]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
c1d7c514 | 2 | |
d1310b2e CM |
3 | #include <linux/bitops.h> |
4 | #include <linux/slab.h> | |
5 | #include <linux/bio.h> | |
6 | #include <linux/mm.h> | |
d1310b2e CM |
7 | #include <linux/pagemap.h> |
8 | #include <linux/page-flags.h> | |
d1310b2e CM |
9 | #include <linux/spinlock.h> |
10 | #include <linux/blkdev.h> | |
11 | #include <linux/swap.h> | |
d1310b2e CM |
12 | #include <linux/writeback.h> |
13 | #include <linux/pagevec.h> | |
268bb0ce | 14 | #include <linux/prefetch.h> |
90a887c9 | 15 | #include <linux/cleancache.h> |
d1310b2e CM |
16 | #include "extent_io.h" |
17 | #include "extent_map.h" | |
902b22f3 DW |
18 | #include "ctree.h" |
19 | #include "btrfs_inode.h" | |
4a54c8c1 | 20 | #include "volumes.h" |
21adbd5c | 21 | #include "check-integrity.h" |
0b32f4bb | 22 | #include "locking.h" |
606686ee | 23 | #include "rcu-string.h" |
fe09e16c | 24 | #include "backref.h" |
6af49dbd | 25 | #include "disk-io.h" |
d1310b2e | 26 | |
d1310b2e CM |
27 | static struct kmem_cache *extent_state_cache; |
28 | static struct kmem_cache *extent_buffer_cache; | |
8ac9f7c1 | 29 | static struct bio_set btrfs_bioset; |
d1310b2e | 30 | |
27a3507d FM |
31 | static inline bool extent_state_in_tree(const struct extent_state *state) |
32 | { | |
33 | return !RB_EMPTY_NODE(&state->rb_node); | |
34 | } | |
35 | ||
6d49ba1b | 36 | #ifdef CONFIG_BTRFS_DEBUG |
d1310b2e CM |
37 | static LIST_HEAD(buffers); |
38 | static LIST_HEAD(states); | |
4bef0848 | 39 | |
d397712b | 40 | static DEFINE_SPINLOCK(leak_lock); |
6d49ba1b ES |
41 | |
42 | static inline | |
43 | void btrfs_leak_debug_add(struct list_head *new, struct list_head *head) | |
44 | { | |
45 | unsigned long flags; | |
46 | ||
47 | spin_lock_irqsave(&leak_lock, flags); | |
48 | list_add(new, head); | |
49 | spin_unlock_irqrestore(&leak_lock, flags); | |
50 | } | |
51 | ||
52 | static inline | |
53 | void btrfs_leak_debug_del(struct list_head *entry) | |
54 | { | |
55 | unsigned long flags; | |
56 | ||
57 | spin_lock_irqsave(&leak_lock, flags); | |
58 | list_del(entry); | |
59 | spin_unlock_irqrestore(&leak_lock, flags); | |
60 | } | |
61 | ||
62 | static inline | |
63 | void btrfs_leak_debug_check(void) | |
64 | { | |
65 | struct extent_state *state; | |
66 | struct extent_buffer *eb; | |
67 | ||
68 | while (!list_empty(&states)) { | |
69 | state = list_entry(states.next, struct extent_state, leak_list); | |
9ee49a04 | 70 | pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n", |
27a3507d FM |
71 | state->start, state->end, state->state, |
72 | extent_state_in_tree(state), | |
b7ac31b7 | 73 | refcount_read(&state->refs)); |
6d49ba1b ES |
74 | list_del(&state->leak_list); |
75 | kmem_cache_free(extent_state_cache, state); | |
76 | } | |
77 | ||
78 | while (!list_empty(&buffers)) { | |
79 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
af2679e4 LB |
80 | pr_err("BTRFS: buffer leak start %llu len %lu refs %d bflags %lu\n", |
81 | eb->start, eb->len, atomic_read(&eb->refs), eb->bflags); | |
6d49ba1b ES |
82 | list_del(&eb->leak_list); |
83 | kmem_cache_free(extent_buffer_cache, eb); | |
84 | } | |
85 | } | |
8d599ae1 | 86 | |
a5dee37d JB |
87 | #define btrfs_debug_check_extent_io_range(tree, start, end) \ |
88 | __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end)) | |
8d599ae1 | 89 | static inline void __btrfs_debug_check_extent_io_range(const char *caller, |
a5dee37d | 90 | struct extent_io_tree *tree, u64 start, u64 end) |
8d599ae1 | 91 | { |
c6100a4b JB |
92 | if (tree->ops && tree->ops->check_extent_io_range) |
93 | tree->ops->check_extent_io_range(tree->private_data, caller, | |
94 | start, end); | |
8d599ae1 | 95 | } |
6d49ba1b ES |
96 | #else |
97 | #define btrfs_leak_debug_add(new, head) do {} while (0) | |
98 | #define btrfs_leak_debug_del(entry) do {} while (0) | |
99 | #define btrfs_leak_debug_check() do {} while (0) | |
8d599ae1 | 100 | #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0) |
4bef0848 | 101 | #endif |
d1310b2e | 102 | |
d1310b2e CM |
103 | #define BUFFER_LRU_MAX 64 |
104 | ||
105 | struct tree_entry { | |
106 | u64 start; | |
107 | u64 end; | |
d1310b2e CM |
108 | struct rb_node rb_node; |
109 | }; | |
110 | ||
111 | struct extent_page_data { | |
112 | struct bio *bio; | |
113 | struct extent_io_tree *tree; | |
771ed689 CM |
114 | /* tells writepage not to lock the state bits for this range |
115 | * it still does the unlocking | |
116 | */ | |
ffbd517d CM |
117 | unsigned int extent_locked:1; |
118 | ||
70fd7614 | 119 | /* tells the submit_bio code to use REQ_SYNC */ |
ffbd517d | 120 | unsigned int sync_io:1; |
d1310b2e CM |
121 | }; |
122 | ||
57599c7e | 123 | static int add_extent_changeset(struct extent_state *state, unsigned bits, |
d38ed27f QW |
124 | struct extent_changeset *changeset, |
125 | int set) | |
126 | { | |
127 | int ret; | |
128 | ||
129 | if (!changeset) | |
57599c7e | 130 | return 0; |
d38ed27f | 131 | if (set && (state->state & bits) == bits) |
57599c7e | 132 | return 0; |
fefdc557 | 133 | if (!set && (state->state & bits) == 0) |
57599c7e | 134 | return 0; |
d38ed27f | 135 | changeset->bytes_changed += state->end - state->start + 1; |
53d32359 | 136 | ret = ulist_add(&changeset->range_changed, state->start, state->end, |
d38ed27f | 137 | GFP_ATOMIC); |
57599c7e | 138 | return ret; |
d38ed27f QW |
139 | } |
140 | ||
aab6e9ed | 141 | static void flush_write_bio(struct extent_page_data *epd); |
e2932ee0 | 142 | |
c2d904e0 JM |
143 | static inline struct btrfs_fs_info * |
144 | tree_fs_info(struct extent_io_tree *tree) | |
145 | { | |
c6100a4b JB |
146 | if (tree->ops) |
147 | return tree->ops->tree_fs_info(tree->private_data); | |
148 | return NULL; | |
c2d904e0 | 149 | } |
0b32f4bb | 150 | |
d1310b2e CM |
151 | int __init extent_io_init(void) |
152 | { | |
837e1972 | 153 | extent_state_cache = kmem_cache_create("btrfs_extent_state", |
9601e3f6 | 154 | sizeof(struct extent_state), 0, |
fba4b697 | 155 | SLAB_MEM_SPREAD, NULL); |
d1310b2e CM |
156 | if (!extent_state_cache) |
157 | return -ENOMEM; | |
158 | ||
837e1972 | 159 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 | 160 | sizeof(struct extent_buffer), 0, |
fba4b697 | 161 | SLAB_MEM_SPREAD, NULL); |
d1310b2e CM |
162 | if (!extent_buffer_cache) |
163 | goto free_state_cache; | |
9be3395b | 164 | |
8ac9f7c1 KO |
165 | if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE, |
166 | offsetof(struct btrfs_io_bio, bio), | |
167 | BIOSET_NEED_BVECS)) | |
9be3395b | 168 | goto free_buffer_cache; |
b208c2f7 | 169 | |
8ac9f7c1 | 170 | if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE)) |
b208c2f7 DW |
171 | goto free_bioset; |
172 | ||
d1310b2e CM |
173 | return 0; |
174 | ||
b208c2f7 | 175 | free_bioset: |
8ac9f7c1 | 176 | bioset_exit(&btrfs_bioset); |
b208c2f7 | 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 | ||
e67c718b | 188 | void __cold extent_io_exit(void) |
d1310b2e | 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); | |
8ac9f7c1 | 199 | bioset_exit(&btrfs_bioset); |
d1310b2e CM |
200 | } |
201 | ||
202 | void extent_io_tree_init(struct extent_io_tree *tree, | |
c6100a4b | 203 | void *private_data) |
d1310b2e | 204 | { |
6bef4d31 | 205 | tree->state = RB_ROOT; |
d1310b2e CM |
206 | tree->ops = NULL; |
207 | tree->dirty_bytes = 0; | |
70dec807 | 208 | spin_lock_init(&tree->lock); |
c6100a4b | 209 | tree->private_data = private_data; |
d1310b2e | 210 | } |
d1310b2e | 211 | |
b2950863 | 212 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
213 | { |
214 | struct extent_state *state; | |
d1310b2e | 215 | |
3ba7ab22 MH |
216 | /* |
217 | * The given mask might be not appropriate for the slab allocator, | |
218 | * drop the unsupported bits | |
219 | */ | |
220 | mask &= ~(__GFP_DMA32|__GFP_HIGHMEM); | |
d1310b2e | 221 | state = kmem_cache_alloc(extent_state_cache, mask); |
2b114d1d | 222 | if (!state) |
d1310b2e CM |
223 | return state; |
224 | state->state = 0; | |
47dc196a | 225 | state->failrec = NULL; |
27a3507d | 226 | RB_CLEAR_NODE(&state->rb_node); |
6d49ba1b | 227 | btrfs_leak_debug_add(&state->leak_list, &states); |
b7ac31b7 | 228 | refcount_set(&state->refs, 1); |
d1310b2e | 229 | init_waitqueue_head(&state->wq); |
143bede5 | 230 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
231 | return state; |
232 | } | |
d1310b2e | 233 | |
4845e44f | 234 | void free_extent_state(struct extent_state *state) |
d1310b2e | 235 | { |
d1310b2e CM |
236 | if (!state) |
237 | return; | |
b7ac31b7 | 238 | if (refcount_dec_and_test(&state->refs)) { |
27a3507d | 239 | WARN_ON(extent_state_in_tree(state)); |
6d49ba1b | 240 | btrfs_leak_debug_del(&state->leak_list); |
143bede5 | 241 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
242 | kmem_cache_free(extent_state_cache, state); |
243 | } | |
244 | } | |
d1310b2e | 245 | |
f2071b21 FM |
246 | static struct rb_node *tree_insert(struct rb_root *root, |
247 | struct rb_node *search_start, | |
248 | u64 offset, | |
12cfbad9 FDBM |
249 | struct rb_node *node, |
250 | struct rb_node ***p_in, | |
251 | struct rb_node **parent_in) | |
d1310b2e | 252 | { |
f2071b21 | 253 | struct rb_node **p; |
d397712b | 254 | struct rb_node *parent = NULL; |
d1310b2e CM |
255 | struct tree_entry *entry; |
256 | ||
12cfbad9 FDBM |
257 | if (p_in && parent_in) { |
258 | p = *p_in; | |
259 | parent = *parent_in; | |
260 | goto do_insert; | |
261 | } | |
262 | ||
f2071b21 | 263 | p = search_start ? &search_start : &root->rb_node; |
d397712b | 264 | while (*p) { |
d1310b2e CM |
265 | parent = *p; |
266 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
267 | ||
268 | if (offset < entry->start) | |
269 | p = &(*p)->rb_left; | |
270 | else if (offset > entry->end) | |
271 | p = &(*p)->rb_right; | |
272 | else | |
273 | return parent; | |
274 | } | |
275 | ||
12cfbad9 | 276 | do_insert: |
d1310b2e CM |
277 | rb_link_node(node, parent, p); |
278 | rb_insert_color(node, root); | |
279 | return NULL; | |
280 | } | |
281 | ||
80ea96b1 | 282 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
12cfbad9 FDBM |
283 | struct rb_node **prev_ret, |
284 | struct rb_node **next_ret, | |
285 | struct rb_node ***p_ret, | |
286 | struct rb_node **parent_ret) | |
d1310b2e | 287 | { |
80ea96b1 | 288 | struct rb_root *root = &tree->state; |
12cfbad9 | 289 | struct rb_node **n = &root->rb_node; |
d1310b2e CM |
290 | struct rb_node *prev = NULL; |
291 | struct rb_node *orig_prev = NULL; | |
292 | struct tree_entry *entry; | |
293 | struct tree_entry *prev_entry = NULL; | |
294 | ||
12cfbad9 FDBM |
295 | while (*n) { |
296 | prev = *n; | |
297 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
d1310b2e CM |
298 | prev_entry = entry; |
299 | ||
300 | if (offset < entry->start) | |
12cfbad9 | 301 | n = &(*n)->rb_left; |
d1310b2e | 302 | else if (offset > entry->end) |
12cfbad9 | 303 | n = &(*n)->rb_right; |
d397712b | 304 | else |
12cfbad9 | 305 | return *n; |
d1310b2e CM |
306 | } |
307 | ||
12cfbad9 FDBM |
308 | if (p_ret) |
309 | *p_ret = n; | |
310 | if (parent_ret) | |
311 | *parent_ret = prev; | |
312 | ||
d1310b2e CM |
313 | if (prev_ret) { |
314 | orig_prev = prev; | |
d397712b | 315 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
316 | prev = rb_next(prev); |
317 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
318 | } | |
319 | *prev_ret = prev; | |
320 | prev = orig_prev; | |
321 | } | |
322 | ||
323 | if (next_ret) { | |
324 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 325 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
326 | prev = rb_prev(prev); |
327 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
328 | } | |
329 | *next_ret = prev; | |
330 | } | |
331 | return NULL; | |
332 | } | |
333 | ||
12cfbad9 FDBM |
334 | static inline struct rb_node * |
335 | tree_search_for_insert(struct extent_io_tree *tree, | |
336 | u64 offset, | |
337 | struct rb_node ***p_ret, | |
338 | struct rb_node **parent_ret) | |
d1310b2e | 339 | { |
70dec807 | 340 | struct rb_node *prev = NULL; |
d1310b2e | 341 | struct rb_node *ret; |
70dec807 | 342 | |
12cfbad9 | 343 | ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret); |
d397712b | 344 | if (!ret) |
d1310b2e CM |
345 | return prev; |
346 | return ret; | |
347 | } | |
348 | ||
12cfbad9 FDBM |
349 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
350 | u64 offset) | |
351 | { | |
352 | return tree_search_for_insert(tree, offset, NULL, NULL); | |
353 | } | |
354 | ||
9ed74f2d JB |
355 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, |
356 | struct extent_state *other) | |
357 | { | |
358 | if (tree->ops && tree->ops->merge_extent_hook) | |
c6100a4b | 359 | tree->ops->merge_extent_hook(tree->private_data, new, other); |
9ed74f2d JB |
360 | } |
361 | ||
d1310b2e CM |
362 | /* |
363 | * utility function to look for merge candidates inside a given range. | |
364 | * Any extents with matching state are merged together into a single | |
365 | * extent in the tree. Extents with EXTENT_IO in their state field | |
366 | * are not merged because the end_io handlers need to be able to do | |
367 | * operations on them without sleeping (or doing allocations/splits). | |
368 | * | |
369 | * This should be called with the tree lock held. | |
370 | */ | |
1bf85046 JM |
371 | static void merge_state(struct extent_io_tree *tree, |
372 | struct extent_state *state) | |
d1310b2e CM |
373 | { |
374 | struct extent_state *other; | |
375 | struct rb_node *other_node; | |
376 | ||
5b21f2ed | 377 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
1bf85046 | 378 | return; |
d1310b2e CM |
379 | |
380 | other_node = rb_prev(&state->rb_node); | |
381 | if (other_node) { | |
382 | other = rb_entry(other_node, struct extent_state, rb_node); | |
383 | if (other->end == state->start - 1 && | |
384 | other->state == state->state) { | |
9ed74f2d | 385 | merge_cb(tree, state, other); |
d1310b2e | 386 | state->start = other->start; |
d1310b2e | 387 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 388 | RB_CLEAR_NODE(&other->rb_node); |
d1310b2e CM |
389 | free_extent_state(other); |
390 | } | |
391 | } | |
392 | other_node = rb_next(&state->rb_node); | |
393 | if (other_node) { | |
394 | other = rb_entry(other_node, struct extent_state, rb_node); | |
395 | if (other->start == state->end + 1 && | |
396 | other->state == state->state) { | |
9ed74f2d | 397 | merge_cb(tree, state, other); |
df98b6e2 | 398 | state->end = other->end; |
df98b6e2 | 399 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 400 | RB_CLEAR_NODE(&other->rb_node); |
df98b6e2 | 401 | free_extent_state(other); |
d1310b2e CM |
402 | } |
403 | } | |
d1310b2e CM |
404 | } |
405 | ||
1bf85046 | 406 | static void set_state_cb(struct extent_io_tree *tree, |
9ee49a04 | 407 | struct extent_state *state, unsigned *bits) |
291d673e | 408 | { |
1bf85046 | 409 | if (tree->ops && tree->ops->set_bit_hook) |
c6100a4b | 410 | tree->ops->set_bit_hook(tree->private_data, state, bits); |
291d673e CM |
411 | } |
412 | ||
413 | static void clear_state_cb(struct extent_io_tree *tree, | |
9ee49a04 | 414 | struct extent_state *state, unsigned *bits) |
291d673e | 415 | { |
9ed74f2d | 416 | if (tree->ops && tree->ops->clear_bit_hook) |
c6100a4b | 417 | tree->ops->clear_bit_hook(tree->private_data, state, bits); |
291d673e CM |
418 | } |
419 | ||
3150b699 | 420 | static void set_state_bits(struct extent_io_tree *tree, |
d38ed27f QW |
421 | struct extent_state *state, unsigned *bits, |
422 | struct extent_changeset *changeset); | |
3150b699 | 423 | |
d1310b2e CM |
424 | /* |
425 | * insert an extent_state struct into the tree. 'bits' are set on the | |
426 | * struct before it is inserted. | |
427 | * | |
428 | * This may return -EEXIST if the extent is already there, in which case the | |
429 | * state struct is freed. | |
430 | * | |
431 | * The tree lock is not taken internally. This is a utility function and | |
432 | * probably isn't what you want to call (see set/clear_extent_bit). | |
433 | */ | |
434 | static int insert_state(struct extent_io_tree *tree, | |
435 | struct extent_state *state, u64 start, u64 end, | |
12cfbad9 FDBM |
436 | struct rb_node ***p, |
437 | struct rb_node **parent, | |
d38ed27f | 438 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e CM |
439 | { |
440 | struct rb_node *node; | |
441 | ||
31b1a2bd | 442 | if (end < start) |
efe120a0 | 443 | WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n", |
c1c9ff7c | 444 | end, start); |
d1310b2e CM |
445 | state->start = start; |
446 | state->end = end; | |
9ed74f2d | 447 | |
d38ed27f | 448 | set_state_bits(tree, state, bits, changeset); |
3150b699 | 449 | |
f2071b21 | 450 | node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent); |
d1310b2e CM |
451 | if (node) { |
452 | struct extent_state *found; | |
453 | found = rb_entry(node, struct extent_state, rb_node); | |
62e85577 | 454 | pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n", |
c1c9ff7c | 455 | found->start, found->end, start, end); |
d1310b2e CM |
456 | return -EEXIST; |
457 | } | |
458 | merge_state(tree, state); | |
459 | return 0; | |
460 | } | |
461 | ||
1bf85046 | 462 | static void split_cb(struct extent_io_tree *tree, struct extent_state *orig, |
9ed74f2d JB |
463 | u64 split) |
464 | { | |
465 | if (tree->ops && tree->ops->split_extent_hook) | |
c6100a4b | 466 | tree->ops->split_extent_hook(tree->private_data, orig, split); |
9ed74f2d JB |
467 | } |
468 | ||
d1310b2e CM |
469 | /* |
470 | * split a given extent state struct in two, inserting the preallocated | |
471 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
472 | * offset inside 'orig' where it should be split. | |
473 | * | |
474 | * Before calling, | |
475 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
476 | * are two extent state structs in the tree: | |
477 | * prealloc: [orig->start, split - 1] | |
478 | * orig: [ split, orig->end ] | |
479 | * | |
480 | * The tree locks are not taken by this function. They need to be held | |
481 | * by the caller. | |
482 | */ | |
483 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
484 | struct extent_state *prealloc, u64 split) | |
485 | { | |
486 | struct rb_node *node; | |
9ed74f2d JB |
487 | |
488 | split_cb(tree, orig, split); | |
489 | ||
d1310b2e CM |
490 | prealloc->start = orig->start; |
491 | prealloc->end = split - 1; | |
492 | prealloc->state = orig->state; | |
493 | orig->start = split; | |
494 | ||
f2071b21 FM |
495 | node = tree_insert(&tree->state, &orig->rb_node, prealloc->end, |
496 | &prealloc->rb_node, NULL, NULL); | |
d1310b2e | 497 | if (node) { |
d1310b2e CM |
498 | free_extent_state(prealloc); |
499 | return -EEXIST; | |
500 | } | |
501 | return 0; | |
502 | } | |
503 | ||
cdc6a395 LZ |
504 | static struct extent_state *next_state(struct extent_state *state) |
505 | { | |
506 | struct rb_node *next = rb_next(&state->rb_node); | |
507 | if (next) | |
508 | return rb_entry(next, struct extent_state, rb_node); | |
509 | else | |
510 | return NULL; | |
511 | } | |
512 | ||
d1310b2e CM |
513 | /* |
514 | * utility function to clear some bits in an extent state struct. | |
1b303fc0 | 515 | * it will optionally wake up any one waiting on this state (wake == 1). |
d1310b2e CM |
516 | * |
517 | * If no bits are set on the state struct after clearing things, the | |
518 | * struct is freed and removed from the tree | |
519 | */ | |
cdc6a395 LZ |
520 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
521 | struct extent_state *state, | |
fefdc557 QW |
522 | unsigned *bits, int wake, |
523 | struct extent_changeset *changeset) | |
d1310b2e | 524 | { |
cdc6a395 | 525 | struct extent_state *next; |
9ee49a04 | 526 | unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS; |
57599c7e | 527 | int ret; |
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); |
57599c7e DS |
535 | ret = add_extent_changeset(state, bits_to_clear, changeset, 0); |
536 | BUG_ON(ret < 0); | |
32c00aff | 537 | state->state &= ~bits_to_clear; |
d1310b2e CM |
538 | if (wake) |
539 | wake_up(&state->wq); | |
0ca1f7ce | 540 | if (state->state == 0) { |
cdc6a395 | 541 | next = next_state(state); |
27a3507d | 542 | if (extent_state_in_tree(state)) { |
d1310b2e | 543 | rb_erase(&state->rb_node, &tree->state); |
27a3507d | 544 | RB_CLEAR_NODE(&state->rb_node); |
d1310b2e CM |
545 | free_extent_state(state); |
546 | } else { | |
547 | WARN_ON(1); | |
548 | } | |
549 | } else { | |
550 | merge_state(tree, state); | |
cdc6a395 | 551 | next = next_state(state); |
d1310b2e | 552 | } |
cdc6a395 | 553 | return next; |
d1310b2e CM |
554 | } |
555 | ||
8233767a XG |
556 | static struct extent_state * |
557 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
558 | { | |
559 | if (!prealloc) | |
560 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
561 | ||
562 | return prealloc; | |
563 | } | |
564 | ||
48a3b636 | 565 | static void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
c2d904e0 | 566 | { |
5d163e0e JM |
567 | btrfs_panic(tree_fs_info(tree), err, |
568 | "Locking error: Extent tree was modified by another thread while locked."); | |
c2d904e0 JM |
569 | } |
570 | ||
d1310b2e CM |
571 | /* |
572 | * clear some bits on a range in the tree. This may require splitting | |
573 | * or inserting elements in the tree, so the gfp mask is used to | |
574 | * indicate which allocations or sleeping are allowed. | |
575 | * | |
576 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
577 | * the given range from the tree regardless of state (ie for truncate). | |
578 | * | |
579 | * the range [start, end] is inclusive. | |
580 | * | |
6763af84 | 581 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e | 582 | */ |
66b0c887 | 583 | int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
fefdc557 QW |
584 | unsigned bits, int wake, int delete, |
585 | struct extent_state **cached_state, | |
586 | gfp_t mask, struct extent_changeset *changeset) | |
d1310b2e CM |
587 | { |
588 | struct extent_state *state; | |
2c64c53d | 589 | struct extent_state *cached; |
d1310b2e CM |
590 | struct extent_state *prealloc = NULL; |
591 | struct rb_node *node; | |
5c939df5 | 592 | u64 last_end; |
d1310b2e | 593 | int err; |
2ac55d41 | 594 | int clear = 0; |
d1310b2e | 595 | |
a5dee37d | 596 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 597 | |
7ee9e440 JB |
598 | if (bits & EXTENT_DELALLOC) |
599 | bits |= EXTENT_NORESERVE; | |
600 | ||
0ca1f7ce YZ |
601 | if (delete) |
602 | bits |= ~EXTENT_CTLBITS; | |
603 | bits |= EXTENT_FIRST_DELALLOC; | |
604 | ||
2ac55d41 JB |
605 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
606 | clear = 1; | |
d1310b2e | 607 | again: |
d0164adc | 608 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
c7bc6319 FM |
609 | /* |
610 | * Don't care for allocation failure here because we might end | |
611 | * up not needing the pre-allocated extent state at all, which | |
612 | * is the case if we only have in the tree extent states that | |
613 | * cover our input range and don't cover too any other range. | |
614 | * If we end up needing a new extent state we allocate it later. | |
615 | */ | |
d1310b2e | 616 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
617 | } |
618 | ||
cad321ad | 619 | spin_lock(&tree->lock); |
2c64c53d CM |
620 | if (cached_state) { |
621 | cached = *cached_state; | |
2ac55d41 JB |
622 | |
623 | if (clear) { | |
624 | *cached_state = NULL; | |
625 | cached_state = NULL; | |
626 | } | |
627 | ||
27a3507d FM |
628 | if (cached && extent_state_in_tree(cached) && |
629 | cached->start <= start && cached->end > start) { | |
2ac55d41 | 630 | if (clear) |
b7ac31b7 | 631 | refcount_dec(&cached->refs); |
2c64c53d | 632 | state = cached; |
42daec29 | 633 | goto hit_next; |
2c64c53d | 634 | } |
2ac55d41 JB |
635 | if (clear) |
636 | free_extent_state(cached); | |
2c64c53d | 637 | } |
d1310b2e CM |
638 | /* |
639 | * this search will find the extents that end after | |
640 | * our range starts | |
641 | */ | |
80ea96b1 | 642 | node = tree_search(tree, start); |
d1310b2e CM |
643 | if (!node) |
644 | goto out; | |
645 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 646 | hit_next: |
d1310b2e CM |
647 | if (state->start > end) |
648 | goto out; | |
649 | WARN_ON(state->end < start); | |
5c939df5 | 650 | last_end = state->end; |
d1310b2e | 651 | |
0449314a | 652 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
653 | if (!(state->state & bits)) { |
654 | state = next_state(state); | |
0449314a | 655 | goto next; |
cdc6a395 | 656 | } |
0449314a | 657 | |
d1310b2e CM |
658 | /* |
659 | * | ---- desired range ---- | | |
660 | * | state | or | |
661 | * | ------------- state -------------- | | |
662 | * | |
663 | * We need to split the extent we found, and may flip | |
664 | * bits on second half. | |
665 | * | |
666 | * If the extent we found extends past our range, we | |
667 | * just split and search again. It'll get split again | |
668 | * the next time though. | |
669 | * | |
670 | * If the extent we found is inside our range, we clear | |
671 | * the desired bit on it. | |
672 | */ | |
673 | ||
674 | if (state->start < start) { | |
8233767a XG |
675 | prealloc = alloc_extent_state_atomic(prealloc); |
676 | BUG_ON(!prealloc); | |
d1310b2e | 677 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
678 | if (err) |
679 | extent_io_tree_panic(tree, err); | |
680 | ||
d1310b2e CM |
681 | prealloc = NULL; |
682 | if (err) | |
683 | goto out; | |
684 | if (state->end <= end) { | |
fefdc557 QW |
685 | state = clear_state_bit(tree, state, &bits, wake, |
686 | changeset); | |
d1ac6e41 | 687 | goto next; |
d1310b2e CM |
688 | } |
689 | goto search_again; | |
690 | } | |
691 | /* | |
692 | * | ---- desired range ---- | | |
693 | * | state | | |
694 | * We need to split the extent, and clear the bit | |
695 | * on the first half | |
696 | */ | |
697 | if (state->start <= end && state->end > end) { | |
8233767a XG |
698 | prealloc = alloc_extent_state_atomic(prealloc); |
699 | BUG_ON(!prealloc); | |
d1310b2e | 700 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
701 | if (err) |
702 | extent_io_tree_panic(tree, err); | |
703 | ||
d1310b2e CM |
704 | if (wake) |
705 | wake_up(&state->wq); | |
42daec29 | 706 | |
fefdc557 | 707 | clear_state_bit(tree, prealloc, &bits, wake, changeset); |
9ed74f2d | 708 | |
d1310b2e CM |
709 | prealloc = NULL; |
710 | goto out; | |
711 | } | |
42daec29 | 712 | |
fefdc557 | 713 | state = clear_state_bit(tree, state, &bits, wake, changeset); |
0449314a | 714 | next: |
5c939df5 YZ |
715 | if (last_end == (u64)-1) |
716 | goto out; | |
717 | start = last_end + 1; | |
cdc6a395 | 718 | if (start <= end && state && !need_resched()) |
692e5759 | 719 | goto hit_next; |
d1310b2e CM |
720 | |
721 | search_again: | |
722 | if (start > end) | |
723 | goto out; | |
cad321ad | 724 | spin_unlock(&tree->lock); |
d0164adc | 725 | if (gfpflags_allow_blocking(mask)) |
d1310b2e CM |
726 | cond_resched(); |
727 | goto again; | |
7ab5cb2a DS |
728 | |
729 | out: | |
730 | spin_unlock(&tree->lock); | |
731 | if (prealloc) | |
732 | free_extent_state(prealloc); | |
733 | ||
734 | return 0; | |
735 | ||
d1310b2e | 736 | } |
d1310b2e | 737 | |
143bede5 JM |
738 | static void wait_on_state(struct extent_io_tree *tree, |
739 | struct extent_state *state) | |
641f5219 CH |
740 | __releases(tree->lock) |
741 | __acquires(tree->lock) | |
d1310b2e CM |
742 | { |
743 | DEFINE_WAIT(wait); | |
744 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 745 | spin_unlock(&tree->lock); |
d1310b2e | 746 | schedule(); |
cad321ad | 747 | spin_lock(&tree->lock); |
d1310b2e | 748 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
749 | } |
750 | ||
751 | /* | |
752 | * waits for one or more bits to clear on a range in the state tree. | |
753 | * The range [start, end] is inclusive. | |
754 | * The tree lock is taken by this function | |
755 | */ | |
41074888 DS |
756 | static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
757 | unsigned long bits) | |
d1310b2e CM |
758 | { |
759 | struct extent_state *state; | |
760 | struct rb_node *node; | |
761 | ||
a5dee37d | 762 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 763 | |
cad321ad | 764 | spin_lock(&tree->lock); |
d1310b2e CM |
765 | again: |
766 | while (1) { | |
767 | /* | |
