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; | |
9be3395b | 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 | { |
c6100a4b JB |
92 | if (tree->ops && tree->ops->check_extent_io_range) |
93 | tree->ops->check_extent_io_range(tree->private_data, caller, | |
94 | start, end); | |
8d599ae1 | 95 | } |
6d49ba1b ES |
96 | #else |
97 | #define btrfs_leak_debug_add(new, head) do {} while (0) | |
98 | #define btrfs_leak_debug_del(entry) do {} while (0) | |
99 | #define btrfs_leak_debug_check() do {} while (0) | |
8d599ae1 | 100 | #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0) |
4bef0848 | 101 | #endif |
d1310b2e | 102 | |
d1310b2e CM |
103 | #define BUFFER_LRU_MAX 64 |
104 | ||
105 | struct tree_entry { | |
106 | u64 start; | |
107 | u64 end; | |
d1310b2e CM |
108 | struct rb_node rb_node; |
109 | }; | |
110 | ||
111 | struct extent_page_data { | |
112 | struct bio *bio; | |
113 | struct extent_io_tree *tree; | |
771ed689 CM |
114 | /* tells writepage not to lock the state bits for this range |
115 | * it still does the unlocking | |
116 | */ | |
ffbd517d CM |
117 | unsigned int extent_locked:1; |
118 | ||
70fd7614 | 119 | /* tells the submit_bio code to use REQ_SYNC */ |
ffbd517d | 120 | unsigned int sync_io:1; |
d1310b2e CM |
121 | }; |
122 | ||
57599c7e | 123 | static int add_extent_changeset(struct extent_state *state, unsigned bits, |
d38ed27f QW |
124 | struct extent_changeset *changeset, |
125 | int set) | |
126 | { | |
127 | int ret; | |
128 | ||
129 | if (!changeset) | |
57599c7e | 130 | return 0; |
d38ed27f | 131 | if (set && (state->state & bits) == bits) |
57599c7e | 132 | return 0; |
fefdc557 | 133 | if (!set && (state->state & bits) == 0) |
57599c7e | 134 | return 0; |
d38ed27f | 135 | changeset->bytes_changed += state->end - state->start + 1; |
53d32359 | 136 | ret = ulist_add(&changeset->range_changed, state->start, state->end, |
d38ed27f | 137 | GFP_ATOMIC); |
57599c7e | 138 | return ret; |
d38ed27f QW |
139 | } |
140 | ||
aab6e9ed | 141 | static void flush_write_bio(struct extent_page_data *epd); |
e2932ee0 | 142 | |
c2d904e0 JM |
143 | static inline struct btrfs_fs_info * |
144 | tree_fs_info(struct extent_io_tree *tree) | |
145 | { | |
c6100a4b JB |
146 | if (tree->ops) |
147 | return tree->ops->tree_fs_info(tree->private_data); | |
148 | return NULL; | |
c2d904e0 | 149 | } |
0b32f4bb | 150 | |
d1310b2e CM |
151 | int __init extent_io_init(void) |
152 | { | |
837e1972 | 153 | extent_state_cache = kmem_cache_create("btrfs_extent_state", |
9601e3f6 | 154 | sizeof(struct extent_state), 0, |
fba4b697 | 155 | SLAB_MEM_SPREAD, NULL); |
d1310b2e CM |
156 | if (!extent_state_cache) |
157 | return -ENOMEM; | |
158 | ||
837e1972 | 159 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 | 160 | sizeof(struct extent_buffer), 0, |
fba4b697 | 161 | SLAB_MEM_SPREAD, NULL); |
d1310b2e CM |
162 | if (!extent_buffer_cache) |
163 | goto free_state_cache; | |
9be3395b CM |
164 | |
165 | btrfs_bioset = bioset_create(BIO_POOL_SIZE, | |
011067b0 N |
166 | offsetof(struct btrfs_io_bio, bio), |
167 | BIOSET_NEED_BVECS); | |
9be3395b CM |
168 | if (!btrfs_bioset) |
169 | goto free_buffer_cache; | |
b208c2f7 DW |
170 | |
171 | if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE)) | |
172 | goto free_bioset; | |
173 | ||
d1310b2e CM |
174 | return 0; |
175 | ||
b208c2f7 DW |
176 | free_bioset: |
177 | bioset_free(btrfs_bioset); | |
178 | btrfs_bioset = NULL; | |
179 | ||
9be3395b CM |
180 | free_buffer_cache: |
181 | kmem_cache_destroy(extent_buffer_cache); | |
182 | extent_buffer_cache = NULL; | |
183 | ||
d1310b2e CM |
184 | free_state_cache: |
185 | kmem_cache_destroy(extent_state_cache); | |
9be3395b | 186 | extent_state_cache = NULL; |
d1310b2e CM |
187 | return -ENOMEM; |
188 | } | |
189 | ||
e67c718b | 190 | void __cold extent_io_exit(void) |
d1310b2e | 191 | { |
6d49ba1b | 192 | btrfs_leak_debug_check(); |
8c0a8537 KS |
193 | |
194 | /* | |
195 | * Make sure all delayed rcu free are flushed before we | |
196 | * destroy caches. | |
197 | */ | |
198 | rcu_barrier(); | |
5598e900 KM |
199 | kmem_cache_destroy(extent_state_cache); |
200 | kmem_cache_destroy(extent_buffer_cache); | |
9be3395b CM |
201 | if (btrfs_bioset) |
202 | bioset_free(btrfs_bioset); | |
d1310b2e CM |
203 | } |
204 | ||
205 | void extent_io_tree_init(struct extent_io_tree *tree, | |
c6100a4b | 206 | void *private_data) |
d1310b2e | 207 | { |
6bef4d31 | 208 | tree->state = RB_ROOT; |
d1310b2e CM |
209 | tree->ops = NULL; |
210 | tree->dirty_bytes = 0; | |
70dec807 | 211 | spin_lock_init(&tree->lock); |
c6100a4b | 212 | tree->private_data = private_data; |
d1310b2e | 213 | } |
d1310b2e | 214 | |
b2950863 | 215 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
216 | { |
217 | struct extent_state *state; | |
d1310b2e | 218 | |
3ba7ab22 MH |
219 | /* |
220 | * The given mask might be not appropriate for the slab allocator, | |
221 | * drop the unsupported bits | |
222 | */ | |
223 | mask &= ~(__GFP_DMA32|__GFP_HIGHMEM); | |
d1310b2e | 224 | state = kmem_cache_alloc(extent_state_cache, mask); |
2b114d1d | 225 | if (!state) |
d1310b2e CM |
226 | return state; |
227 | state->state = 0; | |
47dc196a | 228 | state->failrec = NULL; |
27a3507d | 229 | RB_CLEAR_NODE(&state->rb_node); |
6d49ba1b | 230 | btrfs_leak_debug_add(&state->leak_list, &states); |
b7ac31b7 | 231 | refcount_set(&state->refs, 1); |
d1310b2e | 232 | init_waitqueue_head(&state->wq); |
143bede5 | 233 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
234 | return state; |
235 | } | |
d1310b2e | 236 | |
4845e44f | 237 | void free_extent_state(struct extent_state *state) |
d1310b2e | 238 | { |
d1310b2e CM |
239 | if (!state) |
240 | return; | |
b7ac31b7 | 241 | if (refcount_dec_and_test(&state->refs)) { |
27a3507d | 242 | WARN_ON(extent_state_in_tree(state)); |
6d49ba1b | 243 | btrfs_leak_debug_del(&state->leak_list); |
143bede5 | 244 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
245 | kmem_cache_free(extent_state_cache, state); |
246 | } | |
247 | } | |
d1310b2e | 248 | |
f2071b21 FM |
249 | static struct rb_node *tree_insert(struct rb_root *root, |
250 | struct rb_node *search_start, | |
251 | u64 offset, | |
12cfbad9 FDBM |
252 | struct rb_node *node, |
253 | struct rb_node ***p_in, | |
254 | struct rb_node **parent_in) | |
d1310b2e | 255 | { |
f2071b21 | 256 | struct rb_node **p; |
d397712b | 257 | struct rb_node *parent = NULL; |
d1310b2e CM |
258 | struct tree_entry *entry; |
259 | ||
12cfbad9 FDBM |
260 | if (p_in && parent_in) { |
261 | p = *p_in; | |
262 | parent = *parent_in; | |
263 | goto do_insert; | |
264 | } | |
265 | ||
f2071b21 | 266 | p = search_start ? &search_start : &root->rb_node; |
d397712b | 267 | while (*p) { |
d1310b2e CM |
268 | parent = *p; |
269 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
270 | ||
271 | if (offset < entry->start) | |
272 | p = &(*p)->rb_left; | |
273 | else if (offset > entry->end) | |
274 | p = &(*p)->rb_right; | |
275 | else | |
276 | return parent; | |
277 | } | |
278 | ||
12cfbad9 | 279 | do_insert: |
d1310b2e CM |
280 | rb_link_node(node, parent, p); |
281 | rb_insert_color(node, root); | |
282 | return NULL; | |
283 | } | |
284 | ||
80ea96b1 | 285 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
12cfbad9 FDBM |
286 | struct rb_node **prev_ret, |
287 | struct rb_node **next_ret, | |
288 | struct rb_node ***p_ret, | |
289 | struct rb_node **parent_ret) | |
d1310b2e | 290 | { |
80ea96b1 | 291 | struct rb_root *root = &tree->state; |
12cfbad9 | 292 | struct rb_node **n = &root->rb_node; |
d1310b2e CM |
293 | struct rb_node *prev = NULL; |
294 | struct rb_node *orig_prev = NULL; | |
295 | struct tree_entry *entry; | |
296 | struct tree_entry *prev_entry = NULL; | |
297 | ||
12cfbad9 FDBM |
298 | while (*n) { |
299 | prev = *n; | |
300 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
d1310b2e CM |
301 | prev_entry = entry; |
302 | ||
303 | if (offset < entry->start) | |
12cfbad9 | 304 | n = &(*n)->rb_left; |
d1310b2e | 305 | else if (offset > entry->end) |
12cfbad9 | 306 | n = &(*n)->rb_right; |
d397712b | 307 | else |
12cfbad9 | 308 | return *n; |
d1310b2e CM |
309 | } |
310 | ||
12cfbad9 FDBM |
311 | if (p_ret) |
312 | *p_ret = n; | |
313 | if (parent_ret) | |
314 | *parent_ret = prev; | |
315 | ||
d1310b2e CM |
316 | if (prev_ret) { |
317 | orig_prev = prev; | |
d397712b | 318 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
319 | prev = rb_next(prev); |
320 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
321 | } | |
322 | *prev_ret = prev; | |
323 | prev = orig_prev; | |
324 | } | |
325 | ||
326 | if (next_ret) { | |
327 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 328 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
329 | prev = rb_prev(prev); |
330 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
331 | } | |
332 | *next_ret = prev; | |
333 | } | |
334 | return NULL; | |
335 | } | |
336 | ||
12cfbad9 FDBM |
337 | static inline struct rb_node * |
338 | tree_search_for_insert(struct extent_io_tree *tree, | |
339 | u64 offset, | |
340 | struct rb_node ***p_ret, | |
341 | struct rb_node **parent_ret) | |
d1310b2e | 342 | { |
70dec807 | 343 | struct rb_node *prev = NULL; |
d1310b2e | 344 | struct rb_node *ret; |
70dec807 | 345 | |
12cfbad9 | 346 | ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret); |
d397712b | 347 | if (!ret) |
d1310b2e CM |
348 | return prev; |
349 | return ret; | |
350 | } | |
351 | ||
12cfbad9 FDBM |
352 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
353 | u64 offset) | |
354 | { | |
355 | return tree_search_for_insert(tree, offset, NULL, NULL); | |
356 | } | |
357 | ||
9ed74f2d JB |
358 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, |
359 | struct extent_state *other) | |
360 | { | |
361 | if (tree->ops && tree->ops->merge_extent_hook) | |
c6100a4b | 362 | tree->ops->merge_extent_hook(tree->private_data, new, other); |
9ed74f2d JB |
363 | } |
364 | ||
d1310b2e CM |
365 | /* |
366 | * utility function to look for merge candidates inside a given range. | |
367 | * Any extents with matching state are merged together into a single | |
368 | * extent in the tree. Extents with EXTENT_IO in their state field | |
369 | * are not merged because the end_io handlers need to be able to do | |
370 | * operations on them without sleeping (or doing allocations/splits). | |
371 | * | |
372 | * This should be called with the tree lock held. | |
373 | */ | |
1bf85046 JM |
374 | static void merge_state(struct extent_io_tree *tree, |
375 | struct extent_state *state) | |
d1310b2e CM |
376 | { |
377 | struct extent_state *other; | |
378 | struct rb_node *other_node; | |
379 | ||
5b21f2ed | 380 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
1bf85046 | 381 | return; |
d1310b2e CM |
382 | |
383 | other_node = rb_prev(&state->rb_node); | |
384 | if (other_node) { | |
385 | other = rb_entry(other_node, struct extent_state, rb_node); | |
386 | if (other->end == state->start - 1 && | |
387 | other->state == state->state) { | |
9ed74f2d | 388 | merge_cb(tree, state, other); |
d1310b2e | 389 | state->start = other->start; |
d1310b2e | 390 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 391 | RB_CLEAR_NODE(&other->rb_node); |
d1310b2e CM |
392 | free_extent_state(other); |
393 | } | |
394 | } | |
395 | other_node = rb_next(&state->rb_node); | |
396 | if (other_node) { | |
397 | other = rb_entry(other_node, struct extent_state, rb_node); | |
398 | if (other->start == state->end + 1 && | |
399 | other->state == state->state) { | |
9ed74f2d | 400 | merge_cb(tree, state, other); |
df98b6e2 | 401 | state->end = other->end; |
df98b6e2 | 402 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 403 | RB_CLEAR_NODE(&other->rb_node); |
df98b6e2 | 404 | free_extent_state(other); |
d1310b2e CM |
405 | } |
406 | } | |
d1310b2e CM |
407 | } |
408 | ||
1bf85046 | 409 | static void set_state_cb(struct extent_io_tree *tree, |
9ee49a04 | 410 | struct extent_state *state, unsigned *bits) |
291d673e | 411 | { |
1bf85046 | 412 | if (tree->ops && tree->ops->set_bit_hook) |
c6100a4b | 413 | tree->ops->set_bit_hook(tree->private_data, state, bits); |
291d673e CM |
414 | } |
415 | ||
416 | static void clear_state_cb(struct extent_io_tree *tree, | |
9ee49a04 | 417 | struct extent_state *state, unsigned *bits) |
291d673e | 418 | { |
9ed74f2d | 419 | if (tree->ops && tree->ops->clear_bit_hook) |
c6100a4b | 420 | tree->ops->clear_bit_hook(tree->private_data, state, bits); |
291d673e CM |
421 | } |
422 | ||
3150b699 | 423 | static void set_state_bits(struct extent_io_tree *tree, |
d38ed27f QW |
424 | struct extent_state *state, unsigned *bits, |
425 | struct extent_changeset *changeset); | |
3150b699 | 426 | |
d1310b2e CM |
427 | /* |
428 | * insert an extent_state struct into the tree. 'bits' are set on the | |
429 | * struct before it is inserted. | |
430 | * | |
431 | * This may return -EEXIST if the extent is already there, in which case the | |
432 | * state struct is freed. | |
433 | * | |
434 | * The tree lock is not taken internally. This is a utility function and | |
435 | * probably isn't what you want to call (see set/clear_extent_bit). | |
436 | */ | |
437 | static int insert_state(struct extent_io_tree *tree, | |
438 | struct extent_state *state, u64 start, u64 end, | |
12cfbad9 FDBM |
439 | struct rb_node ***p, |
440 | struct rb_node **parent, | |
d38ed27f | 441 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e CM |
442 | { |
443 | struct rb_node *node; | |
444 | ||
31b1a2bd | 445 | if (end < start) |
efe120a0 | 446 | WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n", |
c1c9ff7c | 447 | end, start); |
d1310b2e CM |
448 | state->start = start; |
449 | state->end = end; | |
9ed74f2d | 450 | |
d38ed27f | 451 | set_state_bits(tree, state, bits, changeset); |
3150b699 | 452 | |
f2071b21 | 453 | node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent); |
d1310b2e CM |
454 | if (node) { |
455 | struct extent_state *found; | |
456 | found = rb_entry(node, struct extent_state, rb_node); | |
62e85577 | 457 | pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n", |
c1c9ff7c | 458 | found->start, found->end, start, end); |
d1310b2e CM |
459 | return -EEXIST; |
460 | } | |
461 | merge_state(tree, state); | |
462 | return 0; | |
463 | } | |
464 | ||
1bf85046 | 465 | static void split_cb(struct extent_io_tree *tree, struct extent_state *orig, |
9ed74f2d JB |
466 | u64 split) |
467 | { | |
468 | if (tree->ops && tree->ops->split_extent_hook) | |
c6100a4b | 469 | tree->ops->split_extent_hook(tree->private_data, orig, split); |
9ed74f2d JB |
470 | } |
471 | ||
d1310b2e CM |
472 | /* |
473 | * split a given extent state struct in two, inserting the preallocated | |
474 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
475 | * offset inside 'orig' where it should be split. | |
476 | * | |
477 | * Before calling, | |
478 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
479 | * are two extent state structs in the tree: | |
480 | * prealloc: [orig->start, split - 1] | |
481 | * orig: [ split, orig->end ] | |
482 | * | |
483 | * The tree locks are not taken by this function. They need to be held | |
484 | * by the caller. | |
485 | */ | |
486 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
487 | struct extent_state *prealloc, u64 split) | |
488 | { | |
489 | struct rb_node *node; | |
9ed74f2d JB |
490 | |
491 | split_cb(tree, orig, split); | |
492 | ||
d1310b2e CM |
493 | prealloc->start = orig->start; |
494 | prealloc->end = split - 1; | |
495 | prealloc->state = orig->state; | |
496 | orig->start = split; | |
497 | ||
f2071b21 FM |
498 | node = tree_insert(&tree->state, &orig->rb_node, prealloc->end, |
499 | &prealloc->rb_node, NULL, NULL); | |
d1310b2e | 500 | if (node) { |
d1310b2e CM |
501 | free_extent_state(prealloc); |
502 | return -EEXIST; | |
503 | } | |
504 | return 0; | |
505 | } | |
506 | ||
cdc6a395 LZ |
507 | static struct extent_state *next_state(struct extent_state *state) |
508 | { | |
509 | struct rb_node *next = rb_next(&state->rb_node); | |
510 | if (next) | |
511 | return rb_entry(next, struct extent_state, rb_node); | |
512 | else | |
513 | return NULL; | |
514 | } | |
515 | ||
d1310b2e CM |
516 | /* |
517 | * utility function to clear some bits in an extent state struct. | |
1b303fc0 | 518 | * it will optionally wake up any one waiting on this state (wake == 1). |
d1310b2e CM |
519 | * |
520 | * If no bits are set on the state struct after clearing things, the | |
521 | * struct is freed and removed from the tree | |
522 | */ | |
cdc6a395 LZ |
523 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
524 | struct extent_state *state, | |
fefdc557 QW |
525 | unsigned *bits, int wake, |
526 | struct extent_changeset *changeset) | |
d1310b2e | 527 | { |
cdc6a395 | 528 | struct extent_state *next; |
9ee49a04 | 529 | unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS; |
57599c7e | 530 | int ret; |
d1310b2e | 531 | |
0ca1f7ce | 532 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
533 | u64 range = state->end - state->start + 1; |
534 | WARN_ON(range > tree->dirty_bytes); | |
535 | tree->dirty_bytes -= range; | |
536 | } | |
291d673e | 537 | clear_state_cb(tree, state, bits); |
57599c7e DS |
538 | ret = add_extent_changeset(state, bits_to_clear, changeset, 0); |
539 | BUG_ON(ret < 0); | |
32c00aff | 540 | state->state &= ~bits_to_clear; |
d1310b2e CM |
541 | if (wake) |
542 | wake_up(&state->wq); | |
0ca1f7ce | 543 | if (state->state == 0) { |
cdc6a395 | 544 | next = next_state(state); |
27a3507d | 545 | if (extent_state_in_tree(state)) { |
d1310b2e | 546 | rb_erase(&state->rb_node, &tree->state); |
27a3507d | 547 | RB_CLEAR_NODE(&state->rb_node); |
d1310b2e CM |
548 | free_extent_state(state); |
549 | } else { | |
550 | WARN_ON(1); | |
551 | } | |
552 | } else { | |
553 | merge_state(tree, state); | |
cdc6a395 | 554 | next = next_state(state); |
d1310b2e | 555 | } |
cdc6a395 | 556 | return next; |
d1310b2e CM |
557 | } |
558 | ||
8233767a XG |
559 | static struct extent_state * |
560 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
561 | { | |
562 | if (!prealloc) | |
563 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
564 | ||
565 | return prealloc; | |
566 | } | |
567 | ||
48a3b636 | 568 | static void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
c2d904e0 | 569 | { |
5d163e0e JM |
570 | btrfs_panic(tree_fs_info(tree), err, |
571 | "Locking error: Extent tree was modified by another thread while locked."); | |
c2d904e0 JM |
572 | } |
573 | ||
d1310b2e CM |
574 | /* |
575 | * clear some bits on a range in the tree. This may require splitting | |
576 | * or inserting elements in the tree, so the gfp mask is used to | |
577 | * indicate which allocations or sleeping are allowed. | |
578 | * | |
579 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
580 | * the given range from the tree regardless of state (ie for truncate). | |
581 | * | |
582 | * the range [start, end] is inclusive. | |
583 | * | |
6763af84 | 584 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e | 585 | */ |
66b0c887 | 586 | int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
fefdc557 QW |
587 | unsigned bits, int wake, int delete, |
588 | struct extent_state **cached_state, | |
589 | gfp_t mask, struct extent_changeset *changeset) | |
d1310b2e CM |
590 | { |
591 | struct extent_state *state; | |
2c64c53d | 592 | struct extent_state *cached; |
d1310b2e CM |
593 | struct extent_state *prealloc = NULL; |
594 | struct rb_node *node; | |
5c939df5 | 595 | u64 last_end; |
d1310b2e | 596 | int err; |
2ac55d41 | 597 | int clear = 0; |
d1310b2e | 598 | |
a5dee37d | 599 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 600 | |
7ee9e440 JB |
601 | if (bits & EXTENT_DELALLOC) |
602 | bits |= EXTENT_NORESERVE; | |
603 | ||
0ca1f7ce YZ |
604 | if (delete) |
605 | bits |= ~EXTENT_CTLBITS; | |
606 | bits |= EXTENT_FIRST_DELALLOC; | |
607 | ||
2ac55d41 JB |
608 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
609 | clear = 1; | |
d1310b2e | 610 | again: |
d0164adc | 611 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
c7bc6319 FM |
612 | /* |
613 | * Don't care for allocation failure here because we might end | |
614 | * up not needing the pre-allocated extent state at all, which | |
615 | * is the case if we only have in the tree extent states that | |
616 | * cover our input range and don't cover too any other range. | |
617 | * If we end up needing a new extent state we allocate it later. | |
618 | */ | |
d1310b2e | 619 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
620 | } |
621 | ||
cad321ad | 622 | spin_lock(&tree->lock); |
2c64c53d CM |
623 | if (cached_state) { |
624 | cached = *cached_state; | |
2ac55d41 JB |
625 | |
626 | if (clear) { | |
627 | *cached_state = NULL; | |
628 | cached_state = NULL; | |
629 | } | |
630 | ||
27a3507d FM |
631 | if (cached && extent_state_in_tree(cached) && |
632 | cached->start <= start && cached->end > start) { | |
2ac55d41 | 633 | if (clear) |
b7ac31b7 | 634 | refcount_dec(&cached->refs); |
2c64c53d | 635 | state = cached; |
42daec29 | 636 | goto hit_next; |
2c64c53d | 637 | } |
2ac55d41 JB |
638 | if (clear) |
639 | free_extent_state(cached); | |
2c64c53d | 640 | } |
d1310b2e CM |
641 | /* |
642 | * this search will find the extents that end after | |
643 | * our range starts | |
644 | */ | |
80ea96b1 | 645 | node = tree_search(tree, start); |
d1310b2e CM |
646 | if (!node) |
647 | goto out; | |
648 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 649 | hit_next: |
d1310b2e CM |
650 | if (state->start > end) |
651 | goto out; | |
652 | WARN_ON(state->end < start); | |
5c939df5 | 653 | last_end = state->end; |
d1310b2e | 654 | |
0449314a | 655 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
656 | if (!(state->state & bits)) { |
657 | state = next_state(state); | |
0449314a | 658 | goto next; |
cdc6a395 | 659 | } |
0449314a | 660 | |
d1310b2e CM |
661 | /* |
662 | * | ---- desired range ---- | | |
663 | * | state | or | |
664 | * | ------------- state -------------- | | |
665 | * | |
666 | * We need to split the extent we found, and may flip | |
667 | * bits on second half. | |
668 | * | |
669 | * If the extent we found extends past our range, we | |
670 | * just split and search again. It'll get split again | |
671 | * the next time though. | |
672 | * | |
673 | * If the extent we found is inside our range, we clear | |
674 | * the desired bit on it. | |
675 | */ | |
676 | ||
677 | if (state->start < start) { | |
8233767a XG |
678 | prealloc = alloc_extent_state_atomic(prealloc); |
679 | BUG_ON(!prealloc); | |
d1310b2e | 680 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
681 | if (err) |
682 | extent_io_tree_panic(tree, err); | |
683 | ||
d1310b2e CM |
684 | prealloc = NULL; |
685 | if (err) | |
686 | goto out; | |
687 | if (state->end <= end) { | |
fefdc557 QW |
688 | state = clear_state_bit(tree, state, &bits, wake, |
689 | changeset); | |
d1ac6e41 | 690 | goto next; |
d1310b2e CM |
691 | } |
692 | goto search_again; | |
693 | } | |
694 | /* | |
695 | * | ---- desired range ---- | | |
696 | * | state | | |
697 | * We need to split the extent, and clear the bit | |
698 | * on the first half | |
699 | */ | |
700 | if (state->start <= end && state->end > end) { | |
8233767a XG |
701 | prealloc = alloc_extent_state_atomic(prealloc); |
702 | BUG_ON(!prealloc); | |
d1310b2e | 703 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
704 | if (err) |
705 | extent_io_tree_panic(tree, err); | |
706 | ||
d1310b2e CM |
707 | if (wake) |
708 | wake_up(&state->wq); | |
42daec29 | 709 | |
fefdc557 | 710 | clear_state_bit(tree, prealloc, &bits, wake, changeset); |
9ed74f2d | 711 | |
d1310b2e CM |
712 | prealloc = NULL; |
713 | goto out; | |
714 | } | |
42daec29 | 715 | |
fefdc557 | 716 | state = clear_state_bit(tree, state, &bits, wake, changeset); |
0449314a | 717 | next: |
5c939df5 YZ |
718 | if (last_end == (u64)-1) |
719 | goto out; | |
720 | start = last_end + 1; | |
cdc6a395 | 721 | if (start <= end && state && !need_resched()) |
692e5759 | 722 | goto hit_next; |
d1310b2e CM |
723 | |
724 | search_again: | |
725 | if (start > end) | |
726 | goto out; | |
cad321ad | 727 | spin_unlock(&tree->lock); |
d0164adc | 728 | if (gfpflags_allow_blocking(mask)) |
d1310b2e CM |
729 | cond_resched(); |
730 | goto again; | |
7ab5cb2a DS |
731 | |
732 | out: | |
733 | spin_unlock(&tree->lock); | |
734 | if (prealloc) | |
735 | free_extent_state(prealloc); | |
736 | ||
737 | return 0; | |
738 | ||
d1310b2e | 739 | } |
d1310b2e | 740 | |
143bede5 JM |
741 | static void wait_on_state(struct extent_io_tree *tree, |
742 | struct extent_state *state) | |
641f5219 CH |
743 | __releases(tree->lock) |
744 | __acquires(tree->lock) | |
d1310b2e CM |
745 | { |
746 | DEFINE_WAIT(wait); | |
747 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 748 | spin_unlock(&tree->lock); |
d1310b2e | 749 | schedule(); |
cad321ad | 750 | spin_lock(&tree->lock); |
d1310b2e | 751 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
752 | } |
753 | ||
754 | /* | |
755 | * waits for one or more bits to clear on a range in the state tree. | |
756 | * The range [start, end] is inclusive. | |
757 | * The tree lock is taken by this function | |
758 | */ | |
41074888 DS |
759 | static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
760 | unsigned long bits) | |
d1310b2e CM |
761 | { |
762 | struct extent_state *state; | |
763 | struct rb_node *node; | |
764 | ||
a5dee37d | 765 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 766 | |
cad321ad | 767 | spin_lock(&tree->lock); |
d1310b2e CM |
768 | again: |
769 | while (1) { | |
770 | /* | |
771 | * this search will find all the extents that end after | |
772 | * our range starts | |
773 | */ | |
80ea96b1 | 774 | node = tree_search(tree, start); |
c50d3e71 | 775 | process_node: |
d1310b2e CM |
776 | if (!node) |
777 | break; | |
778 | ||
779 | state = rb_entry(node, struct extent_state, rb_node); | |
780 | ||
781 | if (state->start > end) | |
782 | goto out; | |
783 | ||
784 | if (state->state & bits) { | |
785 | start = state->start; | |
b7ac31b7 | 786 | refcount_inc(&state->refs); |
d1310b2e CM |
787 | wait_on_state(tree, state); |
788 | free_extent_state(state); | |
789 | goto again; | |
790 | } | |
791 | start = state->end + 1; | |
792 | ||
793 | if (start > end) | |
794 | break; | |
795 | ||
c50d3e71 FM |
796 | if (!cond_resched_lock(&tree->lock)) { |
797 | node = rb_next(node); | |
798 | goto process_node; | |
799 | } | |
d1310b2e CM |
800 | } |
801 | out: | |
cad321ad | 802 | spin_unlock(&tree->lock); |
d1310b2e | 803 | } |
d1310b2e | 804 | |
1bf85046 | 805 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 806 | struct extent_state *state, |
d38ed27f | 807 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e | 808 | { |
9ee49a04 | 809 | unsigned bits_to_set = *bits & ~EXTENT_CTLBITS; |
57599c7e | 810 | int ret; |
9ed74f2d | 811 | |
1bf85046 | 812 | set_state_cb(tree, state, bits); |
0ca1f7ce | 813 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
814 | u64 range = state->end - state->start + 1; |
815 | tree->dirty_bytes += range; | |
816 | } | |
57599c7e DS |
817 | ret = add_extent_changeset(state, bits_to_set, changeset, 1); |
818 | BUG_ON(ret < 0); | |
0ca1f7ce | 819 | state->state |= bits_to_set; |
d1310b2e CM |
820 | } |
821 | ||
e38e2ed7 FM |
822 | static void cache_state_if_flags(struct extent_state *state, |
823 | struct extent_state **cached_ptr, | |
9ee49a04 | 824 | unsigned flags) |
2c64c53d CM |
825 | { |
826 | if (cached_ptr && !(*cached_ptr)) { | |
e38e2ed7 | 827 | if (!flags || (state->state & flags)) { |
2c64c53d | 828 | *cached_ptr = state; |
b7ac31b7 | 829 | refcount_inc(&state->refs); |
2c64c53d CM |
830 | } |
831 | } | |
832 | } | |
833 | ||
e38e2ed7 FM |
834 | static void cache_state(struct extent_state *state, |
835 | struct extent_state **cached_ptr) | |
836 | { | |
837 | return cache_state_if_flags(state, cached_ptr, | |
838 | EXTENT_IOBITS | EXTENT_BOUNDARY); | |
839 | } | |
840 | ||
d1310b2e | 841 | /* |
1edbb734 CM |
