]>
Commit | Line | Data |
---|---|---|
d1310b2e CM |
1 | #include <linux/bitops.h> |
2 | #include <linux/slab.h> | |
3 | #include <linux/bio.h> | |
4 | #include <linux/mm.h> | |
d1310b2e CM |
5 | #include <linux/pagemap.h> |
6 | #include <linux/page-flags.h> | |
d1310b2e CM |
7 | #include <linux/spinlock.h> |
8 | #include <linux/blkdev.h> | |
9 | #include <linux/swap.h> | |
d1310b2e CM |
10 | #include <linux/writeback.h> |
11 | #include <linux/pagevec.h> | |
268bb0ce | 12 | #include <linux/prefetch.h> |
90a887c9 | 13 | #include <linux/cleancache.h> |
d1310b2e CM |
14 | #include "extent_io.h" |
15 | #include "extent_map.h" | |
902b22f3 DW |
16 | #include "ctree.h" |
17 | #include "btrfs_inode.h" | |
4a54c8c1 | 18 | #include "volumes.h" |
21adbd5c | 19 | #include "check-integrity.h" |
0b32f4bb | 20 | #include "locking.h" |
606686ee | 21 | #include "rcu-string.h" |
fe09e16c | 22 | #include "backref.h" |
afce772e | 23 | #include "transaction.h" |
d1310b2e | 24 | |
d1310b2e CM |
25 | static struct kmem_cache *extent_state_cache; |
26 | static struct kmem_cache *extent_buffer_cache; | |
9be3395b | 27 | static struct bio_set *btrfs_bioset; |
d1310b2e | 28 | |
27a3507d FM |
29 | static inline bool extent_state_in_tree(const struct extent_state *state) |
30 | { | |
31 | return !RB_EMPTY_NODE(&state->rb_node); | |
32 | } | |
33 | ||
6d49ba1b | 34 | #ifdef CONFIG_BTRFS_DEBUG |
d1310b2e CM |
35 | static LIST_HEAD(buffers); |
36 | static LIST_HEAD(states); | |
4bef0848 | 37 | |
d397712b | 38 | static DEFINE_SPINLOCK(leak_lock); |
6d49ba1b ES |
39 | |
40 | static inline | |
41 | void btrfs_leak_debug_add(struct list_head *new, struct list_head *head) | |
42 | { | |
43 | unsigned long flags; | |
44 | ||
45 | spin_lock_irqsave(&leak_lock, flags); | |
46 | list_add(new, head); | |
47 | spin_unlock_irqrestore(&leak_lock, flags); | |
48 | } | |
49 | ||
50 | static inline | |
51 | void btrfs_leak_debug_del(struct list_head *entry) | |
52 | { | |
53 | unsigned long flags; | |
54 | ||
55 | spin_lock_irqsave(&leak_lock, flags); | |
56 | list_del(entry); | |
57 | spin_unlock_irqrestore(&leak_lock, flags); | |
58 | } | |
59 | ||
60 | static inline | |
61 | void btrfs_leak_debug_check(void) | |
62 | { | |
63 | struct extent_state *state; | |
64 | struct extent_buffer *eb; | |
65 | ||
66 | while (!list_empty(&states)) { | |
67 | state = list_entry(states.next, struct extent_state, leak_list); | |
9ee49a04 | 68 | pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n", |
27a3507d FM |
69 | state->start, state->end, state->state, |
70 | extent_state_in_tree(state), | |
c1c9ff7c | 71 | atomic_read(&state->refs)); |
6d49ba1b ES |
72 | list_del(&state->leak_list); |
73 | kmem_cache_free(extent_state_cache, state); | |
74 | } | |
75 | ||
76 | while (!list_empty(&buffers)) { | |
77 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
efe120a0 | 78 | printk(KERN_ERR "BTRFS: buffer leak start %llu len %lu " |
c1c9ff7c GU |
79 | "refs %d\n", |
80 | eb->start, eb->len, atomic_read(&eb->refs)); | |
6d49ba1b ES |
81 | list_del(&eb->leak_list); |
82 | kmem_cache_free(extent_buffer_cache, eb); | |
83 | } | |
84 | } | |
8d599ae1 | 85 | |
a5dee37d JB |
86 | #define btrfs_debug_check_extent_io_range(tree, start, end) \ |
87 | __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end)) | |
8d599ae1 | 88 | static inline void __btrfs_debug_check_extent_io_range(const char *caller, |
a5dee37d | 89 | struct extent_io_tree *tree, u64 start, u64 end) |
8d599ae1 | 90 | { |
a5dee37d JB |
91 | struct inode *inode; |
92 | u64 isize; | |
8d599ae1 | 93 | |
a5dee37d JB |
94 | if (!tree->mapping) |
95 | return; | |
8d599ae1 | 96 | |
a5dee37d JB |
97 | inode = tree->mapping->host; |
98 | isize = i_size_read(inode); | |
8d599ae1 | 99 | if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) { |
94647322 DS |
100 | btrfs_debug_rl(BTRFS_I(inode)->root->fs_info, |
101 | "%s: ino %llu isize %llu odd range [%llu,%llu]", | |
c1c9ff7c | 102 | caller, btrfs_ino(inode), isize, start, end); |
8d599ae1 DS |
103 | } |
104 | } | |
6d49ba1b ES |
105 | #else |
106 | #define btrfs_leak_debug_add(new, head) do {} while (0) | |
107 | #define btrfs_leak_debug_del(entry) do {} while (0) | |
108 | #define btrfs_leak_debug_check() do {} while (0) | |
8d599ae1 | 109 | #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0) |
4bef0848 | 110 | #endif |
d1310b2e | 111 | |
d1310b2e CM |
112 | #define BUFFER_LRU_MAX 64 |
113 | ||
114 | struct tree_entry { | |
115 | u64 start; | |
116 | u64 end; | |
d1310b2e CM |
117 | struct rb_node rb_node; |
118 | }; | |
119 | ||
120 | struct extent_page_data { | |
121 | struct bio *bio; | |
122 | struct extent_io_tree *tree; | |
123 | get_extent_t *get_extent; | |
de0022b9 | 124 | unsigned long bio_flags; |
771ed689 CM |
125 | |
126 | /* tells writepage not to lock the state bits for this range | |
127 | * it still does the unlocking | |
128 | */ | |
ffbd517d CM |
129 | unsigned int extent_locked:1; |
130 | ||
131 | /* tells the submit_bio code to use a WRITE_SYNC */ | |
132 | unsigned int sync_io:1; | |
d1310b2e CM |
133 | }; |
134 | ||
d38ed27f QW |
135 | static void add_extent_changeset(struct extent_state *state, unsigned bits, |
136 | struct extent_changeset *changeset, | |
137 | int set) | |
138 | { | |
139 | int ret; | |
140 | ||
141 | if (!changeset) | |
142 | return; | |
143 | if (set && (state->state & bits) == bits) | |
144 | return; | |
fefdc557 QW |
145 | if (!set && (state->state & bits) == 0) |
146 | return; | |
d38ed27f QW |
147 | changeset->bytes_changed += state->end - state->start + 1; |
148 | ret = ulist_add(changeset->range_changed, state->start, state->end, | |
149 | GFP_ATOMIC); | |
150 | /* ENOMEM */ | |
151 | BUG_ON(ret < 0); | |
152 | } | |
153 | ||
0b32f4bb | 154 | static noinline void flush_write_bio(void *data); |
c2d904e0 JM |
155 | static inline struct btrfs_fs_info * |
156 | tree_fs_info(struct extent_io_tree *tree) | |
157 | { | |
a5dee37d JB |
158 | if (!tree->mapping) |
159 | return NULL; | |
c2d904e0 JM |
160 | return btrfs_sb(tree->mapping->host->i_sb); |
161 | } | |
0b32f4bb | 162 | |
d1310b2e CM |
163 | int __init extent_io_init(void) |
164 | { | |
837e1972 | 165 | extent_state_cache = kmem_cache_create("btrfs_extent_state", |
9601e3f6 | 166 | sizeof(struct extent_state), 0, |
fba4b697 | 167 | SLAB_MEM_SPREAD, NULL); |
d1310b2e CM |
168 | if (!extent_state_cache) |
169 | return -ENOMEM; | |
170 | ||
837e1972 | 171 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 | 172 | sizeof(struct extent_buffer), 0, |
fba4b697 | 173 | SLAB_MEM_SPREAD, NULL); |
d1310b2e CM |
174 | if (!extent_buffer_cache) |
175 | goto free_state_cache; | |
9be3395b CM |
176 | |
177 | btrfs_bioset = bioset_create(BIO_POOL_SIZE, | |
178 | offsetof(struct btrfs_io_bio, bio)); | |
179 | if (!btrfs_bioset) | |
180 | goto free_buffer_cache; | |
b208c2f7 DW |
181 | |
182 | if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE)) | |
183 | goto free_bioset; | |
184 | ||
d1310b2e CM |
185 | return 0; |
186 | ||
b208c2f7 DW |
187 | free_bioset: |
188 | bioset_free(btrfs_bioset); | |
189 | btrfs_bioset = NULL; | |
190 | ||
9be3395b CM |
191 | free_buffer_cache: |
192 | kmem_cache_destroy(extent_buffer_cache); | |
193 | extent_buffer_cache = NULL; | |
194 | ||
d1310b2e CM |
195 | free_state_cache: |
196 | kmem_cache_destroy(extent_state_cache); | |
9be3395b | 197 | extent_state_cache = NULL; |
d1310b2e CM |
198 | return -ENOMEM; |
199 | } | |
200 | ||
201 | void extent_io_exit(void) | |
202 | { | |
6d49ba1b | 203 | btrfs_leak_debug_check(); |
8c0a8537 KS |
204 | |
205 | /* | |
206 | * Make sure all delayed rcu free are flushed before we | |
207 | * destroy caches. | |
208 | */ | |
209 | rcu_barrier(); | |
5598e900 KM |
210 | kmem_cache_destroy(extent_state_cache); |
211 | kmem_cache_destroy(extent_buffer_cache); | |
9be3395b CM |
212 | if (btrfs_bioset) |
213 | bioset_free(btrfs_bioset); | |
d1310b2e CM |
214 | } |
215 | ||
216 | void extent_io_tree_init(struct extent_io_tree *tree, | |
f993c883 | 217 | struct address_space *mapping) |
d1310b2e | 218 | { |
6bef4d31 | 219 | tree->state = RB_ROOT; |
d1310b2e CM |
220 | tree->ops = NULL; |
221 | tree->dirty_bytes = 0; | |
70dec807 | 222 | spin_lock_init(&tree->lock); |
d1310b2e | 223 | tree->mapping = mapping; |
d1310b2e | 224 | } |
d1310b2e | 225 | |
b2950863 | 226 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
227 | { |
228 | struct extent_state *state; | |
d1310b2e CM |
229 | |
230 | state = kmem_cache_alloc(extent_state_cache, mask); | |
2b114d1d | 231 | if (!state) |
d1310b2e CM |
232 | return state; |
233 | state->state = 0; | |
47dc196a | 234 | state->failrec = NULL; |
27a3507d | 235 | RB_CLEAR_NODE(&state->rb_node); |
6d49ba1b | 236 | btrfs_leak_debug_add(&state->leak_list, &states); |
d1310b2e CM |
237 | atomic_set(&state->refs, 1); |
238 | init_waitqueue_head(&state->wq); | |
143bede5 | 239 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
240 | return state; |
241 | } | |
d1310b2e | 242 | |
4845e44f | 243 | void free_extent_state(struct extent_state *state) |
d1310b2e | 244 | { |
d1310b2e CM |
245 | if (!state) |
246 | return; | |
247 | if (atomic_dec_and_test(&state->refs)) { | |
27a3507d | 248 | WARN_ON(extent_state_in_tree(state)); |
6d49ba1b | 249 | btrfs_leak_debug_del(&state->leak_list); |
143bede5 | 250 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
251 | kmem_cache_free(extent_state_cache, state); |
252 | } | |
253 | } | |
d1310b2e | 254 | |
f2071b21 FM |
255 | static struct rb_node *tree_insert(struct rb_root *root, |
256 | struct rb_node *search_start, | |
257 | u64 offset, | |
12cfbad9 FDBM |
258 | struct rb_node *node, |
259 | struct rb_node ***p_in, | |
260 | struct rb_node **parent_in) | |
d1310b2e | 261 | { |
f2071b21 | 262 | struct rb_node **p; |
d397712b | 263 | struct rb_node *parent = NULL; |
d1310b2e CM |
264 | struct tree_entry *entry; |
265 | ||
12cfbad9 FDBM |
266 | if (p_in && parent_in) { |
267 | p = *p_in; | |
268 | parent = *parent_in; | |
269 | goto do_insert; | |
270 | } | |
271 | ||
f2071b21 | 272 | p = search_start ? &search_start : &root->rb_node; |
d397712b | 273 | while (*p) { |
d1310b2e CM |
274 | parent = *p; |
275 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
276 | ||
277 | if (offset < entry->start) | |
278 | p = &(*p)->rb_left; | |
279 | else if (offset > entry->end) | |
280 | p = &(*p)->rb_right; | |
281 | else | |
282 | return parent; | |
283 | } | |
284 | ||
12cfbad9 | 285 | do_insert: |
d1310b2e CM |
286 | rb_link_node(node, parent, p); |
287 | rb_insert_color(node, root); | |
288 | return NULL; | |
289 | } | |
290 | ||
80ea96b1 | 291 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
12cfbad9 FDBM |
292 | struct rb_node **prev_ret, |
293 | struct rb_node **next_ret, | |
294 | struct rb_node ***p_ret, | |
295 | struct rb_node **parent_ret) | |
d1310b2e | 296 | { |
80ea96b1 | 297 | struct rb_root *root = &tree->state; |
12cfbad9 | 298 | struct rb_node **n = &root->rb_node; |
d1310b2e CM |
299 | struct rb_node *prev = NULL; |
300 | struct rb_node *orig_prev = NULL; | |
301 | struct tree_entry *entry; | |
302 | struct tree_entry *prev_entry = NULL; | |
303 | ||
12cfbad9 FDBM |
304 | while (*n) { |
305 | prev = *n; | |
306 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
d1310b2e CM |
307 | prev_entry = entry; |
308 | ||
309 | if (offset < entry->start) | |
12cfbad9 | 310 | n = &(*n)->rb_left; |
d1310b2e | 311 | else if (offset > entry->end) |
12cfbad9 | 312 | n = &(*n)->rb_right; |
d397712b | 313 | else |
12cfbad9 | 314 | return *n; |
d1310b2e CM |
315 | } |
316 | ||
12cfbad9 FDBM |
317 | if (p_ret) |
318 | *p_ret = n; | |
319 | if (parent_ret) | |
320 | *parent_ret = prev; | |
321 | ||
d1310b2e CM |
322 | if (prev_ret) { |
323 | orig_prev = prev; | |
d397712b | 324 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
325 | prev = rb_next(prev); |
326 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
327 | } | |
328 | *prev_ret = prev; | |
329 | prev = orig_prev; | |
330 | } | |
331 | ||
332 | if (next_ret) { | |
333 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 334 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
335 | prev = rb_prev(prev); |
336 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
337 | } | |
338 | *next_ret = prev; | |
339 | } | |
340 | return NULL; | |
341 | } | |
342 | ||
12cfbad9 FDBM |
343 | static inline struct rb_node * |
344 | tree_search_for_insert(struct extent_io_tree *tree, | |
345 | u64 offset, | |
346 | struct rb_node ***p_ret, | |
347 | struct rb_node **parent_ret) | |
d1310b2e | 348 | { |
70dec807 | 349 | struct rb_node *prev = NULL; |
d1310b2e | 350 | struct rb_node *ret; |
70dec807 | 351 | |
12cfbad9 | 352 | ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret); |
d397712b | 353 | if (!ret) |
d1310b2e CM |
354 | return prev; |
355 | return ret; | |
356 | } | |
357 | ||
12cfbad9 FDBM |
358 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
359 | u64 offset) | |
360 | { | |
361 | return tree_search_for_insert(tree, offset, NULL, NULL); | |
362 | } | |
363 | ||
9ed74f2d JB |
364 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, |
365 | struct extent_state *other) | |
366 | { | |
367 | if (tree->ops && tree->ops->merge_extent_hook) | |
368 | tree->ops->merge_extent_hook(tree->mapping->host, new, | |
369 | other); | |
370 | } | |
371 | ||
d1310b2e CM |
372 | /* |
373 | * utility function to look for merge candidates inside a given range. | |
374 | * Any extents with matching state are merged together into a single | |
375 | * extent in the tree. Extents with EXTENT_IO in their state field | |
376 | * are not merged because the end_io handlers need to be able to do | |
377 | * operations on them without sleeping (or doing allocations/splits). | |
378 | * | |
379 | * This should be called with the tree lock held. | |
380 | */ | |
1bf85046 JM |
381 | static void merge_state(struct extent_io_tree *tree, |
382 | struct extent_state *state) | |
d1310b2e CM |
383 | { |
384 | struct extent_state *other; | |
385 | struct rb_node *other_node; | |
386 | ||
5b21f2ed | 387 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
1bf85046 | 388 | return; |
d1310b2e CM |
389 | |
390 | other_node = rb_prev(&state->rb_node); | |
391 | if (other_node) { | |
392 | other = rb_entry(other_node, struct extent_state, rb_node); | |
393 | if (other->end == state->start - 1 && | |
394 | other->state == state->state) { | |
9ed74f2d | 395 | merge_cb(tree, state, other); |
d1310b2e | 396 | state->start = other->start; |
d1310b2e | 397 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 398 | RB_CLEAR_NODE(&other->rb_node); |
d1310b2e CM |
399 | free_extent_state(other); |
400 | } | |
401 | } | |
402 | other_node = rb_next(&state->rb_node); | |
403 | if (other_node) { | |
404 | other = rb_entry(other_node, struct extent_state, rb_node); | |
405 | if (other->start == state->end + 1 && | |
406 | other->state == state->state) { | |
9ed74f2d | 407 | merge_cb(tree, state, other); |
df98b6e2 | 408 | state->end = other->end; |
df98b6e2 | 409 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 410 | RB_CLEAR_NODE(&other->rb_node); |
df98b6e2 | 411 | free_extent_state(other); |
d1310b2e CM |
412 | } |
413 | } | |
d1310b2e CM |
414 | } |
415 | ||
1bf85046 | 416 | static void set_state_cb(struct extent_io_tree *tree, |
9ee49a04 | 417 | struct extent_state *state, unsigned *bits) |
291d673e | 418 | { |
1bf85046 JM |
419 | if (tree->ops && tree->ops->set_bit_hook) |
420 | tree->ops->set_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
421 | } |
422 | ||
423 | static void clear_state_cb(struct extent_io_tree *tree, | |
9ee49a04 | 424 | struct extent_state *state, unsigned *bits) |
291d673e | 425 | { |
9ed74f2d JB |
426 | if (tree->ops && tree->ops->clear_bit_hook) |
427 | tree->ops->clear_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
428 | } |
429 | ||
3150b699 | 430 | static void set_state_bits(struct extent_io_tree *tree, |
d38ed27f QW |
431 | struct extent_state *state, unsigned *bits, |
432 | struct extent_changeset *changeset); | |
3150b699 | 433 | |
d1310b2e CM |
434 | /* |
435 | * insert an extent_state struct into the tree. 'bits' are set on the | |
436 | * struct before it is inserted. | |
437 | * | |
438 | * This may return -EEXIST if the extent is already there, in which case the | |
439 | * state struct is freed. | |
440 | * | |
441 | * The tree lock is not taken internally. This is a utility function and | |
442 | * probably isn't what you want to call (see set/clear_extent_bit). | |
443 | */ | |
444 | static int insert_state(struct extent_io_tree *tree, | |
445 | struct extent_state *state, u64 start, u64 end, | |
12cfbad9 FDBM |
446 | struct rb_node ***p, |
447 | struct rb_node **parent, | |
d38ed27f | 448 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e CM |
449 | { |
450 | struct rb_node *node; | |
451 | ||
31b1a2bd | 452 | if (end < start) |
efe120a0 | 453 | WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n", |
c1c9ff7c | 454 | end, start); |
d1310b2e CM |
455 | state->start = start; |
456 | state->end = end; | |
9ed74f2d | 457 | |
d38ed27f | 458 | set_state_bits(tree, state, bits, changeset); |
3150b699 | 459 | |
f2071b21 | 460 | node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent); |
d1310b2e CM |
461 | if (node) { |
462 | struct extent_state *found; | |
463 | found = rb_entry(node, struct extent_state, rb_node); | |
efe120a0 | 464 | printk(KERN_ERR "BTRFS: found node %llu %llu on insert of " |
c1c9ff7c GU |
465 | "%llu %llu\n", |
466 | found->start, found->end, start, end); | |
d1310b2e CM |
467 | return -EEXIST; |
468 | } | |
469 | merge_state(tree, state); | |
470 | return 0; | |
471 | } | |
472 | ||
1bf85046 | 473 | static void split_cb(struct extent_io_tree *tree, struct extent_state *orig, |
9ed74f2d JB |
474 | u64 split) |
475 | { | |
476 | if (tree->ops && tree->ops->split_extent_hook) | |
1bf85046 | 477 | tree->ops->split_extent_hook(tree->mapping->host, orig, split); |
9ed74f2d JB |
478 | } |
479 | ||
d1310b2e CM |
480 | /* |
481 | * split a given extent state struct in two, inserting the preallocated | |
482 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
483 | * offset inside 'orig' where it should be split. | |
484 | * | |
485 | * Before calling, | |
486 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
487 | * are two extent state structs in the tree: | |
488 | * prealloc: [orig->start, split - 1] | |
489 | * orig: [ split, orig->end ] | |
490 | * | |
491 | * The tree locks are not taken by this function. They need to be held | |
492 | * by the caller. | |
493 | */ | |
494 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
495 | struct extent_state *prealloc, u64 split) | |
496 | { | |
497 | struct rb_node *node; | |
9ed74f2d JB |
498 | |
499 | split_cb(tree, orig, split); | |
500 | ||
d1310b2e CM |
501 | prealloc->start = orig->start; |
502 | prealloc->end = split - 1; | |
503 | prealloc->state = orig->state; | |
504 | orig->start = split; | |
505 | ||
f2071b21 FM |
506 | node = tree_insert(&tree->state, &orig->rb_node, prealloc->end, |
507 | &prealloc->rb_node, NULL, NULL); | |
d1310b2e | 508 | if (node) { |
d1310b2e CM |
509 | free_extent_state(prealloc); |
510 | return -EEXIST; | |
511 | } | |
512 | return 0; | |
513 | } | |
514 | ||
cdc6a395 LZ |
515 | static struct extent_state *next_state(struct extent_state *state) |
516 | { | |
517 | struct rb_node *next = rb_next(&state->rb_node); | |
518 | if (next) | |
519 | return rb_entry(next, struct extent_state, rb_node); | |
520 | else | |
521 | return NULL; | |
522 | } | |
523 | ||
d1310b2e CM |
524 | /* |
525 | * utility function to clear some bits in an extent state struct. | |
1b303fc0 | 526 | * it will optionally wake up any one waiting on this state (wake == 1). |
d1310b2e CM |
527 | * |
528 | * If no bits are set on the state struct after clearing things, the | |
529 | * struct is freed and removed from the tree | |
530 | */ | |
cdc6a395 LZ |
531 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
532 | struct extent_state *state, | |
fefdc557 QW |
533 | unsigned *bits, int wake, |
534 | struct extent_changeset *changeset) | |
d1310b2e | 535 | { |
cdc6a395 | 536 | struct extent_state *next; |
9ee49a04 | 537 | unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS; |
d1310b2e | 538 | |
0ca1f7ce | 539 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
540 | u64 range = state->end - state->start + 1; |
541 | WARN_ON(range > tree->dirty_bytes); | |
542 | tree->dirty_bytes -= range; | |
543 | } | |
291d673e | 544 | clear_state_cb(tree, state, bits); |
fefdc557 | 545 | add_extent_changeset(state, bits_to_clear, changeset, 0); |
32c00aff | 546 | state->state &= ~bits_to_clear; |
d1310b2e CM |
547 | if (wake) |
548 | wake_up(&state->wq); | |
0ca1f7ce | 549 | if (state->state == 0) { |
cdc6a395 | 550 | next = next_state(state); |
27a3507d | 551 | if (extent_state_in_tree(state)) { |
d1310b2e | 552 | rb_erase(&state->rb_node, &tree->state); |
27a3507d | 553 | RB_CLEAR_NODE(&state->rb_node); |
d1310b2e CM |
554 | free_extent_state(state); |
555 | } else { | |
556 | WARN_ON(1); | |
557 | } | |
558 | } else { | |
559 | merge_state(tree, state); | |
cdc6a395 | 560 | next = next_state(state); |
d1310b2e | 561 | } |
cdc6a395 | 562 | return next; |
d1310b2e CM |
563 | } |
564 | ||
8233767a XG |
565 | static struct extent_state * |
566 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
567 | { | |
568 | if (!prealloc) | |
569 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
570 | ||
571 | return prealloc; | |
572 | } | |
573 | ||
48a3b636 | 574 | static void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
c2d904e0 JM |
575 | { |
576 | btrfs_panic(tree_fs_info(tree), err, "Locking error: " | |
577 | "Extent tree was modified by another " | |
578 | "thread while locked."); | |
579 | } | |
580 | ||
d1310b2e CM |
581 | /* |
582 | * clear some bits on a range in the tree. This may require splitting | |
583 | * or inserting elements in the tree, so the gfp mask is used to | |
584 | * indicate which allocations or sleeping are allowed. | |
585 | * | |
586 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
587 | * the given range from the tree regardless of state (ie for truncate). | |
588 | * | |
589 | * the range [start, end] is inclusive. | |
590 | * | |
6763af84 | 591 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e | 592 | */ |
fefdc557 QW |
593 | static int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
594 | unsigned bits, int wake, int delete, | |
595 | struct extent_state **cached_state, | |
596 | gfp_t mask, struct extent_changeset *changeset) | |
d1310b2e CM |
597 | { |
598 | struct extent_state *state; | |
2c64c53d | 599 | struct extent_state *cached; |
d1310b2e CM |
600 | struct extent_state *prealloc = NULL; |
601 | struct rb_node *node; | |
5c939df5 | 602 | u64 last_end; |
d1310b2e | 603 | int err; |
2ac55d41 | 604 | int clear = 0; |
d1310b2e | 605 | |
a5dee37d | 606 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 607 | |
7ee9e440 JB |
608 | if (bits & EXTENT_DELALLOC) |
609 | bits |= EXTENT_NORESERVE; | |
610 | ||
0ca1f7ce YZ |
611 | if (delete) |
612 | bits |= ~EXTENT_CTLBITS; | |
613 | bits |= EXTENT_FIRST_DELALLOC; | |
614 | ||
2ac55d41 JB |
615 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
616 | clear = 1; | |
d1310b2e | 617 | again: |
d0164adc | 618 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
c7bc6319 FM |
619 | /* |
620 | * Don't care for allocation failure here because we might end | |
621 | * up not needing the pre-allocated extent state at all, which | |
622 | * is the case if we only have in the tree extent states that | |
623 | * cover our input range and don't cover too any other range. | |
624 | * If we end up needing a new extent state we allocate it later. | |
625 | */ | |
d1310b2e | 626 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
627 | } |
628 | ||
cad321ad | 629 | spin_lock(&tree->lock); |
2c64c53d CM |
630 | if (cached_state) { |
631 | cached = *cached_state; | |
2ac55d41 JB |
632 | |
633 | if (clear) { | |
634 | *cached_state = NULL; | |
635 | cached_state = NULL; | |
636 | } | |
637 | ||
27a3507d FM |
638 | if (cached && extent_state_in_tree(cached) && |
639 | cached->start <= start && cached->end > start) { | |
2ac55d41 JB |
640 | if (clear) |
641 | atomic_dec(&cached->refs); | |
2c64c53d | 642 | state = cached; |
42daec29 | 643 | goto hit_next; |
2c64c53d | 644 | } |
2ac55d41 JB |
645 | if (clear) |
646 | free_extent_state(cached); | |
2c64c53d | 647 | } |
d1310b2e CM |
648 | /* |
649 | * this search will find the extents that end after | |
650 | * our range starts | |
651 | */ | |
80ea96b1 | 652 | node = tree_search(tree, start); |
d1310b2e CM |
653 | if (!node) |
654 | goto out; | |
655 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 656 | hit_next: |
d1310b2e CM |
657 | if (state->start > end) |
658 | goto out; | |
659 | WARN_ON(state->end < start); | |
5c939df5 | 660 | last_end = state->end; |
d1310b2e | 661 | |
0449314a | 662 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
663 | if (!(state->state & bits)) { |
664 | state = next_state(state); | |
0449314a | 665 | goto next; |
cdc6a395 | 666 | } |
0449314a | 667 | |
d1310b2e CM |
668 | /* |
669 | * | ---- desired range ---- | | |
670 | * | state | or | |
671 | * | ------------- state -------------- | | |
672 | * | |
673 | * We need to split the extent we found, and may flip | |
674 | * bits on second half. | |
675 | * | |
676 | * If the extent we found extends past our range, we | |
677 | * just split and search again. It'll get split again | |
678 | * the next time though. | |
679 | * | |
680 | * If the extent we found is inside our range, we clear | |
681 | * the desired bit on it. | |
682 | */ | |
683 | ||
684 | if (state->start < start) { | |
8233767a XG |
685 | prealloc = alloc_extent_state_atomic(prealloc); |
686 | BUG_ON(!prealloc); | |
d1310b2e | 687 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
688 | if (err) |
689 | extent_io_tree_panic(tree, err); | |
690 | ||
d1310b2e CM |
691 | prealloc = NULL; |
692 | if (err) | |
693 | goto out; | |
694 | if (state->end <= end) { | |
fefdc557 QW |
695 | state = clear_state_bit(tree, state, &bits, wake, |
696 | changeset); | |
d1ac6e41 | 697 | goto next; |
d1310b2e CM |
698 | } |
699 | goto search_again; | |
700 | } | |
701 | /* | |
702 | * | ---- desired range ---- | | |
703 | * | state | | |
704 | * We need to split the extent, and clear the bit | |
705 | * on the first half | |
706 | */ | |
707 | if (state->start <= end && state->end > end) { | |
8233767a XG |
708 | prealloc = alloc_extent_state_atomic(prealloc); |
709 | BUG_ON(!prealloc); | |
d1310b2e | 710 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
711 | if (err) |
712 | extent_io_tree_panic(tree, err); | |
713 | ||
d1310b2e CM |
714 | if (wake) |
715 | wake_up(&state->wq); | |
42daec29 | 716 | |
fefdc557 | 717 | clear_state_bit(tree, prealloc, &bits, wake, changeset); |
9ed74f2d | 718 | |
d1310b2e CM |
719 | prealloc = NULL; |
720 | goto out; | |
721 | } | |
42daec29 | 722 | |
fefdc557 | 723 | state = clear_state_bit(tree, state, &bits, wake, changeset); |
0449314a | 724 | next: |
5c939df5 YZ |
725 | if (last_end == (u64)-1) |
726 | goto out; | |
727 | start = last_end + 1; | |
cdc6a395 | 728 | if (start <= end && state && !need_resched()) |
692e5759 | 729 | goto hit_next; |
d1310b2e CM |
730 | |
731 | search_again: | |
732 | if (start > end) | |
733 | goto out; | |
cad321ad | 734 | spin_unlock(&tree->lock); |