768 | * this search will find all the extents that end after | |
769 | * our range starts | |
770 | */ | |
80ea96b1 | 771 | node = tree_search(tree, start); |
c50d3e71 | 772 | process_node: |
d1310b2e CM |
773 | if (!node) |
774 | break; | |
775 | ||
776 | state = rb_entry(node, struct extent_state, rb_node); | |
777 | ||
778 | if (state->start > end) | |
779 | goto out; | |
780 | ||
781 | if (state->state & bits) { | |
782 | start = state->start; | |
b7ac31b7 | 783 | refcount_inc(&state->refs); |
d1310b2e CM |
784 | wait_on_state(tree, state); |
785 | free_extent_state(state); | |
786 | goto again; | |
787 | } | |
788 | start = state->end + 1; | |
789 | ||
790 | if (start > end) | |
791 | break; | |
792 | ||
c50d3e71 FM |
793 | if (!cond_resched_lock(&tree->lock)) { |
794 | node = rb_next(node); | |
795 | goto process_node; | |
796 | } | |
d1310b2e CM |
797 | } |
798 | out: | |
cad321ad | 799 | spin_unlock(&tree->lock); |
d1310b2e | 800 | } |
d1310b2e | 801 | |
1bf85046 | 802 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 803 | struct extent_state *state, |
d38ed27f | 804 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e | 805 | { |
9ee49a04 | 806 | unsigned bits_to_set = *bits & ~EXTENT_CTLBITS; |
57599c7e | 807 | int ret; |
9ed74f2d | 808 | |
1bf85046 | 809 | set_state_cb(tree, state, bits); |
0ca1f7ce | 810 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
811 | u64 range = state->end - state->start + 1; |
812 | tree->dirty_bytes += range; | |
813 | } | |
57599c7e DS |
814 | ret = add_extent_changeset(state, bits_to_set, changeset, 1); |
815 | BUG_ON(ret < 0); | |
0ca1f7ce | 816 | state->state |= bits_to_set; |
d1310b2e CM |
817 | } |
818 | ||
e38e2ed7 FM |
819 | static void cache_state_if_flags(struct extent_state *state, |
820 | struct extent_state **cached_ptr, | |
9ee49a04 | 821 | unsigned flags) |
2c64c53d CM |
822 | { |
823 | if (cached_ptr && !(*cached_ptr)) { | |
e38e2ed7 | 824 | if (!flags || (state->state & flags)) { |
2c64c53d | 825 | *cached_ptr = state; |
b7ac31b7 | 826 | refcount_inc(&state->refs); |
2c64c53d CM |
827 | } |
828 | } | |
829 | } | |
830 | ||
e38e2ed7 FM |
831 | static void cache_state(struct extent_state *state, |
832 | struct extent_state **cached_ptr) | |
833 | { | |
834 | return cache_state_if_flags(state, cached_ptr, | |
835 | EXTENT_IOBITS | EXTENT_BOUNDARY); | |
836 | } | |
837 | ||
d1310b2e | 838 | /* |
1edbb734 CM |
839 | * set some bits on a range in the tree. This may require allocations or |
840 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 841 | * |
1edbb734 CM |
842 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
843 | * part of the range already has the desired bits set. The start of the | |
844 | * existing range is returned in failed_start in this case. | |
d1310b2e | 845 | * |
1edbb734 | 846 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 847 | */ |
1edbb734 | 848 | |
3fbe5c02 JM |
849 | static int __must_check |
850 | __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 851 | unsigned bits, unsigned exclusive_bits, |
41074888 | 852 | u64 *failed_start, struct extent_state **cached_state, |
d38ed27f | 853 | gfp_t mask, struct extent_changeset *changeset) |
d1310b2e CM |
854 | { |
855 | struct extent_state *state; | |
856 | struct extent_state *prealloc = NULL; | |
857 | struct rb_node *node; | |
12cfbad9 FDBM |
858 | struct rb_node **p; |
859 | struct rb_node *parent; | |
d1310b2e | 860 | int err = 0; |
d1310b2e CM |
861 | u64 last_start; |
862 | u64 last_end; | |
42daec29 | 863 | |
a5dee37d | 864 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 865 | |
0ca1f7ce | 866 | bits |= EXTENT_FIRST_DELALLOC; |
d1310b2e | 867 | again: |
d0164adc | 868 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
059f791c DS |
869 | /* |
870 | * Don't care for allocation failure here because we might end | |
871 | * up not needing the pre-allocated extent state at all, which | |
872 | * is the case if we only have in the tree extent states that | |
873 | * cover our input range and don't cover too any other range. | |
874 | * If we end up needing a new extent state we allocate it later. | |
875 | */ | |
d1310b2e | 876 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
877 | } |
878 | ||
cad321ad | 879 | spin_lock(&tree->lock); |
9655d298 CM |
880 | if (cached_state && *cached_state) { |
881 | state = *cached_state; | |
df98b6e2 | 882 | if (state->start <= start && state->end > start && |
27a3507d | 883 | extent_state_in_tree(state)) { |
9655d298 CM |
884 | node = &state->rb_node; |
885 | goto hit_next; | |
886 | } | |
887 | } | |
d1310b2e CM |
888 | /* |
889 | * this search will find all the extents that end after | |
890 | * our range starts. | |
891 | */ | |
12cfbad9 | 892 | node = tree_search_for_insert(tree, start, &p, &parent); |
d1310b2e | 893 | if (!node) { |
8233767a XG |
894 | prealloc = alloc_extent_state_atomic(prealloc); |
895 | BUG_ON(!prealloc); | |
12cfbad9 | 896 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 897 | &p, &parent, &bits, changeset); |
c2d904e0 JM |
898 | if (err) |
899 | extent_io_tree_panic(tree, err); | |
900 | ||
c42ac0bc | 901 | cache_state(prealloc, cached_state); |
d1310b2e | 902 | prealloc = NULL; |
d1310b2e CM |
903 | goto out; |
904 | } | |
d1310b2e | 905 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 906 | hit_next: |
d1310b2e CM |
907 | last_start = state->start; |
908 | last_end = state->end; | |
909 | ||
910 | /* | |
911 | * | ---- desired range ---- | | |
912 | * | state | | |
913 | * | |
914 | * Just lock what we found and keep going | |
915 | */ | |
916 | if (state->start == start && state->end <= end) { | |
1edbb734 | 917 | if (state->state & exclusive_bits) { |
d1310b2e CM |
918 | *failed_start = state->start; |
919 | err = -EEXIST; | |
920 | goto out; | |
921 | } | |
42daec29 | 922 | |
d38ed27f | 923 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 924 | cache_state(state, cached_state); |
d1310b2e | 925 | merge_state(tree, state); |
5c939df5 YZ |
926 | if (last_end == (u64)-1) |
927 | goto out; | |
928 | start = last_end + 1; | |
d1ac6e41 LB |
929 | state = next_state(state); |
930 | if (start < end && state && state->start == start && | |
931 | !need_resched()) | |
932 | goto hit_next; | |
d1310b2e CM |
933 | goto search_again; |
934 | } | |
935 | ||
936 | /* | |
937 | * | ---- desired range ---- | | |
938 | * | state | | |
939 | * or | |
940 | * | ------------- state -------------- | | |
941 | * | |
942 | * We need to split the extent we found, and may flip bits on | |
943 | * second half. | |
944 | * | |
945 | * If the extent we found extends past our | |
946 | * range, we just split and search again. It'll get split | |
947 | * again the next time though. | |
948 | * | |
949 | * If the extent we found is inside our range, we set the | |
950 | * desired bit on it. | |
951 | */ | |
952 | if (state->start < start) { | |
1edbb734 | 953 | if (state->state & exclusive_bits) { |
d1310b2e CM |
954 | *failed_start = start; |
955 | err = -EEXIST; | |
956 | goto out; | |
957 | } | |
8233767a XG |
958 | |
959 | prealloc = alloc_extent_state_atomic(prealloc); | |
960 | BUG_ON(!prealloc); | |
d1310b2e | 961 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
962 | if (err) |
963 | extent_io_tree_panic(tree, err); | |
964 | ||
d1310b2e CM |
965 | prealloc = NULL; |
966 | if (err) | |
967 | goto out; | |
968 | if (state->end <= end) { | |
d38ed27f | 969 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 970 | cache_state(state, cached_state); |
d1310b2e | 971 | merge_state(tree, state); |
5c939df5 YZ |
972 | if (last_end == (u64)-1) |
973 | goto out; | |
974 | start = last_end + 1; | |
d1ac6e41 LB |
975 | state = next_state(state); |
976 | if (start < end && state && state->start == start && | |
977 | !need_resched()) | |
978 | goto hit_next; | |
d1310b2e CM |
979 | } |
980 | goto search_again; | |
981 | } | |
982 | /* | |
983 | * | ---- desired range ---- | | |
984 | * | state | or | state | | |
985 | * | |
986 | * There's a hole, we need to insert something in it and | |
987 | * ignore the extent we found. | |
988 | */ | |
989 | if (state->start > start) { | |
990 | u64 this_end; | |
991 | if (end < last_start) | |
992 | this_end = end; | |
993 | else | |
d397712b | 994 | this_end = last_start - 1; |
8233767a XG |
995 | |
996 | prealloc = alloc_extent_state_atomic(prealloc); | |
997 | BUG_ON(!prealloc); | |
c7f895a2 XG |
998 | |
999 | /* | |
1000 | * Avoid to free 'prealloc' if it can be merged with | |
1001 | * the later extent. | |
1002 | */ | |
d1310b2e | 1003 | err = insert_state(tree, prealloc, start, this_end, |
d38ed27f | 1004 | NULL, NULL, &bits, changeset); |
c2d904e0 JM |
1005 | if (err) |
1006 | extent_io_tree_panic(tree, err); | |
1007 | ||
9ed74f2d JB |
1008 | cache_state(prealloc, cached_state); |
1009 | prealloc = NULL; | |
d1310b2e CM |
1010 | start = this_end + 1; |
1011 | goto search_again; | |
1012 | } | |
1013 | /* | |
1014 | * | ---- desired range ---- | | |
1015 | * | state | | |
1016 | * We need to split the extent, and set the bit | |
1017 | * on the first half | |
1018 | */ | |
1019 | if (state->start <= end && state->end > end) { | |
1edbb734 | 1020 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1021 | *failed_start = start; |
1022 | err = -EEXIST; | |
1023 | goto out; | |
1024 | } | |
8233767a XG |
1025 | |
1026 | prealloc = alloc_extent_state_atomic(prealloc); | |
1027 | BUG_ON(!prealloc); | |
d1310b2e | 1028 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
1029 | if (err) |
1030 | extent_io_tree_panic(tree, err); | |
d1310b2e | 1031 | |
d38ed27f | 1032 | set_state_bits(tree, prealloc, &bits, changeset); |
2c64c53d | 1033 | cache_state(prealloc, cached_state); |
d1310b2e CM |
1034 | merge_state(tree, prealloc); |
1035 | prealloc = NULL; | |
1036 | goto out; | |
1037 | } | |
1038 | ||
b5a4ba14 DS |
1039 | search_again: |
1040 | if (start > end) | |
1041 | goto out; | |
1042 | spin_unlock(&tree->lock); | |
1043 | if (gfpflags_allow_blocking(mask)) | |
1044 | cond_resched(); | |
1045 | goto again; | |
d1310b2e CM |
1046 | |
1047 | out: | |
cad321ad | 1048 | spin_unlock(&tree->lock); |
d1310b2e CM |
1049 | if (prealloc) |
1050 | free_extent_state(prealloc); | |
1051 | ||
1052 | return err; | |
1053 | ||
d1310b2e | 1054 | } |
d1310b2e | 1055 | |
41074888 | 1056 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
9ee49a04 | 1057 | unsigned bits, u64 * failed_start, |
41074888 | 1058 | struct extent_state **cached_state, gfp_t mask) |
3fbe5c02 JM |
1059 | { |
1060 | return __set_extent_bit(tree, start, end, bits, 0, failed_start, | |
d38ed27f | 1061 | cached_state, mask, NULL); |
3fbe5c02 JM |
1062 | } |
1063 | ||
1064 | ||
462d6fac | 1065 | /** |
10983f2e LB |
1066 | * convert_extent_bit - convert all bits in a given range from one bit to |
1067 | * another | |
462d6fac JB |
1068 | * @tree: the io tree to search |
1069 | * @start: the start offset in bytes | |
1070 | * @end: the end offset in bytes (inclusive) | |
1071 | * @bits: the bits to set in this range | |
1072 | * @clear_bits: the bits to clear in this range | |
e6138876 | 1073 | * @cached_state: state that we're going to cache |
462d6fac JB |
1074 | * |
1075 | * This will go through and set bits for the given range. If any states exist | |
1076 | * already in this range they are set with the given bit and cleared of the | |
1077 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
1078 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
1079 | * boundary bits like LOCK. | |
210aa277 DS |
1080 | * |
1081 | * All allocations are done with GFP_NOFS. | |
462d6fac JB |
1082 | */ |
1083 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1084 | unsigned bits, unsigned clear_bits, |
210aa277 | 1085 | struct extent_state **cached_state) |
462d6fac JB |
1086 | { |
1087 | struct extent_state *state; | |
1088 | struct extent_state *prealloc = NULL; | |
1089 | struct rb_node *node; | |
12cfbad9 FDBM |
1090 | struct rb_node **p; |
1091 | struct rb_node *parent; | |
462d6fac JB |
1092 | int err = 0; |
1093 | u64 last_start; | |
1094 | u64 last_end; | |
c8fd3de7 | 1095 | bool first_iteration = true; |
462d6fac | 1096 | |
a5dee37d | 1097 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 1098 | |
462d6fac | 1099 | again: |
210aa277 | 1100 | if (!prealloc) { |
c8fd3de7 FM |
1101 | /* |
1102 | * Best effort, don't worry if extent state allocation fails | |
1103 | * here for the first iteration. We might have a cached state | |
1104 | * that matches exactly the target range, in which case no | |
1105 | * extent state allocations are needed. We'll only know this | |
1106 | * after locking the tree. | |
1107 | */ | |
210aa277 | 1108 | prealloc = alloc_extent_state(GFP_NOFS); |
c8fd3de7 | 1109 | if (!prealloc && !first_iteration) |
462d6fac JB |
1110 | return -ENOMEM; |
1111 | } | |
1112 | ||
1113 | spin_lock(&tree->lock); | |
e6138876 JB |
1114 | if (cached_state && *cached_state) { |
1115 | state = *cached_state; | |
1116 | if (state->start <= start && state->end > start && | |
27a3507d | 1117 | extent_state_in_tree(state)) { |
e6138876 JB |
1118 | node = &state->rb_node; |
1119 | goto hit_next; | |
1120 | } | |
1121 | } | |
1122 | ||
462d6fac JB |
1123 | /* |
1124 | * this search will find all the extents that end after | |
1125 | * our range starts. | |
1126 | */ | |
12cfbad9 | 1127 | node = tree_search_for_insert(tree, start, &p, &parent); |
462d6fac JB |
1128 | if (!node) { |
1129 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1130 | if (!prealloc) { |
1131 | err = -ENOMEM; | |
1132 | goto out; | |
1133 | } | |
12cfbad9 | 1134 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 1135 | &p, &parent, &bits, NULL); |
c2d904e0 JM |
1136 | if (err) |
1137 | extent_io_tree_panic(tree, err); | |
c42ac0bc FDBM |
1138 | cache_state(prealloc, cached_state); |
1139 | prealloc = NULL; | |
462d6fac JB |
1140 | goto out; |
1141 | } | |
1142 | state = rb_entry(node, struct extent_state, rb_node); | |
1143 | hit_next: | |
1144 | last_start = state->start; | |
1145 | last_end = state->end; | |
1146 | ||
1147 | /* | |
1148 | * | ---- desired range ---- | | |
1149 | * | state | | |
1150 | * | |
1151 | * Just lock what we found and keep going | |
1152 | */ | |
1153 | if (state->start == start && state->end <= end) { | |
d38ed27f | 1154 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1155 | cache_state(state, cached_state); |
fefdc557 | 1156 | state = clear_state_bit(tree, state, &clear_bits, 0, NULL); |
462d6fac JB |
1157 | if (last_end == (u64)-1) |
1158 | goto out; | |
462d6fac | 1159 | start = last_end + 1; |
d1ac6e41 LB |
1160 | if (start < end && state && state->start == start && |
1161 | !need_resched()) | |
1162 | goto hit_next; | |
462d6fac JB |
1163 | goto search_again; |
1164 | } | |
1165 | ||
1166 | /* | |
1167 | * | ---- desired range ---- | | |
1168 | * | state | | |
1169 | * or | |
1170 | * | ------------- state -------------- | | |
1171 | * | |
1172 | * We need to split the extent we found, and may flip bits on | |
1173 | * second half. | |
1174 | * | |
1175 | * If the extent we found extends past our | |
1176 | * range, we just split and search again. It'll get split | |
1177 | * again the next time though. | |
1178 | * | |
1179 | * If the extent we found is inside our range, we set the | |
1180 | * desired bit on it. | |
1181 | */ | |
1182 | if (state->start < start) { | |
1183 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1184 | if (!prealloc) { |
1185 | err = -ENOMEM; | |
1186 | goto out; | |
1187 | } | |
462d6fac | 1188 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1189 | if (err) |
1190 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1191 | prealloc = NULL; |
1192 | if (err) | |
1193 | goto out; | |
1194 | if (state->end <= end) { | |
d38ed27f | 1195 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1196 | cache_state(state, cached_state); |
fefdc557 QW |
1197 | state = clear_state_bit(tree, state, &clear_bits, 0, |
1198 | NULL); | |
462d6fac JB |
1199 | if (last_end == (u64)-1) |
1200 | goto out; | |
1201 | start = last_end + 1; | |
d1ac6e41 LB |
1202 | if (start < end && state && state->start == start && |
1203 | !need_resched()) | |
1204 | goto hit_next; | |
462d6fac JB |
1205 | } |
1206 | goto search_again; | |
1207 | } | |
1208 | /* | |
1209 | * | ---- desired range ---- | | |
1210 | * | state | or | state | | |
1211 | * | |
1212 | * There's a hole, we need to insert something in it and | |
1213 | * ignore the extent we found. | |
1214 | */ | |
1215 | if (state->start > start) { | |
1216 | u64 this_end; | |
1217 | if (end < last_start) | |
1218 | this_end = end; | |
1219 | else | |
1220 | this_end = last_start - 1; | |
1221 | ||
1222 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1223 | if (!prealloc) { |
1224 | err = -ENOMEM; | |
1225 | goto out; | |
1226 | } | |
462d6fac JB |
1227 | |
1228 | /* | |
1229 | * Avoid to free 'prealloc' if it can be merged with | |
1230 | * the later extent. | |
1231 | */ | |
1232 | err = insert_state(tree, prealloc, start, this_end, | |
d38ed27f | 1233 | NULL, NULL, &bits, NULL); |
c2d904e0 JM |
1234 | if (err) |
1235 | extent_io_tree_panic(tree, err); | |
e6138876 | 1236 | cache_state(prealloc, cached_state); |
462d6fac JB |
1237 | prealloc = NULL; |
1238 | start = this_end + 1; | |
1239 | goto search_again; | |
1240 | } | |
1241 | /* | |
1242 | * | ---- desired range ---- | | |
1243 | * | state | | |
1244 | * We need to split the extent, and set the bit | |
1245 | * on the first half | |
1246 | */ | |
1247 | if (state->start <= end && state->end > end) { | |
1248 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1249 | if (!prealloc) { |
1250 | err = -ENOMEM; | |
1251 | goto out; | |
1252 | } | |
462d6fac JB |
1253 | |
1254 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1255 | if (err) |
1256 | extent_io_tree_panic(tree, err); | |
462d6fac | 1257 | |
d38ed27f | 1258 | set_state_bits(tree, prealloc, &bits, NULL); |
e6138876 | 1259 | cache_state(prealloc, cached_state); |
fefdc557 | 1260 | clear_state_bit(tree, prealloc, &clear_bits, 0, NULL); |
462d6fac JB |
1261 | prealloc = NULL; |
1262 | goto out; | |
1263 | } | |
1264 | ||
462d6fac JB |
1265 | search_again: |
1266 | if (start > end) | |
1267 | goto out; | |
1268 | spin_unlock(&tree->lock); | |
210aa277 | 1269 | cond_resched(); |
c8fd3de7 | 1270 | first_iteration = false; |
462d6fac | 1271 | goto again; |
462d6fac JB |
1272 | |
1273 | out: | |
1274 | spin_unlock(&tree->lock); | |
1275 | if (prealloc) | |
1276 | free_extent_state(prealloc); | |
1277 | ||
1278 | return err; | |
462d6fac JB |
1279 | } |
1280 | ||
d1310b2e | 1281 | /* wrappers around set/clear extent bit */ |
d38ed27f | 1282 | int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
2c53b912 | 1283 | unsigned bits, struct extent_changeset *changeset) |
d38ed27f QW |
1284 | { |
1285 | /* | |
1286 | * We don't support EXTENT_LOCKED yet, as current changeset will | |
1287 | * record any bits changed, so for EXTENT_LOCKED case, it will | |
1288 | * either fail with -EEXIST or changeset will record the whole | |
1289 | * range. | |
1290 | */ | |
1291 | BUG_ON(bits & EXTENT_LOCKED); | |
1292 | ||
2c53b912 | 1293 | return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS, |
d38ed27f QW |
1294 | changeset); |
1295 | } | |
1296 | ||
fefdc557 QW |
1297 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
1298 | unsigned bits, int wake, int delete, | |
ae0f1625 | 1299 | struct extent_state **cached) |
fefdc557 QW |
1300 | { |
1301 | return __clear_extent_bit(tree, start, end, bits, wake, delete, | |
ae0f1625 | 1302 | cached, GFP_NOFS, NULL); |
fefdc557 QW |
1303 | } |
1304 | ||
fefdc557 | 1305 | int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
f734c44a | 1306 | unsigned bits, struct extent_changeset *changeset) |
fefdc557 QW |
1307 | { |
1308 | /* | |
1309 | * Don't support EXTENT_LOCKED case, same reason as | |
1310 | * set_record_extent_bits(). | |
1311 | */ | |
1312 | BUG_ON(bits & EXTENT_LOCKED); | |
1313 | ||
f734c44a | 1314 | return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS, |
fefdc557 QW |
1315 | changeset); |
1316 | } | |
1317 | ||
d352ac68 CM |
1318 | /* |
1319 | * either insert or lock state struct between start and end use mask to tell | |
1320 | * us if waiting is desired. | |
1321 | */ | |
1edbb734 | 1322 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
ff13db41 | 1323 | struct extent_state **cached_state) |
d1310b2e CM |
1324 | { |
1325 | int err; | |
1326 | u64 failed_start; | |
9ee49a04 | 1327 | |
d1310b2e | 1328 | while (1) { |
ff13db41 | 1329 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, |
3fbe5c02 | 1330 | EXTENT_LOCKED, &failed_start, |
d38ed27f | 1331 | cached_state, GFP_NOFS, NULL); |
d0082371 | 1332 | if (err == -EEXIST) { |
d1310b2e CM |
1333 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1334 | start = failed_start; | |
d0082371 | 1335 | } else |
d1310b2e | 1336 | break; |
d1310b2e CM |
1337 | WARN_ON(start > end); |
1338 | } | |
1339 | return err; | |
1340 | } | |
d1310b2e | 1341 | |
d0082371 | 1342 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1343 | { |
1344 | int err; | |
1345 | u64 failed_start; | |
1346 | ||
3fbe5c02 | 1347 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
d38ed27f | 1348 | &failed_start, NULL, GFP_NOFS, NULL); |
6643558d YZ |
1349 | if (err == -EEXIST) { |
1350 | if (failed_start > start) | |
1351 | clear_extent_bit(tree, start, failed_start - 1, | |
ae0f1625 | 1352 | EXTENT_LOCKED, 1, 0, NULL); |
25179201 | 1353 | return 0; |
6643558d | 1354 | } |
25179201 JB |
1355 | return 1; |
1356 | } | |
25179201 | 1357 | |
bd1fa4f0 | 1358 | void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1359 | { |
09cbfeaf KS |
1360 | unsigned long index = start >> PAGE_SHIFT; |
1361 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1362 | struct page *page; |
1363 | ||
1364 | while (index <= end_index) { | |
1365 | page = find_get_page(inode->i_mapping, index); | |
1366 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1367 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1368 | put_page(page); |
4adaa611 CM |
1369 | index++; |
1370 | } | |
4adaa611 CM |
1371 | } |
1372 | ||
f6311572 | 1373 | void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1374 | { |
09cbfeaf KS |
1375 | unsigned long index = start >> PAGE_SHIFT; |
1376 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1377 | struct page *page; |
1378 | ||
1379 | while (index <= end_index) { | |
1380 | page = find_get_page(inode->i_mapping, index); | |
1381 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
4adaa611 | 1382 | __set_page_dirty_nobuffers(page); |
8d38633c | 1383 | account_page_redirty(page); |
09cbfeaf | 1384 | put_page(page); |
4adaa611 CM |
1385 | index++; |
1386 | } | |
4adaa611 CM |
1387 | } |
1388 | ||
d1310b2e CM |
1389 | /* |
1390 | * helper function to set both pages and extents in the tree writeback | |
1391 | */ | |
35de6db2 | 1392 | static void set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e | 1393 | { |
c6100a4b | 1394 | tree->ops->set_range_writeback(tree->private_data, start, end); |
d1310b2e | 1395 | } |
d1310b2e | 1396 | |
d352ac68 CM |
1397 | /* find the first state struct with 'bits' set after 'start', and |
1398 | * return it. tree->lock must be held. NULL will returned if | |
1399 | * nothing was found after 'start' | |
1400 | */ | |
48a3b636 ES |
1401 | static struct extent_state * |
1402 | find_first_extent_bit_state(struct extent_io_tree *tree, | |
9ee49a04 | 1403 | u64 start, unsigned bits) |
d7fc640e CM |
1404 | { |
1405 | struct rb_node *node; | |
1406 | struct extent_state *state; | |
1407 | ||
1408 | /* | |
1409 | * this search will find all the extents that end after | |
1410 | * our range starts. | |
1411 | */ | |
1412 | node = tree_search(tree, start); | |
d397712b | 1413 | if (!node) |
d7fc640e | 1414 | goto out; |
d7fc640e | 1415 | |
d397712b | 1416 | while (1) { |
d7fc640e | 1417 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1418 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1419 | return state; |
d397712b | 1420 | |
d7fc640e CM |
1421 | node = rb_next(node); |
1422 | if (!node) | |
1423 | break; | |
1424 | } | |
1425 | out: | |
1426 | return NULL; | |
1427 | } | |
d7fc640e | 1428 | |
69261c4b XG |
1429 | /* |
1430 | * find the first offset in the io tree with 'bits' set. zero is | |
1431 | * returned if we find something, and *start_ret and *end_ret are | |
1432 | * set to reflect the state struct that was found. | |
1433 | * | |
477d7eaf | 1434 | * If nothing was found, 1 is returned. If found something, return 0. |
69261c4b XG |
1435 | */ |
1436 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
9ee49a04 | 1437 | u64 *start_ret, u64 *end_ret, unsigned bits, |
e6138876 | 1438 | struct extent_state **cached_state) |
69261c4b XG |
1439 | { |
1440 | struct extent_state *state; | |
e6138876 | 1441 | struct rb_node *n; |
69261c4b XG |
1442 | int ret = 1; |
1443 | ||
1444 | spin_lock(&tree->lock); | |
e6138876 JB |
1445 | if (cached_state && *cached_state) { |
1446 | state = *cached_state; | |
27a3507d | 1447 | if (state->end == start - 1 && extent_state_in_tree(state)) { |
e6138876 JB |
1448 | n = rb_next(&state->rb_node); |
1449 | while (n) { | |
1450 | state = rb_entry(n, struct extent_state, | |
1451 | rb_node); | |
1452 | if (state->state & bits) | |
1453 | goto got_it; | |
1454 | n = rb_next(n); | |
1455 | } | |
1456 | free_extent_state(*cached_state); | |
1457 | *cached_state = NULL; | |
1458 | goto out; | |
1459 | } | |
1460 | free_extent_state(*cached_state); | |
1461 | *cached_state = NULL; | |
1462 | } | |
1463 | ||
69261c4b | 1464 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1465 | got_it: |
69261c4b | 1466 | if (state) { |
e38e2ed7 | 1467 | cache_state_if_flags(state, cached_state, 0); |
69261c4b XG |
1468 | *start_ret = state->start; |
1469 | *end_ret = state->end; | |
1470 | ret = 0; | |
1471 | } | |
e6138876 | 1472 | out: |
69261c4b XG |
1473 | spin_unlock(&tree->lock); |
1474 | return ret; | |
1475 | } | |
1476 | ||
d352ac68 CM |
1477 | /* |
1478 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1479 | * more than 'max_bytes'. start and end are used to return the range, | |
1480 | * | |
1481 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1482 | */ | |
c8b97818 | 1483 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1484 | u64 *start, u64 *end, u64 max_bytes, |
1485 | struct extent_state **cached_state) | |
d1310b2e CM |
1486 | { |
1487 | struct rb_node *node; | |
1488 | struct extent_state *state; | |
1489 | u64 cur_start = *start; | |
1490 | u64 found = 0; | |
1491 | u64 total_bytes = 0; | |
1492 | ||
cad321ad | 1493 | spin_lock(&tree->lock); |
c8b97818 | 1494 | |
d1310b2e CM |
1495 | /* |
1496 | * this search will find all the extents that end after | |
1497 | * our range starts. | |
1498 | */ | |
80ea96b1 | 1499 | node = tree_search(tree, cur_start); |
2b114d1d | 1500 | if (!node) { |
3b951516 CM |
1501 | if (!found) |
1502 | *end = (u64)-1; | |
d1310b2e CM |
1503 | goto out; |
1504 | } | |
1505 | ||
d397712b | 1506 | while (1) { |
d1310b2e | 1507 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1508 | if (found && (state->start != cur_start || |
1509 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1510 | goto out; |
1511 | } | |
1512 | if (!(state->state & EXTENT_DELALLOC)) { | |
1513 | if (!found) | |
1514 | *end = state->end; | |
1515 | goto out; | |
1516 | } | |
c2a128d2 | 1517 | if (!found) { |
d1310b2e | 1518 | *start = state->start; |
c2a128d2 | 1519 | *cached_state = state; |
b7ac31b7 | 1520 | refcount_inc(&state->refs); |
c2a128d2 | 1521 | } |
d1310b2e CM |
1522 | found++; |
1523 | *end = state->end; | |
1524 | cur_start = state->end + 1; | |
1525 | node = rb_next(node); | |
d1310b2e | 1526 | total_bytes += state->end - state->start + 1; |
7bf811a5 | 1527 | if (total_bytes >= max_bytes) |
573aecaf | 1528 | break; |
573aecaf | 1529 | if (!node) |
d1310b2e CM |
1530 | break; |
1531 | } | |
1532 | out: | |
cad321ad | 1533 | spin_unlock(&tree->lock); |
d1310b2e CM |
1534 | return found; |
1535 | } | |
1536 | ||
da2c7009 LB |
1537 | static int __process_pages_contig(struct address_space *mapping, |