842 | * set some bits on a range in the tree. This may require allocations or |
843 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 844 | * |
1edbb734 CM |
845 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
846 | * part of the range already has the desired bits set. The start of the | |
847 | * existing range is returned in failed_start in this case. | |
d1310b2e | 848 | * |
1edbb734 | 849 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 850 | */ |
1edbb734 | 851 | |
3fbe5c02 JM |
852 | static int __must_check |
853 | __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 854 | unsigned bits, unsigned exclusive_bits, |
41074888 | 855 | u64 *failed_start, struct extent_state **cached_state, |
d38ed27f | 856 | gfp_t mask, struct extent_changeset *changeset) |
d1310b2e CM |
857 | { |
858 | struct extent_state *state; | |
859 | struct extent_state *prealloc = NULL; | |
860 | struct rb_node *node; | |
12cfbad9 FDBM |
861 | struct rb_node **p; |
862 | struct rb_node *parent; | |
d1310b2e | 863 | int err = 0; |
d1310b2e CM |
864 | u64 last_start; |
865 | u64 last_end; | |
42daec29 | 866 | |
a5dee37d | 867 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 868 | |
0ca1f7ce | 869 | bits |= EXTENT_FIRST_DELALLOC; |
d1310b2e | 870 | again: |
d0164adc | 871 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
059f791c DS |
872 | /* |
873 | * Don't care for allocation failure here because we might end | |
874 | * up not needing the pre-allocated extent state at all, which | |
875 | * is the case if we only have in the tree extent states that | |
876 | * cover our input range and don't cover too any other range. | |
877 | * If we end up needing a new extent state we allocate it later. | |
878 | */ | |
d1310b2e | 879 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
880 | } |
881 | ||
cad321ad | 882 | spin_lock(&tree->lock); |
9655d298 CM |
883 | if (cached_state && *cached_state) { |
884 | state = *cached_state; | |
df98b6e2 | 885 | if (state->start <= start && state->end > start && |
27a3507d | 886 | extent_state_in_tree(state)) { |
9655d298 CM |
887 | node = &state->rb_node; |
888 | goto hit_next; | |
889 | } | |
890 | } | |
d1310b2e CM |
891 | /* |
892 | * this search will find all the extents that end after | |
893 | * our range starts. | |
894 | */ | |
12cfbad9 | 895 | node = tree_search_for_insert(tree, start, &p, &parent); |
d1310b2e | 896 | if (!node) { |
8233767a XG |
897 | prealloc = alloc_extent_state_atomic(prealloc); |
898 | BUG_ON(!prealloc); | |
12cfbad9 | 899 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 900 | &p, &parent, &bits, changeset); |
c2d904e0 JM |
901 | if (err) |
902 | extent_io_tree_panic(tree, err); | |
903 | ||
c42ac0bc | 904 | cache_state(prealloc, cached_state); |
d1310b2e | 905 | prealloc = NULL; |
d1310b2e CM |
906 | goto out; |
907 | } | |
d1310b2e | 908 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 909 | hit_next: |
d1310b2e CM |
910 | last_start = state->start; |
911 | last_end = state->end; | |
912 | ||
913 | /* | |
914 | * | ---- desired range ---- | | |
915 | * | state | | |
916 | * | |
917 | * Just lock what we found and keep going | |
918 | */ | |
919 | if (state->start == start && state->end <= end) { | |
1edbb734 | 920 | if (state->state & exclusive_bits) { |
d1310b2e CM |
921 | *failed_start = state->start; |
922 | err = -EEXIST; | |
923 | goto out; | |
924 | } | |
42daec29 | 925 | |
d38ed27f | 926 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 927 | cache_state(state, cached_state); |
d1310b2e | 928 | merge_state(tree, state); |
5c939df5 YZ |
929 | if (last_end == (u64)-1) |
930 | goto out; | |
931 | start = last_end + 1; | |
d1ac6e41 LB |
932 | state = next_state(state); |
933 | if (start < end && state && state->start == start && | |
934 | !need_resched()) | |
935 | goto hit_next; | |
d1310b2e CM |
936 | goto search_again; |
937 | } | |
938 | ||
939 | /* | |
940 | * | ---- desired range ---- | | |
941 | * | state | | |
942 | * or | |
943 | * | ------------- state -------------- | | |
944 | * | |
945 | * We need to split the extent we found, and may flip bits on | |
946 | * second half. | |
947 | * | |
948 | * If the extent we found extends past our | |
949 | * range, we just split and search again. It'll get split | |
950 | * again the next time though. | |
951 | * | |
952 | * If the extent we found is inside our range, we set the | |
953 | * desired bit on it. | |
954 | */ | |
955 | if (state->start < start) { | |
1edbb734 | 956 | if (state->state & exclusive_bits) { |
d1310b2e CM |
957 | *failed_start = start; |
958 | err = -EEXIST; | |
959 | goto out; | |
960 | } | |
8233767a XG |
961 | |
962 | prealloc = alloc_extent_state_atomic(prealloc); | |
963 | BUG_ON(!prealloc); | |
d1310b2e | 964 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
965 | if (err) |
966 | extent_io_tree_panic(tree, err); | |
967 | ||
d1310b2e CM |
968 | prealloc = NULL; |
969 | if (err) | |
970 | goto out; | |
971 | if (state->end <= end) { | |
d38ed27f | 972 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 973 | cache_state(state, cached_state); |
d1310b2e | 974 | merge_state(tree, state); |
5c939df5 YZ |
975 | if (last_end == (u64)-1) |
976 | goto out; | |
977 | start = last_end + 1; | |
d1ac6e41 LB |
978 | state = next_state(state); |
979 | if (start < end && state && state->start == start && | |
980 | !need_resched()) | |
981 | goto hit_next; | |
d1310b2e CM |
982 | } |
983 | goto search_again; | |
984 | } | |
985 | /* | |
986 | * | ---- desired range ---- | | |
987 | * | state | or | state | | |
988 | * | |
989 | * There's a hole, we need to insert something in it and | |
990 | * ignore the extent we found. | |
991 | */ | |
992 | if (state->start > start) { | |
993 | u64 this_end; | |
994 | if (end < last_start) | |
995 | this_end = end; | |
996 | else | |
d397712b | 997 | this_end = last_start - 1; |
8233767a XG |
998 | |
999 | prealloc = alloc_extent_state_atomic(prealloc); | |
1000 | BUG_ON(!prealloc); | |
c7f895a2 XG |
1001 | |
1002 | /* | |
1003 | * Avoid to free 'prealloc' if it can be merged with | |
1004 | * the later extent. | |
1005 | */ | |
d1310b2e | 1006 | err = insert_state(tree, prealloc, start, this_end, |
d38ed27f | 1007 | NULL, NULL, &bits, changeset); |
c2d904e0 JM |
1008 | if (err) |
1009 | extent_io_tree_panic(tree, err); | |
1010 | ||
9ed74f2d JB |
1011 | cache_state(prealloc, cached_state); |
1012 | prealloc = NULL; | |
d1310b2e CM |
1013 | start = this_end + 1; |
1014 | goto search_again; | |
1015 | } | |
1016 | /* | |
1017 | * | ---- desired range ---- | | |
1018 | * | state | | |
1019 | * We need to split the extent, and set the bit | |
1020 | * on the first half | |
1021 | */ | |
1022 | if (state->start <= end && state->end > end) { | |
1edbb734 | 1023 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1024 | *failed_start = start; |
1025 | err = -EEXIST; | |
1026 | goto out; | |
1027 | } | |
8233767a XG |
1028 | |
1029 | prealloc = alloc_extent_state_atomic(prealloc); | |
1030 | BUG_ON(!prealloc); | |
d1310b2e | 1031 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
1032 | if (err) |
1033 | extent_io_tree_panic(tree, err); | |
d1310b2e | 1034 | |
d38ed27f | 1035 | set_state_bits(tree, prealloc, &bits, changeset); |
2c64c53d | 1036 | cache_state(prealloc, cached_state); |
d1310b2e CM |
1037 | merge_state(tree, prealloc); |
1038 | prealloc = NULL; | |
1039 | goto out; | |
1040 | } | |
1041 | ||
b5a4ba14 DS |
1042 | search_again: |
1043 | if (start > end) | |
1044 | goto out; | |
1045 | spin_unlock(&tree->lock); | |
1046 | if (gfpflags_allow_blocking(mask)) | |
1047 | cond_resched(); | |
1048 | goto again; | |
d1310b2e CM |
1049 | |
1050 | out: | |
cad321ad | 1051 | spin_unlock(&tree->lock); |
d1310b2e CM |
1052 | if (prealloc) |
1053 | free_extent_state(prealloc); | |
1054 | ||
1055 | return err; | |
1056 | ||
d1310b2e | 1057 | } |
d1310b2e | 1058 | |
41074888 | 1059 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
9ee49a04 | 1060 | unsigned bits, u64 * failed_start, |
41074888 | 1061 | struct extent_state **cached_state, gfp_t mask) |
3fbe5c02 JM |
1062 | { |
1063 | return __set_extent_bit(tree, start, end, bits, 0, failed_start, | |
d38ed27f | 1064 | cached_state, mask, NULL); |
3fbe5c02 JM |
1065 | } |
1066 | ||
1067 | ||
462d6fac | 1068 | /** |
10983f2e LB |
1069 | * convert_extent_bit - convert all bits in a given range from one bit to |
1070 | * another | |
462d6fac JB |
1071 | * @tree: the io tree to search |
1072 | * @start: the start offset in bytes | |
1073 | * @end: the end offset in bytes (inclusive) | |
1074 | * @bits: the bits to set in this range | |
1075 | * @clear_bits: the bits to clear in this range | |
e6138876 | 1076 | * @cached_state: state that we're going to cache |
462d6fac JB |
1077 | * |
1078 | * This will go through and set bits for the given range. If any states exist | |
1079 | * already in this range they are set with the given bit and cleared of the | |
1080 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
1081 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
1082 | * boundary bits like LOCK. | |
210aa277 DS |
1083 | * |
1084 | * All allocations are done with GFP_NOFS. | |
462d6fac JB |
1085 | */ |
1086 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1087 | unsigned bits, unsigned clear_bits, |
210aa277 | 1088 | struct extent_state **cached_state) |
462d6fac JB |
1089 | { |
1090 | struct extent_state *state; | |
1091 | struct extent_state *prealloc = NULL; | |
1092 | struct rb_node *node; | |
12cfbad9 FDBM |
1093 | struct rb_node **p; |
1094 | struct rb_node *parent; | |
462d6fac JB |
1095 | int err = 0; |
1096 | u64 last_start; | |
1097 | u64 last_end; | |
c8fd3de7 | 1098 | bool first_iteration = true; |
462d6fac | 1099 | |
a5dee37d | 1100 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 1101 | |
462d6fac | 1102 | again: |
210aa277 | 1103 | if (!prealloc) { |
c8fd3de7 FM |
1104 | /* |
1105 | * Best effort, don't worry if extent state allocation fails | |
1106 | * here for the first iteration. We might have a cached state | |
1107 | * that matches exactly the target range, in which case no | |
1108 | * extent state allocations are needed. We'll only know this | |
1109 | * after locking the tree. | |
1110 | */ | |
210aa277 | 1111 | prealloc = alloc_extent_state(GFP_NOFS); |
c8fd3de7 | 1112 | if (!prealloc && !first_iteration) |
462d6fac JB |
1113 | return -ENOMEM; |
1114 | } | |
1115 | ||
1116 | spin_lock(&tree->lock); | |
e6138876 JB |
1117 | if (cached_state && *cached_state) { |
1118 | state = *cached_state; | |
1119 | if (state->start <= start && state->end > start && | |
27a3507d | 1120 | extent_state_in_tree(state)) { |
e6138876 JB |
1121 | node = &state->rb_node; |
1122 | goto hit_next; | |
1123 | } | |
1124 | } | |
1125 | ||
462d6fac JB |
1126 | /* |
1127 | * this search will find all the extents that end after | |
1128 | * our range starts. | |
1129 | */ | |
12cfbad9 | 1130 | node = tree_search_for_insert(tree, start, &p, &parent); |
462d6fac JB |
1131 | if (!node) { |
1132 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1133 | if (!prealloc) { |
1134 | err = -ENOMEM; | |
1135 | goto out; | |
1136 | } | |
12cfbad9 | 1137 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 1138 | &p, &parent, &bits, NULL); |
c2d904e0 JM |
1139 | if (err) |
1140 | extent_io_tree_panic(tree, err); | |
c42ac0bc FDBM |
1141 | cache_state(prealloc, cached_state); |
1142 | prealloc = NULL; | |
462d6fac JB |
1143 | goto out; |
1144 | } | |
1145 | state = rb_entry(node, struct extent_state, rb_node); | |
1146 | hit_next: | |
1147 | last_start = state->start; | |
1148 | last_end = state->end; | |
1149 | ||
1150 | /* | |
1151 | * | ---- desired range ---- | | |
1152 | * | state | | |
1153 | * | |
1154 | * Just lock what we found and keep going | |
1155 | */ | |
1156 | if (state->start == start && state->end <= end) { | |
d38ed27f | 1157 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1158 | cache_state(state, cached_state); |
fefdc557 | 1159 | state = clear_state_bit(tree, state, &clear_bits, 0, NULL); |
462d6fac JB |
1160 | if (last_end == (u64)-1) |
1161 | goto out; | |
462d6fac | 1162 | start = last_end + 1; |
d1ac6e41 LB |
1163 | if (start < end && state && state->start == start && |
1164 | !need_resched()) | |
1165 | goto hit_next; | |
462d6fac JB |
1166 | goto search_again; |
1167 | } | |
1168 | ||
1169 | /* | |
1170 | * | ---- desired range ---- | | |
1171 | * | state | | |
1172 | * or | |
1173 | * | ------------- state -------------- | | |
1174 | * | |
1175 | * We need to split the extent we found, and may flip bits on | |
1176 | * second half. | |
1177 | * | |
1178 | * If the extent we found extends past our | |
1179 | * range, we just split and search again. It'll get split | |
1180 | * again the next time though. | |
1181 | * | |
1182 | * If the extent we found is inside our range, we set the | |
1183 | * desired bit on it. | |
1184 | */ | |
1185 | if (state->start < start) { | |
1186 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1187 | if (!prealloc) { |
1188 | err = -ENOMEM; | |
1189 | goto out; | |
1190 | } | |
462d6fac | 1191 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1192 | if (err) |
1193 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1194 | prealloc = NULL; |
1195 | if (err) | |
1196 | goto out; | |
1197 | if (state->end <= end) { | |
d38ed27f | 1198 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1199 | cache_state(state, cached_state); |
fefdc557 QW |
1200 | state = clear_state_bit(tree, state, &clear_bits, 0, |
1201 | NULL); | |
462d6fac JB |
1202 | if (last_end == (u64)-1) |
1203 | goto out; | |
1204 | start = last_end + 1; | |
d1ac6e41 LB |
1205 | if (start < end && state && state->start == start && |
1206 | !need_resched()) | |
1207 | goto hit_next; | |
462d6fac JB |
1208 | } |
1209 | goto search_again; | |
1210 | } | |
1211 | /* | |
1212 | * | ---- desired range ---- | | |
1213 | * | state | or | state | | |
1214 | * | |
1215 | * There's a hole, we need to insert something in it and | |
1216 | * ignore the extent we found. | |
1217 | */ | |
1218 | if (state->start > start) { | |
1219 | u64 this_end; | |
1220 | if (end < last_start) | |
1221 | this_end = end; | |
1222 | else | |
1223 | this_end = last_start - 1; | |
1224 | ||
1225 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1226 | if (!prealloc) { |
1227 | err = -ENOMEM; | |
1228 | goto out; | |
1229 | } | |
462d6fac JB |
1230 | |
1231 | /* | |
1232 | * Avoid to free 'prealloc' if it can be merged with | |
1233 | * the later extent. | |
1234 | */ | |
1235 | err = insert_state(tree, prealloc, start, this_end, | |
d38ed27f | 1236 | NULL, NULL, &bits, NULL); |
c2d904e0 JM |
1237 | if (err) |
1238 | extent_io_tree_panic(tree, err); | |
e6138876 | 1239 | cache_state(prealloc, cached_state); |
462d6fac JB |
1240 | prealloc = NULL; |
1241 | start = this_end + 1; | |
1242 | goto search_again; | |
1243 | } | |
1244 | /* | |
1245 | * | ---- desired range ---- | | |
1246 | * | state | | |
1247 | * We need to split the extent, and set the bit | |
1248 | * on the first half | |
1249 | */ | |
1250 | if (state->start <= end && state->end > end) { | |
1251 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1252 | if (!prealloc) { |
1253 | err = -ENOMEM; | |
1254 | goto out; | |
1255 | } | |
462d6fac JB |
1256 | |
1257 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1258 | if (err) |
1259 | extent_io_tree_panic(tree, err); | |
462d6fac | 1260 | |
d38ed27f | 1261 | set_state_bits(tree, prealloc, &bits, NULL); |
e6138876 | 1262 | cache_state(prealloc, cached_state); |
fefdc557 | 1263 | clear_state_bit(tree, prealloc, &clear_bits, 0, NULL); |
462d6fac JB |
1264 | prealloc = NULL; |
1265 | goto out; | |
1266 | } | |
1267 | ||
462d6fac JB |
1268 | search_again: |
1269 | if (start > end) | |
1270 | goto out; | |
1271 | spin_unlock(&tree->lock); | |
210aa277 | 1272 | cond_resched(); |
c8fd3de7 | 1273 | first_iteration = false; |
462d6fac | 1274 | goto again; |
462d6fac JB |
1275 | |
1276 | out: | |
1277 | spin_unlock(&tree->lock); | |
1278 | if (prealloc) | |
1279 | free_extent_state(prealloc); | |
1280 | ||
1281 | return err; | |
462d6fac JB |
1282 | } |
1283 | ||
d1310b2e | 1284 | /* wrappers around set/clear extent bit */ |
d38ed27f | 1285 | int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
2c53b912 | 1286 | unsigned bits, struct extent_changeset *changeset) |
d38ed27f QW |
1287 | { |
1288 | /* | |
1289 | * We don't support EXTENT_LOCKED yet, as current changeset will | |
1290 | * record any bits changed, so for EXTENT_LOCKED case, it will | |
1291 | * either fail with -EEXIST or changeset will record the whole | |
1292 | * range. | |
1293 | */ | |
1294 | BUG_ON(bits & EXTENT_LOCKED); | |
1295 | ||
2c53b912 | 1296 | return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS, |
d38ed27f QW |
1297 | changeset); |
1298 | } | |
1299 | ||
fefdc557 QW |
1300 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
1301 | unsigned bits, int wake, int delete, | |
ae0f1625 | 1302 | struct extent_state **cached) |
fefdc557 QW |
1303 | { |
1304 | return __clear_extent_bit(tree, start, end, bits, wake, delete, | |
ae0f1625 | 1305 | cached, GFP_NOFS, NULL); |
fefdc557 QW |
1306 | } |
1307 | ||
fefdc557 | 1308 | int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
f734c44a | 1309 | unsigned bits, struct extent_changeset *changeset) |
fefdc557 QW |
1310 | { |
1311 | /* | |
1312 | * Don't support EXTENT_LOCKED case, same reason as | |
1313 | * set_record_extent_bits(). | |
1314 | */ | |
1315 | BUG_ON(bits & EXTENT_LOCKED); | |
1316 | ||
f734c44a | 1317 | return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS, |
fefdc557 QW |
1318 | changeset); |
1319 | } | |
1320 | ||
d352ac68 CM |
1321 | /* |
1322 | * either insert or lock state struct between start and end use mask to tell | |
1323 | * us if waiting is desired. | |
1324 | */ | |
1edbb734 | 1325 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
ff13db41 | 1326 | struct extent_state **cached_state) |
d1310b2e CM |
1327 | { |
1328 | int err; | |
1329 | u64 failed_start; | |
9ee49a04 | 1330 | |
d1310b2e | 1331 | while (1) { |
ff13db41 | 1332 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, |
3fbe5c02 | 1333 | EXTENT_LOCKED, &failed_start, |
d38ed27f | 1334 | cached_state, GFP_NOFS, NULL); |
d0082371 | 1335 | if (err == -EEXIST) { |
d1310b2e CM |
1336 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1337 | start = failed_start; | |
d0082371 | 1338 | } else |
d1310b2e | 1339 | break; |
d1310b2e CM |
1340 | WARN_ON(start > end); |
1341 | } | |
1342 | return err; | |
1343 | } | |
d1310b2e | 1344 | |
d0082371 | 1345 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1346 | { |
1347 | int err; | |
1348 | u64 failed_start; | |
1349 | ||
3fbe5c02 | 1350 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
d38ed27f | 1351 | &failed_start, NULL, GFP_NOFS, NULL); |
6643558d YZ |
1352 | if (err == -EEXIST) { |
1353 | if (failed_start > start) | |
1354 | clear_extent_bit(tree, start, failed_start - 1, | |
ae0f1625 | 1355 | EXTENT_LOCKED, 1, 0, NULL); |
25179201 | 1356 | return 0; |
6643558d | 1357 | } |
25179201 JB |
1358 | return 1; |
1359 | } | |
25179201 | 1360 | |
bd1fa4f0 | 1361 | void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1362 | { |
09cbfeaf KS |
1363 | unsigned long index = start >> PAGE_SHIFT; |
1364 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1365 | struct page *page; |
1366 | ||
1367 | while (index <= end_index) { | |
1368 | page = find_get_page(inode->i_mapping, index); | |
1369 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1370 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1371 | put_page(page); |
4adaa611 CM |
1372 | index++; |
1373 | } | |
4adaa611 CM |
1374 | } |
1375 | ||
f6311572 | 1376 | void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1377 | { |
09cbfeaf KS |
1378 | unsigned long index = start >> PAGE_SHIFT; |
1379 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1380 | struct page *page; |
1381 | ||
1382 | while (index <= end_index) { | |
1383 | page = find_get_page(inode->i_mapping, index); | |
1384 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
4adaa611 | 1385 | __set_page_dirty_nobuffers(page); |
8d38633c | 1386 | account_page_redirty(page); |
09cbfeaf | 1387 | put_page(page); |
4adaa611 CM |
1388 | index++; |
1389 | } | |
4adaa611 CM |
1390 | } |
1391 | ||
d1310b2e CM |
1392 | /* |
1393 | * helper function to set both pages and extents in the tree writeback | |
1394 | */ | |
35de6db2 | 1395 | static void set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e | 1396 | { |
c6100a4b | 1397 | tree->ops->set_range_writeback(tree->private_data, start, end); |
d1310b2e | 1398 | } |
d1310b2e | 1399 | |
d352ac68 CM |
1400 | /* find the first state struct with 'bits' set after 'start', and |
1401 | * return it. tree->lock must be held. NULL will returned if | |
1402 | * nothing was found after 'start' | |
1403 | */ | |
48a3b636 ES |
1404 | static struct extent_state * |
1405 | find_first_extent_bit_state(struct extent_io_tree *tree, | |
9ee49a04 | 1406 | u64 start, unsigned bits) |
d7fc640e CM |
1407 | { |
1408 | struct rb_node *node; | |
1409 | struct extent_state *state; | |
1410 | ||
1411 | /* | |
1412 | * this search will find all the extents that end after | |
1413 | * our range starts. | |
1414 | */ | |
1415 | node = tree_search(tree, start); | |
d397712b | 1416 | if (!node) |
d7fc640e | 1417 | goto out; |
d7fc640e | 1418 | |
d397712b | 1419 | while (1) { |
d7fc640e | 1420 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1421 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1422 | return state; |
d397712b | 1423 | |
d7fc640e CM |
1424 | node = rb_next(node); |
1425 | if (!node) | |
1426 | break; | |
1427 | } | |
1428 | out: | |
1429 | return NULL; | |
1430 | } | |
d7fc640e | 1431 | |
69261c4b XG |
1432 | /* |
1433 | * find the first offset in the io tree with 'bits' set. zero is | |
1434 | * returned if we find something, and *start_ret and *end_ret are | |
1435 | * set to reflect the state struct that was found. | |
1436 | * | |
477d7eaf | 1437 | * If nothing was found, 1 is returned. If found something, return 0. |
69261c4b XG |
1438 | */ |
1439 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
9ee49a04 | 1440 | u64 *start_ret, u64 *end_ret, unsigned bits, |
e6138876 | 1441 | struct extent_state **cached_state) |
69261c4b XG |
1442 | { |
1443 | struct extent_state *state; | |
e6138876 | 1444 | struct rb_node *n; |
69261c4b XG |
1445 | int ret = 1; |
1446 | ||
1447 | spin_lock(&tree->lock); | |
e6138876 JB |
1448 | if (cached_state && *cached_state) { |
1449 | state = *cached_state; | |
27a3507d | 1450 | if (state->end == start - 1 && extent_state_in_tree(state)) { |
e6138876 JB |
1451 | n = rb_next(&state->rb_node); |
1452 | while (n) { | |
1453 | state = rb_entry(n, struct extent_state, | |
1454 | rb_node); | |
1455 | if (state->state & bits) | |
1456 | goto got_it; | |
1457 | n = rb_next(n); | |
1458 | } | |
1459 | free_extent_state(*cached_state); | |
1460 | *cached_state = NULL; | |
1461 | goto out; | |
1462 | } | |
1463 | free_extent_state(*cached_state); | |
1464 | *cached_state = NULL; | |
1465 | } | |
1466 | ||
69261c4b | 1467 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1468 | got_it: |
69261c4b | 1469 | if (state) { |
e38e2ed7 | 1470 | cache_state_if_flags(state, cached_state, 0); |
69261c4b XG |
1471 | *start_ret = state->start; |
1472 | *end_ret = state->end; | |
1473 | ret = 0; | |
1474 | } | |
e6138876 | 1475 | out: |
69261c4b XG |
1476 | spin_unlock(&tree->lock); |
1477 | return ret; | |
1478 | } | |
1479 | ||
d352ac68 CM |
1480 | /* |
1481 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1482 | * more than 'max_bytes'. start and end are used to return the range, | |
1483 | * | |
1484 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1485 | */ | |
c8b97818 | 1486 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1487 | u64 *start, u64 *end, u64 max_bytes, |
1488 | struct extent_state **cached_state) | |
d1310b2e CM |
1489 | { |
1490 | struct rb_node *node; | |
1491 | struct extent_state *state; | |
1492 | u64 cur_start = *start; | |
1493 | u64 found = 0; | |
1494 | u64 total_bytes = 0; | |
1495 | ||
cad321ad | 1496 | spin_lock(&tree->lock); |
c8b97818 | 1497 | |
d1310b2e CM |
1498 | /* |
1499 | * this search will find all the extents that end after | |
1500 | * our range starts. | |
1501 | */ | |
80ea96b1 | 1502 | node = tree_search(tree, cur_start); |
2b114d1d | 1503 | if (!node) { |
3b951516 CM |
1504 | if (!found) |
1505 | *end = (u64)-1; | |
d1310b2e CM |
1506 | goto out; |
1507 | } | |
1508 | ||
d397712b | 1509 | while (1) { |
d1310b2e | 1510 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1511 | if (found && (state->start != cur_start || |
1512 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1513 | goto out; |
1514 | } | |
1515 | if (!(state->state & EXTENT_DELALLOC)) { | |
1516 | if (!found) | |
1517 | *end = state->end; | |
1518 | goto out; | |
1519 | } | |
c2a128d2 | 1520 | if (!found) { |
d1310b2e | 1521 | *start = state->start; |
c2a128d2 | 1522 | *cached_state = state; |
b7ac31b7 | 1523 | refcount_inc(&state->refs); |
c2a128d2 | 1524 | } |
d1310b2e CM |
1525 | found++; |
1526 | *end = state->end; | |
1527 | cur_start = state->end + 1; | |
1528 | node = rb_next(node); | |
d1310b2e | 1529 | total_bytes += state->end - state->start + 1; |
7bf811a5 | 1530 | if (total_bytes >= max_bytes) |
573aecaf | 1531 | break; |
573aecaf | 1532 | if (!node) |
d1310b2e CM |
1533 | break; |
1534 | } | |
1535 | out: | |
cad321ad | 1536 | spin_unlock(&tree->lock); |
d1310b2e CM |
1537 | return found; |
1538 | } | |
1539 | ||
da2c7009 LB |
1540 | static int __process_pages_contig(struct address_space *mapping, |
1541 | struct page *locked_page, | |
1542 | pgoff_t start_index, pgoff_t end_index, | |
1543 | unsigned long page_ops, pgoff_t *index_ret); | |
1544 | ||
143bede5 JM |
1545 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1546 | struct page *locked_page, | |
1547 | u64 start, u64 end) | |
c8b97818 | 1548 | { |
09cbfeaf KS |
1549 | unsigned long index = start >> PAGE_SHIFT; |
1550 | unsigned long end_index = end >> PAGE_SHIFT; | |
c8b97818 | 1551 | |
76c0021d | 1552 | ASSERT(locked_page); |
c8b97818 | 1553 | if (index == locked_page->index && end_index == index) |
143bede5 | 1554 | return; |
c8b97818 | 1555 | |
76c0021d LB |
1556 | __process_pages_contig(inode->i_mapping, locked_page, index, end_index, |
1557 | PAGE_UNLOCK, NULL); | |
c8b97818 CM |
1558 | } |
1559 | ||
1560 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1561 | struct page *locked_page, | |
1562 | u64 delalloc_start, | |
1563 | u64 delalloc_end) | |
1564 | { | |
09cbfeaf | 1565 | unsigned long index = delalloc_start >> PAGE_SHIFT; |
76c0021d | 1566 | unsigned long index_ret = index; |
09cbfeaf | 1567 | unsigned long end_index = delalloc_end >> PAGE_SHIFT; |
c8b97818 | 1568 | int ret; |
c8b97818 | 1569 | |
76c0021d | 1570 | ASSERT(locked_page); |
c8b97818 CM |
1571 | if (index == locked_page->index && index == end_index) |
1572 | return 0; | |
1573 | ||
76c0021d LB |
1574 | ret = __process_pages_contig(inode->i_mapping, locked_page, index, |
1575 | end_index, PAGE_LOCK, &index_ret); | |
1576 | if (ret == -EAGAIN) | |
1577 | __unlock_for_delalloc(inode, locked_page, delalloc_start, | |
1578 | (u64)index_ret << PAGE_SHIFT); | |
c8b97818 CM |
1579 | return ret; |
1580 | } | |
1581 | ||
1582 | /* | |
1583 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1584 | * more than 'max_bytes'. start and end are used to return the range, | |
1585 | * | |
1586 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1587 | */ | |
294e30fe JB |
1588 | STATIC u64 find_lock_delalloc_range(struct inode *inode, |
1589 | struct extent_io_tree *tree, | |
1590 | struct page *locked_page, u64 *start, | |
1591 | u64 *end, u64 max_bytes) | |
c8b97818 CM |
1592 | { |
1593 | u64 delalloc_start; | |
1594 | u64 delalloc_end; | |
1595 | u64 found; | |
9655d298 | 1596 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1597 | int ret; |
