Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
c1d7c514 | 2 | |
d1310b2e CM |
3 | #include <linux/bitops.h> |
4 | #include <linux/slab.h> | |
5 | #include <linux/bio.h> | |
6 | #include <linux/mm.h> | |
d1310b2e CM |
7 | #include <linux/pagemap.h> |
8 | #include <linux/page-flags.h> | |
d1310b2e CM |
9 | #include <linux/spinlock.h> |
10 | #include <linux/blkdev.h> | |
11 | #include <linux/swap.h> | |
d1310b2e CM |
12 | #include <linux/writeback.h> |
13 | #include <linux/pagevec.h> | |
268bb0ce | 14 | #include <linux/prefetch.h> |
90a887c9 | 15 | #include <linux/cleancache.h> |
d1310b2e CM |
16 | #include "extent_io.h" |
17 | #include "extent_map.h" | |
902b22f3 DW |
18 | #include "ctree.h" |
19 | #include "btrfs_inode.h" | |
4a54c8c1 | 20 | #include "volumes.h" |
21adbd5c | 21 | #include "check-integrity.h" |
0b32f4bb | 22 | #include "locking.h" |
606686ee | 23 | #include "rcu-string.h" |
fe09e16c | 24 | #include "backref.h" |
6af49dbd | 25 | #include "disk-io.h" |
d1310b2e | 26 | |
d1310b2e CM |
27 | static struct kmem_cache *extent_state_cache; |
28 | static struct kmem_cache *extent_buffer_cache; | |
8ac9f7c1 | 29 | static struct bio_set btrfs_bioset; |
d1310b2e | 30 | |
27a3507d FM |
31 | static inline bool extent_state_in_tree(const struct extent_state *state) |
32 | { | |
33 | return !RB_EMPTY_NODE(&state->rb_node); | |
34 | } | |
35 | ||
6d49ba1b | 36 | #ifdef CONFIG_BTRFS_DEBUG |
d1310b2e CM |
37 | static LIST_HEAD(buffers); |
38 | static LIST_HEAD(states); | |
4bef0848 | 39 | |
d397712b | 40 | static DEFINE_SPINLOCK(leak_lock); |
6d49ba1b ES |
41 | |
42 | static inline | |
43 | void btrfs_leak_debug_add(struct list_head *new, struct list_head *head) | |
44 | { | |
45 | unsigned long flags; | |
46 | ||
47 | spin_lock_irqsave(&leak_lock, flags); | |
48 | list_add(new, head); | |
49 | spin_unlock_irqrestore(&leak_lock, flags); | |
50 | } | |
51 | ||
52 | static inline | |
53 | void btrfs_leak_debug_del(struct list_head *entry) | |
54 | { | |
55 | unsigned long flags; | |
56 | ||
57 | spin_lock_irqsave(&leak_lock, flags); | |
58 | list_del(entry); | |
59 | spin_unlock_irqrestore(&leak_lock, flags); | |
60 | } | |
61 | ||
62 | static inline | |
63 | void btrfs_leak_debug_check(void) | |
64 | { | |
65 | struct extent_state *state; | |
66 | struct extent_buffer *eb; | |
67 | ||
68 | while (!list_empty(&states)) { | |
69 | state = list_entry(states.next, struct extent_state, leak_list); | |
9ee49a04 | 70 | pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n", |
27a3507d FM |
71 | state->start, state->end, state->state, |
72 | extent_state_in_tree(state), | |
b7ac31b7 | 73 | refcount_read(&state->refs)); |
6d49ba1b ES |
74 | list_del(&state->leak_list); |
75 | kmem_cache_free(extent_state_cache, state); | |
76 | } | |
77 | ||
78 | while (!list_empty(&buffers)) { | |
79 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
af2679e4 LB |
80 | pr_err("BTRFS: buffer leak start %llu len %lu refs %d bflags %lu\n", |
81 | eb->start, eb->len, atomic_read(&eb->refs), eb->bflags); | |
6d49ba1b ES |
82 | list_del(&eb->leak_list); |
83 | kmem_cache_free(extent_buffer_cache, eb); | |
84 | } | |
85 | } | |
8d599ae1 | 86 | |
a5dee37d JB |
87 | #define btrfs_debug_check_extent_io_range(tree, start, end) \ |
88 | __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end)) | |
8d599ae1 | 89 | static inline void __btrfs_debug_check_extent_io_range(const char *caller, |
a5dee37d | 90 | struct extent_io_tree *tree, u64 start, u64 end) |
8d599ae1 | 91 | { |
65a680f6 NB |
92 | struct inode *inode = tree->private_data; |
93 | u64 isize; | |
94 | ||
95 | if (!inode || !is_data_inode(inode)) | |
96 | return; | |
97 | ||
98 | isize = i_size_read(inode); | |
99 | if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) { | |
100 | btrfs_debug_rl(BTRFS_I(inode)->root->fs_info, | |
101 | "%s: ino %llu isize %llu odd range [%llu,%llu]", | |
102 | caller, btrfs_ino(BTRFS_I(inode)), isize, start, end); | |
103 | } | |
8d599ae1 | 104 | } |
6d49ba1b ES |
105 | #else |
106 | #define btrfs_leak_debug_add(new, head) do {} while (0) | |
107 | #define btrfs_leak_debug_del(entry) do {} while (0) | |
108 | #define btrfs_leak_debug_check() do {} while (0) | |
8d599ae1 | 109 | #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0) |
4bef0848 | 110 | #endif |
d1310b2e | 111 | |
d1310b2e CM |
112 | struct tree_entry { |
113 | u64 start; | |
114 | u64 end; | |
d1310b2e CM |
115 | struct rb_node rb_node; |
116 | }; | |
117 | ||
118 | struct extent_page_data { | |
119 | struct bio *bio; | |
120 | struct extent_io_tree *tree; | |
771ed689 CM |
121 | /* tells writepage not to lock the state bits for this range |
122 | * it still does the unlocking | |
123 | */ | |
ffbd517d CM |
124 | unsigned int extent_locked:1; |
125 | ||
70fd7614 | 126 | /* tells the submit_bio code to use REQ_SYNC */ |
ffbd517d | 127 | unsigned int sync_io:1; |
d1310b2e CM |
128 | }; |
129 | ||
57599c7e | 130 | static int add_extent_changeset(struct extent_state *state, unsigned bits, |
d38ed27f QW |
131 | struct extent_changeset *changeset, |
132 | int set) | |
133 | { | |
134 | int ret; | |
135 | ||
136 | if (!changeset) | |
57599c7e | 137 | return 0; |
d38ed27f | 138 | if (set && (state->state & bits) == bits) |
57599c7e | 139 | return 0; |
fefdc557 | 140 | if (!set && (state->state & bits) == 0) |
57599c7e | 141 | return 0; |
d38ed27f | 142 | changeset->bytes_changed += state->end - state->start + 1; |
53d32359 | 143 | ret = ulist_add(&changeset->range_changed, state->start, state->end, |
d38ed27f | 144 | GFP_ATOMIC); |
57599c7e | 145 | return ret; |
d38ed27f QW |
146 | } |
147 | ||
bb58eb9e QW |
148 | static int __must_check submit_one_bio(struct bio *bio, int mirror_num, |
149 | unsigned long bio_flags) | |
150 | { | |
151 | blk_status_t ret = 0; | |
bb58eb9e | 152 | struct extent_io_tree *tree = bio->bi_private; |
bb58eb9e QW |
153 | |
154 | bio->bi_private = NULL; | |
155 | ||
156 | if (tree->ops) | |
157 | ret = tree->ops->submit_bio_hook(tree->private_data, bio, | |
50489a57 | 158 | mirror_num, bio_flags); |
bb58eb9e QW |
159 | else |
160 | btrfsic_submit_bio(bio); | |
161 | ||
162 | return blk_status_to_errno(ret); | |
163 | } | |
164 | ||
3065976b QW |
165 | /* Cleanup unsubmitted bios */ |
166 | static void end_write_bio(struct extent_page_data *epd, int ret) | |
167 | { | |
168 | if (epd->bio) { | |
169 | epd->bio->bi_status = errno_to_blk_status(ret); | |
170 | bio_endio(epd->bio); | |
171 | epd->bio = NULL; | |
172 | } | |
173 | } | |
174 | ||
f4340622 QW |
175 | /* |
176 | * Submit bio from extent page data via submit_one_bio | |
177 | * | |
178 | * Return 0 if everything is OK. | |
179 | * Return <0 for error. | |
180 | */ | |
181 | static int __must_check flush_write_bio(struct extent_page_data *epd) | |
bb58eb9e | 182 | { |
f4340622 | 183 | int ret = 0; |
bb58eb9e | 184 | |
f4340622 | 185 | if (epd->bio) { |
bb58eb9e | 186 | ret = submit_one_bio(epd->bio, 0, 0); |
f4340622 QW |
187 | /* |
188 | * Clean up of epd->bio is handled by its endio function. | |
189 | * And endio is either triggered by successful bio execution | |
190 | * or the error handler of submit bio hook. | |
191 | * So at this point, no matter what happened, we don't need | |
192 | * to clean up epd->bio. | |
193 | */ | |
bb58eb9e QW |
194 | epd->bio = NULL; |
195 | } | |
f4340622 | 196 | return ret; |
bb58eb9e | 197 | } |
e2932ee0 | 198 | |
d1310b2e CM |
199 | int __init extent_io_init(void) |
200 | { | |
837e1972 | 201 | extent_state_cache = kmem_cache_create("btrfs_extent_state", |
9601e3f6 | 202 | sizeof(struct extent_state), 0, |
fba4b697 | 203 | SLAB_MEM_SPREAD, NULL); |
d1310b2e CM |
204 | if (!extent_state_cache) |
205 | return -ENOMEM; | |
206 | ||
837e1972 | 207 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 | 208 | sizeof(struct extent_buffer), 0, |
fba4b697 | 209 | SLAB_MEM_SPREAD, NULL); |
d1310b2e CM |
210 | if (!extent_buffer_cache) |
211 | goto free_state_cache; | |
9be3395b | 212 | |
8ac9f7c1 KO |
213 | if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE, |
214 | offsetof(struct btrfs_io_bio, bio), | |
215 | BIOSET_NEED_BVECS)) | |
9be3395b | 216 | goto free_buffer_cache; |
b208c2f7 | 217 | |
8ac9f7c1 | 218 | if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE)) |
b208c2f7 DW |
219 | goto free_bioset; |
220 | ||
d1310b2e CM |
221 | return 0; |
222 | ||
b208c2f7 | 223 | free_bioset: |
8ac9f7c1 | 224 | bioset_exit(&btrfs_bioset); |
b208c2f7 | 225 | |
9be3395b CM |
226 | free_buffer_cache: |
227 | kmem_cache_destroy(extent_buffer_cache); | |
228 | extent_buffer_cache = NULL; | |
229 | ||
d1310b2e CM |
230 | free_state_cache: |
231 | kmem_cache_destroy(extent_state_cache); | |
9be3395b | 232 | extent_state_cache = NULL; |
d1310b2e CM |
233 | return -ENOMEM; |
234 | } | |
235 | ||
e67c718b | 236 | void __cold extent_io_exit(void) |
d1310b2e | 237 | { |
6d49ba1b | 238 | btrfs_leak_debug_check(); |
8c0a8537 KS |
239 | |
240 | /* | |
241 | * Make sure all delayed rcu free are flushed before we | |
242 | * destroy caches. | |
243 | */ | |
244 | rcu_barrier(); | |
5598e900 KM |
245 | kmem_cache_destroy(extent_state_cache); |
246 | kmem_cache_destroy(extent_buffer_cache); | |
8ac9f7c1 | 247 | bioset_exit(&btrfs_bioset); |
d1310b2e CM |
248 | } |
249 | ||
c258d6e3 | 250 | void extent_io_tree_init(struct btrfs_fs_info *fs_info, |
43eb5f29 QW |
251 | struct extent_io_tree *tree, unsigned int owner, |
252 | void *private_data) | |
d1310b2e | 253 | { |
c258d6e3 | 254 | tree->fs_info = fs_info; |
6bef4d31 | 255 | tree->state = RB_ROOT; |
d1310b2e CM |
256 | tree->ops = NULL; |
257 | tree->dirty_bytes = 0; | |
70dec807 | 258 | spin_lock_init(&tree->lock); |
c6100a4b | 259 | tree->private_data = private_data; |
43eb5f29 | 260 | tree->owner = owner; |
d1310b2e | 261 | } |
d1310b2e | 262 | |
41e7acd3 NB |
263 | void extent_io_tree_release(struct extent_io_tree *tree) |
264 | { | |
265 | spin_lock(&tree->lock); | |
266 | /* | |
267 | * Do a single barrier for the waitqueue_active check here, the state | |
268 | * of the waitqueue should not change once extent_io_tree_release is | |
269 | * called. | |
270 | */ | |
271 | smp_mb(); | |
272 | while (!RB_EMPTY_ROOT(&tree->state)) { | |
273 | struct rb_node *node; | |
274 | struct extent_state *state; | |
275 | ||
276 | node = rb_first(&tree->state); | |
277 | state = rb_entry(node, struct extent_state, rb_node); | |
278 | rb_erase(&state->rb_node, &tree->state); | |
279 | RB_CLEAR_NODE(&state->rb_node); | |
280 | /* | |
281 | * btree io trees aren't supposed to have tasks waiting for | |
282 | * changes in the flags of extent states ever. | |
283 | */ | |
284 | ASSERT(!waitqueue_active(&state->wq)); | |
285 | free_extent_state(state); | |
286 | ||
287 | cond_resched_lock(&tree->lock); | |
288 | } | |
289 | spin_unlock(&tree->lock); | |
290 | } | |
291 | ||
b2950863 | 292 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
293 | { |
294 | struct extent_state *state; | |
d1310b2e | 295 | |
3ba7ab22 MH |
296 | /* |
297 | * The given mask might be not appropriate for the slab allocator, | |
298 | * drop the unsupported bits | |
299 | */ | |
300 | mask &= ~(__GFP_DMA32|__GFP_HIGHMEM); | |
d1310b2e | 301 | state = kmem_cache_alloc(extent_state_cache, mask); |
2b114d1d | 302 | if (!state) |
d1310b2e CM |
303 | return state; |
304 | state->state = 0; | |
47dc196a | 305 | state->failrec = NULL; |
27a3507d | 306 | RB_CLEAR_NODE(&state->rb_node); |
6d49ba1b | 307 | btrfs_leak_debug_add(&state->leak_list, &states); |
b7ac31b7 | 308 | refcount_set(&state->refs, 1); |
d1310b2e | 309 | init_waitqueue_head(&state->wq); |
143bede5 | 310 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
311 | return state; |
312 | } | |
d1310b2e | 313 | |
4845e44f | 314 | void free_extent_state(struct extent_state *state) |
d1310b2e | 315 | { |
d1310b2e CM |
316 | if (!state) |
317 | return; | |
b7ac31b7 | 318 | if (refcount_dec_and_test(&state->refs)) { |
27a3507d | 319 | WARN_ON(extent_state_in_tree(state)); |
6d49ba1b | 320 | btrfs_leak_debug_del(&state->leak_list); |
143bede5 | 321 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
322 | kmem_cache_free(extent_state_cache, state); |
323 | } | |
324 | } | |
d1310b2e | 325 | |
f2071b21 FM |
326 | static struct rb_node *tree_insert(struct rb_root *root, |
327 | struct rb_node *search_start, | |
328 | u64 offset, | |
12cfbad9 FDBM |
329 | struct rb_node *node, |
330 | struct rb_node ***p_in, | |
331 | struct rb_node **parent_in) | |
d1310b2e | 332 | { |
f2071b21 | 333 | struct rb_node **p; |
d397712b | 334 | struct rb_node *parent = NULL; |
d1310b2e CM |
335 | struct tree_entry *entry; |
336 | ||
12cfbad9 FDBM |
337 | if (p_in && parent_in) { |
338 | p = *p_in; | |
339 | parent = *parent_in; | |
340 | goto do_insert; | |
341 | } | |
342 | ||
f2071b21 | 343 | p = search_start ? &search_start : &root->rb_node; |
d397712b | 344 | while (*p) { |
d1310b2e CM |
345 | parent = *p; |
346 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
347 | ||
348 | if (offset < entry->start) | |
349 | p = &(*p)->rb_left; | |
350 | else if (offset > entry->end) | |
351 | p = &(*p)->rb_right; | |
352 | else | |
353 | return parent; | |
354 | } | |
355 | ||
12cfbad9 | 356 | do_insert: |
d1310b2e CM |
357 | rb_link_node(node, parent, p); |
358 | rb_insert_color(node, root); | |
359 | return NULL; | |
360 | } | |
361 | ||
80ea96b1 | 362 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
12cfbad9 | 363 | struct rb_node **next_ret, |
352646c7 | 364 | struct rb_node **prev_ret, |
12cfbad9 FDBM |
365 | struct rb_node ***p_ret, |
366 | struct rb_node **parent_ret) | |
d1310b2e | 367 | { |
80ea96b1 | 368 | struct rb_root *root = &tree->state; |
12cfbad9 | 369 | struct rb_node **n = &root->rb_node; |
d1310b2e CM |
370 | struct rb_node *prev = NULL; |
371 | struct rb_node *orig_prev = NULL; | |
372 | struct tree_entry *entry; | |
373 | struct tree_entry *prev_entry = NULL; | |
374 | ||
12cfbad9 FDBM |
375 | while (*n) { |
376 | prev = *n; | |
377 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
d1310b2e CM |
378 | prev_entry = entry; |
379 | ||
380 | if (offset < entry->start) | |
12cfbad9 | 381 | n = &(*n)->rb_left; |
d1310b2e | 382 | else if (offset > entry->end) |
12cfbad9 | 383 | n = &(*n)->rb_right; |
d397712b | 384 | else |
12cfbad9 | 385 | return *n; |
d1310b2e CM |
386 | } |
387 | ||
12cfbad9 FDBM |
388 | if (p_ret) |
389 | *p_ret = n; | |
390 | if (parent_ret) | |
391 | *parent_ret = prev; | |
392 | ||
352646c7 | 393 | if (next_ret) { |
d1310b2e | 394 | orig_prev = prev; |
d397712b | 395 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
396 | prev = rb_next(prev); |
397 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
398 | } | |
352646c7 | 399 | *next_ret = prev; |
d1310b2e CM |
400 | prev = orig_prev; |
401 | } | |
402 | ||
352646c7 | 403 | if (prev_ret) { |
d1310b2e | 404 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); |
d397712b | 405 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
406 | prev = rb_prev(prev); |
407 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
408 | } | |
352646c7 | 409 | *prev_ret = prev; |
d1310b2e CM |
410 | } |
411 | return NULL; | |
412 | } | |
413 | ||
12cfbad9 FDBM |
414 | static inline struct rb_node * |
415 | tree_search_for_insert(struct extent_io_tree *tree, | |
416 | u64 offset, | |
417 | struct rb_node ***p_ret, | |
418 | struct rb_node **parent_ret) | |
d1310b2e | 419 | { |
352646c7 | 420 | struct rb_node *next= NULL; |
d1310b2e | 421 | struct rb_node *ret; |
70dec807 | 422 | |
352646c7 | 423 | ret = __etree_search(tree, offset, &next, NULL, p_ret, parent_ret); |
d397712b | 424 | if (!ret) |
352646c7 | 425 | return next; |
d1310b2e CM |
426 | return ret; |
427 | } | |
428 | ||
12cfbad9 FDBM |
429 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
430 | u64 offset) | |
431 | { | |
432 | return tree_search_for_insert(tree, offset, NULL, NULL); | |
433 | } | |
434 | ||
d1310b2e CM |
435 | /* |
436 | * utility function to look for merge candidates inside a given range. | |
437 | * Any extents with matching state are merged together into a single | |
438 | * extent in the tree. Extents with EXTENT_IO in their state field | |
439 | * are not merged because the end_io handlers need to be able to do | |
440 | * operations on them without sleeping (or doing allocations/splits). | |
441 | * | |
442 | * This should be called with the tree lock held. | |
443 | */ | |
1bf85046 JM |
444 | static void merge_state(struct extent_io_tree *tree, |
445 | struct extent_state *state) | |
d1310b2e CM |
446 | { |
447 | struct extent_state *other; | |
448 | struct rb_node *other_node; | |
449 | ||
8882679e | 450 | if (state->state & (EXTENT_LOCKED | EXTENT_BOUNDARY)) |
1bf85046 | 451 | return; |
d1310b2e CM |
452 | |
453 | other_node = rb_prev(&state->rb_node); | |
454 | if (other_node) { | |
455 | other = rb_entry(other_node, struct extent_state, rb_node); | |
456 | if (other->end == state->start - 1 && | |
457 | other->state == state->state) { | |
5c848198 NB |
458 | if (tree->private_data && |
459 | is_data_inode(tree->private_data)) | |
460 | btrfs_merge_delalloc_extent(tree->private_data, | |
461 | state, other); | |
d1310b2e | 462 | state->start = other->start; |
d1310b2e | 463 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 464 | RB_CLEAR_NODE(&other->rb_node); |
d1310b2e CM |
465 | free_extent_state(other); |
466 | } | |
467 | } | |
468 | other_node = rb_next(&state->rb_node); | |
469 | if (other_node) { | |
470 | other = rb_entry(other_node, struct extent_state, rb_node); | |
471 | if (other->start == state->end + 1 && | |
472 | other->state == state->state) { | |
5c848198 NB |
473 | if (tree->private_data && |
474 | is_data_inode(tree->private_data)) | |
475 | btrfs_merge_delalloc_extent(tree->private_data, | |
476 | state, other); | |
df98b6e2 | 477 | state->end = other->end; |
df98b6e2 | 478 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 479 | RB_CLEAR_NODE(&other->rb_node); |
df98b6e2 | 480 | free_extent_state(other); |
d1310b2e CM |
481 | } |
482 | } | |
d1310b2e CM |
483 | } |
484 | ||
3150b699 | 485 | static void set_state_bits(struct extent_io_tree *tree, |
d38ed27f QW |
486 | struct extent_state *state, unsigned *bits, |
487 | struct extent_changeset *changeset); | |
3150b699 | 488 | |
d1310b2e CM |
489 | /* |
490 | * insert an extent_state struct into the tree. 'bits' are set on the | |
491 | * struct before it is inserted. | |
492 | * | |
493 | * This may return -EEXIST if the extent is already there, in which case the | |
494 | * state struct is freed. | |
495 | * | |
496 | * The tree lock is not taken internally. This is a utility function and | |
497 | * probably isn't what you want to call (see set/clear_extent_bit). | |
498 | */ | |
499 | static int insert_state(struct extent_io_tree *tree, | |
500 | struct extent_state *state, u64 start, u64 end, | |
12cfbad9 FDBM |
501 | struct rb_node ***p, |
502 | struct rb_node **parent, | |
d38ed27f | 503 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e CM |
504 | { |
505 | struct rb_node *node; | |
506 | ||
31b1a2bd | 507 | if (end < start) |
efe120a0 | 508 | WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n", |
c1c9ff7c | 509 | end, start); |
d1310b2e CM |
510 | state->start = start; |
511 | state->end = end; | |
9ed74f2d | 512 | |
d38ed27f | 513 | set_state_bits(tree, state, bits, changeset); |
3150b699 | 514 | |
f2071b21 | 515 | node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent); |
d1310b2e CM |
516 | if (node) { |
517 | struct extent_state *found; | |
518 | found = rb_entry(node, struct extent_state, rb_node); | |
62e85577 | 519 | pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n", |
c1c9ff7c | 520 | found->start, found->end, start, end); |
d1310b2e CM |
521 | return -EEXIST; |
522 | } | |
523 | merge_state(tree, state); | |
524 | return 0; | |
525 | } | |
526 | ||
527 | /* | |
528 | * split a given extent state struct in two, inserting the preallocated | |
529 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
530 | * offset inside 'orig' where it should be split. | |
531 | * | |
532 | * Before calling, | |
533 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
534 | * are two extent state structs in the tree: | |
535 | * prealloc: [orig->start, split - 1] | |
536 | * orig: [ split, orig->end ] | |
537 | * | |
538 | * The tree locks are not taken by this function. They need to be held | |
539 | * by the caller. | |
540 | */ | |
541 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
542 | struct extent_state *prealloc, u64 split) | |
543 | { | |
544 | struct rb_node *node; | |
9ed74f2d | 545 | |
abbb55f4 NB |
546 | if (tree->private_data && is_data_inode(tree->private_data)) |
547 | btrfs_split_delalloc_extent(tree->private_data, orig, split); | |
9ed74f2d | 548 | |
d1310b2e CM |
549 | prealloc->start = orig->start; |
550 | prealloc->end = split - 1; | |
551 | prealloc->state = orig->state; | |
552 | orig->start = split; | |
553 | ||
f2071b21 FM |
554 | node = tree_insert(&tree->state, &orig->rb_node, prealloc->end, |
555 | &prealloc->rb_node, NULL, NULL); | |
d1310b2e | 556 | if (node) { |
d1310b2e CM |
557 | free_extent_state(prealloc); |
558 | return -EEXIST; | |
559 | } | |
560 | return 0; | |
561 | } | |
562 | ||
cdc6a395 LZ |
563 | static struct extent_state *next_state(struct extent_state *state) |
564 | { | |
565 | struct rb_node *next = rb_next(&state->rb_node); | |
566 | if (next) | |
567 | return rb_entry(next, struct extent_state, rb_node); | |
568 | else | |
569 | return NULL; | |
570 | } | |
571 | ||
d1310b2e CM |
572 | /* |
573 | * utility function to clear some bits in an extent state struct. | |
52042d8e | 574 | * it will optionally wake up anyone waiting on this state (wake == 1). |
d1310b2e CM |
575 | * |
576 | * If no bits are set on the state struct after clearing things, the | |
577 | * struct is freed and removed from the tree | |
578 | */ | |
cdc6a395 LZ |
579 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
580 | struct extent_state *state, | |
fefdc557 QW |
581 | unsigned *bits, int wake, |
582 | struct extent_changeset *changeset) | |
d1310b2e | 583 | { |
cdc6a395 | 584 | struct extent_state *next; |
9ee49a04 | 585 | unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS; |
57599c7e | 586 | int ret; |
d1310b2e | 587 | |
0ca1f7ce | 588 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
589 | u64 range = state->end - state->start + 1; |
590 | WARN_ON(range > tree->dirty_bytes); | |
591 | tree->dirty_bytes -= range; | |
592 | } | |
a36bb5f9 NB |
593 | |
594 | if (tree->private_data && is_data_inode(tree->private_data)) | |
595 | btrfs_clear_delalloc_extent(tree->private_data, state, bits); | |
596 | ||
57599c7e DS |
597 | ret = add_extent_changeset(state, bits_to_clear, changeset, 0); |
598 | BUG_ON(ret < 0); | |
32c00aff | 599 | state->state &= ~bits_to_clear; |
d1310b2e CM |
600 | if (wake) |
601 | wake_up(&state->wq); | |
0ca1f7ce | 602 | if (state->state == 0) { |
cdc6a395 | 603 | next = next_state(state); |
27a3507d | 604 | if (extent_state_in_tree(state)) { |
d1310b2e | 605 | rb_erase(&state->rb_node, &tree->state); |
27a3507d | 606 | RB_CLEAR_NODE(&state->rb_node); |
d1310b2e CM |
607 | free_extent_state(state); |
608 | } else { | |
609 | WARN_ON(1); | |
610 | } | |
611 | } else { | |
612 | merge_state(tree, state); | |
cdc6a395 | 613 | next = next_state(state); |
d1310b2e | 614 | } |
cdc6a395 | 615 | return next; |
d1310b2e CM |
616 | } |
617 | ||
8233767a XG |
618 | static struct extent_state * |
619 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
620 | { | |
621 | if (!prealloc) | |
622 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
623 | ||
624 | return prealloc; | |
625 | } | |
626 | ||
48a3b636 | 627 | static void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
c2d904e0 | 628 | { |
05912a3c DS |
629 | struct inode *inode = tree->private_data; |
630 | ||
631 | btrfs_panic(btrfs_sb(inode->i_sb), err, | |
632 | "locking error: extent tree was modified by another thread while locked"); | |
c2d904e0 JM |
633 | } |
634 | ||
d1310b2e CM |
635 | /* |
636 | * clear some bits on a range in the tree. This may require splitting | |
637 | * or inserting elements in the tree, so the gfp mask is used to | |
638 | * indicate which allocations or sleeping are allowed. | |
639 | * | |
640 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
641 | * the given range from the tree regardless of state (ie for truncate). | |
642 | * | |
643 | * the range [start, end] is inclusive. | |
644 | * | |
6763af84 | 645 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e | 646 | */ |
66b0c887 | 647 | int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
fefdc557 QW |
648 | unsigned bits, int wake, int delete, |
649 | struct extent_state **cached_state, | |
650 | gfp_t mask, struct extent_changeset *changeset) | |
d1310b2e CM |
651 | { |
652 | struct extent_state *state; | |
2c64c53d | 653 | struct extent_state *cached; |
d1310b2e CM |
654 | struct extent_state *prealloc = NULL; |
655 | struct rb_node *node; | |
5c939df5 | 656 | u64 last_end; |
d1310b2e | 657 | int err; |
2ac55d41 | 658 | int clear = 0; |
d1310b2e | 659 | |
a5dee37d | 660 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 | 661 | trace_btrfs_clear_extent_bit(tree, start, end - start + 1, bits); |
8d599ae1 | 662 | |
7ee9e440 JB |
663 | if (bits & EXTENT_DELALLOC) |
664 | bits |= EXTENT_NORESERVE; | |
665 | ||
0ca1f7ce YZ |
666 | if (delete) |
667 | bits |= ~EXTENT_CTLBITS; | |
0ca1f7ce | 668 | |
8882679e | 669 | if (bits & (EXTENT_LOCKED | EXTENT_BOUNDARY)) |
2ac55d41 | 670 | clear = 1; |
d1310b2e | 671 | again: |
d0164adc | 672 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
c7bc6319 FM |
673 | /* |
674 | * Don't care for allocation failure here because we might end | |
675 | * up not needing the pre-allocated extent state at all, which | |
676 | * is the case if we only have in the tree extent states that | |
677 | * cover our input range and don't cover too any other range. | |
678 | * If we end up needing a new extent state we allocate it later. | |
679 | */ | |
d1310b2e | 680 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
681 | } |
682 | ||
cad321ad | 683 | spin_lock(&tree->lock); |
2c64c53d CM |
684 | if (cached_state) { |
685 | cached = *cached_state; | |
2ac55d41 JB |
686 | |
687 | if (clear) { | |
688 | *cached_state = NULL; | |
689 | cached_state = NULL; | |
690 | } | |
691 | ||
27a3507d FM |
692 | if (cached && extent_state_in_tree(cached) && |
693 | cached->start <= start && cached->end > start) { | |
2ac55d41 | 694 | if (clear) |
b7ac31b7 | 695 | refcount_dec(&cached->refs); |
2c64c53d | 696 | state = cached; |
42daec29 | 697 | goto hit_next; |
2c64c53d | 698 | } |
2ac55d41 JB |
699 | if (clear) |
700 | free_extent_state(cached); | |
2c64c53d | 701 | } |
d1310b2e CM |
702 | /* |
703 | * this search will find the extents that end after | |
704 | * our range starts | |
705 | */ | |
80ea96b1 | 706 | node = tree_search(tree, start); |
d1310b2e CM |
707 | if (!node) |
708 | goto out; | |
709 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 710 | hit_next: |
d1310b2e CM |
711 | if (state->start > end) |
712 | goto out; | |
713 | WARN_ON(state->end < start); | |
5c939df5 | 714 | last_end = state->end; |
d1310b2e | 715 | |
0449314a | 716 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
717 | if (!(state->state & bits)) { |
718 | state = next_state(state); | |
0449314a | 719 | goto next; |
cdc6a395 | 720 | } |
0449314a | 721 | |
d1310b2e CM |
722 | /* |
723 | * | ---- desired range ---- | | |
724 | * | state | or | |
725 | * | ------------- state -------------- | | |
726 | * | |
727 | * We need to split the extent we found, and may flip | |
728 | * bits on second half. | |
729 | * | |
730 | * If the extent we found extends past our range, we | |
731 | * just split and search again. It'll get split again | |
732 | * the next time though. | |
733 | * | |
734 | * If the extent we found is inside our range, we clear | |
735 | * the desired bit on it. | |
736 | */ | |
737 | ||
738 | if (state->start < start) { | |
8233767a XG |
739 | prealloc = alloc_extent_state_atomic(prealloc); |
740 | BUG_ON(!prealloc); | |
d1310b2e | 741 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
742 | if (err) |
743 | extent_io_tree_panic(tree, err); | |
744 | ||
d1310b2e CM |
745 | prealloc = NULL; |
746 | if (err) | |
747 | goto out; | |
748 | if (state->end <= end) { | |
fefdc557 QW |
749 | state = clear_state_bit(tree, state, &bits, wake, |
750 | changeset); | |
d1ac6e41 | 751 | goto next; |
d1310b2e CM |
752 | } |
753 | goto search_again; | |
754 | } | |
755 | /* | |
756 | * | ---- desired range ---- | | |
757 | * | state | | |
758 | * We need to split the extent, and clear the bit | |
759 | * on the first half | |
760 | */ | |
761 | if (state->start <= end && state->end > end) { | |
8233767a XG |