1538 | struct page *locked_page, | |
1539 | pgoff_t start_index, pgoff_t end_index, | |
1540 | unsigned long page_ops, pgoff_t *index_ret); | |
1541 | ||
143bede5 JM |
1542 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1543 | struct page *locked_page, | |
1544 | u64 start, u64 end) | |
c8b97818 | 1545 | { |
09cbfeaf KS |
1546 | unsigned long index = start >> PAGE_SHIFT; |
1547 | unsigned long end_index = end >> PAGE_SHIFT; | |
c8b97818 | 1548 | |
76c0021d | 1549 | ASSERT(locked_page); |
c8b97818 | 1550 | if (index == locked_page->index && end_index == index) |
143bede5 | 1551 | return; |
c8b97818 | 1552 | |
76c0021d LB |
1553 | __process_pages_contig(inode->i_mapping, locked_page, index, end_index, |
1554 | PAGE_UNLOCK, NULL); | |
c8b97818 CM |
1555 | } |
1556 | ||
1557 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1558 | struct page *locked_page, | |
1559 | u64 delalloc_start, | |
1560 | u64 delalloc_end) | |
1561 | { | |
09cbfeaf | 1562 | unsigned long index = delalloc_start >> PAGE_SHIFT; |
76c0021d | 1563 | unsigned long index_ret = index; |
09cbfeaf | 1564 | unsigned long end_index = delalloc_end >> PAGE_SHIFT; |
c8b97818 | 1565 | int ret; |
c8b97818 | 1566 | |
76c0021d | 1567 | ASSERT(locked_page); |
c8b97818 CM |
1568 | if (index == locked_page->index && index == end_index) |
1569 | return 0; | |
1570 | ||
76c0021d LB |
1571 | ret = __process_pages_contig(inode->i_mapping, locked_page, index, |
1572 | end_index, PAGE_LOCK, &index_ret); | |
1573 | if (ret == -EAGAIN) | |
1574 | __unlock_for_delalloc(inode, locked_page, delalloc_start, | |
1575 | (u64)index_ret << PAGE_SHIFT); | |
c8b97818 CM |
1576 | return ret; |
1577 | } | |
1578 | ||
1579 | /* | |
1580 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1581 | * more than 'max_bytes'. start and end are used to return the range, | |
1582 | * | |
1583 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1584 | */ | |
294e30fe JB |
1585 | STATIC u64 find_lock_delalloc_range(struct inode *inode, |
1586 | struct extent_io_tree *tree, | |
1587 | struct page *locked_page, u64 *start, | |
1588 | u64 *end, u64 max_bytes) | |
c8b97818 CM |
1589 | { |
1590 | u64 delalloc_start; | |
1591 | u64 delalloc_end; | |
1592 | u64 found; | |
9655d298 | 1593 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1594 | int ret; |
1595 | int loops = 0; | |
1596 | ||
1597 | again: | |
1598 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1599 | delalloc_start = *start; | |
1600 | delalloc_end = 0; | |
1601 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1602 | max_bytes, &cached_state); |
70b99e69 | 1603 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1604 | *start = delalloc_start; |
1605 | *end = delalloc_end; | |
c2a128d2 | 1606 | free_extent_state(cached_state); |
385fe0be | 1607 | return 0; |
c8b97818 CM |
1608 | } |
1609 | ||
70b99e69 CM |
1610 | /* |
1611 | * start comes from the offset of locked_page. We have to lock | |
1612 | * pages in order, so we can't process delalloc bytes before | |
1613 | * locked_page | |
1614 | */ | |
d397712b | 1615 | if (delalloc_start < *start) |
70b99e69 | 1616 | delalloc_start = *start; |
70b99e69 | 1617 | |
c8b97818 CM |
1618 | /* |
1619 | * make sure to limit the number of pages we try to lock down | |
c8b97818 | 1620 | */ |
7bf811a5 JB |
1621 | if (delalloc_end + 1 - delalloc_start > max_bytes) |
1622 | delalloc_end = delalloc_start + max_bytes - 1; | |
d397712b | 1623 | |
c8b97818 CM |
1624 | /* step two, lock all the pages after the page that has start */ |
1625 | ret = lock_delalloc_pages(inode, locked_page, | |
1626 | delalloc_start, delalloc_end); | |
1627 | if (ret == -EAGAIN) { | |
1628 | /* some of the pages are gone, lets avoid looping by | |
1629 | * shortening the size of the delalloc range we're searching | |
1630 | */ | |
9655d298 | 1631 | free_extent_state(cached_state); |
7d788742 | 1632 | cached_state = NULL; |
c8b97818 | 1633 | if (!loops) { |
09cbfeaf | 1634 | max_bytes = PAGE_SIZE; |
c8b97818 CM |
1635 | loops = 1; |
1636 | goto again; | |
1637 | } else { | |
1638 | found = 0; | |
1639 | goto out_failed; | |
1640 | } | |
1641 | } | |
79787eaa | 1642 | BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */ |
c8b97818 CM |
1643 | |
1644 | /* step three, lock the state bits for the whole range */ | |
ff13db41 | 1645 | lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state); |
c8b97818 CM |
1646 | |
1647 | /* then test to make sure it is all still delalloc */ | |
1648 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1649 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1650 | if (!ret) { |
9655d298 | 1651 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
e43bbe5e | 1652 | &cached_state); |
c8b97818 CM |
1653 | __unlock_for_delalloc(inode, locked_page, |
1654 | delalloc_start, delalloc_end); | |
1655 | cond_resched(); | |
1656 | goto again; | |
1657 | } | |
9655d298 | 1658 | free_extent_state(cached_state); |
c8b97818 CM |
1659 | *start = delalloc_start; |
1660 | *end = delalloc_end; | |
1661 | out_failed: | |
1662 | return found; | |
1663 | } | |
1664 | ||
da2c7009 LB |
1665 | static int __process_pages_contig(struct address_space *mapping, |
1666 | struct page *locked_page, | |
1667 | pgoff_t start_index, pgoff_t end_index, | |
1668 | unsigned long page_ops, pgoff_t *index_ret) | |
c8b97818 | 1669 | { |
873695b3 | 1670 | unsigned long nr_pages = end_index - start_index + 1; |
da2c7009 | 1671 | unsigned long pages_locked = 0; |
873695b3 | 1672 | pgoff_t index = start_index; |
c8b97818 | 1673 | struct page *pages[16]; |
873695b3 | 1674 | unsigned ret; |
da2c7009 | 1675 | int err = 0; |
c8b97818 | 1676 | int i; |
771ed689 | 1677 | |
da2c7009 LB |
1678 | if (page_ops & PAGE_LOCK) { |
1679 | ASSERT(page_ops == PAGE_LOCK); | |
1680 | ASSERT(index_ret && *index_ret == start_index); | |
1681 | } | |
1682 | ||
704de49d | 1683 | if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0) |
873695b3 | 1684 | mapping_set_error(mapping, -EIO); |
704de49d | 1685 | |
d397712b | 1686 | while (nr_pages > 0) { |
873695b3 | 1687 | ret = find_get_pages_contig(mapping, index, |
5b050f04 CM |
1688 | min_t(unsigned long, |
1689 | nr_pages, ARRAY_SIZE(pages)), pages); | |
da2c7009 LB |
1690 | if (ret == 0) { |
1691 | /* | |
1692 | * Only if we're going to lock these pages, | |
1693 | * can we find nothing at @index. | |
1694 | */ | |
1695 | ASSERT(page_ops & PAGE_LOCK); | |
49d4a334 LB |
1696 | err = -EAGAIN; |
1697 | goto out; | |
da2c7009 | 1698 | } |
8b62b72b | 1699 | |
da2c7009 | 1700 | for (i = 0; i < ret; i++) { |
c2790a2e | 1701 | if (page_ops & PAGE_SET_PRIVATE2) |
8b62b72b CM |
1702 | SetPagePrivate2(pages[i]); |
1703 | ||
c8b97818 | 1704 | if (pages[i] == locked_page) { |
09cbfeaf | 1705 | put_page(pages[i]); |
da2c7009 | 1706 | pages_locked++; |
c8b97818 CM |
1707 | continue; |
1708 | } | |
c2790a2e | 1709 | if (page_ops & PAGE_CLEAR_DIRTY) |
c8b97818 | 1710 | clear_page_dirty_for_io(pages[i]); |
c2790a2e | 1711 | if (page_ops & PAGE_SET_WRITEBACK) |
c8b97818 | 1712 | set_page_writeback(pages[i]); |
704de49d FM |
1713 | if (page_ops & PAGE_SET_ERROR) |
1714 | SetPageError(pages[i]); | |
c2790a2e | 1715 | if (page_ops & PAGE_END_WRITEBACK) |
c8b97818 | 1716 | end_page_writeback(pages[i]); |
c2790a2e | 1717 | if (page_ops & PAGE_UNLOCK) |
771ed689 | 1718 | unlock_page(pages[i]); |
da2c7009 LB |
1719 | if (page_ops & PAGE_LOCK) { |
1720 | lock_page(pages[i]); | |
1721 | if (!PageDirty(pages[i]) || | |
1722 | pages[i]->mapping != mapping) { | |
1723 | unlock_page(pages[i]); | |
1724 | put_page(pages[i]); | |
1725 | err = -EAGAIN; | |
1726 | goto out; | |
1727 | } | |
1728 | } | |
09cbfeaf | 1729 | put_page(pages[i]); |
da2c7009 | 1730 | pages_locked++; |
c8b97818 CM |
1731 | } |
1732 | nr_pages -= ret; | |
1733 | index += ret; | |
1734 | cond_resched(); | |
1735 | } | |
da2c7009 LB |
1736 | out: |
1737 | if (err && index_ret) | |
1738 | *index_ret = start_index + pages_locked - 1; | |
1739 | return err; | |
c8b97818 | 1740 | } |
c8b97818 | 1741 | |
873695b3 LB |
1742 | void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end, |
1743 | u64 delalloc_end, struct page *locked_page, | |
1744 | unsigned clear_bits, | |
1745 | unsigned long page_ops) | |
1746 | { | |
1747 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0, | |
ae0f1625 | 1748 | NULL); |
873695b3 LB |
1749 | |
1750 | __process_pages_contig(inode->i_mapping, locked_page, | |
1751 | start >> PAGE_SHIFT, end >> PAGE_SHIFT, | |
da2c7009 | 1752 | page_ops, NULL); |
873695b3 LB |
1753 | } |
1754 | ||
d352ac68 CM |
1755 | /* |
1756 | * count the number of bytes in the tree that have a given bit(s) | |
1757 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1758 | * cached. The total number found is returned. | |
1759 | */ | |
d1310b2e CM |
1760 | u64 count_range_bits(struct extent_io_tree *tree, |
1761 | u64 *start, u64 search_end, u64 max_bytes, | |
9ee49a04 | 1762 | unsigned bits, int contig) |
d1310b2e CM |
1763 | { |
1764 | struct rb_node *node; | |
1765 | struct extent_state *state; | |
1766 | u64 cur_start = *start; | |
1767 | u64 total_bytes = 0; | |
ec29ed5b | 1768 | u64 last = 0; |
d1310b2e CM |
1769 | int found = 0; |
1770 | ||
fae7f21c | 1771 | if (WARN_ON(search_end <= cur_start)) |
d1310b2e | 1772 | return 0; |
d1310b2e | 1773 | |
cad321ad | 1774 | spin_lock(&tree->lock); |
d1310b2e CM |
1775 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1776 | total_bytes = tree->dirty_bytes; | |
1777 | goto out; | |
1778 | } | |
1779 | /* | |
1780 | * this search will find all the extents that end after | |
1781 | * our range starts. | |
1782 | */ | |
80ea96b1 | 1783 | node = tree_search(tree, cur_start); |
d397712b | 1784 | if (!node) |
d1310b2e | 1785 | goto out; |
d1310b2e | 1786 | |
d397712b | 1787 | while (1) { |
d1310b2e CM |
1788 | state = rb_entry(node, struct extent_state, rb_node); |
1789 | if (state->start > search_end) | |
1790 | break; | |
ec29ed5b CM |
1791 | if (contig && found && state->start > last + 1) |
1792 | break; | |
1793 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1794 | total_bytes += min(search_end, state->end) + 1 - |
1795 | max(cur_start, state->start); | |
1796 | if (total_bytes >= max_bytes) | |
1797 | break; | |
1798 | if (!found) { | |
af60bed2 | 1799 | *start = max(cur_start, state->start); |
d1310b2e CM |
1800 | found = 1; |
1801 | } | |
ec29ed5b CM |
1802 | last = state->end; |
1803 | } else if (contig && found) { | |
1804 | break; | |
d1310b2e CM |
1805 | } |
1806 | node = rb_next(node); | |
1807 | if (!node) | |
1808 | break; | |
1809 | } | |
1810 | out: | |
cad321ad | 1811 | spin_unlock(&tree->lock); |
d1310b2e CM |
1812 | return total_bytes; |
1813 | } | |
b2950863 | 1814 | |
d352ac68 CM |
1815 | /* |
1816 | * set the private field for a given byte offset in the tree. If there isn't | |
1817 | * an extent_state there already, this does nothing. | |
1818 | */ | |
f827ba9a | 1819 | static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start, |
47dc196a | 1820 | struct io_failure_record *failrec) |
d1310b2e CM |
1821 | { |
1822 | struct rb_node *node; | |
1823 | struct extent_state *state; | |
1824 | int ret = 0; | |
1825 | ||
cad321ad | 1826 | spin_lock(&tree->lock); |
d1310b2e CM |
1827 | /* |
1828 | * this search will find all the extents that end after | |
1829 | * our range starts. | |
1830 | */ | |
80ea96b1 | 1831 | node = tree_search(tree, start); |
2b114d1d | 1832 | if (!node) { |
d1310b2e CM |
1833 | ret = -ENOENT; |
1834 | goto out; | |
1835 | } | |
1836 | state = rb_entry(node, struct extent_state, rb_node); | |
1837 | if (state->start != start) { | |
1838 | ret = -ENOENT; | |
1839 | goto out; | |
1840 | } | |
47dc196a | 1841 | state->failrec = failrec; |
d1310b2e | 1842 | out: |
cad321ad | 1843 | spin_unlock(&tree->lock); |
d1310b2e CM |
1844 | return ret; |
1845 | } | |
1846 | ||
f827ba9a | 1847 | static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start, |
47dc196a | 1848 | struct io_failure_record **failrec) |
d1310b2e CM |
1849 | { |
1850 | struct rb_node *node; | |
1851 | struct extent_state *state; | |
1852 | int ret = 0; | |
1853 | ||
cad321ad | 1854 | spin_lock(&tree->lock); |
d1310b2e CM |
1855 | /* |
1856 | * this search will find all the extents that end after | |
1857 | * our range starts. | |
1858 | */ | |
80ea96b1 | 1859 | node = tree_search(tree, start); |
2b114d1d | 1860 | if (!node) { |
d1310b2e CM |
1861 | ret = -ENOENT; |
1862 | goto out; | |
1863 | } | |
1864 | state = rb_entry(node, struct extent_state, rb_node); | |
1865 | if (state->start != start) { | |
1866 | ret = -ENOENT; | |
1867 | goto out; | |
1868 | } | |
47dc196a | 1869 | *failrec = state->failrec; |
d1310b2e | 1870 | out: |
cad321ad | 1871 | spin_unlock(&tree->lock); |
d1310b2e CM |
1872 | return ret; |
1873 | } | |
1874 | ||
1875 | /* | |
1876 | * searches a range in the state tree for a given mask. | |
70dec807 | 1877 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1878 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1879 | * range is found set. | |
1880 | */ | |
1881 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1882 | unsigned bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1883 | { |
1884 | struct extent_state *state = NULL; | |
1885 | struct rb_node *node; | |
1886 | int bitset = 0; | |
d1310b2e | 1887 | |
cad321ad | 1888 | spin_lock(&tree->lock); |
27a3507d | 1889 | if (cached && extent_state_in_tree(cached) && cached->start <= start && |
df98b6e2 | 1890 | cached->end > start) |
9655d298 CM |
1891 | node = &cached->rb_node; |
1892 | else | |
1893 | node = tree_search(tree, start); | |
d1310b2e CM |
1894 | while (node && start <= end) { |
1895 | state = rb_entry(node, struct extent_state, rb_node); | |
1896 | ||
1897 | if (filled && state->start > start) { | |
1898 | bitset = 0; | |
1899 | break; | |
1900 | } | |
1901 | ||
1902 | if (state->start > end) | |
1903 | break; | |
1904 | ||
1905 | if (state->state & bits) { | |
1906 | bitset = 1; | |
1907 | if (!filled) | |
1908 | break; | |
1909 | } else if (filled) { | |
1910 | bitset = 0; | |
1911 | break; | |
1912 | } | |
46562cec CM |
1913 | |
1914 | if (state->end == (u64)-1) | |
1915 | break; | |
1916 | ||
d1310b2e CM |
1917 | start = state->end + 1; |
1918 | if (start > end) | |
1919 | break; | |
1920 | node = rb_next(node); | |
1921 | if (!node) { | |
1922 | if (filled) | |
1923 | bitset = 0; | |
1924 | break; | |
1925 | } | |
1926 | } | |
cad321ad | 1927 | spin_unlock(&tree->lock); |
d1310b2e CM |
1928 | return bitset; |
1929 | } | |
d1310b2e CM |
1930 | |
1931 | /* | |
1932 | * helper function to set a given page up to date if all the | |
1933 | * extents in the tree for that page are up to date | |
1934 | */ | |
143bede5 | 1935 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e | 1936 | { |
4eee4fa4 | 1937 | u64 start = page_offset(page); |
09cbfeaf | 1938 | u64 end = start + PAGE_SIZE - 1; |
9655d298 | 1939 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 1940 | SetPageUptodate(page); |
d1310b2e CM |
1941 | } |
1942 | ||
7870d082 JB |
1943 | int free_io_failure(struct extent_io_tree *failure_tree, |
1944 | struct extent_io_tree *io_tree, | |
1945 | struct io_failure_record *rec) | |
4a54c8c1 JS |
1946 | { |
1947 | int ret; | |
1948 | int err = 0; | |
4a54c8c1 | 1949 | |
47dc196a | 1950 | set_state_failrec(failure_tree, rec->start, NULL); |
4a54c8c1 JS |
1951 | ret = clear_extent_bits(failure_tree, rec->start, |
1952 | rec->start + rec->len - 1, | |
91166212 | 1953 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 JS |
1954 | if (ret) |
1955 | err = ret; | |
1956 | ||
7870d082 | 1957 | ret = clear_extent_bits(io_tree, rec->start, |
53b381b3 | 1958 | rec->start + rec->len - 1, |
91166212 | 1959 | EXTENT_DAMAGED); |
53b381b3 DW |
1960 | if (ret && !err) |
1961 | err = ret; | |
4a54c8c1 JS |
1962 | |
1963 | kfree(rec); | |
1964 | return err; | |
1965 | } | |
1966 | ||
4a54c8c1 JS |
1967 | /* |
1968 | * this bypasses the standard btrfs submit functions deliberately, as | |
1969 | * the standard behavior is to write all copies in a raid setup. here we only | |
1970 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
1971 | * submit_bio directly. | |
3ec706c8 | 1972 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
1973 | * actually prevents the read that triggered the error from finishing. |
1974 | * currently, there can be no more than two copies of every data bit. thus, | |
1975 | * exactly one rewrite is required. | |
1976 | */ | |
6ec656bc JB |
1977 | int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start, |
1978 | u64 length, u64 logical, struct page *page, | |
1979 | unsigned int pg_offset, int mirror_num) | |
4a54c8c1 JS |
1980 | { |
1981 | struct bio *bio; | |
1982 | struct btrfs_device *dev; | |
4a54c8c1 JS |
1983 | u64 map_length = 0; |
1984 | u64 sector; | |
1985 | struct btrfs_bio *bbio = NULL; | |
1986 | int ret; | |
1987 | ||
1751e8a6 | 1988 | ASSERT(!(fs_info->sb->s_flags & SB_RDONLY)); |
4a54c8c1 JS |
1989 | BUG_ON(!mirror_num); |
1990 | ||
c5e4c3d7 | 1991 | bio = btrfs_io_bio_alloc(1); |
4f024f37 | 1992 | bio->bi_iter.bi_size = 0; |
4a54c8c1 JS |
1993 | map_length = length; |
1994 | ||
b5de8d0d FM |
1995 | /* |
1996 | * Avoid races with device replace and make sure our bbio has devices | |
1997 | * associated to its stripes that don't go away while we are doing the | |
1998 | * read repair operation. | |
1999 | */ | |
2000 | btrfs_bio_counter_inc_blocked(fs_info); | |
e4ff5fb5 | 2001 | if (btrfs_is_parity_mirror(fs_info, logical, length)) { |
c725328c LB |
2002 | /* |
2003 | * Note that we don't use BTRFS_MAP_WRITE because it's supposed | |
2004 | * to update all raid stripes, but here we just want to correct | |
2005 | * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad | |
2006 | * stripe's dev and sector. | |
2007 | */ | |
2008 | ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical, | |
2009 | &map_length, &bbio, 0); | |
2010 | if (ret) { | |
2011 | btrfs_bio_counter_dec(fs_info); | |
2012 | bio_put(bio); | |
2013 | return -EIO; | |
2014 | } | |
2015 | ASSERT(bbio->mirror_num == 1); | |
2016 | } else { | |
2017 | ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, | |
2018 | &map_length, &bbio, mirror_num); | |
2019 | if (ret) { | |
2020 | btrfs_bio_counter_dec(fs_info); | |
2021 | bio_put(bio); | |
2022 | return -EIO; | |
2023 | } | |
2024 | BUG_ON(mirror_num != bbio->mirror_num); | |
4a54c8c1 | 2025 | } |
c725328c LB |
2026 | |
2027 | sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9; | |
4f024f37 | 2028 | bio->bi_iter.bi_sector = sector; |
c725328c | 2029 | dev = bbio->stripes[bbio->mirror_num - 1].dev; |
6e9606d2 | 2030 | btrfs_put_bbio(bbio); |
ebbede42 AJ |
2031 | if (!dev || !dev->bdev || |
2032 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) { | |
b5de8d0d | 2033 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2034 | bio_put(bio); |
2035 | return -EIO; | |
2036 | } | |
74d46992 | 2037 | bio_set_dev(bio, dev->bdev); |
70fd7614 | 2038 | bio->bi_opf = REQ_OP_WRITE | REQ_SYNC; |
ffdd2018 | 2039 | bio_add_page(bio, page, length, pg_offset); |
4a54c8c1 | 2040 | |
4e49ea4a | 2041 | if (btrfsic_submit_bio_wait(bio)) { |
4a54c8c1 | 2042 | /* try to remap that extent elsewhere? */ |
b5de8d0d | 2043 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 | 2044 | bio_put(bio); |
442a4f63 | 2045 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
2046 | return -EIO; |
2047 | } | |
2048 | ||
b14af3b4 DS |
2049 | btrfs_info_rl_in_rcu(fs_info, |
2050 | "read error corrected: ino %llu off %llu (dev %s sector %llu)", | |
6ec656bc | 2051 | ino, start, |
1203b681 | 2052 | rcu_str_deref(dev->name), sector); |
b5de8d0d | 2053 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2054 | bio_put(bio); |
2055 | return 0; | |
2056 | } | |
2057 | ||
2ff7e61e JM |
2058 | int repair_eb_io_failure(struct btrfs_fs_info *fs_info, |
2059 | struct extent_buffer *eb, int mirror_num) | |
ea466794 | 2060 | { |
ea466794 | 2061 | u64 start = eb->start; |
cc5e31a4 | 2062 | int i, num_pages = num_extent_pages(eb); |
d95603b2 | 2063 | int ret = 0; |
ea466794 | 2064 | |
bc98a42c | 2065 | if (sb_rdonly(fs_info->sb)) |
908960c6 ID |
2066 | return -EROFS; |
2067 | ||
ea466794 | 2068 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 2069 | struct page *p = eb->pages[i]; |
1203b681 | 2070 | |
6ec656bc | 2071 | ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p, |
1203b681 | 2072 | start - page_offset(p), mirror_num); |
ea466794 JB |
2073 | if (ret) |
2074 | break; | |
09cbfeaf | 2075 | start += PAGE_SIZE; |
ea466794 JB |
2076 | } |
2077 | ||
2078 | return ret; | |
2079 | } | |
2080 | ||
4a54c8c1 JS |
2081 | /* |
2082 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2083 | * to see if we need to process or clean up an io_failure_record | |
2084 | */ | |
7870d082 JB |
2085 | int clean_io_failure(struct btrfs_fs_info *fs_info, |
2086 | struct extent_io_tree *failure_tree, | |
2087 | struct extent_io_tree *io_tree, u64 start, | |
2088 | struct page *page, u64 ino, unsigned int pg_offset) | |
4a54c8c1 JS |
2089 | { |
2090 | u64 private; | |
4a54c8c1 | 2091 | struct io_failure_record *failrec; |
4a54c8c1 JS |
2092 | struct extent_state *state; |
2093 | int num_copies; | |
4a54c8c1 | 2094 | int ret; |
4a54c8c1 JS |
2095 | |
2096 | private = 0; | |
7870d082 JB |
2097 | ret = count_range_bits(failure_tree, &private, (u64)-1, 1, |
2098 | EXTENT_DIRTY, 0); | |
4a54c8c1 JS |
2099 | if (!ret) |
2100 | return 0; | |
2101 | ||
7870d082 | 2102 | ret = get_state_failrec(failure_tree, start, &failrec); |
4a54c8c1 JS |
2103 | if (ret) |
2104 | return 0; | |
2105 | ||
4a54c8c1 JS |
2106 | BUG_ON(!failrec->this_mirror); |
2107 | ||
2108 | if (failrec->in_validation) { | |
2109 | /* there was no real error, just free the record */ | |
ab8d0fc4 JM |
2110 | btrfs_debug(fs_info, |
2111 | "clean_io_failure: freeing dummy error at %llu", | |
2112 | failrec->start); | |
4a54c8c1 JS |
2113 | goto out; |
2114 | } | |
bc98a42c | 2115 | if (sb_rdonly(fs_info->sb)) |
908960c6 | 2116 | goto out; |
4a54c8c1 | 2117 | |
7870d082 JB |
2118 | spin_lock(&io_tree->lock); |
2119 | state = find_first_extent_bit_state(io_tree, | |
4a54c8c1 JS |
2120 | failrec->start, |
2121 | EXTENT_LOCKED); | |
7870d082 | 2122 | spin_unlock(&io_tree->lock); |
4a54c8c1 | 2123 | |
883d0de4 MX |
2124 | if (state && state->start <= failrec->start && |
2125 | state->end >= failrec->start + failrec->len - 1) { | |
3ec706c8 SB |
2126 | num_copies = btrfs_num_copies(fs_info, failrec->logical, |
2127 | failrec->len); | |
4a54c8c1 | 2128 | if (num_copies > 1) { |
7870d082 JB |
2129 | repair_io_failure(fs_info, ino, start, failrec->len, |
2130 | failrec->logical, page, pg_offset, | |
2131 | failrec->failed_mirror); | |
4a54c8c1 JS |
2132 | } |
2133 | } | |
2134 | ||
2135 | out: | |
7870d082 | 2136 | free_io_failure(failure_tree, io_tree, failrec); |
4a54c8c1 | 2137 | |
454ff3de | 2138 | return 0; |
4a54c8c1 JS |
2139 | } |
2140 | ||
f612496b MX |
2141 | /* |
2142 | * Can be called when | |
2143 | * - hold extent lock | |
2144 | * - under ordered extent | |
2145 | * - the inode is freeing | |
2146 | */ | |
7ab7956e | 2147 | void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end) |
f612496b | 2148 | { |
7ab7956e | 2149 | struct extent_io_tree *failure_tree = &inode->io_failure_tree; |
f612496b MX |
2150 | struct io_failure_record *failrec; |
2151 | struct extent_state *state, *next; | |
2152 | ||
2153 | if (RB_EMPTY_ROOT(&failure_tree->state)) | |
2154 | return; | |
2155 | ||
2156 | spin_lock(&failure_tree->lock); | |
2157 | state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY); | |
2158 | while (state) { | |
2159 | if (state->start > end) | |
2160 | break; | |
2161 | ||
2162 | ASSERT(state->end <= end); | |
2163 | ||
2164 | next = next_state(state); | |
2165 | ||
47dc196a | 2166 | failrec = state->failrec; |
f612496b MX |
2167 | free_extent_state(state); |
2168 | kfree(failrec); | |
2169 | ||
2170 | state = next; | |
2171 | } | |
2172 | spin_unlock(&failure_tree->lock); | |
2173 | } | |
2174 | ||
2fe6303e | 2175 | int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end, |
47dc196a | 2176 | struct io_failure_record **failrec_ret) |
4a54c8c1 | 2177 | { |
ab8d0fc4 | 2178 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e | 2179 | struct io_failure_record *failrec; |
4a54c8c1 | 2180 | struct extent_map *em; |
4a54c8c1 JS |
2181 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2182 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2183 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
4a54c8c1 | 2184 | int ret; |
4a54c8c1 JS |
2185 | u64 logical; |
2186 | ||
47dc196a | 2187 | ret = get_state_failrec(failure_tree, start, &failrec); |
4a54c8c1 JS |
2188 | if (ret) { |
2189 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2190 | if (!failrec) | |
2191 | return -ENOMEM; | |
2fe6303e | 2192 | |
4a54c8c1 JS |
2193 | failrec->start = start; |
2194 | failrec->len = end - start + 1; | |
2195 | failrec->this_mirror = 0; | |
2196 | failrec->bio_flags = 0; | |
2197 | failrec->in_validation = 0; | |
2198 | ||
2199 | read_lock(&em_tree->lock); | |
2200 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2201 | if (!em) { | |
2202 | read_unlock(&em_tree->lock); | |
2203 | kfree(failrec); | |
2204 | return -EIO; | |
2205 | } | |
2206 | ||
68ba990f | 2207 | if (em->start > start || em->start + em->len <= start) { |
4a54c8c1 JS |
2208 | free_extent_map(em); |
2209 | em = NULL; | |
2210 | } | |
2211 | read_unlock(&em_tree->lock); | |
7a2d6a64 | 2212 | if (!em) { |
4a54c8c1 JS |
2213 | kfree(failrec); |
2214 | return -EIO; | |
2215 | } | |
2fe6303e | 2216 | |
4a54c8c1 JS |
2217 | logical = start - em->start; |
2218 | logical = em->block_start + logical; | |
2219 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2220 | logical = em->block_start; | |
2221 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2222 | extent_set_compress_type(&failrec->bio_flags, | |
2223 | em->compress_type); | |
2224 | } | |
2fe6303e | 2225 | |
ab8d0fc4 JM |
2226 | btrfs_debug(fs_info, |
2227 | "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu", | |
2228 | logical, start, failrec->len); | |
2fe6303e | 2229 | |
4a54c8c1 JS |
2230 | failrec->logical = logical; |
2231 | free_extent_map(em); | |
2232 | ||
2233 | /* set the bits in the private failure tree */ | |
2234 | ret = set_extent_bits(failure_tree, start, end, | |
ceeb0ae7 | 2235 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 | 2236 | if (ret >= 0) |
47dc196a | 2237 | ret = set_state_failrec(failure_tree, start, failrec); |
4a54c8c1 JS |
2238 | /* set the bits in the inode's tree */ |
2239 | if (ret >= 0) | |
ceeb0ae7 | 2240 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED); |
4a54c8c1 JS |
2241 | if (ret < 0) { |
2242 | kfree(failrec); | |
2243 | return ret; | |
2244 | } | |
2245 | } else { | |
ab8d0fc4 JM |
2246 | btrfs_debug(fs_info, |
2247 | "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d", | |
2248 | failrec->logical, failrec->start, failrec->len, | |
2249 | failrec->in_validation); | |
4a54c8c1 JS |
2250 | /* |
2251 | * when data can be on disk more than twice, add to failrec here | |
2252 | * (e.g. with a list for failed_mirror) to make | |
2253 | * clean_io_failure() clean all those errors at once. | |
2254 | */ | |
2255 | } | |
2fe6303e MX |
2256 | |
2257 | *failrec_ret = failrec; | |
2258 | ||
2259 | return 0; | |
2260 | } | |
2261 | ||
a0b60d72 | 2262 | bool btrfs_check_repairable(struct inode *inode, unsigned failed_bio_pages, |
2fe6303e MX |
2263 | struct io_failure_record *failrec, int failed_mirror) |
2264 | { | |
ab8d0fc4 | 2265 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e MX |
2266 | int num_copies; |
2267 | ||
ab8d0fc4 | 2268 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
4a54c8c1 JS |
2269 | if (num_copies == 1) { |
2270 | /* | |
2271 | * we only have a single copy of the data, so don't bother with | |
2272 | * all the retry and error correction code that follows. no | |
2273 | * matter what the error is, it is very likely to persist. | |
2274 | */ | |
ab8d0fc4 JM |
2275 | btrfs_debug(fs_info, |
2276 | "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
