]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
d1310b2e CM |
2 | #include <linux/bitops.h> |
3 | #include <linux/slab.h> | |
4 | #include <linux/bio.h> | |
5 | #include <linux/mm.h> | |
d1310b2e CM |
6 | #include <linux/pagemap.h> |
7 | #include <linux/page-flags.h> | |
d1310b2e CM |
8 | #include <linux/spinlock.h> |
9 | #include <linux/blkdev.h> | |
10 | #include <linux/swap.h> | |
d1310b2e CM |
11 | #include <linux/writeback.h> |
12 | #include <linux/pagevec.h> | |
268bb0ce | 13 | #include <linux/prefetch.h> |
90a887c9 | 14 | #include <linux/cleancache.h> |
d1310b2e CM |
15 | #include "extent_io.h" |
16 | #include "extent_map.h" | |
902b22f3 DW |
17 | #include "ctree.h" |
18 | #include "btrfs_inode.h" | |
4a54c8c1 | 19 | #include "volumes.h" |
21adbd5c | 20 | #include "check-integrity.h" |
0b32f4bb | 21 | #include "locking.h" |
606686ee | 22 | #include "rcu-string.h" |
fe09e16c | 23 | #include "backref.h" |
d1310b2e | 24 | |
d1310b2e CM |
25 | static struct kmem_cache *extent_state_cache; |
26 | static struct kmem_cache *extent_buffer_cache; | |
9be3395b | 27 | static struct bio_set *btrfs_bioset; |
d1310b2e | 28 | |
27a3507d FM |
29 | static inline bool extent_state_in_tree(const struct extent_state *state) |
30 | { | |
31 | return !RB_EMPTY_NODE(&state->rb_node); | |
32 | } | |
33 | ||
6d49ba1b | 34 | #ifdef CONFIG_BTRFS_DEBUG |
d1310b2e CM |
35 | static LIST_HEAD(buffers); |
36 | static LIST_HEAD(states); | |
4bef0848 | 37 | |
d397712b | 38 | static DEFINE_SPINLOCK(leak_lock); |
6d49ba1b ES |
39 | |
40 | static inline | |
41 | void btrfs_leak_debug_add(struct list_head *new, struct list_head *head) | |
42 | { | |
43 | unsigned long flags; | |
44 | ||
45 | spin_lock_irqsave(&leak_lock, flags); | |
46 | list_add(new, head); | |
47 | spin_unlock_irqrestore(&leak_lock, flags); | |
48 | } | |
49 | ||
50 | static inline | |
51 | void btrfs_leak_debug_del(struct list_head *entry) | |
52 | { | |
53 | unsigned long flags; | |
54 | ||
55 | spin_lock_irqsave(&leak_lock, flags); | |
56 | list_del(entry); | |
57 | spin_unlock_irqrestore(&leak_lock, flags); | |
58 | } | |
59 | ||
60 | static inline | |
61 | void btrfs_leak_debug_check(void) | |
62 | { | |
63 | struct extent_state *state; | |
64 | struct extent_buffer *eb; | |
65 | ||
66 | while (!list_empty(&states)) { | |
67 | state = list_entry(states.next, struct extent_state, leak_list); | |
9ee49a04 | 68 | pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n", |
27a3507d FM |
69 | state->start, state->end, state->state, |
70 | extent_state_in_tree(state), | |
b7ac31b7 | 71 | refcount_read(&state->refs)); |
6d49ba1b ES |
72 | list_del(&state->leak_list); |
73 | kmem_cache_free(extent_state_cache, state); | |
74 | } | |
75 | ||
76 | while (!list_empty(&buffers)) { | |
77 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
62e85577 | 78 | pr_err("BTRFS: buffer leak start %llu len %lu refs %d\n", |
c1c9ff7c | 79 | eb->start, eb->len, atomic_read(&eb->refs)); |
6d49ba1b ES |
80 | list_del(&eb->leak_list); |
81 | kmem_cache_free(extent_buffer_cache, eb); | |
82 | } | |
83 | } | |
8d599ae1 | 84 | |
a5dee37d JB |
85 | #define btrfs_debug_check_extent_io_range(tree, start, end) \ |
86 | __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end)) | |
8d599ae1 | 87 | static inline void __btrfs_debug_check_extent_io_range(const char *caller, |
a5dee37d | 88 | struct extent_io_tree *tree, u64 start, u64 end) |
8d599ae1 | 89 | { |
c6100a4b JB |
90 | if (tree->ops && tree->ops->check_extent_io_range) |
91 | tree->ops->check_extent_io_range(tree->private_data, caller, | |
92 | start, end); | |
8d599ae1 | 93 | } |
6d49ba1b ES |
94 | #else |
95 | #define btrfs_leak_debug_add(new, head) do {} while (0) | |
96 | #define btrfs_leak_debug_del(entry) do {} while (0) | |
97 | #define btrfs_leak_debug_check() do {} while (0) | |
8d599ae1 | 98 | #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0) |
4bef0848 | 99 | #endif |
d1310b2e | 100 | |
d1310b2e CM |
101 | #define BUFFER_LRU_MAX 64 |
102 | ||
103 | struct tree_entry { | |
104 | u64 start; | |
105 | u64 end; | |
d1310b2e CM |
106 | struct rb_node rb_node; |
107 | }; | |
108 | ||
109 | struct extent_page_data { | |
110 | struct bio *bio; | |
111 | struct extent_io_tree *tree; | |
771ed689 CM |
112 | /* tells writepage not to lock the state bits for this range |
113 | * it still does the unlocking | |
114 | */ | |
ffbd517d CM |
115 | unsigned int extent_locked:1; |
116 | ||
70fd7614 | 117 | /* tells the submit_bio code to use REQ_SYNC */ |
ffbd517d | 118 | unsigned int sync_io:1; |
d1310b2e CM |
119 | }; |
120 | ||
d38ed27f QW |
121 | static void add_extent_changeset(struct extent_state *state, unsigned bits, |
122 | struct extent_changeset *changeset, | |
123 | int set) | |
124 | { | |
125 | int ret; | |
126 | ||
127 | if (!changeset) | |
128 | return; | |
129 | if (set && (state->state & bits) == bits) | |
130 | return; | |
fefdc557 QW |
131 | if (!set && (state->state & bits) == 0) |
132 | return; | |
d38ed27f | 133 | changeset->bytes_changed += state->end - state->start + 1; |
53d32359 | 134 | ret = ulist_add(&changeset->range_changed, state->start, state->end, |
d38ed27f QW |
135 | GFP_ATOMIC); |
136 | /* ENOMEM */ | |
137 | BUG_ON(ret < 0); | |
138 | } | |
139 | ||
0b32f4bb | 140 | static noinline void flush_write_bio(void *data); |
c2d904e0 JM |
141 | static inline struct btrfs_fs_info * |
142 | tree_fs_info(struct extent_io_tree *tree) | |
143 | { | |
c6100a4b JB |
144 | if (tree->ops) |
145 | return tree->ops->tree_fs_info(tree->private_data); | |
146 | return NULL; | |
c2d904e0 | 147 | } |
0b32f4bb | 148 | |
d1310b2e CM |
149 | int __init extent_io_init(void) |
150 | { | |
837e1972 | 151 | extent_state_cache = kmem_cache_create("btrfs_extent_state", |
9601e3f6 | 152 | sizeof(struct extent_state), 0, |
fba4b697 | 153 | SLAB_MEM_SPREAD, NULL); |
d1310b2e CM |
154 | if (!extent_state_cache) |
155 | return -ENOMEM; | |
156 | ||
837e1972 | 157 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 | 158 | sizeof(struct extent_buffer), 0, |
fba4b697 | 159 | SLAB_MEM_SPREAD, NULL); |
d1310b2e CM |
160 | if (!extent_buffer_cache) |
161 | goto free_state_cache; | |
9be3395b CM |
162 | |
163 | btrfs_bioset = bioset_create(BIO_POOL_SIZE, | |
011067b0 N |
164 | offsetof(struct btrfs_io_bio, bio), |
165 | BIOSET_NEED_BVECS); | |
9be3395b CM |
166 | if (!btrfs_bioset) |
167 | goto free_buffer_cache; | |
b208c2f7 DW |
168 | |
169 | if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE)) | |
170 | goto free_bioset; | |
171 | ||
d1310b2e CM |
172 | return 0; |
173 | ||
b208c2f7 DW |
174 | free_bioset: |
175 | bioset_free(btrfs_bioset); | |
176 | btrfs_bioset = NULL; | |
177 | ||
9be3395b CM |
178 | free_buffer_cache: |
179 | kmem_cache_destroy(extent_buffer_cache); | |
180 | extent_buffer_cache = NULL; | |
181 | ||
d1310b2e CM |
182 | free_state_cache: |
183 | kmem_cache_destroy(extent_state_cache); | |
9be3395b | 184 | extent_state_cache = NULL; |
d1310b2e CM |
185 | return -ENOMEM; |
186 | } | |
187 | ||
188 | void extent_io_exit(void) | |
189 | { | |
6d49ba1b | 190 | btrfs_leak_debug_check(); |
8c0a8537 KS |
191 | |
192 | /* | |
193 | * Make sure all delayed rcu free are flushed before we | |
194 | * destroy caches. | |
195 | */ | |
196 | rcu_barrier(); | |
5598e900 KM |
197 | kmem_cache_destroy(extent_state_cache); |
198 | kmem_cache_destroy(extent_buffer_cache); | |
9be3395b CM |
199 | if (btrfs_bioset) |
200 | bioset_free(btrfs_bioset); | |
d1310b2e CM |
201 | } |
202 | ||
203 | void extent_io_tree_init(struct extent_io_tree *tree, | |
c6100a4b | 204 | void *private_data) |
d1310b2e | 205 | { |
6bef4d31 | 206 | tree->state = RB_ROOT; |
d1310b2e CM |
207 | tree->ops = NULL; |
208 | tree->dirty_bytes = 0; | |
70dec807 | 209 | spin_lock_init(&tree->lock); |
c6100a4b | 210 | tree->private_data = private_data; |
d1310b2e | 211 | } |
d1310b2e | 212 | |
b2950863 | 213 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
214 | { |
215 | struct extent_state *state; | |
d1310b2e | 216 | |
3ba7ab22 MH |
217 | /* |
218 | * The given mask might be not appropriate for the slab allocator, | |
219 | * drop the unsupported bits | |
220 | */ | |
221 | mask &= ~(__GFP_DMA32|__GFP_HIGHMEM); | |
d1310b2e | 222 | state = kmem_cache_alloc(extent_state_cache, mask); |
2b114d1d | 223 | if (!state) |
d1310b2e CM |
224 | return state; |
225 | state->state = 0; | |
47dc196a | 226 | state->failrec = NULL; |
27a3507d | 227 | RB_CLEAR_NODE(&state->rb_node); |
6d49ba1b | 228 | btrfs_leak_debug_add(&state->leak_list, &states); |
b7ac31b7 | 229 | refcount_set(&state->refs, 1); |
d1310b2e | 230 | init_waitqueue_head(&state->wq); |
143bede5 | 231 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
232 | return state; |
233 | } | |
d1310b2e | 234 | |
4845e44f | 235 | void free_extent_state(struct extent_state *state) |
d1310b2e | 236 | { |
d1310b2e CM |
237 | if (!state) |
238 | return; | |
b7ac31b7 | 239 | if (refcount_dec_and_test(&state->refs)) { |
27a3507d | 240 | WARN_ON(extent_state_in_tree(state)); |
6d49ba1b | 241 | btrfs_leak_debug_del(&state->leak_list); |
143bede5 | 242 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
243 | kmem_cache_free(extent_state_cache, state); |
244 | } | |
245 | } | |
d1310b2e | 246 | |
f2071b21 FM |
247 | static struct rb_node *tree_insert(struct rb_root *root, |
248 | struct rb_node *search_start, | |
249 | u64 offset, | |
12cfbad9 FDBM |
250 | struct rb_node *node, |
251 | struct rb_node ***p_in, | |
252 | struct rb_node **parent_in) | |
d1310b2e | 253 | { |
f2071b21 | 254 | struct rb_node **p; |
d397712b | 255 | struct rb_node *parent = NULL; |
d1310b2e CM |
256 | struct tree_entry *entry; |
257 | ||
12cfbad9 FDBM |
258 | if (p_in && parent_in) { |
259 | p = *p_in; | |
260 | parent = *parent_in; | |
261 | goto do_insert; | |
262 | } | |
263 | ||
f2071b21 | 264 | p = search_start ? &search_start : &root->rb_node; |
d397712b | 265 | while (*p) { |
d1310b2e CM |
266 | parent = *p; |
267 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
268 | ||
269 | if (offset < entry->start) | |
270 | p = &(*p)->rb_left; | |
271 | else if (offset > entry->end) | |
272 | p = &(*p)->rb_right; | |
273 | else | |
274 | return parent; | |
275 | } | |
276 | ||
12cfbad9 | 277 | do_insert: |
d1310b2e CM |
278 | rb_link_node(node, parent, p); |
279 | rb_insert_color(node, root); | |
280 | return NULL; | |
281 | } | |
282 | ||
80ea96b1 | 283 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
12cfbad9 FDBM |
284 | struct rb_node **prev_ret, |
285 | struct rb_node **next_ret, | |
286 | struct rb_node ***p_ret, | |
287 | struct rb_node **parent_ret) | |
d1310b2e | 288 | { |
80ea96b1 | 289 | struct rb_root *root = &tree->state; |
12cfbad9 | 290 | struct rb_node **n = &root->rb_node; |
d1310b2e CM |
291 | struct rb_node *prev = NULL; |
292 | struct rb_node *orig_prev = NULL; | |
293 | struct tree_entry *entry; | |
294 | struct tree_entry *prev_entry = NULL; | |
295 | ||
12cfbad9 FDBM |
296 | while (*n) { |
297 | prev = *n; | |
298 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
d1310b2e CM |
299 | prev_entry = entry; |
300 | ||
301 | if (offset < entry->start) | |
12cfbad9 | 302 | n = &(*n)->rb_left; |
d1310b2e | 303 | else if (offset > entry->end) |
12cfbad9 | 304 | n = &(*n)->rb_right; |
d397712b | 305 | else |
12cfbad9 | 306 | return *n; |
d1310b2e CM |
307 | } |
308 | ||
12cfbad9 FDBM |
309 | if (p_ret) |
310 | *p_ret = n; | |
311 | if (parent_ret) | |
312 | *parent_ret = prev; | |
313 | ||
d1310b2e CM |
314 | if (prev_ret) { |
315 | orig_prev = prev; | |
d397712b | 316 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
317 | prev = rb_next(prev); |
318 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
319 | } | |
320 | *prev_ret = prev; | |
321 | prev = orig_prev; | |
322 | } | |
323 | ||
324 | if (next_ret) { | |
325 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 326 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
327 | prev = rb_prev(prev); |
328 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
329 | } | |
330 | *next_ret = prev; | |
331 | } | |
332 | return NULL; | |
333 | } | |
334 | ||
12cfbad9 FDBM |
335 | static inline struct rb_node * |
336 | tree_search_for_insert(struct extent_io_tree *tree, | |
337 | u64 offset, | |
338 | struct rb_node ***p_ret, | |
339 | struct rb_node **parent_ret) | |
d1310b2e | 340 | { |
70dec807 | 341 | struct rb_node *prev = NULL; |
d1310b2e | 342 | struct rb_node *ret; |
70dec807 | 343 | |
12cfbad9 | 344 | ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret); |
d397712b | 345 | if (!ret) |
d1310b2e CM |
346 | return prev; |
347 | return ret; | |
348 | } | |
349 | ||
12cfbad9 FDBM |
350 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
351 | u64 offset) | |
352 | { | |
353 | return tree_search_for_insert(tree, offset, NULL, NULL); | |
354 | } | |
355 | ||
9ed74f2d JB |
356 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, |
357 | struct extent_state *other) | |
358 | { | |
359 | if (tree->ops && tree->ops->merge_extent_hook) | |
c6100a4b | 360 | tree->ops->merge_extent_hook(tree->private_data, new, other); |
9ed74f2d JB |
361 | } |
362 | ||
d1310b2e CM |
363 | /* |
364 | * utility function to look for merge candidates inside a given range. | |
365 | * Any extents with matching state are merged together into a single | |
366 | * extent in the tree. Extents with EXTENT_IO in their state field | |
367 | * are not merged because the end_io handlers need to be able to do | |
368 | * operations on them without sleeping (or doing allocations/splits). | |
369 | * | |
370 | * This should be called with the tree lock held. | |
371 | */ | |
1bf85046 JM |
372 | static void merge_state(struct extent_io_tree *tree, |
373 | struct extent_state *state) | |
d1310b2e CM |
374 | { |
375 | struct extent_state *other; | |
376 | struct rb_node *other_node; | |
377 | ||
5b21f2ed | 378 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
1bf85046 | 379 | return; |
d1310b2e CM |
380 | |
381 | other_node = rb_prev(&state->rb_node); | |
382 | if (other_node) { | |
383 | other = rb_entry(other_node, struct extent_state, rb_node); | |
384 | if (other->end == state->start - 1 && | |
385 | other->state == state->state) { | |
9ed74f2d | 386 | merge_cb(tree, state, other); |
d1310b2e | 387 | state->start = other->start; |
d1310b2e | 388 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 389 | RB_CLEAR_NODE(&other->rb_node); |
d1310b2e CM |
390 | free_extent_state(other); |
391 | } | |
392 | } | |
393 | other_node = rb_next(&state->rb_node); | |
394 | if (other_node) { | |
395 | other = rb_entry(other_node, struct extent_state, rb_node); | |
396 | if (other->start == state->end + 1 && | |
397 | other->state == state->state) { | |
9ed74f2d | 398 | merge_cb(tree, state, other); |
df98b6e2 | 399 | state->end = other->end; |
df98b6e2 | 400 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 401 | RB_CLEAR_NODE(&other->rb_node); |
df98b6e2 | 402 | free_extent_state(other); |
d1310b2e CM |
403 | } |
404 | } | |
d1310b2e CM |
405 | } |
406 | ||
1bf85046 | 407 | static void set_state_cb(struct extent_io_tree *tree, |
9ee49a04 | 408 | struct extent_state *state, unsigned *bits) |
291d673e | 409 | { |
1bf85046 | 410 | if (tree->ops && tree->ops->set_bit_hook) |
c6100a4b | 411 | tree->ops->set_bit_hook(tree->private_data, state, bits); |
291d673e CM |
412 | } |
413 | ||
414 | static void clear_state_cb(struct extent_io_tree *tree, | |
9ee49a04 | 415 | struct extent_state *state, unsigned *bits) |
291d673e | 416 | { |
9ed74f2d | 417 | if (tree->ops && tree->ops->clear_bit_hook) |
c6100a4b | 418 | tree->ops->clear_bit_hook(tree->private_data, state, bits); |
291d673e CM |
419 | } |
420 | ||
3150b699 | 421 | static void set_state_bits(struct extent_io_tree *tree, |
d38ed27f QW |
422 | struct extent_state *state, unsigned *bits, |
423 | struct extent_changeset *changeset); | |
3150b699 | 424 | |
d1310b2e CM |
425 | /* |
426 | * insert an extent_state struct into the tree. 'bits' are set on the | |
427 | * struct before it is inserted. | |
428 | * | |
429 | * This may return -EEXIST if the extent is already there, in which case the | |
430 | * state struct is freed. | |
431 | * | |
432 | * The tree lock is not taken internally. This is a utility function and | |
433 | * probably isn't what you want to call (see set/clear_extent_bit). | |
434 | */ | |
435 | static int insert_state(struct extent_io_tree *tree, | |
436 | struct extent_state *state, u64 start, u64 end, | |
12cfbad9 FDBM |
437 | struct rb_node ***p, |
438 | struct rb_node **parent, | |
d38ed27f | 439 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e CM |
440 | { |
441 | struct rb_node *node; | |
442 | ||
31b1a2bd | 443 | if (end < start) |
efe120a0 | 444 | WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n", |
c1c9ff7c | 445 | end, start); |
d1310b2e CM |
446 | state->start = start; |
447 | state->end = end; | |
9ed74f2d | 448 | |
d38ed27f | 449 | set_state_bits(tree, state, bits, changeset); |
3150b699 | 450 | |
f2071b21 | 451 | node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent); |
d1310b2e CM |
452 | if (node) { |
453 | struct extent_state *found; | |
454 | found = rb_entry(node, struct extent_state, rb_node); | |
62e85577 | 455 | pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n", |
c1c9ff7c | 456 | found->start, found->end, start, end); |
d1310b2e CM |
457 | return -EEXIST; |
458 | } | |
459 | merge_state(tree, state); | |
460 | return 0; | |
461 | } | |
462 | ||
1bf85046 | 463 | static void split_cb(struct extent_io_tree *tree, struct extent_state *orig, |
9ed74f2d JB |
464 | u64 split) |
465 | { | |
466 | if (tree->ops && tree->ops->split_extent_hook) | |
c6100a4b | 467 | tree->ops->split_extent_hook(tree->private_data, orig, split); |
9ed74f2d JB |
468 | } |
469 | ||
d1310b2e CM |
470 | /* |
471 | * split a given extent state struct in two, inserting the preallocated | |
472 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
473 | * offset inside 'orig' where it should be split. | |
474 | * | |
475 | * Before calling, | |
476 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
477 | * are two extent state structs in the tree: | |
478 | * prealloc: [orig->start, split - 1] | |
479 | * orig: [ split, orig->end ] | |
480 | * | |
481 | * The tree locks are not taken by this function. They need to be held | |
482 | * by the caller. | |
483 | */ | |
484 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
485 | struct extent_state *prealloc, u64 split) | |
486 | { | |
487 | struct rb_node *node; | |
9ed74f2d JB |
488 | |
489 | split_cb(tree, orig, split); | |
490 | ||
d1310b2e CM |
491 | prealloc->start = orig->start; |
492 | prealloc->end = split - 1; | |
493 | prealloc->state = orig->state; | |
494 | orig->start = split; | |
495 | ||
f2071b21 FM |
496 | node = tree_insert(&tree->state, &orig->rb_node, prealloc->end, |
497 | &prealloc->rb_node, NULL, NULL); | |
d1310b2e | 498 | if (node) { |
d1310b2e CM |
499 | free_extent_state(prealloc); |
500 | return -EEXIST; | |
501 | } | |
502 | return 0; | |
503 | } | |
504 | ||
cdc6a395 LZ |
505 | static struct extent_state *next_state(struct extent_state *state) |
506 | { | |
507 | struct rb_node *next = rb_next(&state->rb_node); | |
508 | if (next) | |
509 | return rb_entry(next, struct extent_state, rb_node); | |
510 | else | |
511 | return NULL; | |
512 | } | |
513 | ||
d1310b2e CM |
514 | /* |
515 | * utility function to clear some bits in an extent state struct. | |
1b303fc0 | 516 | * it will optionally wake up any one waiting on this state (wake == 1). |
d1310b2e CM |
517 | * |
518 | * If no bits are set on the state struct after clearing things, the | |
519 | * struct is freed and removed from the tree | |
520 | */ | |
cdc6a395 LZ |
521 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
522 | struct extent_state *state, | |
fefdc557 QW |
523 | unsigned *bits, int wake, |
524 | struct extent_changeset *changeset) | |
d1310b2e | 525 | { |
cdc6a395 | 526 | struct extent_state *next; |
9ee49a04 | 527 | unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS; |
d1310b2e | 528 | |
0ca1f7ce | 529 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
530 | u64 range = state->end - state->start + 1; |
531 | WARN_ON(range > tree->dirty_bytes); | |
532 | tree->dirty_bytes -= range; | |
533 | } | |
291d673e | 534 | clear_state_cb(tree, state, bits); |
fefdc557 | 535 | add_extent_changeset(state, bits_to_clear, changeset, 0); |
32c00aff | 536 | state->state &= ~bits_to_clear; |
d1310b2e CM |
537 | if (wake) |
538 | wake_up(&state->wq); | |
0ca1f7ce | 539 | if (state->state == 0) { |
cdc6a395 | 540 | next = next_state(state); |
27a3507d | 541 | if (extent_state_in_tree(state)) { |
d1310b2e | 542 | rb_erase(&state->rb_node, &tree->state); |
27a3507d | 543 | RB_CLEAR_NODE(&state->rb_node); |
d1310b2e CM |
544 | free_extent_state(state); |
545 | } else { | |
546 | WARN_ON(1); | |
547 | } | |
548 | } else { | |
549 | merge_state(tree, state); | |
cdc6a395 | 550 | next = next_state(state); |
d1310b2e | 551 | } |
cdc6a395 | 552 | return next; |
d1310b2e CM |
553 | } |
554 | ||
8233767a XG |
555 | static struct extent_state * |
556 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
557 | { | |
558 | if (!prealloc) | |
559 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
560 | ||
561 | return prealloc; | |
562 | } | |
563 | ||
48a3b636 | 564 | static void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
c2d904e0 | 565 | { |
5d163e0e JM |
566 | btrfs_panic(tree_fs_info(tree), err, |
567 | "Locking error: Extent tree was modified by another thread while locked."); | |
c2d904e0 JM |
568 | } |
569 | ||
d1310b2e CM |
570 | /* |
571 | * clear some bits on a range in the tree. This may require splitting | |
572 | * or inserting elements in the tree, so the gfp mask is used to | |
573 | * indicate which allocations or sleeping are allowed. | |
574 | * | |
575 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
576 | * the given range from the tree regardless of state (ie for truncate). | |
577 | * | |
578 | * the range [start, end] is inclusive. | |
579 | * | |
6763af84 | 580 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e | 581 | */ |
66b0c887 | 582 | int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
fefdc557 QW |
583 | unsigned bits, int wake, int delete, |
584 | struct extent_state **cached_state, | |
585 | gfp_t mask, struct extent_changeset *changeset) | |
d1310b2e CM |
586 | { |
587 | struct extent_state *state; | |
2c64c53d | 588 | struct extent_state *cached; |
d1310b2e CM |
589 | struct extent_state *prealloc = NULL; |
590 | struct rb_node *node; | |
5c939df5 | 591 | u64 last_end; |
d1310b2e | 592 | int err; |
2ac55d41 | 593 | int clear = 0; |
d1310b2e | 594 | |
a5dee37d | 595 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 596 | |
7ee9e440 JB |
597 | if (bits & EXTENT_DELALLOC) |
598 | bits |= EXTENT_NORESERVE; | |
599 | ||
0ca1f7ce YZ |
600 | if (delete) |
601 | bits |= ~EXTENT_CTLBITS; | |
602 | bits |= EXTENT_FIRST_DELALLOC; | |
603 | ||
2ac55d41 JB |
604 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
605 | clear = 1; | |
d1310b2e | 606 | again: |
d0164adc | 607 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
c7bc6319 FM |
608 | /* |
609 | * Don't care for allocation failure here because we might end | |
610 | * up not needing the pre-allocated extent state at all, which | |
611 | * is the case if we only have in the tree extent states that | |
612 | * cover our input range and don't cover too any other range. | |
613 | * If we end up needing a new extent state we allocate it later. | |
614 | */ | |
d1310b2e | 615 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
616 | } |
617 | ||
cad321ad | 618 | spin_lock(&tree->lock); |
2c64c53d CM |
619 | if (cached_state) { |
620 | cached = *cached_state; | |
2ac55d41 JB |
621 | |
622 | if (clear) { | |
623 | *cached_state = NULL; | |
624 | cached_state = NULL; | |
625 | } | |
626 | ||
27a3507d FM |
627 | if (cached && extent_state_in_tree(cached) && |
628 | cached->start <= start && cached->end > start) { | |
2ac55d41 | 629 | if (clear) |
b7ac31b7 | 630 | refcount_dec(&cached->refs); |
2c64c53d | 631 | state = cached; |
42daec29 | 632 | goto hit_next; |
2c64c53d | 633 | } |
2ac55d41 JB |
634 | if (clear) |
635 | free_extent_state(cached); | |
2c64c53d | 636 | } |
d1310b2e CM |
637 | /* |
638 | * this search will find the extents that end after | |
639 | * our range starts | |
640 | */ | |
80ea96b1 | 641 | node = tree_search(tree, start); |
d1310b2e CM |
642 | if (!node) |
643 | goto out; | |
644 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 645 | hit_next: |
d1310b2e CM |
646 | if (state->start > end) |
647 | goto out; | |
648 | WARN_ON(state->end < start); | |
5c939df5 | 649 | last_end = state->end; |
d1310b2e | 650 | |
0449314a | 651 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
652 | if (!(state->state & bits)) { |
653 | state = next_state(state); | |
0449314a | 654 | goto next; |
cdc6a395 | 655 | } |
0449314a | 656 | |
d1310b2e CM |
657 | /* |
658 | * | ---- desired range ---- | | |
659 | * | state | or | |
660 | * | ------------- state -------------- | | |
661 | * | |
662 | * We need to split the extent we found, and may flip | |
663 | * bits on second half. | |
664 | * | |
665 | * If the extent we found extends past our range, we | |
666 | * just split and search again. It'll get split again | |
667 | * the next time though. | |
668 | * | |
669 | * If the extent we found is inside our range, we clear | |
670 | * the desired bit on it. | |
671 | */ | |
672 | ||
673 | if (state->start < start) { | |
8233767a XG |
674 | prealloc = alloc_extent_state_atomic(prealloc); |
675 | BUG_ON(!prealloc); | |
d1310b2e | 676 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
677 | if (err) |
678 | extent_io_tree_panic(tree, err); | |
679 | ||
d1310b2e CM |
680 | prealloc = NULL; |
681 | if (err) | |
682 | goto out; | |
683 | if (state->end <= end) { | |
fefdc557 QW |
684 | state = clear_state_bit(tree, state, &bits, wake, |
685 | changeset); | |
d1ac6e41 | 686 | goto next; |
d1310b2e CM |
687 | } |
688 | goto search_again; | |
689 | } | |
690 | /* | |
691 | * | ---- desired range ---- | | |
692 | * | state | | |
693 | * We need to split the extent, and clear the bit | |
694 | * on the first half | |
695 | */ | |
696 | if (state->start <= end && state->end > end) { | |
8233767a XG |
697 | prealloc = alloc_extent_state_atomic(prealloc); |
698 | BUG_ON(!prealloc); | |
d1310b2e | 699 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
700 | if (err) |
701 | extent_io_tree_panic(tree, err); | |
702 | ||
d1310b2e CM |
703 | if (wake) |
704 | wake_up(&state->wq); | |
42daec29 | 705 | |
fefdc557 | 706 | clear_state_bit(tree, prealloc, &bits, wake, changeset); |
9ed74f2d | 707 | |
d1310b2e CM |
708 | prealloc = NULL; |
709 | goto out; | |
710 | } | |
42daec29 | 711 | |
fefdc557 | 712 | state = clear_state_bit(tree, state, &bits, wake, changeset); |
0449314a | 713 | next: |
5c939df5 YZ |
714 | if (last_end == (u64)-1) |
715 | goto out; | |
716 | start = last_end + 1; | |
cdc6a395 | 717 | if (start <= end && state && !need_resched()) |
692e5759 | 718 | goto hit_next; |
d1310b2e CM |
719 | |
720 | search_again: | |
721 | if (start > end) | |
722 | goto out; | |
cad321ad | 723 | spin_unlock(&tree->lock); |
d0164adc | 724 | if (gfpflags_allow_blocking(mask)) |
d1310b2e CM |
725 | cond_resched(); |
726 | goto again; | |
7ab5cb2a DS |
727 | |
728 | out: | |
729 | spin_unlock(&tree->lock); | |
730 | if (prealloc) | |
731 | free_extent_state(prealloc); | |
732 | ||
733 | return 0; | |
734 | ||
d1310b2e | 735 | } |
d1310b2e | 736 | |
143bede5 JM |
737 | static void wait_on_state(struct extent_io_tree *tree, |
738 | struct extent_state *state) | |
641f5219 CH |
739 | __releases(tree->lock) |
740 | __acquires(tree->lock) | |
d1310b2e CM |
741 | { |
742 | DEFINE_WAIT(wait); | |
743 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 744 | spin_unlock(&tree->lock); |
d1310b2e | 745 | schedule(); |
cad321ad | 746 | spin_lock(&tree->lock); |
d1310b2e | 747 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
748 | } |
749 | ||
750 | /* | |
751 | * waits for one or more bits to clear on a range in the state tree. | |
752 | * The range [start, end] is inclusive. | |
753 | * The tree lock is taken by this function | |
754 | */ | |
41074888 DS |
755 | static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
756 | unsigned long bits) | |