762 | prealloc = alloc_extent_state_atomic(prealloc); |
763 | BUG_ON(!prealloc); | |
d1310b2e | 764 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
765 | if (err) |
766 | extent_io_tree_panic(tree, err); | |
767 | ||
d1310b2e CM |
768 | if (wake) |
769 | wake_up(&state->wq); | |
42daec29 | 770 | |
fefdc557 | 771 | clear_state_bit(tree, prealloc, &bits, wake, changeset); |
9ed74f2d | 772 | |
d1310b2e CM |
773 | prealloc = NULL; |
774 | goto out; | |
775 | } | |
42daec29 | 776 | |
fefdc557 | 777 | state = clear_state_bit(tree, state, &bits, wake, changeset); |
0449314a | 778 | next: |
5c939df5 YZ |
779 | if (last_end == (u64)-1) |
780 | goto out; | |
781 | start = last_end + 1; | |
cdc6a395 | 782 | if (start <= end && state && !need_resched()) |
692e5759 | 783 | goto hit_next; |
d1310b2e CM |
784 | |
785 | search_again: | |
786 | if (start > end) | |
787 | goto out; | |
cad321ad | 788 | spin_unlock(&tree->lock); |
d0164adc | 789 | if (gfpflags_allow_blocking(mask)) |
d1310b2e CM |
790 | cond_resched(); |
791 | goto again; | |
7ab5cb2a DS |
792 | |
793 | out: | |
794 | spin_unlock(&tree->lock); | |
795 | if (prealloc) | |
796 | free_extent_state(prealloc); | |
797 | ||
798 | return 0; | |
799 | ||
d1310b2e | 800 | } |
d1310b2e | 801 | |
143bede5 JM |
802 | static void wait_on_state(struct extent_io_tree *tree, |
803 | struct extent_state *state) | |
641f5219 CH |
804 | __releases(tree->lock) |
805 | __acquires(tree->lock) | |
d1310b2e CM |
806 | { |
807 | DEFINE_WAIT(wait); | |
808 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 809 | spin_unlock(&tree->lock); |
d1310b2e | 810 | schedule(); |
cad321ad | 811 | spin_lock(&tree->lock); |
d1310b2e | 812 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
813 | } |
814 | ||
815 | /* | |
816 | * waits for one or more bits to clear on a range in the state tree. | |
817 | * The range [start, end] is inclusive. | |
818 | * The tree lock is taken by this function | |
819 | */ | |
41074888 DS |
820 | static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
821 | unsigned long bits) | |
d1310b2e CM |
822 | { |
823 | struct extent_state *state; | |
824 | struct rb_node *node; | |
825 | ||
a5dee37d | 826 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 827 | |
cad321ad | 828 | spin_lock(&tree->lock); |
d1310b2e CM |
829 | again: |
830 | while (1) { | |
831 | /* | |
832 | * this search will find all the extents that end after | |
833 | * our range starts | |
834 | */ | |
80ea96b1 | 835 | node = tree_search(tree, start); |
c50d3e71 | 836 | process_node: |
d1310b2e CM |
837 | if (!node) |
838 | break; | |
839 | ||
840 | state = rb_entry(node, struct extent_state, rb_node); | |
841 | ||
842 | if (state->start > end) | |
843 | goto out; | |
844 | ||
845 | if (state->state & bits) { | |
846 | start = state->start; | |
b7ac31b7 | 847 | refcount_inc(&state->refs); |
d1310b2e CM |
848 | wait_on_state(tree, state); |
849 | free_extent_state(state); | |
850 | goto again; | |
851 | } | |
852 | start = state->end + 1; | |
853 | ||
854 | if (start > end) | |
855 | break; | |
856 | ||
c50d3e71 FM |
857 | if (!cond_resched_lock(&tree->lock)) { |
858 | node = rb_next(node); | |
859 | goto process_node; | |
860 | } | |
d1310b2e CM |
861 | } |
862 | out: | |
cad321ad | 863 | spin_unlock(&tree->lock); |
d1310b2e | 864 | } |
d1310b2e | 865 | |
1bf85046 | 866 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 867 | struct extent_state *state, |
d38ed27f | 868 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e | 869 | { |
9ee49a04 | 870 | unsigned bits_to_set = *bits & ~EXTENT_CTLBITS; |
57599c7e | 871 | int ret; |
9ed74f2d | 872 | |
e06a1fc9 NB |
873 | if (tree->private_data && is_data_inode(tree->private_data)) |
874 | btrfs_set_delalloc_extent(tree->private_data, state, bits); | |
875 | ||
0ca1f7ce | 876 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
877 | u64 range = state->end - state->start + 1; |
878 | tree->dirty_bytes += range; | |
879 | } | |
57599c7e DS |
880 | ret = add_extent_changeset(state, bits_to_set, changeset, 1); |
881 | BUG_ON(ret < 0); | |
0ca1f7ce | 882 | state->state |= bits_to_set; |
d1310b2e CM |
883 | } |
884 | ||
e38e2ed7 FM |
885 | static void cache_state_if_flags(struct extent_state *state, |
886 | struct extent_state **cached_ptr, | |
9ee49a04 | 887 | unsigned flags) |
2c64c53d CM |
888 | { |
889 | if (cached_ptr && !(*cached_ptr)) { | |
e38e2ed7 | 890 | if (!flags || (state->state & flags)) { |
2c64c53d | 891 | *cached_ptr = state; |
b7ac31b7 | 892 | refcount_inc(&state->refs); |
2c64c53d CM |
893 | } |
894 | } | |
895 | } | |
896 | ||
e38e2ed7 FM |
897 | static void cache_state(struct extent_state *state, |
898 | struct extent_state **cached_ptr) | |
899 | { | |
900 | return cache_state_if_flags(state, cached_ptr, | |
8882679e | 901 | EXTENT_LOCKED | EXTENT_BOUNDARY); |
e38e2ed7 FM |
902 | } |
903 | ||
d1310b2e | 904 | /* |
1edbb734 CM |
905 | * set some bits on a range in the tree. This may require allocations or |
906 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 907 | * |
1edbb734 CM |
908 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
909 | * part of the range already has the desired bits set. The start of the | |
910 | * existing range is returned in failed_start in this case. | |
d1310b2e | 911 | * |
1edbb734 | 912 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 913 | */ |
1edbb734 | 914 | |
3fbe5c02 JM |
915 | static int __must_check |
916 | __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 917 | unsigned bits, unsigned exclusive_bits, |
41074888 | 918 | u64 *failed_start, struct extent_state **cached_state, |
d38ed27f | 919 | gfp_t mask, struct extent_changeset *changeset) |
d1310b2e CM |
920 | { |
921 | struct extent_state *state; | |
922 | struct extent_state *prealloc = NULL; | |
923 | struct rb_node *node; | |
12cfbad9 FDBM |
924 | struct rb_node **p; |
925 | struct rb_node *parent; | |
d1310b2e | 926 | int err = 0; |
d1310b2e CM |
927 | u64 last_start; |
928 | u64 last_end; | |
42daec29 | 929 | |
a5dee37d | 930 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 | 931 | trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits); |
8d599ae1 | 932 | |
d1310b2e | 933 | again: |
d0164adc | 934 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
059f791c DS |
935 | /* |
936 | * Don't care for allocation failure here because we might end | |
937 | * up not needing the pre-allocated extent state at all, which | |
938 | * is the case if we only have in the tree extent states that | |
939 | * cover our input range and don't cover too any other range. | |
940 | * If we end up needing a new extent state we allocate it later. | |
941 | */ | |
d1310b2e | 942 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
943 | } |
944 | ||
cad321ad | 945 | spin_lock(&tree->lock); |
9655d298 CM |
946 | if (cached_state && *cached_state) { |
947 | state = *cached_state; | |
df98b6e2 | 948 | if (state->start <= start && state->end > start && |
27a3507d | 949 | extent_state_in_tree(state)) { |
9655d298 CM |
950 | node = &state->rb_node; |
951 | goto hit_next; | |
952 | } | |
953 | } | |
d1310b2e CM |
954 | /* |
955 | * this search will find all the extents that end after | |
956 | * our range starts. | |
957 | */ | |
12cfbad9 | 958 | node = tree_search_for_insert(tree, start, &p, &parent); |
d1310b2e | 959 | if (!node) { |
8233767a XG |
960 | prealloc = alloc_extent_state_atomic(prealloc); |
961 | BUG_ON(!prealloc); | |
12cfbad9 | 962 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 963 | &p, &parent, &bits, changeset); |
c2d904e0 JM |
964 | if (err) |
965 | extent_io_tree_panic(tree, err); | |
966 | ||
c42ac0bc | 967 | cache_state(prealloc, cached_state); |
d1310b2e | 968 | prealloc = NULL; |
d1310b2e CM |
969 | goto out; |
970 | } | |
d1310b2e | 971 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 972 | hit_next: |
d1310b2e CM |
973 | last_start = state->start; |
974 | last_end = state->end; | |
975 | ||
976 | /* | |
977 | * | ---- desired range ---- | | |
978 | * | state | | |
979 | * | |
980 | * Just lock what we found and keep going | |
981 | */ | |
982 | if (state->start == start && state->end <= end) { | |
1edbb734 | 983 | if (state->state & exclusive_bits) { |
d1310b2e CM |
984 | *failed_start = state->start; |
985 | err = -EEXIST; | |
986 | goto out; | |
987 | } | |
42daec29 | 988 | |
d38ed27f | 989 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 990 | cache_state(state, cached_state); |
d1310b2e | 991 | merge_state(tree, state); |
5c939df5 YZ |
992 | if (last_end == (u64)-1) |
993 | goto out; | |
994 | start = last_end + 1; | |
d1ac6e41 LB |
995 | state = next_state(state); |
996 | if (start < end && state && state->start == start && | |
997 | !need_resched()) | |
998 | goto hit_next; | |
d1310b2e CM |
999 | goto search_again; |
1000 | } | |
1001 | ||
1002 | /* | |
1003 | * | ---- desired range ---- | | |
1004 | * | state | | |
1005 | * or | |
1006 | * | ------------- state -------------- | | |
1007 | * | |
1008 | * We need to split the extent we found, and may flip bits on | |
1009 | * second half. | |
1010 | * | |
1011 | * If the extent we found extends past our | |
1012 | * range, we just split and search again. It'll get split | |
1013 | * again the next time though. | |
1014 | * | |
1015 | * If the extent we found is inside our range, we set the | |
1016 | * desired bit on it. | |
1017 | */ | |
1018 | if (state->start < start) { | |
1edbb734 | 1019 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1020 | *failed_start = start; |
1021 | err = -EEXIST; | |
1022 | goto out; | |
1023 | } | |
8233767a XG |
1024 | |
1025 | prealloc = alloc_extent_state_atomic(prealloc); | |
1026 | BUG_ON(!prealloc); | |
d1310b2e | 1027 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1028 | if (err) |
1029 | extent_io_tree_panic(tree, err); | |
1030 | ||
d1310b2e CM |
1031 | prealloc = NULL; |
1032 | if (err) | |
1033 | goto out; | |
1034 | if (state->end <= end) { | |
d38ed27f | 1035 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 1036 | cache_state(state, cached_state); |
d1310b2e | 1037 | merge_state(tree, state); |
5c939df5 YZ |
1038 | if (last_end == (u64)-1) |
1039 | goto out; | |
1040 | start = last_end + 1; | |
d1ac6e41 LB |
1041 | state = next_state(state); |
1042 | if (start < end && state && state->start == start && | |
1043 | !need_resched()) | |
1044 | goto hit_next; | |
d1310b2e CM |
1045 | } |
1046 | goto search_again; | |
1047 | } | |
1048 | /* | |
1049 | * | ---- desired range ---- | | |
1050 | * | state | or | state | | |
1051 | * | |
1052 | * There's a hole, we need to insert something in it and | |
1053 | * ignore the extent we found. | |
1054 | */ | |
1055 | if (state->start > start) { | |
1056 | u64 this_end; | |
1057 | if (end < last_start) | |
1058 | this_end = end; | |
1059 | else | |
d397712b | 1060 | this_end = last_start - 1; |
8233767a XG |
1061 | |
1062 | prealloc = alloc_extent_state_atomic(prealloc); | |
1063 | BUG_ON(!prealloc); | |
c7f895a2 XG |
1064 | |
1065 | /* | |
1066 | * Avoid to free 'prealloc' if it can be merged with | |
1067 | * the later extent. | |
1068 | */ | |
d1310b2e | 1069 | err = insert_state(tree, prealloc, start, this_end, |
d38ed27f | 1070 | NULL, NULL, &bits, changeset); |
c2d904e0 JM |
1071 | if (err) |
1072 | extent_io_tree_panic(tree, err); | |
1073 | ||
9ed74f2d JB |
1074 | cache_state(prealloc, cached_state); |
1075 | prealloc = NULL; | |
d1310b2e CM |
1076 | start = this_end + 1; |
1077 | goto search_again; | |
1078 | } | |
1079 | /* | |
1080 | * | ---- desired range ---- | | |
1081 | * | state | | |
1082 | * We need to split the extent, and set the bit | |
1083 | * on the first half | |
1084 | */ | |
1085 | if (state->start <= end && state->end > end) { | |
1edbb734 | 1086 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1087 | *failed_start = start; |
1088 | err = -EEXIST; | |
1089 | goto out; | |
1090 | } | |
8233767a XG |
1091 | |
1092 | prealloc = alloc_extent_state_atomic(prealloc); | |
1093 | BUG_ON(!prealloc); | |
d1310b2e | 1094 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
1095 | if (err) |
1096 | extent_io_tree_panic(tree, err); | |
d1310b2e | 1097 | |
d38ed27f | 1098 | set_state_bits(tree, prealloc, &bits, changeset); |
2c64c53d | 1099 | cache_state(prealloc, cached_state); |
d1310b2e CM |
1100 | merge_state(tree, prealloc); |
1101 | prealloc = NULL; | |
1102 | goto out; | |
1103 | } | |
1104 | ||
b5a4ba14 DS |
1105 | search_again: |
1106 | if (start > end) | |
1107 | goto out; | |
1108 | spin_unlock(&tree->lock); | |
1109 | if (gfpflags_allow_blocking(mask)) | |
1110 | cond_resched(); | |
1111 | goto again; | |
d1310b2e CM |
1112 | |
1113 | out: | |
cad321ad | 1114 | spin_unlock(&tree->lock); |
d1310b2e CM |
1115 | if (prealloc) |
1116 | free_extent_state(prealloc); | |
1117 | ||
1118 | return err; | |
1119 | ||
d1310b2e | 1120 | } |
d1310b2e | 1121 | |
41074888 | 1122 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
9ee49a04 | 1123 | unsigned bits, u64 * failed_start, |
41074888 | 1124 | struct extent_state **cached_state, gfp_t mask) |
3fbe5c02 JM |
1125 | { |
1126 | return __set_extent_bit(tree, start, end, bits, 0, failed_start, | |
d38ed27f | 1127 | cached_state, mask, NULL); |
3fbe5c02 JM |
1128 | } |
1129 | ||
1130 | ||
462d6fac | 1131 | /** |
10983f2e LB |
1132 | * convert_extent_bit - convert all bits in a given range from one bit to |
1133 | * another | |
462d6fac JB |
1134 | * @tree: the io tree to search |
1135 | * @start: the start offset in bytes | |
1136 | * @end: the end offset in bytes (inclusive) | |
1137 | * @bits: the bits to set in this range | |
1138 | * @clear_bits: the bits to clear in this range | |
e6138876 | 1139 | * @cached_state: state that we're going to cache |
462d6fac JB |
1140 | * |
1141 | * This will go through and set bits for the given range. If any states exist | |
1142 | * already in this range they are set with the given bit and cleared of the | |
1143 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
1144 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
1145 | * boundary bits like LOCK. | |
210aa277 DS |
1146 | * |
1147 | * All allocations are done with GFP_NOFS. | |
462d6fac JB |
1148 | */ |
1149 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1150 | unsigned bits, unsigned clear_bits, |
210aa277 | 1151 | struct extent_state **cached_state) |
462d6fac JB |
1152 | { |
1153 | struct extent_state *state; | |
1154 | struct extent_state *prealloc = NULL; | |
1155 | struct rb_node *node; | |
12cfbad9 FDBM |
1156 | struct rb_node **p; |
1157 | struct rb_node *parent; | |
462d6fac JB |
1158 | int err = 0; |
1159 | u64 last_start; | |
1160 | u64 last_end; | |
c8fd3de7 | 1161 | bool first_iteration = true; |
462d6fac | 1162 | |
a5dee37d | 1163 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 QW |
1164 | trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits, |
1165 | clear_bits); | |
8d599ae1 | 1166 | |
462d6fac | 1167 | again: |
210aa277 | 1168 | if (!prealloc) { |
c8fd3de7 FM |
1169 | /* |
1170 | * Best effort, don't worry if extent state allocation fails | |
1171 | * here for the first iteration. We might have a cached state | |
1172 | * that matches exactly the target range, in which case no | |
1173 | * extent state allocations are needed. We'll only know this | |
1174 | * after locking the tree. | |
1175 | */ | |
210aa277 | 1176 | prealloc = alloc_extent_state(GFP_NOFS); |
c8fd3de7 | 1177 | if (!prealloc && !first_iteration) |
462d6fac JB |
1178 | return -ENOMEM; |
1179 | } | |
1180 | ||
1181 | spin_lock(&tree->lock); | |
e6138876 JB |
1182 | if (cached_state && *cached_state) { |
1183 | state = *cached_state; | |
1184 | if (state->start <= start && state->end > start && | |
27a3507d | 1185 | extent_state_in_tree(state)) { |
e6138876 JB |
1186 | node = &state->rb_node; |
1187 | goto hit_next; | |
1188 | } | |
1189 | } | |
1190 | ||
462d6fac JB |
1191 | /* |
1192 | * this search will find all the extents that end after | |
1193 | * our range starts. | |
1194 | */ | |
12cfbad9 | 1195 | node = tree_search_for_insert(tree, start, &p, &parent); |
462d6fac JB |
1196 | if (!node) { |
1197 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1198 | if (!prealloc) { |
1199 | err = -ENOMEM; | |
1200 | goto out; | |
1201 | } | |
12cfbad9 | 1202 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 1203 | &p, &parent, &bits, NULL); |
c2d904e0 JM |
1204 | if (err) |
1205 | extent_io_tree_panic(tree, err); | |
c42ac0bc FDBM |
1206 | cache_state(prealloc, cached_state); |
1207 | prealloc = NULL; | |
462d6fac JB |
1208 | goto out; |
1209 | } | |
1210 | state = rb_entry(node, struct extent_state, rb_node); | |
1211 | hit_next: | |
1212 | last_start = state->start; | |
1213 | last_end = state->end; | |
1214 | ||
1215 | /* | |
1216 | * | ---- desired range ---- | | |
1217 | * | state | | |
1218 | * | |
1219 | * Just lock what we found and keep going | |
1220 | */ | |
1221 | if (state->start == start && state->end <= end) { | |
d38ed27f | 1222 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1223 | cache_state(state, cached_state); |
fefdc557 | 1224 | state = clear_state_bit(tree, state, &clear_bits, 0, NULL); |
462d6fac JB |
1225 | if (last_end == (u64)-1) |
1226 | goto out; | |
462d6fac | 1227 | start = last_end + 1; |
d1ac6e41 LB |
1228 | if (start < end && state && state->start == start && |
1229 | !need_resched()) | |
1230 | goto hit_next; | |
462d6fac JB |
1231 | goto search_again; |
1232 | } | |
1233 | ||
1234 | /* | |
1235 | * | ---- desired range ---- | | |
1236 | * | state | | |
1237 | * or | |
1238 | * | ------------- state -------------- | | |
1239 | * | |
1240 | * We need to split the extent we found, and may flip bits on | |
1241 | * second half. | |
1242 | * | |
1243 | * If the extent we found extends past our | |
1244 | * range, we just split and search again. It'll get split | |
1245 | * again the next time though. | |
1246 | * | |
1247 | * If the extent we found is inside our range, we set the | |
1248 | * desired bit on it. | |
1249 | */ | |
1250 | if (state->start < start) { | |
1251 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1252 | if (!prealloc) { |
1253 | err = -ENOMEM; | |
1254 | goto out; | |
1255 | } | |
462d6fac | 1256 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1257 | if (err) |
1258 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1259 | prealloc = NULL; |
1260 | if (err) | |
1261 | goto out; | |
1262 | if (state->end <= end) { | |
d38ed27f | 1263 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1264 | cache_state(state, cached_state); |
fefdc557 QW |
1265 | state = clear_state_bit(tree, state, &clear_bits, 0, |
1266 | NULL); | |
462d6fac JB |
1267 | if (last_end == (u64)-1) |
1268 | goto out; | |
1269 | start = last_end + 1; | |
d1ac6e41 LB |
1270 | if (start < end && state && state->start == start && |
1271 | !need_resched()) | |
1272 | goto hit_next; | |
462d6fac JB |
1273 | } |
1274 | goto search_again; | |
1275 | } | |
1276 | /* | |
1277 | * | ---- desired range ---- | | |
1278 | * | state | or | state | | |
1279 | * | |
1280 | * There's a hole, we need to insert something in it and | |
1281 | * ignore the extent we found. | |
1282 | */ | |
1283 | if (state->start > start) { | |
1284 | u64 this_end; | |
1285 | if (end < last_start) | |
1286 | this_end = end; | |
1287 | else | |
1288 | this_end = last_start - 1; | |
1289 | ||
1290 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1291 | if (!prealloc) { |
1292 | err = -ENOMEM; | |
1293 | goto out; | |
1294 | } | |
462d6fac JB |
1295 | |
1296 | /* | |
1297 | * Avoid to free 'prealloc' if it can be merged with | |
1298 | * the later extent. | |
1299 | */ | |
1300 | err = insert_state(tree, prealloc, start, this_end, | |
d38ed27f | 1301 | NULL, NULL, &bits, NULL); |
c2d904e0 JM |
1302 | if (err) |
1303 | extent_io_tree_panic(tree, err); | |
e6138876 | 1304 | cache_state(prealloc, cached_state); |
462d6fac JB |
1305 | prealloc = NULL; |
1306 | start = this_end + 1; | |
1307 | goto search_again; | |
1308 | } | |
1309 | /* | |
1310 | * | ---- desired range ---- | | |
1311 | * | state | | |
1312 | * We need to split the extent, and set the bit | |
1313 | * on the first half | |
1314 | */ | |
1315 | if (state->start <= end && state->end > end) { | |
1316 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1317 | if (!prealloc) { |
1318 | err = -ENOMEM; | |
1319 | goto out; | |
1320 | } | |
462d6fac JB |
1321 | |
1322 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1323 | if (err) |
1324 | extent_io_tree_panic(tree, err); | |
462d6fac | 1325 | |
d38ed27f | 1326 | set_state_bits(tree, prealloc, &bits, NULL); |
e6138876 | 1327 | cache_state(prealloc, cached_state); |
fefdc557 | 1328 | clear_state_bit(tree, prealloc, &clear_bits, 0, NULL); |
462d6fac JB |
1329 | prealloc = NULL; |
1330 | goto out; | |
1331 | } | |
1332 | ||
462d6fac JB |
1333 | search_again: |
1334 | if (start > end) | |
1335 | goto out; | |
1336 | spin_unlock(&tree->lock); | |
210aa277 | 1337 | cond_resched(); |
c8fd3de7 | 1338 | first_iteration = false; |
462d6fac | 1339 | goto again; |
462d6fac JB |
1340 | |
1341 | out: | |
1342 | spin_unlock(&tree->lock); | |
1343 | if (prealloc) | |
1344 | free_extent_state(prealloc); | |
1345 | ||
1346 | return err; | |
462d6fac JB |
1347 | } |
1348 | ||
d1310b2e | 1349 | /* wrappers around set/clear extent bit */ |
d38ed27f | 1350 | int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
2c53b912 | 1351 | unsigned bits, struct extent_changeset *changeset) |
d38ed27f QW |
1352 | { |
1353 | /* | |
1354 | * We don't support EXTENT_LOCKED yet, as current changeset will | |
1355 | * record any bits changed, so for EXTENT_LOCKED case, it will | |
1356 | * either fail with -EEXIST or changeset will record the whole | |
1357 | * range. | |
1358 | */ | |
1359 | BUG_ON(bits & EXTENT_LOCKED); | |
1360 | ||
2c53b912 | 1361 | return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS, |
d38ed27f QW |
1362 | changeset); |
1363 | } | |
1364 | ||
4ca73656 NB |
1365 | int set_extent_bits_nowait(struct extent_io_tree *tree, u64 start, u64 end, |
1366 | unsigned bits) | |
1367 | { | |
1368 | return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, | |
1369 | GFP_NOWAIT, NULL); | |
1370 | } | |
1371 | ||
fefdc557 QW |
1372 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
1373 | unsigned bits, int wake, int delete, | |
ae0f1625 | 1374 | struct extent_state **cached) |
fefdc557 QW |
1375 | { |
1376 | return __clear_extent_bit(tree, start, end, bits, wake, delete, | |
ae0f1625 | 1377 | cached, GFP_NOFS, NULL); |
fefdc557 QW |
1378 | } |
1379 | ||
fefdc557 | 1380 | int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
f734c44a | 1381 | unsigned bits, struct extent_changeset *changeset) |
fefdc557 QW |
1382 | { |
1383 | /* | |
1384 | * Don't support EXTENT_LOCKED case, same reason as | |
1385 | * set_record_extent_bits(). | |
1386 | */ | |
1387 | BUG_ON(bits & EXTENT_LOCKED); | |
1388 | ||
f734c44a | 1389 | return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS, |
fefdc557 QW |
1390 | changeset); |
1391 | } | |
1392 | ||
d352ac68 CM |
1393 | /* |
1394 | * either insert or lock state struct between start and end use mask to tell | |
1395 | * us if waiting is desired. | |
1396 | */ | |
1edbb734 | 1397 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
ff13db41 | 1398 | struct extent_state **cached_state) |
d1310b2e CM |
1399 | { |
1400 | int err; | |
1401 | u64 failed_start; | |
9ee49a04 | 1402 | |
d1310b2e | 1403 | while (1) { |
ff13db41 | 1404 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, |
3fbe5c02 | 1405 | EXTENT_LOCKED, &failed_start, |
d38ed27f | 1406 | cached_state, GFP_NOFS, NULL); |
d0082371 | 1407 | if (err == -EEXIST) { |
d1310b2e CM |
1408 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1409 | start = failed_start; | |
d0082371 | 1410 | } else |
d1310b2e | 1411 | break; |
d1310b2e CM |
1412 | WARN_ON(start > end); |
1413 | } | |
1414 | return err; | |
1415 | } | |
d1310b2e | 1416 | |
d0082371 | 1417 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1418 | { |
1419 | int err; | |
1420 | u64 failed_start; | |
1421 | ||
3fbe5c02 | 1422 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
d38ed27f | 1423 | &failed_start, NULL, GFP_NOFS, NULL); |
6643558d YZ |
1424 | if (err == -EEXIST) { |
1425 | if (failed_start > start) | |
1426 | clear_extent_bit(tree, start, failed_start - 1, | |
ae0f1625 | 1427 | EXTENT_LOCKED, 1, 0, NULL); |
25179201 | 1428 | return 0; |
6643558d | 1429 | } |
25179201 JB |
1430 | return 1; |
1431 | } | |
25179201 | 1432 | |
bd1fa4f0 | 1433 | void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1434 | { |
09cbfeaf KS |
1435 | unsigned long index = start >> PAGE_SHIFT; |
1436 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1437 | struct page *page; |
1438 | ||
1439 | while (index <= end_index) { | |
1440 | page = find_get_page(inode->i_mapping, index); | |
1441 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1442 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1443 | put_page(page); |
4adaa611 CM |
1444 | index++; |
1445 | } | |
4adaa611 CM |
1446 | } |
1447 | ||
f6311572 | 1448 | void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1449 | { |
09cbfeaf KS |
1450 | unsigned long index = start >> PAGE_SHIFT; |
1451 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1452 | struct page *page; |
1453 | ||
1454 | while (index <= end_index) { | |
1455 | page = find_get_page(inode->i_mapping, index); | |
1456 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
4adaa611 | 1457 | __set_page_dirty_nobuffers(page); |
8d38633c | 1458 | account_page_redirty(page); |
09cbfeaf | 1459 | put_page(page); |
4adaa611 CM |
1460 | index++; |
1461 | } | |
4adaa611 CM |
1462 | } |
1463 | ||
d352ac68 CM |
1464 | /* find the first state struct with 'bits' set after 'start', and |
1465 | * return it. tree->lock must be held. NULL will returned if | |
1466 | * nothing was found after 'start' | |
1467 | */ | |
48a3b636 ES |
1468 | static struct extent_state * |
1469 | find_first_extent_bit_state(struct extent_io_tree *tree, | |
9ee49a04 | 1470 | u64 start, unsigned bits) |
d7fc640e CM |
1471 | { |
1472 | struct rb_node *node; | |
1473 | struct extent_state *state; | |
1474 | ||
1475 | /* | |
1476 | * this search will find all the extents that end after | |
1477 | * our range starts. | |
1478 | */ | |
1479 | node = tree_search(tree, start); | |
d397712b | 1480 | if (!node) |
d7fc640e | 1481 | goto out; |
d7fc640e | 1482 | |
d397712b | 1483 | while (1) { |
d7fc640e | 1484 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1485 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1486 | return state; |
d397712b | 1487 | |
d7fc640e CM |
1488 | node = rb_next(node); |
1489 | if (!node) | |
1490 | break; | |
1491 | } | |
1492 | out: | |
1493 | return NULL; | |
1494 | } | |
d7fc640e | 1495 | |
69261c4b XG |
1496 | /* |
1497 | * find the first offset in the io tree with 'bits' set. zero is | |
1498 | * returned if we find something, and *start_ret and *end_ret are | |
1499 | * set to reflect the state struct that was found. | |
1500 | * | |
477d7eaf | 1501 | * If nothing was found, 1 is returned. If found something, return 0. |
69261c4b XG |
1502 | */ |
1503 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
9ee49a04 | 1504 | u64 *start_ret, u64 *end_ret, unsigned bits, |
e6138876 | 1505 | struct extent_state **cached_state) |
69261c4b XG |
1506 | { |
1507 | struct extent_state *state; | |
1508 | int ret = 1; | |
1509 | ||
1510 | spin_lock(&tree->lock); | |
e6138876 JB |
1511 | if (cached_state && *cached_state) { |
1512 | state = *cached_state; | |
27a3507d | 1513 | if (state->end == start - 1 && extent_state_in_tree(state)) { |
9688e9a9 | 1514 | while ((state = next_state(state)) != NULL) { |
e6138876 JB |
1515 | if (state->state & bits) |
1516 | goto got_it; | |
e6138876 JB |
1517 | } |
1518 | free_extent_state(*cached_state); | |
1519 | *cached_state = NULL; | |
1520 | goto out; | |
1521 | } | |
1522 | free_extent_state(*cached_state); | |
1523 | *cached_state = NULL; | |
1524 | } | |
1525 | ||
69261c4b | 1526 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1527 | got_it: |
69261c4b | 1528 | if (state) { |
e38e2ed7 | 1529 | cache_state_if_flags(state, cached_state, 0); |
69261c4b XG |
1530 | *start_ret = state->start; |
1531 | *end_ret = state->end; | |
1532 | ret = 0; | |
1533 | } | |
e6138876 | 1534 | out: |
69261c4b XG |
1535 | spin_unlock(&tree->lock); |
1536 | return ret; | |
1537 | } | |
1538 | ||
45bfcfc1 NB |
1539 | /** |
1540 | * find_first_clear_extent_bit - finds the first range that has @bits not set | |
1541 | * and that starts after @start | |
1542 | * | |
1543 | * @tree - the tree to search | |
1544 | * @start - the offset at/after which the found extent should start | |
1545 | * @start_ret - records the beginning of the range | |
