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" |
d1310b2e | 23 | |
d1310b2e CM |
24 | static struct kmem_cache *extent_state_cache; |
25 | static struct kmem_cache *extent_buffer_cache; | |
9be3395b | 26 | static struct bio_set *btrfs_bioset; |
d1310b2e | 27 | |
27a3507d FM |
28 | static inline bool extent_state_in_tree(const struct extent_state *state) |
29 | { | |
30 | return !RB_EMPTY_NODE(&state->rb_node); | |
31 | } | |
32 | ||
6d49ba1b | 33 | #ifdef CONFIG_BTRFS_DEBUG |
d1310b2e CM |
34 | static LIST_HEAD(buffers); |
35 | static LIST_HEAD(states); | |
4bef0848 | 36 | |
d397712b | 37 | static DEFINE_SPINLOCK(leak_lock); |
6d49ba1b ES |
38 | |
39 | static inline | |
40 | void btrfs_leak_debug_add(struct list_head *new, struct list_head *head) | |
41 | { | |
42 | unsigned long flags; | |
43 | ||
44 | spin_lock_irqsave(&leak_lock, flags); | |
45 | list_add(new, head); | |
46 | spin_unlock_irqrestore(&leak_lock, flags); | |
47 | } | |
48 | ||
49 | static inline | |
50 | void btrfs_leak_debug_del(struct list_head *entry) | |
51 | { | |
52 | unsigned long flags; | |
53 | ||
54 | spin_lock_irqsave(&leak_lock, flags); | |
55 | list_del(entry); | |
56 | spin_unlock_irqrestore(&leak_lock, flags); | |
57 | } | |
58 | ||
59 | static inline | |
60 | void btrfs_leak_debug_check(void) | |
61 | { | |
62 | struct extent_state *state; | |
63 | struct extent_buffer *eb; | |
64 | ||
65 | while (!list_empty(&states)) { | |
66 | state = list_entry(states.next, struct extent_state, leak_list); | |
27a3507d FM |
67 | pr_err("BTRFS: state leak: start %llu end %llu state %lu in tree %d refs %d\n", |
68 | state->start, state->end, state->state, | |
69 | extent_state_in_tree(state), | |
c1c9ff7c | 70 | atomic_read(&state->refs)); |
6d49ba1b ES |
71 | list_del(&state->leak_list); |
72 | kmem_cache_free(extent_state_cache, state); | |
73 | } | |
74 | ||
75 | while (!list_empty(&buffers)) { | |
76 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
efe120a0 | 77 | printk(KERN_ERR "BTRFS: buffer leak start %llu len %lu " |
c1c9ff7c GU |
78 | "refs %d\n", |
79 | eb->start, eb->len, atomic_read(&eb->refs)); | |
6d49ba1b ES |
80 | list_del(&eb->leak_list); |
81 | kmem_cache_free(extent_buffer_cache, eb); | |
82 | } | |
83 | } | |
8d599ae1 | 84 | |
a5dee37d JB |
85 | #define btrfs_debug_check_extent_io_range(tree, start, end) \ |
86 | __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end)) | |
8d599ae1 | 87 | static inline void __btrfs_debug_check_extent_io_range(const char *caller, |
a5dee37d | 88 | struct extent_io_tree *tree, u64 start, u64 end) |
8d599ae1 | 89 | { |
a5dee37d JB |
90 | struct inode *inode; |
91 | u64 isize; | |
8d599ae1 | 92 | |
a5dee37d JB |
93 | if (!tree->mapping) |
94 | return; | |
8d599ae1 | 95 | |
a5dee37d JB |
96 | inode = tree->mapping->host; |
97 | isize = i_size_read(inode); | |
8d599ae1 DS |
98 | if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) { |
99 | printk_ratelimited(KERN_DEBUG | |
efe120a0 | 100 | "BTRFS: %s: ino %llu isize %llu odd range [%llu,%llu]\n", |
c1c9ff7c | 101 | caller, btrfs_ino(inode), isize, start, end); |
8d599ae1 DS |
102 | } |
103 | } | |
6d49ba1b ES |
104 | #else |
105 | #define btrfs_leak_debug_add(new, head) do {} while (0) | |
106 | #define btrfs_leak_debug_del(entry) do {} while (0) | |
107 | #define btrfs_leak_debug_check() do {} while (0) | |
8d599ae1 | 108 | #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0) |
4bef0848 | 109 | #endif |
d1310b2e | 110 | |
d1310b2e CM |
111 | #define BUFFER_LRU_MAX 64 |
112 | ||
113 | struct tree_entry { | |
114 | u64 start; | |
115 | u64 end; | |