1598 | int loops = 0; | |
1599 | ||
1600 | again: | |
1601 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1602 | delalloc_start = *start; | |
1603 | delalloc_end = 0; | |
1604 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1605 | max_bytes, &cached_state); |
70b99e69 | 1606 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1607 | *start = delalloc_start; |
1608 | *end = delalloc_end; | |
c2a128d2 | 1609 | free_extent_state(cached_state); |
385fe0be | 1610 | return 0; |
c8b97818 CM |
1611 | } |
1612 | ||
70b99e69 CM |
1613 | /* |
1614 | * start comes from the offset of locked_page. We have to lock | |
1615 | * pages in order, so we can't process delalloc bytes before | |
1616 | * locked_page | |
1617 | */ | |
d397712b | 1618 | if (delalloc_start < *start) |
70b99e69 | 1619 | delalloc_start = *start; |
70b99e69 | 1620 | |
c8b97818 CM |
1621 | /* |
1622 | * make sure to limit the number of pages we try to lock down | |
c8b97818 | 1623 | */ |
7bf811a5 JB |
1624 | if (delalloc_end + 1 - delalloc_start > max_bytes) |
1625 | delalloc_end = delalloc_start + max_bytes - 1; | |
d397712b | 1626 | |
c8b97818 CM |
1627 | /* step two, lock all the pages after the page that has start */ |
1628 | ret = lock_delalloc_pages(inode, locked_page, | |
1629 | delalloc_start, delalloc_end); | |
1630 | if (ret == -EAGAIN) { | |
1631 | /* some of the pages are gone, lets avoid looping by | |
1632 | * shortening the size of the delalloc range we're searching | |
1633 | */ | |
9655d298 | 1634 | free_extent_state(cached_state); |
7d788742 | 1635 | cached_state = NULL; |
c8b97818 | 1636 | if (!loops) { |
09cbfeaf | 1637 | max_bytes = PAGE_SIZE; |
c8b97818 CM |
1638 | loops = 1; |
1639 | goto again; | |
1640 | } else { | |
1641 | found = 0; | |
1642 | goto out_failed; | |
1643 | } | |
1644 | } | |
79787eaa | 1645 | BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */ |
c8b97818 CM |
1646 | |
1647 | /* step three, lock the state bits for the whole range */ | |
ff13db41 | 1648 | lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state); |
c8b97818 CM |
1649 | |
1650 | /* then test to make sure it is all still delalloc */ | |
1651 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1652 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1653 | if (!ret) { |
9655d298 | 1654 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
e43bbe5e | 1655 | &cached_state); |
c8b97818 CM |
1656 | __unlock_for_delalloc(inode, locked_page, |
1657 | delalloc_start, delalloc_end); | |
1658 | cond_resched(); | |
1659 | goto again; | |
1660 | } | |
9655d298 | 1661 | free_extent_state(cached_state); |
c8b97818 CM |
1662 | *start = delalloc_start; |
1663 | *end = delalloc_end; | |
1664 | out_failed: | |
1665 | return found; | |
1666 | } | |
1667 | ||
da2c7009 LB |
1668 | static int __process_pages_contig(struct address_space *mapping, |
1669 | struct page *locked_page, | |
1670 | pgoff_t start_index, pgoff_t end_index, | |
1671 | unsigned long page_ops, pgoff_t *index_ret) | |
c8b97818 | 1672 | { |
873695b3 | 1673 | unsigned long nr_pages = end_index - start_index + 1; |
da2c7009 | 1674 | unsigned long pages_locked = 0; |
873695b3 | 1675 | pgoff_t index = start_index; |
c8b97818 | 1676 | struct page *pages[16]; |
873695b3 | 1677 | unsigned ret; |
da2c7009 | 1678 | int err = 0; |
c8b97818 | 1679 | int i; |
771ed689 | 1680 | |
da2c7009 LB |
1681 | if (page_ops & PAGE_LOCK) { |
1682 | ASSERT(page_ops == PAGE_LOCK); | |
1683 | ASSERT(index_ret && *index_ret == start_index); | |
1684 | } | |
1685 | ||
704de49d | 1686 | if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0) |
873695b3 | 1687 | mapping_set_error(mapping, -EIO); |
704de49d | 1688 | |
d397712b | 1689 | while (nr_pages > 0) { |
873695b3 | 1690 | ret = find_get_pages_contig(mapping, index, |
5b050f04 CM |
1691 | min_t(unsigned long, |
1692 | nr_pages, ARRAY_SIZE(pages)), pages); | |
da2c7009 LB |
1693 | if (ret == 0) { |
1694 | /* | |
1695 | * Only if we're going to lock these pages, | |
1696 | * can we find nothing at @index. | |
1697 | */ | |
1698 | ASSERT(page_ops & PAGE_LOCK); | |
49d4a334 LB |
1699 | err = -EAGAIN; |
1700 | goto out; | |
da2c7009 | 1701 | } |
8b62b72b | 1702 | |
da2c7009 | 1703 | for (i = 0; i < ret; i++) { |
c2790a2e | 1704 | if (page_ops & PAGE_SET_PRIVATE2) |
8b62b72b CM |
1705 | SetPagePrivate2(pages[i]); |
1706 | ||
c8b97818 | 1707 | if (pages[i] == locked_page) { |
09cbfeaf | 1708 | put_page(pages[i]); |
da2c7009 | 1709 | pages_locked++; |
c8b97818 CM |
1710 | continue; |
1711 | } | |
c2790a2e | 1712 | if (page_ops & PAGE_CLEAR_DIRTY) |
c8b97818 | 1713 | clear_page_dirty_for_io(pages[i]); |
c2790a2e | 1714 | if (page_ops & PAGE_SET_WRITEBACK) |
c8b97818 | 1715 | set_page_writeback(pages[i]); |
704de49d FM |
1716 | if (page_ops & PAGE_SET_ERROR) |
1717 | SetPageError(pages[i]); | |
c2790a2e | 1718 | if (page_ops & PAGE_END_WRITEBACK) |
c8b97818 | 1719 | end_page_writeback(pages[i]); |
c2790a2e | 1720 | if (page_ops & PAGE_UNLOCK) |
771ed689 | 1721 | unlock_page(pages[i]); |
da2c7009 LB |
1722 | if (page_ops & PAGE_LOCK) { |
1723 | lock_page(pages[i]); | |
1724 | if (!PageDirty(pages[i]) || | |
1725 | pages[i]->mapping != mapping) { | |
1726 | unlock_page(pages[i]); | |
1727 | put_page(pages[i]); | |
1728 | err = -EAGAIN; | |
1729 | goto out; | |
1730 | } | |
1731 | } | |
09cbfeaf | 1732 | put_page(pages[i]); |
da2c7009 | 1733 | pages_locked++; |
c8b97818 CM |
1734 | } |
1735 | nr_pages -= ret; | |
1736 | index += ret; | |
1737 | cond_resched(); | |
1738 | } | |
da2c7009 LB |
1739 | out: |
1740 | if (err && index_ret) | |
1741 | *index_ret = start_index + pages_locked - 1; | |
1742 | return err; | |
c8b97818 | 1743 | } |
c8b97818 | 1744 | |
873695b3 LB |
1745 | void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end, |
1746 | u64 delalloc_end, struct page *locked_page, | |
1747 | unsigned clear_bits, | |
1748 | unsigned long page_ops) | |
1749 | { | |
1750 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0, | |
ae0f1625 | 1751 | NULL); |
873695b3 LB |
1752 | |
1753 | __process_pages_contig(inode->i_mapping, locked_page, | |
1754 | start >> PAGE_SHIFT, end >> PAGE_SHIFT, | |
da2c7009 | 1755 | page_ops, NULL); |
873695b3 LB |
1756 | } |
1757 | ||
d352ac68 CM |
1758 | /* |
1759 | * count the number of bytes in the tree that have a given bit(s) | |
1760 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1761 | * cached. The total number found is returned. | |
1762 | */ | |
d1310b2e CM |
1763 | u64 count_range_bits(struct extent_io_tree *tree, |
1764 | u64 *start, u64 search_end, u64 max_bytes, | |
9ee49a04 | 1765 | unsigned bits, int contig) |
d1310b2e CM |
1766 | { |
1767 | struct rb_node *node; | |
1768 | struct extent_state *state; | |
1769 | u64 cur_start = *start; | |
1770 | u64 total_bytes = 0; | |
ec29ed5b | 1771 | u64 last = 0; |
d1310b2e CM |
1772 | int found = 0; |
1773 | ||
fae7f21c | 1774 | if (WARN_ON(search_end <= cur_start)) |
d1310b2e | 1775 | return 0; |
d1310b2e | 1776 | |
cad321ad | 1777 | spin_lock(&tree->lock); |
d1310b2e CM |
1778 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1779 | total_bytes = tree->dirty_bytes; | |
1780 | goto out; | |
1781 | } | |
1782 | /* | |
1783 | * this search will find all the extents that end after | |
1784 | * our range starts. | |
1785 | */ | |
80ea96b1 | 1786 | node = tree_search(tree, cur_start); |
d397712b | 1787 | if (!node) |
d1310b2e | 1788 | goto out; |
d1310b2e | 1789 | |
d397712b | 1790 | while (1) { |
d1310b2e CM |
1791 | state = rb_entry(node, struct extent_state, rb_node); |
1792 | if (state->start > search_end) | |
1793 | break; | |
ec29ed5b CM |
1794 | if (contig && found && state->start > last + 1) |
1795 | break; | |
1796 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1797 | total_bytes += min(search_end, state->end) + 1 - |
1798 | max(cur_start, state->start); | |
1799 | if (total_bytes >= max_bytes) | |
1800 | break; | |
1801 | if (!found) { | |
af60bed2 | 1802 | *start = max(cur_start, state->start); |
d1310b2e CM |
1803 | found = 1; |
1804 | } | |
ec29ed5b CM |
1805 | last = state->end; |
1806 | } else if (contig && found) { | |
1807 | break; | |
d1310b2e CM |
1808 | } |
1809 | node = rb_next(node); | |
1810 | if (!node) | |
1811 | break; | |
1812 | } | |
1813 | out: | |
cad321ad | 1814 | spin_unlock(&tree->lock); |
d1310b2e CM |
1815 | return total_bytes; |
1816 | } | |
b2950863 | 1817 | |
d352ac68 CM |
1818 | /* |
1819 | * set the private field for a given byte offset in the tree. If there isn't | |
1820 | * an extent_state there already, this does nothing. | |
1821 | */ | |
f827ba9a | 1822 | static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start, |
47dc196a | 1823 | struct io_failure_record *failrec) |
d1310b2e CM |
1824 | { |
1825 | struct rb_node *node; | |
1826 | struct extent_state *state; | |
1827 | int ret = 0; | |
1828 | ||
cad321ad | 1829 | spin_lock(&tree->lock); |
d1310b2e CM |
1830 | /* |
1831 | * this search will find all the extents that end after | |
1832 | * our range starts. | |
1833 | */ | |
80ea96b1 | 1834 | node = tree_search(tree, start); |
2b114d1d | 1835 | if (!node) { |
d1310b2e CM |
1836 | ret = -ENOENT; |
1837 | goto out; | |
1838 | } | |
1839 | state = rb_entry(node, struct extent_state, rb_node); | |
1840 | if (state->start != start) { | |
1841 | ret = -ENOENT; | |
1842 | goto out; | |
1843 | } | |
47dc196a | 1844 | state->failrec = failrec; |
d1310b2e | 1845 | out: |
cad321ad | 1846 | spin_unlock(&tree->lock); |
d1310b2e CM |
1847 | return ret; |
1848 | } | |
1849 | ||
f827ba9a | 1850 | static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start, |
47dc196a | 1851 | struct io_failure_record **failrec) |
d1310b2e CM |
1852 | { |
1853 | struct rb_node *node; | |
1854 | struct extent_state *state; | |
1855 | int ret = 0; | |
1856 | ||
cad321ad | 1857 | spin_lock(&tree->lock); |
d1310b2e CM |
1858 | /* |
1859 | * this search will find all the extents that end after | |
1860 | * our range starts. | |
1861 | */ | |
80ea96b1 | 1862 | node = tree_search(tree, start); |
2b114d1d | 1863 | if (!node) { |
d1310b2e CM |
1864 | ret = -ENOENT; |
1865 | goto out; | |
1866 | } | |
1867 | state = rb_entry(node, struct extent_state, rb_node); | |
1868 | if (state->start != start) { | |
1869 | ret = -ENOENT; | |
1870 | goto out; | |
1871 | } | |
47dc196a | 1872 | *failrec = state->failrec; |
d1310b2e | 1873 | out: |
cad321ad | 1874 | spin_unlock(&tree->lock); |
d1310b2e CM |
1875 | return ret; |
1876 | } | |
1877 | ||
1878 | /* | |
1879 | * searches a range in the state tree for a given mask. | |
70dec807 | 1880 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1881 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1882 | * range is found set. | |
1883 | */ | |
1884 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1885 | unsigned bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1886 | { |
1887 | struct extent_state *state = NULL; | |
1888 | struct rb_node *node; | |
1889 | int bitset = 0; | |
d1310b2e | 1890 | |
cad321ad | 1891 | spin_lock(&tree->lock); |
27a3507d | 1892 | if (cached && extent_state_in_tree(cached) && cached->start <= start && |
df98b6e2 | 1893 | cached->end > start) |
9655d298 CM |
1894 | node = &cached->rb_node; |
1895 | else | |
1896 | node = tree_search(tree, start); | |
d1310b2e CM |
1897 | while (node && start <= end) { |
1898 | state = rb_entry(node, struct extent_state, rb_node); | |
1899 | ||
1900 | if (filled && state->start > start) { | |
1901 | bitset = 0; | |
1902 | break; | |
1903 | } | |
1904 | ||
1905 | if (state->start > end) | |
1906 | break; | |
1907 | ||
1908 | if (state->state & bits) { | |
1909 | bitset = 1; | |
1910 | if (!filled) | |
1911 | break; | |
1912 | } else if (filled) { | |
1913 | bitset = 0; | |
1914 | break; | |
1915 | } | |
46562cec CM |
1916 | |
1917 | if (state->end == (u64)-1) | |
1918 | break; | |
1919 | ||
d1310b2e CM |
1920 | start = state->end + 1; |
1921 | if (start > end) | |
1922 | break; | |
1923 | node = rb_next(node); | |
1924 | if (!node) { | |
1925 | if (filled) | |
1926 | bitset = 0; | |
1927 | break; | |
1928 | } | |
1929 | } | |
cad321ad | 1930 | spin_unlock(&tree->lock); |
d1310b2e CM |
1931 | return bitset; |
1932 | } | |
d1310b2e CM |
1933 | |
1934 | /* | |
1935 | * helper function to set a given page up to date if all the | |
1936 | * extents in the tree for that page are up to date | |
1937 | */ | |
143bede5 | 1938 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e | 1939 | { |
4eee4fa4 | 1940 | u64 start = page_offset(page); |
09cbfeaf | 1941 | u64 end = start + PAGE_SIZE - 1; |
9655d298 | 1942 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 1943 | SetPageUptodate(page); |
d1310b2e CM |
1944 | } |
1945 | ||
7870d082 JB |
1946 | int free_io_failure(struct extent_io_tree *failure_tree, |
1947 | struct extent_io_tree *io_tree, | |
1948 | struct io_failure_record *rec) | |
4a54c8c1 JS |
1949 | { |
1950 | int ret; | |
1951 | int err = 0; | |
4a54c8c1 | 1952 | |
47dc196a | 1953 | set_state_failrec(failure_tree, rec->start, NULL); |
4a54c8c1 JS |
1954 | ret = clear_extent_bits(failure_tree, rec->start, |
1955 | rec->start + rec->len - 1, | |
91166212 | 1956 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 JS |
1957 | if (ret) |
1958 | err = ret; | |
1959 | ||
7870d082 | 1960 | ret = clear_extent_bits(io_tree, rec->start, |
53b381b3 | 1961 | rec->start + rec->len - 1, |
91166212 | 1962 | EXTENT_DAMAGED); |
53b381b3 DW |
1963 | if (ret && !err) |
1964 | err = ret; | |
4a54c8c1 JS |
1965 | |
1966 | kfree(rec); | |
1967 | return err; | |
1968 | } | |
1969 | ||
4a54c8c1 JS |
1970 | /* |
1971 | * this bypasses the standard btrfs submit functions deliberately, as | |
1972 | * the standard behavior is to write all copies in a raid setup. here we only | |
1973 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
1974 | * submit_bio directly. | |
3ec706c8 | 1975 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
1976 | * actually prevents the read that triggered the error from finishing. |
1977 | * currently, there can be no more than two copies of every data bit. thus, | |
1978 | * exactly one rewrite is required. | |
1979 | */ | |
6ec656bc JB |
1980 | int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start, |
1981 | u64 length, u64 logical, struct page *page, | |
1982 | unsigned int pg_offset, int mirror_num) | |
4a54c8c1 JS |
1983 | { |
1984 | struct bio *bio; | |
1985 | struct btrfs_device *dev; | |
4a54c8c1 JS |
1986 | u64 map_length = 0; |
1987 | u64 sector; | |
1988 | struct btrfs_bio *bbio = NULL; | |
1989 | int ret; | |
1990 | ||
1751e8a6 | 1991 | ASSERT(!(fs_info->sb->s_flags & SB_RDONLY)); |
4a54c8c1 JS |
1992 | BUG_ON(!mirror_num); |
1993 | ||
c5e4c3d7 | 1994 | bio = btrfs_io_bio_alloc(1); |
4f024f37 | 1995 | bio->bi_iter.bi_size = 0; |
4a54c8c1 JS |
1996 | map_length = length; |
1997 | ||
b5de8d0d FM |
1998 | /* |
1999 | * Avoid races with device replace and make sure our bbio has devices | |
2000 | * associated to its stripes that don't go away while we are doing the | |
2001 | * read repair operation. | |
2002 | */ | |
2003 | btrfs_bio_counter_inc_blocked(fs_info); | |
e4ff5fb5 | 2004 | if (btrfs_is_parity_mirror(fs_info, logical, length)) { |
c725328c LB |
2005 | /* |
2006 | * Note that we don't use BTRFS_MAP_WRITE because it's supposed | |
2007 | * to update all raid stripes, but here we just want to correct | |
2008 | * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad | |
2009 | * stripe's dev and sector. | |
2010 | */ | |
2011 | ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical, | |
2012 | &map_length, &bbio, 0); | |
2013 | if (ret) { | |
2014 | btrfs_bio_counter_dec(fs_info); | |
2015 | bio_put(bio); | |
2016 | return -EIO; | |
2017 | } | |
2018 | ASSERT(bbio->mirror_num == 1); | |
2019 | } else { | |
2020 | ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, | |
2021 | &map_length, &bbio, mirror_num); | |
2022 | if (ret) { | |
2023 | btrfs_bio_counter_dec(fs_info); | |
2024 | bio_put(bio); | |
2025 | return -EIO; | |
2026 | } | |
2027 | BUG_ON(mirror_num != bbio->mirror_num); | |
4a54c8c1 | 2028 | } |
c725328c LB |
2029 | |
2030 | sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9; | |
4f024f37 | 2031 | bio->bi_iter.bi_sector = sector; |
c725328c | 2032 | dev = bbio->stripes[bbio->mirror_num - 1].dev; |
6e9606d2 | 2033 | btrfs_put_bbio(bbio); |
ebbede42 AJ |
2034 | if (!dev || !dev->bdev || |
2035 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) { | |
b5de8d0d | 2036 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2037 | bio_put(bio); |
2038 | return -EIO; | |
2039 | } | |
74d46992 | 2040 | bio_set_dev(bio, dev->bdev); |
70fd7614 | 2041 | bio->bi_opf = REQ_OP_WRITE | REQ_SYNC; |
ffdd2018 | 2042 | bio_add_page(bio, page, length, pg_offset); |
4a54c8c1 | 2043 | |
4e49ea4a | 2044 | if (btrfsic_submit_bio_wait(bio)) { |
4a54c8c1 | 2045 | /* try to remap that extent elsewhere? */ |
b5de8d0d | 2046 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 | 2047 | bio_put(bio); |
442a4f63 | 2048 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
2049 | return -EIO; |
2050 | } | |
2051 | ||
b14af3b4 DS |
2052 | btrfs_info_rl_in_rcu(fs_info, |
2053 | "read error corrected: ino %llu off %llu (dev %s sector %llu)", | |
6ec656bc | 2054 | ino, start, |
1203b681 | 2055 | rcu_str_deref(dev->name), sector); |
b5de8d0d | 2056 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2057 | bio_put(bio); |
2058 | return 0; | |
2059 | } | |
2060 | ||
2ff7e61e JM |
2061 | int repair_eb_io_failure(struct btrfs_fs_info *fs_info, |
2062 | struct extent_buffer *eb, int mirror_num) | |
ea466794 | 2063 | { |
ea466794 JB |
2064 | u64 start = eb->start; |
2065 | unsigned long i, num_pages = num_extent_pages(eb->start, eb->len); | |
d95603b2 | 2066 | int ret = 0; |
ea466794 | 2067 | |
bc98a42c | 2068 | if (sb_rdonly(fs_info->sb)) |
908960c6 ID |
2069 | return -EROFS; |
2070 | ||
ea466794 | 2071 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 2072 | struct page *p = eb->pages[i]; |
1203b681 | 2073 | |
6ec656bc | 2074 | ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p, |
1203b681 | 2075 | start - page_offset(p), mirror_num); |
ea466794 JB |
2076 | if (ret) |
2077 | break; | |
09cbfeaf | 2078 | start += PAGE_SIZE; |
ea466794 JB |
2079 | } |
2080 | ||
2081 | return ret; | |
2082 | } | |
2083 | ||
4a54c8c1 JS |
2084 | /* |
2085 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2086 | * to see if we need to process or clean up an io_failure_record | |
2087 | */ | |
7870d082 JB |
2088 | int clean_io_failure(struct btrfs_fs_info *fs_info, |
2089 | struct extent_io_tree *failure_tree, | |
2090 | struct extent_io_tree *io_tree, u64 start, | |
2091 | struct page *page, u64 ino, unsigned int pg_offset) | |
4a54c8c1 JS |
2092 | { |
2093 | u64 private; | |
4a54c8c1 | 2094 | struct io_failure_record *failrec; |
4a54c8c1 JS |
2095 | struct extent_state *state; |
2096 | int num_copies; | |
4a54c8c1 | 2097 | int ret; |
4a54c8c1 JS |
2098 | |
2099 | private = 0; | |
7870d082 JB |
2100 | ret = count_range_bits(failure_tree, &private, (u64)-1, 1, |
2101 | EXTENT_DIRTY, 0); | |
4a54c8c1 JS |
2102 | if (!ret) |
2103 | return 0; | |
2104 | ||
7870d082 | 2105 | ret = get_state_failrec(failure_tree, start, &failrec); |
4a54c8c1 JS |
2106 | if (ret) |
2107 | return 0; | |
2108 | ||
4a54c8c1 JS |
2109 | BUG_ON(!failrec->this_mirror); |
2110 | ||
2111 | if (failrec->in_validation) { | |
2112 | /* there was no real error, just free the record */ | |
ab8d0fc4 JM |
2113 | btrfs_debug(fs_info, |
2114 | "clean_io_failure: freeing dummy error at %llu", | |
2115 | failrec->start); | |
4a54c8c1 JS |
2116 | goto out; |
2117 | } | |
bc98a42c | 2118 | if (sb_rdonly(fs_info->sb)) |
908960c6 | 2119 | goto out; |
4a54c8c1 | 2120 | |
7870d082 JB |
2121 | spin_lock(&io_tree->lock); |
2122 | state = find_first_extent_bit_state(io_tree, | |
4a54c8c1 JS |
2123 | failrec->start, |
2124 | EXTENT_LOCKED); | |
7870d082 | 2125 | spin_unlock(&io_tree->lock); |
4a54c8c1 | 2126 | |
883d0de4 MX |
2127 | if (state && state->start <= failrec->start && |
2128 | state->end >= failrec->start + failrec->len - 1) { | |
3ec706c8 SB |
2129 | num_copies = btrfs_num_copies(fs_info, failrec->logical, |
2130 | failrec->len); | |
4a54c8c1 | 2131 | if (num_copies > 1) { |
7870d082 JB |
2132 | repair_io_failure(fs_info, ino, start, failrec->len, |
2133 | failrec->logical, page, pg_offset, | |
2134 | failrec->failed_mirror); | |
4a54c8c1 JS |
2135 | } |
2136 | } | |
2137 | ||
2138 | out: | |
7870d082 | 2139 | free_io_failure(failure_tree, io_tree, failrec); |
4a54c8c1 | 2140 | |
454ff3de | 2141 | return 0; |
4a54c8c1 JS |
2142 | } |
2143 | ||
f612496b MX |
2144 | /* |
2145 | * Can be called when | |
2146 | * - hold extent lock | |
2147 | * - under ordered extent | |
2148 | * - the inode is freeing | |
2149 | */ | |
7ab7956e | 2150 | void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end) |
f612496b | 2151 | { |
7ab7956e | 2152 | struct extent_io_tree *failure_tree = &inode->io_failure_tree; |
f612496b MX |
2153 | struct io_failure_record *failrec; |
2154 | struct extent_state *state, *next; | |
2155 | ||
2156 | if (RB_EMPTY_ROOT(&failure_tree->state)) | |
2157 | return; | |
2158 | ||
2159 | spin_lock(&failure_tree->lock); | |
2160 | state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY); | |
2161 | while (state) { | |
2162 | if (state->start > end) | |
2163 | break; | |
2164 | ||
2165 | ASSERT(state->end <= end); | |
2166 | ||
2167 | next = next_state(state); | |
2168 | ||
47dc196a | 2169 | failrec = state->failrec; |
f612496b MX |
2170 | free_extent_state(state); |
2171 | kfree(failrec); | |
2172 | ||
2173 | state = next; | |
2174 | } | |
2175 | spin_unlock(&failure_tree->lock); | |
2176 | } | |
2177 | ||
2fe6303e | 2178 | int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end, |
47dc196a | 2179 | struct io_failure_record **failrec_ret) |
4a54c8c1 | 2180 | { |
ab8d0fc4 | 2181 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e | 2182 | struct io_failure_record *failrec; |
4a54c8c1 | 2183 | struct extent_map *em; |
4a54c8c1 JS |
2184 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2185 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2186 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
4a54c8c1 | 2187 | int ret; |
4a54c8c1 JS |
2188 | u64 logical; |
2189 | ||
47dc196a | 2190 | ret = get_state_failrec(failure_tree, start, &failrec); |
4a54c8c1 JS |
2191 | if (ret) { |
2192 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2193 | if (!failrec) | |
2194 | return -ENOMEM; | |
2fe6303e | 2195 | |
4a54c8c1 JS |
2196 | failrec->start = start; |
2197 | failrec->len = end - start + 1; | |
2198 | failrec->this_mirror = 0; | |
2199 | failrec->bio_flags = 0; | |
2200 | failrec->in_validation = 0; | |
2201 | ||
2202 | read_lock(&em_tree->lock); | |
2203 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2204 | if (!em) { | |
2205 | read_unlock(&em_tree->lock); | |
2206 | kfree(failrec); | |
2207 | return -EIO; | |
2208 | } | |
2209 | ||
68ba990f | 2210 | if (em->start > start || em->start + em->len <= start) { |
4a54c8c1 JS |
2211 | free_extent_map(em); |
2212 | em = NULL; | |
2213 | } | |
2214 | read_unlock(&em_tree->lock); | |
7a2d6a64 | 2215 | if (!em) { |
4a54c8c1 JS |
2216 | kfree(failrec); |
2217 | return -EIO; | |
2218 | } | |
2fe6303e | 2219 | |
4a54c8c1 JS |
2220 | logical = start - em->start; |
2221 | logical = em->block_start + logical; | |
2222 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2223 | logical = em->block_start; | |
2224 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2225 | extent_set_compress_type(&failrec->bio_flags, | |
2226 | em->compress_type); | |
2227 | } | |
2fe6303e | 2228 | |
ab8d0fc4 JM |
2229 | btrfs_debug(fs_info, |
2230 | "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu", | |
2231 | logical, start, failrec->len); | |
2fe6303e | 2232 | |
4a54c8c1 JS |
2233 | failrec->logical = logical; |
2234 | free_extent_map(em); | |
2235 | ||
2236 | /* set the bits in the private failure tree */ | |
2237 | ret = set_extent_bits(failure_tree, start, end, | |
ceeb0ae7 | 2238 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 | 2239 | if (ret >= 0) |
47dc196a | 2240 | ret = set_state_failrec(failure_tree, start, failrec); |
4a54c8c1 JS |
2241 | /* set the bits in the inode's tree */ |
2242 | if (ret >= 0) | |
ceeb0ae7 | 2243 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED); |
4a54c8c1 JS |
2244 | if (ret < 0) { |
2245 | kfree(failrec); | |
2246 | return ret; | |
2247 | } | |
2248 | } else { | |
ab8d0fc4 JM |
2249 | btrfs_debug(fs_info, |
2250 | "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d", | |
2251 | failrec->logical, failrec->start, failrec->len, | |
2252 | failrec->in_validation); | |
4a54c8c1 JS |
2253 | /* |
2254 | * when data can be on disk more than twice, add to failrec here | |
2255 | * (e.g. with a list for failed_mirror) to make | |
2256 | * clean_io_failure() clean all those errors at once. | |
2257 | */ | |
2258 | } | |
2fe6303e MX |
2259 | |
2260 | *failrec_ret = failrec; | |
2261 | ||
2262 | return 0; | |
2263 | } | |
2264 | ||
a0b60d72 | 2265 | bool btrfs_check_repairable(struct inode *inode, unsigned failed_bio_pages, |
2fe6303e MX |
2266 | struct io_failure_record *failrec, int failed_mirror) |
2267 | { | |
ab8d0fc4 | 2268 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e MX |
2269 | int num_copies; |
2270 | ||
ab8d0fc4 | 2271 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
4a54c8c1 JS |
2272 | if (num_copies == 1) { |
2273 | /* | |
2274 | * we only have a single copy of the data, so don't bother with | |
2275 | * all the retry and error correction code that follows. no | |
2276 | * matter what the error is, it is very likely to persist. | |
2277 | */ | |
ab8d0fc4 JM |
2278 | btrfs_debug(fs_info, |
2279 | "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
2280 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2281 | return false; |
4a54c8c1 JS |
2282 | } |
2283 | ||
4a54c8c1 JS |
2284 | /* |
2285 | * there are two premises: | |
2286 | * a) deliver good data to the caller | |
2287 | * b) correct the bad sectors on disk | |
2288 | */ | |
a0b60d72 | 2289 | if (failed_bio_pages > 1) { |
4a54c8c1 JS |
2290 | /* |
2291 | * to fulfill b), we need to know the exact failing sectors, as | |
2292 | * we don't want to rewrite any more than the failed ones. thus, | |
2293 | * we need separate read requests for the failed bio | |
2294 | * | |
2295 | * if the following BUG_ON triggers, our validation request got | |
2296 | * merged. we need separate requests for our algorithm to work. | |
2297 | */ | |
2298 | BUG_ON(failrec->in_validation); | |
2299 | failrec->in_validation = 1; | |
2300 | failrec->this_mirror = failed_mirror; | |
4a54c8c1 JS |
2301 | } else { |
2302 | /* | |
2303 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2304 | * of the failed sector and if we succeed, we have setup | |
2305 | * everything for repair_io_failure to do the rest for us. | |
2306 | */ | |
2307 | if (failrec->in_validation) { | |
2308 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2309 | failrec->in_validation = 0; | |
2310 | failrec->this_mirror = 0; | |
2311 | } | |
2312 | failrec->failed_mirror = failed_mirror; | |
2313 | failrec->this_mirror++; | |
2314 | if (failrec->this_mirror == failed_mirror) | |
2315 | failrec->this_mirror++; | |
4a54c8c1 JS |
2316 | } |
2317 | ||
facc8a22 | 2318 | if (failrec->this_mirror > num_copies) { |
ab8d0fc4 JM |
2319 | btrfs_debug(fs_info, |
2320 | "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
2321 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2322 | return false; |
4a54c8c1 JS |
2323 | } |
2324 | ||
c3cfb656 | 2325 | return true; |
2fe6303e MX |
2326 | } |
2327 | ||
2328 | ||
2329 | struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio, | |
2330 | struct io_failure_record *failrec, | |
2331 | struct page *page, int pg_offset, int icsum, | |
8b110e39 | 2332 | bio_end_io_t *endio_func, void *data) |
2fe6303e | 2333 | { |