1546 | * @end_ret - records the end of the range (inclusive) | |
1547 | * @bits - the set of bits which must be unset | |
1548 | * | |
1549 | * Since unallocated range is also considered one which doesn't have the bits | |
1550 | * set it's possible that @end_ret contains -1, this happens in case the range | |
1551 | * spans (last_range_end, end of device]. In this case it's up to the caller to | |
1552 | * trim @end_ret to the appropriate size. | |
1553 | */ | |
1554 | void find_first_clear_extent_bit(struct extent_io_tree *tree, u64 start, | |
1555 | u64 *start_ret, u64 *end_ret, unsigned bits) | |
1556 | { | |
1557 | struct extent_state *state; | |
1558 | struct rb_node *node, *prev = NULL, *next; | |
1559 | ||
1560 | spin_lock(&tree->lock); | |
1561 | ||
1562 | /* Find first extent with bits cleared */ | |
1563 | while (1) { | |
1564 | node = __etree_search(tree, start, &next, &prev, NULL, NULL); | |
1565 | if (!node) { | |
1566 | node = next; | |
1567 | if (!node) { | |
1568 | /* | |
1569 | * We are past the last allocated chunk, | |
1570 | * set start at the end of the last extent. The | |
1571 | * device alloc tree should never be empty so | |
1572 | * prev is always set. | |
1573 | */ | |
1574 | ASSERT(prev); | |
1575 | state = rb_entry(prev, struct extent_state, rb_node); | |
1576 | *start_ret = state->end + 1; | |
1577 | *end_ret = -1; | |
1578 | goto out; | |
1579 | } | |
1580 | } | |
1581 | state = rb_entry(node, struct extent_state, rb_node); | |
1582 | if (in_range(start, state->start, state->end - state->start + 1) && | |
1583 | (state->state & bits)) { | |
1584 | start = state->end + 1; | |
1585 | } else { | |
1586 | *start_ret = start; | |
1587 | break; | |
1588 | } | |
1589 | } | |
1590 | ||
1591 | /* | |
1592 | * Find the longest stretch from start until an entry which has the | |
1593 | * bits set | |
1594 | */ | |
1595 | while (1) { | |
1596 | state = rb_entry(node, struct extent_state, rb_node); | |
1597 | if (state->end >= start && !(state->state & bits)) { | |
1598 | *end_ret = state->end; | |
1599 | } else { | |
1600 | *end_ret = state->start - 1; | |
1601 | break; | |
1602 | } | |
1603 | ||
1604 | node = rb_next(node); | |
1605 | if (!node) | |
1606 | break; | |
1607 | } | |
1608 | out: | |
1609 | spin_unlock(&tree->lock); | |
1610 | } | |
1611 | ||
d352ac68 CM |
1612 | /* |
1613 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1614 | * more than 'max_bytes'. start and end are used to return the range, | |
1615 | * | |
3522e903 | 1616 | * true is returned if we find something, false if nothing was in the tree |
d352ac68 | 1617 | */ |
3522e903 | 1618 | static noinline bool find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1619 | u64 *start, u64 *end, u64 max_bytes, |
1620 | struct extent_state **cached_state) | |
d1310b2e CM |
1621 | { |
1622 | struct rb_node *node; | |
1623 | struct extent_state *state; | |
1624 | u64 cur_start = *start; | |
3522e903 | 1625 | bool found = false; |
d1310b2e CM |
1626 | u64 total_bytes = 0; |
1627 | ||
cad321ad | 1628 | spin_lock(&tree->lock); |
c8b97818 | 1629 | |
d1310b2e CM |
1630 | /* |
1631 | * this search will find all the extents that end after | |
1632 | * our range starts. | |
1633 | */ | |
80ea96b1 | 1634 | node = tree_search(tree, cur_start); |
2b114d1d | 1635 | if (!node) { |
3522e903 | 1636 | *end = (u64)-1; |
d1310b2e CM |
1637 | goto out; |
1638 | } | |
1639 | ||
d397712b | 1640 | while (1) { |
d1310b2e | 1641 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1642 | if (found && (state->start != cur_start || |
1643 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1644 | goto out; |
1645 | } | |
1646 | if (!(state->state & EXTENT_DELALLOC)) { | |
1647 | if (!found) | |
1648 | *end = state->end; | |
1649 | goto out; | |
1650 | } | |
c2a128d2 | 1651 | if (!found) { |
d1310b2e | 1652 | *start = state->start; |
c2a128d2 | 1653 | *cached_state = state; |
b7ac31b7 | 1654 | refcount_inc(&state->refs); |
c2a128d2 | 1655 | } |
3522e903 | 1656 | found = true; |
d1310b2e CM |
1657 | *end = state->end; |
1658 | cur_start = state->end + 1; | |
1659 | node = rb_next(node); | |
d1310b2e | 1660 | total_bytes += state->end - state->start + 1; |
7bf811a5 | 1661 | if (total_bytes >= max_bytes) |
573aecaf | 1662 | break; |
573aecaf | 1663 | if (!node) |
d1310b2e CM |
1664 | break; |
1665 | } | |
1666 | out: | |
cad321ad | 1667 | spin_unlock(&tree->lock); |
d1310b2e CM |
1668 | return found; |
1669 | } | |
1670 | ||
da2c7009 LB |
1671 | static int __process_pages_contig(struct address_space *mapping, |
1672 | struct page *locked_page, | |
1673 | pgoff_t start_index, pgoff_t end_index, | |
1674 | unsigned long page_ops, pgoff_t *index_ret); | |
1675 | ||
143bede5 JM |
1676 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1677 | struct page *locked_page, | |
1678 | u64 start, u64 end) | |
c8b97818 | 1679 | { |
09cbfeaf KS |
1680 | unsigned long index = start >> PAGE_SHIFT; |
1681 | unsigned long end_index = end >> PAGE_SHIFT; | |
c8b97818 | 1682 | |
76c0021d | 1683 | ASSERT(locked_page); |
c8b97818 | 1684 | if (index == locked_page->index && end_index == index) |
143bede5 | 1685 | return; |
c8b97818 | 1686 | |
76c0021d LB |
1687 | __process_pages_contig(inode->i_mapping, locked_page, index, end_index, |
1688 | PAGE_UNLOCK, NULL); | |
c8b97818 CM |
1689 | } |
1690 | ||
1691 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1692 | struct page *locked_page, | |
1693 | u64 delalloc_start, | |
1694 | u64 delalloc_end) | |
1695 | { | |
09cbfeaf | 1696 | unsigned long index = delalloc_start >> PAGE_SHIFT; |
76c0021d | 1697 | unsigned long index_ret = index; |
09cbfeaf | 1698 | unsigned long end_index = delalloc_end >> PAGE_SHIFT; |
c8b97818 | 1699 | int ret; |
c8b97818 | 1700 | |
76c0021d | 1701 | ASSERT(locked_page); |
c8b97818 CM |
1702 | if (index == locked_page->index && index == end_index) |
1703 | return 0; | |
1704 | ||
76c0021d LB |
1705 | ret = __process_pages_contig(inode->i_mapping, locked_page, index, |
1706 | end_index, PAGE_LOCK, &index_ret); | |
1707 | if (ret == -EAGAIN) | |
1708 | __unlock_for_delalloc(inode, locked_page, delalloc_start, | |
1709 | (u64)index_ret << PAGE_SHIFT); | |
c8b97818 CM |
1710 | return ret; |
1711 | } | |
1712 | ||
1713 | /* | |
3522e903 LF |
1714 | * Find and lock a contiguous range of bytes in the file marked as delalloc, no |
1715 | * more than @max_bytes. @Start and @end are used to return the range, | |
c8b97818 | 1716 | * |
3522e903 LF |
1717 | * Return: true if we find something |
1718 | * false if nothing was in the tree | |
c8b97818 | 1719 | */ |
ce9f967f | 1720 | EXPORT_FOR_TESTS |
3522e903 | 1721 | noinline_for_stack bool find_lock_delalloc_range(struct inode *inode, |
294e30fe JB |
1722 | struct extent_io_tree *tree, |
1723 | struct page *locked_page, u64 *start, | |
917aacec | 1724 | u64 *end) |
c8b97818 | 1725 | { |
917aacec | 1726 | u64 max_bytes = BTRFS_MAX_EXTENT_SIZE; |
c8b97818 CM |
1727 | u64 delalloc_start; |
1728 | u64 delalloc_end; | |
3522e903 | 1729 | bool found; |
9655d298 | 1730 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1731 | int ret; |
1732 | int loops = 0; | |
1733 | ||
1734 | again: | |
1735 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1736 | delalloc_start = *start; | |
1737 | delalloc_end = 0; | |
1738 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1739 | max_bytes, &cached_state); |
70b99e69 | 1740 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1741 | *start = delalloc_start; |
1742 | *end = delalloc_end; | |
c2a128d2 | 1743 | free_extent_state(cached_state); |
3522e903 | 1744 | return false; |
c8b97818 CM |
1745 | } |
1746 | ||
70b99e69 CM |
1747 | /* |
1748 | * start comes from the offset of locked_page. We have to lock | |
1749 | * pages in order, so we can't process delalloc bytes before | |
1750 | * locked_page | |
1751 | */ | |
d397712b | 1752 | if (delalloc_start < *start) |
70b99e69 | 1753 | delalloc_start = *start; |
70b99e69 | 1754 | |
c8b97818 CM |
1755 | /* |
1756 | * make sure to limit the number of pages we try to lock down | |
c8b97818 | 1757 | */ |
7bf811a5 JB |
1758 | if (delalloc_end + 1 - delalloc_start > max_bytes) |
1759 | delalloc_end = delalloc_start + max_bytes - 1; | |
d397712b | 1760 | |
c8b97818 CM |
1761 | /* step two, lock all the pages after the page that has start */ |
1762 | ret = lock_delalloc_pages(inode, locked_page, | |
1763 | delalloc_start, delalloc_end); | |
9bfd61d9 | 1764 | ASSERT(!ret || ret == -EAGAIN); |
c8b97818 CM |
1765 | if (ret == -EAGAIN) { |
1766 | /* some of the pages are gone, lets avoid looping by | |
1767 | * shortening the size of the delalloc range we're searching | |
1768 | */ | |
9655d298 | 1769 | free_extent_state(cached_state); |
7d788742 | 1770 | cached_state = NULL; |
c8b97818 | 1771 | if (!loops) { |
09cbfeaf | 1772 | max_bytes = PAGE_SIZE; |
c8b97818 CM |
1773 | loops = 1; |
1774 | goto again; | |
1775 | } else { | |
3522e903 | 1776 | found = false; |
c8b97818 CM |
1777 | goto out_failed; |
1778 | } | |
1779 | } | |
c8b97818 CM |
1780 | |
1781 | /* step three, lock the state bits for the whole range */ | |
ff13db41 | 1782 | lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state); |
c8b97818 CM |
1783 | |
1784 | /* then test to make sure it is all still delalloc */ | |
1785 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1786 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1787 | if (!ret) { |
9655d298 | 1788 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
e43bbe5e | 1789 | &cached_state); |
c8b97818 CM |
1790 | __unlock_for_delalloc(inode, locked_page, |
1791 | delalloc_start, delalloc_end); | |
1792 | cond_resched(); | |
1793 | goto again; | |
1794 | } | |
9655d298 | 1795 | free_extent_state(cached_state); |
c8b97818 CM |
1796 | *start = delalloc_start; |
1797 | *end = delalloc_end; | |
1798 | out_failed: | |
1799 | return found; | |
1800 | } | |
1801 | ||
da2c7009 LB |
1802 | static int __process_pages_contig(struct address_space *mapping, |
1803 | struct page *locked_page, | |
1804 | pgoff_t start_index, pgoff_t end_index, | |
1805 | unsigned long page_ops, pgoff_t *index_ret) | |
c8b97818 | 1806 | { |
873695b3 | 1807 | unsigned long nr_pages = end_index - start_index + 1; |
da2c7009 | 1808 | unsigned long pages_locked = 0; |
873695b3 | 1809 | pgoff_t index = start_index; |
c8b97818 | 1810 | struct page *pages[16]; |
873695b3 | 1811 | unsigned ret; |
da2c7009 | 1812 | int err = 0; |
c8b97818 | 1813 | int i; |
771ed689 | 1814 | |
da2c7009 LB |
1815 | if (page_ops & PAGE_LOCK) { |
1816 | ASSERT(page_ops == PAGE_LOCK); | |
1817 | ASSERT(index_ret && *index_ret == start_index); | |
1818 | } | |
1819 | ||
704de49d | 1820 | if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0) |
873695b3 | 1821 | mapping_set_error(mapping, -EIO); |
704de49d | 1822 | |
d397712b | 1823 | while (nr_pages > 0) { |
873695b3 | 1824 | ret = find_get_pages_contig(mapping, index, |
5b050f04 CM |
1825 | min_t(unsigned long, |
1826 | nr_pages, ARRAY_SIZE(pages)), pages); | |
da2c7009 LB |
1827 | if (ret == 0) { |
1828 | /* | |
1829 | * Only if we're going to lock these pages, | |
1830 | * can we find nothing at @index. | |
1831 | */ | |
1832 | ASSERT(page_ops & PAGE_LOCK); | |
49d4a334 LB |
1833 | err = -EAGAIN; |
1834 | goto out; | |
da2c7009 | 1835 | } |
8b62b72b | 1836 | |
da2c7009 | 1837 | for (i = 0; i < ret; i++) { |
c2790a2e | 1838 | if (page_ops & PAGE_SET_PRIVATE2) |
8b62b72b CM |
1839 | SetPagePrivate2(pages[i]); |
1840 | ||
c8b97818 | 1841 | if (pages[i] == locked_page) { |
09cbfeaf | 1842 | put_page(pages[i]); |
da2c7009 | 1843 | pages_locked++; |
c8b97818 CM |
1844 | continue; |
1845 | } | |
c2790a2e | 1846 | if (page_ops & PAGE_CLEAR_DIRTY) |
c8b97818 | 1847 | clear_page_dirty_for_io(pages[i]); |
c2790a2e | 1848 | if (page_ops & PAGE_SET_WRITEBACK) |
c8b97818 | 1849 | set_page_writeback(pages[i]); |
704de49d FM |
1850 | if (page_ops & PAGE_SET_ERROR) |
1851 | SetPageError(pages[i]); | |
c2790a2e | 1852 | if (page_ops & PAGE_END_WRITEBACK) |
c8b97818 | 1853 | end_page_writeback(pages[i]); |
c2790a2e | 1854 | if (page_ops & PAGE_UNLOCK) |
771ed689 | 1855 | unlock_page(pages[i]); |
da2c7009 LB |
1856 | if (page_ops & PAGE_LOCK) { |
1857 | lock_page(pages[i]); | |
1858 | if (!PageDirty(pages[i]) || | |
1859 | pages[i]->mapping != mapping) { | |
1860 | unlock_page(pages[i]); | |
1861 | put_page(pages[i]); | |
1862 | err = -EAGAIN; | |
1863 | goto out; | |
1864 | } | |
1865 | } | |
09cbfeaf | 1866 | put_page(pages[i]); |
da2c7009 | 1867 | pages_locked++; |
c8b97818 CM |
1868 | } |
1869 | nr_pages -= ret; | |
1870 | index += ret; | |
1871 | cond_resched(); | |
1872 | } | |
da2c7009 LB |
1873 | out: |
1874 | if (err && index_ret) | |
1875 | *index_ret = start_index + pages_locked - 1; | |
1876 | return err; | |
c8b97818 | 1877 | } |
c8b97818 | 1878 | |
873695b3 LB |
1879 | void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end, |
1880 | u64 delalloc_end, struct page *locked_page, | |
1881 | unsigned clear_bits, | |
1882 | unsigned long page_ops) | |
1883 | { | |
1884 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0, | |
ae0f1625 | 1885 | NULL); |
873695b3 LB |
1886 | |
1887 | __process_pages_contig(inode->i_mapping, locked_page, | |
1888 | start >> PAGE_SHIFT, end >> PAGE_SHIFT, | |
da2c7009 | 1889 | page_ops, NULL); |
873695b3 LB |
1890 | } |
1891 | ||
d352ac68 CM |
1892 | /* |
1893 | * count the number of bytes in the tree that have a given bit(s) | |
1894 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1895 | * cached. The total number found is returned. | |
1896 | */ | |
d1310b2e CM |
1897 | u64 count_range_bits(struct extent_io_tree *tree, |
1898 | u64 *start, u64 search_end, u64 max_bytes, | |
9ee49a04 | 1899 | unsigned bits, int contig) |
d1310b2e CM |
1900 | { |
1901 | struct rb_node *node; | |
1902 | struct extent_state *state; | |
1903 | u64 cur_start = *start; | |
1904 | u64 total_bytes = 0; | |
ec29ed5b | 1905 | u64 last = 0; |
d1310b2e CM |
1906 | int found = 0; |
1907 | ||
fae7f21c | 1908 | if (WARN_ON(search_end <= cur_start)) |
d1310b2e | 1909 | return 0; |
d1310b2e | 1910 | |
cad321ad | 1911 | spin_lock(&tree->lock); |
d1310b2e CM |
1912 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1913 | total_bytes = tree->dirty_bytes; | |
1914 | goto out; | |
1915 | } | |
1916 | /* | |
1917 | * this search will find all the extents that end after | |
1918 | * our range starts. | |
1919 | */ | |
80ea96b1 | 1920 | node = tree_search(tree, cur_start); |
d397712b | 1921 | if (!node) |
d1310b2e | 1922 | goto out; |
d1310b2e | 1923 | |
d397712b | 1924 | while (1) { |
d1310b2e CM |
1925 | state = rb_entry(node, struct extent_state, rb_node); |
1926 | if (state->start > search_end) | |
1927 | break; | |
ec29ed5b CM |
1928 | if (contig && found && state->start > last + 1) |
1929 | break; | |
1930 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1931 | total_bytes += min(search_end, state->end) + 1 - |
1932 | max(cur_start, state->start); | |
1933 | if (total_bytes >= max_bytes) | |
1934 | break; | |
1935 | if (!found) { | |
af60bed2 | 1936 | *start = max(cur_start, state->start); |
d1310b2e CM |
1937 | found = 1; |
1938 | } | |
ec29ed5b CM |
1939 | last = state->end; |
1940 | } else if (contig && found) { | |
1941 | break; | |
d1310b2e CM |
1942 | } |
1943 | node = rb_next(node); | |
1944 | if (!node) | |
1945 | break; | |
1946 | } | |
1947 | out: | |
cad321ad | 1948 | spin_unlock(&tree->lock); |
d1310b2e CM |
1949 | return total_bytes; |
1950 | } | |
b2950863 | 1951 | |
d352ac68 CM |
1952 | /* |
1953 | * set the private field for a given byte offset in the tree. If there isn't | |
1954 | * an extent_state there already, this does nothing. | |
1955 | */ | |
f827ba9a | 1956 | static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start, |
47dc196a | 1957 | struct io_failure_record *failrec) |
d1310b2e CM |
1958 | { |
1959 | struct rb_node *node; | |
1960 | struct extent_state *state; | |
1961 | int ret = 0; | |
1962 | ||
cad321ad | 1963 | spin_lock(&tree->lock); |
d1310b2e CM |
1964 | /* |
1965 | * this search will find all the extents that end after | |
1966 | * our range starts. | |
1967 | */ | |
80ea96b1 | 1968 | node = tree_search(tree, start); |
2b114d1d | 1969 | if (!node) { |
d1310b2e CM |
1970 | ret = -ENOENT; |
1971 | goto out; | |
1972 | } | |
1973 | state = rb_entry(node, struct extent_state, rb_node); | |
1974 | if (state->start != start) { | |
1975 | ret = -ENOENT; | |
1976 | goto out; | |
1977 | } | |
47dc196a | 1978 | state->failrec = failrec; |
d1310b2e | 1979 | out: |
cad321ad | 1980 | spin_unlock(&tree->lock); |
d1310b2e CM |
1981 | return ret; |
1982 | } | |
1983 | ||
f827ba9a | 1984 | static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start, |
47dc196a | 1985 | struct io_failure_record **failrec) |
d1310b2e CM |
1986 | { |
1987 | struct rb_node *node; | |
1988 | struct extent_state *state; | |
1989 | int ret = 0; | |
1990 | ||
cad321ad | 1991 | spin_lock(&tree->lock); |
d1310b2e CM |
1992 | /* |
1993 | * this search will find all the extents that end after | |
1994 | * our range starts. | |
1995 | */ | |
80ea96b1 | 1996 | node = tree_search(tree, start); |
2b114d1d | 1997 | if (!node) { |
d1310b2e CM |
1998 | ret = -ENOENT; |
1999 | goto out; | |
2000 | } | |
2001 | state = rb_entry(node, struct extent_state, rb_node); | |
2002 | if (state->start != start) { | |
2003 | ret = -ENOENT; | |
2004 | goto out; | |
2005 | } | |
47dc196a | 2006 | *failrec = state->failrec; |
d1310b2e | 2007 | out: |
cad321ad | 2008 | spin_unlock(&tree->lock); |
d1310b2e CM |
2009 | return ret; |
2010 | } | |
2011 | ||
2012 | /* | |
2013 | * searches a range in the state tree for a given mask. | |
70dec807 | 2014 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
2015 | * has the bits set. Otherwise, 1 is returned if any bit in the |
2016 | * range is found set. | |
2017 | */ | |
2018 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 2019 | unsigned bits, int filled, struct extent_state *cached) |
d1310b2e CM |
2020 | { |
2021 | struct extent_state *state = NULL; | |
2022 | struct rb_node *node; | |
2023 | int bitset = 0; | |
d1310b2e | 2024 | |
cad321ad | 2025 | spin_lock(&tree->lock); |
27a3507d | 2026 | if (cached && extent_state_in_tree(cached) && cached->start <= start && |
df98b6e2 | 2027 | cached->end > start) |
9655d298 CM |
2028 | node = &cached->rb_node; |
2029 | else | |
2030 | node = tree_search(tree, start); | |
d1310b2e CM |
2031 | while (node && start <= end) { |
2032 | state = rb_entry(node, struct extent_state, rb_node); | |
2033 | ||
2034 | if (filled && state->start > start) { | |
2035 | bitset = 0; | |
2036 | break; | |
2037 | } | |
2038 | ||
2039 | if (state->start > end) | |
2040 | break; | |
2041 | ||
2042 | if (state->state & bits) { | |
2043 | bitset = 1; | |
2044 | if (!filled) | |
2045 | break; | |
2046 | } else if (filled) { | |
2047 | bitset = 0; | |
2048 | break; | |
2049 | } | |
46562cec CM |
2050 | |
2051 | if (state->end == (u64)-1) | |
2052 | break; | |
2053 | ||
d1310b2e CM |
2054 | start = state->end + 1; |
2055 | if (start > end) | |
2056 | break; | |
2057 | node = rb_next(node); | |
2058 | if (!node) { | |
2059 | if (filled) | |
2060 | bitset = 0; | |
2061 | break; | |
2062 | } | |
2063 | } | |
cad321ad | 2064 | spin_unlock(&tree->lock); |
d1310b2e CM |
2065 | return bitset; |
2066 | } | |
d1310b2e CM |
2067 | |
2068 | /* | |
2069 | * helper function to set a given page up to date if all the | |
2070 | * extents in the tree for that page are up to date | |
2071 | */ | |
143bede5 | 2072 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e | 2073 | { |
4eee4fa4 | 2074 | u64 start = page_offset(page); |
09cbfeaf | 2075 | u64 end = start + PAGE_SIZE - 1; |
9655d298 | 2076 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 2077 | SetPageUptodate(page); |
d1310b2e CM |
2078 | } |
2079 | ||
7870d082 JB |
2080 | int free_io_failure(struct extent_io_tree *failure_tree, |
2081 | struct extent_io_tree *io_tree, | |
2082 | struct io_failure_record *rec) | |
4a54c8c1 JS |
2083 | { |
2084 | int ret; | |
2085 | int err = 0; | |
4a54c8c1 | 2086 | |
47dc196a | 2087 | set_state_failrec(failure_tree, rec->start, NULL); |
4a54c8c1 JS |
2088 | ret = clear_extent_bits(failure_tree, rec->start, |
2089 | rec->start + rec->len - 1, | |
91166212 | 2090 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 JS |
2091 | if (ret) |
2092 | err = ret; | |
2093 | ||
7870d082 | 2094 | ret = clear_extent_bits(io_tree, rec->start, |
53b381b3 | 2095 | rec->start + rec->len - 1, |
91166212 | 2096 | EXTENT_DAMAGED); |
53b381b3 DW |
2097 | if (ret && !err) |
2098 | err = ret; | |
4a54c8c1 JS |
2099 | |
2100 | kfree(rec); | |
2101 | return err; | |
2102 | } | |
2103 | ||
4a54c8c1 JS |
2104 | /* |
2105 | * this bypasses the standard btrfs submit functions deliberately, as | |
2106 | * the standard behavior is to write all copies in a raid setup. here we only | |
2107 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
2108 | * submit_bio directly. | |
3ec706c8 | 2109 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
2110 | * actually prevents the read that triggered the error from finishing. |
2111 | * currently, there can be no more than two copies of every data bit. thus, | |
2112 | * exactly one rewrite is required. | |
2113 | */ | |
6ec656bc JB |
2114 | int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start, |
2115 | u64 length, u64 logical, struct page *page, | |
2116 | unsigned int pg_offset, int mirror_num) | |
4a54c8c1 JS |
2117 | { |
2118 | struct bio *bio; | |
2119 | struct btrfs_device *dev; | |
4a54c8c1 JS |
2120 | u64 map_length = 0; |
2121 | u64 sector; | |
2122 | struct btrfs_bio *bbio = NULL; | |
2123 | int ret; | |
2124 | ||
1751e8a6 | 2125 | ASSERT(!(fs_info->sb->s_flags & SB_RDONLY)); |
4a54c8c1 JS |
2126 | BUG_ON(!mirror_num); |
2127 | ||
c5e4c3d7 | 2128 | bio = btrfs_io_bio_alloc(1); |
4f024f37 | 2129 | bio->bi_iter.bi_size = 0; |
4a54c8c1 JS |
2130 | map_length = length; |
2131 | ||
b5de8d0d FM |
2132 | /* |
2133 | * Avoid races with device replace and make sure our bbio has devices | |
2134 | * associated to its stripes that don't go away while we are doing the | |
2135 | * read repair operation. | |
2136 | */ | |
2137 | btrfs_bio_counter_inc_blocked(fs_info); | |
e4ff5fb5 | 2138 | if (btrfs_is_parity_mirror(fs_info, logical, length)) { |
c725328c LB |
2139 | /* |
2140 | * Note that we don't use BTRFS_MAP_WRITE because it's supposed | |
2141 | * to update all raid stripes, but here we just want to correct | |
2142 | * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad | |
2143 | * stripe's dev and sector. | |
2144 | */ | |
2145 | ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical, | |
2146 | &map_length, &bbio, 0); | |
2147 | if (ret) { | |
2148 | btrfs_bio_counter_dec(fs_info); | |
2149 | bio_put(bio); | |
2150 | return -EIO; | |
2151 | } | |
2152 | ASSERT(bbio->mirror_num == 1); | |
2153 | } else { | |
2154 | ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, | |
2155 | &map_length, &bbio, mirror_num); | |
2156 | if (ret) { | |
2157 | btrfs_bio_counter_dec(fs_info); | |
2158 | bio_put(bio); | |
2159 | return -EIO; | |
2160 | } | |
2161 | BUG_ON(mirror_num != bbio->mirror_num); | |
4a54c8c1 | 2162 | } |
c725328c LB |
2163 | |
2164 | sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9; | |
4f024f37 | 2165 | bio->bi_iter.bi_sector = sector; |
c725328c | 2166 | dev = bbio->stripes[bbio->mirror_num - 1].dev; |
6e9606d2 | 2167 | btrfs_put_bbio(bbio); |
ebbede42 AJ |
2168 | if (!dev || !dev->bdev || |
2169 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) { | |
b5de8d0d | 2170 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2171 | bio_put(bio); |
2172 | return -EIO; | |
2173 | } | |
74d46992 | 2174 | bio_set_dev(bio, dev->bdev); |
70fd7614 | 2175 | bio->bi_opf = REQ_OP_WRITE | REQ_SYNC; |
ffdd2018 | 2176 | bio_add_page(bio, page, length, pg_offset); |
4a54c8c1 | 2177 | |
4e49ea4a | 2178 | if (btrfsic_submit_bio_wait(bio)) { |
4a54c8c1 | 2179 | /* try to remap that extent elsewhere? */ |
b5de8d0d | 2180 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 | 2181 | bio_put(bio); |
442a4f63 | 2182 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
2183 | return -EIO; |
2184 | } | |
2185 | ||
b14af3b4 DS |
2186 | btrfs_info_rl_in_rcu(fs_info, |
2187 | "read error corrected: ino %llu off %llu (dev %s sector %llu)", | |
6ec656bc | 2188 | ino, start, |
1203b681 | 2189 | rcu_str_deref(dev->name), sector); |
b5de8d0d | 2190 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2191 | bio_put(bio); |
2192 | return 0; | |
2193 | } | |
2194 | ||
20a1fbf9 | 2195 | int btrfs_repair_eb_io_failure(struct extent_buffer *eb, int mirror_num) |
ea466794 | 2196 | { |
20a1fbf9 | 2197 | struct btrfs_fs_info *fs_info = eb->fs_info; |
ea466794 | 2198 | u64 start = eb->start; |
cc5e31a4 | 2199 | int i, num_pages = num_extent_pages(eb); |
d95603b2 | 2200 | int ret = 0; |
ea466794 | 2201 | |
bc98a42c | 2202 | if (sb_rdonly(fs_info->sb)) |
908960c6 ID |
2203 | return -EROFS; |
2204 | ||
ea466794 | 2205 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 2206 | struct page *p = eb->pages[i]; |
1203b681 | 2207 | |
6ec656bc | 2208 | ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p, |
1203b681 | 2209 | start - page_offset(p), mirror_num); |
ea466794 JB |
2210 | if (ret) |
2211 | break; | |
09cbfeaf | 2212 | start += PAGE_SIZE; |
ea466794 JB |
2213 | } |
2214 | ||
2215 | return ret; | |
2216 | } | |
2217 | ||
4a54c8c1 JS |
2218 | /* |
2219 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2220 | * to see if we need to process or clean up an io_failure_record | |
2221 | */ | |
7870d082 JB |
2222 | int clean_io_failure(struct btrfs_fs_info *fs_info, |
2223 | struct extent_io_tree *failure_tree, | |
2224 | struct extent_io_tree *io_tree, u64 start, | |
2225 | struct page *page, u64 ino, unsigned int pg_offset) | |
4a54c8c1 JS |
2226 | { |
2227 | u64 private; | |
4a54c8c1 | 2228 | struct io_failure_record *failrec; |
4a54c8c1 JS |
2229 | struct extent_state *state; |
2230 | int num_copies; | |
4a54c8c1 | 2231 | int ret; |
4a54c8c1 JS |
2232 | |
2233 | private = 0; | |
7870d082 JB |
2234 | ret = count_range_bits(failure_tree, &private, (u64)-1, 1, |
2235 | EXTENT_DIRTY, 0); | |
4a54c8c1 JS |
2236 | if (!ret) |
2237 | return 0; | |
2238 | ||
7870d082 | 2239 | ret = get_state_failrec(failure_tree, start, &failrec); |
4a54c8c1 JS |
2240 | if (ret) |
2241 | return 0; | |
2242 | ||
4a54c8c1 JS |
2243 | BUG_ON(!failrec->this_mirror); |
2244 | ||
2245 | if (failrec->in_validation) { | |
2246 | /* there was no real error, just free the record */ | |
ab8d0fc4 JM |
2247 | btrfs_debug(fs_info, |
2248 | "clean_io_failure: freeing dummy error at %llu", | |
2249 | failrec->start); | |
4a54c8c1 JS |
2250 | goto out; |
2251 | } | |
bc98a42c | 2252 | if (sb_rdonly(fs_info->sb)) |
908960c6 | 2253 | goto out; |
4a54c8c1 | 2254 | |
7870d082 JB |
2255 | spin_lock(&io_tree->lock); |
2256 | state = find_first_extent_bit_state(io_tree, | |
4a54c8c1 JS |
2257 | failrec->start, |
2258 | EXTENT_LOCKED); | |
7870d082 | 2259 | spin_unlock(&io_tree->lock); |
4a54c8c1 | 2260 | |
883d0de4 MX |
2261 | if (state && state->start <= failrec->start && |
2262 | state->end >= failrec->start + failrec->len - 1) { | |
3ec706c8 SB |
2263 | num_copies = btrfs_num_copies(fs_info, failrec->logical, |
2264 | failrec->len); | |
4a54c8c1 | 2265 | if (num_copies > 1) { |
7870d082 JB |
2266 | repair_io_failure(fs_info, ino, start, failrec->len, |
2267 | failrec->logical, page, pg_offset, | |
2268 | failrec->failed_mirror); | |
4a54c8c1 JS |
2269 | } |
2270 | } | |
2271 | ||
2272 | out: | |
7870d082 | 2273 | free_io_failure(failure_tree, io_tree, failrec); |
4a54c8c1 | 2274 | |
454ff3de | 2275 | return 0; |
4a54c8c1 JS |
2276 | } |
2277 | ||
f612496b MX |
2278 | /* |
2279 | * Can be called when | |
2280 | * - hold extent lock | |
2281 | * - under ordered extent | |
2282 | * - the inode is freeing | |
2283 | */ | |
7ab7956e | 2284 | void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end) |
f612496b | 2285 | { |
7ab7956e | 2286 | struct extent_io_tree *failure_tree = &inode->io_failure_tree; |
f612496b MX |
2287 | struct io_failure_record *failrec; |
2288 | struct extent_state *state, *next; | |
2289 | ||
2290 | if (RB_EMPTY_ROOT(&failure_tree->state)) | |
2291 | return; | |
2292 | ||
2293 | spin_lock(&failure_tree->lock); | |
2294 | state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY); | |
2295 | while (state) { | |
2296 | if (state->start > end) | |
2297 | break; | |
2298 | ||
2299 | ASSERT(state->end <= end); | |
2300 | ||
2301 | next = next_state(state); | |
2302 | ||
47dc196a | 2303 | failrec = state->failrec; |
f612496b MX |
2304 | free_extent_state(state); |
2305 | kfree(failrec); | |
2306 | ||
2307 | state = next; | |
2308 | } | |
2309 | spin_unlock(&failure_tree->lock); | |
2310 | } | |
2311 | ||
