) |
8d599ae1 | 127 | #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0) |
4bef0848 | 128 | #endif |
d1310b2e | 129 | |
d1310b2e CM |
130 | struct tree_entry { |
131 | u64 start; | |
132 | u64 end; | |
d1310b2e CM |
133 | struct rb_node rb_node; |
134 | }; | |
135 | ||
136 | struct extent_page_data { | |
137 | struct bio *bio; | |
771ed689 CM |
138 | /* tells writepage not to lock the state bits for this range |
139 | * it still does the unlocking | |
140 | */ | |
ffbd517d CM |
141 | unsigned int extent_locked:1; |
142 | ||
70fd7614 | 143 | /* tells the submit_bio code to use REQ_SYNC */ |
ffbd517d | 144 | unsigned int sync_io:1; |
d1310b2e CM |
145 | }; |
146 | ||
f97e27e9 | 147 | static int add_extent_changeset(struct extent_state *state, u32 bits, |
d38ed27f QW |
148 | struct extent_changeset *changeset, |
149 | int set) | |
150 | { | |
151 | int ret; | |
152 | ||
153 | if (!changeset) | |
57599c7e | 154 | return 0; |
d38ed27f | 155 | if (set && (state->state & bits) == bits) |
57599c7e | 156 | return 0; |
fefdc557 | 157 | if (!set && (state->state & bits) == 0) |
57599c7e | 158 | return 0; |
d38ed27f | 159 | changeset->bytes_changed += state->end - state->start + 1; |
53d32359 | 160 | ret = ulist_add(&changeset->range_changed, state->start, state->end, |
d38ed27f | 161 | GFP_ATOMIC); |
57599c7e | 162 | return ret; |
d38ed27f QW |
163 | } |
164 | ||
c1be9c1a NB |
165 | int __must_check submit_one_bio(struct bio *bio, int mirror_num, |
166 | unsigned long bio_flags) | |
bb58eb9e QW |
167 | { |
168 | blk_status_t ret = 0; | |
bb58eb9e | 169 | struct extent_io_tree *tree = bio->bi_private; |
bb58eb9e QW |
170 | |
171 | bio->bi_private = NULL; | |
172 | ||
908930f3 NB |
173 | if (is_data_inode(tree->private_data)) |
174 | ret = btrfs_submit_data_bio(tree->private_data, bio, mirror_num, | |
175 | bio_flags); | |
176 | else | |
1b36294a NB |
177 | ret = btrfs_submit_metadata_bio(tree->private_data, bio, |
178 | mirror_num, bio_flags); | |
bb58eb9e QW |
179 | |
180 | return blk_status_to_errno(ret); | |
181 | } | |
182 | ||
3065976b QW |
183 | /* Cleanup unsubmitted bios */ |
184 | static void end_write_bio(struct extent_page_data *epd, int ret) | |
185 | { | |
186 | if (epd->bio) { | |
187 | epd->bio->bi_status = errno_to_blk_status(ret); | |
188 | bio_endio(epd->bio); | |
189 | epd->bio = NULL; | |
190 | } | |
191 | } | |
192 | ||
f4340622 QW |
193 | /* |
194 | * Submit bio from extent page data via submit_one_bio | |
195 | * | |
196 | * Return 0 if everything is OK. | |
197 | * Return <0 for error. | |
198 | */ | |
199 | static int __must_check flush_write_bio(struct extent_page_data *epd) | |
bb58eb9e | 200 | { |
f4340622 | 201 | int ret = 0; |
bb58eb9e | 202 | |
f4340622 | 203 | if (epd->bio) { |
bb58eb9e | 204 | ret = submit_one_bio(epd->bio, 0, 0); |
f4340622 QW |
205 | /* |
206 | * Clean up of epd->bio is handled by its endio function. | |
207 | * And endio is either triggered by successful bio execution | |
208 | * or the error handler of submit bio hook. | |
209 | * So at this point, no matter what happened, we don't need | |
210 | * to clean up epd->bio. | |
211 | */ | |
bb58eb9e QW |
212 | epd->bio = NULL; |
213 | } | |
f4340622 | 214 | return ret; |
bb58eb9e | 215 | } |
e2932ee0 | 216 | |
6f0d04f8 | 217 | int __init extent_state_cache_init(void) |
d1310b2e | 218 | { |
837e1972 | 219 | extent_state_cache = kmem_cache_create("btrfs_extent_state", |
9601e3f6 | 220 | sizeof(struct extent_state), 0, |
fba4b697 | 221 | SLAB_MEM_SPREAD, NULL); |
d1310b2e CM |
222 | if (!extent_state_cache) |
223 | return -ENOMEM; | |
6f0d04f8 JB |
224 | return 0; |
225 | } | |
d1310b2e | 226 | |
6f0d04f8 JB |
227 | int __init extent_io_init(void) |
228 | { | |
837e1972 | 229 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 | 230 | sizeof(struct extent_buffer), 0, |
fba4b697 | 231 | SLAB_MEM_SPREAD, NULL); |
d1310b2e | 232 | if (!extent_buffer_cache) |
6f0d04f8 | 233 | return -ENOMEM; |
9be3395b | 234 | |
8ac9f7c1 KO |
235 | if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE, |
236 | offsetof(struct btrfs_io_bio, bio), | |
237 | BIOSET_NEED_BVECS)) | |
9be3395b | 238 | goto free_buffer_cache; |
b208c2f7 | 239 | |
8ac9f7c1 | 240 | if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE)) |
b208c2f7 DW |
241 | goto free_bioset; |
242 | ||
d1310b2e CM |
243 | return 0; |
244 | ||
b208c2f7 | 245 | free_bioset: |
8ac9f7c1 | 246 | bioset_exit(&btrfs_bioset); |
b208c2f7 | 247 | |
9be3395b CM |
248 | free_buffer_cache: |
249 | kmem_cache_destroy(extent_buffer_cache); | |
250 | extent_buffer_cache = NULL; | |
6f0d04f8 JB |
251 | return -ENOMEM; |
252 | } | |
9be3395b | 253 | |
6f0d04f8 JB |
254 | void __cold extent_state_cache_exit(void) |
255 | { | |
256 | btrfs_extent_state_leak_debug_check(); | |
d1310b2e | 257 | kmem_cache_destroy(extent_state_cache); |
d1310b2e CM |
258 | } |
259 | ||
e67c718b | 260 | void __cold extent_io_exit(void) |
d1310b2e | 261 | { |
8c0a8537 KS |
262 | /* |
263 | * Make sure all delayed rcu free are flushed before we | |
264 | * destroy caches. | |
265 | */ | |
266 | rcu_barrier(); | |
5598e900 | 267 | kmem_cache_destroy(extent_buffer_cache); |
8ac9f7c1 | 268 | bioset_exit(&btrfs_bioset); |
d1310b2e CM |
269 | } |
270 | ||
41a2ee75 JB |
271 | /* |
272 | * For the file_extent_tree, we want to hold the inode lock when we lookup and | |
273 | * update the disk_i_size, but lockdep will complain because our io_tree we hold | |
274 | * the tree lock and get the inode lock when setting delalloc. These two things | |
275 | * are unrelated, so make a class for the file_extent_tree so we don't get the | |
276 | * two locking patterns mixed up. | |
277 | */ | |
278 | static struct lock_class_key file_extent_tree_class; | |
279 | ||
c258d6e3 | 280 | void extent_io_tree_init(struct btrfs_fs_info *fs_info, |
43eb5f29 QW |
281 | struct extent_io_tree *tree, unsigned int owner, |
282 | void *private_data) | |
d1310b2e | 283 | { |
c258d6e3 | 284 | tree->fs_info = fs_info; |
6bef4d31 | 285 | tree->state = RB_ROOT; |
d1310b2e | 286 | tree->dirty_bytes = 0; |
70dec807 | 287 | spin_lock_init(&tree->lock); |
c6100a4b | 288 | tree->private_data = private_data; |
43eb5f29 | 289 | tree->owner = owner; |
41a2ee75 JB |
290 | if (owner == IO_TREE_INODE_FILE_EXTENT) |
291 | lockdep_set_class(&tree->lock, &file_extent_tree_class); | |
d1310b2e | 292 | } |
d1310b2e | 293 | |
41e7acd3 NB |
294 | void extent_io_tree_release(struct extent_io_tree *tree) |
295 | { | |
296 | spin_lock(&tree->lock); | |
297 | /* | |
298 | * Do a single barrier for the waitqueue_active check here, the state | |
299 | * of the waitqueue should not change once extent_io_tree_release is | |
300 | * called. | |
301 | */ | |
302 | smp_mb(); | |
303 | while (!RB_EMPTY_ROOT(&tree->state)) { | |
304 | struct rb_node *node; | |
305 | struct extent_state *state; | |
306 | ||
307 | node = rb_first(&tree->state); | |
308 | state = rb_entry(node, struct extent_state, rb_node); | |
309 | rb_erase(&state->rb_node, &tree->state); | |
310 | RB_CLEAR_NODE(&state->rb_node); | |
311 | /* | |
312 | * btree io trees aren't supposed to have tasks waiting for | |
313 | * changes in the flags of extent states ever. | |
314 | */ | |
315 | ASSERT(!waitqueue_active(&state->wq)); | |
316 | free_extent_state(state); | |
317 | ||
318 | cond_resched_lock(&tree->lock); | |
319 | } | |
320 | spin_unlock(&tree->lock); | |
321 | } | |
322 | ||
b2950863 | 323 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
324 | { |
325 | struct extent_state *state; | |
d1310b2e | 326 | |
3ba7ab22 MH |
327 | /* |
328 | * The given mask might be not appropriate for the slab allocator, | |
329 | * drop the unsupported bits | |
330 | */ | |
331 | mask &= ~(__GFP_DMA32|__GFP_HIGHMEM); | |
d1310b2e | 332 | state = kmem_cache_alloc(extent_state_cache, mask); |
2b114d1d | 333 | if (!state) |
d1310b2e CM |
334 | return state; |
335 | state->state = 0; | |
47dc196a | 336 | state->failrec = NULL; |
27a3507d | 337 | RB_CLEAR_NODE(&state->rb_node); |
3fd63727 | 338 | btrfs_leak_debug_add(&leak_lock, &state->leak_list, &states); |
b7ac31b7 | 339 | refcount_set(&state->refs, 1); |
d1310b2e | 340 | init_waitqueue_head(&state->wq); |
143bede5 | 341 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
342 | return state; |
343 | } | |
d1310b2e | 344 | |
4845e44f | 345 | void free_extent_state(struct extent_state *state) |
d1310b2e | 346 | { |
d1310b2e CM |
347 | if (!state) |
348 | return; | |
b7ac31b7 | 349 | if (refcount_dec_and_test(&state->refs)) { |
27a3507d | 350 | WARN_ON(extent_state_in_tree(state)); |
3fd63727 | 351 | btrfs_leak_debug_del(&leak_lock, &state->leak_list); |
143bede5 | 352 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
353 | kmem_cache_free(extent_state_cache, state); |
354 | } | |
355 | } | |
d1310b2e | 356 | |
f2071b21 FM |
357 | static struct rb_node *tree_insert(struct rb_root *root, |
358 | struct rb_node *search_start, | |
359 | u64 offset, | |
12cfbad9 FDBM |
360 | struct rb_node *node, |
361 | struct rb_node ***p_in, | |
362 | struct rb_node **parent_in) | |
d1310b2e | 363 | { |
f2071b21 | 364 | struct rb_node **p; |
d397712b | 365 | struct rb_node *parent = NULL; |
d1310b2e CM |
366 | struct tree_entry *entry; |
367 | ||
12cfbad9 FDBM |
368 | if (p_in && parent_in) { |
369 | p = *p_in; | |
370 | parent = *parent_in; | |
371 | goto do_insert; | |
372 | } | |
373 | ||
f2071b21 | 374 | p = search_start ? &search_start : &root->rb_node; |
d397712b | 375 | while (*p) { |
d1310b2e CM |
376 | parent = *p; |
377 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
378 | ||
379 | if (offset < entry->start) | |
380 | p = &(*p)->rb_left; | |
381 | else if (offset > entry->end) | |
382 | p = &(*p)->rb_right; | |
383 | else | |
384 | return parent; | |
385 | } | |
386 | ||
12cfbad9 | 387 | do_insert: |
d1310b2e CM |
388 | rb_link_node(node, parent, p); |
389 | rb_insert_color(node, root); | |
390 | return NULL; | |
391 | } | |
392 | ||
8666e638 | 393 | /** |
3bed2da1 NB |
394 | * Search @tree for an entry that contains @offset. Such entry would have |
395 | * entry->start <= offset && entry->end >= offset. | |
8666e638 | 396 | * |
3bed2da1 NB |
397 | * @tree: the tree to search |
398 | * @offset: offset that should fall within an entry in @tree | |
399 | * @next_ret: pointer to the first entry whose range ends after @offset | |
400 | * @prev_ret: pointer to the first entry whose range begins before @offset | |
401 | * @p_ret: pointer where new node should be anchored (used when inserting an | |
402 | * entry in the tree) | |
403 | * @parent_ret: points to entry which would have been the parent of the entry, | |
8666e638 NB |
404 | * containing @offset |
405 | * | |
406 | * This function returns a pointer to the entry that contains @offset byte | |
407 | * address. If no such entry exists, then NULL is returned and the other | |
408 | * pointer arguments to the function are filled, otherwise the found entry is | |
409 | * returned and other pointers are left untouched. | |
410 | */ | |
80ea96b1 | 411 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
12cfbad9 | 412 | struct rb_node **next_ret, |
352646c7 | 413 | struct rb_node **prev_ret, |
12cfbad9 FDBM |
414 | struct rb_node ***p_ret, |
415 | struct rb_node **parent_ret) | |
d1310b2e | 416 | { |
80ea96b1 | 417 | struct rb_root *root = &tree->state; |
12cfbad9 | 418 | struct rb_node **n = &root->rb_node; |
d1310b2e CM |
419 | struct rb_node *prev = NULL; |
420 | struct rb_node *orig_prev = NULL; | |
421 | struct tree_entry *entry; | |
422 | struct tree_entry *prev_entry = NULL; | |
423 | ||
12cfbad9 FDBM |
424 | while (*n) { |
425 | prev = *n; | |
426 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
d1310b2e CM |
427 | prev_entry = entry; |
428 | ||
429 | if (offset < entry->start) | |
12cfbad9 | 430 | n = &(*n)->rb_left; |
d1310b2e | 431 | else if (offset > entry->end) |
12cfbad9 | 432 | n = &(*n)->rb_right; |
d397712b | 433 | else |
12cfbad9 | 434 | return *n; |
d1310b2e CM |
435 | } |
436 | ||
12cfbad9 FDBM |
437 | if (p_ret) |
438 | *p_ret = n; | |
439 | if (parent_ret) | |
440 | *parent_ret = prev; | |
441 | ||
352646c7 | 442 | if (next_ret) { |
d1310b2e | 443 | orig_prev = prev; |
d397712b | 444 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
445 | prev = rb_next(prev); |
446 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
447 | } | |
352646c7 | 448 | *next_ret = prev; |
d1310b2e CM |
449 | prev = orig_prev; |
450 | } | |
451 | ||
352646c7 | 452 | if (prev_ret) { |
d1310b2e | 453 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); |
d397712b | 454 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
455 | prev = rb_prev(prev); |
456 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
457 | } | |
352646c7 | 458 | *prev_ret = prev; |
d1310b2e CM |
459 | } |
460 | return NULL; | |
461 | } | |
462 | ||
12cfbad9 FDBM |
463 | static inline struct rb_node * |
464 | tree_search_for_insert(struct extent_io_tree *tree, | |
465 | u64 offset, | |
466 | struct rb_node ***p_ret, | |
467 | struct rb_node **parent_ret) | |
d1310b2e | 468 | { |
352646c7 | 469 | struct rb_node *next= NULL; |
d1310b2e | 470 | struct rb_node *ret; |
70dec807 | 471 | |
352646c7 | 472 | ret = __etree_search(tree, offset, &next, NULL, p_ret, parent_ret); |
d397712b | 473 | if (!ret) |
352646c7 | 474 | return next; |
d1310b2e CM |
475 | return ret; |
476 | } | |
477 | ||
12cfbad9 FDBM |
478 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
479 | u64 offset) | |
480 | { | |
481 | return tree_search_for_insert(tree, offset, NULL, NULL); | |
482 | } | |
483 | ||
d1310b2e CM |
484 | /* |
485 | * utility function to look for merge candidates inside a given range. | |
486 | * Any extents with matching state are merged together into a single | |
487 | * extent in the tree. Extents with EXTENT_IO in their state field | |
488 | * are not merged because the end_io handlers need to be able to do | |
489 | * operations on them without sleeping (or doing allocations/splits). | |
490 | * | |
491 | * This should be called with the tree lock held. | |
492 | */ | |
1bf85046 JM |
493 | static void merge_state(struct extent_io_tree *tree, |
494 | struct extent_state *state) | |
d1310b2e CM |
495 | { |
496 | struct extent_state *other; | |
497 | struct rb_node *other_node; | |
498 | ||
8882679e | 499 | if (state->state & (EXTENT_LOCKED | EXTENT_BOUNDARY)) |
1bf85046 | 500 | return; |
d1310b2e CM |
501 | |
502 | other_node = rb_prev(&state->rb_node); | |
503 | if (other_node) { | |
504 | other = rb_entry(other_node, struct extent_state, rb_node); | |
505 | if (other->end == state->start - 1 && | |
506 | other->state == state->state) { | |
5c848198 NB |
507 | if (tree->private_data && |
508 | is_data_inode(tree->private_data)) | |
509 | btrfs_merge_delalloc_extent(tree->private_data, | |
510 | state, other); | |
d1310b2e | 511 | state->start = other->start; |
d1310b2e | 512 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 513 | RB_CLEAR_NODE(&other->rb_node); |
d1310b2e CM |
514 | free_extent_state(other); |
515 | } | |
516 | } | |
517 | other_node = rb_next(&state->rb_node); | |
518 | if (other_node) { | |
519 | other = rb_entry(other_node, struct extent_state, rb_node); | |
520 | if (other->start == state->end + 1 && | |
521 | other->state == state->state) { | |
5c848198 NB |
522 | if (tree->private_data && |
523 | is_data_inode(tree->private_data)) | |
524 | btrfs_merge_delalloc_extent(tree->private_data, | |
525 | state, other); | |
df98b6e2 | 526 | state->end = other->end; |
df98b6e2 | 527 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 528 | RB_CLEAR_NODE(&other->rb_node); |
df98b6e2 | 529 | free_extent_state(other); |
d1310b2e CM |
530 | } |
531 | } | |
d1310b2e CM |
532 | } |
533 | ||
3150b699 | 534 | static void set_state_bits(struct extent_io_tree *tree, |
f97e27e9 | 535 | struct extent_state *state, u32 *bits, |
d38ed27f | 536 | struct extent_changeset *changeset); |
3150b699 | 537 | |
d1310b2e CM |
538 | /* |
539 | * insert an extent_state struct into the tree. 'bits' are set on the | |
540 | * struct before it is inserted. | |
541 | * | |
542 | * This may return -EEXIST if the extent is already there, in which case the | |
543 | * state struct is freed. | |
544 | * | |
545 | * The tree lock is not taken internally. This is a utility function and | |
546 | * probably isn't what you want to call (see set/clear_extent_bit). | |
547 | */ | |
548 | static int insert_state(struct extent_io_tree *tree, | |
549 | struct extent_state *state, u64 start, u64 end, | |
12cfbad9 FDBM |
550 | struct rb_node ***p, |
551 | struct rb_node **parent, | |
f97e27e9 | 552 | u32 *bits, struct extent_changeset *changeset) |
d1310b2e CM |
553 | { |
554 | struct rb_node *node; | |
555 | ||
2792237d DS |
556 | if (end < start) { |
557 | btrfs_err(tree->fs_info, | |
558 | "insert state: end < start %llu %llu", end, start); | |
559 | WARN_ON(1); | |
560 | } | |
d1310b2e CM |
561 | state->start = start; |
562 | state->end = end; | |
9ed74f2d | 563 | |
d38ed27f | 564 | set_state_bits(tree, state, bits, changeset); |
3150b699 | 565 | |
f2071b21 | 566 | node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent); |
d1310b2e CM |
567 | if (node) { |
568 | struct extent_state *found; | |
569 | found = rb_entry(node, struct extent_state, rb_node); | |
2792237d DS |
570 | btrfs_err(tree->fs_info, |
571 | "found node %llu %llu on insert of %llu %llu", | |
c1c9ff7c | 572 | found->start, found->end, start, end); |
d1310b2e CM |
573 | return -EEXIST; |
574 | } | |
575 | merge_state(tree, state); | |
576 | return 0; | |
577 | } | |
578 | ||
579 | /* | |
580 | * split a given extent state struct in two, inserting the preallocated | |
581 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
582 | * offset inside 'orig' where it should be split. | |
583 | * | |
584 | * Before calling, | |
585 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
586 | * are two extent state structs in the tree: | |
587 | * prealloc: [orig->start, split - 1] | |
588 | * orig: [ split, orig->end ] | |
589 | * | |
590 | * The tree locks are not taken by this function. They need to be held | |
591 | * by the caller. | |
592 | */ | |
593 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
594 | struct extent_state *prealloc, u64 split) | |
595 | { | |
596 | struct rb_node *node; | |
9ed74f2d | 597 | |
abbb55f4 NB |
598 | if (tree->private_data && is_data_inode(tree->private_data)) |
599 | btrfs_split_delalloc_extent(tree->private_data, orig, split); | |
9ed74f2d | 600 | |
d1310b2e CM |
601 | prealloc->start = orig->start; |
602 | prealloc->end = split - 1; | |
603 | prealloc->state = orig->state; | |
604 | orig->start = split; | |
605 | ||
f2071b21 FM |
606 | node = tree_insert(&tree->state, &orig->rb_node, prealloc->end, |
607 | &prealloc->rb_node, NULL, NULL); | |
d1310b2e | 608 | if (node) { |
d1310b2e CM |
609 | free_extent_state(prealloc); |
610 | return -EEXIST; | |
611 | } | |
612 | return 0; | |
613 | } | |
614 | ||
cdc6a395 LZ |
615 | static struct extent_state *next_state(struct extent_state *state) |
616 | { | |
617 | struct rb_node *next = rb_next(&state->rb_node); | |
618 | if (next) | |
619 | return rb_entry(next, struct extent_state, rb_node); | |
620 | else | |
621 | return NULL; | |
622 | } | |
623 | ||
d1310b2e CM |
624 | /* |
625 | * utility function to clear some bits in an extent state struct. | |
52042d8e | 626 | * it will optionally wake up anyone waiting on this state (wake == 1). |
d1310b2e CM |
627 | * |
628 | * If no bits are set on the state struct after clearing things, the | |
629 | * struct is freed and removed from the tree | |
630 | */ | |
cdc6a395 LZ |
631 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
632 | struct extent_state *state, | |
f97e27e9 | 633 | u32 *bits, int wake, |
fefdc557 | 634 | struct extent_changeset *changeset) |
d1310b2e | 635 | { |
cdc6a395 | 636 | struct extent_state *next; |
f97e27e9 | 637 | u32 bits_to_clear = *bits & ~EXTENT_CTLBITS; |
57599c7e | 638 | int ret; |
d1310b2e | 639 | |
0ca1f7ce | 640 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
641 | u64 range = state->end - state->start + 1; |
642 | WARN_ON(range > tree->dirty_bytes); | |
643 | tree->dirty_bytes -= range; | |
644 | } | |
a36bb5f9 NB |
645 | |
646 | if (tree->private_data && is_data_inode(tree->private_data)) | |
647 | btrfs_clear_delalloc_extent(tree->private_data, state, bits); | |
648 | ||
57599c7e DS |
649 | ret = add_extent_changeset(state, bits_to_clear, changeset, 0); |
650 | BUG_ON(ret < 0); | |
32c00aff | 651 | state->state &= ~bits_to_clear; |
d1310b2e CM |
652 | if (wake) |
653 | wake_up(&state->wq); | |
0ca1f7ce | 654 | if (state->state == 0) { |
cdc6a395 | 655 | next = next_state(state); |
27a3507d | 656 | if (extent_state_in_tree(state)) { |
d1310b2e | 657 | rb_erase(&state->rb_node, &tree->state); |
27a3507d | 658 | RB_CLEAR_NODE(&state->rb_node); |
d1310b2e CM |
659 | free_extent_state(state); |
660 | } else { | |
661 | WARN_ON(1); | |
662 | } | |
663 | } else { | |
664 | merge_state(tree, state); | |
cdc6a395 | 665 | next = next_state(state); |
d1310b2e | 666 | } |
cdc6a395 | 667 | return next; |
d1310b2e CM |
668 | } |
669 | ||
8233767a XG |
670 | static struct extent_state * |
671 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
672 | { | |
673 | if (!prealloc) | |
674 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
675 | ||
676 | return prealloc; | |
677 | } | |
678 | ||
48a3b636 | 679 | static void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
c2d904e0 | 680 | { |
29b665cc | 681 | btrfs_panic(tree->fs_info, err, |
05912a3c | 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, |
f97e27e9 QW |
698 | u32 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 | 870 | static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 | 871 | u32 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, |
f97e27e9 | 918 | u32 *bits, struct extent_changeset *changeset) |
d1310b2e | 919 | { |
f97e27e9 | 920 | u32 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 | */ |
f97e27e9 QW |
964 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, u32 bits, |
965 | u32 exclusive_bits, u64 *failed_start, | |
1cab5e72 NB |
966 | struct extent_state **cached_state, gfp_t mask, |
967 | struct extent_changeset *changeset) | |
d1310b2e CM |
968 | { |
969 | struct extent_state *state; | |
970 | struct extent_state *prealloc = NULL; | |
971 | struct rb_node *node; | |
12cfbad9 FDBM |
972 | struct rb_node **p; |
973 | struct rb_node *parent; | |
d1310b2e | 974 | int err = 0; |
d1310b2e CM |
975 | u64 last_start; |
976 | u64 last_end; | |
42daec29 | 977 | |
a5dee37d | 978 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 | 979 | trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits); |
8d599ae1 | 980 | |
3f6bb4ae QW |
981 | if (exclusive_bits) |
982 | ASSERT(failed_start); | |
983 | else | |
984 | ASSERT(failed_start == NULL); | |
d1310b2e | 985 | again: |
d0164adc | 986 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
059f791c DS |
987 | /* |
988 | * Don't care for allocation failure here because we might end | |
989 | * up not needing the pre-allocated extent state at all, which | |
990 | * is the case if we only have in the tree extent states that | |
991 | * cover our input range and don't cover too any other range. | |
992 | * If we end up needing a new extent state we allocate it later. | |
993 | */ | |
d1310b2e | 994 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
995 | } |
996 | ||
cad321ad | 997 | spin_lock(&tree->lock); |
9655d298 CM |
998 | if (cached_state && *cached_state) { |
999 | state = *cached_state; | |
df98b6e2 | 1000 | if (state->start <= start && state->end > start && |
27a3507d | 1001 | extent_state_in_tree(state)) { |
9655d298 CM |
1002 | node = &state->rb_node; |
1003 | goto hit_next; | |
1004 | } | |
1005 | } | |
d1310b2e CM |
1006 | /* |
1007 | * this search will find all the extents that end after | |
1008 | * our range starts. | |
1009 | */ | |
12cfbad9 | 1010 | node = tree_search_for_insert(tree, start, &p, &parent); |
d1310b2e | 1011 | if (!node) { |
8233767a XG |
1012 | prealloc = alloc_extent_state_atomic(prealloc); |
1013 | BUG_ON(!prealloc); | |
12cfbad9 | 1014 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 1015 | &p, &parent, &bits, changeset); |
c2d904e0 JM |
1016 | if (err) |
1017 | extent_io_tree_panic(tree, err); | |
1018 | ||
c42ac0bc | 1019 | cache_state(prealloc, cached_state); |
d1310b2e | 1020 | prealloc = NULL; |
d1310b2e CM |
1021 | goto out; |
1022 | } | |
d1310b2e | 1023 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 1024 | hit_next: |
d1310b2e CM |
1025 | last_start = state->start; |
1026 | last_end = state->end; | |
1027 | ||
1028 | /* | |
1029 | * | ---- desired range ---- | | |
1030 | * | state | | |
1031 | * | |
1032 | * Just lock what we found and keep going | |
1033 | */ | |
1034 | if (state->start == start && state->end <= end) { | |
1edbb734 | 1035 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1036 | *failed_start = state->start; |
1037 | err = -EEXIST; | |
1038 | goto out; | |
1039 | } | |
42daec29 | 1040 | |
d38ed27f | 1041 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 1042 | cache_state(state, cached_state); |
d1310b2e | 1043 | merge_state(tree, state); |
5c939df5 YZ |
1044 | if (last_end == (u64)-1) |
1045 | goto out; | |
1046 | start = last_end + 1; | |
d1ac6e41 LB |
1047 | state = next_state(state); |
1048 | if (start < end && state && state->start == start && | |
1049 | !need_resched()) | |
1050 | goto hit_next; | |
d1310b2e CM |
1051 | goto search_again; |
1052 | } | |
1053 | ||
1054 | /* | |
1055 | * | ---- desired range ---- | | |
1056 | * | state | | |
1057 | * or | |
1058 | * | ------------- state -------------- | | |
1059 | * | |
1060 | * We need to split the extent we found, and may flip bits on | |
1061 | * second half. | |
1062 | * | |
1063 | * If the extent we found extends past our | |
1064 | * range, we just split and search again. It'll get split | |
1065 | * again the next time though. | |
1066 | * | |
1067 | * If the extent we found is inside our range, we set the | |
1068 | * desired bit on it. | |
1069 | */ | |
1070 | if (state->start < start) { | |
1edbb734 | 1071 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1072 | *failed_start = start; |
1073 | err = -EEXIST; | |
1074 | goto out; | |
1075 | } | |
8233767a | 1076 | |
55ffaabe FM |
1077 | /* |
1078 | * If this extent already has all the bits we want set, then | |
1079 | * skip it, not necessary to split it or do anything with it. | |
1080 | */ | |
1081 | if ((state->state & bits) == bits) { | |
1082 | start = state->end + 1; | |
1083 | cache_state(state, cached_state); | |
1084 | goto search_again; | |
1085 | } | |
1086 | ||
8233767a XG |
1087 | prealloc = alloc_extent_state_atomic(prealloc); |
1088 | BUG_ON(!prealloc); | |
d1310b2e | 1089 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1090 | if (err) |
1091 | extent_io_tree_panic(tree, err); | |
1092 | ||
d1310b2e CM |
1093 | prealloc = NULL; |
1094 | if (err) | |
1095 | goto out; | |
1096 | if (state->end <= end) { | |
d38ed27f | 1097 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 1098 | cache_state(state, cached_state); |
d1310b2e | 1099 | merge_state(tree, state); |
5c939df5 YZ |
1100 | if (last_end == (u64)-1) |
1101 | goto out; | |
1102 | start = last_end + 1; | |
d1ac6e41 LB |
1103 | state = next_state(state); |
1104 | if (start < end && state && state->start == start && | |
1105 | !need_resched()) | |
1106 | goto hit_next; | |
d1310b2e CM |
1107 | } |
1108 | goto search_again; | |
1109 | } | |
1110 | /* | |
1111 | * | ---- desired range ---- | | |
1112 | * | state | or | state | | |
1113 | * | |
1114 | * There's a hole, we need to insert something in it and | |
1115 | * ignore the extent we found. | |
1116 | */ | |
1117 | if (state->start > start) { | |
1118 | u64 this_end; | |
1119 | if (end < last_start) | |
1120 | this_end = end; | |
1121 | else | |
d397712b | 1122 | this_end = last_start - 1; |
8233767a XG |
1123 | |
1124 | prealloc = alloc_extent_state_atomic(prealloc); | |
1125 | BUG_ON(!prealloc); | |
c7f895a2 XG |
1126 | |
1127 | /* | |
1128 | * Avoid to free 'prealloc' if it can be merged with | |
1129 | * the later extent. | |
1130 | */ | |
d1310b2e | 1131 | err = insert_state(tree, prealloc, start, this_end, |
d38ed27f | 1132 | NULL, NULL, &bits, changeset); |
c2d904e0 JM |
1133 | if (err) |
1134 | extent_io_tree_panic(tree, err); | |
1135 | ||
9ed74f2d JB |
1136 | cache_state(prealloc, cached_state); |
1137 | prealloc = NULL; | |
d1310b2e CM |
1138 | start = this_end + 1; |
1139 | goto search_again; | |
1140 | } | |
1141 | /* | |
1142 | * | ---- desired range ---- | | |
1143 | * | state | | |
1144 | * We need to split the extent, and set the bit | |
1145 | * on the first half | |
1146 | */ | |
1147 | if (state->start <= end && state->end > end) { | |
1edbb734 | 1148 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1149 | *failed_start = start; |
1150 | err = -EEXIST; | |
1151 | goto out; | |
1152 | } | |
8233767a XG |
1153 | |
1154 | prealloc = alloc_extent_state_atomic(prealloc); | |
1155 | BUG_ON(!prealloc); | |
d1310b2e | 1156 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
1157 | if (err) |
1158 | extent_io_tree_panic(tree, err); | |
d1310b2e | 1159 | |
d38ed27f | 1160 | set_state_bits(tree, prealloc, &bits, changeset); |
2c64c53d | 1161 | cache_state(prealloc, cached_state); |
d1310b2e CM |
1162 | merge_state(tree, prealloc); |
1163 | prealloc = NULL; | |
1164 | goto out; | |
1165 | } | |
1166 | ||
b5a4ba14 DS |
1167 | search_again: |
1168 | if (start > end) | |
1169 | goto out; | |
1170 | spin_unlock(&tree->lock); | |
1171 | if (gfpflags_allow_blocking(mask)) | |
1172 | cond_resched(); | |
1173 | goto again; | |
d1310b2e CM |
1174 | |
1175 | out: | |
cad321ad | 1176 | spin_unlock(&tree->lock); |
d1310b2e CM |
1177 | if (prealloc) |
1178 | free_extent_state(prealloc); | |
1179 | ||
1180 | return err; | |
1181 | ||
d1310b2e | 1182 | } |
d1310b2e | 1183 | |
462d6fac | 1184 | /** |
10983f2e LB |
1185 | * convert_extent_bit - convert all bits in a given range from one bit to |
1186 | * another | |
462d6fac JB |
1187 | * @tree: the io tree to search |
1188 | * @start: the start offset in bytes | |
1189 | * @end: the end offset in bytes (inclusive) | |
1190 | * @bits: the bits to set in this range | |
1191 | * @clear_bits: the bits to clear in this range | |
e6138876 | 1192 | * @cached_state: state that we're going to cache |
462d6fac JB |
1193 | * |
1194 | * This will go through and set bits for the given range. If any states exist | |
1195 | * already in this range they are set with the given bit and cleared of the | |
1196 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
1197 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
1198 | * boundary bits like LOCK. | |
210aa277 DS |