0b246afa | 2334 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e MX |
2335 | struct bio *bio; |
2336 | struct btrfs_io_bio *btrfs_failed_bio; | |
2337 | struct btrfs_io_bio *btrfs_bio; | |
2338 | ||
c5e4c3d7 | 2339 | bio = btrfs_io_bio_alloc(1); |
2fe6303e | 2340 | bio->bi_end_io = endio_func; |
4f024f37 | 2341 | bio->bi_iter.bi_sector = failrec->logical >> 9; |
74d46992 | 2342 | bio_set_dev(bio, fs_info->fs_devices->latest_bdev); |
4f024f37 | 2343 | bio->bi_iter.bi_size = 0; |
8b110e39 | 2344 | bio->bi_private = data; |
4a54c8c1 | 2345 | |
facc8a22 MX |
2346 | btrfs_failed_bio = btrfs_io_bio(failed_bio); |
2347 | if (btrfs_failed_bio->csum) { | |
facc8a22 MX |
2348 | u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); |
2349 | ||
2350 | btrfs_bio = btrfs_io_bio(bio); | |
2351 | btrfs_bio->csum = btrfs_bio->csum_inline; | |
2fe6303e MX |
2352 | icsum *= csum_size; |
2353 | memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum, | |
facc8a22 MX |
2354 | csum_size); |
2355 | } | |
2356 | ||
2fe6303e MX |
2357 | bio_add_page(bio, page, failrec->len, pg_offset); |
2358 | ||
2359 | return bio; | |
2360 | } | |
2361 | ||
2362 | /* | |
2363 | * this is a generic handler for readpage errors (default | |
2364 | * readpage_io_failed_hook). if other copies exist, read those and write back | |
2365 | * good data to the failed position. does not investigate in remapping the | |
2366 | * failed extent elsewhere, hoping the device will be smart enough to do this as | |
2367 | * needed | |
2368 | */ | |
2369 | ||
2370 | static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset, | |
2371 | struct page *page, u64 start, u64 end, | |
2372 | int failed_mirror) | |
2373 | { | |
2374 | struct io_failure_record *failrec; | |
2375 | struct inode *inode = page->mapping->host; | |
2376 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
7870d082 | 2377 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2fe6303e | 2378 | struct bio *bio; |
70fd7614 | 2379 | int read_mode = 0; |
4e4cbee9 | 2380 | blk_status_t status; |
2fe6303e | 2381 | int ret; |
a0b60d72 | 2382 | unsigned failed_bio_pages = bio_pages_all(failed_bio); |
2fe6303e | 2383 | |
1f7ad75b | 2384 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
2fe6303e MX |
2385 | |
2386 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
2387 | if (ret) | |
2388 | return ret; | |
2389 | ||
a0b60d72 | 2390 | if (!btrfs_check_repairable(inode, failed_bio_pages, failrec, |
c3cfb656 | 2391 | failed_mirror)) { |
7870d082 | 2392 | free_io_failure(failure_tree, tree, failrec); |
2fe6303e MX |
2393 | return -EIO; |
2394 | } | |
2395 | ||
a0b60d72 | 2396 | if (failed_bio_pages > 1) |
70fd7614 | 2397 | read_mode |= REQ_FAILFAST_DEV; |
2fe6303e MX |
2398 | |
2399 | phy_offset >>= inode->i_sb->s_blocksize_bits; | |
2400 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2401 | start - page_offset(page), | |
8b110e39 MX |
2402 | (int)phy_offset, failed_bio->bi_end_io, |
2403 | NULL); | |
1f7ad75b | 2404 | bio_set_op_attrs(bio, REQ_OP_READ, read_mode); |
4a54c8c1 | 2405 | |
ab8d0fc4 JM |
2406 | btrfs_debug(btrfs_sb(inode->i_sb), |
2407 | "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d", | |
2408 | read_mode, failrec->this_mirror, failrec->in_validation); | |
4a54c8c1 | 2409 | |
8c27cb35 | 2410 | status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror, |
013bd4c3 | 2411 | failrec->bio_flags, 0); |
4e4cbee9 | 2412 | if (status) { |
7870d082 | 2413 | free_io_failure(failure_tree, tree, failrec); |
6c387ab2 | 2414 | bio_put(bio); |
4e4cbee9 | 2415 | ret = blk_status_to_errno(status); |
6c387ab2 MX |
2416 | } |
2417 | ||
013bd4c3 | 2418 | return ret; |
4a54c8c1 JS |
2419 | } |
2420 | ||
d1310b2e CM |
2421 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2422 | ||
b5227c07 | 2423 | void end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
87826df0 JM |
2424 | { |
2425 | int uptodate = (err == 0); | |
2426 | struct extent_io_tree *tree; | |
3e2426bd | 2427 | int ret = 0; |
87826df0 JM |
2428 | |
2429 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
2430 | ||
c3988d63 DS |
2431 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2432 | tree->ops->writepage_end_io_hook(page, start, end, NULL, | |
2433 | uptodate); | |
87826df0 | 2434 | |
87826df0 | 2435 | if (!uptodate) { |
87826df0 JM |
2436 | ClearPageUptodate(page); |
2437 | SetPageError(page); | |
bff5baf8 | 2438 | ret = err < 0 ? err : -EIO; |
5dca6eea | 2439 | mapping_set_error(page->mapping, ret); |
87826df0 | 2440 | } |
87826df0 JM |
2441 | } |
2442 | ||
d1310b2e CM |
2443 | /* |
2444 | * after a writepage IO is done, we need to: | |
2445 | * clear the uptodate bits on error | |
2446 | * clear the writeback bits in the extent tree for this IO | |
2447 | * end_page_writeback if the page has no more pending IO | |
2448 | * | |
2449 | * Scheduling is not allowed, so the extent state tree is expected | |
2450 | * to have one and only one object corresponding to this IO. | |
2451 | */ | |
4246a0b6 | 2452 | static void end_bio_extent_writepage(struct bio *bio) |
d1310b2e | 2453 | { |
4e4cbee9 | 2454 | int error = blk_status_to_errno(bio->bi_status); |
2c30c71b | 2455 | struct bio_vec *bvec; |
d1310b2e CM |
2456 | u64 start; |
2457 | u64 end; | |
2c30c71b | 2458 | int i; |
d1310b2e | 2459 | |
c09abff8 | 2460 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2c30c71b | 2461 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2462 | struct page *page = bvec->bv_page; |
0b246afa JM |
2463 | struct inode *inode = page->mapping->host; |
2464 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
902b22f3 | 2465 | |
17a5adcc AO |
2466 | /* We always issue full-page reads, but if some block |
2467 | * in a page fails to read, blk_update_request() will | |
2468 | * advance bv_offset and adjust bv_len to compensate. | |
2469 | * Print a warning for nonzero offsets, and an error | |
2470 | * if they don't add up to a full page. */ | |
09cbfeaf KS |
2471 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { |
2472 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
0b246afa | 2473 | btrfs_err(fs_info, |
efe120a0 FH |
2474 | "partial page write in btrfs with offset %u and length %u", |
2475 | bvec->bv_offset, bvec->bv_len); | |
2476 | else | |
0b246afa | 2477 | btrfs_info(fs_info, |
5d163e0e | 2478 | "incomplete page write in btrfs with offset %u and length %u", |
efe120a0 FH |
2479 | bvec->bv_offset, bvec->bv_len); |
2480 | } | |
d1310b2e | 2481 | |
17a5adcc AO |
2482 | start = page_offset(page); |
2483 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
d1310b2e | 2484 | |
4e4cbee9 | 2485 | end_extent_writepage(page, error, start, end); |
17a5adcc | 2486 | end_page_writeback(page); |
2c30c71b | 2487 | } |
2b1f55b0 | 2488 | |
d1310b2e | 2489 | bio_put(bio); |
d1310b2e CM |
2490 | } |
2491 | ||
883d0de4 MX |
2492 | static void |
2493 | endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len, | |
2494 | int uptodate) | |
2495 | { | |
2496 | struct extent_state *cached = NULL; | |
2497 | u64 end = start + len - 1; | |
2498 | ||
2499 | if (uptodate && tree->track_uptodate) | |
2500 | set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC); | |
d810a4be | 2501 | unlock_extent_cached_atomic(tree, start, end, &cached); |
883d0de4 MX |
2502 | } |
2503 | ||
d1310b2e CM |
2504 | /* |
2505 | * after a readpage IO is done, we need to: | |
2506 | * clear the uptodate bits on error | |
2507 | * set the uptodate bits if things worked | |
2508 | * set the page up to date if all extents in the tree are uptodate | |
2509 | * clear the lock bit in the extent tree | |
2510 | * unlock the page if there are no other extents locked for it | |
2511 | * | |
2512 | * Scheduling is not allowed, so the extent state tree is expected | |
2513 | * to have one and only one object corresponding to this IO. | |
2514 | */ | |
4246a0b6 | 2515 | static void end_bio_extent_readpage(struct bio *bio) |
d1310b2e | 2516 | { |
2c30c71b | 2517 | struct bio_vec *bvec; |
4e4cbee9 | 2518 | int uptodate = !bio->bi_status; |
facc8a22 | 2519 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); |
7870d082 | 2520 | struct extent_io_tree *tree, *failure_tree; |
facc8a22 | 2521 | u64 offset = 0; |
d1310b2e CM |
2522 | u64 start; |
2523 | u64 end; | |
facc8a22 | 2524 | u64 len; |
883d0de4 MX |
2525 | u64 extent_start = 0; |
2526 | u64 extent_len = 0; | |
5cf1ab56 | 2527 | int mirror; |
d1310b2e | 2528 | int ret; |
2c30c71b | 2529 | int i; |
d1310b2e | 2530 | |
c09abff8 | 2531 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2c30c71b | 2532 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2533 | struct page *page = bvec->bv_page; |
a71754fc | 2534 | struct inode *inode = page->mapping->host; |
ab8d0fc4 | 2535 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
507903b8 | 2536 | |
ab8d0fc4 JM |
2537 | btrfs_debug(fs_info, |
2538 | "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u", | |
4e4cbee9 | 2539 | (u64)bio->bi_iter.bi_sector, bio->bi_status, |
ab8d0fc4 | 2540 | io_bio->mirror_num); |
a71754fc | 2541 | tree = &BTRFS_I(inode)->io_tree; |
7870d082 | 2542 | failure_tree = &BTRFS_I(inode)->io_failure_tree; |
902b22f3 | 2543 | |
17a5adcc AO |
2544 | /* We always issue full-page reads, but if some block |
2545 | * in a page fails to read, blk_update_request() will | |
2546 | * advance bv_offset and adjust bv_len to compensate. | |
2547 | * Print a warning for nonzero offsets, and an error | |
2548 | * if they don't add up to a full page. */ | |
09cbfeaf KS |
2549 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { |
2550 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
ab8d0fc4 JM |
2551 | btrfs_err(fs_info, |
2552 | "partial page read in btrfs with offset %u and length %u", | |
efe120a0 FH |
2553 | bvec->bv_offset, bvec->bv_len); |
2554 | else | |
ab8d0fc4 JM |
2555 | btrfs_info(fs_info, |
2556 | "incomplete page read in btrfs with offset %u and length %u", | |
efe120a0 FH |
2557 | bvec->bv_offset, bvec->bv_len); |
2558 | } | |
d1310b2e | 2559 | |
17a5adcc AO |
2560 | start = page_offset(page); |
2561 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
facc8a22 | 2562 | len = bvec->bv_len; |
d1310b2e | 2563 | |
9be3395b | 2564 | mirror = io_bio->mirror_num; |
20c9801d | 2565 | if (likely(uptodate && tree->ops)) { |
facc8a22 MX |
2566 | ret = tree->ops->readpage_end_io_hook(io_bio, offset, |
2567 | page, start, end, | |
2568 | mirror); | |
5ee0844d | 2569 | if (ret) |
d1310b2e | 2570 | uptodate = 0; |
5ee0844d | 2571 | else |
7870d082 JB |
2572 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
2573 | failure_tree, tree, start, | |
2574 | page, | |
2575 | btrfs_ino(BTRFS_I(inode)), 0); | |
d1310b2e | 2576 | } |
ea466794 | 2577 | |
f2a09da9 MX |
2578 | if (likely(uptodate)) |
2579 | goto readpage_ok; | |
2580 | ||
20a7db8a | 2581 | if (tree->ops) { |
5cf1ab56 | 2582 | ret = tree->ops->readpage_io_failed_hook(page, mirror); |
9d0d1c8b LB |
2583 | if (ret == -EAGAIN) { |
2584 | /* | |
2585 | * Data inode's readpage_io_failed_hook() always | |
2586 | * returns -EAGAIN. | |
2587 | * | |
2588 | * The generic bio_readpage_error handles errors | |
2589 | * the following way: If possible, new read | |
2590 | * requests are created and submitted and will | |
2591 | * end up in end_bio_extent_readpage as well (if | |
2592 | * we're lucky, not in the !uptodate case). In | |
2593 | * that case it returns 0 and we just go on with | |
2594 | * the next page in our bio. If it can't handle | |
2595 | * the error it will return -EIO and we remain | |
2596 | * responsible for that page. | |
2597 | */ | |
2598 | ret = bio_readpage_error(bio, offset, page, | |
2599 | start, end, mirror); | |
2600 | if (ret == 0) { | |
4e4cbee9 | 2601 | uptodate = !bio->bi_status; |
9d0d1c8b LB |
2602 | offset += len; |
2603 | continue; | |
2604 | } | |
2605 | } | |
2606 | ||
f4a8e656 | 2607 | /* |
9d0d1c8b LB |
2608 | * metadata's readpage_io_failed_hook() always returns |
2609 | * -EIO and fixes nothing. -EIO is also returned if | |
2610 | * data inode error could not be fixed. | |
f4a8e656 | 2611 | */ |
9d0d1c8b | 2612 | ASSERT(ret == -EIO); |
7e38326f | 2613 | } |
f2a09da9 | 2614 | readpage_ok: |
883d0de4 | 2615 | if (likely(uptodate)) { |
a71754fc | 2616 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 2617 | pgoff_t end_index = i_size >> PAGE_SHIFT; |
a583c026 | 2618 | unsigned off; |
a71754fc JB |
2619 | |
2620 | /* Zero out the end if this page straddles i_size */ | |
09cbfeaf | 2621 | off = i_size & (PAGE_SIZE-1); |
a583c026 | 2622 | if (page->index == end_index && off) |
09cbfeaf | 2623 | zero_user_segment(page, off, PAGE_SIZE); |
17a5adcc | 2624 | SetPageUptodate(page); |
70dec807 | 2625 | } else { |
17a5adcc AO |
2626 | ClearPageUptodate(page); |
2627 | SetPageError(page); | |
70dec807 | 2628 | } |
17a5adcc | 2629 | unlock_page(page); |
facc8a22 | 2630 | offset += len; |
883d0de4 MX |
2631 | |
2632 | if (unlikely(!uptodate)) { | |
2633 | if (extent_len) { | |
2634 | endio_readpage_release_extent(tree, | |
2635 | extent_start, | |
2636 | extent_len, 1); | |
2637 | extent_start = 0; | |
2638 | extent_len = 0; | |
2639 | } | |
2640 | endio_readpage_release_extent(tree, start, | |
2641 | end - start + 1, 0); | |
2642 | } else if (!extent_len) { | |
2643 | extent_start = start; | |
2644 | extent_len = end + 1 - start; | |
2645 | } else if (extent_start + extent_len == start) { | |
2646 | extent_len += end + 1 - start; | |
2647 | } else { | |
2648 | endio_readpage_release_extent(tree, extent_start, | |
2649 | extent_len, uptodate); | |
2650 | extent_start = start; | |
2651 | extent_len = end + 1 - start; | |
2652 | } | |
2c30c71b | 2653 | } |
d1310b2e | 2654 | |
883d0de4 MX |
2655 | if (extent_len) |
2656 | endio_readpage_release_extent(tree, extent_start, extent_len, | |
2657 | uptodate); | |
facc8a22 | 2658 | if (io_bio->end_io) |
4e4cbee9 | 2659 | io_bio->end_io(io_bio, blk_status_to_errno(bio->bi_status)); |
d1310b2e | 2660 | bio_put(bio); |
d1310b2e CM |
2661 | } |
2662 | ||
9be3395b | 2663 | /* |
184f999e DS |
2664 | * Initialize the members up to but not including 'bio'. Use after allocating a |
2665 | * new bio by bio_alloc_bioset as it does not initialize the bytes outside of | |
2666 | * 'bio' because use of __GFP_ZERO is not supported. | |
9be3395b | 2667 | */ |
184f999e | 2668 | static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio) |
d1310b2e | 2669 | { |
184f999e DS |
2670 | memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio)); |
2671 | } | |
d1310b2e | 2672 | |
9be3395b | 2673 | /* |
6e707bcd DS |
2674 | * The following helpers allocate a bio. As it's backed by a bioset, it'll |
2675 | * never fail. We're returning a bio right now but you can call btrfs_io_bio | |
2676 | * for the appropriate container_of magic | |
9be3395b | 2677 | */ |
c821e7f3 | 2678 | struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte) |
d1310b2e CM |
2679 | { |
2680 | struct bio *bio; | |
d1310b2e | 2681 | |
9f2179a5 | 2682 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, btrfs_bioset); |
74d46992 | 2683 | bio_set_dev(bio, bdev); |
c821e7f3 | 2684 | bio->bi_iter.bi_sector = first_byte >> 9; |
184f999e | 2685 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
d1310b2e CM |
2686 | return bio; |
2687 | } | |
2688 | ||
8b6c1d56 | 2689 | struct bio *btrfs_bio_clone(struct bio *bio) |
9be3395b | 2690 | { |
23ea8e5a MX |
2691 | struct btrfs_io_bio *btrfs_bio; |
2692 | struct bio *new; | |
9be3395b | 2693 | |
6e707bcd | 2694 | /* Bio allocation backed by a bioset does not fail */ |
8b6c1d56 | 2695 | new = bio_clone_fast(bio, GFP_NOFS, btrfs_bioset); |
6e707bcd | 2696 | btrfs_bio = btrfs_io_bio(new); |
184f999e | 2697 | btrfs_io_bio_init(btrfs_bio); |
6e707bcd | 2698 | btrfs_bio->iter = bio->bi_iter; |
23ea8e5a MX |
2699 | return new; |
2700 | } | |
9be3395b | 2701 | |
c5e4c3d7 | 2702 | struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs) |
9be3395b | 2703 | { |
facc8a22 MX |
2704 | struct bio *bio; |
2705 | ||
6e707bcd | 2706 | /* Bio allocation backed by a bioset does not fail */ |
c5e4c3d7 | 2707 | bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, btrfs_bioset); |
184f999e | 2708 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
facc8a22 | 2709 | return bio; |
9be3395b CM |
2710 | } |
2711 | ||
e477094f | 2712 | struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size) |
2f8e9140 LB |
2713 | { |
2714 | struct bio *bio; | |
2715 | struct btrfs_io_bio *btrfs_bio; | |
2716 | ||
2717 | /* this will never fail when it's backed by a bioset */ | |
e477094f | 2718 | bio = bio_clone_fast(orig, GFP_NOFS, btrfs_bioset); |
2f8e9140 LB |
2719 | ASSERT(bio); |
2720 | ||
2721 | btrfs_bio = btrfs_io_bio(bio); | |
184f999e | 2722 | btrfs_io_bio_init(btrfs_bio); |
2f8e9140 LB |
2723 | |
2724 | bio_trim(bio, offset >> 9, size >> 9); | |
17347cec | 2725 | btrfs_bio->iter = bio->bi_iter; |
2f8e9140 LB |
2726 | return bio; |
2727 | } | |
9be3395b | 2728 | |
1f7ad75b MC |
2729 | static int __must_check submit_one_bio(struct bio *bio, int mirror_num, |
2730 | unsigned long bio_flags) | |
d1310b2e | 2731 | { |
4e4cbee9 | 2732 | blk_status_t ret = 0; |
c45a8f2d | 2733 | struct bio_vec *bvec = bio_last_bvec_all(bio); |
70dec807 CM |
2734 | struct page *page = bvec->bv_page; |
2735 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 2736 | u64 start; |
70dec807 | 2737 | |
4eee4fa4 | 2738 | start = page_offset(page) + bvec->bv_offset; |
70dec807 | 2739 | |
902b22f3 | 2740 | bio->bi_private = NULL; |
d1310b2e | 2741 | |
20c9801d | 2742 | if (tree->ops) |
c6100a4b | 2743 | ret = tree->ops->submit_bio_hook(tree->private_data, bio, |
eaf25d93 | 2744 | mirror_num, bio_flags, start); |
0b86a832 | 2745 | else |
4e49ea4a | 2746 | btrfsic_submit_bio(bio); |
4a54c8c1 | 2747 | |
4e4cbee9 | 2748 | return blk_status_to_errno(ret); |
d1310b2e CM |
2749 | } |
2750 | ||
4b81ba48 DS |
2751 | /* |
2752 | * @opf: bio REQ_OP_* and REQ_* flags as one value | |
b8b3d625 DS |
2753 | * @tree: tree so we can call our merge_bio hook |
2754 | * @wbc: optional writeback control for io accounting | |
2755 | * @page: page to add to the bio | |
2756 | * @pg_offset: offset of the new bio or to check whether we are adding | |
2757 | * a contiguous page to the previous one | |
2758 | * @size: portion of page that we want to write | |
2759 | * @offset: starting offset in the page | |
2760 | * @bdev: attach newly created bios to this bdev | |
5c2b1fd7 | 2761 | * @bio_ret: must be valid pointer, newly allocated bio will be stored there |
b8b3d625 DS |
2762 | * @end_io_func: end_io callback for new bio |
2763 | * @mirror_num: desired mirror to read/write | |
2764 | * @prev_bio_flags: flags of previous bio to see if we can merge the current one | |
2765 | * @bio_flags: flags of the current bio to see if we can merge them | |
4b81ba48 DS |
2766 | */ |
2767 | static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree, | |
da2f0f74 | 2768 | struct writeback_control *wbc, |
6273b7f8 | 2769 | struct page *page, u64 offset, |
6c5a4e2c | 2770 | size_t size, unsigned long pg_offset, |
d1310b2e CM |
2771 | struct block_device *bdev, |
2772 | struct bio **bio_ret, | |
f188591e | 2773 | bio_end_io_t end_io_func, |
c8b97818 CM |
2774 | int mirror_num, |
2775 | unsigned long prev_bio_flags, | |
005efedf FM |
2776 | unsigned long bio_flags, |
2777 | bool force_bio_submit) | |
d1310b2e CM |
2778 | { |
2779 | int ret = 0; | |
2780 | struct bio *bio; | |
09cbfeaf | 2781 | size_t page_size = min_t(size_t, size, PAGE_SIZE); |
6273b7f8 | 2782 | sector_t sector = offset >> 9; |
d1310b2e | 2783 | |
5c2b1fd7 DS |
2784 | ASSERT(bio_ret); |
2785 | ||
2786 | if (*bio_ret) { | |
0c8508a6 DS |
2787 | bool contig; |
2788 | bool can_merge = true; | |
2789 | ||
d1310b2e | 2790 | bio = *bio_ret; |
0c8508a6 | 2791 | if (prev_bio_flags & EXTENT_BIO_COMPRESSED) |
4f024f37 | 2792 | contig = bio->bi_iter.bi_sector == sector; |
c8b97818 | 2793 | else |
f73a1c7d | 2794 | contig = bio_end_sector(bio) == sector; |
c8b97818 | 2795 | |
0c8508a6 DS |
2796 | if (tree->ops && tree->ops->merge_bio_hook(page, offset, |
2797 | page_size, bio, bio_flags)) | |
2798 | can_merge = false; | |
2799 | ||
2800 | if (prev_bio_flags != bio_flags || !contig || !can_merge || | |
005efedf | 2801 | force_bio_submit || |
6c5a4e2c | 2802 | bio_add_page(bio, page, page_size, pg_offset) < page_size) { |
1f7ad75b | 2803 | ret = submit_one_bio(bio, mirror_num, prev_bio_flags); |
289454ad NA |
2804 | if (ret < 0) { |
2805 | *bio_ret = NULL; | |
79787eaa | 2806 | return ret; |
289454ad | 2807 | } |
d1310b2e CM |
2808 | bio = NULL; |
2809 | } else { | |
da2f0f74 CM |
2810 | if (wbc) |
2811 | wbc_account_io(wbc, page, page_size); | |
d1310b2e CM |
2812 | return 0; |
2813 | } | |
2814 | } | |
c8b97818 | 2815 | |
6273b7f8 | 2816 | bio = btrfs_bio_alloc(bdev, offset); |
6c5a4e2c | 2817 | bio_add_page(bio, page, page_size, pg_offset); |
d1310b2e CM |
2818 | bio->bi_end_io = end_io_func; |
2819 | bio->bi_private = tree; | |
e6959b93 | 2820 | bio->bi_write_hint = page->mapping->host->i_write_hint; |
4b81ba48 | 2821 | bio->bi_opf = opf; |
da2f0f74 CM |
2822 | if (wbc) { |
2823 | wbc_init_bio(wbc, bio); | |
2824 | wbc_account_io(wbc, page, page_size); | |
2825 | } | |
70dec807 | 2826 | |
5c2b1fd7 | 2827 | *bio_ret = bio; |
d1310b2e CM |
2828 | |
2829 | return ret; | |
2830 | } | |
2831 | ||
48a3b636 ES |
2832 | static void attach_extent_buffer_page(struct extent_buffer *eb, |
2833 | struct page *page) | |
d1310b2e CM |
2834 | { |
2835 | if (!PagePrivate(page)) { | |
2836 | SetPagePrivate(page); | |
09cbfeaf | 2837 | get_page(page); |
4f2de97a JB |
2838 | set_page_private(page, (unsigned long)eb); |
2839 | } else { | |
2840 | WARN_ON(page->private != (unsigned long)eb); | |
d1310b2e CM |
2841 | } |
2842 | } | |
2843 | ||
4f2de97a | 2844 | void set_page_extent_mapped(struct page *page) |
d1310b2e | 2845 | { |
4f2de97a JB |
2846 | if (!PagePrivate(page)) { |
2847 | SetPagePrivate(page); | |
09cbfeaf | 2848 | get_page(page); |
4f2de97a JB |
2849 | set_page_private(page, EXTENT_PAGE_PRIVATE); |
2850 | } | |
d1310b2e CM |
2851 | } |
2852 | ||
125bac01 MX |
2853 | static struct extent_map * |
2854 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
2855 | u64 start, u64 len, get_extent_t *get_extent, | |
2856 | struct extent_map **em_cached) | |
2857 | { | |
2858 | struct extent_map *em; | |
2859 | ||
2860 | if (em_cached && *em_cached) { | |
2861 | em = *em_cached; | |
cbc0e928 | 2862 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 | 2863 | start < extent_map_end(em)) { |
490b54d6 | 2864 | refcount_inc(&em->refs); |
125bac01 MX |
2865 | return em; |
2866 | } | |
2867 | ||
2868 | free_extent_map(em); | |
2869 | *em_cached = NULL; | |
2870 | } | |
2871 | ||
fc4f21b1 | 2872 | em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0); |
125bac01 MX |
2873 | if (em_cached && !IS_ERR_OR_NULL(em)) { |
2874 | BUG_ON(*em_cached); | |
490b54d6 | 2875 | refcount_inc(&em->refs); |
125bac01 MX |
2876 | *em_cached = em; |
2877 | } | |
2878 | return em; | |
2879 | } | |
d1310b2e CM |
2880 | /* |
2881 | * basic readpage implementation. Locked extent state structs are inserted | |
2882 | * into the tree that are removed when the IO is done (by the end_io | |
2883 | * handlers) | |
79787eaa | 2884 | * XXX JDM: This needs looking at to ensure proper page locking |
baf863b9 | 2885 | * return 0 on success, otherwise return error |
d1310b2e | 2886 | */ |
9974090b MX |
2887 | static int __do_readpage(struct extent_io_tree *tree, |
2888 | struct page *page, | |
2889 | get_extent_t *get_extent, | |
125bac01 | 2890 | struct extent_map **em_cached, |
9974090b | 2891 | struct bio **bio, int mirror_num, |
f1c77c55 | 2892 | unsigned long *bio_flags, unsigned int read_flags, |
005efedf | 2893 | u64 *prev_em_start) |
d1310b2e CM |
2894 | { |
2895 | struct inode *inode = page->mapping->host; | |
4eee4fa4 | 2896 | u64 start = page_offset(page); |
8eec8296 | 2897 | const u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
2898 | u64 cur = start; |
2899 | u64 extent_offset; | |
2900 | u64 last_byte = i_size_read(inode); | |
2901 | u64 block_start; | |
2902 | u64 cur_end; | |
d1310b2e CM |
2903 | struct extent_map *em; |
2904 | struct block_device *bdev; | |
baf863b9 | 2905 | int ret = 0; |
d1310b2e | 2906 | int nr = 0; |
306e16ce | 2907 | size_t pg_offset = 0; |
d1310b2e | 2908 | size_t iosize; |
c8b97818 | 2909 | size_t disk_io_size; |
d1310b2e | 2910 | size_t blocksize = inode->i_sb->s_blocksize; |
7f042a83 | 2911 | unsigned long this_bio_flag = 0; |
d1310b2e CM |
2912 | |
2913 | set_page_extent_mapped(page); | |
2914 | ||
90a887c9 DM |
2915 | if (!PageUptodate(page)) { |
2916 | if (cleancache_get_page(page) == 0) { | |
2917 | BUG_ON(blocksize != PAGE_SIZE); | |
9974090b | 2918 | unlock_extent(tree, start, end); |
90a887c9 DM |
2919 | goto out; |
2920 | } | |
2921 | } | |
2922 | ||
09cbfeaf | 2923 | if (page->index == last_byte >> PAGE_SHIFT) { |
c8b97818 | 2924 | char *userpage; |
09cbfeaf | 2925 | size_t zero_offset = last_byte & (PAGE_SIZE - 1); |
c8b97818 CM |
2926 | |
2927 | if (zero_offset) { | |
09cbfeaf | 2928 | iosize = PAGE_SIZE - zero_offset; |
7ac687d9 | 2929 | userpage = kmap_atomic(page); |
c8b97818 CM |
2930 | memset(userpage + zero_offset, 0, iosize); |
2931 | flush_dcache_page(page); | |
7ac687d9 | 2932 | kunmap_atomic(userpage); |
c8b97818 CM |
2933 | } |
2934 | } | |
d1310b2e | 2935 | while (cur <= end) { |
005efedf | 2936 | bool force_bio_submit = false; |
6273b7f8 | 2937 | u64 offset; |
c8f2f24b | 2938 | |
d1310b2e CM |
2939 | if (cur >= last_byte) { |
2940 | char *userpage; | |
507903b8 AJ |
2941 | struct extent_state *cached = NULL; |
2942 | ||
09cbfeaf | 2943 | iosize = PAGE_SIZE - pg_offset; |
7ac687d9 | 2944 | userpage = kmap_atomic(page); |
306e16ce | 2945 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2946 | flush_dcache_page(page); |
7ac687d9 | 2947 | kunmap_atomic(userpage); |
d1310b2e | 2948 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 2949 | &cached, GFP_NOFS); |
7f042a83 | 2950 | unlock_extent_cached(tree, cur, |
e43bbe5e | 2951 | cur + iosize - 1, &cached); |
d1310b2e CM |
2952 | break; |
2953 | } | |
125bac01 MX |
2954 | em = __get_extent_map(inode, page, pg_offset, cur, |
2955 | end - cur + 1, get_extent, em_cached); | |
c704005d | 2956 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 2957 | SetPageError(page); |
7f042a83 | 2958 | unlock_extent(tree, cur, end); |
d1310b2e CM |
2959 | break; |
2960 | } | |
d1310b2e CM |
2961 | extent_offset = cur - em->start; |
2962 | BUG_ON(extent_map_end(em) <= cur); | |
2963 | BUG_ON(end < cur); | |
2964 | ||
261507a0 | 2965 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
4b384318 | 2966 | this_bio_flag |= EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
2967 | extent_set_compress_type(&this_bio_flag, |
2968 | em->compress_type); | |
2969 | } | |
c8b97818 | 2970 | |
d1310b2e CM |
2971 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
2972 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 2973 | iosize = ALIGN(iosize, blocksize); |
c8b97818 CM |
2974 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
2975 | disk_io_size = em->block_len; | |
6273b7f8 | 2976 | offset = em->block_start; |
c8b97818 | 2977 | } else { |
6273b7f8 | 2978 | offset = em->block_start + extent_offset; |
c8b97818 CM |
2979 | disk_io_size = iosize; |
2980 | } | |
d1310b2e CM |
2981 | bdev = em->bdev; |
2982 | block_start = em->block_start; | |
d899e052 YZ |
2983 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2984 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
2985 | |
2986 | /* | |
2987 | * If we have a file range that points to a compressed extent | |
2988 | * and it's followed by a consecutive file range that points to | |
2989 | * to the same compressed extent (possibly with a different | |
2990 | * offset and/or length, so it either points to the whole extent | |
2991 | * or only part of it), we must make sure we do not submit a | |
2992 | * single bio to populate the pages for the 2 ranges because | |
2993 | * this makes the compressed extent read zero out the pages | |