d0164adc | 735 | if (gfpflags_allow_blocking(mask)) |
d1310b2e CM |
736 | cond_resched(); |
737 | goto again; | |
7ab5cb2a DS |
738 | |
739 | out: | |
740 | spin_unlock(&tree->lock); | |
741 | if (prealloc) | |
742 | free_extent_state(prealloc); | |
743 | ||
744 | return 0; | |
745 | ||
d1310b2e | 746 | } |
d1310b2e | 747 | |
143bede5 JM |
748 | static void wait_on_state(struct extent_io_tree *tree, |
749 | struct extent_state *state) | |
641f5219 CH |
750 | __releases(tree->lock) |
751 | __acquires(tree->lock) | |
d1310b2e CM |
752 | { |
753 | DEFINE_WAIT(wait); | |
754 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 755 | spin_unlock(&tree->lock); |
d1310b2e | 756 | schedule(); |
cad321ad | 757 | spin_lock(&tree->lock); |
d1310b2e | 758 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
759 | } |
760 | ||
761 | /* | |
762 | * waits for one or more bits to clear on a range in the state tree. | |
763 | * The range [start, end] is inclusive. | |
764 | * The tree lock is taken by this function | |
765 | */ | |
41074888 DS |
766 | static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
767 | unsigned long bits) | |
d1310b2e CM |
768 | { |
769 | struct extent_state *state; | |
770 | struct rb_node *node; | |
771 | ||
a5dee37d | 772 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 773 | |
cad321ad | 774 | spin_lock(&tree->lock); |
d1310b2e CM |
775 | again: |
776 | while (1) { | |
777 | /* | |
778 | * this search will find all the extents that end after | |
779 | * our range starts | |
780 | */ | |
80ea96b1 | 781 | node = tree_search(tree, start); |
c50d3e71 | 782 | process_node: |
d1310b2e CM |
783 | if (!node) |
784 | break; | |
785 | ||
786 | state = rb_entry(node, struct extent_state, rb_node); | |
787 | ||
788 | if (state->start > end) | |
789 | goto out; | |
790 | ||
791 | if (state->state & bits) { | |
792 | start = state->start; | |
793 | atomic_inc(&state->refs); | |
794 | wait_on_state(tree, state); | |
795 | free_extent_state(state); | |
796 | goto again; | |
797 | } | |
798 | start = state->end + 1; | |
799 | ||
800 | if (start > end) | |
801 | break; | |
802 | ||
c50d3e71 FM |
803 | if (!cond_resched_lock(&tree->lock)) { |
804 | node = rb_next(node); | |
805 | goto process_node; | |
806 | } | |
d1310b2e CM |
807 | } |
808 | out: | |
cad321ad | 809 | spin_unlock(&tree->lock); |
d1310b2e | 810 | } |
d1310b2e | 811 | |
1bf85046 | 812 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 813 | struct extent_state *state, |
d38ed27f | 814 | unsigned *bits, struct extent_changeset *changeset) |
d1310b2e | 815 | { |
9ee49a04 | 816 | unsigned bits_to_set = *bits & ~EXTENT_CTLBITS; |
9ed74f2d | 817 | |
1bf85046 | 818 | set_state_cb(tree, state, bits); |
0ca1f7ce | 819 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
820 | u64 range = state->end - state->start + 1; |
821 | tree->dirty_bytes += range; | |
822 | } | |
d38ed27f | 823 | add_extent_changeset(state, bits_to_set, changeset, 1); |
0ca1f7ce | 824 | state->state |= bits_to_set; |
d1310b2e CM |
825 | } |
826 | ||
e38e2ed7 FM |
827 | static void cache_state_if_flags(struct extent_state *state, |
828 | struct extent_state **cached_ptr, | |
9ee49a04 | 829 | unsigned flags) |
2c64c53d CM |
830 | { |
831 | if (cached_ptr && !(*cached_ptr)) { | |
e38e2ed7 | 832 | if (!flags || (state->state & flags)) { |
2c64c53d CM |
833 | *cached_ptr = state; |
834 | atomic_inc(&state->refs); | |
835 | } | |
836 | } | |
837 | } | |
838 | ||
e38e2ed7 FM |
839 | static void cache_state(struct extent_state *state, |
840 | struct extent_state **cached_ptr) | |
841 | { | |
842 | return cache_state_if_flags(state, cached_ptr, | |
843 | EXTENT_IOBITS | EXTENT_BOUNDARY); | |
844 | } | |
845 | ||
d1310b2e | 846 | /* |
1edbb734 CM |
847 | * set some bits on a range in the tree. This may require allocations or |
848 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 849 | * |
1edbb734 CM |
850 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
851 | * part of the range already has the desired bits set. The start of the | |
852 | * existing range is returned in failed_start in this case. | |
d1310b2e | 853 | * |
1edbb734 | 854 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 855 | */ |
1edbb734 | 856 | |
3fbe5c02 JM |
857 | static int __must_check |
858 | __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 859 | unsigned bits, unsigned exclusive_bits, |
41074888 | 860 | u64 *failed_start, struct extent_state **cached_state, |
d38ed27f | 861 | gfp_t mask, struct extent_changeset *changeset) |
d1310b2e CM |
862 | { |
863 | struct extent_state *state; | |
864 | struct extent_state *prealloc = NULL; | |
865 | struct rb_node *node; | |
12cfbad9 FDBM |
866 | struct rb_node **p; |
867 | struct rb_node *parent; | |
d1310b2e | 868 | int err = 0; |
d1310b2e CM |
869 | u64 last_start; |
870 | u64 last_end; | |
42daec29 | 871 | |
a5dee37d | 872 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 873 | |
0ca1f7ce | 874 | bits |= EXTENT_FIRST_DELALLOC; |
d1310b2e | 875 | again: |
d0164adc | 876 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
059f791c DS |
877 | /* |
878 | * Don't care for allocation failure here because we might end | |
879 | * up not needing the pre-allocated extent state at all, which | |
880 | * is the case if we only have in the tree extent states that | |
881 | * cover our input range and don't cover too any other range. | |
882 | * If we end up needing a new extent state we allocate it later. | |
883 | */ | |
d1310b2e | 884 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
885 | } |
886 | ||
cad321ad | 887 | spin_lock(&tree->lock); |
9655d298 CM |
888 | if (cached_state && *cached_state) { |
889 | state = *cached_state; | |
df98b6e2 | 890 | if (state->start <= start && state->end > start && |
27a3507d | 891 | extent_state_in_tree(state)) { |
9655d298 CM |
892 | node = &state->rb_node; |
893 | goto hit_next; | |
894 | } | |
895 | } | |
d1310b2e CM |
896 | /* |
897 | * this search will find all the extents that end after | |
898 | * our range starts. | |
899 | */ | |
12cfbad9 | 900 | node = tree_search_for_insert(tree, start, &p, &parent); |
d1310b2e | 901 | if (!node) { |
8233767a XG |
902 | prealloc = alloc_extent_state_atomic(prealloc); |
903 | BUG_ON(!prealloc); | |
12cfbad9 | 904 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 905 | &p, &parent, &bits, changeset); |
c2d904e0 JM |
906 | if (err) |
907 | extent_io_tree_panic(tree, err); | |
908 | ||
c42ac0bc | 909 | cache_state(prealloc, cached_state); |
d1310b2e | 910 | prealloc = NULL; |
d1310b2e CM |
911 | goto out; |
912 | } | |
d1310b2e | 913 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 914 | hit_next: |
d1310b2e CM |
915 | last_start = state->start; |
916 | last_end = state->end; | |
917 | ||
918 | /* | |
919 | * | ---- desired range ---- | | |
920 | * | state | | |
921 | * | |
922 | * Just lock what we found and keep going | |
923 | */ | |
924 | if (state->start == start && state->end <= end) { | |
1edbb734 | 925 | if (state->state & exclusive_bits) { |
d1310b2e CM |
926 | *failed_start = state->start; |
927 | err = -EEXIST; | |
928 | goto out; | |
929 | } | |
42daec29 | 930 | |
d38ed27f | 931 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 932 | cache_state(state, cached_state); |
d1310b2e | 933 | merge_state(tree, state); |
5c939df5 YZ |
934 | if (last_end == (u64)-1) |
935 | goto out; | |
936 | start = last_end + 1; | |
d1ac6e41 LB |
937 | state = next_state(state); |
938 | if (start < end && state && state->start == start && | |
939 | !need_resched()) | |
940 | goto hit_next; | |
d1310b2e CM |
941 | goto search_again; |
942 | } | |
943 | ||
944 | /* | |
945 | * | ---- desired range ---- | | |
946 | * | state | | |
947 | * or | |
948 | * | ------------- state -------------- | | |
949 | * | |
950 | * We need to split the extent we found, and may flip bits on | |
951 | * second half. | |
952 | * | |
953 | * If the extent we found extends past our | |
954 | * range, we just split and search again. It'll get split | |
955 | * again the next time though. | |
956 | * | |
957 | * If the extent we found is inside our range, we set the | |
958 | * desired bit on it. | |
959 | */ | |
960 | if (state->start < start) { | |
1edbb734 | 961 | if (state->state & exclusive_bits) { |
d1310b2e CM |
962 | *failed_start = start; |
963 | err = -EEXIST; | |
964 | goto out; | |
965 | } | |
8233767a XG |
966 | |
967 | prealloc = alloc_extent_state_atomic(prealloc); | |
968 | BUG_ON(!prealloc); | |
d1310b2e | 969 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
970 | if (err) |
971 | extent_io_tree_panic(tree, err); | |
972 | ||
d1310b2e CM |
973 | prealloc = NULL; |
974 | if (err) | |
975 | goto out; | |
976 | if (state->end <= end) { | |
d38ed27f | 977 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 978 | cache_state(state, cached_state); |
d1310b2e | 979 | merge_state(tree, state); |
5c939df5 YZ |
980 | if (last_end == (u64)-1) |
981 | goto out; | |
982 | start = last_end + 1; | |
d1ac6e41 LB |
983 | state = next_state(state); |
984 | if (start < end && state && state->start == start && | |
985 | !need_resched()) | |
986 | goto hit_next; | |
d1310b2e CM |
987 | } |
988 | goto search_again; | |
989 | } | |
990 | /* | |
991 | * | ---- desired range ---- | | |
992 | * | state | or | state | | |
993 | * | |
994 | * There's a hole, we need to insert something in it and | |
995 | * ignore the extent we found. | |
996 | */ | |
997 | if (state->start > start) { | |
998 | u64 this_end; | |
999 | if (end < last_start) | |
1000 | this_end = end; | |
1001 | else | |
d397712b | 1002 | this_end = last_start - 1; |
8233767a XG |
1003 | |
1004 | prealloc = alloc_extent_state_atomic(prealloc); | |
1005 | BUG_ON(!prealloc); | |
c7f895a2 XG |
1006 | |
1007 | /* | |
1008 | * Avoid to free 'prealloc' if it can be merged with | |
1009 | * the later extent. | |
1010 | */ | |
d1310b2e | 1011 | err = insert_state(tree, prealloc, start, this_end, |
d38ed27f | 1012 | NULL, NULL, &bits, changeset); |
c2d904e0 JM |
1013 | if (err) |
1014 | extent_io_tree_panic(tree, err); | |
1015 | ||
9ed74f2d JB |
1016 | cache_state(prealloc, cached_state); |
1017 | prealloc = NULL; | |
d1310b2e CM |
1018 | start = this_end + 1; |
1019 | goto search_again; | |
1020 | } | |
1021 | /* | |
1022 | * | ---- desired range ---- | | |
1023 | * | state | | |
1024 | * We need to split the extent, and set the bit | |
1025 | * on the first half | |
1026 | */ | |
1027 | if (state->start <= end && state->end > end) { | |
1edbb734 | 1028 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1029 | *failed_start = start; |
1030 | err = -EEXIST; | |
1031 | goto out; | |
1032 | } | |
8233767a XG |
1033 | |
1034 | prealloc = alloc_extent_state_atomic(prealloc); | |
1035 | BUG_ON(!prealloc); | |
d1310b2e | 1036 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
1037 | if (err) |
1038 | extent_io_tree_panic(tree, err); | |
d1310b2e | 1039 | |
d38ed27f | 1040 | set_state_bits(tree, prealloc, &bits, changeset); |
2c64c53d | 1041 | cache_state(prealloc, cached_state); |
d1310b2e CM |
1042 | merge_state(tree, prealloc); |
1043 | prealloc = NULL; | |
1044 | goto out; | |
1045 | } | |
1046 | ||
b5a4ba14 DS |
1047 | search_again: |
1048 | if (start > end) | |
1049 | goto out; | |
1050 | spin_unlock(&tree->lock); | |
1051 | if (gfpflags_allow_blocking(mask)) | |
1052 | cond_resched(); | |
1053 | goto again; | |
d1310b2e CM |
1054 | |
1055 | out: | |
cad321ad | 1056 | spin_unlock(&tree->lock); |
d1310b2e CM |
1057 | if (prealloc) |
1058 | free_extent_state(prealloc); | |
1059 | ||
1060 | return err; | |
1061 | ||
d1310b2e | 1062 | } |
d1310b2e | 1063 | |
41074888 | 1064 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
9ee49a04 | 1065 | unsigned bits, u64 * failed_start, |
41074888 | 1066 | struct extent_state **cached_state, gfp_t mask) |
3fbe5c02 JM |
1067 | { |
1068 | return __set_extent_bit(tree, start, end, bits, 0, failed_start, | |
d38ed27f | 1069 | cached_state, mask, NULL); |
3fbe5c02 JM |
1070 | } |
1071 | ||
1072 | ||
462d6fac | 1073 | /** |
10983f2e LB |
1074 | * convert_extent_bit - convert all bits in a given range from one bit to |
1075 | * another | |
462d6fac JB |
1076 | * @tree: the io tree to search |
1077 | * @start: the start offset in bytes | |
1078 | * @end: the end offset in bytes (inclusive) | |
1079 | * @bits: the bits to set in this range | |
1080 | * @clear_bits: the bits to clear in this range | |
e6138876 | 1081 | * @cached_state: state that we're going to cache |
462d6fac JB |
1082 | * |
1083 | * This will go through and set bits for the given range. If any states exist | |
1084 | * already in this range they are set with the given bit and cleared of the | |
1085 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
1086 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
1087 | * boundary bits like LOCK. | |
210aa277 DS |
1088 | * |
1089 | * All allocations are done with GFP_NOFS. | |
462d6fac JB |
1090 | */ |
1091 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1092 | unsigned bits, unsigned clear_bits, |
210aa277 | 1093 | struct extent_state **cached_state) |
462d6fac JB |
1094 | { |
1095 | struct extent_state *state; | |
1096 | struct extent_state *prealloc = NULL; | |
1097 | struct rb_node *node; | |
12cfbad9 FDBM |
1098 | struct rb_node **p; |
1099 | struct rb_node *parent; | |
462d6fac JB |
1100 | int err = 0; |
1101 | u64 last_start; | |
1102 | u64 last_end; | |
c8fd3de7 | 1103 | bool first_iteration = true; |
462d6fac | 1104 | |
a5dee37d | 1105 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 1106 | |
462d6fac | 1107 | again: |
210aa277 | 1108 | if (!prealloc) { |
c8fd3de7 FM |
1109 | /* |
1110 | * Best effort, don't worry if extent state allocation fails | |
1111 | * here for the first iteration. We might have a cached state | |
1112 | * that matches exactly the target range, in which case no | |
1113 | * extent state allocations are needed. We'll only know this | |
1114 | * after locking the tree. | |
1115 | */ | |
210aa277 | 1116 | prealloc = alloc_extent_state(GFP_NOFS); |
c8fd3de7 | 1117 | if (!prealloc && !first_iteration) |
462d6fac JB |
1118 | return -ENOMEM; |
1119 | } | |
1120 | ||
1121 | spin_lock(&tree->lock); | |
e6138876 JB |
1122 | if (cached_state && *cached_state) { |
1123 | state = *cached_state; | |
1124 | if (state->start <= start && state->end > start && | |
27a3507d | 1125 | extent_state_in_tree(state)) { |
e6138876 JB |
1126 | node = &state->rb_node; |
1127 | goto hit_next; | |
1128 | } | |
1129 | } | |
1130 | ||
462d6fac JB |
1131 | /* |
1132 | * this search will find all the extents that end after | |
1133 | * our range starts. | |
1134 | */ | |
12cfbad9 | 1135 | node = tree_search_for_insert(tree, start, &p, &parent); |
462d6fac JB |
1136 | if (!node) { |
1137 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1138 | if (!prealloc) { |
1139 | err = -ENOMEM; | |
1140 | goto out; | |
1141 | } | |
12cfbad9 | 1142 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 1143 | &p, &parent, &bits, NULL); |
c2d904e0 JM |
1144 | if (err) |
1145 | extent_io_tree_panic(tree, err); | |
c42ac0bc FDBM |
1146 | cache_state(prealloc, cached_state); |
1147 | prealloc = NULL; | |
462d6fac JB |
1148 | goto out; |
1149 | } | |
1150 | state = rb_entry(node, struct extent_state, rb_node); | |
1151 | hit_next: | |
1152 | last_start = state->start; | |
1153 | last_end = state->end; | |
1154 | ||
1155 | /* | |
1156 | * | ---- desired range ---- | | |
1157 | * | state | | |
1158 | * | |
1159 | * Just lock what we found and keep going | |
1160 | */ | |
1161 | if (state->start == start && state->end <= end) { | |
d38ed27f | 1162 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1163 | cache_state(state, cached_state); |
fefdc557 | 1164 | state = clear_state_bit(tree, state, &clear_bits, 0, NULL); |
462d6fac JB |
1165 | if (last_end == (u64)-1) |
1166 | goto out; | |
462d6fac | 1167 | start = last_end + 1; |
d1ac6e41 LB |
1168 | if (start < end && state && state->start == start && |
1169 | !need_resched()) | |
1170 | goto hit_next; | |
462d6fac JB |
1171 | goto search_again; |
1172 | } | |
1173 | ||
1174 | /* | |
1175 | * | ---- desired range ---- | | |
1176 | * | state | | |
1177 | * or | |
1178 | * | ------------- state -------------- | | |
1179 | * | |
1180 | * We need to split the extent we found, and may flip bits on | |
1181 | * second half. | |
1182 | * | |
1183 | * If the extent we found extends past our | |
1184 | * range, we just split and search again. It'll get split | |
1185 | * again the next time though. | |
1186 | * | |
1187 | * If the extent we found is inside our range, we set the | |
1188 | * desired bit on it. | |
1189 | */ | |
1190 | if (state->start < start) { | |
1191 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1192 | if (!prealloc) { |
1193 | err = -ENOMEM; | |
1194 | goto out; | |
1195 | } | |
462d6fac | 1196 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1197 | if (err) |
1198 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1199 | prealloc = NULL; |
1200 | if (err) | |
1201 | goto out; | |
1202 | if (state->end <= end) { | |
d38ed27f | 1203 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1204 | cache_state(state, cached_state); |
fefdc557 QW |
1205 | state = clear_state_bit(tree, state, &clear_bits, 0, |
1206 | NULL); | |
462d6fac JB |
1207 | if (last_end == (u64)-1) |
1208 | goto out; | |
1209 | start = last_end + 1; | |
d1ac6e41 LB |
1210 | if (start < end && state && state->start == start && |
1211 | !need_resched()) | |
1212 | goto hit_next; | |
462d6fac JB |
1213 | } |
1214 | goto search_again; | |
1215 | } | |
1216 | /* | |
1217 | * | ---- desired range ---- | | |
1218 | * | state | or | state | | |
1219 | * | |
1220 | * There's a hole, we need to insert something in it and | |
1221 | * ignore the extent we found. | |
1222 | */ | |
1223 | if (state->start > start) { | |
1224 | u64 this_end; | |
1225 | if (end < last_start) | |
1226 | this_end = end; | |
1227 | else | |
1228 | this_end = last_start - 1; | |
1229 | ||
1230 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1231 | if (!prealloc) { |
1232 | err = -ENOMEM; | |
1233 | goto out; | |
1234 | } | |
462d6fac JB |
1235 | |
1236 | /* | |
1237 | * Avoid to free 'prealloc' if it can be merged with | |
1238 | * the later extent. | |
1239 | */ | |
1240 | err = insert_state(tree, prealloc, start, this_end, | |
d38ed27f | 1241 | NULL, NULL, &bits, NULL); |
c2d904e0 JM |
1242 | if (err) |
1243 | extent_io_tree_panic(tree, err); | |
e6138876 | 1244 | cache_state(prealloc, cached_state); |
462d6fac JB |
1245 | prealloc = NULL; |
1246 | start = this_end + 1; | |
1247 | goto search_again; | |
1248 | } | |
1249 | /* | |
1250 | * | ---- desired range ---- | | |
1251 | * | state | | |
1252 | * We need to split the extent, and set the bit | |
1253 | * on the first half | |
1254 | */ | |
1255 | if (state->start <= end && state->end > end) { | |
1256 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1257 | if (!prealloc) { |
1258 | err = -ENOMEM; | |
1259 | goto out; | |
1260 | } | |
462d6fac JB |
1261 | |
1262 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1263 | if (err) |
1264 | extent_io_tree_panic(tree, err); | |
462d6fac | 1265 | |
d38ed27f | 1266 | set_state_bits(tree, prealloc, &bits, NULL); |
e6138876 | 1267 | cache_state(prealloc, cached_state); |
fefdc557 | 1268 | clear_state_bit(tree, prealloc, &clear_bits, 0, NULL); |
462d6fac JB |
1269 | prealloc = NULL; |
1270 | goto out; | |
1271 | } | |
1272 | ||
462d6fac JB |
1273 | search_again: |
1274 | if (start > end) | |
1275 | goto out; | |
1276 | spin_unlock(&tree->lock); | |
210aa277 | 1277 | cond_resched(); |
c8fd3de7 | 1278 | first_iteration = false; |
462d6fac | 1279 | goto again; |
462d6fac JB |
1280 | |
1281 | out: | |
1282 | spin_unlock(&tree->lock); | |
1283 | if (prealloc) | |
1284 | free_extent_state(prealloc); | |
1285 | ||
1286 | return err; | |
462d6fac JB |
1287 | } |
1288 | ||
d1310b2e | 1289 | /* wrappers around set/clear extent bit */ |
d38ed27f | 1290 | int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
2c53b912 | 1291 | unsigned bits, struct extent_changeset *changeset) |
d38ed27f QW |
1292 | { |
1293 | /* | |
1294 | * We don't support EXTENT_LOCKED yet, as current changeset will | |
1295 | * record any bits changed, so for EXTENT_LOCKED case, it will | |
1296 | * either fail with -EEXIST or changeset will record the whole | |
1297 | * range. | |
1298 | */ | |
1299 | BUG_ON(bits & EXTENT_LOCKED); | |
1300 | ||
2c53b912 | 1301 | return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS, |
d38ed27f QW |
1302 | changeset); |
1303 | } | |
1304 | ||
fefdc557 QW |
1305 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
1306 | unsigned bits, int wake, int delete, | |
1307 | struct extent_state **cached, gfp_t mask) | |
1308 | { | |
1309 | return __clear_extent_bit(tree, start, end, bits, wake, delete, | |
1310 | cached, mask, NULL); | |
1311 | } | |
1312 | ||
fefdc557 | 1313 | int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
f734c44a | 1314 | unsigned bits, struct extent_changeset *changeset) |
fefdc557 QW |
1315 | { |
1316 | /* | |
1317 | * Don't support EXTENT_LOCKED case, same reason as | |
1318 | * set_record_extent_bits(). | |
1319 | */ | |
1320 | BUG_ON(bits & EXTENT_LOCKED); | |
1321 | ||
f734c44a | 1322 | return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS, |
fefdc557 QW |
1323 | changeset); |
1324 | } | |
1325 | ||
d352ac68 CM |
1326 | /* |
1327 | * either insert or lock state struct between start and end use mask to tell | |
1328 | * us if waiting is desired. | |
1329 | */ | |
1edbb734 | 1330 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
ff13db41 | 1331 | struct extent_state **cached_state) |
d1310b2e CM |
1332 | { |
1333 | int err; | |
1334 | u64 failed_start; | |
9ee49a04 | 1335 | |
d1310b2e | 1336 | while (1) { |
ff13db41 | 1337 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, |
3fbe5c02 | 1338 | EXTENT_LOCKED, &failed_start, |
d38ed27f | 1339 | cached_state, GFP_NOFS, NULL); |
d0082371 | 1340 | if (err == -EEXIST) { |
d1310b2e CM |
1341 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1342 | start = failed_start; | |
d0082371 | 1343 | } else |
d1310b2e | 1344 | break; |
d1310b2e CM |
1345 | WARN_ON(start > end); |
1346 | } | |
1347 | return err; | |
1348 | } | |
d1310b2e | 1349 | |
d0082371 | 1350 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1351 | { |
1352 | int err; | |
1353 | u64 failed_start; | |
1354 | ||
3fbe5c02 | 1355 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
d38ed27f | 1356 | &failed_start, NULL, GFP_NOFS, NULL); |
6643558d YZ |
1357 | if (err == -EEXIST) { |
1358 | if (failed_start > start) | |
1359 | clear_extent_bit(tree, start, failed_start - 1, | |
d0082371 | 1360 | EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS); |
25179201 | 1361 | return 0; |
6643558d | 1362 | } |
25179201 JB |
1363 | return 1; |
1364 | } | |
25179201 | 1365 | |
bd1fa4f0 | 1366 | void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1367 | { |
09cbfeaf KS |
1368 | unsigned long index = start >> PAGE_SHIFT; |
1369 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1370 | struct page *page; |
1371 | ||
1372 | while (index <= end_index) { | |
1373 | page = find_get_page(inode->i_mapping, index); | |
1374 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1375 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1376 | put_page(page); |
4adaa611 CM |
1377 | index++; |
1378 | } | |
4adaa611 CM |
1379 | } |
1380 | ||
f6311572 | 1381 | void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1382 | { |
09cbfeaf KS |
1383 | unsigned long index = start >> PAGE_SHIFT; |
1384 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1385 | struct page *page; |
1386 | ||
1387 | while (index <= end_index) { | |
1388 | page = find_get_page(inode->i_mapping, index); | |
1389 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
4adaa611 | 1390 | __set_page_dirty_nobuffers(page); |
8d38633c | 1391 | account_page_redirty(page); |
09cbfeaf | 1392 | put_page(page); |
4adaa611 CM |
1393 | index++; |
1394 | } | |
4adaa611 CM |
1395 | } |
1396 | ||
d1310b2e CM |
1397 | /* |
1398 | * helper function to set both pages and extents in the tree writeback | |
1399 | */ | |
35de6db2 | 1400 | static void set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e | 1401 | { |
09cbfeaf KS |
1402 | unsigned long index = start >> PAGE_SHIFT; |
1403 | unsigned long end_index = end >> PAGE_SHIFT; | |
d1310b2e CM |
1404 | struct page *page; |
1405 | ||
1406 | while (index <= end_index) { | |
1407 | page = find_get_page(tree->mapping, index); | |
79787eaa | 1408 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ |
d1310b2e | 1409 | set_page_writeback(page); |
09cbfeaf | 1410 | put_page(page); |
d1310b2e CM |
1411 | index++; |
1412 | } | |
d1310b2e | 1413 | } |
d1310b2e | 1414 | |
d352ac68 CM |
1415 | /* find the first state struct with 'bits' set after 'start', and |
1416 | * return it. tree->lock must be held. NULL will returned if | |
1417 | * nothing was found after 'start' | |
1418 | */ | |
48a3b636 ES |
1419 | static struct extent_state * |
1420 | find_first_extent_bit_state(struct extent_io_tree *tree, | |
9ee49a04 | 1421 | u64 start, unsigned bits) |
d7fc640e CM |
1422 | { |
1423 | struct rb_node *node; | |
1424 | struct extent_state *state; | |
1425 | ||
1426 | /* | |
1427 | * this search will find all the extents that end after | |
1428 | * our range starts. | |
1429 | */ | |
1430 | node = tree_search(tree, start); | |
d397712b | 1431 | if (!node) |
d7fc640e | 1432 | goto out; |
d7fc640e | 1433 | |
d397712b | 1434 | while (1) { |
d7fc640e | 1435 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1436 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1437 | return state; |
d397712b | 1438 | |
d7fc640e CM |
1439 | node = rb_next(node); |
1440 | if (!node) | |
1441 | break; | |
1442 | } | |
1443 | out: | |
1444 | return NULL; | |
1445 | } | |
d7fc640e | 1446 | |
69261c4b XG |
1447 | /* |
1448 | * find the first offset in the io tree with 'bits' set. zero is | |
1449 | * returned if we find something, and *start_ret and *end_ret are | |
1450 | * set to reflect the state struct that was found. | |
1451 | * | |
477d7eaf | 1452 | * If nothing was found, 1 is returned. If found something, return 0. |
69261c4b XG |
1453 | */ |
1454 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
9ee49a04 | 1455 | u64 *start_ret, u64 *end_ret, unsigned bits, |
e6138876 | 1456 | struct extent_state **cached_state) |
69261c4b XG |
1457 | { |
1458 | struct extent_state *state; | |
e6138876 | 1459 | struct rb_node *n; |
69261c4b XG |
1460 | int ret = 1; |
1461 | ||
1462 | spin_lock(&tree->lock); | |
e6138876 JB |
1463 | if (cached_state && *cached_state) { |
1464 | state = *cached_state; | |
27a3507d | 1465 | if (state->end == start - 1 && extent_state_in_tree(state)) { |
e6138876 JB |
1466 | n = rb_next(&state->rb_node); |
1467 | while (n) { | |
1468 | state = rb_entry(n, struct extent_state, | |
1469 | rb_node); | |
1470 | if (state->state & bits) | |
1471 | goto got_it; | |
1472 | n = rb_next(n); | |
1473 | } | |
1474 | free_extent_state(*cached_state); | |
1475 | *cached_state = NULL; | |
1476 | goto out; | |
1477 | } | |
1478 | free_extent_state(*cached_state); | |
1479 | *cached_state = NULL; | |
1480 | } | |
1481 | ||
69261c4b | 1482 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1483 | got_it: |
69261c4b | 1484 | if (state) { |
e38e2ed7 | 1485 | cache_state_if_flags(state, cached_state, 0); |