d1310b2e CM |
757 | { |
758 | struct extent_state *state; | |
759 | struct rb_node *node; | |
760 | ||
a5dee37d | 761 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 762 | |
cad321ad | 763 | spin_lock(&tree->lock); |
d1310b2e CM |
764 | again: |
765 | while (1) { | |
766 | /* | |
767 | * this search will find all the extents that end after | |
768 | * our range starts | |
769 | */ | |
80ea96b1 | 770 | node = tree_search(tree, start); |
c50d3e71 | 771 | process_node: |
d1310b2e CM |
772 | if (!node) |
773 | break; | |
774 | ||
775 | state = rb_entry(node, struct extent_state, rb_node); | |
776 | ||
777 | if (state->start > end) | |
778 | goto out; | |
779 | ||
780 | if (state->state & bits) { | |
781 | start = state->start; | |
b7ac31b7 | 782 | refcount_inc(&state->refs); |
d1310b2e CM |
783 | wait_on_state(tree, state); |
784 | free_extent_state(state); | |
785 | goto again; | |
786 | } | |
787 | start = state->end + 1; | |
788 | ||
789 | if (start > end) | |
790 | break; | |
791 | ||
c50d3e71 FM |
792 | if (!cond_resched_lock(&tree->lock)) { |
793 | node = rb_next(node); | |
794 | goto process_node; | |
795 | } | |
d1310b2e CM |
796 | } |
797 | out: | |
cad321ad | 798 | spin_unlock(&tree->lock); |
d1310b2e | 799 | } |
d1310b2e | 800 | |
1bf85046 | 801 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 802 | struct extent_state *state, |
d38ed27f | 803 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e | 804 | { |
9ee49a04 | 805 | unsigned bits_to_set = *bits & ~EXTENT_CTLBITS; |
9ed74f2d | 806 | |
1bf85046 | 807 | set_state_cb(tree, state, bits); |
0ca1f7ce | 808 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
809 | u64 range = state->end - state->start + 1; |
810 | tree->dirty_bytes += range; | |
811 | } | |
d38ed27f | 812 | add_extent_changeset(state, bits_to_set, changeset, 1); |
0ca1f7ce | 813 | state->state |= bits_to_set; |
d1310b2e CM |
814 | } |
815 | ||
e38e2ed7 FM |
816 | static void cache_state_if_flags(struct extent_state *state, |
817 | struct extent_state **cached_ptr, | |
9ee49a04 | 818 | unsigned flags) |
2c64c53d CM |
819 | { |
820 | if (cached_ptr && !(*cached_ptr)) { | |
e38e2ed7 | 821 | if (!flags || (state->state & flags)) { |
2c64c53d | 822 | *cached_ptr = state; |
b7ac31b7 | 823 | refcount_inc(&state->refs); |
2c64c53d CM |
824 | } |
825 | } | |
826 | } | |
827 | ||
e38e2ed7 FM |
828 | static void cache_state(struct extent_state *state, |
829 | struct extent_state **cached_ptr) | |
830 | { | |
831 | return cache_state_if_flags(state, cached_ptr, | |
832 | EXTENT_IOBITS | EXTENT_BOUNDARY); | |
833 | } | |
834 | ||
d1310b2e | 835 | /* |
1edbb734 CM |
836 | * set some bits on a range in the tree. This may require allocations or |
837 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 838 | * |
1edbb734 CM |
839 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
840 | * part of the range already has the desired bits set. The start of the | |
841 | * existing range is returned in failed_start in this case. | |
d1310b2e | 842 | * |
1edbb734 | 843 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 844 | */ |
1edbb734 | 845 | |
3fbe5c02 JM |
846 | static int __must_check |
847 | __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 848 | unsigned bits, unsigned exclusive_bits, |
41074888 | 849 | u64 *failed_start, struct extent_state **cached_state, |
d38ed27f | 850 | gfp_t mask, struct extent_changeset *changeset) |
d1310b2e CM |
851 | { |
852 | struct extent_state *state; | |
853 | struct extent_state *prealloc = NULL; | |
854 | struct rb_node *node; | |
12cfbad9 FDBM |
855 | struct rb_node **p; |
856 | struct rb_node *parent; | |
d1310b2e | 857 | int err = 0; |
d1310b2e CM |
858 | u64 last_start; |
859 | u64 last_end; | |
42daec29 | 860 | |
a5dee37d | 861 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 862 | |
0ca1f7ce | 863 | bits |= EXTENT_FIRST_DELALLOC; |
d1310b2e | 864 | again: |
d0164adc | 865 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
059f791c DS |
866 | /* |
867 | * Don't care for allocation failure here because we might end | |
868 | * up not needing the pre-allocated extent state at all, which | |
869 | * is the case if we only have in the tree extent states that | |
870 | * cover our input range and don't cover too any other range. | |
871 | * If we end up needing a new extent state we allocate it later. | |
872 | */ | |
d1310b2e | 873 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
874 | } |
875 | ||
cad321ad | 876 | spin_lock(&tree->lock); |
9655d298 CM |
877 | if (cached_state && *cached_state) { |
878 | state = *cached_state; | |
df98b6e2 | 879 | if (state->start <= start && state->end > start && |
27a3507d | 880 | extent_state_in_tree(state)) { |
9655d298 CM |
881 | node = &state->rb_node; |
882 | goto hit_next; | |
883 | } | |
884 | } | |
d1310b2e CM |
885 | /* |
886 | * this search will find all the extents that end after | |
887 | * our range starts. | |
888 | */ | |
12cfbad9 | 889 | node = tree_search_for_insert(tree, start, &p, &parent); |
d1310b2e | 890 | if (!node) { |
8233767a XG |
891 | prealloc = alloc_extent_state_atomic(prealloc); |
892 | BUG_ON(!prealloc); | |
12cfbad9 | 893 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 894 | &p, &parent, &bits, changeset); |
c2d904e0 JM |
895 | if (err) |
896 | extent_io_tree_panic(tree, err); | |
897 | ||
c42ac0bc | 898 | cache_state(prealloc, cached_state); |
d1310b2e | 899 | prealloc = NULL; |
d1310b2e CM |
900 | goto out; |
901 | } | |
d1310b2e | 902 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 903 | hit_next: |
d1310b2e CM |
904 | last_start = state->start; |
905 | last_end = state->end; | |
906 | ||
907 | /* | |
908 | * | ---- desired range ---- | | |
909 | * | state | | |
910 | * | |
911 | * Just lock what we found and keep going | |
912 | */ | |
913 | if (state->start == start && state->end <= end) { | |
1edbb734 | 914 | if (state->state & exclusive_bits) { |
d1310b2e CM |
915 | *failed_start = state->start; |
916 | err = -EEXIST; | |
917 | goto out; | |
918 | } | |
42daec29 | 919 | |
d38ed27f | 920 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 921 | cache_state(state, cached_state); |
d1310b2e | 922 | merge_state(tree, state); |
5c939df5 YZ |
923 | if (last_end == (u64)-1) |
924 | goto out; | |
925 | start = last_end + 1; | |
d1ac6e41 LB |
926 | state = next_state(state); |
927 | if (start < end && state && state->start == start && | |
928 | !need_resched()) | |
929 | goto hit_next; | |
d1310b2e CM |
930 | goto search_again; |
931 | } | |
932 | ||
933 | /* | |
934 | * | ---- desired range ---- | | |
935 | * | state | | |
936 | * or | |
937 | * | ------------- state -------------- | | |
938 | * | |
939 | * We need to split the extent we found, and may flip bits on | |
940 | * second half. | |
941 | * | |
942 | * If the extent we found extends past our | |
943 | * range, we just split and search again. It'll get split | |
944 | * again the next time though. | |
945 | * | |
946 | * If the extent we found is inside our range, we set the | |
947 | * desired bit on it. | |
948 | */ | |
949 | if (state->start < start) { | |
1edbb734 | 950 | if (state->state & exclusive_bits) { |
d1310b2e CM |
951 | *failed_start = start; |
952 | err = -EEXIST; | |
953 | goto out; | |
954 | } | |
8233767a XG |
955 | |
956 | prealloc = alloc_extent_state_atomic(prealloc); | |
957 | BUG_ON(!prealloc); | |
d1310b2e | 958 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
959 | if (err) |
960 | extent_io_tree_panic(tree, err); | |
961 | ||
d1310b2e CM |
962 | prealloc = NULL; |
963 | if (err) | |
964 | goto out; | |
965 | if (state->end <= end) { | |
d38ed27f | 966 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 967 | cache_state(state, cached_state); |
d1310b2e | 968 | merge_state(tree, state); |
5c939df5 YZ |
969 | if (last_end == (u64)-1) |
970 | goto out; | |
971 | start = last_end + 1; | |
d1ac6e41 LB |
972 | state = next_state(state); |
973 | if (start < end && state && state->start == start && | |
974 | !need_resched()) | |
975 | goto hit_next; | |
d1310b2e CM |
976 | } |
977 | goto search_again; | |
978 | } | |
979 | /* | |
980 | * | ---- desired range ---- | | |
981 | * | state | or | state | | |
982 | * | |
983 | * There's a hole, we need to insert something in it and | |
984 | * ignore the extent we found. | |
985 | */ | |
986 | if (state->start > start) { | |
987 | u64 this_end; | |
988 | if (end < last_start) | |
989 | this_end = end; | |
990 | else | |
d397712b | 991 | this_end = last_start - 1; |
8233767a XG |
992 | |
993 | prealloc = alloc_extent_state_atomic(prealloc); | |
994 | BUG_ON(!prealloc); | |
c7f895a2 XG |
995 | |
996 | /* | |
997 | * Avoid to free 'prealloc' if it can be merged with | |
998 | * the later extent. | |
999 | */ | |
d1310b2e | 1000 | err = insert_state(tree, prealloc, start, this_end, |
d38ed27f | 1001 | NULL, NULL, &bits, changeset); |
c2d904e0 JM |
1002 | if (err) |
1003 | extent_io_tree_panic(tree, err); | |
1004 | ||
9ed74f2d JB |
1005 | cache_state(prealloc, cached_state); |
1006 | prealloc = NULL; | |
d1310b2e CM |
1007 | start = this_end + 1; |
1008 | goto search_again; | |
1009 | } | |
1010 | /* | |
1011 | * | ---- desired range ---- | | |
1012 | * | state | | |
1013 | * We need to split the extent, and set the bit | |
1014 | * on the first half | |
1015 | */ | |
1016 | if (state->start <= end && state->end > end) { | |
1edbb734 | 1017 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1018 | *failed_start = start; |
1019 | err = -EEXIST; | |
1020 | goto out; | |
1021 | } | |
8233767a XG |
1022 | |
1023 | prealloc = alloc_extent_state_atomic(prealloc); | |
1024 | BUG_ON(!prealloc); | |
d1310b2e | 1025 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
1026 | if (err) |
1027 | extent_io_tree_panic(tree, err); | |
d1310b2e | 1028 | |
d38ed27f | 1029 | set_state_bits(tree, prealloc, &bits, changeset); |
2c64c53d | 1030 | cache_state(prealloc, cached_state); |
d1310b2e CM |
1031 | merge_state(tree, prealloc); |
1032 | prealloc = NULL; | |
1033 | goto out; | |
1034 | } | |
1035 | ||
b5a4ba14 DS |
1036 | search_again: |
1037 | if (start > end) | |
1038 | goto out; | |
1039 | spin_unlock(&tree->lock); | |
1040 | if (gfpflags_allow_blocking(mask)) | |
1041 | cond_resched(); | |
1042 | goto again; | |
d1310b2e CM |
1043 | |
1044 | out: | |
cad321ad | 1045 | spin_unlock(&tree->lock); |
d1310b2e CM |
1046 | if (prealloc) |
1047 | free_extent_state(prealloc); | |
1048 | ||
1049 | return err; | |
1050 | ||
d1310b2e | 1051 | } |
d1310b2e | 1052 | |
41074888 | 1053 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
9ee49a04 | 1054 | unsigned bits, u64 * failed_start, |
41074888 | 1055 | struct extent_state **cached_state, gfp_t mask) |
3fbe5c02 JM |
1056 | { |
1057 | return __set_extent_bit(tree, start, end, bits, 0, failed_start, | |
d38ed27f | 1058 | cached_state, mask, NULL); |
3fbe5c02 JM |
1059 | } |
1060 | ||
1061 | ||
462d6fac | 1062 | /** |
10983f2e LB |
1063 | * convert_extent_bit - convert all bits in a given range from one bit to |
1064 | * another | |
462d6fac JB |
1065 | * @tree: the io tree to search |
1066 | * @start: the start offset in bytes | |
1067 | * @end: the end offset in bytes (inclusive) | |
1068 | * @bits: the bits to set in this range | |
1069 | * @clear_bits: the bits to clear in this range | |
e6138876 | 1070 | * @cached_state: state that we're going to cache |
462d6fac JB |
1071 | * |
1072 | * This will go through and set bits for the given range. If any states exist | |
1073 | * already in this range they are set with the given bit and cleared of the | |
1074 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
1075 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
1076 | * boundary bits like LOCK. | |
210aa277 DS |
1077 | * |
1078 | * All allocations are done with GFP_NOFS. | |
462d6fac JB |
1079 | */ |
1080 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1081 | unsigned bits, unsigned clear_bits, |
210aa277 | 1082 | struct extent_state **cached_state) |
462d6fac JB |
1083 | { |
1084 | struct extent_state *state; | |
1085 | struct extent_state *prealloc = NULL; | |
1086 | struct rb_node *node; | |
12cfbad9 FDBM |
1087 | struct rb_node **p; |
1088 | struct rb_node *parent; | |
462d6fac JB |
1089 | int err = 0; |
1090 | u64 last_start; | |
1091 | u64 last_end; | |
c8fd3de7 | 1092 | bool first_iteration = true; |
462d6fac | 1093 | |
a5dee37d | 1094 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 1095 | |
462d6fac | 1096 | again: |
210aa277 | 1097 | if (!prealloc) { |
c8fd3de7 FM |
1098 | /* |
1099 | * Best effort, don't worry if extent state allocation fails | |
1100 | * here for the first iteration. We might have a cached state | |
1101 | * that matches exactly the target range, in which case no | |
1102 | * extent state allocations are needed. We'll only know this | |
1103 | * after locking the tree. | |
1104 | */ | |
210aa277 | 1105 | prealloc = alloc_extent_state(GFP_NOFS); |
c8fd3de7 | 1106 | if (!prealloc && !first_iteration) |
462d6fac JB |
1107 | return -ENOMEM; |
1108 | } | |
1109 | ||
1110 | spin_lock(&tree->lock); | |
e6138876 JB |
1111 | if (cached_state && *cached_state) { |
1112 | state = *cached_state; | |
1113 | if (state->start <= start && state->end > start && | |
27a3507d | 1114 | extent_state_in_tree(state)) { |
e6138876 JB |
1115 | node = &state->rb_node; |
1116 | goto hit_next; | |
1117 | } | |
1118 | } | |
1119 | ||
462d6fac JB |
1120 | /* |
1121 | * this search will find all the extents that end after | |
1122 | * our range starts. | |
1123 | */ | |
12cfbad9 | 1124 | node = tree_search_for_insert(tree, start, &p, &parent); |
462d6fac JB |
1125 | if (!node) { |
1126 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1127 | if (!prealloc) { |
1128 | err = -ENOMEM; | |
1129 | goto out; | |
1130 | } | |
12cfbad9 | 1131 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 1132 | &p, &parent, &bits, NULL); |
c2d904e0 JM |
1133 | if (err) |
1134 | extent_io_tree_panic(tree, err); | |
c42ac0bc FDBM |
1135 | cache_state(prealloc, cached_state); |
1136 | prealloc = NULL; | |
462d6fac JB |
1137 | goto out; |
1138 | } | |
1139 | state = rb_entry(node, struct extent_state, rb_node); | |
1140 | hit_next: | |
1141 | last_start = state->start; | |
1142 | last_end = state->end; | |
1143 | ||
1144 | /* | |
1145 | * | ---- desired range ---- | | |
1146 | * | state | | |
1147 | * | |
1148 | * Just lock what we found and keep going | |
1149 | */ | |
1150 | if (state->start == start && state->end <= end) { | |
d38ed27f | 1151 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1152 | cache_state(state, cached_state); |
fefdc557 | 1153 | state = clear_state_bit(tree, state, &clear_bits, 0, NULL); |
462d6fac JB |
1154 | if (last_end == (u64)-1) |
1155 | goto out; | |
462d6fac | 1156 | start = last_end + 1; |
d1ac6e41 LB |
1157 | if (start < end && state && state->start == start && |
1158 | !need_resched()) | |
1159 | goto hit_next; | |
462d6fac JB |
1160 | goto search_again; |
1161 | } | |
1162 | ||
1163 | /* | |
1164 | * | ---- desired range ---- | | |
1165 | * | state | | |
1166 | * or | |
1167 | * | ------------- state -------------- | | |
1168 | * | |
1169 | * We need to split the extent we found, and may flip bits on | |
1170 | * second half. | |
1171 | * | |
1172 | * If the extent we found extends past our | |
1173 | * range, we just split and search again. It'll get split | |
1174 | * again the next time though. | |
1175 | * | |
1176 | * If the extent we found is inside our range, we set the | |
1177 | * desired bit on it. | |
1178 | */ | |
1179 | if (state->start < start) { | |
1180 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1181 | if (!prealloc) { |
1182 | err = -ENOMEM; | |
1183 | goto out; | |
1184 | } | |
462d6fac | 1185 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1186 | if (err) |
1187 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1188 | prealloc = NULL; |
1189 | if (err) | |
1190 | goto out; | |
1191 | if (state->end <= end) { | |
d38ed27f | 1192 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1193 | cache_state(state, cached_state); |
fefdc557 QW |
1194 | state = clear_state_bit(tree, state, &clear_bits, 0, |
1195 | NULL); | |
462d6fac JB |
1196 | if (last_end == (u64)-1) |
1197 | goto out; | |
1198 | start = last_end + 1; | |
d1ac6e41 LB |
1199 | if (start < end && state && state->start == start && |
1200 | !need_resched()) | |
1201 | goto hit_next; | |
462d6fac JB |
1202 | } |
1203 | goto search_again; | |
1204 | } | |
1205 | /* | |
1206 | * | ---- desired range ---- | | |
1207 | * | state | or | state | | |
1208 | * | |
1209 | * There's a hole, we need to insert something in it and | |
1210 | * ignore the extent we found. | |
1211 | */ | |
1212 | if (state->start > start) { | |
1213 | u64 this_end; | |
1214 | if (end < last_start) | |
1215 | this_end = end; | |
1216 | else | |
1217 | this_end = last_start - 1; | |
1218 | ||
1219 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1220 | if (!prealloc) { |
1221 | err = -ENOMEM; | |
1222 | goto out; | |
1223 | } | |
462d6fac JB |
1224 | |
1225 | /* | |
1226 | * Avoid to free 'prealloc' if it can be merged with | |
1227 | * the later extent. | |
1228 | */ | |
1229 | err = insert_state(tree, prealloc, start, this_end, | |
d38ed27f | 1230 | NULL, NULL, &bits, NULL); |
c2d904e0 JM |
1231 | if (err) |
1232 | extent_io_tree_panic(tree, err); | |
e6138876 | 1233 | cache_state(prealloc, cached_state); |
462d6fac JB |
1234 | prealloc = NULL; |
1235 | start = this_end + 1; | |
1236 | goto search_again; | |
1237 | } | |
1238 | /* | |
1239 | * | ---- desired range ---- | | |
1240 | * | state | | |
1241 | * We need to split the extent, and set the bit | |
1242 | * on the first half | |
1243 | */ | |
1244 | if (state->start <= end && state->end > end) { | |
1245 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1246 | if (!prealloc) { |
1247 | err = -ENOMEM; | |
1248 | goto out; | |
1249 | } | |
462d6fac JB |
1250 | |
1251 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1252 | if (err) |
1253 | extent_io_tree_panic(tree, err); | |
462d6fac | 1254 | |
d38ed27f | 1255 | set_state_bits(tree, prealloc, &bits, NULL); |
e6138876 | 1256 | cache_state(prealloc, cached_state); |
fefdc557 | 1257 | clear_state_bit(tree, prealloc, &clear_bits, 0, NULL); |
462d6fac JB |
1258 | prealloc = NULL; |
1259 | goto out; | |
1260 | } | |
1261 | ||
462d6fac JB |
1262 | search_again: |
1263 | if (start > end) | |
1264 | goto out; | |
1265 | spin_unlock(&tree->lock); | |
210aa277 | 1266 | cond_resched(); |
c8fd3de7 | 1267 | first_iteration = false; |
462d6fac | 1268 | goto again; |
462d6fac JB |
1269 | |
1270 | out: | |
1271 | spin_unlock(&tree->lock); | |
1272 | if (prealloc) | |
1273 | free_extent_state(prealloc); | |
1274 | ||
1275 | return err; | |
462d6fac JB |
1276 | } |
1277 | ||
d1310b2e | 1278 | /* wrappers around set/clear extent bit */ |
d38ed27f | 1279 | int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
2c53b912 | 1280 | unsigned bits, struct extent_changeset *changeset) |
d38ed27f QW |
1281 | { |
1282 | /* | |
1283 | * We don't support EXTENT_LOCKED yet, as current changeset will | |
1284 | * record any bits changed, so for EXTENT_LOCKED case, it will | |
1285 | * either fail with -EEXIST or changeset will record the whole | |
1286 | * range. | |
1287 | */ | |
1288 | BUG_ON(bits & EXTENT_LOCKED); | |
1289 | ||
2c53b912 | 1290 | return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS, |
d38ed27f QW |
1291 | changeset); |
1292 | } | |
1293 | ||
fefdc557 QW |
1294 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
1295 | unsigned bits, int wake, int delete, | |
ae0f1625 | 1296 | struct extent_state **cached) |
fefdc557 QW |
1297 | { |
1298 | return __clear_extent_bit(tree, start, end, bits, wake, delete, | |
ae0f1625 | 1299 | cached, GFP_NOFS, NULL); |
fefdc557 QW |
1300 | } |
1301 | ||
fefdc557 | 1302 | int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
f734c44a | 1303 | unsigned bits, struct extent_changeset *changeset) |
fefdc557 QW |
1304 | { |
1305 | /* | |
1306 | * Don't support EXTENT_LOCKED case, same reason as | |
1307 | * set_record_extent_bits(). | |
1308 | */ | |
1309 | BUG_ON(bits & EXTENT_LOCKED); | |
1310 | ||
f734c44a | 1311 | return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS, |
fefdc557 QW |
1312 | changeset); |
1313 | } | |
1314 | ||
d352ac68 CM |
1315 | /* |
1316 | * either insert or lock state struct between start and end use mask to tell | |
1317 | * us if waiting is desired. | |
1318 | */ | |
1edbb734 | 1319 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
ff13db41 | 1320 | struct extent_state **cached_state) |
d1310b2e CM |
1321 | { |
1322 | int err; | |
1323 | u64 failed_start; | |
9ee49a04 | 1324 | |
d1310b2e | 1325 | while (1) { |
ff13db41 | 1326 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, |
3fbe5c02 | 1327 | EXTENT_LOCKED, &failed_start, |
d38ed27f | 1328 | cached_state, GFP_NOFS, NULL); |
d0082371 | 1329 | if (err == -EEXIST) { |
d1310b2e CM |
1330 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1331 | start = failed_start; | |
d0082371 | 1332 | } else |
d1310b2e | 1333 | break; |
d1310b2e CM |
1334 | WARN_ON(start > end); |
1335 | } | |
1336 | return err; | |
1337 | } | |
d1310b2e | 1338 | |
d0082371 | 1339 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1340 | { |
1341 | int err; | |
1342 | u64 failed_start; | |
1343 | ||
3fbe5c02 | 1344 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
d38ed27f | 1345 | &failed_start, NULL, GFP_NOFS, NULL); |
6643558d YZ |
1346 | if (err == -EEXIST) { |
1347 | if (failed_start > start) | |
1348 | clear_extent_bit(tree, start, failed_start - 1, | |
ae0f1625 | 1349 | EXTENT_LOCKED, 1, 0, NULL); |
25179201 | 1350 | return 0; |
6643558d | 1351 | } |
25179201 JB |
1352 | return 1; |
1353 | } | |
25179201 | 1354 | |
bd1fa4f0 | 1355 | void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1356 | { |
09cbfeaf KS |
1357 | unsigned long index = start >> PAGE_SHIFT; |
1358 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1359 | struct page *page; |
1360 | ||
1361 | while (index <= end_index) { | |
1362 | page = find_get_page(inode->i_mapping, index); | |
1363 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1364 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1365 | put_page(page); |
4adaa611 CM |
1366 | index++; |
1367 | } | |
4adaa611 CM |
1368 | } |
1369 | ||
f6311572 | 1370 | void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1371 | { |
09cbfeaf KS |
1372 | unsigned long index = start >> PAGE_SHIFT; |
1373 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1374 | struct page *page; |
1375 | ||
1376 | while (index <= end_index) { | |
1377 | page = find_get_page(inode->i_mapping, index); | |
1378 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
4adaa611 | 1379 | __set_page_dirty_nobuffers(page); |
8d38633c | 1380 | account_page_redirty(page); |
09cbfeaf | 1381 | put_page(page); |
4adaa611 CM |
1382 | index++; |
1383 | } | |
4adaa611 CM |
1384 | } |
1385 | ||
d1310b2e CM |
1386 | /* |
1387 | * helper function to set both pages and extents in the tree writeback | |
1388 | */ | |
35de6db2 | 1389 | static void set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e | 1390 | { |
c6100a4b | 1391 | tree->ops->set_range_writeback(tree->private_data, start, end); |
d1310b2e | 1392 | } |
d1310b2e | 1393 | |
d352ac68 CM |
1394 | /* find the first state struct with 'bits' set after 'start', and |
1395 | * return it. tree->lock must be held. NULL will returned if | |
1396 | * nothing was found after 'start' | |
1397 | */ | |
48a3b636 ES |
1398 | static struct extent_state * |
1399 | find_first_extent_bit_state(struct extent_io_tree *tree, | |
9ee49a04 | 1400 | u64 start, unsigned bits) |
d7fc640e CM |
1401 | { |
1402 | struct rb_node *node; | |
1403 | struct extent_state *state; | |
1404 | ||
1405 | /* | |
1406 | * this search will find all the extents that end after | |
1407 | * our range starts. | |
1408 | */ | |
1409 | node = tree_search(tree, start); | |
d397712b | 1410 | if (!node) |
d7fc640e | 1411 | goto out; |
d7fc640e | 1412 | |
d397712b | 1413 | while (1) { |
d7fc640e | 1414 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1415 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1416 | return state; |
d397712b | 1417 | |
d7fc640e CM |
1418 | node = rb_next(node); |
1419 | if (!node) | |
1420 | break; | |
1421 | } | |
1422 | out: | |
1423 | return NULL; | |
1424 | } | |
d7fc640e | 1425 | |
69261c4b XG |
1426 | /* |
1427 | * find the first offset in the io tree with 'bits' set. zero is | |
1428 | * returned if we find something, and *start_ret and *end_ret are | |
1429 | * set to reflect the state struct that was found. | |
1430 | * | |
477d7eaf | 1431 | * If nothing was found, 1 is returned. If found something, return 0. |
69261c4b XG |
1432 | */ |
1433 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
9ee49a04 | 1434 | u64 *start_ret, u64 *end_ret, unsigned bits, |
e6138876 | 1435 | struct extent_state **cached_state) |
69261c4b XG |
1436 | { |
1437 | struct extent_state *state; | |
e6138876 | 1438 | struct rb_node *n; |
69261c4b XG |
1439 | int ret = 1; |
1440 | ||
1441 | spin_lock(&tree->lock); | |
e6138876 JB |
1442 | if (cached_state && *cached_state) { |
1443 | state = *cached_state; | |
27a3507d | 1444 | if (state->end == start - 1 && extent_state_in_tree(state)) { |
e6138876 JB |
1445 | n = rb_next(&state->rb_node); |
1446 | while (n) { | |
1447 | state = rb_entry(n, struct extent_state, | |
1448 | rb_node); | |
1449 | if (state->state & bits) | |
1450 | goto got_it; | |
1451 | n = rb_next(n); | |
1452 | } | |
1453 | free_extent_state(*cached_state); | |
1454 | *cached_state = NULL; | |
1455 | goto out; | |
1456 | } | |
1457 | free_extent_state(*cached_state); | |
1458 | *cached_state = NULL; | |
1459 | } | |
1460 | ||
69261c4b | 1461 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1462 | got_it: |
69261c4b | 1463 | if (state) { |
e38e2ed7 | 1464 | cache_state_if_flags(state, cached_state, 0); |
69261c4b XG |
1465 | *start_ret = state->start; |
1466 | *end_ret = state->end; | |
1467 | ret = 0; | |
1468 | } | |
e6138876 | 1469 | out: |
69261c4b XG |
1470 | spin_unlock(&tree->lock); |
1471 | return ret; | |
1472 | } | |
1473 | ||
d352ac68 CM |
1474 | /* |
1475 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1476 | * more than 'max_bytes'. start and end are used to return the range, | |
1477 | * | |
1478 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1479 | */ | |
c8b97818 | 1480 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1481 | u64 *start, u64 *end, u64 max_bytes, |
1482 | struct extent_state **cached_state) | |
d1310b2e CM |
1483 | { |
1484 | struct rb_node *node; | |
1485 | struct extent_state *state; | |
1486 | u64 cur_start = *start; | |
1487 | u64 found = 0; | |
1488 | u64 total_bytes = 0; | |
1489 | ||
cad321ad | 1490 | spin_lock(&tree->lock); |
c8b97818 | 1491 | |
d1310b2e CM |
1492 | /* |
1493 | * this search will find all the extents that end after | |
1494 | * our range starts. | |
1495 | */ | |
80ea96b1 | 1496 | node = tree_search(tree, cur_start); |
2b114d1d | 1497 | if (!node) { |
3b951516 CM |
1498 | if (!found) |
1499 | *end = (u64)-1; | |
d1310b2e CM |
1500 | goto out; |
1501 | } | |
1502 | ||
d397712b | 1503 | while (1) { |
d1310b2e | 1504 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1505 | if (found && (state->start != cur_start || |
1506 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1507 | goto out; |
1508 | } | |
1509 | if (!(state->state & EXTENT_DELALLOC)) { | |
1510 | if (!found) | |
1511 | *end = state->end; | |
1512 | goto out; | |
1513 | } | |
c2a128d2 | 1514 | if (!found) { |
d1310b2e | 1515 | *start = state->start; |
c2a128d2 | 1516 | *cached_state = state; |
b7ac31b7 | 1517 | refcount_inc(&state->refs); |
c2a128d2 | 1518 | } |
d1310b2e CM |
1519 | found++; |
1520 | *end = state->end; | |
1521 | cur_start = state->end + 1; | |
1522 | node = rb_next(node); | |
d1310b2e | 1523 | total_bytes += state->end - state->start + 1; |
7bf811a5 | 1524 | if (total_bytes >= max_bytes) |
573aecaf | 1525 | break; |
573aecaf | 1526 | if (!node) |
d1310b2e CM |
1527 | break; |
1528 | } | |
1529 | out: | |
cad321ad | 1530 | spin_unlock(&tree->lock); |
d1310b2e CM |
1531 | return found; |
1532 | } | |
1533 | ||
da2c7009 LB |
1534 | static int __process_pages_contig(struct address_space *mapping, |
1535 | struct page *locked_page, | |