1199 | * |
1200 | * All allocations are done with GFP_NOFS. | |
462d6fac JB |
1201 | */ |
1202 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
f97e27e9 | 1203 | u32 bits, u32 clear_bits, |
210aa277 | 1204 | struct extent_state **cached_state) |
462d6fac JB |
1205 | { |
1206 | struct extent_state *state; | |
1207 | struct extent_state *prealloc = NULL; | |
1208 | struct rb_node *node; | |
12cfbad9 FDBM |
1209 | struct rb_node **p; |
1210 | struct rb_node *parent; | |
462d6fac JB |
1211 | int err = 0; |
1212 | u64 last_start; | |
1213 | u64 last_end; | |
c8fd3de7 | 1214 | bool first_iteration = true; |
462d6fac | 1215 | |
a5dee37d | 1216 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 QW |
1217 | trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits, |
1218 | clear_bits); | |
8d599ae1 | 1219 | |
462d6fac | 1220 | again: |
210aa277 | 1221 | if (!prealloc) { |
c8fd3de7 FM |
1222 | /* |
1223 | * Best effort, don't worry if extent state allocation fails | |
1224 | * here for the first iteration. We might have a cached state | |
1225 | * that matches exactly the target range, in which case no | |
1226 | * extent state allocations are needed. We'll only know this | |
1227 | * after locking the tree. | |
1228 | */ | |
210aa277 | 1229 | prealloc = alloc_extent_state(GFP_NOFS); |
c8fd3de7 | 1230 | if (!prealloc && !first_iteration) |
462d6fac JB |
1231 | return -ENOMEM; |
1232 | } | |
1233 | ||
1234 | spin_lock(&tree->lock); | |
e6138876 JB |
1235 | if (cached_state && *cached_state) { |
1236 | state = *cached_state; | |
1237 | if (state->start <= start && state->end > start && | |
27a3507d | 1238 | extent_state_in_tree(state)) { |
e6138876 JB |
1239 | node = &state->rb_node; |
1240 | goto hit_next; | |
1241 | } | |
1242 | } | |
1243 | ||
462d6fac JB |
1244 | /* |
1245 | * this search will find all the extents that end after | |
1246 | * our range starts. | |
1247 | */ | |
12cfbad9 | 1248 | node = tree_search_for_insert(tree, start, &p, &parent); |
462d6fac JB |
1249 | if (!node) { |
1250 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1251 | if (!prealloc) { |
1252 | err = -ENOMEM; | |
1253 | goto out; | |
1254 | } | |
12cfbad9 | 1255 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 1256 | &p, &parent, &bits, NULL); |
c2d904e0 JM |
1257 | if (err) |
1258 | extent_io_tree_panic(tree, err); | |
c42ac0bc FDBM |
1259 | cache_state(prealloc, cached_state); |
1260 | prealloc = NULL; | |
462d6fac JB |
1261 | goto out; |
1262 | } | |
1263 | state = rb_entry(node, struct extent_state, rb_node); | |
1264 | hit_next: | |
1265 | last_start = state->start; | |
1266 | last_end = state->end; | |
1267 | ||
1268 | /* | |
1269 | * | ---- desired range ---- | | |
1270 | * | state | | |
1271 | * | |
1272 | * Just lock what we found and keep going | |
1273 | */ | |
1274 | if (state->start == start && state->end <= end) { | |
d38ed27f | 1275 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1276 | cache_state(state, cached_state); |
fefdc557 | 1277 | state = clear_state_bit(tree, state, &clear_bits, 0, NULL); |
462d6fac JB |
1278 | if (last_end == (u64)-1) |
1279 | goto out; | |
462d6fac | 1280 | start = last_end + 1; |
d1ac6e41 LB |
1281 | if (start < end && state && state->start == start && |
1282 | !need_resched()) | |
1283 | goto hit_next; | |
462d6fac JB |
1284 | goto search_again; |
1285 | } | |
1286 | ||
1287 | /* | |
1288 | * | ---- desired range ---- | | |
1289 | * | state | | |
1290 | * or | |
1291 | * | ------------- state -------------- | | |
1292 | * | |
1293 | * We need to split the extent we found, and may flip bits on | |
1294 | * second half. | |
1295 | * | |
1296 | * If the extent we found extends past our | |
1297 | * range, we just split and search again. It'll get split | |
1298 | * again the next time though. | |
1299 | * | |
1300 | * If the extent we found is inside our range, we set the | |
1301 | * desired bit on it. | |
1302 | */ | |
1303 | if (state->start < start) { | |
1304 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1305 | if (!prealloc) { |
1306 | err = -ENOMEM; | |
1307 | goto out; | |
1308 | } | |
462d6fac | 1309 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1310 | if (err) |
1311 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1312 | prealloc = NULL; |
1313 | if (err) | |
1314 | goto out; | |
1315 | if (state->end <= end) { | |
d38ed27f | 1316 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1317 | cache_state(state, cached_state); |
fefdc557 QW |
1318 | state = clear_state_bit(tree, state, &clear_bits, 0, |
1319 | NULL); | |
462d6fac JB |
1320 | if (last_end == (u64)-1) |
1321 | goto out; | |
1322 | start = last_end + 1; | |
d1ac6e41 LB |
1323 | if (start < end && state && state->start == start && |
1324 | !need_resched()) | |
1325 | goto hit_next; | |
462d6fac JB |
1326 | } |
1327 | goto search_again; | |
1328 | } | |
1329 | /* | |
1330 | * | ---- desired range ---- | | |
1331 | * | state | or | state | | |
1332 | * | |
1333 | * There's a hole, we need to insert something in it and | |
1334 | * ignore the extent we found. | |
1335 | */ | |
1336 | if (state->start > start) { | |
1337 | u64 this_end; | |
1338 | if (end < last_start) | |
1339 | this_end = end; | |
1340 | else | |
1341 | this_end = last_start - 1; | |
1342 | ||
1343 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1344 | if (!prealloc) { |
1345 | err = -ENOMEM; | |
1346 | goto out; | |
1347 | } | |
462d6fac JB |
1348 | |
1349 | /* | |
1350 | * Avoid to free 'prealloc' if it can be merged with | |
1351 | * the later extent. | |
1352 | */ | |
1353 | err = insert_state(tree, prealloc, start, this_end, | |
d38ed27f | 1354 | NULL, NULL, &bits, NULL); |
c2d904e0 JM |
1355 | if (err) |
1356 | extent_io_tree_panic(tree, err); | |
e6138876 | 1357 | cache_state(prealloc, cached_state); |
462d6fac JB |
1358 | prealloc = NULL; |
1359 | start = this_end + 1; | |
1360 | goto search_again; | |
1361 | } | |
1362 | /* | |
1363 | * | ---- desired range ---- | | |
1364 | * | state | | |
1365 | * We need to split the extent, and set the bit | |
1366 | * on the first half | |
1367 | */ | |
1368 | if (state->start <= end && state->end > end) { | |
1369 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1370 | if (!prealloc) { |
1371 | err = -ENOMEM; | |
1372 | goto out; | |
1373 | } | |
462d6fac JB |
1374 | |
1375 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1376 | if (err) |
1377 | extent_io_tree_panic(tree, err); | |
462d6fac | 1378 | |
d38ed27f | 1379 | set_state_bits(tree, prealloc, &bits, NULL); |
e6138876 | 1380 | cache_state(prealloc, cached_state); |
fefdc557 | 1381 | clear_state_bit(tree, prealloc, &clear_bits, 0, NULL); |
462d6fac JB |
1382 | prealloc = NULL; |
1383 | goto out; | |
1384 | } | |
1385 | ||
462d6fac JB |
1386 | search_again: |
1387 | if (start > end) | |
1388 | goto out; | |
1389 | spin_unlock(&tree->lock); | |
210aa277 | 1390 | cond_resched(); |
c8fd3de7 | 1391 | first_iteration = false; |
462d6fac | 1392 | goto again; |
462d6fac JB |
1393 | |
1394 | out: | |
1395 | spin_unlock(&tree->lock); | |
1396 | if (prealloc) | |
1397 | free_extent_state(prealloc); | |
1398 | ||
1399 | return err; | |
462d6fac JB |
1400 | } |
1401 | ||
d1310b2e | 1402 | /* wrappers around set/clear extent bit */ |
d38ed27f | 1403 | int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 | 1404 | u32 bits, struct extent_changeset *changeset) |
d38ed27f QW |
1405 | { |
1406 | /* | |
1407 | * We don't support EXTENT_LOCKED yet, as current changeset will | |
1408 | * record any bits changed, so for EXTENT_LOCKED case, it will | |
1409 | * either fail with -EEXIST or changeset will record the whole | |
1410 | * range. | |
1411 | */ | |
1412 | BUG_ON(bits & EXTENT_LOCKED); | |
1413 | ||
1cab5e72 NB |
1414 | return set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS, |
1415 | changeset); | |
d38ed27f QW |
1416 | } |
1417 | ||
4ca73656 | 1418 | int set_extent_bits_nowait(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 | 1419 | u32 bits) |
4ca73656 | 1420 | { |
1cab5e72 NB |
1421 | return set_extent_bit(tree, start, end, bits, 0, NULL, NULL, |
1422 | GFP_NOWAIT, NULL); | |
4ca73656 NB |
1423 | } |
1424 | ||
fefdc557 | 1425 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 | 1426 | u32 bits, int wake, int delete, |
ae0f1625 | 1427 | struct extent_state **cached) |
fefdc557 QW |
1428 | { |
1429 | return __clear_extent_bit(tree, start, end, bits, wake, delete, | |
ae0f1625 | 1430 | cached, GFP_NOFS, NULL); |
fefdc557 QW |
1431 | } |
1432 | ||
fefdc557 | 1433 | int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 | 1434 | u32 bits, struct extent_changeset *changeset) |
fefdc557 QW |
1435 | { |
1436 | /* | |
1437 | * Don't support EXTENT_LOCKED case, same reason as | |
1438 | * set_record_extent_bits(). | |
1439 | */ | |
1440 | BUG_ON(bits & EXTENT_LOCKED); | |
1441 | ||
f734c44a | 1442 | return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS, |
fefdc557 QW |
1443 | changeset); |
1444 | } | |
1445 | ||
d352ac68 CM |
1446 | /* |
1447 | * either insert or lock state struct between start and end use mask to tell | |
1448 | * us if waiting is desired. | |
1449 | */ | |
1edbb734 | 1450 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
ff13db41 | 1451 | struct extent_state **cached_state) |
d1310b2e CM |
1452 | { |
1453 | int err; | |
1454 | u64 failed_start; | |
9ee49a04 | 1455 | |
d1310b2e | 1456 | while (1) { |
1cab5e72 NB |
1457 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED, |
1458 | EXTENT_LOCKED, &failed_start, | |
1459 | cached_state, GFP_NOFS, NULL); | |
d0082371 | 1460 | if (err == -EEXIST) { |
d1310b2e CM |
1461 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1462 | start = failed_start; | |
d0082371 | 1463 | } else |
d1310b2e | 1464 | break; |
d1310b2e CM |
1465 | WARN_ON(start > end); |
1466 | } | |
1467 | return err; | |
1468 | } | |
d1310b2e | 1469 | |
d0082371 | 1470 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1471 | { |
1472 | int err; | |
1473 | u64 failed_start; | |
1474 | ||
1cab5e72 NB |
1475 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
1476 | &failed_start, NULL, GFP_NOFS, NULL); | |
6643558d YZ |
1477 | if (err == -EEXIST) { |
1478 | if (failed_start > start) | |
1479 | clear_extent_bit(tree, start, failed_start - 1, | |
ae0f1625 | 1480 | EXTENT_LOCKED, 1, 0, NULL); |
25179201 | 1481 | return 0; |
6643558d | 1482 | } |
25179201 JB |
1483 | return 1; |
1484 | } | |
25179201 | 1485 | |
bd1fa4f0 | 1486 | void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1487 | { |
09cbfeaf KS |
1488 | unsigned long index = start >> PAGE_SHIFT; |
1489 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1490 | struct page *page; |
1491 | ||
1492 | while (index <= end_index) { | |
1493 | page = find_get_page(inode->i_mapping, index); | |
1494 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1495 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1496 | put_page(page); |
4adaa611 CM |
1497 | index++; |
1498 | } | |
4adaa611 CM |
1499 | } |
1500 | ||
f6311572 | 1501 | void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1502 | { |
09cbfeaf KS |
1503 | unsigned long index = start >> PAGE_SHIFT; |
1504 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1505 | struct page *page; |
1506 | ||
1507 | while (index <= end_index) { | |
1508 | page = find_get_page(inode->i_mapping, index); | |
1509 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
4adaa611 | 1510 | __set_page_dirty_nobuffers(page); |
8d38633c | 1511 | account_page_redirty(page); |
09cbfeaf | 1512 | put_page(page); |
4adaa611 CM |
1513 | index++; |
1514 | } | |
4adaa611 CM |
1515 | } |
1516 | ||
d352ac68 CM |
1517 | /* find the first state struct with 'bits' set after 'start', and |
1518 | * return it. tree->lock must be held. NULL will returned if | |
1519 | * nothing was found after 'start' | |
1520 | */ | |
48a3b636 | 1521 | static struct extent_state * |
f97e27e9 | 1522 | find_first_extent_bit_state(struct extent_io_tree *tree, u64 start, u32 bits) |
d7fc640e CM |
1523 | { |
1524 | struct rb_node *node; | |
1525 | struct extent_state *state; | |
1526 | ||
1527 | /* | |
1528 | * this search will find all the extents that end after | |
1529 | * our range starts. | |
1530 | */ | |
1531 | node = tree_search(tree, start); | |
d397712b | 1532 | if (!node) |
d7fc640e | 1533 | goto out; |
d7fc640e | 1534 | |
d397712b | 1535 | while (1) { |
d7fc640e | 1536 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1537 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1538 | return state; |
d397712b | 1539 | |
d7fc640e CM |
1540 | node = rb_next(node); |
1541 | if (!node) | |
1542 | break; | |
1543 | } | |
1544 | out: | |
1545 | return NULL; | |
1546 | } | |
d7fc640e | 1547 | |
69261c4b | 1548 | /* |
03509b78 | 1549 | * Find the first offset in the io tree with one or more @bits set. |
69261c4b | 1550 | * |
03509b78 QW |
1551 | * Note: If there are multiple bits set in @bits, any of them will match. |
1552 | * | |
1553 | * Return 0 if we find something, and update @start_ret and @end_ret. | |
1554 | * Return 1 if we found nothing. | |
69261c4b XG |
1555 | */ |
1556 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
f97e27e9 | 1557 | u64 *start_ret, u64 *end_ret, u32 bits, |
e6138876 | 1558 | struct extent_state **cached_state) |
69261c4b XG |
1559 | { |
1560 | struct extent_state *state; | |
1561 | int ret = 1; | |
1562 | ||
1563 | spin_lock(&tree->lock); | |
e6138876 JB |
1564 | if (cached_state && *cached_state) { |
1565 | state = *cached_state; | |
27a3507d | 1566 | if (state->end == start - 1 && extent_state_in_tree(state)) { |
9688e9a9 | 1567 | while ((state = next_state(state)) != NULL) { |
e6138876 JB |
1568 | if (state->state & bits) |
1569 | goto got_it; | |
e6138876 JB |
1570 | } |
1571 | free_extent_state(*cached_state); | |
1572 | *cached_state = NULL; | |
1573 | goto out; | |
1574 | } | |
1575 | free_extent_state(*cached_state); | |
1576 | *cached_state = NULL; | |
1577 | } | |
1578 | ||
69261c4b | 1579 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1580 | got_it: |
69261c4b | 1581 | if (state) { |
e38e2ed7 | 1582 | cache_state_if_flags(state, cached_state, 0); |
69261c4b XG |
1583 | *start_ret = state->start; |
1584 | *end_ret = state->end; | |
1585 | ret = 0; | |
1586 | } | |
e6138876 | 1587 | out: |
69261c4b XG |
1588 | spin_unlock(&tree->lock); |
1589 | return ret; | |
1590 | } | |
1591 | ||
41a2ee75 | 1592 | /** |
3bed2da1 NB |
1593 | * Find a contiguous area of bits |
1594 | * | |
1595 | * @tree: io tree to check | |
1596 | * @start: offset to start the search from | |
1597 | * @start_ret: the first offset we found with the bits set | |
1598 | * @end_ret: the final contiguous range of the bits that were set | |
1599 | * @bits: bits to look for | |
41a2ee75 JB |
1600 | * |
1601 | * set_extent_bit and clear_extent_bit can temporarily split contiguous ranges | |
1602 | * to set bits appropriately, and then merge them again. During this time it | |
1603 | * will drop the tree->lock, so use this helper if you want to find the actual | |
1604 | * contiguous area for given bits. We will search to the first bit we find, and | |
1605 | * then walk down the tree until we find a non-contiguous area. The area | |
1606 | * returned will be the full contiguous area with the bits set. | |
1607 | */ | |
1608 | int find_contiguous_extent_bit(struct extent_io_tree *tree, u64 start, | |
f97e27e9 | 1609 | u64 *start_ret, u64 *end_ret, u32 bits) |
41a2ee75 JB |
1610 | { |
1611 | struct extent_state *state; | |
1612 | int ret = 1; | |
1613 | ||
1614 | spin_lock(&tree->lock); | |
1615 | state = find_first_extent_bit_state(tree, start, bits); | |
1616 | if (state) { | |
1617 | *start_ret = state->start; | |
1618 | *end_ret = state->end; | |
1619 | while ((state = next_state(state)) != NULL) { | |
1620 | if (state->start > (*end_ret + 1)) | |
1621 | break; | |
1622 | *end_ret = state->end; | |
1623 | } | |
1624 | ret = 0; | |
1625 | } | |
1626 | spin_unlock(&tree->lock); | |
1627 | return ret; | |
1628 | } | |
1629 | ||
45bfcfc1 | 1630 | /** |
3bed2da1 NB |
1631 | * Find the first range that has @bits not set. This range could start before |
1632 | * @start. | |
45bfcfc1 | 1633 | * |
3bed2da1 NB |
1634 | * @tree: the tree to search |
1635 | * @start: offset at/after which the found extent should start | |
1636 | * @start_ret: records the beginning of the range | |
1637 | * @end_ret: records the end of the range (inclusive) | |
1638 | * @bits: the set of bits which must be unset | |
45bfcfc1 NB |
1639 | * |
1640 | * Since unallocated range is also considered one which doesn't have the bits | |
1641 | * set it's possible that @end_ret contains -1, this happens in case the range | |
1642 | * spans (last_range_end, end of device]. In this case it's up to the caller to | |
1643 | * trim @end_ret to the appropriate size. | |
1644 | */ | |
1645 | void find_first_clear_extent_bit(struct extent_io_tree *tree, u64 start, | |
f97e27e9 | 1646 | u64 *start_ret, u64 *end_ret, u32 bits) |
45bfcfc1 NB |
1647 | { |
1648 | struct extent_state *state; | |
1649 | struct rb_node *node, *prev = NULL, *next; | |
1650 | ||
1651 | spin_lock(&tree->lock); | |
1652 | ||
1653 | /* Find first extent with bits cleared */ | |
1654 | while (1) { | |
1655 | node = __etree_search(tree, start, &next, &prev, NULL, NULL); | |
5750c375 NB |
1656 | if (!node && !next && !prev) { |
1657 | /* | |
1658 | * Tree is completely empty, send full range and let | |
1659 | * caller deal with it | |
1660 | */ | |
1661 | *start_ret = 0; | |
1662 | *end_ret = -1; | |
1663 | goto out; | |
1664 | } else if (!node && !next) { | |
1665 | /* | |
1666 | * We are past the last allocated chunk, set start at | |
1667 | * the end of the last extent. | |
1668 | */ | |
1669 | state = rb_entry(prev, struct extent_state, rb_node); | |
1670 | *start_ret = state->end + 1; | |
1671 | *end_ret = -1; | |
1672 | goto out; | |
1673 | } else if (!node) { | |
45bfcfc1 | 1674 | node = next; |
45bfcfc1 | 1675 | } |
1eaebb34 NB |
1676 | /* |
1677 | * At this point 'node' either contains 'start' or start is | |
1678 | * before 'node' | |
1679 | */ | |
45bfcfc1 | 1680 | state = rb_entry(node, struct extent_state, rb_node); |
1eaebb34 NB |
1681 | |
1682 | if (in_range(start, state->start, state->end - state->start + 1)) { | |
1683 | if (state->state & bits) { | |
1684 | /* | |
1685 | * |--range with bits sets--| | |
1686 | * | | |
1687 | * start | |
1688 | */ | |
1689 | start = state->end + 1; | |
1690 | } else { | |
1691 | /* | |
1692 | * 'start' falls within a range that doesn't | |
1693 | * have the bits set, so take its start as | |
1694 | * the beginning of the desired range | |
1695 | * | |
1696 | * |--range with bits cleared----| | |
1697 | * | | |
1698 | * start | |
1699 | */ | |
1700 | *start_ret = state->start; | |
1701 | break; | |
1702 | } | |
45bfcfc1 | 1703 | } else { |
1eaebb34 NB |
1704 | /* |
1705 | * |---prev range---|---hole/unset---|---node range---| | |
1706 | * | | |
1707 | * start | |
1708 | * | |
1709 | * or | |
1710 | * | |
1711 | * |---hole/unset--||--first node--| | |
1712 | * 0 | | |
1713 | * start | |
1714 | */ | |
1715 | if (prev) { | |
1716 | state = rb_entry(prev, struct extent_state, | |
1717 | rb_node); | |
1718 | *start_ret = state->end + 1; | |
1719 | } else { | |
1720 | *start_ret = 0; | |
1721 | } | |
45bfcfc1 NB |
1722 | break; |
1723 | } | |
1724 | } | |
1725 | ||
1726 | /* | |
1727 | * Find the longest stretch from start until an entry which has the | |
1728 | * bits set | |
1729 | */ | |
1730 | while (1) { | |
1731 | state = rb_entry(node, struct extent_state, rb_node); | |
1732 | if (state->end >= start && !(state->state & bits)) { | |
1733 | *end_ret = state->end; | |
1734 | } else { | |
1735 | *end_ret = state->start - 1; | |
1736 | break; | |
1737 | } | |
1738 | ||
1739 | node = rb_next(node); | |
1740 | if (!node) | |
1741 | break; | |
1742 | } | |
1743 | out: | |
1744 | spin_unlock(&tree->lock); | |
1745 | } | |
1746 | ||
d352ac68 CM |
1747 | /* |
1748 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1749 | * more than 'max_bytes'. start and end are used to return the range, | |
1750 | * | |
3522e903 | 1751 | * true is returned if we find something, false if nothing was in the tree |
d352ac68 | 1752 | */ |
083e75e7 JB |
1753 | bool btrfs_find_delalloc_range(struct extent_io_tree *tree, u64 *start, |
1754 | u64 *end, u64 max_bytes, | |
1755 | struct extent_state **cached_state) | |
d1310b2e CM |
1756 | { |
1757 | struct rb_node *node; | |
1758 | struct extent_state *state; | |
1759 | u64 cur_start = *start; | |
3522e903 | 1760 | bool found = false; |
d1310b2e CM |
1761 | u64 total_bytes = 0; |
1762 | ||
cad321ad | 1763 | spin_lock(&tree->lock); |
c8b97818 | 1764 | |
d1310b2e CM |
1765 | /* |
1766 | * this search will find all the extents that end after | |
1767 | * our range starts. | |
1768 | */ | |
80ea96b1 | 1769 | node = tree_search(tree, cur_start); |
2b114d1d | 1770 | if (!node) { |
3522e903 | 1771 | *end = (u64)-1; |
d1310b2e CM |
1772 | goto out; |
1773 | } | |
1774 | ||
d397712b | 1775 | while (1) { |
d1310b2e | 1776 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1777 | if (found && (state->start != cur_start || |
1778 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1779 | goto out; |
1780 | } | |
1781 | if (!(state->state & EXTENT_DELALLOC)) { | |
1782 | if (!found) | |
1783 | *end = state->end; | |
1784 | goto out; | |
1785 | } | |
c2a128d2 | 1786 | if (!found) { |
d1310b2e | 1787 | *start = state->start; |
c2a128d2 | 1788 | *cached_state = state; |
b7ac31b7 | 1789 | refcount_inc(&state->refs); |
c2a128d2 | 1790 | } |
3522e903 | 1791 | found = true; |
d1310b2e CM |
1792 | *end = state->end; |
1793 | cur_start = state->end + 1; | |
1794 | node = rb_next(node); | |
d1310b2e | 1795 | total_bytes += state->end - state->start + 1; |
7bf811a5 | 1796 | if (total_bytes >= max_bytes) |
573aecaf | 1797 | break; |
573aecaf | 1798 | if (!node) |
d1310b2e CM |
1799 | break; |
1800 | } | |
1801 | out: | |
cad321ad | 1802 | spin_unlock(&tree->lock); |
d1310b2e CM |
1803 | return found; |
1804 | } | |
1805 | ||
da2c7009 LB |
1806 | static int __process_pages_contig(struct address_space *mapping, |
1807 | struct page *locked_page, | |
1808 | pgoff_t start_index, pgoff_t end_index, | |
1809 | unsigned long page_ops, pgoff_t *index_ret); | |
1810 | ||
143bede5 JM |
1811 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1812 | struct page *locked_page, | |
1813 | u64 start, u64 end) | |
c8b97818 | 1814 | { |
09cbfeaf KS |
1815 | unsigned long index = start >> PAGE_SHIFT; |
1816 | unsigned long end_index = end >> PAGE_SHIFT; | |
c8b97818 | 1817 | |
76c0021d | 1818 | ASSERT(locked_page); |
c8b97818 | 1819 | if (index == locked_page->index && end_index == index) |
143bede5 | 1820 | return; |
c8b97818 | 1821 | |
76c0021d LB |
1822 | __process_pages_contig(inode->i_mapping, locked_page, index, end_index, |
1823 | PAGE_UNLOCK, NULL); | |
c8b97818 CM |
1824 | } |
1825 | ||
1826 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1827 | struct page *locked_page, | |
1828 | u64 delalloc_start, | |
1829 | u64 delalloc_end) | |
1830 | { | |
09cbfeaf | 1831 | unsigned long index = delalloc_start >> PAGE_SHIFT; |
76c0021d | 1832 | unsigned long index_ret = index; |
09cbfeaf | 1833 | unsigned long end_index = delalloc_end >> PAGE_SHIFT; |
c8b97818 | 1834 | int ret; |
c8b97818 | 1835 | |
76c0021d | 1836 | ASSERT(locked_page); |
c8b97818 CM |
1837 | if (index == locked_page->index && index == end_index) |
1838 | return 0; | |
1839 | ||
76c0021d LB |
1840 | ret = __process_pages_contig(inode->i_mapping, locked_page, index, |
1841 | end_index, PAGE_LOCK, &index_ret); | |
1842 | if (ret == -EAGAIN) | |
1843 | __unlock_for_delalloc(inode, locked_page, delalloc_start, | |
1844 | (u64)index_ret << PAGE_SHIFT); | |
c8b97818 CM |
1845 | return ret; |
1846 | } | |
1847 | ||
1848 | /* | |
3522e903 LF |
1849 | * Find and lock a contiguous range of bytes in the file marked as delalloc, no |
1850 | * more than @max_bytes. @Start and @end are used to return the range, | |
c8b97818 | 1851 | * |
3522e903 LF |
1852 | * Return: true if we find something |
1853 | * false if nothing was in the tree | |
c8b97818 | 1854 | */ |
ce9f967f | 1855 | EXPORT_FOR_TESTS |
3522e903 | 1856 | noinline_for_stack bool find_lock_delalloc_range(struct inode *inode, |
294e30fe | 1857 | struct page *locked_page, u64 *start, |
917aacec | 1858 | u64 *end) |
c8b97818 | 1859 | { |
9978059b | 1860 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
917aacec | 1861 | u64 max_bytes = BTRFS_MAX_EXTENT_SIZE; |
c8b97818 CM |
1862 | u64 delalloc_start; |
1863 | u64 delalloc_end; | |
3522e903 | 1864 | bool found; |
9655d298 | 1865 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1866 | int ret; |
1867 | int loops = 0; | |
1868 | ||
1869 | again: | |
1870 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1871 | delalloc_start = *start; | |
1872 | delalloc_end = 0; | |
083e75e7 JB |
1873 | found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end, |
1874 | max_bytes, &cached_state); | |
70b99e69 | 1875 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1876 | *start = delalloc_start; |
1877 | *end = delalloc_end; | |
c2a128d2 | 1878 | free_extent_state(cached_state); |
3522e903 | 1879 | return false; |
c8b97818 CM |
1880 | } |
1881 | ||
70b99e69 CM |
1882 | /* |
1883 | * start comes from the offset of locked_page. We have to lock | |
1884 | * pages in order, so we can't process delalloc bytes before | |
1885 | * locked_page | |
1886 | */ | |
d397712b | 1887 | if (delalloc_start < *start) |
70b99e69 | 1888 | delalloc_start = *start; |
70b99e69 | 1889 | |
c8b97818 CM |
1890 | /* |
1891 | * make sure to limit the number of pages we try to lock down | |
c8b97818 | 1892 | */ |
7bf811a5 JB |
1893 | if (delalloc_end + 1 - delalloc_start > max_bytes) |
1894 | delalloc_end = delalloc_start + max_bytes - 1; | |
d397712b | 1895 | |
c8b97818 CM |
1896 | /* step two, lock all the pages after the page that has start */ |
1897 | ret = lock_delalloc_pages(inode, locked_page, | |
1898 | delalloc_start, delalloc_end); | |
9bfd61d9 | 1899 | ASSERT(!ret || ret == -EAGAIN); |
c8b97818 CM |
1900 | if (ret == -EAGAIN) { |
1901 | /* some of the pages are gone, lets avoid looping by | |
1902 | * shortening the size of the delalloc range we're searching | |
1903 | */ | |
9655d298 | 1904 | free_extent_state(cached_state); |
7d788742 | 1905 | cached_state = NULL; |
c8b97818 | 1906 | if (!loops) { |
09cbfeaf | 1907 | max_bytes = PAGE_SIZE; |
c8b97818 CM |
1908 | loops = 1; |
1909 | goto again; | |
1910 | } else { | |
3522e903 | 1911 | found = false; |
c8b97818 CM |
1912 | goto out_failed; |
1913 | } | |
1914 | } | |
c8b97818 CM |
1915 | |
1916 | /* step three, lock the state bits for the whole range */ | |
ff13db41 | 1917 | lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state); |
c8b97818 CM |
1918 | |
1919 | /* then test to make sure it is all still delalloc */ | |
1920 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1921 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1922 | if (!ret) { |
9655d298 | 1923 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
e43bbe5e | 1924 | &cached_state); |
c8b97818 CM |
1925 | __unlock_for_delalloc(inode, locked_page, |
1926 | delalloc_start, delalloc_end); | |
1927 | cond_resched(); | |
1928 | goto again; | |
1929 | } | |
9655d298 | 1930 | free_extent_state(cached_state); |
c8b97818 CM |
1931 | *start = delalloc_start; |
1932 | *end = delalloc_end; | |
1933 | out_failed: | |
1934 | return found; | |
1935 | } | |
1936 | ||
da2c7009 LB |
1937 | static int __process_pages_contig(struct address_space *mapping, |
1938 | struct page *locked_page, | |
1939 | pgoff_t start_index, pgoff_t end_index, | |
1940 | unsigned long page_ops, pgoff_t *index_ret) | |
c8b97818 | 1941 | { |
873695b3 | 1942 | unsigned long nr_pages = end_index - start_index + 1; |
12e3360f | 1943 | unsigned long pages_processed = 0; |
873695b3 | 1944 | pgoff_t index = start_index; |
c8b97818 | 1945 | struct page *pages[16]; |
873695b3 | 1946 | unsigned ret; |
da2c7009 | 1947 | int err = 0; |
c8b97818 | 1948 | int i; |
771ed689 | 1949 | |
da2c7009 LB |
1950 | if (page_ops & PAGE_LOCK) { |
1951 | ASSERT(page_ops == PAGE_LOCK); | |
1952 | ASSERT(index_ret && *index_ret == start_index); | |
1953 | } | |
1954 | ||
704de49d | 1955 | if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0) |
873695b3 | 1956 | mapping_set_error(mapping, -EIO); |
704de49d | 1957 | |
d397712b | 1958 | while (nr_pages > 0) { |
873695b3 | 1959 | ret = find_get_pages_contig(mapping, index, |
5b050f04 CM |
1960 | min_t(unsigned long, |
1961 | nr_pages, ARRAY_SIZE(pages)), pages); | |
da2c7009 LB |
1962 | if (ret == 0) { |
1963 | /* | |
1964 | * Only if we're going to lock these pages, | |
1965 | * can we find nothing at @index. | |
1966 | */ | |
1967 | ASSERT(page_ops & PAGE_LOCK); | |
49d4a334 LB |
1968 | err = -EAGAIN; |
1969 | goto out; | |
da2c7009 | 1970 | } |
8b62b72b | 1971 | |
da2c7009 | 1972 | for (i = 0; i < ret; i++) { |
c2790a2e | 1973 | if (page_ops & PAGE_SET_PRIVATE2) |
8b62b72b CM |
1974 | SetPagePrivate2(pages[i]); |
1975 | ||
1d53c9e6 | 1976 | if (locked_page && pages[i] == locked_page) { |
09cbfeaf | 1977 | put_page(pages[i]); |
12e3360f | 1978 | pages_processed++; |
c8b97818 CM |
1979 | continue; |
1980 | } | |
6869b0a8 | 1981 | if (page_ops & PAGE_START_WRITEBACK) { |
c8b97818 | 1982 | clear_page_dirty_for_io(pages[i]); |
c8b97818 | 1983 | set_page_writeback(pages[i]); |
6869b0a8 | 1984 | } |
704de49d FM |
1985 | if (page_ops & PAGE_SET_ERROR) |
1986 | SetPageError(pages[i]); | |
c2790a2e | 1987 | if (page_ops & PAGE_END_WRITEBACK) |
c8b97818 | 1988 | end_page_writeback(pages[i]); |
c2790a2e | 1989 | if (page_ops & PAGE_UNLOCK) |
771ed689 | 1990 | unlock_page(pages[i]); |
da2c7009 LB |
1991 | if (page_ops & PAGE_LOCK) { |
1992 | lock_page(pages[i]); | |
1993 | if (!PageDirty(pages[i]) || | |
1994 | pages[i]->mapping != mapping) { | |
1995 | unlock_page(pages[i]); | |
5909ca11 RK |
1996 | for (; i < ret; i++) |
1997 | put_page(pages[i]); | |
da2c7009 LB |
1998 | err = -EAGAIN; |
1999 | goto out; | |
2000 | } | |
2001 | } | |
09cbfeaf | 2002 | put_page(pages[i]); |
12e3360f | 2003 | pages_processed++; |
c8b97818 CM |
2004 | } |
2005 | nr_pages -= ret; | |
2006 | index += ret; | |
2007 | cond_resched(); | |
2008 | } | |
da2c7009 LB |
2009 | out: |
2010 | if (err && index_ret) | |
12e3360f | 2011 | *index_ret = start_index + pages_processed - 1; |
da2c7009 | 2012 | return err; |
c8b97818 | 2013 | } |
c8b97818 | 2014 | |
ad7ff17b | 2015 | void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end, |
74e9194a | 2016 | struct page *locked_page, |
f97e27e9 | 2017 | u32 clear_bits, unsigned long page_ops) |
873695b3 | 2018 | { |
ad7ff17b | 2019 | clear_extent_bit(&inode->io_tree, start, end, clear_bits, 1, 0, NULL); |