2994 | * belonging to the 2nd range. Imagine the following scenario: | |
2995 | * | |
2996 | * File layout | |
2997 | * [0 - 8K] [8K - 24K] | |
2998 | * | | | |
2999 | * | | | |
3000 | * points to extent X, points to extent X, | |
3001 | * offset 4K, length of 8K offset 0, length 16K | |
3002 | * | |
3003 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
3004 | * | |
3005 | * If the bio to read the compressed extent covers both ranges, | |
3006 | * it will decompress extent X into the pages belonging to the | |
3007 | * first range and then it will stop, zeroing out the remaining | |
3008 | * pages that belong to the other range that points to extent X. | |
3009 | * So here we make sure we submit 2 bios, one for the first | |
3010 | * range and another one for the third range. Both will target | |
3011 | * the same physical extent from disk, but we can't currently | |
3012 | * make the compressed bio endio callback populate the pages | |
3013 | * for both ranges because each compressed bio is tightly | |
3014 | * coupled with a single extent map, and each range can have | |
3015 | * an extent map with a different offset value relative to the | |
3016 | * uncompressed data of our extent and different lengths. This | |
3017 | * is a corner case so we prioritize correctness over | |
3018 | * non-optimal behavior (submitting 2 bios for the same extent). | |
3019 | */ | |
3020 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
3021 | prev_em_start && *prev_em_start != (u64)-1 && | |
3022 | *prev_em_start != em->orig_start) | |
3023 | force_bio_submit = true; | |
3024 | ||
3025 | if (prev_em_start) | |
3026 | *prev_em_start = em->orig_start; | |
3027 | ||
d1310b2e CM |
3028 | free_extent_map(em); |
3029 | em = NULL; | |
3030 | ||
3031 | /* we've found a hole, just zero and go on */ | |
3032 | if (block_start == EXTENT_MAP_HOLE) { | |
3033 | char *userpage; | |
507903b8 AJ |
3034 | struct extent_state *cached = NULL; |
3035 | ||
7ac687d9 | 3036 | userpage = kmap_atomic(page); |
306e16ce | 3037 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3038 | flush_dcache_page(page); |
7ac687d9 | 3039 | kunmap_atomic(userpage); |
d1310b2e CM |
3040 | |
3041 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 | 3042 | &cached, GFP_NOFS); |
7f042a83 | 3043 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3044 | cur + iosize - 1, &cached); |
d1310b2e | 3045 | cur = cur + iosize; |
306e16ce | 3046 | pg_offset += iosize; |
d1310b2e CM |
3047 | continue; |
3048 | } | |
3049 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
3050 | if (test_range_bit(tree, cur, cur_end, |
3051 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 3052 | check_page_uptodate(tree, page); |
7f042a83 | 3053 | unlock_extent(tree, cur, cur + iosize - 1); |
d1310b2e | 3054 | cur = cur + iosize; |
306e16ce | 3055 | pg_offset += iosize; |
d1310b2e CM |
3056 | continue; |
3057 | } | |
70dec807 CM |
3058 | /* we have an inline extent but it didn't get marked up |
3059 | * to date. Error out | |
3060 | */ | |
3061 | if (block_start == EXTENT_MAP_INLINE) { | |
3062 | SetPageError(page); | |
7f042a83 | 3063 | unlock_extent(tree, cur, cur + iosize - 1); |
70dec807 | 3064 | cur = cur + iosize; |
306e16ce | 3065 | pg_offset += iosize; |
70dec807 CM |
3066 | continue; |
3067 | } | |
d1310b2e | 3068 | |
4b81ba48 | 3069 | ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL, |
6273b7f8 DS |
3070 | page, offset, disk_io_size, |
3071 | pg_offset, bdev, bio, | |
c8b97818 CM |
3072 | end_bio_extent_readpage, mirror_num, |
3073 | *bio_flags, | |
005efedf FM |
3074 | this_bio_flag, |
3075 | force_bio_submit); | |
c8f2f24b JB |
3076 | if (!ret) { |
3077 | nr++; | |
3078 | *bio_flags = this_bio_flag; | |
3079 | } else { | |
d1310b2e | 3080 | SetPageError(page); |
7f042a83 | 3081 | unlock_extent(tree, cur, cur + iosize - 1); |
baf863b9 | 3082 | goto out; |
edd33c99 | 3083 | } |
d1310b2e | 3084 | cur = cur + iosize; |
306e16ce | 3085 | pg_offset += iosize; |
d1310b2e | 3086 | } |
90a887c9 | 3087 | out: |
d1310b2e CM |
3088 | if (!nr) { |
3089 | if (!PageError(page)) | |
3090 | SetPageUptodate(page); | |
3091 | unlock_page(page); | |
3092 | } | |
baf863b9 | 3093 | return ret; |
d1310b2e CM |
3094 | } |
3095 | ||
9974090b MX |
3096 | static inline void __do_contiguous_readpages(struct extent_io_tree *tree, |
3097 | struct page *pages[], int nr_pages, | |
3098 | u64 start, u64 end, | |
125bac01 | 3099 | struct extent_map **em_cached, |
d3fac6ba | 3100 | struct bio **bio, |
1f7ad75b | 3101 | unsigned long *bio_flags, |
808f80b4 | 3102 | u64 *prev_em_start) |
9974090b MX |
3103 | { |
3104 | struct inode *inode; | |
3105 | struct btrfs_ordered_extent *ordered; | |
3106 | int index; | |
3107 | ||
3108 | inode = pages[0]->mapping->host; | |
3109 | while (1) { | |
3110 | lock_extent(tree, start, end); | |
a776c6fa | 3111 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
9974090b MX |
3112 | end - start + 1); |
3113 | if (!ordered) | |
3114 | break; | |
3115 | unlock_extent(tree, start, end); | |
3116 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3117 | btrfs_put_ordered_extent(ordered); | |
3118 | } | |
3119 | ||
3120 | for (index = 0; index < nr_pages; index++) { | |
4ef77695 DS |
3121 | __do_readpage(tree, pages[index], btrfs_get_extent, em_cached, |
3122 | bio, 0, bio_flags, 0, prev_em_start); | |
09cbfeaf | 3123 | put_page(pages[index]); |
9974090b MX |
3124 | } |
3125 | } | |
3126 | ||
3127 | static void __extent_readpages(struct extent_io_tree *tree, | |
3128 | struct page *pages[], | |
e4d17ef5 | 3129 | int nr_pages, |
125bac01 | 3130 | struct extent_map **em_cached, |
d3fac6ba | 3131 | struct bio **bio, unsigned long *bio_flags, |
808f80b4 | 3132 | u64 *prev_em_start) |
9974090b | 3133 | { |
35a3621b | 3134 | u64 start = 0; |
9974090b MX |
3135 | u64 end = 0; |
3136 | u64 page_start; | |
3137 | int index; | |
35a3621b | 3138 | int first_index = 0; |
9974090b MX |
3139 | |
3140 | for (index = 0; index < nr_pages; index++) { | |
3141 | page_start = page_offset(pages[index]); | |
3142 | if (!end) { | |
3143 | start = page_start; | |
09cbfeaf | 3144 | end = start + PAGE_SIZE - 1; |
9974090b MX |
3145 | first_index = index; |
3146 | } else if (end + 1 == page_start) { | |
09cbfeaf | 3147 | end += PAGE_SIZE; |
9974090b MX |
3148 | } else { |
3149 | __do_contiguous_readpages(tree, &pages[first_index], | |
3150 | index - first_index, start, | |
4ef77695 | 3151 | end, em_cached, |
d3fac6ba | 3152 | bio, bio_flags, |
1f7ad75b | 3153 | prev_em_start); |
9974090b | 3154 | start = page_start; |
09cbfeaf | 3155 | end = start + PAGE_SIZE - 1; |
9974090b MX |
3156 | first_index = index; |
3157 | } | |
3158 | } | |
3159 | ||
3160 | if (end) | |
3161 | __do_contiguous_readpages(tree, &pages[first_index], | |
3162 | index - first_index, start, | |
4ef77695 | 3163 | end, em_cached, bio, |
d3fac6ba | 3164 | bio_flags, prev_em_start); |
9974090b MX |
3165 | } |
3166 | ||
3167 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
3168 | struct page *page, | |
3169 | get_extent_t *get_extent, | |
3170 | struct bio **bio, int mirror_num, | |
f1c77c55 DS |
3171 | unsigned long *bio_flags, |
3172 | unsigned int read_flags) | |
9974090b MX |
3173 | { |
3174 | struct inode *inode = page->mapping->host; | |
3175 | struct btrfs_ordered_extent *ordered; | |
3176 | u64 start = page_offset(page); | |
09cbfeaf | 3177 | u64 end = start + PAGE_SIZE - 1; |
9974090b MX |
3178 | int ret; |
3179 | ||
3180 | while (1) { | |
3181 | lock_extent(tree, start, end); | |
a776c6fa | 3182 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
09cbfeaf | 3183 | PAGE_SIZE); |
9974090b MX |
3184 | if (!ordered) |
3185 | break; | |
3186 | unlock_extent(tree, start, end); | |
3187 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3188 | btrfs_put_ordered_extent(ordered); | |
3189 | } | |
3190 | ||
125bac01 | 3191 | ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num, |
1f7ad75b | 3192 | bio_flags, read_flags, NULL); |
9974090b MX |
3193 | return ret; |
3194 | } | |
3195 | ||
d1310b2e | 3196 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, |
8ddc7d9c | 3197 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3198 | { |
3199 | struct bio *bio = NULL; | |
c8b97818 | 3200 | unsigned long bio_flags = 0; |
d1310b2e CM |
3201 | int ret; |
3202 | ||
8ddc7d9c | 3203 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, |
1f7ad75b | 3204 | &bio_flags, 0); |
d1310b2e | 3205 | if (bio) |
1f7ad75b | 3206 | ret = submit_one_bio(bio, mirror_num, bio_flags); |
d1310b2e CM |
3207 | return ret; |
3208 | } | |
d1310b2e | 3209 | |
3d4b9496 | 3210 | static void update_nr_written(struct writeback_control *wbc, |
a9132667 | 3211 | unsigned long nr_written) |
11c8349b CM |
3212 | { |
3213 | wbc->nr_to_write -= nr_written; | |
11c8349b CM |
3214 | } |
3215 | ||
d1310b2e | 3216 | /* |
40f76580 CM |
3217 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
3218 | * | |
3219 | * This returns 1 if our fill_delalloc function did all the work required | |
3220 | * to write the page (copy into inline extent). In this case the IO has | |
3221 | * been started and the page is already unlocked. | |
3222 | * | |
3223 | * This returns 0 if all went well (page still locked) | |
3224 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 3225 | */ |
40f76580 CM |
3226 | static noinline_for_stack int writepage_delalloc(struct inode *inode, |
3227 | struct page *page, struct writeback_control *wbc, | |
3228 | struct extent_page_data *epd, | |
3229 | u64 delalloc_start, | |
3230 | unsigned long *nr_written) | |
3231 | { | |
3232 | struct extent_io_tree *tree = epd->tree; | |
09cbfeaf | 3233 | u64 page_end = delalloc_start + PAGE_SIZE - 1; |
40f76580 CM |
3234 | u64 nr_delalloc; |
3235 | u64 delalloc_to_write = 0; | |
3236 | u64 delalloc_end = 0; | |
3237 | int ret; | |
3238 | int page_started = 0; | |
3239 | ||
3240 | if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc) | |
3241 | return 0; | |
3242 | ||
3243 | while (delalloc_end < page_end) { | |
3244 | nr_delalloc = find_lock_delalloc_range(inode, tree, | |
3245 | page, | |
3246 | &delalloc_start, | |
3247 | &delalloc_end, | |
dcab6a3b | 3248 | BTRFS_MAX_EXTENT_SIZE); |
40f76580 CM |
3249 | if (nr_delalloc == 0) { |
3250 | delalloc_start = delalloc_end + 1; | |
3251 | continue; | |
3252 | } | |
3253 | ret = tree->ops->fill_delalloc(inode, page, | |
3254 | delalloc_start, | |
3255 | delalloc_end, | |
3256 | &page_started, | |
f82b7359 | 3257 | nr_written, wbc); |
40f76580 CM |
3258 | /* File system has been set read-only */ |
3259 | if (ret) { | |
3260 | SetPageError(page); | |
3261 | /* fill_delalloc should be return < 0 for error | |
3262 | * but just in case, we use > 0 here meaning the | |
3263 | * IO is started, so we don't want to return > 0 | |
3264 | * unless things are going well. | |
3265 | */ | |
3266 | ret = ret < 0 ? ret : -EIO; | |
3267 | goto done; | |
3268 | } | |
3269 | /* | |
ea1754a0 KS |
3270 | * delalloc_end is already one less than the total length, so |
3271 | * we don't subtract one from PAGE_SIZE | |
40f76580 CM |
3272 | */ |
3273 | delalloc_to_write += (delalloc_end - delalloc_start + | |
ea1754a0 | 3274 | PAGE_SIZE) >> PAGE_SHIFT; |
40f76580 CM |
3275 | delalloc_start = delalloc_end + 1; |
3276 | } | |
3277 | if (wbc->nr_to_write < delalloc_to_write) { | |
3278 | int thresh = 8192; | |
3279 | ||
3280 | if (delalloc_to_write < thresh * 2) | |
3281 | thresh = delalloc_to_write; | |
3282 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3283 | thresh); | |
3284 | } | |
3285 | ||
3286 | /* did the fill delalloc function already unlock and start | |
3287 | * the IO? | |
3288 | */ | |
3289 | if (page_started) { | |
3290 | /* | |
3291 | * we've unlocked the page, so we can't update | |
3292 | * the mapping's writeback index, just update | |
3293 | * nr_to_write. | |
3294 | */ | |
3295 | wbc->nr_to_write -= *nr_written; | |
3296 | return 1; | |
3297 | } | |
3298 | ||
3299 | ret = 0; | |
3300 | ||
3301 | done: | |
3302 | return ret; | |
3303 | } | |
3304 | ||
3305 | /* | |
3306 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3307 | * and does the loop to map the page into extents and bios. | |
3308 | * | |
3309 | * We return 1 if the IO is started and the page is unlocked, | |
3310 | * 0 if all went well (page still locked) | |
3311 | * < 0 if there were errors (page still locked) | |
3312 | */ | |
3313 | static noinline_for_stack int __extent_writepage_io(struct inode *inode, | |
3314 | struct page *page, | |
3315 | struct writeback_control *wbc, | |
3316 | struct extent_page_data *epd, | |
3317 | loff_t i_size, | |
3318 | unsigned long nr_written, | |
f1c77c55 | 3319 | unsigned int write_flags, int *nr_ret) |
d1310b2e | 3320 | { |
d1310b2e | 3321 | struct extent_io_tree *tree = epd->tree; |
4eee4fa4 | 3322 | u64 start = page_offset(page); |
09cbfeaf | 3323 | u64 page_end = start + PAGE_SIZE - 1; |
d1310b2e CM |
3324 | u64 end; |
3325 | u64 cur = start; | |
3326 | u64 extent_offset; | |
d1310b2e CM |
3327 | u64 block_start; |
3328 | u64 iosize; | |
d1310b2e CM |
3329 | struct extent_map *em; |
3330 | struct block_device *bdev; | |
7f3c74fb | 3331 | size_t pg_offset = 0; |
d1310b2e | 3332 | size_t blocksize; |
40f76580 CM |
3333 | int ret = 0; |
3334 | int nr = 0; | |
3335 | bool compressed; | |
c8b97818 | 3336 | |
247e743c | 3337 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
3338 | ret = tree->ops->writepage_start_hook(page, start, |
3339 | page_end); | |
87826df0 JM |
3340 | if (ret) { |
3341 | /* Fixup worker will requeue */ | |
3342 | if (ret == -EBUSY) | |
3343 | wbc->pages_skipped++; | |
3344 | else | |
3345 | redirty_page_for_writepage(wbc, page); | |
40f76580 | 3346 | |
3d4b9496 | 3347 | update_nr_written(wbc, nr_written); |
247e743c | 3348 | unlock_page(page); |
bcf93489 | 3349 | return 1; |
247e743c CM |
3350 | } |
3351 | } | |
3352 | ||
11c8349b CM |
3353 | /* |
3354 | * we don't want to touch the inode after unlocking the page, | |
3355 | * so we update the mapping writeback index now | |
3356 | */ | |
3d4b9496 | 3357 | update_nr_written(wbc, nr_written + 1); |
771ed689 | 3358 | |
d1310b2e | 3359 | end = page_end; |
40f76580 | 3360 | if (i_size <= start) { |
e6dcd2dc CM |
3361 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3362 | tree->ops->writepage_end_io_hook(page, start, | |
3363 | page_end, NULL, 1); | |
d1310b2e CM |
3364 | goto done; |
3365 | } | |
3366 | ||
d1310b2e CM |
3367 | blocksize = inode->i_sb->s_blocksize; |
3368 | ||
3369 | while (cur <= end) { | |
40f76580 | 3370 | u64 em_end; |
6273b7f8 | 3371 | u64 offset; |
58409edd | 3372 | |
40f76580 | 3373 | if (cur >= i_size) { |
e6dcd2dc CM |
3374 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3375 | tree->ops->writepage_end_io_hook(page, cur, | |
3376 | page_end, NULL, 1); | |
d1310b2e CM |
3377 | break; |
3378 | } | |
3c98c62f | 3379 | em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, cur, |
d1310b2e | 3380 | end - cur + 1, 1); |
c704005d | 3381 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3382 | SetPageError(page); |
61391d56 | 3383 | ret = PTR_ERR_OR_ZERO(em); |
d1310b2e CM |
3384 | break; |
3385 | } | |
3386 | ||
3387 | extent_offset = cur - em->start; | |
40f76580 CM |
3388 | em_end = extent_map_end(em); |
3389 | BUG_ON(em_end <= cur); | |
d1310b2e | 3390 | BUG_ON(end < cur); |
40f76580 | 3391 | iosize = min(em_end - cur, end - cur + 1); |
fda2832f | 3392 | iosize = ALIGN(iosize, blocksize); |
6273b7f8 | 3393 | offset = em->block_start + extent_offset; |
d1310b2e CM |
3394 | bdev = em->bdev; |
3395 | block_start = em->block_start; | |
c8b97818 | 3396 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
3397 | free_extent_map(em); |
3398 | em = NULL; | |
3399 | ||
c8b97818 CM |
3400 | /* |
3401 | * compressed and inline extents are written through other | |
3402 | * paths in the FS | |
3403 | */ | |
3404 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3405 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
3406 | /* |
3407 | * end_io notification does not happen here for | |
3408 | * compressed extents | |
3409 | */ | |
3410 | if (!compressed && tree->ops && | |
3411 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
3412 | tree->ops->writepage_end_io_hook(page, cur, |
3413 | cur + iosize - 1, | |
3414 | NULL, 1); | |
c8b97818 CM |
3415 | else if (compressed) { |
3416 | /* we don't want to end_page_writeback on | |
3417 | * a compressed extent. this happens | |
3418 | * elsewhere | |
3419 | */ | |
3420 | nr++; | |
3421 | } | |
3422 | ||
3423 | cur += iosize; | |
7f3c74fb | 3424 | pg_offset += iosize; |
d1310b2e CM |
3425 | continue; |
3426 | } | |
c8b97818 | 3427 | |
58409edd DS |
3428 | set_range_writeback(tree, cur, cur + iosize - 1); |
3429 | if (!PageWriteback(page)) { | |
3430 | btrfs_err(BTRFS_I(inode)->root->fs_info, | |
3431 | "page %lu not writeback, cur %llu end %llu", | |
3432 | page->index, cur, end); | |
d1310b2e | 3433 | } |
7f3c74fb | 3434 | |
4b81ba48 | 3435 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc, |
6273b7f8 | 3436 | page, offset, iosize, pg_offset, |
c2df8bb4 | 3437 | bdev, &epd->bio, |
58409edd DS |
3438 | end_bio_extent_writepage, |
3439 | 0, 0, 0, false); | |
fe01aa65 | 3440 | if (ret) { |
58409edd | 3441 | SetPageError(page); |
fe01aa65 TK |
3442 | if (PageWriteback(page)) |
3443 | end_page_writeback(page); | |
3444 | } | |
d1310b2e | 3445 | |
d1310b2e | 3446 | cur = cur + iosize; |
7f3c74fb | 3447 | pg_offset += iosize; |
d1310b2e CM |
3448 | nr++; |
3449 | } | |
40f76580 CM |
3450 | done: |
3451 | *nr_ret = nr; | |
40f76580 CM |
3452 | return ret; |
3453 | } | |
3454 | ||
3455 | /* | |
3456 | * the writepage semantics are similar to regular writepage. extent | |
3457 | * records are inserted to lock ranges in the tree, and as dirty areas | |
3458 | * are found, they are marked writeback. Then the lock bits are removed | |
3459 | * and the end_io handler clears the writeback ranges | |
3460 | */ | |
3461 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
aab6e9ed | 3462 | struct extent_page_data *epd) |
40f76580 CM |
3463 | { |
3464 | struct inode *inode = page->mapping->host; | |
40f76580 | 3465 | u64 start = page_offset(page); |
09cbfeaf | 3466 | u64 page_end = start + PAGE_SIZE - 1; |
40f76580 CM |
3467 | int ret; |
3468 | int nr = 0; | |
3469 | size_t pg_offset = 0; | |
3470 | loff_t i_size = i_size_read(inode); | |
09cbfeaf | 3471 | unsigned long end_index = i_size >> PAGE_SHIFT; |
f1c77c55 | 3472 | unsigned int write_flags = 0; |
40f76580 CM |
3473 | unsigned long nr_written = 0; |
3474 | ||
ff40adf7 | 3475 | write_flags = wbc_to_write_flags(wbc); |
40f76580 CM |
3476 | |
3477 | trace___extent_writepage(page, inode, wbc); | |
3478 | ||
3479 | WARN_ON(!PageLocked(page)); | |
3480 | ||
3481 | ClearPageError(page); | |
3482 | ||
09cbfeaf | 3483 | pg_offset = i_size & (PAGE_SIZE - 1); |
40f76580 CM |
3484 | if (page->index > end_index || |
3485 | (page->index == end_index && !pg_offset)) { | |
09cbfeaf | 3486 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); |
40f76580 CM |
3487 | unlock_page(page); |
3488 | return 0; | |
3489 | } | |
3490 | ||
3491 | if (page->index == end_index) { | |
3492 | char *userpage; | |
3493 | ||
3494 | userpage = kmap_atomic(page); | |
3495 | memset(userpage + pg_offset, 0, | |
09cbfeaf | 3496 | PAGE_SIZE - pg_offset); |
40f76580 CM |
3497 | kunmap_atomic(userpage); |
3498 | flush_dcache_page(page); | |
3499 | } | |
3500 | ||
3501 | pg_offset = 0; | |
3502 | ||
3503 | set_page_extent_mapped(page); | |
3504 | ||
3505 | ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written); | |
3506 | if (ret == 1) | |
3507 | goto done_unlocked; | |
3508 | if (ret) | |
3509 | goto done; | |
3510 | ||
3511 | ret = __extent_writepage_io(inode, page, wbc, epd, | |
3512 | i_size, nr_written, write_flags, &nr); | |
3513 | if (ret == 1) | |
3514 | goto done_unlocked; | |
3515 | ||
d1310b2e CM |
3516 | done: |
3517 | if (nr == 0) { | |
3518 | /* make sure the mapping tag for page dirty gets cleared */ | |
3519 | set_page_writeback(page); | |
3520 | end_page_writeback(page); | |
3521 | } | |
61391d56 FM |
3522 | if (PageError(page)) { |
3523 | ret = ret < 0 ? ret : -EIO; | |
3524 | end_extent_writepage(page, ret, start, page_end); | |
3525 | } | |
d1310b2e | 3526 | unlock_page(page); |
40f76580 | 3527 | return ret; |
771ed689 | 3528 | |
11c8349b | 3529 | done_unlocked: |
d1310b2e CM |
3530 | return 0; |
3531 | } | |
3532 | ||
fd8b2b61 | 3533 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 3534 | { |
74316201 N |
3535 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
3536 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
3537 | } |
3538 | ||
0e378df1 CM |
3539 | static noinline_for_stack int |
3540 | lock_extent_buffer_for_io(struct extent_buffer *eb, | |
3541 | struct btrfs_fs_info *fs_info, | |
3542 | struct extent_page_data *epd) | |
0b32f4bb JB |
3543 | { |
3544 | unsigned long i, num_pages; | |
3545 | int flush = 0; | |
3546 | int ret = 0; | |
3547 | ||
3548 | if (!btrfs_try_tree_write_lock(eb)) { | |
3549 | flush = 1; | |
3550 | flush_write_bio(epd); | |
3551 | btrfs_tree_lock(eb); | |
3552 | } | |
3553 | ||
3554 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3555 | btrfs_tree_unlock(eb); | |
3556 | if (!epd->sync_io) | |
3557 | return 0; | |
3558 | if (!flush) { | |
3559 | flush_write_bio(epd); | |
3560 | flush = 1; | |
3561 | } | |
a098d8e8 CM |
3562 | while (1) { |
3563 | wait_on_extent_buffer_writeback(eb); | |
3564 | btrfs_tree_lock(eb); | |
3565 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3566 | break; | |
0b32f4bb | 3567 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3568 | } |
3569 | } | |
3570 | ||
51561ffe JB |
3571 | /* |
3572 | * We need to do this to prevent races in people who check if the eb is | |
3573 | * under IO since we can end up having no IO bits set for a short period | |
3574 | * of time. | |
3575 | */ | |
3576 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
3577 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3578 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 3579 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 3580 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
104b4e51 NB |
3581 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
3582 | -eb->len, | |
3583 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 3584 | ret = 1; |
51561ffe JB |
3585 | } else { |
3586 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
3587 | } |
3588 | ||
3589 | btrfs_tree_unlock(eb); | |
3590 | ||
3591 | if (!ret) | |
3592 | return ret; | |
3593 | ||
3594 | num_pages = num_extent_pages(eb->start, eb->len); | |
3595 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 3596 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3597 | |
3598 | if (!trylock_page(p)) { | |
3599 | if (!flush) { | |
3600 | flush_write_bio(epd); | |
3601 | flush = 1; | |
3602 | } | |
3603 | lock_page(p); | |
3604 | } | |
3605 | } | |
3606 | ||
3607 | return ret; | |
3608 | } | |
3609 | ||
3610 | static void end_extent_buffer_writeback(struct extent_buffer *eb) | |
3611 | { | |
3612 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
4e857c58 | 3613 | smp_mb__after_atomic(); |
0b32f4bb JB |
3614 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); |
3615 | } | |
3616 | ||
656f30db FM |
3617 | static void set_btree_ioerr(struct page *page) |
3618 | { | |
3619 | struct extent_buffer *eb = (struct extent_buffer *)page->private; | |
656f30db FM |
3620 | |
3621 | SetPageError(page); | |
3622 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) | |
3623 | return; | |
3624 | ||
3625 | /* | |
3626 | * If writeback for a btree extent that doesn't belong to a log tree | |
3627 | * failed, increment the counter transaction->eb_write_errors. | |
3628 | * We do this because while the transaction is running and before it's | |
3629 | * committing (when we call filemap_fdata[write|wait]_range against | |
3630 | * the btree inode), we might have | |
3631 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
3632 | * returns an error or an error happens during writeback, when we're | |
3633 | * committing the transaction we wouldn't know about it, since the pages | |
3634 | * can be no longer dirty nor marked anymore for writeback (if a | |
3635 | * subsequent modification to the extent buffer didn't happen before the | |
3636 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
3637 | * able to find the pages tagged with SetPageError at transaction | |
3638 | * commit time. So if this happens we must abort the transaction, | |
3639 | * otherwise we commit a super block with btree roots that point to | |
3640 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
3641 | * or the content of some node/leaf from a past generation that got | |
3642 | * cowed or deleted and is no longer valid. | |
3643 | * | |
3644 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
3645 | * not be enough - we need to distinguish between log tree extents vs | |
3646 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
3647 | * will catch and clear such errors in the mapping - and that call might | |
3648 | * be from a log sync and not from a transaction commit. Also, checking | |
3649 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
3650 | * not done and would not be reliable - the eb might have been released | |
3651 | * from memory and reading it back again means that flag would not be | |
3652 | * set (since it's a runtime flag, not persisted on disk). | |
3653 | * | |
3654 | * Using the flags below in the btree inode also makes us achieve the | |
3655 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
3656 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
3657 | * is called, the writeback for all dirty pages had already finished | |
3658 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
3659 | * filemap_fdatawait_range() would return success, as it could not know | |
3660 | * that writeback errors happened (the pages were no longer tagged for | |
3661 | * writeback). | |
3662 | */ | |
3663 | switch (eb->log_index) { | |
3664 | case -1: | |
afcdd129 | 3665 | set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags); |
656f30db FM |
3666 | break; |
3667 | case 0: | |
afcdd129 | 3668 | set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags); |
656f30db FM |
3669 | break; |
3670 | case 1: | |
afcdd129 | 3671 | set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags); |
656f30db FM |
3672 | break; |
3673 | default: | |
3674 | BUG(); /* unexpected, logic error */ | |
3675 | } | |
3676 | } | |
3677 | ||
4246a0b6 | 3678 | static void end_bio_extent_buffer_writepage(struct bio *bio) |
0b32f4bb | 3679 | { |
2c30c71b | 3680 | struct bio_vec *bvec; |
0b32f4bb | 3681 | struct extent_buffer *eb; |
2c30c71b | 3682 | int i, done; |
0b32f4bb | 3683 | |
c09abff8 | 3684 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2c30c71b | 3685 | bio_for_each_segment_all(bvec, bio, i) { |
0b32f4bb JB |
3686 | struct page *page = bvec->bv_page; |
3687 | ||
0b32f4bb JB |
3688 | eb = (struct extent_buffer *)page->private; |
3689 | BUG_ON(!eb); | |
3690 | done = atomic_dec_and_test(&eb->io_pages); | |
3691 | ||
4e4cbee9 | 3692 | if (bio->bi_status || |
4246a0b6 | 3693 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { |
0b32f4bb | 3694 | ClearPageUptodate(page); |
656f30db | 3695 | set_btree_ioerr(page); |
0b32f4bb JB |
3696 | } |
3697 | ||
3698 | end_page_writeback(page); | |
3699 | ||
3700 | if (!done) | |
3701 | continue; | |
3702 | ||
3703 | end_extent_buffer_writeback(eb); | |
2c30c71b | 3704 | } |
0b32f4bb JB |
3705 | |
3706 | bio_put(bio); | |
0b32f4bb JB |
3707 | } |
3708 | ||
0e378df1 | 3709 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb JB |
3710 | struct btrfs_fs_info *fs_info, |
3711 | struct writeback_control *wbc, | |
3712 | struct extent_page_data *epd) | |
3713 | { | |
3714 | struct block_device *bdev = fs_info->fs_devices->latest_bdev; | |
f28491e0 | 3715 | struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree; |