d1310b2e CM |
116 | struct rb_node rb_node; |
117 | }; | |
118 | ||
119 | struct extent_page_data { | |
120 | struct bio *bio; | |
121 | struct extent_io_tree *tree; | |
122 | get_extent_t *get_extent; | |
de0022b9 | 123 | unsigned long bio_flags; |
771ed689 CM |
124 | |
125 | /* tells writepage not to lock the state bits for this range | |
126 | * it still does the unlocking | |
127 | */ | |
ffbd517d CM |
128 | unsigned int extent_locked:1; |
129 | ||
130 | /* tells the submit_bio code to use a WRITE_SYNC */ | |
131 | unsigned int sync_io:1; | |
d1310b2e CM |
132 | }; |
133 | ||
0b32f4bb | 134 | static noinline void flush_write_bio(void *data); |
c2d904e0 JM |
135 | static inline struct btrfs_fs_info * |
136 | tree_fs_info(struct extent_io_tree *tree) | |
137 | { | |
a5dee37d JB |
138 | if (!tree->mapping) |
139 | return NULL; | |
c2d904e0 JM |
140 | return btrfs_sb(tree->mapping->host->i_sb); |
141 | } | |
0b32f4bb | 142 | |
d1310b2e CM |
143 | int __init extent_io_init(void) |
144 | { | |
837e1972 | 145 | extent_state_cache = kmem_cache_create("btrfs_extent_state", |
9601e3f6 CH |
146 | sizeof(struct extent_state), 0, |
147 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
148 | if (!extent_state_cache) |
149 | return -ENOMEM; | |
150 | ||
837e1972 | 151 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 CH |
152 | sizeof(struct extent_buffer), 0, |
153 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
154 | if (!extent_buffer_cache) |
155 | goto free_state_cache; | |
9be3395b CM |
156 | |
157 | btrfs_bioset = bioset_create(BIO_POOL_SIZE, | |
158 | offsetof(struct btrfs_io_bio, bio)); | |
159 | if (!btrfs_bioset) | |
160 | goto free_buffer_cache; | |
b208c2f7 DW |
161 | |
162 | if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE)) | |
163 | goto free_bioset; | |
164 | ||
d1310b2e CM |
165 | return 0; |
166 | ||
b208c2f7 DW |
167 | free_bioset: |
168 | bioset_free(btrfs_bioset); | |
169 | btrfs_bioset = NULL; | |
170 | ||
9be3395b CM |
171 | free_buffer_cache: |
172 | kmem_cache_destroy(extent_buffer_cache); | |
173 | extent_buffer_cache = NULL; | |
174 | ||
d1310b2e CM |
175 | free_state_cache: |
176 | kmem_cache_destroy(extent_state_cache); | |
9be3395b | 177 | extent_state_cache = NULL; |
d1310b2e CM |
178 | return -ENOMEM; |
179 | } | |
180 | ||
181 | void extent_io_exit(void) | |
182 | { | |
6d49ba1b | 183 | btrfs_leak_debug_check(); |
8c0a8537 KS |
184 | |
185 | /* | |
186 | * Make sure all delayed rcu free are flushed before we | |
187 | * destroy caches. | |
188 | */ | |
189 | rcu_barrier(); | |
d1310b2e CM |
190 | if (extent_state_cache) |
191 | kmem_cache_destroy(extent_state_cache); | |
192 | if (extent_buffer_cache) | |
193 | kmem_cache_destroy(extent_buffer_cache); | |
9be3395b CM |
194 | if (btrfs_bioset) |
195 | bioset_free(btrfs_bioset); | |
d1310b2e CM |
196 | } |
197 | ||
198 | void extent_io_tree_init(struct extent_io_tree *tree, | |
f993c883 | 199 | struct address_space *mapping) |
d1310b2e | 200 | { |
6bef4d31 | 201 | tree->state = RB_ROOT; |
d1310b2e CM |
202 | tree->ops = NULL; |
203 | tree->dirty_bytes = 0; | |
70dec807 | 204 | spin_lock_init(&tree->lock); |
d1310b2e | 205 | tree->mapping = mapping; |
d1310b2e | 206 | } |
d1310b2e | 207 | |
b2950863 | 208 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
209 | { |
210 | struct extent_state *state; | |
d1310b2e CM |
211 | |
212 | state = kmem_cache_alloc(extent_state_cache, mask); | |
2b114d1d | 213 | if (!state) |
d1310b2e CM |
214 | return state; |
215 | state->state = 0; | |
d1310b2e | 216 | state->private = 0; |