873695b3 | 2020 | |
ad7ff17b | 2021 | __process_pages_contig(inode->vfs_inode.i_mapping, locked_page, |
873695b3 | 2022 | start >> PAGE_SHIFT, end >> PAGE_SHIFT, |
da2c7009 | 2023 | page_ops, NULL); |
873695b3 LB |
2024 | } |
2025 | ||
d352ac68 CM |
2026 | /* |
2027 | * count the number of bytes in the tree that have a given bit(s) | |
2028 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
2029 | * cached. The total number found is returned. | |
2030 | */ | |
d1310b2e CM |
2031 | u64 count_range_bits(struct extent_io_tree *tree, |
2032 | u64 *start, u64 search_end, u64 max_bytes, | |
f97e27e9 | 2033 | u32 bits, int contig) |
d1310b2e CM |
2034 | { |
2035 | struct rb_node *node; | |
2036 | struct extent_state *state; | |
2037 | u64 cur_start = *start; | |
2038 | u64 total_bytes = 0; | |
ec29ed5b | 2039 | u64 last = 0; |
d1310b2e CM |
2040 | int found = 0; |
2041 | ||
fae7f21c | 2042 | if (WARN_ON(search_end <= cur_start)) |
d1310b2e | 2043 | return 0; |
d1310b2e | 2044 | |
cad321ad | 2045 | spin_lock(&tree->lock); |
d1310b2e CM |
2046 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
2047 | total_bytes = tree->dirty_bytes; | |
2048 | goto out; | |
2049 | } | |
2050 | /* | |
2051 | * this search will find all the extents that end after | |
2052 | * our range starts. | |
2053 | */ | |
80ea96b1 | 2054 | node = tree_search(tree, cur_start); |
d397712b | 2055 | if (!node) |
d1310b2e | 2056 | goto out; |
d1310b2e | 2057 | |
d397712b | 2058 | while (1) { |
d1310b2e CM |
2059 | state = rb_entry(node, struct extent_state, rb_node); |
2060 | if (state->start > search_end) | |
2061 | break; | |
ec29ed5b CM |
2062 | if (contig && found && state->start > last + 1) |
2063 | break; | |
2064 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
2065 | total_bytes += min(search_end, state->end) + 1 - |
2066 | max(cur_start, state->start); | |
2067 | if (total_bytes >= max_bytes) | |
2068 | break; | |
2069 | if (!found) { | |
af60bed2 | 2070 | *start = max(cur_start, state->start); |
d1310b2e CM |
2071 | found = 1; |
2072 | } | |
ec29ed5b CM |
2073 | last = state->end; |
2074 | } else if (contig && found) { | |
2075 | break; | |
d1310b2e CM |
2076 | } |
2077 | node = rb_next(node); | |
2078 | if (!node) | |
2079 | break; | |
2080 | } | |
2081 | out: | |
cad321ad | 2082 | spin_unlock(&tree->lock); |
d1310b2e CM |
2083 | return total_bytes; |
2084 | } | |
b2950863 | 2085 | |
d352ac68 CM |
2086 | /* |
2087 | * set the private field for a given byte offset in the tree. If there isn't | |
2088 | * an extent_state there already, this does nothing. | |
2089 | */ | |
b3f167aa JB |
2090 | int set_state_failrec(struct extent_io_tree *tree, u64 start, |
2091 | struct io_failure_record *failrec) | |
d1310b2e CM |
2092 | { |
2093 | struct rb_node *node; | |
2094 | struct extent_state *state; | |
2095 | int ret = 0; | |
2096 | ||
cad321ad | 2097 | spin_lock(&tree->lock); |
d1310b2e CM |
2098 | /* |
2099 | * this search will find all the extents that end after | |
2100 | * our range starts. | |
2101 | */ | |
80ea96b1 | 2102 | node = tree_search(tree, start); |
2b114d1d | 2103 | if (!node) { |
d1310b2e CM |
2104 | ret = -ENOENT; |
2105 | goto out; | |
2106 | } | |
2107 | state = rb_entry(node, struct extent_state, rb_node); | |
2108 | if (state->start != start) { | |
2109 | ret = -ENOENT; | |
2110 | goto out; | |
2111 | } | |
47dc196a | 2112 | state->failrec = failrec; |
d1310b2e | 2113 | out: |
cad321ad | 2114 | spin_unlock(&tree->lock); |
d1310b2e CM |
2115 | return ret; |
2116 | } | |
2117 | ||
2279a270 | 2118 | struct io_failure_record *get_state_failrec(struct extent_io_tree *tree, u64 start) |
d1310b2e CM |
2119 | { |
2120 | struct rb_node *node; | |
2121 | struct extent_state *state; | |
2279a270 | 2122 | struct io_failure_record *failrec; |
d1310b2e | 2123 | |
cad321ad | 2124 | spin_lock(&tree->lock); |
d1310b2e CM |
2125 | /* |
2126 | * this search will find all the extents that end after | |
2127 | * our range starts. | |
2128 | */ | |
80ea96b1 | 2129 | node = tree_search(tree, start); |
2b114d1d | 2130 | if (!node) { |
2279a270 | 2131 | failrec = ERR_PTR(-ENOENT); |
d1310b2e CM |
2132 | goto out; |
2133 | } | |
2134 | state = rb_entry(node, struct extent_state, rb_node); | |
2135 | if (state->start != start) { | |
2279a270 | 2136 | failrec = ERR_PTR(-ENOENT); |
d1310b2e CM |
2137 | goto out; |
2138 | } | |
2279a270 NB |
2139 | |
2140 | failrec = state->failrec; | |
d1310b2e | 2141 | out: |
cad321ad | 2142 | spin_unlock(&tree->lock); |
2279a270 | 2143 | return failrec; |
d1310b2e CM |
2144 | } |
2145 | ||
2146 | /* | |
2147 | * searches a range in the state tree for a given mask. | |
70dec807 | 2148 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
2149 | * has the bits set. Otherwise, 1 is returned if any bit in the |
2150 | * range is found set. | |
2151 | */ | |
2152 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
f97e27e9 | 2153 | u32 bits, int filled, struct extent_state *cached) |
d1310b2e CM |
2154 | { |
2155 | struct extent_state *state = NULL; | |
2156 | struct rb_node *node; | |
2157 | int bitset = 0; | |
d1310b2e | 2158 | |
cad321ad | 2159 | spin_lock(&tree->lock); |
27a3507d | 2160 | if (cached && extent_state_in_tree(cached) && cached->start <= start && |
df98b6e2 | 2161 | cached->end > start) |
9655d298 CM |
2162 | node = &cached->rb_node; |
2163 | else | |
2164 | node = tree_search(tree, start); | |
d1310b2e CM |
2165 | while (node && start <= end) { |
2166 | state = rb_entry(node, struct extent_state, rb_node); | |
2167 | ||
2168 | if (filled && state->start > start) { | |
2169 | bitset = 0; | |
2170 | break; | |
2171 | } | |
2172 | ||
2173 | if (state->start > end) | |
2174 | break; | |
2175 | ||
2176 | if (state->state & bits) { | |
2177 | bitset = 1; | |
2178 | if (!filled) | |
2179 | break; | |
2180 | } else if (filled) { | |
2181 | bitset = 0; | |
2182 | break; | |
2183 | } | |
46562cec CM |
2184 | |
2185 | if (state->end == (u64)-1) | |
2186 | break; | |
2187 | ||
d1310b2e CM |
2188 | start = state->end + 1; |
2189 | if (start > end) | |
2190 | break; | |
2191 | node = rb_next(node); | |
2192 | if (!node) { | |
2193 | if (filled) | |
2194 | bitset = 0; | |
2195 | break; | |
2196 | } | |
2197 | } | |
cad321ad | 2198 | spin_unlock(&tree->lock); |
d1310b2e CM |
2199 | return bitset; |
2200 | } | |
d1310b2e CM |
2201 | |
2202 | /* | |
2203 | * helper function to set a given page up to date if all the | |
2204 | * extents in the tree for that page are up to date | |
2205 | */ | |
143bede5 | 2206 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e | 2207 | { |
4eee4fa4 | 2208 | u64 start = page_offset(page); |
09cbfeaf | 2209 | u64 end = start + PAGE_SIZE - 1; |
9655d298 | 2210 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 2211 | SetPageUptodate(page); |
d1310b2e CM |
2212 | } |
2213 | ||
7870d082 JB |
2214 | int free_io_failure(struct extent_io_tree *failure_tree, |
2215 | struct extent_io_tree *io_tree, | |
2216 | struct io_failure_record *rec) | |
4a54c8c1 JS |
2217 | { |
2218 | int ret; | |
2219 | int err = 0; | |
4a54c8c1 | 2220 | |
47dc196a | 2221 | set_state_failrec(failure_tree, rec->start, NULL); |
4a54c8c1 JS |
2222 | ret = clear_extent_bits(failure_tree, rec->start, |
2223 | rec->start + rec->len - 1, | |
91166212 | 2224 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 JS |
2225 | if (ret) |
2226 | err = ret; | |
2227 | ||
7870d082 | 2228 | ret = clear_extent_bits(io_tree, rec->start, |
53b381b3 | 2229 | rec->start + rec->len - 1, |
91166212 | 2230 | EXTENT_DAMAGED); |
53b381b3 DW |
2231 | if (ret && !err) |
2232 | err = ret; | |
4a54c8c1 JS |
2233 | |
2234 | kfree(rec); | |
2235 | return err; | |
2236 | } | |
2237 | ||
4a54c8c1 JS |
2238 | /* |
2239 | * this bypasses the standard btrfs submit functions deliberately, as | |
2240 | * the standard behavior is to write all copies in a raid setup. here we only | |
2241 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
2242 | * submit_bio directly. | |
3ec706c8 | 2243 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
2244 | * actually prevents the read that triggered the error from finishing. |
2245 | * currently, there can be no more than two copies of every data bit. thus, | |
2246 | * exactly one rewrite is required. | |
2247 | */ | |
6ec656bc JB |
2248 | int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start, |
2249 | u64 length, u64 logical, struct page *page, | |
2250 | unsigned int pg_offset, int mirror_num) | |
4a54c8c1 JS |
2251 | { |
2252 | struct bio *bio; | |
2253 | struct btrfs_device *dev; | |
4a54c8c1 JS |
2254 | u64 map_length = 0; |
2255 | u64 sector; | |
2256 | struct btrfs_bio *bbio = NULL; | |
2257 | int ret; | |
2258 | ||
1751e8a6 | 2259 | ASSERT(!(fs_info->sb->s_flags & SB_RDONLY)); |
4a54c8c1 JS |
2260 | BUG_ON(!mirror_num); |
2261 | ||
c5e4c3d7 | 2262 | bio = btrfs_io_bio_alloc(1); |
4f024f37 | 2263 | bio->bi_iter.bi_size = 0; |
4a54c8c1 JS |
2264 | map_length = length; |
2265 | ||
b5de8d0d FM |
2266 | /* |
2267 | * Avoid races with device replace and make sure our bbio has devices | |
2268 | * associated to its stripes that don't go away while we are doing the | |
2269 | * read repair operation. | |
2270 | */ | |
2271 | btrfs_bio_counter_inc_blocked(fs_info); | |
e4ff5fb5 | 2272 | if (btrfs_is_parity_mirror(fs_info, logical, length)) { |
c725328c LB |
2273 | /* |
2274 | * Note that we don't use BTRFS_MAP_WRITE because it's supposed | |
2275 | * to update all raid stripes, but here we just want to correct | |
2276 | * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad | |
2277 | * stripe's dev and sector. | |
2278 | */ | |
2279 | ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical, | |
2280 | &map_length, &bbio, 0); | |
2281 | if (ret) { | |
2282 | btrfs_bio_counter_dec(fs_info); | |
2283 | bio_put(bio); | |
2284 | return -EIO; | |
2285 | } | |
2286 | ASSERT(bbio->mirror_num == 1); | |
2287 | } else { | |
2288 | ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, | |
2289 | &map_length, &bbio, mirror_num); | |
2290 | if (ret) { | |
2291 | btrfs_bio_counter_dec(fs_info); | |
2292 | bio_put(bio); | |
2293 | return -EIO; | |
2294 | } | |
2295 | BUG_ON(mirror_num != bbio->mirror_num); | |
4a54c8c1 | 2296 | } |
c725328c LB |
2297 | |
2298 | sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9; | |
4f024f37 | 2299 | bio->bi_iter.bi_sector = sector; |
c725328c | 2300 | dev = bbio->stripes[bbio->mirror_num - 1].dev; |
6e9606d2 | 2301 | btrfs_put_bbio(bbio); |
ebbede42 AJ |
2302 | if (!dev || !dev->bdev || |
2303 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) { | |
b5de8d0d | 2304 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2305 | bio_put(bio); |
2306 | return -EIO; | |
2307 | } | |
74d46992 | 2308 | bio_set_dev(bio, dev->bdev); |
70fd7614 | 2309 | bio->bi_opf = REQ_OP_WRITE | REQ_SYNC; |
ffdd2018 | 2310 | bio_add_page(bio, page, length, pg_offset); |
4a54c8c1 | 2311 | |
4e49ea4a | 2312 | if (btrfsic_submit_bio_wait(bio)) { |
4a54c8c1 | 2313 | /* try to remap that extent elsewhere? */ |
b5de8d0d | 2314 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 | 2315 | bio_put(bio); |
442a4f63 | 2316 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
2317 | return -EIO; |
2318 | } | |
2319 | ||
b14af3b4 DS |
2320 | btrfs_info_rl_in_rcu(fs_info, |
2321 | "read error corrected: ino %llu off %llu (dev %s sector %llu)", | |
6ec656bc | 2322 | ino, start, |
1203b681 | 2323 | rcu_str_deref(dev->name), sector); |
b5de8d0d | 2324 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2325 | bio_put(bio); |
2326 | return 0; | |
2327 | } | |
2328 | ||
2b48966a | 2329 | int btrfs_repair_eb_io_failure(const struct extent_buffer *eb, int mirror_num) |
ea466794 | 2330 | { |
20a1fbf9 | 2331 | struct btrfs_fs_info *fs_info = eb->fs_info; |
ea466794 | 2332 | u64 start = eb->start; |
cc5e31a4 | 2333 | int i, num_pages = num_extent_pages(eb); |
d95603b2 | 2334 | int ret = 0; |
ea466794 | 2335 | |
bc98a42c | 2336 | if (sb_rdonly(fs_info->sb)) |
908960c6 ID |
2337 | return -EROFS; |
2338 | ||
ea466794 | 2339 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 2340 | struct page *p = eb->pages[i]; |
1203b681 | 2341 | |
6ec656bc | 2342 | ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p, |
1203b681 | 2343 | start - page_offset(p), mirror_num); |
ea466794 JB |
2344 | if (ret) |
2345 | break; | |
09cbfeaf | 2346 | start += PAGE_SIZE; |
ea466794 JB |
2347 | } |
2348 | ||
2349 | return ret; | |
2350 | } | |
2351 | ||
4a54c8c1 JS |
2352 | /* |
2353 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2354 | * to see if we need to process or clean up an io_failure_record | |
2355 | */ | |
7870d082 JB |
2356 | int clean_io_failure(struct btrfs_fs_info *fs_info, |
2357 | struct extent_io_tree *failure_tree, | |
2358 | struct extent_io_tree *io_tree, u64 start, | |
2359 | struct page *page, u64 ino, unsigned int pg_offset) | |
4a54c8c1 JS |
2360 | { |
2361 | u64 private; | |
4a54c8c1 | 2362 | struct io_failure_record *failrec; |
4a54c8c1 JS |
2363 | struct extent_state *state; |
2364 | int num_copies; | |
4a54c8c1 | 2365 | int ret; |
4a54c8c1 JS |
2366 | |
2367 | private = 0; | |
7870d082 JB |
2368 | ret = count_range_bits(failure_tree, &private, (u64)-1, 1, |
2369 | EXTENT_DIRTY, 0); | |
4a54c8c1 JS |
2370 | if (!ret) |
2371 | return 0; | |
2372 | ||
2279a270 NB |
2373 | failrec = get_state_failrec(failure_tree, start); |
2374 | if (IS_ERR(failrec)) | |
4a54c8c1 JS |
2375 | return 0; |
2376 | ||
4a54c8c1 JS |
2377 | BUG_ON(!failrec->this_mirror); |
2378 | ||
2379 | if (failrec->in_validation) { | |
2380 | /* there was no real error, just free the record */ | |
ab8d0fc4 JM |
2381 | btrfs_debug(fs_info, |
2382 | "clean_io_failure: freeing dummy error at %llu", | |
2383 | failrec->start); | |
4a54c8c1 JS |
2384 | goto out; |
2385 | } | |
bc98a42c | 2386 | if (sb_rdonly(fs_info->sb)) |
908960c6 | 2387 | goto out; |
4a54c8c1 | 2388 | |
7870d082 JB |
2389 | spin_lock(&io_tree->lock); |
2390 | state = find_first_extent_bit_state(io_tree, | |
4a54c8c1 JS |
2391 | failrec->start, |
2392 | EXTENT_LOCKED); | |
7870d082 | 2393 | spin_unlock(&io_tree->lock); |
4a54c8c1 | 2394 | |
883d0de4 MX |
2395 | if (state && state->start <= failrec->start && |
2396 | state->end >= failrec->start + failrec->len - 1) { | |
3ec706c8 SB |
2397 | num_copies = btrfs_num_copies(fs_info, failrec->logical, |
2398 | failrec->len); | |
4a54c8c1 | 2399 | if (num_copies > 1) { |
7870d082 JB |
2400 | repair_io_failure(fs_info, ino, start, failrec->len, |
2401 | failrec->logical, page, pg_offset, | |
2402 | failrec->failed_mirror); | |
4a54c8c1 JS |
2403 | } |
2404 | } | |
2405 | ||
2406 | out: | |
7870d082 | 2407 | free_io_failure(failure_tree, io_tree, failrec); |
4a54c8c1 | 2408 | |
454ff3de | 2409 | return 0; |
4a54c8c1 JS |
2410 | } |
2411 | ||
f612496b MX |
2412 | /* |
2413 | * Can be called when | |
2414 | * - hold extent lock | |
2415 | * - under ordered extent | |
2416 | * - the inode is freeing | |
2417 | */ | |
7ab7956e | 2418 | void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end) |
f612496b | 2419 | { |
7ab7956e | 2420 | struct extent_io_tree *failure_tree = &inode->io_failure_tree; |
f612496b MX |
2421 | struct io_failure_record *failrec; |
2422 | struct extent_state *state, *next; | |
2423 | ||
2424 | if (RB_EMPTY_ROOT(&failure_tree->state)) | |
2425 | return; | |
2426 | ||
2427 | spin_lock(&failure_tree->lock); | |
2428 | state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY); | |
2429 | while (state) { | |
2430 | if (state->start > end) | |
2431 | break; | |
2432 | ||
2433 | ASSERT(state->end <= end); | |
2434 | ||
2435 | next = next_state(state); | |
2436 | ||
47dc196a | 2437 | failrec = state->failrec; |
f612496b MX |
2438 | free_extent_state(state); |
2439 | kfree(failrec); | |
2440 | ||
2441 | state = next; | |
2442 | } | |
2443 | spin_unlock(&failure_tree->lock); | |
2444 | } | |
2445 | ||
3526302f NB |
2446 | static struct io_failure_record *btrfs_get_io_failure_record(struct inode *inode, |
2447 | u64 start, u64 end) | |
4a54c8c1 | 2448 | { |
ab8d0fc4 | 2449 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e | 2450 | struct io_failure_record *failrec; |
4a54c8c1 | 2451 | struct extent_map *em; |
4a54c8c1 JS |
2452 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2453 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2454 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
4a54c8c1 | 2455 | int ret; |
4a54c8c1 JS |
2456 | u64 logical; |
2457 | ||
2279a270 | 2458 | failrec = get_state_failrec(failure_tree, start); |
3526302f | 2459 | if (!IS_ERR(failrec)) { |
ab8d0fc4 JM |
2460 | btrfs_debug(fs_info, |
2461 | "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d", | |
2462 | failrec->logical, failrec->start, failrec->len, | |
2463 | failrec->in_validation); | |
4a54c8c1 JS |
2464 | /* |
2465 | * when data can be on disk more than twice, add to failrec here | |
2466 | * (e.g. with a list for failed_mirror) to make | |
2467 | * clean_io_failure() clean all those errors at once. | |
2468 | */ | |
3526302f NB |
2469 | |
2470 | return failrec; | |
4a54c8c1 | 2471 | } |
2fe6303e | 2472 | |
3526302f NB |
2473 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); |
2474 | if (!failrec) | |
2475 | return ERR_PTR(-ENOMEM); | |
2fe6303e | 2476 | |
3526302f NB |
2477 | failrec->start = start; |
2478 | failrec->len = end - start + 1; | |
2479 | failrec->this_mirror = 0; | |
2480 | failrec->bio_flags = 0; | |
2481 | failrec->in_validation = 0; | |
2482 | ||
2483 | read_lock(&em_tree->lock); | |
2484 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2485 | if (!em) { | |
2486 | read_unlock(&em_tree->lock); | |
2487 | kfree(failrec); | |
2488 | return ERR_PTR(-EIO); | |
2489 | } | |
2490 | ||
2491 | if (em->start > start || em->start + em->len <= start) { | |
2492 | free_extent_map(em); | |
2493 | em = NULL; | |
2494 | } | |
2495 | read_unlock(&em_tree->lock); | |
2496 | if (!em) { | |
2497 | kfree(failrec); | |
2498 | return ERR_PTR(-EIO); | |
2499 | } | |
2500 | ||
2501 | logical = start - em->start; | |
2502 | logical = em->block_start + logical; | |
2503 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2504 | logical = em->block_start; | |
2505 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2506 | extent_set_compress_type(&failrec->bio_flags, em->compress_type); | |
2507 | } | |
2508 | ||
2509 | btrfs_debug(fs_info, | |
2510 | "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu", | |
2511 | logical, start, failrec->len); | |
2512 | ||
2513 | failrec->logical = logical; | |
2514 | free_extent_map(em); | |
2515 | ||
2516 | /* Set the bits in the private failure tree */ | |
2517 | ret = set_extent_bits(failure_tree, start, end, | |
2518 | EXTENT_LOCKED | EXTENT_DIRTY); | |
2519 | if (ret >= 0) { | |
2520 | ret = set_state_failrec(failure_tree, start, failrec); | |
2521 | /* Set the bits in the inode's tree */ | |
2522 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED); | |
2523 | } else if (ret < 0) { | |
2524 | kfree(failrec); | |
2525 | return ERR_PTR(ret); | |
2526 | } | |
2527 | ||
2528 | return failrec; | |
2fe6303e MX |
2529 | } |
2530 | ||
ce06d3ec OS |
2531 | static bool btrfs_check_repairable(struct inode *inode, bool needs_validation, |
2532 | struct io_failure_record *failrec, | |
2533 | int failed_mirror) | |
2fe6303e | 2534 | { |
ab8d0fc4 | 2535 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e MX |
2536 | int num_copies; |
2537 | ||
ab8d0fc4 | 2538 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
4a54c8c1 JS |
2539 | if (num_copies == 1) { |
2540 | /* | |
2541 | * we only have a single copy of the data, so don't bother with | |
2542 | * all the retry and error correction code that follows. no | |
2543 | * matter what the error is, it is very likely to persist. | |
2544 | */ | |
ab8d0fc4 JM |
2545 | btrfs_debug(fs_info, |
2546 | "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
2547 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2548 | return false; |
4a54c8c1 JS |
2549 | } |
2550 | ||
4a54c8c1 JS |
2551 | /* |
2552 | * there are two premises: | |
2553 | * a) deliver good data to the caller | |
2554 | * b) correct the bad sectors on disk | |
2555 | */ | |
c7333972 | 2556 | if (needs_validation) { |
4a54c8c1 JS |
2557 | /* |
2558 | * to fulfill b), we need to know the exact failing sectors, as | |
2559 | * we don't want to rewrite any more than the failed ones. thus, | |
2560 | * we need separate read requests for the failed bio | |
2561 | * | |
2562 | * if the following BUG_ON triggers, our validation request got | |
2563 | * merged. we need separate requests for our algorithm to work. | |
2564 | */ | |
2565 | BUG_ON(failrec->in_validation); | |
2566 | failrec->in_validation = 1; | |
2567 | failrec->this_mirror = failed_mirror; | |
4a54c8c1 JS |
2568 | } else { |
2569 | /* | |
2570 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2571 | * of the failed sector and if we succeed, we have setup | |
2572 | * everything for repair_io_failure to do the rest for us. | |
2573 | */ | |
2574 | if (failrec->in_validation) { | |
2575 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2576 | failrec->in_validation = 0; | |
2577 | failrec->this_mirror = 0; | |
2578 | } | |
2579 | failrec->failed_mirror = failed_mirror; | |
2580 | failrec->this_mirror++; | |
2581 | if (failrec->this_mirror == failed_mirror) | |
2582 | failrec->this_mirror++; | |
4a54c8c1 JS |
2583 | } |
2584 | ||
facc8a22 | 2585 | if (failrec->this_mirror > num_copies) { |
ab8d0fc4 JM |
2586 | btrfs_debug(fs_info, |
2587 | "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
2588 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2589 | return false; |
4a54c8c1 JS |
2590 | } |
2591 | ||
c3cfb656 | 2592 | return true; |
2fe6303e MX |
2593 | } |
2594 | ||
c7333972 | 2595 | static bool btrfs_io_needs_validation(struct inode *inode, struct bio *bio) |
2fe6303e | 2596 | { |
c7333972 | 2597 | u64 len = 0; |
77d5d689 | 2598 | const u32 blocksize = inode->i_sb->s_blocksize; |
2fe6303e | 2599 | |
f337bd74 OS |
2600 | /* |
2601 | * If bi_status is BLK_STS_OK, then this was a checksum error, not an | |
2602 | * I/O error. In this case, we already know exactly which sector was | |
2603 | * bad, so we don't need to validate. | |
2604 | */ | |
2605 | if (bio->bi_status == BLK_STS_OK) | |
2606 | return false; | |
4a54c8c1 | 2607 | |
c7333972 OS |
2608 | /* |
2609 | * We need to validate each sector individually if the failed I/O was | |
2610 | * for multiple sectors. | |
77d5d689 OS |
2611 | * |
2612 | * There are a few possible bios that can end up here: | |
2613 | * 1. A buffered read bio, which is not cloned. | |
2614 | * 2. A direct I/O read bio, which is cloned. | |
2615 | * 3. A (buffered or direct) repair bio, which is not cloned. | |
2616 | * | |
2617 | * For cloned bios (case 2), we can get the size from | |
2618 | * btrfs_io_bio->iter; for non-cloned bios (cases 1 and 3), we can get | |
2619 | * it from the bvecs. | |
c7333972 | 2620 | */ |
77d5d689 OS |
2621 | if (bio_flagged(bio, BIO_CLONED)) { |
2622 | if (btrfs_io_bio(bio)->iter.bi_size > blocksize) | |
c7333972 | 2623 | return true; |
77d5d689 OS |
2624 | } else { |
2625 | struct bio_vec *bvec; | |
2626 | int i; | |
facc8a22 | 2627 | |
77d5d689 OS |
2628 | bio_for_each_bvec_all(bvec, bio, i) { |
2629 | len += bvec->bv_len; | |
2630 | if (len > blocksize) | |
2631 | return true; | |
2632 | } | |
facc8a22 | 2633 | } |
c7333972 | 2634 | return false; |
2fe6303e MX |
2635 | } |
2636 | ||
77d5d689 | 2637 | blk_status_t btrfs_submit_read_repair(struct inode *inode, |
7ffd27e3 | 2638 | struct bio *failed_bio, u32 bio_offset, |
77d5d689 OS |
2639 | struct page *page, unsigned int pgoff, |
2640 | u64 start, u64 end, int failed_mirror, | |
2641 | submit_bio_hook_t *submit_bio_hook) | |
2fe6303e MX |
2642 | { |
2643 | struct io_failure_record *failrec; | |
77d5d689 | 2644 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e | 2645 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
7870d082 | 2646 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
77d5d689 | 2647 | struct btrfs_io_bio *failed_io_bio = btrfs_io_bio(failed_bio); |
7ffd27e3 | 2648 | const int icsum = bio_offset >> fs_info->sectorsize_bits; |
c7333972 | 2649 | bool need_validation; |
77d5d689 OS |
2650 | struct bio *repair_bio; |
2651 | struct btrfs_io_bio *repair_io_bio; | |
4e4cbee9 | 2652 | blk_status_t status; |
2fe6303e | 2653 | |
77d5d689 OS |
2654 | btrfs_debug(fs_info, |
2655 | "repair read error: read error at %llu", start); | |
2fe6303e | 2656 | |
1f7ad75b | 2657 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
2fe6303e | 2658 | |
3526302f NB |
2659 | failrec = btrfs_get_io_failure_record(inode, start, end); |
2660 | if (IS_ERR(failrec)) | |
2661 | return errno_to_blk_status(PTR_ERR(failrec)); | |
2fe6303e | 2662 | |
c7333972 | 2663 | need_validation = btrfs_io_needs_validation(inode, failed_bio); |
2fe6303e | 2664 | |
c7333972 | 2665 | if (!btrfs_check_repairable(inode, need_validation, failrec, |
c3cfb656 | 2666 | failed_mirror)) { |
7870d082 | 2667 | free_io_failure(failure_tree, tree, failrec); |
77d5d689 | 2668 | return BLK_STS_IOERR; |
2fe6303e MX |
2669 | } |
2670 | ||
77d5d689 OS |
2671 | repair_bio = btrfs_io_bio_alloc(1); |
2672 | repair_io_bio = btrfs_io_bio(repair_bio); | |
2673 | repair_bio->bi_opf = REQ_OP_READ; | |
c7333972 | 2674 | if (need_validation) |
77d5d689 OS |
2675 | repair_bio->bi_opf |= REQ_FAILFAST_DEV; |
2676 | repair_bio->bi_end_io = failed_bio->bi_end_io; | |
2677 | repair_bio->bi_iter.bi_sector = failrec->logical >> 9; | |
2678 | repair_bio->bi_private = failed_bio->bi_private; | |
2fe6303e | 2679 | |
77d5d689 | 2680 | if (failed_io_bio->csum) { |
223486c2 | 2681 | const u32 csum_size = fs_info->csum_size; |
77d5d689 OS |
2682 | |
2683 | repair_io_bio->csum = repair_io_bio->csum_inline; | |
2684 | memcpy(repair_io_bio->csum, | |
2685 | failed_io_bio->csum + csum_size * icsum, csum_size); | |
2686 | } | |
2fe6303e | 2687 | |
77d5d689 OS |
2688 | bio_add_page(repair_bio, page, failrec->len, pgoff); |
2689 | repair_io_bio->logical = failrec->start; | |
2690 | repair_io_bio->iter = repair_bio->bi_iter; | |
4a54c8c1 | 2691 | |
ab8d0fc4 | 2692 | btrfs_debug(btrfs_sb(inode->i_sb), |
77d5d689 OS |
2693 | "repair read error: submitting new read to mirror %d, in_validation=%d", |
2694 | failrec->this_mirror, failrec->in_validation); | |
4a54c8c1 | 2695 | |
77d5d689 OS |
2696 | status = submit_bio_hook(inode, repair_bio, failrec->this_mirror, |
2697 | failrec->bio_flags); | |
4e4cbee9 | 2698 | if (status) { |
7870d082 | 2699 | free_io_failure(failure_tree, tree, failrec); |
77d5d689 | 2700 | bio_put(repair_bio); |
6c387ab2 | 2701 | } |
77d5d689 | 2702 | return status; |
4a54c8c1 JS |
2703 | } |
2704 | ||
d1310b2e CM |
2705 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2706 | ||
b5227c07 | 2707 | void end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
87826df0 JM |
2708 | { |
2709 | int uptodate = (err == 0); | |
3e2426bd | 2710 | int ret = 0; |
87826df0 | 2711 | |
c629732d | 2712 | btrfs_writepage_endio_finish_ordered(page, start, end, uptodate); |
87826df0 | 2713 | |
87826df0 | 2714 | if (!uptodate) { |
87826df0 JM |
2715 | ClearPageUptodate(page); |
2716 | SetPageError(page); | |
bff5baf8 | 2717 | ret = err < 0 ? err : -EIO; |
5dca6eea | 2718 | mapping_set_error(page->mapping, ret); |
87826df0 | 2719 | } |
87826df0 JM |
2720 | } |
2721 | ||
d1310b2e CM |
2722 | /* |
2723 | * after a writepage IO is done, we need to: | |
2724 | * clear the uptodate bits on error | |
2725 | * clear the writeback bits in the extent tree for this IO | |
2726 | * end_page_writeback if the page has no more pending IO | |
2727 | * | |
2728 | * Scheduling is not allowed, so the extent state tree is expected | |
2729 | * to have one and only one object corresponding to this IO. | |
2730 | */ | |
4246a0b6 | 2731 | static void end_bio_extent_writepage(struct bio *bio) |
d1310b2e | 2732 | { |
4e4cbee9 | 2733 | int error = blk_status_to_errno(bio->bi_status); |
2c30c71b | 2734 | struct bio_vec *bvec; |
d1310b2e CM |
2735 | u64 start; |
2736 | u64 end; | |
6dc4f100 | 2737 | struct bvec_iter_all iter_all; |
d1310b2e | 2738 | |
c09abff8 | 2739 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 2740 | bio_for_each_segment_all(bvec, bio, iter_all) { |
d1310b2e | 2741 | struct page *page = bvec->bv_page; |
0b246afa JM |
2742 | struct inode *inode = page->mapping->host; |
2743 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
902b22f3 | 2744 | |
17a5adcc AO |
2745 | /* We always issue full-page reads, but if some block |
2746 | * in a page fails to read, blk_update_request() will | |
2747 | * advance bv_offset and adjust bv_len to compensate. | |
2748 | * Print a warning for nonzero offsets, and an error | |
2749 | * if they don't add up to a full page. */ | |
09cbfeaf KS |
2750 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { |
2751 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
0b246afa | 2752 | btrfs_err(fs_info, |
efe120a0 FH |
2753 | "partial page write in btrfs with offset %u and length %u", |
2754 | bvec->bv_offset, bvec->bv_len); | |
2755 | else | |
0b246afa | 2756 | btrfs_info(fs_info, |
5d163e0e | 2757 | "incomplete page write in btrfs with offset %u and length %u", |
efe120a0 FH |
2758 | bvec->bv_offset, bvec->bv_len); |
2759 | } | |
d1310b2e | 2760 | |
17a5adcc AO |
2761 | start = page_offset(page); |
2762 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
d1310b2e | 2763 | |
4e4cbee9 | 2764 | end_extent_writepage(page, error, start, end); |
17a5adcc | 2765 | end_page_writeback(page); |
2c30c71b | 2766 | } |
2b1f55b0 | 2767 | |
d1310b2e | 2768 | bio_put(bio); |
d1310b2e CM |
2769 | } |
2770 | ||
94e8c95c QW |
2771 | /* |
2772 | * Record previously processed extent range | |
2773 | * | |
2774 | * For endio_readpage_release_extent() to handle a full extent range, reducing | |
2775 | * the extent io operations. | |
2776 | */ | |
2777 | struct processed_extent { | |
2778 | struct btrfs_inode *inode; | |
2779 | /* Start of the range in @inode */ | |
2780 | u64 start; | |
2e626e56 | 2781 | /* End of the range in @inode */ |
94e8c95c QW |
2782 | u64 end; |
2783 | bool uptodate; | |
2784 | }; | |
2785 | ||
2786 | /* | |
2787 | * Try to release processed extent range | |
2788 | * | |
2789 | * May not release the extent range right now if the current range is | |
2790 | * contiguous to processed extent. | |
2791 | * | |
2792 | * Will release processed extent when any of @inode, @uptodate, the range is | |
2793 | * no longer contiguous to the processed range. | |
2794 | * | |
2795 | * Passing @inode == NULL will force processed extent to be released. | |
2796 | */ | |
2797 | static void endio_readpage_release_extent(struct processed_extent *processed, | |
2798 | struct btrfs_inode *inode, u64 start, u64 end, | |
2799 | bool uptodate) | |