2277 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2278 | return false; |
4a54c8c1 JS |
2279 | } |
2280 | ||
4a54c8c1 JS |
2281 | /* |
2282 | * there are two premises: | |
2283 | * a) deliver good data to the caller | |
2284 | * b) correct the bad sectors on disk | |
2285 | */ | |
a0b60d72 | 2286 | if (failed_bio_pages > 1) { |
4a54c8c1 JS |
2287 | /* |
2288 | * to fulfill b), we need to know the exact failing sectors, as | |
2289 | * we don't want to rewrite any more than the failed ones. thus, | |
2290 | * we need separate read requests for the failed bio | |
2291 | * | |
2292 | * if the following BUG_ON triggers, our validation request got | |
2293 | * merged. we need separate requests for our algorithm to work. | |
2294 | */ | |
2295 | BUG_ON(failrec->in_validation); | |
2296 | failrec->in_validation = 1; | |
2297 | failrec->this_mirror = failed_mirror; | |
4a54c8c1 JS |
2298 | } else { |
2299 | /* | |
2300 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2301 | * of the failed sector and if we succeed, we have setup | |
2302 | * everything for repair_io_failure to do the rest for us. | |
2303 | */ | |
2304 | if (failrec->in_validation) { | |
2305 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2306 | failrec->in_validation = 0; | |
2307 | failrec->this_mirror = 0; | |
2308 | } | |
2309 | failrec->failed_mirror = failed_mirror; | |
2310 | failrec->this_mirror++; | |
2311 | if (failrec->this_mirror == failed_mirror) | |
2312 | failrec->this_mirror++; | |
4a54c8c1 JS |
2313 | } |
2314 | ||
facc8a22 | 2315 | if (failrec->this_mirror > num_copies) { |
ab8d0fc4 JM |
2316 | btrfs_debug(fs_info, |
2317 | "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
2318 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2319 | return false; |
4a54c8c1 JS |
2320 | } |
2321 | ||
c3cfb656 | 2322 | return true; |
2fe6303e MX |
2323 | } |
2324 | ||
2325 | ||
2326 | struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio, | |
2327 | struct io_failure_record *failrec, | |
2328 | struct page *page, int pg_offset, int icsum, | |
8b110e39 | 2329 | bio_end_io_t *endio_func, void *data) |
2fe6303e | 2330 | { |
0b246afa | 2331 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e MX |
2332 | struct bio *bio; |
2333 | struct btrfs_io_bio *btrfs_failed_bio; | |
2334 | struct btrfs_io_bio *btrfs_bio; | |
2335 | ||
c5e4c3d7 | 2336 | bio = btrfs_io_bio_alloc(1); |
2fe6303e | 2337 | bio->bi_end_io = endio_func; |
4f024f37 | 2338 | bio->bi_iter.bi_sector = failrec->logical >> 9; |
74d46992 | 2339 | bio_set_dev(bio, fs_info->fs_devices->latest_bdev); |
4f024f37 | 2340 | bio->bi_iter.bi_size = 0; |
8b110e39 | 2341 | bio->bi_private = data; |
4a54c8c1 | 2342 | |
facc8a22 MX |
2343 | btrfs_failed_bio = btrfs_io_bio(failed_bio); |
2344 | if (btrfs_failed_bio->csum) { | |
facc8a22 MX |
2345 | u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); |
2346 | ||
2347 | btrfs_bio = btrfs_io_bio(bio); | |
2348 | btrfs_bio->csum = btrfs_bio->csum_inline; | |
2fe6303e MX |
2349 | icsum *= csum_size; |
2350 | memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum, | |
facc8a22 MX |
2351 | csum_size); |
2352 | } | |
2353 | ||
2fe6303e MX |
2354 | bio_add_page(bio, page, failrec->len, pg_offset); |
2355 | ||
2356 | return bio; | |
2357 | } | |
2358 | ||
2359 | /* | |
2360 | * this is a generic handler for readpage errors (default | |
2361 | * readpage_io_failed_hook). if other copies exist, read those and write back | |
2362 | * good data to the failed position. does not investigate in remapping the | |
2363 | * failed extent elsewhere, hoping the device will be smart enough to do this as | |
2364 | * needed | |
2365 | */ | |
2366 | ||
2367 | static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset, | |
2368 | struct page *page, u64 start, u64 end, | |
2369 | int failed_mirror) | |
2370 | { | |
2371 | struct io_failure_record *failrec; | |
2372 | struct inode *inode = page->mapping->host; | |
2373 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
7870d082 | 2374 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2fe6303e | 2375 | struct bio *bio; |
70fd7614 | 2376 | int read_mode = 0; |
4e4cbee9 | 2377 | blk_status_t status; |
2fe6303e | 2378 | int ret; |
a0b60d72 | 2379 | unsigned failed_bio_pages = bio_pages_all(failed_bio); |
2fe6303e | 2380 | |
1f7ad75b | 2381 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
2fe6303e MX |
2382 | |
2383 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
2384 | if (ret) | |
2385 | return ret; | |
2386 | ||
a0b60d72 | 2387 | if (!btrfs_check_repairable(inode, failed_bio_pages, failrec, |
c3cfb656 | 2388 | failed_mirror)) { |
7870d082 | 2389 | free_io_failure(failure_tree, tree, failrec); |
2fe6303e MX |
2390 | return -EIO; |
2391 | } | |
2392 | ||
a0b60d72 | 2393 | if (failed_bio_pages > 1) |
70fd7614 | 2394 | read_mode |= REQ_FAILFAST_DEV; |
2fe6303e MX |
2395 | |
2396 | phy_offset >>= inode->i_sb->s_blocksize_bits; | |
2397 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2398 | start - page_offset(page), | |
8b110e39 MX |
2399 | (int)phy_offset, failed_bio->bi_end_io, |
2400 | NULL); | |
ebcc3263 | 2401 | bio->bi_opf = REQ_OP_READ | read_mode; |
4a54c8c1 | 2402 | |
ab8d0fc4 JM |
2403 | btrfs_debug(btrfs_sb(inode->i_sb), |
2404 | "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d", | |
2405 | read_mode, failrec->this_mirror, failrec->in_validation); | |
4a54c8c1 | 2406 | |
8c27cb35 | 2407 | status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror, |
013bd4c3 | 2408 | failrec->bio_flags, 0); |
4e4cbee9 | 2409 | if (status) { |
7870d082 | 2410 | free_io_failure(failure_tree, tree, failrec); |
6c387ab2 | 2411 | bio_put(bio); |
4e4cbee9 | 2412 | ret = blk_status_to_errno(status); |
6c387ab2 MX |
2413 | } |
2414 | ||
013bd4c3 | 2415 | return ret; |
4a54c8c1 JS |
2416 | } |
2417 | ||
d1310b2e CM |
2418 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2419 | ||
b5227c07 | 2420 | void end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
87826df0 JM |
2421 | { |
2422 | int uptodate = (err == 0); | |
2423 | struct extent_io_tree *tree; | |
3e2426bd | 2424 | int ret = 0; |
87826df0 JM |
2425 | |
2426 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
2427 | ||
c3988d63 DS |
2428 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2429 | tree->ops->writepage_end_io_hook(page, start, end, NULL, | |
2430 | uptodate); | |
87826df0 | 2431 | |
87826df0 | 2432 | if (!uptodate) { |
87826df0 JM |
2433 | ClearPageUptodate(page); |
2434 | SetPageError(page); | |
bff5baf8 | 2435 | ret = err < 0 ? err : -EIO; |
5dca6eea | 2436 | mapping_set_error(page->mapping, ret); |
87826df0 | 2437 | } |
87826df0 JM |
2438 | } |
2439 | ||
d1310b2e CM |
2440 | /* |
2441 | * after a writepage IO is done, we need to: | |
2442 | * clear the uptodate bits on error | |
2443 | * clear the writeback bits in the extent tree for this IO | |
2444 | * end_page_writeback if the page has no more pending IO | |
2445 | * | |
2446 | * Scheduling is not allowed, so the extent state tree is expected | |
2447 | * to have one and only one object corresponding to this IO. | |
2448 | */ | |
4246a0b6 | 2449 | static void end_bio_extent_writepage(struct bio *bio) |
d1310b2e | 2450 | { |
4e4cbee9 | 2451 | int error = blk_status_to_errno(bio->bi_status); |
2c30c71b | 2452 | struct bio_vec *bvec; |
d1310b2e CM |
2453 | u64 start; |
2454 | u64 end; | |
2c30c71b | 2455 | int i; |
d1310b2e | 2456 | |
c09abff8 | 2457 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2c30c71b | 2458 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2459 | struct page *page = bvec->bv_page; |
0b246afa JM |
2460 | struct inode *inode = page->mapping->host; |
2461 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
902b22f3 | 2462 | |
17a5adcc AO |
2463 | /* We always issue full-page reads, but if some block |
2464 | * in a page fails to read, blk_update_request() will | |
2465 | * advance bv_offset and adjust bv_len to compensate. | |
2466 | * Print a warning for nonzero offsets, and an error | |
2467 | * if they don't add up to a full page. */ | |
09cbfeaf KS |
2468 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { |
2469 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
0b246afa | 2470 | btrfs_err(fs_info, |
efe120a0 FH |
2471 | "partial page write in btrfs with offset %u and length %u", |
2472 | bvec->bv_offset, bvec->bv_len); | |
2473 | else | |
0b246afa | 2474 | btrfs_info(fs_info, |
5d163e0e | 2475 | "incomplete page write in btrfs with offset %u and length %u", |
efe120a0 FH |
2476 | bvec->bv_offset, bvec->bv_len); |
2477 | } | |
d1310b2e | 2478 | |
17a5adcc AO |
2479 | start = page_offset(page); |
2480 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
d1310b2e | 2481 | |
4e4cbee9 | 2482 | end_extent_writepage(page, error, start, end); |
17a5adcc | 2483 | end_page_writeback(page); |
2c30c71b | 2484 | } |
2b1f55b0 | 2485 | |
d1310b2e | 2486 | bio_put(bio); |
d1310b2e CM |
2487 | } |
2488 | ||
883d0de4 MX |
2489 | static void |
2490 | endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len, | |
2491 | int uptodate) | |
2492 | { | |
2493 | struct extent_state *cached = NULL; | |
2494 | u64 end = start + len - 1; | |
2495 | ||
2496 | if (uptodate && tree->track_uptodate) | |
2497 | set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC); | |
d810a4be | 2498 | unlock_extent_cached_atomic(tree, start, end, &cached); |
883d0de4 MX |
2499 | } |
2500 | ||
d1310b2e CM |
2501 | /* |
2502 | * after a readpage IO is done, we need to: | |
2503 | * clear the uptodate bits on error | |
2504 | * set the uptodate bits if things worked | |
2505 | * set the page up to date if all extents in the tree are uptodate | |
2506 | * clear the lock bit in the extent tree | |
2507 | * unlock the page if there are no other extents locked for it | |
2508 | * | |
2509 | * Scheduling is not allowed, so the extent state tree is expected | |
2510 | * to have one and only one object corresponding to this IO. | |
2511 | */ | |
4246a0b6 | 2512 | static void end_bio_extent_readpage(struct bio *bio) |
d1310b2e | 2513 | { |
2c30c71b | 2514 | struct bio_vec *bvec; |
4e4cbee9 | 2515 | int uptodate = !bio->bi_status; |
facc8a22 | 2516 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); |
7870d082 | 2517 | struct extent_io_tree *tree, *failure_tree; |
facc8a22 | 2518 | u64 offset = 0; |
d1310b2e CM |
2519 | u64 start; |
2520 | u64 end; | |
facc8a22 | 2521 | u64 len; |
883d0de4 MX |
2522 | u64 extent_start = 0; |
2523 | u64 extent_len = 0; | |
5cf1ab56 | 2524 | int mirror; |
d1310b2e | 2525 | int ret; |
2c30c71b | 2526 | int i; |
d1310b2e | 2527 | |
c09abff8 | 2528 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2c30c71b | 2529 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2530 | struct page *page = bvec->bv_page; |
a71754fc | 2531 | struct inode *inode = page->mapping->host; |
ab8d0fc4 | 2532 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
507903b8 | 2533 | |
ab8d0fc4 JM |
2534 | btrfs_debug(fs_info, |
2535 | "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u", | |
4e4cbee9 | 2536 | (u64)bio->bi_iter.bi_sector, bio->bi_status, |
ab8d0fc4 | 2537 | io_bio->mirror_num); |
a71754fc | 2538 | tree = &BTRFS_I(inode)->io_tree; |
7870d082 | 2539 | failure_tree = &BTRFS_I(inode)->io_failure_tree; |
902b22f3 | 2540 | |
17a5adcc AO |
2541 | /* We always issue full-page reads, but if some block |
2542 | * in a page fails to read, blk_update_request() will | |
2543 | * advance bv_offset and adjust bv_len to compensate. | |
2544 | * Print a warning for nonzero offsets, and an error | |
2545 | * if they don't add up to a full page. */ | |
09cbfeaf KS |
2546 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { |
2547 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
ab8d0fc4 JM |
2548 | btrfs_err(fs_info, |
2549 | "partial page read in btrfs with offset %u and length %u", | |
efe120a0 FH |
2550 | bvec->bv_offset, bvec->bv_len); |
2551 | else | |
ab8d0fc4 JM |
2552 | btrfs_info(fs_info, |
2553 | "incomplete page read in btrfs with offset %u and length %u", | |
efe120a0 FH |
2554 | bvec->bv_offset, bvec->bv_len); |
2555 | } | |
d1310b2e | 2556 | |
17a5adcc AO |
2557 | start = page_offset(page); |
2558 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
facc8a22 | 2559 | len = bvec->bv_len; |
d1310b2e | 2560 | |
9be3395b | 2561 | mirror = io_bio->mirror_num; |
20c9801d | 2562 | if (likely(uptodate && tree->ops)) { |
facc8a22 MX |
2563 | ret = tree->ops->readpage_end_io_hook(io_bio, offset, |
2564 | page, start, end, | |
2565 | mirror); | |
5ee0844d | 2566 | if (ret) |
d1310b2e | 2567 | uptodate = 0; |
5ee0844d | 2568 | else |
7870d082 JB |
2569 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
2570 | failure_tree, tree, start, | |
2571 | page, | |
2572 | btrfs_ino(BTRFS_I(inode)), 0); | |
d1310b2e | 2573 | } |
ea466794 | 2574 | |
f2a09da9 MX |
2575 | if (likely(uptodate)) |
2576 | goto readpage_ok; | |
2577 | ||
20a7db8a | 2578 | if (tree->ops) { |
5cf1ab56 | 2579 | ret = tree->ops->readpage_io_failed_hook(page, mirror); |
9d0d1c8b LB |
2580 | if (ret == -EAGAIN) { |
2581 | /* | |
2582 | * Data inode's readpage_io_failed_hook() always | |
2583 | * returns -EAGAIN. | |
2584 | * | |
2585 | * The generic bio_readpage_error handles errors | |
2586 | * the following way: If possible, new read | |
2587 | * requests are created and submitted and will | |
2588 | * end up in end_bio_extent_readpage as well (if | |
2589 | * we're lucky, not in the !uptodate case). In | |
2590 | * that case it returns 0 and we just go on with | |
2591 | * the next page in our bio. If it can't handle | |
2592 | * the error it will return -EIO and we remain | |
2593 | * responsible for that page. | |
2594 | */ | |
2595 | ret = bio_readpage_error(bio, offset, page, | |
2596 | start, end, mirror); | |
2597 | if (ret == 0) { | |
4e4cbee9 | 2598 | uptodate = !bio->bi_status; |
9d0d1c8b LB |
2599 | offset += len; |
2600 | continue; | |
2601 | } | |
2602 | } | |
2603 | ||
f4a8e656 | 2604 | /* |
9d0d1c8b LB |
2605 | * metadata's readpage_io_failed_hook() always returns |
2606 | * -EIO and fixes nothing. -EIO is also returned if | |
2607 | * data inode error could not be fixed. | |
f4a8e656 | 2608 | */ |
9d0d1c8b | 2609 | ASSERT(ret == -EIO); |
7e38326f | 2610 | } |
f2a09da9 | 2611 | readpage_ok: |
883d0de4 | 2612 | if (likely(uptodate)) { |
a71754fc | 2613 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 2614 | pgoff_t end_index = i_size >> PAGE_SHIFT; |
a583c026 | 2615 | unsigned off; |
a71754fc JB |
2616 | |
2617 | /* Zero out the end if this page straddles i_size */ | |
09cbfeaf | 2618 | off = i_size & (PAGE_SIZE-1); |
a583c026 | 2619 | if (page->index == end_index && off) |
09cbfeaf | 2620 | zero_user_segment(page, off, PAGE_SIZE); |
17a5adcc | 2621 | SetPageUptodate(page); |
70dec807 | 2622 | } else { |
17a5adcc AO |
2623 | ClearPageUptodate(page); |
2624 | SetPageError(page); | |
70dec807 | 2625 | } |
17a5adcc | 2626 | unlock_page(page); |
facc8a22 | 2627 | offset += len; |
883d0de4 MX |
2628 | |
2629 | if (unlikely(!uptodate)) { | |
2630 | if (extent_len) { | |
2631 | endio_readpage_release_extent(tree, | |
2632 | extent_start, | |
2633 | extent_len, 1); | |
2634 | extent_start = 0; | |
2635 | extent_len = 0; | |
2636 | } | |
2637 | endio_readpage_release_extent(tree, start, | |
2638 | end - start + 1, 0); | |
2639 | } else if (!extent_len) { | |
2640 | extent_start = start; | |
2641 | extent_len = end + 1 - start; | |
2642 | } else if (extent_start + extent_len == start) { | |
2643 | extent_len += end + 1 - start; | |
2644 | } else { | |
2645 | endio_readpage_release_extent(tree, extent_start, | |
2646 | extent_len, uptodate); | |
2647 | extent_start = start; | |
2648 | extent_len = end + 1 - start; | |
2649 | } | |
2c30c71b | 2650 | } |
d1310b2e | 2651 | |
883d0de4 MX |
2652 | if (extent_len) |
2653 | endio_readpage_release_extent(tree, extent_start, extent_len, | |
2654 | uptodate); | |
facc8a22 | 2655 | if (io_bio->end_io) |
4e4cbee9 | 2656 | io_bio->end_io(io_bio, blk_status_to_errno(bio->bi_status)); |
d1310b2e | 2657 | bio_put(bio); |
d1310b2e CM |
2658 | } |
2659 | ||
9be3395b | 2660 | /* |
184f999e DS |
2661 | * Initialize the members up to but not including 'bio'. Use after allocating a |
2662 | * new bio by bio_alloc_bioset as it does not initialize the bytes outside of | |
2663 | * 'bio' because use of __GFP_ZERO is not supported. | |
9be3395b | 2664 | */ |
184f999e | 2665 | static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio) |
d1310b2e | 2666 | { |
184f999e DS |
2667 | memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio)); |
2668 | } | |
d1310b2e | 2669 | |
9be3395b | 2670 | /* |
6e707bcd DS |
2671 | * The following helpers allocate a bio. As it's backed by a bioset, it'll |
2672 | * never fail. We're returning a bio right now but you can call btrfs_io_bio | |
2673 | * for the appropriate container_of magic | |
9be3395b | 2674 | */ |
c821e7f3 | 2675 | struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte) |
d1310b2e CM |
2676 | { |
2677 | struct bio *bio; | |
d1310b2e | 2678 | |
8ac9f7c1 | 2679 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &btrfs_bioset); |
74d46992 | 2680 | bio_set_dev(bio, bdev); |
c821e7f3 | 2681 | bio->bi_iter.bi_sector = first_byte >> 9; |
184f999e | 2682 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
d1310b2e CM |
2683 | return bio; |
2684 | } | |
2685 | ||
8b6c1d56 | 2686 | struct bio *btrfs_bio_clone(struct bio *bio) |
9be3395b | 2687 | { |
23ea8e5a MX |
2688 | struct btrfs_io_bio *btrfs_bio; |
2689 | struct bio *new; | |
9be3395b | 2690 | |
6e707bcd | 2691 | /* Bio allocation backed by a bioset does not fail */ |
8ac9f7c1 | 2692 | new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset); |
6e707bcd | 2693 | btrfs_bio = btrfs_io_bio(new); |
184f999e | 2694 | btrfs_io_bio_init(btrfs_bio); |
6e707bcd | 2695 | btrfs_bio->iter = bio->bi_iter; |
23ea8e5a MX |
2696 | return new; |
2697 | } | |
9be3395b | 2698 | |
c5e4c3d7 | 2699 | struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs) |
9be3395b | 2700 | { |
facc8a22 MX |
2701 | struct bio *bio; |
2702 | ||
6e707bcd | 2703 | /* Bio allocation backed by a bioset does not fail */ |
8ac9f7c1 | 2704 | bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset); |
184f999e | 2705 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
facc8a22 | 2706 | return bio; |
9be3395b CM |
2707 | } |
2708 | ||
e477094f | 2709 | struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size) |
2f8e9140 LB |
2710 | { |
2711 | struct bio *bio; | |
2712 | struct btrfs_io_bio *btrfs_bio; | |
2713 | ||
2714 | /* this will never fail when it's backed by a bioset */ | |
8ac9f7c1 | 2715 | bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset); |
2f8e9140 LB |
2716 | ASSERT(bio); |
2717 | ||
2718 | btrfs_bio = btrfs_io_bio(bio); | |
184f999e | 2719 | btrfs_io_bio_init(btrfs_bio); |
2f8e9140 LB |
2720 | |
2721 | bio_trim(bio, offset >> 9, size >> 9); | |
17347cec | 2722 | btrfs_bio->iter = bio->bi_iter; |
2f8e9140 LB |
2723 | return bio; |
2724 | } | |
9be3395b | 2725 | |
1f7ad75b MC |
2726 | static int __must_check submit_one_bio(struct bio *bio, int mirror_num, |
2727 | unsigned long bio_flags) | |
d1310b2e | 2728 | { |
4e4cbee9 | 2729 | blk_status_t ret = 0; |
c45a8f2d | 2730 | struct bio_vec *bvec = bio_last_bvec_all(bio); |
70dec807 CM |
2731 | struct page *page = bvec->bv_page; |
2732 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 2733 | u64 start; |
70dec807 | 2734 | |
4eee4fa4 | 2735 | start = page_offset(page) + bvec->bv_offset; |
70dec807 | 2736 | |
902b22f3 | 2737 | bio->bi_private = NULL; |
d1310b2e | 2738 | |
20c9801d | 2739 | if (tree->ops) |
c6100a4b | 2740 | ret = tree->ops->submit_bio_hook(tree->private_data, bio, |
eaf25d93 | 2741 | mirror_num, bio_flags, start); |
0b86a832 | 2742 | else |
4e49ea4a | 2743 | btrfsic_submit_bio(bio); |
4a54c8c1 | 2744 | |
4e4cbee9 | 2745 | return blk_status_to_errno(ret); |
d1310b2e CM |
2746 | } |
2747 | ||
4b81ba48 DS |
2748 | /* |
2749 | * @opf: bio REQ_OP_* and REQ_* flags as one value | |
b8b3d625 DS |
2750 | * @tree: tree so we can call our merge_bio hook |
2751 | * @wbc: optional writeback control for io accounting | |
2752 | * @page: page to add to the bio | |
2753 | * @pg_offset: offset of the new bio or to check whether we are adding | |
2754 | * a contiguous page to the previous one | |
2755 | * @size: portion of page that we want to write | |
2756 | * @offset: starting offset in the page | |
2757 | * @bdev: attach newly created bios to this bdev | |
5c2b1fd7 | 2758 | * @bio_ret: must be valid pointer, newly allocated bio will be stored there |
b8b3d625 DS |
2759 | * @end_io_func: end_io callback for new bio |
2760 | * @mirror_num: desired mirror to read/write | |
2761 | * @prev_bio_flags: flags of previous bio to see if we can merge the current one | |
2762 | * @bio_flags: flags of the current bio to see if we can merge them | |
4b81ba48 DS |
2763 | */ |
2764 | static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree, | |
da2f0f74 | 2765 | struct writeback_control *wbc, |
6273b7f8 | 2766 | struct page *page, u64 offset, |
6c5a4e2c | 2767 | size_t size, unsigned long pg_offset, |
d1310b2e CM |
2768 | struct block_device *bdev, |
2769 | struct bio **bio_ret, | |
f188591e | 2770 | bio_end_io_t end_io_func, |
c8b97818 CM |
2771 | int mirror_num, |
2772 | unsigned long prev_bio_flags, | |
005efedf FM |
2773 | unsigned long bio_flags, |
2774 | bool force_bio_submit) | |
d1310b2e CM |
2775 | { |
2776 | int ret = 0; | |
2777 | struct bio *bio; | |
09cbfeaf | 2778 | size_t page_size = min_t(size_t, size, PAGE_SIZE); |
6273b7f8 | 2779 | sector_t sector = offset >> 9; |
d1310b2e | 2780 | |
5c2b1fd7 DS |
2781 | ASSERT(bio_ret); |
2782 | ||
2783 | if (*bio_ret) { | |
0c8508a6 DS |
2784 | bool contig; |
2785 | bool can_merge = true; | |
2786 | ||
d1310b2e | 2787 | bio = *bio_ret; |
0c8508a6 | 2788 | if (prev_bio_flags & EXTENT_BIO_COMPRESSED) |
4f024f37 | 2789 | contig = bio->bi_iter.bi_sector == sector; |
c8b97818 | 2790 | else |
f73a1c7d | 2791 | contig = bio_end_sector(bio) == sector; |
c8b97818 | 2792 | |
0c8508a6 DS |
2793 | if (tree->ops && tree->ops->merge_bio_hook(page, offset, |
2794 | page_size, bio, bio_flags)) | |
2795 | can_merge = false; | |
2796 | ||
2797 | if (prev_bio_flags != bio_flags || !contig || !can_merge || | |
005efedf | 2798 | force_bio_submit || |
6c5a4e2c | 2799 | bio_add_page(bio, page, page_size, pg_offset) < page_size) { |
1f7ad75b | 2800 | ret = submit_one_bio(bio, mirror_num, prev_bio_flags); |
289454ad NA |
2801 | if (ret < 0) { |
2802 | *bio_ret = NULL; | |
79787eaa | 2803 | return ret; |
289454ad | 2804 | } |
d1310b2e CM |
2805 | bio = NULL; |
2806 | } else { | |
da2f0f74 CM |
2807 | if (wbc) |
2808 | wbc_account_io(wbc, page, page_size); | |
d1310b2e CM |
2809 | return 0; |
2810 | } | |
2811 | } | |
c8b97818 | 2812 | |
6273b7f8 | 2813 | bio = btrfs_bio_alloc(bdev, offset); |
6c5a4e2c | 2814 | bio_add_page(bio, page, page_size, pg_offset); |
d1310b2e CM |
2815 | bio->bi_end_io = end_io_func; |
2816 | bio->bi_private = tree; | |
e6959b93 | 2817 | bio->bi_write_hint = page->mapping->host->i_write_hint; |
4b81ba48 | 2818 | bio->bi_opf = opf; |
da2f0f74 CM |
2819 | if (wbc) { |
2820 | wbc_init_bio(wbc, bio); | |
2821 | wbc_account_io(wbc, page, page_size); | |
2822 | } | |
70dec807 | 2823 | |
5c2b1fd7 | 2824 | *bio_ret = bio; |
d1310b2e CM |
2825 | |
2826 | return ret; | |
2827 | } | |
2828 | ||
48a3b636 ES |
2829 | static void attach_extent_buffer_page(struct extent_buffer *eb, |
2830 | struct page *page) | |
d1310b2e CM |
2831 | { |
2832 | if (!PagePrivate(page)) { | |
2833 | SetPagePrivate(page); | |
09cbfeaf | 2834 | get_page(page); |
4f2de97a JB |
2835 | set_page_private(page, (unsigned long)eb); |
2836 | } else { | |
2837 | WARN_ON(page->private != (unsigned long)eb); | |
d1310b2e CM |
2838 | } |
2839 | } | |
2840 | ||
4f2de97a | 2841 | void set_page_extent_mapped(struct page *page) |
d1310b2e | 2842 | { |
4f2de97a JB |
2843 | if (!PagePrivate(page)) { |
2844 | SetPagePrivate(page); | |
09cbfeaf | 2845 | get_page(page); |
4f2de97a JB |
2846 | set_page_private(page, EXTENT_PAGE_PRIVATE); |
2847 | } | |
d1310b2e CM |
2848 | } |
2849 | ||
125bac01 MX |
2850 | static struct extent_map * |
2851 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
2852 | u64 start, u64 len, get_extent_t *get_extent, | |
2853 | struct extent_map **em_cached) | |
2854 | { | |
2855 | struct extent_map *em; | |
2856 | ||
2857 | if (em_cached && *em_cached) { | |
2858 | em = *em_cached; | |
cbc0e928 | 2859 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 | 2860 | start < extent_map_end(em)) { |
490b54d6 | 2861 | refcount_inc(&em->refs); |
125bac01 MX |
2862 | return em; |
2863 | } | |
2864 | ||
2865 | free_extent_map(em); | |
2866 | *em_cached = NULL; | |
2867 | } | |
2868 | ||
fc4f21b1 | 2869 | em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0); |
125bac01 MX |
2870 | if (em_cached && !IS_ERR_OR_NULL(em)) { |
2871 | BUG_ON(*em_cached); | |
490b54d6 | 2872 | refcount_inc(&em->refs); |
125bac01 MX |
2873 | *em_cached = em; |
2874 | } | |
2875 | return em; | |
2876 | } | |
d1310b2e CM |
2877 | /* |
2878 | * basic readpage implementation. Locked extent state structs are inserted | |
2879 | * into the tree that are removed when the IO is done (by the end_io | |
2880 | * handlers) | |
79787eaa | 2881 | * XXX JDM: This needs looking at to ensure proper page locking |
baf863b9 | 2882 | * return 0 on success, otherwise return error |
d1310b2e | 2883 | */ |
9974090b MX |
2884 | static int __do_readpage(struct extent_io_tree *tree, |
2885 | struct page *page, | |
2886 | get_extent_t *get_extent, | |
125bac01 | 2887 | struct extent_map **em_cached, |
9974090b | 2888 | struct bio **bio, int mirror_num, |
f1c77c55 | 2889 | unsigned long *bio_flags, unsigned int read_flags, |
005efedf | 2890 | u64 *prev_em_start) |
d1310b2e CM |
2891 | { |
2892 | struct inode *inode = page->mapping->host; | |
4eee4fa4 | 2893 | u64 start = page_offset(page); |
8eec8296 | 2894 | const u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
2895 | u64 cur = start; |
2896 | u64 extent_offset; | |
2897 | u64 last_byte = i_size_read(inode); | |
2898 | u64 block_start; | |
2899 | u64 cur_end; | |
d1310b2e CM |
2900 | struct extent_map *em; |
2901 | struct block_device *bdev; | |
baf863b9 | 2902 | int ret = 0; |
d1310b2e | 2903 | int nr = 0; |
306e16ce | 2904 | size_t pg_offset = 0; |
d1310b2e | 2905 | size_t iosize; |
c8b97818 | 2906 | size_t disk_io_size; |
d1310b2e | 2907 | size_t blocksize = inode->i_sb->s_blocksize; |
7f042a83 | 2908 | unsigned long this_bio_flag = 0; |
d1310b2e CM |
2909 | |
2910 | set_page_extent_mapped(page); | |
2911 | ||
90a887c9 DM |
2912 | if (!PageUptodate(page)) { |
2913 | if (cleancache_get_page(page) == 0) { | |
2914 | BUG_ON(blocksize != PAGE_SIZE); | |
9974090b | 2915 | unlock_extent(tree, start, end); |
90a887c9 DM |
2916 | goto out; |
2917 | } | |
2918 | } | |
2919 | ||
09cbfeaf | 2920 | if (page->index == last_byte >> PAGE_SHIFT) { |
c8b97818 | 2921 | char *userpage; |
09cbfeaf | 2922 | size_t zero_offset = last_byte & (PAGE_SIZE - 1); |
c8b97818 CM |
2923 | |
2924 | if (zero_offset) { | |
09cbfeaf | 2925 | iosize = PAGE_SIZE - zero_offset; |
7ac687d9 | 2926 | userpage = kmap_atomic(page); |
c8b97818 CM |
2927 | memset(userpage + zero_offset, 0, iosize); |
2928 | flush_dcache_page(page); | |
7ac687d9 | 2929 | kunmap_atomic(userpage); |
c8b97818 CM |
2930 | } |
2931 | } | |
d1310b2e | 2932 | while (cur <= end) { |
005efedf | 2933 | bool force_bio_submit = false; |
6273b7f8 | 2934 | u64 offset; |
c8f2f24b | 2935 | |
d1310b2e CM |
2936 | if (cur >= last_byte) { |
2937 | char *userpage; | |
507903b8 AJ |
2938 | struct extent_state *cached = NULL; |
2939 | ||
09cbfeaf | 2940 | iosize = PAGE_SIZE - pg_offset; |
7ac687d9 | 2941 | userpage = kmap_atomic(page); |
306e16ce | 2942 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2943 | flush_dcache_page(page); |
7ac687d9 | 2944 | kunmap_atomic(userpage); |
d1310b2e | 2945 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 2946 | &cached, GFP_NOFS); |