2fe6303e | 2312 | int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end, |
47dc196a | 2313 | struct io_failure_record **failrec_ret) |
4a54c8c1 | 2314 | { |
ab8d0fc4 | 2315 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e | 2316 | struct io_failure_record *failrec; |
4a54c8c1 | 2317 | struct extent_map *em; |
4a54c8c1 JS |
2318 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2319 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2320 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
4a54c8c1 | 2321 | int ret; |
4a54c8c1 JS |
2322 | u64 logical; |
2323 | ||
47dc196a | 2324 | ret = get_state_failrec(failure_tree, start, &failrec); |
4a54c8c1 JS |
2325 | if (ret) { |
2326 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2327 | if (!failrec) | |
2328 | return -ENOMEM; | |
2fe6303e | 2329 | |
4a54c8c1 JS |
2330 | failrec->start = start; |
2331 | failrec->len = end - start + 1; | |
2332 | failrec->this_mirror = 0; | |
2333 | failrec->bio_flags = 0; | |
2334 | failrec->in_validation = 0; | |
2335 | ||
2336 | read_lock(&em_tree->lock); | |
2337 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2338 | if (!em) { | |
2339 | read_unlock(&em_tree->lock); | |
2340 | kfree(failrec); | |
2341 | return -EIO; | |
2342 | } | |
2343 | ||
68ba990f | 2344 | if (em->start > start || em->start + em->len <= start) { |
4a54c8c1 JS |
2345 | free_extent_map(em); |
2346 | em = NULL; | |
2347 | } | |
2348 | read_unlock(&em_tree->lock); | |
7a2d6a64 | 2349 | if (!em) { |
4a54c8c1 JS |
2350 | kfree(failrec); |
2351 | return -EIO; | |
2352 | } | |
2fe6303e | 2353 | |
4a54c8c1 JS |
2354 | logical = start - em->start; |
2355 | logical = em->block_start + logical; | |
2356 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2357 | logical = em->block_start; | |
2358 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2359 | extent_set_compress_type(&failrec->bio_flags, | |
2360 | em->compress_type); | |
2361 | } | |
2fe6303e | 2362 | |
ab8d0fc4 JM |
2363 | btrfs_debug(fs_info, |
2364 | "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu", | |
2365 | logical, start, failrec->len); | |
2fe6303e | 2366 | |
4a54c8c1 JS |
2367 | failrec->logical = logical; |
2368 | free_extent_map(em); | |
2369 | ||
2370 | /* set the bits in the private failure tree */ | |
2371 | ret = set_extent_bits(failure_tree, start, end, | |
ceeb0ae7 | 2372 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 | 2373 | if (ret >= 0) |
47dc196a | 2374 | ret = set_state_failrec(failure_tree, start, failrec); |
4a54c8c1 JS |
2375 | /* set the bits in the inode's tree */ |
2376 | if (ret >= 0) | |
ceeb0ae7 | 2377 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED); |
4a54c8c1 JS |
2378 | if (ret < 0) { |
2379 | kfree(failrec); | |
2380 | return ret; | |
2381 | } | |
2382 | } else { | |
ab8d0fc4 JM |
2383 | btrfs_debug(fs_info, |
2384 | "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d", | |
2385 | failrec->logical, failrec->start, failrec->len, | |
2386 | failrec->in_validation); | |
4a54c8c1 JS |
2387 | /* |
2388 | * when data can be on disk more than twice, add to failrec here | |
2389 | * (e.g. with a list for failed_mirror) to make | |
2390 | * clean_io_failure() clean all those errors at once. | |
2391 | */ | |
2392 | } | |
2fe6303e MX |
2393 | |
2394 | *failrec_ret = failrec; | |
2395 | ||
2396 | return 0; | |
2397 | } | |
2398 | ||
a0b60d72 | 2399 | bool btrfs_check_repairable(struct inode *inode, unsigned failed_bio_pages, |
2fe6303e MX |
2400 | struct io_failure_record *failrec, int failed_mirror) |
2401 | { | |
ab8d0fc4 | 2402 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e MX |
2403 | int num_copies; |
2404 | ||
ab8d0fc4 | 2405 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
4a54c8c1 JS |
2406 | if (num_copies == 1) { |
2407 | /* | |
2408 | * we only have a single copy of the data, so don't bother with | |
2409 | * all the retry and error correction code that follows. no | |
2410 | * matter what the error is, it is very likely to persist. | |
2411 | */ | |
ab8d0fc4 JM |
2412 | btrfs_debug(fs_info, |
2413 | "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
2414 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2415 | return false; |
4a54c8c1 JS |
2416 | } |
2417 | ||
4a54c8c1 JS |
2418 | /* |
2419 | * there are two premises: | |
2420 | * a) deliver good data to the caller | |
2421 | * b) correct the bad sectors on disk | |
2422 | */ | |
a0b60d72 | 2423 | if (failed_bio_pages > 1) { |
4a54c8c1 JS |
2424 | /* |
2425 | * to fulfill b), we need to know the exact failing sectors, as | |
2426 | * we don't want to rewrite any more than the failed ones. thus, | |
2427 | * we need separate read requests for the failed bio | |
2428 | * | |
2429 | * if the following BUG_ON triggers, our validation request got | |
2430 | * merged. we need separate requests for our algorithm to work. | |
2431 | */ | |
2432 | BUG_ON(failrec->in_validation); | |
2433 | failrec->in_validation = 1; | |
2434 | failrec->this_mirror = failed_mirror; | |
4a54c8c1 JS |
2435 | } else { |
2436 | /* | |
2437 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2438 | * of the failed sector and if we succeed, we have setup | |
2439 | * everything for repair_io_failure to do the rest for us. | |
2440 | */ | |
2441 | if (failrec->in_validation) { | |
2442 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2443 | failrec->in_validation = 0; | |
2444 | failrec->this_mirror = 0; | |
2445 | } | |
2446 | failrec->failed_mirror = failed_mirror; | |
2447 | failrec->this_mirror++; | |
2448 | if (failrec->this_mirror == failed_mirror) | |
2449 | failrec->this_mirror++; | |
4a54c8c1 JS |
2450 | } |
2451 | ||
facc8a22 | 2452 | if (failrec->this_mirror > num_copies) { |
ab8d0fc4 JM |
2453 | btrfs_debug(fs_info, |
2454 | "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
2455 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2456 | return false; |
4a54c8c1 JS |
2457 | } |
2458 | ||
c3cfb656 | 2459 | return true; |
2fe6303e MX |
2460 | } |
2461 | ||
2462 | ||
2463 | struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio, | |
2464 | struct io_failure_record *failrec, | |
2465 | struct page *page, int pg_offset, int icsum, | |
8b110e39 | 2466 | bio_end_io_t *endio_func, void *data) |
2fe6303e | 2467 | { |
0b246afa | 2468 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e MX |
2469 | struct bio *bio; |
2470 | struct btrfs_io_bio *btrfs_failed_bio; | |
2471 | struct btrfs_io_bio *btrfs_bio; | |
2472 | ||
c5e4c3d7 | 2473 | bio = btrfs_io_bio_alloc(1); |
2fe6303e | 2474 | bio->bi_end_io = endio_func; |
4f024f37 | 2475 | bio->bi_iter.bi_sector = failrec->logical >> 9; |
74d46992 | 2476 | bio_set_dev(bio, fs_info->fs_devices->latest_bdev); |
4f024f37 | 2477 | bio->bi_iter.bi_size = 0; |
8b110e39 | 2478 | bio->bi_private = data; |
4a54c8c1 | 2479 | |
facc8a22 MX |
2480 | btrfs_failed_bio = btrfs_io_bio(failed_bio); |
2481 | if (btrfs_failed_bio->csum) { | |
facc8a22 MX |
2482 | u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); |
2483 | ||
2484 | btrfs_bio = btrfs_io_bio(bio); | |
2485 | btrfs_bio->csum = btrfs_bio->csum_inline; | |
2fe6303e MX |
2486 | icsum *= csum_size; |
2487 | memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum, | |
facc8a22 MX |
2488 | csum_size); |
2489 | } | |
2490 | ||
2fe6303e MX |
2491 | bio_add_page(bio, page, failrec->len, pg_offset); |
2492 | ||
2493 | return bio; | |
2494 | } | |
2495 | ||
2496 | /* | |
78e62c02 NB |
2497 | * This is a generic handler for readpage errors. If other copies exist, read |
2498 | * those and write back good data to the failed position. Does not investigate | |
2499 | * in remapping the failed extent elsewhere, hoping the device will be smart | |
2500 | * enough to do this as needed | |
2fe6303e | 2501 | */ |
2fe6303e MX |
2502 | static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset, |
2503 | struct page *page, u64 start, u64 end, | |
2504 | int failed_mirror) | |
2505 | { | |
2506 | struct io_failure_record *failrec; | |
2507 | struct inode *inode = page->mapping->host; | |
2508 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
7870d082 | 2509 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2fe6303e | 2510 | struct bio *bio; |
70fd7614 | 2511 | int read_mode = 0; |
4e4cbee9 | 2512 | blk_status_t status; |
2fe6303e | 2513 | int ret; |
8a2ee44a | 2514 | unsigned failed_bio_pages = failed_bio->bi_iter.bi_size >> PAGE_SHIFT; |
2fe6303e | 2515 | |
1f7ad75b | 2516 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
2fe6303e MX |
2517 | |
2518 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
2519 | if (ret) | |
2520 | return ret; | |
2521 | ||
a0b60d72 | 2522 | if (!btrfs_check_repairable(inode, failed_bio_pages, failrec, |
c3cfb656 | 2523 | failed_mirror)) { |
7870d082 | 2524 | free_io_failure(failure_tree, tree, failrec); |
2fe6303e MX |
2525 | return -EIO; |
2526 | } | |
2527 | ||
a0b60d72 | 2528 | if (failed_bio_pages > 1) |
70fd7614 | 2529 | read_mode |= REQ_FAILFAST_DEV; |
2fe6303e MX |
2530 | |
2531 | phy_offset >>= inode->i_sb->s_blocksize_bits; | |
2532 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2533 | start - page_offset(page), | |
8b110e39 MX |
2534 | (int)phy_offset, failed_bio->bi_end_io, |
2535 | NULL); | |
ebcc3263 | 2536 | bio->bi_opf = REQ_OP_READ | read_mode; |
4a54c8c1 | 2537 | |
ab8d0fc4 JM |
2538 | btrfs_debug(btrfs_sb(inode->i_sb), |
2539 | "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d", | |
2540 | read_mode, failrec->this_mirror, failrec->in_validation); | |
4a54c8c1 | 2541 | |
8c27cb35 | 2542 | status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror, |
50489a57 | 2543 | failrec->bio_flags); |
4e4cbee9 | 2544 | if (status) { |
7870d082 | 2545 | free_io_failure(failure_tree, tree, failrec); |
6c387ab2 | 2546 | bio_put(bio); |
4e4cbee9 | 2547 | ret = blk_status_to_errno(status); |
6c387ab2 MX |
2548 | } |
2549 | ||
013bd4c3 | 2550 | return ret; |
4a54c8c1 JS |
2551 | } |
2552 | ||
d1310b2e CM |
2553 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2554 | ||
b5227c07 | 2555 | void end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
87826df0 JM |
2556 | { |
2557 | int uptodate = (err == 0); | |
3e2426bd | 2558 | int ret = 0; |
87826df0 | 2559 | |
c629732d | 2560 | btrfs_writepage_endio_finish_ordered(page, start, end, uptodate); |
87826df0 | 2561 | |
87826df0 | 2562 | if (!uptodate) { |
87826df0 JM |
2563 | ClearPageUptodate(page); |
2564 | SetPageError(page); | |
bff5baf8 | 2565 | ret = err < 0 ? err : -EIO; |
5dca6eea | 2566 | mapping_set_error(page->mapping, ret); |
87826df0 | 2567 | } |
87826df0 JM |
2568 | } |
2569 | ||
d1310b2e CM |
2570 | /* |
2571 | * after a writepage IO is done, we need to: | |
2572 | * clear the uptodate bits on error | |
2573 | * clear the writeback bits in the extent tree for this IO | |
2574 | * end_page_writeback if the page has no more pending IO | |
2575 | * | |
2576 | * Scheduling is not allowed, so the extent state tree is expected | |
2577 | * to have one and only one object corresponding to this IO. | |
2578 | */ | |
4246a0b6 | 2579 | static void end_bio_extent_writepage(struct bio *bio) |
d1310b2e | 2580 | { |
4e4cbee9 | 2581 | int error = blk_status_to_errno(bio->bi_status); |
2c30c71b | 2582 | struct bio_vec *bvec; |
d1310b2e CM |
2583 | u64 start; |
2584 | u64 end; | |
2c30c71b | 2585 | int i; |
6dc4f100 | 2586 | struct bvec_iter_all iter_all; |
d1310b2e | 2587 | |
c09abff8 | 2588 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
6dc4f100 | 2589 | bio_for_each_segment_all(bvec, bio, i, iter_all) { |
d1310b2e | 2590 | struct page *page = bvec->bv_page; |
0b246afa JM |
2591 | struct inode *inode = page->mapping->host; |
2592 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
902b22f3 | 2593 | |
17a5adcc AO |
2594 | /* We always issue full-page reads, but if some block |
2595 | * in a page fails to read, blk_update_request() will | |
2596 | * advance bv_offset and adjust bv_len to compensate. | |
2597 | * Print a warning for nonzero offsets, and an error | |
2598 | * if they don't add up to a full page. */ | |
09cbfeaf KS |
2599 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { |
2600 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
0b246afa | 2601 | btrfs_err(fs_info, |
efe120a0 FH |
2602 | "partial page write in btrfs with offset %u and length %u", |
2603 | bvec->bv_offset, bvec->bv_len); | |
2604 | else | |
0b246afa | 2605 | btrfs_info(fs_info, |
5d163e0e | 2606 | "incomplete page write in btrfs with offset %u and length %u", |
efe120a0 FH |
2607 | bvec->bv_offset, bvec->bv_len); |
2608 | } | |
d1310b2e | 2609 | |
17a5adcc AO |
2610 | start = page_offset(page); |
2611 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
d1310b2e | 2612 | |
4e4cbee9 | 2613 | end_extent_writepage(page, error, start, end); |
17a5adcc | 2614 | end_page_writeback(page); |
2c30c71b | 2615 | } |
2b1f55b0 | 2616 | |
d1310b2e | 2617 | bio_put(bio); |
d1310b2e CM |
2618 | } |
2619 | ||
883d0de4 MX |
2620 | static void |
2621 | endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len, | |
2622 | int uptodate) | |
2623 | { | |
2624 | struct extent_state *cached = NULL; | |
2625 | u64 end = start + len - 1; | |
2626 | ||
2627 | if (uptodate && tree->track_uptodate) | |
2628 | set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC); | |
d810a4be | 2629 | unlock_extent_cached_atomic(tree, start, end, &cached); |
883d0de4 MX |
2630 | } |
2631 | ||
d1310b2e CM |
2632 | /* |
2633 | * after a readpage IO is done, we need to: | |
2634 | * clear the uptodate bits on error | |
2635 | * set the uptodate bits if things worked | |
2636 | * set the page up to date if all extents in the tree are uptodate | |
2637 | * clear the lock bit in the extent tree | |
2638 | * unlock the page if there are no other extents locked for it | |
2639 | * | |
2640 | * Scheduling is not allowed, so the extent state tree is expected | |
2641 | * to have one and only one object corresponding to this IO. | |
2642 | */ | |
4246a0b6 | 2643 | static void end_bio_extent_readpage(struct bio *bio) |
d1310b2e | 2644 | { |
2c30c71b | 2645 | struct bio_vec *bvec; |
4e4cbee9 | 2646 | int uptodate = !bio->bi_status; |
facc8a22 | 2647 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); |
7870d082 | 2648 | struct extent_io_tree *tree, *failure_tree; |
facc8a22 | 2649 | u64 offset = 0; |
d1310b2e CM |
2650 | u64 start; |
2651 | u64 end; | |
facc8a22 | 2652 | u64 len; |
883d0de4 MX |
2653 | u64 extent_start = 0; |
2654 | u64 extent_len = 0; | |
5cf1ab56 | 2655 | int mirror; |
d1310b2e | 2656 | int ret; |
2c30c71b | 2657 | int i; |
6dc4f100 | 2658 | struct bvec_iter_all iter_all; |
d1310b2e | 2659 | |
c09abff8 | 2660 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
6dc4f100 | 2661 | bio_for_each_segment_all(bvec, bio, i, iter_all) { |
d1310b2e | 2662 | struct page *page = bvec->bv_page; |
a71754fc | 2663 | struct inode *inode = page->mapping->host; |
ab8d0fc4 | 2664 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
78e62c02 NB |
2665 | bool data_inode = btrfs_ino(BTRFS_I(inode)) |
2666 | != BTRFS_BTREE_INODE_OBJECTID; | |
507903b8 | 2667 | |
ab8d0fc4 JM |
2668 | btrfs_debug(fs_info, |
2669 | "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u", | |
4e4cbee9 | 2670 | (u64)bio->bi_iter.bi_sector, bio->bi_status, |
ab8d0fc4 | 2671 | io_bio->mirror_num); |
a71754fc | 2672 | tree = &BTRFS_I(inode)->io_tree; |
7870d082 | 2673 | failure_tree = &BTRFS_I(inode)->io_failure_tree; |
902b22f3 | 2674 | |
17a5adcc AO |
2675 | /* We always issue full-page reads, but if some block |
2676 | * in a page fails to read, blk_update_request() will | |
2677 | * advance bv_offset and adjust bv_len to compensate. | |
2678 | * Print a warning for nonzero offsets, and an error | |
2679 | * if they don't add up to a full page. */ | |
09cbfeaf KS |
2680 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { |
2681 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
ab8d0fc4 JM |
2682 | btrfs_err(fs_info, |
2683 | "partial page read in btrfs with offset %u and length %u", | |
efe120a0 FH |
2684 | bvec->bv_offset, bvec->bv_len); |
2685 | else | |
ab8d0fc4 JM |
2686 | btrfs_info(fs_info, |
2687 | "incomplete page read in btrfs with offset %u and length %u", | |
efe120a0 FH |
2688 | bvec->bv_offset, bvec->bv_len); |
2689 | } | |
d1310b2e | 2690 | |
17a5adcc AO |
2691 | start = page_offset(page); |
2692 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
facc8a22 | 2693 | len = bvec->bv_len; |
d1310b2e | 2694 | |
9be3395b | 2695 | mirror = io_bio->mirror_num; |
78e62c02 | 2696 | if (likely(uptodate)) { |
facc8a22 MX |
2697 | ret = tree->ops->readpage_end_io_hook(io_bio, offset, |
2698 | page, start, end, | |
2699 | mirror); | |
5ee0844d | 2700 | if (ret) |
d1310b2e | 2701 | uptodate = 0; |
5ee0844d | 2702 | else |
7870d082 JB |
2703 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
2704 | failure_tree, tree, start, | |
2705 | page, | |
2706 | btrfs_ino(BTRFS_I(inode)), 0); | |
d1310b2e | 2707 | } |
ea466794 | 2708 | |
f2a09da9 MX |
2709 | if (likely(uptodate)) |
2710 | goto readpage_ok; | |
2711 | ||
78e62c02 | 2712 | if (data_inode) { |
9d0d1c8b | 2713 | |
f4a8e656 | 2714 | /* |
78e62c02 NB |
2715 | * The generic bio_readpage_error handles errors the |
2716 | * following way: If possible, new read requests are | |
2717 | * created and submitted and will end up in | |
2718 | * end_bio_extent_readpage as well (if we're lucky, | |
2719 | * not in the !uptodate case). In that case it returns | |
2720 | * 0 and we just go on with the next page in our bio. | |
2721 | * If it can't handle the error it will return -EIO and | |
2722 | * we remain responsible for that page. | |
f4a8e656 | 2723 | */ |
78e62c02 NB |
2724 | ret = bio_readpage_error(bio, offset, page, start, end, |
2725 | mirror); | |
2726 | if (ret == 0) { | |
2727 | uptodate = !bio->bi_status; | |
2728 | offset += len; | |
2729 | continue; | |
2730 | } | |
2731 | } else { | |
2732 | struct extent_buffer *eb; | |
2733 | ||
2734 | eb = (struct extent_buffer *)page->private; | |
2735 | set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); | |
2736 | eb->read_mirror = mirror; | |
2737 | atomic_dec(&eb->io_pages); | |
2738 | if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, | |
2739 | &eb->bflags)) | |
2740 | btree_readahead_hook(eb, -EIO); | |
7e38326f | 2741 | } |
f2a09da9 | 2742 | readpage_ok: |
883d0de4 | 2743 | if (likely(uptodate)) { |
a71754fc | 2744 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 2745 | pgoff_t end_index = i_size >> PAGE_SHIFT; |
a583c026 | 2746 | unsigned off; |
a71754fc JB |
2747 | |
2748 | /* Zero out the end if this page straddles i_size */ | |
7073017a | 2749 | off = offset_in_page(i_size); |
a583c026 | 2750 | if (page->index == end_index && off) |
09cbfeaf | 2751 | zero_user_segment(page, off, PAGE_SIZE); |
17a5adcc | 2752 | SetPageUptodate(page); |
70dec807 | 2753 | } else { |
17a5adcc AO |
2754 | ClearPageUptodate(page); |
2755 | SetPageError(page); | |
70dec807 | 2756 | } |
17a5adcc | 2757 | unlock_page(page); |
facc8a22 | 2758 | offset += len; |
883d0de4 MX |
2759 | |
2760 | if (unlikely(!uptodate)) { | |
2761 | if (extent_len) { | |
2762 | endio_readpage_release_extent(tree, | |
2763 | extent_start, | |
2764 | extent_len, 1); | |
2765 | extent_start = 0; | |
2766 | extent_len = 0; | |
2767 | } | |
2768 | endio_readpage_release_extent(tree, start, | |
2769 | end - start + 1, 0); | |
2770 | } else if (!extent_len) { | |
2771 | extent_start = start; | |
2772 | extent_len = end + 1 - start; | |
2773 | } else if (extent_start + extent_len == start) { | |
2774 | extent_len += end + 1 - start; | |
2775 | } else { | |
2776 | endio_readpage_release_extent(tree, extent_start, | |
2777 | extent_len, uptodate); | |
2778 | extent_start = start; | |
2779 | extent_len = end + 1 - start; | |
2780 | } | |
2c30c71b | 2781 | } |
d1310b2e | 2782 | |
883d0de4 MX |
2783 | if (extent_len) |
2784 | endio_readpage_release_extent(tree, extent_start, extent_len, | |
2785 | uptodate); | |
b3a0dd50 | 2786 | btrfs_io_bio_free_csum(io_bio); |
d1310b2e | 2787 | bio_put(bio); |
d1310b2e CM |
2788 | } |
2789 | ||
9be3395b | 2790 | /* |
184f999e DS |
2791 | * Initialize the members up to but not including 'bio'. Use after allocating a |
2792 | * new bio by bio_alloc_bioset as it does not initialize the bytes outside of | |
2793 | * 'bio' because use of __GFP_ZERO is not supported. | |
9be3395b | 2794 | */ |
184f999e | 2795 | static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio) |
d1310b2e | 2796 | { |
184f999e DS |
2797 | memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio)); |
2798 | } | |
d1310b2e | 2799 | |
9be3395b | 2800 | /* |
6e707bcd DS |
2801 | * The following helpers allocate a bio. As it's backed by a bioset, it'll |
2802 | * never fail. We're returning a bio right now but you can call btrfs_io_bio | |
2803 | * for the appropriate container_of magic | |
9be3395b | 2804 | */ |
c821e7f3 | 2805 | struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte) |
d1310b2e CM |
2806 | { |
2807 | struct bio *bio; | |
d1310b2e | 2808 | |
8ac9f7c1 | 2809 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &btrfs_bioset); |
74d46992 | 2810 | bio_set_dev(bio, bdev); |
c821e7f3 | 2811 | bio->bi_iter.bi_sector = first_byte >> 9; |
184f999e | 2812 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
d1310b2e CM |
2813 | return bio; |
2814 | } | |
2815 | ||
8b6c1d56 | 2816 | struct bio *btrfs_bio_clone(struct bio *bio) |
9be3395b | 2817 | { |
23ea8e5a MX |
2818 | struct btrfs_io_bio *btrfs_bio; |
2819 | struct bio *new; | |
9be3395b | 2820 | |
6e707bcd | 2821 | /* Bio allocation backed by a bioset does not fail */ |
8ac9f7c1 | 2822 | new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset); |
6e707bcd | 2823 | btrfs_bio = btrfs_io_bio(new); |
184f999e | 2824 | btrfs_io_bio_init(btrfs_bio); |
6e707bcd | 2825 | btrfs_bio->iter = bio->bi_iter; |
23ea8e5a MX |
2826 | return new; |
2827 | } | |
9be3395b | 2828 | |
c5e4c3d7 | 2829 | struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs) |
9be3395b | 2830 | { |
facc8a22 MX |
2831 | struct bio *bio; |
2832 | ||
6e707bcd | 2833 | /* Bio allocation backed by a bioset does not fail */ |
8ac9f7c1 | 2834 | bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset); |
184f999e | 2835 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
facc8a22 | 2836 | return bio; |
9be3395b CM |
2837 | } |
2838 | ||
e477094f | 2839 | struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size) |
2f8e9140 LB |
2840 | { |
2841 | struct bio *bio; | |
2842 | struct btrfs_io_bio *btrfs_bio; | |
2843 | ||
2844 | /* this will never fail when it's backed by a bioset */ | |
8ac9f7c1 | 2845 | bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset); |
2f8e9140 LB |
2846 | ASSERT(bio); |
2847 | ||
2848 | btrfs_bio = btrfs_io_bio(bio); | |
184f999e | 2849 | btrfs_io_bio_init(btrfs_bio); |
2f8e9140 LB |
2850 | |
2851 | bio_trim(bio, offset >> 9, size >> 9); | |
17347cec | 2852 | btrfs_bio->iter = bio->bi_iter; |
2f8e9140 LB |
2853 | return bio; |
2854 | } | |
9be3395b | 2855 | |
4b81ba48 DS |
2856 | /* |
2857 | * @opf: bio REQ_OP_* and REQ_* flags as one value | |
b8b3d625 DS |
2858 | * @tree: tree so we can call our merge_bio hook |
2859 | * @wbc: optional writeback control for io accounting | |
2860 | * @page: page to add to the bio | |
2861 | * @pg_offset: offset of the new bio or to check whether we are adding | |
2862 | * a contiguous page to the previous one | |
2863 | * @size: portion of page that we want to write | |
2864 | * @offset: starting offset in the page | |
2865 | * @bdev: attach newly created bios to this bdev | |
5c2b1fd7 | 2866 | * @bio_ret: must be valid pointer, newly allocated bio will be stored there |
b8b3d625 DS |
2867 | * @end_io_func: end_io callback for new bio |
2868 | * @mirror_num: desired mirror to read/write | |
2869 | * @prev_bio_flags: flags of previous bio to see if we can merge the current one | |
2870 | * @bio_flags: flags of the current bio to see if we can merge them | |
4b81ba48 DS |
2871 | */ |
2872 | static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree, | |
da2f0f74 | 2873 | struct writeback_control *wbc, |
6273b7f8 | 2874 | struct page *page, u64 offset, |
6c5a4e2c | 2875 | size_t size, unsigned long pg_offset, |
d1310b2e CM |
2876 | struct block_device *bdev, |
2877 | struct bio **bio_ret, | |
f188591e | 2878 | bio_end_io_t end_io_func, |
c8b97818 CM |
2879 | int mirror_num, |
2880 | unsigned long prev_bio_flags, | |
005efedf FM |
2881 | unsigned long bio_flags, |
2882 | bool force_bio_submit) | |
d1310b2e CM |
2883 | { |
2884 | int ret = 0; | |
2885 | struct bio *bio; | |
09cbfeaf | 2886 | size_t page_size = min_t(size_t, size, PAGE_SIZE); |
6273b7f8 | 2887 | sector_t sector = offset >> 9; |
d1310b2e | 2888 | |
5c2b1fd7 DS |
2889 | ASSERT(bio_ret); |
2890 | ||
2891 | if (*bio_ret) { | |
0c8508a6 DS |
2892 | bool contig; |
2893 | bool can_merge = true; | |
2894 | ||
d1310b2e | 2895 | bio = *bio_ret; |
0c8508a6 | 2896 | if (prev_bio_flags & EXTENT_BIO_COMPRESSED) |
4f024f37 | 2897 | contig = bio->bi_iter.bi_sector == sector; |
c8b97818 | 2898 | else |
f73a1c7d | 2899 | contig = bio_end_sector(bio) == sector; |
c8b97818 | 2900 | |
da12fe54 NB |
2901 | ASSERT(tree->ops); |
2902 | if (btrfs_bio_fits_in_stripe(page, page_size, bio, bio_flags)) | |
0c8508a6 DS |
2903 | can_merge = false; |
2904 | ||
2905 | if (prev_bio_flags != bio_flags || !contig || !can_merge || | |
005efedf | 2906 | force_bio_submit || |
6c5a4e2c | 2907 | bio_add_page(bio, page, page_size, pg_offset) < page_size) { |
1f7ad75b | 2908 | ret = submit_one_bio(bio, mirror_num, prev_bio_flags); |
289454ad NA |
2909 | if (ret < 0) { |
2910 | *bio_ret = NULL; | |
79787eaa | 2911 | return ret; |
289454ad | 2912 | } |
d1310b2e CM |
2913 | bio = NULL; |
2914 | } else { | |
da2f0f74 CM |
2915 | if (wbc) |
2916 | wbc_account_io(wbc, page, page_size); | |
d1310b2e CM |
2917 | return 0; |
2918 | } | |
2919 | } | |
c8b97818 | 2920 | |
6273b7f8 | 2921 | bio = btrfs_bio_alloc(bdev, offset); |
6c5a4e2c | 2922 | bio_add_page(bio, page, page_size, pg_offset); |
d1310b2e CM |
2923 | bio->bi_end_io = end_io_func; |
2924 | bio->bi_private = tree; | |
e6959b93 | 2925 | bio->bi_write_hint = page->mapping->host->i_write_hint; |
4b81ba48 | 2926 | bio->bi_opf = opf; |
da2f0f74 CM |
2927 | if (wbc) { |
2928 | wbc_init_bio(wbc, bio); | |
2929 | wbc_account_io(wbc, page, page_size); | |
2930 | } | |
70dec807 | 2931 | |
5c2b1fd7 | 2932 | *bio_ret = bio; |
d1310b2e CM |
2933 | |
2934 | return ret; | |
2935 | } | |
2936 | ||
48a3b636 ES |
2937 | static void attach_extent_buffer_page(struct extent_buffer *eb, |
2938 | struct page *page) | |
d1310b2e CM |
2939 | { |
2940 | if (!PagePrivate(page)) { | |
2941 | SetPagePrivate(page); | |
09cbfeaf | 2942 | get_page(page); |
4f2de97a JB |
2943 | set_page_private(page, (unsigned long)eb); |
2944 | } else { | |
2945 | WARN_ON(page->private != (unsigned long)eb); | |
d1310b2e CM |
2946 | } |
2947 | } | |
2948 | ||
4f2de97a | 2949 | void set_page_extent_mapped(struct page *page) |
d1310b2e | 2950 | { |
4f2de97a JB |
2951 | if (!PagePrivate(page)) { |
2952 | SetPagePrivate(page); | |
09cbfeaf | 2953 | get_page(page); |
4f2de97a JB |
2954 | set_page_private(page, EXTENT_PAGE_PRIVATE); |
2955 | } | |
d1310b2e CM |
2956 | } |
2957 | ||
125bac01 MX |
2958 | static struct extent_map * |
2959 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
2960 | u64 start, u64 len, get_extent_t *get_extent, | |
2961 | struct extent_map **em_cached) | |
2962 | { | |
2963 | struct extent_map *em; | |
2964 | ||
2965 | if (em_cached && *em_cached) { | |
2966 | em = *em_cached; | |
cbc0e928 | 2967 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 | 2968 | start < extent_map_end(em)) { |
490b54d6 | 2969 | refcount_inc(&em->refs); |
125bac01 MX |
2970 | return em; |
2971 | } | |
2972 | ||
2973 | free_extent_map(em); | |
2974 | *em_cached = NULL; | |
2975 | } | |
2976 | ||
fc4f21b1 | 2977 | em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0); |
125bac01 MX |
2978 | if (em_cached && !IS_ERR_OR_NULL(em)) { |
2979 | BUG_ON(*em_cached); | |
490b54d6 | 2980 | refcount_inc(&em->refs); |
125bac01 MX |
2981 | *em_cached = em; |
2982 | } | |
2983 | return em; | |
2984 | } | |
d1310b2e CM |
2985 | /* |
2986 | * basic readpage implementation. Locked extent state structs are inserted | |
2987 | * into the tree that are removed when the IO is done (by the end_io | |
2988 | * handlers) | |
79787eaa | 2989 | * XXX JDM: This needs looking at to ensure proper page locking |
baf863b9 | 2990 | * return 0 on success, otherwise return error |
d1310b2e | 2991 | */ |
9974090b MX |
2992 | static int __do_readpage(struct extent_io_tree *tree, |
2993 | struct page *page, | |
2994 | get_extent_t *get_extent, | |
125bac01 | 2995 | struct extent_map **em_cached, |
9974090b | 2996 | struct bio **bio, int mirror_num, |