883d0de4 MX |
2800 | { |
2801 | struct extent_state *cached = NULL; | |
94e8c95c QW |
2802 | struct extent_io_tree *tree; |
2803 | ||
2804 | /* The first extent, initialize @processed */ | |
2805 | if (!processed->inode) | |
2806 | goto update; | |
883d0de4 | 2807 | |
94e8c95c QW |
2808 | /* |
2809 | * Contiguous to processed extent, just uptodate the end. | |
2810 | * | |
2811 | * Several things to notice: | |
2812 | * | |
2813 | * - bio can be merged as long as on-disk bytenr is contiguous | |
2814 | * This means we can have page belonging to other inodes, thus need to | |
2815 | * check if the inode still matches. | |
2816 | * - bvec can contain range beyond current page for multi-page bvec | |
2817 | * Thus we need to do processed->end + 1 >= start check | |
2818 | */ | |
2819 | if (processed->inode == inode && processed->uptodate == uptodate && | |
2820 | processed->end + 1 >= start && end >= processed->end) { | |
2821 | processed->end = end; | |
2822 | return; | |
2823 | } | |
2824 | ||
2825 | tree = &processed->inode->io_tree; | |
2826 | /* | |
2827 | * Now we don't have range contiguous to the processed range, release | |
2828 | * the processed range now. | |
2829 | */ | |
2830 | if (processed->uptodate && tree->track_uptodate) | |
2831 | set_extent_uptodate(tree, processed->start, processed->end, | |
2832 | &cached, GFP_ATOMIC); | |
2833 | unlock_extent_cached_atomic(tree, processed->start, processed->end, | |
2834 | &cached); | |
2835 | ||
2836 | update: | |
2837 | /* Update processed to current range */ | |
2838 | processed->inode = inode; | |
2839 | processed->start = start; | |
2840 | processed->end = end; | |
2841 | processed->uptodate = uptodate; | |
883d0de4 MX |
2842 | } |
2843 | ||
92082d40 QW |
2844 | static void begin_page_read(struct btrfs_fs_info *fs_info, struct page *page) |
2845 | { | |
2846 | ASSERT(PageLocked(page)); | |
2847 | if (fs_info->sectorsize == PAGE_SIZE) | |
2848 | return; | |
2849 | ||
2850 | ASSERT(PagePrivate(page)); | |
2851 | btrfs_subpage_start_reader(fs_info, page, page_offset(page), PAGE_SIZE); | |
2852 | } | |
2853 | ||
2854 | static void end_page_read(struct page *page, bool uptodate, u64 start, u32 len) | |
e09caaf9 | 2855 | { |
4325cb22 QW |
2856 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); |
2857 | ||
2858 | ASSERT(page_offset(page) <= start && | |
2859 | start + len <= page_offset(page) + PAGE_SIZE); | |
2860 | ||
e09caaf9 | 2861 | if (uptodate) { |
4325cb22 | 2862 | btrfs_page_set_uptodate(fs_info, page, start, len); |
e09caaf9 | 2863 | } else { |
4325cb22 QW |
2864 | btrfs_page_clear_uptodate(fs_info, page, start, len); |
2865 | btrfs_page_set_error(fs_info, page, start, len); | |
e09caaf9 | 2866 | } |
4325cb22 QW |
2867 | |
2868 | if (fs_info->sectorsize == PAGE_SIZE) | |
2869 | unlock_page(page); | |
92082d40 QW |
2870 | else if (is_data_inode(page->mapping->host)) |
2871 | /* | |
2872 | * For subpage data, unlock the page if we're the last reader. | |
2873 | * For subpage metadata, page lock is not utilized for read. | |
2874 | */ | |
2875 | btrfs_subpage_end_reader(fs_info, page, start, len); | |
e09caaf9 QW |
2876 | } |
2877 | ||
d1310b2e CM |
2878 | /* |
2879 | * after a readpage IO is done, we need to: | |
2880 | * clear the uptodate bits on error | |
2881 | * set the uptodate bits if things worked | |
2882 | * set the page up to date if all extents in the tree are uptodate | |
2883 | * clear the lock bit in the extent tree | |
2884 | * unlock the page if there are no other extents locked for it | |
2885 | * | |
2886 | * Scheduling is not allowed, so the extent state tree is expected | |
2887 | * to have one and only one object corresponding to this IO. | |
2888 | */ | |
4246a0b6 | 2889 | static void end_bio_extent_readpage(struct bio *bio) |
d1310b2e | 2890 | { |
2c30c71b | 2891 | struct bio_vec *bvec; |
4e4cbee9 | 2892 | int uptodate = !bio->bi_status; |
facc8a22 | 2893 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); |
7870d082 | 2894 | struct extent_io_tree *tree, *failure_tree; |
94e8c95c | 2895 | struct processed_extent processed = { 0 }; |
7ffd27e3 QW |
2896 | /* |
2897 | * The offset to the beginning of a bio, since one bio can never be | |
2898 | * larger than UINT_MAX, u32 here is enough. | |
2899 | */ | |
2900 | u32 bio_offset = 0; | |
5cf1ab56 | 2901 | int mirror; |
d1310b2e | 2902 | int ret; |
6dc4f100 | 2903 | struct bvec_iter_all iter_all; |
d1310b2e | 2904 | |
c09abff8 | 2905 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 2906 | bio_for_each_segment_all(bvec, bio, iter_all) { |
d1310b2e | 2907 | struct page *page = bvec->bv_page; |
a71754fc | 2908 | struct inode *inode = page->mapping->host; |
ab8d0fc4 | 2909 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
7ffd27e3 QW |
2910 | const u32 sectorsize = fs_info->sectorsize; |
2911 | u64 start; | |
2912 | u64 end; | |
2913 | u32 len; | |
507903b8 | 2914 | |
ab8d0fc4 JM |
2915 | btrfs_debug(fs_info, |
2916 | "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u", | |
1201b58b | 2917 | bio->bi_iter.bi_sector, bio->bi_status, |
ab8d0fc4 | 2918 | io_bio->mirror_num); |
a71754fc | 2919 | tree = &BTRFS_I(inode)->io_tree; |
7870d082 | 2920 | failure_tree = &BTRFS_I(inode)->io_failure_tree; |
902b22f3 | 2921 | |
8b8bbd46 QW |
2922 | /* |
2923 | * We always issue full-sector reads, but if some block in a | |
2924 | * page fails to read, blk_update_request() will advance | |
2925 | * bv_offset and adjust bv_len to compensate. Print a warning | |
2926 | * for unaligned offsets, and an error if they don't add up to | |
2927 | * a full sector. | |
2928 | */ | |
2929 | if (!IS_ALIGNED(bvec->bv_offset, sectorsize)) | |
2930 | btrfs_err(fs_info, | |
2931 | "partial page read in btrfs with offset %u and length %u", | |
2932 | bvec->bv_offset, bvec->bv_len); | |
2933 | else if (!IS_ALIGNED(bvec->bv_offset + bvec->bv_len, | |
2934 | sectorsize)) | |
2935 | btrfs_info(fs_info, | |
2936 | "incomplete page read with offset %u and length %u", | |
2937 | bvec->bv_offset, bvec->bv_len); | |
2938 | ||
2939 | start = page_offset(page) + bvec->bv_offset; | |
2940 | end = start + bvec->bv_len - 1; | |
facc8a22 | 2941 | len = bvec->bv_len; |
d1310b2e | 2942 | |
9be3395b | 2943 | mirror = io_bio->mirror_num; |
78e62c02 | 2944 | if (likely(uptodate)) { |
be17b3af | 2945 | if (is_data_inode(inode)) |
7ffd27e3 QW |
2946 | ret = btrfs_verify_data_csum(io_bio, |
2947 | bio_offset, page, start, end, | |
2948 | mirror); | |
9a446d6a NB |
2949 | else |
2950 | ret = btrfs_validate_metadata_buffer(io_bio, | |
8e1dc982 | 2951 | page, start, end, mirror); |
5ee0844d | 2952 | if (ret) |
d1310b2e | 2953 | uptodate = 0; |
5ee0844d | 2954 | else |
7870d082 JB |
2955 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
2956 | failure_tree, tree, start, | |
2957 | page, | |
2958 | btrfs_ino(BTRFS_I(inode)), 0); | |
d1310b2e | 2959 | } |
ea466794 | 2960 | |
f2a09da9 MX |
2961 | if (likely(uptodate)) |
2962 | goto readpage_ok; | |
2963 | ||
be17b3af | 2964 | if (is_data_inode(inode)) { |
9d0d1c8b | 2965 | |
f4a8e656 | 2966 | /* |
78e62c02 NB |
2967 | * The generic bio_readpage_error handles errors the |
2968 | * following way: If possible, new read requests are | |
2969 | * created and submitted and will end up in | |
2970 | * end_bio_extent_readpage as well (if we're lucky, | |
2971 | * not in the !uptodate case). In that case it returns | |
2972 | * 0 and we just go on with the next page in our bio. | |
2973 | * If it can't handle the error it will return -EIO and | |
2974 | * we remain responsible for that page. | |
f4a8e656 | 2975 | */ |
7ffd27e3 QW |
2976 | if (!btrfs_submit_read_repair(inode, bio, bio_offset, |
2977 | page, | |
77d5d689 OS |
2978 | start - page_offset(page), |
2979 | start, end, mirror, | |
908930f3 | 2980 | btrfs_submit_data_bio)) { |
78e62c02 | 2981 | uptodate = !bio->bi_status; |
7ffd27e3 QW |
2982 | ASSERT(bio_offset + len > bio_offset); |
2983 | bio_offset += len; | |
78e62c02 NB |
2984 | continue; |
2985 | } | |
2986 | } else { | |
2987 | struct extent_buffer *eb; | |
2988 | ||
2989 | eb = (struct extent_buffer *)page->private; | |
2990 | set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); | |
2991 | eb->read_mirror = mirror; | |
2992 | atomic_dec(&eb->io_pages); | |
2993 | if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, | |
2994 | &eb->bflags)) | |
2995 | btree_readahead_hook(eb, -EIO); | |
7e38326f | 2996 | } |
f2a09da9 | 2997 | readpage_ok: |
883d0de4 | 2998 | if (likely(uptodate)) { |
a71754fc | 2999 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 3000 | pgoff_t end_index = i_size >> PAGE_SHIFT; |
a583c026 | 3001 | unsigned off; |
a71754fc JB |
3002 | |
3003 | /* Zero out the end if this page straddles i_size */ | |
7073017a | 3004 | off = offset_in_page(i_size); |
a583c026 | 3005 | if (page->index == end_index && off) |
09cbfeaf | 3006 | zero_user_segment(page, off, PAGE_SIZE); |
70dec807 | 3007 | } |
7ffd27e3 QW |
3008 | ASSERT(bio_offset + len > bio_offset); |
3009 | bio_offset += len; | |
883d0de4 | 3010 | |
e09caaf9 | 3011 | /* Update page status and unlock */ |
92082d40 | 3012 | end_page_read(page, uptodate, start, len); |
94e8c95c QW |
3013 | endio_readpage_release_extent(&processed, BTRFS_I(inode), |
3014 | start, end, uptodate); | |
2c30c71b | 3015 | } |
94e8c95c QW |
3016 | /* Release the last extent */ |
3017 | endio_readpage_release_extent(&processed, NULL, 0, 0, false); | |
b3a0dd50 | 3018 | btrfs_io_bio_free_csum(io_bio); |
d1310b2e | 3019 | bio_put(bio); |
d1310b2e CM |
3020 | } |
3021 | ||
9be3395b | 3022 | /* |
184f999e DS |
3023 | * Initialize the members up to but not including 'bio'. Use after allocating a |
3024 | * new bio by bio_alloc_bioset as it does not initialize the bytes outside of | |
3025 | * 'bio' because use of __GFP_ZERO is not supported. | |
9be3395b | 3026 | */ |
184f999e | 3027 | static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio) |
d1310b2e | 3028 | { |
184f999e DS |
3029 | memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio)); |
3030 | } | |
d1310b2e | 3031 | |
9be3395b | 3032 | /* |
6e707bcd DS |
3033 | * The following helpers allocate a bio. As it's backed by a bioset, it'll |
3034 | * never fail. We're returning a bio right now but you can call btrfs_io_bio | |
3035 | * for the appropriate container_of magic | |
9be3395b | 3036 | */ |
e749af44 | 3037 | struct bio *btrfs_bio_alloc(u64 first_byte) |
d1310b2e CM |
3038 | { |
3039 | struct bio *bio; | |
d1310b2e | 3040 | |
8ac9f7c1 | 3041 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &btrfs_bioset); |
c821e7f3 | 3042 | bio->bi_iter.bi_sector = first_byte >> 9; |
184f999e | 3043 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
d1310b2e CM |
3044 | return bio; |
3045 | } | |
3046 | ||
8b6c1d56 | 3047 | struct bio *btrfs_bio_clone(struct bio *bio) |
9be3395b | 3048 | { |
23ea8e5a MX |
3049 | struct btrfs_io_bio *btrfs_bio; |
3050 | struct bio *new; | |
9be3395b | 3051 | |
6e707bcd | 3052 | /* Bio allocation backed by a bioset does not fail */ |
8ac9f7c1 | 3053 | new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset); |
6e707bcd | 3054 | btrfs_bio = btrfs_io_bio(new); |
184f999e | 3055 | btrfs_io_bio_init(btrfs_bio); |
6e707bcd | 3056 | btrfs_bio->iter = bio->bi_iter; |
23ea8e5a MX |
3057 | return new; |
3058 | } | |
9be3395b | 3059 | |
c5e4c3d7 | 3060 | struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs) |
9be3395b | 3061 | { |
facc8a22 MX |
3062 | struct bio *bio; |
3063 | ||
6e707bcd | 3064 | /* Bio allocation backed by a bioset does not fail */ |
8ac9f7c1 | 3065 | bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset); |
184f999e | 3066 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
facc8a22 | 3067 | return bio; |
9be3395b CM |
3068 | } |
3069 | ||
e477094f | 3070 | struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size) |
2f8e9140 LB |
3071 | { |
3072 | struct bio *bio; | |
3073 | struct btrfs_io_bio *btrfs_bio; | |
3074 | ||
3075 | /* this will never fail when it's backed by a bioset */ | |
8ac9f7c1 | 3076 | bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset); |
2f8e9140 LB |
3077 | ASSERT(bio); |
3078 | ||
3079 | btrfs_bio = btrfs_io_bio(bio); | |
184f999e | 3080 | btrfs_io_bio_init(btrfs_bio); |
2f8e9140 LB |
3081 | |
3082 | bio_trim(bio, offset >> 9, size >> 9); | |
17347cec | 3083 | btrfs_bio->iter = bio->bi_iter; |
2f8e9140 LB |
3084 | return bio; |
3085 | } | |
9be3395b | 3086 | |
953651eb NA |
3087 | /** |
3088 | * Attempt to add a page to bio | |
3089 | * | |
3090 | * @bio: destination bio | |
3091 | * @page: page to add to the bio | |
3092 | * @disk_bytenr: offset of the new bio or to check whether we are adding | |
3093 | * a contiguous page to the previous one | |
3094 | * @pg_offset: starting offset in the page | |
3095 | * @size: portion of page that we want to write | |
3096 | * @prev_bio_flags: flags of previous bio to see if we can merge the current one | |
3097 | * @bio_flags: flags of the current bio to see if we can merge them | |
3098 | * @return: true if page was added, false otherwise | |
3099 | * | |
3100 | * Attempt to add a page to bio considering stripe alignment etc. | |
3101 | * | |
3102 | * Return true if successfully page added. Otherwise, return false. | |
3103 | */ | |
3104 | static bool btrfs_bio_add_page(struct bio *bio, struct page *page, | |
3105 | u64 disk_bytenr, unsigned int size, | |
3106 | unsigned int pg_offset, | |
3107 | unsigned long prev_bio_flags, | |
3108 | unsigned long bio_flags) | |
3109 | { | |
3110 | const sector_t sector = disk_bytenr >> SECTOR_SHIFT; | |
3111 | bool contig; | |
e1326f03 | 3112 | int ret; |
953651eb NA |
3113 | |
3114 | if (prev_bio_flags != bio_flags) | |
3115 | return false; | |
3116 | ||
3117 | if (prev_bio_flags & EXTENT_BIO_COMPRESSED) | |
3118 | contig = bio->bi_iter.bi_sector == sector; | |
3119 | else | |
3120 | contig = bio_end_sector(bio) == sector; | |
3121 | if (!contig) | |
3122 | return false; | |
3123 | ||
3124 | if (btrfs_bio_fits_in_stripe(page, size, bio, bio_flags)) | |
3125 | return false; | |
3126 | ||
cacb2cea JT |
3127 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) { |
3128 | struct page *first_page = bio_first_bvec_all(bio)->bv_page; | |
3129 | ||
3130 | if (!btrfs_bio_fits_in_ordered_extent(first_page, bio, size)) | |
3131 | return false; | |
e1326f03 | 3132 | ret = bio_add_zone_append_page(bio, page, size, pg_offset); |
cacb2cea | 3133 | } else { |
e1326f03 | 3134 | ret = bio_add_page(bio, page, size, pg_offset); |
cacb2cea | 3135 | } |
e1326f03 NA |
3136 | |
3137 | return ret == size; | |
953651eb NA |
3138 | } |
3139 | ||
4b81ba48 DS |
3140 | /* |
3141 | * @opf: bio REQ_OP_* and REQ_* flags as one value | |
b8b3d625 DS |
3142 | * @wbc: optional writeback control for io accounting |
3143 | * @page: page to add to the bio | |
0c64c33c QW |
3144 | * @disk_bytenr: logical bytenr where the write will be |
3145 | * @size: portion of page that we want to write to | |
b8b3d625 DS |
3146 | * @pg_offset: offset of the new bio or to check whether we are adding |
3147 | * a contiguous page to the previous one | |
5c2b1fd7 | 3148 | * @bio_ret: must be valid pointer, newly allocated bio will be stored there |
b8b3d625 DS |
3149 | * @end_io_func: end_io callback for new bio |
3150 | * @mirror_num: desired mirror to read/write | |
3151 | * @prev_bio_flags: flags of previous bio to see if we can merge the current one | |
3152 | * @bio_flags: flags of the current bio to see if we can merge them | |
4b81ba48 | 3153 | */ |
0ceb34bf | 3154 | static int submit_extent_page(unsigned int opf, |
da2f0f74 | 3155 | struct writeback_control *wbc, |
0c64c33c | 3156 | struct page *page, u64 disk_bytenr, |
6c5a4e2c | 3157 | size_t size, unsigned long pg_offset, |
d1310b2e | 3158 | struct bio **bio_ret, |
f188591e | 3159 | bio_end_io_t end_io_func, |
c8b97818 CM |
3160 | int mirror_num, |
3161 | unsigned long prev_bio_flags, | |
005efedf FM |
3162 | unsigned long bio_flags, |
3163 | bool force_bio_submit) | |
d1310b2e CM |
3164 | { |
3165 | int ret = 0; | |
3166 | struct bio *bio; | |
e940e9a7 | 3167 | size_t io_size = min_t(size_t, size, PAGE_SIZE); |
e1326f03 NA |
3168 | struct btrfs_inode *inode = BTRFS_I(page->mapping->host); |
3169 | struct extent_io_tree *tree = &inode->io_tree; | |
3170 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
d1310b2e | 3171 | |
5c2b1fd7 DS |
3172 | ASSERT(bio_ret); |
3173 | ||
3174 | if (*bio_ret) { | |
d1310b2e | 3175 | bio = *bio_ret; |
953651eb NA |
3176 | if (force_bio_submit || |
3177 | !btrfs_bio_add_page(bio, page, disk_bytenr, io_size, | |
3178 | pg_offset, prev_bio_flags, bio_flags)) { | |
1f7ad75b | 3179 | ret = submit_one_bio(bio, mirror_num, prev_bio_flags); |
289454ad NA |
3180 | if (ret < 0) { |
3181 | *bio_ret = NULL; | |
79787eaa | 3182 | return ret; |
289454ad | 3183 | } |
d1310b2e CM |
3184 | bio = NULL; |
3185 | } else { | |
da2f0f74 | 3186 | if (wbc) |
e940e9a7 | 3187 | wbc_account_cgroup_owner(wbc, page, io_size); |
d1310b2e CM |
3188 | return 0; |
3189 | } | |
3190 | } | |
c8b97818 | 3191 | |
0c64c33c | 3192 | bio = btrfs_bio_alloc(disk_bytenr); |
e940e9a7 | 3193 | bio_add_page(bio, page, io_size, pg_offset); |
d1310b2e CM |
3194 | bio->bi_end_io = end_io_func; |
3195 | bio->bi_private = tree; | |
e6959b93 | 3196 | bio->bi_write_hint = page->mapping->host->i_write_hint; |
4b81ba48 | 3197 | bio->bi_opf = opf; |
da2f0f74 | 3198 | if (wbc) { |
429aebc0 DS |
3199 | struct block_device *bdev; |
3200 | ||
e1326f03 | 3201 | bdev = fs_info->fs_devices->latest_bdev; |
429aebc0 | 3202 | bio_set_dev(bio, bdev); |
da2f0f74 | 3203 | wbc_init_bio(wbc, bio); |
e940e9a7 | 3204 | wbc_account_cgroup_owner(wbc, page, io_size); |
da2f0f74 | 3205 | } |
e1326f03 NA |
3206 | if (btrfs_is_zoned(fs_info) && bio_op(bio) == REQ_OP_ZONE_APPEND) { |
3207 | struct extent_map *em; | |
3208 | struct map_lookup *map; | |
3209 | ||
3210 | em = btrfs_get_chunk_map(fs_info, disk_bytenr, io_size); | |
3211 | if (IS_ERR(em)) | |
3212 | return PTR_ERR(em); | |
3213 | ||
3214 | map = em->map_lookup; | |
3215 | /* We only support single profile for now */ | |
3216 | ASSERT(map->num_stripes == 1); | |
3217 | btrfs_io_bio(bio)->device = map->stripes[0].dev; | |
3218 | ||
3219 | free_extent_map(em); | |
3220 | } | |
70dec807 | 3221 | |
5c2b1fd7 | 3222 | *bio_ret = bio; |
d1310b2e CM |
3223 | |
3224 | return ret; | |
3225 | } | |
3226 | ||
760f991f QW |
3227 | static int attach_extent_buffer_page(struct extent_buffer *eb, |
3228 | struct page *page, | |
3229 | struct btrfs_subpage *prealloc) | |
d1310b2e | 3230 | { |
760f991f QW |
3231 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3232 | int ret = 0; | |
3233 | ||
0d01e247 QW |
3234 | /* |
3235 | * If the page is mapped to btree inode, we should hold the private | |
3236 | * lock to prevent race. | |
3237 | * For cloned or dummy extent buffers, their pages are not mapped and | |
3238 | * will not race with any other ebs. | |
3239 | */ | |
3240 | if (page->mapping) | |
3241 | lockdep_assert_held(&page->mapping->private_lock); | |
3242 | ||
760f991f QW |
3243 | if (fs_info->sectorsize == PAGE_SIZE) { |
3244 | if (!PagePrivate(page)) | |
3245 | attach_page_private(page, eb); | |
3246 | else | |
3247 | WARN_ON(page->private != (unsigned long)eb); | |
3248 | return 0; | |
3249 | } | |
3250 | ||
3251 | /* Already mapped, just free prealloc */ | |
3252 | if (PagePrivate(page)) { | |
3253 | btrfs_free_subpage(prealloc); | |
3254 | return 0; | |
3255 | } | |
3256 | ||
3257 | if (prealloc) | |
3258 | /* Has preallocated memory for subpage */ | |
3259 | attach_page_private(page, prealloc); | |
d1b89bc0 | 3260 | else |
760f991f QW |
3261 | /* Do new allocation to attach subpage */ |
3262 | ret = btrfs_attach_subpage(fs_info, page, | |
3263 | BTRFS_SUBPAGE_METADATA); | |
3264 | return ret; | |
d1310b2e CM |
3265 | } |
3266 | ||
32443de3 | 3267 | int set_page_extent_mapped(struct page *page) |
d1310b2e | 3268 | { |
32443de3 QW |
3269 | struct btrfs_fs_info *fs_info; |
3270 | ||
3271 | ASSERT(page->mapping); | |
3272 | ||
3273 | if (PagePrivate(page)) | |
3274 | return 0; | |
3275 | ||
3276 | fs_info = btrfs_sb(page->mapping->host->i_sb); | |
3277 | ||
3278 | if (fs_info->sectorsize < PAGE_SIZE) | |
3279 | return btrfs_attach_subpage(fs_info, page, BTRFS_SUBPAGE_DATA); | |
3280 | ||
3281 | attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE); | |
3282 | return 0; | |
3283 | } | |
3284 | ||
3285 | void clear_page_extent_mapped(struct page *page) | |
3286 | { | |
3287 | struct btrfs_fs_info *fs_info; | |
3288 | ||
3289 | ASSERT(page->mapping); | |
3290 | ||
d1b89bc0 | 3291 | if (!PagePrivate(page)) |
32443de3 QW |
3292 | return; |
3293 | ||
3294 | fs_info = btrfs_sb(page->mapping->host->i_sb); | |
3295 | if (fs_info->sectorsize < PAGE_SIZE) | |
3296 | return btrfs_detach_subpage(fs_info, page); | |
3297 | ||
3298 | detach_page_private(page); | |
d1310b2e CM |
3299 | } |
3300 | ||
125bac01 MX |
3301 | static struct extent_map * |
3302 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
1a5ee1e6 | 3303 | u64 start, u64 len, struct extent_map **em_cached) |
125bac01 MX |
3304 | { |
3305 | struct extent_map *em; | |
3306 | ||
3307 | if (em_cached && *em_cached) { | |
3308 | em = *em_cached; | |
cbc0e928 | 3309 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 | 3310 | start < extent_map_end(em)) { |
490b54d6 | 3311 | refcount_inc(&em->refs); |
125bac01 MX |
3312 | return em; |
3313 | } | |
3314 | ||
3315 | free_extent_map(em); | |
3316 | *em_cached = NULL; | |
3317 | } | |
3318 | ||
1a5ee1e6 | 3319 | em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len); |
125bac01 MX |
3320 | if (em_cached && !IS_ERR_OR_NULL(em)) { |
3321 | BUG_ON(*em_cached); | |
490b54d6 | 3322 | refcount_inc(&em->refs); |
125bac01 MX |
3323 | *em_cached = em; |
3324 | } | |
3325 | return em; | |
3326 | } | |
d1310b2e CM |
3327 | /* |
3328 | * basic readpage implementation. Locked extent state structs are inserted | |
3329 | * into the tree that are removed when the IO is done (by the end_io | |
3330 | * handlers) | |
79787eaa | 3331 | * XXX JDM: This needs looking at to ensure proper page locking |
baf863b9 | 3332 | * return 0 on success, otherwise return error |
d1310b2e | 3333 | */ |
0f208812 NB |
3334 | int btrfs_do_readpage(struct page *page, struct extent_map **em_cached, |
3335 | struct bio **bio, unsigned long *bio_flags, | |
3336 | unsigned int read_flags, u64 *prev_em_start) | |
d1310b2e CM |
3337 | { |
3338 | struct inode *inode = page->mapping->host; | |
92082d40 | 3339 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4eee4fa4 | 3340 | u64 start = page_offset(page); |
8eec8296 | 3341 | const u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
3342 | u64 cur = start; |
3343 | u64 extent_offset; | |
3344 | u64 last_byte = i_size_read(inode); | |
3345 | u64 block_start; | |
3346 | u64 cur_end; | |
d1310b2e | 3347 | struct extent_map *em; |
baf863b9 | 3348 | int ret = 0; |
d1310b2e | 3349 | int nr = 0; |
306e16ce | 3350 | size_t pg_offset = 0; |
d1310b2e CM |
3351 | size_t iosize; |
3352 | size_t blocksize = inode->i_sb->s_blocksize; | |
7f042a83 | 3353 | unsigned long this_bio_flag = 0; |
f657a31c | 3354 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
ae6957eb | 3355 | |
32443de3 QW |
3356 | ret = set_page_extent_mapped(page); |
3357 | if (ret < 0) { | |
3358 | unlock_extent(tree, start, end); | |
92082d40 QW |
3359 | btrfs_page_set_error(fs_info, page, start, PAGE_SIZE); |
3360 | unlock_page(page); | |
32443de3 QW |
3361 | goto out; |
3362 | } | |
d1310b2e | 3363 | |
90a887c9 DM |
3364 | if (!PageUptodate(page)) { |
3365 | if (cleancache_get_page(page) == 0) { | |
3366 | BUG_ON(blocksize != PAGE_SIZE); | |
9974090b | 3367 | unlock_extent(tree, start, end); |
92082d40 | 3368 | unlock_page(page); |
90a887c9 DM |
3369 | goto out; |
3370 | } | |
3371 | } | |
3372 | ||
09cbfeaf | 3373 | if (page->index == last_byte >> PAGE_SHIFT) { |
c8b97818 | 3374 | char *userpage; |
7073017a | 3375 | size_t zero_offset = offset_in_page(last_byte); |
c8b97818 CM |
3376 | |
3377 | if (zero_offset) { | |
09cbfeaf | 3378 | iosize = PAGE_SIZE - zero_offset; |
7ac687d9 | 3379 | userpage = kmap_atomic(page); |
c8b97818 CM |
3380 | memset(userpage + zero_offset, 0, iosize); |
3381 | flush_dcache_page(page); | |
7ac687d9 | 3382 | kunmap_atomic(userpage); |
c8b97818 CM |
3383 | } |
3384 | } | |
92082d40 | 3385 | begin_page_read(fs_info, page); |
d1310b2e | 3386 | while (cur <= end) { |
005efedf | 3387 | bool force_bio_submit = false; |
0c64c33c | 3388 | u64 disk_bytenr; |
c8f2f24b | 3389 | |
d1310b2e CM |
3390 | if (cur >= last_byte) { |
3391 | char *userpage; | |
507903b8 AJ |
3392 | struct extent_state *cached = NULL; |
3393 | ||
09cbfeaf | 3394 | iosize = PAGE_SIZE - pg_offset; |
7ac687d9 | 3395 | userpage = kmap_atomic(page); |
306e16ce | 3396 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3397 | flush_dcache_page(page); |
7ac687d9 | 3398 | kunmap_atomic(userpage); |
d1310b2e | 3399 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 3400 | &cached, GFP_NOFS); |
7f042a83 | 3401 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3402 | cur + iosize - 1, &cached); |
92082d40 | 3403 | end_page_read(page, true, cur, iosize); |
d1310b2e CM |
3404 | break; |
3405 | } | |
125bac01 | 3406 | em = __get_extent_map(inode, page, pg_offset, cur, |
1a5ee1e6 | 3407 | end - cur + 1, em_cached); |
c704005d | 3408 | if (IS_ERR_OR_NULL(em)) { |
7f042a83 | 3409 | unlock_extent(tree, cur, end); |
92082d40 | 3410 | end_page_read(page, false, cur, end + 1 - cur); |
d1310b2e CM |
3411 | break; |
3412 | } | |
d1310b2e CM |
3413 | extent_offset = cur - em->start; |
3414 | BUG_ON(extent_map_end(em) <= cur); | |
3415 | BUG_ON(end < cur); | |
3416 | ||
261507a0 | 3417 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
4b384318 | 3418 | this_bio_flag |= EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
3419 | extent_set_compress_type(&this_bio_flag, |
3420 | em->compress_type); | |
3421 | } | |
c8b97818 | 3422 | |
d1310b2e CM |
3423 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
3424 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 3425 | iosize = ALIGN(iosize, blocksize); |
949b3273 | 3426 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) |
0c64c33c | 3427 | disk_bytenr = em->block_start; |
949b3273 | 3428 | else |
0c64c33c | 3429 | disk_bytenr = em->block_start + extent_offset; |
d1310b2e | 3430 | block_start = em->block_start; |
d899e052 YZ |
3431 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
3432 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
3433 | |
3434 | /* | |
3435 | * If we have a file range that points to a compressed extent | |
260db43c | 3436 | * and it's followed by a consecutive file range that points |
005efedf FM |
3437 | * to the same compressed extent (possibly with a different |
3438 | * offset and/or length, so it either points to the whole extent | |
3439 | * or only part of it), we must make sure we do not submit a | |
3440 | * single bio to populate the pages for the 2 ranges because | |
3441 | * this makes the compressed extent read zero out the pages | |
3442 | * belonging to the 2nd range. Imagine the following scenario: | |
3443 | * | |
3444 | * File layout | |
3445 | * [0 - 8K] [8K - 24K] | |
3446 | * | | | |
3447 | * | | | |
3448 | * points to extent X, points to extent X, | |
3449 | * offset 4K, length of 8K offset 0, length 16K | |
3450 | * | |
3451 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
3452 | * | |
3453 | * If the bio to read the compressed extent covers both ranges, | |
3454 | * it will decompress extent X into the pages belonging to the | |
3455 | * first range and then it will stop, zeroing out the remaining | |
3456 | * pages that belong to the other range that points to extent X. | |
3457 | * So here we make sure we submit 2 bios, one for the first | |
3458 | * range and another one for the third range. Both will target | |
3459 | * the same physical extent from disk, but we can't currently | |
3460 | * make the compressed bio endio callback populate the pages | |
3461 | * for both ranges because each compressed bio is tightly | |
3462 | * coupled with a single extent map, and each range can have | |
3463 | * an extent map with a different offset value relative to the | |
3464 | * uncompressed data of our extent and different lengths. This | |
3465 | * is a corner case so we prioritize correctness over | |
3466 | * non-optimal behavior (submitting 2 bios for the same extent). | |
3467 | */ | |
3468 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
3469 | prev_em_start && *prev_em_start != (u64)-1 && | |
8e928218 | 3470 | *prev_em_start != em->start) |
005efedf FM |
3471 | force_bio_submit = true; |
3472 | ||
3473 | if (prev_em_start) | |
8e928218 | 3474 | *prev_em_start = em->start; |
005efedf | 3475 | |
d1310b2e CM |
3476 | free_extent_map(em); |
3477 | em = NULL; | |
3478 | ||
3479 | /* we've found a hole, just zero and go on */ | |
3480 | if (block_start == EXTENT_MAP_HOLE) { | |
3481 | char *userpage; | |
507903b8 AJ |
3482 | struct extent_state *cached = NULL; |
3483 | ||
7ac687d9 | 3484 | userpage = kmap_atomic(page); |
306e16ce | 3485 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3486 | flush_dcache_page(page); |
7ac687d9 | 3487 | kunmap_atomic(userpage); |
d1310b2e CM |
3488 | |
3489 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 | 3490 | &cached, GFP_NOFS); |
7f042a83 | 3491 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3492 | cur + iosize - 1, &cached); |
92082d40 | 3493 | end_page_read(page, true, cur, iosize); |
d1310b2e | 3494 | cur = cur + iosize; |
306e16ce | 3495 | pg_offset += iosize; |
d1310b2e CM |
3496 | continue; |
3497 | } | |
3498 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
3499 | if (test_range_bit(tree, cur, cur_end, |
3500 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 3501 | check_page_uptodate(tree, page); |
7f042a83 | 3502 | unlock_extent(tree, cur, cur + iosize - 1); |
92082d40 | 3503 | end_page_read(page, true, cur, iosize); |
d1310b2e | 3504 | cur = cur + iosize; |
306e16ce | 3505 | pg_offset += iosize; |
d1310b2e CM |
3506 | continue; |
3507 | } | |
70dec807 CM |
3508 | /* we have an inline extent but it didn't get marked up |
3509 | * to date. Error out | |
3510 | */ | |
3511 | if (block_start == EXTENT_MAP_INLINE) { | |
7f042a83 | 3512 | unlock_extent(tree, cur, cur + iosize - 1); |
92082d40 | 3513 | end_page_read(page, false, cur, iosize); |
70dec807 | 3514 | cur = cur + iosize; |
306e16ce | 3515 | pg_offset += iosize; |
70dec807 CM |
3516 | continue; |
3517 | } | |
d1310b2e | 3518 | |
0ceb34bf | 3519 | ret = submit_extent_page(REQ_OP_READ | read_flags, NULL, |
0c64c33c | 3520 | page, disk_bytenr, iosize, |
fa17ed06 | 3521 | pg_offset, bio, |
fd513000 | 3522 | end_bio_extent_readpage, 0, |
c8b97818 | 3523 | *bio_flags, |
005efedf FM |