7f042a83 | 2947 | unlock_extent_cached(tree, cur, |
e43bbe5e | 2948 | cur + iosize - 1, &cached); |
d1310b2e CM |
2949 | break; |
2950 | } | |
125bac01 MX |
2951 | em = __get_extent_map(inode, page, pg_offset, cur, |
2952 | end - cur + 1, get_extent, em_cached); | |
c704005d | 2953 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 2954 | SetPageError(page); |
7f042a83 | 2955 | unlock_extent(tree, cur, end); |
d1310b2e CM |
2956 | break; |
2957 | } | |
d1310b2e CM |
2958 | extent_offset = cur - em->start; |
2959 | BUG_ON(extent_map_end(em) <= cur); | |
2960 | BUG_ON(end < cur); | |
2961 | ||
261507a0 | 2962 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
4b384318 | 2963 | this_bio_flag |= EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
2964 | extent_set_compress_type(&this_bio_flag, |
2965 | em->compress_type); | |
2966 | } | |
c8b97818 | 2967 | |
d1310b2e CM |
2968 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
2969 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 2970 | iosize = ALIGN(iosize, blocksize); |
c8b97818 CM |
2971 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
2972 | disk_io_size = em->block_len; | |
6273b7f8 | 2973 | offset = em->block_start; |
c8b97818 | 2974 | } else { |
6273b7f8 | 2975 | offset = em->block_start + extent_offset; |
c8b97818 CM |
2976 | disk_io_size = iosize; |
2977 | } | |
d1310b2e CM |
2978 | bdev = em->bdev; |
2979 | block_start = em->block_start; | |
d899e052 YZ |
2980 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2981 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
2982 | |
2983 | /* | |
2984 | * If we have a file range that points to a compressed extent | |
2985 | * and it's followed by a consecutive file range that points to | |
2986 | * to the same compressed extent (possibly with a different | |
2987 | * offset and/or length, so it either points to the whole extent | |
2988 | * or only part of it), we must make sure we do not submit a | |
2989 | * single bio to populate the pages for the 2 ranges because | |
2990 | * this makes the compressed extent read zero out the pages | |
2991 | * belonging to the 2nd range. Imagine the following scenario: | |
2992 | * | |
2993 | * File layout | |
2994 | * [0 - 8K] [8K - 24K] | |
2995 | * | | | |
2996 | * | | | |
2997 | * points to extent X, points to extent X, | |
2998 | * offset 4K, length of 8K offset 0, length 16K | |
2999 | * | |
3000 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
3001 | * | |
3002 | * If the bio to read the compressed extent covers both ranges, | |
3003 | * it will decompress extent X into the pages belonging to the | |
3004 | * first range and then it will stop, zeroing out the remaining | |
3005 | * pages that belong to the other range that points to extent X. | |
3006 | * So here we make sure we submit 2 bios, one for the first | |
3007 | * range and another one for the third range. Both will target | |
3008 | * the same physical extent from disk, but we can't currently | |
3009 | * make the compressed bio endio callback populate the pages | |
3010 | * for both ranges because each compressed bio is tightly | |
3011 | * coupled with a single extent map, and each range can have | |
3012 | * an extent map with a different offset value relative to the | |
3013 | * uncompressed data of our extent and different lengths. This | |
3014 | * is a corner case so we prioritize correctness over | |
3015 | * non-optimal behavior (submitting 2 bios for the same extent). | |
3016 | */ | |
3017 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
3018 | prev_em_start && *prev_em_start != (u64)-1 && | |
3019 | *prev_em_start != em->orig_start) | |
3020 | force_bio_submit = true; | |
3021 | ||
3022 | if (prev_em_start) | |
3023 | *prev_em_start = em->orig_start; | |
3024 | ||
d1310b2e CM |
3025 | free_extent_map(em); |
3026 | em = NULL; | |
3027 | ||
3028 | /* we've found a hole, just zero and go on */ | |
3029 | if (block_start == EXTENT_MAP_HOLE) { | |
3030 | char *userpage; | |
507903b8 AJ |
3031 | struct extent_state *cached = NULL; |
3032 | ||
7ac687d9 | 3033 | userpage = kmap_atomic(page); |
306e16ce | 3034 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3035 | flush_dcache_page(page); |
7ac687d9 | 3036 | kunmap_atomic(userpage); |
d1310b2e CM |
3037 | |
3038 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 | 3039 | &cached, GFP_NOFS); |
7f042a83 | 3040 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3041 | cur + iosize - 1, &cached); |
d1310b2e | 3042 | cur = cur + iosize; |
306e16ce | 3043 | pg_offset += iosize; |
d1310b2e CM |
3044 | continue; |
3045 | } | |
3046 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
3047 | if (test_range_bit(tree, cur, cur_end, |
3048 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 3049 | check_page_uptodate(tree, page); |
7f042a83 | 3050 | unlock_extent(tree, cur, cur + iosize - 1); |
d1310b2e | 3051 | cur = cur + iosize; |
306e16ce | 3052 | pg_offset += iosize; |
d1310b2e CM |
3053 | continue; |
3054 | } | |
70dec807 CM |
3055 | /* we have an inline extent but it didn't get marked up |
3056 | * to date. Error out | |
3057 | */ | |
3058 | if (block_start == EXTENT_MAP_INLINE) { | |
3059 | SetPageError(page); | |
7f042a83 | 3060 | unlock_extent(tree, cur, cur + iosize - 1); |
70dec807 | 3061 | cur = cur + iosize; |
306e16ce | 3062 | pg_offset += iosize; |
70dec807 CM |
3063 | continue; |
3064 | } | |
d1310b2e | 3065 | |
4b81ba48 | 3066 | ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL, |
6273b7f8 DS |
3067 | page, offset, disk_io_size, |
3068 | pg_offset, bdev, bio, | |
c8b97818 CM |
3069 | end_bio_extent_readpage, mirror_num, |
3070 | *bio_flags, | |
005efedf FM |
3071 | this_bio_flag, |
3072 | force_bio_submit); | |
c8f2f24b JB |
3073 | if (!ret) { |
3074 | nr++; | |
3075 | *bio_flags = this_bio_flag; | |
3076 | } else { | |
d1310b2e | 3077 | SetPageError(page); |
7f042a83 | 3078 | unlock_extent(tree, cur, cur + iosize - 1); |
baf863b9 | 3079 | goto out; |
edd33c99 | 3080 | } |
d1310b2e | 3081 | cur = cur + iosize; |
306e16ce | 3082 | pg_offset += iosize; |
d1310b2e | 3083 | } |
90a887c9 | 3084 | out: |
d1310b2e CM |
3085 | if (!nr) { |
3086 | if (!PageError(page)) | |
3087 | SetPageUptodate(page); | |
3088 | unlock_page(page); | |
3089 | } | |
baf863b9 | 3090 | return ret; |
d1310b2e CM |
3091 | } |
3092 | ||
9974090b MX |
3093 | static inline void __do_contiguous_readpages(struct extent_io_tree *tree, |
3094 | struct page *pages[], int nr_pages, | |
3095 | u64 start, u64 end, | |
125bac01 | 3096 | struct extent_map **em_cached, |
d3fac6ba | 3097 | struct bio **bio, |
1f7ad75b | 3098 | unsigned long *bio_flags, |
808f80b4 | 3099 | u64 *prev_em_start) |
9974090b MX |
3100 | { |
3101 | struct inode *inode; | |
3102 | struct btrfs_ordered_extent *ordered; | |
3103 | int index; | |
3104 | ||
3105 | inode = pages[0]->mapping->host; | |
3106 | while (1) { | |
3107 | lock_extent(tree, start, end); | |
a776c6fa | 3108 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
9974090b MX |
3109 | end - start + 1); |
3110 | if (!ordered) | |
3111 | break; | |
3112 | unlock_extent(tree, start, end); | |
3113 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3114 | btrfs_put_ordered_extent(ordered); | |
3115 | } | |
3116 | ||
3117 | for (index = 0; index < nr_pages; index++) { | |
4ef77695 DS |
3118 | __do_readpage(tree, pages[index], btrfs_get_extent, em_cached, |
3119 | bio, 0, bio_flags, 0, prev_em_start); | |
09cbfeaf | 3120 | put_page(pages[index]); |
9974090b MX |
3121 | } |
3122 | } | |
3123 | ||
3124 | static void __extent_readpages(struct extent_io_tree *tree, | |
3125 | struct page *pages[], | |
e4d17ef5 | 3126 | int nr_pages, |
125bac01 | 3127 | struct extent_map **em_cached, |
d3fac6ba | 3128 | struct bio **bio, unsigned long *bio_flags, |
808f80b4 | 3129 | u64 *prev_em_start) |
9974090b | 3130 | { |
35a3621b | 3131 | u64 start = 0; |
9974090b MX |
3132 | u64 end = 0; |
3133 | u64 page_start; | |
3134 | int index; | |
35a3621b | 3135 | int first_index = 0; |
9974090b MX |
3136 | |
3137 | for (index = 0; index < nr_pages; index++) { | |
3138 | page_start = page_offset(pages[index]); | |
3139 | if (!end) { | |
3140 | start = page_start; | |
09cbfeaf | 3141 | end = start + PAGE_SIZE - 1; |
9974090b MX |
3142 | first_index = index; |
3143 | } else if (end + 1 == page_start) { | |
09cbfeaf | 3144 | end += PAGE_SIZE; |
9974090b MX |
3145 | } else { |
3146 | __do_contiguous_readpages(tree, &pages[first_index], | |
3147 | index - first_index, start, | |
4ef77695 | 3148 | end, em_cached, |
d3fac6ba | 3149 | bio, bio_flags, |
1f7ad75b | 3150 | prev_em_start); |
9974090b | 3151 | start = page_start; |
09cbfeaf | 3152 | end = start + PAGE_SIZE - 1; |
9974090b MX |
3153 | first_index = index; |
3154 | } | |
3155 | } | |
3156 | ||
3157 | if (end) | |
3158 | __do_contiguous_readpages(tree, &pages[first_index], | |
3159 | index - first_index, start, | |
4ef77695 | 3160 | end, em_cached, bio, |
d3fac6ba | 3161 | bio_flags, prev_em_start); |
9974090b MX |
3162 | } |
3163 | ||
3164 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
3165 | struct page *page, | |
3166 | get_extent_t *get_extent, | |
3167 | struct bio **bio, int mirror_num, | |
f1c77c55 DS |
3168 | unsigned long *bio_flags, |
3169 | unsigned int read_flags) | |
9974090b MX |
3170 | { |
3171 | struct inode *inode = page->mapping->host; | |
3172 | struct btrfs_ordered_extent *ordered; | |
3173 | u64 start = page_offset(page); | |
09cbfeaf | 3174 | u64 end = start + PAGE_SIZE - 1; |
9974090b MX |
3175 | int ret; |
3176 | ||
3177 | while (1) { | |
3178 | lock_extent(tree, start, end); | |
a776c6fa | 3179 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
09cbfeaf | 3180 | PAGE_SIZE); |
9974090b MX |
3181 | if (!ordered) |
3182 | break; | |
3183 | unlock_extent(tree, start, end); | |
3184 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3185 | btrfs_put_ordered_extent(ordered); | |
3186 | } | |
3187 | ||
125bac01 | 3188 | ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num, |
1f7ad75b | 3189 | bio_flags, read_flags, NULL); |
9974090b MX |
3190 | return ret; |
3191 | } | |
3192 | ||
d1310b2e | 3193 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, |
8ddc7d9c | 3194 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3195 | { |
3196 | struct bio *bio = NULL; | |
c8b97818 | 3197 | unsigned long bio_flags = 0; |
d1310b2e CM |
3198 | int ret; |
3199 | ||
8ddc7d9c | 3200 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, |
1f7ad75b | 3201 | &bio_flags, 0); |
d1310b2e | 3202 | if (bio) |
1f7ad75b | 3203 | ret = submit_one_bio(bio, mirror_num, bio_flags); |
d1310b2e CM |
3204 | return ret; |
3205 | } | |
d1310b2e | 3206 | |
3d4b9496 | 3207 | static void update_nr_written(struct writeback_control *wbc, |
a9132667 | 3208 | unsigned long nr_written) |
11c8349b CM |
3209 | { |
3210 | wbc->nr_to_write -= nr_written; | |
11c8349b CM |
3211 | } |
3212 | ||
d1310b2e | 3213 | /* |
40f76580 CM |
3214 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
3215 | * | |
3216 | * This returns 1 if our fill_delalloc function did all the work required | |
3217 | * to write the page (copy into inline extent). In this case the IO has | |
3218 | * been started and the page is already unlocked. | |
3219 | * | |
3220 | * This returns 0 if all went well (page still locked) | |
3221 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 3222 | */ |
40f76580 CM |
3223 | static noinline_for_stack int writepage_delalloc(struct inode *inode, |
3224 | struct page *page, struct writeback_control *wbc, | |
3225 | struct extent_page_data *epd, | |
3226 | u64 delalloc_start, | |
3227 | unsigned long *nr_written) | |
3228 | { | |
3229 | struct extent_io_tree *tree = epd->tree; | |
09cbfeaf | 3230 | u64 page_end = delalloc_start + PAGE_SIZE - 1; |
40f76580 CM |
3231 | u64 nr_delalloc; |
3232 | u64 delalloc_to_write = 0; | |
3233 | u64 delalloc_end = 0; | |
3234 | int ret; | |
3235 | int page_started = 0; | |
3236 | ||
3237 | if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc) | |
3238 | return 0; | |
3239 | ||
3240 | while (delalloc_end < page_end) { | |
3241 | nr_delalloc = find_lock_delalloc_range(inode, tree, | |
3242 | page, | |
3243 | &delalloc_start, | |
3244 | &delalloc_end, | |
dcab6a3b | 3245 | BTRFS_MAX_EXTENT_SIZE); |
40f76580 CM |
3246 | if (nr_delalloc == 0) { |
3247 | delalloc_start = delalloc_end + 1; | |
3248 | continue; | |
3249 | } | |
3250 | ret = tree->ops->fill_delalloc(inode, page, | |
3251 | delalloc_start, | |
3252 | delalloc_end, | |
3253 | &page_started, | |
f82b7359 | 3254 | nr_written, wbc); |
40f76580 CM |
3255 | /* File system has been set read-only */ |
3256 | if (ret) { | |
3257 | SetPageError(page); | |
3258 | /* fill_delalloc should be return < 0 for error | |
3259 | * but just in case, we use > 0 here meaning the | |
3260 | * IO is started, so we don't want to return > 0 | |
3261 | * unless things are going well. | |
3262 | */ | |
3263 | ret = ret < 0 ? ret : -EIO; | |
3264 | goto done; | |
3265 | } | |
3266 | /* | |
ea1754a0 KS |
3267 | * delalloc_end is already one less than the total length, so |
3268 | * we don't subtract one from PAGE_SIZE | |
40f76580 CM |
3269 | */ |
3270 | delalloc_to_write += (delalloc_end - delalloc_start + | |
ea1754a0 | 3271 | PAGE_SIZE) >> PAGE_SHIFT; |
40f76580 CM |
3272 | delalloc_start = delalloc_end + 1; |
3273 | } | |
3274 | if (wbc->nr_to_write < delalloc_to_write) { | |
3275 | int thresh = 8192; | |
3276 | ||
3277 | if (delalloc_to_write < thresh * 2) | |
3278 | thresh = delalloc_to_write; | |
3279 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3280 | thresh); | |
3281 | } | |
3282 | ||
3283 | /* did the fill delalloc function already unlock and start | |
3284 | * the IO? | |
3285 | */ | |
3286 | if (page_started) { | |
3287 | /* | |
3288 | * we've unlocked the page, so we can't update | |
3289 | * the mapping's writeback index, just update | |
3290 | * nr_to_write. | |
3291 | */ | |
3292 | wbc->nr_to_write -= *nr_written; | |
3293 | return 1; | |
3294 | } | |
3295 | ||
3296 | ret = 0; | |
3297 | ||
3298 | done: | |
3299 | return ret; | |
3300 | } | |
3301 | ||
3302 | /* | |
3303 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3304 | * and does the loop to map the page into extents and bios. | |
3305 | * | |
3306 | * We return 1 if the IO is started and the page is unlocked, | |
3307 | * 0 if all went well (page still locked) | |
3308 | * < 0 if there were errors (page still locked) | |
3309 | */ | |
3310 | static noinline_for_stack int __extent_writepage_io(struct inode *inode, | |
3311 | struct page *page, | |
3312 | struct writeback_control *wbc, | |
3313 | struct extent_page_data *epd, | |
3314 | loff_t i_size, | |
3315 | unsigned long nr_written, | |
f1c77c55 | 3316 | unsigned int write_flags, int *nr_ret) |
d1310b2e | 3317 | { |
d1310b2e | 3318 | struct extent_io_tree *tree = epd->tree; |
4eee4fa4 | 3319 | u64 start = page_offset(page); |
09cbfeaf | 3320 | u64 page_end = start + PAGE_SIZE - 1; |
d1310b2e CM |
3321 | u64 end; |
3322 | u64 cur = start; | |
3323 | u64 extent_offset; | |
d1310b2e CM |
3324 | u64 block_start; |
3325 | u64 iosize; | |
d1310b2e CM |
3326 | struct extent_map *em; |
3327 | struct block_device *bdev; | |
7f3c74fb | 3328 | size_t pg_offset = 0; |
d1310b2e | 3329 | size_t blocksize; |
40f76580 CM |
3330 | int ret = 0; |
3331 | int nr = 0; | |
3332 | bool compressed; | |
c8b97818 | 3333 | |
247e743c | 3334 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
3335 | ret = tree->ops->writepage_start_hook(page, start, |
3336 | page_end); | |
87826df0 JM |
3337 | if (ret) { |
3338 | /* Fixup worker will requeue */ | |
3339 | if (ret == -EBUSY) | |
3340 | wbc->pages_skipped++; | |
3341 | else | |
3342 | redirty_page_for_writepage(wbc, page); | |
40f76580 | 3343 | |
3d4b9496 | 3344 | update_nr_written(wbc, nr_written); |
247e743c | 3345 | unlock_page(page); |
bcf93489 | 3346 | return 1; |
247e743c CM |
3347 | } |
3348 | } | |
3349 | ||
11c8349b CM |
3350 | /* |
3351 | * we don't want to touch the inode after unlocking the page, | |
3352 | * so we update the mapping writeback index now | |
3353 | */ | |
3d4b9496 | 3354 | update_nr_written(wbc, nr_written + 1); |
771ed689 | 3355 | |
d1310b2e | 3356 | end = page_end; |
40f76580 | 3357 | if (i_size <= start) { |
e6dcd2dc CM |
3358 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3359 | tree->ops->writepage_end_io_hook(page, start, | |
3360 | page_end, NULL, 1); | |
d1310b2e CM |
3361 | goto done; |
3362 | } | |
3363 | ||
d1310b2e CM |
3364 | blocksize = inode->i_sb->s_blocksize; |
3365 | ||
3366 | while (cur <= end) { | |
40f76580 | 3367 | u64 em_end; |
6273b7f8 | 3368 | u64 offset; |
58409edd | 3369 | |
40f76580 | 3370 | if (cur >= i_size) { |
e6dcd2dc CM |
3371 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3372 | tree->ops->writepage_end_io_hook(page, cur, | |
3373 | page_end, NULL, 1); | |
d1310b2e CM |
3374 | break; |
3375 | } | |
3c98c62f | 3376 | em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, cur, |
d1310b2e | 3377 | end - cur + 1, 1); |
c704005d | 3378 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3379 | SetPageError(page); |
61391d56 | 3380 | ret = PTR_ERR_OR_ZERO(em); |
d1310b2e CM |
3381 | break; |
3382 | } | |
3383 | ||
3384 | extent_offset = cur - em->start; | |
40f76580 CM |
3385 | em_end = extent_map_end(em); |
3386 | BUG_ON(em_end <= cur); | |
d1310b2e | 3387 | BUG_ON(end < cur); |
40f76580 | 3388 | iosize = min(em_end - cur, end - cur + 1); |
fda2832f | 3389 | iosize = ALIGN(iosize, blocksize); |
6273b7f8 | 3390 | offset = em->block_start + extent_offset; |
d1310b2e CM |
3391 | bdev = em->bdev; |
3392 | block_start = em->block_start; | |
c8b97818 | 3393 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
3394 | free_extent_map(em); |
3395 | em = NULL; | |
3396 | ||
c8b97818 CM |
3397 | /* |
3398 | * compressed and inline extents are written through other | |
3399 | * paths in the FS | |
3400 | */ | |
3401 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3402 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
3403 | /* |
3404 | * end_io notification does not happen here for | |
3405 | * compressed extents | |
3406 | */ | |
3407 | if (!compressed && tree->ops && | |
3408 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
3409 | tree->ops->writepage_end_io_hook(page, cur, |
3410 | cur + iosize - 1, | |
3411 | NULL, 1); | |
c8b97818 CM |
3412 | else if (compressed) { |
3413 | /* we don't want to end_page_writeback on | |
3414 | * a compressed extent. this happens | |
3415 | * elsewhere | |
3416 | */ | |
3417 | nr++; | |
3418 | } | |
3419 | ||
3420 | cur += iosize; | |
7f3c74fb | 3421 | pg_offset += iosize; |
d1310b2e CM |
3422 | continue; |
3423 | } | |
c8b97818 | 3424 | |
58409edd DS |
3425 | set_range_writeback(tree, cur, cur + iosize - 1); |
3426 | if (!PageWriteback(page)) { | |
3427 | btrfs_err(BTRFS_I(inode)->root->fs_info, | |
3428 | "page %lu not writeback, cur %llu end %llu", | |
3429 | page->index, cur, end); | |
d1310b2e | 3430 | } |
7f3c74fb | 3431 | |
4b81ba48 | 3432 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc, |
6273b7f8 | 3433 | page, offset, iosize, pg_offset, |
c2df8bb4 | 3434 | bdev, &epd->bio, |
58409edd DS |
3435 | end_bio_extent_writepage, |
3436 | 0, 0, 0, false); | |
fe01aa65 | 3437 | if (ret) { |
58409edd | 3438 | SetPageError(page); |
fe01aa65 TK |
3439 | if (PageWriteback(page)) |
3440 | end_page_writeback(page); | |
3441 | } | |
d1310b2e | 3442 | |
d1310b2e | 3443 | cur = cur + iosize; |
7f3c74fb | 3444 | pg_offset += iosize; |
d1310b2e CM |
3445 | nr++; |
3446 | } | |
40f76580 CM |
3447 | done: |
3448 | *nr_ret = nr; | |
40f76580 CM |
3449 | return ret; |
3450 | } | |
3451 | ||
3452 | /* | |
3453 | * the writepage semantics are similar to regular writepage. extent | |
3454 | * records are inserted to lock ranges in the tree, and as dirty areas | |
3455 | * are found, they are marked writeback. Then the lock bits are removed | |
3456 | * and the end_io handler clears the writeback ranges | |
3457 | */ | |
3458 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
aab6e9ed | 3459 | struct extent_page_data *epd) |
40f76580 CM |
3460 | { |
3461 | struct inode *inode = page->mapping->host; | |
40f76580 | 3462 | u64 start = page_offset(page); |
09cbfeaf | 3463 | u64 page_end = start + PAGE_SIZE - 1; |
40f76580 CM |
3464 | int ret; |
3465 | int nr = 0; | |
3466 | size_t pg_offset = 0; | |
3467 | loff_t i_size = i_size_read(inode); | |
09cbfeaf | 3468 | unsigned long end_index = i_size >> PAGE_SHIFT; |
f1c77c55 | 3469 | unsigned int write_flags = 0; |
40f76580 CM |
3470 | unsigned long nr_written = 0; |
3471 | ||
ff40adf7 | 3472 | write_flags = wbc_to_write_flags(wbc); |
40f76580 CM |
3473 | |
3474 | trace___extent_writepage(page, inode, wbc); | |
3475 | ||
3476 | WARN_ON(!PageLocked(page)); | |
3477 | ||
3478 | ClearPageError(page); | |
3479 | ||
09cbfeaf | 3480 | pg_offset = i_size & (PAGE_SIZE - 1); |
40f76580 CM |
3481 | if (page->index > end_index || |
3482 | (page->index == end_index && !pg_offset)) { | |
09cbfeaf | 3483 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); |
40f76580 CM |
3484 | unlock_page(page); |
3485 | return 0; | |
3486 | } | |
3487 | ||
3488 | if (page->index == end_index) { | |
3489 | char *userpage; | |
3490 | ||
3491 | userpage = kmap_atomic(page); | |
3492 | memset(userpage + pg_offset, 0, | |
09cbfeaf | 3493 | PAGE_SIZE - pg_offset); |
40f76580 CM |
3494 | kunmap_atomic(userpage); |
3495 | flush_dcache_page(page); | |
3496 | } | |
3497 | ||
3498 | pg_offset = 0; | |
3499 | ||
3500 | set_page_extent_mapped(page); | |
3501 | ||
3502 | ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written); | |
3503 | if (ret == 1) | |
3504 | goto done_unlocked; | |
3505 | if (ret) | |
3506 | goto done; | |
3507 | ||
3508 | ret = __extent_writepage_io(inode, page, wbc, epd, | |
3509 | i_size, nr_written, write_flags, &nr); | |
3510 | if (ret == 1) | |
3511 | goto done_unlocked; | |
3512 | ||
d1310b2e CM |
3513 | done: |
3514 | if (nr == 0) { | |
3515 | /* make sure the mapping tag for page dirty gets cleared */ | |
3516 | set_page_writeback(page); | |
3517 | end_page_writeback(page); | |
3518 | } | |
61391d56 FM |
3519 | if (PageError(page)) { |
3520 | ret = ret < 0 ? ret : -EIO; | |
3521 | end_extent_writepage(page, ret, start, page_end); | |
3522 | } | |
d1310b2e | 3523 | unlock_page(page); |
40f76580 | 3524 | return ret; |
771ed689 | 3525 | |
11c8349b | 3526 | done_unlocked: |
d1310b2e CM |
3527 | return 0; |
3528 | } | |
3529 | ||
fd8b2b61 | 3530 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 3531 | { |
74316201 N |
3532 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
3533 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
3534 | } |
3535 | ||
0e378df1 CM |
3536 | static noinline_for_stack int |
3537 | lock_extent_buffer_for_io(struct extent_buffer *eb, | |
3538 | struct btrfs_fs_info *fs_info, | |
3539 | struct extent_page_data *epd) | |
0b32f4bb | 3540 | { |
cc5e31a4 | 3541 | int i, num_pages; |
0b32f4bb JB |
3542 | int flush = 0; |
3543 | int ret = 0; | |
3544 | ||
3545 | if (!btrfs_try_tree_write_lock(eb)) { | |
3546 | flush = 1; | |
3547 | flush_write_bio(epd); | |
3548 | btrfs_tree_lock(eb); | |
3549 | } | |
3550 | ||
3551 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3552 | btrfs_tree_unlock(eb); | |
3553 | if (!epd->sync_io) | |
3554 | return 0; | |
3555 | if (!flush) { | |
3556 | flush_write_bio(epd); | |
3557 | flush = 1; | |
3558 | } | |
a098d8e8 CM |
3559 | while (1) { |
3560 | wait_on_extent_buffer_writeback(eb); | |
3561 | btrfs_tree_lock(eb); | |
3562 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3563 | break; | |
0b32f4bb | 3564 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3565 | } |
3566 | } | |
3567 | ||
51561ffe JB |
3568 | /* |
3569 | * We need to do this to prevent races in people who check if the eb is | |
3570 | * under IO since we can end up having no IO bits set for a short period | |
3571 | * of time. | |
3572 | */ | |
3573 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
3574 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3575 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 3576 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 3577 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
104b4e51 NB |
3578 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
3579 | -eb->len, | |
3580 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 3581 | ret = 1; |
51561ffe JB |
3582 | } else { |
3583 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
3584 | } |
3585 | ||
3586 | btrfs_tree_unlock(eb); | |
3587 | ||
3588 | if (!ret) | |
3589 | return ret; | |
3590 | ||
65ad0104 | 3591 | num_pages = num_extent_pages(eb); |
0b32f4bb | 3592 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3593 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3594 | |
3595 | if (!trylock_page(p)) { | |
3596 | if (!flush) { | |
3597 | flush_write_bio(epd); | |
3598 | flush = 1; | |
3599 | } | |
3600 | lock_page(p); | |
3601 | } | |
3602 | } | |
3603 | ||
3604 | return ret; | |
3605 | } | |
3606 | ||
3607 | static void end_extent_buffer_writeback(struct extent_buffer *eb) | |
3608 | { | |
3609 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
4e857c58 | 3610 | smp_mb__after_atomic(); |
0b32f4bb JB |
3611 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); |
3612 | } | |
3613 | ||
656f30db FM |
3614 | static void set_btree_ioerr(struct page *page) |
3615 | { | |
3616 | struct extent_buffer *eb = (struct extent_buffer *)page->private; | |
656f30db FM |
3617 | |
3618 | SetPageError(page); | |
3619 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) | |
3620 | return; | |
3621 | ||
3622 | /* | |
3623 | * If writeback for a btree extent that doesn't belong to a log tree | |
3624 | * failed, increment the counter transaction->eb_write_errors. | |
3625 | * We do this because while the transaction is running and before it's | |
3626 | * committing (when we call filemap_fdata[write|wait]_range against | |
3627 | * the btree inode), we might have | |
3628 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
3629 | * returns an error or an error happens during writeback, when we're | |
3630 | * committing the transaction we wouldn't know about it, since the pages | |
3631 | * can be no longer dirty nor marked anymore for writeback (if a | |
3632 | * subsequent modification to the extent buffer didn't happen before the | |
3633 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
3634 | * able to find the pages tagged with SetPageError at transaction | |
3635 | * commit time. So if this happens we must abort the transaction, | |
3636 | * otherwise we commit a super block with btree roots that point to | |
3637 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
3638 | * or the content of some node/leaf from a past generation that got | |
3639 | * cowed or deleted and is no longer valid. | |
3640 | * | |
3641 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
3642 | * not be enough - we need to distinguish between log tree extents vs | |
3643 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
3644 | * will catch and clear such errors in the mapping - and that call might | |
3645 | * be from a log sync and not from a transaction commit. Also, checking | |
3646 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
3647 | * not done and would not be reliable - the eb might have been released | |
3648 | * from memory and reading it back again means that flag would not be | |
3649 | * set (since it's a runtime flag, not persisted on disk). | |
3650 | * | |
3651 | * Using the flags below in the btree inode also makes us achieve the | |
3652 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
3653 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
3654 | * is called, the writeback for all dirty pages had already finished | |
3655 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
3656 | * filemap_fdatawait_range() would return success, as it could not know | |
3657 | * that writeback errors happened (the pages were no longer tagged for | |
3658 | * writeback). | |
3659 | */ | |
3660 | switch (eb->log_index) { | |