f1c77c55 | 2997 | unsigned long *bio_flags, unsigned int read_flags, |
005efedf | 2998 | u64 *prev_em_start) |
d1310b2e CM |
2999 | { |
3000 | struct inode *inode = page->mapping->host; | |
4eee4fa4 | 3001 | u64 start = page_offset(page); |
8eec8296 | 3002 | const u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
3003 | u64 cur = start; |
3004 | u64 extent_offset; | |
3005 | u64 last_byte = i_size_read(inode); | |
3006 | u64 block_start; | |
3007 | u64 cur_end; | |
d1310b2e CM |
3008 | struct extent_map *em; |
3009 | struct block_device *bdev; | |
baf863b9 | 3010 | int ret = 0; |
d1310b2e | 3011 | int nr = 0; |
306e16ce | 3012 | size_t pg_offset = 0; |
d1310b2e | 3013 | size_t iosize; |
c8b97818 | 3014 | size_t disk_io_size; |
d1310b2e | 3015 | size_t blocksize = inode->i_sb->s_blocksize; |
7f042a83 | 3016 | unsigned long this_bio_flag = 0; |
d1310b2e CM |
3017 | |
3018 | set_page_extent_mapped(page); | |
3019 | ||
90a887c9 DM |
3020 | if (!PageUptodate(page)) { |
3021 | if (cleancache_get_page(page) == 0) { | |
3022 | BUG_ON(blocksize != PAGE_SIZE); | |
9974090b | 3023 | unlock_extent(tree, start, end); |
90a887c9 DM |
3024 | goto out; |
3025 | } | |
3026 | } | |
3027 | ||
09cbfeaf | 3028 | if (page->index == last_byte >> PAGE_SHIFT) { |
c8b97818 | 3029 | char *userpage; |
7073017a | 3030 | size_t zero_offset = offset_in_page(last_byte); |
c8b97818 CM |
3031 | |
3032 | if (zero_offset) { | |
09cbfeaf | 3033 | iosize = PAGE_SIZE - zero_offset; |
7ac687d9 | 3034 | userpage = kmap_atomic(page); |
c8b97818 CM |
3035 | memset(userpage + zero_offset, 0, iosize); |
3036 | flush_dcache_page(page); | |
7ac687d9 | 3037 | kunmap_atomic(userpage); |
c8b97818 CM |
3038 | } |
3039 | } | |
d1310b2e | 3040 | while (cur <= end) { |
005efedf | 3041 | bool force_bio_submit = false; |
6273b7f8 | 3042 | u64 offset; |
c8f2f24b | 3043 | |
d1310b2e CM |
3044 | if (cur >= last_byte) { |
3045 | char *userpage; | |
507903b8 AJ |
3046 | struct extent_state *cached = NULL; |
3047 | ||
09cbfeaf | 3048 | iosize = PAGE_SIZE - pg_offset; |
7ac687d9 | 3049 | userpage = kmap_atomic(page); |
306e16ce | 3050 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3051 | flush_dcache_page(page); |
7ac687d9 | 3052 | kunmap_atomic(userpage); |
d1310b2e | 3053 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 3054 | &cached, GFP_NOFS); |
7f042a83 | 3055 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3056 | cur + iosize - 1, &cached); |
d1310b2e CM |
3057 | break; |
3058 | } | |
125bac01 MX |
3059 | em = __get_extent_map(inode, page, pg_offset, cur, |
3060 | end - cur + 1, get_extent, em_cached); | |
c704005d | 3061 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3062 | SetPageError(page); |
7f042a83 | 3063 | unlock_extent(tree, cur, end); |
d1310b2e CM |
3064 | break; |
3065 | } | |
d1310b2e CM |
3066 | extent_offset = cur - em->start; |
3067 | BUG_ON(extent_map_end(em) <= cur); | |
3068 | BUG_ON(end < cur); | |
3069 | ||
261507a0 | 3070 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
4b384318 | 3071 | this_bio_flag |= EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
3072 | extent_set_compress_type(&this_bio_flag, |
3073 | em->compress_type); | |
3074 | } | |
c8b97818 | 3075 | |
d1310b2e CM |
3076 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
3077 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 3078 | iosize = ALIGN(iosize, blocksize); |
c8b97818 CM |
3079 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
3080 | disk_io_size = em->block_len; | |
6273b7f8 | 3081 | offset = em->block_start; |
c8b97818 | 3082 | } else { |
6273b7f8 | 3083 | offset = em->block_start + extent_offset; |
c8b97818 CM |
3084 | disk_io_size = iosize; |
3085 | } | |
d1310b2e CM |
3086 | bdev = em->bdev; |
3087 | block_start = em->block_start; | |
d899e052 YZ |
3088 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
3089 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
3090 | |
3091 | /* | |
3092 | * If we have a file range that points to a compressed extent | |
3093 | * and it's followed by a consecutive file range that points to | |
3094 | * to the same compressed extent (possibly with a different | |
3095 | * offset and/or length, so it either points to the whole extent | |
3096 | * or only part of it), we must make sure we do not submit a | |
3097 | * single bio to populate the pages for the 2 ranges because | |
3098 | * this makes the compressed extent read zero out the pages | |
3099 | * belonging to the 2nd range. Imagine the following scenario: | |
3100 | * | |
3101 | * File layout | |
3102 | * [0 - 8K] [8K - 24K] | |
3103 | * | | | |
3104 | * | | | |
3105 | * points to extent X, points to extent X, | |
3106 | * offset 4K, length of 8K offset 0, length 16K | |
3107 | * | |
3108 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
3109 | * | |
3110 | * If the bio to read the compressed extent covers both ranges, | |
3111 | * it will decompress extent X into the pages belonging to the | |
3112 | * first range and then it will stop, zeroing out the remaining | |
3113 | * pages that belong to the other range that points to extent X. | |
3114 | * So here we make sure we submit 2 bios, one for the first | |
3115 | * range and another one for the third range. Both will target | |
3116 | * the same physical extent from disk, but we can't currently | |
3117 | * make the compressed bio endio callback populate the pages | |
3118 | * for both ranges because each compressed bio is tightly | |
3119 | * coupled with a single extent map, and each range can have | |
3120 | * an extent map with a different offset value relative to the | |
3121 | * uncompressed data of our extent and different lengths. This | |
3122 | * is a corner case so we prioritize correctness over | |
3123 | * non-optimal behavior (submitting 2 bios for the same extent). | |
3124 | */ | |
3125 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
3126 | prev_em_start && *prev_em_start != (u64)-1 && | |
8e928218 | 3127 | *prev_em_start != em->start) |
005efedf FM |
3128 | force_bio_submit = true; |
3129 | ||
3130 | if (prev_em_start) | |
8e928218 | 3131 | *prev_em_start = em->start; |
005efedf | 3132 | |
d1310b2e CM |
3133 | free_extent_map(em); |
3134 | em = NULL; | |
3135 | ||
3136 | /* we've found a hole, just zero and go on */ | |
3137 | if (block_start == EXTENT_MAP_HOLE) { | |
3138 | char *userpage; | |
507903b8 AJ |
3139 | struct extent_state *cached = NULL; |
3140 | ||
7ac687d9 | 3141 | userpage = kmap_atomic(page); |
306e16ce | 3142 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3143 | flush_dcache_page(page); |
7ac687d9 | 3144 | kunmap_atomic(userpage); |
d1310b2e CM |
3145 | |
3146 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 | 3147 | &cached, GFP_NOFS); |
7f042a83 | 3148 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3149 | cur + iosize - 1, &cached); |
d1310b2e | 3150 | cur = cur + iosize; |
306e16ce | 3151 | pg_offset += iosize; |
d1310b2e CM |
3152 | continue; |
3153 | } | |
3154 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
3155 | if (test_range_bit(tree, cur, cur_end, |
3156 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 3157 | check_page_uptodate(tree, page); |
7f042a83 | 3158 | unlock_extent(tree, cur, cur + iosize - 1); |
d1310b2e | 3159 | cur = cur + iosize; |
306e16ce | 3160 | pg_offset += iosize; |
d1310b2e CM |
3161 | continue; |
3162 | } | |
70dec807 CM |
3163 | /* we have an inline extent but it didn't get marked up |
3164 | * to date. Error out | |
3165 | */ | |
3166 | if (block_start == EXTENT_MAP_INLINE) { | |
3167 | SetPageError(page); | |
7f042a83 | 3168 | unlock_extent(tree, cur, cur + iosize - 1); |
70dec807 | 3169 | cur = cur + iosize; |
306e16ce | 3170 | pg_offset += iosize; |
70dec807 CM |
3171 | continue; |
3172 | } | |
d1310b2e | 3173 | |
4b81ba48 | 3174 | ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL, |
6273b7f8 DS |
3175 | page, offset, disk_io_size, |
3176 | pg_offset, bdev, bio, | |
c8b97818 CM |
3177 | end_bio_extent_readpage, mirror_num, |
3178 | *bio_flags, | |
005efedf FM |
3179 | this_bio_flag, |
3180 | force_bio_submit); | |
c8f2f24b JB |
3181 | if (!ret) { |
3182 | nr++; | |
3183 | *bio_flags = this_bio_flag; | |
3184 | } else { | |
d1310b2e | 3185 | SetPageError(page); |
7f042a83 | 3186 | unlock_extent(tree, cur, cur + iosize - 1); |
baf863b9 | 3187 | goto out; |
edd33c99 | 3188 | } |
d1310b2e | 3189 | cur = cur + iosize; |
306e16ce | 3190 | pg_offset += iosize; |
d1310b2e | 3191 | } |
90a887c9 | 3192 | out: |
d1310b2e CM |
3193 | if (!nr) { |
3194 | if (!PageError(page)) | |
3195 | SetPageUptodate(page); | |
3196 | unlock_page(page); | |
3197 | } | |
baf863b9 | 3198 | return ret; |
d1310b2e CM |
3199 | } |
3200 | ||
e65ef21e | 3201 | static inline void contiguous_readpages(struct extent_io_tree *tree, |
9974090b MX |
3202 | struct page *pages[], int nr_pages, |
3203 | u64 start, u64 end, | |
125bac01 | 3204 | struct extent_map **em_cached, |
d3fac6ba | 3205 | struct bio **bio, |
1f7ad75b | 3206 | unsigned long *bio_flags, |
808f80b4 | 3207 | u64 *prev_em_start) |
9974090b MX |
3208 | { |
3209 | struct inode *inode; | |
3210 | struct btrfs_ordered_extent *ordered; | |
3211 | int index; | |
3212 | ||
3213 | inode = pages[0]->mapping->host; | |
3214 | while (1) { | |
3215 | lock_extent(tree, start, end); | |
a776c6fa | 3216 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
9974090b MX |
3217 | end - start + 1); |
3218 | if (!ordered) | |
3219 | break; | |
3220 | unlock_extent(tree, start, end); | |
3221 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3222 | btrfs_put_ordered_extent(ordered); | |
3223 | } | |
3224 | ||
3225 | for (index = 0; index < nr_pages; index++) { | |
4ef77695 | 3226 | __do_readpage(tree, pages[index], btrfs_get_extent, em_cached, |
5e9d3982 | 3227 | bio, 0, bio_flags, REQ_RAHEAD, prev_em_start); |
09cbfeaf | 3228 | put_page(pages[index]); |
9974090b MX |
3229 | } |
3230 | } | |
3231 | ||
9974090b MX |
3232 | static int __extent_read_full_page(struct extent_io_tree *tree, |
3233 | struct page *page, | |
3234 | get_extent_t *get_extent, | |
3235 | struct bio **bio, int mirror_num, | |
f1c77c55 DS |
3236 | unsigned long *bio_flags, |
3237 | unsigned int read_flags) | |
9974090b MX |
3238 | { |
3239 | struct inode *inode = page->mapping->host; | |
3240 | struct btrfs_ordered_extent *ordered; | |
3241 | u64 start = page_offset(page); | |
09cbfeaf | 3242 | u64 end = start + PAGE_SIZE - 1; |
9974090b MX |
3243 | int ret; |
3244 | ||
3245 | while (1) { | |
3246 | lock_extent(tree, start, end); | |
a776c6fa | 3247 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
09cbfeaf | 3248 | PAGE_SIZE); |
9974090b MX |
3249 | if (!ordered) |
3250 | break; | |
3251 | unlock_extent(tree, start, end); | |
3252 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3253 | btrfs_put_ordered_extent(ordered); | |
3254 | } | |
3255 | ||
125bac01 | 3256 | ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num, |
1f7ad75b | 3257 | bio_flags, read_flags, NULL); |
9974090b MX |
3258 | return ret; |
3259 | } | |
3260 | ||
d1310b2e | 3261 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, |
8ddc7d9c | 3262 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3263 | { |
3264 | struct bio *bio = NULL; | |
c8b97818 | 3265 | unsigned long bio_flags = 0; |
d1310b2e CM |
3266 | int ret; |
3267 | ||
8ddc7d9c | 3268 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, |
1f7ad75b | 3269 | &bio_flags, 0); |
d1310b2e | 3270 | if (bio) |
1f7ad75b | 3271 | ret = submit_one_bio(bio, mirror_num, bio_flags); |
d1310b2e CM |
3272 | return ret; |
3273 | } | |
d1310b2e | 3274 | |
3d4b9496 | 3275 | static void update_nr_written(struct writeback_control *wbc, |
a9132667 | 3276 | unsigned long nr_written) |
11c8349b CM |
3277 | { |
3278 | wbc->nr_to_write -= nr_written; | |
11c8349b CM |
3279 | } |
3280 | ||
d1310b2e | 3281 | /* |
40f76580 CM |
3282 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
3283 | * | |
5eaad97a | 3284 | * This returns 1 if btrfs_run_delalloc_range function did all the work required |
40f76580 CM |
3285 | * to write the page (copy into inline extent). In this case the IO has |
3286 | * been started and the page is already unlocked. | |
3287 | * | |
3288 | * This returns 0 if all went well (page still locked) | |
3289 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 3290 | */ |
40f76580 | 3291 | static noinline_for_stack int writepage_delalloc(struct inode *inode, |
8cc0237a NB |
3292 | struct page *page, struct writeback_control *wbc, |
3293 | u64 delalloc_start, unsigned long *nr_written) | |
40f76580 | 3294 | { |
8cc0237a | 3295 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
09cbfeaf | 3296 | u64 page_end = delalloc_start + PAGE_SIZE - 1; |
3522e903 | 3297 | bool found; |
40f76580 CM |
3298 | u64 delalloc_to_write = 0; |
3299 | u64 delalloc_end = 0; | |
3300 | int ret; | |
3301 | int page_started = 0; | |
3302 | ||
40f76580 CM |
3303 | |
3304 | while (delalloc_end < page_end) { | |
3522e903 | 3305 | found = find_lock_delalloc_range(inode, tree, |
40f76580 CM |
3306 | page, |
3307 | &delalloc_start, | |
917aacec | 3308 | &delalloc_end); |
3522e903 | 3309 | if (!found) { |
40f76580 CM |
3310 | delalloc_start = delalloc_end + 1; |
3311 | continue; | |
3312 | } | |
5eaad97a NB |
3313 | ret = btrfs_run_delalloc_range(inode, page, delalloc_start, |
3314 | delalloc_end, &page_started, nr_written, wbc); | |
40f76580 CM |
3315 | /* File system has been set read-only */ |
3316 | if (ret) { | |
3317 | SetPageError(page); | |
5eaad97a NB |
3318 | /* |
3319 | * btrfs_run_delalloc_range should return < 0 for error | |
3320 | * but just in case, we use > 0 here meaning the IO is | |
3321 | * started, so we don't want to return > 0 unless | |
3322 | * things are going well. | |
40f76580 CM |
3323 | */ |
3324 | ret = ret < 0 ? ret : -EIO; | |
3325 | goto done; | |
3326 | } | |
3327 | /* | |
ea1754a0 KS |
3328 | * delalloc_end is already one less than the total length, so |
3329 | * we don't subtract one from PAGE_SIZE | |
40f76580 CM |
3330 | */ |
3331 | delalloc_to_write += (delalloc_end - delalloc_start + | |
ea1754a0 | 3332 | PAGE_SIZE) >> PAGE_SHIFT; |
40f76580 CM |
3333 | delalloc_start = delalloc_end + 1; |
3334 | } | |
3335 | if (wbc->nr_to_write < delalloc_to_write) { | |
3336 | int thresh = 8192; | |
3337 | ||
3338 | if (delalloc_to_write < thresh * 2) | |
3339 | thresh = delalloc_to_write; | |
3340 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3341 | thresh); | |
3342 | } | |
3343 | ||
3344 | /* did the fill delalloc function already unlock and start | |
3345 | * the IO? | |
3346 | */ | |
3347 | if (page_started) { | |
3348 | /* | |
3349 | * we've unlocked the page, so we can't update | |
3350 | * the mapping's writeback index, just update | |
3351 | * nr_to_write. | |
3352 | */ | |
3353 | wbc->nr_to_write -= *nr_written; | |
3354 | return 1; | |
3355 | } | |
3356 | ||
3357 | ret = 0; | |
3358 | ||
3359 | done: | |
3360 | return ret; | |
3361 | } | |
3362 | ||
3363 | /* | |
3364 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3365 | * and does the loop to map the page into extents and bios. | |
3366 | * | |
3367 | * We return 1 if the IO is started and the page is unlocked, | |
3368 | * 0 if all went well (page still locked) | |
3369 | * < 0 if there were errors (page still locked) | |
3370 | */ | |
3371 | static noinline_for_stack int __extent_writepage_io(struct inode *inode, | |
3372 | struct page *page, | |
3373 | struct writeback_control *wbc, | |
3374 | struct extent_page_data *epd, | |
3375 | loff_t i_size, | |
3376 | unsigned long nr_written, | |
f1c77c55 | 3377 | unsigned int write_flags, int *nr_ret) |
d1310b2e | 3378 | { |
d1310b2e | 3379 | struct extent_io_tree *tree = epd->tree; |
4eee4fa4 | 3380 | u64 start = page_offset(page); |
09cbfeaf | 3381 | u64 page_end = start + PAGE_SIZE - 1; |
d1310b2e CM |
3382 | u64 end; |
3383 | u64 cur = start; | |
3384 | u64 extent_offset; | |
d1310b2e CM |
3385 | u64 block_start; |
3386 | u64 iosize; | |
d1310b2e CM |
3387 | struct extent_map *em; |
3388 | struct block_device *bdev; | |
7f3c74fb | 3389 | size_t pg_offset = 0; |
d1310b2e | 3390 | size_t blocksize; |
40f76580 CM |
3391 | int ret = 0; |
3392 | int nr = 0; | |
3393 | bool compressed; | |
c8b97818 | 3394 | |
d75855b4 NB |
3395 | ret = btrfs_writepage_cow_fixup(page, start, page_end); |
3396 | if (ret) { | |
3397 | /* Fixup worker will requeue */ | |
3398 | if (ret == -EBUSY) | |
3399 | wbc->pages_skipped++; | |
3400 | else | |
3401 | redirty_page_for_writepage(wbc, page); | |
40f76580 | 3402 | |
d75855b4 NB |
3403 | update_nr_written(wbc, nr_written); |
3404 | unlock_page(page); | |
3405 | return 1; | |
247e743c CM |
3406 | } |
3407 | ||
11c8349b CM |
3408 | /* |
3409 | * we don't want to touch the inode after unlocking the page, | |
3410 | * so we update the mapping writeback index now | |
3411 | */ | |
3d4b9496 | 3412 | update_nr_written(wbc, nr_written + 1); |
771ed689 | 3413 | |
d1310b2e | 3414 | end = page_end; |
40f76580 | 3415 | if (i_size <= start) { |
c629732d | 3416 | btrfs_writepage_endio_finish_ordered(page, start, page_end, 1); |
d1310b2e CM |
3417 | goto done; |
3418 | } | |
3419 | ||
d1310b2e CM |
3420 | blocksize = inode->i_sb->s_blocksize; |
3421 | ||
3422 | while (cur <= end) { | |
40f76580 | 3423 | u64 em_end; |
6273b7f8 | 3424 | u64 offset; |
58409edd | 3425 | |
40f76580 | 3426 | if (cur >= i_size) { |
7087a9d8 | 3427 | btrfs_writepage_endio_finish_ordered(page, cur, |
c629732d | 3428 | page_end, 1); |
d1310b2e CM |
3429 | break; |
3430 | } | |
3c98c62f | 3431 | em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, cur, |
d1310b2e | 3432 | end - cur + 1, 1); |
c704005d | 3433 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3434 | SetPageError(page); |
61391d56 | 3435 | ret = PTR_ERR_OR_ZERO(em); |
d1310b2e CM |
3436 | break; |
3437 | } | |
3438 | ||
3439 | extent_offset = cur - em->start; | |
40f76580 CM |
3440 | em_end = extent_map_end(em); |
3441 | BUG_ON(em_end <= cur); | |
d1310b2e | 3442 | BUG_ON(end < cur); |
40f76580 | 3443 | iosize = min(em_end - cur, end - cur + 1); |
fda2832f | 3444 | iosize = ALIGN(iosize, blocksize); |
6273b7f8 | 3445 | offset = em->block_start + extent_offset; |
d1310b2e CM |
3446 | bdev = em->bdev; |
3447 | block_start = em->block_start; | |
c8b97818 | 3448 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
3449 | free_extent_map(em); |
3450 | em = NULL; | |
3451 | ||
c8b97818 CM |
3452 | /* |
3453 | * compressed and inline extents are written through other | |
3454 | * paths in the FS | |
3455 | */ | |
3456 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3457 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
3458 | /* |
3459 | * end_io notification does not happen here for | |
3460 | * compressed extents | |
3461 | */ | |
7087a9d8 NB |
3462 | if (!compressed) |
3463 | btrfs_writepage_endio_finish_ordered(page, cur, | |
3464 | cur + iosize - 1, | |
c629732d | 3465 | 1); |
c8b97818 CM |
3466 | else if (compressed) { |
3467 | /* we don't want to end_page_writeback on | |
3468 | * a compressed extent. this happens | |
3469 | * elsewhere | |
3470 | */ | |
3471 | nr++; | |
3472 | } | |
3473 | ||
3474 | cur += iosize; | |
7f3c74fb | 3475 | pg_offset += iosize; |
d1310b2e CM |
3476 | continue; |
3477 | } | |
c8b97818 | 3478 | |
5cdc84bf | 3479 | btrfs_set_range_writeback(tree, cur, cur + iosize - 1); |
58409edd DS |
3480 | if (!PageWriteback(page)) { |
3481 | btrfs_err(BTRFS_I(inode)->root->fs_info, | |
3482 | "page %lu not writeback, cur %llu end %llu", | |
3483 | page->index, cur, end); | |
d1310b2e | 3484 | } |
7f3c74fb | 3485 | |
4b81ba48 | 3486 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc, |
6273b7f8 | 3487 | page, offset, iosize, pg_offset, |
c2df8bb4 | 3488 | bdev, &epd->bio, |
58409edd DS |
3489 | end_bio_extent_writepage, |
3490 | 0, 0, 0, false); | |
fe01aa65 | 3491 | if (ret) { |
58409edd | 3492 | SetPageError(page); |
fe01aa65 TK |
3493 | if (PageWriteback(page)) |
3494 | end_page_writeback(page); | |
3495 | } | |
d1310b2e | 3496 | |
d1310b2e | 3497 | cur = cur + iosize; |
7f3c74fb | 3498 | pg_offset += iosize; |
d1310b2e CM |
3499 | nr++; |
3500 | } | |
40f76580 CM |
3501 | done: |
3502 | *nr_ret = nr; | |
40f76580 CM |
3503 | return ret; |
3504 | } | |
3505 | ||
3506 | /* | |
3507 | * the writepage semantics are similar to regular writepage. extent | |
3508 | * records are inserted to lock ranges in the tree, and as dirty areas | |
3509 | * are found, they are marked writeback. Then the lock bits are removed | |
3510 | * and the end_io handler clears the writeback ranges | |
3065976b QW |
3511 | * |
3512 | * Return 0 if everything goes well. | |
3513 | * Return <0 for error. | |
40f76580 CM |
3514 | */ |
3515 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
aab6e9ed | 3516 | struct extent_page_data *epd) |
40f76580 CM |
3517 | { |
3518 | struct inode *inode = page->mapping->host; | |
40f76580 | 3519 | u64 start = page_offset(page); |
09cbfeaf | 3520 | u64 page_end = start + PAGE_SIZE - 1; |
40f76580 CM |
3521 | int ret; |
3522 | int nr = 0; | |
3523 | size_t pg_offset = 0; | |
3524 | loff_t i_size = i_size_read(inode); | |
09cbfeaf | 3525 | unsigned long end_index = i_size >> PAGE_SHIFT; |
f1c77c55 | 3526 | unsigned int write_flags = 0; |
40f76580 CM |
3527 | unsigned long nr_written = 0; |
3528 | ||
ff40adf7 | 3529 | write_flags = wbc_to_write_flags(wbc); |
40f76580 CM |
3530 | |
3531 | trace___extent_writepage(page, inode, wbc); | |
3532 | ||
3533 | WARN_ON(!PageLocked(page)); | |
3534 | ||
3535 | ClearPageError(page); | |
3536 | ||
7073017a | 3537 | pg_offset = offset_in_page(i_size); |
40f76580 CM |
3538 | if (page->index > end_index || |
3539 | (page->index == end_index && !pg_offset)) { | |
09cbfeaf | 3540 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); |
40f76580 CM |
3541 | unlock_page(page); |
3542 | return 0; | |
3543 | } | |
3544 | ||
3545 | if (page->index == end_index) { | |
3546 | char *userpage; | |
3547 | ||
3548 | userpage = kmap_atomic(page); | |
3549 | memset(userpage + pg_offset, 0, | |
09cbfeaf | 3550 | PAGE_SIZE - pg_offset); |
40f76580 CM |
3551 | kunmap_atomic(userpage); |
3552 | flush_dcache_page(page); | |
3553 | } | |
3554 | ||
3555 | pg_offset = 0; | |
3556 | ||
3557 | set_page_extent_mapped(page); | |
3558 | ||
7789a55a | 3559 | if (!epd->extent_locked) { |
8cc0237a | 3560 | ret = writepage_delalloc(inode, page, wbc, start, &nr_written); |
7789a55a NB |
3561 | if (ret == 1) |
3562 | goto done_unlocked; | |
3563 | if (ret) | |
3564 | goto done; | |
3565 | } | |
40f76580 CM |
3566 | |
3567 | ret = __extent_writepage_io(inode, page, wbc, epd, | |
3568 | i_size, nr_written, write_flags, &nr); | |
3569 | if (ret == 1) | |
3570 | goto done_unlocked; | |
3571 | ||
d1310b2e CM |
3572 | done: |
3573 | if (nr == 0) { | |
3574 | /* make sure the mapping tag for page dirty gets cleared */ | |
3575 | set_page_writeback(page); | |
3576 | end_page_writeback(page); | |
3577 | } | |
61391d56 FM |
3578 | if (PageError(page)) { |
3579 | ret = ret < 0 ? ret : -EIO; | |
3580 | end_extent_writepage(page, ret, start, page_end); | |
3581 | } | |
d1310b2e | 3582 | unlock_page(page); |
3065976b | 3583 | ASSERT(ret <= 0); |
40f76580 | 3584 | return ret; |
771ed689 | 3585 | |
11c8349b | 3586 | done_unlocked: |
d1310b2e CM |
3587 | return 0; |
3588 | } | |
3589 | ||
fd8b2b61 | 3590 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 3591 | { |
74316201 N |
3592 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
3593 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
3594 | } |
3595 | ||
2e3c2513 QW |
3596 | /* |
3597 | * Lock eb pages and flush the bio if we can't the locks | |
3598 | * | |
3599 | * Return 0 if nothing went wrong | |
3600 | * Return >0 is same as 0, except bio is not submitted | |
3601 | * Return <0 if something went wrong, no page is locked | |
3602 | */ | |
9df76fb5 | 3603 | static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb, |
0e378df1 | 3604 | struct extent_page_data *epd) |
0b32f4bb | 3605 | { |
9df76fb5 | 3606 | struct btrfs_fs_info *fs_info = eb->fs_info; |
2e3c2513 | 3607 | int i, num_pages, failed_page_nr; |
0b32f4bb JB |
3608 | int flush = 0; |
3609 | int ret = 0; | |
3610 | ||
3611 | if (!btrfs_try_tree_write_lock(eb)) { | |
f4340622 | 3612 | ret = flush_write_bio(epd); |
2e3c2513 QW |
3613 | if (ret < 0) |
3614 | return ret; | |
3615 | flush = 1; | |
0b32f4bb JB |
3616 | btrfs_tree_lock(eb); |
3617 | } | |
3618 | ||
3619 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3620 | btrfs_tree_unlock(eb); | |
3621 | if (!epd->sync_io) | |
3622 | return 0; | |
3623 | if (!flush) { | |
f4340622 | 3624 | ret = flush_write_bio(epd); |
2e3c2513 QW |
3625 | if (ret < 0) |
3626 | return ret; | |
0b32f4bb JB |
3627 | flush = 1; |
3628 | } | |
a098d8e8 CM |
3629 | while (1) { |
3630 | wait_on_extent_buffer_writeback(eb); | |
3631 | btrfs_tree_lock(eb); | |
3632 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3633 | break; | |
0b32f4bb | 3634 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3635 | } |
3636 | } | |
3637 | ||
51561ffe JB |
3638 | /* |
3639 | * We need to do this to prevent races in people who check if the eb is | |
3640 | * under IO since we can end up having no IO bits set for a short period | |
3641 | * of time. | |
3642 | */ | |
3643 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
3644 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3645 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 3646 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 3647 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
104b4e51 NB |
3648 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
3649 | -eb->len, | |
3650 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 3651 | ret = 1; |
51561ffe JB |
3652 | } else { |
3653 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
3654 | } |
3655 | ||
3656 | btrfs_tree_unlock(eb); | |
3657 | ||
3658 | if (!ret) | |
3659 | return ret; | |
3660 | ||
65ad0104 | 3661 | num_pages = num_extent_pages(eb); |
0b32f4bb | 3662 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3663 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3664 | |
3665 | if (!trylock_page(p)) { | |
3666 | if (!flush) { | |
f4340622 | 3667 | ret = flush_write_bio(epd); |
2e3c2513 QW |
3668 | if (ret < 0) { |
3669 | failed_page_nr = i; | |
3670 | goto err_unlock; | |
3671 | } | |
0b32f4bb JB |
3672 | flush = 1; |
3673 | } | |
3674 | lock_page(p); | |
3675 | } | |
3676 | } | |
3677 | ||
3678 | return ret; | |
2e3c2513 QW |
3679 | err_unlock: |
3680 | /* Unlock already locked pages */ | |
3681 | for (i = 0; i < failed_page_nr; i++) | |
3682 | unlock_page(eb->pages[i]); | |
3683 | return ret; | |
0b32f4bb JB |
3684 | } |
3685 | ||
3686 | static void end_extent_buffer_writeback(struct extent_buffer *eb) | |
3687 | { | |
3688 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
4e857c58 | 3689 | smp_mb__after_atomic(); |
0b32f4bb JB |
3690 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); |
3691 | } | |
3692 | ||
656f30db FM |
3693 | static void set_btree_ioerr(struct page *page) |
3694 | { | |
3695 | struct extent_buffer *eb = (struct extent_buffer *)page->private; | |
656f30db FM |
3696 | |
3697 | SetPageError(page); | |
3698 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) | |
3699 | return; | |
3700 | ||
3701 | /* | |
3702 | * If writeback for a btree extent that doesn't belong to a log tree | |
3703 | * failed, increment the counter transaction->eb_write_errors. | |
3704 | * We do this because while the transaction is running and before it's | |
3705 | * committing (when we call filemap_fdata[write|wait]_range against | |
3706 | * the btree inode), we might have | |
3707 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
3708 | * returns an error or an error happens during writeback, when we're | |
3709 | * committing the transaction we wouldn't know about it, since the pages | |
3710 | * can be no longer dirty nor marked anymore for writeback (if a | |
3711 | * subsequent modification to the extent buffer didn't happen before the | |
3712 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
3713 | * able to find the pages tagged with SetPageError at transaction | |
3714 | * commit time. So if this happens we must abort the transaction, | |
3715 | * otherwise we commit a super block with btree roots that point to | |
3716 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
3717 | * or the content of some node/leaf from a past generation that got | |
3718 | * cowed or deleted and is no longer valid. | |
3719 | * | |
3720 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
3721 | * not be enough - we need to distinguish between log tree extents vs | |
3722 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
3723 | * will catch and clear such errors in the mapping - and that call might | |