3524 | this_bio_flag, |
3525 | force_bio_submit); | |
c8f2f24b JB |
3526 | if (!ret) { |
3527 | nr++; | |
3528 | *bio_flags = this_bio_flag; | |
3529 | } else { | |
7f042a83 | 3530 | unlock_extent(tree, cur, cur + iosize - 1); |
92082d40 | 3531 | end_page_read(page, false, cur, iosize); |
baf863b9 | 3532 | goto out; |
edd33c99 | 3533 | } |
d1310b2e | 3534 | cur = cur + iosize; |
306e16ce | 3535 | pg_offset += iosize; |
d1310b2e | 3536 | } |
90a887c9 | 3537 | out: |
baf863b9 | 3538 | return ret; |
d1310b2e CM |
3539 | } |
3540 | ||
b6660e80 | 3541 | static inline void contiguous_readpages(struct page *pages[], int nr_pages, |
9974090b | 3542 | u64 start, u64 end, |
125bac01 | 3543 | struct extent_map **em_cached, |
d3fac6ba | 3544 | struct bio **bio, |
1f7ad75b | 3545 | unsigned long *bio_flags, |
808f80b4 | 3546 | u64 *prev_em_start) |
9974090b | 3547 | { |
23d31bd4 | 3548 | struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host); |
9974090b MX |
3549 | int index; |
3550 | ||
b272ae22 | 3551 | btrfs_lock_and_flush_ordered_range(inode, start, end, NULL); |
9974090b MX |
3552 | |
3553 | for (index = 0; index < nr_pages; index++) { | |
0f208812 NB |
3554 | btrfs_do_readpage(pages[index], em_cached, bio, bio_flags, |
3555 | REQ_RAHEAD, prev_em_start); | |
09cbfeaf | 3556 | put_page(pages[index]); |
9974090b MX |
3557 | } |
3558 | } | |
3559 | ||
3d4b9496 | 3560 | static void update_nr_written(struct writeback_control *wbc, |
a9132667 | 3561 | unsigned long nr_written) |
11c8349b CM |
3562 | { |
3563 | wbc->nr_to_write -= nr_written; | |
11c8349b CM |
3564 | } |
3565 | ||
d1310b2e | 3566 | /* |
40f76580 CM |
3567 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
3568 | * | |
5eaad97a | 3569 | * This returns 1 if btrfs_run_delalloc_range function did all the work required |
40f76580 CM |
3570 | * to write the page (copy into inline extent). In this case the IO has |
3571 | * been started and the page is already unlocked. | |
3572 | * | |
3573 | * This returns 0 if all went well (page still locked) | |
3574 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 3575 | */ |
cd4c0bf9 | 3576 | static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode, |
8cc0237a NB |
3577 | struct page *page, struct writeback_control *wbc, |
3578 | u64 delalloc_start, unsigned long *nr_written) | |
40f76580 | 3579 | { |
09cbfeaf | 3580 | u64 page_end = delalloc_start + PAGE_SIZE - 1; |
3522e903 | 3581 | bool found; |
40f76580 CM |
3582 | u64 delalloc_to_write = 0; |
3583 | u64 delalloc_end = 0; | |
3584 | int ret; | |
3585 | int page_started = 0; | |
3586 | ||
40f76580 CM |
3587 | |
3588 | while (delalloc_end < page_end) { | |
cd4c0bf9 | 3589 | found = find_lock_delalloc_range(&inode->vfs_inode, page, |
40f76580 | 3590 | &delalloc_start, |
917aacec | 3591 | &delalloc_end); |
3522e903 | 3592 | if (!found) { |
40f76580 CM |
3593 | delalloc_start = delalloc_end + 1; |
3594 | continue; | |
3595 | } | |
cd4c0bf9 | 3596 | ret = btrfs_run_delalloc_range(inode, page, delalloc_start, |
5eaad97a | 3597 | delalloc_end, &page_started, nr_written, wbc); |
40f76580 CM |
3598 | if (ret) { |
3599 | SetPageError(page); | |
5eaad97a NB |
3600 | /* |
3601 | * btrfs_run_delalloc_range should return < 0 for error | |
3602 | * but just in case, we use > 0 here meaning the IO is | |
3603 | * started, so we don't want to return > 0 unless | |
3604 | * things are going well. | |
40f76580 | 3605 | */ |
b69d1ee9 | 3606 | return ret < 0 ? ret : -EIO; |
40f76580 CM |
3607 | } |
3608 | /* | |
ea1754a0 KS |
3609 | * delalloc_end is already one less than the total length, so |
3610 | * we don't subtract one from PAGE_SIZE | |
40f76580 CM |
3611 | */ |
3612 | delalloc_to_write += (delalloc_end - delalloc_start + | |
ea1754a0 | 3613 | PAGE_SIZE) >> PAGE_SHIFT; |
40f76580 CM |
3614 | delalloc_start = delalloc_end + 1; |
3615 | } | |
3616 | if (wbc->nr_to_write < delalloc_to_write) { | |
3617 | int thresh = 8192; | |
3618 | ||
3619 | if (delalloc_to_write < thresh * 2) | |
3620 | thresh = delalloc_to_write; | |
3621 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3622 | thresh); | |
3623 | } | |
3624 | ||
3625 | /* did the fill delalloc function already unlock and start | |
3626 | * the IO? | |
3627 | */ | |
3628 | if (page_started) { | |
3629 | /* | |
3630 | * we've unlocked the page, so we can't update | |
3631 | * the mapping's writeback index, just update | |
3632 | * nr_to_write. | |
3633 | */ | |
3634 | wbc->nr_to_write -= *nr_written; | |
3635 | return 1; | |
3636 | } | |
3637 | ||
b69d1ee9 | 3638 | return 0; |
40f76580 CM |
3639 | } |
3640 | ||
3641 | /* | |
3642 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3643 | * and does the loop to map the page into extents and bios. | |
3644 | * | |
3645 | * We return 1 if the IO is started and the page is unlocked, | |
3646 | * 0 if all went well (page still locked) | |
3647 | * < 0 if there were errors (page still locked) | |
3648 | */ | |
d4580fe2 | 3649 | static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode, |
40f76580 CM |
3650 | struct page *page, |
3651 | struct writeback_control *wbc, | |
3652 | struct extent_page_data *epd, | |
3653 | loff_t i_size, | |
3654 | unsigned long nr_written, | |
57e5ffeb | 3655 | int *nr_ret) |
d1310b2e | 3656 | { |
6bc5636a | 3657 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
d4580fe2 | 3658 | struct extent_io_tree *tree = &inode->io_tree; |
4eee4fa4 | 3659 | u64 start = page_offset(page); |
6bc5636a | 3660 | u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
3661 | u64 cur = start; |
3662 | u64 extent_offset; | |
d1310b2e | 3663 | u64 block_start; |
d1310b2e | 3664 | struct extent_map *em; |
40f76580 CM |
3665 | int ret = 0; |
3666 | int nr = 0; | |
57e5ffeb | 3667 | const unsigned int write_flags = wbc_to_write_flags(wbc); |
40f76580 | 3668 | bool compressed; |
c8b97818 | 3669 | |
6bc5636a | 3670 | ret = btrfs_writepage_cow_fixup(page, start, end); |
d75855b4 NB |
3671 | if (ret) { |
3672 | /* Fixup worker will requeue */ | |
5ab58055 | 3673 | redirty_page_for_writepage(wbc, page); |
d75855b4 NB |
3674 | update_nr_written(wbc, nr_written); |
3675 | unlock_page(page); | |
3676 | return 1; | |
247e743c CM |
3677 | } |
3678 | ||
11c8349b CM |
3679 | /* |
3680 | * we don't want to touch the inode after unlocking the page, | |
3681 | * so we update the mapping writeback index now | |
3682 | */ | |
3d4b9496 | 3683 | update_nr_written(wbc, nr_written + 1); |
771ed689 | 3684 | |
d1310b2e | 3685 | while (cur <= end) { |
0c64c33c | 3686 | u64 disk_bytenr; |
40f76580 | 3687 | u64 em_end; |
6bc5636a | 3688 | u32 iosize; |
58409edd | 3689 | |
40f76580 | 3690 | if (cur >= i_size) { |
6bc5636a | 3691 | btrfs_writepage_endio_finish_ordered(page, cur, end, 1); |
d1310b2e CM |
3692 | break; |
3693 | } | |
d4580fe2 | 3694 | em = btrfs_get_extent(inode, NULL, 0, cur, end - cur + 1); |
c704005d | 3695 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3696 | SetPageError(page); |
61391d56 | 3697 | ret = PTR_ERR_OR_ZERO(em); |
d1310b2e CM |
3698 | break; |
3699 | } | |
3700 | ||
3701 | extent_offset = cur - em->start; | |
40f76580 | 3702 | em_end = extent_map_end(em); |
6bc5636a QW |
3703 | ASSERT(cur <= em_end); |
3704 | ASSERT(cur < end); | |
3705 | ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize)); | |
3706 | ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize)); | |
d1310b2e | 3707 | block_start = em->block_start; |
c8b97818 | 3708 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
6bc5636a QW |
3709 | disk_bytenr = em->block_start + extent_offset; |
3710 | ||
3711 | /* Note that em_end from extent_map_end() is exclusive */ | |
3712 | iosize = min(em_end, end + 1) - cur; | |
d1310b2e CM |
3713 | free_extent_map(em); |
3714 | em = NULL; | |
3715 | ||
c8b97818 CM |
3716 | /* |
3717 | * compressed and inline extents are written through other | |
3718 | * paths in the FS | |
3719 | */ | |
3720 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3721 | block_start == EXTENT_MAP_INLINE) { |
c8b04030 | 3722 | if (compressed) |
c8b97818 | 3723 | nr++; |
c8b04030 OS |
3724 | else |
3725 | btrfs_writepage_endio_finish_ordered(page, cur, | |
3726 | cur + iosize - 1, 1); | |
c8b97818 | 3727 | cur += iosize; |
d1310b2e CM |
3728 | continue; |
3729 | } | |
c8b97818 | 3730 | |
5cdc84bf | 3731 | btrfs_set_range_writeback(tree, cur, cur + iosize - 1); |
58409edd | 3732 | if (!PageWriteback(page)) { |
d4580fe2 | 3733 | btrfs_err(inode->root->fs_info, |
58409edd DS |
3734 | "page %lu not writeback, cur %llu end %llu", |
3735 | page->index, cur, end); | |
d1310b2e | 3736 | } |
7f3c74fb | 3737 | |
0ceb34bf | 3738 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc, |
6bc5636a QW |
3739 | page, disk_bytenr, iosize, |
3740 | cur - page_offset(page), &epd->bio, | |
58409edd DS |
3741 | end_bio_extent_writepage, |
3742 | 0, 0, 0, false); | |
fe01aa65 | 3743 | if (ret) { |
58409edd | 3744 | SetPageError(page); |
fe01aa65 TK |
3745 | if (PageWriteback(page)) |
3746 | end_page_writeback(page); | |
3747 | } | |
d1310b2e | 3748 | |
6bc5636a | 3749 | cur += iosize; |
d1310b2e CM |
3750 | nr++; |
3751 | } | |
40f76580 | 3752 | *nr_ret = nr; |
40f76580 CM |
3753 | return ret; |
3754 | } | |
3755 | ||
3756 | /* | |
3757 | * the writepage semantics are similar to regular writepage. extent | |
3758 | * records are inserted to lock ranges in the tree, and as dirty areas | |
3759 | * are found, they are marked writeback. Then the lock bits are removed | |
3760 | * and the end_io handler clears the writeback ranges | |
3065976b QW |
3761 | * |
3762 | * Return 0 if everything goes well. | |
3763 | * Return <0 for error. | |
40f76580 CM |
3764 | */ |
3765 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
aab6e9ed | 3766 | struct extent_page_data *epd) |
40f76580 CM |
3767 | { |
3768 | struct inode *inode = page->mapping->host; | |
40f76580 | 3769 | u64 start = page_offset(page); |
09cbfeaf | 3770 | u64 page_end = start + PAGE_SIZE - 1; |
40f76580 CM |
3771 | int ret; |
3772 | int nr = 0; | |
eb70d222 | 3773 | size_t pg_offset; |
40f76580 | 3774 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 3775 | unsigned long end_index = i_size >> PAGE_SHIFT; |
40f76580 CM |
3776 | unsigned long nr_written = 0; |
3777 | ||
40f76580 CM |
3778 | trace___extent_writepage(page, inode, wbc); |
3779 | ||
3780 | WARN_ON(!PageLocked(page)); | |
3781 | ||
3782 | ClearPageError(page); | |
3783 | ||
7073017a | 3784 | pg_offset = offset_in_page(i_size); |
40f76580 CM |
3785 | if (page->index > end_index || |
3786 | (page->index == end_index && !pg_offset)) { | |
09cbfeaf | 3787 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); |
40f76580 CM |
3788 | unlock_page(page); |
3789 | return 0; | |
3790 | } | |
3791 | ||
3792 | if (page->index == end_index) { | |
3793 | char *userpage; | |
3794 | ||
3795 | userpage = kmap_atomic(page); | |
3796 | memset(userpage + pg_offset, 0, | |
09cbfeaf | 3797 | PAGE_SIZE - pg_offset); |
40f76580 CM |
3798 | kunmap_atomic(userpage); |
3799 | flush_dcache_page(page); | |
3800 | } | |
3801 | ||
32443de3 QW |
3802 | ret = set_page_extent_mapped(page); |
3803 | if (ret < 0) { | |
3804 | SetPageError(page); | |
3805 | goto done; | |
3806 | } | |
40f76580 | 3807 | |
7789a55a | 3808 | if (!epd->extent_locked) { |
cd4c0bf9 NB |
3809 | ret = writepage_delalloc(BTRFS_I(inode), page, wbc, start, |
3810 | &nr_written); | |
7789a55a | 3811 | if (ret == 1) |
169d2c87 | 3812 | return 0; |
7789a55a NB |
3813 | if (ret) |
3814 | goto done; | |
3815 | } | |
40f76580 | 3816 | |
d4580fe2 NB |
3817 | ret = __extent_writepage_io(BTRFS_I(inode), page, wbc, epd, i_size, |
3818 | nr_written, &nr); | |
40f76580 | 3819 | if (ret == 1) |
169d2c87 | 3820 | return 0; |
40f76580 | 3821 | |
d1310b2e CM |
3822 | done: |
3823 | if (nr == 0) { | |
3824 | /* make sure the mapping tag for page dirty gets cleared */ | |
3825 | set_page_writeback(page); | |
3826 | end_page_writeback(page); | |
3827 | } | |
61391d56 FM |
3828 | if (PageError(page)) { |
3829 | ret = ret < 0 ? ret : -EIO; | |
3830 | end_extent_writepage(page, ret, start, page_end); | |
3831 | } | |
d1310b2e | 3832 | unlock_page(page); |
3065976b | 3833 | ASSERT(ret <= 0); |
40f76580 | 3834 | return ret; |
d1310b2e CM |
3835 | } |
3836 | ||
fd8b2b61 | 3837 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 3838 | { |
74316201 N |
3839 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
3840 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
3841 | } |
3842 | ||
18dfa711 FM |
3843 | static void end_extent_buffer_writeback(struct extent_buffer *eb) |
3844 | { | |
3845 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
3846 | smp_mb__after_atomic(); | |
3847 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); | |
3848 | } | |
3849 | ||
2e3c2513 | 3850 | /* |
a3efb2f0 | 3851 | * Lock extent buffer status and pages for writeback. |
2e3c2513 | 3852 | * |
a3efb2f0 QW |
3853 | * May try to flush write bio if we can't get the lock. |
3854 | * | |
3855 | * Return 0 if the extent buffer doesn't need to be submitted. | |
3856 | * (E.g. the extent buffer is not dirty) | |
3857 | * Return >0 is the extent buffer is submitted to bio. | |
3858 | * Return <0 if something went wrong, no page is locked. | |
2e3c2513 | 3859 | */ |
9df76fb5 | 3860 | static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb, |
0e378df1 | 3861 | struct extent_page_data *epd) |
0b32f4bb | 3862 | { |
9df76fb5 | 3863 | struct btrfs_fs_info *fs_info = eb->fs_info; |
2e3c2513 | 3864 | int i, num_pages, failed_page_nr; |
0b32f4bb JB |
3865 | int flush = 0; |
3866 | int ret = 0; | |
3867 | ||
3868 | if (!btrfs_try_tree_write_lock(eb)) { | |
f4340622 | 3869 | ret = flush_write_bio(epd); |
2e3c2513 QW |
3870 | if (ret < 0) |
3871 | return ret; | |
3872 | flush = 1; | |
0b32f4bb JB |
3873 | btrfs_tree_lock(eb); |
3874 | } | |
3875 | ||
3876 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3877 | btrfs_tree_unlock(eb); | |
3878 | if (!epd->sync_io) | |
3879 | return 0; | |
3880 | if (!flush) { | |
f4340622 | 3881 | ret = flush_write_bio(epd); |
2e3c2513 QW |
3882 | if (ret < 0) |
3883 | return ret; | |
0b32f4bb JB |
3884 | flush = 1; |
3885 | } | |
a098d8e8 CM |
3886 | while (1) { |
3887 | wait_on_extent_buffer_writeback(eb); | |
3888 | btrfs_tree_lock(eb); | |
3889 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3890 | break; | |
0b32f4bb | 3891 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3892 | } |
3893 | } | |
3894 | ||
51561ffe JB |
3895 | /* |
3896 | * We need to do this to prevent races in people who check if the eb is | |
3897 | * under IO since we can end up having no IO bits set for a short period | |
3898 | * of time. | |
3899 | */ | |
3900 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
3901 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3902 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 3903 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 3904 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
104b4e51 NB |
3905 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
3906 | -eb->len, | |
3907 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 3908 | ret = 1; |
51561ffe JB |
3909 | } else { |
3910 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
3911 | } |
3912 | ||
3913 | btrfs_tree_unlock(eb); | |
3914 | ||
3915 | if (!ret) | |
3916 | return ret; | |
3917 | ||
65ad0104 | 3918 | num_pages = num_extent_pages(eb); |
0b32f4bb | 3919 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3920 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3921 | |
3922 | if (!trylock_page(p)) { | |
3923 | if (!flush) { | |
18dfa711 FM |
3924 | int err; |
3925 | ||
3926 | err = flush_write_bio(epd); | |
3927 | if (err < 0) { | |
3928 | ret = err; | |
2e3c2513 QW |
3929 | failed_page_nr = i; |
3930 | goto err_unlock; | |
3931 | } | |
0b32f4bb JB |
3932 | flush = 1; |
3933 | } | |
3934 | lock_page(p); | |
3935 | } | |
3936 | } | |
3937 | ||
3938 | return ret; | |
2e3c2513 QW |
3939 | err_unlock: |
3940 | /* Unlock already locked pages */ | |
3941 | for (i = 0; i < failed_page_nr; i++) | |
3942 | unlock_page(eb->pages[i]); | |
18dfa711 FM |
3943 | /* |
3944 | * Clear EXTENT_BUFFER_WRITEBACK and wake up anyone waiting on it. | |
3945 | * Also set back EXTENT_BUFFER_DIRTY so future attempts to this eb can | |
3946 | * be made and undo everything done before. | |
3947 | */ | |
3948 | btrfs_tree_lock(eb); | |
3949 | spin_lock(&eb->refs_lock); | |
3950 | set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); | |
3951 | end_extent_buffer_writeback(eb); | |
3952 | spin_unlock(&eb->refs_lock); | |
3953 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, eb->len, | |
3954 | fs_info->dirty_metadata_batch); | |
3955 | btrfs_clear_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); | |
3956 | btrfs_tree_unlock(eb); | |
2e3c2513 | 3957 | return ret; |
0b32f4bb JB |
3958 | } |
3959 | ||
656f30db FM |
3960 | static void set_btree_ioerr(struct page *page) |
3961 | { | |
3962 | struct extent_buffer *eb = (struct extent_buffer *)page->private; | |
eb5b64f1 | 3963 | struct btrfs_fs_info *fs_info; |
656f30db FM |
3964 | |
3965 | SetPageError(page); | |
3966 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) | |
3967 | return; | |
3968 | ||
eb5b64f1 DZ |
3969 | /* |
3970 | * If we error out, we should add back the dirty_metadata_bytes | |
3971 | * to make it consistent. | |
3972 | */ | |
3973 | fs_info = eb->fs_info; | |
3974 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, | |
3975 | eb->len, fs_info->dirty_metadata_batch); | |
3976 | ||
656f30db FM |
3977 | /* |
3978 | * If writeback for a btree extent that doesn't belong to a log tree | |
3979 | * failed, increment the counter transaction->eb_write_errors. | |
3980 | * We do this because while the transaction is running and before it's | |
3981 | * committing (when we call filemap_fdata[write|wait]_range against | |
3982 | * the btree inode), we might have | |
3983 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
3984 | * returns an error or an error happens during writeback, when we're | |
3985 | * committing the transaction we wouldn't know about it, since the pages | |
3986 | * can be no longer dirty nor marked anymore for writeback (if a | |
3987 | * subsequent modification to the extent buffer didn't happen before the | |
3988 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
3989 | * able to find the pages tagged with SetPageError at transaction | |
3990 | * commit time. So if this happens we must abort the transaction, | |
3991 | * otherwise we commit a super block with btree roots that point to | |
3992 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
3993 | * or the content of some node/leaf from a past generation that got | |
3994 | * cowed or deleted and is no longer valid. | |
3995 | * | |
3996 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
3997 | * not be enough - we need to distinguish between log tree extents vs | |
3998 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
3999 | * will catch and clear such errors in the mapping - and that call might | |
4000 | * be from a log sync and not from a transaction commit. Also, checking | |
4001 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
4002 | * not done and would not be reliable - the eb might have been released | |
4003 | * from memory and reading it back again means that flag would not be | |
4004 | * set (since it's a runtime flag, not persisted on disk). | |
4005 | * | |
4006 | * Using the flags below in the btree inode also makes us achieve the | |
4007 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
4008 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
4009 | * is called, the writeback for all dirty pages had already finished | |
4010 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
4011 | * filemap_fdatawait_range() would return success, as it could not know | |
4012 | * that writeback errors happened (the pages were no longer tagged for | |
4013 | * writeback). | |
4014 | */ | |
4015 | switch (eb->log_index) { | |
4016 | case -1: | |
afcdd129 | 4017 | set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags); |
656f30db FM |
4018 | break; |
4019 | case 0: | |
afcdd129 | 4020 | set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags); |
656f30db FM |
4021 | break; |
4022 | case 1: | |
afcdd129 | 4023 | set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags); |
656f30db FM |
4024 | break; |
4025 | default: | |
4026 | BUG(); /* unexpected, logic error */ | |
4027 | } | |
4028 | } | |
4029 | ||
4246a0b6 | 4030 | static void end_bio_extent_buffer_writepage(struct bio *bio) |
0b32f4bb | 4031 | { |
2c30c71b | 4032 | struct bio_vec *bvec; |
0b32f4bb | 4033 | struct extent_buffer *eb; |
2b070cfe | 4034 | int done; |
6dc4f100 | 4035 | struct bvec_iter_all iter_all; |
0b32f4bb | 4036 | |
c09abff8 | 4037 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 4038 | bio_for_each_segment_all(bvec, bio, iter_all) { |
0b32f4bb JB |
4039 | struct page *page = bvec->bv_page; |
4040 | ||
0b32f4bb JB |
4041 | eb = (struct extent_buffer *)page->private; |
4042 | BUG_ON(!eb); | |
4043 | done = atomic_dec_and_test(&eb->io_pages); | |
4044 | ||
4e4cbee9 | 4045 | if (bio->bi_status || |
4246a0b6 | 4046 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { |
0b32f4bb | 4047 | ClearPageUptodate(page); |
656f30db | 4048 | set_btree_ioerr(page); |
0b32f4bb JB |
4049 | } |
4050 | ||
4051 | end_page_writeback(page); | |
4052 | ||
4053 | if (!done) | |
4054 | continue; | |
4055 | ||
4056 | end_extent_buffer_writeback(eb); | |
2c30c71b | 4057 | } |
0b32f4bb JB |
4058 | |
4059 | bio_put(bio); | |
0b32f4bb JB |
4060 | } |
4061 | ||
0e378df1 | 4062 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb JB |
4063 | struct writeback_control *wbc, |
4064 | struct extent_page_data *epd) | |
4065 | { | |
0c64c33c | 4066 | u64 disk_bytenr = eb->start; |
851cd173 | 4067 | u32 nritems; |
cc5e31a4 | 4068 | int i, num_pages; |
851cd173 | 4069 | unsigned long start, end; |
ff40adf7 | 4070 | unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META; |
d7dbe9e7 | 4071 | int ret = 0; |
0b32f4bb | 4072 | |
656f30db | 4073 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); |
65ad0104 | 4074 | num_pages = num_extent_pages(eb); |
0b32f4bb | 4075 | atomic_set(&eb->io_pages, num_pages); |
de0022b9 | 4076 | |
851cd173 LB |
4077 | /* set btree blocks beyond nritems with 0 to avoid stale content. */ |
4078 | nritems = btrfs_header_nritems(eb); | |
3eb548ee | 4079 | if (btrfs_header_level(eb) > 0) { |
3eb548ee LB |
4080 | end = btrfs_node_key_ptr_offset(nritems); |
4081 | ||
b159fa28 | 4082 | memzero_extent_buffer(eb, end, eb->len - end); |
851cd173 LB |
4083 | } else { |
4084 | /* | |
4085 | * leaf: | |
4086 | * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0 | |
4087 | */ | |
4088 | start = btrfs_item_nr_offset(nritems); | |
8f881e8c | 4089 | end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(eb); |
b159fa28 | 4090 | memzero_extent_buffer(eb, start, end - start); |
3eb548ee LB |
4091 | } |
4092 | ||
0b32f4bb | 4093 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 4094 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
4095 | |
4096 | clear_page_dirty_for_io(p); | |
4097 | set_page_writeback(p); | |
0ceb34bf | 4098 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc, |
0c64c33c | 4099 | p, disk_bytenr, PAGE_SIZE, 0, |
c2df8bb4 | 4100 | &epd->bio, |
1f7ad75b | 4101 | end_bio_extent_buffer_writepage, |
18fdc679 | 4102 | 0, 0, 0, false); |
0b32f4bb | 4103 | if (ret) { |
656f30db | 4104 | set_btree_ioerr(p); |
fe01aa65 TK |
4105 | if (PageWriteback(p)) |
4106 | end_page_writeback(p); | |
0b32f4bb JB |
4107 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
4108 | end_extent_buffer_writeback(eb); | |
4109 | ret = -EIO; | |
4110 | break; | |
4111 | } | |
0c64c33c | 4112 | disk_bytenr += PAGE_SIZE; |
3d4b9496 | 4113 | update_nr_written(wbc, 1); |
0b32f4bb JB |
4114 | unlock_page(p); |
4115 | } | |
4116 | ||
4117 | if (unlikely(ret)) { | |
4118 | for (; i < num_pages; i++) { | |
bbf65cf0 | 4119 | struct page *p = eb->pages[i]; |
81465028 | 4120 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
4121 | unlock_page(p); |
4122 | } | |
4123 | } | |
4124 | ||
4125 | return ret; | |
4126 | } | |
4127 | ||
f91e0d0c QW |
4128 | /* |
4129 | * Submit all page(s) of one extent buffer. | |
4130 | * | |
4131 | * @page: the page of one extent buffer | |
4132 | * @eb_context: to determine if we need to submit this page, if current page | |
4133 | * belongs to this eb, we don't need to submit | |
4134 | * | |
4135 | * The caller should pass each page in their bytenr order, and here we use | |
4136 | * @eb_context to determine if we have submitted pages of one extent buffer. | |
4137 | * | |
4138 | * If we have, we just skip until we hit a new page that doesn't belong to | |
4139 | * current @eb_context. | |
4140 | * | |
4141 | * If not, we submit all the page(s) of the extent buffer. | |
4142 | * | |
4143 | * Return >0 if we have submitted the extent buffer successfully. | |
4144 | * Return 0 if we don't need to submit the page, as it's already submitted by | |
4145 | * previous call. | |
4146 | * Return <0 for fatal error. | |
4147 | */ | |
4148 | static int submit_eb_page(struct page *page, struct writeback_control *wbc, | |
4149 | struct extent_page_data *epd, | |
4150 | struct extent_buffer **eb_context) | |
4151 | { | |
4152 | struct address_space *mapping = page->mapping; | |
4153 | struct extent_buffer *eb; | |
4154 | int ret; | |
4155 | ||
4156 | if (!PagePrivate(page)) | |
4157 | return 0; | |
4158 | ||
4159 | spin_lock(&mapping->private_lock); | |
4160 | if (!PagePrivate(page)) { | |
4161 | spin_unlock(&mapping->private_lock); | |
4162 | return 0; | |
4163 | } | |
4164 | ||
4165 | eb = (struct extent_buffer *)page->private; | |
4166 | ||
4167 | /* | |
4168 | * Shouldn't happen and normally this would be a BUG_ON but no point | |
4169 | * crashing the machine for something we can survive anyway. | |
4170 | */ | |
4171 | if (WARN_ON(!eb)) { | |
4172 | spin_unlock(&mapping->private_lock); | |
4173 | return 0; | |
4174 | } | |
4175 | ||
4176 | if (eb == *eb_context) { | |
4177 | spin_unlock(&mapping->private_lock); | |
4178 | return 0; | |
4179 | } | |
4180 | ret = atomic_inc_not_zero(&eb->refs); | |
4181 | spin_unlock(&mapping->private_lock); | |
4182 | if (!ret) | |
4183 | return 0; | |
4184 | ||
4185 | *eb_context = eb; | |
4186 | ||
4187 | ret = lock_extent_buffer_for_io(eb, epd); | |
4188 | if (ret <= 0) { | |
4189 | free_extent_buffer(eb); | |
4190 | return ret; | |
4191 | } | |
4192 | ret = write_one_eb(eb, wbc, epd); | |
4193 | free_extent_buffer(eb); | |
4194 | if (ret < 0) | |
4195 | return ret; | |
4196 | return 1; | |
4197 | } | |
4198 | ||
0b32f4bb JB |
4199 | int btree_write_cache_pages(struct address_space *mapping, |
4200 | struct writeback_control *wbc) | |
4201 | { | |
f91e0d0c | 4202 | struct extent_buffer *eb_context = NULL; |
0b32f4bb JB |
4203 | struct extent_page_data epd = { |
4204 | .bio = NULL, | |
0b32f4bb JB |
4205 | .extent_locked = 0, |
4206 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
4207 | }; | |
b3ff8f1d | 4208 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; |
0b32f4bb JB |
4209 | int ret = 0; |
4210 | int done = 0; | |
4211 | int nr_to_write_done = 0; | |
4212 | struct pagevec pvec; | |
4213 | int nr_pages; | |
4214 | pgoff_t index; | |
4215 | pgoff_t end; /* Inclusive */ | |
4216 | int scanned = 0; | |
10bbd235 | 4217 | xa_mark_t tag; |
0b32f4bb | 4218 | |
86679820 | 4219 | pagevec_init(&pvec); |
0b32f4bb JB |
4220 | if (wbc->range_cyclic) { |
4221 | index = mapping->writeback_index; /* Start from prev offset */ | |
4222 | end = -1; | |
556755a8 JB |
4223 | /* |
4224 | * Start from the beginning does not need to cycle over the | |
4225 | * range, mark it as scanned. | |
4226 | */ | |
4227 | scanned = (index == 0); | |
0b32f4bb | 4228 | } else { |
09cbfeaf KS |
4229 | index = wbc->range_start >> PAGE_SHIFT; |
4230 | end = wbc->range_end >> PAGE_SHIFT; | |
0b32f4bb JB |
4231 | scanned = 1; |
4232 | } | |
4233 | if (wbc->sync_mode == WB_SYNC_ALL) | |
4234 | tag = PAGECACHE_TAG_TOWRITE; | |
4235 | else | |
4236 | tag = PAGECACHE_TAG_DIRTY; | |
4237 | retry: | |
4238 | if (wbc->sync_mode == WB_SYNC_ALL) | |
4239 | tag_pages_for_writeback(mapping, index, end); | |
4240 | while (!done && !nr_to_write_done && (index <= end) && | |
4006f437 | 4241 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 4242 | tag))) { |
0b32f4bb JB |
4243 | unsigned i; |
4244 | ||
0b32f4bb JB |
4245 | for (i = 0; i < nr_pages; i++) { |
4246 | struct page *page = pvec.pages[i]; | |
4247 | ||
f91e0d0c QW |
4248 | ret = submit_eb_page(page, wbc, &epd, &eb_context); |
4249 | if (ret == 0) | |
0b32f4bb | 4250 | continue; |
f91e0d0c | 4251 | if (ret < 0) { |
0b32f4bb | 4252 | done = 1; |
0b32f4bb JB |
4253 | break; |
4254 | } | |
0b32f4bb JB |
4255 | |
4256 | /* | |
4257 | * the filesystem may choose to bump up nr_to_write. | |
4258 | * We have to make sure to honor the new nr_to_write | |
4259 | * at any time | |
4260 | */ | |
4261 | nr_to_write_done = wbc->nr_to_write <= 0; | |
4262 | } | |
4263 | pagevec_release(&pvec); | |
4264 | cond_resched(); | |
4265 | } | |
4266 | if (!scanned && !done) { | |
4267 | /* | |
4268 | * We hit the last page and there is more work to be done: wrap | |
4269 | * back to the start of the file | |
4270 | */ | |
4271 | scanned = 1; | |
4272 | index = 0; | |
4273 | goto retry; | |
4274 | } | |
2b952eea QW |
4275 | if (ret < 0) { |
4276 | end_write_bio(&epd, ret); | |
4277 | return ret; | |
4278 | } | |
b3ff8f1d QW |
4279 | /* |
4280 | * If something went wrong, don't allow any metadata write bio to be | |
4281 | * submitted. | |
4282 | * | |
4283 | * This would prevent use-after-free if we had dirty pages not | |
4284 | * cleaned up, which can still happen by fuzzed images. | |
4285 | * | |
4286 | * - Bad extent tree | |
4287 | * Allowing existing tree block to be allocated for other trees. | |
4288 | * | |
4289 | * - Log tree operations | |
4290 | * Exiting tree blocks get allocated to log tree, bumps its | |
4291 | * generation, then get cleaned in tree re-balance. | |
4292 | * Such tree block will not be written back, since it's clean, | |
4293 | * thus no WRITTEN flag set. | |
4294 | * And after log writes back, this tree block is not traced by | |
4295 | * any dirty extent_io_tree. | |
4296 | * | |
4297 | * - Offending tree block gets re-dirtied from its original owner | |
4298 | * Since it has bumped generation, no WRITTEN flag, it can be | |
4299 | * reused without COWing. This tree block will not be traced | |
4300 | * by btrfs_transaction::dirty_pages. | |
4301 | * | |
4302 | * Now such dirty tree block will not be cleaned by any dirty | |
4303 | * extent io tree. Thus we don't want to submit such wild eb | |
4304 | * if the fs already has error. | |