27a3507d | 217 | RB_CLEAR_NODE(&state->rb_node); |
6d49ba1b | 218 | btrfs_leak_debug_add(&state->leak_list, &states); |
d1310b2e CM |
219 | atomic_set(&state->refs, 1); |
220 | init_waitqueue_head(&state->wq); | |
143bede5 | 221 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
222 | return state; |
223 | } | |
d1310b2e | 224 | |
4845e44f | 225 | void free_extent_state(struct extent_state *state) |
d1310b2e | 226 | { |
d1310b2e CM |
227 | if (!state) |
228 | return; | |
229 | if (atomic_dec_and_test(&state->refs)) { | |
27a3507d | 230 | WARN_ON(extent_state_in_tree(state)); |
6d49ba1b | 231 | btrfs_leak_debug_del(&state->leak_list); |
143bede5 | 232 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
233 | kmem_cache_free(extent_state_cache, state); |
234 | } | |
235 | } | |
d1310b2e | 236 | |
f2071b21 FM |
237 | static struct rb_node *tree_insert(struct rb_root *root, |
238 | struct rb_node *search_start, | |
239 | u64 offset, | |
12cfbad9 FDBM |
240 | struct rb_node *node, |
241 | struct rb_node ***p_in, | |
242 | struct rb_node **parent_in) | |
d1310b2e | 243 | { |
f2071b21 | 244 | struct rb_node **p; |
d397712b | 245 | struct rb_node *parent = NULL; |
d1310b2e CM |
246 | struct tree_entry *entry; |
247 | ||
12cfbad9 FDBM |
248 | if (p_in && parent_in) { |
249 | p = *p_in; | |
250 | parent = *parent_in; | |
251 | goto do_insert; | |
252 | } | |
253 | ||
f2071b21 | 254 | p = search_start ? &search_start : &root->rb_node; |
d397712b | 255 | while (*p) { |
d1310b2e CM |
256 | parent = *p; |
257 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
258 | ||
259 | if (offset < entry->start) | |
260 | p = &(*p)->rb_left; | |
261 | else if (offset > entry->end) | |
262 | p = &(*p)->rb_right; | |
263 | else | |
264 | return parent; | |
265 | } | |
266 | ||
12cfbad9 | 267 | do_insert: |
d1310b2e CM |
268 | rb_link_node(node, parent, p); |
269 | rb_insert_color(node, root); | |
270 | return NULL; | |
271 | } | |
272 | ||
80ea96b1 | 273 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
12cfbad9 FDBM |
274 | struct rb_node **prev_ret, |
275 | struct rb_node **next_ret, | |
276 | struct rb_node ***p_ret, | |
277 | struct rb_node **parent_ret) | |
d1310b2e | 278 | { |
80ea96b1 | 279 | struct rb_root *root = &tree->state; |
12cfbad9 | 280 | struct rb_node **n = &root->rb_node; |
d1310b2e CM |
281 | struct rb_node *prev = NULL; |
282 | struct rb_node *orig_prev = NULL; | |
283 | struct tree_entry *entry; | |
284 | struct tree_entry *prev_entry = NULL; | |
285 | ||
12cfbad9 FDBM |
286 | while (*n) { |
287 | prev = *n; | |
288 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
d1310b2e CM |
289 | prev_entry = entry; |
290 | ||
291 | if (offset < entry->start) | |
12cfbad9 | 292 | n = &(*n)->rb_left; |
d1310b2e | 293 | else if (offset > entry->end) |
12cfbad9 | 294 | n = &(*n)->rb_right; |
d397712b | 295 | else |
12cfbad9 | 296 | return *n; |
d1310b2e CM |
297 | } |
298 | ||
12cfbad9 FDBM |
299 | if (p_ret) |
300 | *p_ret = n; | |
301 | if (parent_ret) | |
302 | *parent_ret = prev; | |
303 | ||
d1310b2e CM |
304 | if (prev_ret) { |
305 | orig_prev = prev; | |
d397712b | 306 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
307 | prev = rb_next(prev); |
308 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
309 | } | |
310 | *prev_ret = prev; | |
311 | prev = orig_prev; | |
312 | } | |
313 | ||
314 | if (next_ret) { | |
315 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 316 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
317 | prev = rb_prev(prev); |
318 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
319 | } | |