3661 | case -1: | |
afcdd129 | 3662 | set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags); |
656f30db FM |
3663 | break; |
3664 | case 0: | |
afcdd129 | 3665 | set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags); |
656f30db FM |
3666 | break; |
3667 | case 1: | |
afcdd129 | 3668 | set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags); |
656f30db FM |
3669 | break; |
3670 | default: | |
3671 | BUG(); /* unexpected, logic error */ | |
3672 | } | |
3673 | } | |
3674 | ||
4246a0b6 | 3675 | static void end_bio_extent_buffer_writepage(struct bio *bio) |
0b32f4bb | 3676 | { |
2c30c71b | 3677 | struct bio_vec *bvec; |
0b32f4bb | 3678 | struct extent_buffer *eb; |
2c30c71b | 3679 | int i, done; |
0b32f4bb | 3680 | |
c09abff8 | 3681 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2c30c71b | 3682 | bio_for_each_segment_all(bvec, bio, i) { |
0b32f4bb JB |
3683 | struct page *page = bvec->bv_page; |
3684 | ||
0b32f4bb JB |
3685 | eb = (struct extent_buffer *)page->private; |
3686 | BUG_ON(!eb); | |
3687 | done = atomic_dec_and_test(&eb->io_pages); | |
3688 | ||
4e4cbee9 | 3689 | if (bio->bi_status || |
4246a0b6 | 3690 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { |
0b32f4bb | 3691 | ClearPageUptodate(page); |
656f30db | 3692 | set_btree_ioerr(page); |
0b32f4bb JB |
3693 | } |
3694 | ||
3695 | end_page_writeback(page); | |
3696 | ||
3697 | if (!done) | |
3698 | continue; | |
3699 | ||
3700 | end_extent_buffer_writeback(eb); | |
2c30c71b | 3701 | } |
0b32f4bb JB |
3702 | |
3703 | bio_put(bio); | |
0b32f4bb JB |
3704 | } |
3705 | ||
0e378df1 | 3706 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb JB |
3707 | struct btrfs_fs_info *fs_info, |
3708 | struct writeback_control *wbc, | |
3709 | struct extent_page_data *epd) | |
3710 | { | |
3711 | struct block_device *bdev = fs_info->fs_devices->latest_bdev; | |
f28491e0 | 3712 | struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree; |
0b32f4bb | 3713 | u64 offset = eb->start; |
851cd173 | 3714 | u32 nritems; |
cc5e31a4 | 3715 | int i, num_pages; |
851cd173 | 3716 | unsigned long start, end; |
ff40adf7 | 3717 | unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META; |
d7dbe9e7 | 3718 | int ret = 0; |
0b32f4bb | 3719 | |
656f30db | 3720 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); |
65ad0104 | 3721 | num_pages = num_extent_pages(eb); |
0b32f4bb | 3722 | atomic_set(&eb->io_pages, num_pages); |
de0022b9 | 3723 | |
851cd173 LB |
3724 | /* set btree blocks beyond nritems with 0 to avoid stale content. */ |
3725 | nritems = btrfs_header_nritems(eb); | |
3eb548ee | 3726 | if (btrfs_header_level(eb) > 0) { |
3eb548ee LB |
3727 | end = btrfs_node_key_ptr_offset(nritems); |
3728 | ||
b159fa28 | 3729 | memzero_extent_buffer(eb, end, eb->len - end); |
851cd173 LB |
3730 | } else { |
3731 | /* | |
3732 | * leaf: | |
3733 | * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0 | |
3734 | */ | |
3735 | start = btrfs_item_nr_offset(nritems); | |
3d9ec8c4 | 3736 | end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, eb); |
b159fa28 | 3737 | memzero_extent_buffer(eb, start, end - start); |
3eb548ee LB |
3738 | } |
3739 | ||
0b32f4bb | 3740 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3741 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3742 | |
3743 | clear_page_dirty_for_io(p); | |
3744 | set_page_writeback(p); | |
4b81ba48 | 3745 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc, |
6273b7f8 | 3746 | p, offset, PAGE_SIZE, 0, bdev, |
c2df8bb4 | 3747 | &epd->bio, |
1f7ad75b | 3748 | end_bio_extent_buffer_writepage, |
18fdc679 | 3749 | 0, 0, 0, false); |
0b32f4bb | 3750 | if (ret) { |
656f30db | 3751 | set_btree_ioerr(p); |
fe01aa65 TK |
3752 | if (PageWriteback(p)) |
3753 | end_page_writeback(p); | |
0b32f4bb JB |
3754 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
3755 | end_extent_buffer_writeback(eb); | |
3756 | ret = -EIO; | |
3757 | break; | |
3758 | } | |
09cbfeaf | 3759 | offset += PAGE_SIZE; |
3d4b9496 | 3760 | update_nr_written(wbc, 1); |
0b32f4bb JB |
3761 | unlock_page(p); |
3762 | } | |
3763 | ||
3764 | if (unlikely(ret)) { | |
3765 | for (; i < num_pages; i++) { | |
bbf65cf0 | 3766 | struct page *p = eb->pages[i]; |
81465028 | 3767 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
3768 | unlock_page(p); |
3769 | } | |
3770 | } | |
3771 | ||
3772 | return ret; | |
3773 | } | |
3774 | ||
3775 | int btree_write_cache_pages(struct address_space *mapping, | |
3776 | struct writeback_control *wbc) | |
3777 | { | |
3778 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; | |
3779 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; | |
3780 | struct extent_buffer *eb, *prev_eb = NULL; | |
3781 | struct extent_page_data epd = { | |
3782 | .bio = NULL, | |
3783 | .tree = tree, | |
3784 | .extent_locked = 0, | |
3785 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
3786 | }; | |
3787 | int ret = 0; | |
3788 | int done = 0; | |
3789 | int nr_to_write_done = 0; | |
3790 | struct pagevec pvec; | |
3791 | int nr_pages; | |
3792 | pgoff_t index; | |
3793 | pgoff_t end; /* Inclusive */ | |
3794 | int scanned = 0; | |
3795 | int tag; | |
3796 | ||
86679820 | 3797 | pagevec_init(&pvec); |
0b32f4bb JB |
3798 | if (wbc->range_cyclic) { |
3799 | index = mapping->writeback_index; /* Start from prev offset */ | |
3800 | end = -1; | |
3801 | } else { | |
09cbfeaf KS |
3802 | index = wbc->range_start >> PAGE_SHIFT; |
3803 | end = wbc->range_end >> PAGE_SHIFT; | |
0b32f4bb JB |
3804 | scanned = 1; |
3805 | } | |
3806 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3807 | tag = PAGECACHE_TAG_TOWRITE; | |
3808 | else | |
3809 | tag = PAGECACHE_TAG_DIRTY; | |
3810 | retry: | |
3811 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3812 | tag_pages_for_writeback(mapping, index, end); | |
3813 | while (!done && !nr_to_write_done && (index <= end) && | |
4006f437 | 3814 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 3815 | tag))) { |
0b32f4bb JB |
3816 | unsigned i; |
3817 | ||
3818 | scanned = 1; | |
3819 | for (i = 0; i < nr_pages; i++) { | |
3820 | struct page *page = pvec.pages[i]; | |
3821 | ||
3822 | if (!PagePrivate(page)) | |
3823 | continue; | |
3824 | ||
b5bae261 JB |
3825 | spin_lock(&mapping->private_lock); |
3826 | if (!PagePrivate(page)) { | |
3827 | spin_unlock(&mapping->private_lock); | |
3828 | continue; | |
3829 | } | |
3830 | ||
0b32f4bb | 3831 | eb = (struct extent_buffer *)page->private; |
b5bae261 JB |
3832 | |
3833 | /* | |
3834 | * Shouldn't happen and normally this would be a BUG_ON | |
3835 | * but no sense in crashing the users box for something | |
3836 | * we can survive anyway. | |
3837 | */ | |
fae7f21c | 3838 | if (WARN_ON(!eb)) { |
b5bae261 | 3839 | spin_unlock(&mapping->private_lock); |
0b32f4bb JB |
3840 | continue; |
3841 | } | |
3842 | ||
b5bae261 JB |
3843 | if (eb == prev_eb) { |
3844 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 3845 | continue; |
b5bae261 | 3846 | } |
0b32f4bb | 3847 | |
b5bae261 JB |
3848 | ret = atomic_inc_not_zero(&eb->refs); |
3849 | spin_unlock(&mapping->private_lock); | |
3850 | if (!ret) | |
0b32f4bb | 3851 | continue; |
0b32f4bb JB |
3852 | |
3853 | prev_eb = eb; | |
3854 | ret = lock_extent_buffer_for_io(eb, fs_info, &epd); | |
3855 | if (!ret) { | |
3856 | free_extent_buffer(eb); | |
3857 | continue; | |
3858 | } | |
3859 | ||
3860 | ret = write_one_eb(eb, fs_info, wbc, &epd); | |
3861 | if (ret) { | |
3862 | done = 1; | |
3863 | free_extent_buffer(eb); | |
3864 | break; | |
3865 | } | |
3866 | free_extent_buffer(eb); | |
3867 | ||
3868 | /* | |
3869 | * the filesystem may choose to bump up nr_to_write. | |
3870 | * We have to make sure to honor the new nr_to_write | |
3871 | * at any time | |
3872 | */ | |
3873 | nr_to_write_done = wbc->nr_to_write <= 0; | |
3874 | } | |
3875 | pagevec_release(&pvec); | |
3876 | cond_resched(); | |
3877 | } | |
3878 | if (!scanned && !done) { | |
3879 | /* | |
3880 | * We hit the last page and there is more work to be done: wrap | |
3881 | * back to the start of the file | |
3882 | */ | |
3883 | scanned = 1; | |
3884 | index = 0; | |
3885 | goto retry; | |
3886 | } | |
3887 | flush_write_bio(&epd); | |
3888 | return ret; | |
3889 | } | |
3890 | ||
d1310b2e | 3891 | /** |
4bef0848 | 3892 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
3893 | * @mapping: address space structure to write |
3894 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
935db853 | 3895 | * @data: data passed to __extent_writepage function |
d1310b2e CM |
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, |
aab6e9ed | 3907 | struct extent_page_data *epd) |
d1310b2e | 3908 | { |
7fd1a3f7 | 3909 | struct inode *inode = mapping->host; |
d1310b2e CM |
3910 | int ret = 0; |
3911 | int done = 0; | |
f85d7d6c | 3912 | int nr_to_write_done = 0; |
d1310b2e CM |
3913 | struct pagevec pvec; |
3914 | int nr_pages; | |
3915 | pgoff_t index; | |
3916 | pgoff_t end; /* Inclusive */ | |
a9132667 LB |
3917 | pgoff_t done_index; |
3918 | int range_whole = 0; | |
d1310b2e | 3919 | int scanned = 0; |
f7aaa06b | 3920 | int tag; |
d1310b2e | 3921 | |
7fd1a3f7 JB |
3922 | /* |
3923 | * We have to hold onto the inode so that ordered extents can do their | |
3924 | * work when the IO finishes. The alternative to this is failing to add | |
3925 | * an ordered extent if the igrab() fails there and that is a huge pain | |
3926 | * to deal with, so instead just hold onto the inode throughout the | |
3927 | * writepages operation. If it fails here we are freeing up the inode | |
3928 | * anyway and we'd rather not waste our time writing out stuff that is | |
3929 | * going to be truncated anyway. | |
3930 | */ | |
3931 | if (!igrab(inode)) | |
3932 | return 0; | |
3933 | ||
86679820 | 3934 | pagevec_init(&pvec); |
d1310b2e CM |
3935 | if (wbc->range_cyclic) { |
3936 | index = mapping->writeback_index; /* Start from prev offset */ | |
3937 | end = -1; | |
3938 | } else { | |
09cbfeaf KS |
3939 | index = wbc->range_start >> PAGE_SHIFT; |
3940 | end = wbc->range_end >> PAGE_SHIFT; | |
a9132667 LB |
3941 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
3942 | range_whole = 1; | |
d1310b2e CM |
3943 | scanned = 1; |
3944 | } | |
f7aaa06b JB |
3945 | if (wbc->sync_mode == WB_SYNC_ALL) |
3946 | tag = PAGECACHE_TAG_TOWRITE; | |
3947 | else | |
3948 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 3949 | retry: |
f7aaa06b JB |
3950 | if (wbc->sync_mode == WB_SYNC_ALL) |
3951 | tag_pages_for_writeback(mapping, index, end); | |
a9132667 | 3952 | done_index = index; |
f85d7d6c | 3953 | while (!done && !nr_to_write_done && (index <= end) && |
67fd707f JK |
3954 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, |
3955 | &index, end, tag))) { | |
d1310b2e CM |
3956 | unsigned i; |
3957 | ||
3958 | scanned = 1; | |
3959 | for (i = 0; i < nr_pages; i++) { | |
3960 | struct page *page = pvec.pages[i]; | |
3961 | ||
a9132667 | 3962 | done_index = page->index; |
d1310b2e | 3963 | /* |
b93b0163 MW |
3964 | * At this point we hold neither the i_pages lock nor |
3965 | * the page lock: the page may be truncated or | |
3966 | * invalidated (changing page->mapping to NULL), | |
3967 | * or even swizzled back from swapper_space to | |
3968 | * tmpfs file mapping | |
d1310b2e | 3969 | */ |
c8f2f24b | 3970 | if (!trylock_page(page)) { |
aab6e9ed | 3971 | flush_write_bio(epd); |
c8f2f24b | 3972 | lock_page(page); |
01d658f2 | 3973 | } |
d1310b2e CM |
3974 | |
3975 | if (unlikely(page->mapping != mapping)) { | |
3976 | unlock_page(page); | |
3977 | continue; | |
3978 | } | |
3979 | ||
d2c3f4f6 | 3980 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 | 3981 | if (PageWriteback(page)) |
aab6e9ed | 3982 | flush_write_bio(epd); |
d1310b2e | 3983 | wait_on_page_writeback(page); |
d2c3f4f6 | 3984 | } |
d1310b2e CM |
3985 | |
3986 | if (PageWriteback(page) || | |
3987 | !clear_page_dirty_for_io(page)) { | |
3988 | unlock_page(page); | |
3989 | continue; | |
3990 | } | |
3991 | ||
aab6e9ed | 3992 | ret = __extent_writepage(page, wbc, epd); |
d1310b2e CM |
3993 | |
3994 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
3995 | unlock_page(page); | |
3996 | ret = 0; | |
3997 | } | |
a9132667 LB |
3998 | if (ret < 0) { |
3999 | /* | |
4000 | * done_index is set past this page, | |
4001 | * so media errors will not choke | |
4002 | * background writeout for the entire | |
4003 | * file. This has consequences for | |
4004 | * range_cyclic semantics (ie. it may | |
4005 | * not be suitable for data integrity | |
4006 | * writeout). | |
4007 | */ | |
4008 | done_index = page->index + 1; | |
4009 | done = 1; | |
4010 | break; | |
4011 | } | |
f85d7d6c CM |
4012 | |
4013 | /* | |
4014 | * the filesystem may choose to bump up nr_to_write. | |
4015 | * We have to make sure to honor the new nr_to_write | |
4016 | * at any time | |
4017 | */ | |
4018 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
4019 | } |
4020 | pagevec_release(&pvec); | |
4021 | cond_resched(); | |
4022 | } | |
894b36e3 | 4023 | if (!scanned && !done) { |
d1310b2e CM |
4024 | /* |
4025 | * We hit the last page and there is more work to be done: wrap | |
4026 | * back to the start of the file | |
4027 | */ | |
4028 | scanned = 1; | |
4029 | index = 0; | |
4030 | goto retry; | |
4031 | } | |
a9132667 LB |
4032 | |
4033 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole)) | |
4034 | mapping->writeback_index = done_index; | |
4035 | ||
7fd1a3f7 | 4036 | btrfs_add_delayed_iput(inode); |
894b36e3 | 4037 | return ret; |
d1310b2e | 4038 | } |
d1310b2e | 4039 | |
aab6e9ed | 4040 | static void flush_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 4041 | { |
d2c3f4f6 | 4042 | if (epd->bio) { |
355808c2 JM |
4043 | int ret; |
4044 | ||
18fdc679 | 4045 | ret = submit_one_bio(epd->bio, 0, 0); |
79787eaa | 4046 | BUG_ON(ret < 0); /* -ENOMEM */ |
d2c3f4f6 CM |
4047 | epd->bio = NULL; |
4048 | } | |
4049 | } | |
4050 | ||
0a9b0e53 | 4051 | int extent_write_full_page(struct page *page, struct writeback_control *wbc) |
d1310b2e CM |
4052 | { |
4053 | int ret; | |
d1310b2e CM |
4054 | struct extent_page_data epd = { |
4055 | .bio = NULL, | |
0a9b0e53 | 4056 | .tree = &BTRFS_I(page->mapping->host)->io_tree, |
771ed689 | 4057 | .extent_locked = 0, |
ffbd517d | 4058 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e | 4059 | }; |
d1310b2e | 4060 | |
d1310b2e CM |
4061 | ret = __extent_writepage(page, wbc, &epd); |
4062 | ||
e2932ee0 | 4063 | flush_write_bio(&epd); |
d1310b2e CM |
4064 | return ret; |
4065 | } | |
d1310b2e | 4066 | |
5e3ee236 | 4067 | int extent_write_locked_range(struct inode *inode, u64 start, u64 end, |
771ed689 CM |
4068 | int mode) |
4069 | { | |
4070 | int ret = 0; | |
4071 | struct address_space *mapping = inode->i_mapping; | |
5e3ee236 | 4072 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
771ed689 | 4073 | struct page *page; |
09cbfeaf KS |
4074 | unsigned long nr_pages = (end - start + PAGE_SIZE) >> |
4075 | PAGE_SHIFT; | |
771ed689 CM |
4076 | |
4077 | struct extent_page_data epd = { | |
4078 | .bio = NULL, | |
4079 | .tree = tree, | |
771ed689 | 4080 | .extent_locked = 1, |
ffbd517d | 4081 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
4082 | }; |
4083 | struct writeback_control wbc_writepages = { | |
771ed689 | 4084 | .sync_mode = mode, |
771ed689 CM |
4085 | .nr_to_write = nr_pages * 2, |
4086 | .range_start = start, | |
4087 | .range_end = end + 1, | |
4088 | }; | |
4089 | ||
d397712b | 4090 | while (start <= end) { |
09cbfeaf | 4091 | page = find_get_page(mapping, start >> PAGE_SHIFT); |
771ed689 CM |
4092 | if (clear_page_dirty_for_io(page)) |
4093 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
4094 | else { | |
4095 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
4096 | tree->ops->writepage_end_io_hook(page, start, | |
09cbfeaf | 4097 | start + PAGE_SIZE - 1, |
771ed689 CM |
4098 | NULL, 1); |
4099 | unlock_page(page); | |
4100 | } | |
09cbfeaf KS |
4101 | put_page(page); |
4102 | start += PAGE_SIZE; | |
771ed689 CM |
4103 | } |
4104 | ||
e2932ee0 | 4105 | flush_write_bio(&epd); |
771ed689 CM |
4106 | return ret; |
4107 | } | |
d1310b2e | 4108 | |
8ae225a8 | 4109 | int extent_writepages(struct address_space *mapping, |
d1310b2e CM |
4110 | struct writeback_control *wbc) |
4111 | { | |
4112 | int ret = 0; | |
4113 | struct extent_page_data epd = { | |
4114 | .bio = NULL, | |
8ae225a8 | 4115 | .tree = &BTRFS_I(mapping->host)->io_tree, |
771ed689 | 4116 | .extent_locked = 0, |
ffbd517d | 4117 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
4118 | }; |
4119 | ||
935db853 | 4120 | ret = extent_write_cache_pages(mapping, wbc, &epd); |
e2932ee0 | 4121 | flush_write_bio(&epd); |
d1310b2e CM |
4122 | return ret; |
4123 | } | |
d1310b2e | 4124 | |
2a3ff0ad NB |
4125 | int extent_readpages(struct address_space *mapping, struct list_head *pages, |
4126 | unsigned nr_pages) | |
d1310b2e CM |
4127 | { |
4128 | struct bio *bio = NULL; | |
4129 | unsigned page_idx; | |
c8b97818 | 4130 | unsigned long bio_flags = 0; |
67c9684f LB |
4131 | struct page *pagepool[16]; |
4132 | struct page *page; | |
125bac01 | 4133 | struct extent_map *em_cached = NULL; |
2a3ff0ad | 4134 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; |
67c9684f | 4135 | int nr = 0; |
808f80b4 | 4136 | u64 prev_em_start = (u64)-1; |
d1310b2e | 4137 | |
d1310b2e | 4138 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
67c9684f | 4139 | page = list_entry(pages->prev, struct page, lru); |
d1310b2e CM |
4140 | |
4141 | prefetchw(&page->flags); | |
4142 | list_del(&page->lru); | |
67c9684f | 4143 | if (add_to_page_cache_lru(page, mapping, |
8a5c743e MH |
4144 | page->index, |
4145 | readahead_gfp_mask(mapping))) { | |
09cbfeaf | 4146 | put_page(page); |
67c9684f | 4147 | continue; |
d1310b2e | 4148 | } |
67c9684f LB |
4149 | |
4150 | pagepool[nr++] = page; | |
4151 | if (nr < ARRAY_SIZE(pagepool)) | |
4152 | continue; | |
e4d17ef5 DS |
4153 | __extent_readpages(tree, pagepool, nr, &em_cached, &bio, |
4154 | &bio_flags, &prev_em_start); | |
67c9684f | 4155 | nr = 0; |
d1310b2e | 4156 | } |
9974090b | 4157 | if (nr) |
e4d17ef5 DS |
4158 | __extent_readpages(tree, pagepool, nr, &em_cached, &bio, |
4159 | &bio_flags, &prev_em_start); | |
67c9684f | 4160 | |
125bac01 MX |
4161 | if (em_cached) |
4162 | free_extent_map(em_cached); | |
4163 | ||
d1310b2e CM |
4164 | BUG_ON(!list_empty(pages)); |
4165 | if (bio) | |
1f7ad75b | 4166 | return submit_one_bio(bio, 0, bio_flags); |
d1310b2e CM |
4167 | return 0; |
4168 | } | |
d1310b2e CM |
4169 | |
4170 | /* | |
4171 | * basic invalidatepage code, this waits on any locked or writeback | |
4172 | * ranges corresponding to the page, and then deletes any extent state | |
4173 | * records from the tree | |
4174 | */ | |
4175 | int extent_invalidatepage(struct extent_io_tree *tree, | |
4176 | struct page *page, unsigned long offset) | |
4177 | { | |
2ac55d41 | 4178 | struct extent_state *cached_state = NULL; |
4eee4fa4 | 4179 | u64 start = page_offset(page); |
09cbfeaf | 4180 | u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
4181 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; |
4182 | ||
fda2832f | 4183 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
4184 | if (start > end) |
4185 | return 0; | |
4186 | ||
ff13db41 | 4187 | lock_extent_bits(tree, start, end, &cached_state); |
1edbb734 | 4188 | wait_on_page_writeback(page); |
d1310b2e | 4189 | clear_extent_bit(tree, start, end, |
32c00aff JB |
4190 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
4191 | EXTENT_DO_ACCOUNTING, | |
ae0f1625 | 4192 | 1, 1, &cached_state); |
d1310b2e CM |
4193 | return 0; |
4194 | } | |
d1310b2e | 4195 | |
7b13b7b1 CM |
4196 | /* |
4197 | * a helper for releasepage, this tests for areas of the page that | |
4198 | * are locked or under IO and drops the related state bits if it is safe | |
4199 | * to drop the page. | |
4200 | */ | |
29c68b2d | 4201 | static int try_release_extent_state(struct extent_io_tree *tree, |
48a3b636 | 4202 | struct page *page, gfp_t mask) |
7b13b7b1 | 4203 | { |
4eee4fa4 | 4204 | u64 start = page_offset(page); |
09cbfeaf | 4205 | u64 end = start + PAGE_SIZE - 1; |
7b13b7b1 CM |
4206 | int ret = 1; |
4207 | ||
211f90e6 | 4208 | if (test_range_bit(tree, start, end, |
8b62b72b | 4209 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
4210 | ret = 0; |
4211 | else { | |
11ef160f CM |
4212 | /* |
4213 | * at this point we can safely clear everything except the | |
4214 | * locked bit and the nodatasum bit | |
4215 | */ | |
66b0c887 | 4216 | ret = __clear_extent_bit(tree, start, end, |
11ef160f | 4217 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
66b0c887 | 4218 | 0, 0, NULL, mask, NULL); |
e3f24cc5 CM |
4219 | |
4220 | /* if clear_extent_bit failed for enomem reasons, | |
4221 | * we can't allow the release to continue. | |
4222 | */ | |
4223 | if (ret < 0) | |
4224 | ret = 0; | |
4225 | else | |
4226 | ret = 1; | |
7b13b7b1 CM |
4227 | } |
4228 | return ret; | |
4229 | } | |
7b13b7b1 | 4230 | |
d1310b2e CM |
4231 | /* |
4232 | * a helper for releasepage. As long as there are no locked extents | |
4233 | * in the range corresponding to the page, both state records and extent | |
4234 | * map records are removed | |
4235 | */ | |
477a30ba | 4236 | int try_release_extent_mapping(struct page *page, gfp_t mask) |
d1310b2e CM |
4237 | { |
4238 | struct extent_map *em; | |
4eee4fa4 | 4239 | u64 start = page_offset(page); |
09cbfeaf | 4240 | u64 end = start + PAGE_SIZE - 1; |
bd3599a0 FM |
4241 | struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host); |
4242 | struct extent_io_tree *tree = &btrfs_inode->io_tree; | |
4243 | struct extent_map_tree *map = &btrfs_inode->extent_tree; | |
7b13b7b1 | 4244 | |
d0164adc | 4245 | if (gfpflags_allow_blocking(mask) && |
ee22184b | 4246 | page->mapping->host->i_size > SZ_16M) { |
39b5637f | 4247 | u64 len; |
70dec807 | 4248 | while (start <= end) { |
39b5637f | 4249 | len = end - start + 1; |
890871be | 4250 | write_lock(&map->lock); |
39b5637f | 4251 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 4252 | if (!em) { |
890871be | 4253 | write_unlock(&map->lock); |
70dec807 CM |
4254 | break; |
4255 | } | |
7f3c74fb CM |
4256 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
4257 | em->start != start) { | |
890871be | 4258 | write_unlock(&map->lock); |
70dec807 CM |
4259 | free_extent_map(em); |
4260 | break; | |
4261 | } | |
4262 | if (!test_range_bit(tree, em->start, | |
4263 | extent_map_end(em) - 1, | |
8b62b72b | 4264 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 4265 | 0, NULL)) { |
bd3599a0 FM |
4266 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
4267 | &btrfs_inode->runtime_flags); | |
70dec807 CM |
4268 | remove_extent_mapping(map, em); |
4269 | /* once for the rb tree */ | |
4270 | free_extent_map(em); | |
4271 | } | |
4272 | start = extent_map_end(em); | |
890871be | 4273 | write_unlock(&map->lock); |
70dec807 CM |
4274 | |
4275 | /* once for us */ | |
d1310b2e CM |
4276 | free_extent_map(em); |
4277 | } | |
d1310b2e | 4278 | } |
29c68b2d | 4279 | return try_release_extent_state(tree, page, mask); |
d1310b2e | 4280 | } |
d1310b2e | 4281 | |
ec29ed5b CM |
4282 | /* |
4283 | * helper function for fiemap, which doesn't want to see any holes. | |
4284 | * This maps until we find something past 'last' | |
4285 | */ | |
4286 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
e3350e16 | 4287 | u64 offset, u64 last) |
ec29ed5b | 4288 | { |
da17066c | 4289 | u64 sectorsize = btrfs_inode_sectorsize(inode); |
ec29ed5b CM |
4290 | struct extent_map *em; |
4291 | u64 len; | |
4292 | ||
4293 | if (offset >= last) | |
4294 | return NULL; | |
4295 | ||
67871254 | 4296 | while (1) { |
ec29ed5b CM |
4297 | len = last - offset; |
4298 | if (len == 0) | |
4299 | break; | |
fda2832f | 4300 | len = ALIGN(len, sectorsize); |
e3350e16 DS |
4301 | em = btrfs_get_extent_fiemap(BTRFS_I(inode), NULL, 0, offset, |
4302 | len, 0); | |
c704005d | 4303 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
4304 | return em; |
4305 | ||
4306 | /* if this isn't a hole return it */ | |
4a2d25cd | 4307 | if (em->block_start != EXTENT_MAP_HOLE) |
ec29ed5b | 4308 | return em; |
ec29ed5b CM |
4309 | |
4310 | /* this is a hole, advance to the next extent */ | |
4311 | offset = extent_map_end(em); | |
4312 | free_extent_map(em); | |
4313 | if (offset >= last) | |
4314 | break; | |
4315 | } | |
4316 | return NULL; | |
4317 | } | |
4318 | ||
4751832d QW |
4319 | /* |
4320 | * To cache previous fiemap extent | |
4321 | * | |
4322 | * Will be used for merging fiemap extent | |
4323 | */ | |
4324 | struct fiemap_cache { | |
4325 | u64 offset; | |
4326 | u64 phys; | |
4327 | u64 len; | |
4328 | u32 flags; | |
4329 | bool cached; | |
4330 | }; | |
4331 | ||
4332 | /* | |
4333 | * Helper to submit fiemap extent. | |
4334 | * | |
4335 | * Will try to merge current fiemap extent specified by @offset, @phys, | |
4336 | * @len and @flags with cached one. | |
4337 | * And only when we fails to merge, cached one will be submitted as | |
4338 | * fiemap extent. | |
4339 | * | |
4340 | * Return value is the same as fiemap_fill_next_extent(). | |
4341 | */ | |
4342 | static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, | |
4343 | struct fiemap_cache *cache, | |
4344 | u64 offset, u64 phys, u64 len, u32 flags) | |
4345 | { | |
4346 | int ret = 0; | |
4347 | ||
4348 | if (!cache->cached) | |
4349 | goto assign; | |
4350 | ||
4351 | /* | |
4352 | * Sanity check, extent_fiemap() should have ensured that new | |
4353 | * fiemap extent won't overlap with cahced one. | |
4354 | * Not recoverable. | |
4355 | * | |
4356 | * NOTE: Physical address can overlap, due to compression | |
4357 | */ | |
4358 | if (cache->offset + cache->len > offset) { | |
4359 | WARN_ON(1); | |
4360 | return -EINVAL; | |
4361 | } | |
4362 | ||
4363 | /* | |
4364 | * Only merges fiemap extents if | |
4365 | * 1) Their logical addresses are continuous | |
4366 | * | |
4367 | * 2) Their physical addresses are continuous | |
4368 | * So truly compressed (physical size smaller than logical size) | |
4369 | * extents won't get merged with each other | |
4370 | * | |
4371 | * 3) Share same flags except FIEMAP_EXTENT_LAST | |
4372 | * So regular extent won't get merged with prealloc extent | |
4373 | */ | |
4374 | if (cache->offset + cache->len == offset && | |
4375 | cache->phys + cache->len == phys && | |
4376 | (cache->flags & ~FIEMAP_EXTENT_LAST) == | |
4377 | (flags & ~FIEMAP_EXTENT_LAST)) { | |
4378 | cache->len += len; | |
4379 | cache->flags |= flags; | |
4380 | goto try_submit_last; | |
4381 | } | |
4382 | ||
4383 | /* Not mergeable, need to submit cached one */ | |
4384 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, | |
4385 | cache->len, cache->flags); | |
4386 | cache->cached = false; | |
4387 | if (ret) | |
4388 | return ret; | |
4389 | assign: | |
4390 | cache->cached = true; | |
4391 | cache->offset = offset; | |
4392 | cache->phys = phys; | |
4393 | cache->len = len; | |
4394 | cache->flags = flags; | |
4395 | try_submit_last: | |
4396 | if (cache->flags & FIEMAP_EXTENT_LAST) { | |
4397 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, | |
4398 | cache->phys, cache->len, cache->flags); | |
4399 | cache->cached = false; | |
4400 | } | |
4401 | return ret; | |
4402 | } | |
4403 | ||
4404 | /* | |
848c23b7 | 4405 | * Emit last fiemap cache |
4751832d | 4406 | * |
848c23b7 QW |
4407 | * The last fiemap cache may still be cached in the following case: |
4408 | * 0 4k 8k | |
4409 | * |<- Fiemap range ->| | |
4410 | * |<------------ First extent ----------->| | |
4411 | * | |
4412 | * In this case, the first extent range will be cached but not emitted. | |
4413 | * So we must emit it before ending extent_fiemap(). | |
4751832d | 4414 | */ |
848c23b7 QW |
4415 | static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info, |
4416 | struct fiemap_extent_info *fieinfo, | |
4417 | struct fiemap_cache *cache) | |
4751832d QW |
4418 | { |
4419 | int ret; | |
4420 | ||
4421 | if (!cache->cached) | |
4422 | return 0; | |
4423 | ||
4751832d QW |
4424 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, |
4425 | cache->len, cache->flags); | |
4426 | cache->cached = false; | |