1536 | pgoff_t start_index, pgoff_t end_index, | |
1537 | unsigned long page_ops, pgoff_t *index_ret); | |
1538 | ||
143bede5 JM |
1539 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1540 | struct page *locked_page, | |
1541 | u64 start, u64 end) | |
c8b97818 | 1542 | { |
09cbfeaf KS |
1543 | unsigned long index = start >> PAGE_SHIFT; |
1544 | unsigned long end_index = end >> PAGE_SHIFT; | |
c8b97818 | 1545 | |
76c0021d | 1546 | ASSERT(locked_page); |
c8b97818 | 1547 | if (index == locked_page->index && end_index == index) |
143bede5 | 1548 | return; |
c8b97818 | 1549 | |
76c0021d LB |
1550 | __process_pages_contig(inode->i_mapping, locked_page, index, end_index, |
1551 | PAGE_UNLOCK, NULL); | |
c8b97818 CM |
1552 | } |
1553 | ||
1554 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1555 | struct page *locked_page, | |
1556 | u64 delalloc_start, | |
1557 | u64 delalloc_end) | |
1558 | { | |
09cbfeaf | 1559 | unsigned long index = delalloc_start >> PAGE_SHIFT; |
76c0021d | 1560 | unsigned long index_ret = index; |
09cbfeaf | 1561 | unsigned long end_index = delalloc_end >> PAGE_SHIFT; |
c8b97818 | 1562 | int ret; |
c8b97818 | 1563 | |
76c0021d | 1564 | ASSERT(locked_page); |
c8b97818 CM |
1565 | if (index == locked_page->index && index == end_index) |
1566 | return 0; | |
1567 | ||
76c0021d LB |
1568 | ret = __process_pages_contig(inode->i_mapping, locked_page, index, |
1569 | end_index, PAGE_LOCK, &index_ret); | |
1570 | if (ret == -EAGAIN) | |
1571 | __unlock_for_delalloc(inode, locked_page, delalloc_start, | |
1572 | (u64)index_ret << PAGE_SHIFT); | |
c8b97818 CM |
1573 | return ret; |
1574 | } | |
1575 | ||
1576 | /* | |
1577 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1578 | * more than 'max_bytes'. start and end are used to return the range, | |
1579 | * | |
1580 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1581 | */ | |
294e30fe JB |
1582 | STATIC u64 find_lock_delalloc_range(struct inode *inode, |
1583 | struct extent_io_tree *tree, | |
1584 | struct page *locked_page, u64 *start, | |
1585 | u64 *end, u64 max_bytes) | |
c8b97818 CM |
1586 | { |
1587 | u64 delalloc_start; | |
1588 | u64 delalloc_end; | |
1589 | u64 found; | |
9655d298 | 1590 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1591 | int ret; |
1592 | int loops = 0; | |
1593 | ||
1594 | again: | |
1595 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1596 | delalloc_start = *start; | |
1597 | delalloc_end = 0; | |
1598 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1599 | max_bytes, &cached_state); |
70b99e69 | 1600 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1601 | *start = delalloc_start; |
1602 | *end = delalloc_end; | |
c2a128d2 | 1603 | free_extent_state(cached_state); |
385fe0be | 1604 | return 0; |
c8b97818 CM |
1605 | } |
1606 | ||
70b99e69 CM |
1607 | /* |
1608 | * start comes from the offset of locked_page. We have to lock | |
1609 | * pages in order, so we can't process delalloc bytes before | |
1610 | * locked_page | |
1611 | */ | |
d397712b | 1612 | if (delalloc_start < *start) |
70b99e69 | 1613 | delalloc_start = *start; |
70b99e69 | 1614 | |
c8b97818 CM |
1615 | /* |
1616 | * make sure to limit the number of pages we try to lock down | |
c8b97818 | 1617 | */ |
7bf811a5 JB |
1618 | if (delalloc_end + 1 - delalloc_start > max_bytes) |
1619 | delalloc_end = delalloc_start + max_bytes - 1; | |
d397712b | 1620 | |
c8b97818 CM |
1621 | /* step two, lock all the pages after the page that has start */ |
1622 | ret = lock_delalloc_pages(inode, locked_page, | |
1623 | delalloc_start, delalloc_end); | |
1624 | if (ret == -EAGAIN) { | |
1625 | /* some of the pages are gone, lets avoid looping by | |
1626 | * shortening the size of the delalloc range we're searching | |
1627 | */ | |
9655d298 | 1628 | free_extent_state(cached_state); |
7d788742 | 1629 | cached_state = NULL; |
c8b97818 | 1630 | if (!loops) { |
09cbfeaf | 1631 | max_bytes = PAGE_SIZE; |
c8b97818 CM |
1632 | loops = 1; |
1633 | goto again; | |
1634 | } else { | |
1635 | found = 0; | |
1636 | goto out_failed; | |
1637 | } | |
1638 | } | |
79787eaa | 1639 | BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */ |
c8b97818 CM |
1640 | |
1641 | /* step three, lock the state bits for the whole range */ | |
ff13db41 | 1642 | lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state); |
c8b97818 CM |
1643 | |
1644 | /* then test to make sure it is all still delalloc */ | |
1645 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1646 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1647 | if (!ret) { |
9655d298 CM |
1648 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
1649 | &cached_state, GFP_NOFS); | |
c8b97818 CM |
1650 | __unlock_for_delalloc(inode, locked_page, |
1651 | delalloc_start, delalloc_end); | |
1652 | cond_resched(); | |
1653 | goto again; | |
1654 | } | |
9655d298 | 1655 | free_extent_state(cached_state); |
c8b97818 CM |
1656 | *start = delalloc_start; |
1657 | *end = delalloc_end; | |
1658 | out_failed: | |
1659 | return found; | |
1660 | } | |
1661 | ||
da2c7009 LB |
1662 | static int __process_pages_contig(struct address_space *mapping, |
1663 | struct page *locked_page, | |
1664 | pgoff_t start_index, pgoff_t end_index, | |
1665 | unsigned long page_ops, pgoff_t *index_ret) | |
c8b97818 | 1666 | { |
873695b3 | 1667 | unsigned long nr_pages = end_index - start_index + 1; |
da2c7009 | 1668 | unsigned long pages_locked = 0; |
873695b3 | 1669 | pgoff_t index = start_index; |
c8b97818 | 1670 | struct page *pages[16]; |
873695b3 | 1671 | unsigned ret; |
da2c7009 | 1672 | int err = 0; |
c8b97818 | 1673 | int i; |
771ed689 | 1674 | |
da2c7009 LB |
1675 | if (page_ops & PAGE_LOCK) { |
1676 | ASSERT(page_ops == PAGE_LOCK); | |
1677 | ASSERT(index_ret && *index_ret == start_index); | |
1678 | } | |
1679 | ||
704de49d | 1680 | if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0) |
873695b3 | 1681 | mapping_set_error(mapping, -EIO); |
704de49d | 1682 | |
d397712b | 1683 | while (nr_pages > 0) { |
873695b3 | 1684 | ret = find_get_pages_contig(mapping, index, |
5b050f04 CM |
1685 | min_t(unsigned long, |
1686 | nr_pages, ARRAY_SIZE(pages)), pages); | |
da2c7009 LB |
1687 | if (ret == 0) { |
1688 | /* | |
1689 | * Only if we're going to lock these pages, | |
1690 | * can we find nothing at @index. | |
1691 | */ | |
1692 | ASSERT(page_ops & PAGE_LOCK); | |
49d4a334 LB |
1693 | err = -EAGAIN; |
1694 | goto out; | |
da2c7009 | 1695 | } |
8b62b72b | 1696 | |
da2c7009 | 1697 | for (i = 0; i < ret; i++) { |
c2790a2e | 1698 | if (page_ops & PAGE_SET_PRIVATE2) |
8b62b72b CM |
1699 | SetPagePrivate2(pages[i]); |
1700 | ||
c8b97818 | 1701 | if (pages[i] == locked_page) { |
09cbfeaf | 1702 | put_page(pages[i]); |
da2c7009 | 1703 | pages_locked++; |
c8b97818 CM |
1704 | continue; |
1705 | } | |
c2790a2e | 1706 | if (page_ops & PAGE_CLEAR_DIRTY) |
c8b97818 | 1707 | clear_page_dirty_for_io(pages[i]); |
c2790a2e | 1708 | if (page_ops & PAGE_SET_WRITEBACK) |
c8b97818 | 1709 | set_page_writeback(pages[i]); |
704de49d FM |
1710 | if (page_ops & PAGE_SET_ERROR) |
1711 | SetPageError(pages[i]); | |
c2790a2e | 1712 | if (page_ops & PAGE_END_WRITEBACK) |
c8b97818 | 1713 | end_page_writeback(pages[i]); |
c2790a2e | 1714 | if (page_ops & PAGE_UNLOCK) |
771ed689 | 1715 | unlock_page(pages[i]); |
da2c7009 LB |
1716 | if (page_ops & PAGE_LOCK) { |
1717 | lock_page(pages[i]); | |
1718 | if (!PageDirty(pages[i]) || | |
1719 | pages[i]->mapping != mapping) { | |
1720 | unlock_page(pages[i]); | |
1721 | put_page(pages[i]); | |
1722 | err = -EAGAIN; | |
1723 | goto out; | |
1724 | } | |
1725 | } | |
09cbfeaf | 1726 | put_page(pages[i]); |
da2c7009 | 1727 | pages_locked++; |
c8b97818 CM |
1728 | } |
1729 | nr_pages -= ret; | |
1730 | index += ret; | |
1731 | cond_resched(); | |
1732 | } | |
da2c7009 LB |
1733 | out: |
1734 | if (err && index_ret) | |
1735 | *index_ret = start_index + pages_locked - 1; | |
1736 | return err; | |
c8b97818 | 1737 | } |
c8b97818 | 1738 | |
873695b3 LB |
1739 | void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end, |
1740 | u64 delalloc_end, struct page *locked_page, | |
1741 | unsigned clear_bits, | |
1742 | unsigned long page_ops) | |
1743 | { | |
1744 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0, | |
ae0f1625 | 1745 | NULL); |
873695b3 LB |
1746 | |
1747 | __process_pages_contig(inode->i_mapping, locked_page, | |
1748 | start >> PAGE_SHIFT, end >> PAGE_SHIFT, | |
da2c7009 | 1749 | page_ops, NULL); |
873695b3 LB |
1750 | } |
1751 | ||
d352ac68 CM |
1752 | /* |
1753 | * count the number of bytes in the tree that have a given bit(s) | |
1754 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1755 | * cached. The total number found is returned. | |
1756 | */ | |
d1310b2e CM |
1757 | u64 count_range_bits(struct extent_io_tree *tree, |
1758 | u64 *start, u64 search_end, u64 max_bytes, | |
9ee49a04 | 1759 | unsigned bits, int contig) |
d1310b2e CM |
1760 | { |
1761 | struct rb_node *node; | |
1762 | struct extent_state *state; | |
1763 | u64 cur_start = *start; | |
1764 | u64 total_bytes = 0; | |
ec29ed5b | 1765 | u64 last = 0; |
d1310b2e CM |
1766 | int found = 0; |
1767 | ||
fae7f21c | 1768 | if (WARN_ON(search_end <= cur_start)) |
d1310b2e | 1769 | return 0; |
d1310b2e | 1770 | |
cad321ad | 1771 | spin_lock(&tree->lock); |
d1310b2e CM |
1772 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1773 | total_bytes = tree->dirty_bytes; | |
1774 | goto out; | |
1775 | } | |
1776 | /* | |
1777 | * this search will find all the extents that end after | |
1778 | * our range starts. | |
1779 | */ | |
80ea96b1 | 1780 | node = tree_search(tree, cur_start); |
d397712b | 1781 | if (!node) |
d1310b2e | 1782 | goto out; |
d1310b2e | 1783 | |
d397712b | 1784 | while (1) { |
d1310b2e CM |
1785 | state = rb_entry(node, struct extent_state, rb_node); |
1786 | if (state->start > search_end) | |
1787 | break; | |
ec29ed5b CM |
1788 | if (contig && found && state->start > last + 1) |
1789 | break; | |
1790 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1791 | total_bytes += min(search_end, state->end) + 1 - |
1792 | max(cur_start, state->start); | |
1793 | if (total_bytes >= max_bytes) | |
1794 | break; | |
1795 | if (!found) { | |
af60bed2 | 1796 | *start = max(cur_start, state->start); |
d1310b2e CM |
1797 | found = 1; |
1798 | } | |
ec29ed5b CM |
1799 | last = state->end; |
1800 | } else if (contig && found) { | |
1801 | break; | |
d1310b2e CM |
1802 | } |
1803 | node = rb_next(node); | |
1804 | if (!node) | |
1805 | break; | |
1806 | } | |
1807 | out: | |
cad321ad | 1808 | spin_unlock(&tree->lock); |
d1310b2e CM |
1809 | return total_bytes; |
1810 | } | |
b2950863 | 1811 | |
d352ac68 CM |
1812 | /* |
1813 | * set the private field for a given byte offset in the tree. If there isn't | |
1814 | * an extent_state there already, this does nothing. | |
1815 | */ | |
f827ba9a | 1816 | static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start, |
47dc196a | 1817 | struct io_failure_record *failrec) |
d1310b2e CM |
1818 | { |
1819 | struct rb_node *node; | |
1820 | struct extent_state *state; | |
1821 | int ret = 0; | |
1822 | ||
cad321ad | 1823 | spin_lock(&tree->lock); |
d1310b2e CM |
1824 | /* |
1825 | * this search will find all the extents that end after | |
1826 | * our range starts. | |
1827 | */ | |
80ea96b1 | 1828 | node = tree_search(tree, start); |
2b114d1d | 1829 | if (!node) { |
d1310b2e CM |
1830 | ret = -ENOENT; |
1831 | goto out; | |
1832 | } | |
1833 | state = rb_entry(node, struct extent_state, rb_node); | |
1834 | if (state->start != start) { | |
1835 | ret = -ENOENT; | |
1836 | goto out; | |
1837 | } | |
47dc196a | 1838 | state->failrec = failrec; |
d1310b2e | 1839 | out: |
cad321ad | 1840 | spin_unlock(&tree->lock); |
d1310b2e CM |
1841 | return ret; |
1842 | } | |
1843 | ||
f827ba9a | 1844 | static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start, |
47dc196a | 1845 | struct io_failure_record **failrec) |
d1310b2e CM |
1846 | { |
1847 | struct rb_node *node; | |
1848 | struct extent_state *state; | |
1849 | int ret = 0; | |
1850 | ||
cad321ad | 1851 | spin_lock(&tree->lock); |
d1310b2e CM |
1852 | /* |
1853 | * this search will find all the extents that end after | |
1854 | * our range starts. | |
1855 | */ | |
80ea96b1 | 1856 | node = tree_search(tree, start); |
2b114d1d | 1857 | if (!node) { |
d1310b2e CM |
1858 | ret = -ENOENT; |
1859 | goto out; | |
1860 | } | |
1861 | state = rb_entry(node, struct extent_state, rb_node); | |
1862 | if (state->start != start) { | |
1863 | ret = -ENOENT; | |
1864 | goto out; | |
1865 | } | |
47dc196a | 1866 | *failrec = state->failrec; |
d1310b2e | 1867 | out: |
cad321ad | 1868 | spin_unlock(&tree->lock); |
d1310b2e CM |
1869 | return ret; |
1870 | } | |
1871 | ||
1872 | /* | |
1873 | * searches a range in the state tree for a given mask. | |
70dec807 | 1874 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1875 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1876 | * range is found set. | |
1877 | */ | |
1878 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1879 | unsigned bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1880 | { |
1881 | struct extent_state *state = NULL; | |
1882 | struct rb_node *node; | |
1883 | int bitset = 0; | |
d1310b2e | 1884 | |
cad321ad | 1885 | spin_lock(&tree->lock); |
27a3507d | 1886 | if (cached && extent_state_in_tree(cached) && cached->start <= start && |
df98b6e2 | 1887 | cached->end > start) |
9655d298 CM |
1888 | node = &cached->rb_node; |
1889 | else | |
1890 | node = tree_search(tree, start); | |
d1310b2e CM |
1891 | while (node && start <= end) { |
1892 | state = rb_entry(node, struct extent_state, rb_node); | |
1893 | ||
1894 | if (filled && state->start > start) { | |
1895 | bitset = 0; | |
1896 | break; | |
1897 | } | |
1898 | ||
1899 | if (state->start > end) | |
1900 | break; | |
1901 | ||
1902 | if (state->state & bits) { | |
1903 | bitset = 1; | |
1904 | if (!filled) | |
1905 | break; | |
1906 | } else if (filled) { | |
1907 | bitset = 0; | |
1908 | break; | |
1909 | } | |
46562cec CM |
1910 | |
1911 | if (state->end == (u64)-1) | |
1912 | break; | |
1913 | ||
d1310b2e CM |
1914 | start = state->end + 1; |
1915 | if (start > end) | |
1916 | break; | |
1917 | node = rb_next(node); | |
1918 | if (!node) { | |
1919 | if (filled) | |
1920 | bitset = 0; | |
1921 | break; | |
1922 | } | |
1923 | } | |
cad321ad | 1924 | spin_unlock(&tree->lock); |
d1310b2e CM |
1925 | return bitset; |
1926 | } | |
d1310b2e CM |
1927 | |
1928 | /* | |
1929 | * helper function to set a given page up to date if all the | |
1930 | * extents in the tree for that page are up to date | |
1931 | */ | |
143bede5 | 1932 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e | 1933 | { |
4eee4fa4 | 1934 | u64 start = page_offset(page); |
09cbfeaf | 1935 | u64 end = start + PAGE_SIZE - 1; |
9655d298 | 1936 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 1937 | SetPageUptodate(page); |
d1310b2e CM |
1938 | } |
1939 | ||
7870d082 JB |
1940 | int free_io_failure(struct extent_io_tree *failure_tree, |
1941 | struct extent_io_tree *io_tree, | |
1942 | struct io_failure_record *rec) | |
4a54c8c1 JS |
1943 | { |
1944 | int ret; | |
1945 | int err = 0; | |
4a54c8c1 | 1946 | |
47dc196a | 1947 | set_state_failrec(failure_tree, rec->start, NULL); |
4a54c8c1 JS |
1948 | ret = clear_extent_bits(failure_tree, rec->start, |
1949 | rec->start + rec->len - 1, | |
91166212 | 1950 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 JS |
1951 | if (ret) |
1952 | err = ret; | |
1953 | ||
7870d082 | 1954 | ret = clear_extent_bits(io_tree, rec->start, |
53b381b3 | 1955 | rec->start + rec->len - 1, |
91166212 | 1956 | EXTENT_DAMAGED); |
53b381b3 DW |
1957 | if (ret && !err) |
1958 | err = ret; | |
4a54c8c1 JS |
1959 | |
1960 | kfree(rec); | |
1961 | return err; | |
1962 | } | |
1963 | ||
4a54c8c1 JS |
1964 | /* |
1965 | * this bypasses the standard btrfs submit functions deliberately, as | |
1966 | * the standard behavior is to write all copies in a raid setup. here we only | |
1967 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
1968 | * submit_bio directly. | |
3ec706c8 | 1969 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
1970 | * actually prevents the read that triggered the error from finishing. |
1971 | * currently, there can be no more than two copies of every data bit. thus, | |
1972 | * exactly one rewrite is required. | |
1973 | */ | |
6ec656bc JB |
1974 | int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start, |
1975 | u64 length, u64 logical, struct page *page, | |
1976 | unsigned int pg_offset, int mirror_num) | |
4a54c8c1 JS |
1977 | { |
1978 | struct bio *bio; | |
1979 | struct btrfs_device *dev; | |
4a54c8c1 JS |
1980 | u64 map_length = 0; |
1981 | u64 sector; | |
1982 | struct btrfs_bio *bbio = NULL; | |
1983 | int ret; | |
1984 | ||
1751e8a6 | 1985 | ASSERT(!(fs_info->sb->s_flags & SB_RDONLY)); |
4a54c8c1 JS |
1986 | BUG_ON(!mirror_num); |
1987 | ||
c5e4c3d7 | 1988 | bio = btrfs_io_bio_alloc(1); |
4f024f37 | 1989 | bio->bi_iter.bi_size = 0; |
4a54c8c1 JS |
1990 | map_length = length; |
1991 | ||
b5de8d0d FM |
1992 | /* |
1993 | * Avoid races with device replace and make sure our bbio has devices | |
1994 | * associated to its stripes that don't go away while we are doing the | |
1995 | * read repair operation. | |
1996 | */ | |
1997 | btrfs_bio_counter_inc_blocked(fs_info); | |
e4ff5fb5 | 1998 | if (btrfs_is_parity_mirror(fs_info, logical, length)) { |
c725328c LB |
1999 | /* |
2000 | * Note that we don't use BTRFS_MAP_WRITE because it's supposed | |
2001 | * to update all raid stripes, but here we just want to correct | |
2002 | * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad | |
2003 | * stripe's dev and sector. | |
2004 | */ | |
2005 | ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical, | |
2006 | &map_length, &bbio, 0); | |
2007 | if (ret) { | |
2008 | btrfs_bio_counter_dec(fs_info); | |
2009 | bio_put(bio); | |
2010 | return -EIO; | |
2011 | } | |
2012 | ASSERT(bbio->mirror_num == 1); | |
2013 | } else { | |
2014 | ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, | |
2015 | &map_length, &bbio, mirror_num); | |
2016 | if (ret) { | |
2017 | btrfs_bio_counter_dec(fs_info); | |
2018 | bio_put(bio); | |
2019 | return -EIO; | |
2020 | } | |
2021 | BUG_ON(mirror_num != bbio->mirror_num); | |
4a54c8c1 | 2022 | } |
c725328c LB |
2023 | |
2024 | sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9; | |
4f024f37 | 2025 | bio->bi_iter.bi_sector = sector; |
c725328c | 2026 | dev = bbio->stripes[bbio->mirror_num - 1].dev; |
6e9606d2 | 2027 | btrfs_put_bbio(bbio); |
4a54c8c1 | 2028 | if (!dev || !dev->bdev || !dev->writeable) { |
b5de8d0d | 2029 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2030 | bio_put(bio); |
2031 | return -EIO; | |
2032 | } | |
74d46992 | 2033 | bio_set_dev(bio, dev->bdev); |
70fd7614 | 2034 | bio->bi_opf = REQ_OP_WRITE | REQ_SYNC; |
ffdd2018 | 2035 | bio_add_page(bio, page, length, pg_offset); |
4a54c8c1 | 2036 | |
4e49ea4a | 2037 | if (btrfsic_submit_bio_wait(bio)) { |
4a54c8c1 | 2038 | /* try to remap that extent elsewhere? */ |
b5de8d0d | 2039 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 | 2040 | bio_put(bio); |
442a4f63 | 2041 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
2042 | return -EIO; |
2043 | } | |
2044 | ||
b14af3b4 DS |
2045 | btrfs_info_rl_in_rcu(fs_info, |
2046 | "read error corrected: ino %llu off %llu (dev %s sector %llu)", | |
6ec656bc | 2047 | ino, start, |
1203b681 | 2048 | rcu_str_deref(dev->name), sector); |
b5de8d0d | 2049 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2050 | bio_put(bio); |
2051 | return 0; | |
2052 | } | |
2053 | ||
2ff7e61e JM |
2054 | int repair_eb_io_failure(struct btrfs_fs_info *fs_info, |
2055 | struct extent_buffer *eb, int mirror_num) | |
ea466794 | 2056 | { |
ea466794 JB |
2057 | u64 start = eb->start; |
2058 | unsigned long i, num_pages = num_extent_pages(eb->start, eb->len); | |
d95603b2 | 2059 | int ret = 0; |
ea466794 | 2060 | |
bc98a42c | 2061 | if (sb_rdonly(fs_info->sb)) |
908960c6 ID |
2062 | return -EROFS; |
2063 | ||
ea466794 | 2064 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 2065 | struct page *p = eb->pages[i]; |
1203b681 | 2066 | |
6ec656bc | 2067 | ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p, |
1203b681 | 2068 | start - page_offset(p), mirror_num); |
ea466794 JB |
2069 | if (ret) |
2070 | break; | |
09cbfeaf | 2071 | start += PAGE_SIZE; |
ea466794 JB |
2072 | } |
2073 | ||
2074 | return ret; | |
2075 | } | |
2076 | ||
4a54c8c1 JS |
2077 | /* |
2078 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2079 | * to see if we need to process or clean up an io_failure_record | |
2080 | */ | |
7870d082 JB |
2081 | int clean_io_failure(struct btrfs_fs_info *fs_info, |
2082 | struct extent_io_tree *failure_tree, | |
2083 | struct extent_io_tree *io_tree, u64 start, | |
2084 | struct page *page, u64 ino, unsigned int pg_offset) | |
4a54c8c1 JS |
2085 | { |
2086 | u64 private; | |
4a54c8c1 | 2087 | struct io_failure_record *failrec; |
4a54c8c1 JS |
2088 | struct extent_state *state; |
2089 | int num_copies; | |
4a54c8c1 | 2090 | int ret; |
4a54c8c1 JS |
2091 | |
2092 | private = 0; | |
7870d082 JB |
2093 | ret = count_range_bits(failure_tree, &private, (u64)-1, 1, |
2094 | EXTENT_DIRTY, 0); | |
4a54c8c1 JS |
2095 | if (!ret) |
2096 | return 0; | |
2097 | ||
7870d082 | 2098 | ret = get_state_failrec(failure_tree, start, &failrec); |
4a54c8c1 JS |
2099 | if (ret) |
2100 | return 0; | |
2101 | ||
4a54c8c1 JS |
2102 | BUG_ON(!failrec->this_mirror); |
2103 | ||
2104 | if (failrec->in_validation) { | |
2105 | /* there was no real error, just free the record */ | |
ab8d0fc4 JM |
2106 | btrfs_debug(fs_info, |
2107 | "clean_io_failure: freeing dummy error at %llu", | |
2108 | failrec->start); | |
4a54c8c1 JS |
2109 | goto out; |
2110 | } | |
bc98a42c | 2111 | if (sb_rdonly(fs_info->sb)) |
908960c6 | 2112 | goto out; |
4a54c8c1 | 2113 | |
7870d082 JB |
2114 | spin_lock(&io_tree->lock); |
2115 | state = find_first_extent_bit_state(io_tree, | |
4a54c8c1 JS |
2116 | failrec->start, |
2117 | EXTENT_LOCKED); | |
7870d082 | 2118 | spin_unlock(&io_tree->lock); |
4a54c8c1 | 2119 | |
883d0de4 MX |
2120 | if (state && state->start <= failrec->start && |
2121 | state->end >= failrec->start + failrec->len - 1) { | |
3ec706c8 SB |
2122 | num_copies = btrfs_num_copies(fs_info, failrec->logical, |
2123 | failrec->len); | |
4a54c8c1 | 2124 | if (num_copies > 1) { |
7870d082 JB |
2125 | repair_io_failure(fs_info, ino, start, failrec->len, |
2126 | failrec->logical, page, pg_offset, | |
2127 | failrec->failed_mirror); | |
4a54c8c1 JS |
2128 | } |
2129 | } | |
2130 | ||
2131 | out: | |
7870d082 | 2132 | free_io_failure(failure_tree, io_tree, failrec); |
4a54c8c1 | 2133 | |
454ff3de | 2134 | return 0; |
4a54c8c1 JS |
2135 | } |
2136 | ||
f612496b MX |
2137 | /* |
2138 | * Can be called when | |
2139 | * - hold extent lock | |
2140 | * - under ordered extent | |
2141 | * - the inode is freeing | |
2142 | */ | |
7ab7956e | 2143 | void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end) |
f612496b | 2144 | { |
7ab7956e | 2145 | struct extent_io_tree *failure_tree = &inode->io_failure_tree; |
f612496b MX |
2146 | struct io_failure_record *failrec; |
2147 | struct extent_state *state, *next; | |
2148 | ||
2149 | if (RB_EMPTY_ROOT(&failure_tree->state)) | |
2150 | return; | |
2151 | ||
2152 | spin_lock(&failure_tree->lock); | |
2153 | state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY); | |
2154 | while (state) { | |
2155 | if (state->start > end) | |
2156 | break; | |
2157 | ||
2158 | ASSERT(state->end <= end); | |
2159 | ||
2160 | next = next_state(state); | |
2161 | ||
47dc196a | 2162 | failrec = state->failrec; |
f612496b MX |
2163 | free_extent_state(state); |
2164 | kfree(failrec); | |
2165 | ||
2166 | state = next; | |
2167 | } | |
2168 | spin_unlock(&failure_tree->lock); | |
2169 | } | |
2170 | ||
2fe6303e | 2171 | int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end, |
47dc196a | 2172 | struct io_failure_record **failrec_ret) |
4a54c8c1 | 2173 | { |
ab8d0fc4 | 2174 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e | 2175 | struct io_failure_record *failrec; |
4a54c8c1 | 2176 | struct extent_map *em; |
4a54c8c1 JS |
2177 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2178 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2179 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
4a54c8c1 | 2180 | int ret; |
4a54c8c1 JS |
2181 | u64 logical; |
2182 | ||
47dc196a | 2183 | ret = get_state_failrec(failure_tree, start, &failrec); |
4a54c8c1 JS |
2184 | if (ret) { |
2185 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2186 | if (!failrec) | |
2187 | return -ENOMEM; | |
2fe6303e | 2188 | |
4a54c8c1 JS |
2189 | failrec->start = start; |
2190 | failrec->len = end - start + 1; | |
2191 | failrec->this_mirror = 0; | |
2192 | failrec->bio_flags = 0; | |
2193 | failrec->in_validation = 0; | |
2194 | ||
2195 | read_lock(&em_tree->lock); | |
2196 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2197 | if (!em) { | |
2198 | read_unlock(&em_tree->lock); | |
2199 | kfree(failrec); | |
2200 | return -EIO; | |
2201 | } | |
2202 | ||
68ba990f | 2203 | if (em->start > start || em->start + em->len <= start) { |
4a54c8c1 JS |
2204 | free_extent_map(em); |
2205 | em = NULL; | |
2206 | } | |
2207 | read_unlock(&em_tree->lock); | |
7a2d6a64 | 2208 | if (!em) { |
4a54c8c1 JS |
2209 | kfree(failrec); |
2210 | return -EIO; | |
2211 | } | |
2fe6303e | 2212 | |
4a54c8c1 JS |
2213 | logical = start - em->start; |
2214 | logical = em->block_start + logical; | |
2215 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2216 | logical = em->block_start; | |
2217 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2218 | extent_set_compress_type(&failrec->bio_flags, | |
2219 | em->compress_type); | |
2220 | } | |
2fe6303e | 2221 | |
ab8d0fc4 JM |
2222 | btrfs_debug(fs_info, |
2223 | "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu", | |
2224 | logical, start, failrec->len); | |
2fe6303e | 2225 | |
4a54c8c1 JS |
2226 | failrec->logical = logical; |
2227 | free_extent_map(em); | |
2228 | ||
2229 | /* set the bits in the private failure tree */ | |
2230 | ret = set_extent_bits(failure_tree, start, end, | |
ceeb0ae7 | 2231 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 | 2232 | if (ret >= 0) |
47dc196a | 2233 | ret = set_state_failrec(failure_tree, start, failrec); |
4a54c8c1 JS |
2234 | /* set the bits in the inode's tree */ |
2235 | if (ret >= 0) | |
ceeb0ae7 | 2236 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED); |
4a54c8c1 JS |
2237 | if (ret < 0) { |
2238 | kfree(failrec); | |
2239 | return ret; | |
2240 | } | |
2241 | } else { | |
ab8d0fc4 JM |
2242 | btrfs_debug(fs_info, |
2243 | "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d", | |
2244 | failrec->logical, failrec->start, failrec->len, | |
2245 | failrec->in_validation); | |
4a54c8c1 JS |
2246 | /* |
2247 | * when data can be on disk more than twice, add to failrec here | |
2248 | * (e.g. with a list for failed_mirror) to make | |
2249 | * clean_io_failure() clean all those errors at once. | |
2250 | */ | |
2251 | } | |
2fe6303e MX |
2252 | |
2253 | *failrec_ret = failrec; | |
2254 | ||
2255 | return 0; | |
2256 | } | |
2257 | ||
c3cfb656 | 2258 | bool btrfs_check_repairable(struct inode *inode, struct bio *failed_bio, |
2fe6303e MX |
2259 | struct io_failure_record *failrec, int failed_mirror) |
2260 | { | |
ab8d0fc4 | 2261 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e MX |
2262 | int num_copies; |
2263 | ||
ab8d0fc4 | 2264 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
4a54c8c1 JS |
2265 | if (num_copies == 1) { |
2266 | /* | |
2267 | * we only have a single copy of the data, so don't bother with | |
2268 | * all the retry and error correction code that follows. no | |
2269 | * matter what the error is, it is very likely to persist. | |
2270 | */ | |
ab8d0fc4 JM |
2271 | btrfs_debug(fs_info, |
2272 | "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
2273 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2274 | return false; |
4a54c8c1 JS |
2275 | } |
2276 | ||
4a54c8c1 JS |
2277 | /* |
2278 | * there are two premises: | |
2279 | * a) deliver good data to the caller | |
2280 | * b) correct the bad sectors on disk | |
2281 | */ | |
2282 | if (failed_bio->bi_vcnt > 1) { | |
2283 | /* | |
2284 | * to fulfill b), we need to know the exact failing sectors, as | |