320 | *next_ret = prev; | |
321 | } | |
322 | return NULL; | |
323 | } | |
324 | ||
12cfbad9 FDBM |
325 | static inline struct rb_node * |
326 | tree_search_for_insert(struct extent_io_tree *tree, | |
327 | u64 offset, | |
328 | struct rb_node ***p_ret, | |
329 | struct rb_node **parent_ret) | |
d1310b2e | 330 | { |
70dec807 | 331 | struct rb_node *prev = NULL; |
d1310b2e | 332 | struct rb_node *ret; |
70dec807 | 333 | |
12cfbad9 | 334 | ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret); |
d397712b | 335 | if (!ret) |
d1310b2e CM |
336 | return prev; |
337 | return ret; | |
338 | } | |
339 | ||
12cfbad9 FDBM |
340 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
341 | u64 offset) | |
342 | { | |
343 | return tree_search_for_insert(tree, offset, NULL, NULL); | |
344 | } | |
345 | ||
9ed74f2d JB |
346 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, |
347 | struct extent_state *other) | |
348 | { | |
349 | if (tree->ops && tree->ops->merge_extent_hook) | |
350 | tree->ops->merge_extent_hook(tree->mapping->host, new, | |
351 | other); | |
352 | } | |
353 | ||
d1310b2e CM |
354 | /* |
355 | * utility function to look for merge candidates inside a given range. | |
356 | * Any extents with matching state are merged together into a single | |
357 | * extent in the tree. Extents with EXTENT_IO in their state field | |
358 | * are not merged because the end_io handlers need to be able to do | |
359 | * operations on them without sleeping (or doing allocations/splits). | |
360 | * | |
361 | * This should be called with the tree lock held. | |
362 | */ | |
1bf85046 JM |
363 | static void merge_state(struct extent_io_tree *tree, |
364 | struct extent_state *state) | |
d1310b2e CM |
365 | { |
366 | struct extent_state *other; | |
367 | struct rb_node *other_node; | |
368 | ||
5b21f2ed | 369 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
1bf85046 | 370 | return; |
d1310b2e CM |
371 | |
372 | other_node = rb_prev(&state->rb_node); | |
373 | if (other_node) { | |
374 | other = rb_entry(other_node, struct extent_state, rb_node); | |
375 | if (other->end == state->start - 1 && | |
376 | other->state == state->state) { | |
9ed74f2d | 377 | merge_cb(tree, state, other); |
d1310b2e | 378 | state->start = other->start; |
d1310b2e | 379 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 380 | RB_CLEAR_NODE(&other->rb_node); |
d1310b2e CM |
381 | free_extent_state(other); |
382 | } | |
383 | } | |
384 | other_node = rb_next(&state->rb_node); | |
385 | if (other_node) { | |
386 | other = rb_entry(other_node, struct extent_state, rb_node); | |
387 | if (other->start == state->end + 1 && | |
388 | other->state == state->state) { | |
9ed74f2d | 389 | merge_cb(tree, state, other); |
df98b6e2 | 390 | state->end = other->end; |
df98b6e2 | 391 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 392 | RB_CLEAR_NODE(&other->rb_node); |
df98b6e2 | 393 | free_extent_state(other); |
d1310b2e CM |
394 | } |
395 | } | |
d1310b2e CM |
396 | } |
397 | ||
1bf85046 | 398 | static void set_state_cb(struct extent_io_tree *tree, |
41074888 | 399 | struct extent_state *state, unsigned long *bits) |
291d673e | 400 | { |
1bf85046 JM |
401 | if (tree->ops && tree->ops->set_bit_hook) |
402 | tree->ops->set_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
403 | } |
404 | ||
405 | static void clear_state_cb(struct extent_io_tree *tree, | |
41074888 | 406 | struct extent_state *state, unsigned long *bits) |
291d673e | 407 | { |
9ed74f2d JB |
408 | if (tree->ops && tree->ops->clear_bit_hook) |
409 | tree->ops->clear_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