69261c4b XG |
1486 | *start_ret = state->start; |
1487 | *end_ret = state->end; | |
1488 | ret = 0; | |
1489 | } | |
e6138876 | 1490 | out: |
69261c4b XG |
1491 | spin_unlock(&tree->lock); |
1492 | return ret; | |
1493 | } | |
1494 | ||
d352ac68 CM |
1495 | /* |
1496 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1497 | * more than 'max_bytes'. start and end are used to return the range, | |
1498 | * | |
1499 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1500 | */ | |
c8b97818 | 1501 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1502 | u64 *start, u64 *end, u64 max_bytes, |
1503 | struct extent_state **cached_state) | |
d1310b2e CM |
1504 | { |
1505 | struct rb_node *node; | |
1506 | struct extent_state *state; | |
1507 | u64 cur_start = *start; | |
1508 | u64 found = 0; | |
1509 | u64 total_bytes = 0; | |
1510 | ||
cad321ad | 1511 | spin_lock(&tree->lock); |
c8b97818 | 1512 | |
d1310b2e CM |
1513 | /* |
1514 | * this search will find all the extents that end after | |
1515 | * our range starts. | |
1516 | */ | |
80ea96b1 | 1517 | node = tree_search(tree, cur_start); |
2b114d1d | 1518 | if (!node) { |
3b951516 CM |
1519 | if (!found) |
1520 | *end = (u64)-1; | |
d1310b2e CM |
1521 | goto out; |
1522 | } | |
1523 | ||
d397712b | 1524 | while (1) { |
d1310b2e | 1525 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1526 | if (found && (state->start != cur_start || |
1527 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1528 | goto out; |
1529 | } | |
1530 | if (!(state->state & EXTENT_DELALLOC)) { | |
1531 | if (!found) | |
1532 | *end = state->end; | |
1533 | goto out; | |
1534 | } | |
c2a128d2 | 1535 | if (!found) { |
d1310b2e | 1536 | *start = state->start; |
c2a128d2 JB |
1537 | *cached_state = state; |
1538 | atomic_inc(&state->refs); | |
1539 | } | |
d1310b2e CM |
1540 | found++; |
1541 | *end = state->end; | |
1542 | cur_start = state->end + 1; | |
1543 | node = rb_next(node); | |
d1310b2e | 1544 | total_bytes += state->end - state->start + 1; |
7bf811a5 | 1545 | if (total_bytes >= max_bytes) |
573aecaf | 1546 | break; |
573aecaf | 1547 | if (!node) |
d1310b2e CM |
1548 | break; |
1549 | } | |
1550 | out: | |
cad321ad | 1551 | spin_unlock(&tree->lock); |
d1310b2e CM |
1552 | return found; |
1553 | } | |
1554 | ||
143bede5 JM |
1555 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1556 | struct page *locked_page, | |
1557 | u64 start, u64 end) | |
c8b97818 CM |
1558 | { |
1559 | int ret; | |
1560 | struct page *pages[16]; | |
09cbfeaf KS |
1561 | unsigned long index = start >> PAGE_SHIFT; |
1562 | unsigned long end_index = end >> PAGE_SHIFT; | |
c8b97818 CM |
1563 | unsigned long nr_pages = end_index - index + 1; |
1564 | int i; | |
1565 | ||
1566 | if (index == locked_page->index && end_index == index) | |
143bede5 | 1567 | return; |
c8b97818 | 1568 | |
d397712b | 1569 | while (nr_pages > 0) { |
c8b97818 | 1570 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1571 | min_t(unsigned long, nr_pages, |
1572 | ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1573 | for (i = 0; i < ret; i++) { |
1574 | if (pages[i] != locked_page) | |
1575 | unlock_page(pages[i]); | |
09cbfeaf | 1576 | put_page(pages[i]); |
c8b97818 CM |
1577 | } |
1578 | nr_pages -= ret; | |
1579 | index += ret; | |
1580 | cond_resched(); | |
1581 | } | |
c8b97818 CM |
1582 | } |
1583 | ||
1584 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1585 | struct page *locked_page, | |
1586 | u64 delalloc_start, | |
1587 | u64 delalloc_end) | |
1588 | { | |
09cbfeaf | 1589 | unsigned long index = delalloc_start >> PAGE_SHIFT; |
c8b97818 | 1590 | unsigned long start_index = index; |
09cbfeaf | 1591 | unsigned long end_index = delalloc_end >> PAGE_SHIFT; |
c8b97818 CM |
1592 | unsigned long pages_locked = 0; |
1593 | struct page *pages[16]; | |
1594 | unsigned long nrpages; | |
1595 | int ret; | |
1596 | int i; | |
1597 | ||
1598 | /* the caller is responsible for locking the start index */ | |
1599 | if (index == locked_page->index && index == end_index) | |
1600 | return 0; | |
1601 | ||
1602 | /* skip the page at the start index */ | |
1603 | nrpages = end_index - index + 1; | |
d397712b | 1604 | while (nrpages > 0) { |
c8b97818 | 1605 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1606 | min_t(unsigned long, |
1607 | nrpages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1608 | if (ret == 0) { |
1609 | ret = -EAGAIN; | |
1610 | goto done; | |
1611 | } | |
1612 | /* now we have an array of pages, lock them all */ | |
1613 | for (i = 0; i < ret; i++) { | |
1614 | /* | |
1615 | * the caller is taking responsibility for | |
1616 | * locked_page | |
1617 | */ | |
771ed689 | 1618 | if (pages[i] != locked_page) { |
c8b97818 | 1619 | lock_page(pages[i]); |
f2b1c41c CM |
1620 | if (!PageDirty(pages[i]) || |
1621 | pages[i]->mapping != inode->i_mapping) { | |
771ed689 CM |
1622 | ret = -EAGAIN; |
1623 | unlock_page(pages[i]); | |
09cbfeaf | 1624 | put_page(pages[i]); |
771ed689 CM |
1625 | goto done; |
1626 | } | |
1627 | } | |
09cbfeaf | 1628 | put_page(pages[i]); |
771ed689 | 1629 | pages_locked++; |
c8b97818 | 1630 | } |
c8b97818 CM |
1631 | nrpages -= ret; |
1632 | index += ret; | |
1633 | cond_resched(); | |
1634 | } | |
1635 | ret = 0; | |
1636 | done: | |
1637 | if (ret && pages_locked) { | |
1638 | __unlock_for_delalloc(inode, locked_page, | |
1639 | delalloc_start, | |
1640 | ((u64)(start_index + pages_locked - 1)) << | |
09cbfeaf | 1641 | PAGE_SHIFT); |
c8b97818 CM |
1642 | } |
1643 | return ret; | |
1644 | } | |
1645 | ||
1646 | /* | |
1647 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1648 | * more than 'max_bytes'. start and end are used to return the range, | |
1649 | * | |
1650 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1651 | */ | |
294e30fe JB |
1652 | STATIC u64 find_lock_delalloc_range(struct inode *inode, |
1653 | struct extent_io_tree *tree, | |
1654 | struct page *locked_page, u64 *start, | |
1655 | u64 *end, u64 max_bytes) | |
c8b97818 CM |
1656 | { |
1657 | u64 delalloc_start; | |
1658 | u64 delalloc_end; | |
1659 | u64 found; | |
9655d298 | 1660 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1661 | int ret; |
1662 | int loops = 0; | |
1663 | ||
1664 | again: | |
1665 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1666 | delalloc_start = *start; | |
1667 | delalloc_end = 0; | |
1668 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1669 | max_bytes, &cached_state); |
70b99e69 | 1670 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1671 | *start = delalloc_start; |
1672 | *end = delalloc_end; | |
c2a128d2 | 1673 | free_extent_state(cached_state); |
385fe0be | 1674 | return 0; |
c8b97818 CM |
1675 | } |
1676 | ||
70b99e69 CM |
1677 | /* |
1678 | * start comes from the offset of locked_page. We have to lock | |
1679 | * pages in order, so we can't process delalloc bytes before | |
1680 | * locked_page | |
1681 | */ | |
d397712b | 1682 | if (delalloc_start < *start) |
70b99e69 | 1683 | delalloc_start = *start; |
70b99e69 | 1684 | |
c8b97818 CM |
1685 | /* |
1686 | * make sure to limit the number of pages we try to lock down | |
c8b97818 | 1687 | */ |
7bf811a5 JB |
1688 | if (delalloc_end + 1 - delalloc_start > max_bytes) |
1689 | delalloc_end = delalloc_start + max_bytes - 1; | |
d397712b | 1690 | |
c8b97818 CM |
1691 | /* step two, lock all the pages after the page that has start */ |
1692 | ret = lock_delalloc_pages(inode, locked_page, | |
1693 | delalloc_start, delalloc_end); | |
1694 | if (ret == -EAGAIN) { | |
1695 | /* some of the pages are gone, lets avoid looping by | |
1696 | * shortening the size of the delalloc range we're searching | |
1697 | */ | |
9655d298 | 1698 | free_extent_state(cached_state); |
7d788742 | 1699 | cached_state = NULL; |
c8b97818 | 1700 | if (!loops) { |
09cbfeaf | 1701 | max_bytes = PAGE_SIZE; |
c8b97818 CM |
1702 | loops = 1; |
1703 | goto again; | |
1704 | } else { | |
1705 | found = 0; | |
1706 | goto out_failed; | |
1707 | } | |
1708 | } | |
79787eaa | 1709 | BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */ |
c8b97818 CM |
1710 | |
1711 | /* step three, lock the state bits for the whole range */ | |
ff13db41 | 1712 | lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state); |
c8b97818 CM |
1713 | |
1714 | /* then test to make sure it is all still delalloc */ | |
1715 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1716 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1717 | if (!ret) { |
9655d298 CM |
1718 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
1719 | &cached_state, GFP_NOFS); | |
c8b97818 CM |
1720 | __unlock_for_delalloc(inode, locked_page, |
1721 | delalloc_start, delalloc_end); | |
1722 | cond_resched(); | |
1723 | goto again; | |
1724 | } | |
9655d298 | 1725 | free_extent_state(cached_state); |
c8b97818 CM |
1726 | *start = delalloc_start; |
1727 | *end = delalloc_end; | |
1728 | out_failed: | |
1729 | return found; | |
1730 | } | |
1731 | ||
a9d93e17 | 1732 | void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end, |
ba8b04c1 | 1733 | u64 delalloc_end, struct page *locked_page, |
9ee49a04 | 1734 | unsigned clear_bits, |
c2790a2e | 1735 | unsigned long page_ops) |
c8b97818 | 1736 | { |
c2790a2e | 1737 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
c8b97818 CM |
1738 | int ret; |
1739 | struct page *pages[16]; | |
09cbfeaf KS |
1740 | unsigned long index = start >> PAGE_SHIFT; |
1741 | unsigned long end_index = end >> PAGE_SHIFT; | |
c8b97818 CM |
1742 | unsigned long nr_pages = end_index - index + 1; |
1743 | int i; | |
771ed689 | 1744 | |
2c64c53d | 1745 | clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); |
c2790a2e | 1746 | if (page_ops == 0) |
a9d93e17 | 1747 | return; |
c8b97818 | 1748 | |
704de49d FM |
1749 | if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0) |
1750 | mapping_set_error(inode->i_mapping, -EIO); | |
1751 | ||
d397712b | 1752 | while (nr_pages > 0) { |
c8b97818 | 1753 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1754 | min_t(unsigned long, |
1755 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 | 1756 | for (i = 0; i < ret; i++) { |
8b62b72b | 1757 | |
c2790a2e | 1758 | if (page_ops & PAGE_SET_PRIVATE2) |
8b62b72b CM |
1759 | SetPagePrivate2(pages[i]); |
1760 | ||
c8b97818 | 1761 | if (pages[i] == locked_page) { |
09cbfeaf | 1762 | put_page(pages[i]); |
c8b97818 CM |
1763 | continue; |
1764 | } | |
c2790a2e | 1765 | if (page_ops & PAGE_CLEAR_DIRTY) |
c8b97818 | 1766 | clear_page_dirty_for_io(pages[i]); |
c2790a2e | 1767 | if (page_ops & PAGE_SET_WRITEBACK) |
c8b97818 | 1768 | set_page_writeback(pages[i]); |
704de49d FM |
1769 | if (page_ops & PAGE_SET_ERROR) |
1770 | SetPageError(pages[i]); | |
c2790a2e | 1771 | if (page_ops & PAGE_END_WRITEBACK) |
c8b97818 | 1772 | end_page_writeback(pages[i]); |
c2790a2e | 1773 | if (page_ops & PAGE_UNLOCK) |
771ed689 | 1774 | unlock_page(pages[i]); |
09cbfeaf | 1775 | put_page(pages[i]); |
c8b97818 CM |
1776 | } |
1777 | nr_pages -= ret; | |
1778 | index += ret; | |
1779 | cond_resched(); | |
1780 | } | |
c8b97818 | 1781 | } |
c8b97818 | 1782 | |
d352ac68 CM |
1783 | /* |
1784 | * count the number of bytes in the tree that have a given bit(s) | |
1785 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1786 | * cached. The total number found is returned. | |
1787 | */ | |
d1310b2e CM |
1788 | u64 count_range_bits(struct extent_io_tree *tree, |
1789 | u64 *start, u64 search_end, u64 max_bytes, | |
9ee49a04 | 1790 | unsigned bits, int contig) |
d1310b2e CM |
1791 | { |
1792 | struct rb_node *node; | |
1793 | struct extent_state *state; | |
1794 | u64 cur_start = *start; | |
1795 | u64 total_bytes = 0; | |
ec29ed5b | 1796 | u64 last = 0; |
d1310b2e CM |
1797 | int found = 0; |
1798 | ||
fae7f21c | 1799 | if (WARN_ON(search_end <= cur_start)) |
d1310b2e | 1800 | return 0; |
d1310b2e | 1801 | |
cad321ad | 1802 | spin_lock(&tree->lock); |
d1310b2e CM |
1803 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1804 | total_bytes = tree->dirty_bytes; | |
1805 | goto out; | |
1806 | } | |
1807 | /* | |
1808 | * this search will find all the extents that end after | |
1809 | * our range starts. | |
1810 | */ | |
80ea96b1 | 1811 | node = tree_search(tree, cur_start); |
d397712b | 1812 | if (!node) |
d1310b2e | 1813 | goto out; |
d1310b2e | 1814 | |
d397712b | 1815 | while (1) { |
d1310b2e CM |
1816 | state = rb_entry(node, struct extent_state, rb_node); |
1817 | if (state->start > search_end) | |
1818 | break; | |
ec29ed5b CM |
1819 | if (contig && found && state->start > last + 1) |
1820 | break; | |
1821 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1822 | total_bytes += min(search_end, state->end) + 1 - |
1823 | max(cur_start, state->start); | |
1824 | if (total_bytes >= max_bytes) | |
1825 | break; | |
1826 | if (!found) { | |
af60bed2 | 1827 | *start = max(cur_start, state->start); |
d1310b2e CM |
1828 | found = 1; |
1829 | } | |
ec29ed5b CM |
1830 | last = state->end; |
1831 | } else if (contig && found) { | |
1832 | break; | |
d1310b2e CM |
1833 | } |
1834 | node = rb_next(node); | |
1835 | if (!node) | |
1836 | break; | |
1837 | } | |
1838 | out: | |
cad321ad | 1839 | spin_unlock(&tree->lock); |
d1310b2e CM |
1840 | return total_bytes; |
1841 | } | |
b2950863 | 1842 | |
d352ac68 CM |
1843 | /* |
1844 | * set the private field for a given byte offset in the tree. If there isn't | |
1845 | * an extent_state there already, this does nothing. | |
1846 | */ | |
f827ba9a | 1847 | static noinline int set_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 | state->failrec = failrec; |
d1310b2e | 1870 | out: |
cad321ad | 1871 | spin_unlock(&tree->lock); |
d1310b2e CM |
1872 | return ret; |
1873 | } | |
1874 | ||
f827ba9a | 1875 | static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start, |
47dc196a | 1876 | struct io_failure_record **failrec) |
d1310b2e CM |
1877 | { |
1878 | struct rb_node *node; | |
1879 | struct extent_state *state; | |
1880 | int ret = 0; | |
1881 | ||
cad321ad | 1882 | spin_lock(&tree->lock); |
d1310b2e CM |
1883 | /* |
1884 | * this search will find all the extents that end after | |
1885 | * our range starts. | |
1886 | */ | |
80ea96b1 | 1887 | node = tree_search(tree, start); |
2b114d1d | 1888 | if (!node) { |
d1310b2e CM |
1889 | ret = -ENOENT; |
1890 | goto out; | |
1891 | } | |
1892 | state = rb_entry(node, struct extent_state, rb_node); | |
1893 | if (state->start != start) { | |
1894 | ret = -ENOENT; | |
1895 | goto out; | |
1896 | } | |
47dc196a | 1897 | *failrec = state->failrec; |
d1310b2e | 1898 | out: |
cad321ad | 1899 | spin_unlock(&tree->lock); |
d1310b2e CM |
1900 | return ret; |
1901 | } | |
1902 | ||
1903 | /* | |
1904 | * searches a range in the state tree for a given mask. | |
70dec807 | 1905 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1906 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1907 | * range is found set. | |
1908 | */ | |
1909 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9ee49a04 | 1910 | unsigned bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1911 | { |
1912 | struct extent_state *state = NULL; | |
1913 | struct rb_node *node; | |
1914 | int bitset = 0; | |
d1310b2e | 1915 | |
cad321ad | 1916 | spin_lock(&tree->lock); |
27a3507d | 1917 | if (cached && extent_state_in_tree(cached) && cached->start <= start && |
df98b6e2 | 1918 | cached->end > start) |
9655d298 CM |
1919 | node = &cached->rb_node; |
1920 | else | |
1921 | node = tree_search(tree, start); | |
d1310b2e CM |
1922 | while (node && start <= end) { |
1923 | state = rb_entry(node, struct extent_state, rb_node); | |
1924 | ||
1925 | if (filled && state->start > start) { | |
1926 | bitset = 0; | |
1927 | break; | |
1928 | } | |
1929 | ||
1930 | if (state->start > end) | |
1931 | break; | |
1932 | ||
1933 | if (state->state & bits) { | |
1934 | bitset = 1; | |
1935 | if (!filled) | |
1936 | break; | |
1937 | } else if (filled) { | |
1938 | bitset = 0; | |
1939 | break; | |
1940 | } | |
46562cec CM |
1941 | |
1942 | if (state->end == (u64)-1) | |
1943 | break; | |
1944 | ||
d1310b2e CM |
1945 | start = state->end + 1; |
1946 | if (start > end) | |
1947 | break; | |
1948 | node = rb_next(node); | |
1949 | if (!node) { | |
1950 | if (filled) | |
1951 | bitset = 0; | |
1952 | break; | |
1953 | } | |
1954 | } | |
cad321ad | 1955 | spin_unlock(&tree->lock); |
d1310b2e CM |
1956 | return bitset; |
1957 | } | |
d1310b2e CM |
1958 | |
1959 | /* | |
1960 | * helper function to set a given page up to date if all the | |
1961 | * extents in the tree for that page are up to date | |
1962 | */ | |
143bede5 | 1963 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e | 1964 | { |
4eee4fa4 | 1965 | u64 start = page_offset(page); |
09cbfeaf | 1966 | u64 end = start + PAGE_SIZE - 1; |
9655d298 | 1967 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 1968 | SetPageUptodate(page); |
d1310b2e CM |
1969 | } |
1970 | ||
8b110e39 | 1971 | int free_io_failure(struct inode *inode, struct io_failure_record *rec) |
4a54c8c1 JS |
1972 | { |
1973 | int ret; | |
1974 | int err = 0; | |
1975 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
1976 | ||
47dc196a | 1977 | set_state_failrec(failure_tree, rec->start, NULL); |
4a54c8c1 JS |
1978 | ret = clear_extent_bits(failure_tree, rec->start, |
1979 | rec->start + rec->len - 1, | |
91166212 | 1980 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 JS |
1981 | if (ret) |
1982 | err = ret; | |
1983 | ||
53b381b3 DW |
1984 | ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start, |
1985 | rec->start + rec->len - 1, | |
91166212 | 1986 | EXTENT_DAMAGED); |
53b381b3 DW |
1987 | if (ret && !err) |
1988 | err = ret; | |
4a54c8c1 JS |
1989 | |
1990 | kfree(rec); | |
1991 | return err; | |
1992 | } | |
1993 | ||
4a54c8c1 JS |
1994 | /* |
1995 | * this bypasses the standard btrfs submit functions deliberately, as | |
1996 | * the standard behavior is to write all copies in a raid setup. here we only | |
1997 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
1998 | * submit_bio directly. | |
3ec706c8 | 1999 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
2000 | * actually prevents the read that triggered the error from finishing. |
2001 | * currently, there can be no more than two copies of every data bit. thus, | |
2002 | * exactly one rewrite is required. | |
2003 | */ | |
1203b681 MX |
2004 | int repair_io_failure(struct inode *inode, u64 start, u64 length, u64 logical, |
2005 | struct page *page, unsigned int pg_offset, int mirror_num) | |
4a54c8c1 | 2006 | { |
1203b681 | 2007 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; |
4a54c8c1 JS |
2008 | struct bio *bio; |
2009 | struct btrfs_device *dev; | |
4a54c8c1 JS |
2010 | u64 map_length = 0; |
2011 | u64 sector; | |
2012 | struct btrfs_bio *bbio = NULL; | |
53b381b3 | 2013 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
4a54c8c1 JS |
2014 | int ret; |
2015 | ||
908960c6 | 2016 | ASSERT(!(fs_info->sb->s_flags & MS_RDONLY)); |
4a54c8c1 JS |
2017 | BUG_ON(!mirror_num); |
2018 | ||
53b381b3 DW |
2019 | /* we can't repair anything in raid56 yet */ |
2020 | if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num)) | |
2021 | return 0; | |
2022 | ||
9be3395b | 2023 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
4a54c8c1 JS |
2024 | if (!bio) |
2025 | return -EIO; | |
4f024f37 | 2026 | bio->bi_iter.bi_size = 0; |
4a54c8c1 JS |
2027 | map_length = length; |
2028 | ||
b5de8d0d FM |
2029 | /* |
2030 | * Avoid races with device replace and make sure our bbio has devices | |
2031 | * associated to its stripes that don't go away while we are doing the | |
2032 | * read repair operation. | |
2033 | */ | |
2034 | btrfs_bio_counter_inc_blocked(fs_info); | |
3ec706c8 | 2035 | ret = btrfs_map_block(fs_info, WRITE, logical, |
4a54c8c1 JS |
2036 | &map_length, &bbio, mirror_num); |
2037 | if (ret) { | |
b5de8d0d | 2038 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2039 | bio_put(bio); |
2040 | return -EIO; | |
2041 | } | |
2042 | BUG_ON(mirror_num != bbio->mirror_num); | |
2043 | sector = bbio->stripes[mirror_num-1].physical >> 9; | |
4f024f37 | 2044 | bio->bi_iter.bi_sector = sector; |
4a54c8c1 | 2045 | dev = bbio->stripes[mirror_num-1].dev; |
6e9606d2 | 2046 | btrfs_put_bbio(bbio); |
4a54c8c1 | 2047 | if (!dev || !dev->bdev || !dev->writeable) { |
b5de8d0d | 2048 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2049 | bio_put(bio); |
2050 | return -EIO; | |
2051 | } | |
2052 | bio->bi_bdev = dev->bdev; | |
b571bc60 | 2053 | bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_SYNC); |
ffdd2018 | 2054 | bio_add_page(bio, page, length, pg_offset); |
4a54c8c1 | 2055 | |
4e49ea4a | 2056 | if (btrfsic_submit_bio_wait(bio)) { |
4a54c8c1 | 2057 | /* try to remap that extent elsewhere? */ |
b5de8d0d | 2058 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 | 2059 | bio_put(bio); |
442a4f63 | 2060 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
2061 | return -EIO; |
2062 | } | |
2063 | ||
b14af3b4 DS |
2064 | btrfs_info_rl_in_rcu(fs_info, |
2065 | "read error corrected: ino %llu off %llu (dev %s sector %llu)", | |
1203b681 MX |
2066 | btrfs_ino(inode), start, |
2067 | rcu_str_deref(dev->name), sector); | |
b5de8d0d | 2068 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2069 | bio_put(bio); |
2070 | return 0; | |
2071 | } | |
2072 | ||
ea466794 JB |
2073 | int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb, |
2074 | int mirror_num) | |
2075 | { | |
ea466794 JB |
2076 | u64 start = eb->start; |
2077 | unsigned long i, num_pages = num_extent_pages(eb->start, eb->len); | |
d95603b2 | 2078 | int ret = 0; |
ea466794 | 2079 | |
908960c6 ID |
2080 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
2081 | return -EROFS; | |
2082 | ||
ea466794 | 2083 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 2084 | struct page *p = eb->pages[i]; |
1203b681 MX |
2085 | |
2086 | ret = repair_io_failure(root->fs_info->btree_inode, start, | |
09cbfeaf | 2087 | PAGE_SIZE, start, p, |
1203b681 | 2088 | start - page_offset(p), mirror_num); |
ea466794 JB |
2089 | if (ret) |
2090 | break; | |
09cbfeaf | 2091 | start += PAGE_SIZE; |
ea466794 JB |
2092 | } |
2093 | ||
2094 | return ret; | |
2095 | } | |
2096 | ||
4a54c8c1 JS |
2097 | /* |
2098 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2099 | * to see if we need to process or clean up an io_failure_record | |
2100 | */ | |
8b110e39 MX |
2101 | int clean_io_failure(struct inode *inode, u64 start, struct page *page, |
2102 | unsigned int pg_offset) | |
4a54c8c1 JS |
2103 | { |
2104 | u64 private; | |
4a54c8c1 | 2105 | struct io_failure_record *failrec; |
908960c6 | 2106 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; |
4a54c8c1 JS |
2107 | struct extent_state *state; |
2108 | int num_copies; | |
4a54c8c1 | 2109 | int ret; |
4a54c8c1 JS |
2110 | |
2111 | private = 0; | |
2112 | ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
2113 | (u64)-1, 1, EXTENT_DIRTY, 0); | |
2114 | if (!ret) | |
2115 | return 0; | |
2116 | ||
47dc196a DS |
2117 | ret = get_state_failrec(&BTRFS_I(inode)->io_failure_tree, start, |
2118 | &failrec); | |
4a54c8c1 JS |
2119 | if (ret) |
2120 | return 0; | |
2121 | ||
4a54c8c1 JS |
2122 | BUG_ON(!failrec->this_mirror); |
2123 | ||
2124 | if (failrec->in_validation) { | |
2125 | /* there was no real error, just free the record */ | |
2126 | pr_debug("clean_io_failure: freeing dummy error at %llu\n", | |
2127 | failrec->start); | |
4a54c8c1 JS |
2128 | goto out; |
2129 | } | |
908960c6 ID |
2130 | if (fs_info->sb->s_flags & MS_RDONLY) |
2131 | goto out; | |
4a54c8c1 JS |
2132 | |
2133 | spin_lock(&BTRFS_I(inode)->io_tree.lock); | |
2134 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, | |
2135 | failrec->start, | |
2136 | EXTENT_LOCKED); | |
2137 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); | |
2138 | ||
883d0de4 MX |
2139 | if (state && state->start <= failrec->start && |
2140 | state->end >= failrec->start + failrec->len - 1) { | |
3ec706c8 SB |
2141 | num_copies = btrfs_num_copies(fs_info, failrec->logical, |
2142 | failrec->len); | |
4a54c8c1 | 2143 | if (num_copies > 1) { |
1203b681 | 2144 | repair_io_failure(inode, start, failrec->len, |
454ff3de | 2145 | failrec->logical, page, |
1203b681 | 2146 | pg_offset, failrec->failed_mirror); |
4a54c8c1 JS |
2147 | } |
2148 | } | |
2149 | ||
2150 | out: | |
454ff3de | 2151 | free_io_failure(inode, failrec); |
4a54c8c1 | 2152 | |
454ff3de | 2153 | return 0; |
4a54c8c1 JS |
2154 | } |
2155 | ||
f612496b MX |
2156 | /* |
2157 | * Can be called when | |
2158 | * - hold extent lock | |
2159 | * - under ordered extent | |
2160 | * - the inode is freeing | |
2161 | */ | |
2162 | void btrfs_free_io_failure_record(struct inode *inode, u64 start, u64 end) | |
2163 | { | |
2164 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
2165 | struct io_failure_record *failrec; | |
2166 | struct extent_state *state, *next; | |
2167 | ||
2168 | if (RB_EMPTY_ROOT(&failure_tree->state)) | |
2169 | return; | |
2170 | ||
2171 | spin_lock(&failure_tree->lock); | |
2172 | state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY); | |
2173 | while (state) { | |
2174 | if (state->start > end) | |
2175 | break; | |
2176 | ||
2177 | ASSERT(state->end <= end); | |
2178 | ||
2179 | next = next_state(state); | |
2180 | ||
47dc196a | 2181 | failrec = state->failrec; |
f612496b MX |
2182 | free_extent_state(state); |
2183 | kfree(failrec); | |
2184 | ||
2185 | state = next; | |
2186 | } | |
2187 | spin_unlock(&failure_tree->lock); | |
2188 | } | |
2189 | ||
2fe6303e | 2190 | int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end, |
47dc196a | 2191 | struct io_failure_record **failrec_ret) |
4a54c8c1 | 2192 | { |
2fe6303e | 2193 | struct io_failure_record *failrec; |
4a54c8c1 | 2194 | struct extent_map *em; |
4a54c8c1 JS |
2195 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2196 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2197 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
4a54c8c1 | 2198 | int ret; |
4a54c8c1 JS |
2199 | u64 logical; |
2200 | ||
47dc196a | 2201 | ret = get_state_failrec(failure_tree, start, &failrec); |
4a54c8c1 JS |
2202 | if (ret) { |
2203 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2204 | if (!failrec) | |
2205 | return -ENOMEM; | |
2fe6303e | 2206 | |
4a54c8c1 JS |
2207 | failrec->start = start; |
2208 | failrec->len = end - start + 1; | |
2209 | failrec->this_mirror = 0; | |
2210 | failrec->bio_flags = 0; | |
2211 | failrec->in_validation = 0; | |
2212 | ||
2213 | read_lock(&em_tree->lock); | |
2214 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2215 | if (!em) { | |
2216 | read_unlock(&em_tree->lock); | |
2217 | kfree(failrec); | |
2218 | return -EIO; | |
2219 | } | |
2220 | ||
68ba990f | 2221 | if (em->start > start || em->start + em->len <= start) { |
4a54c8c1 JS |
2222 | free_extent_map(em); |
2223 | em = NULL; | |
2224 | } | |
2225 | read_unlock(&em_tree->lock); | |
7a2d6a64 | 2226 | if (!em) { |
4a54c8c1 JS |
2227 | kfree(failrec); |
2228 | return -EIO; | |
2229 | } | |
2fe6303e | 2230 | |
4a54c8c1 JS |
2231 | logical = start - em->start; |
2232 | logical = em->block_start + logical; | |
2233 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2234 | logical = em->block_start; | |
2235 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2236 | extent_set_compress_type(&failrec->bio_flags, | |
2237 | em->compress_type); | |
2238 | } | |
2fe6303e MX |
2239 | |
2240 | pr_debug("Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu\n", | |
2241 | logical, start, failrec->len); | |