4305 | */ | |
4306 | if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { | |
4307 | ret = flush_write_bio(&epd); | |
4308 | } else { | |
fbabd4a3 | 4309 | ret = -EROFS; |
b3ff8f1d QW |
4310 | end_write_bio(&epd, ret); |
4311 | } | |
0b32f4bb JB |
4312 | return ret; |
4313 | } | |
4314 | ||
d1310b2e | 4315 | /** |
3bed2da1 NB |
4316 | * Walk the list of dirty pages of the given address space and write all of them. |
4317 | * | |
d1310b2e | 4318 | * @mapping: address space structure to write |
3bed2da1 NB |
4319 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write |
4320 | * @epd: holds context for the write, namely the bio | |
d1310b2e CM |
4321 | * |
4322 | * If a page is already under I/O, write_cache_pages() skips it, even | |
4323 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
4324 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
4325 | * and msync() need to guarantee that all the data which was dirty at the time | |
4326 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
4327 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
4328 | * existing IO to complete. | |
4329 | */ | |
4242b64a | 4330 | static int extent_write_cache_pages(struct address_space *mapping, |
4bef0848 | 4331 | struct writeback_control *wbc, |
aab6e9ed | 4332 | struct extent_page_data *epd) |
d1310b2e | 4333 | { |
7fd1a3f7 | 4334 | struct inode *inode = mapping->host; |
d1310b2e CM |
4335 | int ret = 0; |
4336 | int done = 0; | |
f85d7d6c | 4337 | int nr_to_write_done = 0; |
d1310b2e CM |
4338 | struct pagevec pvec; |
4339 | int nr_pages; | |
4340 | pgoff_t index; | |
4341 | pgoff_t end; /* Inclusive */ | |
a9132667 LB |
4342 | pgoff_t done_index; |
4343 | int range_whole = 0; | |
d1310b2e | 4344 | int scanned = 0; |
10bbd235 | 4345 | xa_mark_t tag; |
d1310b2e | 4346 | |
7fd1a3f7 JB |
4347 | /* |
4348 | * We have to hold onto the inode so that ordered extents can do their | |
4349 | * work when the IO finishes. The alternative to this is failing to add | |
4350 | * an ordered extent if the igrab() fails there and that is a huge pain | |
4351 | * to deal with, so instead just hold onto the inode throughout the | |
4352 | * writepages operation. If it fails here we are freeing up the inode | |
4353 | * anyway and we'd rather not waste our time writing out stuff that is | |
4354 | * going to be truncated anyway. | |
4355 | */ | |
4356 | if (!igrab(inode)) | |
4357 | return 0; | |
4358 | ||
86679820 | 4359 | pagevec_init(&pvec); |
d1310b2e CM |
4360 | if (wbc->range_cyclic) { |
4361 | index = mapping->writeback_index; /* Start from prev offset */ | |
4362 | end = -1; | |
556755a8 JB |
4363 | /* |
4364 | * Start from the beginning does not need to cycle over the | |
4365 | * range, mark it as scanned. | |
4366 | */ | |
4367 | scanned = (index == 0); | |
d1310b2e | 4368 | } else { |
09cbfeaf KS |
4369 | index = wbc->range_start >> PAGE_SHIFT; |
4370 | end = wbc->range_end >> PAGE_SHIFT; | |
a9132667 LB |
4371 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
4372 | range_whole = 1; | |
d1310b2e CM |
4373 | scanned = 1; |
4374 | } | |
3cd24c69 EL |
4375 | |
4376 | /* | |
4377 | * We do the tagged writepage as long as the snapshot flush bit is set | |
4378 | * and we are the first one who do the filemap_flush() on this inode. | |
4379 | * | |
4380 | * The nr_to_write == LONG_MAX is needed to make sure other flushers do | |
4381 | * not race in and drop the bit. | |
4382 | */ | |
4383 | if (range_whole && wbc->nr_to_write == LONG_MAX && | |
4384 | test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
4385 | &BTRFS_I(inode)->runtime_flags)) | |
4386 | wbc->tagged_writepages = 1; | |
4387 | ||
4388 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) | |
f7aaa06b JB |
4389 | tag = PAGECACHE_TAG_TOWRITE; |
4390 | else | |
4391 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 4392 | retry: |
3cd24c69 | 4393 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
f7aaa06b | 4394 | tag_pages_for_writeback(mapping, index, end); |
a9132667 | 4395 | done_index = index; |
f85d7d6c | 4396 | while (!done && !nr_to_write_done && (index <= end) && |
67fd707f JK |
4397 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, |
4398 | &index, end, tag))) { | |
d1310b2e CM |
4399 | unsigned i; |
4400 | ||
d1310b2e CM |
4401 | for (i = 0; i < nr_pages; i++) { |
4402 | struct page *page = pvec.pages[i]; | |
4403 | ||
f7bddf1e | 4404 | done_index = page->index + 1; |
d1310b2e | 4405 | /* |
b93b0163 MW |
4406 | * At this point we hold neither the i_pages lock nor |
4407 | * the page lock: the page may be truncated or | |
4408 | * invalidated (changing page->mapping to NULL), | |
4409 | * or even swizzled back from swapper_space to | |
4410 | * tmpfs file mapping | |
d1310b2e | 4411 | */ |
c8f2f24b | 4412 | if (!trylock_page(page)) { |
f4340622 QW |
4413 | ret = flush_write_bio(epd); |
4414 | BUG_ON(ret < 0); | |
c8f2f24b | 4415 | lock_page(page); |
01d658f2 | 4416 | } |
d1310b2e CM |
4417 | |
4418 | if (unlikely(page->mapping != mapping)) { | |
4419 | unlock_page(page); | |
4420 | continue; | |
4421 | } | |
4422 | ||
d2c3f4f6 | 4423 | if (wbc->sync_mode != WB_SYNC_NONE) { |
f4340622 QW |
4424 | if (PageWriteback(page)) { |
4425 | ret = flush_write_bio(epd); | |
4426 | BUG_ON(ret < 0); | |
4427 | } | |
d1310b2e | 4428 | wait_on_page_writeback(page); |
d2c3f4f6 | 4429 | } |
d1310b2e CM |
4430 | |
4431 | if (PageWriteback(page) || | |
4432 | !clear_page_dirty_for_io(page)) { | |
4433 | unlock_page(page); | |
4434 | continue; | |
4435 | } | |
4436 | ||
aab6e9ed | 4437 | ret = __extent_writepage(page, wbc, epd); |
a9132667 | 4438 | if (ret < 0) { |
a9132667 LB |
4439 | done = 1; |
4440 | break; | |
4441 | } | |
f85d7d6c CM |
4442 | |
4443 | /* | |
4444 | * the filesystem may choose to bump up nr_to_write. | |
4445 | * We have to make sure to honor the new nr_to_write | |
4446 | * at any time | |
4447 | */ | |
4448 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
4449 | } |
4450 | pagevec_release(&pvec); | |
4451 | cond_resched(); | |
4452 | } | |
894b36e3 | 4453 | if (!scanned && !done) { |
d1310b2e CM |
4454 | /* |
4455 | * We hit the last page and there is more work to be done: wrap | |
4456 | * back to the start of the file | |
4457 | */ | |
4458 | scanned = 1; | |
4459 | index = 0; | |
42ffb0bf JB |
4460 | |
4461 | /* | |
4462 | * If we're looping we could run into a page that is locked by a | |
4463 | * writer and that writer could be waiting on writeback for a | |
4464 | * page in our current bio, and thus deadlock, so flush the | |
4465 | * write bio here. | |
4466 | */ | |
4467 | ret = flush_write_bio(epd); | |
4468 | if (!ret) | |
4469 | goto retry; | |
d1310b2e | 4470 | } |
a9132667 LB |
4471 | |
4472 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole)) | |
4473 | mapping->writeback_index = done_index; | |
4474 | ||
7fd1a3f7 | 4475 | btrfs_add_delayed_iput(inode); |
894b36e3 | 4476 | return ret; |
d1310b2e | 4477 | } |
d1310b2e | 4478 | |
0a9b0e53 | 4479 | int extent_write_full_page(struct page *page, struct writeback_control *wbc) |
d1310b2e CM |
4480 | { |
4481 | int ret; | |
d1310b2e CM |
4482 | struct extent_page_data epd = { |
4483 | .bio = NULL, | |
771ed689 | 4484 | .extent_locked = 0, |
ffbd517d | 4485 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e | 4486 | }; |
d1310b2e | 4487 | |
d1310b2e | 4488 | ret = __extent_writepage(page, wbc, &epd); |
3065976b QW |
4489 | ASSERT(ret <= 0); |
4490 | if (ret < 0) { | |
4491 | end_write_bio(&epd, ret); | |
4492 | return ret; | |
4493 | } | |
d1310b2e | 4494 | |
3065976b QW |
4495 | ret = flush_write_bio(&epd); |
4496 | ASSERT(ret <= 0); | |
d1310b2e CM |
4497 | return ret; |
4498 | } | |
d1310b2e | 4499 | |
5e3ee236 | 4500 | int extent_write_locked_range(struct inode *inode, u64 start, u64 end, |
771ed689 CM |
4501 | int mode) |
4502 | { | |
4503 | int ret = 0; | |
4504 | struct address_space *mapping = inode->i_mapping; | |
4505 | struct page *page; | |
09cbfeaf KS |
4506 | unsigned long nr_pages = (end - start + PAGE_SIZE) >> |
4507 | PAGE_SHIFT; | |
771ed689 CM |
4508 | |
4509 | struct extent_page_data epd = { | |
4510 | .bio = NULL, | |
771ed689 | 4511 | .extent_locked = 1, |
ffbd517d | 4512 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
4513 | }; |
4514 | struct writeback_control wbc_writepages = { | |
771ed689 | 4515 | .sync_mode = mode, |
771ed689 CM |
4516 | .nr_to_write = nr_pages * 2, |
4517 | .range_start = start, | |
4518 | .range_end = end + 1, | |
ec39f769 CM |
4519 | /* We're called from an async helper function */ |
4520 | .punt_to_cgroup = 1, | |
4521 | .no_cgroup_owner = 1, | |
771ed689 CM |
4522 | }; |
4523 | ||
dbb70bec | 4524 | wbc_attach_fdatawrite_inode(&wbc_writepages, inode); |
d397712b | 4525 | while (start <= end) { |
09cbfeaf | 4526 | page = find_get_page(mapping, start >> PAGE_SHIFT); |
771ed689 CM |
4527 | if (clear_page_dirty_for_io(page)) |
4528 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
4529 | else { | |
7087a9d8 | 4530 | btrfs_writepage_endio_finish_ordered(page, start, |
c629732d | 4531 | start + PAGE_SIZE - 1, 1); |
771ed689 CM |
4532 | unlock_page(page); |
4533 | } | |
09cbfeaf KS |
4534 | put_page(page); |
4535 | start += PAGE_SIZE; | |
771ed689 CM |
4536 | } |
4537 | ||
02c6db4f | 4538 | ASSERT(ret <= 0); |
dbb70bec CM |
4539 | if (ret == 0) |
4540 | ret = flush_write_bio(&epd); | |
4541 | else | |
02c6db4f | 4542 | end_write_bio(&epd, ret); |
dbb70bec CM |
4543 | |
4544 | wbc_detach_inode(&wbc_writepages); | |
771ed689 CM |
4545 | return ret; |
4546 | } | |
d1310b2e | 4547 | |
8ae225a8 | 4548 | int extent_writepages(struct address_space *mapping, |
d1310b2e CM |
4549 | struct writeback_control *wbc) |
4550 | { | |
4551 | int ret = 0; | |
4552 | struct extent_page_data epd = { | |
4553 | .bio = NULL, | |
771ed689 | 4554 | .extent_locked = 0, |
ffbd517d | 4555 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
4556 | }; |
4557 | ||
935db853 | 4558 | ret = extent_write_cache_pages(mapping, wbc, &epd); |
a2a72fbd QW |
4559 | ASSERT(ret <= 0); |
4560 | if (ret < 0) { | |
4561 | end_write_bio(&epd, ret); | |
4562 | return ret; | |
4563 | } | |
4564 | ret = flush_write_bio(&epd); | |
d1310b2e CM |
4565 | return ret; |
4566 | } | |
d1310b2e | 4567 | |
ba206a02 | 4568 | void extent_readahead(struct readahead_control *rac) |
d1310b2e CM |
4569 | { |
4570 | struct bio *bio = NULL; | |
c8b97818 | 4571 | unsigned long bio_flags = 0; |
67c9684f | 4572 | struct page *pagepool[16]; |
125bac01 | 4573 | struct extent_map *em_cached = NULL; |
808f80b4 | 4574 | u64 prev_em_start = (u64)-1; |
ba206a02 | 4575 | int nr; |
d1310b2e | 4576 | |
ba206a02 MWO |
4577 | while ((nr = readahead_page_batch(rac, pagepool))) { |
4578 | u64 contig_start = page_offset(pagepool[0]); | |
4579 | u64 contig_end = page_offset(pagepool[nr - 1]) + PAGE_SIZE - 1; | |
e65ef21e | 4580 | |
ba206a02 | 4581 | ASSERT(contig_start + nr * PAGE_SIZE - 1 == contig_end); |
e65ef21e | 4582 | |
ba206a02 MWO |
4583 | contiguous_readpages(pagepool, nr, contig_start, contig_end, |
4584 | &em_cached, &bio, &bio_flags, &prev_em_start); | |
d1310b2e | 4585 | } |
67c9684f | 4586 | |
125bac01 MX |
4587 | if (em_cached) |
4588 | free_extent_map(em_cached); | |
4589 | ||
ba206a02 MWO |
4590 | if (bio) { |
4591 | if (submit_one_bio(bio, 0, bio_flags)) | |
4592 | return; | |
4593 | } | |
d1310b2e | 4594 | } |
d1310b2e CM |
4595 | |
4596 | /* | |
4597 | * basic invalidatepage code, this waits on any locked or writeback | |
4598 | * ranges corresponding to the page, and then deletes any extent state | |
4599 | * records from the tree | |
4600 | */ | |
4601 | int extent_invalidatepage(struct extent_io_tree *tree, | |
4602 | struct page *page, unsigned long offset) | |
4603 | { | |
2ac55d41 | 4604 | struct extent_state *cached_state = NULL; |
4eee4fa4 | 4605 | u64 start = page_offset(page); |
09cbfeaf | 4606 | u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
4607 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; |
4608 | ||
829ddec9 QW |
4609 | /* This function is only called for the btree inode */ |
4610 | ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO); | |
4611 | ||
fda2832f | 4612 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
4613 | if (start > end) |
4614 | return 0; | |
4615 | ||
ff13db41 | 4616 | lock_extent_bits(tree, start, end, &cached_state); |
1edbb734 | 4617 | wait_on_page_writeback(page); |
829ddec9 QW |
4618 | |
4619 | /* | |
4620 | * Currently for btree io tree, only EXTENT_LOCKED is utilized, | |
4621 | * so here we only need to unlock the extent range to free any | |
4622 | * existing extent state. | |
4623 | */ | |
4624 | unlock_extent_cached(tree, start, end, &cached_state); | |
d1310b2e CM |
4625 | return 0; |
4626 | } | |
d1310b2e | 4627 | |
7b13b7b1 CM |
4628 | /* |
4629 | * a helper for releasepage, this tests for areas of the page that | |
4630 | * are locked or under IO and drops the related state bits if it is safe | |
4631 | * to drop the page. | |
4632 | */ | |
29c68b2d | 4633 | static int try_release_extent_state(struct extent_io_tree *tree, |
48a3b636 | 4634 | struct page *page, gfp_t mask) |
7b13b7b1 | 4635 | { |
4eee4fa4 | 4636 | u64 start = page_offset(page); |
09cbfeaf | 4637 | u64 end = start + PAGE_SIZE - 1; |
7b13b7b1 CM |
4638 | int ret = 1; |
4639 | ||
8882679e | 4640 | if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) { |
7b13b7b1 | 4641 | ret = 0; |
8882679e | 4642 | } else { |
11ef160f | 4643 | /* |
2766ff61 FM |
4644 | * At this point we can safely clear everything except the |
4645 | * locked bit, the nodatasum bit and the delalloc new bit. | |
4646 | * The delalloc new bit will be cleared by ordered extent | |
4647 | * completion. | |
11ef160f | 4648 | */ |
66b0c887 | 4649 | ret = __clear_extent_bit(tree, start, end, |
2766ff61 FM |
4650 | ~(EXTENT_LOCKED | EXTENT_NODATASUM | EXTENT_DELALLOC_NEW), |
4651 | 0, 0, NULL, mask, NULL); | |
e3f24cc5 CM |
4652 | |
4653 | /* if clear_extent_bit failed for enomem reasons, | |
4654 | * we can't allow the release to continue. | |
4655 | */ | |
4656 | if (ret < 0) | |
4657 | ret = 0; | |
4658 | else | |
4659 | ret = 1; | |
7b13b7b1 CM |
4660 | } |
4661 | return ret; | |
4662 | } | |
7b13b7b1 | 4663 | |
d1310b2e CM |
4664 | /* |
4665 | * a helper for releasepage. As long as there are no locked extents | |
4666 | * in the range corresponding to the page, both state records and extent | |
4667 | * map records are removed | |
4668 | */ | |
477a30ba | 4669 | int try_release_extent_mapping(struct page *page, gfp_t mask) |
d1310b2e CM |
4670 | { |
4671 | struct extent_map *em; | |
4eee4fa4 | 4672 | u64 start = page_offset(page); |
09cbfeaf | 4673 | u64 end = start + PAGE_SIZE - 1; |
bd3599a0 FM |
4674 | struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host); |
4675 | struct extent_io_tree *tree = &btrfs_inode->io_tree; | |
4676 | struct extent_map_tree *map = &btrfs_inode->extent_tree; | |
7b13b7b1 | 4677 | |
d0164adc | 4678 | if (gfpflags_allow_blocking(mask) && |
ee22184b | 4679 | page->mapping->host->i_size > SZ_16M) { |
39b5637f | 4680 | u64 len; |
70dec807 | 4681 | while (start <= end) { |
fbc2bd7e FM |
4682 | struct btrfs_fs_info *fs_info; |
4683 | u64 cur_gen; | |
4684 | ||
39b5637f | 4685 | len = end - start + 1; |
890871be | 4686 | write_lock(&map->lock); |
39b5637f | 4687 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 4688 | if (!em) { |
890871be | 4689 | write_unlock(&map->lock); |
70dec807 CM |
4690 | break; |
4691 | } | |
7f3c74fb CM |
4692 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
4693 | em->start != start) { | |
890871be | 4694 | write_unlock(&map->lock); |
70dec807 CM |
4695 | free_extent_map(em); |
4696 | break; | |
4697 | } | |
3d6448e6 FM |
4698 | if (test_range_bit(tree, em->start, |
4699 | extent_map_end(em) - 1, | |
4700 | EXTENT_LOCKED, 0, NULL)) | |
4701 | goto next; | |
4702 | /* | |
4703 | * If it's not in the list of modified extents, used | |
4704 | * by a fast fsync, we can remove it. If it's being | |
4705 | * logged we can safely remove it since fsync took an | |
4706 | * extra reference on the em. | |
4707 | */ | |
4708 | if (list_empty(&em->list) || | |
fbc2bd7e FM |
4709 | test_bit(EXTENT_FLAG_LOGGING, &em->flags)) |
4710 | goto remove_em; | |
4711 | /* | |
4712 | * If it's in the list of modified extents, remove it | |
4713 | * only if its generation is older then the current one, | |
4714 | * in which case we don't need it for a fast fsync. | |
4715 | * Otherwise don't remove it, we could be racing with an | |
4716 | * ongoing fast fsync that could miss the new extent. | |
4717 | */ | |
4718 | fs_info = btrfs_inode->root->fs_info; | |
4719 | spin_lock(&fs_info->trans_lock); | |
4720 | cur_gen = fs_info->generation; | |
4721 | spin_unlock(&fs_info->trans_lock); | |
4722 | if (em->generation >= cur_gen) | |
4723 | goto next; | |
4724 | remove_em: | |
5e548b32 FM |
4725 | /* |
4726 | * We only remove extent maps that are not in the list of | |
4727 | * modified extents or that are in the list but with a | |
4728 | * generation lower then the current generation, so there | |
4729 | * is no need to set the full fsync flag on the inode (it | |
4730 | * hurts the fsync performance for workloads with a data | |
4731 | * size that exceeds or is close to the system's memory). | |
4732 | */ | |
fbc2bd7e FM |
4733 | remove_extent_mapping(map, em); |
4734 | /* once for the rb tree */ | |
4735 | free_extent_map(em); | |
3d6448e6 | 4736 | next: |
70dec807 | 4737 | start = extent_map_end(em); |
890871be | 4738 | write_unlock(&map->lock); |
70dec807 CM |
4739 | |
4740 | /* once for us */ | |
d1310b2e | 4741 | free_extent_map(em); |
9f47eb54 PM |
4742 | |
4743 | cond_resched(); /* Allow large-extent preemption. */ | |
d1310b2e | 4744 | } |
d1310b2e | 4745 | } |
29c68b2d | 4746 | return try_release_extent_state(tree, page, mask); |
d1310b2e | 4747 | } |
d1310b2e | 4748 | |
ec29ed5b CM |
4749 | /* |
4750 | * helper function for fiemap, which doesn't want to see any holes. | |
4751 | * This maps until we find something past 'last' | |
4752 | */ | |
f1bbde8d | 4753 | static struct extent_map *get_extent_skip_holes(struct btrfs_inode *inode, |
e3350e16 | 4754 | u64 offset, u64 last) |
ec29ed5b | 4755 | { |
f1bbde8d | 4756 | u64 sectorsize = btrfs_inode_sectorsize(inode); |
ec29ed5b CM |
4757 | struct extent_map *em; |
4758 | u64 len; | |
4759 | ||
4760 | if (offset >= last) | |
4761 | return NULL; | |
4762 | ||
67871254 | 4763 | while (1) { |
ec29ed5b CM |
4764 | len = last - offset; |
4765 | if (len == 0) | |
4766 | break; | |
fda2832f | 4767 | len = ALIGN(len, sectorsize); |
f1bbde8d | 4768 | em = btrfs_get_extent_fiemap(inode, offset, len); |
c704005d | 4769 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
4770 | return em; |
4771 | ||
4772 | /* if this isn't a hole return it */ | |
4a2d25cd | 4773 | if (em->block_start != EXTENT_MAP_HOLE) |
ec29ed5b | 4774 | return em; |
ec29ed5b CM |
4775 | |
4776 | /* this is a hole, advance to the next extent */ | |
4777 | offset = extent_map_end(em); | |
4778 | free_extent_map(em); | |
4779 | if (offset >= last) | |
4780 | break; | |
4781 | } | |
4782 | return NULL; | |
4783 | } | |
4784 | ||
4751832d QW |
4785 | /* |
4786 | * To cache previous fiemap extent | |
4787 | * | |
4788 | * Will be used for merging fiemap extent | |
4789 | */ | |
4790 | struct fiemap_cache { | |
4791 | u64 offset; | |
4792 | u64 phys; | |
4793 | u64 len; | |
4794 | u32 flags; | |
4795 | bool cached; | |
4796 | }; | |
4797 | ||
4798 | /* | |
4799 | * Helper to submit fiemap extent. | |
4800 | * | |
4801 | * Will try to merge current fiemap extent specified by @offset, @phys, | |
4802 | * @len and @flags with cached one. | |
4803 | * And only when we fails to merge, cached one will be submitted as | |
4804 | * fiemap extent. | |
4805 | * | |
4806 | * Return value is the same as fiemap_fill_next_extent(). | |
4807 | */ | |
4808 | static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, | |
4809 | struct fiemap_cache *cache, | |
4810 | u64 offset, u64 phys, u64 len, u32 flags) | |
4811 | { | |
4812 | int ret = 0; | |
4813 | ||
4814 | if (!cache->cached) | |
4815 | goto assign; | |
4816 | ||
4817 | /* | |
4818 | * Sanity check, extent_fiemap() should have ensured that new | |
52042d8e | 4819 | * fiemap extent won't overlap with cached one. |
4751832d QW |
4820 | * Not recoverable. |
4821 | * | |
4822 | * NOTE: Physical address can overlap, due to compression | |
4823 | */ | |
4824 | if (cache->offset + cache->len > offset) { | |
4825 | WARN_ON(1); | |
4826 | return -EINVAL; | |
4827 | } | |
4828 | ||
4829 | /* | |
4830 | * Only merges fiemap extents if | |
4831 | * 1) Their logical addresses are continuous | |
4832 | * | |
4833 | * 2) Their physical addresses are continuous | |
4834 | * So truly compressed (physical size smaller than logical size) | |
4835 | * extents won't get merged with each other | |
4836 | * | |
4837 | * 3) Share same flags except FIEMAP_EXTENT_LAST | |
4838 | * So regular extent won't get merged with prealloc extent | |
4839 | */ | |
4840 | if (cache->offset + cache->len == offset && | |
4841 | cache->phys + cache->len == phys && | |
4842 | (cache->flags & ~FIEMAP_EXTENT_LAST) == | |
4843 | (flags & ~FIEMAP_EXTENT_LAST)) { | |
4844 | cache->len += len; | |
4845 | cache->flags |= flags; | |
4846 | goto try_submit_last; | |
4847 | } | |
4848 | ||
4849 | /* Not mergeable, need to submit cached one */ | |
4850 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, | |
4851 | cache->len, cache->flags); | |
4852 | cache->cached = false; | |
4853 | if (ret) | |
4854 | return ret; | |
4855 | assign: | |
4856 | cache->cached = true; | |
4857 | cache->offset = offset; | |
4858 | cache->phys = phys; | |
4859 | cache->len = len; | |
4860 | cache->flags = flags; | |
4861 | try_submit_last: | |
4862 | if (cache->flags & FIEMAP_EXTENT_LAST) { | |
4863 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, | |
4864 | cache->phys, cache->len, cache->flags); | |
4865 | cache->cached = false; | |
4866 | } | |
4867 | return ret; | |
4868 | } | |
4869 | ||
4870 | /* | |
848c23b7 | 4871 | * Emit last fiemap cache |
4751832d | 4872 | * |
848c23b7 QW |
4873 | * The last fiemap cache may still be cached in the following case: |
4874 | * 0 4k 8k | |
4875 | * |<- Fiemap range ->| | |
4876 | * |<------------ First extent ----------->| | |
4877 | * | |
4878 | * In this case, the first extent range will be cached but not emitted. | |
4879 | * So we must emit it before ending extent_fiemap(). | |
4751832d | 4880 | */ |
5c5aff98 | 4881 | static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo, |
848c23b7 | 4882 | struct fiemap_cache *cache) |
4751832d QW |
4883 | { |
4884 | int ret; | |
4885 | ||
4886 | if (!cache->cached) | |
4887 | return 0; | |
4888 | ||
4751832d QW |
4889 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, |
4890 | cache->len, cache->flags); | |
4891 | cache->cached = false; | |
4892 | if (ret > 0) | |
4893 | ret = 0; | |
4894 | return ret; | |
4895 | } | |
4896 | ||
facee0a0 | 4897 | int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo, |
bab16e21 | 4898 | u64 start, u64 len) |
1506fcc8 | 4899 | { |
975f84fe | 4900 | int ret = 0; |
1506fcc8 YS |
4901 | u64 off = start; |
4902 | u64 max = start + len; | |
4903 | u32 flags = 0; | |
975f84fe JB |
4904 | u32 found_type; |
4905 | u64 last; | |
ec29ed5b | 4906 | u64 last_for_get_extent = 0; |
1506fcc8 | 4907 | u64 disko = 0; |
facee0a0 | 4908 | u64 isize = i_size_read(&inode->vfs_inode); |
975f84fe | 4909 | struct btrfs_key found_key; |
1506fcc8 | 4910 | struct extent_map *em = NULL; |
2ac55d41 | 4911 | struct extent_state *cached_state = NULL; |
975f84fe | 4912 | struct btrfs_path *path; |
facee0a0 | 4913 | struct btrfs_root *root = inode->root; |
4751832d | 4914 | struct fiemap_cache cache = { 0 }; |
5911c8fe DS |
4915 | struct ulist *roots; |
4916 | struct ulist *tmp_ulist; | |
1506fcc8 | 4917 | int end = 0; |
ec29ed5b CM |
4918 | u64 em_start = 0; |
4919 | u64 em_len = 0; | |
4920 | u64 em_end = 0; | |
1506fcc8 YS |
4921 | |
4922 | if (len == 0) | |
4923 | return -EINVAL; | |
4924 | ||
975f84fe JB |
4925 | path = btrfs_alloc_path(); |
4926 | if (!path) | |
4927 | return -ENOMEM; | |
975f84fe | 4928 | |
5911c8fe DS |
4929 | roots = ulist_alloc(GFP_KERNEL); |
4930 | tmp_ulist = ulist_alloc(GFP_KERNEL); | |
4931 | if (!roots || !tmp_ulist) { | |
4932 | ret = -ENOMEM; | |
4933 | goto out_free_ulist; | |
4934 | } | |
4935 | ||
facee0a0 NB |
4936 | start = round_down(start, btrfs_inode_sectorsize(inode)); |
4937 | len = round_up(max, btrfs_inode_sectorsize(inode)) - start; | |
4d479cf0 | 4938 | |
ec29ed5b CM |
4939 | /* |
4940 | * lookup the last file extent. We're not using i_size here | |
4941 | * because there might be preallocation past i_size | |
4942 | */ | |
facee0a0 NB |
4943 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1, |
4944 | 0); | |
975f84fe | 4945 | if (ret < 0) { |
5911c8fe | 4946 | goto out_free_ulist; |
2d324f59 LB |
4947 | } else { |
4948 | WARN_ON(!ret); | |
4949 | if (ret == 1) | |
4950 | ret = 0; | |
975f84fe | 4951 | } |
2d324f59 | 4952 | |
975f84fe | 4953 | path->slots[0]--; |
975f84fe | 4954 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
962a298f | 4955 | found_type = found_key.type; |
975f84fe | 4956 | |
ec29ed5b | 4957 | /* No extents, but there might be delalloc bits */ |
facee0a0 | 4958 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 4959 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
4960 | /* have to trust i_size as the end */ |
4961 | last = (u64)-1; | |
4962 | last_for_get_extent = isize; | |
4963 | } else { | |
4964 | /* | |
4965 | * remember the start of the last extent. There are a | |
4966 | * bunch of different factors that go into the length of the | |
4967 | * extent, so its much less complex to remember where it started | |
4968 | */ | |
4969 | last = found_key.offset; | |
4970 | last_for_get_extent = last + 1; | |
975f84fe | 4971 | } |
fe09e16c | 4972 | btrfs_release_path(path); |
975f84fe | 4973 | |
ec29ed5b CM |
4974 | /* |
4975 | * we might have some extents allocated but more delalloc past those | |
4976 | * extents. so, we trust isize unless the start of the last extent is | |
4977 | * beyond isize | |
4978 | */ | |
4979 | if (last < isize) { | |
4980 | last = (u64)-1; | |
4981 | last_for_get_extent = isize; | |
4982 | } | |
4983 | ||
facee0a0 | 4984 | lock_extent_bits(&inode->io_tree, start, start + len - 1, |
d0082371 | 4985 | &cached_state); |
ec29ed5b | 4986 | |
facee0a0 | 4987 | em = get_extent_skip_holes(inode, start, last_for_get_extent); |
1506fcc8 YS |
4988 | if (!em) |
4989 | goto out; | |
4990 | if (IS_ERR(em)) { | |
4991 | ret = PTR_ERR(em); | |
4992 | goto out; | |
4993 | } | |
975f84fe | 4994 | |
1506fcc8 | 4995 | while (!end) { |
b76bb701 | 4996 | u64 offset_in_extent = 0; |
ea8efc74 CM |
4997 | |
4998 | /* break if the extent we found is outside the range */ | |
4999 | if (em->start >= max || extent_map_end(em) < off) | |
5000 | break; | |
5001 | ||
5002 | /* | |
5003 | * get_extent may return an extent that starts before our | |
5004 | * requested range. We have to make sure the ranges | |
5005 | * we return to fiemap always move forward and don't | |
5006 | * overlap, so adjust the offsets here | |
5007 | */ | |
5008 | em_start = max(em->start, off); | |
1506fcc8 | 5009 | |
ea8efc74 CM |
5010 | /* |
5011 | * record the offset from the start of the extent | |
b76bb701 JB |
5012 | * for adjusting the disk offset below. Only do this if the |
5013 | * extent isn't compressed since our in ram offset may be past | |
5014 | * what we have actually allocated on disk. | |
ea8efc74 | 5015 | */ |
b76bb701 JB |
5016 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
5017 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 5018 | em_end = extent_map_end(em); |
ea8efc74 | 5019 | em_len = em_end - em_start; |
1506fcc8 | 5020 | flags = 0; |
f0986318 FM |
5021 | if (em->block_start < EXTENT_MAP_LAST_BYTE) |
5022 | disko = em->block_start + offset_in_extent; | |
5023 | else | |
5024 | disko = 0; | |
1506fcc8 | 5025 | |
ea8efc74 CM |
5026 | /* |
5027 | * bump off for our next call to get_extent | |
5028 | */ | |
5029 | off = extent_map_end(em); | |
5030 | if (off >= max) | |
5031 | end = 1; | |
5032 | ||
93dbfad7 | 5033 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
5034 | end = 1; |
5035 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 5036 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
5037 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
5038 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 5039 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
5040 | flags |= (FIEMAP_EXTENT_DELALLOC | |
5041 | FIEMAP_EXTENT_UNKNOWN); | |
dc046b10 JB |
5042 | } else if (fieinfo->fi_extents_max) { |
5043 | u64 bytenr = em->block_start - | |
5044 | (em->start - em->orig_start); | |
fe09e16c | 5045 | |
fe09e16c LB |
5046 | /* |
5047 | * As btrfs supports shared space, this information | |
5048 | * can be exported to userspace tools via | |
dc046b10 JB |
5049 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 |
5050 | * then we're just getting a count and we can skip the | |
5051 | * lookup stuff. | |
fe09e16c | 5052 | */ |
facee0a0 | 5053 | ret = btrfs_check_shared(root, btrfs_ino(inode), |
5911c8fe | 5054 | bytenr, roots, tmp_ulist); |
dc046b10 | 5055 | if (ret < 0) |
fe09e16c | 5056 | goto out_free; |
dc046b10 | 5057 | if (ret) |
fe09e16c | 5058 | flags |= FIEMAP_EXTENT_SHARED; |
dc046b10 | 5059 | ret = 0; |
1506fcc8 YS |
5060 | } |
5061 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
5062 | flags |= FIEMAP_EXTENT_ENCODED; | |
0d2b2372 JB |
5063 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
5064 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
1506fcc8 | 5065 | |
1506fcc8 YS |
5066 | free_extent_map(em); |
5067 | em = NULL; | |
ec29ed5b CM |
5068 | if ((em_start >= last) || em_len == (u64)-1 || |
5069 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
5070 | flags |= FIEMAP_EXTENT_LAST; |
5071 | end = 1; | |
5072 | } | |
5073 | ||
ec29ed5b | 5074 | /* now scan forward to see if this is really the last extent. */ |
facee0a0 | 5075 | em = get_extent_skip_holes(inode, off, last_for_get_extent); |
ec29ed5b CM |
5076 | if (IS_ERR(em)) { |
5077 | ret = PTR_ERR(em); | |
5078 | goto out; | |
5079 | } | |
5080 | if (!em) { | |
975f84fe JB |
5081 | flags |= FIEMAP_EXTENT_LAST; |
5082 | end = 1; | |
5083 | } | |
4751832d QW |
5084 | ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko, |
5085 | em_len, flags); | |
26e726af CS |
5086 | if (ret) { |
5087 | if (ret == 1) | |
5088 | ret = 0; | |
ec29ed5b | 5089 | goto out_free; |
26e726af | 5090 | } |
1506fcc8 YS |
5091 | } |
5092 | out_free: | |
4751832d | 5093 | if (!ret) |
5c5aff98 | 5094 | ret = emit_last_fiemap_cache(fieinfo, &cache); |
1506fcc8 YS |
5095 | free_extent_map(em); |