2285 | * we don't want to rewrite any more than the failed ones. thus, | |
2286 | * we need separate read requests for the failed bio | |
2287 | * | |
2288 | * if the following BUG_ON triggers, our validation request got | |
2289 | * merged. we need separate requests for our algorithm to work. | |
2290 | */ | |
2291 | BUG_ON(failrec->in_validation); | |
2292 | failrec->in_validation = 1; | |
2293 | failrec->this_mirror = failed_mirror; | |
4a54c8c1 JS |
2294 | } else { |
2295 | /* | |
2296 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2297 | * of the failed sector and if we succeed, we have setup | |
2298 | * everything for repair_io_failure to do the rest for us. | |
2299 | */ | |
2300 | if (failrec->in_validation) { | |
2301 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2302 | failrec->in_validation = 0; | |
2303 | failrec->this_mirror = 0; | |
2304 | } | |
2305 | failrec->failed_mirror = failed_mirror; | |
2306 | failrec->this_mirror++; | |
2307 | if (failrec->this_mirror == failed_mirror) | |
2308 | failrec->this_mirror++; | |
4a54c8c1 JS |
2309 | } |
2310 | ||
facc8a22 | 2311 | if (failrec->this_mirror > num_copies) { |
ab8d0fc4 JM |
2312 | btrfs_debug(fs_info, |
2313 | "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
2314 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2315 | return false; |
4a54c8c1 JS |
2316 | } |
2317 | ||
c3cfb656 | 2318 | return true; |
2fe6303e MX |
2319 | } |
2320 | ||
2321 | ||
2322 | struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio, | |
2323 | struct io_failure_record *failrec, | |
2324 | struct page *page, int pg_offset, int icsum, | |
8b110e39 | 2325 | bio_end_io_t *endio_func, void *data) |
2fe6303e | 2326 | { |
0b246afa | 2327 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e MX |
2328 | struct bio *bio; |
2329 | struct btrfs_io_bio *btrfs_failed_bio; | |
2330 | struct btrfs_io_bio *btrfs_bio; | |
2331 | ||
c5e4c3d7 | 2332 | bio = btrfs_io_bio_alloc(1); |
2fe6303e | 2333 | bio->bi_end_io = endio_func; |
4f024f37 | 2334 | bio->bi_iter.bi_sector = failrec->logical >> 9; |
74d46992 | 2335 | bio_set_dev(bio, fs_info->fs_devices->latest_bdev); |
4f024f37 | 2336 | bio->bi_iter.bi_size = 0; |
8b110e39 | 2337 | bio->bi_private = data; |
4a54c8c1 | 2338 | |
facc8a22 MX |
2339 | btrfs_failed_bio = btrfs_io_bio(failed_bio); |
2340 | if (btrfs_failed_bio->csum) { | |
facc8a22 MX |
2341 | u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); |
2342 | ||
2343 | btrfs_bio = btrfs_io_bio(bio); | |
2344 | btrfs_bio->csum = btrfs_bio->csum_inline; | |
2fe6303e MX |
2345 | icsum *= csum_size; |
2346 | memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum, | |
facc8a22 MX |
2347 | csum_size); |
2348 | } | |
2349 | ||
2fe6303e MX |
2350 | bio_add_page(bio, page, failrec->len, pg_offset); |
2351 | ||
2352 | return bio; | |
2353 | } | |
2354 | ||
2355 | /* | |
2356 | * this is a generic handler for readpage errors (default | |
2357 | * readpage_io_failed_hook). if other copies exist, read those and write back | |
2358 | * good data to the failed position. does not investigate in remapping the | |
2359 | * failed extent elsewhere, hoping the device will be smart enough to do this as | |
2360 | * needed | |
2361 | */ | |
2362 | ||
2363 | static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset, | |
2364 | struct page *page, u64 start, u64 end, | |
2365 | int failed_mirror) | |
2366 | { | |
2367 | struct io_failure_record *failrec; | |
2368 | struct inode *inode = page->mapping->host; | |
2369 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
7870d082 | 2370 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2fe6303e | 2371 | struct bio *bio; |
70fd7614 | 2372 | int read_mode = 0; |
4e4cbee9 | 2373 | blk_status_t status; |
2fe6303e MX |
2374 | int ret; |
2375 | ||
1f7ad75b | 2376 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
2fe6303e MX |
2377 | |
2378 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
2379 | if (ret) | |
2380 | return ret; | |
2381 | ||
c3cfb656 LB |
2382 | if (!btrfs_check_repairable(inode, failed_bio, failrec, |
2383 | failed_mirror)) { | |
7870d082 | 2384 | free_io_failure(failure_tree, tree, failrec); |
2fe6303e MX |
2385 | return -EIO; |
2386 | } | |
2387 | ||
2388 | if (failed_bio->bi_vcnt > 1) | |
70fd7614 | 2389 | read_mode |= REQ_FAILFAST_DEV; |
2fe6303e MX |
2390 | |
2391 | phy_offset >>= inode->i_sb->s_blocksize_bits; | |
2392 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2393 | start - page_offset(page), | |
8b110e39 MX |
2394 | (int)phy_offset, failed_bio->bi_end_io, |
2395 | NULL); | |
1f7ad75b | 2396 | bio_set_op_attrs(bio, REQ_OP_READ, read_mode); |
4a54c8c1 | 2397 | |
ab8d0fc4 JM |
2398 | btrfs_debug(btrfs_sb(inode->i_sb), |
2399 | "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d", | |
2400 | read_mode, failrec->this_mirror, failrec->in_validation); | |
4a54c8c1 | 2401 | |
8c27cb35 | 2402 | status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror, |
013bd4c3 | 2403 | failrec->bio_flags, 0); |
4e4cbee9 | 2404 | if (status) { |
7870d082 | 2405 | free_io_failure(failure_tree, tree, failrec); |
6c387ab2 | 2406 | bio_put(bio); |
4e4cbee9 | 2407 | ret = blk_status_to_errno(status); |
6c387ab2 MX |
2408 | } |
2409 | ||
013bd4c3 | 2410 | return ret; |
4a54c8c1 JS |
2411 | } |
2412 | ||
d1310b2e CM |
2413 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2414 | ||
b5227c07 | 2415 | void end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
87826df0 JM |
2416 | { |
2417 | int uptodate = (err == 0); | |
2418 | struct extent_io_tree *tree; | |
3e2426bd | 2419 | int ret = 0; |
87826df0 JM |
2420 | |
2421 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
2422 | ||
c3988d63 DS |
2423 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2424 | tree->ops->writepage_end_io_hook(page, start, end, NULL, | |
2425 | uptodate); | |
87826df0 | 2426 | |
87826df0 | 2427 | if (!uptodate) { |
87826df0 JM |
2428 | ClearPageUptodate(page); |
2429 | SetPageError(page); | |
bff5baf8 | 2430 | ret = err < 0 ? err : -EIO; |
5dca6eea | 2431 | mapping_set_error(page->mapping, ret); |
87826df0 | 2432 | } |
87826df0 JM |
2433 | } |
2434 | ||
d1310b2e CM |
2435 | /* |
2436 | * after a writepage IO is done, we need to: | |
2437 | * clear the uptodate bits on error | |
2438 | * clear the writeback bits in the extent tree for this IO | |
2439 | * end_page_writeback if the page has no more pending IO | |
2440 | * | |
2441 | * Scheduling is not allowed, so the extent state tree is expected | |
2442 | * to have one and only one object corresponding to this IO. | |
2443 | */ | |
4246a0b6 | 2444 | static void end_bio_extent_writepage(struct bio *bio) |
d1310b2e | 2445 | { |
4e4cbee9 | 2446 | int error = blk_status_to_errno(bio->bi_status); |
2c30c71b | 2447 | struct bio_vec *bvec; |
d1310b2e CM |
2448 | u64 start; |
2449 | u64 end; | |
2c30c71b | 2450 | int i; |
d1310b2e | 2451 | |
c09abff8 | 2452 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2c30c71b | 2453 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2454 | struct page *page = bvec->bv_page; |
0b246afa JM |
2455 | struct inode *inode = page->mapping->host; |
2456 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
902b22f3 | 2457 | |
17a5adcc AO |
2458 | /* We always issue full-page reads, but if some block |
2459 | * in a page fails to read, blk_update_request() will | |
2460 | * advance bv_offset and adjust bv_len to compensate. | |
2461 | * Print a warning for nonzero offsets, and an error | |
2462 | * if they don't add up to a full page. */ | |
09cbfeaf KS |
2463 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { |
2464 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
0b246afa | 2465 | btrfs_err(fs_info, |
efe120a0 FH |
2466 | "partial page write in btrfs with offset %u and length %u", |
2467 | bvec->bv_offset, bvec->bv_len); | |
2468 | else | |
0b246afa | 2469 | btrfs_info(fs_info, |
5d163e0e | 2470 | "incomplete page write in btrfs with offset %u and length %u", |
efe120a0 FH |
2471 | bvec->bv_offset, bvec->bv_len); |
2472 | } | |
d1310b2e | 2473 | |
17a5adcc AO |
2474 | start = page_offset(page); |
2475 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
d1310b2e | 2476 | |
4e4cbee9 | 2477 | end_extent_writepage(page, error, start, end); |
17a5adcc | 2478 | end_page_writeback(page); |
2c30c71b | 2479 | } |
2b1f55b0 | 2480 | |
d1310b2e | 2481 | bio_put(bio); |
d1310b2e CM |
2482 | } |
2483 | ||
883d0de4 MX |
2484 | static void |
2485 | endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len, | |
2486 | int uptodate) | |
2487 | { | |
2488 | struct extent_state *cached = NULL; | |
2489 | u64 end = start + len - 1; | |
2490 | ||
2491 | if (uptodate && tree->track_uptodate) | |
2492 | set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC); | |
2493 | unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC); | |
2494 | } | |
2495 | ||
d1310b2e CM |
2496 | /* |
2497 | * after a readpage IO is done, we need to: | |
2498 | * clear the uptodate bits on error | |
2499 | * set the uptodate bits if things worked | |
2500 | * set the page up to date if all extents in the tree are uptodate | |
2501 | * clear the lock bit in the extent tree | |
2502 | * unlock the page if there are no other extents locked for it | |
2503 | * | |
2504 | * Scheduling is not allowed, so the extent state tree is expected | |
2505 | * to have one and only one object corresponding to this IO. | |
2506 | */ | |
4246a0b6 | 2507 | static void end_bio_extent_readpage(struct bio *bio) |
d1310b2e | 2508 | { |
2c30c71b | 2509 | struct bio_vec *bvec; |
4e4cbee9 | 2510 | int uptodate = !bio->bi_status; |
facc8a22 | 2511 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); |
7870d082 | 2512 | struct extent_io_tree *tree, *failure_tree; |
facc8a22 | 2513 | u64 offset = 0; |
d1310b2e CM |
2514 | u64 start; |
2515 | u64 end; | |
facc8a22 | 2516 | u64 len; |
883d0de4 MX |
2517 | u64 extent_start = 0; |
2518 | u64 extent_len = 0; | |
5cf1ab56 | 2519 | int mirror; |
d1310b2e | 2520 | int ret; |
2c30c71b | 2521 | int i; |
d1310b2e | 2522 | |
c09abff8 | 2523 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2c30c71b | 2524 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2525 | struct page *page = bvec->bv_page; |
a71754fc | 2526 | struct inode *inode = page->mapping->host; |
ab8d0fc4 | 2527 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
507903b8 | 2528 | |
ab8d0fc4 JM |
2529 | btrfs_debug(fs_info, |
2530 | "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u", | |
4e4cbee9 | 2531 | (u64)bio->bi_iter.bi_sector, bio->bi_status, |
ab8d0fc4 | 2532 | io_bio->mirror_num); |
a71754fc | 2533 | tree = &BTRFS_I(inode)->io_tree; |
7870d082 | 2534 | failure_tree = &BTRFS_I(inode)->io_failure_tree; |
902b22f3 | 2535 | |
17a5adcc AO |
2536 | /* We always issue full-page reads, but if some block |
2537 | * in a page fails to read, blk_update_request() will | |
2538 | * advance bv_offset and adjust bv_len to compensate. | |
2539 | * Print a warning for nonzero offsets, and an error | |
2540 | * if they don't add up to a full page. */ | |
09cbfeaf KS |
2541 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { |
2542 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
ab8d0fc4 JM |
2543 | btrfs_err(fs_info, |
2544 | "partial page read in btrfs with offset %u and length %u", | |
efe120a0 FH |
2545 | bvec->bv_offset, bvec->bv_len); |
2546 | else | |
ab8d0fc4 JM |
2547 | btrfs_info(fs_info, |
2548 | "incomplete page read in btrfs with offset %u and length %u", | |
efe120a0 FH |
2549 | bvec->bv_offset, bvec->bv_len); |
2550 | } | |
d1310b2e | 2551 | |
17a5adcc AO |
2552 | start = page_offset(page); |
2553 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
facc8a22 | 2554 | len = bvec->bv_len; |
d1310b2e | 2555 | |
9be3395b | 2556 | mirror = io_bio->mirror_num; |
20c9801d | 2557 | if (likely(uptodate && tree->ops)) { |
facc8a22 MX |
2558 | ret = tree->ops->readpage_end_io_hook(io_bio, offset, |
2559 | page, start, end, | |
2560 | mirror); | |
5ee0844d | 2561 | if (ret) |
d1310b2e | 2562 | uptodate = 0; |
5ee0844d | 2563 | else |
7870d082 JB |
2564 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
2565 | failure_tree, tree, start, | |
2566 | page, | |
2567 | btrfs_ino(BTRFS_I(inode)), 0); | |
d1310b2e | 2568 | } |
ea466794 | 2569 | |
f2a09da9 MX |
2570 | if (likely(uptodate)) |
2571 | goto readpage_ok; | |
2572 | ||
20a7db8a | 2573 | if (tree->ops) { |
5cf1ab56 | 2574 | ret = tree->ops->readpage_io_failed_hook(page, mirror); |
9d0d1c8b LB |
2575 | if (ret == -EAGAIN) { |
2576 | /* | |
2577 | * Data inode's readpage_io_failed_hook() always | |
2578 | * returns -EAGAIN. | |
2579 | * | |
2580 | * The generic bio_readpage_error handles errors | |
2581 | * the following way: If possible, new read | |
2582 | * requests are created and submitted and will | |
2583 | * end up in end_bio_extent_readpage as well (if | |
2584 | * we're lucky, not in the !uptodate case). In | |
2585 | * that case it returns 0 and we just go on with | |
2586 | * the next page in our bio. If it can't handle | |
2587 | * the error it will return -EIO and we remain | |
2588 | * responsible for that page. | |
2589 | */ | |
2590 | ret = bio_readpage_error(bio, offset, page, | |
2591 | start, end, mirror); | |
2592 | if (ret == 0) { | |
4e4cbee9 | 2593 | uptodate = !bio->bi_status; |
9d0d1c8b LB |
2594 | offset += len; |
2595 | continue; | |
2596 | } | |
2597 | } | |
2598 | ||
f4a8e656 | 2599 | /* |
9d0d1c8b LB |
2600 | * metadata's readpage_io_failed_hook() always returns |
2601 | * -EIO and fixes nothing. -EIO is also returned if | |
2602 | * data inode error could not be fixed. | |
f4a8e656 | 2603 | */ |
9d0d1c8b | 2604 | ASSERT(ret == -EIO); |
7e38326f | 2605 | } |
f2a09da9 | 2606 | readpage_ok: |
883d0de4 | 2607 | if (likely(uptodate)) { |
a71754fc | 2608 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 2609 | pgoff_t end_index = i_size >> PAGE_SHIFT; |
a583c026 | 2610 | unsigned off; |
a71754fc JB |
2611 | |
2612 | /* Zero out the end if this page straddles i_size */ | |
09cbfeaf | 2613 | off = i_size & (PAGE_SIZE-1); |
a583c026 | 2614 | if (page->index == end_index && off) |
09cbfeaf | 2615 | zero_user_segment(page, off, PAGE_SIZE); |
17a5adcc | 2616 | SetPageUptodate(page); |
70dec807 | 2617 | } else { |
17a5adcc AO |
2618 | ClearPageUptodate(page); |
2619 | SetPageError(page); | |
70dec807 | 2620 | } |
17a5adcc | 2621 | unlock_page(page); |
facc8a22 | 2622 | offset += len; |
883d0de4 MX |
2623 | |
2624 | if (unlikely(!uptodate)) { | |
2625 | if (extent_len) { | |
2626 | endio_readpage_release_extent(tree, | |
2627 | extent_start, | |
2628 | extent_len, 1); | |
2629 | extent_start = 0; | |
2630 | extent_len = 0; | |
2631 | } | |
2632 | endio_readpage_release_extent(tree, start, | |
2633 | end - start + 1, 0); | |
2634 | } else if (!extent_len) { | |
2635 | extent_start = start; | |
2636 | extent_len = end + 1 - start; | |
2637 | } else if (extent_start + extent_len == start) { | |
2638 | extent_len += end + 1 - start; | |
2639 | } else { | |
2640 | endio_readpage_release_extent(tree, extent_start, | |
2641 | extent_len, uptodate); | |
2642 | extent_start = start; | |
2643 | extent_len = end + 1 - start; | |
2644 | } | |
2c30c71b | 2645 | } |
d1310b2e | 2646 | |
883d0de4 MX |
2647 | if (extent_len) |
2648 | endio_readpage_release_extent(tree, extent_start, extent_len, | |
2649 | uptodate); | |
facc8a22 | 2650 | if (io_bio->end_io) |
4e4cbee9 | 2651 | io_bio->end_io(io_bio, blk_status_to_errno(bio->bi_status)); |
d1310b2e | 2652 | bio_put(bio); |
d1310b2e CM |
2653 | } |
2654 | ||
9be3395b | 2655 | /* |
184f999e DS |
2656 | * Initialize the members up to but not including 'bio'. Use after allocating a |
2657 | * new bio by bio_alloc_bioset as it does not initialize the bytes outside of | |
2658 | * 'bio' because use of __GFP_ZERO is not supported. | |
9be3395b | 2659 | */ |
184f999e | 2660 | static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio) |
d1310b2e | 2661 | { |
184f999e DS |
2662 | memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio)); |
2663 | } | |
d1310b2e | 2664 | |
9be3395b | 2665 | /* |
6e707bcd DS |
2666 | * The following helpers allocate a bio. As it's backed by a bioset, it'll |
2667 | * never fail. We're returning a bio right now but you can call btrfs_io_bio | |
2668 | * for the appropriate container_of magic | |
9be3395b | 2669 | */ |
c821e7f3 | 2670 | struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte) |
d1310b2e CM |
2671 | { |
2672 | struct bio *bio; | |
d1310b2e | 2673 | |
9f2179a5 | 2674 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, btrfs_bioset); |
74d46992 | 2675 | bio_set_dev(bio, bdev); |
c821e7f3 | 2676 | bio->bi_iter.bi_sector = first_byte >> 9; |
184f999e | 2677 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
d1310b2e CM |
2678 | return bio; |
2679 | } | |
2680 | ||
8b6c1d56 | 2681 | struct bio *btrfs_bio_clone(struct bio *bio) |
9be3395b | 2682 | { |
23ea8e5a MX |
2683 | struct btrfs_io_bio *btrfs_bio; |
2684 | struct bio *new; | |
9be3395b | 2685 | |
6e707bcd | 2686 | /* Bio allocation backed by a bioset does not fail */ |
8b6c1d56 | 2687 | new = bio_clone_fast(bio, GFP_NOFS, btrfs_bioset); |
6e707bcd | 2688 | btrfs_bio = btrfs_io_bio(new); |
184f999e | 2689 | btrfs_io_bio_init(btrfs_bio); |
6e707bcd | 2690 | btrfs_bio->iter = bio->bi_iter; |
23ea8e5a MX |
2691 | return new; |
2692 | } | |
9be3395b | 2693 | |
c5e4c3d7 | 2694 | struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs) |
9be3395b | 2695 | { |
facc8a22 MX |
2696 | struct bio *bio; |
2697 | ||
6e707bcd | 2698 | /* Bio allocation backed by a bioset does not fail */ |
c5e4c3d7 | 2699 | bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, btrfs_bioset); |
184f999e | 2700 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
facc8a22 | 2701 | return bio; |
9be3395b CM |
2702 | } |
2703 | ||
e477094f | 2704 | struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size) |
2f8e9140 LB |
2705 | { |
2706 | struct bio *bio; | |
2707 | struct btrfs_io_bio *btrfs_bio; | |
2708 | ||
2709 | /* this will never fail when it's backed by a bioset */ | |
e477094f | 2710 | bio = bio_clone_fast(orig, GFP_NOFS, btrfs_bioset); |
2f8e9140 LB |
2711 | ASSERT(bio); |
2712 | ||
2713 | btrfs_bio = btrfs_io_bio(bio); | |
184f999e | 2714 | btrfs_io_bio_init(btrfs_bio); |
2f8e9140 LB |
2715 | |
2716 | bio_trim(bio, offset >> 9, size >> 9); | |
17347cec | 2717 | btrfs_bio->iter = bio->bi_iter; |
2f8e9140 LB |
2718 | return bio; |
2719 | } | |
9be3395b | 2720 | |
1f7ad75b MC |
2721 | static int __must_check submit_one_bio(struct bio *bio, int mirror_num, |
2722 | unsigned long bio_flags) | |
d1310b2e | 2723 | { |
4e4cbee9 | 2724 | blk_status_t ret = 0; |
70dec807 CM |
2725 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
2726 | struct page *page = bvec->bv_page; | |
2727 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 2728 | u64 start; |
70dec807 | 2729 | |
4eee4fa4 | 2730 | start = page_offset(page) + bvec->bv_offset; |
70dec807 | 2731 | |
902b22f3 | 2732 | bio->bi_private = NULL; |
d1310b2e CM |
2733 | bio_get(bio); |
2734 | ||
20c9801d | 2735 | if (tree->ops) |
c6100a4b | 2736 | ret = tree->ops->submit_bio_hook(tree->private_data, bio, |
eaf25d93 | 2737 | mirror_num, bio_flags, start); |
0b86a832 | 2738 | else |
4e49ea4a | 2739 | btrfsic_submit_bio(bio); |
4a54c8c1 | 2740 | |
d1310b2e | 2741 | bio_put(bio); |
4e4cbee9 | 2742 | return blk_status_to_errno(ret); |
d1310b2e CM |
2743 | } |
2744 | ||
1f7ad75b | 2745 | static int merge_bio(struct extent_io_tree *tree, struct page *page, |
3444a972 JM |
2746 | unsigned long offset, size_t size, struct bio *bio, |
2747 | unsigned long bio_flags) | |
2748 | { | |
2749 | int ret = 0; | |
20c9801d | 2750 | if (tree->ops) |
81a75f67 | 2751 | ret = tree->ops->merge_bio_hook(page, offset, size, bio, |
3444a972 | 2752 | bio_flags); |
3444a972 JM |
2753 | return ret; |
2754 | ||
2755 | } | |
2756 | ||
4b81ba48 DS |
2757 | /* |
2758 | * @opf: bio REQ_OP_* and REQ_* flags as one value | |
2759 | */ | |
2760 | static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree, | |
da2f0f74 | 2761 | struct writeback_control *wbc, |
6273b7f8 | 2762 | struct page *page, u64 offset, |
6c5a4e2c | 2763 | size_t size, unsigned long pg_offset, |
d1310b2e CM |
2764 | struct block_device *bdev, |
2765 | struct bio **bio_ret, | |
f188591e | 2766 | bio_end_io_t end_io_func, |
c8b97818 CM |
2767 | int mirror_num, |
2768 | unsigned long prev_bio_flags, | |
005efedf FM |
2769 | unsigned long bio_flags, |
2770 | bool force_bio_submit) | |
d1310b2e CM |
2771 | { |
2772 | int ret = 0; | |
2773 | struct bio *bio; | |
c8b97818 | 2774 | int contig = 0; |
c8b97818 | 2775 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; |
09cbfeaf | 2776 | size_t page_size = min_t(size_t, size, PAGE_SIZE); |
6273b7f8 | 2777 | sector_t sector = offset >> 9; |
d1310b2e CM |
2778 | |
2779 | if (bio_ret && *bio_ret) { | |
2780 | bio = *bio_ret; | |
c8b97818 | 2781 | if (old_compressed) |
4f024f37 | 2782 | contig = bio->bi_iter.bi_sector == sector; |
c8b97818 | 2783 | else |
f73a1c7d | 2784 | contig = bio_end_sector(bio) == sector; |
c8b97818 CM |
2785 | |
2786 | if (prev_bio_flags != bio_flags || !contig || | |
005efedf | 2787 | force_bio_submit || |
6c5a4e2c DS |
2788 | merge_bio(tree, page, pg_offset, page_size, bio, bio_flags) || |
2789 | bio_add_page(bio, page, page_size, pg_offset) < page_size) { | |
1f7ad75b | 2790 | ret = submit_one_bio(bio, mirror_num, prev_bio_flags); |
289454ad NA |
2791 | if (ret < 0) { |
2792 | *bio_ret = NULL; | |
79787eaa | 2793 | return ret; |
289454ad | 2794 | } |
d1310b2e CM |
2795 | bio = NULL; |
2796 | } else { | |
da2f0f74 CM |
2797 | if (wbc) |
2798 | wbc_account_io(wbc, page, page_size); | |
d1310b2e CM |
2799 | return 0; |
2800 | } | |
2801 | } | |
c8b97818 | 2802 | |
6273b7f8 | 2803 | bio = btrfs_bio_alloc(bdev, offset); |
6c5a4e2c | 2804 | bio_add_page(bio, page, page_size, pg_offset); |
d1310b2e CM |
2805 | bio->bi_end_io = end_io_func; |
2806 | bio->bi_private = tree; | |
e6959b93 | 2807 | bio->bi_write_hint = page->mapping->host->i_write_hint; |
4b81ba48 | 2808 | bio->bi_opf = opf; |
da2f0f74 CM |
2809 | if (wbc) { |
2810 | wbc_init_bio(wbc, bio); | |
2811 | wbc_account_io(wbc, page, page_size); | |
2812 | } | |
70dec807 | 2813 | |
d397712b | 2814 | if (bio_ret) |
d1310b2e | 2815 | *bio_ret = bio; |
d397712b | 2816 | else |
1f7ad75b | 2817 | ret = submit_one_bio(bio, mirror_num, bio_flags); |
d1310b2e CM |
2818 | |
2819 | return ret; | |
2820 | } | |
2821 | ||
48a3b636 ES |
2822 | static void attach_extent_buffer_page(struct extent_buffer *eb, |
2823 | struct page *page) | |
d1310b2e CM |
2824 | { |
2825 | if (!PagePrivate(page)) { | |
2826 | SetPagePrivate(page); | |
09cbfeaf | 2827 | get_page(page); |
4f2de97a JB |
2828 | set_page_private(page, (unsigned long)eb); |
2829 | } else { | |
2830 | WARN_ON(page->private != (unsigned long)eb); | |
d1310b2e CM |
2831 | } |
2832 | } | |
2833 | ||
4f2de97a | 2834 | void set_page_extent_mapped(struct page *page) |
d1310b2e | 2835 | { |
4f2de97a JB |
2836 | if (!PagePrivate(page)) { |
2837 | SetPagePrivate(page); | |
09cbfeaf | 2838 | get_page(page); |
4f2de97a JB |
2839 | set_page_private(page, EXTENT_PAGE_PRIVATE); |
2840 | } | |
d1310b2e CM |
2841 | } |
2842 | ||
125bac01 MX |
2843 | static struct extent_map * |
2844 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
2845 | u64 start, u64 len, get_extent_t *get_extent, | |
2846 | struct extent_map **em_cached) | |
2847 | { | |
2848 | struct extent_map *em; | |
2849 | ||
2850 | if (em_cached && *em_cached) { | |
2851 | em = *em_cached; | |
cbc0e928 | 2852 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 | 2853 | start < extent_map_end(em)) { |
490b54d6 | 2854 | refcount_inc(&em->refs); |
125bac01 MX |
2855 | return em; |
2856 | } | |
2857 | ||
2858 | free_extent_map(em); | |
2859 | *em_cached = NULL; | |
2860 | } | |
2861 | ||
fc4f21b1 | 2862 | em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0); |
125bac01 MX |
2863 | if (em_cached && !IS_ERR_OR_NULL(em)) { |
2864 | BUG_ON(*em_cached); | |
490b54d6 | 2865 | refcount_inc(&em->refs); |
125bac01 MX |
2866 | *em_cached = em; |
2867 | } | |
2868 | return em; | |
2869 | } | |
d1310b2e CM |
2870 | /* |
2871 | * basic readpage implementation. Locked extent state structs are inserted | |
2872 | * into the tree that are removed when the IO is done (by the end_io | |
2873 | * handlers) | |
79787eaa | 2874 | * XXX JDM: This needs looking at to ensure proper page locking |
baf863b9 | 2875 | * return 0 on success, otherwise return error |
d1310b2e | 2876 | */ |
9974090b MX |
2877 | static int __do_readpage(struct extent_io_tree *tree, |
2878 | struct page *page, | |
2879 | get_extent_t *get_extent, | |
125bac01 | 2880 | struct extent_map **em_cached, |
9974090b | 2881 | struct bio **bio, int mirror_num, |
f1c77c55 | 2882 | unsigned long *bio_flags, unsigned int read_flags, |
005efedf | 2883 | u64 *prev_em_start) |
d1310b2e CM |
2884 | { |
2885 | struct inode *inode = page->mapping->host; | |
4eee4fa4 | 2886 | u64 start = page_offset(page); |
09cbfeaf | 2887 | u64 page_end = start + PAGE_SIZE - 1; |
d1310b2e CM |
2888 | u64 end; |
2889 | u64 cur = start; | |
2890 | u64 extent_offset; | |
2891 | u64 last_byte = i_size_read(inode); | |
2892 | u64 block_start; | |
2893 | u64 cur_end; | |
d1310b2e CM |
2894 | struct extent_map *em; |
2895 | struct block_device *bdev; | |
baf863b9 | 2896 | int ret = 0; |
d1310b2e | 2897 | int nr = 0; |
306e16ce | 2898 | size_t pg_offset = 0; |
d1310b2e | 2899 | size_t iosize; |
c8b97818 | 2900 | size_t disk_io_size; |
d1310b2e | 2901 | size_t blocksize = inode->i_sb->s_blocksize; |
7f042a83 | 2902 | unsigned long this_bio_flag = 0; |
d1310b2e CM |
2903 | |
2904 | set_page_extent_mapped(page); | |
2905 | ||
9974090b | 2906 | end = page_end; |
90a887c9 DM |
2907 | if (!PageUptodate(page)) { |
2908 | if (cleancache_get_page(page) == 0) { | |
2909 | BUG_ON(blocksize != PAGE_SIZE); | |
9974090b | 2910 | unlock_extent(tree, start, end); |
90a887c9 DM |
2911 | goto out; |
2912 | } | |
2913 | } | |
2914 | ||
09cbfeaf | 2915 | if (page->index == last_byte >> PAGE_SHIFT) { |
c8b97818 | 2916 | char *userpage; |
09cbfeaf | 2917 | size_t zero_offset = last_byte & (PAGE_SIZE - 1); |
c8b97818 CM |
2918 | |
2919 | if (zero_offset) { | |
09cbfeaf | 2920 | iosize = PAGE_SIZE - zero_offset; |
7ac687d9 | 2921 | userpage = kmap_atomic(page); |
c8b97818 CM |
2922 | memset(userpage + zero_offset, 0, iosize); |
2923 | flush_dcache_page(page); | |
7ac687d9 | 2924 | kunmap_atomic(userpage); |
c8b97818 CM |
2925 | } |
2926 | } | |
d1310b2e | 2927 | while (cur <= end) { |
005efedf | 2928 | bool force_bio_submit = false; |
6273b7f8 | 2929 | u64 offset; |
c8f2f24b | 2930 | |
d1310b2e CM |
2931 | if (cur >= last_byte) { |
2932 | char *userpage; | |
507903b8 AJ |
2933 | struct extent_state *cached = NULL; |
2934 | ||
09cbfeaf | 2935 | iosize = PAGE_SIZE - pg_offset; |
7ac687d9 | 2936 | userpage = kmap_atomic(page); |
306e16ce | 2937 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2938 | flush_dcache_page(page); |
7ac687d9 | 2939 | kunmap_atomic(userpage); |
d1310b2e | 2940 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 2941 | &cached, GFP_NOFS); |
7f042a83 FM |
2942 | unlock_extent_cached(tree, cur, |
2943 | cur + iosize - 1, | |
2944 | &cached, GFP_NOFS); | |
d1310b2e CM |
2945 | break; |
2946 | } | |
125bac01 MX |
2947 | em = __get_extent_map(inode, page, pg_offset, cur, |
2948 | end - cur + 1, get_extent, em_cached); | |
c704005d | 2949 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 2950 | SetPageError(page); |
7f042a83 | 2951 | unlock_extent(tree, cur, end); |
d1310b2e CM |
2952 | break; |
2953 | } | |
d1310b2e CM |
2954 | extent_offset = cur - em->start; |
2955 | BUG_ON(extent_map_end(em) <= cur); | |
2956 | BUG_ON(end < cur); | |
2957 | ||
261507a0 | 2958 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
4b384318 | 2959 | this_bio_flag |= EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
2960 | extent_set_compress_type(&this_bio_flag, |
2961 | em->compress_type); | |
2962 | } | |
c8b97818 | 2963 | |
d1310b2e CM |
2964 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
2965 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 2966 | iosize = ALIGN(iosize, blocksize); |
c8b97818 CM |
2967 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
2968 | disk_io_size = em->block_len; | |