2242 | ||
4a54c8c1 JS |
2243 | failrec->logical = logical; |
2244 | free_extent_map(em); | |
2245 | ||
2246 | /* set the bits in the private failure tree */ | |
2247 | ret = set_extent_bits(failure_tree, start, end, | |
ceeb0ae7 | 2248 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 | 2249 | if (ret >= 0) |
47dc196a | 2250 | ret = set_state_failrec(failure_tree, start, failrec); |
4a54c8c1 JS |
2251 | /* set the bits in the inode's tree */ |
2252 | if (ret >= 0) | |
ceeb0ae7 | 2253 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED); |
4a54c8c1 JS |
2254 | if (ret < 0) { |
2255 | kfree(failrec); | |
2256 | return ret; | |
2257 | } | |
2258 | } else { | |
2fe6303e | 2259 | pr_debug("Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d\n", |
4a54c8c1 JS |
2260 | failrec->logical, failrec->start, failrec->len, |
2261 | failrec->in_validation); | |
2262 | /* | |
2263 | * when data can be on disk more than twice, add to failrec here | |
2264 | * (e.g. with a list for failed_mirror) to make | |
2265 | * clean_io_failure() clean all those errors at once. | |
2266 | */ | |
2267 | } | |
2fe6303e MX |
2268 | |
2269 | *failrec_ret = failrec; | |
2270 | ||
2271 | return 0; | |
2272 | } | |
2273 | ||
2274 | int btrfs_check_repairable(struct inode *inode, struct bio *failed_bio, | |
2275 | struct io_failure_record *failrec, int failed_mirror) | |
2276 | { | |
2277 | int num_copies; | |
2278 | ||
5d964051 SB |
2279 | num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info, |
2280 | failrec->logical, failrec->len); | |
4a54c8c1 JS |
2281 | if (num_copies == 1) { |
2282 | /* | |
2283 | * we only have a single copy of the data, so don't bother with | |
2284 | * all the retry and error correction code that follows. no | |
2285 | * matter what the error is, it is very likely to persist. | |
2286 | */ | |
2fe6303e | 2287 | pr_debug("Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n", |
09a7f7a2 | 2288 | num_copies, failrec->this_mirror, failed_mirror); |
2fe6303e | 2289 | return 0; |
4a54c8c1 JS |
2290 | } |
2291 | ||
4a54c8c1 JS |
2292 | /* |
2293 | * there are two premises: | |
2294 | * a) deliver good data to the caller | |
2295 | * b) correct the bad sectors on disk | |
2296 | */ | |
2297 | if (failed_bio->bi_vcnt > 1) { | |
2298 | /* | |
2299 | * to fulfill b), we need to know the exact failing sectors, as | |
2300 | * we don't want to rewrite any more than the failed ones. thus, | |
2301 | * we need separate read requests for the failed bio | |
2302 | * | |
2303 | * if the following BUG_ON triggers, our validation request got | |
2304 | * merged. we need separate requests for our algorithm to work. | |
2305 | */ | |
2306 | BUG_ON(failrec->in_validation); | |
2307 | failrec->in_validation = 1; | |
2308 | failrec->this_mirror = failed_mirror; | |
4a54c8c1 JS |
2309 | } else { |
2310 | /* | |
2311 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2312 | * of the failed sector and if we succeed, we have setup | |
2313 | * everything for repair_io_failure to do the rest for us. | |
2314 | */ | |
2315 | if (failrec->in_validation) { | |
2316 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2317 | failrec->in_validation = 0; | |
2318 | failrec->this_mirror = 0; | |
2319 | } | |
2320 | failrec->failed_mirror = failed_mirror; | |
2321 | failrec->this_mirror++; | |
2322 | if (failrec->this_mirror == failed_mirror) | |
2323 | failrec->this_mirror++; | |
4a54c8c1 JS |
2324 | } |
2325 | ||
facc8a22 | 2326 | if (failrec->this_mirror > num_copies) { |
2fe6303e | 2327 | pr_debug("Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n", |
4a54c8c1 | 2328 | num_copies, failrec->this_mirror, failed_mirror); |
2fe6303e | 2329 | return 0; |
4a54c8c1 JS |
2330 | } |
2331 | ||
2fe6303e MX |
2332 | return 1; |
2333 | } | |
2334 | ||
2335 | ||
2336 | struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio, | |
2337 | struct io_failure_record *failrec, | |
2338 | struct page *page, int pg_offset, int icsum, | |
8b110e39 | 2339 | bio_end_io_t *endio_func, void *data) |
2fe6303e MX |
2340 | { |
2341 | struct bio *bio; | |
2342 | struct btrfs_io_bio *btrfs_failed_bio; | |
2343 | struct btrfs_io_bio *btrfs_bio; | |
2344 | ||
9be3395b | 2345 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
2fe6303e MX |
2346 | if (!bio) |
2347 | return NULL; | |
2348 | ||
2349 | bio->bi_end_io = endio_func; | |
4f024f37 | 2350 | bio->bi_iter.bi_sector = failrec->logical >> 9; |
4a54c8c1 | 2351 | bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; |
4f024f37 | 2352 | bio->bi_iter.bi_size = 0; |
8b110e39 | 2353 | bio->bi_private = data; |
4a54c8c1 | 2354 | |
facc8a22 MX |
2355 | btrfs_failed_bio = btrfs_io_bio(failed_bio); |
2356 | if (btrfs_failed_bio->csum) { | |
2357 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2358 | u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); | |
2359 | ||
2360 | btrfs_bio = btrfs_io_bio(bio); | |
2361 | btrfs_bio->csum = btrfs_bio->csum_inline; | |
2fe6303e MX |
2362 | icsum *= csum_size; |
2363 | memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum, | |
facc8a22 MX |
2364 | csum_size); |
2365 | } | |
2366 | ||
2fe6303e MX |
2367 | bio_add_page(bio, page, failrec->len, pg_offset); |
2368 | ||
2369 | return bio; | |
2370 | } | |
2371 | ||
2372 | /* | |
2373 | * this is a generic handler for readpage errors (default | |
2374 | * readpage_io_failed_hook). if other copies exist, read those and write back | |
2375 | * good data to the failed position. does not investigate in remapping the | |
2376 | * failed extent elsewhere, hoping the device will be smart enough to do this as | |
2377 | * needed | |
2378 | */ | |
2379 | ||
2380 | static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset, | |
2381 | struct page *page, u64 start, u64 end, | |
2382 | int failed_mirror) | |
2383 | { | |
2384 | struct io_failure_record *failrec; | |
2385 | struct inode *inode = page->mapping->host; | |
2386 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2387 | struct bio *bio; | |
2388 | int read_mode; | |
2389 | int ret; | |
2390 | ||
1f7ad75b | 2391 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
2fe6303e MX |
2392 | |
2393 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
2394 | if (ret) | |
2395 | return ret; | |
2396 | ||
2397 | ret = btrfs_check_repairable(inode, failed_bio, failrec, failed_mirror); | |
2398 | if (!ret) { | |
2399 | free_io_failure(inode, failrec); | |
2400 | return -EIO; | |
2401 | } | |
2402 | ||
2403 | if (failed_bio->bi_vcnt > 1) | |
2404 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; | |
2405 | else | |
2406 | read_mode = READ_SYNC; | |
2407 | ||
2408 | phy_offset >>= inode->i_sb->s_blocksize_bits; | |
2409 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2410 | start - page_offset(page), | |
8b110e39 MX |
2411 | (int)phy_offset, failed_bio->bi_end_io, |
2412 | NULL); | |
2fe6303e MX |
2413 | if (!bio) { |
2414 | free_io_failure(inode, failrec); | |
2415 | return -EIO; | |
2416 | } | |
1f7ad75b | 2417 | bio_set_op_attrs(bio, REQ_OP_READ, read_mode); |
4a54c8c1 | 2418 | |
2fe6303e MX |
2419 | pr_debug("Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d\n", |
2420 | read_mode, failrec->this_mirror, failrec->in_validation); | |
4a54c8c1 | 2421 | |
81a75f67 | 2422 | ret = tree->ops->submit_bio_hook(inode, bio, failrec->this_mirror, |
013bd4c3 | 2423 | failrec->bio_flags, 0); |
6c387ab2 | 2424 | if (ret) { |
454ff3de | 2425 | free_io_failure(inode, failrec); |
6c387ab2 MX |
2426 | bio_put(bio); |
2427 | } | |
2428 | ||
013bd4c3 | 2429 | return ret; |
4a54c8c1 JS |
2430 | } |
2431 | ||
d1310b2e CM |
2432 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2433 | ||
b5227c07 | 2434 | void end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
87826df0 JM |
2435 | { |
2436 | int uptodate = (err == 0); | |
2437 | struct extent_io_tree *tree; | |
3e2426bd | 2438 | int ret = 0; |
87826df0 JM |
2439 | |
2440 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
2441 | ||
2442 | if (tree->ops && tree->ops->writepage_end_io_hook) { | |
2443 | ret = tree->ops->writepage_end_io_hook(page, start, | |
2444 | end, NULL, uptodate); | |
2445 | if (ret) | |
2446 | uptodate = 0; | |
2447 | } | |
2448 | ||
87826df0 | 2449 | if (!uptodate) { |
87826df0 JM |
2450 | ClearPageUptodate(page); |
2451 | SetPageError(page); | |
5dca6eea LB |
2452 | ret = ret < 0 ? ret : -EIO; |
2453 | mapping_set_error(page->mapping, ret); | |
87826df0 | 2454 | } |
87826df0 JM |
2455 | } |
2456 | ||
d1310b2e CM |
2457 | /* |
2458 | * after a writepage IO is done, we need to: | |
2459 | * clear the uptodate bits on error | |
2460 | * clear the writeback bits in the extent tree for this IO | |
2461 | * end_page_writeback if the page has no more pending IO | |
2462 | * | |
2463 | * Scheduling is not allowed, so the extent state tree is expected | |
2464 | * to have one and only one object corresponding to this IO. | |
2465 | */ | |
4246a0b6 | 2466 | static void end_bio_extent_writepage(struct bio *bio) |
d1310b2e | 2467 | { |
2c30c71b | 2468 | struct bio_vec *bvec; |
d1310b2e CM |
2469 | u64 start; |
2470 | u64 end; | |
2c30c71b | 2471 | int i; |
d1310b2e | 2472 | |
2c30c71b | 2473 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2474 | struct page *page = bvec->bv_page; |
902b22f3 | 2475 | |
17a5adcc AO |
2476 | /* We always issue full-page reads, but if some block |
2477 | * in a page fails to read, blk_update_request() will | |
2478 | * advance bv_offset and adjust bv_len to compensate. | |
2479 | * Print a warning for nonzero offsets, and an error | |
2480 | * if they don't add up to a full page. */ | |
09cbfeaf KS |
2481 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { |
2482 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
efe120a0 FH |
2483 | btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info, |
2484 | "partial page write in btrfs with offset %u and length %u", | |
2485 | bvec->bv_offset, bvec->bv_len); | |
2486 | else | |
2487 | btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info, | |
2488 | "incomplete page write in btrfs with offset %u and " | |
2489 | "length %u", | |
2490 | bvec->bv_offset, bvec->bv_len); | |
2491 | } | |
d1310b2e | 2492 | |
17a5adcc AO |
2493 | start = page_offset(page); |
2494 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
d1310b2e | 2495 | |
b5227c07 | 2496 | end_extent_writepage(page, bio->bi_error, start, end); |
17a5adcc | 2497 | end_page_writeback(page); |
2c30c71b | 2498 | } |
2b1f55b0 | 2499 | |
d1310b2e | 2500 | bio_put(bio); |
d1310b2e CM |
2501 | } |
2502 | ||
883d0de4 MX |
2503 | static void |
2504 | endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len, | |
2505 | int uptodate) | |
2506 | { | |
2507 | struct extent_state *cached = NULL; | |
2508 | u64 end = start + len - 1; | |
2509 | ||
2510 | if (uptodate && tree->track_uptodate) | |
2511 | set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC); | |
2512 | unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC); | |
2513 | } | |
2514 | ||
d1310b2e CM |
2515 | /* |
2516 | * after a readpage IO is done, we need to: | |
2517 | * clear the uptodate bits on error | |
2518 | * set the uptodate bits if things worked | |
2519 | * set the page up to date if all extents in the tree are uptodate | |
2520 | * clear the lock bit in the extent tree | |
2521 | * unlock the page if there are no other extents locked for it | |
2522 | * | |
2523 | * Scheduling is not allowed, so the extent state tree is expected | |
2524 | * to have one and only one object corresponding to this IO. | |
2525 | */ | |
4246a0b6 | 2526 | static void end_bio_extent_readpage(struct bio *bio) |
d1310b2e | 2527 | { |
2c30c71b | 2528 | struct bio_vec *bvec; |
4246a0b6 | 2529 | int uptodate = !bio->bi_error; |
facc8a22 | 2530 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); |
902b22f3 | 2531 | struct extent_io_tree *tree; |
facc8a22 | 2532 | u64 offset = 0; |
d1310b2e CM |
2533 | u64 start; |
2534 | u64 end; | |
facc8a22 | 2535 | u64 len; |
883d0de4 MX |
2536 | u64 extent_start = 0; |
2537 | u64 extent_len = 0; | |
5cf1ab56 | 2538 | int mirror; |
d1310b2e | 2539 | int ret; |
2c30c71b | 2540 | int i; |
d1310b2e | 2541 | |
2c30c71b | 2542 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2543 | struct page *page = bvec->bv_page; |
a71754fc | 2544 | struct inode *inode = page->mapping->host; |
507903b8 | 2545 | |
be3940c0 | 2546 | pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, " |
4246a0b6 CH |
2547 | "mirror=%u\n", (u64)bio->bi_iter.bi_sector, |
2548 | bio->bi_error, io_bio->mirror_num); | |
a71754fc | 2549 | tree = &BTRFS_I(inode)->io_tree; |
902b22f3 | 2550 | |
17a5adcc AO |
2551 | /* We always issue full-page reads, but if some block |
2552 | * in a page fails to read, blk_update_request() will | |
2553 | * advance bv_offset and adjust bv_len to compensate. | |
2554 | * Print a warning for nonzero offsets, and an error | |
2555 | * if they don't add up to a full page. */ | |
09cbfeaf KS |
2556 | if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) { |
2557 | if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE) | |
efe120a0 FH |
2558 | btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info, |
2559 | "partial page read in btrfs with offset %u and length %u", | |
2560 | bvec->bv_offset, bvec->bv_len); | |
2561 | else | |
2562 | btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info, | |
2563 | "incomplete page read in btrfs with offset %u and " | |
2564 | "length %u", | |
2565 | bvec->bv_offset, bvec->bv_len); | |
2566 | } | |
d1310b2e | 2567 | |
17a5adcc AO |
2568 | start = page_offset(page); |
2569 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
facc8a22 | 2570 | len = bvec->bv_len; |
d1310b2e | 2571 | |
9be3395b | 2572 | mirror = io_bio->mirror_num; |
f2a09da9 MX |
2573 | if (likely(uptodate && tree->ops && |
2574 | tree->ops->readpage_end_io_hook)) { | |
facc8a22 MX |
2575 | ret = tree->ops->readpage_end_io_hook(io_bio, offset, |
2576 | page, start, end, | |
2577 | mirror); | |
5ee0844d | 2578 | if (ret) |
d1310b2e | 2579 | uptodate = 0; |
5ee0844d | 2580 | else |
1203b681 | 2581 | clean_io_failure(inode, start, page, 0); |
d1310b2e | 2582 | } |
ea466794 | 2583 | |
f2a09da9 MX |
2584 | if (likely(uptodate)) |
2585 | goto readpage_ok; | |
2586 | ||
2587 | if (tree->ops && tree->ops->readpage_io_failed_hook) { | |
5cf1ab56 | 2588 | ret = tree->ops->readpage_io_failed_hook(page, mirror); |
4246a0b6 | 2589 | if (!ret && !bio->bi_error) |
ea466794 | 2590 | uptodate = 1; |
f2a09da9 | 2591 | } else { |
f4a8e656 JS |
2592 | /* |
2593 | * The generic bio_readpage_error handles errors the | |
2594 | * following way: If possible, new read requests are | |
2595 | * created and submitted and will end up in | |
2596 | * end_bio_extent_readpage as well (if we're lucky, not | |
2597 | * in the !uptodate case). In that case it returns 0 and | |
2598 | * we just go on with the next page in our bio. If it | |
2599 | * can't handle the error it will return -EIO and we | |
2600 | * remain responsible for that page. | |
2601 | */ | |
facc8a22 MX |
2602 | ret = bio_readpage_error(bio, offset, page, start, end, |
2603 | mirror); | |
7e38326f | 2604 | if (ret == 0) { |
4246a0b6 | 2605 | uptodate = !bio->bi_error; |
38c1c2e4 | 2606 | offset += len; |
7e38326f CM |
2607 | continue; |
2608 | } | |
2609 | } | |
f2a09da9 | 2610 | readpage_ok: |
883d0de4 | 2611 | if (likely(uptodate)) { |
a71754fc | 2612 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 2613 | pgoff_t end_index = i_size >> PAGE_SHIFT; |
a583c026 | 2614 | unsigned off; |
a71754fc JB |
2615 | |
2616 | /* Zero out the end if this page straddles i_size */ | |
09cbfeaf | 2617 | off = i_size & (PAGE_SIZE-1); |
a583c026 | 2618 | if (page->index == end_index && off) |
09cbfeaf | 2619 | zero_user_segment(page, off, PAGE_SIZE); |
17a5adcc | 2620 | SetPageUptodate(page); |
70dec807 | 2621 | } else { |
17a5adcc AO |
2622 | ClearPageUptodate(page); |
2623 | SetPageError(page); | |
70dec807 | 2624 | } |
17a5adcc | 2625 | unlock_page(page); |
facc8a22 | 2626 | offset += len; |
883d0de4 MX |
2627 | |
2628 | if (unlikely(!uptodate)) { | |
2629 | if (extent_len) { | |
2630 | endio_readpage_release_extent(tree, | |
2631 | extent_start, | |
2632 | extent_len, 1); | |
2633 | extent_start = 0; | |
2634 | extent_len = 0; | |
2635 | } | |
2636 | endio_readpage_release_extent(tree, start, | |
2637 | end - start + 1, 0); | |
2638 | } else if (!extent_len) { | |
2639 | extent_start = start; | |
2640 | extent_len = end + 1 - start; | |
2641 | } else if (extent_start + extent_len == start) { | |
2642 | extent_len += end + 1 - start; | |
2643 | } else { | |
2644 | endio_readpage_release_extent(tree, extent_start, | |
2645 | extent_len, uptodate); | |
2646 | extent_start = start; | |
2647 | extent_len = end + 1 - start; | |
2648 | } | |
2c30c71b | 2649 | } |
d1310b2e | 2650 | |
883d0de4 MX |
2651 | if (extent_len) |
2652 | endio_readpage_release_extent(tree, extent_start, extent_len, | |
2653 | uptodate); | |
facc8a22 | 2654 | if (io_bio->end_io) |
4246a0b6 | 2655 | io_bio->end_io(io_bio, bio->bi_error); |
d1310b2e | 2656 | bio_put(bio); |
d1310b2e CM |
2657 | } |
2658 | ||
9be3395b CM |
2659 | /* |
2660 | * this allocates from the btrfs_bioset. We're returning a bio right now | |
2661 | * but you can call btrfs_io_bio for the appropriate container_of magic | |
2662 | */ | |
88f794ed MX |
2663 | struct bio * |
2664 | btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
2665 | gfp_t gfp_flags) | |
d1310b2e | 2666 | { |
facc8a22 | 2667 | struct btrfs_io_bio *btrfs_bio; |
d1310b2e CM |
2668 | struct bio *bio; |
2669 | ||
9be3395b | 2670 | bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset); |
d1310b2e CM |
2671 | |
2672 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
9be3395b CM |
2673 | while (!bio && (nr_vecs /= 2)) { |
2674 | bio = bio_alloc_bioset(gfp_flags, | |
2675 | nr_vecs, btrfs_bioset); | |
2676 | } | |
d1310b2e CM |
2677 | } |
2678 | ||
2679 | if (bio) { | |
2680 | bio->bi_bdev = bdev; | |
4f024f37 | 2681 | bio->bi_iter.bi_sector = first_sector; |
facc8a22 MX |
2682 | btrfs_bio = btrfs_io_bio(bio); |
2683 | btrfs_bio->csum = NULL; | |
2684 | btrfs_bio->csum_allocated = NULL; | |
2685 | btrfs_bio->end_io = NULL; | |
d1310b2e CM |
2686 | } |
2687 | return bio; | |
2688 | } | |
2689 | ||
9be3395b CM |
2690 | struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask) |
2691 | { | |
23ea8e5a MX |
2692 | struct btrfs_io_bio *btrfs_bio; |
2693 | struct bio *new; | |
9be3395b | 2694 | |
23ea8e5a MX |
2695 | new = bio_clone_bioset(bio, gfp_mask, btrfs_bioset); |
2696 | if (new) { | |
2697 | btrfs_bio = btrfs_io_bio(new); | |
2698 | btrfs_bio->csum = NULL; | |
2699 | btrfs_bio->csum_allocated = NULL; | |
2700 | btrfs_bio->end_io = NULL; | |
2701 | } | |
2702 | return new; | |
2703 | } | |
9be3395b CM |
2704 | |
2705 | /* this also allocates from the btrfs_bioset */ | |
2706 | struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs) | |
2707 | { | |
facc8a22 MX |
2708 | struct btrfs_io_bio *btrfs_bio; |
2709 | struct bio *bio; | |
2710 | ||
2711 | bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset); | |
2712 | if (bio) { | |
2713 | btrfs_bio = btrfs_io_bio(bio); | |
2714 | btrfs_bio->csum = NULL; | |
2715 | btrfs_bio->csum_allocated = NULL; | |
2716 | btrfs_bio->end_io = NULL; | |
2717 | } | |
2718 | return bio; | |
9be3395b CM |
2719 | } |
2720 | ||
2721 | ||
1f7ad75b MC |
2722 | static int __must_check submit_one_bio(struct bio *bio, int mirror_num, |
2723 | unsigned long bio_flags) | |
d1310b2e | 2724 | { |
d1310b2e | 2725 | int ret = 0; |
70dec807 CM |
2726 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
2727 | struct page *page = bvec->bv_page; | |
2728 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 2729 | u64 start; |
70dec807 | 2730 | |
4eee4fa4 | 2731 | start = page_offset(page) + bvec->bv_offset; |
70dec807 | 2732 | |
902b22f3 | 2733 | bio->bi_private = NULL; |
d1310b2e CM |
2734 | bio_get(bio); |
2735 | ||
065631f6 | 2736 | if (tree->ops && tree->ops->submit_bio_hook) |
81a75f67 | 2737 | ret = tree->ops->submit_bio_hook(page->mapping->host, bio, |
eaf25d93 | 2738 | mirror_num, bio_flags, start); |
0b86a832 | 2739 | else |
4e49ea4a | 2740 | btrfsic_submit_bio(bio); |
4a54c8c1 | 2741 | |
d1310b2e CM |
2742 | bio_put(bio); |
2743 | return ret; | |
2744 | } | |
2745 | ||
1f7ad75b | 2746 | static int merge_bio(struct extent_io_tree *tree, struct page *page, |
3444a972 JM |
2747 | unsigned long offset, size_t size, struct bio *bio, |
2748 | unsigned long bio_flags) | |
2749 | { | |
2750 | int ret = 0; | |
2751 | if (tree->ops && tree->ops->merge_bio_hook) | |
81a75f67 | 2752 | ret = tree->ops->merge_bio_hook(page, offset, size, bio, |
3444a972 | 2753 | bio_flags); |
3444a972 JM |
2754 | return ret; |
2755 | ||
2756 | } | |
2757 | ||
1f7ad75b | 2758 | static int submit_extent_page(int op, int op_flags, struct extent_io_tree *tree, |
da2f0f74 | 2759 | struct writeback_control *wbc, |
d1310b2e CM |
2760 | struct page *page, sector_t sector, |
2761 | size_t size, unsigned long offset, | |
2762 | struct block_device *bdev, | |
2763 | struct bio **bio_ret, | |
2764 | unsigned long max_pages, | |
f188591e | 2765 | bio_end_io_t end_io_func, |
c8b97818 CM |
2766 | int mirror_num, |
2767 | unsigned long prev_bio_flags, | |
005efedf FM |
2768 | unsigned long bio_flags, |
2769 | bool force_bio_submit) | |
d1310b2e CM |
2770 | { |
2771 | int ret = 0; | |
2772 | struct bio *bio; | |
c8b97818 | 2773 | int contig = 0; |
c8b97818 | 2774 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; |
09cbfeaf | 2775 | size_t page_size = min_t(size_t, size, PAGE_SIZE); |
d1310b2e CM |
2776 | |
2777 | if (bio_ret && *bio_ret) { | |
2778 | bio = *bio_ret; | |
c8b97818 | 2779 | if (old_compressed) |
4f024f37 | 2780 | contig = bio->bi_iter.bi_sector == sector; |
c8b97818 | 2781 | else |
f73a1c7d | 2782 | contig = bio_end_sector(bio) == sector; |
c8b97818 CM |
2783 | |
2784 | if (prev_bio_flags != bio_flags || !contig || | |
005efedf | 2785 | force_bio_submit || |
1f7ad75b | 2786 | merge_bio(tree, page, offset, page_size, bio, bio_flags) || |
c8b97818 | 2787 | bio_add_page(bio, page, page_size, offset) < page_size) { |
1f7ad75b | 2788 | ret = submit_one_bio(bio, mirror_num, prev_bio_flags); |
289454ad NA |
2789 | if (ret < 0) { |
2790 | *bio_ret = NULL; | |
79787eaa | 2791 | return ret; |
289454ad | 2792 | } |
d1310b2e CM |
2793 | bio = NULL; |
2794 | } else { | |
da2f0f74 CM |
2795 | if (wbc) |
2796 | wbc_account_io(wbc, page, page_size); | |
d1310b2e CM |
2797 | return 0; |
2798 | } | |
2799 | } | |
c8b97818 | 2800 | |
b54ffb73 KO |
2801 | bio = btrfs_bio_alloc(bdev, sector, BIO_MAX_PAGES, |
2802 | GFP_NOFS | __GFP_HIGH); | |
5df67083 TI |
2803 | if (!bio) |
2804 | return -ENOMEM; | |
70dec807 | 2805 | |
c8b97818 | 2806 | bio_add_page(bio, page, page_size, offset); |
d1310b2e CM |
2807 | bio->bi_end_io = end_io_func; |
2808 | bio->bi_private = tree; | |
1f7ad75b | 2809 | bio_set_op_attrs(bio, op, op_flags); |
da2f0f74 CM |
2810 | if (wbc) { |
2811 | wbc_init_bio(wbc, bio); | |
2812 | wbc_account_io(wbc, page, page_size); | |
2813 | } | |
70dec807 | 2814 | |
d397712b | 2815 | if (bio_ret) |
d1310b2e | 2816 | *bio_ret = bio; |
d397712b | 2817 | else |
1f7ad75b | 2818 | ret = submit_one_bio(bio, mirror_num, bio_flags); |
d1310b2e CM |
2819 | |
2820 | return ret; | |
2821 | } | |
2822 | ||
48a3b636 ES |
2823 | static void attach_extent_buffer_page(struct extent_buffer *eb, |
2824 | struct page *page) | |
d1310b2e CM |
2825 | { |
2826 | if (!PagePrivate(page)) { | |
2827 | SetPagePrivate(page); | |
09cbfeaf | 2828 | get_page(page); |
4f2de97a JB |
2829 | set_page_private(page, (unsigned long)eb); |
2830 | } else { | |
2831 | WARN_ON(page->private != (unsigned long)eb); | |
d1310b2e CM |
2832 | } |
2833 | } | |
2834 | ||
4f2de97a | 2835 | void set_page_extent_mapped(struct page *page) |
d1310b2e | 2836 | { |
4f2de97a JB |
2837 | if (!PagePrivate(page)) { |
2838 | SetPagePrivate(page); | |
09cbfeaf | 2839 | get_page(page); |
4f2de97a JB |
2840 | set_page_private(page, EXTENT_PAGE_PRIVATE); |
2841 | } | |
d1310b2e CM |
2842 | } |
2843 | ||
125bac01 MX |
2844 | static struct extent_map * |
2845 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
2846 | u64 start, u64 len, get_extent_t *get_extent, | |
2847 | struct extent_map **em_cached) | |
2848 | { | |
2849 | struct extent_map *em; | |
2850 | ||
2851 | if (em_cached && *em_cached) { | |
2852 | em = *em_cached; | |
cbc0e928 | 2853 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 MX |
2854 | start < extent_map_end(em)) { |
2855 | atomic_inc(&em->refs); | |
2856 | return em; | |
2857 | } | |
2858 | ||
2859 | free_extent_map(em); | |
2860 | *em_cached = NULL; | |
2861 | } | |
2862 | ||
2863 | em = get_extent(inode, page, pg_offset, start, len, 0); | |
2864 | if (em_cached && !IS_ERR_OR_NULL(em)) { | |
2865 | BUG_ON(*em_cached); | |
2866 | atomic_inc(&em->refs); | |
2867 | *em_cached = em; | |
2868 | } | |
2869 | return em; | |
2870 | } | |
d1310b2e CM |
2871 | /* |
2872 | * basic readpage implementation. Locked extent state structs are inserted | |
2873 | * into the tree that are removed when the IO is done (by the end_io | |
2874 | * handlers) | |
79787eaa | 2875 | * XXX JDM: This needs looking at to ensure proper page locking |
baf863b9 | 2876 | * return 0 on success, otherwise return error |
d1310b2e | 2877 | */ |
9974090b MX |
2878 | static int __do_readpage(struct extent_io_tree *tree, |
2879 | struct page *page, | |
2880 | get_extent_t *get_extent, | |
125bac01 | 2881 | struct extent_map **em_cached, |
9974090b | 2882 | struct bio **bio, int mirror_num, |
1f7ad75b | 2883 | unsigned long *bio_flags, int read_flags, |
005efedf | 2884 | u64 *prev_em_start) |
d1310b2e CM |
2885 | { |
2886 | struct inode *inode = page->mapping->host; | |
4eee4fa4 | 2887 | u64 start = page_offset(page); |
09cbfeaf | 2888 | u64 page_end = start + PAGE_SIZE - 1; |
d1310b2e CM |
2889 | u64 end; |
2890 | u64 cur = start; | |
2891 | u64 extent_offset; | |
2892 | u64 last_byte = i_size_read(inode); | |
2893 | u64 block_start; | |
2894 | u64 cur_end; | |
2895 | sector_t sector; | |
2896 | struct extent_map *em; | |
2897 | struct block_device *bdev; | |
baf863b9 | 2898 | int ret = 0; |
d1310b2e | 2899 | int nr = 0; |
306e16ce | 2900 | size_t pg_offset = 0; |
d1310b2e | 2901 | size_t iosize; |
c8b97818 | 2902 | size_t disk_io_size; |
d1310b2e | 2903 | size_t blocksize = inode->i_sb->s_blocksize; |
7f042a83 | 2904 | unsigned long this_bio_flag = 0; |
d1310b2e CM |
2905 | |
2906 | set_page_extent_mapped(page); | |
2907 | ||
9974090b | 2908 | end = page_end; |
90a887c9 DM |
2909 | if (!PageUptodate(page)) { |
2910 | if (cleancache_get_page(page) == 0) { | |
2911 | BUG_ON(blocksize != PAGE_SIZE); | |
9974090b | 2912 | unlock_extent(tree, start, end); |
90a887c9 DM |
2913 | goto out; |
2914 | } | |
2915 | } | |
2916 | ||
09cbfeaf | 2917 | if (page->index == last_byte >> PAGE_SHIFT) { |
c8b97818 | 2918 | char *userpage; |
09cbfeaf | 2919 | size_t zero_offset = last_byte & (PAGE_SIZE - 1); |
c8b97818 CM |
2920 | |
2921 | if (zero_offset) { | |
09cbfeaf | 2922 | iosize = PAGE_SIZE - zero_offset; |
7ac687d9 | 2923 | userpage = kmap_atomic(page); |
c8b97818 CM |
2924 | memset(userpage + zero_offset, 0, iosize); |
2925 | flush_dcache_page(page); | |
7ac687d9 | 2926 | kunmap_atomic(userpage); |
c8b97818 CM |
2927 | } |
2928 | } | |
d1310b2e | 2929 | while (cur <= end) { |
09cbfeaf | 2930 | unsigned long pnr = (last_byte >> PAGE_SHIFT) + 1; |