5096 | out: | |
facee0a0 | 5097 | unlock_extent_cached(&inode->io_tree, start, start + len - 1, |
e43bbe5e | 5098 | &cached_state); |
5911c8fe DS |
5099 | |
5100 | out_free_ulist: | |
e02d48ea | 5101 | btrfs_free_path(path); |
5911c8fe DS |
5102 | ulist_free(roots); |
5103 | ulist_free(tmp_ulist); | |
1506fcc8 YS |
5104 | return ret; |
5105 | } | |
5106 | ||
727011e0 CM |
5107 | static void __free_extent_buffer(struct extent_buffer *eb) |
5108 | { | |
727011e0 CM |
5109 | kmem_cache_free(extent_buffer_cache, eb); |
5110 | } | |
5111 | ||
2b48966a | 5112 | int extent_buffer_under_io(const struct extent_buffer *eb) |
db7f3436 JB |
5113 | { |
5114 | return (atomic_read(&eb->io_pages) || | |
5115 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
5116 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
5117 | } | |
5118 | ||
8ff8466d | 5119 | static bool page_range_has_eb(struct btrfs_fs_info *fs_info, struct page *page) |
db7f3436 | 5120 | { |
8ff8466d | 5121 | struct btrfs_subpage *subpage; |
db7f3436 | 5122 | |
8ff8466d | 5123 | lockdep_assert_held(&page->mapping->private_lock); |
db7f3436 | 5124 | |
8ff8466d QW |
5125 | if (PagePrivate(page)) { |
5126 | subpage = (struct btrfs_subpage *)page->private; | |
5127 | if (atomic_read(&subpage->eb_refs)) | |
5128 | return true; | |
5129 | } | |
5130 | return false; | |
5131 | } | |
db7f3436 | 5132 | |
8ff8466d QW |
5133 | static void detach_extent_buffer_page(struct extent_buffer *eb, struct page *page) |
5134 | { | |
5135 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
5136 | const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); | |
5137 | ||
5138 | /* | |
5139 | * For mapped eb, we're going to change the page private, which should | |
5140 | * be done under the private_lock. | |
5141 | */ | |
5142 | if (mapped) | |
5143 | spin_lock(&page->mapping->private_lock); | |
5144 | ||
5145 | if (!PagePrivate(page)) { | |
5d2361db | 5146 | if (mapped) |
8ff8466d QW |
5147 | spin_unlock(&page->mapping->private_lock); |
5148 | return; | |
5149 | } | |
5150 | ||
5151 | if (fs_info->sectorsize == PAGE_SIZE) { | |
5d2361db FL |
5152 | /* |
5153 | * We do this since we'll remove the pages after we've | |
5154 | * removed the eb from the radix tree, so we could race | |
5155 | * and have this page now attached to the new eb. So | |
5156 | * only clear page_private if it's still connected to | |
5157 | * this eb. | |
5158 | */ | |
5159 | if (PagePrivate(page) && | |
5160 | page->private == (unsigned long)eb) { | |
5161 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
5162 | BUG_ON(PageDirty(page)); | |
5163 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 5164 | /* |
5d2361db FL |
5165 | * We need to make sure we haven't be attached |
5166 | * to a new eb. | |
db7f3436 | 5167 | */ |
d1b89bc0 | 5168 | detach_page_private(page); |
db7f3436 | 5169 | } |
5d2361db FL |
5170 | if (mapped) |
5171 | spin_unlock(&page->mapping->private_lock); | |
8ff8466d QW |
5172 | return; |
5173 | } | |
5174 | ||
5175 | /* | |
5176 | * For subpage, we can have dummy eb with page private. In this case, | |
5177 | * we can directly detach the private as such page is only attached to | |
5178 | * one dummy eb, no sharing. | |
5179 | */ | |
5180 | if (!mapped) { | |
5181 | btrfs_detach_subpage(fs_info, page); | |
5182 | return; | |
5183 | } | |
5184 | ||
5185 | btrfs_page_dec_eb_refs(fs_info, page); | |
5186 | ||
5187 | /* | |
5188 | * We can only detach the page private if there are no other ebs in the | |
5189 | * page range. | |
5190 | */ | |
5191 | if (!page_range_has_eb(fs_info, page)) | |
5192 | btrfs_detach_subpage(fs_info, page); | |
5193 | ||
5194 | spin_unlock(&page->mapping->private_lock); | |
5195 | } | |
5196 | ||
5197 | /* Release all pages attached to the extent buffer */ | |
5198 | static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb) | |
5199 | { | |
5200 | int i; | |
5201 | int num_pages; | |
5202 | ||
5203 | ASSERT(!extent_buffer_under_io(eb)); | |
5204 | ||
5205 | num_pages = num_extent_pages(eb); | |
5206 | for (i = 0; i < num_pages; i++) { | |
5207 | struct page *page = eb->pages[i]; | |
5208 | ||
5209 | if (!page) | |
5210 | continue; | |
5211 | ||
5212 | detach_extent_buffer_page(eb, page); | |
5d2361db | 5213 | |
01327610 | 5214 | /* One for when we allocated the page */ |
09cbfeaf | 5215 | put_page(page); |
d64766fd | 5216 | } |
db7f3436 JB |
5217 | } |
5218 | ||
5219 | /* | |
5220 | * Helper for releasing the extent buffer. | |
5221 | */ | |
5222 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
5223 | { | |
55ac0139 | 5224 | btrfs_release_extent_buffer_pages(eb); |
8c38938c | 5225 | btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list); |
db7f3436 JB |
5226 | __free_extent_buffer(eb); |
5227 | } | |
5228 | ||
f28491e0 JB |
5229 | static struct extent_buffer * |
5230 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 5231 | unsigned long len) |
d1310b2e CM |
5232 | { |
5233 | struct extent_buffer *eb = NULL; | |
5234 | ||
d1b5c567 | 5235 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
5236 | eb->start = start; |
5237 | eb->len = len; | |
f28491e0 | 5238 | eb->fs_info = fs_info; |
815a51c7 | 5239 | eb->bflags = 0; |
196d59ab | 5240 | init_rwsem(&eb->lock); |
b4ce94de | 5241 | |
3fd63727 JB |
5242 | btrfs_leak_debug_add(&fs_info->eb_leak_lock, &eb->leak_list, |
5243 | &fs_info->allocated_ebs); | |
d3575156 | 5244 | INIT_LIST_HEAD(&eb->release_list); |
6d49ba1b | 5245 | |
3083ee2e | 5246 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 5247 | atomic_set(&eb->refs, 1); |
0b32f4bb | 5248 | atomic_set(&eb->io_pages, 0); |
727011e0 | 5249 | |
deb67895 | 5250 | ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE); |
d1310b2e CM |
5251 | |
5252 | return eb; | |
5253 | } | |
5254 | ||
2b48966a | 5255 | struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src) |
815a51c7 | 5256 | { |
cc5e31a4 | 5257 | int i; |
815a51c7 JS |
5258 | struct page *p; |
5259 | struct extent_buffer *new; | |
cc5e31a4 | 5260 | int num_pages = num_extent_pages(src); |
815a51c7 | 5261 | |
3f556f78 | 5262 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
5263 | if (new == NULL) |
5264 | return NULL; | |
5265 | ||
62c053fb QW |
5266 | /* |
5267 | * Set UNMAPPED before calling btrfs_release_extent_buffer(), as | |
5268 | * btrfs_release_extent_buffer() have different behavior for | |
5269 | * UNMAPPED subpage extent buffer. | |
5270 | */ | |
5271 | set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags); | |
5272 | ||
815a51c7 | 5273 | for (i = 0; i < num_pages; i++) { |
760f991f QW |
5274 | int ret; |
5275 | ||
9ec72677 | 5276 | p = alloc_page(GFP_NOFS); |
db7f3436 JB |
5277 | if (!p) { |
5278 | btrfs_release_extent_buffer(new); | |
5279 | return NULL; | |
5280 | } | |
760f991f QW |
5281 | ret = attach_extent_buffer_page(new, p, NULL); |
5282 | if (ret < 0) { | |
5283 | put_page(p); | |
5284 | btrfs_release_extent_buffer(new); | |
5285 | return NULL; | |
5286 | } | |
815a51c7 | 5287 | WARN_ON(PageDirty(p)); |
815a51c7 | 5288 | new->pages[i] = p; |
fba1acf9 | 5289 | copy_page(page_address(p), page_address(src->pages[i])); |
815a51c7 | 5290 | } |
92d83e94 | 5291 | set_extent_buffer_uptodate(new); |
815a51c7 JS |
5292 | |
5293 | return new; | |
5294 | } | |
5295 | ||
0f331229 OS |
5296 | struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
5297 | u64 start, unsigned long len) | |
815a51c7 JS |
5298 | { |
5299 | struct extent_buffer *eb; | |
cc5e31a4 DS |
5300 | int num_pages; |
5301 | int i; | |
815a51c7 | 5302 | |
3f556f78 | 5303 | eb = __alloc_extent_buffer(fs_info, start, len); |
815a51c7 JS |
5304 | if (!eb) |
5305 | return NULL; | |
5306 | ||
65ad0104 | 5307 | num_pages = num_extent_pages(eb); |
815a51c7 | 5308 | for (i = 0; i < num_pages; i++) { |
09bc1f0f QW |
5309 | int ret; |
5310 | ||
9ec72677 | 5311 | eb->pages[i] = alloc_page(GFP_NOFS); |
815a51c7 JS |
5312 | if (!eb->pages[i]) |
5313 | goto err; | |
09bc1f0f QW |
5314 | ret = attach_extent_buffer_page(eb, eb->pages[i], NULL); |
5315 | if (ret < 0) | |
5316 | goto err; | |
815a51c7 JS |
5317 | } |
5318 | set_extent_buffer_uptodate(eb); | |
5319 | btrfs_set_header_nritems(eb, 0); | |
b0132a3b | 5320 | set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
815a51c7 JS |
5321 | |
5322 | return eb; | |
5323 | err: | |
09bc1f0f QW |
5324 | for (; i > 0; i--) { |
5325 | detach_extent_buffer_page(eb, eb->pages[i - 1]); | |
84167d19 | 5326 | __free_page(eb->pages[i - 1]); |
09bc1f0f | 5327 | } |
815a51c7 JS |
5328 | __free_extent_buffer(eb); |
5329 | return NULL; | |
5330 | } | |
5331 | ||
0f331229 | 5332 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
da17066c | 5333 | u64 start) |
0f331229 | 5334 | { |
da17066c | 5335 | return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize); |
0f331229 OS |
5336 | } |
5337 | ||
0b32f4bb JB |
5338 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
5339 | { | |
242e18c7 | 5340 | int refs; |
6bf9cd2e BB |
5341 | /* |
5342 | * The TREE_REF bit is first set when the extent_buffer is added | |
5343 | * to the radix tree. It is also reset, if unset, when a new reference | |
5344 | * is created by find_extent_buffer. | |
0b32f4bb | 5345 | * |
6bf9cd2e BB |
5346 | * It is only cleared in two cases: freeing the last non-tree |
5347 | * reference to the extent_buffer when its STALE bit is set or | |
5348 | * calling releasepage when the tree reference is the only reference. | |
0b32f4bb | 5349 | * |
6bf9cd2e BB |
5350 | * In both cases, care is taken to ensure that the extent_buffer's |
5351 | * pages are not under io. However, releasepage can be concurrently | |
5352 | * called with creating new references, which is prone to race | |
5353 | * conditions between the calls to check_buffer_tree_ref in those | |
5354 | * codepaths and clearing TREE_REF in try_release_extent_buffer. | |
0b32f4bb | 5355 | * |
6bf9cd2e BB |
5356 | * The actual lifetime of the extent_buffer in the radix tree is |
5357 | * adequately protected by the refcount, but the TREE_REF bit and | |
5358 | * its corresponding reference are not. To protect against this | |
5359 | * class of races, we call check_buffer_tree_ref from the codepaths | |
5360 | * which trigger io after they set eb->io_pages. Note that once io is | |
5361 | * initiated, TREE_REF can no longer be cleared, so that is the | |
5362 | * moment at which any such race is best fixed. | |
0b32f4bb | 5363 | */ |
242e18c7 CM |
5364 | refs = atomic_read(&eb->refs); |
5365 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
5366 | return; | |
5367 | ||
594831c4 JB |
5368 | spin_lock(&eb->refs_lock); |
5369 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 5370 | atomic_inc(&eb->refs); |
594831c4 | 5371 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
5372 | } |
5373 | ||
2457aec6 MG |
5374 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
5375 | struct page *accessed) | |
5df4235e | 5376 | { |
cc5e31a4 | 5377 | int num_pages, i; |
5df4235e | 5378 | |
0b32f4bb JB |
5379 | check_buffer_tree_ref(eb); |
5380 | ||
65ad0104 | 5381 | num_pages = num_extent_pages(eb); |
5df4235e | 5382 | for (i = 0; i < num_pages; i++) { |
fb85fc9a DS |
5383 | struct page *p = eb->pages[i]; |
5384 | ||
2457aec6 MG |
5385 | if (p != accessed) |
5386 | mark_page_accessed(p); | |
5df4235e JB |
5387 | } |
5388 | } | |
5389 | ||
f28491e0 JB |
5390 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
5391 | u64 start) | |
452c75c3 CS |
5392 | { |
5393 | struct extent_buffer *eb; | |
5394 | ||
5395 | rcu_read_lock(); | |
f28491e0 | 5396 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
478ef886 | 5397 | start >> fs_info->sectorsize_bits); |
452c75c3 CS |
5398 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
5399 | rcu_read_unlock(); | |
062c19e9 FM |
5400 | /* |
5401 | * Lock our eb's refs_lock to avoid races with | |
5402 | * free_extent_buffer. When we get our eb it might be flagged | |
5403 | * with EXTENT_BUFFER_STALE and another task running | |
5404 | * free_extent_buffer might have seen that flag set, | |
5405 | * eb->refs == 2, that the buffer isn't under IO (dirty and | |
5406 | * writeback flags not set) and it's still in the tree (flag | |
5407 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process | |
5408 | * of decrementing the extent buffer's reference count twice. | |
5409 | * So here we could race and increment the eb's reference count, | |
5410 | * clear its stale flag, mark it as dirty and drop our reference | |
5411 | * before the other task finishes executing free_extent_buffer, | |
5412 | * which would later result in an attempt to free an extent | |
5413 | * buffer that is dirty. | |
5414 | */ | |
5415 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
5416 | spin_lock(&eb->refs_lock); | |
5417 | spin_unlock(&eb->refs_lock); | |
5418 | } | |
2457aec6 | 5419 | mark_extent_buffer_accessed(eb, NULL); |
452c75c3 CS |
5420 | return eb; |
5421 | } | |
5422 | rcu_read_unlock(); | |
5423 | ||
5424 | return NULL; | |
5425 | } | |
5426 | ||
faa2dbf0 JB |
5427 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
5428 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
da17066c | 5429 | u64 start) |
faa2dbf0 JB |
5430 | { |
5431 | struct extent_buffer *eb, *exists = NULL; | |
5432 | int ret; | |
5433 | ||
5434 | eb = find_extent_buffer(fs_info, start); | |
5435 | if (eb) | |
5436 | return eb; | |
da17066c | 5437 | eb = alloc_dummy_extent_buffer(fs_info, start); |
faa2dbf0 | 5438 | if (!eb) |
b6293c82 | 5439 | return ERR_PTR(-ENOMEM); |
faa2dbf0 JB |
5440 | eb->fs_info = fs_info; |
5441 | again: | |
e1860a77 | 5442 | ret = radix_tree_preload(GFP_NOFS); |
b6293c82 DC |
5443 | if (ret) { |
5444 | exists = ERR_PTR(ret); | |
faa2dbf0 | 5445 | goto free_eb; |
b6293c82 | 5446 | } |
faa2dbf0 JB |
5447 | spin_lock(&fs_info->buffer_lock); |
5448 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
478ef886 | 5449 | start >> fs_info->sectorsize_bits, eb); |
faa2dbf0 JB |
5450 | spin_unlock(&fs_info->buffer_lock); |
5451 | radix_tree_preload_end(); | |
5452 | if (ret == -EEXIST) { | |
5453 | exists = find_extent_buffer(fs_info, start); | |
5454 | if (exists) | |
5455 | goto free_eb; | |
5456 | else | |
5457 | goto again; | |
5458 | } | |
5459 | check_buffer_tree_ref(eb); | |
5460 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
5461 | ||
faa2dbf0 JB |
5462 | return eb; |
5463 | free_eb: | |
5464 | btrfs_release_extent_buffer(eb); | |
5465 | return exists; | |
5466 | } | |
5467 | #endif | |
5468 | ||
81982210 QW |
5469 | static struct extent_buffer *grab_extent_buffer( |
5470 | struct btrfs_fs_info *fs_info, struct page *page) | |
c0f0a9e7 QW |
5471 | { |
5472 | struct extent_buffer *exists; | |
5473 | ||
81982210 QW |
5474 | /* |
5475 | * For subpage case, we completely rely on radix tree to ensure we | |
5476 | * don't try to insert two ebs for the same bytenr. So here we always | |
5477 | * return NULL and just continue. | |
5478 | */ | |
5479 | if (fs_info->sectorsize < PAGE_SIZE) | |
5480 | return NULL; | |
5481 | ||
c0f0a9e7 QW |
5482 | /* Page not yet attached to an extent buffer */ |
5483 | if (!PagePrivate(page)) | |
5484 | return NULL; | |
5485 | ||
5486 | /* | |
5487 | * We could have already allocated an eb for this page and attached one | |
5488 | * so lets see if we can get a ref on the existing eb, and if we can we | |
5489 | * know it's good and we can just return that one, else we know we can | |
5490 | * just overwrite page->private. | |
5491 | */ | |
5492 | exists = (struct extent_buffer *)page->private; | |
5493 | if (atomic_inc_not_zero(&exists->refs)) | |
5494 | return exists; | |
5495 | ||
5496 | WARN_ON(PageDirty(page)); | |
5497 | detach_page_private(page); | |
5498 | return NULL; | |
5499 | } | |
5500 | ||
f28491e0 | 5501 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
3fbaf258 | 5502 | u64 start, u64 owner_root, int level) |
d1310b2e | 5503 | { |
da17066c | 5504 | unsigned long len = fs_info->nodesize; |
cc5e31a4 DS |
5505 | int num_pages; |
5506 | int i; | |
09cbfeaf | 5507 | unsigned long index = start >> PAGE_SHIFT; |
d1310b2e | 5508 | struct extent_buffer *eb; |
6af118ce | 5509 | struct extent_buffer *exists = NULL; |
d1310b2e | 5510 | struct page *p; |
f28491e0 | 5511 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
d1310b2e | 5512 | int uptodate = 1; |
19fe0a8b | 5513 | int ret; |
d1310b2e | 5514 | |
da17066c | 5515 | if (!IS_ALIGNED(start, fs_info->sectorsize)) { |
c871b0f2 LB |
5516 | btrfs_err(fs_info, "bad tree block start %llu", start); |
5517 | return ERR_PTR(-EINVAL); | |
5518 | } | |
5519 | ||
1aaac38c QW |
5520 | if (fs_info->sectorsize < PAGE_SIZE && |
5521 | offset_in_page(start) + len > PAGE_SIZE) { | |
5522 | btrfs_err(fs_info, | |
5523 | "tree block crosses page boundary, start %llu nodesize %lu", | |
5524 | start, len); | |
5525 | return ERR_PTR(-EINVAL); | |
5526 | } | |
5527 | ||
f28491e0 | 5528 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 5529 | if (eb) |
6af118ce | 5530 | return eb; |
6af118ce | 5531 | |
23d79d81 | 5532 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 5533 | if (!eb) |
c871b0f2 | 5534 | return ERR_PTR(-ENOMEM); |
e114c545 | 5535 | btrfs_set_buffer_lockdep_class(owner_root, eb, level); |
d1310b2e | 5536 | |
65ad0104 | 5537 | num_pages = num_extent_pages(eb); |
727011e0 | 5538 | for (i = 0; i < num_pages; i++, index++) { |
760f991f QW |
5539 | struct btrfs_subpage *prealloc = NULL; |
5540 | ||
d1b5c567 | 5541 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
c871b0f2 LB |
5542 | if (!p) { |
5543 | exists = ERR_PTR(-ENOMEM); | |
6af118ce | 5544 | goto free_eb; |
c871b0f2 | 5545 | } |
4f2de97a | 5546 | |
760f991f QW |
5547 | /* |
5548 | * Preallocate page->private for subpage case, so that we won't | |
5549 | * allocate memory with private_lock hold. The memory will be | |
5550 | * freed by attach_extent_buffer_page() or freed manually if | |
5551 | * we exit earlier. | |
5552 | * | |
5553 | * Although we have ensured one subpage eb can only have one | |
5554 | * page, but it may change in the future for 16K page size | |
5555 | * support, so we still preallocate the memory in the loop. | |
5556 | */ | |
5557 | ret = btrfs_alloc_subpage(fs_info, &prealloc, | |
5558 | BTRFS_SUBPAGE_METADATA); | |
5559 | if (ret < 0) { | |
5560 | unlock_page(p); | |
5561 | put_page(p); | |
5562 | exists = ERR_PTR(ret); | |
5563 | goto free_eb; | |
5564 | } | |
5565 | ||
4f2de97a | 5566 | spin_lock(&mapping->private_lock); |
81982210 | 5567 | exists = grab_extent_buffer(fs_info, p); |
c0f0a9e7 QW |
5568 | if (exists) { |
5569 | spin_unlock(&mapping->private_lock); | |
5570 | unlock_page(p); | |
5571 | put_page(p); | |
5572 | mark_extent_buffer_accessed(exists, p); | |
760f991f | 5573 | btrfs_free_subpage(prealloc); |
c0f0a9e7 | 5574 | goto free_eb; |
d1310b2e | 5575 | } |
760f991f QW |
5576 | /* Should not fail, as we have preallocated the memory */ |
5577 | ret = attach_extent_buffer_page(eb, p, prealloc); | |
5578 | ASSERT(!ret); | |
8ff8466d QW |
5579 | /* |
5580 | * To inform we have extra eb under allocation, so that | |
5581 | * detach_extent_buffer_page() won't release the page private | |
5582 | * when the eb hasn't yet been inserted into radix tree. | |
5583 | * | |
5584 | * The ref will be decreased when the eb released the page, in | |
5585 | * detach_extent_buffer_page(). | |
5586 | * Thus needs no special handling in error path. | |
5587 | */ | |
5588 | btrfs_page_inc_eb_refs(fs_info, p); | |
4f2de97a | 5589 | spin_unlock(&mapping->private_lock); |
760f991f | 5590 | |
0b32f4bb | 5591 | WARN_ON(PageDirty(p)); |
727011e0 | 5592 | eb->pages[i] = p; |
d1310b2e CM |
5593 | if (!PageUptodate(p)) |
5594 | uptodate = 0; | |
eb14ab8e CM |
5595 | |
5596 | /* | |
b16d011e NB |
5597 | * We can't unlock the pages just yet since the extent buffer |
5598 | * hasn't been properly inserted in the radix tree, this | |
5599 | * opens a race with btree_releasepage which can free a page | |
5600 | * while we are still filling in all pages for the buffer and | |
5601 | * we could crash. | |
eb14ab8e | 5602 | */ |
d1310b2e CM |
5603 | } |
5604 | if (uptodate) | |
b4ce94de | 5605 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 5606 | again: |
e1860a77 | 5607 | ret = radix_tree_preload(GFP_NOFS); |
c871b0f2 LB |
5608 | if (ret) { |
5609 | exists = ERR_PTR(ret); | |
19fe0a8b | 5610 | goto free_eb; |
c871b0f2 | 5611 | } |
19fe0a8b | 5612 | |
f28491e0 JB |
5613 | spin_lock(&fs_info->buffer_lock); |
5614 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
478ef886 | 5615 | start >> fs_info->sectorsize_bits, eb); |
f28491e0 | 5616 | spin_unlock(&fs_info->buffer_lock); |
452c75c3 | 5617 | radix_tree_preload_end(); |
19fe0a8b | 5618 | if (ret == -EEXIST) { |
f28491e0 | 5619 | exists = find_extent_buffer(fs_info, start); |
452c75c3 CS |
5620 | if (exists) |
5621 | goto free_eb; | |
5622 | else | |
115391d2 | 5623 | goto again; |
6af118ce | 5624 | } |
6af118ce | 5625 | /* add one reference for the tree */ |
0b32f4bb | 5626 | check_buffer_tree_ref(eb); |
34b41ace | 5627 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
5628 | |
5629 | /* | |
b16d011e NB |
5630 | * Now it's safe to unlock the pages because any calls to |
5631 | * btree_releasepage will correctly detect that a page belongs to a | |
5632 | * live buffer and won't free them prematurely. | |
eb14ab8e | 5633 | */ |
28187ae5 NB |
5634 | for (i = 0; i < num_pages; i++) |
5635 | unlock_page(eb->pages[i]); | |
d1310b2e CM |
5636 | return eb; |
5637 | ||
6af118ce | 5638 | free_eb: |
5ca64f45 | 5639 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
5640 | for (i = 0; i < num_pages; i++) { |
5641 | if (eb->pages[i]) | |
5642 | unlock_page(eb->pages[i]); | |
5643 | } | |
eb14ab8e | 5644 | |
897ca6e9 | 5645 | btrfs_release_extent_buffer(eb); |
6af118ce | 5646 | return exists; |
d1310b2e | 5647 | } |
d1310b2e | 5648 | |
3083ee2e JB |
5649 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
5650 | { | |
5651 | struct extent_buffer *eb = | |
5652 | container_of(head, struct extent_buffer, rcu_head); | |
5653 | ||
5654 | __free_extent_buffer(eb); | |
5655 | } | |
5656 | ||
f7a52a40 | 5657 | static int release_extent_buffer(struct extent_buffer *eb) |
5ce48d0f | 5658 | __releases(&eb->refs_lock) |
3083ee2e | 5659 | { |
07e21c4d NB |
5660 | lockdep_assert_held(&eb->refs_lock); |
5661 | ||
3083ee2e JB |
5662 | WARN_ON(atomic_read(&eb->refs) == 0); |
5663 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 5664 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 5665 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 5666 | |
815a51c7 | 5667 | spin_unlock(&eb->refs_lock); |
3083ee2e | 5668 | |
f28491e0 JB |
5669 | spin_lock(&fs_info->buffer_lock); |
5670 | radix_tree_delete(&fs_info->buffer_radix, | |
478ef886 | 5671 | eb->start >> fs_info->sectorsize_bits); |
f28491e0 | 5672 | spin_unlock(&fs_info->buffer_lock); |
34b41ace JB |
5673 | } else { |
5674 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 5675 | } |
3083ee2e | 5676 | |
8c38938c | 5677 | btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list); |
3083ee2e | 5678 | /* Should be safe to release our pages at this point */ |
55ac0139 | 5679 | btrfs_release_extent_buffer_pages(eb); |
bcb7e449 | 5680 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
b0132a3b | 5681 | if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) { |
bcb7e449 JB |
5682 | __free_extent_buffer(eb); |
5683 | return 1; | |
5684 | } | |
5685 | #endif | |
3083ee2e | 5686 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 5687 | return 1; |
3083ee2e JB |
5688 | } |
5689 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
5690 | |
5691 | return 0; | |
3083ee2e JB |
5692 | } |
5693 | ||
d1310b2e CM |
5694 | void free_extent_buffer(struct extent_buffer *eb) |
5695 | { | |
242e18c7 CM |
5696 | int refs; |
5697 | int old; | |
d1310b2e CM |
5698 | if (!eb) |
5699 | return; | |
5700 | ||
242e18c7 CM |
5701 | while (1) { |
5702 | refs = atomic_read(&eb->refs); | |
46cc775e NB |
5703 | if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3) |
5704 | || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && | |
5705 | refs == 1)) | |
242e18c7 CM |
5706 | break; |
5707 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
5708 | if (old == refs) | |
5709 | return; | |
5710 | } | |
5711 | ||
3083ee2e JB |
5712 | spin_lock(&eb->refs_lock); |
5713 | if (atomic_read(&eb->refs) == 2 && | |
5714 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 5715 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
5716 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5717 | atomic_dec(&eb->refs); | |
5718 | ||
5719 | /* | |
5720 | * I know this is terrible, but it's temporary until we stop tracking | |
5721 | * the uptodate bits and such for the extent buffers. | |
5722 | */ | |
f7a52a40 | 5723 | release_extent_buffer(eb); |
3083ee2e JB |
5724 | } |
5725 | ||
5726 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
5727 | { | |
5728 | if (!eb) | |
d1310b2e CM |
5729 | return; |
5730 | ||
3083ee2e JB |
5731 | spin_lock(&eb->refs_lock); |
5732 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
5733 | ||
0b32f4bb | 5734 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
5735 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5736 | atomic_dec(&eb->refs); | |
f7a52a40 | 5737 | release_extent_buffer(eb); |
d1310b2e | 5738 | } |
d1310b2e | 5739 | |
2b48966a | 5740 | void clear_extent_buffer_dirty(const struct extent_buffer *eb) |
d1310b2e | 5741 | { |
cc5e31a4 DS |
5742 | int i; |
5743 | int num_pages; | |
d1310b2e CM |
5744 | struct page *page; |
5745 | ||
65ad0104 | 5746 | num_pages = num_extent_pages(eb); |
d1310b2e CM |
5747 | |
5748 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 5749 | page = eb->pages[i]; |
b9473439 | 5750 | if (!PageDirty(page)) |
d2c3f4f6 CM |
5751 | continue; |
5752 | ||
a61e6f29 | 5753 | lock_page(page); |
eb14ab8e CM |
5754 | WARN_ON(!PagePrivate(page)); |
5755 | ||
d1310b2e | 5756 | clear_page_dirty_for_io(page); |
b93b0163 | 5757 | xa_lock_irq(&page->mapping->i_pages); |
0a943c65 MW |
5758 | if (!PageDirty(page)) |
5759 | __xa_clear_mark(&page->mapping->i_pages, | |
5760 | page_index(page), PAGECACHE_TAG_DIRTY); | |
b93b0163 | 5761 | xa_unlock_irq(&page->mapping->i_pages); |
bf0da8c1 | 5762 | ClearPageError(page); |
a61e6f29 | 5763 | unlock_page(page); |
d1310b2e | 5764 | } |
0b32f4bb | 5765 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 5766 | } |
d1310b2e | 5767 | |
abb57ef3 | 5768 | bool set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5769 | { |
cc5e31a4 DS |
5770 | int i; |
5771 | int num_pages; | |
abb57ef3 | 5772 | bool was_dirty; |
d1310b2e | 5773 | |
0b32f4bb JB |
5774 | check_buffer_tree_ref(eb); |
5775 | ||
b9473439 | 5776 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 5777 | |
65ad0104 | 5778 | num_pages = num_extent_pages(eb); |
3083ee2e | 5779 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
5780 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
5781 | ||
abb57ef3 LB |
5782 | if (!was_dirty) |
5783 | for (i = 0; i < num_pages; i++) | |
5784 | set_page_dirty(eb->pages[i]); | |
51995c39 LB |
5785 | |
5786 | #ifdef CONFIG_BTRFS_DEBUG | |
5787 | for (i = 0; i < num_pages; i++) | |
5788 | ASSERT(PageDirty(eb->pages[i])); | |
5789 | #endif | |
5790 | ||
b9473439 | 5791 | return was_dirty; |
d1310b2e | 5792 | } |
d1310b2e | 5793 | |
69ba3927 | 5794 | void clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 | 5795 | { |
251f2acc | 5796 | struct btrfs_fs_info *fs_info = eb->fs_info; |
1259ab75 | 5797 | struct page *page; |
cc5e31a4 | 5798 | int num_pages; |
251f2acc | 5799 | int i; |
1259ab75 | 5800 | |
b4ce94de | 5801 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 5802 | num_pages = num_extent_pages(eb); |
1259ab75 | 5803 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5804 | page = eb->pages[i]; |
33958dc6 | 5805 | if (page) |
251f2acc QW |
5806 | btrfs_page_clear_uptodate(fs_info, page, |
5807 | eb->start, eb->len); | |
1259ab75 | 5808 | } |
1259ab75 CM |
5809 | } |
5810 | ||
09c25a8c | 5811 | void set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 5812 | { |
251f2acc | 5813 | struct btrfs_fs_info *fs_info = eb->fs_info; |
d1310b2e | 5814 | struct page *page; |
cc5e31a4 | 5815 | int num_pages; |
251f2acc | 5816 | int i; |
d1310b2e | 5817 | |
0b32f4bb | 5818 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 5819 | num_pages = num_extent_pages(eb); |
d1310b2e | 5820 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5821 | page = eb->pages[i]; |
251f2acc | 5822 | btrfs_page_set_uptodate(fs_info, page, eb->start, eb->len); |
d1310b2e | 5823 | } |
d1310b2e | 5824 | } |
d1310b2e | 5825 | |
4012daf7 QW |
5826 | static int read_extent_buffer_subpage(struct extent_buffer *eb, int wait, |
5827 | int mirror_num) | |
5828 | { | |
5829 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
5830 | struct extent_io_tree *io_tree; | |
5831 | struct page *page = eb->pages[0]; | |
5832 | struct bio *bio = NULL; | |
5833 | int ret = 0; | |
5834 | ||
5835 | ASSERT(!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags)); | |
5836 | ASSERT(PagePrivate(page)); | |
5837 | io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; | |
5838 | ||
5839 | if (wait == WAIT_NONE) { | |
5840 | ret = try_lock_extent(io_tree, eb->start, | |
5841 | eb->start + eb->len - 1); | |
5842 | if (ret <= 0) | |
5843 | return ret; | |
5844 | } else { | |
5845 | ret = lock_extent(io_tree, eb->start, eb->start + eb->len - 1); | |
5846 | if (ret < 0) | |
5847 | return ret; | |
5848 | } | |
5849 | ||
5850 | ret = 0; | |
5851 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags) || | |
5852 | PageUptodate(page) || | |
5853 | btrfs_subpage_test_uptodate(fs_info, page, eb->start, eb->len)) { | |
5854 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
5855 | unlock_extent(io_tree, eb->start, eb->start + eb->len - 1); | |
5856 | return ret; | |
5857 | } | |
5858 | ||
5859 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); | |
5860 | eb->read_mirror = 0; | |
5861 | atomic_set(&eb->io_pages, 1); | |
5862 | check_buffer_tree_ref(eb); | |
5863 | btrfs_subpage_clear_error(fs_info, page, eb->start, eb->len); | |
5864 | ||
5865 | ret = submit_extent_page(REQ_OP_READ | REQ_META, NULL, page, eb->start, | |
5866 | eb->len, eb->start - page_offset(page), &bio, | |
5867 | end_bio_extent_readpage, mirror_num, 0, 0, | |
5868 | true); | |
5869 | if (ret) { | |
5870 | /* | |
5871 | * In the endio function, if we hit something wrong we will | |
5872 | * increase the io_pages, so here we need to decrease it for | |
5873 | * error path. | |
5874 | */ | |
5875 | atomic_dec(&eb->io_pages); | |
5876 | } | |
5877 | if (bio) { | |
5878 | int tmp; | |
5879 | ||
5880 | tmp = submit_one_bio(bio, mirror_num, 0); | |
5881 | if (tmp < 0) | |
5882 | return tmp; | |
5883 | } | |
5884 | if (ret || wait != WAIT_COMPLETE) | |
5885 | return ret; | |