6273b7f8 | 2969 | offset = em->block_start; |
c8b97818 | 2970 | } else { |
6273b7f8 | 2971 | offset = em->block_start + extent_offset; |
c8b97818 CM |
2972 | disk_io_size = iosize; |
2973 | } | |
d1310b2e CM |
2974 | bdev = em->bdev; |
2975 | block_start = em->block_start; | |
d899e052 YZ |
2976 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2977 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
2978 | |
2979 | /* | |
2980 | * If we have a file range that points to a compressed extent | |
2981 | * and it's followed by a consecutive file range that points to | |
2982 | * to the same compressed extent (possibly with a different | |
2983 | * offset and/or length, so it either points to the whole extent | |
2984 | * or only part of it), we must make sure we do not submit a | |
2985 | * single bio to populate the pages for the 2 ranges because | |
2986 | * this makes the compressed extent read zero out the pages | |
2987 | * belonging to the 2nd range. Imagine the following scenario: | |
2988 | * | |
2989 | * File layout | |
2990 | * [0 - 8K] [8K - 24K] | |
2991 | * | | | |
2992 | * | | | |
2993 | * points to extent X, points to extent X, | |
2994 | * offset 4K, length of 8K offset 0, length 16K | |
2995 | * | |
2996 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
2997 | * | |
2998 | * If the bio to read the compressed extent covers both ranges, | |
2999 | * it will decompress extent X into the pages belonging to the | |
3000 | * first range and then it will stop, zeroing out the remaining | |
3001 | * pages that belong to the other range that points to extent X. | |
3002 | * So here we make sure we submit 2 bios, one for the first | |
3003 | * range and another one for the third range. Both will target | |
3004 | * the same physical extent from disk, but we can't currently | |
3005 | * make the compressed bio endio callback populate the pages | |
3006 | * for both ranges because each compressed bio is tightly | |
3007 | * coupled with a single extent map, and each range can have | |
3008 | * an extent map with a different offset value relative to the | |
3009 | * uncompressed data of our extent and different lengths. This | |
3010 | * is a corner case so we prioritize correctness over | |
3011 | * non-optimal behavior (submitting 2 bios for the same extent). | |
3012 | */ | |
3013 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
3014 | prev_em_start && *prev_em_start != (u64)-1 && | |
3015 | *prev_em_start != em->orig_start) | |
3016 | force_bio_submit = true; | |
3017 | ||
3018 | if (prev_em_start) | |
3019 | *prev_em_start = em->orig_start; | |
3020 | ||
d1310b2e CM |
3021 | free_extent_map(em); |
3022 | em = NULL; | |
3023 | ||
3024 | /* we've found a hole, just zero and go on */ | |
3025 | if (block_start == EXTENT_MAP_HOLE) { | |
3026 | char *userpage; | |
507903b8 AJ |
3027 | struct extent_state *cached = NULL; |
3028 | ||
7ac687d9 | 3029 | userpage = kmap_atomic(page); |
306e16ce | 3030 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3031 | flush_dcache_page(page); |
7ac687d9 | 3032 | kunmap_atomic(userpage); |
d1310b2e CM |
3033 | |
3034 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 | 3035 | &cached, GFP_NOFS); |
7f042a83 FM |
3036 | unlock_extent_cached(tree, cur, |
3037 | cur + iosize - 1, | |
3038 | &cached, GFP_NOFS); | |
d1310b2e | 3039 | cur = cur + iosize; |
306e16ce | 3040 | pg_offset += iosize; |
d1310b2e CM |
3041 | continue; |
3042 | } | |
3043 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
3044 | if (test_range_bit(tree, cur, cur_end, |
3045 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 3046 | check_page_uptodate(tree, page); |
7f042a83 | 3047 | unlock_extent(tree, cur, cur + iosize - 1); |
d1310b2e | 3048 | cur = cur + iosize; |
306e16ce | 3049 | pg_offset += iosize; |
d1310b2e CM |
3050 | continue; |
3051 | } | |
70dec807 CM |
3052 | /* we have an inline extent but it didn't get marked up |
3053 | * to date. Error out | |
3054 | */ | |
3055 | if (block_start == EXTENT_MAP_INLINE) { | |
3056 | SetPageError(page); | |
7f042a83 | 3057 | unlock_extent(tree, cur, cur + iosize - 1); |
70dec807 | 3058 | cur = cur + iosize; |
306e16ce | 3059 | pg_offset += iosize; |
70dec807 CM |
3060 | continue; |
3061 | } | |
d1310b2e | 3062 | |
4b81ba48 | 3063 | ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL, |
6273b7f8 DS |
3064 | page, offset, disk_io_size, |
3065 | pg_offset, bdev, bio, | |
c8b97818 CM |
3066 | end_bio_extent_readpage, mirror_num, |
3067 | *bio_flags, | |
005efedf FM |
3068 | this_bio_flag, |
3069 | force_bio_submit); | |
c8f2f24b JB |
3070 | if (!ret) { |
3071 | nr++; | |
3072 | *bio_flags = this_bio_flag; | |
3073 | } else { | |
d1310b2e | 3074 | SetPageError(page); |
7f042a83 | 3075 | unlock_extent(tree, cur, cur + iosize - 1); |
baf863b9 | 3076 | goto out; |
edd33c99 | 3077 | } |
d1310b2e | 3078 | cur = cur + iosize; |
306e16ce | 3079 | pg_offset += iosize; |
d1310b2e | 3080 | } |
90a887c9 | 3081 | out: |
d1310b2e CM |
3082 | if (!nr) { |
3083 | if (!PageError(page)) | |
3084 | SetPageUptodate(page); | |
3085 | unlock_page(page); | |
3086 | } | |
baf863b9 | 3087 | return ret; |
d1310b2e CM |
3088 | } |
3089 | ||
9974090b MX |
3090 | static inline void __do_contiguous_readpages(struct extent_io_tree *tree, |
3091 | struct page *pages[], int nr_pages, | |
3092 | u64 start, u64 end, | |
3093 | get_extent_t *get_extent, | |
125bac01 | 3094 | struct extent_map **em_cached, |
d3fac6ba | 3095 | struct bio **bio, |
1f7ad75b | 3096 | unsigned long *bio_flags, |
808f80b4 | 3097 | u64 *prev_em_start) |
9974090b MX |
3098 | { |
3099 | struct inode *inode; | |
3100 | struct btrfs_ordered_extent *ordered; | |
3101 | int index; | |
3102 | ||
3103 | inode = pages[0]->mapping->host; | |
3104 | while (1) { | |
3105 | lock_extent(tree, start, end); | |
a776c6fa | 3106 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
9974090b MX |
3107 | end - start + 1); |
3108 | if (!ordered) | |
3109 | break; | |
3110 | unlock_extent(tree, start, end); | |
3111 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3112 | btrfs_put_ordered_extent(ordered); | |
3113 | } | |
3114 | ||
3115 | for (index = 0; index < nr_pages; index++) { | |
125bac01 | 3116 | __do_readpage(tree, pages[index], get_extent, em_cached, bio, |
d3fac6ba | 3117 | 0, bio_flags, 0, prev_em_start); |
09cbfeaf | 3118 | put_page(pages[index]); |
9974090b MX |
3119 | } |
3120 | } | |
3121 | ||
3122 | static void __extent_readpages(struct extent_io_tree *tree, | |
3123 | struct page *pages[], | |
3124 | int nr_pages, get_extent_t *get_extent, | |
125bac01 | 3125 | struct extent_map **em_cached, |
d3fac6ba | 3126 | struct bio **bio, unsigned long *bio_flags, |
808f80b4 | 3127 | u64 *prev_em_start) |
9974090b | 3128 | { |
35a3621b | 3129 | u64 start = 0; |
9974090b MX |
3130 | u64 end = 0; |
3131 | u64 page_start; | |
3132 | int index; | |
35a3621b | 3133 | int first_index = 0; |
9974090b MX |
3134 | |
3135 | for (index = 0; index < nr_pages; index++) { | |
3136 | page_start = page_offset(pages[index]); | |
3137 | if (!end) { | |
3138 | start = page_start; | |
09cbfeaf | 3139 | end = start + PAGE_SIZE - 1; |
9974090b MX |
3140 | first_index = index; |
3141 | } else if (end + 1 == page_start) { | |
09cbfeaf | 3142 | end += PAGE_SIZE; |
9974090b MX |
3143 | } else { |
3144 | __do_contiguous_readpages(tree, &pages[first_index], | |
3145 | index - first_index, start, | |
125bac01 | 3146 | end, get_extent, em_cached, |
d3fac6ba | 3147 | bio, bio_flags, |
1f7ad75b | 3148 | prev_em_start); |
9974090b | 3149 | start = page_start; |
09cbfeaf | 3150 | end = start + PAGE_SIZE - 1; |
9974090b MX |
3151 | first_index = index; |
3152 | } | |
3153 | } | |
3154 | ||
3155 | if (end) | |
3156 | __do_contiguous_readpages(tree, &pages[first_index], | |
3157 | index - first_index, start, | |
125bac01 | 3158 | end, get_extent, em_cached, bio, |
d3fac6ba | 3159 | bio_flags, prev_em_start); |
9974090b MX |
3160 | } |
3161 | ||
3162 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
3163 | struct page *page, | |
3164 | get_extent_t *get_extent, | |
3165 | struct bio **bio, int mirror_num, | |
f1c77c55 DS |
3166 | unsigned long *bio_flags, |
3167 | unsigned int read_flags) | |
9974090b MX |
3168 | { |
3169 | struct inode *inode = page->mapping->host; | |
3170 | struct btrfs_ordered_extent *ordered; | |
3171 | u64 start = page_offset(page); | |
09cbfeaf | 3172 | u64 end = start + PAGE_SIZE - 1; |
9974090b MX |
3173 | int ret; |
3174 | ||
3175 | while (1) { | |
3176 | lock_extent(tree, start, end); | |
a776c6fa | 3177 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
09cbfeaf | 3178 | PAGE_SIZE); |
9974090b MX |
3179 | if (!ordered) |
3180 | break; | |
3181 | unlock_extent(tree, start, end); | |
3182 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3183 | btrfs_put_ordered_extent(ordered); | |
3184 | } | |
3185 | ||
125bac01 | 3186 | ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num, |
1f7ad75b | 3187 | bio_flags, read_flags, NULL); |
9974090b MX |
3188 | return ret; |
3189 | } | |
3190 | ||
d1310b2e | 3191 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, |
8ddc7d9c | 3192 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3193 | { |
3194 | struct bio *bio = NULL; | |
c8b97818 | 3195 | unsigned long bio_flags = 0; |
d1310b2e CM |
3196 | int ret; |
3197 | ||
8ddc7d9c | 3198 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, |
1f7ad75b | 3199 | &bio_flags, 0); |
d1310b2e | 3200 | if (bio) |
1f7ad75b | 3201 | ret = submit_one_bio(bio, mirror_num, bio_flags); |
d1310b2e CM |
3202 | return ret; |
3203 | } | |
d1310b2e | 3204 | |
3d4b9496 | 3205 | static void update_nr_written(struct writeback_control *wbc, |
a9132667 | 3206 | unsigned long nr_written) |
11c8349b CM |
3207 | { |
3208 | wbc->nr_to_write -= nr_written; | |
11c8349b CM |
3209 | } |
3210 | ||
d1310b2e | 3211 | /* |
40f76580 CM |
3212 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
3213 | * | |
3214 | * This returns 1 if our fill_delalloc function did all the work required | |
3215 | * to write the page (copy into inline extent). In this case the IO has | |
3216 | * been started and the page is already unlocked. | |
3217 | * | |
3218 | * This returns 0 if all went well (page still locked) | |
3219 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 3220 | */ |
40f76580 CM |
3221 | static noinline_for_stack int writepage_delalloc(struct inode *inode, |
3222 | struct page *page, struct writeback_control *wbc, | |
3223 | struct extent_page_data *epd, | |
3224 | u64 delalloc_start, | |
3225 | unsigned long *nr_written) | |
3226 | { | |
3227 | struct extent_io_tree *tree = epd->tree; | |
09cbfeaf | 3228 | u64 page_end = delalloc_start + PAGE_SIZE - 1; |
40f76580 CM |
3229 | u64 nr_delalloc; |
3230 | u64 delalloc_to_write = 0; | |
3231 | u64 delalloc_end = 0; | |
3232 | int ret; | |
3233 | int page_started = 0; | |
3234 | ||
3235 | if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc) | |
3236 | return 0; | |
3237 | ||
3238 | while (delalloc_end < page_end) { | |
3239 | nr_delalloc = find_lock_delalloc_range(inode, tree, | |
3240 | page, | |
3241 | &delalloc_start, | |
3242 | &delalloc_end, | |
dcab6a3b | 3243 | BTRFS_MAX_EXTENT_SIZE); |
40f76580 CM |
3244 | if (nr_delalloc == 0) { |
3245 | delalloc_start = delalloc_end + 1; | |
3246 | continue; | |
3247 | } | |
3248 | ret = tree->ops->fill_delalloc(inode, page, | |
3249 | delalloc_start, | |
3250 | delalloc_end, | |
3251 | &page_started, | |
f82b7359 | 3252 | nr_written, wbc); |
40f76580 CM |
3253 | /* File system has been set read-only */ |
3254 | if (ret) { | |
3255 | SetPageError(page); | |
3256 | /* fill_delalloc should be return < 0 for error | |
3257 | * but just in case, we use > 0 here meaning the | |
3258 | * IO is started, so we don't want to return > 0 | |
3259 | * unless things are going well. | |
3260 | */ | |
3261 | ret = ret < 0 ? ret : -EIO; | |
3262 | goto done; | |
3263 | } | |
3264 | /* | |
ea1754a0 KS |
3265 | * delalloc_end is already one less than the total length, so |
3266 | * we don't subtract one from PAGE_SIZE | |
40f76580 CM |
3267 | */ |
3268 | delalloc_to_write += (delalloc_end - delalloc_start + | |
ea1754a0 | 3269 | PAGE_SIZE) >> PAGE_SHIFT; |
40f76580 CM |
3270 | delalloc_start = delalloc_end + 1; |
3271 | } | |
3272 | if (wbc->nr_to_write < delalloc_to_write) { | |
3273 | int thresh = 8192; | |
3274 | ||
3275 | if (delalloc_to_write < thresh * 2) | |
3276 | thresh = delalloc_to_write; | |
3277 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3278 | thresh); | |
3279 | } | |
3280 | ||
3281 | /* did the fill delalloc function already unlock and start | |
3282 | * the IO? | |
3283 | */ | |
3284 | if (page_started) { | |
3285 | /* | |
3286 | * we've unlocked the page, so we can't update | |
3287 | * the mapping's writeback index, just update | |
3288 | * nr_to_write. | |
3289 | */ | |
3290 | wbc->nr_to_write -= *nr_written; | |
3291 | return 1; | |
3292 | } | |
3293 | ||
3294 | ret = 0; | |
3295 | ||
3296 | done: | |
3297 | return ret; | |
3298 | } | |
3299 | ||
3300 | /* | |
3301 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3302 | * and does the loop to map the page into extents and bios. | |
3303 | * | |
3304 | * We return 1 if the IO is started and the page is unlocked, | |
3305 | * 0 if all went well (page still locked) | |
3306 | * < 0 if there were errors (page still locked) | |
3307 | */ | |
3308 | static noinline_for_stack int __extent_writepage_io(struct inode *inode, | |
3309 | struct page *page, | |
3310 | struct writeback_control *wbc, | |
3311 | struct extent_page_data *epd, | |
3312 | loff_t i_size, | |
3313 | unsigned long nr_written, | |
f1c77c55 | 3314 | unsigned int write_flags, int *nr_ret) |
d1310b2e | 3315 | { |
d1310b2e | 3316 | struct extent_io_tree *tree = epd->tree; |
4eee4fa4 | 3317 | u64 start = page_offset(page); |
09cbfeaf | 3318 | u64 page_end = start + PAGE_SIZE - 1; |
d1310b2e CM |
3319 | u64 end; |
3320 | u64 cur = start; | |
3321 | u64 extent_offset; | |
d1310b2e CM |
3322 | u64 block_start; |
3323 | u64 iosize; | |
d1310b2e CM |
3324 | struct extent_map *em; |
3325 | struct block_device *bdev; | |
7f3c74fb | 3326 | size_t pg_offset = 0; |
d1310b2e | 3327 | size_t blocksize; |
40f76580 CM |
3328 | int ret = 0; |
3329 | int nr = 0; | |
3330 | bool compressed; | |
c8b97818 | 3331 | |
247e743c | 3332 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
3333 | ret = tree->ops->writepage_start_hook(page, start, |
3334 | page_end); | |
87826df0 JM |
3335 | if (ret) { |
3336 | /* Fixup worker will requeue */ | |
3337 | if (ret == -EBUSY) | |
3338 | wbc->pages_skipped++; | |
3339 | else | |
3340 | redirty_page_for_writepage(wbc, page); | |
40f76580 | 3341 | |
3d4b9496 | 3342 | update_nr_written(wbc, nr_written); |
247e743c | 3343 | unlock_page(page); |
bcf93489 | 3344 | return 1; |
247e743c CM |
3345 | } |
3346 | } | |
3347 | ||
11c8349b CM |
3348 | /* |
3349 | * we don't want to touch the inode after unlocking the page, | |
3350 | * so we update the mapping writeback index now | |
3351 | */ | |
3d4b9496 | 3352 | update_nr_written(wbc, nr_written + 1); |
771ed689 | 3353 | |
d1310b2e | 3354 | end = page_end; |
40f76580 | 3355 | if (i_size <= start) { |
e6dcd2dc CM |
3356 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3357 | tree->ops->writepage_end_io_hook(page, start, | |
3358 | page_end, NULL, 1); | |
d1310b2e CM |
3359 | goto done; |
3360 | } | |
3361 | ||
d1310b2e CM |
3362 | blocksize = inode->i_sb->s_blocksize; |
3363 | ||
3364 | while (cur <= end) { | |
40f76580 | 3365 | u64 em_end; |
6273b7f8 | 3366 | u64 offset; |
58409edd | 3367 | |
40f76580 | 3368 | if (cur >= i_size) { |
e6dcd2dc CM |
3369 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3370 | tree->ops->writepage_end_io_hook(page, cur, | |
3371 | page_end, NULL, 1); | |
d1310b2e CM |
3372 | break; |
3373 | } | |
3c98c62f | 3374 | em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, cur, |
d1310b2e | 3375 | end - cur + 1, 1); |
c704005d | 3376 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3377 | SetPageError(page); |
61391d56 | 3378 | ret = PTR_ERR_OR_ZERO(em); |
d1310b2e CM |
3379 | break; |
3380 | } | |
3381 | ||
3382 | extent_offset = cur - em->start; | |
40f76580 CM |
3383 | em_end = extent_map_end(em); |
3384 | BUG_ON(em_end <= cur); | |
d1310b2e | 3385 | BUG_ON(end < cur); |
40f76580 | 3386 | iosize = min(em_end - cur, end - cur + 1); |
fda2832f | 3387 | iosize = ALIGN(iosize, blocksize); |
6273b7f8 | 3388 | offset = em->block_start + extent_offset; |
d1310b2e CM |
3389 | bdev = em->bdev; |
3390 | block_start = em->block_start; | |
c8b97818 | 3391 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
3392 | free_extent_map(em); |
3393 | em = NULL; | |
3394 | ||
c8b97818 CM |
3395 | /* |
3396 | * compressed and inline extents are written through other | |
3397 | * paths in the FS | |
3398 | */ | |
3399 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3400 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
3401 | /* |
3402 | * end_io notification does not happen here for | |
3403 | * compressed extents | |
3404 | */ | |
3405 | if (!compressed && tree->ops && | |
3406 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
3407 | tree->ops->writepage_end_io_hook(page, cur, |
3408 | cur + iosize - 1, | |
3409 | NULL, 1); | |
c8b97818 CM |
3410 | else if (compressed) { |
3411 | /* we don't want to end_page_writeback on | |
3412 | * a compressed extent. this happens | |
3413 | * elsewhere | |
3414 | */ | |
3415 | nr++; | |
3416 | } | |
3417 | ||
3418 | cur += iosize; | |
7f3c74fb | 3419 | pg_offset += iosize; |
d1310b2e CM |
3420 | continue; |
3421 | } | |
c8b97818 | 3422 | |
58409edd DS |
3423 | set_range_writeback(tree, cur, cur + iosize - 1); |
3424 | if (!PageWriteback(page)) { | |
3425 | btrfs_err(BTRFS_I(inode)->root->fs_info, | |
3426 | "page %lu not writeback, cur %llu end %llu", | |
3427 | page->index, cur, end); | |
d1310b2e | 3428 | } |
7f3c74fb | 3429 | |
4b81ba48 | 3430 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc, |
6273b7f8 | 3431 | page, offset, iosize, pg_offset, |
c2df8bb4 | 3432 | bdev, &epd->bio, |
58409edd DS |
3433 | end_bio_extent_writepage, |
3434 | 0, 0, 0, false); | |
fe01aa65 | 3435 | if (ret) { |
58409edd | 3436 | SetPageError(page); |
fe01aa65 TK |
3437 | if (PageWriteback(page)) |
3438 | end_page_writeback(page); | |
3439 | } | |
d1310b2e | 3440 | |
d1310b2e | 3441 | cur = cur + iosize; |
7f3c74fb | 3442 | pg_offset += iosize; |
d1310b2e CM |
3443 | nr++; |
3444 | } | |
40f76580 CM |
3445 | done: |
3446 | *nr_ret = nr; | |
40f76580 CM |
3447 | return ret; |
3448 | } | |
3449 | ||
3450 | /* | |
3451 | * the writepage semantics are similar to regular writepage. extent | |
3452 | * records are inserted to lock ranges in the tree, and as dirty areas | |
3453 | * are found, they are marked writeback. Then the lock bits are removed | |
3454 | * and the end_io handler clears the writeback ranges | |
3455 | */ | |
3456 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
3457 | void *data) | |
3458 | { | |
3459 | struct inode *inode = page->mapping->host; | |
3460 | struct extent_page_data *epd = data; | |
3461 | u64 start = page_offset(page); | |
09cbfeaf | 3462 | u64 page_end = start + PAGE_SIZE - 1; |
40f76580 CM |
3463 | int ret; |
3464 | int nr = 0; | |
3465 | size_t pg_offset = 0; | |
3466 | loff_t i_size = i_size_read(inode); | |
09cbfeaf | 3467 | unsigned long end_index = i_size >> PAGE_SHIFT; |
f1c77c55 | 3468 | unsigned int write_flags = 0; |
40f76580 CM |
3469 | unsigned long nr_written = 0; |
3470 | ||
ff40adf7 | 3471 | write_flags = wbc_to_write_flags(wbc); |
40f76580 CM |
3472 | |
3473 | trace___extent_writepage(page, inode, wbc); | |
3474 | ||
3475 | WARN_ON(!PageLocked(page)); | |
3476 | ||
3477 | ClearPageError(page); | |
3478 | ||
09cbfeaf | 3479 | pg_offset = i_size & (PAGE_SIZE - 1); |
40f76580 CM |
3480 | if (page->index > end_index || |
3481 | (page->index == end_index && !pg_offset)) { | |
09cbfeaf | 3482 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); |
40f76580 CM |
3483 | unlock_page(page); |
3484 | return 0; | |
3485 | } | |
3486 | ||
3487 | if (page->index == end_index) { | |
3488 | char *userpage; | |
3489 | ||
3490 | userpage = kmap_atomic(page); | |
3491 | memset(userpage + pg_offset, 0, | |
09cbfeaf | 3492 | PAGE_SIZE - pg_offset); |
40f76580 CM |
3493 | kunmap_atomic(userpage); |
3494 | flush_dcache_page(page); | |
3495 | } | |
3496 | ||
3497 | pg_offset = 0; | |
3498 | ||
3499 | set_page_extent_mapped(page); | |
3500 | ||
3501 | ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written); | |
3502 | if (ret == 1) | |
3503 | goto done_unlocked; | |
3504 | if (ret) | |
3505 | goto done; | |
3506 | ||
3507 | ret = __extent_writepage_io(inode, page, wbc, epd, | |
3508 | i_size, nr_written, write_flags, &nr); | |
3509 | if (ret == 1) | |
3510 | goto done_unlocked; | |
3511 | ||
d1310b2e CM |
3512 | done: |
3513 | if (nr == 0) { | |
3514 | /* make sure the mapping tag for page dirty gets cleared */ | |
3515 | set_page_writeback(page); | |
3516 | end_page_writeback(page); | |
3517 | } | |
61391d56 FM |
3518 | if (PageError(page)) { |
3519 | ret = ret < 0 ? ret : -EIO; | |
3520 | end_extent_writepage(page, ret, start, page_end); | |
3521 | } | |
d1310b2e | 3522 | unlock_page(page); |
40f76580 | 3523 | return ret; |
771ed689 | 3524 | |
11c8349b | 3525 | done_unlocked: |
d1310b2e CM |
3526 | return 0; |
3527 | } | |
3528 | ||
fd8b2b61 | 3529 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 3530 | { |
74316201 N |
3531 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
3532 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
3533 | } |
3534 | ||
0e378df1 CM |
3535 | static noinline_for_stack int |
3536 | lock_extent_buffer_for_io(struct extent_buffer *eb, | |
3537 | struct btrfs_fs_info *fs_info, | |
3538 | struct extent_page_data *epd) | |
0b32f4bb JB |
3539 | { |
3540 | unsigned long i, num_pages; | |
3541 | int flush = 0; | |
3542 | int ret = 0; | |
3543 | ||
3544 | if (!btrfs_try_tree_write_lock(eb)) { | |
3545 | flush = 1; | |
3546 | flush_write_bio(epd); | |
3547 | btrfs_tree_lock(eb); | |
3548 | } | |
3549 | ||
3550 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3551 | btrfs_tree_unlock(eb); | |
3552 | if (!epd->sync_io) | |
3553 | return 0; | |
3554 | if (!flush) { | |
3555 | flush_write_bio(epd); | |
3556 | flush = 1; | |
3557 | } | |
a098d8e8 CM |
3558 | while (1) { |
3559 | wait_on_extent_buffer_writeback(eb); | |
3560 | btrfs_tree_lock(eb); | |
3561 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3562 | break; | |
0b32f4bb | 3563 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3564 | } |
3565 | } | |
3566 | ||
51561ffe JB |
3567 | /* |
3568 | * We need to do this to prevent races in people who check if the eb is | |
3569 | * under IO since we can end up having no IO bits set for a short period | |
3570 | * of time. | |
3571 | */ | |
3572 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
3573 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3574 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 3575 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 3576 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
104b4e51 NB |
3577 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
3578 | -eb->len, | |
3579 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 3580 | ret = 1; |
51561ffe JB |
3581 | } else { |
3582 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
3583 | } |
3584 | ||
3585 | btrfs_tree_unlock(eb); | |
3586 | ||
3587 | if (!ret) | |
3588 | return ret; | |
3589 | ||
3590 | num_pages = num_extent_pages(eb->start, eb->len); | |
3591 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 3592 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3593 | |
3594 | if (!trylock_page(p)) { | |
3595 | if (!flush) { | |
3596 | flush_write_bio(epd); | |
3597 | flush = 1; | |
3598 | } | |
3599 | lock_page(p); | |
3600 | } | |
3601 | } | |
3602 | ||
3603 | return ret; | |
3604 | } | |
3605 | ||
3606 | static void end_extent_buffer_writeback(struct extent_buffer *eb) | |
3607 | { | |
3608 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
4e857c58 | 3609 | smp_mb__after_atomic(); |
0b32f4bb JB |
3610 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); |
3611 | } | |
3612 | ||
656f30db FM |
3613 | static void set_btree_ioerr(struct page *page) |
3614 | { | |
3615 | struct extent_buffer *eb = (struct extent_buffer *)page->private; | |
656f30db FM |
3616 | |
3617 | SetPageError(page); | |
3618 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) | |
3619 | return; | |
3620 | ||
3621 | /* | |
3622 | * If writeback for a btree extent that doesn't belong to a log tree | |
3623 | * failed, increment the counter transaction->eb_write_errors. | |
3624 | * We do this because while the transaction is running and before it's | |
3625 | * committing (when we call filemap_fdata[write|wait]_range against | |
3626 | * the btree inode), we might have | |
3627 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
3628 | * returns an error or an error happens during writeback, when we're | |
3629 | * committing the transaction we wouldn't know about it, since the pages | |
3630 | * can be no longer dirty nor marked anymore for writeback (if a | |
3631 | * subsequent modification to the extent buffer didn't happen before the | |
3632 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
3633 | * able to find the pages tagged with SetPageError at transaction | |
3634 | * commit time. So if this happens we must abort the transaction, | |
3635 | * otherwise we commit a super block with btree roots that point to | |
3636 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
3637 | * or the content of some node/leaf from a past generation that got | |
3638 | * cowed or deleted and is no longer valid. | |
3639 | * | |
3640 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
3641 | * not be enough - we need to distinguish between log tree extents vs | |
3642 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
3643 | * will catch and clear such errors in the mapping - and that call might | |
3644 | * be from a log sync and not from a transaction commit. Also, checking | |
3645 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
3646 | * not done and would not be reliable - the eb might have been released | |
3647 | * from memory and reading it back again means that flag would not be | |
3648 | * set (since it's a runtime flag, not persisted on disk). | |
3649 | * | |
3650 | * Using the flags below in the btree inode also makes us achieve the | |
3651 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
3652 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
3653 | * is called, the writeback for all dirty pages had already finished | |
3654 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
3655 | * filemap_fdatawait_range() would return success, as it could not know | |
3656 | * that writeback errors happened (the pages were no longer tagged for | |
3657 | * writeback). | |
3658 | */ | |
3659 | switch (eb->log_index) { | |
3660 | case -1: | |
afcdd129 | 3661 | set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags); |
656f30db FM |
3662 | break; |
3663 | case 0: | |
afcdd129 | 3664 | set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags); |
656f30db FM |
3665 | break; |
3666 | case 1: | |
afcdd129 | 3667 | set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags); |
656f30db FM |
3668 | break; |
3669 | default: | |
3670 | BUG(); /* unexpected, logic error */ | |
3671 | } | |
3672 | } | |
3673 | ||
4246a0b6 | 3674 | static void end_bio_extent_buffer_writepage(struct bio *bio) |
0b32f4bb | 3675 | { |
2c30c71b | 3676 | struct bio_vec *bvec; |
0b32f4bb | 3677 | struct extent_buffer *eb; |
2c30c71b | 3678 | int i, done; |
0b32f4bb | 3679 | |
c09abff8 | 3680 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2c30c71b | 3681 | bio_for_each_segment_all(bvec, bio, i) { |
0b32f4bb JB |
3682 | struct page *page = bvec->bv_page; |
3683 | ||
0b32f4bb JB |
3684 | eb = (struct extent_buffer *)page->private; |
3685 | BUG_ON(!eb); | |
3686 | done = atomic_dec_and_test(&eb->io_pages); | |
3687 | ||
4e4cbee9 | 3688 | if (bio->bi_status || |
4246a0b6 | 3689 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { |
0b32f4bb | 3690 | ClearPageUptodate(page); |
656f30db | 3691 | set_btree_ioerr(page); |
0b32f4bb JB |
3692 | } |
3693 | ||
3694 | end_page_writeback(page); | |
3695 | ||