005efedf | 2931 | bool force_bio_submit = false; |
c8f2f24b | 2932 | |
d1310b2e CM |
2933 | if (cur >= last_byte) { |
2934 | char *userpage; | |
507903b8 AJ |
2935 | struct extent_state *cached = NULL; |
2936 | ||
09cbfeaf | 2937 | iosize = PAGE_SIZE - pg_offset; |
7ac687d9 | 2938 | userpage = kmap_atomic(page); |
306e16ce | 2939 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2940 | flush_dcache_page(page); |
7ac687d9 | 2941 | kunmap_atomic(userpage); |
d1310b2e | 2942 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 2943 | &cached, GFP_NOFS); |
7f042a83 FM |
2944 | unlock_extent_cached(tree, cur, |
2945 | cur + iosize - 1, | |
2946 | &cached, GFP_NOFS); | |
d1310b2e CM |
2947 | break; |
2948 | } | |
125bac01 MX |
2949 | em = __get_extent_map(inode, page, pg_offset, cur, |
2950 | end - cur + 1, get_extent, em_cached); | |
c704005d | 2951 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 2952 | SetPageError(page); |
7f042a83 | 2953 | unlock_extent(tree, cur, end); |
d1310b2e CM |
2954 | break; |
2955 | } | |
d1310b2e CM |
2956 | extent_offset = cur - em->start; |
2957 | BUG_ON(extent_map_end(em) <= cur); | |
2958 | BUG_ON(end < cur); | |
2959 | ||
261507a0 | 2960 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
4b384318 | 2961 | this_bio_flag |= EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
2962 | extent_set_compress_type(&this_bio_flag, |
2963 | em->compress_type); | |
2964 | } | |
c8b97818 | 2965 | |
d1310b2e CM |
2966 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
2967 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 2968 | iosize = ALIGN(iosize, blocksize); |
c8b97818 CM |
2969 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
2970 | disk_io_size = em->block_len; | |
2971 | sector = em->block_start >> 9; | |
2972 | } else { | |
2973 | sector = (em->block_start + extent_offset) >> 9; | |
2974 | disk_io_size = iosize; | |
2975 | } | |
d1310b2e CM |
2976 | bdev = em->bdev; |
2977 | block_start = em->block_start; | |
d899e052 YZ |
2978 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2979 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
2980 | |
2981 | /* | |
2982 | * If we have a file range that points to a compressed extent | |
2983 | * and it's followed by a consecutive file range that points to | |
2984 | * to the same compressed extent (possibly with a different | |
2985 | * offset and/or length, so it either points to the whole extent | |
2986 | * or only part of it), we must make sure we do not submit a | |
2987 | * single bio to populate the pages for the 2 ranges because | |
2988 | * this makes the compressed extent read zero out the pages | |
2989 | * belonging to the 2nd range. Imagine the following scenario: | |
2990 | * | |
2991 | * File layout | |
2992 | * [0 - 8K] [8K - 24K] | |
2993 | * | | | |
2994 | * | | | |
2995 | * points to extent X, points to extent X, | |
2996 | * offset 4K, length of 8K offset 0, length 16K | |
2997 | * | |
2998 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
2999 | * | |
3000 | * If the bio to read the compressed extent covers both ranges, | |
3001 | * it will decompress extent X into the pages belonging to the | |
3002 | * first range and then it will stop, zeroing out the remaining | |
3003 | * pages that belong to the other range that points to extent X. | |
3004 | * So here we make sure we submit 2 bios, one for the first | |
3005 | * range and another one for the third range. Both will target | |
3006 | * the same physical extent from disk, but we can't currently | |
3007 | * make the compressed bio endio callback populate the pages | |
3008 | * for both ranges because each compressed bio is tightly | |
3009 | * coupled with a single extent map, and each range can have | |
3010 | * an extent map with a different offset value relative to the | |
3011 | * uncompressed data of our extent and different lengths. This | |
3012 | * is a corner case so we prioritize correctness over | |
3013 | * non-optimal behavior (submitting 2 bios for the same extent). | |
3014 | */ | |
3015 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
3016 | prev_em_start && *prev_em_start != (u64)-1 && | |
3017 | *prev_em_start != em->orig_start) | |
3018 | force_bio_submit = true; | |
3019 | ||
3020 | if (prev_em_start) | |
3021 | *prev_em_start = em->orig_start; | |
3022 | ||
d1310b2e CM |
3023 | free_extent_map(em); |
3024 | em = NULL; | |
3025 | ||
3026 | /* we've found a hole, just zero and go on */ | |
3027 | if (block_start == EXTENT_MAP_HOLE) { | |
3028 | char *userpage; | |
507903b8 AJ |
3029 | struct extent_state *cached = NULL; |
3030 | ||
7ac687d9 | 3031 | userpage = kmap_atomic(page); |
306e16ce | 3032 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 3033 | flush_dcache_page(page); |
7ac687d9 | 3034 | kunmap_atomic(userpage); |
d1310b2e CM |
3035 | |
3036 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 | 3037 | &cached, GFP_NOFS); |
7f042a83 FM |
3038 | unlock_extent_cached(tree, cur, |
3039 | cur + iosize - 1, | |
3040 | &cached, GFP_NOFS); | |
d1310b2e | 3041 | cur = cur + iosize; |
306e16ce | 3042 | pg_offset += iosize; |
d1310b2e CM |
3043 | continue; |
3044 | } | |
3045 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
3046 | if (test_range_bit(tree, cur, cur_end, |
3047 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 3048 | check_page_uptodate(tree, page); |
7f042a83 | 3049 | unlock_extent(tree, cur, cur + iosize - 1); |
d1310b2e | 3050 | cur = cur + iosize; |
306e16ce | 3051 | pg_offset += iosize; |
d1310b2e CM |
3052 | continue; |
3053 | } | |
70dec807 CM |
3054 | /* we have an inline extent but it didn't get marked up |
3055 | * to date. Error out | |
3056 | */ | |
3057 | if (block_start == EXTENT_MAP_INLINE) { | |
3058 | SetPageError(page); | |
7f042a83 | 3059 | unlock_extent(tree, cur, cur + iosize - 1); |
70dec807 | 3060 | cur = cur + iosize; |
306e16ce | 3061 | pg_offset += iosize; |
70dec807 CM |
3062 | continue; |
3063 | } | |
d1310b2e | 3064 | |
c8f2f24b | 3065 | pnr -= page->index; |
1f7ad75b MC |
3066 | ret = submit_extent_page(REQ_OP_READ, read_flags, tree, NULL, |
3067 | page, sector, disk_io_size, pg_offset, | |
89642229 | 3068 | bdev, bio, pnr, |
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, | |
3096 | get_extent_t *get_extent, | |
125bac01 | 3097 | struct extent_map **em_cached, |
9974090b | 3098 | struct bio **bio, int mirror_num, |
1f7ad75b | 3099 | unsigned long *bio_flags, |
808f80b4 | 3100 | u64 *prev_em_start) |
9974090b MX |
3101 | { |
3102 | struct inode *inode; | |
3103 | struct btrfs_ordered_extent *ordered; | |
3104 | int index; | |
3105 | ||
3106 | inode = pages[0]->mapping->host; | |
3107 | while (1) { | |
3108 | lock_extent(tree, start, end); | |
3109 | ordered = btrfs_lookup_ordered_range(inode, start, | |
3110 | end - start + 1); | |
3111 | if (!ordered) | |
3112 | break; | |
3113 | unlock_extent(tree, start, end); | |
3114 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3115 | btrfs_put_ordered_extent(ordered); | |
3116 | } | |
3117 | ||
3118 | for (index = 0; index < nr_pages; index++) { | |
125bac01 | 3119 | __do_readpage(tree, pages[index], get_extent, em_cached, bio, |
1f7ad75b | 3120 | mirror_num, bio_flags, 0, prev_em_start); |
09cbfeaf | 3121 | put_page(pages[index]); |
9974090b MX |
3122 | } |
3123 | } | |
3124 | ||
3125 | static void __extent_readpages(struct extent_io_tree *tree, | |
3126 | struct page *pages[], | |
3127 | int nr_pages, get_extent_t *get_extent, | |
125bac01 | 3128 | struct extent_map **em_cached, |
9974090b | 3129 | struct bio **bio, int mirror_num, |
1f7ad75b | 3130 | unsigned long *bio_flags, |
808f80b4 | 3131 | u64 *prev_em_start) |
9974090b | 3132 | { |
35a3621b | 3133 | u64 start = 0; |
9974090b MX |
3134 | u64 end = 0; |
3135 | u64 page_start; | |
3136 | int index; | |
35a3621b | 3137 | int first_index = 0; |
9974090b MX |
3138 | |
3139 | for (index = 0; index < nr_pages; index++) { | |
3140 | page_start = page_offset(pages[index]); | |
3141 | if (!end) { | |
3142 | start = page_start; | |
09cbfeaf | 3143 | end = start + PAGE_SIZE - 1; |
9974090b MX |
3144 | first_index = index; |
3145 | } else if (end + 1 == page_start) { | |
09cbfeaf | 3146 | end += PAGE_SIZE; |
9974090b MX |
3147 | } else { |
3148 | __do_contiguous_readpages(tree, &pages[first_index], | |
3149 | index - first_index, start, | |
125bac01 MX |
3150 | end, get_extent, em_cached, |
3151 | bio, mirror_num, bio_flags, | |
1f7ad75b | 3152 | prev_em_start); |
9974090b | 3153 | start = page_start; |
09cbfeaf | 3154 | end = start + PAGE_SIZE - 1; |
9974090b MX |
3155 | first_index = index; |
3156 | } | |
3157 | } | |
3158 | ||
3159 | if (end) | |
3160 | __do_contiguous_readpages(tree, &pages[first_index], | |
3161 | index - first_index, start, | |
125bac01 | 3162 | end, get_extent, em_cached, bio, |
1f7ad75b | 3163 | mirror_num, bio_flags, |
808f80b4 | 3164 | prev_em_start); |
9974090b MX |
3165 | } |
3166 | ||
3167 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
3168 | struct page *page, | |
3169 | get_extent_t *get_extent, | |
3170 | struct bio **bio, int mirror_num, | |
1f7ad75b | 3171 | unsigned long *bio_flags, int read_flags) |
9974090b MX |
3172 | { |
3173 | struct inode *inode = page->mapping->host; | |
3174 | struct btrfs_ordered_extent *ordered; | |
3175 | u64 start = page_offset(page); | |
09cbfeaf | 3176 | u64 end = start + PAGE_SIZE - 1; |
9974090b MX |
3177 | int ret; |
3178 | ||
3179 | while (1) { | |
3180 | lock_extent(tree, start, end); | |
dbfdb6d1 | 3181 | ordered = btrfs_lookup_ordered_range(inode, start, |
09cbfeaf | 3182 | PAGE_SIZE); |
9974090b MX |
3183 | if (!ordered) |
3184 | break; | |
3185 | unlock_extent(tree, start, end); | |
3186 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3187 | btrfs_put_ordered_extent(ordered); | |
3188 | } | |
3189 | ||
125bac01 | 3190 | ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num, |
1f7ad75b | 3191 | bio_flags, read_flags, NULL); |
9974090b MX |
3192 | return ret; |
3193 | } | |
3194 | ||
d1310b2e | 3195 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, |
8ddc7d9c | 3196 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3197 | { |
3198 | struct bio *bio = NULL; | |
c8b97818 | 3199 | unsigned long bio_flags = 0; |
d1310b2e CM |
3200 | int ret; |
3201 | ||
8ddc7d9c | 3202 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, |
1f7ad75b | 3203 | &bio_flags, 0); |
d1310b2e | 3204 | if (bio) |
1f7ad75b | 3205 | ret = submit_one_bio(bio, mirror_num, bio_flags); |
d1310b2e CM |
3206 | return ret; |
3207 | } | |
d1310b2e | 3208 | |
a9132667 LB |
3209 | static void update_nr_written(struct page *page, struct writeback_control *wbc, |
3210 | unsigned long nr_written) | |
11c8349b CM |
3211 | { |
3212 | wbc->nr_to_write -= nr_written; | |
11c8349b CM |
3213 | } |
3214 | ||
d1310b2e | 3215 | /* |
40f76580 CM |
3216 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
3217 | * | |
3218 | * This returns 1 if our fill_delalloc function did all the work required | |
3219 | * to write the page (copy into inline extent). In this case the IO has | |
3220 | * been started and the page is already unlocked. | |
3221 | * | |
3222 | * This returns 0 if all went well (page still locked) | |
3223 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 3224 | */ |
40f76580 CM |
3225 | static noinline_for_stack int writepage_delalloc(struct inode *inode, |
3226 | struct page *page, struct writeback_control *wbc, | |
3227 | struct extent_page_data *epd, | |
3228 | u64 delalloc_start, | |
3229 | unsigned long *nr_written) | |
3230 | { | |
3231 | struct extent_io_tree *tree = epd->tree; | |
09cbfeaf | 3232 | u64 page_end = delalloc_start + PAGE_SIZE - 1; |
40f76580 CM |
3233 | u64 nr_delalloc; |
3234 | u64 delalloc_to_write = 0; | |
3235 | u64 delalloc_end = 0; | |
3236 | int ret; | |
3237 | int page_started = 0; | |
3238 | ||
3239 | if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc) | |
3240 | return 0; | |
3241 | ||
3242 | while (delalloc_end < page_end) { | |
3243 | nr_delalloc = find_lock_delalloc_range(inode, tree, | |
3244 | page, | |
3245 | &delalloc_start, | |
3246 | &delalloc_end, | |
dcab6a3b | 3247 | BTRFS_MAX_EXTENT_SIZE); |
40f76580 CM |
3248 | if (nr_delalloc == 0) { |
3249 | delalloc_start = delalloc_end + 1; | |
3250 | continue; | |
3251 | } | |
3252 | ret = tree->ops->fill_delalloc(inode, page, | |
3253 | delalloc_start, | |
3254 | delalloc_end, | |
3255 | &page_started, | |
3256 | nr_written); | |
3257 | /* File system has been set read-only */ | |
3258 | if (ret) { | |
3259 | SetPageError(page); | |
3260 | /* fill_delalloc should be return < 0 for error | |
3261 | * but just in case, we use > 0 here meaning the | |
3262 | * IO is started, so we don't want to return > 0 | |
3263 | * unless things are going well. | |
3264 | */ | |
3265 | ret = ret < 0 ? ret : -EIO; | |
3266 | goto done; | |
3267 | } | |
3268 | /* | |
ea1754a0 KS |
3269 | * delalloc_end is already one less than the total length, so |
3270 | * we don't subtract one from PAGE_SIZE | |
40f76580 CM |
3271 | */ |
3272 | delalloc_to_write += (delalloc_end - delalloc_start + | |
ea1754a0 | 3273 | PAGE_SIZE) >> PAGE_SHIFT; |
40f76580 CM |
3274 | delalloc_start = delalloc_end + 1; |
3275 | } | |
3276 | if (wbc->nr_to_write < delalloc_to_write) { | |
3277 | int thresh = 8192; | |
3278 | ||
3279 | if (delalloc_to_write < thresh * 2) | |
3280 | thresh = delalloc_to_write; | |
3281 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3282 | thresh); | |
3283 | } | |
3284 | ||
3285 | /* did the fill delalloc function already unlock and start | |
3286 | * the IO? | |
3287 | */ | |
3288 | if (page_started) { | |
3289 | /* | |
3290 | * we've unlocked the page, so we can't update | |
3291 | * the mapping's writeback index, just update | |
3292 | * nr_to_write. | |
3293 | */ | |
3294 | wbc->nr_to_write -= *nr_written; | |
3295 | return 1; | |
3296 | } | |
3297 | ||
3298 | ret = 0; | |
3299 | ||
3300 | done: | |
3301 | return ret; | |
3302 | } | |
3303 | ||
3304 | /* | |
3305 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3306 | * and does the loop to map the page into extents and bios. | |
3307 | * | |
3308 | * We return 1 if the IO is started and the page is unlocked, | |
3309 | * 0 if all went well (page still locked) | |
3310 | * < 0 if there were errors (page still locked) | |
3311 | */ | |
3312 | static noinline_for_stack int __extent_writepage_io(struct inode *inode, | |
3313 | struct page *page, | |
3314 | struct writeback_control *wbc, | |
3315 | struct extent_page_data *epd, | |
3316 | loff_t i_size, | |
3317 | unsigned long nr_written, | |
3318 | int write_flags, int *nr_ret) | |
d1310b2e | 3319 | { |
d1310b2e | 3320 | struct extent_io_tree *tree = epd->tree; |
4eee4fa4 | 3321 | u64 start = page_offset(page); |
09cbfeaf | 3322 | u64 page_end = start + PAGE_SIZE - 1; |
d1310b2e CM |
3323 | u64 end; |
3324 | u64 cur = start; | |
3325 | u64 extent_offset; | |
d1310b2e CM |
3326 | u64 block_start; |
3327 | u64 iosize; | |
3328 | sector_t sector; | |
2c64c53d | 3329 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
3330 | struct extent_map *em; |
3331 | struct block_device *bdev; | |
7f3c74fb | 3332 | size_t pg_offset = 0; |
d1310b2e | 3333 | size_t blocksize; |
40f76580 CM |
3334 | int ret = 0; |
3335 | int nr = 0; | |
3336 | bool compressed; | |
c8b97818 | 3337 | |
247e743c | 3338 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
3339 | ret = tree->ops->writepage_start_hook(page, start, |
3340 | page_end); | |
87826df0 JM |
3341 | if (ret) { |
3342 | /* Fixup worker will requeue */ | |
3343 | if (ret == -EBUSY) | |
3344 | wbc->pages_skipped++; | |
3345 | else | |
3346 | redirty_page_for_writepage(wbc, page); | |
40f76580 | 3347 | |
11c8349b | 3348 | update_nr_written(page, wbc, nr_written); |
247e743c | 3349 | unlock_page(page); |
40f76580 | 3350 | ret = 1; |
11c8349b | 3351 | goto done_unlocked; |
247e743c CM |
3352 | } |
3353 | } | |
3354 | ||
11c8349b CM |
3355 | /* |
3356 | * we don't want to touch the inode after unlocking the page, | |
3357 | * so we update the mapping writeback index now | |
3358 | */ | |
3359 | update_nr_written(page, wbc, nr_written + 1); | |
771ed689 | 3360 | |
d1310b2e | 3361 | end = page_end; |
40f76580 | 3362 | if (i_size <= start) { |
e6dcd2dc CM |
3363 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3364 | tree->ops->writepage_end_io_hook(page, start, | |
3365 | page_end, NULL, 1); | |
d1310b2e CM |
3366 | goto done; |
3367 | } | |
3368 | ||
d1310b2e CM |
3369 | blocksize = inode->i_sb->s_blocksize; |
3370 | ||
3371 | while (cur <= end) { | |
40f76580 | 3372 | u64 em_end; |
58409edd DS |
3373 | unsigned long max_nr; |
3374 | ||
40f76580 | 3375 | if (cur >= i_size) { |
e6dcd2dc CM |
3376 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3377 | tree->ops->writepage_end_io_hook(page, cur, | |
3378 | page_end, NULL, 1); | |
d1310b2e CM |
3379 | break; |
3380 | } | |
7f3c74fb | 3381 | em = epd->get_extent(inode, page, pg_offset, cur, |
d1310b2e | 3382 | end - cur + 1, 1); |
c704005d | 3383 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3384 | SetPageError(page); |
61391d56 | 3385 | ret = PTR_ERR_OR_ZERO(em); |
d1310b2e CM |
3386 | break; |
3387 | } | |
3388 | ||
3389 | extent_offset = cur - em->start; | |
40f76580 CM |
3390 | em_end = extent_map_end(em); |
3391 | BUG_ON(em_end <= cur); | |
d1310b2e | 3392 | BUG_ON(end < cur); |
40f76580 | 3393 | iosize = min(em_end - cur, end - cur + 1); |
fda2832f | 3394 | iosize = ALIGN(iosize, blocksize); |
d1310b2e CM |
3395 | sector = (em->block_start + extent_offset) >> 9; |
3396 | bdev = em->bdev; | |
3397 | block_start = em->block_start; | |
c8b97818 | 3398 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
3399 | free_extent_map(em); |
3400 | em = NULL; | |
3401 | ||
c8b97818 CM |
3402 | /* |
3403 | * compressed and inline extents are written through other | |
3404 | * paths in the FS | |
3405 | */ | |
3406 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3407 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
3408 | /* |
3409 | * end_io notification does not happen here for | |
3410 | * compressed extents | |
3411 | */ | |
3412 | if (!compressed && tree->ops && | |
3413 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
3414 | tree->ops->writepage_end_io_hook(page, cur, |
3415 | cur + iosize - 1, | |
3416 | NULL, 1); | |
c8b97818 CM |
3417 | else if (compressed) { |
3418 | /* we don't want to end_page_writeback on | |
3419 | * a compressed extent. this happens | |
3420 | * elsewhere | |
3421 | */ | |
3422 | nr++; | |
3423 | } | |
3424 | ||
3425 | cur += iosize; | |
7f3c74fb | 3426 | pg_offset += iosize; |
d1310b2e CM |
3427 | continue; |
3428 | } | |
c8b97818 | 3429 | |
58409edd DS |
3430 | max_nr = (i_size >> PAGE_SHIFT) + 1; |
3431 | ||
3432 | set_range_writeback(tree, cur, cur + iosize - 1); | |
3433 | if (!PageWriteback(page)) { | |
3434 | btrfs_err(BTRFS_I(inode)->root->fs_info, | |
3435 | "page %lu not writeback, cur %llu end %llu", | |
3436 | page->index, cur, end); | |
d1310b2e | 3437 | } |
7f3c74fb | 3438 | |
1f7ad75b MC |
3439 | ret = submit_extent_page(REQ_OP_WRITE, write_flags, tree, wbc, |
3440 | page, sector, iosize, pg_offset, | |
58409edd DS |
3441 | bdev, &epd->bio, max_nr, |
3442 | end_bio_extent_writepage, | |
3443 | 0, 0, 0, false); | |
3444 | if (ret) | |
3445 | SetPageError(page); | |
d1310b2e | 3446 | |
d1310b2e | 3447 | cur = cur + iosize; |
7f3c74fb | 3448 | pg_offset += iosize; |
d1310b2e CM |
3449 | nr++; |
3450 | } | |
40f76580 CM |
3451 | done: |
3452 | *nr_ret = nr; | |
3453 | ||
3454 | done_unlocked: | |
3455 | ||
3456 | /* drop our reference on any cached states */ | |
3457 | free_extent_state(cached_state); | |
3458 | return ret; | |
3459 | } | |
3460 | ||
3461 | /* | |
3462 | * the writepage semantics are similar to regular writepage. extent | |
3463 | * records are inserted to lock ranges in the tree, and as dirty areas | |
3464 | * are found, they are marked writeback. Then the lock bits are removed | |
3465 | * and the end_io handler clears the writeback ranges | |
3466 | */ | |
3467 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
3468 | void *data) | |
3469 | { | |
3470 | struct inode *inode = page->mapping->host; | |
3471 | struct extent_page_data *epd = data; | |
3472 | u64 start = page_offset(page); | |
09cbfeaf | 3473 | u64 page_end = start + PAGE_SIZE - 1; |
40f76580 CM |
3474 | int ret; |
3475 | int nr = 0; | |
3476 | size_t pg_offset = 0; | |
3477 | loff_t i_size = i_size_read(inode); | |
09cbfeaf | 3478 | unsigned long end_index = i_size >> PAGE_SHIFT; |
1f7ad75b | 3479 | int write_flags = 0; |
40f76580 CM |
3480 | unsigned long nr_written = 0; |
3481 | ||
3482 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3483 | write_flags = WRITE_SYNC; | |
40f76580 CM |
3484 | |
3485 | trace___extent_writepage(page, inode, wbc); | |
3486 | ||
3487 | WARN_ON(!PageLocked(page)); | |
3488 | ||
3489 | ClearPageError(page); | |
3490 | ||
09cbfeaf | 3491 | pg_offset = i_size & (PAGE_SIZE - 1); |
40f76580 CM |
3492 | if (page->index > end_index || |
3493 | (page->index == end_index && !pg_offset)) { | |
09cbfeaf | 3494 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); |
40f76580 CM |
3495 | unlock_page(page); |
3496 | return 0; | |
3497 | } | |
3498 | ||
3499 | if (page->index == end_index) { | |
3500 | char *userpage; | |
3501 | ||
3502 | userpage = kmap_atomic(page); | |
3503 | memset(userpage + pg_offset, 0, | |
09cbfeaf | 3504 | PAGE_SIZE - pg_offset); |
40f76580 CM |
3505 | kunmap_atomic(userpage); |
3506 | flush_dcache_page(page); | |
3507 | } | |
3508 | ||
3509 | pg_offset = 0; | |
3510 | ||
3511 | set_page_extent_mapped(page); | |
3512 | ||
3513 | ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written); | |
3514 | if (ret == 1) | |
3515 | goto done_unlocked; | |
3516 | if (ret) | |
3517 | goto done; | |
3518 | ||
3519 | ret = __extent_writepage_io(inode, page, wbc, epd, | |
3520 | i_size, nr_written, write_flags, &nr); | |
3521 | if (ret == 1) | |
3522 | goto done_unlocked; | |
3523 | ||
d1310b2e CM |
3524 | done: |
3525 | if (nr == 0) { | |
3526 | /* make sure the mapping tag for page dirty gets cleared */ | |
3527 | set_page_writeback(page); | |
3528 | end_page_writeback(page); | |
3529 | } | |
61391d56 FM |
3530 | if (PageError(page)) { |
3531 | ret = ret < 0 ? ret : -EIO; | |
3532 | end_extent_writepage(page, ret, start, page_end); | |
3533 | } | |
d1310b2e | 3534 | unlock_page(page); |
40f76580 | 3535 | return ret; |
771ed689 | 3536 | |
11c8349b | 3537 | done_unlocked: |
d1310b2e CM |
3538 | return 0; |
3539 | } | |
3540 | ||
fd8b2b61 | 3541 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 3542 | { |
74316201 N |
3543 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
3544 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
3545 | } |
3546 | ||
0e378df1 CM |
3547 | static noinline_for_stack int |
3548 | lock_extent_buffer_for_io(struct extent_buffer *eb, | |
3549 | struct btrfs_fs_info *fs_info, | |
3550 | struct extent_page_data *epd) | |
0b32f4bb JB |
3551 | { |
3552 | unsigned long i, num_pages; | |
3553 | int flush = 0; | |
3554 | int ret = 0; | |
3555 | ||
3556 | if (!btrfs_try_tree_write_lock(eb)) { | |
3557 | flush = 1; | |
3558 | flush_write_bio(epd); | |
3559 | btrfs_tree_lock(eb); | |
3560 | } | |
3561 | ||
3562 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3563 | btrfs_tree_unlock(eb); | |
3564 | if (!epd->sync_io) | |
3565 | return 0; | |
3566 | if (!flush) { | |
3567 | flush_write_bio(epd); | |
3568 | flush = 1; | |
3569 | } | |
a098d8e8 CM |
3570 | while (1) { |
3571 | wait_on_extent_buffer_writeback(eb); | |
3572 | btrfs_tree_lock(eb); | |
3573 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3574 | break; | |
0b32f4bb | 3575 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3576 | } |
3577 | } | |
3578 | ||
51561ffe JB |
3579 | /* |
3580 | * We need to do this to prevent races in people who check if the eb is | |
3581 | * under IO since we can end up having no IO bits set for a short period | |
3582 | * of time. | |
3583 | */ | |
3584 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
3585 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3586 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 3587 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 3588 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
e2d84521 MX |
3589 | __percpu_counter_add(&fs_info->dirty_metadata_bytes, |
3590 | -eb->len, | |
3591 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 3592 | ret = 1; |
51561ffe JB |
3593 | } else { |
3594 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
3595 | } |
3596 | ||
3597 | btrfs_tree_unlock(eb); | |
3598 | ||
3599 | if (!ret) | |
3600 | return ret; | |
3601 | ||
3602 | num_pages = num_extent_pages(eb->start, eb->len); | |
3603 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 3604 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3605 | |
3606 | if (!trylock_page(p)) { | |
3607 | if (!flush) { | |
3608 | flush_write_bio(epd); | |
3609 | flush = 1; | |
3610 | } | |
3611 | lock_page(p); | |
3612 | } | |
3613 | } | |
3614 | ||
3615 | return ret; | |
3616 | } | |
3617 | ||
3618 | static void end_extent_buffer_writeback(struct extent_buffer *eb) | |
3619 | { | |
3620 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
4e857c58 | 3621 | smp_mb__after_atomic(); |
0b32f4bb JB |
3622 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); |
3623 | } | |
3624 | ||
656f30db FM |
3625 | static void set_btree_ioerr(struct page *page) |
3626 | { | |
3627 | struct extent_buffer *eb = (struct extent_buffer *)page->private; | |
656f30db FM |
3628 | |
3629 | SetPageError(page); | |
3630 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) | |
3631 | return; | |
3632 | ||
3633 | /* | |
3634 | * If writeback for a btree extent that doesn't belong to a log tree | |
3635 | * failed, increment the counter transaction->eb_write_errors. | |
3636 | * We do this because while the transaction is running and before it's | |
3637 | * committing (when we call filemap_fdata[write|wait]_range against | |
3638 | * the btree inode), we might have | |
3639 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
3640 | * returns an error or an error happens during writeback, when we're | |
3641 | * committing the transaction we wouldn't know about it, since the pages | |
3642 | * can be no longer dirty nor marked anymore for writeback (if a | |
3643 | * subsequent modification to the extent buffer didn't happen before the | |
3644 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
3645 | * able to find the pages tagged with SetPageError at transaction | |
3646 | * commit time. So if this happens we must abort the transaction, | |
3647 | * otherwise we commit a super block with btree roots that point to | |
3648 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
3649 | * or the content of some node/leaf from a past generation that got | |
3650 | * cowed or deleted and is no longer valid. | |
3651 | * | |
3652 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
3653 | * not be enough - we need to distinguish between log tree extents vs | |
3654 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
3655 | * will catch and clear such errors in the mapping - and that call might | |
3656 | * be from a log sync and not from a transaction commit. Also, checking | |
3657 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
3658 | * not done and would not be reliable - the eb might have been released | |