3724 | * be from a log sync and not from a transaction commit. Also, checking | |
3725 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
3726 | * not done and would not be reliable - the eb might have been released | |
3727 | * from memory and reading it back again means that flag would not be | |
3728 | * set (since it's a runtime flag, not persisted on disk). | |
3729 | * | |
3730 | * Using the flags below in the btree inode also makes us achieve the | |
3731 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
3732 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
3733 | * is called, the writeback for all dirty pages had already finished | |
3734 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
3735 | * filemap_fdatawait_range() would return success, as it could not know | |
3736 | * that writeback errors happened (the pages were no longer tagged for | |
3737 | * writeback). | |
3738 | */ | |
3739 | switch (eb->log_index) { | |
3740 | case -1: | |
afcdd129 | 3741 | set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags); |
656f30db FM |
3742 | break; |
3743 | case 0: | |
afcdd129 | 3744 | set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags); |
656f30db FM |
3745 | break; |
3746 | case 1: | |
afcdd129 | 3747 | set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags); |
656f30db FM |
3748 | break; |
3749 | default: | |
3750 | BUG(); /* unexpected, logic error */ | |
3751 | } | |
3752 | } | |
3753 | ||
4246a0b6 | 3754 | static void end_bio_extent_buffer_writepage(struct bio *bio) |
0b32f4bb | 3755 | { |
2c30c71b | 3756 | struct bio_vec *bvec; |
0b32f4bb | 3757 | struct extent_buffer *eb; |
2c30c71b | 3758 | int i, done; |
6dc4f100 | 3759 | struct bvec_iter_all iter_all; |
0b32f4bb | 3760 | |
c09abff8 | 3761 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
6dc4f100 | 3762 | bio_for_each_segment_all(bvec, bio, i, iter_all) { |
0b32f4bb JB |
3763 | struct page *page = bvec->bv_page; |
3764 | ||
0b32f4bb JB |
3765 | eb = (struct extent_buffer *)page->private; |
3766 | BUG_ON(!eb); | |
3767 | done = atomic_dec_and_test(&eb->io_pages); | |
3768 | ||
4e4cbee9 | 3769 | if (bio->bi_status || |
4246a0b6 | 3770 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { |
0b32f4bb | 3771 | ClearPageUptodate(page); |
656f30db | 3772 | set_btree_ioerr(page); |
0b32f4bb JB |
3773 | } |
3774 | ||
3775 | end_page_writeback(page); | |
3776 | ||
3777 | if (!done) | |
3778 | continue; | |
3779 | ||
3780 | end_extent_buffer_writeback(eb); | |
2c30c71b | 3781 | } |
0b32f4bb JB |
3782 | |
3783 | bio_put(bio); | |
0b32f4bb JB |
3784 | } |
3785 | ||
0e378df1 | 3786 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb JB |
3787 | struct writeback_control *wbc, |
3788 | struct extent_page_data *epd) | |
3789 | { | |
0ab02063 | 3790 | struct btrfs_fs_info *fs_info = eb->fs_info; |
0b32f4bb | 3791 | struct block_device *bdev = fs_info->fs_devices->latest_bdev; |
f28491e0 | 3792 | struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree; |
0b32f4bb | 3793 | u64 offset = eb->start; |
851cd173 | 3794 | u32 nritems; |
cc5e31a4 | 3795 | int i, num_pages; |
851cd173 | 3796 | unsigned long start, end; |
ff40adf7 | 3797 | unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META; |
d7dbe9e7 | 3798 | int ret = 0; |
0b32f4bb | 3799 | |
656f30db | 3800 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); |
65ad0104 | 3801 | num_pages = num_extent_pages(eb); |
0b32f4bb | 3802 | atomic_set(&eb->io_pages, num_pages); |
de0022b9 | 3803 | |
851cd173 LB |
3804 | /* set btree blocks beyond nritems with 0 to avoid stale content. */ |
3805 | nritems = btrfs_header_nritems(eb); | |
3eb548ee | 3806 | if (btrfs_header_level(eb) > 0) { |
3eb548ee LB |
3807 | end = btrfs_node_key_ptr_offset(nritems); |
3808 | ||
b159fa28 | 3809 | memzero_extent_buffer(eb, end, eb->len - end); |
851cd173 LB |
3810 | } else { |
3811 | /* | |
3812 | * leaf: | |
3813 | * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0 | |
3814 | */ | |
3815 | start = btrfs_item_nr_offset(nritems); | |
8f881e8c | 3816 | end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(eb); |
b159fa28 | 3817 | memzero_extent_buffer(eb, start, end - start); |
3eb548ee LB |
3818 | } |
3819 | ||
0b32f4bb | 3820 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3821 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3822 | |
3823 | clear_page_dirty_for_io(p); | |
3824 | set_page_writeback(p); | |
4b81ba48 | 3825 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc, |
6273b7f8 | 3826 | p, offset, PAGE_SIZE, 0, bdev, |
c2df8bb4 | 3827 | &epd->bio, |
1f7ad75b | 3828 | end_bio_extent_buffer_writepage, |
18fdc679 | 3829 | 0, 0, 0, false); |
0b32f4bb | 3830 | if (ret) { |
656f30db | 3831 | set_btree_ioerr(p); |
fe01aa65 TK |
3832 | if (PageWriteback(p)) |
3833 | end_page_writeback(p); | |
0b32f4bb JB |
3834 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
3835 | end_extent_buffer_writeback(eb); | |
3836 | ret = -EIO; | |
3837 | break; | |
3838 | } | |
09cbfeaf | 3839 | offset += PAGE_SIZE; |
3d4b9496 | 3840 | update_nr_written(wbc, 1); |
0b32f4bb JB |
3841 | unlock_page(p); |
3842 | } | |
3843 | ||
3844 | if (unlikely(ret)) { | |
3845 | for (; i < num_pages; i++) { | |
bbf65cf0 | 3846 | struct page *p = eb->pages[i]; |
81465028 | 3847 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
3848 | unlock_page(p); |
3849 | } | |
3850 | } | |
3851 | ||
3852 | return ret; | |
3853 | } | |
3854 | ||
3855 | int btree_write_cache_pages(struct address_space *mapping, | |
3856 | struct writeback_control *wbc) | |
3857 | { | |
3858 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; | |
0b32f4bb JB |
3859 | struct extent_buffer *eb, *prev_eb = NULL; |
3860 | struct extent_page_data epd = { | |
3861 | .bio = NULL, | |
3862 | .tree = tree, | |
3863 | .extent_locked = 0, | |
3864 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
3865 | }; | |
3866 | int ret = 0; | |
3867 | int done = 0; | |
3868 | int nr_to_write_done = 0; | |
3869 | struct pagevec pvec; | |
3870 | int nr_pages; | |
3871 | pgoff_t index; | |
3872 | pgoff_t end; /* Inclusive */ | |
3873 | int scanned = 0; | |
10bbd235 | 3874 | xa_mark_t tag; |
0b32f4bb | 3875 | |
86679820 | 3876 | pagevec_init(&pvec); |
0b32f4bb JB |
3877 | if (wbc->range_cyclic) { |
3878 | index = mapping->writeback_index; /* Start from prev offset */ | |
3879 | end = -1; | |
3880 | } else { | |
09cbfeaf KS |
3881 | index = wbc->range_start >> PAGE_SHIFT; |
3882 | end = wbc->range_end >> PAGE_SHIFT; | |
0b32f4bb JB |
3883 | scanned = 1; |
3884 | } | |
3885 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3886 | tag = PAGECACHE_TAG_TOWRITE; | |
3887 | else | |
3888 | tag = PAGECACHE_TAG_DIRTY; | |
3889 | retry: | |
3890 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3891 | tag_pages_for_writeback(mapping, index, end); | |
3892 | while (!done && !nr_to_write_done && (index <= end) && | |
4006f437 | 3893 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 3894 | tag))) { |
0b32f4bb JB |
3895 | unsigned i; |
3896 | ||
3897 | scanned = 1; | |
3898 | for (i = 0; i < nr_pages; i++) { | |
3899 | struct page *page = pvec.pages[i]; | |
3900 | ||
3901 | if (!PagePrivate(page)) | |
3902 | continue; | |
3903 | ||
b5bae261 JB |
3904 | spin_lock(&mapping->private_lock); |
3905 | if (!PagePrivate(page)) { | |
3906 | spin_unlock(&mapping->private_lock); | |
3907 | continue; | |
3908 | } | |
3909 | ||
0b32f4bb | 3910 | eb = (struct extent_buffer *)page->private; |
b5bae261 JB |
3911 | |
3912 | /* | |
3913 | * Shouldn't happen and normally this would be a BUG_ON | |
3914 | * but no sense in crashing the users box for something | |
3915 | * we can survive anyway. | |
3916 | */ | |
fae7f21c | 3917 | if (WARN_ON(!eb)) { |
b5bae261 | 3918 | spin_unlock(&mapping->private_lock); |
0b32f4bb JB |
3919 | continue; |
3920 | } | |
3921 | ||
b5bae261 JB |
3922 | if (eb == prev_eb) { |
3923 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 3924 | continue; |
b5bae261 | 3925 | } |
0b32f4bb | 3926 | |
b5bae261 JB |
3927 | ret = atomic_inc_not_zero(&eb->refs); |
3928 | spin_unlock(&mapping->private_lock); | |
3929 | if (!ret) | |
0b32f4bb | 3930 | continue; |
0b32f4bb JB |
3931 | |
3932 | prev_eb = eb; | |
9df76fb5 | 3933 | ret = lock_extent_buffer_for_io(eb, &epd); |
0b32f4bb JB |
3934 | if (!ret) { |
3935 | free_extent_buffer(eb); | |
3936 | continue; | |
3937 | } | |
3938 | ||
0ab02063 | 3939 | ret = write_one_eb(eb, wbc, &epd); |
0b32f4bb JB |
3940 | if (ret) { |
3941 | done = 1; | |
3942 | free_extent_buffer(eb); | |
3943 | break; | |
3944 | } | |
3945 | free_extent_buffer(eb); | |
3946 | ||
3947 | /* | |
3948 | * the filesystem may choose to bump up nr_to_write. | |
3949 | * We have to make sure to honor the new nr_to_write | |
3950 | * at any time | |
3951 | */ | |
3952 | nr_to_write_done = wbc->nr_to_write <= 0; | |
3953 | } | |
3954 | pagevec_release(&pvec); | |
3955 | cond_resched(); | |
3956 | } | |
3957 | if (!scanned && !done) { | |
3958 | /* | |
3959 | * We hit the last page and there is more work to be done: wrap | |
3960 | * back to the start of the file | |
3961 | */ | |
3962 | scanned = 1; | |
3963 | index = 0; | |
3964 | goto retry; | |
3965 | } | |
2b952eea QW |
3966 | ASSERT(ret <= 0); |
3967 | if (ret < 0) { | |
3968 | end_write_bio(&epd, ret); | |
3969 | return ret; | |
3970 | } | |
3971 | ret = flush_write_bio(&epd); | |
0b32f4bb JB |
3972 | return ret; |
3973 | } | |
3974 | ||
d1310b2e | 3975 | /** |
4bef0848 | 3976 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
3977 | * @mapping: address space structure to write |
3978 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
935db853 | 3979 | * @data: data passed to __extent_writepage function |
d1310b2e CM |
3980 | * |
3981 | * If a page is already under I/O, write_cache_pages() skips it, even | |
3982 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
3983 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
3984 | * and msync() need to guarantee that all the data which was dirty at the time | |
3985 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
3986 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
3987 | * existing IO to complete. | |
3988 | */ | |
4242b64a | 3989 | static int extent_write_cache_pages(struct address_space *mapping, |
4bef0848 | 3990 | struct writeback_control *wbc, |
aab6e9ed | 3991 | struct extent_page_data *epd) |
d1310b2e | 3992 | { |
7fd1a3f7 | 3993 | struct inode *inode = mapping->host; |
d1310b2e CM |
3994 | int ret = 0; |
3995 | int done = 0; | |
f85d7d6c | 3996 | int nr_to_write_done = 0; |
d1310b2e CM |
3997 | struct pagevec pvec; |
3998 | int nr_pages; | |
3999 | pgoff_t index; | |
4000 | pgoff_t end; /* Inclusive */ | |
a9132667 LB |
4001 | pgoff_t done_index; |
4002 | int range_whole = 0; | |
d1310b2e | 4003 | int scanned = 0; |
10bbd235 | 4004 | xa_mark_t tag; |
d1310b2e | 4005 | |
7fd1a3f7 JB |
4006 | /* |
4007 | * We have to hold onto the inode so that ordered extents can do their | |
4008 | * work when the IO finishes. The alternative to this is failing to add | |
4009 | * an ordered extent if the igrab() fails there and that is a huge pain | |
4010 | * to deal with, so instead just hold onto the inode throughout the | |
4011 | * writepages operation. If it fails here we are freeing up the inode | |
4012 | * anyway and we'd rather not waste our time writing out stuff that is | |
4013 | * going to be truncated anyway. | |
4014 | */ | |
4015 | if (!igrab(inode)) | |
4016 | return 0; | |
4017 | ||
86679820 | 4018 | pagevec_init(&pvec); |
d1310b2e CM |
4019 | if (wbc->range_cyclic) { |
4020 | index = mapping->writeback_index; /* Start from prev offset */ | |
4021 | end = -1; | |
4022 | } else { | |
09cbfeaf KS |
4023 | index = wbc->range_start >> PAGE_SHIFT; |
4024 | end = wbc->range_end >> PAGE_SHIFT; | |
a9132667 LB |
4025 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
4026 | range_whole = 1; | |
d1310b2e CM |
4027 | scanned = 1; |
4028 | } | |
3cd24c69 EL |
4029 | |
4030 | /* | |
4031 | * We do the tagged writepage as long as the snapshot flush bit is set | |
4032 | * and we are the first one who do the filemap_flush() on this inode. | |
4033 | * | |
4034 | * The nr_to_write == LONG_MAX is needed to make sure other flushers do | |
4035 | * not race in and drop the bit. | |
4036 | */ | |
4037 | if (range_whole && wbc->nr_to_write == LONG_MAX && | |
4038 | test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
4039 | &BTRFS_I(inode)->runtime_flags)) | |
4040 | wbc->tagged_writepages = 1; | |
4041 | ||
4042 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) | |
f7aaa06b JB |
4043 | tag = PAGECACHE_TAG_TOWRITE; |
4044 | else | |
4045 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 4046 | retry: |
3cd24c69 | 4047 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
f7aaa06b | 4048 | tag_pages_for_writeback(mapping, index, end); |
a9132667 | 4049 | done_index = index; |
f85d7d6c | 4050 | while (!done && !nr_to_write_done && (index <= end) && |
67fd707f JK |
4051 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, |
4052 | &index, end, tag))) { | |
d1310b2e CM |
4053 | unsigned i; |
4054 | ||
4055 | scanned = 1; | |
4056 | for (i = 0; i < nr_pages; i++) { | |
4057 | struct page *page = pvec.pages[i]; | |
4058 | ||
a9132667 | 4059 | done_index = page->index; |
d1310b2e | 4060 | /* |
b93b0163 MW |
4061 | * At this point we hold neither the i_pages lock nor |
4062 | * the page lock: the page may be truncated or | |
4063 | * invalidated (changing page->mapping to NULL), | |
4064 | * or even swizzled back from swapper_space to | |
4065 | * tmpfs file mapping | |
d1310b2e | 4066 | */ |
c8f2f24b | 4067 | if (!trylock_page(page)) { |
f4340622 QW |
4068 | ret = flush_write_bio(epd); |
4069 | BUG_ON(ret < 0); | |
c8f2f24b | 4070 | lock_page(page); |
01d658f2 | 4071 | } |
d1310b2e CM |
4072 | |
4073 | if (unlikely(page->mapping != mapping)) { | |
4074 | unlock_page(page); | |
4075 | continue; | |
4076 | } | |
4077 | ||
d2c3f4f6 | 4078 | if (wbc->sync_mode != WB_SYNC_NONE) { |
f4340622 QW |
4079 | if (PageWriteback(page)) { |
4080 | ret = flush_write_bio(epd); | |
4081 | BUG_ON(ret < 0); | |
4082 | } | |
d1310b2e | 4083 | wait_on_page_writeback(page); |
d2c3f4f6 | 4084 | } |
d1310b2e CM |
4085 | |
4086 | if (PageWriteback(page) || | |
4087 | !clear_page_dirty_for_io(page)) { | |
4088 | unlock_page(page); | |
4089 | continue; | |
4090 | } | |
4091 | ||
aab6e9ed | 4092 | ret = __extent_writepage(page, wbc, epd); |
a9132667 LB |
4093 | if (ret < 0) { |
4094 | /* | |
4095 | * done_index is set past this page, | |
4096 | * so media errors will not choke | |
4097 | * background writeout for the entire | |
4098 | * file. This has consequences for | |
4099 | * range_cyclic semantics (ie. it may | |
4100 | * not be suitable for data integrity | |
4101 | * writeout). | |
4102 | */ | |
4103 | done_index = page->index + 1; | |
4104 | done = 1; | |
4105 | break; | |
4106 | } | |
f85d7d6c CM |
4107 | |
4108 | /* | |
4109 | * the filesystem may choose to bump up nr_to_write. | |
4110 | * We have to make sure to honor the new nr_to_write | |
4111 | * at any time | |
4112 | */ | |
4113 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
4114 | } |
4115 | pagevec_release(&pvec); | |
4116 | cond_resched(); | |
4117 | } | |
894b36e3 | 4118 | if (!scanned && !done) { |
d1310b2e CM |
4119 | /* |
4120 | * We hit the last page and there is more work to be done: wrap | |
4121 | * back to the start of the file | |
4122 | */ | |
4123 | scanned = 1; | |
4124 | index = 0; | |
4125 | goto retry; | |
4126 | } | |
a9132667 LB |
4127 | |
4128 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole)) | |
4129 | mapping->writeback_index = done_index; | |
4130 | ||
7fd1a3f7 | 4131 | btrfs_add_delayed_iput(inode); |
894b36e3 | 4132 | return ret; |
d1310b2e | 4133 | } |
d1310b2e | 4134 | |
0a9b0e53 | 4135 | int extent_write_full_page(struct page *page, struct writeback_control *wbc) |
d1310b2e CM |
4136 | { |
4137 | int ret; | |
d1310b2e CM |
4138 | struct extent_page_data epd = { |
4139 | .bio = NULL, | |
0a9b0e53 | 4140 | .tree = &BTRFS_I(page->mapping->host)->io_tree, |
771ed689 | 4141 | .extent_locked = 0, |
ffbd517d | 4142 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e | 4143 | }; |
d1310b2e | 4144 | |
d1310b2e | 4145 | ret = __extent_writepage(page, wbc, &epd); |
3065976b QW |
4146 | ASSERT(ret <= 0); |
4147 | if (ret < 0) { | |
4148 | end_write_bio(&epd, ret); | |
4149 | return ret; | |
4150 | } | |
d1310b2e | 4151 | |
3065976b QW |
4152 | ret = flush_write_bio(&epd); |
4153 | ASSERT(ret <= 0); | |
d1310b2e CM |
4154 | return ret; |
4155 | } | |
d1310b2e | 4156 | |
5e3ee236 | 4157 | int extent_write_locked_range(struct inode *inode, u64 start, u64 end, |
771ed689 CM |
4158 | int mode) |
4159 | { | |
4160 | int ret = 0; | |
4161 | struct address_space *mapping = inode->i_mapping; | |
5e3ee236 | 4162 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
771ed689 | 4163 | struct page *page; |
09cbfeaf KS |
4164 | unsigned long nr_pages = (end - start + PAGE_SIZE) >> |
4165 | PAGE_SHIFT; | |
771ed689 CM |
4166 | |
4167 | struct extent_page_data epd = { | |
4168 | .bio = NULL, | |
4169 | .tree = tree, | |
771ed689 | 4170 | .extent_locked = 1, |
ffbd517d | 4171 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
4172 | }; |
4173 | struct writeback_control wbc_writepages = { | |
771ed689 | 4174 | .sync_mode = mode, |
771ed689 CM |
4175 | .nr_to_write = nr_pages * 2, |
4176 | .range_start = start, | |
4177 | .range_end = end + 1, | |
4178 | }; | |
4179 | ||
d397712b | 4180 | while (start <= end) { |
09cbfeaf | 4181 | page = find_get_page(mapping, start >> PAGE_SHIFT); |
771ed689 CM |
4182 | if (clear_page_dirty_for_io(page)) |
4183 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
4184 | else { | |
7087a9d8 | 4185 | btrfs_writepage_endio_finish_ordered(page, start, |
c629732d | 4186 | start + PAGE_SIZE - 1, 1); |
771ed689 CM |
4187 | unlock_page(page); |
4188 | } | |
09cbfeaf KS |
4189 | put_page(page); |
4190 | start += PAGE_SIZE; | |
771ed689 CM |
4191 | } |
4192 | ||
02c6db4f QW |
4193 | ASSERT(ret <= 0); |
4194 | if (ret < 0) { | |
4195 | end_write_bio(&epd, ret); | |
4196 | return ret; | |
4197 | } | |
4198 | ret = flush_write_bio(&epd); | |
771ed689 CM |
4199 | return ret; |
4200 | } | |
d1310b2e | 4201 | |
8ae225a8 | 4202 | int extent_writepages(struct address_space *mapping, |
d1310b2e CM |
4203 | struct writeback_control *wbc) |
4204 | { | |
4205 | int ret = 0; | |
4206 | struct extent_page_data epd = { | |
4207 | .bio = NULL, | |
8ae225a8 | 4208 | .tree = &BTRFS_I(mapping->host)->io_tree, |
771ed689 | 4209 | .extent_locked = 0, |
ffbd517d | 4210 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
4211 | }; |
4212 | ||
935db853 | 4213 | ret = extent_write_cache_pages(mapping, wbc, &epd); |
a2a72fbd QW |
4214 | ASSERT(ret <= 0); |
4215 | if (ret < 0) { | |
4216 | end_write_bio(&epd, ret); | |
4217 | return ret; | |
4218 | } | |
4219 | ret = flush_write_bio(&epd); | |
d1310b2e CM |
4220 | return ret; |
4221 | } | |
d1310b2e | 4222 | |
2a3ff0ad NB |
4223 | int extent_readpages(struct address_space *mapping, struct list_head *pages, |
4224 | unsigned nr_pages) | |
d1310b2e CM |
4225 | { |
4226 | struct bio *bio = NULL; | |
c8b97818 | 4227 | unsigned long bio_flags = 0; |
67c9684f | 4228 | struct page *pagepool[16]; |
125bac01 | 4229 | struct extent_map *em_cached = NULL; |
2a3ff0ad | 4230 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; |
67c9684f | 4231 | int nr = 0; |
808f80b4 | 4232 | u64 prev_em_start = (u64)-1; |
d1310b2e | 4233 | |
61ed3a14 | 4234 | while (!list_empty(pages)) { |
e65ef21e NB |
4235 | u64 contig_end = 0; |
4236 | ||
61ed3a14 | 4237 | for (nr = 0; nr < ARRAY_SIZE(pagepool) && !list_empty(pages);) { |
f86196ea | 4238 | struct page *page = lru_to_page(pages); |
d1310b2e | 4239 | |
61ed3a14 NB |
4240 | prefetchw(&page->flags); |
4241 | list_del(&page->lru); | |
4242 | if (add_to_page_cache_lru(page, mapping, page->index, | |
4243 | readahead_gfp_mask(mapping))) { | |
4244 | put_page(page); | |
e65ef21e | 4245 | break; |
61ed3a14 NB |
4246 | } |
4247 | ||
4248 | pagepool[nr++] = page; | |
e65ef21e | 4249 | contig_end = page_offset(page) + PAGE_SIZE - 1; |
d1310b2e | 4250 | } |
67c9684f | 4251 | |
e65ef21e NB |
4252 | if (nr) { |
4253 | u64 contig_start = page_offset(pagepool[0]); | |
4254 | ||
4255 | ASSERT(contig_start + nr * PAGE_SIZE - 1 == contig_end); | |
4256 | ||
4257 | contiguous_readpages(tree, pagepool, nr, contig_start, | |
4258 | contig_end, &em_cached, &bio, &bio_flags, | |
4259 | &prev_em_start); | |
4260 | } | |
d1310b2e | 4261 | } |
67c9684f | 4262 | |
125bac01 MX |
4263 | if (em_cached) |
4264 | free_extent_map(em_cached); | |
4265 | ||
d1310b2e | 4266 | if (bio) |
1f7ad75b | 4267 | return submit_one_bio(bio, 0, bio_flags); |
d1310b2e CM |
4268 | return 0; |
4269 | } | |
d1310b2e CM |
4270 | |
4271 | /* | |
4272 | * basic invalidatepage code, this waits on any locked or writeback | |
4273 | * ranges corresponding to the page, and then deletes any extent state | |
4274 | * records from the tree | |
4275 | */ | |
4276 | int extent_invalidatepage(struct extent_io_tree *tree, | |
4277 | struct page *page, unsigned long offset) | |
4278 | { | |
2ac55d41 | 4279 | struct extent_state *cached_state = NULL; |
4eee4fa4 | 4280 | u64 start = page_offset(page); |
09cbfeaf | 4281 | u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
4282 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; |
4283 | ||
fda2832f | 4284 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
4285 | if (start > end) |
4286 | return 0; | |
4287 | ||
ff13db41 | 4288 | lock_extent_bits(tree, start, end, &cached_state); |
1edbb734 | 4289 | wait_on_page_writeback(page); |
d1310b2e | 4290 | clear_extent_bit(tree, start, end, |
32c00aff JB |
4291 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
4292 | EXTENT_DO_ACCOUNTING, | |
ae0f1625 | 4293 | 1, 1, &cached_state); |
d1310b2e CM |
4294 | return 0; |
4295 | } | |
d1310b2e | 4296 | |
7b13b7b1 CM |
4297 | /* |
4298 | * a helper for releasepage, this tests for areas of the page that | |
4299 | * are locked or under IO and drops the related state bits if it is safe | |
4300 | * to drop the page. | |
4301 | */ | |
29c68b2d | 4302 | static int try_release_extent_state(struct extent_io_tree *tree, |
48a3b636 | 4303 | struct page *page, gfp_t mask) |
7b13b7b1 | 4304 | { |
4eee4fa4 | 4305 | u64 start = page_offset(page); |
09cbfeaf | 4306 | u64 end = start + PAGE_SIZE - 1; |
7b13b7b1 CM |
4307 | int ret = 1; |
4308 | ||
8882679e | 4309 | if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) { |
7b13b7b1 | 4310 | ret = 0; |
8882679e | 4311 | } else { |
11ef160f CM |
4312 | /* |
4313 | * at this point we can safely clear everything except the | |
4314 | * locked bit and the nodatasum bit | |
4315 | */ | |
66b0c887 | 4316 | ret = __clear_extent_bit(tree, start, end, |
11ef160f | 4317 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
66b0c887 | 4318 | 0, 0, NULL, mask, NULL); |
e3f24cc5 CM |
4319 | |
4320 | /* if clear_extent_bit failed for enomem reasons, | |
4321 | * we can't allow the release to continue. | |
4322 | */ | |
4323 | if (ret < 0) | |
4324 | ret = 0; | |
4325 | else | |
4326 | ret = 1; | |
7b13b7b1 CM |
4327 | } |
4328 | return ret; | |
4329 | } | |
7b13b7b1 | 4330 | |
d1310b2e CM |
4331 | /* |
4332 | * a helper for releasepage. As long as there are no locked extents | |
4333 | * in the range corresponding to the page, both state records and extent | |
4334 | * map records are removed | |
4335 | */ | |
477a30ba | 4336 | int try_release_extent_mapping(struct page *page, gfp_t mask) |
d1310b2e CM |
4337 | { |
4338 | struct extent_map *em; | |
4eee4fa4 | 4339 | u64 start = page_offset(page); |
09cbfeaf | 4340 | u64 end = start + PAGE_SIZE - 1; |
bd3599a0 FM |
4341 | struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host); |
4342 | struct extent_io_tree *tree = &btrfs_inode->io_tree; | |
4343 | struct extent_map_tree *map = &btrfs_inode->extent_tree; | |
7b13b7b1 | 4344 | |
d0164adc | 4345 | if (gfpflags_allow_blocking(mask) && |
ee22184b | 4346 | page->mapping->host->i_size > SZ_16M) { |
39b5637f | 4347 | u64 len; |
70dec807 | 4348 | while (start <= end) { |
39b5637f | 4349 | len = end - start + 1; |
890871be | 4350 | write_lock(&map->lock); |
39b5637f | 4351 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 4352 | if (!em) { |
890871be | 4353 | write_unlock(&map->lock); |
70dec807 CM |
4354 | break; |
4355 | } | |
7f3c74fb CM |
4356 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
4357 | em->start != start) { | |
890871be | 4358 | write_unlock(&map->lock); |
70dec807 CM |
4359 | free_extent_map(em); |
4360 | break; | |
4361 | } | |
4362 | if (!test_range_bit(tree, em->start, | |
4363 | extent_map_end(em) - 1, | |
4e586ca3 | 4364 | EXTENT_LOCKED, 0, NULL)) { |
bd3599a0 FM |
4365 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
4366 | &btrfs_inode->runtime_flags); | |
70dec807 CM |
4367 | remove_extent_mapping(map, em); |
4368 | /* once for the rb tree */ | |
4369 | free_extent_map(em); | |
4370 | } | |
4371 | start = extent_map_end(em); | |
890871be | 4372 | write_unlock(&map->lock); |
70dec807 CM |
4373 | |
4374 | /* once for us */ | |
d1310b2e CM |
4375 | free_extent_map(em); |
4376 | } | |
d1310b2e | 4377 | } |
29c68b2d | 4378 | return try_release_extent_state(tree, page, mask); |
d1310b2e | 4379 | } |
d1310b2e | 4380 | |
ec29ed5b CM |
4381 | /* |
4382 | * helper function for fiemap, which doesn't want to see any holes. | |
4383 | * This maps until we find something past 'last' | |
4384 | */ | |
4385 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
e3350e16 | 4386 | u64 offset, u64 last) |
ec29ed5b | 4387 | { |
da17066c | 4388 | u64 sectorsize = btrfs_inode_sectorsize(inode); |
ec29ed5b CM |
4389 | struct extent_map *em; |
4390 | u64 len; | |
4391 | ||
4392 | if (offset >= last) | |
4393 | return NULL; | |
4394 | ||
67871254 | 4395 | while (1) { |
ec29ed5b CM |
4396 | len = last - offset; |
4397 | if (len == 0) | |
4398 | break; | |
fda2832f | 4399 | len = ALIGN(len, sectorsize); |
4ab47a8d | 4400 | em = btrfs_get_extent_fiemap(BTRFS_I(inode), offset, len); |
c704005d | 4401 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
4402 | return em; |
4403 | ||
4404 | /* if this isn't a hole return it */ | |
4a2d25cd | 4405 | if (em->block_start != EXTENT_MAP_HOLE) |
ec29ed5b | 4406 | return em; |
ec29ed5b CM |
4407 | |
4408 | /* this is a hole, advance to the next extent */ | |
4409 | offset = extent_map_end(em); | |
4410 | free_extent_map(em); | |
4411 | if (offset >= last) | |
4412 | break; | |
4413 | } | |
4414 | return NULL; | |
4415 | } | |
4416 | ||
4751832d QW |
4417 | /* |
4418 | * To cache previous fiemap extent | |
4419 | * | |
4420 | * Will be used for merging fiemap extent | |
4421 | */ | |
4422 | struct fiemap_cache { | |
4423 | u64 offset; | |
4424 | u64 phys; | |
4425 | u64 len; | |
4426 | u32 flags; | |
4427 | bool cached; | |
4428 | }; | |
4429 | ||
4430 | /* | |
4431 | * Helper to submit fiemap extent. | |
4432 | * | |
4433 | * Will try to merge current fiemap extent specified by @offset, @phys, | |
4434 | * @len and @flags with cached one. | |
4435 | * And only when we fails to merge, cached one will be submitted as | |
4436 | * fiemap extent. | |
4437 | * | |
4438 | * Return value is the same as fiemap_fill_next_extent(). | |
4439 | */ | |
4440 | static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, | |
4441 | struct fiemap_cache *cache, | |
4442 | u64 offset, u64 phys, u64 len, u32 flags) | |
4443 | { | |
4444 | int ret = 0; | |
4445 | ||
4446 | if (!cache->cached) | |
4447 | goto assign; | |
4448 | ||
4449 | /* | |
4450 | * Sanity check, extent_fiemap() should have ensured that new | |
52042d8e | 4451 | * fiemap extent won't overlap with cached one. |
4751832d QW |
4452 | * Not recoverable. |
4453 | * | |
4454 | * NOTE: Physical address can overlap, due to compression | |
4455 | */ | |
4456 | if (cache->offset + cache->len > offset) { | |
4457 | WARN_ON(1); | |
4458 | return -EINVAL; | |
4459 | } | |
4460 | ||
4461 | /* | |
4462 | * Only merges fiemap extents if | |
4463 | * 1) Their logical addresses are continuous | |
4464 | * | |
4465 | * 2) Their physical addresses are continuous | |
4466 | * So truly compressed (physical size smaller than logical size) | |
4467 | * extents won't get merged with each other | |
4468 | * | |
4469 | * 3) Share same flags except FIEMAP_EXTENT_LAST | |
4470 | * So regular extent won't get merged with prealloc extent | |
4471 | */ | |
4472 | if (cache->offset + cache->len == offset && | |
4473 | cache->phys + cache->len == phys && | |
4474 | (cache->flags & ~FIEMAP_EXTENT_LAST) == | |
4475 | (flags & ~FIEMAP_EXTENT_LAST)) { | |
4476 | cache->len += len; | |
4477 | cache->flags |= flags; | |
4478 | goto try_submit_last; | |
4479 | } | |