4427 | if (ret > 0) | |
4428 | ret = 0; | |
4429 | return ret; | |
4430 | } | |
4431 | ||
1506fcc8 | 4432 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
2135fb9b | 4433 | __u64 start, __u64 len) |
1506fcc8 | 4434 | { |
975f84fe | 4435 | int ret = 0; |
1506fcc8 YS |
4436 | u64 off = start; |
4437 | u64 max = start + len; | |
4438 | u32 flags = 0; | |
975f84fe JB |
4439 | u32 found_type; |
4440 | u64 last; | |
ec29ed5b | 4441 | u64 last_for_get_extent = 0; |
1506fcc8 | 4442 | u64 disko = 0; |
ec29ed5b | 4443 | u64 isize = i_size_read(inode); |
975f84fe | 4444 | struct btrfs_key found_key; |
1506fcc8 | 4445 | struct extent_map *em = NULL; |
2ac55d41 | 4446 | struct extent_state *cached_state = NULL; |
975f84fe | 4447 | struct btrfs_path *path; |
dc046b10 | 4448 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4751832d | 4449 | struct fiemap_cache cache = { 0 }; |
1506fcc8 | 4450 | int end = 0; |
ec29ed5b CM |
4451 | u64 em_start = 0; |
4452 | u64 em_len = 0; | |
4453 | u64 em_end = 0; | |
1506fcc8 YS |
4454 | |
4455 | if (len == 0) | |
4456 | return -EINVAL; | |
4457 | ||
975f84fe JB |
4458 | path = btrfs_alloc_path(); |
4459 | if (!path) | |
4460 | return -ENOMEM; | |
4461 | path->leave_spinning = 1; | |
4462 | ||
da17066c JM |
4463 | start = round_down(start, btrfs_inode_sectorsize(inode)); |
4464 | len = round_up(max, btrfs_inode_sectorsize(inode)) - start; | |
4d479cf0 | 4465 | |
ec29ed5b CM |
4466 | /* |
4467 | * lookup the last file extent. We're not using i_size here | |
4468 | * because there might be preallocation past i_size | |
4469 | */ | |
f85b7379 DS |
4470 | ret = btrfs_lookup_file_extent(NULL, root, path, |
4471 | btrfs_ino(BTRFS_I(inode)), -1, 0); | |
975f84fe JB |
4472 | if (ret < 0) { |
4473 | btrfs_free_path(path); | |
4474 | return ret; | |
2d324f59 LB |
4475 | } else { |
4476 | WARN_ON(!ret); | |
4477 | if (ret == 1) | |
4478 | ret = 0; | |
975f84fe | 4479 | } |
2d324f59 | 4480 | |
975f84fe | 4481 | path->slots[0]--; |
975f84fe | 4482 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
962a298f | 4483 | found_type = found_key.type; |
975f84fe | 4484 | |
ec29ed5b | 4485 | /* No extents, but there might be delalloc bits */ |
4a0cc7ca | 4486 | if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) || |
975f84fe | 4487 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
4488 | /* have to trust i_size as the end */ |
4489 | last = (u64)-1; | |
4490 | last_for_get_extent = isize; | |
4491 | } else { | |
4492 | /* | |
4493 | * remember the start of the last extent. There are a | |
4494 | * bunch of different factors that go into the length of the | |
4495 | * extent, so its much less complex to remember where it started | |
4496 | */ | |
4497 | last = found_key.offset; | |
4498 | last_for_get_extent = last + 1; | |
975f84fe | 4499 | } |
fe09e16c | 4500 | btrfs_release_path(path); |
975f84fe | 4501 | |
ec29ed5b CM |
4502 | /* |
4503 | * we might have some extents allocated but more delalloc past those | |
4504 | * extents. so, we trust isize unless the start of the last extent is | |
4505 | * beyond isize | |
4506 | */ | |
4507 | if (last < isize) { | |
4508 | last = (u64)-1; | |
4509 | last_for_get_extent = isize; | |
4510 | } | |
4511 | ||
ff13db41 | 4512 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
d0082371 | 4513 | &cached_state); |
ec29ed5b | 4514 | |
e3350e16 | 4515 | em = get_extent_skip_holes(inode, start, last_for_get_extent); |
1506fcc8 YS |
4516 | if (!em) |
4517 | goto out; | |
4518 | if (IS_ERR(em)) { | |
4519 | ret = PTR_ERR(em); | |
4520 | goto out; | |
4521 | } | |
975f84fe | 4522 | |
1506fcc8 | 4523 | while (!end) { |
b76bb701 | 4524 | u64 offset_in_extent = 0; |
ea8efc74 CM |
4525 | |
4526 | /* break if the extent we found is outside the range */ | |
4527 | if (em->start >= max || extent_map_end(em) < off) | |
4528 | break; | |
4529 | ||
4530 | /* | |
4531 | * get_extent may return an extent that starts before our | |
4532 | * requested range. We have to make sure the ranges | |
4533 | * we return to fiemap always move forward and don't | |
4534 | * overlap, so adjust the offsets here | |
4535 | */ | |
4536 | em_start = max(em->start, off); | |
1506fcc8 | 4537 | |
ea8efc74 CM |
4538 | /* |
4539 | * record the offset from the start of the extent | |
b76bb701 JB |
4540 | * for adjusting the disk offset below. Only do this if the |
4541 | * extent isn't compressed since our in ram offset may be past | |
4542 | * what we have actually allocated on disk. | |
ea8efc74 | 4543 | */ |
b76bb701 JB |
4544 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
4545 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 4546 | em_end = extent_map_end(em); |
ea8efc74 | 4547 | em_len = em_end - em_start; |
1506fcc8 | 4548 | flags = 0; |
f0986318 FM |
4549 | if (em->block_start < EXTENT_MAP_LAST_BYTE) |
4550 | disko = em->block_start + offset_in_extent; | |
4551 | else | |
4552 | disko = 0; | |
1506fcc8 | 4553 | |
ea8efc74 CM |
4554 | /* |
4555 | * bump off for our next call to get_extent | |
4556 | */ | |
4557 | off = extent_map_end(em); | |
4558 | if (off >= max) | |
4559 | end = 1; | |
4560 | ||
93dbfad7 | 4561 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
4562 | end = 1; |
4563 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 4564 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
4565 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
4566 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 4567 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
4568 | flags |= (FIEMAP_EXTENT_DELALLOC | |
4569 | FIEMAP_EXTENT_UNKNOWN); | |
dc046b10 JB |
4570 | } else if (fieinfo->fi_extents_max) { |
4571 | u64 bytenr = em->block_start - | |
4572 | (em->start - em->orig_start); | |
fe09e16c | 4573 | |
fe09e16c LB |
4574 | /* |
4575 | * As btrfs supports shared space, this information | |
4576 | * can be exported to userspace tools via | |
dc046b10 JB |
4577 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 |
4578 | * then we're just getting a count and we can skip the | |
4579 | * lookup stuff. | |
fe09e16c | 4580 | */ |
bb739cf0 EN |
4581 | ret = btrfs_check_shared(root, |
4582 | btrfs_ino(BTRFS_I(inode)), | |
4583 | bytenr); | |
dc046b10 | 4584 | if (ret < 0) |
fe09e16c | 4585 | goto out_free; |
dc046b10 | 4586 | if (ret) |
fe09e16c | 4587 | flags |= FIEMAP_EXTENT_SHARED; |
dc046b10 | 4588 | ret = 0; |
1506fcc8 YS |
4589 | } |
4590 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
4591 | flags |= FIEMAP_EXTENT_ENCODED; | |
0d2b2372 JB |
4592 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
4593 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
1506fcc8 | 4594 | |
1506fcc8 YS |
4595 | free_extent_map(em); |
4596 | em = NULL; | |
ec29ed5b CM |
4597 | if ((em_start >= last) || em_len == (u64)-1 || |
4598 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
4599 | flags |= FIEMAP_EXTENT_LAST; |
4600 | end = 1; | |
4601 | } | |
4602 | ||
ec29ed5b | 4603 | /* now scan forward to see if this is really the last extent. */ |
e3350e16 | 4604 | em = get_extent_skip_holes(inode, off, last_for_get_extent); |
ec29ed5b CM |
4605 | if (IS_ERR(em)) { |
4606 | ret = PTR_ERR(em); | |
4607 | goto out; | |
4608 | } | |
4609 | if (!em) { | |
975f84fe JB |
4610 | flags |= FIEMAP_EXTENT_LAST; |
4611 | end = 1; | |
4612 | } | |
4751832d QW |
4613 | ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko, |
4614 | em_len, flags); | |
26e726af CS |
4615 | if (ret) { |
4616 | if (ret == 1) | |
4617 | ret = 0; | |
ec29ed5b | 4618 | goto out_free; |
26e726af | 4619 | } |
1506fcc8 YS |
4620 | } |
4621 | out_free: | |
4751832d | 4622 | if (!ret) |
848c23b7 | 4623 | ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache); |
1506fcc8 YS |
4624 | free_extent_map(em); |
4625 | out: | |
fe09e16c | 4626 | btrfs_free_path(path); |
a52f4cd2 | 4627 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
e43bbe5e | 4628 | &cached_state); |
1506fcc8 YS |
4629 | return ret; |
4630 | } | |
4631 | ||
727011e0 CM |
4632 | static void __free_extent_buffer(struct extent_buffer *eb) |
4633 | { | |
6d49ba1b | 4634 | btrfs_leak_debug_del(&eb->leak_list); |
727011e0 CM |
4635 | kmem_cache_free(extent_buffer_cache, eb); |
4636 | } | |
4637 | ||
a26e8c9f | 4638 | int extent_buffer_under_io(struct extent_buffer *eb) |
db7f3436 JB |
4639 | { |
4640 | return (atomic_read(&eb->io_pages) || | |
4641 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4642 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4643 | } | |
4644 | ||
4645 | /* | |
55ac0139 | 4646 | * Release all pages attached to the extent buffer. |
db7f3436 | 4647 | */ |
55ac0139 | 4648 | static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb) |
db7f3436 | 4649 | { |
d64766fd NB |
4650 | int i; |
4651 | int num_pages; | |
b0132a3b | 4652 | int mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
db7f3436 JB |
4653 | |
4654 | BUG_ON(extent_buffer_under_io(eb)); | |
4655 | ||
d64766fd NB |
4656 | num_pages = num_extent_pages(eb); |
4657 | for (i = 0; i < num_pages; i++) { | |
4658 | struct page *page = eb->pages[i]; | |
db7f3436 | 4659 | |
5d2361db FL |
4660 | if (!page) |
4661 | continue; | |
4662 | if (mapped) | |
db7f3436 | 4663 | spin_lock(&page->mapping->private_lock); |
5d2361db FL |
4664 | /* |
4665 | * We do this since we'll remove the pages after we've | |
4666 | * removed the eb from the radix tree, so we could race | |
4667 | * and have this page now attached to the new eb. So | |
4668 | * only clear page_private if it's still connected to | |
4669 | * this eb. | |
4670 | */ | |
4671 | if (PagePrivate(page) && | |
4672 | page->private == (unsigned long)eb) { | |
4673 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4674 | BUG_ON(PageDirty(page)); | |
4675 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 4676 | /* |
5d2361db FL |
4677 | * We need to make sure we haven't be attached |
4678 | * to a new eb. | |
db7f3436 | 4679 | */ |
5d2361db FL |
4680 | ClearPagePrivate(page); |
4681 | set_page_private(page, 0); | |
4682 | /* One for the page private */ | |
09cbfeaf | 4683 | put_page(page); |
db7f3436 | 4684 | } |
5d2361db FL |
4685 | |
4686 | if (mapped) | |
4687 | spin_unlock(&page->mapping->private_lock); | |
4688 | ||
01327610 | 4689 | /* One for when we allocated the page */ |
09cbfeaf | 4690 | put_page(page); |
d64766fd | 4691 | } |
db7f3436 JB |
4692 | } |
4693 | ||
4694 | /* | |
4695 | * Helper for releasing the extent buffer. | |
4696 | */ | |
4697 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4698 | { | |
55ac0139 | 4699 | btrfs_release_extent_buffer_pages(eb); |
db7f3436 JB |
4700 | __free_extent_buffer(eb); |
4701 | } | |
4702 | ||
f28491e0 JB |
4703 | static struct extent_buffer * |
4704 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 4705 | unsigned long len) |
d1310b2e CM |
4706 | { |
4707 | struct extent_buffer *eb = NULL; | |
4708 | ||
d1b5c567 | 4709 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
4710 | eb->start = start; |
4711 | eb->len = len; | |
f28491e0 | 4712 | eb->fs_info = fs_info; |
815a51c7 | 4713 | eb->bflags = 0; |
bd681513 CM |
4714 | rwlock_init(&eb->lock); |
4715 | atomic_set(&eb->write_locks, 0); | |
4716 | atomic_set(&eb->read_locks, 0); | |
4717 | atomic_set(&eb->blocking_readers, 0); | |
4718 | atomic_set(&eb->blocking_writers, 0); | |
4719 | atomic_set(&eb->spinning_readers, 0); | |
4720 | atomic_set(&eb->spinning_writers, 0); | |
5b25f70f | 4721 | eb->lock_nested = 0; |
bd681513 CM |
4722 | init_waitqueue_head(&eb->write_lock_wq); |
4723 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 4724 | |
6d49ba1b ES |
4725 | btrfs_leak_debug_add(&eb->leak_list, &buffers); |
4726 | ||
3083ee2e | 4727 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 4728 | atomic_set(&eb->refs, 1); |
0b32f4bb | 4729 | atomic_set(&eb->io_pages, 0); |
727011e0 | 4730 | |
b8dae313 DS |
4731 | /* |
4732 | * Sanity checks, currently the maximum is 64k covered by 16x 4k pages | |
4733 | */ | |
4734 | BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE | |
4735 | > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
4736 | BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
d1310b2e CM |
4737 | |
4738 | return eb; | |
4739 | } | |
4740 | ||
815a51c7 JS |
4741 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) |
4742 | { | |
cc5e31a4 | 4743 | int i; |
815a51c7 JS |
4744 | struct page *p; |
4745 | struct extent_buffer *new; | |
cc5e31a4 | 4746 | int num_pages = num_extent_pages(src); |
815a51c7 | 4747 | |
3f556f78 | 4748 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
4749 | if (new == NULL) |
4750 | return NULL; | |
4751 | ||
4752 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4753 | p = alloc_page(GFP_NOFS); |
db7f3436 JB |
4754 | if (!p) { |
4755 | btrfs_release_extent_buffer(new); | |
4756 | return NULL; | |
4757 | } | |
815a51c7 JS |
4758 | attach_extent_buffer_page(new, p); |
4759 | WARN_ON(PageDirty(p)); | |
4760 | SetPageUptodate(p); | |
4761 | new->pages[i] = p; | |
fba1acf9 | 4762 | copy_page(page_address(p), page_address(src->pages[i])); |
815a51c7 JS |
4763 | } |
4764 | ||
815a51c7 | 4765 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); |
b0132a3b | 4766 | set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags); |
815a51c7 JS |
4767 | |
4768 | return new; | |
4769 | } | |
4770 | ||
0f331229 OS |
4771 | struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
4772 | u64 start, unsigned long len) | |
815a51c7 JS |
4773 | { |
4774 | struct extent_buffer *eb; | |
cc5e31a4 DS |
4775 | int num_pages; |
4776 | int i; | |
815a51c7 | 4777 | |
3f556f78 | 4778 | eb = __alloc_extent_buffer(fs_info, start, len); |
815a51c7 JS |
4779 | if (!eb) |
4780 | return NULL; | |
4781 | ||
65ad0104 | 4782 | num_pages = num_extent_pages(eb); |
815a51c7 | 4783 | for (i = 0; i < num_pages; i++) { |
9ec72677 | 4784 | eb->pages[i] = alloc_page(GFP_NOFS); |
815a51c7 JS |
4785 | if (!eb->pages[i]) |
4786 | goto err; | |
4787 | } | |
4788 | set_extent_buffer_uptodate(eb); | |
4789 | btrfs_set_header_nritems(eb, 0); | |
b0132a3b | 4790 | set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
815a51c7 JS |
4791 | |
4792 | return eb; | |
4793 | err: | |
84167d19 SB |
4794 | for (; i > 0; i--) |
4795 | __free_page(eb->pages[i - 1]); | |
815a51c7 JS |
4796 | __free_extent_buffer(eb); |
4797 | return NULL; | |
4798 | } | |
4799 | ||
0f331229 | 4800 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
da17066c | 4801 | u64 start) |
0f331229 | 4802 | { |
da17066c | 4803 | return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize); |
0f331229 OS |
4804 | } |
4805 | ||
0b32f4bb JB |
4806 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
4807 | { | |
242e18c7 | 4808 | int refs; |
0b32f4bb JB |
4809 | /* the ref bit is tricky. We have to make sure it is set |
4810 | * if we have the buffer dirty. Otherwise the | |
4811 | * code to free a buffer can end up dropping a dirty | |
4812 | * page | |
4813 | * | |
4814 | * Once the ref bit is set, it won't go away while the | |
4815 | * buffer is dirty or in writeback, and it also won't | |
4816 | * go away while we have the reference count on the | |
4817 | * eb bumped. | |
4818 | * | |
4819 | * We can't just set the ref bit without bumping the | |
4820 | * ref on the eb because free_extent_buffer might | |
4821 | * see the ref bit and try to clear it. If this happens | |
4822 | * free_extent_buffer might end up dropping our original | |
4823 | * ref by mistake and freeing the page before we are able | |
4824 | * to add one more ref. | |
4825 | * | |
4826 | * So bump the ref count first, then set the bit. If someone | |
4827 | * beat us to it, drop the ref we added. | |
4828 | */ | |
242e18c7 CM |
4829 | refs = atomic_read(&eb->refs); |
4830 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
4831 | return; | |
4832 | ||
594831c4 JB |
4833 | spin_lock(&eb->refs_lock); |
4834 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 4835 | atomic_inc(&eb->refs); |
594831c4 | 4836 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
4837 | } |
4838 | ||
2457aec6 MG |
4839 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
4840 | struct page *accessed) | |
5df4235e | 4841 | { |
cc5e31a4 | 4842 | int num_pages, i; |
5df4235e | 4843 | |
0b32f4bb JB |
4844 | check_buffer_tree_ref(eb); |
4845 | ||
65ad0104 | 4846 | num_pages = num_extent_pages(eb); |
5df4235e | 4847 | for (i = 0; i < num_pages; i++) { |
fb85fc9a DS |
4848 | struct page *p = eb->pages[i]; |
4849 | ||
2457aec6 MG |
4850 | if (p != accessed) |
4851 | mark_page_accessed(p); | |
5df4235e JB |
4852 | } |
4853 | } | |
4854 | ||
f28491e0 JB |
4855 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
4856 | u64 start) | |
452c75c3 CS |
4857 | { |
4858 | struct extent_buffer *eb; | |
4859 | ||
4860 | rcu_read_lock(); | |
f28491e0 | 4861 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
09cbfeaf | 4862 | start >> PAGE_SHIFT); |
452c75c3 CS |
4863 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
4864 | rcu_read_unlock(); | |
062c19e9 FM |
4865 | /* |
4866 | * Lock our eb's refs_lock to avoid races with | |
4867 | * free_extent_buffer. When we get our eb it might be flagged | |
4868 | * with EXTENT_BUFFER_STALE and another task running | |
4869 | * free_extent_buffer might have seen that flag set, | |
4870 | * eb->refs == 2, that the buffer isn't under IO (dirty and | |
4871 | * writeback flags not set) and it's still in the tree (flag | |
4872 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process | |
4873 | * of decrementing the extent buffer's reference count twice. | |
4874 | * So here we could race and increment the eb's reference count, | |
4875 | * clear its stale flag, mark it as dirty and drop our reference | |
4876 | * before the other task finishes executing free_extent_buffer, | |
4877 | * which would later result in an attempt to free an extent | |
4878 | * buffer that is dirty. | |
4879 | */ | |
4880 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
4881 | spin_lock(&eb->refs_lock); | |
4882 | spin_unlock(&eb->refs_lock); | |
4883 | } | |
2457aec6 | 4884 | mark_extent_buffer_accessed(eb, NULL); |
452c75c3 CS |
4885 | return eb; |
4886 | } | |
4887 | rcu_read_unlock(); | |
4888 | ||
4889 | return NULL; | |
4890 | } | |
4891 | ||
faa2dbf0 JB |
4892 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
4893 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
da17066c | 4894 | u64 start) |
faa2dbf0 JB |
4895 | { |
4896 | struct extent_buffer *eb, *exists = NULL; | |
4897 | int ret; | |
4898 | ||
4899 | eb = find_extent_buffer(fs_info, start); | |
4900 | if (eb) | |
4901 | return eb; | |
da17066c | 4902 | eb = alloc_dummy_extent_buffer(fs_info, start); |
faa2dbf0 JB |
4903 | if (!eb) |
4904 | return NULL; | |
4905 | eb->fs_info = fs_info; | |
4906 | again: | |
e1860a77 | 4907 | ret = radix_tree_preload(GFP_NOFS); |
faa2dbf0 JB |
4908 | if (ret) |
4909 | goto free_eb; | |
4910 | spin_lock(&fs_info->buffer_lock); | |
4911 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
09cbfeaf | 4912 | start >> PAGE_SHIFT, eb); |
faa2dbf0 JB |
4913 | spin_unlock(&fs_info->buffer_lock); |
4914 | radix_tree_preload_end(); | |
4915 | if (ret == -EEXIST) { | |
4916 | exists = find_extent_buffer(fs_info, start); | |
4917 | if (exists) | |
4918 | goto free_eb; | |
4919 | else | |
4920 | goto again; | |
4921 | } | |
4922 | check_buffer_tree_ref(eb); | |
4923 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
4924 | ||
4925 | /* | |
4926 | * We will free dummy extent buffer's if they come into | |
4927 | * free_extent_buffer with a ref count of 2, but if we are using this we | |
4928 | * want the buffers to stay in memory until we're done with them, so | |
4929 | * bump the ref count again. | |
4930 | */ | |
4931 | atomic_inc(&eb->refs); | |
4932 | return eb; | |
4933 | free_eb: | |
4934 | btrfs_release_extent_buffer(eb); | |
4935 | return exists; | |
4936 | } | |
4937 | #endif | |
4938 | ||
f28491e0 | 4939 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
ce3e6984 | 4940 | u64 start) |
d1310b2e | 4941 | { |
da17066c | 4942 | unsigned long len = fs_info->nodesize; |
cc5e31a4 DS |
4943 | int num_pages; |
4944 | int i; | |
09cbfeaf | 4945 | unsigned long index = start >> PAGE_SHIFT; |
d1310b2e | 4946 | struct extent_buffer *eb; |
6af118ce | 4947 | struct extent_buffer *exists = NULL; |
d1310b2e | 4948 | struct page *p; |
f28491e0 | 4949 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
d1310b2e | 4950 | int uptodate = 1; |
19fe0a8b | 4951 | int ret; |
d1310b2e | 4952 | |
da17066c | 4953 | if (!IS_ALIGNED(start, fs_info->sectorsize)) { |
c871b0f2 LB |
4954 | btrfs_err(fs_info, "bad tree block start %llu", start); |
4955 | return ERR_PTR(-EINVAL); | |
4956 | } | |
4957 | ||
f28491e0 | 4958 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 4959 | if (eb) |
6af118ce | 4960 | return eb; |
6af118ce | 4961 | |
23d79d81 | 4962 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 4963 | if (!eb) |
c871b0f2 | 4964 | return ERR_PTR(-ENOMEM); |
d1310b2e | 4965 | |
65ad0104 | 4966 | num_pages = num_extent_pages(eb); |
727011e0 | 4967 | for (i = 0; i < num_pages; i++, index++) { |
d1b5c567 | 4968 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
c871b0f2 LB |
4969 | if (!p) { |
4970 | exists = ERR_PTR(-ENOMEM); | |
6af118ce | 4971 | goto free_eb; |
c871b0f2 | 4972 | } |
4f2de97a JB |
4973 | |
4974 | spin_lock(&mapping->private_lock); | |
4975 | if (PagePrivate(p)) { | |
4976 | /* | |
4977 | * We could have already allocated an eb for this page | |
4978 | * and attached one so lets see if we can get a ref on | |
4979 | * the existing eb, and if we can we know it's good and | |
4980 | * we can just return that one, else we know we can just | |
4981 | * overwrite page->private. | |
4982 | */ | |
4983 | exists = (struct extent_buffer *)p->private; | |
4984 | if (atomic_inc_not_zero(&exists->refs)) { | |
4985 | spin_unlock(&mapping->private_lock); | |
4986 | unlock_page(p); | |
09cbfeaf | 4987 | put_page(p); |
2457aec6 | 4988 | mark_extent_buffer_accessed(exists, p); |
4f2de97a JB |
4989 | goto free_eb; |
4990 | } | |
5ca64f45 | 4991 | exists = NULL; |
4f2de97a | 4992 | |
0b32f4bb | 4993 | /* |
4f2de97a JB |
4994 | * Do this so attach doesn't complain and we need to |
4995 | * drop the ref the old guy had. | |
4996 | */ | |
4997 | ClearPagePrivate(p); | |
0b32f4bb | 4998 | WARN_ON(PageDirty(p)); |
09cbfeaf | 4999 | put_page(p); |
d1310b2e | 5000 | } |
4f2de97a JB |
5001 | attach_extent_buffer_page(eb, p); |
5002 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 5003 | WARN_ON(PageDirty(p)); |
727011e0 | 5004 | eb->pages[i] = p; |
d1310b2e CM |
5005 | if (!PageUptodate(p)) |
5006 | uptodate = 0; | |
eb14ab8e CM |
5007 | |
5008 | /* | |
b16d011e NB |
5009 | * We can't unlock the pages just yet since the extent buffer |
5010 | * hasn't been properly inserted in the radix tree, this | |
5011 | * opens a race with btree_releasepage which can free a page | |
5012 | * while we are still filling in all pages for the buffer and | |
5013 | * we could crash. | |
eb14ab8e | 5014 | */ |
d1310b2e CM |
5015 | } |
5016 | if (uptodate) | |
b4ce94de | 5017 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 5018 | again: |
e1860a77 | 5019 | ret = radix_tree_preload(GFP_NOFS); |
c871b0f2 LB |
5020 | if (ret) { |
5021 | exists = ERR_PTR(ret); | |
19fe0a8b | 5022 | goto free_eb; |
c871b0f2 | 5023 | } |
19fe0a8b | 5024 | |
f28491e0 JB |
5025 | spin_lock(&fs_info->buffer_lock); |
5026 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
09cbfeaf | 5027 | start >> PAGE_SHIFT, eb); |
f28491e0 | 5028 | spin_unlock(&fs_info->buffer_lock); |
452c75c3 | 5029 | radix_tree_preload_end(); |
19fe0a8b | 5030 | if (ret == -EEXIST) { |
f28491e0 | 5031 | exists = find_extent_buffer(fs_info, start); |
452c75c3 CS |
5032 | if (exists) |
5033 | goto free_eb; | |
5034 | else | |
115391d2 | 5035 | goto again; |
6af118ce | 5036 | } |
6af118ce | 5037 | /* add one reference for the tree */ |
0b32f4bb | 5038 | check_buffer_tree_ref(eb); |
34b41ace | 5039 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
5040 | |
5041 | /* | |
b16d011e NB |
5042 | * Now it's safe to unlock the pages because any calls to |
5043 | * btree_releasepage will correctly detect that a page belongs to a | |
5044 | * live buffer and won't free them prematurely. | |
eb14ab8e | 5045 | */ |
28187ae5 NB |
5046 | for (i = 0; i < num_pages; i++) |
5047 | unlock_page(eb->pages[i]); | |
d1310b2e CM |
5048 | return eb; |
5049 | ||
6af118ce | 5050 | free_eb: |
5ca64f45 | 5051 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
5052 | for (i = 0; i < num_pages; i++) { |
5053 | if (eb->pages[i]) | |
5054 | unlock_page(eb->pages[i]); | |
5055 | } | |
eb14ab8e | 5056 | |
897ca6e9 | 5057 | btrfs_release_extent_buffer(eb); |
6af118ce | 5058 | return exists; |
d1310b2e | 5059 | } |
d1310b2e | 5060 | |
3083ee2e JB |
5061 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
5062 | { | |
5063 | struct extent_buffer *eb = | |
5064 | container_of(head, struct extent_buffer, rcu_head); | |
5065 | ||
5066 | __free_extent_buffer(eb); | |
5067 | } | |
5068 | ||
f7a52a40 | 5069 | static int release_extent_buffer(struct extent_buffer *eb) |
3083ee2e | 5070 | { |
07e21c4d NB |
5071 | lockdep_assert_held(&eb->refs_lock); |
5072 | ||
3083ee2e JB |
5073 | WARN_ON(atomic_read(&eb->refs) == 0); |
5074 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 5075 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 5076 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 5077 | |
815a51c7 | 5078 | spin_unlock(&eb->refs_lock); |
3083ee2e | 5079 | |
f28491e0 JB |
5080 | spin_lock(&fs_info->buffer_lock); |
5081 | radix_tree_delete(&fs_info->buffer_radix, | |
09cbfeaf | 5082 | eb->start >> PAGE_SHIFT); |
f28491e0 | 5083 | spin_unlock(&fs_info->buffer_lock); |
34b41ace JB |
5084 | } else { |
5085 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 5086 | } |
3083ee2e JB |
5087 | |
5088 | /* Should be safe to release our pages at this point */ | |
55ac0139 | 5089 | btrfs_release_extent_buffer_pages(eb); |
bcb7e449 | 5090 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
b0132a3b | 5091 | if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) { |
bcb7e449 JB |
5092 | __free_extent_buffer(eb); |
5093 | return 1; | |
5094 | } | |
5095 | #endif | |
3083ee2e | 5096 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 5097 | return 1; |
3083ee2e JB |
5098 | } |
5099 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
5100 | |
5101 | return 0; | |
3083ee2e JB |
5102 | } |
5103 | ||
d1310b2e CM |
5104 | void free_extent_buffer(struct extent_buffer *eb) |
5105 | { | |
242e18c7 CM |
5106 | int refs; |
5107 | int old; | |
d1310b2e CM |
5108 | if (!eb) |
5109 | return; | |
5110 | ||
242e18c7 CM |
5111 | while (1) { |
5112 | refs = atomic_read(&eb->refs); | |
5113 | if (refs <= 3) | |
5114 | break; | |
5115 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
5116 | if (old == refs) | |
5117 | return; | |
5118 | } | |
5119 | ||
3083ee2e | 5120 | spin_lock(&eb->refs_lock); |
815a51c7 | 5121 | if (atomic_read(&eb->refs) == 2 && |
b0132a3b | 5122 | test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags)) |
815a51c7 JS |
5123 | atomic_dec(&eb->refs); |
5124 | ||
3083ee2e JB |
5125 | if (atomic_read(&eb->refs) == 2 && |
5126 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 5127 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
5128 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5129 | atomic_dec(&eb->refs); | |
5130 | ||
5131 | /* | |
5132 | * I know this is terrible, but it's temporary until we stop tracking | |
5133 | * the uptodate bits and such for the extent buffers. | |
5134 | */ | |
f7a52a40 | 5135 | release_extent_buffer(eb); |
3083ee2e JB |
5136 | } |
5137 | ||
5138 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
5139 | { | |
5140 | if (!eb) | |
d1310b2e CM |
5141 | return; |
5142 | ||
3083ee2e JB |
5143 | spin_lock(&eb->refs_lock); |
5144 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
5145 | ||
0b32f4bb | 5146 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
5147 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5148 | atomic_dec(&eb->refs); | |