3659 | * from memory and reading it back again means that flag would not be | |
3660 | * set (since it's a runtime flag, not persisted on disk). | |
3661 | * | |
3662 | * Using the flags below in the btree inode also makes us achieve the | |
3663 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
3664 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
3665 | * is called, the writeback for all dirty pages had already finished | |
3666 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
3667 | * filemap_fdatawait_range() would return success, as it could not know | |
3668 | * that writeback errors happened (the pages were no longer tagged for | |
3669 | * writeback). | |
3670 | */ | |
3671 | switch (eb->log_index) { | |
3672 | case -1: | |
afcdd129 | 3673 | set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags); |
656f30db FM |
3674 | break; |
3675 | case 0: | |
afcdd129 | 3676 | set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags); |
656f30db FM |
3677 | break; |
3678 | case 1: | |
afcdd129 | 3679 | set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags); |
656f30db FM |
3680 | break; |
3681 | default: | |
3682 | BUG(); /* unexpected, logic error */ | |
3683 | } | |
3684 | } | |
3685 | ||
4246a0b6 | 3686 | static void end_bio_extent_buffer_writepage(struct bio *bio) |
0b32f4bb | 3687 | { |
2c30c71b | 3688 | struct bio_vec *bvec; |
0b32f4bb | 3689 | struct extent_buffer *eb; |
2c30c71b | 3690 | int i, done; |
0b32f4bb | 3691 | |
2c30c71b | 3692 | bio_for_each_segment_all(bvec, bio, i) { |
0b32f4bb JB |
3693 | struct page *page = bvec->bv_page; |
3694 | ||
0b32f4bb JB |
3695 | eb = (struct extent_buffer *)page->private; |
3696 | BUG_ON(!eb); | |
3697 | done = atomic_dec_and_test(&eb->io_pages); | |
3698 | ||
4246a0b6 CH |
3699 | if (bio->bi_error || |
3700 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { | |
0b32f4bb | 3701 | ClearPageUptodate(page); |
656f30db | 3702 | set_btree_ioerr(page); |
0b32f4bb JB |
3703 | } |
3704 | ||
3705 | end_page_writeback(page); | |
3706 | ||
3707 | if (!done) | |
3708 | continue; | |
3709 | ||
3710 | end_extent_buffer_writeback(eb); | |
2c30c71b | 3711 | } |
0b32f4bb JB |
3712 | |
3713 | bio_put(bio); | |
0b32f4bb JB |
3714 | } |
3715 | ||
0e378df1 | 3716 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb JB |
3717 | struct btrfs_fs_info *fs_info, |
3718 | struct writeback_control *wbc, | |
3719 | struct extent_page_data *epd) | |
3720 | { | |
3721 | struct block_device *bdev = fs_info->fs_devices->latest_bdev; | |
f28491e0 | 3722 | struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree; |
0b32f4bb JB |
3723 | u64 offset = eb->start; |
3724 | unsigned long i, num_pages; | |
de0022b9 | 3725 | unsigned long bio_flags = 0; |
1f7ad75b | 3726 | int write_flags = (epd->sync_io ? WRITE_SYNC : 0) | REQ_META; |
d7dbe9e7 | 3727 | int ret = 0; |
0b32f4bb | 3728 | |
656f30db | 3729 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); |
0b32f4bb JB |
3730 | num_pages = num_extent_pages(eb->start, eb->len); |
3731 | atomic_set(&eb->io_pages, num_pages); | |
de0022b9 JB |
3732 | if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID) |
3733 | bio_flags = EXTENT_BIO_TREE_LOG; | |
3734 | ||
0b32f4bb | 3735 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 3736 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
3737 | |
3738 | clear_page_dirty_for_io(p); | |
3739 | set_page_writeback(p); | |
1f7ad75b MC |
3740 | ret = submit_extent_page(REQ_OP_WRITE, write_flags, tree, wbc, |
3741 | p, offset >> 9, PAGE_SIZE, 0, bdev, | |
3742 | &epd->bio, -1, | |
3743 | end_bio_extent_buffer_writepage, | |
005efedf | 3744 | 0, epd->bio_flags, bio_flags, false); |
de0022b9 | 3745 | epd->bio_flags = bio_flags; |
0b32f4bb | 3746 | if (ret) { |
656f30db | 3747 | set_btree_ioerr(p); |
55e3bd2e | 3748 | end_page_writeback(p); |
0b32f4bb JB |
3749 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
3750 | end_extent_buffer_writeback(eb); | |
3751 | ret = -EIO; | |
3752 | break; | |
3753 | } | |
09cbfeaf | 3754 | offset += PAGE_SIZE; |
0b32f4bb JB |
3755 | update_nr_written(p, wbc, 1); |
3756 | unlock_page(p); | |
3757 | } | |
3758 | ||
3759 | if (unlikely(ret)) { | |
3760 | for (; i < num_pages; i++) { | |
bbf65cf0 | 3761 | struct page *p = eb->pages[i]; |
81465028 | 3762 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
3763 | unlock_page(p); |
3764 | } | |
3765 | } | |
3766 | ||
3767 | return ret; | |
3768 | } | |
3769 | ||
3770 | int btree_write_cache_pages(struct address_space *mapping, | |
3771 | struct writeback_control *wbc) | |
3772 | { | |
3773 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; | |
3774 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; | |
3775 | struct extent_buffer *eb, *prev_eb = NULL; | |
3776 | struct extent_page_data epd = { | |
3777 | .bio = NULL, | |
3778 | .tree = tree, | |
3779 | .extent_locked = 0, | |
3780 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
de0022b9 | 3781 | .bio_flags = 0, |
0b32f4bb JB |
3782 | }; |
3783 | int ret = 0; | |
3784 | int done = 0; | |
3785 | int nr_to_write_done = 0; | |
3786 | struct pagevec pvec; | |
3787 | int nr_pages; | |
3788 | pgoff_t index; | |
3789 | pgoff_t end; /* Inclusive */ | |
3790 | int scanned = 0; | |
3791 | int tag; | |
3792 | ||
3793 | pagevec_init(&pvec, 0); | |
3794 | if (wbc->range_cyclic) { | |
3795 | index = mapping->writeback_index; /* Start from prev offset */ | |
3796 | end = -1; | |
3797 | } else { | |
09cbfeaf KS |
3798 | index = wbc->range_start >> PAGE_SHIFT; |
3799 | end = wbc->range_end >> PAGE_SHIFT; | |
0b32f4bb JB |
3800 | scanned = 1; |
3801 | } | |
3802 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3803 | tag = PAGECACHE_TAG_TOWRITE; | |
3804 | else | |
3805 | tag = PAGECACHE_TAG_DIRTY; | |
3806 | retry: | |
3807 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3808 | tag_pages_for_writeback(mapping, index, end); | |
3809 | while (!done && !nr_to_write_done && (index <= end) && | |
3810 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, | |
3811 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
3812 | unsigned i; | |
3813 | ||
3814 | scanned = 1; | |
3815 | for (i = 0; i < nr_pages; i++) { | |
3816 | struct page *page = pvec.pages[i]; | |
3817 | ||
3818 | if (!PagePrivate(page)) | |
3819 | continue; | |
3820 | ||
3821 | if (!wbc->range_cyclic && page->index > end) { | |
3822 | done = 1; | |
3823 | break; | |
3824 | } | |
3825 | ||
b5bae261 JB |
3826 | spin_lock(&mapping->private_lock); |
3827 | if (!PagePrivate(page)) { | |
3828 | spin_unlock(&mapping->private_lock); | |
3829 | continue; | |
3830 | } | |
3831 | ||
0b32f4bb | 3832 | eb = (struct extent_buffer *)page->private; |
b5bae261 JB |
3833 | |
3834 | /* | |
3835 | * Shouldn't happen and normally this would be a BUG_ON | |
3836 | * but no sense in crashing the users box for something | |
3837 | * we can survive anyway. | |
3838 | */ | |
fae7f21c | 3839 | if (WARN_ON(!eb)) { |
b5bae261 | 3840 | spin_unlock(&mapping->private_lock); |
0b32f4bb JB |
3841 | continue; |
3842 | } | |
3843 | ||
b5bae261 JB |
3844 | if (eb == prev_eb) { |
3845 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 3846 | continue; |
b5bae261 | 3847 | } |
0b32f4bb | 3848 | |
b5bae261 JB |
3849 | ret = atomic_inc_not_zero(&eb->refs); |
3850 | spin_unlock(&mapping->private_lock); | |
3851 | if (!ret) | |
0b32f4bb | 3852 | continue; |
0b32f4bb JB |
3853 | |
3854 | prev_eb = eb; | |
3855 | ret = lock_extent_buffer_for_io(eb, fs_info, &epd); | |
3856 | if (!ret) { | |
3857 | free_extent_buffer(eb); | |
3858 | continue; | |
3859 | } | |
3860 | ||
3861 | ret = write_one_eb(eb, fs_info, wbc, &epd); | |
3862 | if (ret) { | |
3863 | done = 1; | |
3864 | free_extent_buffer(eb); | |
3865 | break; | |
3866 | } | |
3867 | free_extent_buffer(eb); | |
3868 | ||
3869 | /* | |
3870 | * the filesystem may choose to bump up nr_to_write. | |
3871 | * We have to make sure to honor the new nr_to_write | |
3872 | * at any time | |
3873 | */ | |
3874 | nr_to_write_done = wbc->nr_to_write <= 0; | |
3875 | } | |
3876 | pagevec_release(&pvec); | |
3877 | cond_resched(); | |
3878 | } | |
3879 | if (!scanned && !done) { | |
3880 | /* | |
3881 | * We hit the last page and there is more work to be done: wrap | |
3882 | * back to the start of the file | |
3883 | */ | |
3884 | scanned = 1; | |
3885 | index = 0; | |
3886 | goto retry; | |
3887 | } | |
3888 | flush_write_bio(&epd); | |
3889 | return ret; | |
3890 | } | |
3891 | ||
d1310b2e | 3892 | /** |
4bef0848 | 3893 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
3894 | * @mapping: address space structure to write |
3895 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
3896 | * @writepage: function called for each page | |
3897 | * @data: data passed to writepage function | |
3898 | * | |
3899 | * If a page is already under I/O, write_cache_pages() skips it, even | |
3900 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
3901 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
3902 | * and msync() need to guarantee that all the data which was dirty at the time | |
3903 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
3904 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
3905 | * existing IO to complete. | |
3906 | */ | |
b2950863 | 3907 | static int extent_write_cache_pages(struct extent_io_tree *tree, |
4bef0848 CM |
3908 | struct address_space *mapping, |
3909 | struct writeback_control *wbc, | |
d2c3f4f6 CM |
3910 | writepage_t writepage, void *data, |
3911 | void (*flush_fn)(void *)) | |
d1310b2e | 3912 | { |
7fd1a3f7 | 3913 | struct inode *inode = mapping->host; |
d1310b2e CM |
3914 | int ret = 0; |
3915 | int done = 0; | |
f85d7d6c | 3916 | int nr_to_write_done = 0; |
d1310b2e CM |
3917 | struct pagevec pvec; |
3918 | int nr_pages; | |
3919 | pgoff_t index; | |
3920 | pgoff_t end; /* Inclusive */ | |
a9132667 LB |
3921 | pgoff_t done_index; |
3922 | int range_whole = 0; | |
d1310b2e | 3923 | int scanned = 0; |
f7aaa06b | 3924 | int tag; |
d1310b2e | 3925 | |
7fd1a3f7 JB |
3926 | /* |
3927 | * We have to hold onto the inode so that ordered extents can do their | |
3928 | * work when the IO finishes. The alternative to this is failing to add | |
3929 | * an ordered extent if the igrab() fails there and that is a huge pain | |
3930 | * to deal with, so instead just hold onto the inode throughout the | |
3931 | * writepages operation. If it fails here we are freeing up the inode | |
3932 | * anyway and we'd rather not waste our time writing out stuff that is | |
3933 | * going to be truncated anyway. | |
3934 | */ | |
3935 | if (!igrab(inode)) | |
3936 | return 0; | |
3937 | ||
d1310b2e CM |
3938 | pagevec_init(&pvec, 0); |
3939 | if (wbc->range_cyclic) { | |
3940 | index = mapping->writeback_index; /* Start from prev offset */ | |
3941 | end = -1; | |
3942 | } else { | |
09cbfeaf KS |
3943 | index = wbc->range_start >> PAGE_SHIFT; |
3944 | end = wbc->range_end >> PAGE_SHIFT; | |
a9132667 LB |
3945 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
3946 | range_whole = 1; | |
d1310b2e CM |
3947 | scanned = 1; |
3948 | } | |
f7aaa06b JB |
3949 | if (wbc->sync_mode == WB_SYNC_ALL) |
3950 | tag = PAGECACHE_TAG_TOWRITE; | |
3951 | else | |
3952 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 3953 | retry: |
f7aaa06b JB |
3954 | if (wbc->sync_mode == WB_SYNC_ALL) |
3955 | tag_pages_for_writeback(mapping, index, end); | |
a9132667 | 3956 | done_index = index; |
f85d7d6c | 3957 | while (!done && !nr_to_write_done && (index <= end) && |
f7aaa06b JB |
3958 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
3959 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
d1310b2e CM |
3960 | unsigned i; |
3961 | ||
3962 | scanned = 1; | |
3963 | for (i = 0; i < nr_pages; i++) { | |
3964 | struct page *page = pvec.pages[i]; | |
3965 | ||
a9132667 | 3966 | done_index = page->index; |
d1310b2e CM |
3967 | /* |
3968 | * At this point we hold neither mapping->tree_lock nor | |
3969 | * lock on the page itself: the page may be truncated or | |
3970 | * invalidated (changing page->mapping to NULL), or even | |
3971 | * swizzled back from swapper_space to tmpfs file | |
3972 | * mapping | |
3973 | */ | |
c8f2f24b JB |
3974 | if (!trylock_page(page)) { |
3975 | flush_fn(data); | |
3976 | lock_page(page); | |
01d658f2 | 3977 | } |
d1310b2e CM |
3978 | |
3979 | if (unlikely(page->mapping != mapping)) { | |
3980 | unlock_page(page); | |
3981 | continue; | |
3982 | } | |
3983 | ||
3984 | if (!wbc->range_cyclic && page->index > end) { | |
3985 | done = 1; | |
3986 | unlock_page(page); | |
3987 | continue; | |
3988 | } | |
3989 | ||
d2c3f4f6 | 3990 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 CM |
3991 | if (PageWriteback(page)) |
3992 | flush_fn(data); | |
d1310b2e | 3993 | wait_on_page_writeback(page); |
d2c3f4f6 | 3994 | } |
d1310b2e CM |
3995 | |
3996 | if (PageWriteback(page) || | |
3997 | !clear_page_dirty_for_io(page)) { | |
3998 | unlock_page(page); | |
3999 | continue; | |
4000 | } | |
4001 | ||
4002 | ret = (*writepage)(page, wbc, data); | |
4003 | ||
4004 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
4005 | unlock_page(page); | |
4006 | ret = 0; | |
4007 | } | |
a9132667 LB |
4008 | if (ret < 0) { |
4009 | /* | |
4010 | * done_index is set past this page, | |
4011 | * so media errors will not choke | |
4012 | * background writeout for the entire | |
4013 | * file. This has consequences for | |
4014 | * range_cyclic semantics (ie. it may | |
4015 | * not be suitable for data integrity | |
4016 | * writeout). | |
4017 | */ | |
4018 | done_index = page->index + 1; | |
4019 | done = 1; | |
4020 | break; | |
4021 | } | |
f85d7d6c CM |
4022 | |
4023 | /* | |
4024 | * the filesystem may choose to bump up nr_to_write. | |
4025 | * We have to make sure to honor the new nr_to_write | |
4026 | * at any time | |
4027 | */ | |
4028 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
4029 | } |
4030 | pagevec_release(&pvec); | |
4031 | cond_resched(); | |
4032 | } | |
894b36e3 | 4033 | if (!scanned && !done) { |
d1310b2e CM |
4034 | /* |
4035 | * We hit the last page and there is more work to be done: wrap | |
4036 | * back to the start of the file | |
4037 | */ | |
4038 | scanned = 1; | |
4039 | index = 0; | |
4040 | goto retry; | |
4041 | } | |
a9132667 LB |
4042 | |
4043 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole)) | |
4044 | mapping->writeback_index = done_index; | |
4045 | ||
7fd1a3f7 | 4046 | btrfs_add_delayed_iput(inode); |
894b36e3 | 4047 | return ret; |
d1310b2e | 4048 | } |
d1310b2e | 4049 | |
ffbd517d | 4050 | static void flush_epd_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 4051 | { |
d2c3f4f6 | 4052 | if (epd->bio) { |
355808c2 JM |
4053 | int ret; |
4054 | ||
1f7ad75b MC |
4055 | bio_set_op_attrs(epd->bio, REQ_OP_WRITE, |
4056 | epd->sync_io ? WRITE_SYNC : 0); | |
355808c2 | 4057 | |
1f7ad75b | 4058 | ret = submit_one_bio(epd->bio, 0, epd->bio_flags); |
79787eaa | 4059 | BUG_ON(ret < 0); /* -ENOMEM */ |
d2c3f4f6 CM |
4060 | epd->bio = NULL; |
4061 | } | |
4062 | } | |
4063 | ||
ffbd517d CM |
4064 | static noinline void flush_write_bio(void *data) |
4065 | { | |
4066 | struct extent_page_data *epd = data; | |
4067 | flush_epd_write_bio(epd); | |
4068 | } | |
4069 | ||
d1310b2e CM |
4070 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, |
4071 | get_extent_t *get_extent, | |
4072 | struct writeback_control *wbc) | |
4073 | { | |
4074 | int ret; | |
d1310b2e CM |
4075 | struct extent_page_data epd = { |
4076 | .bio = NULL, | |
4077 | .tree = tree, | |
4078 | .get_extent = get_extent, | |
771ed689 | 4079 | .extent_locked = 0, |
ffbd517d | 4080 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
de0022b9 | 4081 | .bio_flags = 0, |
d1310b2e | 4082 | }; |
d1310b2e | 4083 | |
d1310b2e CM |
4084 | ret = __extent_writepage(page, wbc, &epd); |
4085 | ||
ffbd517d | 4086 | flush_epd_write_bio(&epd); |
d1310b2e CM |
4087 | return ret; |
4088 | } | |
d1310b2e | 4089 | |
771ed689 CM |
4090 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, |
4091 | u64 start, u64 end, get_extent_t *get_extent, | |
4092 | int mode) | |
4093 | { | |
4094 | int ret = 0; | |
4095 | struct address_space *mapping = inode->i_mapping; | |
4096 | struct page *page; | |
09cbfeaf KS |
4097 | unsigned long nr_pages = (end - start + PAGE_SIZE) >> |
4098 | PAGE_SHIFT; | |
771ed689 CM |
4099 | |
4100 | struct extent_page_data epd = { | |
4101 | .bio = NULL, | |
4102 | .tree = tree, | |
4103 | .get_extent = get_extent, | |
4104 | .extent_locked = 1, | |
ffbd517d | 4105 | .sync_io = mode == WB_SYNC_ALL, |
de0022b9 | 4106 | .bio_flags = 0, |
771ed689 CM |
4107 | }; |
4108 | struct writeback_control wbc_writepages = { | |
771ed689 | 4109 | .sync_mode = mode, |
771ed689 CM |
4110 | .nr_to_write = nr_pages * 2, |
4111 | .range_start = start, | |
4112 | .range_end = end + 1, | |
4113 | }; | |
4114 | ||
d397712b | 4115 | while (start <= end) { |
09cbfeaf | 4116 | page = find_get_page(mapping, start >> PAGE_SHIFT); |
771ed689 CM |
4117 | if (clear_page_dirty_for_io(page)) |
4118 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
4119 | else { | |
4120 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
4121 | tree->ops->writepage_end_io_hook(page, start, | |
09cbfeaf | 4122 | start + PAGE_SIZE - 1, |
771ed689 CM |
4123 | NULL, 1); |
4124 | unlock_page(page); | |
4125 | } | |
09cbfeaf KS |
4126 | put_page(page); |
4127 | start += PAGE_SIZE; | |
771ed689 CM |
4128 | } |
4129 | ||
ffbd517d | 4130 | flush_epd_write_bio(&epd); |
771ed689 CM |
4131 | return ret; |
4132 | } | |
d1310b2e CM |
4133 | |
4134 | int extent_writepages(struct extent_io_tree *tree, | |
4135 | struct address_space *mapping, | |
4136 | get_extent_t *get_extent, | |
4137 | struct writeback_control *wbc) | |
4138 | { | |
4139 | int ret = 0; | |
4140 | struct extent_page_data epd = { | |
4141 | .bio = NULL, | |
4142 | .tree = tree, | |
4143 | .get_extent = get_extent, | |
771ed689 | 4144 | .extent_locked = 0, |
ffbd517d | 4145 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
de0022b9 | 4146 | .bio_flags = 0, |
d1310b2e CM |
4147 | }; |
4148 | ||
4bef0848 | 4149 | ret = extent_write_cache_pages(tree, mapping, wbc, |
d2c3f4f6 CM |
4150 | __extent_writepage, &epd, |
4151 | flush_write_bio); | |
ffbd517d | 4152 | flush_epd_write_bio(&epd); |
d1310b2e CM |
4153 | return ret; |
4154 | } | |
d1310b2e CM |
4155 | |
4156 | int extent_readpages(struct extent_io_tree *tree, | |
4157 | struct address_space *mapping, | |
4158 | struct list_head *pages, unsigned nr_pages, | |
4159 | get_extent_t get_extent) | |
4160 | { | |
4161 | struct bio *bio = NULL; | |
4162 | unsigned page_idx; | |
c8b97818 | 4163 | unsigned long bio_flags = 0; |
67c9684f LB |
4164 | struct page *pagepool[16]; |
4165 | struct page *page; | |
125bac01 | 4166 | struct extent_map *em_cached = NULL; |
67c9684f | 4167 | int nr = 0; |
808f80b4 | 4168 | u64 prev_em_start = (u64)-1; |
d1310b2e | 4169 | |
d1310b2e | 4170 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
67c9684f | 4171 | page = list_entry(pages->prev, struct page, lru); |
d1310b2e CM |
4172 | |
4173 | prefetchw(&page->flags); | |
4174 | list_del(&page->lru); | |
67c9684f | 4175 | if (add_to_page_cache_lru(page, mapping, |
8a5c743e MH |
4176 | page->index, |
4177 | readahead_gfp_mask(mapping))) { | |
09cbfeaf | 4178 | put_page(page); |
67c9684f | 4179 | continue; |
d1310b2e | 4180 | } |
67c9684f LB |
4181 | |
4182 | pagepool[nr++] = page; | |
4183 | if (nr < ARRAY_SIZE(pagepool)) | |
4184 | continue; | |
125bac01 | 4185 | __extent_readpages(tree, pagepool, nr, get_extent, &em_cached, |
1f7ad75b | 4186 | &bio, 0, &bio_flags, &prev_em_start); |
67c9684f | 4187 | nr = 0; |
d1310b2e | 4188 | } |
9974090b | 4189 | if (nr) |
125bac01 | 4190 | __extent_readpages(tree, pagepool, nr, get_extent, &em_cached, |
1f7ad75b | 4191 | &bio, 0, &bio_flags, &prev_em_start); |
67c9684f | 4192 | |
125bac01 MX |
4193 | if (em_cached) |
4194 | free_extent_map(em_cached); | |
4195 | ||
d1310b2e CM |
4196 | BUG_ON(!list_empty(pages)); |
4197 | if (bio) | |
1f7ad75b | 4198 | return submit_one_bio(bio, 0, bio_flags); |
d1310b2e CM |
4199 | return 0; |
4200 | } | |
d1310b2e CM |
4201 | |
4202 | /* | |
4203 | * basic invalidatepage code, this waits on any locked or writeback | |
4204 | * ranges corresponding to the page, and then deletes any extent state | |
4205 | * records from the tree | |
4206 | */ | |
4207 | int extent_invalidatepage(struct extent_io_tree *tree, | |
4208 | struct page *page, unsigned long offset) | |
4209 | { | |
2ac55d41 | 4210 | struct extent_state *cached_state = NULL; |
4eee4fa4 | 4211 | u64 start = page_offset(page); |
09cbfeaf | 4212 | u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
4213 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; |
4214 | ||
fda2832f | 4215 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
4216 | if (start > end) |
4217 | return 0; | |
4218 | ||
ff13db41 | 4219 | lock_extent_bits(tree, start, end, &cached_state); |
1edbb734 | 4220 | wait_on_page_writeback(page); |
d1310b2e | 4221 | clear_extent_bit(tree, start, end, |
32c00aff JB |
4222 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
4223 | EXTENT_DO_ACCOUNTING, | |
2ac55d41 | 4224 | 1, 1, &cached_state, GFP_NOFS); |
d1310b2e CM |
4225 | return 0; |
4226 | } | |
d1310b2e | 4227 | |
7b13b7b1 CM |
4228 | /* |
4229 | * a helper for releasepage, this tests for areas of the page that | |
4230 | * are locked or under IO and drops the related state bits if it is safe | |
4231 | * to drop the page. | |
4232 | */ | |
48a3b636 ES |
4233 | static int try_release_extent_state(struct extent_map_tree *map, |
4234 | struct extent_io_tree *tree, | |
4235 | struct page *page, gfp_t mask) | |
7b13b7b1 | 4236 | { |
4eee4fa4 | 4237 | u64 start = page_offset(page); |
09cbfeaf | 4238 | u64 end = start + PAGE_SIZE - 1; |
7b13b7b1 CM |
4239 | int ret = 1; |
4240 | ||
211f90e6 | 4241 | if (test_range_bit(tree, start, end, |
8b62b72b | 4242 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
4243 | ret = 0; |
4244 | else { | |
4245 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
4246 | mask = GFP_NOFS; | |
11ef160f CM |
4247 | /* |
4248 | * at this point we can safely clear everything except the | |
4249 | * locked bit and the nodatasum bit | |
4250 | */ | |
e3f24cc5 | 4251 | ret = clear_extent_bit(tree, start, end, |
11ef160f CM |
4252 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
4253 | 0, 0, NULL, mask); | |
e3f24cc5 CM |
4254 | |
4255 | /* if clear_extent_bit failed for enomem reasons, | |
4256 | * we can't allow the release to continue. | |
4257 | */ | |
4258 | if (ret < 0) | |
4259 | ret = 0; | |
4260 | else | |
4261 | ret = 1; | |
7b13b7b1 CM |
4262 | } |
4263 | return ret; | |
4264 | } | |
7b13b7b1 | 4265 | |
d1310b2e CM |
4266 | /* |
4267 | * a helper for releasepage. As long as there are no locked extents | |
4268 | * in the range corresponding to the page, both state records and extent | |
4269 | * map records are removed | |
4270 | */ | |
4271 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
4272 | struct extent_io_tree *tree, struct page *page, |
4273 | gfp_t mask) | |
d1310b2e CM |
4274 | { |
4275 | struct extent_map *em; | |
4eee4fa4 | 4276 | u64 start = page_offset(page); |
09cbfeaf | 4277 | u64 end = start + PAGE_SIZE - 1; |
7b13b7b1 | 4278 | |
d0164adc | 4279 | if (gfpflags_allow_blocking(mask) && |
ee22184b | 4280 | page->mapping->host->i_size > SZ_16M) { |
39b5637f | 4281 | u64 len; |
70dec807 | 4282 | while (start <= end) { |
39b5637f | 4283 | len = end - start + 1; |
890871be | 4284 | write_lock(&map->lock); |
39b5637f | 4285 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 4286 | if (!em) { |
890871be | 4287 | write_unlock(&map->lock); |
70dec807 CM |
4288 | break; |
4289 | } | |
7f3c74fb CM |
4290 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
4291 | em->start != start) { | |
890871be | 4292 | write_unlock(&map->lock); |
70dec807 CM |
4293 | free_extent_map(em); |
4294 | break; | |
4295 | } | |
4296 | if (!test_range_bit(tree, em->start, | |
4297 | extent_map_end(em) - 1, | |
8b62b72b | 4298 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 4299 | 0, NULL)) { |
70dec807 CM |
4300 | remove_extent_mapping(map, em); |
4301 | /* once for the rb tree */ | |
4302 | free_extent_map(em); | |
4303 | } | |
4304 | start = extent_map_end(em); | |
890871be | 4305 | write_unlock(&map->lock); |
70dec807 CM |
4306 | |
4307 | /* once for us */ | |
d1310b2e CM |
4308 | free_extent_map(em); |
4309 | } | |
d1310b2e | 4310 | } |
7b13b7b1 | 4311 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 4312 | } |
d1310b2e | 4313 | |
ec29ed5b CM |
4314 | /* |
4315 | * helper function for fiemap, which doesn't want to see any holes. | |
4316 | * This maps until we find something past 'last' | |
4317 | */ | |
4318 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
4319 | u64 offset, | |
4320 | u64 last, | |
4321 | get_extent_t *get_extent) | |
4322 | { | |
4323 | u64 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
4324 | struct extent_map *em; | |
4325 | u64 len; | |
4326 | ||
4327 | if (offset >= last) | |
4328 | return NULL; | |
4329 | ||
67871254 | 4330 | while (1) { |
ec29ed5b CM |
4331 | len = last - offset; |
4332 | if (len == 0) | |
4333 | break; | |
fda2832f | 4334 | len = ALIGN(len, sectorsize); |
ec29ed5b | 4335 | em = get_extent(inode, NULL, 0, offset, len, 0); |
c704005d | 4336 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
4337 | return em; |
4338 | ||
4339 | /* if this isn't a hole return it */ | |
4340 | if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) && | |
4341 | em->block_start != EXTENT_MAP_HOLE) { | |
4342 | return em; | |
4343 | } | |
4344 | ||
4345 | /* this is a hole, advance to the next extent */ | |
4346 | offset = extent_map_end(em); | |
4347 | free_extent_map(em); | |
4348 | if (offset >= last) | |
4349 | break; | |
4350 | } | |
4351 | return NULL; | |
4352 | } | |
4353 | ||
1506fcc8 YS |
4354 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4355 | __u64 start, __u64 len, get_extent_t *get_extent) | |
4356 | { | |
975f84fe | 4357 | int ret = 0; |
1506fcc8 YS |
4358 | u64 off = start; |
4359 | u64 max = start + len; | |
4360 | u32 flags = 0; | |
975f84fe JB |
4361 | u32 found_type; |
4362 | u64 last; | |
ec29ed5b | 4363 | u64 last_for_get_extent = 0; |
1506fcc8 | 4364 | u64 disko = 0; |
ec29ed5b | 4365 | u64 isize = i_size_read(inode); |
975f84fe | 4366 | struct btrfs_key found_key; |
1506fcc8 | 4367 | struct extent_map *em = NULL; |
2ac55d41 | 4368 | struct extent_state *cached_state = NULL; |
975f84fe | 4369 | struct btrfs_path *path; |
dc046b10 | 4370 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1506fcc8 | 4371 | int end = 0; |
ec29ed5b CM |
4372 | u64 em_start = 0; |
4373 | u64 em_len = 0; | |
4374 | u64 em_end = 0; | |
1506fcc8 YS |
4375 | |
4376 | if (len == 0) | |
4377 | return -EINVAL; | |
4378 | ||
975f84fe JB |
4379 | path = btrfs_alloc_path(); |
4380 | if (!path) | |
4381 | return -ENOMEM; | |
4382 | path->leave_spinning = 1; | |
4383 | ||
2c91943b QW |
4384 | start = round_down(start, BTRFS_I(inode)->root->sectorsize); |
4385 | len = round_up(max, BTRFS_I(inode)->root->sectorsize) - start; | |
4d479cf0 | 4386 | |
ec29ed5b CM |
4387 | /* |
4388 | * lookup the last file extent. We're not using i_size here | |
4389 | * because there might be preallocation past i_size | |
4390 | */ | |
dc046b10 JB |
4391 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1, |
4392 | 0); | |
975f84fe JB |
4393 | if (ret < 0) { |