4480 | ||
4481 | /* Not mergeable, need to submit cached one */ | |
4482 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, | |
4483 | cache->len, cache->flags); | |
4484 | cache->cached = false; | |
4485 | if (ret) | |
4486 | return ret; | |
4487 | assign: | |
4488 | cache->cached = true; | |
4489 | cache->offset = offset; | |
4490 | cache->phys = phys; | |
4491 | cache->len = len; | |
4492 | cache->flags = flags; | |
4493 | try_submit_last: | |
4494 | if (cache->flags & FIEMAP_EXTENT_LAST) { | |
4495 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, | |
4496 | cache->phys, cache->len, cache->flags); | |
4497 | cache->cached = false; | |
4498 | } | |
4499 | return ret; | |
4500 | } | |
4501 | ||
4502 | /* | |
848c23b7 | 4503 | * Emit last fiemap cache |
4751832d | 4504 | * |
848c23b7 QW |
4505 | * The last fiemap cache may still be cached in the following case: |
4506 | * 0 4k 8k | |
4507 | * |<- Fiemap range ->| | |
4508 | * |<------------ First extent ----------->| | |
4509 | * | |
4510 | * In this case, the first extent range will be cached but not emitted. | |
4511 | * So we must emit it before ending extent_fiemap(). | |
4751832d | 4512 | */ |
848c23b7 QW |
4513 | static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info, |
4514 | struct fiemap_extent_info *fieinfo, | |
4515 | struct fiemap_cache *cache) | |
4751832d QW |
4516 | { |
4517 | int ret; | |
4518 | ||
4519 | if (!cache->cached) | |
4520 | return 0; | |
4521 | ||
4751832d QW |
4522 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, |
4523 | cache->len, cache->flags); | |
4524 | cache->cached = false; | |
4525 | if (ret > 0) | |
4526 | ret = 0; | |
4527 | return ret; | |
4528 | } | |
4529 | ||
1506fcc8 | 4530 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
2135fb9b | 4531 | __u64 start, __u64 len) |
1506fcc8 | 4532 | { |
975f84fe | 4533 | int ret = 0; |
1506fcc8 YS |
4534 | u64 off = start; |
4535 | u64 max = start + len; | |
4536 | u32 flags = 0; | |
975f84fe JB |
4537 | u32 found_type; |
4538 | u64 last; | |
ec29ed5b | 4539 | u64 last_for_get_extent = 0; |
1506fcc8 | 4540 | u64 disko = 0; |
ec29ed5b | 4541 | u64 isize = i_size_read(inode); |
975f84fe | 4542 | struct btrfs_key found_key; |
1506fcc8 | 4543 | struct extent_map *em = NULL; |
2ac55d41 | 4544 | struct extent_state *cached_state = NULL; |
975f84fe | 4545 | struct btrfs_path *path; |
dc046b10 | 4546 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4751832d | 4547 | struct fiemap_cache cache = { 0 }; |
1506fcc8 | 4548 | int end = 0; |
ec29ed5b CM |
4549 | u64 em_start = 0; |
4550 | u64 em_len = 0; | |
4551 | u64 em_end = 0; | |
1506fcc8 YS |
4552 | |
4553 | if (len == 0) | |
4554 | return -EINVAL; | |
4555 | ||
975f84fe JB |
4556 | path = btrfs_alloc_path(); |
4557 | if (!path) | |
4558 | return -ENOMEM; | |
4559 | path->leave_spinning = 1; | |
4560 | ||
da17066c JM |
4561 | start = round_down(start, btrfs_inode_sectorsize(inode)); |
4562 | len = round_up(max, btrfs_inode_sectorsize(inode)) - start; | |
4d479cf0 | 4563 | |
ec29ed5b CM |
4564 | /* |
4565 | * lookup the last file extent. We're not using i_size here | |
4566 | * because there might be preallocation past i_size | |
4567 | */ | |
f85b7379 DS |
4568 | ret = btrfs_lookup_file_extent(NULL, root, path, |
4569 | btrfs_ino(BTRFS_I(inode)), -1, 0); | |
975f84fe JB |
4570 | if (ret < 0) { |
4571 | btrfs_free_path(path); | |
4572 | return ret; | |
2d324f59 LB |
4573 | } else { |
4574 | WARN_ON(!ret); | |
4575 | if (ret == 1) | |
4576 | ret = 0; | |
975f84fe | 4577 | } |
2d324f59 | 4578 | |
975f84fe | 4579 | path->slots[0]--; |
975f84fe | 4580 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
962a298f | 4581 | found_type = found_key.type; |
975f84fe | 4582 | |
ec29ed5b | 4583 | /* No extents, but there might be delalloc bits */ |
4a0cc7ca | 4584 | if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) || |
975f84fe | 4585 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
4586 | /* have to trust i_size as the end */ |
4587 | last = (u64)-1; | |
4588 | last_for_get_extent = isize; | |
4589 | } else { | |
4590 | /* | |
4591 | * remember the start of the last extent. There are a | |
4592 | * bunch of different factors that go into the length of the | |
4593 | * extent, so its much less complex to remember where it started | |
4594 | */ | |
4595 | last = found_key.offset; | |
4596 | last_for_get_extent = last + 1; | |
975f84fe | 4597 | } |
fe09e16c | 4598 | btrfs_release_path(path); |
975f84fe | 4599 | |
ec29ed5b CM |
4600 | /* |
4601 | * we might have some extents allocated but more delalloc past those | |
4602 | * extents. so, we trust isize unless the start of the last extent is | |
4603 | * beyond isize | |
4604 | */ | |
4605 | if (last < isize) { | |
4606 | last = (u64)-1; | |
4607 | last_for_get_extent = isize; | |
4608 | } | |
4609 | ||
ff13db41 | 4610 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
d0082371 | 4611 | &cached_state); |
ec29ed5b | 4612 | |
e3350e16 | 4613 | em = get_extent_skip_holes(inode, start, last_for_get_extent); |
1506fcc8 YS |
4614 | if (!em) |
4615 | goto out; | |
4616 | if (IS_ERR(em)) { | |
4617 | ret = PTR_ERR(em); | |
4618 | goto out; | |
4619 | } | |
975f84fe | 4620 | |
1506fcc8 | 4621 | while (!end) { |
b76bb701 | 4622 | u64 offset_in_extent = 0; |
ea8efc74 CM |
4623 | |
4624 | /* break if the extent we found is outside the range */ | |
4625 | if (em->start >= max || extent_map_end(em) < off) | |
4626 | break; | |
4627 | ||
4628 | /* | |
4629 | * get_extent may return an extent that starts before our | |
4630 | * requested range. We have to make sure the ranges | |
4631 | * we return to fiemap always move forward and don't | |
4632 | * overlap, so adjust the offsets here | |
4633 | */ | |
4634 | em_start = max(em->start, off); | |
1506fcc8 | 4635 | |
ea8efc74 CM |
4636 | /* |
4637 | * record the offset from the start of the extent | |
b76bb701 JB |
4638 | * for adjusting the disk offset below. Only do this if the |
4639 | * extent isn't compressed since our in ram offset may be past | |
4640 | * what we have actually allocated on disk. | |
ea8efc74 | 4641 | */ |
b76bb701 JB |
4642 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
4643 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 4644 | em_end = extent_map_end(em); |
ea8efc74 | 4645 | em_len = em_end - em_start; |
1506fcc8 | 4646 | flags = 0; |
f0986318 FM |
4647 | if (em->block_start < EXTENT_MAP_LAST_BYTE) |
4648 | disko = em->block_start + offset_in_extent; | |
4649 | else | |
4650 | disko = 0; | |
1506fcc8 | 4651 | |
ea8efc74 CM |
4652 | /* |
4653 | * bump off for our next call to get_extent | |
4654 | */ | |
4655 | off = extent_map_end(em); | |
4656 | if (off >= max) | |
4657 | end = 1; | |
4658 | ||
93dbfad7 | 4659 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
4660 | end = 1; |
4661 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 4662 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
4663 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
4664 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 4665 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
4666 | flags |= (FIEMAP_EXTENT_DELALLOC | |
4667 | FIEMAP_EXTENT_UNKNOWN); | |
dc046b10 JB |
4668 | } else if (fieinfo->fi_extents_max) { |
4669 | u64 bytenr = em->block_start - | |
4670 | (em->start - em->orig_start); | |
fe09e16c | 4671 | |
fe09e16c LB |
4672 | /* |
4673 | * As btrfs supports shared space, this information | |
4674 | * can be exported to userspace tools via | |
dc046b10 JB |
4675 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 |
4676 | * then we're just getting a count and we can skip the | |
4677 | * lookup stuff. | |
fe09e16c | 4678 | */ |
bb739cf0 EN |
4679 | ret = btrfs_check_shared(root, |
4680 | btrfs_ino(BTRFS_I(inode)), | |
4681 | bytenr); | |
dc046b10 | 4682 | if (ret < 0) |
fe09e16c | 4683 | goto out_free; |
dc046b10 | 4684 | if (ret) |
fe09e16c | 4685 | flags |= FIEMAP_EXTENT_SHARED; |
dc046b10 | 4686 | ret = 0; |
1506fcc8 YS |
4687 | } |
4688 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
4689 | flags |= FIEMAP_EXTENT_ENCODED; | |
0d2b2372 JB |
4690 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
4691 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
1506fcc8 | 4692 | |
1506fcc8 YS |
4693 | free_extent_map(em); |
4694 | em = NULL; | |
ec29ed5b CM |
4695 | if ((em_start >= last) || em_len == (u64)-1 || |
4696 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
4697 | flags |= FIEMAP_EXTENT_LAST; |
4698 | end = 1; | |
4699 | } | |
4700 | ||
ec29ed5b | 4701 | /* now scan forward to see if this is really the last extent. */ |
e3350e16 | 4702 | em = get_extent_skip_holes(inode, off, last_for_get_extent); |
ec29ed5b CM |
4703 | if (IS_ERR(em)) { |
4704 | ret = PTR_ERR(em); | |
4705 | goto out; | |
4706 | } | |
4707 | if (!em) { | |
975f84fe JB |
4708 | flags |= FIEMAP_EXTENT_LAST; |
4709 | end = 1; | |
4710 | } | |
4751832d QW |
4711 | ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko, |
4712 | em_len, flags); | |
26e726af CS |
4713 | if (ret) { |
4714 | if (ret == 1) | |
4715 | ret = 0; | |
ec29ed5b | 4716 | goto out_free; |
26e726af | 4717 | } |
1506fcc8 YS |
4718 | } |
4719 | out_free: | |
4751832d | 4720 | if (!ret) |
848c23b7 | 4721 | ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache); |
1506fcc8 YS |
4722 | free_extent_map(em); |
4723 | out: | |
fe09e16c | 4724 | btrfs_free_path(path); |
a52f4cd2 | 4725 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
e43bbe5e | 4726 | &cached_state); |
1506fcc8 YS |
4727 | return ret; |
4728 | } | |
4729 | ||
727011e0 CM |
4730 | static void __free_extent_buffer(struct extent_buffer *eb) |
4731 | { | |
6d49ba1b | 4732 | btrfs_leak_debug_del(&eb->leak_list); |
727011e0 CM |
4733 | kmem_cache_free(extent_buffer_cache, eb); |
4734 | } | |
4735 | ||
a26e8c9f | 4736 | int extent_buffer_under_io(struct extent_buffer *eb) |
db7f3436 JB |
4737 | { |
4738 | return (atomic_read(&eb->io_pages) || | |
4739 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4740 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4741 | } | |
4742 | ||
4743 | /* | |
55ac0139 | 4744 | * Release all pages attached to the extent buffer. |
db7f3436 | 4745 | */ |
55ac0139 | 4746 | static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb) |
db7f3436 | 4747 | { |
d64766fd NB |
4748 | int i; |
4749 | int num_pages; | |
b0132a3b | 4750 | int mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
db7f3436 JB |
4751 | |
4752 | BUG_ON(extent_buffer_under_io(eb)); | |
4753 | ||
d64766fd NB |
4754 | num_pages = num_extent_pages(eb); |
4755 | for (i = 0; i < num_pages; i++) { | |
4756 | struct page *page = eb->pages[i]; | |
db7f3436 | 4757 | |
5d2361db FL |
4758 | if (!page) |
4759 | continue; | |
4760 | if (mapped) | |
db7f3436 | 4761 | spin_lock(&page->mapping->private_lock); |
5d2361db FL |
4762 | /* |
4763 | * We do this since we'll remove the pages after we've | |
4764 | * removed the eb from the radix tree, so we could race | |
4765 | * and have this page now attached to the new eb. So | |
4766 | * only clear page_private if it's still connected to | |
4767 | * this eb. | |
4768 | */ | |
4769 | if (PagePrivate(page) && | |
4770 | page->private == (unsigned long)eb) { | |
4771 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4772 | BUG_ON(PageDirty(page)); | |
4773 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 4774 | /* |
5d2361db FL |
4775 | * We need to make sure we haven't be attached |
4776 | * to a new eb. | |
db7f3436 | 4777 | */ |
5d2361db FL |
4778 | ClearPagePrivate(page); |
4779 | set_page_private(page, 0); | |
4780 | /* One for the page private */ | |
09cbfeaf | 4781 | put_page(page); |
db7f3436 | 4782 | } |
5d2361db FL |
4783 | |
4784 | if (mapped) | |
4785 | spin_unlock(&page->mapping->private_lock); | |
4786 | ||
01327610 | 4787 | /* One for when we allocated the page */ |
09cbfeaf | 4788 | put_page(page); |
d64766fd | 4789 | } |
db7f3436 JB |
4790 | } |
4791 | ||
4792 | /* | |
4793 | * Helper for releasing the extent buffer. | |
4794 | */ | |
4795 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4796 | { | |
55ac0139 | 4797 | btrfs_release_extent_buffer_pages(eb); |
db7f3436 JB |
4798 | __free_extent_buffer(eb); |
4799 | } | |
4800 | ||
f28491e0 JB |
4801 | static struct extent_buffer * |
4802 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 4803 | unsigned long len) |
d1310b2e CM |
4804 | { |
4805 | struct extent_buffer *eb = NULL; | |
4806 | ||
d1b5c567 | 4807 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
4808 | eb->start = start; |
4809 | eb->len = len; | |
f28491e0 | 4810 | eb->fs_info = fs_info; |
815a51c7 | 4811 | eb->bflags = 0; |
bd681513 | 4812 | rwlock_init(&eb->lock); |
bd681513 CM |
4813 | atomic_set(&eb->blocking_readers, 0); |
4814 | atomic_set(&eb->blocking_writers, 0); | |
ed1b4ed7 | 4815 | eb->lock_nested = false; |
bd681513 CM |
4816 | init_waitqueue_head(&eb->write_lock_wq); |
4817 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 4818 | |
6d49ba1b ES |
4819 | btrfs_leak_debug_add(&eb->leak_list, &buffers); |
4820 | ||
3083ee2e | 4821 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 4822 | atomic_set(&eb->refs, 1); |
0b32f4bb | 4823 | atomic_set(&eb->io_pages, 0); |
727011e0 | 4824 | |
b8dae313 DS |
4825 | /* |
4826 | * Sanity checks, currently the maximum is 64k covered by 16x 4k pages | |
4827 | */ | |
4828 | BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE | |
4829 | > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
4830 | BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
d1310b2e | 4831 | |
843ccf9f DS |
4832 | #ifdef CONFIG_BTRFS_DEBUG |
4833 | atomic_set(&eb->spinning_writers, 0); | |
afd495a8 | 4834 | atomic_set(&eb->spinning_readers, 0); |
5c9c799a | 4835 | atomic_set(&eb->read_locks, 0); |
c79adfc0 | 4836 | atomic_set(&eb->write_locks, 0); |
843ccf9f DS |
4837 | #endif |
4838 | ||
d1310b2e CM |
4839 | return eb; |
4840 | } | |
4841 | ||
815a51c7 JS |
4842 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) |
4843 | { | |
cc5e31a4 | 4844 | int i; |
815a51c7 JS |
4845 | struct page *p; |
4846 | struct extent_buffer *new; | |
cc5e31a4 | 4847 | int num_pages = num_extent_pages(src); |
815a51c7 | 4848 | |
3f556f78 | 4849 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
4850 | if (new == NULL) |
4851 | return NULL; | |
4852 | ||
4853 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4854 | p = alloc_page(GFP_NOFS); |
db7f3436 JB |
4855 | if (!p) { |
4856 | btrfs_release_extent_buffer(new); | |
4857 | return NULL; | |
4858 | } | |
815a51c7 JS |
4859 | attach_extent_buffer_page(new, p); |
4860 | WARN_ON(PageDirty(p)); | |
4861 | SetPageUptodate(p); | |
4862 | new->pages[i] = p; | |
fba1acf9 | 4863 | copy_page(page_address(p), page_address(src->pages[i])); |
815a51c7 JS |
4864 | } |
4865 | ||
815a51c7 | 4866 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); |
b0132a3b | 4867 | set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags); |
815a51c7 JS |
4868 | |
4869 | return new; | |
4870 | } | |
4871 | ||
0f331229 OS |
4872 | struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
4873 | u64 start, unsigned long len) | |
815a51c7 JS |
4874 | { |
4875 | struct extent_buffer *eb; | |
cc5e31a4 DS |
4876 | int num_pages; |
4877 | int i; | |
815a51c7 | 4878 | |
3f556f78 | 4879 | eb = __alloc_extent_buffer(fs_info, start, len); |
815a51c7 JS |
4880 | if (!eb) |
4881 | return NULL; | |
4882 | ||
65ad0104 | 4883 | num_pages = num_extent_pages(eb); |
815a51c7 | 4884 | for (i = 0; i < num_pages; i++) { |
9ec72677 | 4885 | eb->pages[i] = alloc_page(GFP_NOFS); |
815a51c7 JS |
4886 | if (!eb->pages[i]) |
4887 | goto err; | |
4888 | } | |
4889 | set_extent_buffer_uptodate(eb); | |
4890 | btrfs_set_header_nritems(eb, 0); | |
b0132a3b | 4891 | set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
815a51c7 JS |
4892 | |
4893 | return eb; | |
4894 | err: | |
84167d19 SB |
4895 | for (; i > 0; i--) |
4896 | __free_page(eb->pages[i - 1]); | |
815a51c7 JS |
4897 | __free_extent_buffer(eb); |
4898 | return NULL; | |
4899 | } | |
4900 | ||
0f331229 | 4901 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
da17066c | 4902 | u64 start) |
0f331229 | 4903 | { |
da17066c | 4904 | return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize); |
0f331229 OS |
4905 | } |
4906 | ||
0b32f4bb JB |
4907 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
4908 | { | |
242e18c7 | 4909 | int refs; |
0b32f4bb JB |
4910 | /* the ref bit is tricky. We have to make sure it is set |
4911 | * if we have the buffer dirty. Otherwise the | |
4912 | * code to free a buffer can end up dropping a dirty | |
4913 | * page | |
4914 | * | |
4915 | * Once the ref bit is set, it won't go away while the | |
4916 | * buffer is dirty or in writeback, and it also won't | |
4917 | * go away while we have the reference count on the | |
4918 | * eb bumped. | |
4919 | * | |
4920 | * We can't just set the ref bit without bumping the | |
4921 | * ref on the eb because free_extent_buffer might | |
4922 | * see the ref bit and try to clear it. If this happens | |
4923 | * free_extent_buffer might end up dropping our original | |
4924 | * ref by mistake and freeing the page before we are able | |
4925 | * to add one more ref. | |
4926 | * | |
4927 | * So bump the ref count first, then set the bit. If someone | |
4928 | * beat us to it, drop the ref we added. | |
4929 | */ | |
242e18c7 CM |
4930 | refs = atomic_read(&eb->refs); |
4931 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
4932 | return; | |
4933 | ||
594831c4 JB |
4934 | spin_lock(&eb->refs_lock); |
4935 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 4936 | atomic_inc(&eb->refs); |
594831c4 | 4937 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
4938 | } |
4939 | ||
2457aec6 MG |
4940 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
4941 | struct page *accessed) | |
5df4235e | 4942 | { |
cc5e31a4 | 4943 | int num_pages, i; |
5df4235e | 4944 | |
0b32f4bb JB |
4945 | check_buffer_tree_ref(eb); |
4946 | ||
65ad0104 | 4947 | num_pages = num_extent_pages(eb); |
5df4235e | 4948 | for (i = 0; i < num_pages; i++) { |
fb85fc9a DS |
4949 | struct page *p = eb->pages[i]; |
4950 | ||
2457aec6 MG |
4951 | if (p != accessed) |
4952 | mark_page_accessed(p); | |
5df4235e JB |
4953 | } |
4954 | } | |
4955 | ||
f28491e0 JB |
4956 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
4957 | u64 start) | |
452c75c3 CS |
4958 | { |
4959 | struct extent_buffer *eb; | |
4960 | ||
4961 | rcu_read_lock(); | |
f28491e0 | 4962 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
09cbfeaf | 4963 | start >> PAGE_SHIFT); |
452c75c3 CS |
4964 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
4965 | rcu_read_unlock(); | |
062c19e9 FM |
4966 | /* |
4967 | * Lock our eb's refs_lock to avoid races with | |
4968 | * free_extent_buffer. When we get our eb it might be flagged | |
4969 | * with EXTENT_BUFFER_STALE and another task running | |
4970 | * free_extent_buffer might have seen that flag set, | |
4971 | * eb->refs == 2, that the buffer isn't under IO (dirty and | |
4972 | * writeback flags not set) and it's still in the tree (flag | |
4973 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process | |
4974 | * of decrementing the extent buffer's reference count twice. | |
4975 | * So here we could race and increment the eb's reference count, | |
4976 | * clear its stale flag, mark it as dirty and drop our reference | |
4977 | * before the other task finishes executing free_extent_buffer, | |
4978 | * which would later result in an attempt to free an extent | |
4979 | * buffer that is dirty. | |
4980 | */ | |
4981 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
4982 | spin_lock(&eb->refs_lock); | |
4983 | spin_unlock(&eb->refs_lock); | |
4984 | } | |
2457aec6 | 4985 | mark_extent_buffer_accessed(eb, NULL); |
452c75c3 CS |
4986 | return eb; |
4987 | } | |
4988 | rcu_read_unlock(); | |
4989 | ||
4990 | return NULL; | |
4991 | } | |
4992 | ||
faa2dbf0 JB |
4993 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
4994 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
da17066c | 4995 | u64 start) |
faa2dbf0 JB |
4996 | { |
4997 | struct extent_buffer *eb, *exists = NULL; | |
4998 | int ret; | |
4999 | ||
5000 | eb = find_extent_buffer(fs_info, start); | |
5001 | if (eb) | |
5002 | return eb; | |
da17066c | 5003 | eb = alloc_dummy_extent_buffer(fs_info, start); |
faa2dbf0 JB |
5004 | if (!eb) |
5005 | return NULL; | |
5006 | eb->fs_info = fs_info; | |
5007 | again: | |
e1860a77 | 5008 | ret = radix_tree_preload(GFP_NOFS); |
faa2dbf0 JB |
5009 | if (ret) |
5010 | goto free_eb; | |
5011 | spin_lock(&fs_info->buffer_lock); | |
5012 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
09cbfeaf | 5013 | start >> PAGE_SHIFT, eb); |
faa2dbf0 JB |
5014 | spin_unlock(&fs_info->buffer_lock); |
5015 | radix_tree_preload_end(); | |
5016 | if (ret == -EEXIST) { | |
5017 | exists = find_extent_buffer(fs_info, start); | |
5018 | if (exists) | |
5019 | goto free_eb; | |
5020 | else | |
5021 | goto again; | |
5022 | } | |
5023 | check_buffer_tree_ref(eb); | |
5024 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
5025 | ||
faa2dbf0 JB |
5026 | return eb; |
5027 | free_eb: | |
5028 | btrfs_release_extent_buffer(eb); | |
5029 | return exists; | |
5030 | } | |
5031 | #endif | |
5032 | ||
f28491e0 | 5033 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
ce3e6984 | 5034 | u64 start) |
d1310b2e | 5035 | { |
da17066c | 5036 | unsigned long len = fs_info->nodesize; |
cc5e31a4 DS |
5037 | int num_pages; |
5038 | int i; | |
09cbfeaf | 5039 | unsigned long index = start >> PAGE_SHIFT; |
d1310b2e | 5040 | struct extent_buffer *eb; |
6af118ce | 5041 | struct extent_buffer *exists = NULL; |
d1310b2e | 5042 | struct page *p; |
f28491e0 | 5043 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
d1310b2e | 5044 | int uptodate = 1; |
19fe0a8b | 5045 | int ret; |
d1310b2e | 5046 | |
da17066c | 5047 | if (!IS_ALIGNED(start, fs_info->sectorsize)) { |
c871b0f2 LB |
5048 | btrfs_err(fs_info, "bad tree block start %llu", start); |
5049 | return ERR_PTR(-EINVAL); | |
5050 | } | |
5051 | ||
f28491e0 | 5052 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 5053 | if (eb) |
6af118ce | 5054 | return eb; |
6af118ce | 5055 | |
23d79d81 | 5056 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 5057 | if (!eb) |
c871b0f2 | 5058 | return ERR_PTR(-ENOMEM); |
d1310b2e | 5059 | |
65ad0104 | 5060 | num_pages = num_extent_pages(eb); |
727011e0 | 5061 | for (i = 0; i < num_pages; i++, index++) { |
d1b5c567 | 5062 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
c871b0f2 LB |
5063 | if (!p) { |
5064 | exists = ERR_PTR(-ENOMEM); | |
6af118ce | 5065 | goto free_eb; |
c871b0f2 | 5066 | } |
4f2de97a JB |
5067 | |
5068 | spin_lock(&mapping->private_lock); | |
5069 | if (PagePrivate(p)) { | |
5070 | /* | |
5071 | * We could have already allocated an eb for this page | |
5072 | * and attached one so lets see if we can get a ref on | |
5073 | * the existing eb, and if we can we know it's good and | |
5074 | * we can just return that one, else we know we can just | |
5075 | * overwrite page->private. | |
5076 | */ | |
5077 | exists = (struct extent_buffer *)p->private; | |
5078 | if (atomic_inc_not_zero(&exists->refs)) { | |
5079 | spin_unlock(&mapping->private_lock); | |
5080 | unlock_page(p); | |
09cbfeaf | 5081 | put_page(p); |
2457aec6 | 5082 | mark_extent_buffer_accessed(exists, p); |
4f2de97a JB |
5083 | goto free_eb; |
5084 | } | |
5ca64f45 | 5085 | exists = NULL; |
4f2de97a | 5086 | |
0b32f4bb | 5087 | /* |
4f2de97a JB |
5088 | * Do this so attach doesn't complain and we need to |
5089 | * drop the ref the old guy had. | |
5090 | */ | |
5091 | ClearPagePrivate(p); | |
0b32f4bb | 5092 | WARN_ON(PageDirty(p)); |
09cbfeaf | 5093 | put_page(p); |
d1310b2e | 5094 | } |
4f2de97a JB |
5095 | attach_extent_buffer_page(eb, p); |
5096 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 5097 | WARN_ON(PageDirty(p)); |
727011e0 | 5098 | eb->pages[i] = p; |
d1310b2e CM |
5099 | if (!PageUptodate(p)) |
5100 | uptodate = 0; | |
eb14ab8e CM |
5101 | |
5102 | /* | |
b16d011e NB |
5103 | * We can't unlock the pages just yet since the extent buffer |
5104 | * hasn't been properly inserted in the radix tree, this | |
5105 | * opens a race with btree_releasepage which can free a page | |
5106 | * while we are still filling in all pages for the buffer and | |
5107 | * we could crash. | |
eb14ab8e | 5108 | */ |
d1310b2e CM |
5109 | } |
5110 | if (uptodate) | |
b4ce94de | 5111 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 5112 | again: |
e1860a77 | 5113 | ret = radix_tree_preload(GFP_NOFS); |
c871b0f2 LB |
5114 | if (ret) { |
5115 | exists = ERR_PTR(ret); | |
19fe0a8b | 5116 | goto free_eb; |
c871b0f2 | 5117 | } |
19fe0a8b | 5118 | |
f28491e0 JB |
5119 | spin_lock(&fs_info->buffer_lock); |
5120 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
09cbfeaf | 5121 | start >> PAGE_SHIFT, eb); |
f28491e0 | 5122 | spin_unlock(&fs_info->buffer_lock); |
452c75c3 | 5123 | radix_tree_preload_end(); |
19fe0a8b | 5124 | if (ret == -EEXIST) { |
f28491e0 | 5125 | exists = find_extent_buffer(fs_info, start); |
452c75c3 CS |
5126 | if (exists) |
5127 | goto free_eb; | |
5128 | else | |
115391d2 | 5129 | goto again; |
6af118ce | 5130 | } |
6af118ce | 5131 | /* add one reference for the tree */ |
0b32f4bb | 5132 | check_buffer_tree_ref(eb); |
34b41ace | 5133 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
5134 | |
5135 | /* | |
b16d011e NB |
5136 | * Now it's safe to unlock the pages because any calls to |
5137 | * btree_releasepage will correctly detect that a page belongs to a | |
5138 | * live buffer and won't free them prematurely. | |
eb14ab8e | 5139 | */ |
28187ae5 NB |
5140 | for (i = 0; i < num_pages; i++) |
5141 | unlock_page(eb->pages[i]); | |
d1310b2e CM |
5142 | return eb; |
5143 | ||
6af118ce | 5144 | free_eb: |
5ca64f45 | 5145 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
5146 | for (i = 0; i < num_pages; i++) { |
5147 | if (eb->pages[i]) | |
5148 | unlock_page(eb->pages[i]); | |
5149 | } | |
eb14ab8e | 5150 | |
897ca6e9 | 5151 | btrfs_release_extent_buffer(eb); |
6af118ce | 5152 | return exists; |
d1310b2e | 5153 | } |
d1310b2e | 5154 | |
3083ee2e JB |
5155 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
5156 | { | |
5157 | struct extent_buffer *eb = | |
5158 | container_of(head, struct extent_buffer, rcu_head); | |
5159 | ||
5160 | __free_extent_buffer(eb); | |
5161 | } | |
5162 | ||
f7a52a40 | 5163 | static int release_extent_buffer(struct extent_buffer *eb) |
3083ee2e | 5164 | { |
07e21c4d NB |
5165 | lockdep_assert_held(&eb->refs_lock); |
5166 | ||
3083ee2e JB |
5167 | WARN_ON(atomic_read(&eb->refs) == 0); |
5168 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 5169 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 5170 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 5171 | |
815a51c7 | 5172 | spin_unlock(&eb->refs_lock); |
3083ee2e | 5173 | |
f28491e0 JB |
5174 | spin_lock(&fs_info->buffer_lock); |
5175 | radix_tree_delete(&fs_info->buffer_radix, | |
09cbfeaf | 5176 | eb->start >> PAGE_SHIFT); |
f28491e0 | 5177 | spin_unlock(&fs_info->buffer_lock); |
34b41ace JB |
5178 | } else { |
5179 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 5180 | } |
3083ee2e JB |
5181 | |
5182 | /* Should be safe to release our pages at this point */ | |
55ac0139 | 5183 | btrfs_release_extent_buffer_pages(eb); |
bcb7e449 | 5184 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
b0132a3b | 5185 | if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) { |
bcb7e449 JB |
5186 | __free_extent_buffer(eb); |
5187 | return 1; | |
5188 | } | |
5189 | #endif | |
3083ee2e | 5190 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 5191 | return 1; |
3083ee2e JB |
5192 | } |
5193 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
5194 | |
5195 | return 0; | |
3083ee2e JB |
5196 | } |
5197 | ||
d1310b2e CM |
5198 | void free_extent_buffer(struct extent_buffer *eb) |
5199 | { | |
242e18c7 CM |
5200 | int refs; |
5201 | int old; | |
d1310b2e CM |
5202 | if (!eb) |
5203 | return; | |
5204 | ||
242e18c7 CM |
5205 | while (1) { |
5206 | refs = atomic_read(&eb->refs); | |
46cc775e NB |
5207 | if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3) |
5208 | || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && | |
5209 | refs == 1)) | |
242e18c7 CM |
5210 | break; |
5211 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
5212 | if (old == refs) | |
5213 | return; | |
5214 | } | |
5215 | ||
3083ee2e JB |
5216 | spin_lock(&eb->refs_lock); |
5217 | if (atomic_read(&eb->refs) == 2 && | |
5218 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 5219 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
5220 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5221 | atomic_dec(&eb->refs); | |
5222 | ||
5223 | /* | |