3696 | if (!done) | |
3697 | continue; | |
3698 | ||
3699 | end_extent_buffer_writeback(eb); | |
2c30c71b | 3700 | } |
0b32f4bb JB |
3701 | |
3702 | bio_put(bio); | |
0b32f4bb JB |
3703 | } |
3704 | ||
0e378df1 | 3705 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb JB |
3706 | struct btrfs_fs_info *fs_info, |
3707 | struct writeback_control *wbc, | |
3708 | struct extent_page_data *epd) | |
3709 | { | |
3710 | struct block_device *bdev = fs_info->fs_devices->latest_bdev; | |
f28491e0 | 3711 | struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree; |
0b32f4bb | 3712 | u64 offset = eb->start; |
851cd173 | 3713 | u32 nritems; |
0b32f4bb | 3714 | unsigned long i, num_pages; |
851cd173 | 3715 | unsigned long start, end; |
ff40adf7 | 3716 | unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META; |
d7dbe9e7 | 3717 | int ret = 0; |
0b32f4bb | 3718 | |
656f30db | 3719 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); |
0b32f4bb JB |
3720 | num_pages = num_extent_pages(eb->start, eb->len); |
3721 | atomic_set(&eb->io_pages, num_pages); | |
de0022b9 | 3722 | |
851cd173 LB |
3723 | /* set btree blocks beyond nritems with 0 to avoid stale content. */ |
3724 | nritems = btrfs_header_nritems(eb); | |
3eb548ee | 3725 | if (btrfs_header_level(eb) > 0) { |
3eb548ee LB |
3726 | end = btrfs_node_key_ptr_offset(nritems); |
3727 | ||
b159fa28 | 3728 | memzero_extent_buffer(eb, end, eb->len - end); |
851cd173 LB |
3729 | } else { |
3730 | /* | |
3731 | * leaf: | |
3732 | * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0 | |
3733 | */ | |
3734 | start = btrfs_item_nr_offset(nritems); | |
3d9ec8c4 | 3735 | end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, eb); |
b159fa28 | 3736 | memzero_extent_buffer(eb, start, end - start); |
3eb548ee LB |
3737 | } |
3738 | ||
0b32f4bb | 3739 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3740 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3741 | |
3742 | clear_page_dirty_for_io(p); | |
3743 | set_page_writeback(p); | |
4b81ba48 | 3744 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc, |
6273b7f8 | 3745 | p, offset, PAGE_SIZE, 0, bdev, |
c2df8bb4 | 3746 | &epd->bio, |
1f7ad75b | 3747 | end_bio_extent_buffer_writepage, |
18fdc679 | 3748 | 0, 0, 0, false); |
0b32f4bb | 3749 | if (ret) { |
656f30db | 3750 | set_btree_ioerr(p); |
fe01aa65 TK |
3751 | if (PageWriteback(p)) |
3752 | end_page_writeback(p); | |
0b32f4bb JB |
3753 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
3754 | end_extent_buffer_writeback(eb); | |
3755 | ret = -EIO; | |
3756 | break; | |
3757 | } | |
09cbfeaf | 3758 | offset += PAGE_SIZE; |
3d4b9496 | 3759 | update_nr_written(wbc, 1); |
0b32f4bb JB |
3760 | unlock_page(p); |
3761 | } | |
3762 | ||
3763 | if (unlikely(ret)) { | |
3764 | for (; i < num_pages; i++) { | |
bbf65cf0 | 3765 | struct page *p = eb->pages[i]; |
81465028 | 3766 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
3767 | unlock_page(p); |
3768 | } | |
3769 | } | |
3770 | ||
3771 | return ret; | |
3772 | } | |
3773 | ||
3774 | int btree_write_cache_pages(struct address_space *mapping, | |
3775 | struct writeback_control *wbc) | |
3776 | { | |
3777 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; | |
3778 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; | |
3779 | struct extent_buffer *eb, *prev_eb = NULL; | |
3780 | struct extent_page_data epd = { | |
3781 | .bio = NULL, | |
3782 | .tree = tree, | |
3783 | .extent_locked = 0, | |
3784 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
3785 | }; | |
3786 | int ret = 0; | |
3787 | int done = 0; | |
3788 | int nr_to_write_done = 0; | |
3789 | struct pagevec pvec; | |
3790 | int nr_pages; | |
3791 | pgoff_t index; | |
3792 | pgoff_t end; /* Inclusive */ | |
3793 | int scanned = 0; | |
3794 | int tag; | |
3795 | ||
86679820 | 3796 | pagevec_init(&pvec); |
0b32f4bb JB |
3797 | if (wbc->range_cyclic) { |
3798 | index = mapping->writeback_index; /* Start from prev offset */ | |
3799 | end = -1; | |
3800 | } else { | |
09cbfeaf KS |
3801 | index = wbc->range_start >> PAGE_SHIFT; |
3802 | end = wbc->range_end >> PAGE_SHIFT; | |
0b32f4bb JB |
3803 | scanned = 1; |
3804 | } | |
3805 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3806 | tag = PAGECACHE_TAG_TOWRITE; | |
3807 | else | |
3808 | tag = PAGECACHE_TAG_DIRTY; | |
3809 | retry: | |
3810 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3811 | tag_pages_for_writeback(mapping, index, end); | |
3812 | while (!done && !nr_to_write_done && (index <= end) && | |
4006f437 | 3813 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 3814 | tag))) { |
0b32f4bb JB |
3815 | unsigned i; |
3816 | ||
3817 | scanned = 1; | |
3818 | for (i = 0; i < nr_pages; i++) { | |
3819 | struct page *page = pvec.pages[i]; | |
3820 | ||
3821 | if (!PagePrivate(page)) | |
3822 | continue; | |
3823 | ||
b5bae261 JB |
3824 | spin_lock(&mapping->private_lock); |
3825 | if (!PagePrivate(page)) { | |
3826 | spin_unlock(&mapping->private_lock); | |
3827 | continue; | |
3828 | } | |
3829 | ||
0b32f4bb | 3830 | eb = (struct extent_buffer *)page->private; |
b5bae261 JB |
3831 | |
3832 | /* | |
3833 | * Shouldn't happen and normally this would be a BUG_ON | |
3834 | * but no sense in crashing the users box for something | |
3835 | * we can survive anyway. | |
3836 | */ | |
fae7f21c | 3837 | if (WARN_ON(!eb)) { |
b5bae261 | 3838 | spin_unlock(&mapping->private_lock); |
0b32f4bb JB |
3839 | continue; |
3840 | } | |
3841 | ||
b5bae261 JB |
3842 | if (eb == prev_eb) { |
3843 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 3844 | continue; |
b5bae261 | 3845 | } |
0b32f4bb | 3846 | |
b5bae261 JB |
3847 | ret = atomic_inc_not_zero(&eb->refs); |
3848 | spin_unlock(&mapping->private_lock); | |
3849 | if (!ret) | |
0b32f4bb | 3850 | continue; |
0b32f4bb JB |
3851 | |
3852 | prev_eb = eb; | |
3853 | ret = lock_extent_buffer_for_io(eb, fs_info, &epd); | |
3854 | if (!ret) { | |
3855 | free_extent_buffer(eb); | |
3856 | continue; | |
3857 | } | |
3858 | ||
3859 | ret = write_one_eb(eb, fs_info, wbc, &epd); | |
3860 | if (ret) { | |
3861 | done = 1; | |
3862 | free_extent_buffer(eb); | |
3863 | break; | |
3864 | } | |
3865 | free_extent_buffer(eb); | |
3866 | ||
3867 | /* | |
3868 | * the filesystem may choose to bump up nr_to_write. | |
3869 | * We have to make sure to honor the new nr_to_write | |
3870 | * at any time | |
3871 | */ | |
3872 | nr_to_write_done = wbc->nr_to_write <= 0; | |
3873 | } | |
3874 | pagevec_release(&pvec); | |
3875 | cond_resched(); | |
3876 | } | |
3877 | if (!scanned && !done) { | |
3878 | /* | |
3879 | * We hit the last page and there is more work to be done: wrap | |
3880 | * back to the start of the file | |
3881 | */ | |
3882 | scanned = 1; | |
3883 | index = 0; | |
3884 | goto retry; | |
3885 | } | |
3886 | flush_write_bio(&epd); | |
3887 | return ret; | |
3888 | } | |
3889 | ||
d1310b2e | 3890 | /** |
4bef0848 | 3891 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
3892 | * @mapping: address space structure to write |
3893 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
3894 | * @writepage: function called for each page | |
3895 | * @data: data passed to writepage function | |
3896 | * | |
3897 | * If a page is already under I/O, write_cache_pages() skips it, even | |
3898 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
3899 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
3900 | * and msync() need to guarantee that all the data which was dirty at the time | |
3901 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
3902 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
3903 | * existing IO to complete. | |
3904 | */ | |
4242b64a | 3905 | static int extent_write_cache_pages(struct address_space *mapping, |
4bef0848 | 3906 | struct writeback_control *wbc, |
d2c3f4f6 CM |
3907 | writepage_t writepage, void *data, |
3908 | void (*flush_fn)(void *)) | |
d1310b2e | 3909 | { |
7fd1a3f7 | 3910 | struct inode *inode = mapping->host; |
d1310b2e CM |
3911 | int ret = 0; |
3912 | int done = 0; | |
f85d7d6c | 3913 | int nr_to_write_done = 0; |
d1310b2e CM |
3914 | struct pagevec pvec; |
3915 | int nr_pages; | |
3916 | pgoff_t index; | |
3917 | pgoff_t end; /* Inclusive */ | |
a9132667 LB |
3918 | pgoff_t done_index; |
3919 | int range_whole = 0; | |
d1310b2e | 3920 | int scanned = 0; |
f7aaa06b | 3921 | int tag; |
d1310b2e | 3922 | |
7fd1a3f7 JB |
3923 | /* |
3924 | * We have to hold onto the inode so that ordered extents can do their | |
3925 | * work when the IO finishes. The alternative to this is failing to add | |
3926 | * an ordered extent if the igrab() fails there and that is a huge pain | |
3927 | * to deal with, so instead just hold onto the inode throughout the | |
3928 | * writepages operation. If it fails here we are freeing up the inode | |
3929 | * anyway and we'd rather not waste our time writing out stuff that is | |
3930 | * going to be truncated anyway. | |
3931 | */ | |
3932 | if (!igrab(inode)) | |
3933 | return 0; | |
3934 | ||
86679820 | 3935 | pagevec_init(&pvec); |
d1310b2e CM |
3936 | if (wbc->range_cyclic) { |
3937 | index = mapping->writeback_index; /* Start from prev offset */ | |
3938 | end = -1; | |
3939 | } else { | |
09cbfeaf KS |
3940 | index = wbc->range_start >> PAGE_SHIFT; |
3941 | end = wbc->range_end >> PAGE_SHIFT; | |
a9132667 LB |
3942 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
3943 | range_whole = 1; | |
d1310b2e CM |
3944 | scanned = 1; |
3945 | } | |
f7aaa06b JB |
3946 | if (wbc->sync_mode == WB_SYNC_ALL) |
3947 | tag = PAGECACHE_TAG_TOWRITE; | |
3948 | else | |
3949 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 3950 | retry: |
f7aaa06b JB |
3951 | if (wbc->sync_mode == WB_SYNC_ALL) |
3952 | tag_pages_for_writeback(mapping, index, end); | |
a9132667 | 3953 | done_index = index; |
f85d7d6c | 3954 | while (!done && !nr_to_write_done && (index <= end) && |
67fd707f JK |
3955 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, |
3956 | &index, end, tag))) { | |
d1310b2e CM |
3957 | unsigned i; |
3958 | ||
3959 | scanned = 1; | |
3960 | for (i = 0; i < nr_pages; i++) { | |
3961 | struct page *page = pvec.pages[i]; | |
3962 | ||
a9132667 | 3963 | done_index = page->index; |
d1310b2e CM |
3964 | /* |
3965 | * At this point we hold neither mapping->tree_lock nor | |
3966 | * lock on the page itself: the page may be truncated or | |
3967 | * invalidated (changing page->mapping to NULL), or even | |
3968 | * swizzled back from swapper_space to tmpfs file | |
3969 | * mapping | |
3970 | */ | |
c8f2f24b JB |
3971 | if (!trylock_page(page)) { |
3972 | flush_fn(data); | |
3973 | lock_page(page); | |
01d658f2 | 3974 | } |
d1310b2e CM |
3975 | |
3976 | if (unlikely(page->mapping != mapping)) { | |
3977 | unlock_page(page); | |
3978 | continue; | |
3979 | } | |
3980 | ||
d2c3f4f6 | 3981 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 CM |
3982 | if (PageWriteback(page)) |
3983 | flush_fn(data); | |
d1310b2e | 3984 | wait_on_page_writeback(page); |
d2c3f4f6 | 3985 | } |
d1310b2e CM |
3986 | |
3987 | if (PageWriteback(page) || | |
3988 | !clear_page_dirty_for_io(page)) { | |
3989 | unlock_page(page); | |
3990 | continue; | |
3991 | } | |
3992 | ||
3993 | ret = (*writepage)(page, wbc, data); | |
3994 | ||
3995 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
3996 | unlock_page(page); | |
3997 | ret = 0; | |
3998 | } | |
a9132667 LB |
3999 | if (ret < 0) { |
4000 | /* | |
4001 | * done_index is set past this page, | |
4002 | * so media errors will not choke | |
4003 | * background writeout for the entire | |
4004 | * file. This has consequences for | |
4005 | * range_cyclic semantics (ie. it may | |
4006 | * not be suitable for data integrity | |
4007 | * writeout). | |
4008 | */ | |
4009 | done_index = page->index + 1; | |
4010 | done = 1; | |
4011 | break; | |
4012 | } | |
f85d7d6c CM |
4013 | |
4014 | /* | |
4015 | * the filesystem may choose to bump up nr_to_write. | |
4016 | * We have to make sure to honor the new nr_to_write | |
4017 | * at any time | |
4018 | */ | |
4019 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
4020 | } |
4021 | pagevec_release(&pvec); | |
4022 | cond_resched(); | |
4023 | } | |
894b36e3 | 4024 | if (!scanned && !done) { |
d1310b2e CM |
4025 | /* |
4026 | * We hit the last page and there is more work to be done: wrap | |
4027 | * back to the start of the file | |
4028 | */ | |
4029 | scanned = 1; | |
4030 | index = 0; | |
4031 | goto retry; | |
4032 | } | |
a9132667 LB |
4033 | |
4034 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole)) | |
4035 | mapping->writeback_index = done_index; | |
4036 | ||
7fd1a3f7 | 4037 | btrfs_add_delayed_iput(inode); |
894b36e3 | 4038 | return ret; |
d1310b2e | 4039 | } |
d1310b2e | 4040 | |
ffbd517d | 4041 | static void flush_epd_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 4042 | { |
d2c3f4f6 | 4043 | if (epd->bio) { |
355808c2 JM |
4044 | int ret; |
4045 | ||
18fdc679 | 4046 | ret = submit_one_bio(epd->bio, 0, 0); |
79787eaa | 4047 | BUG_ON(ret < 0); /* -ENOMEM */ |
d2c3f4f6 CM |
4048 | epd->bio = NULL; |
4049 | } | |
4050 | } | |
4051 | ||
ffbd517d CM |
4052 | static noinline void flush_write_bio(void *data) |
4053 | { | |
4054 | struct extent_page_data *epd = data; | |
4055 | flush_epd_write_bio(epd); | |
4056 | } | |
4057 | ||
d1310b2e | 4058 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, |
d1310b2e CM |
4059 | struct writeback_control *wbc) |
4060 | { | |
4061 | int ret; | |
d1310b2e CM |
4062 | struct extent_page_data epd = { |
4063 | .bio = NULL, | |
4064 | .tree = tree, | |
771ed689 | 4065 | .extent_locked = 0, |
ffbd517d | 4066 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e | 4067 | }; |
d1310b2e | 4068 | |
d1310b2e CM |
4069 | ret = __extent_writepage(page, wbc, &epd); |
4070 | ||
ffbd517d | 4071 | flush_epd_write_bio(&epd); |
d1310b2e CM |
4072 | return ret; |
4073 | } | |
d1310b2e | 4074 | |
771ed689 | 4075 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, |
916b9298 | 4076 | u64 start, u64 end, int mode) |
771ed689 CM |
4077 | { |
4078 | int ret = 0; | |
4079 | struct address_space *mapping = inode->i_mapping; | |
4080 | struct page *page; | |
09cbfeaf KS |
4081 | unsigned long nr_pages = (end - start + PAGE_SIZE) >> |
4082 | PAGE_SHIFT; | |
771ed689 CM |
4083 | |
4084 | struct extent_page_data epd = { | |
4085 | .bio = NULL, | |
4086 | .tree = tree, | |
771ed689 | 4087 | .extent_locked = 1, |
ffbd517d | 4088 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
4089 | }; |
4090 | struct writeback_control wbc_writepages = { | |
771ed689 | 4091 | .sync_mode = mode, |
771ed689 CM |
4092 | .nr_to_write = nr_pages * 2, |
4093 | .range_start = start, | |
4094 | .range_end = end + 1, | |
4095 | }; | |
4096 | ||
d397712b | 4097 | while (start <= end) { |
09cbfeaf | 4098 | page = find_get_page(mapping, start >> PAGE_SHIFT); |
771ed689 CM |
4099 | if (clear_page_dirty_for_io(page)) |
4100 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
4101 | else { | |
4102 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
4103 | tree->ops->writepage_end_io_hook(page, start, | |
09cbfeaf | 4104 | start + PAGE_SIZE - 1, |
771ed689 CM |
4105 | NULL, 1); |
4106 | unlock_page(page); | |
4107 | } | |
09cbfeaf KS |
4108 | put_page(page); |
4109 | start += PAGE_SIZE; | |
771ed689 CM |
4110 | } |
4111 | ||
ffbd517d | 4112 | flush_epd_write_bio(&epd); |
771ed689 CM |
4113 | return ret; |
4114 | } | |
d1310b2e CM |
4115 | |
4116 | int extent_writepages(struct extent_io_tree *tree, | |
4117 | struct address_space *mapping, | |
d1310b2e CM |
4118 | struct writeback_control *wbc) |
4119 | { | |
4120 | int ret = 0; | |
4121 | struct extent_page_data epd = { | |
4122 | .bio = NULL, | |
4123 | .tree = tree, | |
771ed689 | 4124 | .extent_locked = 0, |
ffbd517d | 4125 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
4126 | }; |
4127 | ||
4242b64a | 4128 | ret = extent_write_cache_pages(mapping, wbc, __extent_writepage, &epd, |
d2c3f4f6 | 4129 | flush_write_bio); |
ffbd517d | 4130 | flush_epd_write_bio(&epd); |
d1310b2e CM |
4131 | return ret; |
4132 | } | |
d1310b2e CM |
4133 | |
4134 | int extent_readpages(struct extent_io_tree *tree, | |
4135 | struct address_space *mapping, | |
0932584b | 4136 | struct list_head *pages, unsigned nr_pages) |
d1310b2e CM |
4137 | { |
4138 | struct bio *bio = NULL; | |
4139 | unsigned page_idx; | |
c8b97818 | 4140 | unsigned long bio_flags = 0; |
67c9684f LB |
4141 | struct page *pagepool[16]; |
4142 | struct page *page; | |
125bac01 | 4143 | struct extent_map *em_cached = NULL; |
67c9684f | 4144 | int nr = 0; |
808f80b4 | 4145 | u64 prev_em_start = (u64)-1; |
d1310b2e | 4146 | |
d1310b2e | 4147 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
67c9684f | 4148 | page = list_entry(pages->prev, struct page, lru); |
d1310b2e CM |
4149 | |
4150 | prefetchw(&page->flags); | |
4151 | list_del(&page->lru); | |
67c9684f | 4152 | if (add_to_page_cache_lru(page, mapping, |
8a5c743e MH |
4153 | page->index, |
4154 | readahead_gfp_mask(mapping))) { | |
09cbfeaf | 4155 | put_page(page); |
67c9684f | 4156 | continue; |
d1310b2e | 4157 | } |
67c9684f LB |
4158 | |
4159 | pagepool[nr++] = page; | |
4160 | if (nr < ARRAY_SIZE(pagepool)) | |
4161 | continue; | |
0932584b DS |
4162 | __extent_readpages(tree, pagepool, nr, btrfs_get_extent, |
4163 | &em_cached, &bio, &bio_flags, &prev_em_start); | |
67c9684f | 4164 | nr = 0; |
d1310b2e | 4165 | } |
9974090b | 4166 | if (nr) |
0932584b DS |
4167 | __extent_readpages(tree, pagepool, nr, btrfs_get_extent, |
4168 | &em_cached, &bio, &bio_flags, &prev_em_start); | |
67c9684f | 4169 | |
125bac01 MX |
4170 | if (em_cached) |
4171 | free_extent_map(em_cached); | |
4172 | ||
d1310b2e CM |
4173 | BUG_ON(!list_empty(pages)); |
4174 | if (bio) | |
1f7ad75b | 4175 | return submit_one_bio(bio, 0, bio_flags); |
d1310b2e CM |
4176 | return 0; |
4177 | } | |
d1310b2e CM |
4178 | |
4179 | /* | |
4180 | * basic invalidatepage code, this waits on any locked or writeback | |
4181 | * ranges corresponding to the page, and then deletes any extent state | |
4182 | * records from the tree | |
4183 | */ | |
4184 | int extent_invalidatepage(struct extent_io_tree *tree, | |
4185 | struct page *page, unsigned long offset) | |
4186 | { | |
2ac55d41 | 4187 | struct extent_state *cached_state = NULL; |
4eee4fa4 | 4188 | u64 start = page_offset(page); |
09cbfeaf | 4189 | u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
4190 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; |
4191 | ||
fda2832f | 4192 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
4193 | if (start > end) |
4194 | return 0; | |
4195 | ||
ff13db41 | 4196 | lock_extent_bits(tree, start, end, &cached_state); |
1edbb734 | 4197 | wait_on_page_writeback(page); |
d1310b2e | 4198 | clear_extent_bit(tree, start, end, |
32c00aff JB |
4199 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
4200 | EXTENT_DO_ACCOUNTING, | |
ae0f1625 | 4201 | 1, 1, &cached_state); |
d1310b2e CM |
4202 | return 0; |
4203 | } | |
d1310b2e | 4204 | |
7b13b7b1 CM |
4205 | /* |
4206 | * a helper for releasepage, this tests for areas of the page that | |
4207 | * are locked or under IO and drops the related state bits if it is safe | |
4208 | * to drop the page. | |
4209 | */ | |
48a3b636 ES |
4210 | static int try_release_extent_state(struct extent_map_tree *map, |
4211 | struct extent_io_tree *tree, | |
4212 | struct page *page, gfp_t mask) | |
7b13b7b1 | 4213 | { |
4eee4fa4 | 4214 | u64 start = page_offset(page); |
09cbfeaf | 4215 | u64 end = start + PAGE_SIZE - 1; |
7b13b7b1 CM |
4216 | int ret = 1; |
4217 | ||
211f90e6 | 4218 | if (test_range_bit(tree, start, end, |
8b62b72b | 4219 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
4220 | ret = 0; |
4221 | else { | |
11ef160f CM |
4222 | /* |
4223 | * at this point we can safely clear everything except the | |
4224 | * locked bit and the nodatasum bit | |
4225 | */ | |
66b0c887 | 4226 | ret = __clear_extent_bit(tree, start, end, |
11ef160f | 4227 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
66b0c887 | 4228 | 0, 0, NULL, mask, NULL); |
e3f24cc5 CM |
4229 | |
4230 | /* if clear_extent_bit failed for enomem reasons, | |
4231 | * we can't allow the release to continue. | |
4232 | */ | |
4233 | if (ret < 0) | |
4234 | ret = 0; | |
4235 | else | |
4236 | ret = 1; | |
7b13b7b1 CM |
4237 | } |
4238 | return ret; | |
4239 | } | |
7b13b7b1 | 4240 | |
d1310b2e CM |
4241 | /* |
4242 | * a helper for releasepage. As long as there are no locked extents | |
4243 | * in the range corresponding to the page, both state records and extent | |
4244 | * map records are removed | |
4245 | */ | |
4246 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
4247 | struct extent_io_tree *tree, struct page *page, |
4248 | gfp_t mask) | |
d1310b2e CM |
4249 | { |
4250 | struct extent_map *em; | |
4eee4fa4 | 4251 | u64 start = page_offset(page); |
09cbfeaf | 4252 | u64 end = start + PAGE_SIZE - 1; |
7b13b7b1 | 4253 | |
d0164adc | 4254 | if (gfpflags_allow_blocking(mask) && |
ee22184b | 4255 | page->mapping->host->i_size > SZ_16M) { |
39b5637f | 4256 | u64 len; |
70dec807 | 4257 | while (start <= end) { |
39b5637f | 4258 | len = end - start + 1; |
890871be | 4259 | write_lock(&map->lock); |
39b5637f | 4260 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 4261 | if (!em) { |
890871be | 4262 | write_unlock(&map->lock); |
70dec807 CM |
4263 | break; |
4264 | } | |
7f3c74fb CM |
4265 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
4266 | em->start != start) { | |
890871be | 4267 | write_unlock(&map->lock); |
70dec807 CM |
4268 | free_extent_map(em); |
4269 | break; | |
4270 | } | |
4271 | if (!test_range_bit(tree, em->start, | |
4272 | extent_map_end(em) - 1, | |
8b62b72b | 4273 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 4274 | 0, NULL)) { |
70dec807 CM |
4275 | remove_extent_mapping(map, em); |
4276 | /* once for the rb tree */ | |
4277 | free_extent_map(em); | |
4278 | } | |
4279 | start = extent_map_end(em); | |
890871be | 4280 | write_unlock(&map->lock); |
70dec807 CM |
4281 | |
4282 | /* once for us */ | |
d1310b2e CM |
4283 | free_extent_map(em); |
4284 | } | |
d1310b2e | 4285 | } |
7b13b7b1 | 4286 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 4287 | } |
d1310b2e | 4288 | |
ec29ed5b CM |
4289 | /* |
4290 | * helper function for fiemap, which doesn't want to see any holes. | |
4291 | * This maps until we find something past 'last' | |
4292 | */ | |
4293 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
e3350e16 | 4294 | u64 offset, u64 last) |
ec29ed5b | 4295 | { |
da17066c | 4296 | u64 sectorsize = btrfs_inode_sectorsize(inode); |
ec29ed5b CM |
4297 | struct extent_map *em; |
4298 | u64 len; | |
4299 | ||
4300 | if (offset >= last) | |
4301 | return NULL; | |
4302 | ||
67871254 | 4303 | while (1) { |
ec29ed5b CM |
4304 | len = last - offset; |
4305 | if (len == 0) | |
4306 | break; | |
fda2832f | 4307 | len = ALIGN(len, sectorsize); |
e3350e16 DS |
4308 | em = btrfs_get_extent_fiemap(BTRFS_I(inode), NULL, 0, offset, |
4309 | len, 0); | |
c704005d | 4310 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
4311 | return em; |
4312 | ||
4313 | /* if this isn't a hole return it */ | |
4314 | if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) && | |
4315 | em->block_start != EXTENT_MAP_HOLE) { | |
4316 | return em; | |
4317 | } | |
4318 | ||
4319 | /* this is a hole, advance to the next extent */ | |
4320 | offset = extent_map_end(em); | |
4321 | free_extent_map(em); | |
4322 | if (offset >= last) | |
4323 | break; | |
4324 | } | |
4325 | return NULL; | |
4326 | } | |
4327 | ||
4751832d QW |
4328 | /* |
4329 | * To cache previous fiemap extent | |
4330 | * | |
4331 | * Will be used for merging fiemap extent | |
4332 | */ | |
4333 | struct fiemap_cache { | |
4334 | u64 offset; | |
4335 | u64 phys; | |
4336 | u64 len; | |
4337 | u32 flags; | |
4338 | bool cached; | |
4339 | }; | |
4340 | ||
4341 | /* | |
4342 | * Helper to submit fiemap extent. | |
4343 | * | |
4344 | * Will try to merge current fiemap extent specified by @offset, @phys, | |
4345 | * @len and @flags with cached one. | |
4346 | * And only when we fails to merge, cached one will be submitted as | |
4347 | * fiemap extent. | |
4348 | * | |
4349 | * Return value is the same as fiemap_fill_next_extent(). | |
4350 | */ | |
4351 | static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, | |
4352 | struct fiemap_cache *cache, | |
4353 | u64 offset, u64 phys, u64 len, u32 flags) | |
4354 | { | |
4355 | int ret = 0; | |
4356 | ||
4357 | if (!cache->cached) | |
4358 | goto assign; | |
4359 | ||
4360 | /* | |
4361 | * Sanity check, extent_fiemap() should have ensured that new | |
4362 | * fiemap extent won't overlap with cahced one. | |
4363 | * Not recoverable. | |
4364 | * | |
4365 | * NOTE: Physical address can overlap, due to compression | |
4366 | */ | |
4367 | if (cache->offset + cache->len > offset) { | |
4368 | WARN_ON(1); | |
4369 | return -EINVAL; | |
4370 | } | |
4371 | ||
4372 | /* | |
4373 | * Only merges fiemap extents if | |
4374 | * 1) Their logical addresses are continuous | |
4375 | * | |
4376 | * 2) Their physical addresses are continuous | |
4377 | * So truly compressed (physical size smaller than logical size) | |
4378 | * extents won't get merged with each other | |
4379 | * | |
4380 | * 3) Share same flags except FIEMAP_EXTENT_LAST | |
4381 | * So regular extent won't get merged with prealloc extent | |
4382 | */ | |
4383 | if (cache->offset + cache->len == offset && | |
4384 | cache->phys + cache->len == phys && | |
4385 | (cache->flags & ~FIEMAP_EXTENT_LAST) == | |
4386 | (flags & ~FIEMAP_EXTENT_LAST)) { | |
4387 | cache->len += len; | |
4388 | cache->flags |= flags; | |
4389 | goto try_submit_last; | |
4390 | } | |
4391 | ||
4392 | /* Not mergeable, need to submit cached one */ | |
4393 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, | |
4394 | cache->len, cache->flags); | |
4395 | cache->cached = false; | |
4396 | if (ret) | |
4397 | return ret; | |
4398 | assign: | |
4399 | cache->cached = true; | |
4400 | cache->offset = offset; | |
4401 | cache->phys = phys; | |
4402 | cache->len = len; | |
4403 | cache->flags = flags; | |
4404 | try_submit_last: | |
4405 | if (cache->flags & FIEMAP_EXTENT_LAST) { | |
4406 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, | |
4407 | cache->phys, cache->len, cache->flags); | |
4408 | cache->cached = false; | |
4409 | } | |
4410 | return ret; | |
4411 | } | |
4412 | ||
4413 | /* | |
848c23b7 | 4414 | * Emit last fiemap cache |
4751832d | 4415 | * |
848c23b7 QW |
4416 | * The last fiemap cache may still be cached in the following case: |
4417 | * 0 4k 8k | |
4418 | * |<- Fiemap range ->| | |
4419 | * |<------------ First extent ----------->| | |
4420 | * | |
4421 | * In this case, the first extent range will be cached but not emitted. | |
4422 | * So we must emit it before ending extent_fiemap(). | |
4751832d | 4423 | */ |
848c23b7 QW |
4424 | static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info, |
4425 | struct fiemap_extent_info *fieinfo, | |
4426 | struct fiemap_cache *cache) | |
4751832d QW |
4427 | { |
4428 | int ret; | |
4429 | ||
4430 | if (!cache->cached) | |
4431 | return 0; | |
4432 | ||
4751832d QW |
4433 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, |
4434 | cache->len, cache->flags); | |
4435 | cache->cached = false; | |
4436 | if (ret > 0) | |
4437 | ret = 0; | |
4438 | return ret; | |
4439 | } | |
4440 | ||
1506fcc8 | 4441 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
2135fb9b | 4442 | __u64 start, __u64 len) |
1506fcc8 | 4443 | { |
975f84fe | 4444 | int ret = 0; |
1506fcc8 YS |
4445 | u64 off = start; |
4446 | u64 max = start + len; | |
4447 | u32 flags = 0; | |
975f84fe JB |
4448 | u32 found_type; |
4449 | u64 last; | |
ec29ed5b | 4450 | u64 last_for_get_extent = 0; |
1506fcc8 | 4451 | u64 disko = 0; |
ec29ed5b | 4452 | u64 isize = i_size_read(inode); |
975f84fe | 4453 | struct btrfs_key found_key; |
1506fcc8 | 4454 | struct extent_map *em = NULL; |
2ac55d41 | 4455 | struct extent_state *cached_state = NULL; |
975f84fe | 4456 | struct btrfs_path *path; |
dc046b10 | 4457 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4751832d | 4458 | struct fiemap_cache cache = { 0 }; |
1506fcc8 | 4459 | int end = 0; |
ec29ed5b CM |
4460 | u64 em_start = 0; |
4461 | u64 em_len = 0; | |
4462 | u64 em_end = 0; | |
1506fcc8 YS |
4463 | |
4464 | if (len == 0) | |
4465 | return -EINVAL; | |
4466 | ||
975f84fe JB |