f7a52a40 | 5149 | release_extent_buffer(eb); |
d1310b2e | 5150 | } |
d1310b2e | 5151 | |
1d4284bd | 5152 | void clear_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5153 | { |
cc5e31a4 DS |
5154 | int i; |
5155 | int num_pages; | |
d1310b2e CM |
5156 | struct page *page; |
5157 | ||
65ad0104 | 5158 | num_pages = num_extent_pages(eb); |
d1310b2e CM |
5159 | |
5160 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 5161 | page = eb->pages[i]; |
b9473439 | 5162 | if (!PageDirty(page)) |
d2c3f4f6 CM |
5163 | continue; |
5164 | ||
a61e6f29 | 5165 | lock_page(page); |
eb14ab8e CM |
5166 | WARN_ON(!PagePrivate(page)); |
5167 | ||
d1310b2e | 5168 | clear_page_dirty_for_io(page); |
b93b0163 | 5169 | xa_lock_irq(&page->mapping->i_pages); |
d1310b2e | 5170 | if (!PageDirty(page)) { |
b93b0163 | 5171 | radix_tree_tag_clear(&page->mapping->i_pages, |
d1310b2e CM |
5172 | page_index(page), |
5173 | PAGECACHE_TAG_DIRTY); | |
5174 | } | |
b93b0163 | 5175 | xa_unlock_irq(&page->mapping->i_pages); |
bf0da8c1 | 5176 | ClearPageError(page); |
a61e6f29 | 5177 | unlock_page(page); |
d1310b2e | 5178 | } |
0b32f4bb | 5179 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 5180 | } |
d1310b2e | 5181 | |
0b32f4bb | 5182 | int set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5183 | { |
cc5e31a4 DS |
5184 | int i; |
5185 | int num_pages; | |
b9473439 | 5186 | int was_dirty = 0; |
d1310b2e | 5187 | |
0b32f4bb JB |
5188 | check_buffer_tree_ref(eb); |
5189 | ||
b9473439 | 5190 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 5191 | |
65ad0104 | 5192 | num_pages = num_extent_pages(eb); |
3083ee2e | 5193 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
5194 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
5195 | ||
b9473439 | 5196 | for (i = 0; i < num_pages; i++) |
fb85fc9a | 5197 | set_page_dirty(eb->pages[i]); |
b9473439 | 5198 | return was_dirty; |
d1310b2e | 5199 | } |
d1310b2e | 5200 | |
69ba3927 | 5201 | void clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 | 5202 | { |
cc5e31a4 | 5203 | int i; |
1259ab75 | 5204 | struct page *page; |
cc5e31a4 | 5205 | int num_pages; |
1259ab75 | 5206 | |
b4ce94de | 5207 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 5208 | num_pages = num_extent_pages(eb); |
1259ab75 | 5209 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5210 | page = eb->pages[i]; |
33958dc6 CM |
5211 | if (page) |
5212 | ClearPageUptodate(page); | |
1259ab75 | 5213 | } |
1259ab75 CM |
5214 | } |
5215 | ||
09c25a8c | 5216 | void set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 5217 | { |
cc5e31a4 | 5218 | int i; |
d1310b2e | 5219 | struct page *page; |
cc5e31a4 | 5220 | int num_pages; |
d1310b2e | 5221 | |
0b32f4bb | 5222 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 5223 | num_pages = num_extent_pages(eb); |
d1310b2e | 5224 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5225 | page = eb->pages[i]; |
d1310b2e CM |
5226 | SetPageUptodate(page); |
5227 | } | |
d1310b2e | 5228 | } |
d1310b2e | 5229 | |
d1310b2e | 5230 | int read_extent_buffer_pages(struct extent_io_tree *tree, |
6af49dbd | 5231 | struct extent_buffer *eb, int wait, int mirror_num) |
d1310b2e | 5232 | { |
cc5e31a4 | 5233 | int i; |
d1310b2e CM |
5234 | struct page *page; |
5235 | int err; | |
5236 | int ret = 0; | |
ce9adaa5 CM |
5237 | int locked_pages = 0; |
5238 | int all_uptodate = 1; | |
cc5e31a4 | 5239 | int num_pages; |
727011e0 | 5240 | unsigned long num_reads = 0; |
a86c12c7 | 5241 | struct bio *bio = NULL; |
c8b97818 | 5242 | unsigned long bio_flags = 0; |
a86c12c7 | 5243 | |
b4ce94de | 5244 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
5245 | return 0; |
5246 | ||
65ad0104 | 5247 | num_pages = num_extent_pages(eb); |
8436ea91 | 5248 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5249 | page = eb->pages[i]; |
bb82ab88 | 5250 | if (wait == WAIT_NONE) { |
2db04966 | 5251 | if (!trylock_page(page)) |
ce9adaa5 | 5252 | goto unlock_exit; |
d1310b2e CM |
5253 | } else { |
5254 | lock_page(page); | |
5255 | } | |
ce9adaa5 | 5256 | locked_pages++; |
2571e739 LB |
5257 | } |
5258 | /* | |
5259 | * We need to firstly lock all pages to make sure that | |
5260 | * the uptodate bit of our pages won't be affected by | |
5261 | * clear_extent_buffer_uptodate(). | |
5262 | */ | |
8436ea91 | 5263 | for (i = 0; i < num_pages; i++) { |
2571e739 | 5264 | page = eb->pages[i]; |
727011e0 CM |
5265 | if (!PageUptodate(page)) { |
5266 | num_reads++; | |
ce9adaa5 | 5267 | all_uptodate = 0; |
727011e0 | 5268 | } |
ce9adaa5 | 5269 | } |
2571e739 | 5270 | |
ce9adaa5 | 5271 | if (all_uptodate) { |
8436ea91 | 5272 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
5273 | goto unlock_exit; |
5274 | } | |
5275 | ||
656f30db | 5276 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 5277 | eb->read_mirror = 0; |
0b32f4bb | 5278 | atomic_set(&eb->io_pages, num_reads); |
8436ea91 | 5279 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5280 | page = eb->pages[i]; |
baf863b9 | 5281 | |
ce9adaa5 | 5282 | if (!PageUptodate(page)) { |
baf863b9 LB |
5283 | if (ret) { |
5284 | atomic_dec(&eb->io_pages); | |
5285 | unlock_page(page); | |
5286 | continue; | |
5287 | } | |
5288 | ||
f188591e | 5289 | ClearPageError(page); |
a86c12c7 | 5290 | err = __extent_read_full_page(tree, page, |
6af49dbd | 5291 | btree_get_extent, &bio, |
d4c7ca86 | 5292 | mirror_num, &bio_flags, |
1f7ad75b | 5293 | REQ_META); |
baf863b9 | 5294 | if (err) { |
d1310b2e | 5295 | ret = err; |
baf863b9 LB |
5296 | /* |
5297 | * We use &bio in above __extent_read_full_page, | |
5298 | * so we ensure that if it returns error, the | |
5299 | * current page fails to add itself to bio and | |
5300 | * it's been unlocked. | |
5301 | * | |
5302 | * We must dec io_pages by ourselves. | |
5303 | */ | |
5304 | atomic_dec(&eb->io_pages); | |
5305 | } | |
d1310b2e CM |
5306 | } else { |
5307 | unlock_page(page); | |
5308 | } | |
5309 | } | |
5310 | ||
355808c2 | 5311 | if (bio) { |
1f7ad75b | 5312 | err = submit_one_bio(bio, mirror_num, bio_flags); |
79787eaa JM |
5313 | if (err) |
5314 | return err; | |
355808c2 | 5315 | } |
a86c12c7 | 5316 | |
bb82ab88 | 5317 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 5318 | return ret; |
d397712b | 5319 | |
8436ea91 | 5320 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5321 | page = eb->pages[i]; |
d1310b2e | 5322 | wait_on_page_locked(page); |
d397712b | 5323 | if (!PageUptodate(page)) |
d1310b2e | 5324 | ret = -EIO; |
d1310b2e | 5325 | } |
d397712b | 5326 | |
d1310b2e | 5327 | return ret; |
ce9adaa5 CM |
5328 | |
5329 | unlock_exit: | |
d397712b | 5330 | while (locked_pages > 0) { |
ce9adaa5 | 5331 | locked_pages--; |
8436ea91 JB |
5332 | page = eb->pages[locked_pages]; |
5333 | unlock_page(page); | |
ce9adaa5 CM |
5334 | } |
5335 | return ret; | |
d1310b2e | 5336 | } |
d1310b2e | 5337 | |
1cbb1f45 JM |
5338 | void read_extent_buffer(const struct extent_buffer *eb, void *dstv, |
5339 | unsigned long start, unsigned long len) | |
d1310b2e CM |
5340 | { |
5341 | size_t cur; | |
5342 | size_t offset; | |
5343 | struct page *page; | |
5344 | char *kaddr; | |
5345 | char *dst = (char *)dstv; | |
09cbfeaf KS |
5346 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5347 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e | 5348 | |
f716abd5 LB |
5349 | if (start + len > eb->len) { |
5350 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n", | |
5351 | eb->start, eb->len, start, len); | |
5352 | memset(dst, 0, len); | |
5353 | return; | |
5354 | } | |
d1310b2e | 5355 | |
09cbfeaf | 5356 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
d1310b2e | 5357 | |
d397712b | 5358 | while (len > 0) { |
fb85fc9a | 5359 | page = eb->pages[i]; |
d1310b2e | 5360 | |
09cbfeaf | 5361 | cur = min(len, (PAGE_SIZE - offset)); |
a6591715 | 5362 | kaddr = page_address(page); |
d1310b2e | 5363 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
5364 | |
5365 | dst += cur; | |
5366 | len -= cur; | |
5367 | offset = 0; | |
5368 | i++; | |
5369 | } | |
5370 | } | |
d1310b2e | 5371 | |
1cbb1f45 JM |
5372 | int read_extent_buffer_to_user(const struct extent_buffer *eb, |
5373 | void __user *dstv, | |
5374 | unsigned long start, unsigned long len) | |
550ac1d8 GH |
5375 | { |
5376 | size_t cur; | |
5377 | size_t offset; | |
5378 | struct page *page; | |
5379 | char *kaddr; | |
5380 | char __user *dst = (char __user *)dstv; | |
09cbfeaf KS |
5381 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5382 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
550ac1d8 GH |
5383 | int ret = 0; |
5384 | ||
5385 | WARN_ON(start > eb->len); | |
5386 | WARN_ON(start + len > eb->start + eb->len); | |
5387 | ||
09cbfeaf | 5388 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
550ac1d8 GH |
5389 | |
5390 | while (len > 0) { | |
fb85fc9a | 5391 | page = eb->pages[i]; |
550ac1d8 | 5392 | |
09cbfeaf | 5393 | cur = min(len, (PAGE_SIZE - offset)); |
550ac1d8 GH |
5394 | kaddr = page_address(page); |
5395 | if (copy_to_user(dst, kaddr + offset, cur)) { | |
5396 | ret = -EFAULT; | |
5397 | break; | |
5398 | } | |
5399 | ||
5400 | dst += cur; | |
5401 | len -= cur; | |
5402 | offset = 0; | |
5403 | i++; | |
5404 | } | |
5405 | ||
5406 | return ret; | |
5407 | } | |
5408 | ||
415b35a5 LB |
5409 | /* |
5410 | * return 0 if the item is found within a page. | |
5411 | * return 1 if the item spans two pages. | |
5412 | * return -EINVAL otherwise. | |
5413 | */ | |
1cbb1f45 JM |
5414 | int map_private_extent_buffer(const struct extent_buffer *eb, |
5415 | unsigned long start, unsigned long min_len, | |
5416 | char **map, unsigned long *map_start, | |
5417 | unsigned long *map_len) | |
d1310b2e | 5418 | { |
09cbfeaf | 5419 | size_t offset = start & (PAGE_SIZE - 1); |
d1310b2e CM |
5420 | char *kaddr; |
5421 | struct page *p; | |
09cbfeaf KS |
5422 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5423 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e | 5424 | unsigned long end_i = (start_offset + start + min_len - 1) >> |
09cbfeaf | 5425 | PAGE_SHIFT; |
d1310b2e | 5426 | |
f716abd5 LB |
5427 | if (start + min_len > eb->len) { |
5428 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n", | |
5429 | eb->start, eb->len, start, min_len); | |
5430 | return -EINVAL; | |
5431 | } | |
5432 | ||
d1310b2e | 5433 | if (i != end_i) |
415b35a5 | 5434 | return 1; |
d1310b2e CM |
5435 | |
5436 | if (i == 0) { | |
5437 | offset = start_offset; | |
5438 | *map_start = 0; | |
5439 | } else { | |
5440 | offset = 0; | |
09cbfeaf | 5441 | *map_start = ((u64)i << PAGE_SHIFT) - start_offset; |
d1310b2e | 5442 | } |
d397712b | 5443 | |
fb85fc9a | 5444 | p = eb->pages[i]; |
a6591715 | 5445 | kaddr = page_address(p); |
d1310b2e | 5446 | *map = kaddr + offset; |
09cbfeaf | 5447 | *map_len = PAGE_SIZE - offset; |
d1310b2e CM |
5448 | return 0; |
5449 | } | |
d1310b2e | 5450 | |
1cbb1f45 JM |
5451 | int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv, |
5452 | unsigned long start, unsigned long len) | |
d1310b2e CM |
5453 | { |
5454 | size_t cur; | |
5455 | size_t offset; | |
5456 | struct page *page; | |
5457 | char *kaddr; | |
5458 | char *ptr = (char *)ptrv; | |
09cbfeaf KS |
5459 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5460 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e CM |
5461 | int ret = 0; |
5462 | ||
5463 | WARN_ON(start > eb->len); | |
5464 | WARN_ON(start + len > eb->start + eb->len); | |
5465 | ||
09cbfeaf | 5466 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
d1310b2e | 5467 | |
d397712b | 5468 | while (len > 0) { |
fb85fc9a | 5469 | page = eb->pages[i]; |
d1310b2e | 5470 | |
09cbfeaf | 5471 | cur = min(len, (PAGE_SIZE - offset)); |
d1310b2e | 5472 | |
a6591715 | 5473 | kaddr = page_address(page); |
d1310b2e | 5474 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
5475 | if (ret) |
5476 | break; | |
5477 | ||
5478 | ptr += cur; | |
5479 | len -= cur; | |
5480 | offset = 0; | |
5481 | i++; | |
5482 | } | |
5483 | return ret; | |
5484 | } | |
d1310b2e | 5485 | |
f157bf76 DS |
5486 | void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb, |
5487 | const void *srcv) | |
5488 | { | |
5489 | char *kaddr; | |
5490 | ||
5491 | WARN_ON(!PageUptodate(eb->pages[0])); | |
5492 | kaddr = page_address(eb->pages[0]); | |
5493 | memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv, | |
5494 | BTRFS_FSID_SIZE); | |
5495 | } | |
5496 | ||
5497 | void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv) | |
5498 | { | |
5499 | char *kaddr; | |
5500 | ||
5501 | WARN_ON(!PageUptodate(eb->pages[0])); | |
5502 | kaddr = page_address(eb->pages[0]); | |
5503 | memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv, | |
5504 | BTRFS_FSID_SIZE); | |
5505 | } | |
5506 | ||
d1310b2e CM |
5507 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, |
5508 | unsigned long start, unsigned long len) | |
5509 | { | |
5510 | size_t cur; | |
5511 | size_t offset; | |
5512 | struct page *page; | |
5513 | char *kaddr; | |
5514 | char *src = (char *)srcv; | |
09cbfeaf KS |
5515 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5516 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e CM |
5517 | |
5518 | WARN_ON(start > eb->len); | |
5519 | WARN_ON(start + len > eb->start + eb->len); | |
5520 | ||
09cbfeaf | 5521 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
d1310b2e | 5522 | |
d397712b | 5523 | while (len > 0) { |
fb85fc9a | 5524 | page = eb->pages[i]; |
d1310b2e CM |
5525 | WARN_ON(!PageUptodate(page)); |
5526 | ||
09cbfeaf | 5527 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 5528 | kaddr = page_address(page); |
d1310b2e | 5529 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
5530 | |
5531 | src += cur; | |
5532 | len -= cur; | |
5533 | offset = 0; | |
5534 | i++; | |
5535 | } | |
5536 | } | |
d1310b2e | 5537 | |
b159fa28 DS |
5538 | void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start, |
5539 | unsigned long len) | |
d1310b2e CM |
5540 | { |
5541 | size_t cur; | |
5542 | size_t offset; | |
5543 | struct page *page; | |
5544 | char *kaddr; | |
09cbfeaf KS |
5545 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5546 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e CM |
5547 | |
5548 | WARN_ON(start > eb->len); | |
5549 | WARN_ON(start + len > eb->start + eb->len); | |
5550 | ||
09cbfeaf | 5551 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
d1310b2e | 5552 | |
d397712b | 5553 | while (len > 0) { |
fb85fc9a | 5554 | page = eb->pages[i]; |
d1310b2e CM |
5555 | WARN_ON(!PageUptodate(page)); |
5556 | ||
09cbfeaf | 5557 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 5558 | kaddr = page_address(page); |
b159fa28 | 5559 | memset(kaddr + offset, 0, cur); |
d1310b2e CM |
5560 | |
5561 | len -= cur; | |
5562 | offset = 0; | |
5563 | i++; | |
5564 | } | |
5565 | } | |
d1310b2e | 5566 | |
58e8012c DS |
5567 | void copy_extent_buffer_full(struct extent_buffer *dst, |
5568 | struct extent_buffer *src) | |
5569 | { | |
5570 | int i; | |
cc5e31a4 | 5571 | int num_pages; |
58e8012c DS |
5572 | |
5573 | ASSERT(dst->len == src->len); | |
5574 | ||
65ad0104 | 5575 | num_pages = num_extent_pages(dst); |
58e8012c DS |
5576 | for (i = 0; i < num_pages; i++) |
5577 | copy_page(page_address(dst->pages[i]), | |
5578 | page_address(src->pages[i])); | |
5579 | } | |
5580 | ||
d1310b2e CM |
5581 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, |
5582 | unsigned long dst_offset, unsigned long src_offset, | |
5583 | unsigned long len) | |
5584 | { | |
5585 | u64 dst_len = dst->len; | |
5586 | size_t cur; | |
5587 | size_t offset; | |
5588 | struct page *page; | |
5589 | char *kaddr; | |
09cbfeaf KS |
5590 | size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1); |
5591 | unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT; | |
d1310b2e CM |
5592 | |
5593 | WARN_ON(src->len != dst_len); | |
5594 | ||
5595 | offset = (start_offset + dst_offset) & | |
09cbfeaf | 5596 | (PAGE_SIZE - 1); |
d1310b2e | 5597 | |
d397712b | 5598 | while (len > 0) { |
fb85fc9a | 5599 | page = dst->pages[i]; |
d1310b2e CM |
5600 | WARN_ON(!PageUptodate(page)); |
5601 | ||
09cbfeaf | 5602 | cur = min(len, (unsigned long)(PAGE_SIZE - offset)); |
d1310b2e | 5603 | |
a6591715 | 5604 | kaddr = page_address(page); |
d1310b2e | 5605 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
5606 | |
5607 | src_offset += cur; | |
5608 | len -= cur; | |
5609 | offset = 0; | |
5610 | i++; | |
5611 | } | |
5612 | } | |
d1310b2e | 5613 | |
3e1e8bb7 OS |
5614 | /* |
5615 | * eb_bitmap_offset() - calculate the page and offset of the byte containing the | |
5616 | * given bit number | |
5617 | * @eb: the extent buffer | |
5618 | * @start: offset of the bitmap item in the extent buffer | |
5619 | * @nr: bit number | |
5620 | * @page_index: return index of the page in the extent buffer that contains the | |
5621 | * given bit number | |
5622 | * @page_offset: return offset into the page given by page_index | |
5623 | * | |
5624 | * This helper hides the ugliness of finding the byte in an extent buffer which | |
5625 | * contains a given bit. | |
5626 | */ | |
5627 | static inline void eb_bitmap_offset(struct extent_buffer *eb, | |
5628 | unsigned long start, unsigned long nr, | |
5629 | unsigned long *page_index, | |
5630 | size_t *page_offset) | |
5631 | { | |
09cbfeaf | 5632 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
3e1e8bb7 OS |
5633 | size_t byte_offset = BIT_BYTE(nr); |
5634 | size_t offset; | |
5635 | ||
5636 | /* | |
5637 | * The byte we want is the offset of the extent buffer + the offset of | |
5638 | * the bitmap item in the extent buffer + the offset of the byte in the | |
5639 | * bitmap item. | |
5640 | */ | |
5641 | offset = start_offset + start + byte_offset; | |
5642 | ||
09cbfeaf KS |
5643 | *page_index = offset >> PAGE_SHIFT; |
5644 | *page_offset = offset & (PAGE_SIZE - 1); | |
3e1e8bb7 OS |
5645 | } |
5646 | ||
5647 | /** | |
5648 | * extent_buffer_test_bit - determine whether a bit in a bitmap item is set | |
5649 | * @eb: the extent buffer | |
5650 | * @start: offset of the bitmap item in the extent buffer | |
5651 | * @nr: bit number to test | |
5652 | */ | |
5653 | int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start, | |
5654 | unsigned long nr) | |
5655 | { | |
2fe1d551 | 5656 | u8 *kaddr; |
3e1e8bb7 OS |
5657 | struct page *page; |
5658 | unsigned long i; | |
5659 | size_t offset; | |
5660 | ||
5661 | eb_bitmap_offset(eb, start, nr, &i, &offset); | |
5662 | page = eb->pages[i]; | |
5663 | WARN_ON(!PageUptodate(page)); | |
5664 | kaddr = page_address(page); | |
5665 | return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1))); | |
5666 | } | |
5667 | ||
5668 | /** | |
5669 | * extent_buffer_bitmap_set - set an area of a bitmap | |
5670 | * @eb: the extent buffer | |
5671 | * @start: offset of the bitmap item in the extent buffer | |
5672 | * @pos: bit number of the first bit | |
5673 | * @len: number of bits to set | |
5674 | */ | |
5675 | void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start, | |
5676 | unsigned long pos, unsigned long len) | |
5677 | { | |
2fe1d551 | 5678 | u8 *kaddr; |
3e1e8bb7 OS |
5679 | struct page *page; |
5680 | unsigned long i; | |
5681 | size_t offset; | |
5682 | const unsigned int size = pos + len; | |
5683 | int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 5684 | u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
5685 | |
5686 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5687 | page = eb->pages[i]; | |
5688 | WARN_ON(!PageUptodate(page)); | |
5689 | kaddr = page_address(page); | |
5690 | ||
5691 | while (len >= bits_to_set) { | |
5692 | kaddr[offset] |= mask_to_set; | |
5693 | len -= bits_to_set; | |
5694 | bits_to_set = BITS_PER_BYTE; | |
9c894696 | 5695 | mask_to_set = ~0; |
09cbfeaf | 5696 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5697 | offset = 0; |
5698 | page = eb->pages[++i]; | |
5699 | WARN_ON(!PageUptodate(page)); | |
5700 | kaddr = page_address(page); | |
5701 | } | |
5702 | } | |
5703 | if (len) { | |
5704 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
5705 | kaddr[offset] |= mask_to_set; | |
5706 | } | |
5707 | } | |
5708 | ||
5709 | ||
5710 | /** | |
5711 | * extent_buffer_bitmap_clear - clear an area of a bitmap | |
5712 | * @eb: the extent buffer | |
5713 | * @start: offset of the bitmap item in the extent buffer | |
5714 | * @pos: bit number of the first bit | |
5715 | * @len: number of bits to clear | |
5716 | */ | |
5717 | void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start, | |
5718 | unsigned long pos, unsigned long len) | |
5719 | { | |
2fe1d551 | 5720 | u8 *kaddr; |
3e1e8bb7 OS |
5721 | struct page *page; |
5722 | unsigned long i; | |
5723 | size_t offset; | |
5724 | const unsigned int size = pos + len; | |
5725 | int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 5726 | u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
5727 | |
5728 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5729 | page = eb->pages[i]; | |
5730 | WARN_ON(!PageUptodate(page)); | |
5731 | kaddr = page_address(page); | |
5732 | ||
5733 | while (len >= bits_to_clear) { | |
5734 | kaddr[offset] &= ~mask_to_clear; | |
5735 | len -= bits_to_clear; | |
5736 | bits_to_clear = BITS_PER_BYTE; | |
9c894696 | 5737 | mask_to_clear = ~0; |
09cbfeaf | 5738 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5739 | offset = 0; |
5740 | page = eb->pages[++i]; | |
5741 | WARN_ON(!PageUptodate(page)); | |
5742 | kaddr = page_address(page); | |
5743 | } | |
5744 | } | |
5745 | if (len) { | |
5746 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
5747 | kaddr[offset] &= ~mask_to_clear; | |
5748 | } | |
5749 | } | |
5750 | ||
3387206f ST |
5751 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
5752 | { | |
5753 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
5754 | return distance < len; | |
5755 | } | |
5756 | ||
d1310b2e CM |
5757 | static void copy_pages(struct page *dst_page, struct page *src_page, |
5758 | unsigned long dst_off, unsigned long src_off, | |
5759 | unsigned long len) | |
5760 | { | |
a6591715 | 5761 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 5762 | char *src_kaddr; |
727011e0 | 5763 | int must_memmove = 0; |
d1310b2e | 5764 | |
3387206f | 5765 | if (dst_page != src_page) { |
a6591715 | 5766 | src_kaddr = page_address(src_page); |
3387206f | 5767 | } else { |
d1310b2e | 5768 | src_kaddr = dst_kaddr; |
727011e0 CM |
5769 | if (areas_overlap(src_off, dst_off, len)) |
5770 | must_memmove = 1; | |
3387206f | 5771 | } |
d1310b2e | 5772 | |
727011e0 CM |
5773 | if (must_memmove) |
5774 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
5775 | else | |
5776 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
5777 | } |
5778 | ||
5779 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5780 | unsigned long src_offset, unsigned long len) | |
5781 | { | |
0b246afa | 5782 | struct btrfs_fs_info *fs_info = dst->fs_info; |
d1310b2e CM |
5783 | size_t cur; |
5784 | size_t dst_off_in_page; | |
5785 | size_t src_off_in_page; | |
09cbfeaf | 5786 | size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1); |
d1310b2e CM |
5787 | unsigned long dst_i; |
5788 | unsigned long src_i; | |
5789 | ||
5790 | if (src_offset + len > dst->len) { | |
0b246afa | 5791 | btrfs_err(fs_info, |
5d163e0e JM |
5792 | "memmove bogus src_offset %lu move len %lu dst len %lu", |
5793 | src_offset, len, dst->len); | |
d1310b2e CM |
5794 | BUG_ON(1); |
5795 | } | |
5796 | if (dst_offset + len > dst->len) { | |
0b246afa | 5797 | btrfs_err(fs_info, |
5d163e0e JM |
5798 | "memmove bogus dst_offset %lu move len %lu dst len %lu", |
5799 | dst_offset, len, dst->len); | |
d1310b2e CM |
5800 | BUG_ON(1); |
5801 | } | |
5802 | ||
d397712b | 5803 | while (len > 0) { |
d1310b2e | 5804 | dst_off_in_page = (start_offset + dst_offset) & |
09cbfeaf | 5805 | (PAGE_SIZE - 1); |
d1310b2e | 5806 | src_off_in_page = (start_offset + src_offset) & |
09cbfeaf | 5807 | (PAGE_SIZE - 1); |
d1310b2e | 5808 | |
09cbfeaf KS |
5809 | dst_i = (start_offset + dst_offset) >> PAGE_SHIFT; |
5810 | src_i = (start_offset + src_offset) >> PAGE_SHIFT; | |
d1310b2e | 5811 | |
09cbfeaf | 5812 | cur = min(len, (unsigned long)(PAGE_SIZE - |
d1310b2e CM |
5813 | src_off_in_page)); |
5814 | cur = min_t(unsigned long, cur, | |
09cbfeaf | 5815 | (unsigned long)(PAGE_SIZE - dst_off_in_page)); |
d1310b2e | 5816 | |
fb85fc9a | 5817 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5818 | dst_off_in_page, src_off_in_page, cur); |
5819 | ||
5820 | src_offset += cur; | |
5821 | dst_offset += cur; | |
5822 | len -= cur; | |
5823 | } | |
5824 | } | |
d1310b2e CM |
5825 | |
5826 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5827 | unsigned long src_offset, unsigned long len) | |
5828 | { | |
0b246afa | 5829 | struct btrfs_fs_info *fs_info = dst->fs_info; |
d1310b2e CM |
5830 | size_t cur; |
5831 | size_t dst_off_in_page; | |
5832 | size_t src_off_in_page; | |
5833 | unsigned long dst_end = dst_offset + len - 1; | |
5834 | unsigned long src_end = src_offset + len - 1; | |
09cbfeaf | 5835 | size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1); |
d1310b2e CM |
5836 | unsigned long dst_i; |
5837 | unsigned long src_i; | |
5838 | ||
5839 | if (src_offset + len > dst->len) { | |
0b246afa | 5840 | btrfs_err(fs_info, |
5d163e0e JM |
5841 | "memmove bogus src_offset %lu move len %lu len %lu", |
5842 | src_offset, len, dst->len); | |
d1310b2e CM |
5843 | BUG_ON(1); |
5844 | } | |
5845 | if (dst_offset + len > dst->len) { | |
0b246afa | 5846 | btrfs_err(fs_info, |
5d163e0e JM |
5847 | "memmove bogus dst_offset %lu move len %lu len %lu", |
5848 | dst_offset, len, dst->len); | |
d1310b2e CM |
5849 | BUG_ON(1); |
5850 | } | |
727011e0 | 5851 | if (dst_offset < src_offset) { |
d1310b2e CM |
5852 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
5853 | return; | |
5854 | } | |
d397712b | 5855 | while (len > 0) { |
09cbfeaf KS |
5856 | dst_i = (start_offset + dst_end) >> PAGE_SHIFT; |
5857 | src_i = (start_offset + src_end) >> PAGE_SHIFT; | |
d1310b2e CM |
5858 | |
5859 | dst_off_in_page = (start_offset + dst_end) & | |
09cbfeaf | 5860 | (PAGE_SIZE - 1); |
d1310b2e | 5861 | src_off_in_page = (start_offset + src_end) & |
09cbfeaf | 5862 | (PAGE_SIZE - 1); |
d1310b2e CM |
5863 | |
5864 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
5865 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 5866 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5867 | dst_off_in_page - cur + 1, |
5868 | src_off_in_page - cur + 1, cur); | |
5869 | ||
5870 | dst_end -= cur; | |
5871 | src_end -= cur; | |
5872 | len -= cur; | |
5873 | } | |
5874 | } | |
6af118ce | 5875 | |
f7a52a40 | 5876 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 5877 | { |
6af118ce | 5878 | struct extent_buffer *eb; |
6af118ce | 5879 | |
3083ee2e | 5880 | /* |
01327610 | 5881 | * We need to make sure nobody is attaching this page to an eb right |
3083ee2e JB |
5882 | * now. |
5883 | */ | |
5884 | spin_lock(&page->mapping->private_lock); | |
5885 | if (!PagePrivate(page)) { | |
5886 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 5887 | return 1; |
45f49bce | 5888 | } |
6af118ce | 5889 | |
3083ee2e JB |
5890 | eb = (struct extent_buffer *)page->private; |
5891 | BUG_ON(!eb); | |
19fe0a8b MX |
5892 | |
5893 | /* | |
3083ee2e JB |
5894 | * This is a little awful but should be ok, we need to make sure that |
5895 | * the eb doesn't disappear out from under us while we're looking at | |
5896 | * this page. | |
19fe0a8b | 5897 | */ |
3083ee2e | 5898 | spin_lock(&eb->refs_lock); |
0b32f4bb | 5899 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
5900 | spin_unlock(&eb->refs_lock); |
5901 | spin_unlock(&page->mapping->private_lock); | |
5902 | return 0; | |
b9473439 | 5903 | } |
3083ee2e | 5904 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 5905 | |
19fe0a8b | 5906 | /* |
3083ee2e JB |
5907 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
5908 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 5909 | */ |
3083ee2e JB |
5910 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
5911 | spin_unlock(&eb->refs_lock); | |
5912 | return 0; | |
b9473439 | 5913 | } |
19fe0a8b | 5914 | |
f7a52a40 | 5915 | return release_extent_buffer(eb); |
6af118ce | 5916 | } |