5224 | * I know this is terrible, but it's temporary until we stop tracking | |
5225 | * the uptodate bits and such for the extent buffers. | |
5226 | */ | |
f7a52a40 | 5227 | release_extent_buffer(eb); |
3083ee2e JB |
5228 | } |
5229 | ||
5230 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
5231 | { | |
5232 | if (!eb) | |
d1310b2e CM |
5233 | return; |
5234 | ||
3083ee2e JB |
5235 | spin_lock(&eb->refs_lock); |
5236 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
5237 | ||
0b32f4bb | 5238 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
5239 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5240 | atomic_dec(&eb->refs); | |
f7a52a40 | 5241 | release_extent_buffer(eb); |
d1310b2e | 5242 | } |
d1310b2e | 5243 | |
1d4284bd | 5244 | void clear_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5245 | { |
cc5e31a4 DS |
5246 | int i; |
5247 | int num_pages; | |
d1310b2e CM |
5248 | struct page *page; |
5249 | ||
65ad0104 | 5250 | num_pages = num_extent_pages(eb); |
d1310b2e CM |
5251 | |
5252 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 5253 | page = eb->pages[i]; |
b9473439 | 5254 | if (!PageDirty(page)) |
d2c3f4f6 CM |
5255 | continue; |
5256 | ||
a61e6f29 | 5257 | lock_page(page); |
eb14ab8e CM |
5258 | WARN_ON(!PagePrivate(page)); |
5259 | ||
d1310b2e | 5260 | clear_page_dirty_for_io(page); |
b93b0163 | 5261 | xa_lock_irq(&page->mapping->i_pages); |
0a943c65 MW |
5262 | if (!PageDirty(page)) |
5263 | __xa_clear_mark(&page->mapping->i_pages, | |
5264 | page_index(page), PAGECACHE_TAG_DIRTY); | |
b93b0163 | 5265 | xa_unlock_irq(&page->mapping->i_pages); |
bf0da8c1 | 5266 | ClearPageError(page); |
a61e6f29 | 5267 | unlock_page(page); |
d1310b2e | 5268 | } |
0b32f4bb | 5269 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 5270 | } |
d1310b2e | 5271 | |
abb57ef3 | 5272 | bool set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5273 | { |
cc5e31a4 DS |
5274 | int i; |
5275 | int num_pages; | |
abb57ef3 | 5276 | bool was_dirty; |
d1310b2e | 5277 | |
0b32f4bb JB |
5278 | check_buffer_tree_ref(eb); |
5279 | ||
b9473439 | 5280 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 5281 | |
65ad0104 | 5282 | num_pages = num_extent_pages(eb); |
3083ee2e | 5283 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
5284 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
5285 | ||
abb57ef3 LB |
5286 | if (!was_dirty) |
5287 | for (i = 0; i < num_pages; i++) | |
5288 | set_page_dirty(eb->pages[i]); | |
51995c39 LB |
5289 | |
5290 | #ifdef CONFIG_BTRFS_DEBUG | |
5291 | for (i = 0; i < num_pages; i++) | |
5292 | ASSERT(PageDirty(eb->pages[i])); | |
5293 | #endif | |
5294 | ||
b9473439 | 5295 | return was_dirty; |
d1310b2e | 5296 | } |
d1310b2e | 5297 | |
69ba3927 | 5298 | void clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 | 5299 | { |
cc5e31a4 | 5300 | int i; |
1259ab75 | 5301 | struct page *page; |
cc5e31a4 | 5302 | int num_pages; |
1259ab75 | 5303 | |
b4ce94de | 5304 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 5305 | num_pages = num_extent_pages(eb); |
1259ab75 | 5306 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5307 | page = eb->pages[i]; |
33958dc6 CM |
5308 | if (page) |
5309 | ClearPageUptodate(page); | |
1259ab75 | 5310 | } |
1259ab75 CM |
5311 | } |
5312 | ||
09c25a8c | 5313 | void set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 5314 | { |
cc5e31a4 | 5315 | int i; |
d1310b2e | 5316 | struct page *page; |
cc5e31a4 | 5317 | int num_pages; |
d1310b2e | 5318 | |
0b32f4bb | 5319 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 5320 | num_pages = num_extent_pages(eb); |
d1310b2e | 5321 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5322 | page = eb->pages[i]; |
d1310b2e CM |
5323 | SetPageUptodate(page); |
5324 | } | |
d1310b2e | 5325 | } |
d1310b2e | 5326 | |
c2ccfbc6 | 5327 | int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num) |
d1310b2e | 5328 | { |
cc5e31a4 | 5329 | int i; |
d1310b2e CM |
5330 | struct page *page; |
5331 | int err; | |
5332 | int ret = 0; | |
ce9adaa5 CM |
5333 | int locked_pages = 0; |
5334 | int all_uptodate = 1; | |
cc5e31a4 | 5335 | int num_pages; |
727011e0 | 5336 | unsigned long num_reads = 0; |
a86c12c7 | 5337 | struct bio *bio = NULL; |
c8b97818 | 5338 | unsigned long bio_flags = 0; |
c2ccfbc6 | 5339 | struct extent_io_tree *tree = &BTRFS_I(eb->fs_info->btree_inode)->io_tree; |
a86c12c7 | 5340 | |
b4ce94de | 5341 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
5342 | return 0; |
5343 | ||
65ad0104 | 5344 | num_pages = num_extent_pages(eb); |
8436ea91 | 5345 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5346 | page = eb->pages[i]; |
bb82ab88 | 5347 | if (wait == WAIT_NONE) { |
2db04966 | 5348 | if (!trylock_page(page)) |
ce9adaa5 | 5349 | goto unlock_exit; |
d1310b2e CM |
5350 | } else { |
5351 | lock_page(page); | |
5352 | } | |
ce9adaa5 | 5353 | locked_pages++; |
2571e739 LB |
5354 | } |
5355 | /* | |
5356 | * We need to firstly lock all pages to make sure that | |
5357 | * the uptodate bit of our pages won't be affected by | |
5358 | * clear_extent_buffer_uptodate(). | |
5359 | */ | |
8436ea91 | 5360 | for (i = 0; i < num_pages; i++) { |
2571e739 | 5361 | page = eb->pages[i]; |
727011e0 CM |
5362 | if (!PageUptodate(page)) { |
5363 | num_reads++; | |
ce9adaa5 | 5364 | all_uptodate = 0; |
727011e0 | 5365 | } |
ce9adaa5 | 5366 | } |
2571e739 | 5367 | |
ce9adaa5 | 5368 | if (all_uptodate) { |
8436ea91 | 5369 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
5370 | goto unlock_exit; |
5371 | } | |
5372 | ||
656f30db | 5373 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 5374 | eb->read_mirror = 0; |
0b32f4bb | 5375 | atomic_set(&eb->io_pages, num_reads); |
8436ea91 | 5376 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5377 | page = eb->pages[i]; |
baf863b9 | 5378 | |
ce9adaa5 | 5379 | if (!PageUptodate(page)) { |
baf863b9 LB |
5380 | if (ret) { |
5381 | atomic_dec(&eb->io_pages); | |
5382 | unlock_page(page); | |
5383 | continue; | |
5384 | } | |
5385 | ||
f188591e | 5386 | ClearPageError(page); |
a86c12c7 | 5387 | err = __extent_read_full_page(tree, page, |
6af49dbd | 5388 | btree_get_extent, &bio, |
d4c7ca86 | 5389 | mirror_num, &bio_flags, |
1f7ad75b | 5390 | REQ_META); |
baf863b9 | 5391 | if (err) { |
d1310b2e | 5392 | ret = err; |
baf863b9 LB |
5393 | /* |
5394 | * We use &bio in above __extent_read_full_page, | |
5395 | * so we ensure that if it returns error, the | |
5396 | * current page fails to add itself to bio and | |
5397 | * it's been unlocked. | |
5398 | * | |
5399 | * We must dec io_pages by ourselves. | |
5400 | */ | |
5401 | atomic_dec(&eb->io_pages); | |
5402 | } | |
d1310b2e CM |
5403 | } else { |
5404 | unlock_page(page); | |
5405 | } | |
5406 | } | |
5407 | ||
355808c2 | 5408 | if (bio) { |
1f7ad75b | 5409 | err = submit_one_bio(bio, mirror_num, bio_flags); |
79787eaa JM |
5410 | if (err) |
5411 | return err; | |
355808c2 | 5412 | } |
a86c12c7 | 5413 | |
bb82ab88 | 5414 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 5415 | return ret; |
d397712b | 5416 | |
8436ea91 | 5417 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5418 | page = eb->pages[i]; |
d1310b2e | 5419 | wait_on_page_locked(page); |
d397712b | 5420 | if (!PageUptodate(page)) |
d1310b2e | 5421 | ret = -EIO; |
d1310b2e | 5422 | } |
d397712b | 5423 | |
d1310b2e | 5424 | return ret; |
ce9adaa5 CM |
5425 | |
5426 | unlock_exit: | |
d397712b | 5427 | while (locked_pages > 0) { |
ce9adaa5 | 5428 | locked_pages--; |
8436ea91 JB |
5429 | page = eb->pages[locked_pages]; |
5430 | unlock_page(page); | |
ce9adaa5 CM |
5431 | } |
5432 | return ret; | |
d1310b2e | 5433 | } |
d1310b2e | 5434 | |
1cbb1f45 JM |
5435 | void read_extent_buffer(const struct extent_buffer *eb, void *dstv, |
5436 | unsigned long start, unsigned long len) | |
d1310b2e CM |
5437 | { |
5438 | size_t cur; | |
5439 | size_t offset; | |
5440 | struct page *page; | |
5441 | char *kaddr; | |
5442 | char *dst = (char *)dstv; | |
7073017a | 5443 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5444 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
d1310b2e | 5445 | |
f716abd5 LB |
5446 | if (start + len > eb->len) { |
5447 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n", | |
5448 | eb->start, eb->len, start, len); | |
5449 | memset(dst, 0, len); | |
5450 | return; | |
5451 | } | |
d1310b2e | 5452 | |
7073017a | 5453 | offset = offset_in_page(start_offset + start); |
d1310b2e | 5454 | |
d397712b | 5455 | while (len > 0) { |
fb85fc9a | 5456 | page = eb->pages[i]; |
d1310b2e | 5457 | |
09cbfeaf | 5458 | cur = min(len, (PAGE_SIZE - offset)); |
a6591715 | 5459 | kaddr = page_address(page); |
d1310b2e | 5460 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
5461 | |
5462 | dst += cur; | |
5463 | len -= cur; | |
5464 | offset = 0; | |
5465 | i++; | |
5466 | } | |
5467 | } | |
d1310b2e | 5468 | |
1cbb1f45 JM |
5469 | int read_extent_buffer_to_user(const struct extent_buffer *eb, |
5470 | void __user *dstv, | |
5471 | unsigned long start, unsigned long len) | |
550ac1d8 GH |
5472 | { |
5473 | size_t cur; | |
5474 | size_t offset; | |
5475 | struct page *page; | |
5476 | char *kaddr; | |
5477 | char __user *dst = (char __user *)dstv; | |
7073017a | 5478 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5479 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
550ac1d8 GH |
5480 | int ret = 0; |
5481 | ||
5482 | WARN_ON(start > eb->len); | |
5483 | WARN_ON(start + len > eb->start + eb->len); | |
5484 | ||
7073017a | 5485 | offset = offset_in_page(start_offset + start); |
550ac1d8 GH |
5486 | |
5487 | while (len > 0) { | |
fb85fc9a | 5488 | page = eb->pages[i]; |
550ac1d8 | 5489 | |
09cbfeaf | 5490 | cur = min(len, (PAGE_SIZE - offset)); |
550ac1d8 GH |
5491 | kaddr = page_address(page); |
5492 | if (copy_to_user(dst, kaddr + offset, cur)) { | |
5493 | ret = -EFAULT; | |
5494 | break; | |
5495 | } | |
5496 | ||
5497 | dst += cur; | |
5498 | len -= cur; | |
5499 | offset = 0; | |
5500 | i++; | |
5501 | } | |
5502 | ||
5503 | return ret; | |
5504 | } | |
5505 | ||
415b35a5 LB |
5506 | /* |
5507 | * return 0 if the item is found within a page. | |
5508 | * return 1 if the item spans two pages. | |
5509 | * return -EINVAL otherwise. | |
5510 | */ | |
1cbb1f45 JM |
5511 | int map_private_extent_buffer(const struct extent_buffer *eb, |
5512 | unsigned long start, unsigned long min_len, | |
5513 | char **map, unsigned long *map_start, | |
5514 | unsigned long *map_len) | |
d1310b2e | 5515 | { |
cc2c39d6 | 5516 | size_t offset; |
d1310b2e CM |
5517 | char *kaddr; |
5518 | struct page *p; | |
7073017a | 5519 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5520 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
d1310b2e | 5521 | unsigned long end_i = (start_offset + start + min_len - 1) >> |
09cbfeaf | 5522 | PAGE_SHIFT; |
d1310b2e | 5523 | |
f716abd5 LB |
5524 | if (start + min_len > eb->len) { |
5525 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n", | |
5526 | eb->start, eb->len, start, min_len); | |
5527 | return -EINVAL; | |
5528 | } | |
5529 | ||
d1310b2e | 5530 | if (i != end_i) |
415b35a5 | 5531 | return 1; |
d1310b2e CM |
5532 | |
5533 | if (i == 0) { | |
5534 | offset = start_offset; | |
5535 | *map_start = 0; | |
5536 | } else { | |
5537 | offset = 0; | |
09cbfeaf | 5538 | *map_start = ((u64)i << PAGE_SHIFT) - start_offset; |
d1310b2e | 5539 | } |
d397712b | 5540 | |
fb85fc9a | 5541 | p = eb->pages[i]; |
a6591715 | 5542 | kaddr = page_address(p); |
d1310b2e | 5543 | *map = kaddr + offset; |
09cbfeaf | 5544 | *map_len = PAGE_SIZE - offset; |
d1310b2e CM |
5545 | return 0; |
5546 | } | |
d1310b2e | 5547 | |
1cbb1f45 JM |
5548 | int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv, |
5549 | unsigned long start, unsigned long len) | |
d1310b2e CM |
5550 | { |
5551 | size_t cur; | |
5552 | size_t offset; | |
5553 | struct page *page; | |
5554 | char *kaddr; | |
5555 | char *ptr = (char *)ptrv; | |
7073017a | 5556 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5557 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
d1310b2e CM |
5558 | int ret = 0; |
5559 | ||
5560 | WARN_ON(start > eb->len); | |
5561 | WARN_ON(start + len > eb->start + eb->len); | |
5562 | ||
7073017a | 5563 | offset = offset_in_page(start_offset + start); |
d1310b2e | 5564 | |
d397712b | 5565 | while (len > 0) { |
fb85fc9a | 5566 | page = eb->pages[i]; |
d1310b2e | 5567 | |
09cbfeaf | 5568 | cur = min(len, (PAGE_SIZE - offset)); |
d1310b2e | 5569 | |
a6591715 | 5570 | kaddr = page_address(page); |
d1310b2e | 5571 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
5572 | if (ret) |
5573 | break; | |
5574 | ||
5575 | ptr += cur; | |
5576 | len -= cur; | |
5577 | offset = 0; | |
5578 | i++; | |
5579 | } | |
5580 | return ret; | |
5581 | } | |
d1310b2e | 5582 | |
f157bf76 DS |
5583 | void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb, |
5584 | const void *srcv) | |
5585 | { | |
5586 | char *kaddr; | |
5587 | ||
5588 | WARN_ON(!PageUptodate(eb->pages[0])); | |
5589 | kaddr = page_address(eb->pages[0]); | |
5590 | memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv, | |
5591 | BTRFS_FSID_SIZE); | |
5592 | } | |
5593 | ||
5594 | void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv) | |
5595 | { | |
5596 | char *kaddr; | |
5597 | ||
5598 | WARN_ON(!PageUptodate(eb->pages[0])); | |
5599 | kaddr = page_address(eb->pages[0]); | |
5600 | memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv, | |
5601 | BTRFS_FSID_SIZE); | |
5602 | } | |
5603 | ||
d1310b2e CM |
5604 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, |
5605 | unsigned long start, unsigned long len) | |
5606 | { | |
5607 | size_t cur; | |
5608 | size_t offset; | |
5609 | struct page *page; | |
5610 | char *kaddr; | |
5611 | char *src = (char *)srcv; | |
7073017a | 5612 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5613 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
d1310b2e CM |
5614 | |
5615 | WARN_ON(start > eb->len); | |
5616 | WARN_ON(start + len > eb->start + eb->len); | |
5617 | ||
7073017a | 5618 | offset = offset_in_page(start_offset + start); |
d1310b2e | 5619 | |
d397712b | 5620 | while (len > 0) { |
fb85fc9a | 5621 | page = eb->pages[i]; |
d1310b2e CM |
5622 | WARN_ON(!PageUptodate(page)); |
5623 | ||
09cbfeaf | 5624 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 5625 | kaddr = page_address(page); |
d1310b2e | 5626 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
5627 | |
5628 | src += cur; | |
5629 | len -= cur; | |
5630 | offset = 0; | |
5631 | i++; | |
5632 | } | |
5633 | } | |
d1310b2e | 5634 | |
b159fa28 DS |
5635 | void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start, |
5636 | unsigned long len) | |
d1310b2e CM |
5637 | { |
5638 | size_t cur; | |
5639 | size_t offset; | |
5640 | struct page *page; | |
5641 | char *kaddr; | |
7073017a | 5642 | size_t start_offset = offset_in_page(eb->start); |
09cbfeaf | 5643 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; |
d1310b2e CM |
5644 | |
5645 | WARN_ON(start > eb->len); | |
5646 | WARN_ON(start + len > eb->start + eb->len); | |
5647 | ||
7073017a | 5648 | offset = offset_in_page(start_offset + start); |
d1310b2e | 5649 | |
d397712b | 5650 | while (len > 0) { |
fb85fc9a | 5651 | page = eb->pages[i]; |
d1310b2e CM |
5652 | WARN_ON(!PageUptodate(page)); |
5653 | ||
09cbfeaf | 5654 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 5655 | kaddr = page_address(page); |
b159fa28 | 5656 | memset(kaddr + offset, 0, cur); |
d1310b2e CM |
5657 | |
5658 | len -= cur; | |
5659 | offset = 0; | |
5660 | i++; | |
5661 | } | |
5662 | } | |
d1310b2e | 5663 | |
58e8012c DS |
5664 | void copy_extent_buffer_full(struct extent_buffer *dst, |
5665 | struct extent_buffer *src) | |
5666 | { | |
5667 | int i; | |
cc5e31a4 | 5668 | int num_pages; |
58e8012c DS |
5669 | |
5670 | ASSERT(dst->len == src->len); | |
5671 | ||
65ad0104 | 5672 | num_pages = num_extent_pages(dst); |
58e8012c DS |
5673 | for (i = 0; i < num_pages; i++) |
5674 | copy_page(page_address(dst->pages[i]), | |
5675 | page_address(src->pages[i])); | |
5676 | } | |
5677 | ||
d1310b2e CM |
5678 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, |
5679 | unsigned long dst_offset, unsigned long src_offset, | |
5680 | unsigned long len) | |
5681 | { | |
5682 | u64 dst_len = dst->len; | |
5683 | size_t cur; | |
5684 | size_t offset; | |
5685 | struct page *page; | |
5686 | char *kaddr; | |
7073017a | 5687 | size_t start_offset = offset_in_page(dst->start); |
09cbfeaf | 5688 | unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT; |
d1310b2e CM |
5689 | |
5690 | WARN_ON(src->len != dst_len); | |
5691 | ||
7073017a | 5692 | offset = offset_in_page(start_offset + dst_offset); |
d1310b2e | 5693 | |
d397712b | 5694 | while (len > 0) { |
fb85fc9a | 5695 | page = dst->pages[i]; |
d1310b2e CM |
5696 | WARN_ON(!PageUptodate(page)); |
5697 | ||
09cbfeaf | 5698 | cur = min(len, (unsigned long)(PAGE_SIZE - offset)); |
d1310b2e | 5699 | |
a6591715 | 5700 | kaddr = page_address(page); |
d1310b2e | 5701 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
5702 | |
5703 | src_offset += cur; | |
5704 | len -= cur; | |
5705 | offset = 0; | |
5706 | i++; | |
5707 | } | |
5708 | } | |
d1310b2e | 5709 | |
3e1e8bb7 OS |
5710 | /* |
5711 | * eb_bitmap_offset() - calculate the page and offset of the byte containing the | |
5712 | * given bit number | |
5713 | * @eb: the extent buffer | |
5714 | * @start: offset of the bitmap item in the extent buffer | |
5715 | * @nr: bit number | |
5716 | * @page_index: return index of the page in the extent buffer that contains the | |
5717 | * given bit number | |
5718 | * @page_offset: return offset into the page given by page_index | |
5719 | * | |
5720 | * This helper hides the ugliness of finding the byte in an extent buffer which | |
5721 | * contains a given bit. | |
5722 | */ | |
5723 | static inline void eb_bitmap_offset(struct extent_buffer *eb, | |
5724 | unsigned long start, unsigned long nr, | |
5725 | unsigned long *page_index, | |
5726 | size_t *page_offset) | |
5727 | { | |
7073017a | 5728 | size_t start_offset = offset_in_page(eb->start); |
3e1e8bb7 OS |
5729 | size_t byte_offset = BIT_BYTE(nr); |
5730 | size_t offset; | |
5731 | ||
5732 | /* | |
5733 | * The byte we want is the offset of the extent buffer + the offset of | |
5734 | * the bitmap item in the extent buffer + the offset of the byte in the | |
5735 | * bitmap item. | |
5736 | */ | |
5737 | offset = start_offset + start + byte_offset; | |
5738 | ||
09cbfeaf | 5739 | *page_index = offset >> PAGE_SHIFT; |
7073017a | 5740 | *page_offset = offset_in_page(offset); |
3e1e8bb7 OS |
5741 | } |
5742 | ||
5743 | /** | |
5744 | * extent_buffer_test_bit - determine whether a bit in a bitmap item is set | |
5745 | * @eb: the extent buffer | |
5746 | * @start: offset of the bitmap item in the extent buffer | |
5747 | * @nr: bit number to test | |
5748 | */ | |
5749 | int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start, | |
5750 | unsigned long nr) | |
5751 | { | |
2fe1d551 | 5752 | u8 *kaddr; |
3e1e8bb7 OS |
5753 | struct page *page; |
5754 | unsigned long i; | |
5755 | size_t offset; | |
5756 | ||
5757 | eb_bitmap_offset(eb, start, nr, &i, &offset); | |
5758 | page = eb->pages[i]; | |
5759 | WARN_ON(!PageUptodate(page)); | |
5760 | kaddr = page_address(page); | |
5761 | return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1))); | |
5762 | } | |
5763 | ||
5764 | /** | |
5765 | * extent_buffer_bitmap_set - set an area of a bitmap | |
5766 | * @eb: the extent buffer | |
5767 | * @start: offset of the bitmap item in the extent buffer | |
5768 | * @pos: bit number of the first bit | |
5769 | * @len: number of bits to set | |
5770 | */ | |
5771 | void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start, | |
5772 | unsigned long pos, unsigned long len) | |
5773 | { | |
2fe1d551 | 5774 | u8 *kaddr; |
3e1e8bb7 OS |
5775 | struct page *page; |
5776 | unsigned long i; | |
5777 | size_t offset; | |
5778 | const unsigned int size = pos + len; | |
5779 | int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 5780 | u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
5781 | |
5782 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5783 | page = eb->pages[i]; | |
5784 | WARN_ON(!PageUptodate(page)); | |
5785 | kaddr = page_address(page); | |
5786 | ||
5787 | while (len >= bits_to_set) { | |
5788 | kaddr[offset] |= mask_to_set; | |
5789 | len -= bits_to_set; | |
5790 | bits_to_set = BITS_PER_BYTE; | |
9c894696 | 5791 | mask_to_set = ~0; |
09cbfeaf | 5792 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5793 | offset = 0; |
5794 | page = eb->pages[++i]; | |
5795 | WARN_ON(!PageUptodate(page)); | |
5796 | kaddr = page_address(page); | |
5797 | } | |
5798 | } | |
5799 | if (len) { | |
5800 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
5801 | kaddr[offset] |= mask_to_set; | |
5802 | } | |
5803 | } | |
5804 | ||
5805 | ||
5806 | /** | |
5807 | * extent_buffer_bitmap_clear - clear an area of a bitmap | |
5808 | * @eb: the extent buffer | |
5809 | * @start: offset of the bitmap item in the extent buffer | |
5810 | * @pos: bit number of the first bit | |
5811 | * @len: number of bits to clear | |
5812 | */ | |
5813 | void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start, | |
5814 | unsigned long pos, unsigned long len) | |
5815 | { | |
2fe1d551 | 5816 | u8 *kaddr; |
3e1e8bb7 OS |
5817 | struct page *page; |
5818 | unsigned long i; | |
5819 | size_t offset; | |
5820 | const unsigned int size = pos + len; | |
5821 | int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 5822 | u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
5823 | |
5824 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5825 | page = eb->pages[i]; | |
5826 | WARN_ON(!PageUptodate(page)); | |
5827 | kaddr = page_address(page); | |
5828 | ||
5829 | while (len >= bits_to_clear) { | |
5830 | kaddr[offset] &= ~mask_to_clear; | |
5831 | len -= bits_to_clear; | |
5832 | bits_to_clear = BITS_PER_BYTE; | |
9c894696 | 5833 | mask_to_clear = ~0; |
09cbfeaf | 5834 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5835 | offset = 0; |
5836 | page = eb->pages[++i]; | |
5837 | WARN_ON(!PageUptodate(page)); | |
5838 | kaddr = page_address(page); | |
5839 | } | |
5840 | } | |
5841 | if (len) { | |
5842 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
5843 | kaddr[offset] &= ~mask_to_clear; | |
5844 | } | |
5845 | } | |
5846 | ||
3387206f ST |
5847 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
5848 | { | |
5849 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
5850 | return distance < len; | |
5851 | } | |
5852 | ||
d1310b2e CM |
5853 | static void copy_pages(struct page *dst_page, struct page *src_page, |
5854 | unsigned long dst_off, unsigned long src_off, | |
5855 | unsigned long len) | |
5856 | { | |
a6591715 | 5857 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 5858 | char *src_kaddr; |
727011e0 | 5859 | int must_memmove = 0; |
d1310b2e | 5860 | |
3387206f | 5861 | if (dst_page != src_page) { |
a6591715 | 5862 | src_kaddr = page_address(src_page); |
3387206f | 5863 | } else { |
d1310b2e | 5864 | src_kaddr = dst_kaddr; |
727011e0 CM |
5865 | if (areas_overlap(src_off, dst_off, len)) |
5866 | must_memmove = 1; | |
3387206f | 5867 | } |
d1310b2e | 5868 | |
727011e0 CM |
5869 | if (must_memmove) |
5870 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
5871 | else | |
5872 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
5873 | } |
5874 | ||
5875 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5876 | unsigned long src_offset, unsigned long len) | |
5877 | { | |
0b246afa | 5878 | struct btrfs_fs_info *fs_info = dst->fs_info; |
d1310b2e CM |
5879 | size_t cur; |
5880 | size_t dst_off_in_page; | |
5881 | size_t src_off_in_page; | |
7073017a | 5882 | size_t start_offset = offset_in_page(dst->start); |
d1310b2e CM |
5883 | unsigned long dst_i; |
5884 | unsigned long src_i; | |
5885 | ||
5886 | if (src_offset + len > dst->len) { | |
0b246afa | 5887 | btrfs_err(fs_info, |
5d163e0e JM |
5888 | "memmove bogus src_offset %lu move len %lu dst len %lu", |
5889 | src_offset, len, dst->len); | |
290342f6 | 5890 | BUG(); |
d1310b2e CM |
5891 | } |
5892 | if (dst_offset + len > dst->len) { | |
0b246afa | 5893 | btrfs_err(fs_info, |
5d163e0e JM |
5894 | "memmove bogus dst_offset %lu move len %lu dst len %lu", |
5895 | dst_offset, len, dst->len); | |
290342f6 | 5896 | BUG(); |
d1310b2e CM |
5897 | } |
5898 | ||
d397712b | 5899 | while (len > 0) { |
7073017a JT |
5900 | dst_off_in_page = offset_in_page(start_offset + dst_offset); |
5901 | src_off_in_page = offset_in_page(start_offset + src_offset); | |
d1310b2e | 5902 | |
09cbfeaf KS |
5903 | dst_i = (start_offset + dst_offset) >> PAGE_SHIFT; |
5904 | src_i = (start_offset + src_offset) >> PAGE_SHIFT; | |
d1310b2e | 5905 | |
09cbfeaf | 5906 | cur = min(len, (unsigned long)(PAGE_SIZE - |
d1310b2e CM |
5907 | src_off_in_page)); |
5908 | cur = min_t(unsigned long, cur, | |
09cbfeaf | 5909 | (unsigned long)(PAGE_SIZE - dst_off_in_page)); |
d1310b2e | 5910 | |
fb85fc9a | 5911 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5912 | dst_off_in_page, src_off_in_page, cur); |
5913 | ||
5914 | src_offset += cur; | |
5915 | dst_offset += cur; | |
5916 | len -= cur; | |
5917 | } | |
5918 | } | |
d1310b2e CM |
5919 | |
5920 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5921 | unsigned long src_offset, unsigned long len) | |
5922 | { | |
0b246afa | 5923 | struct btrfs_fs_info *fs_info = dst->fs_info; |
d1310b2e CM |
5924 | size_t cur; |
5925 | size_t dst_off_in_page; | |
5926 | size_t src_off_in_page; | |
5927 | unsigned long dst_end = dst_offset + len - 1; | |
5928 | unsigned long src_end = src_offset + len - 1; | |
7073017a | 5929 | size_t start_offset = offset_in_page(dst->start); |
d1310b2e CM |
5930 | unsigned long dst_i; |
5931 | unsigned long src_i; | |
5932 | ||
5933 | if (src_offset + len > dst->len) { | |
0b246afa | 5934 | btrfs_err(fs_info, |
5d163e0e JM |
5935 | "memmove bogus src_offset %lu move len %lu len %lu", |
5936 | src_offset, len, dst->len); | |
290342f6 | 5937 | BUG(); |
d1310b2e CM |
5938 | } |
5939 | if (dst_offset + len > dst->len) { | |
0b246afa | 5940 | btrfs_err(fs_info, |
5d163e0e JM |
5941 | "memmove bogus dst_offset %lu move len %lu len %lu", |
5942 | dst_offset, len, dst->len); | |
290342f6 | 5943 | BUG(); |
d1310b2e | 5944 | } |
727011e0 | 5945 | if (dst_offset < src_offset) { |
d1310b2e CM |
5946 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
5947 | return; | |
5948 | } | |
d397712b | 5949 | while (len > 0) { |
09cbfeaf KS |
5950 | dst_i = (start_offset + dst_end) >> PAGE_SHIFT; |
5951 | src_i = (start_offset + src_end) >> PAGE_SHIFT; | |
d1310b2e | 5952 | |
7073017a JT |
5953 | dst_off_in_page = offset_in_page(start_offset + dst_end); |
5954 | src_off_in_page = offset_in_page(start_offset + src_end); | |
d1310b2e CM |
5955 | |
5956 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
5957 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 5958 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5959 | dst_off_in_page - cur + 1, |
5960 | src_off_in_page - cur + 1, cur); | |
5961 | ||
5962 | dst_end -= cur; | |
5963 | src_end -= cur; | |
5964 | len -= cur; | |
5965 | } | |
5966 | } | |
6af118ce | 5967 | |
f7a52a40 | 5968 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 5969 | { |
6af118ce | 5970 | struct extent_buffer *eb; |
6af118ce | 5971 | |
3083ee2e | 5972 | /* |
01327610 | 5973 | * We need to make sure nobody is attaching this page to an eb right |
3083ee2e JB |
5974 | * now. |
5975 | */ | |
5976 | spin_lock(&page->mapping->private_lock); | |
5977 | if (!PagePrivate(page)) { | |
5978 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 5979 | return 1; |
45f49bce | 5980 | } |
6af118ce | 5981 | |
3083ee2e JB |
5982 | eb = (struct extent_buffer *)page->private; |
5983 | BUG_ON(!eb); | |
19fe0a8b MX |
5984 | |
5985 | /* | |
3083ee2e JB |
5986 | * This is a little awful but should be ok, we need to make sure that |
5987 | * the eb doesn't disappear out from under us while we're looking at | |
5988 | * this page. | |
19fe0a8b | 5989 | */ |
3083ee2e | 5990 | spin_lock(&eb->refs_lock); |
0b32f4bb | 5991 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
5992 | spin_unlock(&eb->refs_lock); |
5993 | spin_unlock(&page->mapping->private_lock); | |
5994 | return 0; | |
b9473439 | 5995 | } |
3083ee2e | 5996 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 5997 | |
19fe0a8b | 5998 | /* |
3083ee2e JB |
5999 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
6000 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 6001 | */ |
3083ee2e JB |
6002 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
6003 | spin_unlock(&eb->refs_lock); | |
6004 | return 0; | |
b9473439 | 6005 | } |
19fe0a8b | 6006 | |
f7a52a40 | 6007 | return release_extent_buffer(eb); |
6af118ce | 6008 | } |