4394 | btrfs_free_path(path); | |
4395 | return ret; | |
2d324f59 LB |
4396 | } else { |
4397 | WARN_ON(!ret); | |
4398 | if (ret == 1) | |
4399 | ret = 0; | |
975f84fe | 4400 | } |
2d324f59 | 4401 | |
975f84fe | 4402 | path->slots[0]--; |
975f84fe | 4403 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
962a298f | 4404 | found_type = found_key.type; |
975f84fe | 4405 | |
ec29ed5b | 4406 | /* No extents, but there might be delalloc bits */ |
33345d01 | 4407 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 4408 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
4409 | /* have to trust i_size as the end */ |
4410 | last = (u64)-1; | |
4411 | last_for_get_extent = isize; | |
4412 | } else { | |
4413 | /* | |
4414 | * remember the start of the last extent. There are a | |
4415 | * bunch of different factors that go into the length of the | |
4416 | * extent, so its much less complex to remember where it started | |
4417 | */ | |
4418 | last = found_key.offset; | |
4419 | last_for_get_extent = last + 1; | |
975f84fe | 4420 | } |
fe09e16c | 4421 | btrfs_release_path(path); |
975f84fe | 4422 | |
ec29ed5b CM |
4423 | /* |
4424 | * we might have some extents allocated but more delalloc past those | |
4425 | * extents. so, we trust isize unless the start of the last extent is | |
4426 | * beyond isize | |
4427 | */ | |
4428 | if (last < isize) { | |
4429 | last = (u64)-1; | |
4430 | last_for_get_extent = isize; | |
4431 | } | |
4432 | ||
ff13db41 | 4433 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
d0082371 | 4434 | &cached_state); |
ec29ed5b | 4435 | |
4d479cf0 | 4436 | em = get_extent_skip_holes(inode, start, last_for_get_extent, |
ec29ed5b | 4437 | get_extent); |
1506fcc8 YS |
4438 | if (!em) |
4439 | goto out; | |
4440 | if (IS_ERR(em)) { | |
4441 | ret = PTR_ERR(em); | |
4442 | goto out; | |
4443 | } | |
975f84fe | 4444 | |
1506fcc8 | 4445 | while (!end) { |
b76bb701 | 4446 | u64 offset_in_extent = 0; |
ea8efc74 CM |
4447 | |
4448 | /* break if the extent we found is outside the range */ | |
4449 | if (em->start >= max || extent_map_end(em) < off) | |
4450 | break; | |
4451 | ||
4452 | /* | |
4453 | * get_extent may return an extent that starts before our | |
4454 | * requested range. We have to make sure the ranges | |
4455 | * we return to fiemap always move forward and don't | |
4456 | * overlap, so adjust the offsets here | |
4457 | */ | |
4458 | em_start = max(em->start, off); | |
1506fcc8 | 4459 | |
ea8efc74 CM |
4460 | /* |
4461 | * record the offset from the start of the extent | |
b76bb701 JB |
4462 | * for adjusting the disk offset below. Only do this if the |
4463 | * extent isn't compressed since our in ram offset may be past | |
4464 | * what we have actually allocated on disk. | |
ea8efc74 | 4465 | */ |
b76bb701 JB |
4466 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
4467 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 4468 | em_end = extent_map_end(em); |
ea8efc74 | 4469 | em_len = em_end - em_start; |
1506fcc8 YS |
4470 | disko = 0; |
4471 | flags = 0; | |
4472 | ||
ea8efc74 CM |
4473 | /* |
4474 | * bump off for our next call to get_extent | |
4475 | */ | |
4476 | off = extent_map_end(em); | |
4477 | if (off >= max) | |
4478 | end = 1; | |
4479 | ||
93dbfad7 | 4480 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
4481 | end = 1; |
4482 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 4483 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
4484 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
4485 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 4486 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
4487 | flags |= (FIEMAP_EXTENT_DELALLOC | |
4488 | FIEMAP_EXTENT_UNKNOWN); | |
dc046b10 | 4489 | } else if (fieinfo->fi_extents_max) { |
afce772e LF |
4490 | struct btrfs_trans_handle *trans; |
4491 | ||
dc046b10 JB |
4492 | u64 bytenr = em->block_start - |
4493 | (em->start - em->orig_start); | |
fe09e16c | 4494 | |
ea8efc74 | 4495 | disko = em->block_start + offset_in_extent; |
fe09e16c | 4496 | |
afce772e LF |
4497 | /* |
4498 | * We need a trans handle to get delayed refs | |
4499 | */ | |
4500 | trans = btrfs_join_transaction(root); | |
4501 | /* | |
4502 | * It's OK if we can't start a trans we can still check | |
4503 | * from commit_root | |
4504 | */ | |
4505 | if (IS_ERR(trans)) | |
4506 | trans = NULL; | |
4507 | ||
fe09e16c LB |
4508 | /* |
4509 | * As btrfs supports shared space, this information | |
4510 | * can be exported to userspace tools via | |
dc046b10 JB |
4511 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 |
4512 | * then we're just getting a count and we can skip the | |
4513 | * lookup stuff. | |
fe09e16c | 4514 | */ |
afce772e | 4515 | ret = btrfs_check_shared(trans, root->fs_info, |
dc046b10 JB |
4516 | root->objectid, |
4517 | btrfs_ino(inode), bytenr); | |
afce772e LF |
4518 | if (trans) |
4519 | btrfs_end_transaction(trans, root); | |
dc046b10 | 4520 | if (ret < 0) |
fe09e16c | 4521 | goto out_free; |
dc046b10 | 4522 | if (ret) |
fe09e16c | 4523 | flags |= FIEMAP_EXTENT_SHARED; |
dc046b10 | 4524 | ret = 0; |
1506fcc8 YS |
4525 | } |
4526 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
4527 | flags |= FIEMAP_EXTENT_ENCODED; | |
0d2b2372 JB |
4528 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
4529 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
1506fcc8 | 4530 | |
1506fcc8 YS |
4531 | free_extent_map(em); |
4532 | em = NULL; | |
ec29ed5b CM |
4533 | if ((em_start >= last) || em_len == (u64)-1 || |
4534 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
4535 | flags |= FIEMAP_EXTENT_LAST; |
4536 | end = 1; | |
4537 | } | |
4538 | ||
ec29ed5b CM |
4539 | /* now scan forward to see if this is really the last extent. */ |
4540 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
4541 | get_extent); | |
4542 | if (IS_ERR(em)) { | |
4543 | ret = PTR_ERR(em); | |
4544 | goto out; | |
4545 | } | |
4546 | if (!em) { | |
975f84fe JB |
4547 | flags |= FIEMAP_EXTENT_LAST; |
4548 | end = 1; | |
4549 | } | |
ec29ed5b CM |
4550 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, |
4551 | em_len, flags); | |
26e726af CS |
4552 | if (ret) { |
4553 | if (ret == 1) | |
4554 | ret = 0; | |
ec29ed5b | 4555 | goto out_free; |
26e726af | 4556 | } |
1506fcc8 YS |
4557 | } |
4558 | out_free: | |
4559 | free_extent_map(em); | |
4560 | out: | |
fe09e16c | 4561 | btrfs_free_path(path); |
a52f4cd2 | 4562 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
2ac55d41 | 4563 | &cached_state, GFP_NOFS); |
1506fcc8 YS |
4564 | return ret; |
4565 | } | |
4566 | ||
727011e0 CM |
4567 | static void __free_extent_buffer(struct extent_buffer *eb) |
4568 | { | |
6d49ba1b | 4569 | btrfs_leak_debug_del(&eb->leak_list); |
727011e0 CM |
4570 | kmem_cache_free(extent_buffer_cache, eb); |
4571 | } | |
4572 | ||
a26e8c9f | 4573 | int extent_buffer_under_io(struct extent_buffer *eb) |
db7f3436 JB |
4574 | { |
4575 | return (atomic_read(&eb->io_pages) || | |
4576 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4577 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4578 | } | |
4579 | ||
4580 | /* | |
4581 | * Helper for releasing extent buffer page. | |
4582 | */ | |
a50924e3 | 4583 | static void btrfs_release_extent_buffer_page(struct extent_buffer *eb) |
db7f3436 JB |
4584 | { |
4585 | unsigned long index; | |
db7f3436 JB |
4586 | struct page *page; |
4587 | int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4588 | ||
4589 | BUG_ON(extent_buffer_under_io(eb)); | |
4590 | ||
a50924e3 DS |
4591 | index = num_extent_pages(eb->start, eb->len); |
4592 | if (index == 0) | |
db7f3436 JB |
4593 | return; |
4594 | ||
4595 | do { | |
4596 | index--; | |
fb85fc9a | 4597 | page = eb->pages[index]; |
5d2361db FL |
4598 | if (!page) |
4599 | continue; | |
4600 | if (mapped) | |
db7f3436 | 4601 | spin_lock(&page->mapping->private_lock); |
5d2361db FL |
4602 | /* |
4603 | * We do this since we'll remove the pages after we've | |
4604 | * removed the eb from the radix tree, so we could race | |
4605 | * and have this page now attached to the new eb. So | |
4606 | * only clear page_private if it's still connected to | |
4607 | * this eb. | |
4608 | */ | |
4609 | if (PagePrivate(page) && | |
4610 | page->private == (unsigned long)eb) { | |
4611 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4612 | BUG_ON(PageDirty(page)); | |
4613 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 4614 | /* |
5d2361db FL |
4615 | * We need to make sure we haven't be attached |
4616 | * to a new eb. | |
db7f3436 | 4617 | */ |
5d2361db FL |
4618 | ClearPagePrivate(page); |
4619 | set_page_private(page, 0); | |
4620 | /* One for the page private */ | |
09cbfeaf | 4621 | put_page(page); |
db7f3436 | 4622 | } |
5d2361db FL |
4623 | |
4624 | if (mapped) | |
4625 | spin_unlock(&page->mapping->private_lock); | |
4626 | ||
01327610 | 4627 | /* One for when we allocated the page */ |
09cbfeaf | 4628 | put_page(page); |
a50924e3 | 4629 | } while (index != 0); |
db7f3436 JB |
4630 | } |
4631 | ||
4632 | /* | |
4633 | * Helper for releasing the extent buffer. | |
4634 | */ | |
4635 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4636 | { | |
a50924e3 | 4637 | btrfs_release_extent_buffer_page(eb); |
db7f3436 JB |
4638 | __free_extent_buffer(eb); |
4639 | } | |
4640 | ||
f28491e0 JB |
4641 | static struct extent_buffer * |
4642 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 4643 | unsigned long len) |
d1310b2e CM |
4644 | { |
4645 | struct extent_buffer *eb = NULL; | |
4646 | ||
d1b5c567 | 4647 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
4648 | eb->start = start; |
4649 | eb->len = len; | |
f28491e0 | 4650 | eb->fs_info = fs_info; |
815a51c7 | 4651 | eb->bflags = 0; |
bd681513 CM |
4652 | rwlock_init(&eb->lock); |
4653 | atomic_set(&eb->write_locks, 0); | |
4654 | atomic_set(&eb->read_locks, 0); | |
4655 | atomic_set(&eb->blocking_readers, 0); | |
4656 | atomic_set(&eb->blocking_writers, 0); | |
4657 | atomic_set(&eb->spinning_readers, 0); | |
4658 | atomic_set(&eb->spinning_writers, 0); | |
5b25f70f | 4659 | eb->lock_nested = 0; |
bd681513 CM |
4660 | init_waitqueue_head(&eb->write_lock_wq); |
4661 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 4662 | |
6d49ba1b ES |
4663 | btrfs_leak_debug_add(&eb->leak_list, &buffers); |
4664 | ||
3083ee2e | 4665 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 4666 | atomic_set(&eb->refs, 1); |
0b32f4bb | 4667 | atomic_set(&eb->io_pages, 0); |
727011e0 | 4668 | |
b8dae313 DS |
4669 | /* |
4670 | * Sanity checks, currently the maximum is 64k covered by 16x 4k pages | |
4671 | */ | |
4672 | BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE | |
4673 | > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
4674 | BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
d1310b2e CM |
4675 | |
4676 | return eb; | |
4677 | } | |
4678 | ||
815a51c7 JS |
4679 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) |
4680 | { | |
4681 | unsigned long i; | |
4682 | struct page *p; | |
4683 | struct extent_buffer *new; | |
4684 | unsigned long num_pages = num_extent_pages(src->start, src->len); | |
4685 | ||
3f556f78 | 4686 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
4687 | if (new == NULL) |
4688 | return NULL; | |
4689 | ||
4690 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4691 | p = alloc_page(GFP_NOFS); |
db7f3436 JB |
4692 | if (!p) { |
4693 | btrfs_release_extent_buffer(new); | |
4694 | return NULL; | |
4695 | } | |
815a51c7 JS |
4696 | attach_extent_buffer_page(new, p); |
4697 | WARN_ON(PageDirty(p)); | |
4698 | SetPageUptodate(p); | |
4699 | new->pages[i] = p; | |
4700 | } | |
4701 | ||
4702 | copy_extent_buffer(new, src, 0, 0, src->len); | |
4703 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); | |
4704 | set_bit(EXTENT_BUFFER_DUMMY, &new->bflags); | |
4705 | ||
4706 | return new; | |
4707 | } | |
4708 | ||
0f331229 OS |
4709 | struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
4710 | u64 start, unsigned long len) | |
815a51c7 JS |
4711 | { |
4712 | struct extent_buffer *eb; | |
3f556f78 | 4713 | unsigned long num_pages; |
815a51c7 JS |
4714 | unsigned long i; |
4715 | ||
0f331229 | 4716 | num_pages = num_extent_pages(start, len); |
3f556f78 DS |
4717 | |
4718 | eb = __alloc_extent_buffer(fs_info, start, len); | |
815a51c7 JS |
4719 | if (!eb) |
4720 | return NULL; | |
4721 | ||
4722 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4723 | eb->pages[i] = alloc_page(GFP_NOFS); |
815a51c7 JS |
4724 | if (!eb->pages[i]) |
4725 | goto err; | |
4726 | } | |
4727 | set_extent_buffer_uptodate(eb); | |
4728 | btrfs_set_header_nritems(eb, 0); | |
4729 | set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4730 | ||
4731 | return eb; | |
4732 | err: | |
84167d19 SB |
4733 | for (; i > 0; i--) |
4734 | __free_page(eb->pages[i - 1]); | |
815a51c7 JS |
4735 | __free_extent_buffer(eb); |
4736 | return NULL; | |
4737 | } | |
4738 | ||
0f331229 | 4739 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
b9ef22de | 4740 | u64 start, u32 nodesize) |
0f331229 OS |
4741 | { |
4742 | unsigned long len; | |
4743 | ||
4744 | if (!fs_info) { | |
4745 | /* | |
4746 | * Called only from tests that don't always have a fs_info | |
b9ef22de | 4747 | * available |
0f331229 | 4748 | */ |
b9ef22de | 4749 | len = nodesize; |
0f331229 OS |
4750 | } else { |
4751 | len = fs_info->tree_root->nodesize; | |
4752 | } | |
4753 | ||
4754 | return __alloc_dummy_extent_buffer(fs_info, start, len); | |
4755 | } | |
4756 | ||
0b32f4bb JB |
4757 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
4758 | { | |
242e18c7 | 4759 | int refs; |
0b32f4bb JB |
4760 | /* the ref bit is tricky. We have to make sure it is set |
4761 | * if we have the buffer dirty. Otherwise the | |
4762 | * code to free a buffer can end up dropping a dirty | |
4763 | * page | |
4764 | * | |
4765 | * Once the ref bit is set, it won't go away while the | |
4766 | * buffer is dirty or in writeback, and it also won't | |
4767 | * go away while we have the reference count on the | |
4768 | * eb bumped. | |
4769 | * | |
4770 | * We can't just set the ref bit without bumping the | |
4771 | * ref on the eb because free_extent_buffer might | |
4772 | * see the ref bit and try to clear it. If this happens | |
4773 | * free_extent_buffer might end up dropping our original | |
4774 | * ref by mistake and freeing the page before we are able | |
4775 | * to add one more ref. | |
4776 | * | |
4777 | * So bump the ref count first, then set the bit. If someone | |
4778 | * beat us to it, drop the ref we added. | |
4779 | */ | |
242e18c7 CM |
4780 | refs = atomic_read(&eb->refs); |
4781 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
4782 | return; | |
4783 | ||
594831c4 JB |
4784 | spin_lock(&eb->refs_lock); |
4785 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 4786 | atomic_inc(&eb->refs); |
594831c4 | 4787 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
4788 | } |
4789 | ||
2457aec6 MG |
4790 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
4791 | struct page *accessed) | |
5df4235e JB |
4792 | { |
4793 | unsigned long num_pages, i; | |
4794 | ||
0b32f4bb JB |
4795 | check_buffer_tree_ref(eb); |
4796 | ||
5df4235e JB |
4797 | num_pages = num_extent_pages(eb->start, eb->len); |
4798 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a DS |
4799 | struct page *p = eb->pages[i]; |
4800 | ||
2457aec6 MG |
4801 | if (p != accessed) |
4802 | mark_page_accessed(p); | |
5df4235e JB |
4803 | } |
4804 | } | |
4805 | ||
f28491e0 JB |
4806 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
4807 | u64 start) | |
452c75c3 CS |
4808 | { |
4809 | struct extent_buffer *eb; | |
4810 | ||
4811 | rcu_read_lock(); | |
f28491e0 | 4812 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
09cbfeaf | 4813 | start >> PAGE_SHIFT); |
452c75c3 CS |
4814 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
4815 | rcu_read_unlock(); | |
062c19e9 FM |
4816 | /* |
4817 | * Lock our eb's refs_lock to avoid races with | |
4818 | * free_extent_buffer. When we get our eb it might be flagged | |
4819 | * with EXTENT_BUFFER_STALE and another task running | |
4820 | * free_extent_buffer might have seen that flag set, | |
4821 | * eb->refs == 2, that the buffer isn't under IO (dirty and | |
4822 | * writeback flags not set) and it's still in the tree (flag | |
4823 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process | |
4824 | * of decrementing the extent buffer's reference count twice. | |
4825 | * So here we could race and increment the eb's reference count, | |
4826 | * clear its stale flag, mark it as dirty and drop our reference | |
4827 | * before the other task finishes executing free_extent_buffer, | |
4828 | * which would later result in an attempt to free an extent | |
4829 | * buffer that is dirty. | |
4830 | */ | |
4831 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
4832 | spin_lock(&eb->refs_lock); | |
4833 | spin_unlock(&eb->refs_lock); | |
4834 | } | |
2457aec6 | 4835 | mark_extent_buffer_accessed(eb, NULL); |
452c75c3 CS |
4836 | return eb; |
4837 | } | |
4838 | rcu_read_unlock(); | |
4839 | ||
4840 | return NULL; | |
4841 | } | |
4842 | ||
faa2dbf0 JB |
4843 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
4844 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
b9ef22de | 4845 | u64 start, u32 nodesize) |
faa2dbf0 JB |
4846 | { |
4847 | struct extent_buffer *eb, *exists = NULL; | |
4848 | int ret; | |
4849 | ||
4850 | eb = find_extent_buffer(fs_info, start); | |
4851 | if (eb) | |
4852 | return eb; | |
b9ef22de | 4853 | eb = alloc_dummy_extent_buffer(fs_info, start, nodesize); |
faa2dbf0 JB |
4854 | if (!eb) |
4855 | return NULL; | |
4856 | eb->fs_info = fs_info; | |
4857 | again: | |
e1860a77 | 4858 | ret = radix_tree_preload(GFP_NOFS); |
faa2dbf0 JB |
4859 | if (ret) |
4860 | goto free_eb; | |
4861 | spin_lock(&fs_info->buffer_lock); | |
4862 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
09cbfeaf | 4863 | start >> PAGE_SHIFT, eb); |
faa2dbf0 JB |
4864 | spin_unlock(&fs_info->buffer_lock); |
4865 | radix_tree_preload_end(); | |
4866 | if (ret == -EEXIST) { | |
4867 | exists = find_extent_buffer(fs_info, start); | |
4868 | if (exists) | |
4869 | goto free_eb; | |
4870 | else | |
4871 | goto again; | |
4872 | } | |
4873 | check_buffer_tree_ref(eb); | |
4874 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
4875 | ||
4876 | /* | |
4877 | * We will free dummy extent buffer's if they come into | |
4878 | * free_extent_buffer with a ref count of 2, but if we are using this we | |
4879 | * want the buffers to stay in memory until we're done with them, so | |
4880 | * bump the ref count again. | |
4881 | */ | |
4882 | atomic_inc(&eb->refs); | |
4883 | return eb; | |
4884 | free_eb: | |
4885 | btrfs_release_extent_buffer(eb); | |
4886 | return exists; | |
4887 | } | |
4888 | #endif | |
4889 | ||
f28491e0 | 4890 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
ce3e6984 | 4891 | u64 start) |
d1310b2e | 4892 | { |
ce3e6984 | 4893 | unsigned long len = fs_info->tree_root->nodesize; |
d1310b2e CM |
4894 | unsigned long num_pages = num_extent_pages(start, len); |
4895 | unsigned long i; | |
09cbfeaf | 4896 | unsigned long index = start >> PAGE_SHIFT; |
d1310b2e | 4897 | struct extent_buffer *eb; |
6af118ce | 4898 | struct extent_buffer *exists = NULL; |
d1310b2e | 4899 | struct page *p; |
f28491e0 | 4900 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
d1310b2e | 4901 | int uptodate = 1; |
19fe0a8b | 4902 | int ret; |
d1310b2e | 4903 | |
c871b0f2 LB |
4904 | if (!IS_ALIGNED(start, fs_info->tree_root->sectorsize)) { |
4905 | btrfs_err(fs_info, "bad tree block start %llu", start); | |
4906 | return ERR_PTR(-EINVAL); | |
4907 | } | |
4908 | ||
f28491e0 | 4909 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 4910 | if (eb) |
6af118ce | 4911 | return eb; |
6af118ce | 4912 | |
23d79d81 | 4913 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 4914 | if (!eb) |
c871b0f2 | 4915 | return ERR_PTR(-ENOMEM); |
d1310b2e | 4916 | |
727011e0 | 4917 | for (i = 0; i < num_pages; i++, index++) { |
d1b5c567 | 4918 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
c871b0f2 LB |
4919 | if (!p) { |
4920 | exists = ERR_PTR(-ENOMEM); | |
6af118ce | 4921 | goto free_eb; |
c871b0f2 | 4922 | } |
4f2de97a JB |
4923 | |
4924 | spin_lock(&mapping->private_lock); | |
4925 | if (PagePrivate(p)) { | |
4926 | /* | |
4927 | * We could have already allocated an eb for this page | |
4928 | * and attached one so lets see if we can get a ref on | |
4929 | * the existing eb, and if we can we know it's good and | |
4930 | * we can just return that one, else we know we can just | |
4931 | * overwrite page->private. | |
4932 | */ | |
4933 | exists = (struct extent_buffer *)p->private; | |
4934 | if (atomic_inc_not_zero(&exists->refs)) { | |
4935 | spin_unlock(&mapping->private_lock); | |
4936 | unlock_page(p); | |
09cbfeaf | 4937 | put_page(p); |
2457aec6 | 4938 | mark_extent_buffer_accessed(exists, p); |
4f2de97a JB |
4939 | goto free_eb; |
4940 | } | |
5ca64f45 | 4941 | exists = NULL; |
4f2de97a | 4942 | |
0b32f4bb | 4943 | /* |
4f2de97a JB |
4944 | * Do this so attach doesn't complain and we need to |
4945 | * drop the ref the old guy had. | |
4946 | */ | |
4947 | ClearPagePrivate(p); | |
0b32f4bb | 4948 | WARN_ON(PageDirty(p)); |
09cbfeaf | 4949 | put_page(p); |
d1310b2e | 4950 | } |
4f2de97a JB |
4951 | attach_extent_buffer_page(eb, p); |
4952 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 4953 | WARN_ON(PageDirty(p)); |
727011e0 | 4954 | eb->pages[i] = p; |
d1310b2e CM |
4955 | if (!PageUptodate(p)) |
4956 | uptodate = 0; | |
eb14ab8e CM |
4957 | |
4958 | /* | |
4959 | * see below about how we avoid a nasty race with release page | |
4960 | * and why we unlock later | |
4961 | */ | |
d1310b2e CM |
4962 | } |
4963 | if (uptodate) | |
b4ce94de | 4964 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 4965 | again: |
e1860a77 | 4966 | ret = radix_tree_preload(GFP_NOFS); |
c871b0f2 LB |
4967 | if (ret) { |
4968 | exists = ERR_PTR(ret); | |
19fe0a8b | 4969 | goto free_eb; |
c871b0f2 | 4970 | } |
19fe0a8b | 4971 | |
f28491e0 JB |
4972 | spin_lock(&fs_info->buffer_lock); |
4973 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
09cbfeaf | 4974 | start >> PAGE_SHIFT, eb); |
f28491e0 | 4975 | spin_unlock(&fs_info->buffer_lock); |
452c75c3 | 4976 | radix_tree_preload_end(); |
19fe0a8b | 4977 | if (ret == -EEXIST) { |
f28491e0 | 4978 | exists = find_extent_buffer(fs_info, start); |
452c75c3 CS |
4979 | if (exists) |
4980 | goto free_eb; | |
4981 | else | |
115391d2 | 4982 | goto again; |
6af118ce | 4983 | } |
6af118ce | 4984 | /* add one reference for the tree */ |
0b32f4bb | 4985 | check_buffer_tree_ref(eb); |
34b41ace | 4986 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
4987 | |
4988 | /* | |
4989 | * there is a race where release page may have | |
4990 | * tried to find this extent buffer in the radix | |
4991 | * but failed. It will tell the VM it is safe to | |
4992 | * reclaim the, and it will clear the page private bit. | |
4993 | * We must make sure to set the page private bit properly | |
4994 | * after the extent buffer is in the radix tree so | |
4995 | * it doesn't get lost | |
4996 | */ | |
727011e0 CM |
4997 | SetPageChecked(eb->pages[0]); |
4998 | for (i = 1; i < num_pages; i++) { | |
fb85fc9a | 4999 | p = eb->pages[i]; |
727011e0 CM |
5000 | ClearPageChecked(p); |
5001 | unlock_page(p); | |
5002 | } | |
5003 | unlock_page(eb->pages[0]); | |
d1310b2e CM |
5004 | return eb; |
5005 | ||
6af118ce | 5006 | free_eb: |
5ca64f45 | 5007 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
5008 | for (i = 0; i < num_pages; i++) { |
5009 | if (eb->pages[i]) | |
5010 | unlock_page(eb->pages[i]); | |
5011 | } | |
eb14ab8e | 5012 | |
897ca6e9 | 5013 | btrfs_release_extent_buffer(eb); |
6af118ce | 5014 | return exists; |
d1310b2e | 5015 | } |
d1310b2e | 5016 | |
3083ee2e JB |
5017 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
5018 | { | |
5019 | struct extent_buffer *eb = | |
5020 | container_of(head, struct extent_buffer, rcu_head); | |
5021 | ||
5022 | __free_extent_buffer(eb); | |
5023 | } | |
5024 | ||
3083ee2e | 5025 | /* Expects to have eb->eb_lock already held */ |
f7a52a40 | 5026 | static int release_extent_buffer(struct extent_buffer *eb) |
3083ee2e JB |
5027 | { |
5028 | WARN_ON(atomic_read(&eb->refs) == 0); | |
5029 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 5030 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 5031 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 5032 | |
815a51c7 | 5033 | spin_unlock(&eb->refs_lock); |
3083ee2e | 5034 | |
f28491e0 JB |
5035 | spin_lock(&fs_info->buffer_lock); |
5036 | radix_tree_delete(&fs_info->buffer_radix, | |
09cbfeaf | 5037 | eb->start >> PAGE_SHIFT); |
f28491e0 | 5038 | spin_unlock(&fs_info->buffer_lock); |
34b41ace JB |
5039 | } else { |
5040 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 5041 | } |
3083ee2e JB |
5042 | |
5043 | /* Should be safe to release our pages at this point */ | |
a50924e3 | 5044 | btrfs_release_extent_buffer_page(eb); |
bcb7e449 JB |
5045 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
5046 | if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) { | |
5047 | __free_extent_buffer(eb); | |
5048 | return 1; | |
5049 | } | |
5050 | #endif | |
3083ee2e | 5051 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 5052 | return 1; |
3083ee2e JB |
5053 | } |
5054 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
5055 | |
5056 | return 0; | |
3083ee2e JB |
5057 | } |
5058 | ||
d1310b2e CM |
5059 | void free_extent_buffer(struct extent_buffer *eb) |
5060 | { | |
242e18c7 CM |
5061 | int refs; |
5062 | int old; | |
d1310b2e CM |
5063 | if (!eb) |
5064 | return; | |
5065 | ||
242e18c7 CM |
5066 | while (1) { |
5067 | refs = atomic_read(&eb->refs); | |
5068 | if (refs <= 3) | |
5069 | break; | |
5070 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
5071 | if (old == refs) | |
5072 | return; | |
5073 | } | |
5074 | ||
3083ee2e | 5075 | spin_lock(&eb->refs_lock); |
815a51c7 JS |
5076 | if (atomic_read(&eb->refs) == 2 && |
5077 | test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) | |
5078 | atomic_dec(&eb->refs); | |
5079 | ||
3083ee2e JB |
5080 | if (atomic_read(&eb->refs) == 2 && |
5081 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 5082 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
5083 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5084 | atomic_dec(&eb->refs); | |
5085 | ||
5086 | /* | |
5087 | * I know this is terrible, but it's temporary until we stop tracking | |
5088 | * the uptodate bits and such for the extent buffers. | |
5089 | */ | |
f7a52a40 | 5090 | release_extent_buffer(eb); |
3083ee2e JB |
5091 | } |
5092 | ||
5093 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
5094 | { | |
5095 | if (!eb) | |
d1310b2e CM |
5096 | return; |
5097 | ||
3083ee2e JB |
5098 | spin_lock(&eb->refs_lock); |
5099 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
5100 | ||
0b32f4bb | 5101 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
5102 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
5103 | atomic_dec(&eb->refs); | |
f7a52a40 | 5104 | release_extent_buffer(eb); |
d1310b2e | 5105 | } |
d1310b2e | 5106 | |
1d4284bd | 5107 | void clear_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 5108 | { |
d1310b2e CM |
5109 | unsigned long i; |
5110 | unsigned long num_pages; | |
5111 | struct page *page; | |
5112 | ||
d1310b2e CM |
5113 | num_pages = num_extent_pages(eb->start, eb->len); |