410 | } |
411 | ||
3150b699 | 412 | static void set_state_bits(struct extent_io_tree *tree, |
41074888 | 413 | struct extent_state *state, unsigned long *bits); |
3150b699 | 414 | |
d1310b2e CM |
415 | /* |
416 | * insert an extent_state struct into the tree. 'bits' are set on the | |
417 | * struct before it is inserted. | |
418 | * | |
419 | * This may return -EEXIST if the extent is already there, in which case the | |
420 | * state struct is freed. | |
421 | * | |
422 | * The tree lock is not taken internally. This is a utility function and | |
423 | * probably isn't what you want to call (see set/clear_extent_bit). | |
424 | */ | |
425 | static int insert_state(struct extent_io_tree *tree, | |
426 | struct extent_state *state, u64 start, u64 end, | |
12cfbad9 FDBM |
427 | struct rb_node ***p, |
428 | struct rb_node **parent, | |
41074888 | 429 | unsigned long *bits) |
d1310b2e CM |
430 | { |
431 | struct rb_node *node; | |
432 | ||
31b1a2bd | 433 | if (end < start) |
efe120a0 | 434 | WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n", |
c1c9ff7c | 435 | end, start); |
d1310b2e CM |
436 | state->start = start; |
437 | state->end = end; | |
9ed74f2d | 438 | |
3150b699 XG |
439 | set_state_bits(tree, state, bits); |
440 | ||
f2071b21 | 441 | node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent); |
d1310b2e CM |
442 | if (node) { |
443 | struct extent_state *found; | |
444 | found = rb_entry(node, struct extent_state, rb_node); | |
efe120a0 | 445 | printk(KERN_ERR "BTRFS: found node %llu %llu on insert of " |
c1c9ff7c GU |
446 | "%llu %llu\n", |
447 | found->start, found->end, start, end); | |
d1310b2e CM |
448 | return -EEXIST; |
449 | } | |
450 | merge_state(tree, state); | |
451 | return 0; | |
452 | } | |
453 | ||
1bf85046 | 454 | static void split_cb(struct extent_io_tree *tree, struct extent_state *orig, |
9ed74f2d JB |
455 | u64 split) |
456 | { | |
457 | if (tree->ops && tree->ops->split_extent_hook) | |
1bf85046 | 458 | tree->ops->split_extent_hook(tree->mapping->host, orig, split); |
9ed74f2d JB |
459 | } |
460 | ||
d1310b2e CM |
461 | /* |
462 | * split a given extent state struct in two, inserting the preallocated | |
463 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
464 | * offset inside 'orig' where it should be split. | |
465 | * | |
466 | * Before calling, | |
467 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
468 | * are two extent state structs in the tree: | |
469 | * prealloc: [orig->start, split - 1] | |
470 | * orig: [ split, orig->end ] | |
471 | * | |
472 | * The tree locks are not taken by this function. They need to be held | |
473 | * by the caller. | |
474 | */ | |
475 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
476 | struct extent_state *prealloc, u64 split) | |
477 | { | |
478 | struct rb_node *node; | |
9ed74f2d JB |
479 | |
480 | split_cb(tree, orig, split); | |
481 | ||
d1310b2e CM |
482 | prealloc->start = orig->start; |
483 | prealloc->end = split - 1; | |
484 | prealloc->state = orig->state; | |
485 | orig->start = split; | |
486 | ||
f2071b21 FM |
487 | node = tree_insert(&tree->state, &orig->rb_node, prealloc->end, |
488 | &prealloc->rb_node, NULL, NULL); | |
d1310b2e | 489 | if (node) { |
d1310b2e CM |
490 | free_extent_state(prealloc); |
491 | return -EEXIST; | |
492 | } | |
493 | return 0; | |
494 | } | |
495 | ||
cdc6a395 LZ |
496 | static struct extent_state *next_state(struct extent_state *state) |
497 | { | |
498 | struct rb_node *next = rb_next(&state->rb_node); | |
499 | if (next) | |
500 | return rb_entry(next, struct extent_state, rb_node); | |