0b32f4bb | 3716 | u64 offset = eb->start; |
851cd173 | 3717 | u32 nritems; |
0b32f4bb | 3718 | unsigned long i, num_pages; |
851cd173 | 3719 | unsigned long start, end; |
ff40adf7 | 3720 | unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META; |
d7dbe9e7 | 3721 | int ret = 0; |
0b32f4bb | 3722 | |
656f30db | 3723 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); |
0b32f4bb JB |
3724 | num_pages = num_extent_pages(eb->start, eb->len); |
3725 | atomic_set(&eb->io_pages, num_pages); | |
de0022b9 | 3726 | |
851cd173 LB |
3727 | /* set btree blocks beyond nritems with 0 to avoid stale content. */ |
3728 | nritems = btrfs_header_nritems(eb); | |
3eb548ee | 3729 | if (btrfs_header_level(eb) > 0) { |
3eb548ee LB |
3730 | end = btrfs_node_key_ptr_offset(nritems); |
3731 | ||
b159fa28 | 3732 | memzero_extent_buffer(eb, end, eb->len - end); |
851cd173 LB |
3733 | } else { |
3734 | /* | |
3735 | * leaf: | |
3736 | * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0 | |
3737 | */ | |
3738 | start = btrfs_item_nr_offset(nritems); | |
3d9ec8c4 | 3739 | end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, eb); |
b159fa28 | 3740 | memzero_extent_buffer(eb, start, end - start); |
3eb548ee LB |
3741 | } |
3742 | ||
0b32f4bb | 3743 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3744 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3745 | |
3746 | clear_page_dirty_for_io(p); | |
3747 | set_page_writeback(p); | |
4b81ba48 | 3748 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc, |
6273b7f8 | 3749 | p, offset, PAGE_SIZE, 0, bdev, |
c2df8bb4 | 3750 | &epd->bio, |
1f7ad75b | 3751 | end_bio_extent_buffer_writepage, |
18fdc679 | 3752 | 0, 0, 0, false); |
0b32f4bb | 3753 | if (ret) { |
656f30db | 3754 | set_btree_ioerr(p); |
fe01aa65 TK |
3755 | if (PageWriteback(p)) |
3756 | end_page_writeback(p); | |
0b32f4bb JB |
3757 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
3758 | end_extent_buffer_writeback(eb); | |
3759 | ret = -EIO; | |
3760 | break; | |
3761 | } | |
09cbfeaf | 3762 | offset += PAGE_SIZE; |
3d4b9496 | 3763 | update_nr_written(wbc, 1); |
0b32f4bb JB |
3764 | unlock_page(p); |
3765 | } | |
3766 | ||
3767 | if (unlikely(ret)) { | |
3768 | for (; i < num_pages; i++) { | |
bbf65cf0 | 3769 | struct page *p = eb->pages[i]; |
81465028 | 3770 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
3771 | unlock_page(p); |
3772 | } | |
3773 | } | |
3774 | ||
3775 | return ret; | |
3776 | } | |
3777 | ||
3778 | int btree_write_cache_pages(struct address_space *mapping, | |
3779 | struct writeback_control *wbc) | |
3780 | { | |
3781 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; | |
3782 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; | |
3783 | struct extent_buffer *eb, *prev_eb = NULL; | |
3784 | struct extent_page_data epd = { | |
3785 | .bio = NULL, | |
3786 | .tree = tree, | |
3787 | .extent_locked = 0, | |
3788 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
3789 | }; | |
3790 | int ret = 0; | |
3791 | int done = 0; | |
3792 | int nr_to_write_done = 0; | |
3793 | struct pagevec pvec; | |
3794 | int nr_pages; | |
3795 | pgoff_t index; | |
3796 | pgoff_t end; /* Inclusive */ | |
3797 | int scanned = 0; | |
3798 | int tag; | |
3799 | ||
86679820 | 3800 | pagevec_init(&pvec); |
0b32f4bb JB |
3801 | if (wbc->range_cyclic) { |
3802 | index = mapping->writeback_index; /* Start from prev offset */ | |
3803 | end = -1; | |
3804 | } else { | |
09cbfeaf KS |
3805 | index = wbc->range_start >> PAGE_SHIFT; |
3806 | end = wbc->range_end >> PAGE_SHIFT; | |
0b32f4bb JB |
3807 | scanned = 1; |
3808 | } | |
3809 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3810 | tag = PAGECACHE_TAG_TOWRITE; | |
3811 | else | |
3812 | tag = PAGECACHE_TAG_DIRTY; | |
3813 | retry: | |
3814 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3815 | tag_pages_for_writeback(mapping, index, end); | |
3816 | while (!done && !nr_to_write_done && (index <= end) && | |
4006f437 | 3817 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 3818 | tag))) { |
0b32f4bb JB |
3819 | unsigned i; |
3820 | ||
3821 | scanned = 1; | |
3822 | for (i = 0; i < nr_pages; i++) { | |
3823 | struct page *page = pvec.pages[i]; | |
3824 | ||
3825 | if (!PagePrivate(page)) | |
3826 | continue; | |
3827 | ||
b5bae261 JB |
3828 | spin_lock(&mapping->private_lock); |
3829 | if (!PagePrivate(page)) { | |
3830 | spin_unlock(&mapping->private_lock); | |
3831 | continue; | |
3832 | } | |
3833 | ||
0b32f4bb | 3834 | eb = (struct extent_buffer *)page->private; |
b5bae261 JB |
3835 | |
3836 | /* | |
3837 | * Shouldn't happen and normally this would be a BUG_ON | |
3838 | * but no sense in crashing the users box for something | |
3839 | * we can survive anyway. | |
3840 | */ | |
fae7f21c | 3841 | if (WARN_ON(!eb)) { |
b5bae261 | 3842 | spin_unlock(&mapping->private_lock); |
0b32f4bb JB |
3843 | continue; |
3844 | } | |
3845 | ||
b5bae261 JB |
3846 | if (eb == prev_eb) { |
3847 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 3848 | continue; |
b5bae261 | 3849 | } |
0b32f4bb | 3850 | |
b5bae261 JB |
3851 | ret = atomic_inc_not_zero(&eb->refs); |
3852 | spin_unlock(&mapping->private_lock); | |
3853 | if (!ret) | |
0b32f4bb | 3854 | continue; |
0b32f4bb JB |
3855 | |
3856 | prev_eb = eb; | |
3857 | ret = lock_extent_buffer_for_io(eb, fs_info, &epd); | |
3858 | if (!ret) { | |
3859 | free_extent_buffer(eb); | |
3860 | continue; | |
3861 | } | |
3862 | ||
3863 | ret = write_one_eb(eb, fs_info, wbc, &epd); | |
3864 | if (ret) { | |
3865 | done = 1; | |
3866 | free_extent_buffer(eb); | |
3867 | break; | |
3868 | } | |
3869 | free_extent_buffer(eb); | |
3870 | ||
3871 | /* | |
3872 | * the filesystem may choose to bump up nr_to_write. | |
3873 | * We have to make sure to honor the new nr_to_write | |
3874 | * at any time | |
3875 | */ | |
3876 | nr_to_write_done = wbc->nr_to_write <= 0; | |
3877 | } | |
3878 | pagevec_release(&pvec); | |
3879 | cond_resched(); | |
3880 | } | |
3881 | if (!scanned && !done) { | |
3882 | /* | |
3883 | * We hit the last page and there is more work to be done: wrap | |
3884 | * back to the start of the file | |
3885 | */ | |
3886 | scanned = 1; | |
3887 | index = 0; | |
3888 | goto retry; | |
3889 | } | |
3890 | flush_write_bio(&epd); | |
3891 | return ret; | |
3892 | } | |
3893 | ||
d1310b2e | 3894 | /** |
4bef0848 | 3895 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
3896 | * @mapping: address space structure to write |
3897 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
935db853 | 3898 | * @data: data passed to __extent_writepage function |
d1310b2e CM |
3899 | * |
3900 | * If a page is already under I/O, write_cache_pages() skips it, even | |
3901 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
3902 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
3903 | * and msync() need to guarantee that all the data which was dirty at the time | |
3904 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
3905 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
3906 | * existing IO to complete. | |
3907 | */ | |
4242b64a | 3908 | static int extent_write_cache_pages(struct address_space *mapping, |
4bef0848 | 3909 | struct writeback_control *wbc, |
aab6e9ed | 3910 | struct extent_page_data *epd) |
d1310b2e | 3911 | { |
7fd1a3f7 | 3912 | struct inode *inode = mapping->host; |
d1310b2e CM |
3913 | int ret = 0; |
3914 | int done = 0; | |
f85d7d6c | 3915 | int nr_to_write_done = 0; |
d1310b2e CM |
3916 | struct pagevec pvec; |
3917 | int nr_pages; | |
3918 | pgoff_t index; | |
3919 | pgoff_t end; /* Inclusive */ | |
a9132667 LB |
3920 | pgoff_t done_index; |
3921 | int range_whole = 0; | |
d1310b2e | 3922 | int scanned = 0; |
f7aaa06b | 3923 | int tag; |
d1310b2e | 3924 | |
7fd1a3f7 JB |
3925 | /* |
3926 | * We have to hold onto the inode so that ordered extents can do their | |
3927 | * work when the IO finishes. The alternative to this is failing to add | |
3928 | * an ordered extent if the igrab() fails there and that is a huge pain | |
3929 | * to deal with, so instead just hold onto the inode throughout the | |
3930 | * writepages operation. If it fails here we are freeing up the inode | |
3931 | * anyway and we'd rather not waste our time writing out stuff that is | |
3932 | * going to be truncated anyway. | |
3933 | */ | |
3934 | if (!igrab(inode)) | |
3935 | return 0; | |
3936 | ||
86679820 | 3937 | pagevec_init(&pvec); |
d1310b2e CM |
3938 | if (wbc->range_cyclic) { |
3939 | index = mapping->writeback_index; /* Start from prev offset */ | |
3940 | end = -1; | |
3941 | } else { | |
09cbfeaf KS |
3942 | index = wbc->range_start >> PAGE_SHIFT; |
3943 | end = wbc->range_end >> PAGE_SHIFT; | |
a9132667 LB |
3944 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
3945 | range_whole = 1; | |
d1310b2e CM |
3946 | scanned = 1; |
3947 | } | |
f7aaa06b JB |
3948 | if (wbc->sync_mode == WB_SYNC_ALL) |
3949 | tag = PAGECACHE_TAG_TOWRITE; | |
3950 | else | |
3951 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 3952 | retry: |
f7aaa06b JB |
3953 | if (wbc->sync_mode == WB_SYNC_ALL) |
3954 | tag_pages_for_writeback(mapping, index, end); | |
a9132667 | 3955 | done_index = index; |
f85d7d6c | 3956 | while (!done && !nr_to_write_done && (index <= end) && |
67fd707f JK |
3957 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, |
3958 | &index, end, tag))) { | |
d1310b2e CM |
3959 | unsigned i; |
3960 | ||
3961 | scanned = 1; | |
3962 | for (i = 0; i < nr_pages; i++) { | |
3963 | struct page *page = pvec.pages[i]; | |
3964 | ||
a9132667 | 3965 | done_index = page->index; |
d1310b2e | 3966 | /* |
b93b0163 MW |
3967 | * At this point we hold neither the i_pages lock nor |
3968 | * the page lock: the page may be truncated or | |
3969 | * invalidated (changing page->mapping to NULL), | |
3970 | * or even swizzled back from swapper_space to | |
3971 | * tmpfs file mapping | |
d1310b2e | 3972 | */ |
c8f2f24b | 3973 | if (!trylock_page(page)) { |
aab6e9ed | 3974 | flush_write_bio(epd); |
c8f2f24b | 3975 | lock_page(page); |
01d658f2 | 3976 | } |
d1310b2e CM |
3977 | |
3978 | if (unlikely(page->mapping != mapping)) { | |
3979 | unlock_page(page); | |
3980 | continue; | |
3981 | } | |
3982 | ||
d2c3f4f6 | 3983 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 | 3984 | if (PageWriteback(page)) |
aab6e9ed | 3985 | flush_write_bio(epd); |
d1310b2e | 3986 | wait_on_page_writeback(page); |
d2c3f4f6 | 3987 | } |
d1310b2e CM |
3988 | |
3989 | if (PageWriteback(page) || | |
3990 | !clear_page_dirty_for_io(page)) { | |
3991 | unlock_page(page); | |
3992 | continue; | |
3993 | } | |
3994 | ||
aab6e9ed | 3995 | ret = __extent_writepage(page, wbc, epd); |
d1310b2e CM |
3996 | |
3997 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
3998 | unlock_page(page); | |
3999 | ret = 0; | |
4000 | } | |
a9132667 LB |
4001 | if (ret < 0) { |
4002 | /* | |
4003 | * done_index is set past this page, | |
4004 | * so media errors will not choke | |
4005 | * background writeout for the entire | |
4006 | * file. This has consequences for | |
4007 | * range_cyclic semantics (ie. it may | |
4008 | * not be suitable for data integrity | |
4009 | * writeout). | |
4010 | */ | |
4011 | done_index = page->index + 1; | |
4012 | done = 1; | |
4013 | break; | |
4014 | } | |
f85d7d6c CM |
4015 | |
4016 | /* | |
4017 | * the filesystem may choose to bump up nr_to_write. | |
4018 | * We have to make sure to honor the new nr_to_write | |
4019 | * at any time | |
4020 | */ | |
4021 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
4022 | } |
4023 | pagevec_release(&pvec); | |
4024 | cond_resched(); | |
4025 | } | |
894b36e3 | 4026 | if (!scanned && !done) { |
d1310b2e CM |
4027 | /* |
4028 | * We hit the last page and there is more work to be done: wrap | |
4029 | * back to the start of the file | |
4030 | */ | |
4031 | scanned = 1; | |
4032 | index = 0; | |
4033 | goto retry; | |
4034 | } | |
a9132667 LB |
4035 | |
4036 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole)) | |
4037 | mapping->writeback_index = done_index; | |
4038 | ||
7fd1a3f7 | 4039 | btrfs_add_delayed_iput(inode); |
894b36e3 | 4040 | return ret; |
d1310b2e | 4041 | } |
d1310b2e | 4042 | |
aab6e9ed | 4043 | static void flush_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 4044 | { |
d2c3f4f6 | 4045 | if (epd->bio) { |
355808c2 JM |
4046 | int ret; |
4047 | ||
18fdc679 | 4048 | ret = submit_one_bio(epd->bio, 0, 0); |
79787eaa | 4049 | BUG_ON(ret < 0); /* -ENOMEM */ |
d2c3f4f6 CM |
4050 | epd->bio = NULL; |
4051 | } | |
4052 | } | |
4053 | ||
0a9b0e53 | 4054 | int extent_write_full_page(struct page *page, struct writeback_control *wbc) |
d1310b2e CM |
4055 | { |
4056 | int ret; | |
d1310b2e CM |
4057 | struct extent_page_data epd = { |
4058 | .bio = NULL, | |
0a9b0e53 | 4059 | .tree = &BTRFS_I(page->mapping->host)->io_tree, |
771ed689 | 4060 | .extent_locked = 0, |
ffbd517d | 4061 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e | 4062 | }; |
d1310b2e | 4063 | |
d1310b2e CM |
4064 | ret = __extent_writepage(page, wbc, &epd); |
4065 | ||
e2932ee0 | 4066 | flush_write_bio(&epd); |
d1310b2e CM |
4067 | return ret; |
4068 | } | |
d1310b2e | 4069 | |
5e3ee236 | 4070 | int extent_write_locked_range(struct inode *inode, u64 start, u64 end, |
771ed689 CM |
4071 | int mode) |
4072 | { | |
4073 | int ret = 0; | |
4074 | struct address_space *mapping = inode->i_mapping; | |
5e3ee236 | 4075 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
771ed689 | 4076 | struct page *page; |
09cbfeaf KS |
4077 | unsigned long nr_pages = (end - start + PAGE_SIZE) >> |
4078 | PAGE_SHIFT; | |
771ed689 CM |
4079 | |
4080 | struct extent_page_data epd = { | |
4081 | .bio = NULL, | |
4082 | .tree = tree, | |
771ed689 | 4083 | .extent_locked = 1, |
ffbd517d | 4084 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
4085 | }; |
4086 | struct writeback_control wbc_writepages = { | |
771ed689 | 4087 | .sync_mode = mode, |
771ed689 CM |
4088 | .nr_to_write = nr_pages * 2, |
4089 | .range_start = start, | |
4090 | .range_end = end + 1, | |
4091 | }; | |
4092 | ||
d397712b | 4093 | while (start <= end) { |
09cbfeaf | 4094 | page = find_get_page(mapping, start >> PAGE_SHIFT); |
771ed689 CM |
4095 | if (clear_page_dirty_for_io(page)) |
4096 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
4097 | else { | |
4098 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
4099 | tree->ops->writepage_end_io_hook(page, start, | |
09cbfeaf | 4100 | start + PAGE_SIZE - 1, |
771ed689 CM |
4101 | NULL, 1); |
4102 | unlock_page(page); | |
4103 | } | |
09cbfeaf KS |
4104 | put_page(page); |
4105 | start += PAGE_SIZE; | |
771ed689 CM |
4106 | } |
4107 | ||
e2932ee0 | 4108 | flush_write_bio(&epd); |
771ed689 CM |
4109 | return ret; |
4110 | } | |
d1310b2e CM |
4111 | |
4112 | int extent_writepages(struct extent_io_tree *tree, | |
4113 | struct address_space *mapping, | |
d1310b2e CM |
4114 | struct writeback_control *wbc) |
4115 | { | |
4116 | int ret = 0; | |
4117 | struct extent_page_data epd = { | |
4118 | .bio = NULL, | |
4119 | .tree = tree, | |
771ed689 | 4120 | .extent_locked = 0, |
ffbd517d | 4121 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
4122 | }; |
4123 | ||
935db853 | 4124 | ret = extent_write_cache_pages(mapping, wbc, &epd); |
e2932ee0 | 4125 | flush_write_bio(&epd); |
d1310b2e CM |
4126 | return ret; |
4127 | } | |
d1310b2e CM |
4128 | |
4129 | int extent_readpages(struct extent_io_tree *tree, | |
4130 | struct address_space *mapping, | |
0932584b | 4131 | struct list_head *pages, unsigned nr_pages) |
d1310b2e CM |
4132 | { |
4133 | struct bio *bio = NULL; | |
4134 | unsigned page_idx; | |
c8b97818 | 4135 | unsigned long bio_flags = 0; |
67c9684f LB |
4136 | struct page *pagepool[16]; |
4137 | struct page *page; | |
125bac01 | 4138 | struct extent_map *em_cached = NULL; |
67c9684f | 4139 | int nr = 0; |
808f80b4 | 4140 | u64 prev_em_start = (u64)-1; |
d1310b2e | 4141 | |
d1310b2e | 4142 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
67c9684f | 4143 | page = list_entry(pages->prev, struct page, lru); |
d1310b2e CM |
4144 | |
4145 | prefetchw(&page->flags); | |
4146 | list_del(&page->lru); | |
67c9684f | 4147 | if (add_to_page_cache_lru(page, mapping, |
8a5c743e MH |
4148 | page->index, |
4149 | readahead_gfp_mask(mapping))) { | |
09cbfeaf | 4150 | put_page(page); |
67c9684f | 4151 | continue; |
d1310b2e | 4152 | } |
67c9684f LB |
4153 | |
4154 | pagepool[nr++] = page; | |
4155 | if (nr < ARRAY_SIZE(pagepool)) | |
4156 | continue; | |
e4d17ef5 DS |
4157 | __extent_readpages(tree, pagepool, nr, &em_cached, &bio, |
4158 | &bio_flags, &prev_em_start); | |
67c9684f | 4159 | nr = 0; |
d1310b2e | 4160 | } |
9974090b | 4161 | if (nr) |
e4d17ef5 DS |
4162 | __extent_readpages(tree, pagepool, nr, &em_cached, &bio, |
4163 | &bio_flags, &prev_em_start); | |
67c9684f | 4164 | |
125bac01 MX |
4165 | if (em_cached) |
4166 | free_extent_map(em_cached); | |
4167 | ||
d1310b2e CM |
4168 | BUG_ON(!list_empty(pages)); |
4169 | if (bio) | |
1f7ad75b | 4170 | return submit_one_bio(bio, 0, bio_flags); |
d1310b2e CM |
4171 | return 0; |
4172 | } | |
d1310b2e CM |
4173 | |
4174 | /* | |
4175 | * basic invalidatepage code, this waits on any locked or writeback | |
4176 | * ranges corresponding to the page, and then deletes any extent state | |
4177 | * records from the tree | |
4178 | */ | |
4179 | int extent_invalidatepage(struct extent_io_tree *tree, | |
4180 | struct page *page, unsigned long offset) | |
4181 | { | |
2ac55d41 | 4182 | struct extent_state *cached_state = NULL; |
4eee4fa4 | 4183 | u64 start = page_offset(page); |
09cbfeaf | 4184 | u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
4185 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; |
4186 | ||
fda2832f | 4187 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
4188 | if (start > end) |
4189 | return 0; | |
4190 | ||
ff13db41 | 4191 | lock_extent_bits(tree, start, end, &cached_state); |
1edbb734 | 4192 | wait_on_page_writeback(page); |
d1310b2e | 4193 | clear_extent_bit(tree, start, end, |
32c00aff JB |
4194 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
4195 | EXTENT_DO_ACCOUNTING, | |
ae0f1625 | 4196 | 1, 1, &cached_state); |
d1310b2e CM |
4197 | return 0; |
4198 | } | |
d1310b2e | 4199 | |
7b13b7b1 CM |
4200 | /* |
4201 | * a helper for releasepage, this tests for areas of the page that | |
4202 | * are locked or under IO and drops the related state bits if it is safe | |
4203 | * to drop the page. | |
4204 | */ | |
48a3b636 ES |
4205 | static int try_release_extent_state(struct extent_map_tree *map, |
4206 | struct extent_io_tree *tree, | |
4207 | struct page *page, gfp_t mask) | |
7b13b7b1 | 4208 | { |
4eee4fa4 | 4209 | u64 start = page_offset(page); |
09cbfeaf | 4210 | u64 end = start + PAGE_SIZE - 1; |
7b13b7b1 CM |
4211 | int ret = 1; |
4212 | ||
211f90e6 | 4213 | if (test_range_bit(tree, start, end, |
8b62b72b | 4214 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
4215 | ret = 0; |
4216 | else { | |
11ef160f CM |
4217 | /* |
4218 | * at this point we can safely clear everything except the | |
4219 | * locked bit and the nodatasum bit | |
4220 | */ | |
66b0c887 | 4221 | ret = __clear_extent_bit(tree, start, end, |
11ef160f | 4222 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
66b0c887 | 4223 | 0, 0, NULL, mask, NULL); |
e3f24cc5 CM |
4224 | |
4225 | /* if clear_extent_bit failed for enomem reasons, | |
4226 | * we can't allow the release to continue. | |
4227 | */ | |
4228 | if (ret < 0) | |
4229 | ret = 0; | |
4230 | else | |
4231 | ret = 1; | |
7b13b7b1 CM |
4232 | } |
4233 | return ret; | |
4234 | } | |
7b13b7b1 | 4235 | |
d1310b2e CM |
4236 | /* |
4237 | * a helper for releasepage. As long as there are no locked extents | |
4238 | * in the range corresponding to the page, both state records and extent | |
4239 | * map records are removed | |
4240 | */ | |
4241 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
4242 | struct extent_io_tree *tree, struct page *page, |
4243 | gfp_t mask) | |
d1310b2e CM |
4244 | { |
4245 | struct extent_map *em; | |
4eee4fa4 | 4246 | u64 start = page_offset(page); |
09cbfeaf | 4247 | u64 end = start + PAGE_SIZE - 1; |
7b13b7b1 | 4248 | |
d0164adc | 4249 | if (gfpflags_allow_blocking(mask) && |
ee22184b | 4250 | page->mapping->host->i_size > SZ_16M) { |
39b5637f | 4251 | u64 len; |
70dec807 | 4252 | while (start <= end) { |
39b5637f | 4253 | len = end - start + 1; |
890871be | 4254 | write_lock(&map->lock); |
39b5637f | 4255 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 4256 | if (!em) { |
890871be | 4257 | write_unlock(&map->lock); |
70dec807 CM |
4258 | break; |
4259 | } | |
7f3c74fb CM |
4260 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
4261 | em->start != start) { | |
890871be | 4262 | write_unlock(&map->lock); |
70dec807 CM |
4263 | free_extent_map(em); |
4264 | break; | |
4265 | } | |
4266 | if (!test_range_bit(tree, em->start, | |
4267 | extent_map_end(em) - 1, | |
8b62b72b | 4268 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 4269 | 0, NULL)) { |
70dec807 CM |
4270 | remove_extent_mapping(map, em); |
4271 | /* once for the rb tree */ | |
4272 | free_extent_map(em); | |
4273 | } | |
4274 | start = extent_map_end(em); | |
890871be | 4275 | write_unlock(&map->lock); |
70dec807 CM |
4276 | |
4277 | /* once for us */ | |
d1310b2e CM |
4278 | free_extent_map(em); |
4279 | } | |
d1310b2e | 4280 | } |
7b13b7b1 | 4281 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 4282 | } |
d1310b2e | 4283 | |
ec29ed5b CM |
4284 | /* |
4285 | * helper function for fiemap, which doesn't want to see any holes. | |
4286 | * This maps until we find something past 'last' | |
4287 | */ | |
4288 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
e3350e16 | 4289 | u64 offset, u64 last) |
ec29ed5b | 4290 | { |
da17066c | 4291 | u64 sectorsize = btrfs_inode_sectorsize(inode); |
ec29ed5b CM |
4292 | struct extent_map *em; |
4293 | u64 len; | |
4294 | ||
4295 | if (offset >= last) | |
4296 | return NULL; | |
4297 | ||
67871254 | 4298 | while (1) { |
ec29ed5b CM |
4299 | len = last - offset; |
4300 | if (len == 0) | |
4301 | break; | |
fda2832f | 4302 | len = ALIGN(len, sectorsize); |
e3350e16 DS |
4303 | em = btrfs_get_extent_fiemap(BTRFS_I(inode), NULL, 0, offset, |
4304 | len, 0); | |
c704005d | 4305 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
4306 | return em; |
4307 | ||
4308 | /* if this isn't a hole return it */ | |
4a2d25cd | 4309 | if (em->block_start != EXTENT_MAP_HOLE) |
ec29ed5b | 4310 | return em; |
ec29ed5b CM |
4311 | |
4312 | /* this is a hole, advance to the next extent */ | |
4313 | offset = extent_map_end(em); | |
4314 | free_extent_map(em); | |
4315 | if (offset >= last) | |
4316 | break; | |
4317 | } | |
4318 | return NULL; | |
4319 | } | |
4320 | ||
4751832d QW |
4321 | /* |
4322 | * To cache previous fiemap extent | |
4323 | * | |
4324 | * Will be used for merging fiemap extent | |
4325 | */ | |
4326 | struct fiemap_cache { | |
4327 | u64 offset; | |
4328 | u64 phys; | |
4329 | u64 len; | |
4330 | u32 flags; | |
4331 | bool cached; | |
4332 | }; | |
4333 | ||
4334 | /* | |
4335 | * Helper to submit fiemap extent. | |
4336 | * | |
4337 | * Will try to merge current fiemap extent specified by @offset, @phys, | |
4338 | * @len and @flags with cached one. | |
4339 | * And only when we fails to merge, cached one will be submitted as | |
4340 | * fiemap extent. | |
4341 | * | |
4342 | * Return value is the same as fiemap_fill_next_extent(). | |
4343 | */ | |
4344 | static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, | |
4345 | struct fiemap_cache *cache, | |
4346 | u64 offset, u64 phys, u64 len, u32 flags) | |
4347 | { | |
4348 | int ret = 0; | |
4349 | ||
4350 | if (!cache->cached) | |
4351 | goto assign; | |
4352 | ||
4353 | /* | |
4354 | * Sanity check, extent_fiemap() should have ensured that new | |
4355 | * fiemap extent won't overlap with cahced one. | |
4356 | * Not recoverable. | |
4357 | * | |
4358 | * NOTE: Physical address can overlap, due to compression | |
4359 | */ | |
4360 | if (cache->offset + cache->len > offset) { | |
4361 | WARN_ON(1); | |
4362 | return -EINVAL; | |
4363 | } | |
4364 | ||
4365 | /* | |
4366 | * Only merges fiemap extents if | |
4367 | * 1) Their logical addresses are continuous | |
4368 | * | |
4369 | * 2) Their physical addresses are continuous | |
4370 | * So truly compressed (physical size smaller than logical size) | |
4371 | * extents won't get merged with each other | |
4372 | * | |
4373 | * 3) Share same flags except FIEMAP_EXTENT_LAST | |
4374 | * So regular extent won't get merged with prealloc extent | |
4375 | */ | |
4376 | if (cache->offset + cache->len == offset && | |
4377 | cache->phys + cache->len == phys && | |
4378 | (cache->flags & ~FIEMAP_EXTENT_LAST) == | |
4379 | (flags & ~FIEMAP_EXTENT_LAST)) { | |
4380 | cache->len += len; | |
4381 | cache->flags |= flags; | |
4382 | goto try_submit_last; | |
4383 | } | |
4384 | ||
4385 | /* Not mergeable, need to submit cached one */ | |
4386 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, | |
4387 | cache->len, cache->flags); | |
4388 | cache->cached = false; | |
4389 | if (ret) | |
4390 | return ret; | |
4391 | assign: | |
4392 | cache->cached = true; | |
4393 | cache->offset = offset; | |
4394 | cache->phys = phys; | |
4395 | cache->len = len; | |
4396 | cache->flags = flags; | |
4397 | try_submit_last: | |
4398 | if (cache->flags & FIEMAP_EXTENT_LAST) { | |
4399 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, | |
4400 | cache->phys, cache->len, cache->flags); | |
4401 | cache->cached = false; | |
4402 | } | |
4403 | return ret; | |
4404 | } | |
4405 | ||
4406 | /* | |
848c23b7 | 4407 | * Emit last fiemap cache |
4751832d | 4408 | * |
848c23b7 QW |
4409 | * The last fiemap cache may still be cached in the following case: |
4410 | * 0 4k 8k | |
4411 | * |<- Fiemap range ->| | |
4412 | * |<------------ First extent ----------->| | |
4413 | * | |
4414 | * In this case, the first extent range will be cached but not emitted. | |
4415 | * So we must emit it before ending extent_fiemap(). | |
4751832d | 4416 | */ |
848c23b7 QW |
4417 | static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info, |
4418 | struct fiemap_extent_info *fieinfo, | |
4419 | struct fiemap_cache *cache) | |
4751832d QW |
4420 | { |
4421 | int ret; | |
4422 | ||
4423 | if (!cache->cached) | |
4424 | return 0; | |
4425 | ||
4751832d QW |
4426 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, |
4427 | cache->len, cache->flags); | |
4428 | cache->cached = false; | |
4429 | if (ret > 0) | |
4430 | ret = 0; | |
4431 | return ret; | |
4432 | } | |
4433 | ||
1506fcc8 | 4434 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
2135fb9b | 4435 | __u64 start, __u64 len) |
1506fcc8 | 4436 | { |
975f84fe | 4437 | int ret = 0; |
1506fcc8 YS |
4438 | u64 off = start; |
4439 | u64 max = start + len; | |
4440 | u32 flags = 0; | |
975f84fe JB |
4441 | u32 found_type; |
4442 | u64 last; | |
ec29ed5b | 4443 | u64 last_for_get_extent = 0; |
1506fcc8 | 4444 | u64 disko = 0; |
ec29ed5b | 4445 | u64 isize = i_size_read(inode); |
975f84fe | 4446 | struct btrfs_key found_key; |
1506fcc8 | 4447 | struct extent_map *em = NULL; |
2ac55d41 | 4448 | struct extent_state *cached_state = NULL; |
975f84fe | 4449 | struct btrfs_path *path; |
dc046b10 | 4450 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4751832d | 4451 | struct fiemap_cache cache = { 0 }; |
1506fcc8 | 4452 | int end = 0; |
ec29ed5b CM |
4453 | u64 em_start = 0; |
4454 | u64 em_len = 0; | |
4455 | u64 em_end = 0; | |
1506fcc8 YS |
4456 | |
4457 | if (len == 0) | |
4458 | return -EINVAL; | |
4459 | ||
975f84fe JB |
4460 | path = btrfs_alloc_path(); |
4461 | if (!path) | |
4462 | return -ENOMEM; | |
4463 | path->leave_spinning = 1; | |
4464 | ||
da17066c JM |
4465 | start = round_down(start, btrfs_inode_sectorsize(inode)); |
4466 | len = round_up(max, btrfs_inode_sectorsize(inode)) - start; | |
4d479cf0 | 4467 | |
ec29ed5b CM |
4468 | /* |
4469 | * lookup the last file extent. We're not using i_size here | |