4467 | path = btrfs_alloc_path(); |
4468 | if (!path) | |
4469 | return -ENOMEM; | |
4470 | path->leave_spinning = 1; | |
4471 | ||
da17066c JM |
4472 | start = round_down(start, btrfs_inode_sectorsize(inode)); |
4473 | len = round_up(max, btrfs_inode_sectorsize(inode)) - start; | |
4d479cf0 | 4474 | |
ec29ed5b CM |
4475 | /* |
4476 | * lookup the last file extent. We're not using i_size here | |
4477 | * because there might be preallocation past i_size | |
4478 | */ | |
f85b7379 DS |
4479 | ret = btrfs_lookup_file_extent(NULL, root, path, |
4480 | btrfs_ino(BTRFS_I(inode)), -1, 0); | |
975f84fe JB |
4481 | if (ret < 0) { |
4482 | btrfs_free_path(path); | |
4483 | return ret; | |
2d324f59 LB |
4484 | } else { |
4485 | WARN_ON(!ret); | |
4486 | if (ret == 1) | |
4487 | ret = 0; | |
975f84fe | 4488 | } |
2d324f59 | 4489 | |
975f84fe | 4490 | path->slots[0]--; |
975f84fe | 4491 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
962a298f | 4492 | found_type = found_key.type; |
975f84fe | 4493 | |
ec29ed5b | 4494 | /* No extents, but there might be delalloc bits */ |
4a0cc7ca | 4495 | if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) || |
975f84fe | 4496 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
4497 | /* have to trust i_size as the end */ |
4498 | last = (u64)-1; | |
4499 | last_for_get_extent = isize; | |
4500 | } else { | |
4501 | /* | |
4502 | * remember the start of the last extent. There are a | |
4503 | * bunch of different factors that go into the length of the | |
4504 | * extent, so its much less complex to remember where it started | |
4505 | */ | |
4506 | last = found_key.offset; | |
4507 | last_for_get_extent = last + 1; | |
975f84fe | 4508 | } |
fe09e16c | 4509 | btrfs_release_path(path); |
975f84fe | 4510 | |
ec29ed5b CM |
4511 | /* |
4512 | * we might have some extents allocated but more delalloc past those | |
4513 | * extents. so, we trust isize unless the start of the last extent is | |
4514 | * beyond isize | |
4515 | */ | |
4516 | if (last < isize) { | |
4517 | last = (u64)-1; | |
4518 | last_for_get_extent = isize; | |
4519 | } | |
4520 | ||
ff13db41 | 4521 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
d0082371 | 4522 | &cached_state); |
ec29ed5b | 4523 | |
e3350e16 | 4524 | em = get_extent_skip_holes(inode, start, last_for_get_extent); |
1506fcc8 YS |
4525 | if (!em) |
4526 | goto out; | |
4527 | if (IS_ERR(em)) { | |
4528 | ret = PTR_ERR(em); | |
4529 | goto out; | |
4530 | } | |
975f84fe | 4531 | |
1506fcc8 | 4532 | while (!end) { |
b76bb701 | 4533 | u64 offset_in_extent = 0; |
ea8efc74 CM |
4534 | |
4535 | /* break if the extent we found is outside the range */ | |
4536 | if (em->start >= max || extent_map_end(em) < off) | |
4537 | break; | |
4538 | ||
4539 | /* | |
4540 | * get_extent may return an extent that starts before our | |
4541 | * requested range. We have to make sure the ranges | |
4542 | * we return to fiemap always move forward and don't | |
4543 | * overlap, so adjust the offsets here | |
4544 | */ | |
4545 | em_start = max(em->start, off); | |
1506fcc8 | 4546 | |
ea8efc74 CM |
4547 | /* |
4548 | * record the offset from the start of the extent | |
b76bb701 JB |
4549 | * for adjusting the disk offset below. Only do this if the |
4550 | * extent isn't compressed since our in ram offset may be past | |
4551 | * what we have actually allocated on disk. | |
ea8efc74 | 4552 | */ |
b76bb701 JB |
4553 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
4554 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 4555 | em_end = extent_map_end(em); |
ea8efc74 | 4556 | em_len = em_end - em_start; |
1506fcc8 YS |
4557 | disko = 0; |
4558 | flags = 0; | |
4559 | ||
ea8efc74 CM |
4560 | /* |
4561 | * bump off for our next call to get_extent | |
4562 | */ | |
4563 | off = extent_map_end(em); | |
4564 | if (off >= max) | |
4565 | end = 1; | |
4566 | ||
93dbfad7 | 4567 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
4568 | end = 1; |
4569 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 4570 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
4571 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
4572 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 4573 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
4574 | flags |= (FIEMAP_EXTENT_DELALLOC | |
4575 | FIEMAP_EXTENT_UNKNOWN); | |
dc046b10 JB |
4576 | } else if (fieinfo->fi_extents_max) { |
4577 | u64 bytenr = em->block_start - | |
4578 | (em->start - em->orig_start); | |
fe09e16c | 4579 | |
ea8efc74 | 4580 | disko = em->block_start + offset_in_extent; |
fe09e16c LB |
4581 | |
4582 | /* | |
4583 | * As btrfs supports shared space, this information | |
4584 | * can be exported to userspace tools via | |
dc046b10 JB |
4585 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 |
4586 | * then we're just getting a count and we can skip the | |
4587 | * lookup stuff. | |
fe09e16c | 4588 | */ |
bb739cf0 EN |
4589 | ret = btrfs_check_shared(root, |
4590 | btrfs_ino(BTRFS_I(inode)), | |
4591 | bytenr); | |
dc046b10 | 4592 | if (ret < 0) |
fe09e16c | 4593 | goto out_free; |
dc046b10 | 4594 | if (ret) |
fe09e16c | 4595 | flags |= FIEMAP_EXTENT_SHARED; |
dc046b10 | 4596 | ret = 0; |
1506fcc8 YS |
4597 | } |
4598 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
4599 | flags |= FIEMAP_EXTENT_ENCODED; | |
0d2b2372 JB |
4600 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
4601 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
1506fcc8 | 4602 | |
1506fcc8 YS |
4603 | free_extent_map(em); |
4604 | em = NULL; | |
ec29ed5b CM |
4605 | if ((em_start >= last) || em_len == (u64)-1 || |
4606 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
4607 | flags |= FIEMAP_EXTENT_LAST; |
4608 | end = 1; | |
4609 | } | |
4610 | ||
ec29ed5b | 4611 | /* now scan forward to see if this is really the last extent. */ |
e3350e16 | 4612 | em = get_extent_skip_holes(inode, off, last_for_get_extent); |
ec29ed5b CM |
4613 | if (IS_ERR(em)) { |
4614 | ret = PTR_ERR(em); | |
4615 | goto out; | |
4616 | } | |
4617 | if (!em) { | |
975f84fe JB |
4618 | flags |= FIEMAP_EXTENT_LAST; |
4619 | end = 1; | |
4620 | } | |
4751832d QW |
4621 | ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko, |
4622 | em_len, flags); | |
26e726af CS |
4623 | if (ret) { |
4624 | if (ret == 1) | |
4625 | ret = 0; | |
ec29ed5b | 4626 | goto out_free; |
26e726af | 4627 | } |
1506fcc8 YS |
4628 | } |
4629 | out_free: | |
4751832d | 4630 | if (!ret) |
848c23b7 | 4631 | ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache); |
1506fcc8 YS |
4632 | free_extent_map(em); |
4633 | out: | |
fe09e16c | 4634 | btrfs_free_path(path); |
a52f4cd2 | 4635 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
2ac55d41 | 4636 | &cached_state, GFP_NOFS); |
1506fcc8 YS |
4637 | return ret; |
4638 | } | |
4639 | ||
727011e0 CM |
4640 | static void __free_extent_buffer(struct extent_buffer *eb) |
4641 | { | |
6d49ba1b | 4642 | btrfs_leak_debug_del(&eb->leak_list); |
727011e0 CM |
4643 | kmem_cache_free(extent_buffer_cache, eb); |
4644 | } | |
4645 | ||
a26e8c9f | 4646 | int extent_buffer_under_io(struct extent_buffer *eb) |
db7f3436 JB |
4647 | { |
4648 | return (atomic_read(&eb->io_pages) || | |
4649 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4650 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4651 | } | |
4652 | ||
4653 | /* | |
4654 | * Helper for releasing extent buffer page. | |
4655 | */ | |
a50924e3 | 4656 | static void btrfs_release_extent_buffer_page(struct extent_buffer *eb) |
db7f3436 JB |
4657 | { |
4658 | unsigned long index; | |
db7f3436 JB |
4659 | struct page *page; |
4660 | int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4661 | ||
4662 | BUG_ON(extent_buffer_under_io(eb)); | |
4663 | ||
a50924e3 DS |
4664 | index = num_extent_pages(eb->start, eb->len); |
4665 | if (index == 0) | |
db7f3436 JB |
4666 | return; |
4667 | ||
4668 | do { | |
4669 | index--; | |
fb85fc9a | 4670 | page = eb->pages[index]; |
5d2361db FL |
4671 | if (!page) |
4672 | continue; | |
4673 | if (mapped) | |
db7f3436 | 4674 | spin_lock(&page->mapping->private_lock); |
5d2361db FL |
4675 | /* |
4676 | * We do this since we'll remove the pages after we've | |
4677 | * removed the eb from the radix tree, so we could race | |
4678 | * and have this page now attached to the new eb. So | |
4679 | * only clear page_private if it's still connected to | |
4680 | * this eb. | |
4681 | */ | |
4682 | if (PagePrivate(page) && | |
4683 | page->private == (unsigned long)eb) { | |
4684 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4685 | BUG_ON(PageDirty(page)); | |
4686 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 4687 | /* |
5d2361db FL |
4688 | * We need to make sure we haven't be attached |
4689 | * to a new eb. | |
db7f3436 | 4690 | */ |
5d2361db FL |
4691 | ClearPagePrivate(page); |
4692 | set_page_private(page, 0); | |
4693 | /* One for the page private */ | |
09cbfeaf | 4694 | put_page(page); |
db7f3436 | 4695 | } |
5d2361db FL |
4696 | |
4697 | if (mapped) | |
4698 | spin_unlock(&page->mapping->private_lock); | |
4699 | ||
01327610 | 4700 | /* One for when we allocated the page */ |
09cbfeaf | 4701 | put_page(page); |
a50924e3 | 4702 | } while (index != 0); |
db7f3436 JB |
4703 | } |
4704 | ||
4705 | /* | |
4706 | * Helper for releasing the extent buffer. | |
4707 | */ | |
4708 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4709 | { | |
a50924e3 | 4710 | btrfs_release_extent_buffer_page(eb); |
db7f3436 JB |
4711 | __free_extent_buffer(eb); |
4712 | } | |
4713 | ||
f28491e0 JB |
4714 | static struct extent_buffer * |
4715 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 4716 | unsigned long len) |
d1310b2e CM |
4717 | { |
4718 | struct extent_buffer *eb = NULL; | |
4719 | ||
d1b5c567 | 4720 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
4721 | eb->start = start; |
4722 | eb->len = len; | |
f28491e0 | 4723 | eb->fs_info = fs_info; |
815a51c7 | 4724 | eb->bflags = 0; |
bd681513 CM |
4725 | rwlock_init(&eb->lock); |
4726 | atomic_set(&eb->write_locks, 0); | |
4727 | atomic_set(&eb->read_locks, 0); | |
4728 | atomic_set(&eb->blocking_readers, 0); | |
4729 | atomic_set(&eb->blocking_writers, 0); | |
4730 | atomic_set(&eb->spinning_readers, 0); | |
4731 | atomic_set(&eb->spinning_writers, 0); | |
5b25f70f | 4732 | eb->lock_nested = 0; |
bd681513 CM |
4733 | init_waitqueue_head(&eb->write_lock_wq); |
4734 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 4735 | |
6d49ba1b ES |
4736 | btrfs_leak_debug_add(&eb->leak_list, &buffers); |
4737 | ||
3083ee2e | 4738 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 4739 | atomic_set(&eb->refs, 1); |
0b32f4bb | 4740 | atomic_set(&eb->io_pages, 0); |
727011e0 | 4741 | |
b8dae313 DS |
4742 | /* |
4743 | * Sanity checks, currently the maximum is 64k covered by 16x 4k pages | |
4744 | */ | |
4745 | BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE | |
4746 | > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
4747 | BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
d1310b2e CM |
4748 | |
4749 | return eb; | |
4750 | } | |
4751 | ||
815a51c7 JS |
4752 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) |
4753 | { | |
4754 | unsigned long i; | |
4755 | struct page *p; | |
4756 | struct extent_buffer *new; | |
4757 | unsigned long num_pages = num_extent_pages(src->start, src->len); | |
4758 | ||
3f556f78 | 4759 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
4760 | if (new == NULL) |
4761 | return NULL; | |
4762 | ||
4763 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4764 | p = alloc_page(GFP_NOFS); |
db7f3436 JB |
4765 | if (!p) { |
4766 | btrfs_release_extent_buffer(new); | |
4767 | return NULL; | |
4768 | } | |
815a51c7 JS |
4769 | attach_extent_buffer_page(new, p); |
4770 | WARN_ON(PageDirty(p)); | |
4771 | SetPageUptodate(p); | |
4772 | new->pages[i] = p; | |
fba1acf9 | 4773 | copy_page(page_address(p), page_address(src->pages[i])); |
815a51c7 JS |
4774 | } |
4775 | ||
815a51c7 JS |
4776 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); |
4777 | set_bit(EXTENT_BUFFER_DUMMY, &new->bflags); | |
4778 | ||
4779 | return new; | |
4780 | } | |
4781 | ||
0f331229 OS |
4782 | struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
4783 | u64 start, unsigned long len) | |
815a51c7 JS |
4784 | { |
4785 | struct extent_buffer *eb; | |
3f556f78 | 4786 | unsigned long num_pages; |
815a51c7 JS |
4787 | unsigned long i; |
4788 | ||
0f331229 | 4789 | num_pages = num_extent_pages(start, len); |
3f556f78 DS |
4790 | |
4791 | eb = __alloc_extent_buffer(fs_info, start, len); | |
815a51c7 JS |
4792 | if (!eb) |
4793 | return NULL; | |
4794 | ||
4795 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4796 | eb->pages[i] = alloc_page(GFP_NOFS); |
815a51c7 JS |
4797 | if (!eb->pages[i]) |
4798 | goto err; | |
4799 | } | |
4800 | set_extent_buffer_uptodate(eb); | |
4801 | btrfs_set_header_nritems(eb, 0); | |
4802 | set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4803 | ||
4804 | return eb; | |
4805 | err: | |
84167d19 SB |
4806 | for (; i > 0; i--) |
4807 | __free_page(eb->pages[i - 1]); | |
815a51c7 JS |
4808 | __free_extent_buffer(eb); |
4809 | return NULL; | |
4810 | } | |
4811 | ||
0f331229 | 4812 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
da17066c | 4813 | u64 start) |
0f331229 | 4814 | { |
da17066c | 4815 | return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize); |
0f331229 OS |
4816 | } |
4817 | ||
0b32f4bb JB |
4818 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
4819 | { | |
242e18c7 | 4820 | int refs; |
0b32f4bb JB |
4821 | /* the ref bit is tricky. We have to make sure it is set |
4822 | * if we have the buffer dirty. Otherwise the | |
4823 | * code to free a buffer can end up dropping a dirty | |
4824 | * page | |
4825 | * | |
4826 | * Once the ref bit is set, it won't go away while the | |
4827 | * buffer is dirty or in writeback, and it also won't | |
4828 | * go away while we have the reference count on the | |
4829 | * eb bumped. | |
4830 | * | |
4831 | * We can't just set the ref bit without bumping the | |
4832 | * ref on the eb because free_extent_buffer might | |
4833 | * see the ref bit and try to clear it. If this happens | |
4834 | * free_extent_buffer might end up dropping our original | |
4835 | * ref by mistake and freeing the page before we are able | |
4836 | * to add one more ref. | |
4837 | * | |
4838 | * So bump the ref count first, then set the bit. If someone | |
4839 | * beat us to it, drop the ref we added. | |
4840 | */ | |
242e18c7 CM |
4841 | refs = atomic_read(&eb->refs); |
4842 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
4843 | return; | |
4844 | ||
594831c4 JB |
4845 | spin_lock(&eb->refs_lock); |
4846 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 4847 | atomic_inc(&eb->refs); |
594831c4 | 4848 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
4849 | } |
4850 | ||
2457aec6 MG |
4851 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
4852 | struct page *accessed) | |
5df4235e JB |
4853 | { |
4854 | unsigned long num_pages, i; | |
4855 | ||
0b32f4bb JB |
4856 | check_buffer_tree_ref(eb); |
4857 | ||
5df4235e JB |
4858 | num_pages = num_extent_pages(eb->start, eb->len); |
4859 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a DS |
4860 | struct page *p = eb->pages[i]; |
4861 | ||
2457aec6 MG |
4862 | if (p != accessed) |
4863 | mark_page_accessed(p); | |
5df4235e JB |
4864 | } |
4865 | } | |
4866 | ||
f28491e0 JB |
4867 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
4868 | u64 start) | |
452c75c3 CS |
4869 | { |
4870 | struct extent_buffer *eb; | |
4871 | ||
4872 | rcu_read_lock(); | |
f28491e0 | 4873 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
09cbfeaf | 4874 | start >> PAGE_SHIFT); |
452c75c3 CS |
4875 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
4876 | rcu_read_unlock(); | |
062c19e9 FM |
4877 | /* |
4878 | * Lock our eb's refs_lock to avoid races with | |
4879 | * free_extent_buffer. When we get our eb it might be flagged | |
4880 | * with EXTENT_BUFFER_STALE and another task running | |
4881 | * free_extent_buffer might have seen that flag set, | |
4882 | * eb->refs == 2, that the buffer isn't under IO (dirty and | |
4883 | * writeback flags not set) and it's still in the tree (flag | |
4884 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process | |
4885 | * of decrementing the extent buffer's reference count twice. | |
4886 | * So here we could race and increment the eb's reference count, | |
4887 | * clear its stale flag, mark it as dirty and drop our reference | |
4888 | * before the other task finishes executing free_extent_buffer, | |
4889 | * which would later result in an attempt to free an extent | |
4890 | * buffer that is dirty. | |
4891 | */ | |
4892 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
4893 | spin_lock(&eb->refs_lock); | |
4894 | spin_unlock(&eb->refs_lock); | |
4895 | } | |
2457aec6 | 4896 | mark_extent_buffer_accessed(eb, NULL); |
452c75c3 CS |
4897 | return eb; |
4898 | } | |
4899 | rcu_read_unlock(); | |
4900 | ||
4901 | return NULL; | |
4902 | } | |
4903 | ||
faa2dbf0 JB |
4904 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
4905 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
da17066c | 4906 | u64 start) |
faa2dbf0 JB |
4907 | { |
4908 | struct extent_buffer *eb, *exists = NULL; | |
4909 | int ret; | |
4910 | ||
4911 | eb = find_extent_buffer(fs_info, start); | |
4912 | if (eb) | |
4913 | return eb; | |
da17066c | 4914 | eb = alloc_dummy_extent_buffer(fs_info, start); |
faa2dbf0 JB |
4915 | if (!eb) |
4916 | return NULL; | |
4917 | eb->fs_info = fs_info; | |
4918 | again: | |
e1860a77 | 4919 | ret = radix_tree_preload(GFP_NOFS); |
faa2dbf0 JB |
4920 | if (ret) |
4921 | goto free_eb; | |
4922 | spin_lock(&fs_info->buffer_lock); | |
4923 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
09cbfeaf | 4924 | start >> PAGE_SHIFT, eb); |
faa2dbf0 JB |
4925 | spin_unlock(&fs_info->buffer_lock); |
4926 | radix_tree_preload_end(); | |
4927 | if (ret == -EEXIST) { | |
4928 | exists = find_extent_buffer(fs_info, start); | |
4929 | if (exists) | |
4930 | goto free_eb; | |
4931 | else | |
4932 | goto again; | |
4933 | } | |
4934 | check_buffer_tree_ref(eb); | |
4935 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
4936 | ||
4937 | /* | |
4938 | * We will free dummy extent buffer's if they come into | |
4939 | * free_extent_buffer with a ref count of 2, but if we are using this we | |
4940 | * want the buffers to stay in memory until we're done with them, so | |
4941 | * bump the ref count again. | |
4942 | */ | |
4943 | atomic_inc(&eb->refs); | |
4944 | return eb; | |
4945 | free_eb: | |
4946 | btrfs_release_extent_buffer(eb); | |
4947 | return exists; | |
4948 | } | |
4949 | #endif | |
4950 | ||
f28491e0 | 4951 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
ce3e6984 | 4952 | u64 start) |
d1310b2e | 4953 | { |
da17066c | 4954 | unsigned long len = fs_info->nodesize; |
d1310b2e CM |
4955 | unsigned long num_pages = num_extent_pages(start, len); |
4956 | unsigned long i; | |
09cbfeaf | 4957 | unsigned long index = start >> PAGE_SHIFT; |
d1310b2e | 4958 | struct extent_buffer *eb; |
6af118ce | 4959 | struct extent_buffer *exists = NULL; |
d1310b2e | 4960 | struct page *p; |
f28491e0 | 4961 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
d1310b2e | 4962 | int uptodate = 1; |
19fe0a8b | 4963 | int ret; |
d1310b2e | 4964 | |
da17066c | 4965 | if (!IS_ALIGNED(start, fs_info->sectorsize)) { |
c871b0f2 LB |
4966 | btrfs_err(fs_info, "bad tree block start %llu", start); |
4967 | return ERR_PTR(-EINVAL); | |
4968 | } | |
4969 | ||
f28491e0 | 4970 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 4971 | if (eb) |
6af118ce | 4972 | return eb; |
6af118ce | 4973 | |
23d79d81 | 4974 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 4975 | if (!eb) |
c871b0f2 | 4976 | return ERR_PTR(-ENOMEM); |
d1310b2e | 4977 | |
727011e0 | 4978 | for (i = 0; i < num_pages; i++, index++) { |
d1b5c567 | 4979 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
c871b0f2 LB |
4980 | if (!p) { |
4981 | exists = ERR_PTR(-ENOMEM); | |
6af118ce | 4982 | goto free_eb; |
c871b0f2 | 4983 | } |
4f2de97a JB |
4984 | |
4985 | spin_lock(&mapping->private_lock); | |
4986 | if (PagePrivate(p)) { | |
4987 | /* | |
4988 | * We could have already allocated an eb for this page | |
4989 | * and attached one so lets see if we can get a ref on | |
4990 | * the existing eb, and if we can we know it's good and | |
4991 | * we can just return that one, else we know we can just | |
4992 | * overwrite page->private. | |
4993 | */ | |
4994 | exists = (struct extent_buffer *)p->private; | |
4995 | if (atomic_inc_not_zero(&exists->refs)) { | |
4996 | spin_unlock(&mapping->private_lock); | |
4997 | unlock_page(p); | |
09cbfeaf | 4998 | put_page(p); |
2457aec6 | 4999 | mark_extent_buffer_accessed(exists, p); |
4f2de97a JB |
5000 | goto free_eb; |
5001 | } | |
5ca64f45 | 5002 | exists = NULL; |
4f2de97a | 5003 | |
0b32f4bb | 5004 | /* |
4f2de97a JB |
5005 | * Do this so attach doesn't complain and we need to |
5006 | * drop the ref the old guy had. | |
5007 | */ | |
5008 | ClearPagePrivate(p); | |
0b32f4bb | 5009 | WARN_ON(PageDirty(p)); |
09cbfeaf | 5010 | put_page(p); |
d1310b2e | 5011 | } |
4f2de97a JB |
5012 | attach_extent_buffer_page(eb, p); |
5013 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 5014 | WARN_ON(PageDirty(p)); |
727011e0 | 5015 | eb->pages[i] = p; |
d1310b2e CM |
5016 | if (!PageUptodate(p)) |
5017 | uptodate = 0; | |
eb14ab8e CM |
5018 | |
5019 | /* | |
5020 | * see below about how we avoid a nasty race with release page | |
5021 | * and why we unlock later | |
5022 | */ | |
d1310b2e CM |
5023 | } |
5024 | if (uptodate) | |
b4ce94de | 5025 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 5026 | again: |
e1860a77 | 5027 | ret = radix_tree_preload(GFP_NOFS); |
c871b0f2 LB |
5028 | if (ret) { |
5029 | exists = ERR_PTR(ret); | |
19fe0a8b | 5030 | goto free_eb; |
c871b0f2 | 5031 | } |
19fe0a8b | 5032 | |
f28491e0 JB |
5033 | spin_lock(&fs_info->buffer_lock); |
5034 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
09cbfeaf | 5035 | start >> PAGE_SHIFT, eb); |
f28491e0 | 5036 | spin_unlock(&fs_info->buffer_lock); |
452c75c3 | 5037 | radix_tree_preload_end(); |
19fe0a8b | 5038 | if (ret == -EEXIST) { |
f28491e0 | 5039 | exists = find_extent_buffer(fs_info, start); |
452c75c3 CS |
5040 | if (exists) |
5041 | goto free_eb; | |
5042 | else | |
115391d2 | 5043 | goto again; |
6af118ce | 5044 | } |
6af118ce | 5045 | /* add one reference for the tree */ |
0b32f4bb | 5046 | check_buffer_tree_ref(eb); |
34b41ace | 5047 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
5048 | |
5049 | /* | |
5050 | * there is a race where release page may have | |
5051 | * tried to find this extent buffer in the radix | |
5052 | * but failed. It will tell the VM it is safe to | |
5053 | * reclaim the, and it will clear the page private bit. | |
5054 | * We must make sure to set the page private bit properly | |
5055 | * after the extent buffer is in the radix tree so | |
5056 | * it doesn't get lost | |
5057 | */ | |
727011e0 CM |
5058 | SetPageChecked(eb->pages[0]); |
5059 | for (i = 1; i < num_pages; i++) { | |
fb85fc9a | 5060 | p = eb->pages[i]; |
727011e0 CM |
5061 | ClearPageChecked(p); |
5062 | unlock_page(p); | |
5063 | } | |
5064 | unlock_page(eb->pages[0]); | |
d1310b2e CM |
5065 | return eb; |
5066 | ||
6af118ce | 5067 | free_eb: |
5ca64f45 | 5068 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
5069 | for (i = 0; i < num_pages; i++) { |
5070 | if (eb->pages[i]) | |
5071 | unlock_page(eb->pages[i]); | |
5072 | } | |
eb14ab8e | 5073 | |
897ca6e9 | 5074 | btrfs_release_extent_buffer(eb); |
6af118ce | 5075 | return exists; |
d1310b2e | 5076 | } |
d1310b2e | 5077 | |
3083ee2e JB |
5078 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
5079 | { | |
5080 | struct extent_buffer *eb = | |
5081 | container_of(head, struct extent_buffer, rcu_head); | |
5082 | ||
5083 | __free_extent_buffer(eb); | |
5084 | } | |
5085 | ||
3083ee2e | 5086 | /* Expects to have eb->eb_lock already held */ |
f7a52a40 | 5087 | static int release_extent_buffer(struct extent_buffer *eb) |
3083ee2e JB |
5088 | { |
5089 | WARN_ON(atomic_read(&eb->refs) == 0); | |
5090 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 5091 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 5092 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 5093 | |
815a51c7 | 5094 | spin_unlock(&eb->refs_lock); |
3083ee2e | 5095 | |
f28491e0 JB |
5096 | spin_lock(&fs_info->buffer_lock); |
5097 | radix_tree_delete(&fs_info->buffer_radix, | |
09cbfeaf | 5098 | eb->start >> PAGE_SHIFT); |
f28491e0 | 5099 | spin_unlock(&fs_info->buffer_lock); |
34b41ace JB |
5100 | } else { |
5101 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 5102 | } |
3083ee2e JB |
5103 | |
5104 | /* Should be safe to release our pages at this point */ | |
a50924e3 | 5105 | btrfs_release_extent_buffer_page(eb); |
bcb7e449 JB |
5106 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
5107 | if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) { | |
5108 | __free_extent_buffer(eb); | |
5109 | return 1; | |
5110 | } | |
5111 | #endif | |
3083ee2e | 5112 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 5113 | return 1; |
3083ee2e JB |
5114 | } |
5115 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
5116 | |
5117 | return 0; | |
3083ee2e JB |
5118 | } |
5119 | ||
d1310b2e CM |
5120 | void free_extent_buffer(struct extent_buffer *eb) |
5121 | { | |
242e18c7 CM |
5122 | int refs; |
5123 | int old; | |
d1310b2e CM |
5124 | if (!eb) |
5125 | return; | |
5126 | ||
242e18c7 CM |
5127 | while (1) { |
5128 | refs = atomic_read(&eb->refs); | |
5129 | if (refs <= 3) | |
5130 | break; | |
5131 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
5132 | if (old == refs) | |
5133 | return; | |
5134 | } | |
5135 | ||
3083ee2e | 5136 | spin_lock(&eb->refs_lock); |
815a51c7 JS |
5137 | if (atomic_read(&eb->refs) == 2 && |
5138 | test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) | |
5139 | atomic_dec(&eb->refs); | |
5140 | ||
3083ee2e JB |
5141 | if (atomic_read(&eb->refs) == 2 && |
5142 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 5143 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
5144 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5145 | atomic_dec(&eb->refs); | |
5146 | ||
5147 | /* | |
5148 | * I know this is terrible, but it's temporary until we stop tracking | |
5149 | * the uptodate bits and such for the extent buffers. | |
5150 | */ | |
f7a52a40 | 5151 | release_extent_buffer(eb); |
3083ee2e JB |
5152 | } |
5153 | ||
5154 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
5155 | { | |
5156 | if (!eb) | |
d1310b2e CM |
5157 | return; |
5158 | ||
3083ee2e JB |
5159 | spin_lock(&eb->refs_lock); |
5160 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
5161 | ||
0b32f4bb | 5162 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
5163 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5164 | atomic_dec(&eb->refs); | |
f7a52a40 | 5165 | release_extent_buffer(eb); |
d1310b2e | 5166 | } |
d1310b2e | 5167 | |
1d4284bd | 5168 | void clear_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5169 | { |
d1310b2e CM |
5170 | unsigned long i; |
5171 | unsigned long num_pages; | |
5172 | struct page *page; | |
5173 | ||
d1310b2e CM |
5174 | num_pages = num_extent_pages(eb->start, eb->len); |
5175 | ||
5176 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 5177 | page = eb->pages[i]; |
b9473439 | 5178 | if (!PageDirty(page)) |
d2c3f4f6 CM |
5179 | continue; |
5180 | ||
a61e6f29 | 5181 | lock_page(page); |
eb14ab8e CM |
5182 | WARN_ON(!PagePrivate(page)); |
5183 | ||
d1310b2e | 5184 | clear_page_dirty_for_io(page); |
0ee0fda0 | 5185 | spin_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
5186 | if (!PageDirty(page)) { |
5187 | radix_tree_tag_clear(&page->mapping->page_tree, | |
5188 | page_index(page), | |
5189 | PAGECACHE_TAG_DIRTY); | |
5190 | } | |
0ee0fda0 | 5191 | spin_unlock_irq(&page->mapping->tree_lock); |
bf0da8c1 | 5192 | ClearPageError(page); |
a61e6f29 | 5193 | unlock_page(page); |
d1310b2e | 5194 | } |
0b32f4bb | 5195 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 5196 | } |
d1310b2e | 5197 | |
0b32f4bb | 5198 | int set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e CM |
5199 | { |
5200 | unsigned long i; | |
5201 | unsigned long num_pages; | |
b9473439 | 5202 | int was_dirty = 0; |