501 | else | |
502 | return NULL; | |
503 | } | |
504 | ||
d1310b2e CM |
505 | /* |
506 | * utility function to clear some bits in an extent state struct. | |
1b303fc0 | 507 | * it will optionally wake up any one waiting on this state (wake == 1). |
d1310b2e CM |
508 | * |
509 | * If no bits are set on the state struct after clearing things, the | |
510 | * struct is freed and removed from the tree | |
511 | */ | |
cdc6a395 LZ |
512 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
513 | struct extent_state *state, | |
41074888 | 514 | unsigned long *bits, int wake) |
d1310b2e | 515 | { |
cdc6a395 | 516 | struct extent_state *next; |
41074888 | 517 | unsigned long bits_to_clear = *bits & ~EXTENT_CTLBITS; |
d1310b2e | 518 | |
0ca1f7ce | 519 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
520 | u64 range = state->end - state->start + 1; |
521 | WARN_ON(range > tree->dirty_bytes); | |
522 | tree->dirty_bytes -= range; | |
523 | } | |
291d673e | 524 | clear_state_cb(tree, state, bits); |
32c00aff | 525 | state->state &= ~bits_to_clear; |
d1310b2e CM |
526 | if (wake) |
527 | wake_up(&state->wq); | |
0ca1f7ce | 528 | if (state->state == 0) { |
cdc6a395 | 529 | next = next_state(state); |
27a3507d | 530 | if (extent_state_in_tree(state)) { |
d1310b2e | 531 | rb_erase(&state->rb_node, &tree->state); |
27a3507d | 532 | RB_CLEAR_NODE(&state->rb_node); |
d1310b2e CM |
533 | free_extent_state(state); |
534 | } else { | |
535 | WARN_ON(1); | |
536 | } | |
537 | } else { | |
538 | merge_state(tree, state); | |
cdc6a395 | 539 | next = next_state(state); |
d1310b2e | 540 | } |
cdc6a395 | 541 | return next; |
d1310b2e CM |
542 | } |
543 | ||
8233767a XG |
544 | static struct extent_state * |
545 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
546 | { | |
547 | if (!prealloc) | |
548 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
549 | ||
550 | return prealloc; | |
551 | } | |
552 | ||
48a3b636 | 553 | static void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
c2d904e0 JM |
554 | { |
555 | btrfs_panic(tree_fs_info(tree), err, "Locking error: " | |
556 | "Extent tree was modified by another " | |
557 | "thread while locked."); | |
558 | } | |
559 | ||
d1310b2e CM |
560 | /* |
561 | * clear some bits on a range in the tree. This may require splitting | |
562 | * or inserting elements in the tree, so the gfp mask is used to | |
563 | * indicate which allocations or sleeping are allowed. | |
564 | * | |
565 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
566 | * the given range from the tree regardless of state (ie for truncate). | |
567 | * | |
568 | * the range [start, end] is inclusive. | |
569 | * | |
6763af84 | 570 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e CM |
571 | */ |
572 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
41074888 | 573 | unsigned long bits, int wake, int delete, |
2c64c53d CM |
574 | struct extent_state **cached_state, |
575 | gfp_t mask) | |
d1310b2e CM |
576 | { |
577 | struct extent_state *state; | |
2c64c53d | 578 | struct extent_state *cached; |
d1310b2e CM |
579 | struct extent_state *prealloc = NULL; |
580 | struct rb_node *node; | |
5c939df5 | 581 | u64 last_end; |
d1310b2e | 582 | int err; |
2ac55d41 | 583 | int clear = 0; |
d1310b2e | 584 | |
a5dee37d | 585 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 586 | |
7ee9e440 JB |
587 | if (bits & EXTENT_DELALLOC) |
588 | bits |= EXTENT_NORESERVE; | |
589 | ||
0ca1f7ce YZ |
590 | if (delete) |
591 | bits |= ~EXTENT_CTLBITS; | |
592 | bits |= EXTENT_FIRST_DELALLOC; | |
593 | ||
2ac55d41 JB |