5114 | ||
5115 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 5116 | page = eb->pages[i]; |
b9473439 | 5117 | if (!PageDirty(page)) |
d2c3f4f6 CM |
5118 | continue; |
5119 | ||
a61e6f29 | 5120 | lock_page(page); |
eb14ab8e CM |
5121 | WARN_ON(!PagePrivate(page)); |
5122 | ||
d1310b2e | 5123 | clear_page_dirty_for_io(page); |
0ee0fda0 | 5124 | spin_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
5125 | if (!PageDirty(page)) { |
5126 | radix_tree_tag_clear(&page->mapping->page_tree, | |
5127 | page_index(page), | |
5128 | PAGECACHE_TAG_DIRTY); | |
5129 | } | |
0ee0fda0 | 5130 | spin_unlock_irq(&page->mapping->tree_lock); |
bf0da8c1 | 5131 | ClearPageError(page); |
a61e6f29 | 5132 | unlock_page(page); |
d1310b2e | 5133 | } |
0b32f4bb | 5134 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 5135 | } |
d1310b2e | 5136 | |
0b32f4bb | 5137 | int set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e CM |
5138 | { |
5139 | unsigned long i; | |
5140 | unsigned long num_pages; | |
b9473439 | 5141 | int was_dirty = 0; |
d1310b2e | 5142 | |
0b32f4bb JB |
5143 | check_buffer_tree_ref(eb); |
5144 | ||
b9473439 | 5145 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 5146 | |
d1310b2e | 5147 | num_pages = num_extent_pages(eb->start, eb->len); |
3083ee2e | 5148 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
5149 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
5150 | ||
b9473439 | 5151 | for (i = 0; i < num_pages; i++) |
fb85fc9a | 5152 | set_page_dirty(eb->pages[i]); |
b9473439 | 5153 | return was_dirty; |
d1310b2e | 5154 | } |
d1310b2e | 5155 | |
69ba3927 | 5156 | void clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 CM |
5157 | { |
5158 | unsigned long i; | |
5159 | struct page *page; | |
5160 | unsigned long num_pages; | |
5161 | ||
b4ce94de | 5162 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
0b32f4bb | 5163 | num_pages = num_extent_pages(eb->start, eb->len); |
1259ab75 | 5164 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5165 | page = eb->pages[i]; |
33958dc6 CM |
5166 | if (page) |
5167 | ClearPageUptodate(page); | |
1259ab75 | 5168 | } |
1259ab75 CM |
5169 | } |
5170 | ||
09c25a8c | 5171 | void set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e CM |
5172 | { |
5173 | unsigned long i; | |
5174 | struct page *page; | |
5175 | unsigned long num_pages; | |
5176 | ||
0b32f4bb | 5177 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 5178 | num_pages = num_extent_pages(eb->start, eb->len); |
d1310b2e | 5179 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 5180 | page = eb->pages[i]; |
d1310b2e CM |
5181 | SetPageUptodate(page); |
5182 | } | |
d1310b2e | 5183 | } |
d1310b2e | 5184 | |
0b32f4bb | 5185 | int extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 5186 | { |
0b32f4bb | 5187 | return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 5188 | } |
d1310b2e CM |
5189 | |
5190 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
bb82ab88 | 5191 | struct extent_buffer *eb, u64 start, int wait, |
f188591e | 5192 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
5193 | { |
5194 | unsigned long i; | |
5195 | unsigned long start_i; | |
5196 | struct page *page; | |
5197 | int err; | |
5198 | int ret = 0; | |
ce9adaa5 CM |
5199 | int locked_pages = 0; |
5200 | int all_uptodate = 1; | |
d1310b2e | 5201 | unsigned long num_pages; |
727011e0 | 5202 | unsigned long num_reads = 0; |
a86c12c7 | 5203 | struct bio *bio = NULL; |
c8b97818 | 5204 | unsigned long bio_flags = 0; |
a86c12c7 | 5205 | |
b4ce94de | 5206 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
5207 | return 0; |
5208 | ||
d1310b2e CM |
5209 | if (start) { |
5210 | WARN_ON(start < eb->start); | |
09cbfeaf KS |
5211 | start_i = (start >> PAGE_SHIFT) - |
5212 | (eb->start >> PAGE_SHIFT); | |
d1310b2e CM |
5213 | } else { |
5214 | start_i = 0; | |
5215 | } | |
5216 | ||
5217 | num_pages = num_extent_pages(eb->start, eb->len); | |
5218 | for (i = start_i; i < num_pages; i++) { | |
fb85fc9a | 5219 | page = eb->pages[i]; |
bb82ab88 | 5220 | if (wait == WAIT_NONE) { |
2db04966 | 5221 | if (!trylock_page(page)) |
ce9adaa5 | 5222 | goto unlock_exit; |
d1310b2e CM |
5223 | } else { |
5224 | lock_page(page); | |
5225 | } | |
ce9adaa5 | 5226 | locked_pages++; |
2571e739 LB |
5227 | } |
5228 | /* | |
5229 | * We need to firstly lock all pages to make sure that | |
5230 | * the uptodate bit of our pages won't be affected by | |
5231 | * clear_extent_buffer_uptodate(). | |
5232 | */ | |
5233 | for (i = start_i; i < num_pages; i++) { | |
5234 | page = eb->pages[i]; | |
727011e0 CM |
5235 | if (!PageUptodate(page)) { |
5236 | num_reads++; | |
ce9adaa5 | 5237 | all_uptodate = 0; |
727011e0 | 5238 | } |
ce9adaa5 | 5239 | } |
2571e739 | 5240 | |
ce9adaa5 CM |
5241 | if (all_uptodate) { |
5242 | if (start_i == 0) | |
b4ce94de | 5243 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
5244 | goto unlock_exit; |
5245 | } | |
5246 | ||
656f30db | 5247 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 5248 | eb->read_mirror = 0; |
0b32f4bb | 5249 | atomic_set(&eb->io_pages, num_reads); |
ce9adaa5 | 5250 | for (i = start_i; i < num_pages; i++) { |
fb85fc9a | 5251 | page = eb->pages[i]; |
baf863b9 | 5252 | |
ce9adaa5 | 5253 | if (!PageUptodate(page)) { |
baf863b9 LB |
5254 | if (ret) { |
5255 | atomic_dec(&eb->io_pages); | |
5256 | unlock_page(page); | |
5257 | continue; | |
5258 | } | |
5259 | ||
f188591e | 5260 | ClearPageError(page); |
a86c12c7 | 5261 | err = __extent_read_full_page(tree, page, |
f188591e | 5262 | get_extent, &bio, |
d4c7ca86 | 5263 | mirror_num, &bio_flags, |
1f7ad75b | 5264 | REQ_META); |
baf863b9 | 5265 | if (err) { |
d1310b2e | 5266 | ret = err; |
baf863b9 LB |
5267 | /* |
5268 | * We use &bio in above __extent_read_full_page, | |
5269 | * so we ensure that if it returns error, the | |
5270 | * current page fails to add itself to bio and | |
5271 | * it's been unlocked. | |
5272 | * | |
5273 | * We must dec io_pages by ourselves. | |
5274 | */ | |
5275 | atomic_dec(&eb->io_pages); | |
5276 | } | |
d1310b2e CM |
5277 | } else { |
5278 | unlock_page(page); | |
5279 | } | |
5280 | } | |
5281 | ||
355808c2 | 5282 | if (bio) { |
1f7ad75b | 5283 | err = submit_one_bio(bio, mirror_num, bio_flags); |
79787eaa JM |
5284 | if (err) |
5285 | return err; | |
355808c2 | 5286 | } |
a86c12c7 | 5287 | |
bb82ab88 | 5288 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 5289 | return ret; |
d397712b | 5290 | |
d1310b2e | 5291 | for (i = start_i; i < num_pages; i++) { |
fb85fc9a | 5292 | page = eb->pages[i]; |
d1310b2e | 5293 | wait_on_page_locked(page); |
d397712b | 5294 | if (!PageUptodate(page)) |
d1310b2e | 5295 | ret = -EIO; |
d1310b2e | 5296 | } |
d397712b | 5297 | |
d1310b2e | 5298 | return ret; |
ce9adaa5 CM |
5299 | |
5300 | unlock_exit: | |
5301 | i = start_i; | |
d397712b | 5302 | while (locked_pages > 0) { |
fb85fc9a | 5303 | page = eb->pages[i]; |
ce9adaa5 CM |
5304 | i++; |
5305 | unlock_page(page); | |
5306 | locked_pages--; | |
5307 | } | |
5308 | return ret; | |
d1310b2e | 5309 | } |
d1310b2e CM |
5310 | |
5311 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
5312 | unsigned long start, | |
5313 | unsigned long len) | |
5314 | { | |
5315 | size_t cur; | |
5316 | size_t offset; | |
5317 | struct page *page; | |
5318 | char *kaddr; | |
5319 | char *dst = (char *)dstv; | |
09cbfeaf KS |
5320 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5321 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e CM |
5322 | |
5323 | WARN_ON(start > eb->len); | |
5324 | WARN_ON(start + len > eb->start + eb->len); | |
5325 | ||
09cbfeaf | 5326 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
d1310b2e | 5327 | |
d397712b | 5328 | while (len > 0) { |
fb85fc9a | 5329 | page = eb->pages[i]; |
d1310b2e | 5330 | |
09cbfeaf | 5331 | cur = min(len, (PAGE_SIZE - offset)); |
a6591715 | 5332 | kaddr = page_address(page); |
d1310b2e | 5333 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
5334 | |
5335 | dst += cur; | |
5336 | len -= cur; | |
5337 | offset = 0; | |
5338 | i++; | |
5339 | } | |
5340 | } | |
d1310b2e | 5341 | |
550ac1d8 GH |
5342 | int read_extent_buffer_to_user(struct extent_buffer *eb, void __user *dstv, |
5343 | unsigned long start, | |
5344 | unsigned long len) | |
5345 | { | |
5346 | size_t cur; | |
5347 | size_t offset; | |
5348 | struct page *page; | |
5349 | char *kaddr; | |
5350 | char __user *dst = (char __user *)dstv; | |
09cbfeaf KS |
5351 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5352 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
550ac1d8 GH |
5353 | int ret = 0; |
5354 | ||
5355 | WARN_ON(start > eb->len); | |
5356 | WARN_ON(start + len > eb->start + eb->len); | |
5357 | ||
09cbfeaf | 5358 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
550ac1d8 GH |
5359 | |
5360 | while (len > 0) { | |
fb85fc9a | 5361 | page = eb->pages[i]; |
550ac1d8 | 5362 | |
09cbfeaf | 5363 | cur = min(len, (PAGE_SIZE - offset)); |
550ac1d8 GH |
5364 | kaddr = page_address(page); |
5365 | if (copy_to_user(dst, kaddr + offset, cur)) { | |
5366 | ret = -EFAULT; | |
5367 | break; | |
5368 | } | |
5369 | ||
5370 | dst += cur; | |
5371 | len -= cur; | |
5372 | offset = 0; | |
5373 | i++; | |
5374 | } | |
5375 | ||
5376 | return ret; | |
5377 | } | |
5378 | ||
415b35a5 LB |
5379 | /* |
5380 | * return 0 if the item is found within a page. | |
5381 | * return 1 if the item spans two pages. | |
5382 | * return -EINVAL otherwise. | |
5383 | */ | |
d1310b2e | 5384 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, |
a6591715 | 5385 | unsigned long min_len, char **map, |
d1310b2e | 5386 | unsigned long *map_start, |
a6591715 | 5387 | unsigned long *map_len) |
d1310b2e | 5388 | { |
09cbfeaf | 5389 | size_t offset = start & (PAGE_SIZE - 1); |
d1310b2e CM |
5390 | char *kaddr; |
5391 | struct page *p; | |
09cbfeaf KS |
5392 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5393 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e | 5394 | unsigned long end_i = (start_offset + start + min_len - 1) >> |
09cbfeaf | 5395 | PAGE_SHIFT; |
d1310b2e CM |
5396 | |
5397 | if (i != end_i) | |
415b35a5 | 5398 | return 1; |
d1310b2e CM |
5399 | |
5400 | if (i == 0) { | |
5401 | offset = start_offset; | |
5402 | *map_start = 0; | |
5403 | } else { | |
5404 | offset = 0; | |
09cbfeaf | 5405 | *map_start = ((u64)i << PAGE_SHIFT) - start_offset; |
d1310b2e | 5406 | } |
d397712b | 5407 | |
d1310b2e | 5408 | if (start + min_len > eb->len) { |
31b1a2bd | 5409 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, " |
c1c9ff7c GU |
5410 | "wanted %lu %lu\n", |
5411 | eb->start, eb->len, start, min_len); | |
85026533 | 5412 | return -EINVAL; |
d1310b2e CM |
5413 | } |
5414 | ||
fb85fc9a | 5415 | p = eb->pages[i]; |
a6591715 | 5416 | kaddr = page_address(p); |
d1310b2e | 5417 | *map = kaddr + offset; |
09cbfeaf | 5418 | *map_len = PAGE_SIZE - offset; |
d1310b2e CM |
5419 | return 0; |
5420 | } | |
d1310b2e | 5421 | |
d1310b2e CM |
5422 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, |
5423 | unsigned long start, | |
5424 | unsigned long len) | |
5425 | { | |
5426 | size_t cur; | |
5427 | size_t offset; | |
5428 | struct page *page; | |
5429 | char *kaddr; | |
5430 | char *ptr = (char *)ptrv; | |
09cbfeaf KS |
5431 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5432 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e CM |
5433 | int ret = 0; |
5434 | ||
5435 | WARN_ON(start > eb->len); | |
5436 | WARN_ON(start + len > eb->start + eb->len); | |
5437 | ||
09cbfeaf | 5438 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
d1310b2e | 5439 | |
d397712b | 5440 | while (len > 0) { |
fb85fc9a | 5441 | page = eb->pages[i]; |
d1310b2e | 5442 | |
09cbfeaf | 5443 | cur = min(len, (PAGE_SIZE - offset)); |
d1310b2e | 5444 | |
a6591715 | 5445 | kaddr = page_address(page); |
d1310b2e | 5446 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
5447 | if (ret) |
5448 | break; | |
5449 | ||
5450 | ptr += cur; | |
5451 | len -= cur; | |
5452 | offset = 0; | |
5453 | i++; | |
5454 | } | |
5455 | return ret; | |
5456 | } | |
d1310b2e CM |
5457 | |
5458 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
5459 | unsigned long start, unsigned long len) | |
5460 | { | |
5461 | size_t cur; | |
5462 | size_t offset; | |
5463 | struct page *page; | |
5464 | char *kaddr; | |
5465 | char *src = (char *)srcv; | |
09cbfeaf KS |
5466 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5467 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e CM |
5468 | |
5469 | WARN_ON(start > eb->len); | |
5470 | WARN_ON(start + len > eb->start + eb->len); | |
5471 | ||
09cbfeaf | 5472 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
d1310b2e | 5473 | |
d397712b | 5474 | while (len > 0) { |
fb85fc9a | 5475 | page = eb->pages[i]; |
d1310b2e CM |
5476 | WARN_ON(!PageUptodate(page)); |
5477 | ||
09cbfeaf | 5478 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 5479 | kaddr = page_address(page); |
d1310b2e | 5480 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
5481 | |
5482 | src += cur; | |
5483 | len -= cur; | |
5484 | offset = 0; | |
5485 | i++; | |
5486 | } | |
5487 | } | |
d1310b2e CM |
5488 | |
5489 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
5490 | unsigned long start, unsigned long len) | |
5491 | { | |
5492 | size_t cur; | |
5493 | size_t offset; | |
5494 | struct page *page; | |
5495 | char *kaddr; | |
09cbfeaf KS |
5496 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
5497 | unsigned long i = (start_offset + start) >> PAGE_SHIFT; | |
d1310b2e CM |
5498 | |
5499 | WARN_ON(start > eb->len); | |
5500 | WARN_ON(start + len > eb->start + eb->len); | |
5501 | ||
09cbfeaf | 5502 | offset = (start_offset + start) & (PAGE_SIZE - 1); |
d1310b2e | 5503 | |
d397712b | 5504 | while (len > 0) { |
fb85fc9a | 5505 | page = eb->pages[i]; |
d1310b2e CM |
5506 | WARN_ON(!PageUptodate(page)); |
5507 | ||
09cbfeaf | 5508 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 5509 | kaddr = page_address(page); |
d1310b2e | 5510 | memset(kaddr + offset, c, cur); |
d1310b2e CM |
5511 | |
5512 | len -= cur; | |
5513 | offset = 0; | |
5514 | i++; | |
5515 | } | |
5516 | } | |
d1310b2e CM |
5517 | |
5518 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
5519 | unsigned long dst_offset, unsigned long src_offset, | |
5520 | unsigned long len) | |
5521 | { | |
5522 | u64 dst_len = dst->len; | |
5523 | size_t cur; | |
5524 | size_t offset; | |
5525 | struct page *page; | |
5526 | char *kaddr; | |
09cbfeaf KS |
5527 | size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1); |
5528 | unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT; | |
d1310b2e CM |
5529 | |
5530 | WARN_ON(src->len != dst_len); | |
5531 | ||
5532 | offset = (start_offset + dst_offset) & | |
09cbfeaf | 5533 | (PAGE_SIZE - 1); |
d1310b2e | 5534 | |
d397712b | 5535 | while (len > 0) { |
fb85fc9a | 5536 | page = dst->pages[i]; |
d1310b2e CM |
5537 | WARN_ON(!PageUptodate(page)); |
5538 | ||
09cbfeaf | 5539 | cur = min(len, (unsigned long)(PAGE_SIZE - offset)); |
d1310b2e | 5540 | |
a6591715 | 5541 | kaddr = page_address(page); |
d1310b2e | 5542 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
5543 | |
5544 | src_offset += cur; | |
5545 | len -= cur; | |
5546 | offset = 0; | |
5547 | i++; | |
5548 | } | |
5549 | } | |
d1310b2e | 5550 | |
3e1e8bb7 OS |
5551 | /* |
5552 | * The extent buffer bitmap operations are done with byte granularity because | |
5553 | * bitmap items are not guaranteed to be aligned to a word and therefore a | |
5554 | * single word in a bitmap may straddle two pages in the extent buffer. | |
5555 | */ | |
5556 | #define BIT_BYTE(nr) ((nr) / BITS_PER_BYTE) | |
5557 | #define BYTE_MASK ((1 << BITS_PER_BYTE) - 1) | |
5558 | #define BITMAP_FIRST_BYTE_MASK(start) \ | |
5559 | ((BYTE_MASK << ((start) & (BITS_PER_BYTE - 1))) & BYTE_MASK) | |
5560 | #define BITMAP_LAST_BYTE_MASK(nbits) \ | |
5561 | (BYTE_MASK >> (-(nbits) & (BITS_PER_BYTE - 1))) | |
5562 | ||
5563 | /* | |
5564 | * eb_bitmap_offset() - calculate the page and offset of the byte containing the | |
5565 | * given bit number | |
5566 | * @eb: the extent buffer | |
5567 | * @start: offset of the bitmap item in the extent buffer | |
5568 | * @nr: bit number | |
5569 | * @page_index: return index of the page in the extent buffer that contains the | |
5570 | * given bit number | |
5571 | * @page_offset: return offset into the page given by page_index | |
5572 | * | |
5573 | * This helper hides the ugliness of finding the byte in an extent buffer which | |
5574 | * contains a given bit. | |
5575 | */ | |
5576 | static inline void eb_bitmap_offset(struct extent_buffer *eb, | |
5577 | unsigned long start, unsigned long nr, | |
5578 | unsigned long *page_index, | |
5579 | size_t *page_offset) | |
5580 | { | |
09cbfeaf | 5581 | size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1); |
3e1e8bb7 OS |
5582 | size_t byte_offset = BIT_BYTE(nr); |
5583 | size_t offset; | |
5584 | ||
5585 | /* | |
5586 | * The byte we want is the offset of the extent buffer + the offset of | |
5587 | * the bitmap item in the extent buffer + the offset of the byte in the | |
5588 | * bitmap item. | |
5589 | */ | |
5590 | offset = start_offset + start + byte_offset; | |
5591 | ||
09cbfeaf KS |
5592 | *page_index = offset >> PAGE_SHIFT; |
5593 | *page_offset = offset & (PAGE_SIZE - 1); | |
3e1e8bb7 OS |
5594 | } |
5595 | ||
5596 | /** | |
5597 | * extent_buffer_test_bit - determine whether a bit in a bitmap item is set | |
5598 | * @eb: the extent buffer | |
5599 | * @start: offset of the bitmap item in the extent buffer | |
5600 | * @nr: bit number to test | |
5601 | */ | |
5602 | int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start, | |
5603 | unsigned long nr) | |
5604 | { | |
5605 | char *kaddr; | |
5606 | struct page *page; | |
5607 | unsigned long i; | |
5608 | size_t offset; | |
5609 | ||
5610 | eb_bitmap_offset(eb, start, nr, &i, &offset); | |
5611 | page = eb->pages[i]; | |
5612 | WARN_ON(!PageUptodate(page)); | |
5613 | kaddr = page_address(page); | |
5614 | return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1))); | |
5615 | } | |
5616 | ||
5617 | /** | |
5618 | * extent_buffer_bitmap_set - set an area of a bitmap | |
5619 | * @eb: the extent buffer | |
5620 | * @start: offset of the bitmap item in the extent buffer | |
5621 | * @pos: bit number of the first bit | |
5622 | * @len: number of bits to set | |
5623 | */ | |
5624 | void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start, | |
5625 | unsigned long pos, unsigned long len) | |
5626 | { | |
5627 | char *kaddr; | |
5628 | struct page *page; | |
5629 | unsigned long i; | |
5630 | size_t offset; | |
5631 | const unsigned int size = pos + len; | |
5632 | int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
5633 | unsigned int mask_to_set = BITMAP_FIRST_BYTE_MASK(pos); | |
5634 | ||
5635 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5636 | page = eb->pages[i]; | |
5637 | WARN_ON(!PageUptodate(page)); | |
5638 | kaddr = page_address(page); | |
5639 | ||
5640 | while (len >= bits_to_set) { | |
5641 | kaddr[offset] |= mask_to_set; | |
5642 | len -= bits_to_set; | |
5643 | bits_to_set = BITS_PER_BYTE; | |
5644 | mask_to_set = ~0U; | |
09cbfeaf | 5645 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5646 | offset = 0; |
5647 | page = eb->pages[++i]; | |
5648 | WARN_ON(!PageUptodate(page)); | |
5649 | kaddr = page_address(page); | |
5650 | } | |
5651 | } | |
5652 | if (len) { | |
5653 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
5654 | kaddr[offset] |= mask_to_set; | |
5655 | } | |
5656 | } | |
5657 | ||
5658 | ||
5659 | /** | |
5660 | * extent_buffer_bitmap_clear - clear an area of a bitmap | |
5661 | * @eb: the extent buffer | |
5662 | * @start: offset of the bitmap item in the extent buffer | |
5663 | * @pos: bit number of the first bit | |
5664 | * @len: number of bits to clear | |
5665 | */ | |
5666 | void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start, | |
5667 | unsigned long pos, unsigned long len) | |
5668 | { | |
5669 | char *kaddr; | |
5670 | struct page *page; | |
5671 | unsigned long i; | |
5672 | size_t offset; | |
5673 | const unsigned int size = pos + len; | |
5674 | int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
5675 | unsigned int mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos); | |
5676 | ||
5677 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5678 | page = eb->pages[i]; | |
5679 | WARN_ON(!PageUptodate(page)); | |
5680 | kaddr = page_address(page); | |
5681 | ||
5682 | while (len >= bits_to_clear) { | |
5683 | kaddr[offset] &= ~mask_to_clear; | |
5684 | len -= bits_to_clear; | |
5685 | bits_to_clear = BITS_PER_BYTE; | |
5686 | mask_to_clear = ~0U; | |
09cbfeaf | 5687 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5688 | offset = 0; |
5689 | page = eb->pages[++i]; | |
5690 | WARN_ON(!PageUptodate(page)); | |
5691 | kaddr = page_address(page); | |
5692 | } | |
5693 | } | |
5694 | if (len) { | |
5695 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
5696 | kaddr[offset] &= ~mask_to_clear; | |
5697 | } | |
5698 | } | |
5699 | ||
3387206f ST |
5700 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
5701 | { | |
5702 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
5703 | return distance < len; | |
5704 | } | |
5705 | ||
d1310b2e CM |
5706 | static void copy_pages(struct page *dst_page, struct page *src_page, |
5707 | unsigned long dst_off, unsigned long src_off, | |
5708 | unsigned long len) | |
5709 | { | |
a6591715 | 5710 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 5711 | char *src_kaddr; |
727011e0 | 5712 | int must_memmove = 0; |
d1310b2e | 5713 | |
3387206f | 5714 | if (dst_page != src_page) { |
a6591715 | 5715 | src_kaddr = page_address(src_page); |
3387206f | 5716 | } else { |
d1310b2e | 5717 | src_kaddr = dst_kaddr; |
727011e0 CM |
5718 | if (areas_overlap(src_off, dst_off, len)) |
5719 | must_memmove = 1; | |
3387206f | 5720 | } |
d1310b2e | 5721 | |
727011e0 CM |
5722 | if (must_memmove) |
5723 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
5724 | else | |
5725 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
5726 | } |
5727 | ||
5728 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5729 | unsigned long src_offset, unsigned long len) | |
5730 | { | |
5731 | size_t cur; | |
5732 | size_t dst_off_in_page; | |
5733 | size_t src_off_in_page; | |
09cbfeaf | 5734 | size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1); |
d1310b2e CM |
5735 | unsigned long dst_i; |
5736 | unsigned long src_i; | |
5737 | ||
5738 | if (src_offset + len > dst->len) { | |
f14d104d DS |
5739 | btrfs_err(dst->fs_info, |
5740 | "memmove bogus src_offset %lu move " | |
5741 | "len %lu dst len %lu", src_offset, len, dst->len); | |
d1310b2e CM |
5742 | BUG_ON(1); |
5743 | } | |
5744 | if (dst_offset + len > dst->len) { | |
f14d104d DS |
5745 | btrfs_err(dst->fs_info, |
5746 | "memmove bogus dst_offset %lu move " | |
5747 | "len %lu dst len %lu", dst_offset, len, dst->len); | |
d1310b2e CM |
5748 | BUG_ON(1); |
5749 | } | |
5750 | ||
d397712b | 5751 | while (len > 0) { |
d1310b2e | 5752 | dst_off_in_page = (start_offset + dst_offset) & |
09cbfeaf | 5753 | (PAGE_SIZE - 1); |
d1310b2e | 5754 | src_off_in_page = (start_offset + src_offset) & |
09cbfeaf | 5755 | (PAGE_SIZE - 1); |
d1310b2e | 5756 | |
09cbfeaf KS |
5757 | dst_i = (start_offset + dst_offset) >> PAGE_SHIFT; |
5758 | src_i = (start_offset + src_offset) >> PAGE_SHIFT; | |
d1310b2e | 5759 | |
09cbfeaf | 5760 | cur = min(len, (unsigned long)(PAGE_SIZE - |
d1310b2e CM |
5761 | src_off_in_page)); |
5762 | cur = min_t(unsigned long, cur, | |
09cbfeaf | 5763 | (unsigned long)(PAGE_SIZE - dst_off_in_page)); |
d1310b2e | 5764 | |
fb85fc9a | 5765 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5766 | dst_off_in_page, src_off_in_page, cur); |
5767 | ||
5768 | src_offset += cur; | |
5769 | dst_offset += cur; | |
5770 | len -= cur; | |
5771 | } | |
5772 | } | |
d1310b2e CM |
5773 | |
5774 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5775 | unsigned long src_offset, unsigned long len) | |
5776 | { | |
5777 | size_t cur; | |
5778 | size_t dst_off_in_page; | |
5779 | size_t src_off_in_page; | |
5780 | unsigned long dst_end = dst_offset + len - 1; | |
5781 | unsigned long src_end = src_offset + len - 1; | |
09cbfeaf | 5782 | size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1); |
d1310b2e CM |
5783 | unsigned long dst_i; |
5784 | unsigned long src_i; | |
5785 | ||
5786 | if (src_offset + len > dst->len) { | |
f14d104d DS |
5787 | btrfs_err(dst->fs_info, "memmove bogus src_offset %lu move " |
5788 | "len %lu len %lu", src_offset, len, dst->len); | |
d1310b2e CM |
5789 | BUG_ON(1); |
5790 | } | |
5791 | if (dst_offset + len > dst->len) { | |
f14d104d DS |
5792 | btrfs_err(dst->fs_info, "memmove bogus dst_offset %lu move " |
5793 | "len %lu len %lu", dst_offset, len, dst->len); | |
d1310b2e CM |
5794 | BUG_ON(1); |
5795 | } | |
727011e0 | 5796 | if (dst_offset < src_offset) { |
d1310b2e CM |
5797 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
5798 | return; | |
5799 | } | |
d397712b | 5800 | while (len > 0) { |
09cbfeaf KS |
5801 | dst_i = (start_offset + dst_end) >> PAGE_SHIFT; |
5802 | src_i = (start_offset + src_end) >> PAGE_SHIFT; | |
d1310b2e CM |
5803 | |
5804 | dst_off_in_page = (start_offset + dst_end) & | |
09cbfeaf | 5805 | (PAGE_SIZE - 1); |
d1310b2e | 5806 | src_off_in_page = (start_offset + src_end) & |
09cbfeaf | 5807 | (PAGE_SIZE - 1); |
d1310b2e CM |
5808 | |
5809 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
5810 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 5811 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5812 | dst_off_in_page - cur + 1, |
5813 | src_off_in_page - cur + 1, cur); | |
5814 | ||
5815 | dst_end -= cur; | |
5816 | src_end -= cur; | |
5817 | len -= cur; | |
5818 | } | |
5819 | } | |
6af118ce | 5820 | |
f7a52a40 | 5821 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 5822 | { |
6af118ce | 5823 | struct extent_buffer *eb; |
6af118ce | 5824 | |
3083ee2e | 5825 | /* |
01327610 | 5826 | * We need to make sure nobody is attaching this page to an eb right |
3083ee2e JB |
5827 | * now. |
5828 | */ | |
5829 | spin_lock(&page->mapping->private_lock); | |
5830 | if (!PagePrivate(page)) { | |
5831 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 5832 | return 1; |
45f49bce | 5833 | } |
6af118ce | 5834 | |
3083ee2e JB |
5835 | eb = (struct extent_buffer *)page->private; |
5836 | BUG_ON(!eb); | |
19fe0a8b MX |
5837 | |
5838 | /* | |
3083ee2e JB |
5839 | * This is a little awful but should be ok, we need to make sure that |
5840 | * the eb doesn't disappear out from under us while we're looking at | |
5841 | * this page. | |
19fe0a8b | 5842 | */ |
3083ee2e | 5843 | spin_lock(&eb->refs_lock); |
0b32f4bb | 5844 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
5845 | spin_unlock(&eb->refs_lock); |
5846 | spin_unlock(&page->mapping->private_lock); | |
5847 | return 0; | |
b9473439 | 5848 | } |
3083ee2e | 5849 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 5850 | |
19fe0a8b | 5851 | /* |
3083ee2e JB |
5852 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
5853 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 5854 | */ |
3083ee2e JB |
5855 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
5856 | spin_unlock(&eb->refs_lock); | |
5857 | return 0; | |
b9473439 | 5858 | } |
19fe0a8b | 5859 | |
f7a52a40 | 5860 | return release_extent_buffer(eb); |
6af118ce | 5861 | } |