5886 | ||
5887 | wait_extent_bit(io_tree, eb->start, eb->start + eb->len - 1, EXTENT_LOCKED); | |
5888 | if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) | |
5889 | ret = -EIO; | |
5890 | return ret; | |
5891 | } | |
5892 | ||
c2ccfbc6 | 5893 | int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num) |
d1310b2e | 5894 | { |
cc5e31a4 | 5895 | int i; |
d1310b2e CM |
5896 | struct page *page; |
5897 | int err; | |
5898 | int ret = 0; | |
ce9adaa5 CM |
5899 | int locked_pages = 0; |
5900 | int all_uptodate = 1; | |
cc5e31a4 | 5901 | int num_pages; |
727011e0 | 5902 | unsigned long num_reads = 0; |
a86c12c7 | 5903 | struct bio *bio = NULL; |
c8b97818 | 5904 | unsigned long bio_flags = 0; |
a86c12c7 | 5905 | |
b4ce94de | 5906 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
5907 | return 0; |
5908 | ||
4012daf7 QW |
5909 | if (eb->fs_info->sectorsize < PAGE_SIZE) |
5910 | return read_extent_buffer_subpage(eb, wait, mirror_num); | |
5911 | ||
65ad0104 | 5912 | num_pages = num_extent_pages(eb); |
8436ea91 | 5913 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5914 | page = eb->pages[i]; |
bb82ab88 | 5915 | if (wait == WAIT_NONE) { |
2c4d8cb7 QW |
5916 | /* |
5917 | * WAIT_NONE is only utilized by readahead. If we can't | |
5918 | * acquire the lock atomically it means either the eb | |
5919 | * is being read out or under modification. | |
5920 | * Either way the eb will be or has been cached, | |
5921 | * readahead can exit safely. | |
5922 | */ | |
2db04966 | 5923 | if (!trylock_page(page)) |
ce9adaa5 | 5924 | goto unlock_exit; |
d1310b2e CM |
5925 | } else { |
5926 | lock_page(page); | |
5927 | } | |
ce9adaa5 | 5928 | locked_pages++; |
2571e739 LB |
5929 | } |
5930 | /* | |
5931 | * We need to firstly lock all pages to make sure that | |
5932 | * the uptodate bit of our pages won't be affected by | |
5933 | * clear_extent_buffer_uptodate(). | |
5934 | */ | |
8436ea91 | 5935 | for (i = 0; i < num_pages; i++) { |
2571e739 | 5936 | page = eb->pages[i]; |
727011e0 CM |
5937 | if (!PageUptodate(page)) { |
5938 | num_reads++; | |
ce9adaa5 | 5939 | all_uptodate = 0; |
727011e0 | 5940 | } |
ce9adaa5 | 5941 | } |
2571e739 | 5942 | |
ce9adaa5 | 5943 | if (all_uptodate) { |
8436ea91 | 5944 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
5945 | goto unlock_exit; |
5946 | } | |
5947 | ||
656f30db | 5948 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 5949 | eb->read_mirror = 0; |
0b32f4bb | 5950 | atomic_set(&eb->io_pages, num_reads); |
6bf9cd2e BB |
5951 | /* |
5952 | * It is possible for releasepage to clear the TREE_REF bit before we | |
5953 | * set io_pages. See check_buffer_tree_ref for a more detailed comment. | |
5954 | */ | |
5955 | check_buffer_tree_ref(eb); | |
8436ea91 | 5956 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5957 | page = eb->pages[i]; |
baf863b9 | 5958 | |
ce9adaa5 | 5959 | if (!PageUptodate(page)) { |
baf863b9 LB |
5960 | if (ret) { |
5961 | atomic_dec(&eb->io_pages); | |
5962 | unlock_page(page); | |
5963 | continue; | |
5964 | } | |
5965 | ||
f188591e | 5966 | ClearPageError(page); |
0420177c NB |
5967 | err = submit_extent_page(REQ_OP_READ | REQ_META, NULL, |
5968 | page, page_offset(page), PAGE_SIZE, 0, | |
5969 | &bio, end_bio_extent_readpage, | |
5970 | mirror_num, 0, 0, false); | |
baf863b9 | 5971 | if (err) { |
baf863b9 | 5972 | /* |
0420177c NB |
5973 | * We failed to submit the bio so it's the |
5974 | * caller's responsibility to perform cleanup | |
5975 | * i.e unlock page/set error bit. | |
baf863b9 | 5976 | */ |
0420177c NB |
5977 | ret = err; |
5978 | SetPageError(page); | |
5979 | unlock_page(page); | |
baf863b9 LB |
5980 | atomic_dec(&eb->io_pages); |
5981 | } | |
d1310b2e CM |
5982 | } else { |
5983 | unlock_page(page); | |
5984 | } | |
5985 | } | |
5986 | ||
355808c2 | 5987 | if (bio) { |
1f7ad75b | 5988 | err = submit_one_bio(bio, mirror_num, bio_flags); |
79787eaa JM |
5989 | if (err) |
5990 | return err; | |
355808c2 | 5991 | } |
a86c12c7 | 5992 | |
bb82ab88 | 5993 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 5994 | return ret; |
d397712b | 5995 | |
8436ea91 | 5996 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5997 | page = eb->pages[i]; |
d1310b2e | 5998 | wait_on_page_locked(page); |
d397712b | 5999 | if (!PageUptodate(page)) |
d1310b2e | 6000 | ret = -EIO; |
d1310b2e | 6001 | } |
d397712b | 6002 | |
d1310b2e | 6003 | return ret; |
ce9adaa5 CM |
6004 | |
6005 | unlock_exit: | |
d397712b | 6006 | while (locked_pages > 0) { |
ce9adaa5 | 6007 | locked_pages--; |
8436ea91 JB |
6008 | page = eb->pages[locked_pages]; |
6009 | unlock_page(page); | |
ce9adaa5 CM |
6010 | } |
6011 | return ret; | |
d1310b2e | 6012 | } |
d1310b2e | 6013 | |
f98b6215 QW |
6014 | static bool report_eb_range(const struct extent_buffer *eb, unsigned long start, |
6015 | unsigned long len) | |
6016 | { | |
6017 | btrfs_warn(eb->fs_info, | |
6018 | "access to eb bytenr %llu len %lu out of range start %lu len %lu", | |
6019 | eb->start, eb->len, start, len); | |
6020 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
6021 | ||
6022 | return true; | |
6023 | } | |
6024 | ||
6025 | /* | |
6026 | * Check if the [start, start + len) range is valid before reading/writing | |
6027 | * the eb. | |
6028 | * NOTE: @start and @len are offset inside the eb, not logical address. | |
6029 | * | |
6030 | * Caller should not touch the dst/src memory if this function returns error. | |
6031 | */ | |
6032 | static inline int check_eb_range(const struct extent_buffer *eb, | |
6033 | unsigned long start, unsigned long len) | |
6034 | { | |
6035 | unsigned long offset; | |
6036 | ||
6037 | /* start, start + len should not go beyond eb->len nor overflow */ | |
6038 | if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len)) | |
6039 | return report_eb_range(eb, start, len); | |
6040 | ||
6041 | return false; | |
6042 | } | |
6043 | ||
1cbb1f45 JM |
6044 | void read_extent_buffer(const struct extent_buffer *eb, void *dstv, |
6045 | unsigned long start, unsigned long len) | |
d1310b2e CM |
6046 | { |
6047 | size_t cur; | |
6048 | size_t offset; | |
6049 | struct page *page; | |
6050 | char *kaddr; | |
6051 | char *dst = (char *)dstv; | |
884b07d0 | 6052 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 6053 | |
f98b6215 | 6054 | if (check_eb_range(eb, start, len)) |
f716abd5 | 6055 | return; |
d1310b2e | 6056 | |
884b07d0 | 6057 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 6058 | |
d397712b | 6059 | while (len > 0) { |
fb85fc9a | 6060 | page = eb->pages[i]; |
d1310b2e | 6061 | |
09cbfeaf | 6062 | cur = min(len, (PAGE_SIZE - offset)); |
a6591715 | 6063 | kaddr = page_address(page); |
d1310b2e | 6064 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
6065 | |
6066 | dst += cur; | |
6067 | len -= cur; | |
6068 | offset = 0; | |
6069 | i++; | |
6070 | } | |
6071 | } | |
d1310b2e | 6072 | |
a48b73ec JB |
6073 | int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb, |
6074 | void __user *dstv, | |
6075 | unsigned long start, unsigned long len) | |
550ac1d8 GH |
6076 | { |
6077 | size_t cur; | |
6078 | size_t offset; | |
6079 | struct page *page; | |
6080 | char *kaddr; | |
6081 | char __user *dst = (char __user *)dstv; | |
884b07d0 | 6082 | unsigned long i = get_eb_page_index(start); |
550ac1d8 GH |
6083 | int ret = 0; |
6084 | ||
6085 | WARN_ON(start > eb->len); | |
6086 | WARN_ON(start + len > eb->start + eb->len); | |
6087 | ||
884b07d0 | 6088 | offset = get_eb_offset_in_page(eb, start); |
550ac1d8 GH |
6089 | |
6090 | while (len > 0) { | |
fb85fc9a | 6091 | page = eb->pages[i]; |
550ac1d8 | 6092 | |
09cbfeaf | 6093 | cur = min(len, (PAGE_SIZE - offset)); |
550ac1d8 | 6094 | kaddr = page_address(page); |
a48b73ec | 6095 | if (copy_to_user_nofault(dst, kaddr + offset, cur)) { |
550ac1d8 GH |
6096 | ret = -EFAULT; |
6097 | break; | |
6098 | } | |
6099 | ||
6100 | dst += cur; | |
6101 | len -= cur; | |
6102 | offset = 0; | |
6103 | i++; | |
6104 | } | |
6105 | ||
6106 | return ret; | |
6107 | } | |
6108 | ||
1cbb1f45 JM |
6109 | int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv, |
6110 | unsigned long start, unsigned long len) | |
d1310b2e CM |
6111 | { |
6112 | size_t cur; | |
6113 | size_t offset; | |
6114 | struct page *page; | |
6115 | char *kaddr; | |
6116 | char *ptr = (char *)ptrv; | |
884b07d0 | 6117 | unsigned long i = get_eb_page_index(start); |
d1310b2e CM |
6118 | int ret = 0; |
6119 | ||
f98b6215 QW |
6120 | if (check_eb_range(eb, start, len)) |
6121 | return -EINVAL; | |
d1310b2e | 6122 | |
884b07d0 | 6123 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 6124 | |
d397712b | 6125 | while (len > 0) { |
fb85fc9a | 6126 | page = eb->pages[i]; |
d1310b2e | 6127 | |
09cbfeaf | 6128 | cur = min(len, (PAGE_SIZE - offset)); |
d1310b2e | 6129 | |
a6591715 | 6130 | kaddr = page_address(page); |
d1310b2e | 6131 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
6132 | if (ret) |
6133 | break; | |
6134 | ||
6135 | ptr += cur; | |
6136 | len -= cur; | |
6137 | offset = 0; | |
6138 | i++; | |
6139 | } | |
6140 | return ret; | |
6141 | } | |
d1310b2e | 6142 | |
2b48966a | 6143 | void write_extent_buffer_chunk_tree_uuid(const struct extent_buffer *eb, |
f157bf76 DS |
6144 | const void *srcv) |
6145 | { | |
6146 | char *kaddr; | |
6147 | ||
6148 | WARN_ON(!PageUptodate(eb->pages[0])); | |
884b07d0 | 6149 | kaddr = page_address(eb->pages[0]) + get_eb_offset_in_page(eb, 0); |
f157bf76 DS |
6150 | memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv, |
6151 | BTRFS_FSID_SIZE); | |
6152 | } | |
6153 | ||
2b48966a | 6154 | void write_extent_buffer_fsid(const struct extent_buffer *eb, const void *srcv) |
f157bf76 DS |
6155 | { |
6156 | char *kaddr; | |
6157 | ||
6158 | WARN_ON(!PageUptodate(eb->pages[0])); | |
884b07d0 | 6159 | kaddr = page_address(eb->pages[0]) + get_eb_offset_in_page(eb, 0); |
f157bf76 DS |
6160 | memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv, |
6161 | BTRFS_FSID_SIZE); | |
6162 | } | |
6163 | ||
2b48966a | 6164 | void write_extent_buffer(const struct extent_buffer *eb, const void *srcv, |
d1310b2e CM |
6165 | unsigned long start, unsigned long len) |
6166 | { | |
6167 | size_t cur; | |
6168 | size_t offset; | |
6169 | struct page *page; | |
6170 | char *kaddr; | |
6171 | char *src = (char *)srcv; | |
884b07d0 | 6172 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 6173 | |
d3575156 NA |
6174 | WARN_ON(test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)); |
6175 | ||
f98b6215 QW |
6176 | if (check_eb_range(eb, start, len)) |
6177 | return; | |
d1310b2e | 6178 | |
884b07d0 | 6179 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 6180 | |
d397712b | 6181 | while (len > 0) { |
fb85fc9a | 6182 | page = eb->pages[i]; |
d1310b2e CM |
6183 | WARN_ON(!PageUptodate(page)); |
6184 | ||
09cbfeaf | 6185 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 6186 | kaddr = page_address(page); |
d1310b2e | 6187 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
6188 | |
6189 | src += cur; | |
6190 | len -= cur; | |
6191 | offset = 0; | |
6192 | i++; | |
6193 | } | |
6194 | } | |
d1310b2e | 6195 | |
2b48966a | 6196 | void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start, |
b159fa28 | 6197 | unsigned long len) |
d1310b2e CM |
6198 | { |
6199 | size_t cur; | |
6200 | size_t offset; | |
6201 | struct page *page; | |
6202 | char *kaddr; | |
884b07d0 | 6203 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 6204 | |
f98b6215 QW |
6205 | if (check_eb_range(eb, start, len)) |
6206 | return; | |
d1310b2e | 6207 | |
884b07d0 | 6208 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 6209 | |
d397712b | 6210 | while (len > 0) { |
fb85fc9a | 6211 | page = eb->pages[i]; |
d1310b2e CM |
6212 | WARN_ON(!PageUptodate(page)); |
6213 | ||
09cbfeaf | 6214 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 6215 | kaddr = page_address(page); |
b159fa28 | 6216 | memset(kaddr + offset, 0, cur); |
d1310b2e CM |
6217 | |
6218 | len -= cur; | |
6219 | offset = 0; | |
6220 | i++; | |
6221 | } | |
6222 | } | |
d1310b2e | 6223 | |
2b48966a DS |
6224 | void copy_extent_buffer_full(const struct extent_buffer *dst, |
6225 | const struct extent_buffer *src) | |
58e8012c DS |
6226 | { |
6227 | int i; | |
cc5e31a4 | 6228 | int num_pages; |
58e8012c DS |
6229 | |
6230 | ASSERT(dst->len == src->len); | |
6231 | ||
884b07d0 QW |
6232 | if (dst->fs_info->sectorsize == PAGE_SIZE) { |
6233 | num_pages = num_extent_pages(dst); | |
6234 | for (i = 0; i < num_pages; i++) | |
6235 | copy_page(page_address(dst->pages[i]), | |
6236 | page_address(src->pages[i])); | |
6237 | } else { | |
6238 | size_t src_offset = get_eb_offset_in_page(src, 0); | |
6239 | size_t dst_offset = get_eb_offset_in_page(dst, 0); | |
6240 | ||
6241 | ASSERT(src->fs_info->sectorsize < PAGE_SIZE); | |
6242 | memcpy(page_address(dst->pages[0]) + dst_offset, | |
6243 | page_address(src->pages[0]) + src_offset, | |
6244 | src->len); | |
6245 | } | |
58e8012c DS |
6246 | } |
6247 | ||
2b48966a DS |
6248 | void copy_extent_buffer(const struct extent_buffer *dst, |
6249 | const struct extent_buffer *src, | |
d1310b2e CM |
6250 | unsigned long dst_offset, unsigned long src_offset, |
6251 | unsigned long len) | |
6252 | { | |
6253 | u64 dst_len = dst->len; | |
6254 | size_t cur; | |
6255 | size_t offset; | |
6256 | struct page *page; | |
6257 | char *kaddr; | |
884b07d0 | 6258 | unsigned long i = get_eb_page_index(dst_offset); |
d1310b2e | 6259 | |
f98b6215 QW |
6260 | if (check_eb_range(dst, dst_offset, len) || |
6261 | check_eb_range(src, src_offset, len)) | |
6262 | return; | |
6263 | ||
d1310b2e CM |
6264 | WARN_ON(src->len != dst_len); |
6265 | ||
884b07d0 | 6266 | offset = get_eb_offset_in_page(dst, dst_offset); |
d1310b2e | 6267 | |
d397712b | 6268 | while (len > 0) { |
fb85fc9a | 6269 | page = dst->pages[i]; |
d1310b2e CM |
6270 | WARN_ON(!PageUptodate(page)); |
6271 | ||
09cbfeaf | 6272 | cur = min(len, (unsigned long)(PAGE_SIZE - offset)); |
d1310b2e | 6273 | |
a6591715 | 6274 | kaddr = page_address(page); |
d1310b2e | 6275 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
6276 | |
6277 | src_offset += cur; | |
6278 | len -= cur; | |
6279 | offset = 0; | |
6280 | i++; | |
6281 | } | |
6282 | } | |
d1310b2e | 6283 | |
3e1e8bb7 OS |
6284 | /* |
6285 | * eb_bitmap_offset() - calculate the page and offset of the byte containing the | |
6286 | * given bit number | |
6287 | * @eb: the extent buffer | |
6288 | * @start: offset of the bitmap item in the extent buffer | |
6289 | * @nr: bit number | |
6290 | * @page_index: return index of the page in the extent buffer that contains the | |
6291 | * given bit number | |
6292 | * @page_offset: return offset into the page given by page_index | |
6293 | * | |
6294 | * This helper hides the ugliness of finding the byte in an extent buffer which | |
6295 | * contains a given bit. | |
6296 | */ | |
2b48966a | 6297 | static inline void eb_bitmap_offset(const struct extent_buffer *eb, |
3e1e8bb7 OS |
6298 | unsigned long start, unsigned long nr, |
6299 | unsigned long *page_index, | |
6300 | size_t *page_offset) | |
6301 | { | |
3e1e8bb7 OS |
6302 | size_t byte_offset = BIT_BYTE(nr); |
6303 | size_t offset; | |
6304 | ||
6305 | /* | |
6306 | * The byte we want is the offset of the extent buffer + the offset of | |
6307 | * the bitmap item in the extent buffer + the offset of the byte in the | |
6308 | * bitmap item. | |
6309 | */ | |
884b07d0 | 6310 | offset = start + offset_in_page(eb->start) + byte_offset; |
3e1e8bb7 | 6311 | |
09cbfeaf | 6312 | *page_index = offset >> PAGE_SHIFT; |
7073017a | 6313 | *page_offset = offset_in_page(offset); |
3e1e8bb7 OS |
6314 | } |
6315 | ||
6316 | /** | |
6317 | * extent_buffer_test_bit - determine whether a bit in a bitmap item is set | |
6318 | * @eb: the extent buffer | |
6319 | * @start: offset of the bitmap item in the extent buffer | |
6320 | * @nr: bit number to test | |
6321 | */ | |
2b48966a | 6322 | int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start, |
3e1e8bb7 OS |
6323 | unsigned long nr) |
6324 | { | |
2fe1d551 | 6325 | u8 *kaddr; |
3e1e8bb7 OS |
6326 | struct page *page; |
6327 | unsigned long i; | |
6328 | size_t offset; | |
6329 | ||
6330 | eb_bitmap_offset(eb, start, nr, &i, &offset); | |
6331 | page = eb->pages[i]; | |
6332 | WARN_ON(!PageUptodate(page)); | |
6333 | kaddr = page_address(page); | |
6334 | return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1))); | |
6335 | } | |
6336 | ||
6337 | /** | |
6338 | * extent_buffer_bitmap_set - set an area of a bitmap | |
6339 | * @eb: the extent buffer | |
6340 | * @start: offset of the bitmap item in the extent buffer | |
6341 | * @pos: bit number of the first bit | |
6342 | * @len: number of bits to set | |
6343 | */ | |
2b48966a | 6344 | void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start, |
3e1e8bb7 OS |
6345 | unsigned long pos, unsigned long len) |
6346 | { | |
2fe1d551 | 6347 | u8 *kaddr; |
3e1e8bb7 OS |
6348 | struct page *page; |
6349 | unsigned long i; | |
6350 | size_t offset; | |
6351 | const unsigned int size = pos + len; | |
6352 | int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 6353 | u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
6354 | |
6355 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
6356 | page = eb->pages[i]; | |
6357 | WARN_ON(!PageUptodate(page)); | |
6358 | kaddr = page_address(page); | |
6359 | ||
6360 | while (len >= bits_to_set) { | |
6361 | kaddr[offset] |= mask_to_set; | |
6362 | len -= bits_to_set; | |
6363 | bits_to_set = BITS_PER_BYTE; | |
9c894696 | 6364 | mask_to_set = ~0; |
09cbfeaf | 6365 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
6366 | offset = 0; |
6367 | page = eb->pages[++i]; | |
6368 | WARN_ON(!PageUptodate(page)); | |
6369 | kaddr = page_address(page); | |
6370 | } | |
6371 | } | |
6372 | if (len) { | |
6373 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
6374 | kaddr[offset] |= mask_to_set; | |
6375 | } | |
6376 | } | |
6377 | ||
6378 | ||
6379 | /** | |
6380 | * extent_buffer_bitmap_clear - clear an area of a bitmap | |
6381 | * @eb: the extent buffer | |
6382 | * @start: offset of the bitmap item in the extent buffer | |
6383 | * @pos: bit number of the first bit | |
6384 | * @len: number of bits to clear | |
6385 | */ | |
2b48966a DS |
6386 | void extent_buffer_bitmap_clear(const struct extent_buffer *eb, |
6387 | unsigned long start, unsigned long pos, | |
6388 | unsigned long len) | |
3e1e8bb7 | 6389 | { |
2fe1d551 | 6390 | u8 *kaddr; |
3e1e8bb7 OS |
6391 | struct page *page; |
6392 | unsigned long i; | |
6393 | size_t offset; | |
6394 | const unsigned int size = pos + len; | |
6395 | int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 6396 | u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
6397 | |
6398 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
6399 | page = eb->pages[i]; | |
6400 | WARN_ON(!PageUptodate(page)); | |
6401 | kaddr = page_address(page); | |
6402 | ||
6403 | while (len >= bits_to_clear) { | |
6404 | kaddr[offset] &= ~mask_to_clear; | |
6405 | len -= bits_to_clear; | |
6406 | bits_to_clear = BITS_PER_BYTE; | |
9c894696 | 6407 | mask_to_clear = ~0; |
09cbfeaf | 6408 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
6409 | offset = 0; |
6410 | page = eb->pages[++i]; | |
6411 | WARN_ON(!PageUptodate(page)); | |
6412 | kaddr = page_address(page); | |
6413 | } | |
6414 | } | |
6415 | if (len) { | |
6416 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
6417 | kaddr[offset] &= ~mask_to_clear; | |
6418 | } | |
6419 | } | |
6420 | ||
3387206f ST |
6421 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
6422 | { | |
6423 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
6424 | return distance < len; | |
6425 | } | |
6426 | ||
d1310b2e CM |
6427 | static void copy_pages(struct page *dst_page, struct page *src_page, |
6428 | unsigned long dst_off, unsigned long src_off, | |
6429 | unsigned long len) | |
6430 | { | |
a6591715 | 6431 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 6432 | char *src_kaddr; |
727011e0 | 6433 | int must_memmove = 0; |
d1310b2e | 6434 | |
3387206f | 6435 | if (dst_page != src_page) { |
a6591715 | 6436 | src_kaddr = page_address(src_page); |
3387206f | 6437 | } else { |
d1310b2e | 6438 | src_kaddr = dst_kaddr; |
727011e0 CM |
6439 | if (areas_overlap(src_off, dst_off, len)) |
6440 | must_memmove = 1; | |
3387206f | 6441 | } |
d1310b2e | 6442 | |
727011e0 CM |
6443 | if (must_memmove) |
6444 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
6445 | else | |
6446 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
6447 | } |
6448 | ||
2b48966a DS |
6449 | void memcpy_extent_buffer(const struct extent_buffer *dst, |
6450 | unsigned long dst_offset, unsigned long src_offset, | |
6451 | unsigned long len) | |
d1310b2e CM |
6452 | { |
6453 | size_t cur; | |
6454 | size_t dst_off_in_page; | |
6455 | size_t src_off_in_page; | |
d1310b2e CM |
6456 | unsigned long dst_i; |
6457 | unsigned long src_i; | |
6458 | ||
f98b6215 QW |
6459 | if (check_eb_range(dst, dst_offset, len) || |
6460 | check_eb_range(dst, src_offset, len)) | |
6461 | return; | |
d1310b2e | 6462 | |
d397712b | 6463 | while (len > 0) { |
884b07d0 QW |
6464 | dst_off_in_page = get_eb_offset_in_page(dst, dst_offset); |
6465 | src_off_in_page = get_eb_offset_in_page(dst, src_offset); | |
d1310b2e | 6466 | |
884b07d0 QW |
6467 | dst_i = get_eb_page_index(dst_offset); |
6468 | src_i = get_eb_page_index(src_offset); | |
d1310b2e | 6469 | |
09cbfeaf | 6470 | cur = min(len, (unsigned long)(PAGE_SIZE - |
d1310b2e CM |
6471 | src_off_in_page)); |
6472 | cur = min_t(unsigned long, cur, | |
09cbfeaf | 6473 | (unsigned long)(PAGE_SIZE - dst_off_in_page)); |
d1310b2e | 6474 | |
fb85fc9a | 6475 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
6476 | dst_off_in_page, src_off_in_page, cur); |
6477 | ||
6478 | src_offset += cur; | |
6479 | dst_offset += cur; | |
6480 | len -= cur; | |
6481 | } | |
6482 | } | |
d1310b2e | 6483 | |
2b48966a DS |
6484 | void memmove_extent_buffer(const struct extent_buffer *dst, |
6485 | unsigned long dst_offset, unsigned long src_offset, | |
6486 | unsigned long len) | |
d1310b2e CM |
6487 | { |
6488 | size_t cur; | |
6489 | size_t dst_off_in_page; | |
6490 | size_t src_off_in_page; | |
6491 | unsigned long dst_end = dst_offset + len - 1; | |
6492 | unsigned long src_end = src_offset + len - 1; | |
d1310b2e CM |
6493 | unsigned long dst_i; |
6494 | unsigned long src_i; | |
6495 | ||
f98b6215 QW |
6496 | if (check_eb_range(dst, dst_offset, len) || |
6497 | check_eb_range(dst, src_offset, len)) | |
6498 | return; | |
727011e0 | 6499 | if (dst_offset < src_offset) { |
d1310b2e CM |
6500 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
6501 | return; | |
6502 | } | |
d397712b | 6503 | while (len > 0) { |
884b07d0 QW |
6504 | dst_i = get_eb_page_index(dst_end); |
6505 | src_i = get_eb_page_index(src_end); | |
d1310b2e | 6506 | |
884b07d0 QW |
6507 | dst_off_in_page = get_eb_offset_in_page(dst, dst_end); |
6508 | src_off_in_page = get_eb_offset_in_page(dst, src_end); | |
d1310b2e CM |
6509 | |
6510 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
6511 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 6512 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
6513 | dst_off_in_page - cur + 1, |
6514 | src_off_in_page - cur + 1, cur); | |
6515 | ||
6516 | dst_end -= cur; | |
6517 | src_end -= cur; | |
6518 | len -= cur; | |
6519 | } | |
6520 | } | |
6af118ce | 6521 | |
d1e86e3f QW |
6522 | static struct extent_buffer *get_next_extent_buffer( |
6523 | struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr) | |
6524 | { | |
6525 | struct extent_buffer *gang[BTRFS_SUBPAGE_BITMAP_SIZE]; | |
6526 | struct extent_buffer *found = NULL; | |
6527 | u64 page_start = page_offset(page); | |
6528 | int ret; | |
6529 | int i; | |
6530 | ||
6531 | ASSERT(in_range(bytenr, page_start, PAGE_SIZE)); | |
6532 | ASSERT(PAGE_SIZE / fs_info->nodesize <= BTRFS_SUBPAGE_BITMAP_SIZE); | |
6533 | lockdep_assert_held(&fs_info->buffer_lock); | |
6534 | ||
6535 | ret = radix_tree_gang_lookup(&fs_info->buffer_radix, (void **)gang, | |
6536 | bytenr >> fs_info->sectorsize_bits, | |
6537 | PAGE_SIZE / fs_info->nodesize); | |
6538 | for (i = 0; i < ret; i++) { | |
6539 | /* Already beyond page end */ | |
6540 | if (gang[i]->start >= page_start + PAGE_SIZE) | |
6541 | break; | |
6542 | /* Found one */ | |
6543 | if (gang[i]->start >= bytenr) { | |
6544 | found = gang[i]; | |
6545 | break; | |
6546 | } | |
6547 | } | |
6548 | return found; | |
6549 | } | |
6550 | ||
6551 | static int try_release_subpage_extent_buffer(struct page *page) | |
6552 | { | |
6553 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
6554 | u64 cur = page_offset(page); | |
6555 | const u64 end = page_offset(page) + PAGE_SIZE; | |
6556 | int ret; | |
6557 | ||
6558 | while (cur < end) { | |
6559 | struct extent_buffer *eb = NULL; | |
6560 | ||
6561 | /* | |
6562 | * Unlike try_release_extent_buffer() which uses page->private | |
6563 | * to grab buffer, for subpage case we rely on radix tree, thus | |
6564 | * we need to ensure radix tree consistency. | |
6565 | * | |
6566 | * We also want an atomic snapshot of the radix tree, thus go | |
6567 | * with spinlock rather than RCU. | |
6568 | */ | |
6569 | spin_lock(&fs_info->buffer_lock); | |
6570 | eb = get_next_extent_buffer(fs_info, page, cur); | |
6571 | if (!eb) { | |
6572 | /* No more eb in the page range after or at cur */ | |
6573 | spin_unlock(&fs_info->buffer_lock); | |
6574 | break; | |
6575 | } | |
6576 | cur = eb->start + eb->len; | |
6577 | ||
6578 | /* | |
6579 | * The same as try_release_extent_buffer(), to ensure the eb | |
6580 | * won't disappear out from under us. | |
6581 | */ | |
6582 | spin_lock(&eb->refs_lock); | |
6583 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { | |
6584 | spin_unlock(&eb->refs_lock); | |
6585 | spin_unlock(&fs_info->buffer_lock); | |
6586 | break; | |
6587 | } | |
6588 | spin_unlock(&fs_info->buffer_lock); | |
6589 | ||
6590 | /* | |
6591 | * If tree ref isn't set then we know the ref on this eb is a | |
6592 | * real ref, so just return, this eb will likely be freed soon | |
6593 | * anyway. | |
6594 | */ | |
6595 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { | |
6596 | spin_unlock(&eb->refs_lock); | |
6597 | break; | |
6598 | } | |
6599 | ||
6600 | /* | |
6601 | * Here we don't care about the return value, we will always | |
6602 | * check the page private at the end. And | |
6603 | * release_extent_buffer() will release the refs_lock. | |
6604 | */ | |
6605 | release_extent_buffer(eb); | |
6606 | } | |
6607 | /* | |
6608 | * Finally to check if we have cleared page private, as if we have | |
6609 | * released all ebs in the page, the page private should be cleared now. | |
6610 | */ | |
6611 | spin_lock(&page->mapping->private_lock); | |
6612 | if (!PagePrivate(page)) | |
6613 | ret = 1; | |
6614 | else | |
6615 | ret = 0; | |
6616 | spin_unlock(&page->mapping->private_lock); | |
6617 | return ret; | |
6618 | ||
6619 | } | |
6620 | ||
f7a52a40 | 6621 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 6622 | { |
6af118ce | 6623 | struct extent_buffer *eb; |
6af118ce | 6624 | |
d1e86e3f QW |
6625 | if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE) |
6626 | return try_release_subpage_extent_buffer(page); | |
6627 | ||
3083ee2e | 6628 | /* |
d1e86e3f QW |
6629 | * We need to make sure nobody is changing page->private, as we rely on |
6630 | * page->private as the pointer to extent buffer. | |
3083ee2e JB |
6631 | */ |
6632 | spin_lock(&page->mapping->private_lock); | |
6633 | if (!PagePrivate(page)) { | |
6634 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 6635 | return 1; |
45f49bce | 6636 | } |
6af118ce | 6637 | |
3083ee2e JB |
6638 | eb = (struct extent_buffer *)page->private; |
6639 | BUG_ON(!eb); | |
19fe0a8b MX |
6640 | |
6641 | /* | |
3083ee2e JB |
6642 | * This is a little awful but should be ok, we need to make sure that |
6643 | * the eb doesn't disappear out from under us while we're looking at | |
6644 | * this page. | |
19fe0a8b | 6645 | */ |
3083ee2e | 6646 | spin_lock(&eb->refs_lock); |
0b32f4bb | 6647 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
6648 | spin_unlock(&eb->refs_lock); |
6649 | spin_unlock(&page->mapping->private_lock); | |
6650 | return 0; | |
b9473439 | 6651 | } |
3083ee2e | 6652 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 6653 | |
19fe0a8b | 6654 | /* |
3083ee2e JB |
6655 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
6656 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 6657 | */ |
3083ee2e JB |
6658 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
6659 | spin_unlock(&eb->refs_lock); | |
6660 | return 0; | |
b9473439 | 6661 | } |
19fe0a8b | 6662 | |
f7a52a40 | 6663 | return release_extent_buffer(eb); |
6af118ce | 6664 | } |
bfb484d9 JB |
6665 | |
6666 | /* | |
6667 | * btrfs_readahead_tree_block - attempt to readahead a child block | |
6668 | * @fs_info: the fs_info | |
6669 | * @bytenr: bytenr to read | |
3fbaf258 | 6670 | * @owner_root: objectid of the root that owns this eb |
bfb484d9 | 6671 | * @gen: generation for the uptodate check, can be 0 |
3fbaf258 | 6672 | * @level: level for the eb |
bfb484d9 JB |
6673 | * |
6674 | * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a | |
6675 | * normal uptodate check of the eb, without checking the generation. If we have | |
6676 | * to read the block we will not block on anything. | |
6677 | */ | |
6678 | void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info, | |
3fbaf258 | 6679 | u64 bytenr, u64 owner_root, u64 gen, int level) |
bfb484d9 JB |
6680 | { |
6681 | struct extent_buffer *eb; | |
6682 | int ret; | |
6683 | ||
3fbaf258 | 6684 | eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level); |
bfb484d9 JB |
6685 | if (IS_ERR(eb)) |
6686 | return; | |
6687 | ||
6688 | if (btrfs_buffer_uptodate(eb, gen, 1)) { | |
6689 | free_extent_buffer(eb); | |
6690 | return; | |
6691 | } | |
6692 | ||
6693 | ret = read_extent_buffer_pages(eb, WAIT_NONE, 0); | |
6694 | if (ret < 0) | |
6695 | free_extent_buffer_stale(eb); | |
6696 | else | |
6697 | free_extent_buffer(eb); | |
6698 | } | |
6699 | ||
6700 | /* | |
6701 | * btrfs_readahead_node_child - readahead a node's child block | |
6702 | * @node: parent node we're reading from | |
6703 | * @slot: slot in the parent node for the child we want to read | |
6704 | * | |
6705 | * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at | |
6706 | * the slot in the node provided. | |
6707 | */ | |
6708 | void btrfs_readahead_node_child(struct extent_buffer *node, int slot) | |
6709 | { | |
6710 | btrfs_readahead_tree_block(node->fs_info, | |
6711 | btrfs_node_blockptr(node, slot), | |
3fbaf258 JB |
6712 | btrfs_header_owner(node), |
6713 | btrfs_node_ptr_generation(node, slot), | |
6714 | btrfs_header_level(node) - 1); | |
bfb484d9 | 6715 | } |