4470 | * because there might be preallocation past i_size | |
4471 | */ | |
f85b7379 DS |
4472 | ret = btrfs_lookup_file_extent(NULL, root, path, |
4473 | btrfs_ino(BTRFS_I(inode)), -1, 0); | |
975f84fe JB |
4474 | if (ret < 0) { |
4475 | btrfs_free_path(path); | |
4476 | return ret; | |
2d324f59 LB |
4477 | } else { |
4478 | WARN_ON(!ret); | |
4479 | if (ret == 1) | |
4480 | ret = 0; | |
975f84fe | 4481 | } |
2d324f59 | 4482 | |
975f84fe | 4483 | path->slots[0]--; |
975f84fe | 4484 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
962a298f | 4485 | found_type = found_key.type; |
975f84fe | 4486 | |
ec29ed5b | 4487 | /* No extents, but there might be delalloc bits */ |
4a0cc7ca | 4488 | if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) || |
975f84fe | 4489 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
4490 | /* have to trust i_size as the end */ |
4491 | last = (u64)-1; | |
4492 | last_for_get_extent = isize; | |
4493 | } else { | |
4494 | /* | |
4495 | * remember the start of the last extent. There are a | |
4496 | * bunch of different factors that go into the length of the | |
4497 | * extent, so its much less complex to remember where it started | |
4498 | */ | |
4499 | last = found_key.offset; | |
4500 | last_for_get_extent = last + 1; | |
975f84fe | 4501 | } |
fe09e16c | 4502 | btrfs_release_path(path); |
975f84fe | 4503 | |
ec29ed5b CM |
4504 | /* |
4505 | * we might have some extents allocated but more delalloc past those | |
4506 | * extents. so, we trust isize unless the start of the last extent is | |
4507 | * beyond isize | |
4508 | */ | |
4509 | if (last < isize) { | |
4510 | last = (u64)-1; | |
4511 | last_for_get_extent = isize; | |
4512 | } | |
4513 | ||
ff13db41 | 4514 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
d0082371 | 4515 | &cached_state); |
ec29ed5b | 4516 | |
e3350e16 | 4517 | em = get_extent_skip_holes(inode, start, last_for_get_extent); |
1506fcc8 YS |
4518 | if (!em) |
4519 | goto out; | |
4520 | if (IS_ERR(em)) { | |
4521 | ret = PTR_ERR(em); | |
4522 | goto out; | |
4523 | } | |
975f84fe | 4524 | |
1506fcc8 | 4525 | while (!end) { |
b76bb701 | 4526 | u64 offset_in_extent = 0; |
ea8efc74 CM |
4527 | |
4528 | /* break if the extent we found is outside the range */ | |
4529 | if (em->start >= max || extent_map_end(em) < off) | |
4530 | break; | |
4531 | ||
4532 | /* | |
4533 | * get_extent may return an extent that starts before our | |
4534 | * requested range. We have to make sure the ranges | |
4535 | * we return to fiemap always move forward and don't | |
4536 | * overlap, so adjust the offsets here | |
4537 | */ | |
4538 | em_start = max(em->start, off); | |
1506fcc8 | 4539 | |
ea8efc74 CM |
4540 | /* |
4541 | * record the offset from the start of the extent | |
b76bb701 JB |
4542 | * for adjusting the disk offset below. Only do this if the |
4543 | * extent isn't compressed since our in ram offset may be past | |
4544 | * what we have actually allocated on disk. | |
ea8efc74 | 4545 | */ |
b76bb701 JB |
4546 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
4547 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 4548 | em_end = extent_map_end(em); |
ea8efc74 | 4549 | em_len = em_end - em_start; |
1506fcc8 YS |
4550 | disko = 0; |
4551 | flags = 0; | |
4552 | ||
ea8efc74 CM |
4553 | /* |
4554 | * bump off for our next call to get_extent | |
4555 | */ | |
4556 | off = extent_map_end(em); | |
4557 | if (off >= max) | |
4558 | end = 1; | |
4559 | ||
93dbfad7 | 4560 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
4561 | end = 1; |
4562 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 4563 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
4564 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
4565 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 4566 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
4567 | flags |= (FIEMAP_EXTENT_DELALLOC | |
4568 | FIEMAP_EXTENT_UNKNOWN); | |
dc046b10 JB |
4569 | } else if (fieinfo->fi_extents_max) { |
4570 | u64 bytenr = em->block_start - | |
4571 | (em->start - em->orig_start); | |
fe09e16c | 4572 | |
ea8efc74 | 4573 | disko = em->block_start + offset_in_extent; |
fe09e16c LB |
4574 | |
4575 | /* | |
4576 | * As btrfs supports shared space, this information | |
4577 | * can be exported to userspace tools via | |
dc046b10 JB |
4578 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 |
4579 | * then we're just getting a count and we can skip the | |
4580 | * lookup stuff. | |
fe09e16c | 4581 | */ |
bb739cf0 EN |
4582 | ret = btrfs_check_shared(root, |
4583 | btrfs_ino(BTRFS_I(inode)), | |
4584 | bytenr); | |
dc046b10 | 4585 | if (ret < 0) |
fe09e16c | 4586 | goto out_free; |
dc046b10 | 4587 | if (ret) |
fe09e16c | 4588 | flags |= FIEMAP_EXTENT_SHARED; |
dc046b10 | 4589 | ret = 0; |
1506fcc8 YS |
4590 | } |
4591 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
4592 | flags |= FIEMAP_EXTENT_ENCODED; | |
0d2b2372 JB |
4593 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
4594 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
1506fcc8 | 4595 | |
1506fcc8 YS |
4596 | free_extent_map(em); |
4597 | em = NULL; | |
ec29ed5b CM |
4598 | if ((em_start >= last) || em_len == (u64)-1 || |
4599 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
4600 | flags |= FIEMAP_EXTENT_LAST; |
4601 | end = 1; | |
4602 | } | |
4603 | ||
ec29ed5b | 4604 | /* now scan forward to see if this is really the last extent. */ |
e3350e16 | 4605 | em = get_extent_skip_holes(inode, off, last_for_get_extent); |
ec29ed5b CM |
4606 | if (IS_ERR(em)) { |
4607 | ret = PTR_ERR(em); | |
4608 | goto out; | |
4609 | } | |
4610 | if (!em) { | |
975f84fe JB |
4611 | flags |= FIEMAP_EXTENT_LAST; |
4612 | end = 1; | |
4613 | } | |
4751832d QW |
4614 | ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko, |
4615 | em_len, flags); | |
26e726af CS |
4616 | if (ret) { |
4617 | if (ret == 1) | |
4618 | ret = 0; | |
ec29ed5b | 4619 | goto out_free; |
26e726af | 4620 | } |
1506fcc8 YS |
4621 | } |
4622 | out_free: | |
4751832d | 4623 | if (!ret) |
848c23b7 | 4624 | ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache); |
1506fcc8 YS |
4625 | free_extent_map(em); |
4626 | out: | |
fe09e16c | 4627 | btrfs_free_path(path); |
a52f4cd2 | 4628 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
e43bbe5e | 4629 | &cached_state); |
1506fcc8 YS |
4630 | return ret; |
4631 | } | |
4632 | ||
727011e0 CM |
4633 | static void __free_extent_buffer(struct extent_buffer *eb) |
4634 | { | |
6d49ba1b | 4635 | btrfs_leak_debug_del(&eb->leak_list); |
727011e0 CM |
4636 | kmem_cache_free(extent_buffer_cache, eb); |
4637 | } | |
4638 | ||
a26e8c9f | 4639 | int extent_buffer_under_io(struct extent_buffer *eb) |
db7f3436 JB |
4640 | { |
4641 | return (atomic_read(&eb->io_pages) || | |
4642 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4643 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4644 | } | |
4645 | ||
4646 | /* | |
4647 | * Helper for releasing extent buffer page. | |
4648 | */ | |
a50924e3 | 4649 | static void btrfs_release_extent_buffer_page(struct extent_buffer *eb) |
db7f3436 JB |
4650 | { |
4651 | unsigned long index; | |
db7f3436 JB |
4652 | struct page *page; |
4653 | int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4654 | ||
4655 | BUG_ON(extent_buffer_under_io(eb)); | |
4656 | ||
a50924e3 DS |
4657 | index = num_extent_pages(eb->start, eb->len); |
4658 | if (index == 0) | |
db7f3436 JB |
4659 | return; |
4660 | ||
4661 | do { | |
4662 | index--; | |
fb85fc9a | 4663 | page = eb->pages[index]; |
5d2361db FL |
4664 | if (!page) |
4665 | continue; | |
4666 | if (mapped) | |
db7f3436 | 4667 | spin_lock(&page->mapping->private_lock); |
5d2361db FL |
4668 | /* |
4669 | * We do this since we'll remove the pages after we've | |
4670 | * removed the eb from the radix tree, so we could race | |
4671 | * and have this page now attached to the new eb. So | |
4672 | * only clear page_private if it's still connected to | |
4673 | * this eb. | |
4674 | */ | |
4675 | if (PagePrivate(page) && | |
4676 | page->private == (unsigned long)eb) { | |
4677 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4678 | BUG_ON(PageDirty(page)); | |
4679 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 4680 | /* |
5d2361db FL |
4681 | * We need to make sure we haven't be attached |
4682 | * to a new eb. | |
db7f3436 | 4683 | */ |
5d2361db FL |
4684 | ClearPagePrivate(page); |
4685 | set_page_private(page, 0); | |
4686 | /* One for the page private */ | |
09cbfeaf | 4687 | put_page(page); |
db7f3436 | 4688 | } |
5d2361db FL |
4689 | |
4690 | if (mapped) | |
4691 | spin_unlock(&page->mapping->private_lock); | |
4692 | ||
01327610 | 4693 | /* One for when we allocated the page */ |
09cbfeaf | 4694 | put_page(page); |
a50924e3 | 4695 | } while (index != 0); |
db7f3436 JB |
4696 | } |
4697 | ||
4698 | /* | |
4699 | * Helper for releasing the extent buffer. | |
4700 | */ | |
4701 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4702 | { | |
a50924e3 | 4703 | btrfs_release_extent_buffer_page(eb); |
db7f3436 JB |
4704 | __free_extent_buffer(eb); |
4705 | } | |
4706 | ||
f28491e0 JB |
4707 | static struct extent_buffer * |
4708 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 4709 | unsigned long len) |
d1310b2e CM |
4710 | { |
4711 | struct extent_buffer *eb = NULL; | |
4712 | ||
d1b5c567 | 4713 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
4714 | eb->start = start; |
4715 | eb->len = len; | |
f28491e0 | 4716 | eb->fs_info = fs_info; |
815a51c7 | 4717 | eb->bflags = 0; |
bd681513 CM |
4718 | rwlock_init(&eb->lock); |
4719 | atomic_set(&eb->write_locks, 0); | |
4720 | atomic_set(&eb->read_locks, 0); | |
4721 | atomic_set(&eb->blocking_readers, 0); | |
4722 | atomic_set(&eb->blocking_writers, 0); | |
4723 | atomic_set(&eb->spinning_readers, 0); | |
4724 | atomic_set(&eb->spinning_writers, 0); | |
5b25f70f | 4725 | eb->lock_nested = 0; |
bd681513 CM |
4726 | init_waitqueue_head(&eb->write_lock_wq); |
4727 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 4728 | |
6d49ba1b ES |
4729 | btrfs_leak_debug_add(&eb->leak_list, &buffers); |
4730 | ||
3083ee2e | 4731 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 4732 | atomic_set(&eb->refs, 1); |
0b32f4bb | 4733 | atomic_set(&eb->io_pages, 0); |
727011e0 | 4734 | |
b8dae313 DS |
4735 | /* |
4736 | * Sanity checks, currently the maximum is 64k covered by 16x 4k pages | |
4737 | */ | |
4738 | BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE | |
4739 | > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
4740 | BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
d1310b2e CM |
4741 | |
4742 | return eb; | |
4743 | } | |
4744 | ||
815a51c7 JS |
4745 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) |
4746 | { | |
4747 | unsigned long i; | |
4748 | struct page *p; | |
4749 | struct extent_buffer *new; | |
4750 | unsigned long num_pages = num_extent_pages(src->start, src->len); | |
4751 | ||
3f556f78 | 4752 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
4753 | if (new == NULL) |
4754 | return NULL; | |
4755 | ||
4756 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4757 | p = alloc_page(GFP_NOFS); |
db7f3436 JB |
4758 | if (!p) { |
4759 | btrfs_release_extent_buffer(new); | |
4760 | return NULL; | |
4761 | } | |
815a51c7 JS |
4762 | attach_extent_buffer_page(new, p); |
4763 | WARN_ON(PageDirty(p)); | |
4764 | SetPageUptodate(p); | |
4765 | new->pages[i] = p; | |
fba1acf9 | 4766 | copy_page(page_address(p), page_address(src->pages[i])); |
815a51c7 JS |
4767 | } |
4768 | ||
815a51c7 JS |
4769 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); |
4770 | set_bit(EXTENT_BUFFER_DUMMY, &new->bflags); | |
4771 | ||
4772 | return new; | |
4773 | } | |
4774 | ||
0f331229 OS |
4775 | struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
4776 | u64 start, unsigned long len) | |
815a51c7 JS |
4777 | { |
4778 | struct extent_buffer *eb; | |
3f556f78 | 4779 | unsigned long num_pages; |
815a51c7 JS |
4780 | unsigned long i; |
4781 | ||
0f331229 | 4782 | num_pages = num_extent_pages(start, len); |
3f556f78 DS |
4783 | |
4784 | eb = __alloc_extent_buffer(fs_info, start, len); | |
815a51c7 JS |
4785 | if (!eb) |
4786 | return NULL; | |
4787 | ||
4788 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4789 | eb->pages[i] = alloc_page(GFP_NOFS); |
815a51c7 JS |
4790 | if (!eb->pages[i]) |
4791 | goto err; | |
4792 | } | |
4793 | set_extent_buffer_uptodate(eb); | |
4794 | btrfs_set_header_nritems(eb, 0); | |
4795 | set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4796 | ||
4797 | return eb; | |
4798 | err: | |
84167d19 SB |
4799 | for (; i > 0; i--) |
4800 | __free_page(eb->pages[i - 1]); | |
815a51c7 JS |
4801 | __free_extent_buffer(eb); |
4802 | return NULL; | |
4803 | } | |
4804 | ||
0f331229 | 4805 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
da17066c | 4806 | u64 start) |
0f331229 | 4807 | { |
da17066c | 4808 | return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize); |
0f331229 OS |
4809 | } |
4810 | ||
0b32f4bb JB |
4811 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
4812 | { | |
242e18c7 | 4813 | int refs; |
0b32f4bb JB |
4814 | /* the ref bit is tricky. We have to make sure it is set |
4815 | * if we have the buffer dirty. Otherwise the | |
4816 | * code to free a buffer can end up dropping a dirty | |
4817 | * page | |
4818 | * | |
4819 | * Once the ref bit is set, it won't go away while the | |
4820 | * buffer is dirty or in writeback, and it also won't | |
4821 | * go away while we have the reference count on the | |
4822 | * eb bumped. | |
4823 | * | |
4824 | * We can't just set the ref bit without bumping the | |
4825 | * ref on the eb because free_extent_buffer might | |
4826 | * see the ref bit and try to clear it. If this happens | |
4827 | * free_extent_buffer might end up dropping our original | |
4828 | * ref by mistake and freeing the page before we are able | |
4829 | * to add one more ref. | |
4830 | * | |
4831 | * So bump the ref count first, then set the bit. If someone | |
4832 | * beat us to it, drop the ref we added. | |
4833 | */ | |
242e18c7 CM |
4834 | refs = atomic_read(&eb->refs); |
4835 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
4836 | return; | |
4837 | ||
594831c4 JB |
4838 | spin_lock(&eb->refs_lock); |
4839 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 4840 | atomic_inc(&eb->refs); |
594831c4 | 4841 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
4842 | } |
4843 | ||
2457aec6 MG |
4844 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
4845 | struct page *accessed) | |
5df4235e JB |
4846 | { |
4847 | unsigned long num_pages, i; | |
4848 | ||
0b32f4bb JB |
4849 | check_buffer_tree_ref(eb); |
4850 | ||
5df4235e JB |
4851 | num_pages = num_extent_pages(eb->start, eb->len); |
4852 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a DS |
4853 | struct page *p = eb->pages[i]; |
4854 | ||
2457aec6 MG |
4855 | if (p != accessed) |
4856 | mark_page_accessed(p); | |
5df4235e JB |
4857 | } |
4858 | } | |
4859 | ||
f28491e0 JB |
4860 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
4861 | u64 start) | |
452c75c3 CS |
4862 | { |
4863 | struct extent_buffer *eb; | |
4864 | ||
4865 | rcu_read_lock(); | |
f28491e0 | 4866 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
09cbfeaf | 4867 | start >> PAGE_SHIFT); |
452c75c3 CS |
4868 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
4869 | rcu_read_unlock(); | |
062c19e9 FM |
4870 | /* |
4871 | * Lock our eb's refs_lock to avoid races with | |
4872 | * free_extent_buffer. When we get our eb it might be flagged | |
4873 | * with EXTENT_BUFFER_STALE and another task running | |
4874 | * free_extent_buffer might have seen that flag set, | |
4875 | * eb->refs == 2, that the buffer isn't under IO (dirty and | |
4876 | * writeback flags not set) and it's still in the tree (flag | |
4877 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process | |
4878 | * of decrementing the extent buffer's reference count twice. | |
4879 | * So here we could race and increment the eb's reference count, | |
4880 | * clear its stale flag, mark it as dirty and drop our reference | |
4881 | * before the other task finishes executing free_extent_buffer, | |
4882 | * which would later result in an attempt to free an extent | |
4883 | * buffer that is dirty. | |
4884 | */ | |
4885 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
4886 | spin_lock(&eb->refs_lock); | |
4887 | spin_unlock(&eb->refs_lock); | |
4888 | } | |
2457aec6 | 4889 | mark_extent_buffer_accessed(eb, NULL); |
452c75c3 CS |
4890 | return eb; |
4891 | } | |
4892 | rcu_read_unlock(); | |
4893 | ||
4894 | return NULL; | |
4895 | } | |
4896 | ||
faa2dbf0 JB |
4897 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
4898 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
da17066c | 4899 | u64 start) |
faa2dbf0 JB |
4900 | { |
4901 | struct extent_buffer *eb, *exists = NULL; | |
4902 | int ret; | |
4903 | ||
4904 | eb = find_extent_buffer(fs_info, start); | |
4905 | if (eb) | |
4906 | return eb; | |
da17066c | 4907 | eb = alloc_dummy_extent_buffer(fs_info, start); |
faa2dbf0 JB |
4908 | if (!eb) |
4909 | return NULL; | |
4910 | eb->fs_info = fs_info; | |
4911 | again: | |
e1860a77 | 4912 | ret = radix_tree_preload(GFP_NOFS); |
faa2dbf0 JB |
4913 | if (ret) |
4914 | goto free_eb; | |
4915 | spin_lock(&fs_info->buffer_lock); | |
4916 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
09cbfeaf | 4917 | start >> PAGE_SHIFT, eb); |
faa2dbf0 JB |
4918 | spin_unlock(&fs_info->buffer_lock); |
4919 | radix_tree_preload_end(); | |
4920 | if (ret == -EEXIST) { | |
4921 | exists = find_extent_buffer(fs_info, start); | |
4922 | if (exists) | |
4923 | goto free_eb; | |
4924 | else | |
4925 | goto again; | |
4926 | } | |
4927 | check_buffer_tree_ref(eb); | |
4928 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
4929 | ||
4930 | /* | |
4931 | * We will free dummy extent buffer's if they come into | |
4932 | * free_extent_buffer with a ref count of 2, but if we are using this we | |
4933 | * want the buffers to stay in memory until we're done with them, so | |
4934 | * bump the ref count again. | |
4935 | */ | |
4936 | atomic_inc(&eb->refs); | |
4937 | return eb; | |
4938 | free_eb: | |
4939 | btrfs_release_extent_buffer(eb); | |
4940 | return exists; | |
4941 | } | |
4942 | #endif | |
4943 | ||
f28491e0 | 4944 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
ce3e6984 | 4945 | u64 start) |
d1310b2e | 4946 | { |
da17066c | 4947 | unsigned long len = fs_info->nodesize; |
d1310b2e CM |
4948 | unsigned long num_pages = num_extent_pages(start, len); |
4949 | unsigned long i; | |
09cbfeaf | 4950 | unsigned long index = start >> PAGE_SHIFT; |
d1310b2e | 4951 | struct extent_buffer *eb; |
6af118ce | 4952 | struct extent_buffer *exists = NULL; |
d1310b2e | 4953 | struct page *p; |
f28491e0 | 4954 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
d1310b2e | 4955 | int uptodate = 1; |
19fe0a8b | 4956 | int ret; |
d1310b2e | 4957 | |
da17066c | 4958 | if (!IS_ALIGNED(start, fs_info->sectorsize)) { |
c871b0f2 LB |
4959 | btrfs_err(fs_info, "bad tree block start %llu", start); |
4960 | return ERR_PTR(-EINVAL); | |
4961 | } | |
4962 | ||
f28491e0 | 4963 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 4964 | if (eb) |
6af118ce | 4965 | return eb; |
6af118ce | 4966 | |
23d79d81 | 4967 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 4968 | if (!eb) |
c871b0f2 | 4969 | return ERR_PTR(-ENOMEM); |
d1310b2e | 4970 | |
727011e0 | 4971 | for (i = 0; i < num_pages; i++, index++) { |
d1b5c567 | 4972 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
c871b0f2 LB |
4973 | if (!p) { |
4974 | exists = ERR_PTR(-ENOMEM); | |
6af118ce | 4975 | goto free_eb; |
c871b0f2 | 4976 | } |
4f2de97a JB |
4977 | |
4978 | spin_lock(&mapping->private_lock); | |
4979 | if (PagePrivate(p)) { | |
4980 | /* | |
4981 | * We could have already allocated an eb for this page | |
4982 | * and attached one so lets see if we can get a ref on | |
4983 | * the existing eb, and if we can we know it's good and | |
4984 | * we can just return that one, else we know we can just | |
4985 | * overwrite page->private. | |
4986 | */ | |
4987 | exists = (struct extent_buffer *)p->private; | |
4988 | if (atomic_inc_not_zero(&exists->refs)) { | |
4989 | spin_unlock(&mapping->private_lock); | |
4990 | unlock_page(p); | |
09cbfeaf | 4991 | put_page(p); |
2457aec6 | 4992 | mark_extent_buffer_accessed(exists, p); |
4f2de97a JB |
4993 | goto free_eb; |
4994 | } | |
5ca64f45 | 4995 | exists = NULL; |
4f2de97a | 4996 | |
0b32f4bb | 4997 | /* |
4f2de97a JB |
4998 | * Do this so attach doesn't complain and we need to |
4999 | * drop the ref the old guy had. | |
5000 | */ | |
5001 | ClearPagePrivate(p); | |
0b32f4bb | 5002 | WARN_ON(PageDirty(p)); |
09cbfeaf | 5003 | put_page(p); |
d1310b2e | 5004 | } |
4f2de97a JB |
5005 | attach_extent_buffer_page(eb, p); |
5006 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 5007 | WARN_ON(PageDirty(p)); |
727011e0 | 5008 | eb->pages[i] = p; |
d1310b2e CM |
5009 | if (!PageUptodate(p)) |
5010 | uptodate = 0; | |
eb14ab8e CM |
5011 | |
5012 | /* | |
5013 | * see below about how we avoid a nasty race with release page | |
5014 | * and why we unlock later | |
5015 | */ | |
d1310b2e CM |
5016 | } |
5017 | if (uptodate) | |
b4ce94de | 5018 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 5019 | again: |
e1860a77 | 5020 | ret = radix_tree_preload(GFP_NOFS); |
c871b0f2 LB |
5021 | if (ret) { |
5022 | exists = ERR_PTR(ret); | |
19fe0a8b | 5023 | goto free_eb; |
c871b0f2 | 5024 | } |
19fe0a8b | 5025 | |
f28491e0 JB |
5026 | spin_lock(&fs_info->buffer_lock); |
5027 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
09cbfeaf | 5028 | start >> PAGE_SHIFT, eb); |
f28491e0 | 5029 | spin_unlock(&fs_info->buffer_lock); |
452c75c3 | 5030 | radix_tree_preload_end(); |
19fe0a8b | 5031 | if (ret == -EEXIST) { |
f28491e0 | 5032 | exists = find_extent_buffer(fs_info, start); |
452c75c3 CS |
5033 | if (exists) |
5034 | goto free_eb; | |
5035 | else | |
115391d2 | 5036 | goto again; |
6af118ce | 5037 | } |
6af118ce | 5038 | /* add one reference for the tree */ |
0b32f4bb | 5039 | check_buffer_tree_ref(eb); |
34b41ace | 5040 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
5041 | |
5042 | /* | |
5043 | * there is a race where release page may have | |
5044 | * tried to find this extent buffer in the radix | |
5045 | * but failed. It will tell the VM it is safe to | |
5046 | * reclaim the, and it will clear the page private bit. | |
5047 | * We must make sure to set the page private bit properly | |
5048 | * after the extent buffer is in the radix tree so | |
5049 | * it doesn't get lost | |
5050 | */ | |
727011e0 CM |
5051 | SetPageChecked(eb->pages[0]); |
5052 | for (i = 1; i < num_pages; i++) { | |
fb85fc9a | 5053 | p = eb->pages[i]; |
727011e0 CM |
5054 | ClearPageChecked(p); |
5055 | unlock_page(p); | |
5056 | } | |
5057 | unlock_page(eb->pages[0]); | |
d1310b2e CM |
5058 | return eb; |
5059 | ||
6af118ce | 5060 | free_eb: |
5ca64f45 | 5061 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
5062 | for (i = 0; i < num_pages; i++) { |
5063 | if (eb->pages[i]) | |
5064 | unlock_page(eb->pages[i]); | |
5065 | } | |
eb14ab8e | 5066 | |
897ca6e9 | 5067 | btrfs_release_extent_buffer(eb); |
6af118ce | 5068 | return exists; |
d1310b2e | 5069 | } |
d1310b2e | 5070 | |
3083ee2e JB |
5071 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
5072 | { | |
5073 | struct extent_buffer *eb = | |
5074 | container_of(head, struct extent_buffer, rcu_head); | |
5075 | ||
5076 | __free_extent_buffer(eb); | |
5077 | } | |
5078 | ||
3083ee2e | 5079 | /* Expects to have eb->eb_lock already held */ |
f7a52a40 | 5080 | static int release_extent_buffer(struct extent_buffer *eb) |
3083ee2e JB |
5081 | { |
5082 | WARN_ON(atomic_read(&eb->refs) == 0); | |
5083 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 5084 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 5085 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 5086 | |
815a51c7 | 5087 | spin_unlock(&eb->refs_lock); |
3083ee2e | 5088 | |
f28491e0 JB |
5089 | spin_lock(&fs_info->buffer_lock); |
5090 | radix_tree_delete(&fs_info->buffer_radix, | |
09cbfeaf | 5091 | eb->start >> PAGE_SHIFT); |
f28491e0 | 5092 | spin_unlock(&fs_info->buffer_lock); |
34b41ace JB |
5093 | } else { |
5094 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 5095 | } |
3083ee2e JB |
5096 | |
5097 | /* Should be safe to release our pages at this point */ | |
a50924e3 | 5098 | btrfs_release_extent_buffer_page(eb); |
bcb7e449 JB |
5099 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
5100 | if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) { | |
5101 | __free_extent_buffer(eb); | |
5102 | return 1; | |
5103 | } | |
5104 | #endif | |
3083ee2e | 5105 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 5106 | return 1; |
3083ee2e JB |
5107 | } |
5108 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
5109 | |
5110 | return 0; | |
3083ee2e JB |
5111 | } |
5112 | ||
d1310b2e CM |
5113 | void free_extent_buffer(struct extent_buffer *eb) |
5114 | { | |
242e18c7 CM |
5115 | int refs; |
5116 | int old; | |
d1310b2e CM |
5117 | if (!eb) |
5118 | return; | |
5119 | ||
242e18c7 CM |
5120 | while (1) { |
5121 | refs = atomic_read(&eb->refs); | |
5122 | if (refs <= 3) | |
5123 | break; | |
5124 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
5125 | if (old == refs) | |
5126 | return; | |
5127 | } | |
5128 | ||
3083ee2e | 5129 | spin_lock(&eb->refs_lock); |
815a51c7 JS |
5130 | if (atomic_read(&eb->refs) == 2 && |
5131 | test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) | |
5132 | atomic_dec(&eb->refs); | |
5133 | ||
3083ee2e JB |
5134 | if (atomic_read(&eb->refs) == 2 && |
5135 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 5136 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
5137 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5138 | atomic_dec(&eb->refs); | |
5139 | ||
5140 | /* | |
5141 | * I know this is terrible, but it's temporary until we stop tracking | |
5142 | * the uptodate bits and such for the extent buffers. | |
5143 | */ | |
f7a52a40 | 5144 | release_extent_buffer(eb); |
3083ee2e JB |
5145 | } |
5146 | ||
5147 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
5148 | { | |
5149 | if (!eb) | |
d1310b2e CM |
5150 | return; |
5151 | ||
3083ee2e JB |
5152 | spin_lock(&eb->refs_lock); |
5153 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
5154 | ||
0b32f4bb | 5155 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
5156 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5157 | atomic_dec(&eb->refs); | |
f7a52a40 | 5158 | release_extent_buffer(eb); |
d1310b2e | 5159 | } |
d1310b2e | 5160 | |
1d4284bd | 5161 | void clear_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5162 | { |
d1310b2e CM |
5163 | unsigned long i; |
5164 | unsigned long num_pages; | |
5165 | struct page *page; | |
5166 | ||
d1310b2e CM |
5167 | num_pages = num_extent_pages(eb->start, eb->len); |
5168 | ||
5169 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 5170 | page = eb->pages[i]; |
b9473439 | 5171 | if (!PageDirty(page)) |
d2c3f4f6 CM |
5172 | continue; |
5173 | ||
a61e6f29 | 5174 | lock_page(page); |
eb14ab8e CM |
5175 | WARN_ON(!PagePrivate(page)); |
5176 | ||
d1310b2e | 5177 | clear_page_dirty_for_io(page); |
b93b0163 | 5178 | xa_lock_irq(&page->mapping->i_pages); |
d1310b2e | 5179 | if (!PageDirty(page)) { |
b93b0163 | 5180 | radix_tree_tag_clear(&page->mapping->i_pages, |
d1310b2e CM |
5181 | page_index(page), |
5182 | PAGECACHE_TAG_DIRTY); | |
5183 | } | |
b93b0163 | 5184 | xa_unlock_irq(&page->mapping->i_pages); |
bf0da8c1 | 5185 | ClearPageError(page); |
a61e6f29 | 5186 | unlock_page(page); |
d1310b2e | 5187 | } |
0b32f4bb | 5188 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 5189 | } |
d1310b2e | 5190 | |
0b32f4bb | 5191 | int set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e CM |
5192 | { |
5193 | unsigned long i; | |
5194 | unsigned long num_pages; | |
b9473439 | 5195 | int was_dirty = 0; |