d1310b2e | 5203 | |
0b32f4bb JB |
5204 | check_buffer_tree_ref(eb); |
5205 | ||
b9473439 | 5206 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 5207 | |
d1310b2e | 5208 | num_pages = num_extent_pages(eb->start, eb->len); |
3083ee2e | 5209 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
5210 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
5211 | ||
b9473439 | 5212 | for (i = 0; i < num_pages; i++) |
fb85fc9a | 5213 | set_page_dirty(eb->pages[i]); |
b9473439 | 5214 | return was_dirty; |
d1310b2e | 5215 | } |
d1310b2e | 5216 | |
69ba3927 | 5217 | void clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 CM |
5218 | { |
5219 | unsigned long i; | |
5220 | struct page *page; | |
5221 | unsigned long num_pages; | |
5222 | ||
b4ce94de | 5223 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
0b32f4bb | 5224 | num_pages = num_extent_pages(eb->start, eb->len); |
1259ab75 | 5225 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5226 | page = eb->pages[i]; |
33958dc6 CM |
5227 | if (page) |
5228 | ClearPageUptodate(page); | |
1259ab75 | 5229 | } |
1259ab75 CM |
5230 | } |
5231 | ||
09c25a8c | 5232 | void set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e CM |
5233 | { |
5234 | unsigned long i; | |
5235 | struct page *page; | |
5236 | unsigned long num_pages; | |
5237 | ||
0b32f4bb | 5238 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 5239 | num_pages = num_extent_pages(eb->start, eb->len); |
d1310b2e | 5240 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5241 | page = eb->pages[i]; |
d1310b2e CM |
5242 | SetPageUptodate(page); |
5243 | } | |
d1310b2e | 5244 | } |
d1310b2e | 5245 | |
0b32f4bb | 5246 | int extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 5247 | { |
0b32f4bb | 5248 | return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 5249 | } |
d1310b2e CM |
5250 | |
5251 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
8436ea91 | 5252 | struct extent_buffer *eb, int wait, |
f188591e | 5253 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
5254 | { |
5255 | unsigned long i; | |
d1310b2e CM |
5256 | struct page *page; |
5257 | int err; | |
5258 | int ret = 0; | |
ce9adaa5 CM |
5259 | int locked_pages = 0; |
5260 | int all_uptodate = 1; | |
d1310b2e | 5261 | unsigned long num_pages; |
727011e0 | 5262 | unsigned long num_reads = 0; |
a86c12c7 | 5263 | struct bio *bio = NULL; |
c8b97818 | 5264 | unsigned long bio_flags = 0; |
a86c12c7 | 5265 | |
b4ce94de | 5266 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
5267 | return 0; |
5268 | ||
d1310b2e | 5269 | num_pages = num_extent_pages(eb->start, eb->len); |
8436ea91 | 5270 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5271 | page = eb->pages[i]; |
bb82ab88 | 5272 | if (wait == WAIT_NONE) { |
2db04966 | 5273 | if (!trylock_page(page)) |
ce9adaa5 | 5274 | goto unlock_exit; |
d1310b2e CM |
5275 | } else { |
5276 | lock_page(page); | |
5277 | } | |
ce9adaa5 | 5278 | locked_pages++; |
2571e739 LB |
5279 | } |
5280 | /* | |
5281 | * We need to firstly lock all pages to make sure that | |
5282 | * the uptodate bit of our pages won't be affected by | |
5283 | * clear_extent_buffer_uptodate(). | |
5284 | */ | |
8436ea91 | 5285 | for (i = 0; i < num_pages; i++) { |
2571e739 | 5286 | page = eb->pages[i]; |
727011e0 CM |
5287 | if (!PageUptodate(page)) { |
5288 | num_reads++; | |
ce9adaa5 | 5289 | all_uptodate = 0; |
727011e0 | 5290 | } |
ce9adaa5 | 5291 | } |
2571e739 | 5292 | |
ce9adaa5 | 5293 | if (all_uptodate) { |
8436ea91 | 5294 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
5295 | goto unlock_exit; |
5296 | } | |
5297 | ||
656f30db | 5298 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 5299 | eb->read_mirror = 0; |
0b32f4bb | 5300 | atomic_set(&eb->io_pages, num_reads); |
8436ea91 | 5301 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5302 | page = eb->pages[i]; |
baf863b9 | 5303 | |
ce9adaa5 | 5304 | if (!PageUptodate(page)) { |
baf863b9 LB |
5305 | if (ret) { |
5306 | atomic_dec(&eb->io_pages); | |
5307 | unlock_page(page); | |
5308 | continue; | |
5309 | } | |
5310 | ||
f188591e | 5311 | ClearPageError(page); |
a86c12c7 | 5312 | err = __extent_read_full_page(tree, page, |
f188591e | 5313 | get_extent, &bio, |
d4c7ca86 | 5314 | mirror_num, &bio_flags, |
1f7ad75b | 5315 | REQ_META); |
baf863b9 | 5316 | if (err) { |
d1310b2e | 5317 | ret = err; |
baf863b9 LB |
5318 | /* |
5319 | * We use &bio in above __extent_read_full_page, | |
5320 | * so we ensure that if it returns error, the | |
5321 | * current page fails to add itself to bio and | |
5322 | * it's been unlocked. | |
5323 | * | |
5324 | * We must dec io_pages by ourselves. | |
5325 | */ | |
5326 | atomic_dec(&eb->io_pages); | |
5327 | } | |
d1310b2e CM |
5328 | } else { |
5329 | unlock_page(page); | |
5330 | } | |
5331 | } | |
5332 | ||
355808c2 | 5333 | if (bio) { |
1f7ad75b | 5334 | err = submit_one_bio(bio, mirror_num, bio_flags); |
79787eaa JM |
5335 | if (err) |
5336 | return err; | |
355808c2 | 5337 | } |
a86c12c7 | 5338 | |
bb82ab88 | 5339 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 5340 | return ret; |
d397712b | 5341 | |
8436ea91 | 5342 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5343 | page = eb->pages[i]; |
d1310b2e | 5344 | wait_on_page_locked(page); |
d397712b | 5345 | if (!PageUptodate(page)) |
d1310b2e | 5346 | ret = -EIO; |
d1310b2e | 5347 | } |
d397712b | 5348 | |
d1310b2e | 5349 | return ret; |
ce9adaa5 CM |
5350 | |
5351 | unlock_exit: | |
d397712b | 5352 | while (locked_pages > 0) { |
ce9adaa5 | 5353 | locked_pages--; |
8436ea91 JB |
5354 | page = eb->pages[locked_pages]; |
5355 | unlock_page(page); | |
ce9adaa5 CM |
5356 | } |
5357 | return ret; | |
d1310b2e | 5358 | } |
d1310b2e | 5359 | |
1cbb1f45 JM |
5360 | void read_extent_buffer(const struct extent_buffer *eb, void *dstv, |
5361 | unsigned long start, unsigned long len) | |
d1310b2e CM |
5362 | { |
5363 | size_t cur; | |
5364 | size_t offset; | |
5365 | struct page *page; | |
5366 | char *kaddr; | |
5367 | char *dst = (char *)dstv; | |
09cbfeaf KS |
5368 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5369 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e | 5370 | |
f716abd5 LB |
5371 | if (start + len > eb->len) { |
5372 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n", | |
5373 | eb->start, eb->len, start, len); | |
5374 | memset(dst, 0, len); | |
5375 | return; | |
5376 | } | |
d1310b2e | 5377 | |
09cbfeaf | 5378 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
d1310b2e | 5379 | |
d397712b | 5380 | while (len > 0) { |
fb85fc9a | 5381 | page = eb->pages[i]; |
d1310b2e | 5382 | |
09cbfeaf | 5383 | cur = min(len, (PAGE_SIZE - offset)); |
a6591715 | 5384 | kaddr = page_address(page); |
d1310b2e | 5385 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
5386 | |
5387 | dst += cur; | |
5388 | len -= cur; | |
5389 | offset = 0; | |
5390 | i++; | |
5391 | } | |
5392 | } | |
d1310b2e | 5393 | |
1cbb1f45 JM |
5394 | int read_extent_buffer_to_user(const struct extent_buffer *eb, |
5395 | void __user *dstv, | |
5396 | unsigned long start, unsigned long len) | |
550ac1d8 GH |
5397 | { |
5398 | size_t cur; | |
5399 | size_t offset; | |
5400 | struct page *page; | |
5401 | char *kaddr; | |
5402 | char __user *dst = (char __user *)dstv; | |
09cbfeaf KS |
5403 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5404 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
550ac1d8 GH |
5405 | int ret = 0; |
5406 | ||
5407 | WARN_ON(start > eb->len); | |
5408 | WARN_ON(start + len > eb->start + eb->len); | |
5409 | ||
09cbfeaf | 5410 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
550ac1d8 GH |
5411 | |
5412 | while (len > 0) { | |
fb85fc9a | 5413 | page = eb->pages[i]; |
550ac1d8 | 5414 | |
09cbfeaf | 5415 | cur = min(len, (PAGE_SIZE - offset)); |
550ac1d8 GH |
5416 | kaddr = page_address(page); |
5417 | if (copy_to_user(dst, kaddr + offset, cur)) { | |
5418 | ret = -EFAULT; | |
5419 | break; | |
5420 | } | |
5421 | ||
5422 | dst += cur; | |
5423 | len -= cur; | |
5424 | offset = 0; | |
5425 | i++; | |
5426 | } | |
5427 | ||
5428 | return ret; | |
5429 | } | |
5430 | ||
415b35a5 LB |
5431 | /* |
5432 | * return 0 if the item is found within a page. | |
5433 | * return 1 if the item spans two pages. | |
5434 | * return -EINVAL otherwise. | |
5435 | */ | |
1cbb1f45 JM |
5436 | int map_private_extent_buffer(const struct extent_buffer *eb, |
5437 | unsigned long start, unsigned long min_len, | |
5438 | char **map, unsigned long *map_start, | |
5439 | unsigned long *map_len) | |
d1310b2e | 5440 | { |
09cbfeaf | 5441 | size_t offset = start & (PAGE_SIZE - 1); |
d1310b2e CM |
5442 | char *kaddr; |
5443 | struct page *p; | |
09cbfeaf KS |
5444 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5445 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e | 5446 | unsigned long end_i = (start_offset + start + min_len - 1) >> |
09cbfeaf | 5447 | PAGE_SHIFT; |
d1310b2e | 5448 | |
f716abd5 LB |
5449 | if (start + min_len > eb->len) { |
5450 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n", | |
5451 | eb->start, eb->len, start, min_len); | |
5452 | return -EINVAL; | |
5453 | } | |
5454 | ||
d1310b2e | 5455 | if (i != end_i) |
415b35a5 | 5456 | return 1; |
d1310b2e CM |
5457 | |
5458 | if (i == 0) { | |
5459 | offset = start_offset; | |
5460 | *map_start = 0; | |
5461 | } else { | |
5462 | offset = 0; | |
09cbfeaf | 5463 | *map_start = ((u64)i << PAGE_SHIFT) - start_offset; |
d1310b2e | 5464 | } |
d397712b | 5465 | |
fb85fc9a | 5466 | p = eb->pages[i]; |
a6591715 | 5467 | kaddr = page_address(p); |
d1310b2e | 5468 | *map = kaddr + offset; |
09cbfeaf | 5469 | *map_len = PAGE_SIZE - offset; |
d1310b2e CM |
5470 | return 0; |
5471 | } | |
d1310b2e | 5472 | |
1cbb1f45 JM |
5473 | int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv, |
5474 | unsigned long start, unsigned long len) | |
d1310b2e CM |
5475 | { |
5476 | size_t cur; | |
5477 | size_t offset; | |
5478 | struct page *page; | |
5479 | char *kaddr; | |
5480 | char *ptr = (char *)ptrv; | |
09cbfeaf KS |
5481 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5482 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e CM |
5483 | int ret = 0; |
5484 | ||
5485 | WARN_ON(start > eb->len); | |
5486 | WARN_ON(start + len > eb->start + eb->len); | |
5487 | ||
09cbfeaf | 5488 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
d1310b2e | 5489 | |
d397712b | 5490 | while (len > 0) { |
fb85fc9a | 5491 | page = eb->pages[i]; |
d1310b2e | 5492 | |
09cbfeaf | 5493 | cur = min(len, (PAGE_SIZE - offset)); |
d1310b2e | 5494 | |
a6591715 | 5495 | kaddr = page_address(page); |
d1310b2e | 5496 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
5497 | if (ret) |
5498 | break; | |
5499 | ||
5500 | ptr += cur; | |
5501 | len -= cur; | |
5502 | offset = 0; | |
5503 | i++; | |
5504 | } | |
5505 | return ret; | |
5506 | } | |
d1310b2e | 5507 | |
f157bf76 DS |
5508 | void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb, |
5509 | const void *srcv) | |
5510 | { | |
5511 | char *kaddr; | |
5512 | ||
5513 | WARN_ON(!PageUptodate(eb->pages[0])); | |
5514 | kaddr = page_address(eb->pages[0]); | |
5515 | memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv, | |
5516 | BTRFS_FSID_SIZE); | |
5517 | } | |
5518 | ||
5519 | void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv) | |
5520 | { | |
5521 | char *kaddr; | |
5522 | ||
5523 | WARN_ON(!PageUptodate(eb->pages[0])); | |
5524 | kaddr = page_address(eb->pages[0]); | |
5525 | memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv, | |
5526 | BTRFS_FSID_SIZE); | |
5527 | } | |
5528 | ||
d1310b2e CM |
5529 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, |
5530 | unsigned long start, unsigned long len) | |
5531 | { | |
5532 | size_t cur; | |
5533 | size_t offset; | |
5534 | struct page *page; | |
5535 | char *kaddr; | |
5536 | char *src = (char *)srcv; | |
09cbfeaf KS |
5537 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5538 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e CM |
5539 | |
5540 | WARN_ON(start > eb->len); | |
5541 | WARN_ON(start + len > eb->start + eb->len); | |
5542 | ||
09cbfeaf | 5543 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
d1310b2e | 5544 | |
d397712b | 5545 | while (len > 0) { |
fb85fc9a | 5546 | page = eb->pages[i]; |
d1310b2e CM |
5547 | WARN_ON(!PageUptodate(page)); |
5548 | ||
09cbfeaf | 5549 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 5550 | kaddr = page_address(page); |
d1310b2e | 5551 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
5552 | |
5553 | src += cur; | |
5554 | len -= cur; | |
5555 | offset = 0; | |
5556 | i++; | |
5557 | } | |
5558 | } | |
d1310b2e | 5559 | |
b159fa28 DS |
5560 | void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start, |
5561 | unsigned long len) | |
d1310b2e CM |
5562 | { |
5563 | size_t cur; | |
5564 | size_t offset; | |
5565 | struct page *page; | |
5566 | char *kaddr; | |
09cbfeaf KS |
5567 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5568 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e CM |
5569 | |
5570 | WARN_ON(start > eb->len); | |
5571 | WARN_ON(start + len > eb->start + eb->len); | |
5572 | ||
09cbfeaf | 5573 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
d1310b2e | 5574 | |
d397712b | 5575 | while (len > 0) { |
fb85fc9a | 5576 | page = eb->pages[i]; |
d1310b2e CM |
5577 | WARN_ON(!PageUptodate(page)); |
5578 | ||
09cbfeaf | 5579 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 5580 | kaddr = page_address(page); |
b159fa28 | 5581 | memset(kaddr + offset, 0, cur); |
d1310b2e CM |
5582 | |
5583 | len -= cur; | |
5584 | offset = 0; | |
5585 | i++; | |
5586 | } | |
5587 | } | |
d1310b2e | 5588 | |
58e8012c DS |
5589 | void copy_extent_buffer_full(struct extent_buffer *dst, |
5590 | struct extent_buffer *src) | |
5591 | { | |
5592 | int i; | |
5593 | unsigned num_pages; | |
5594 | ||
5595 | ASSERT(dst->len == src->len); | |
5596 | ||
5597 | num_pages = num_extent_pages(dst->start, dst->len); | |
5598 | for (i = 0; i < num_pages; i++) | |
5599 | copy_page(page_address(dst->pages[i]), | |
5600 | page_address(src->pages[i])); | |
5601 | } | |
5602 | ||
d1310b2e CM |
5603 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, |
5604 | unsigned long dst_offset, unsigned long src_offset, | |
5605 | unsigned long len) | |
5606 | { | |
5607 | u64 dst_len = dst->len; | |
5608 | size_t cur; | |
5609 | size_t offset; | |
5610 | struct page *page; | |
5611 | char *kaddr; | |
09cbfeaf KS |
5612 | size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1); |
5613 | unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT; | |
d1310b2e CM |
5614 | |
5615 | WARN_ON(src->len != dst_len); | |
5616 | ||
5617 | offset = (start_offset + dst_offset) & | |
09cbfeaf | 5618 | (PAGE_SIZE - 1); |
d1310b2e | 5619 | |
d397712b | 5620 | while (len > 0) { |
fb85fc9a | 5621 | page = dst->pages[i]; |
d1310b2e CM |
5622 | WARN_ON(!PageUptodate(page)); |
5623 | ||
09cbfeaf | 5624 | cur = min(len, (unsigned long)(PAGE_SIZE - offset)); |
d1310b2e | 5625 | |
a6591715 | 5626 | kaddr = page_address(page); |
d1310b2e | 5627 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
5628 | |
5629 | src_offset += cur; | |
5630 | len -= cur; | |
5631 | offset = 0; | |
5632 | i++; | |
5633 | } | |
5634 | } | |
d1310b2e | 5635 | |
2fe1d551 OS |
5636 | void le_bitmap_set(u8 *map, unsigned int start, int len) |
5637 | { | |
5638 | u8 *p = map + BIT_BYTE(start); | |
5639 | const unsigned int size = start + len; | |
5640 | int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE); | |
5641 | u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start); | |
5642 | ||
5643 | while (len - bits_to_set >= 0) { | |
5644 | *p |= mask_to_set; | |
5645 | len -= bits_to_set; | |
5646 | bits_to_set = BITS_PER_BYTE; | |
9c894696 | 5647 | mask_to_set = ~0; |
2fe1d551 OS |
5648 | p++; |
5649 | } | |
5650 | if (len) { | |
5651 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
5652 | *p |= mask_to_set; | |
5653 | } | |
5654 | } | |
5655 | ||
5656 | void le_bitmap_clear(u8 *map, unsigned int start, int len) | |
5657 | { | |
5658 | u8 *p = map + BIT_BYTE(start); | |
5659 | const unsigned int size = start + len; | |
5660 | int bits_to_clear = BITS_PER_BYTE - (start % BITS_PER_BYTE); | |
5661 | u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(start); | |
5662 | ||
5663 | while (len - bits_to_clear >= 0) { | |
5664 | *p &= ~mask_to_clear; | |
5665 | len -= bits_to_clear; | |
5666 | bits_to_clear = BITS_PER_BYTE; | |
9c894696 | 5667 | mask_to_clear = ~0; |
2fe1d551 OS |
5668 | p++; |
5669 | } | |
5670 | if (len) { | |
5671 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
5672 | *p &= ~mask_to_clear; | |
5673 | } | |
5674 | } | |
3e1e8bb7 OS |
5675 | |
5676 | /* | |
5677 | * eb_bitmap_offset() - calculate the page and offset of the byte containing the | |
5678 | * given bit number | |
5679 | * @eb: the extent buffer | |
5680 | * @start: offset of the bitmap item in the extent buffer | |
5681 | * @nr: bit number | |
5682 | * @page_index: return index of the page in the extent buffer that contains the | |
5683 | * given bit number | |
5684 | * @page_offset: return offset into the page given by page_index | |
5685 | * | |
5686 | * This helper hides the ugliness of finding the byte in an extent buffer which | |
5687 | * contains a given bit. | |
5688 | */ | |
5689 | static inline void eb_bitmap_offset(struct extent_buffer *eb, | |
5690 | unsigned long start, unsigned long nr, | |
5691 | unsigned long *page_index, | |
5692 | size_t *page_offset) | |
5693 | { | |
09cbfeaf | 5694 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
3e1e8bb7 OS |
5695 | size_t byte_offset = BIT_BYTE(nr); |
5696 | size_t offset; | |
5697 | ||
5698 | /* | |
5699 | * The byte we want is the offset of the extent buffer + the offset of | |
5700 | * the bitmap item in the extent buffer + the offset of the byte in the | |
5701 | * bitmap item. | |
5702 | */ | |
5703 | offset = start_offset + start + byte_offset; | |
5704 | ||
09cbfeaf KS |
5705 | *page_index = offset >> PAGE_SHIFT; |
5706 | *page_offset = offset & (PAGE_SIZE - 1); | |
3e1e8bb7 OS |
5707 | } |
5708 | ||
5709 | /** | |
5710 | * extent_buffer_test_bit - determine whether a bit in a bitmap item is set | |
5711 | * @eb: the extent buffer | |
5712 | * @start: offset of the bitmap item in the extent buffer | |
5713 | * @nr: bit number to test | |
5714 | */ | |
5715 | int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start, | |
5716 | unsigned long nr) | |
5717 | { | |
2fe1d551 | 5718 | u8 *kaddr; |
3e1e8bb7 OS |
5719 | struct page *page; |
5720 | unsigned long i; | |
5721 | size_t offset; | |
5722 | ||
5723 | eb_bitmap_offset(eb, start, nr, &i, &offset); | |
5724 | page = eb->pages[i]; | |
5725 | WARN_ON(!PageUptodate(page)); | |
5726 | kaddr = page_address(page); | |
5727 | return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1))); | |
5728 | } | |
5729 | ||
5730 | /** | |
5731 | * extent_buffer_bitmap_set - set an area of a bitmap | |
5732 | * @eb: the extent buffer | |
5733 | * @start: offset of the bitmap item in the extent buffer | |
5734 | * @pos: bit number of the first bit | |
5735 | * @len: number of bits to set | |
5736 | */ | |
5737 | void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start, | |
5738 | unsigned long pos, unsigned long len) | |
5739 | { | |
2fe1d551 | 5740 | u8 *kaddr; |
3e1e8bb7 OS |
5741 | struct page *page; |
5742 | unsigned long i; | |
5743 | size_t offset; | |
5744 | const unsigned int size = pos + len; | |
5745 | int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 5746 | u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
5747 | |
5748 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5749 | page = eb->pages[i]; | |
5750 | WARN_ON(!PageUptodate(page)); | |
5751 | kaddr = page_address(page); | |
5752 | ||
5753 | while (len >= bits_to_set) { | |
5754 | kaddr[offset] |= mask_to_set; | |
5755 | len -= bits_to_set; | |
5756 | bits_to_set = BITS_PER_BYTE; | |
9c894696 | 5757 | mask_to_set = ~0; |
09cbfeaf | 5758 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5759 | offset = 0; |
5760 | page = eb->pages[++i]; | |
5761 | WARN_ON(!PageUptodate(page)); | |
5762 | kaddr = page_address(page); | |
5763 | } | |
5764 | } | |
5765 | if (len) { | |
5766 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
5767 | kaddr[offset] |= mask_to_set; | |
5768 | } | |
5769 | } | |
5770 | ||
5771 | ||
5772 | /** | |
5773 | * extent_buffer_bitmap_clear - clear an area of a bitmap | |
5774 | * @eb: the extent buffer | |
5775 | * @start: offset of the bitmap item in the extent buffer | |
5776 | * @pos: bit number of the first bit | |
5777 | * @len: number of bits to clear | |
5778 | */ | |
5779 | void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start, | |
5780 | unsigned long pos, unsigned long len) | |
5781 | { | |
2fe1d551 | 5782 | u8 *kaddr; |
3e1e8bb7 OS |
5783 | struct page *page; |
5784 | unsigned long i; | |
5785 | size_t offset; | |
5786 | const unsigned int size = pos + len; | |
5787 | int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 5788 | u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
5789 | |
5790 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5791 | page = eb->pages[i]; | |
5792 | WARN_ON(!PageUptodate(page)); | |
5793 | kaddr = page_address(page); | |
5794 | ||
5795 | while (len >= bits_to_clear) { | |
5796 | kaddr[offset] &= ~mask_to_clear; | |
5797 | len -= bits_to_clear; | |
5798 | bits_to_clear = BITS_PER_BYTE; | |
9c894696 | 5799 | mask_to_clear = ~0; |
09cbfeaf | 5800 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5801 | offset = 0; |
5802 | page = eb->pages[++i]; | |
5803 | WARN_ON(!PageUptodate(page)); | |
5804 | kaddr = page_address(page); | |
5805 | } | |
5806 | } | |
5807 | if (len) { | |
5808 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
5809 | kaddr[offset] &= ~mask_to_clear; | |
5810 | } | |
5811 | } | |
5812 | ||
3387206f ST |
5813 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
5814 | { | |
5815 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
5816 | return distance < len; | |
5817 | } | |
5818 | ||
d1310b2e CM |
5819 | static void copy_pages(struct page *dst_page, struct page *src_page, |
5820 | unsigned long dst_off, unsigned long src_off, | |
5821 | unsigned long len) | |
5822 | { | |
a6591715 | 5823 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 5824 | char *src_kaddr; |
727011e0 | 5825 | int must_memmove = 0; |
d1310b2e | 5826 | |
3387206f | 5827 | if (dst_page != src_page) { |
a6591715 | 5828 | src_kaddr = page_address(src_page); |
3387206f | 5829 | } else { |
d1310b2e | 5830 | src_kaddr = dst_kaddr; |
727011e0 CM |
5831 | if (areas_overlap(src_off, dst_off, len)) |
5832 | must_memmove = 1; | |
3387206f | 5833 | } |
d1310b2e | 5834 | |
727011e0 CM |
5835 | if (must_memmove) |
5836 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
5837 | else | |
5838 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
5839 | } |
5840 | ||
5841 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5842 | unsigned long src_offset, unsigned long len) | |
5843 | { | |
0b246afa | 5844 | struct btrfs_fs_info *fs_info = dst->fs_info; |
d1310b2e CM |
5845 | size_t cur; |
5846 | size_t dst_off_in_page; | |
5847 | size_t src_off_in_page; | |
09cbfeaf | 5848 | size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1); |
d1310b2e CM |
5849 | unsigned long dst_i; |
5850 | unsigned long src_i; | |
5851 | ||
5852 | if (src_offset + len > dst->len) { | |
0b246afa | 5853 | btrfs_err(fs_info, |
5d163e0e JM |
5854 | "memmove bogus src_offset %lu move len %lu dst len %lu", |
5855 | src_offset, len, dst->len); | |
d1310b2e CM |
5856 | BUG_ON(1); |
5857 | } | |
5858 | if (dst_offset + len > dst->len) { | |
0b246afa | 5859 | btrfs_err(fs_info, |
5d163e0e JM |
5860 | "memmove bogus dst_offset %lu move len %lu dst len %lu", |
5861 | dst_offset, len, dst->len); | |
d1310b2e CM |
5862 | BUG_ON(1); |
5863 | } | |
5864 | ||
d397712b | 5865 | while (len > 0) { |
d1310b2e | 5866 | dst_off_in_page = (start_offset + dst_offset) & |
09cbfeaf | 5867 | (PAGE_SIZE - 1); |
d1310b2e | 5868 | src_off_in_page = (start_offset + src_offset) & |
09cbfeaf | 5869 | (PAGE_SIZE - 1); |
d1310b2e | 5870 | |
09cbfeaf KS |
5871 | dst_i = (start_offset + dst_offset) >> PAGE_SHIFT; |
5872 | src_i = (start_offset + src_offset) >> PAGE_SHIFT; | |
d1310b2e | 5873 | |
09cbfeaf | 5874 | cur = min(len, (unsigned long)(PAGE_SIZE - |
d1310b2e CM |
5875 | src_off_in_page)); |
5876 | cur = min_t(unsigned long, cur, | |
09cbfeaf | 5877 | (unsigned long)(PAGE_SIZE - dst_off_in_page)); |
d1310b2e | 5878 | |
fb85fc9a | 5879 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5880 | dst_off_in_page, src_off_in_page, cur); |
5881 | ||
5882 | src_offset += cur; | |
5883 | dst_offset += cur; | |
5884 | len -= cur; | |
5885 | } | |
5886 | } | |
d1310b2e CM |
5887 | |
5888 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5889 | unsigned long src_offset, unsigned long len) | |
5890 | { | |
0b246afa | 5891 | struct btrfs_fs_info *fs_info = dst->fs_info; |
d1310b2e CM |
5892 | size_t cur; |
5893 | size_t dst_off_in_page; | |
5894 | size_t src_off_in_page; | |
5895 | unsigned long dst_end = dst_offset + len - 1; | |
5896 | unsigned long src_end = src_offset + len - 1; | |
09cbfeaf | 5897 | size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1); |
d1310b2e CM |
5898 | unsigned long dst_i; |
5899 | unsigned long src_i; | |
5900 | ||
5901 | if (src_offset + len > dst->len) { | |
0b246afa | 5902 | btrfs_err(fs_info, |
5d163e0e JM |
5903 | "memmove bogus src_offset %lu move len %lu len %lu", |
5904 | src_offset, len, dst->len); | |
d1310b2e CM |
5905 | BUG_ON(1); |
5906 | } | |
5907 | if (dst_offset + len > dst->len) { | |
0b246afa | 5908 | btrfs_err(fs_info, |
5d163e0e JM |
5909 | "memmove bogus dst_offset %lu move len %lu len %lu", |
5910 | dst_offset, len, dst->len); | |
d1310b2e CM |
5911 | BUG_ON(1); |
5912 | } | |
727011e0 | 5913 | if (dst_offset < src_offset) { |
d1310b2e CM |
5914 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
5915 | return; | |
5916 | } | |
d397712b | 5917 | while (len > 0) { |
09cbfeaf KS |
5918 | dst_i = (start_offset + dst_end) >> PAGE_SHIFT; |
5919 | src_i = (start_offset + src_end) >> PAGE_SHIFT; | |
d1310b2e CM |
5920 | |
5921 | dst_off_in_page = (start_offset + dst_end) & | |
09cbfeaf | 5922 | (PAGE_SIZE - 1); |
d1310b2e | 5923 | src_off_in_page = (start_offset + src_end) & |
09cbfeaf | 5924 | (PAGE_SIZE - 1); |
d1310b2e CM |
5925 | |
5926 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
5927 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 5928 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5929 | dst_off_in_page - cur + 1, |
5930 | src_off_in_page - cur + 1, cur); | |
5931 | ||
5932 | dst_end -= cur; | |
5933 | src_end -= cur; | |
5934 | len -= cur; | |
5935 | } | |
5936 | } | |
6af118ce | 5937 | |
f7a52a40 | 5938 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 5939 | { |
6af118ce | 5940 | struct extent_buffer *eb; |
6af118ce | 5941 | |
3083ee2e | 5942 | /* |
01327610 | 5943 | * We need to make sure nobody is attaching this page to an eb right |
3083ee2e JB |
5944 | * now. |
5945 | */ | |
5946 | spin_lock(&page->mapping->private_lock); | |
5947 | if (!PagePrivate(page)) { | |
5948 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 5949 | return 1; |
45f49bce | 5950 | } |
6af118ce | 5951 | |
3083ee2e JB |
5952 | eb = (struct extent_buffer *)page->private; |
5953 | BUG_ON(!eb); | |
19fe0a8b MX |
5954 | |
5955 | /* | |
3083ee2e JB |
5956 | * This is a little awful but should be ok, we need to make sure that |
5957 | * the eb doesn't disappear out from under us while we're looking at | |
5958 | * this page. | |
19fe0a8b | 5959 | */ |
3083ee2e | 5960 | spin_lock(&eb->refs_lock); |
0b32f4bb | 5961 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
5962 | spin_unlock(&eb->refs_lock); |
5963 | spin_unlock(&page->mapping->private_lock); | |
5964 | return 0; | |
b9473439 | 5965 | } |
3083ee2e | 5966 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 5967 | |
19fe0a8b | 5968 | /* |
3083ee2e JB |
5969 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
5970 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 5971 | */ |
3083ee2e JB |
5972 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
5973 | spin_unlock(&eb->refs_lock); | |
5974 | return 0; | |
b9473439 | 5975 | } |
19fe0a8b | 5976 | |
f7a52a40 | 5977 | return release_extent_buffer(eb); |
6af118ce | 5978 | } |