d1310b2e | 5196 | |
0b32f4bb JB |
5197 | check_buffer_tree_ref(eb); |
5198 | ||
b9473439 | 5199 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 5200 | |
d1310b2e | 5201 | num_pages = num_extent_pages(eb->start, eb->len); |
3083ee2e | 5202 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
5203 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
5204 | ||
b9473439 | 5205 | for (i = 0; i < num_pages; i++) |
fb85fc9a | 5206 | set_page_dirty(eb->pages[i]); |
b9473439 | 5207 | return was_dirty; |
d1310b2e | 5208 | } |
d1310b2e | 5209 | |
69ba3927 | 5210 | void clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 CM |
5211 | { |
5212 | unsigned long i; | |
5213 | struct page *page; | |
5214 | unsigned long num_pages; | |
5215 | ||
b4ce94de | 5216 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
0b32f4bb | 5217 | num_pages = num_extent_pages(eb->start, eb->len); |
1259ab75 | 5218 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5219 | page = eb->pages[i]; |
33958dc6 CM |
5220 | if (page) |
5221 | ClearPageUptodate(page); | |
1259ab75 | 5222 | } |
1259ab75 CM |
5223 | } |
5224 | ||
09c25a8c | 5225 | void set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e CM |
5226 | { |
5227 | unsigned long i; | |
5228 | struct page *page; | |
5229 | unsigned long num_pages; | |
5230 | ||
0b32f4bb | 5231 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 5232 | num_pages = num_extent_pages(eb->start, eb->len); |
d1310b2e | 5233 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5234 | page = eb->pages[i]; |
d1310b2e CM |
5235 | SetPageUptodate(page); |
5236 | } | |
d1310b2e | 5237 | } |
d1310b2e | 5238 | |
d1310b2e | 5239 | int read_extent_buffer_pages(struct extent_io_tree *tree, |
6af49dbd | 5240 | struct extent_buffer *eb, int wait, int mirror_num) |
d1310b2e CM |
5241 | { |
5242 | unsigned long i; | |
d1310b2e CM |
5243 | struct page *page; |
5244 | int err; | |
5245 | int ret = 0; | |
ce9adaa5 CM |
5246 | int locked_pages = 0; |
5247 | int all_uptodate = 1; | |
d1310b2e | 5248 | unsigned long num_pages; |
727011e0 | 5249 | unsigned long num_reads = 0; |
a86c12c7 | 5250 | struct bio *bio = NULL; |
c8b97818 | 5251 | unsigned long bio_flags = 0; |
a86c12c7 | 5252 | |
b4ce94de | 5253 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
5254 | return 0; |
5255 | ||
d1310b2e | 5256 | num_pages = num_extent_pages(eb->start, eb->len); |
8436ea91 | 5257 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5258 | page = eb->pages[i]; |
bb82ab88 | 5259 | if (wait == WAIT_NONE) { |
2db04966 | 5260 | if (!trylock_page(page)) |
ce9adaa5 | 5261 | goto unlock_exit; |
d1310b2e CM |
5262 | } else { |
5263 | lock_page(page); | |
5264 | } | |
ce9adaa5 | 5265 | locked_pages++; |
2571e739 LB |
5266 | } |
5267 | /* | |
5268 | * We need to firstly lock all pages to make sure that | |
5269 | * the uptodate bit of our pages won't be affected by | |
5270 | * clear_extent_buffer_uptodate(). | |
5271 | */ | |
8436ea91 | 5272 | for (i = 0; i < num_pages; i++) { |
2571e739 | 5273 | page = eb->pages[i]; |
727011e0 CM |
5274 | if (!PageUptodate(page)) { |
5275 | num_reads++; | |
ce9adaa5 | 5276 | all_uptodate = 0; |
727011e0 | 5277 | } |
ce9adaa5 | 5278 | } |
2571e739 | 5279 | |
ce9adaa5 | 5280 | if (all_uptodate) { |
8436ea91 | 5281 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
5282 | goto unlock_exit; |
5283 | } | |
5284 | ||
656f30db | 5285 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 5286 | eb->read_mirror = 0; |
0b32f4bb | 5287 | atomic_set(&eb->io_pages, num_reads); |
8436ea91 | 5288 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5289 | page = eb->pages[i]; |
baf863b9 | 5290 | |
ce9adaa5 | 5291 | if (!PageUptodate(page)) { |
baf863b9 LB |
5292 | if (ret) { |
5293 | atomic_dec(&eb->io_pages); | |
5294 | unlock_page(page); | |
5295 | continue; | |
5296 | } | |
5297 | ||
f188591e | 5298 | ClearPageError(page); |
a86c12c7 | 5299 | err = __extent_read_full_page(tree, page, |
6af49dbd | 5300 | btree_get_extent, &bio, |
d4c7ca86 | 5301 | mirror_num, &bio_flags, |
1f7ad75b | 5302 | REQ_META); |
baf863b9 | 5303 | if (err) { |
d1310b2e | 5304 | ret = err; |
baf863b9 LB |
5305 | /* |
5306 | * We use &bio in above __extent_read_full_page, | |
5307 | * so we ensure that if it returns error, the | |
5308 | * current page fails to add itself to bio and | |
5309 | * it's been unlocked. | |
5310 | * | |
5311 | * We must dec io_pages by ourselves. | |
5312 | */ | |
5313 | atomic_dec(&eb->io_pages); | |
5314 | } | |
d1310b2e CM |
5315 | } else { |
5316 | unlock_page(page); | |
5317 | } | |
5318 | } | |
5319 | ||
355808c2 | 5320 | if (bio) { |
1f7ad75b | 5321 | err = submit_one_bio(bio, mirror_num, bio_flags); |
79787eaa JM |
5322 | if (err) |
5323 | return err; | |
355808c2 | 5324 | } |
a86c12c7 | 5325 | |
bb82ab88 | 5326 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 5327 | return ret; |
d397712b | 5328 | |
8436ea91 | 5329 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5330 | page = eb->pages[i]; |
d1310b2e | 5331 | wait_on_page_locked(page); |
d397712b | 5332 | if (!PageUptodate(page)) |
d1310b2e | 5333 | ret = -EIO; |
d1310b2e | 5334 | } |
d397712b | 5335 | |
d1310b2e | 5336 | return ret; |
ce9adaa5 CM |
5337 | |
5338 | unlock_exit: | |
d397712b | 5339 | while (locked_pages > 0) { |
ce9adaa5 | 5340 | locked_pages--; |
8436ea91 JB |
5341 | page = eb->pages[locked_pages]; |
5342 | unlock_page(page); | |
ce9adaa5 CM |
5343 | } |
5344 | return ret; | |
d1310b2e | 5345 | } |
d1310b2e | 5346 | |
1cbb1f45 JM |
5347 | void read_extent_buffer(const struct extent_buffer *eb, void *dstv, |
5348 | unsigned long start, unsigned long len) | |
d1310b2e CM |
5349 | { |
5350 | size_t cur; | |
5351 | size_t offset; | |
5352 | struct page *page; | |
5353 | char *kaddr; | |
5354 | char *dst = (char *)dstv; | |
09cbfeaf KS |
5355 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5356 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e | 5357 | |
f716abd5 LB |
5358 | if (start + len > eb->len) { |
5359 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n", | |
5360 | eb->start, eb->len, start, len); | |
5361 | memset(dst, 0, len); | |
5362 | return; | |
5363 | } | |
d1310b2e | 5364 | |
09cbfeaf | 5365 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
d1310b2e | 5366 | |
d397712b | 5367 | while (len > 0) { |
fb85fc9a | 5368 | page = eb->pages[i]; |
d1310b2e | 5369 | |
09cbfeaf | 5370 | cur = min(len, (PAGE_SIZE - offset)); |
a6591715 | 5371 | kaddr = page_address(page); |
d1310b2e | 5372 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
5373 | |
5374 | dst += cur; | |
5375 | len -= cur; | |
5376 | offset = 0; | |
5377 | i++; | |
5378 | } | |
5379 | } | |
d1310b2e | 5380 | |
1cbb1f45 JM |
5381 | int read_extent_buffer_to_user(const struct extent_buffer *eb, |
5382 | void __user *dstv, | |
5383 | unsigned long start, unsigned long len) | |
550ac1d8 GH |
5384 | { |
5385 | size_t cur; | |
5386 | size_t offset; | |
5387 | struct page *page; | |
5388 | char *kaddr; | |
5389 | char __user *dst = (char __user *)dstv; | |
09cbfeaf KS |
5390 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5391 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
550ac1d8 GH |
5392 | int ret = 0; |
5393 | ||
5394 | WARN_ON(start > eb->len); | |
5395 | WARN_ON(start + len > eb->start + eb->len); | |
5396 | ||
09cbfeaf | 5397 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
550ac1d8 GH |
5398 | |
5399 | while (len > 0) { | |
fb85fc9a | 5400 | page = eb->pages[i]; |
550ac1d8 | 5401 | |
09cbfeaf | 5402 | cur = min(len, (PAGE_SIZE - offset)); |
550ac1d8 GH |
5403 | kaddr = page_address(page); |
5404 | if (copy_to_user(dst, kaddr + offset, cur)) { | |
5405 | ret = -EFAULT; | |
5406 | break; | |
5407 | } | |
5408 | ||
5409 | dst += cur; | |
5410 | len -= cur; | |
5411 | offset = 0; | |
5412 | i++; | |
5413 | } | |
5414 | ||
5415 | return ret; | |
5416 | } | |
5417 | ||
415b35a5 LB |
5418 | /* |
5419 | * return 0 if the item is found within a page. | |
5420 | * return 1 if the item spans two pages. | |
5421 | * return -EINVAL otherwise. | |
5422 | */ | |
1cbb1f45 JM |
5423 | int map_private_extent_buffer(const struct extent_buffer *eb, |
5424 | unsigned long start, unsigned long min_len, | |
5425 | char **map, unsigned long *map_start, | |
5426 | unsigned long *map_len) | |
d1310b2e | 5427 | { |
09cbfeaf | 5428 | size_t offset = start & (PAGE_SIZE - 1); |
d1310b2e CM |
5429 | char *kaddr; |
5430 | struct page *p; | |
09cbfeaf KS |
5431 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5432 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e | 5433 | unsigned long end_i = (start_offset + start + min_len - 1) >> |
09cbfeaf | 5434 | PAGE_SHIFT; |
d1310b2e | 5435 | |
f716abd5 LB |
5436 | if (start + min_len > eb->len) { |
5437 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n", | |
5438 | eb->start, eb->len, start, min_len); | |
5439 | return -EINVAL; | |
5440 | } | |
5441 | ||
d1310b2e | 5442 | if (i != end_i) |
415b35a5 | 5443 | return 1; |
d1310b2e CM |
5444 | |
5445 | if (i == 0) { | |
5446 | offset = start_offset; | |
5447 | *map_start = 0; | |
5448 | } else { | |
5449 | offset = 0; | |
09cbfeaf | 5450 | *map_start = ((u64)i << PAGE_SHIFT) - start_offset; |
d1310b2e | 5451 | } |
d397712b | 5452 | |
fb85fc9a | 5453 | p = eb->pages[i]; |
a6591715 | 5454 | kaddr = page_address(p); |
d1310b2e | 5455 | *map = kaddr + offset; |
09cbfeaf | 5456 | *map_len = PAGE_SIZE - offset; |
d1310b2e CM |
5457 | return 0; |
5458 | } | |
d1310b2e | 5459 | |
1cbb1f45 JM |
5460 | int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv, |
5461 | unsigned long start, unsigned long len) | |
d1310b2e CM |
5462 | { |
5463 | size_t cur; | |
5464 | size_t offset; | |
5465 | struct page *page; | |
5466 | char *kaddr; | |
5467 | char *ptr = (char *)ptrv; | |
09cbfeaf KS |
5468 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5469 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e CM |
5470 | int ret = 0; |
5471 | ||
5472 | WARN_ON(start > eb->len); | |
5473 | WARN_ON(start + len > eb->start + eb->len); | |
5474 | ||
09cbfeaf | 5475 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
d1310b2e | 5476 | |
d397712b | 5477 | while (len > 0) { |
fb85fc9a | 5478 | page = eb->pages[i]; |
d1310b2e | 5479 | |
09cbfeaf | 5480 | cur = min(len, (PAGE_SIZE - offset)); |
d1310b2e | 5481 | |
a6591715 | 5482 | kaddr = page_address(page); |
d1310b2e | 5483 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
5484 | if (ret) |
5485 | break; | |
5486 | ||
5487 | ptr += cur; | |
5488 | len -= cur; | |
5489 | offset = 0; | |
5490 | i++; | |
5491 | } | |
5492 | return ret; | |
5493 | } | |
d1310b2e | 5494 | |
f157bf76 DS |
5495 | void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb, |
5496 | const void *srcv) | |
5497 | { | |
5498 | char *kaddr; | |
5499 | ||
5500 | WARN_ON(!PageUptodate(eb->pages[0])); | |
5501 | kaddr = page_address(eb->pages[0]); | |
5502 | memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv, | |
5503 | BTRFS_FSID_SIZE); | |
5504 | } | |
5505 | ||
5506 | void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv) | |
5507 | { | |
5508 | char *kaddr; | |
5509 | ||
5510 | WARN_ON(!PageUptodate(eb->pages[0])); | |
5511 | kaddr = page_address(eb->pages[0]); | |
5512 | memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv, | |
5513 | BTRFS_FSID_SIZE); | |
5514 | } | |
5515 | ||
d1310b2e CM |
5516 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, |
5517 | unsigned long start, unsigned long len) | |
5518 | { | |
5519 | size_t cur; | |
5520 | size_t offset; | |
5521 | struct page *page; | |
5522 | char *kaddr; | |
5523 | char *src = (char *)srcv; | |
09cbfeaf KS |
5524 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5525 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e CM |
5526 | |
5527 | WARN_ON(start > eb->len); | |
5528 | WARN_ON(start + len > eb->start + eb->len); | |
5529 | ||
09cbfeaf | 5530 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
d1310b2e | 5531 | |
d397712b | 5532 | while (len > 0) { |
fb85fc9a | 5533 | page = eb->pages[i]; |
d1310b2e CM |
5534 | WARN_ON(!PageUptodate(page)); |
5535 | ||
09cbfeaf | 5536 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 5537 | kaddr = page_address(page); |
d1310b2e | 5538 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
5539 | |
5540 | src += cur; | |
5541 | len -= cur; | |
5542 | offset = 0; | |
5543 | i++; | |
5544 | } | |
5545 | } | |
d1310b2e | 5546 | |
b159fa28 DS |
5547 | void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start, |
5548 | unsigned long len) | |
d1310b2e CM |
5549 | { |
5550 | size_t cur; | |
5551 | size_t offset; | |
5552 | struct page *page; | |
5553 | char *kaddr; | |
09cbfeaf KS |
5554 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5555 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e CM |
5556 | |
5557 | WARN_ON(start > eb->len); | |
5558 | WARN_ON(start + len > eb->start + eb->len); | |
5559 | ||
09cbfeaf | 5560 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
d1310b2e | 5561 | |
d397712b | 5562 | while (len > 0) { |
fb85fc9a | 5563 | page = eb->pages[i]; |
d1310b2e CM |
5564 | WARN_ON(!PageUptodate(page)); |
5565 | ||
09cbfeaf | 5566 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 5567 | kaddr = page_address(page); |
b159fa28 | 5568 | memset(kaddr + offset, 0, cur); |
d1310b2e CM |
5569 | |
5570 | len -= cur; | |
5571 | offset = 0; | |
5572 | i++; | |
5573 | } | |
5574 | } | |
d1310b2e | 5575 | |
58e8012c DS |
5576 | void copy_extent_buffer_full(struct extent_buffer *dst, |
5577 | struct extent_buffer *src) | |
5578 | { | |
5579 | int i; | |
5580 | unsigned num_pages; | |
5581 | ||
5582 | ASSERT(dst->len == src->len); | |
5583 | ||
5584 | num_pages = num_extent_pages(dst->start, dst->len); | |
5585 | for (i = 0; i < num_pages; i++) | |
5586 | copy_page(page_address(dst->pages[i]), | |
5587 | page_address(src->pages[i])); | |
5588 | } | |
5589 | ||
d1310b2e CM |
5590 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, |
5591 | unsigned long dst_offset, unsigned long src_offset, | |
5592 | unsigned long len) | |
5593 | { | |
5594 | u64 dst_len = dst->len; | |
5595 | size_t cur; | |
5596 | size_t offset; | |
5597 | struct page *page; | |
5598 | char *kaddr; | |
09cbfeaf KS |
5599 | size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1); |
5600 | unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT; | |
d1310b2e CM |
5601 | |
5602 | WARN_ON(src->len != dst_len); | |
5603 | ||
5604 | offset = (start_offset + dst_offset) & | |
09cbfeaf | 5605 | (PAGE_SIZE - 1); |
d1310b2e | 5606 | |
d397712b | 5607 | while (len > 0) { |
fb85fc9a | 5608 | page = dst->pages[i]; |
d1310b2e CM |
5609 | WARN_ON(!PageUptodate(page)); |
5610 | ||
09cbfeaf | 5611 | cur = min(len, (unsigned long)(PAGE_SIZE - offset)); |
d1310b2e | 5612 | |
a6591715 | 5613 | kaddr = page_address(page); |
d1310b2e | 5614 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
5615 | |
5616 | src_offset += cur; | |
5617 | len -= cur; | |
5618 | offset = 0; | |
5619 | i++; | |
5620 | } | |
5621 | } | |
d1310b2e | 5622 | |
2fe1d551 OS |
5623 | void le_bitmap_set(u8 *map, unsigned int start, int len) |
5624 | { | |
5625 | u8 *p = map + BIT_BYTE(start); | |
5626 | const unsigned int size = start + len; | |
5627 | int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE); | |
5628 | u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start); | |
5629 | ||
5630 | while (len - bits_to_set >= 0) { | |
5631 | *p |= mask_to_set; | |
5632 | len -= bits_to_set; | |
5633 | bits_to_set = BITS_PER_BYTE; | |
9c894696 | 5634 | mask_to_set = ~0; |
2fe1d551 OS |
5635 | p++; |
5636 | } | |
5637 | if (len) { | |
5638 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
5639 | *p |= mask_to_set; | |
5640 | } | |
5641 | } | |
5642 | ||
5643 | void le_bitmap_clear(u8 *map, unsigned int start, int len) | |
5644 | { | |
5645 | u8 *p = map + BIT_BYTE(start); | |
5646 | const unsigned int size = start + len; | |
5647 | int bits_to_clear = BITS_PER_BYTE - (start % BITS_PER_BYTE); | |
5648 | u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(start); | |
5649 | ||
5650 | while (len - bits_to_clear >= 0) { | |
5651 | *p &= ~mask_to_clear; | |
5652 | len -= bits_to_clear; | |
5653 | bits_to_clear = BITS_PER_BYTE; | |
9c894696 | 5654 | mask_to_clear = ~0; |
2fe1d551 OS |
5655 | p++; |
5656 | } | |
5657 | if (len) { | |
5658 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
5659 | *p &= ~mask_to_clear; | |
5660 | } | |
5661 | } | |
3e1e8bb7 OS |
5662 | |
5663 | /* | |
5664 | * eb_bitmap_offset() - calculate the page and offset of the byte containing the | |
5665 | * given bit number | |
5666 | * @eb: the extent buffer | |
5667 | * @start: offset of the bitmap item in the extent buffer | |
5668 | * @nr: bit number | |
5669 | * @page_index: return index of the page in the extent buffer that contains the | |
5670 | * given bit number | |
5671 | * @page_offset: return offset into the page given by page_index | |
5672 | * | |
5673 | * This helper hides the ugliness of finding the byte in an extent buffer which | |
5674 | * contains a given bit. | |
5675 | */ | |
5676 | static inline void eb_bitmap_offset(struct extent_buffer *eb, | |
5677 | unsigned long start, unsigned long nr, | |
5678 | unsigned long *page_index, | |
5679 | size_t *page_offset) | |
5680 | { | |
09cbfeaf | 5681 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
3e1e8bb7 OS |
5682 | size_t byte_offset = BIT_BYTE(nr); |
5683 | size_t offset; | |
5684 | ||
5685 | /* | |
5686 | * The byte we want is the offset of the extent buffer + the offset of | |
5687 | * the bitmap item in the extent buffer + the offset of the byte in the | |
5688 | * bitmap item. | |
5689 | */ | |
5690 | offset = start_offset + start + byte_offset; | |
5691 | ||
09cbfeaf KS |
5692 | *page_index = offset >> PAGE_SHIFT; |
5693 | *page_offset = offset & (PAGE_SIZE - 1); | |
3e1e8bb7 OS |
5694 | } |
5695 | ||
5696 | /** | |
5697 | * extent_buffer_test_bit - determine whether a bit in a bitmap item is set | |
5698 | * @eb: the extent buffer | |
5699 | * @start: offset of the bitmap item in the extent buffer | |
5700 | * @nr: bit number to test | |
5701 | */ | |
5702 | int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start, | |
5703 | unsigned long nr) | |
5704 | { | |
2fe1d551 | 5705 | u8 *kaddr; |
3e1e8bb7 OS |
5706 | struct page *page; |
5707 | unsigned long i; | |
5708 | size_t offset; | |
5709 | ||
5710 | eb_bitmap_offset(eb, start, nr, &i, &offset); | |
5711 | page = eb->pages[i]; | |
5712 | WARN_ON(!PageUptodate(page)); | |
5713 | kaddr = page_address(page); | |
5714 | return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1))); | |
5715 | } | |
5716 | ||
5717 | /** | |
5718 | * extent_buffer_bitmap_set - set an area of a bitmap | |
5719 | * @eb: the extent buffer | |
5720 | * @start: offset of the bitmap item in the extent buffer | |
5721 | * @pos: bit number of the first bit | |
5722 | * @len: number of bits to set | |
5723 | */ | |
5724 | void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start, | |
5725 | unsigned long pos, unsigned long len) | |
5726 | { | |
2fe1d551 | 5727 | u8 *kaddr; |
3e1e8bb7 OS |
5728 | struct page *page; |
5729 | unsigned long i; | |
5730 | size_t offset; | |
5731 | const unsigned int size = pos + len; | |
5732 | int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 5733 | u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
5734 | |
5735 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5736 | page = eb->pages[i]; | |
5737 | WARN_ON(!PageUptodate(page)); | |
5738 | kaddr = page_address(page); | |
5739 | ||
5740 | while (len >= bits_to_set) { | |
5741 | kaddr[offset] |= mask_to_set; | |
5742 | len -= bits_to_set; | |
5743 | bits_to_set = BITS_PER_BYTE; | |
9c894696 | 5744 | mask_to_set = ~0; |
09cbfeaf | 5745 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5746 | offset = 0; |
5747 | page = eb->pages[++i]; | |
5748 | WARN_ON(!PageUptodate(page)); | |
5749 | kaddr = page_address(page); | |
5750 | } | |
5751 | } | |
5752 | if (len) { | |
5753 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
5754 | kaddr[offset] |= mask_to_set; | |
5755 | } | |
5756 | } | |
5757 | ||
5758 | ||
5759 | /** | |
5760 | * extent_buffer_bitmap_clear - clear an area of a bitmap | |
5761 | * @eb: the extent buffer | |
5762 | * @start: offset of the bitmap item in the extent buffer | |
5763 | * @pos: bit number of the first bit | |
5764 | * @len: number of bits to clear | |
5765 | */ | |
5766 | void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start, | |
5767 | unsigned long pos, unsigned long len) | |
5768 | { | |
2fe1d551 | 5769 | u8 *kaddr; |
3e1e8bb7 OS |
5770 | struct page *page; |
5771 | unsigned long i; | |
5772 | size_t offset; | |
5773 | const unsigned int size = pos + len; | |
5774 | int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 5775 | u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
5776 | |
5777 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5778 | page = eb->pages[i]; | |
5779 | WARN_ON(!PageUptodate(page)); | |
5780 | kaddr = page_address(page); | |
5781 | ||
5782 | while (len >= bits_to_clear) { | |
5783 | kaddr[offset] &= ~mask_to_clear; | |
5784 | len -= bits_to_clear; | |
5785 | bits_to_clear = BITS_PER_BYTE; | |
9c894696 | 5786 | mask_to_clear = ~0; |
09cbfeaf | 5787 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5788 | offset = 0; |
5789 | page = eb->pages[++i]; | |
5790 | WARN_ON(!PageUptodate(page)); | |
5791 | kaddr = page_address(page); | |
5792 | } | |
5793 | } | |
5794 | if (len) { | |
5795 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
5796 | kaddr[offset] &= ~mask_to_clear; | |
5797 | } | |
5798 | } | |
5799 | ||
3387206f ST |
5800 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
5801 | { | |
5802 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
5803 | return distance < len; | |
5804 | } | |
5805 | ||
d1310b2e CM |
5806 | static void copy_pages(struct page *dst_page, struct page *src_page, |
5807 | unsigned long dst_off, unsigned long src_off, | |
5808 | unsigned long len) | |
5809 | { | |
a6591715 | 5810 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 5811 | char *src_kaddr; |
727011e0 | 5812 | int must_memmove = 0; |
d1310b2e | 5813 | |
3387206f | 5814 | if (dst_page != src_page) { |
a6591715 | 5815 | src_kaddr = page_address(src_page); |
3387206f | 5816 | } else { |
d1310b2e | 5817 | src_kaddr = dst_kaddr; |
727011e0 CM |
5818 | if (areas_overlap(src_off, dst_off, len)) |
5819 | must_memmove = 1; | |
3387206f | 5820 | } |
d1310b2e | 5821 | |
727011e0 CM |
5822 | if (must_memmove) |
5823 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
5824 | else | |
5825 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
5826 | } |
5827 | ||
5828 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5829 | unsigned long src_offset, unsigned long len) | |
5830 | { | |
0b246afa | 5831 | struct btrfs_fs_info *fs_info = dst->fs_info; |
d1310b2e CM |
5832 | size_t cur; |
5833 | size_t dst_off_in_page; | |
5834 | size_t src_off_in_page; | |
09cbfeaf | 5835 | size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1); |
d1310b2e CM |
5836 | unsigned long dst_i; |
5837 | unsigned long src_i; | |
5838 | ||
5839 | if (src_offset + len > dst->len) { | |
0b246afa | 5840 | btrfs_err(fs_info, |
5d163e0e JM |
5841 | "memmove bogus src_offset %lu move len %lu dst len %lu", |
5842 | src_offset, len, dst->len); | |
d1310b2e CM |
5843 | BUG_ON(1); |
5844 | } | |
5845 | if (dst_offset + len > dst->len) { | |
0b246afa | 5846 | btrfs_err(fs_info, |
5d163e0e JM |
5847 | "memmove bogus dst_offset %lu move len %lu dst len %lu", |
5848 | dst_offset, len, dst->len); | |
d1310b2e CM |
5849 | BUG_ON(1); |
5850 | } | |
5851 | ||
d397712b | 5852 | while (len > 0) { |
d1310b2e | 5853 | dst_off_in_page = (start_offset + dst_offset) & |
09cbfeaf | 5854 | (PAGE_SIZE - 1); |
d1310b2e | 5855 | src_off_in_page = (start_offset + src_offset) & |
09cbfeaf | 5856 | (PAGE_SIZE - 1); |
d1310b2e | 5857 | |
09cbfeaf KS |
5858 | dst_i = (start_offset + dst_offset) >> PAGE_SHIFT; |
5859 | src_i = (start_offset + src_offset) >> PAGE_SHIFT; | |
d1310b2e | 5860 | |
09cbfeaf | 5861 | cur = min(len, (unsigned long)(PAGE_SIZE - |
d1310b2e CM |
5862 | src_off_in_page)); |
5863 | cur = min_t(unsigned long, cur, | |
09cbfeaf | 5864 | (unsigned long)(PAGE_SIZE - dst_off_in_page)); |
d1310b2e | 5865 | |
fb85fc9a | 5866 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5867 | dst_off_in_page, src_off_in_page, cur); |
5868 | ||
5869 | src_offset += cur; | |
5870 | dst_offset += cur; | |
5871 | len -= cur; | |
5872 | } | |
5873 | } | |
d1310b2e CM |
5874 | |
5875 | void memmove_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; | |
5882 | unsigned long dst_end = dst_offset + len - 1; | |
5883 | unsigned long src_end = src_offset + len - 1; | |
09cbfeaf | 5884 | size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1); |
d1310b2e CM |
5885 | unsigned long dst_i; |
5886 | unsigned long src_i; | |
5887 | ||
5888 | if (src_offset + len > dst->len) { | |
0b246afa | 5889 | btrfs_err(fs_info, |
5d163e0e JM |
5890 | "memmove bogus src_offset %lu move len %lu len %lu", |
5891 | src_offset, len, dst->len); | |
d1310b2e CM |
5892 | BUG_ON(1); |
5893 | } | |
5894 | if (dst_offset + len > dst->len) { | |
0b246afa | 5895 | btrfs_err(fs_info, |
5d163e0e JM |
5896 | "memmove bogus dst_offset %lu move len %lu len %lu", |
5897 | dst_offset, len, dst->len); | |
d1310b2e CM |
5898 | BUG_ON(1); |
5899 | } | |
727011e0 | 5900 | if (dst_offset < src_offset) { |
d1310b2e CM |
5901 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
5902 | return; | |
5903 | } | |
d397712b | 5904 | while (len > 0) { |
09cbfeaf KS |
5905 | dst_i = (start_offset + dst_end) >> PAGE_SHIFT; |
5906 | src_i = (start_offset + src_end) >> PAGE_SHIFT; | |
d1310b2e CM |
5907 | |
5908 | dst_off_in_page = (start_offset + dst_end) & | |
09cbfeaf | 5909 | (PAGE_SIZE - 1); |
d1310b2e | 5910 | src_off_in_page = (start_offset + src_end) & |
09cbfeaf | 5911 | (PAGE_SIZE - 1); |
d1310b2e CM |
5912 | |
5913 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
5914 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 5915 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5916 | dst_off_in_page - cur + 1, |
5917 | src_off_in_page - cur + 1, cur); | |
5918 | ||
5919 | dst_end -= cur; | |
5920 | src_end -= cur; | |
5921 | len -= cur; | |
5922 | } | |
5923 | } | |
6af118ce | 5924 | |
f7a52a40 | 5925 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 5926 | { |
6af118ce | 5927 | struct extent_buffer *eb; |
6af118ce | 5928 | |
3083ee2e | 5929 | /* |
01327610 | 5930 | * We need to make sure nobody is attaching this page to an eb right |
3083ee2e JB |
5931 | * now. |
5932 | */ | |
5933 | spin_lock(&page->mapping->private_lock); | |
5934 | if (!PagePrivate(page)) { | |
5935 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 5936 | return 1; |
45f49bce | 5937 | } |
6af118ce | 5938 | |
3083ee2e JB |
5939 | eb = (struct extent_buffer *)page->private; |
5940 | BUG_ON(!eb); | |
19fe0a8b MX |
5941 | |
5942 | /* | |
3083ee2e JB |
5943 | * This is a little awful but should be ok, we need to make sure that |
5944 | * the eb doesn't disappear out from under us while we're looking at | |
5945 | * this page. | |
19fe0a8b | 5946 | */ |
3083ee2e | 5947 | spin_lock(&eb->refs_lock); |
0b32f4bb | 5948 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
5949 | spin_unlock(&eb->refs_lock); |
5950 | spin_unlock(&page->mapping->private_lock); | |
5951 | return 0; | |
b9473439 | 5952 | } |
3083ee2e | 5953 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 5954 | |
19fe0a8b | 5955 | /* |
3083ee2e JB |
5956 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
5957 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 5958 | */ |
3083ee2e JB |
5959 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
5960 | spin_unlock(&eb->refs_lock); | |
5961 | return 0; | |
b9473439 | 5962 | } |
19fe0a8b | 5963 | |
f7a52a40 | 5964 | return release_extent_buffer(eb); |
6af118ce | 5965 | } |