594 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
595 | clear = 1; | |
d1310b2e CM |
596 | again: |
597 | if (!prealloc && (mask & __GFP_WAIT)) { | |
598 | prealloc = alloc_extent_state(mask); | |
599 | if (!prealloc) | |
600 | return -ENOMEM; | |
601 | } | |
602 | ||
cad321ad | 603 | spin_lock(&tree->lock); |
2c64c53d CM |
604 | if (cached_state) { |
605 | cached = *cached_state; | |
2ac55d41 JB |
606 | |
607 | if (clear) { | |
608 | *cached_state = NULL; | |
609 | cached_state = NULL; | |
610 | } | |
611 | ||
27a3507d FM |
612 | if (cached && extent_state_in_tree(cached) && |
613 | cached->start <= start && cached->end > start) { | |
2ac55d41 JB |
614 | if (clear) |
615 | atomic_dec(&cached->refs); | |
2c64c53d | 616 | state = cached; |
42daec29 | 617 | goto hit_next; |
2c64c53d | 618 | } |
2ac55d41 JB |
619 | if (clear) |
620 | free_extent_state(cached); | |
2c64c53d | 621 | } |
d1310b2e CM |
622 | /* |
623 | * this search will find the extents that end after | |
624 | * our range starts | |
625 | */ | |
80ea96b1 | 626 | node = tree_search(tree, start); |
d1310b2e CM |
627 | if (!node) |
628 | goto out; | |
629 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 630 | hit_next: |
d1310b2e CM |
631 | if (state->start > end) |
632 | goto out; | |
633 | WARN_ON(state->end < start); | |
5c939df5 | 634 | last_end = state->end; |
d1310b2e | 635 | |
0449314a | 636 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
637 | if (!(state->state & bits)) { |
638 | state = next_state(state); | |
0449314a | 639 | goto next; |
cdc6a395 | 640 | } |
0449314a | 641 | |
d1310b2e CM |
642 | /* |
643 | * | ---- desired range ---- | | |
644 | * | state | or | |
645 | * | ------------- state -------------- | | |
646 | * | |
647 | * We need to split the extent we found, and may flip | |
648 | * bits on second half. | |
649 | * | |
650 | * If the extent we found extends past our range, we | |
651 | * just split and search again. It'll get split again | |
652 | * the next time though. | |
653 | * | |
654 | * If the extent we found is inside our range, we clear | |
655 | * the desired bit on it. | |
656 | */ | |
657 | ||
658 | if (state->start < start) { | |
8233767a XG |
659 | prealloc = alloc_extent_state_atomic(prealloc); |
660 | BUG_ON(!prealloc); | |
d1310b2e | 661 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
662 | if (err) |
663 | extent_io_tree_panic(tree, err); | |
664 | ||
d1310b2e CM |
665 | prealloc = NULL; |
666 | if (err) | |
667 | goto out; | |
668 | if (state->end <= end) { | |
d1ac6e41 LB |
669 | state = clear_state_bit(tree, state, &bits, wake); |
670 | goto next; | |
d1310b2e CM |
671 | } |
672 | goto search_again; | |
673 | } | |
674 | /* | |
675 | * | ---- desired range ---- | | |
676 | * | state | | |
677 | * We need to split the extent, and clear the bit | |
678 | * on the first half | |
679 | */ | |
680 | if (state->start <= end && state->end > end) { | |
8233767a XG |
681 | prealloc = alloc_extent_state_atomic(prealloc); |
682 | BUG_ON(!prealloc); | |
d1310b2e | 683 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
684 | if (err) |
685 | extent_io_tree_panic(tree, err); | |
686 | ||
d1310b2e CM |
687 | if (wake) |
688 | wake_up(&state->wq); | |
42daec29 | 689 | |
6763af84 | 690 | clear_state_bit(tree, prealloc, &bits, wake); |
9ed74f2d | 691 | |
d1310b2e CM |
692 | prealloc = NULL; |
693 | goto out; | |
694 | } | |
42daec29 | 695 | |
cdc6a395 | 696 | state = clear_state_bit(tree, state, &bits, wake); |
0449314a | 697 | next: |
5c939df5 YZ |
698 | if (last_end == (u64)-1) |
699 | goto out; | |
700 | start = last_end + 1; | |
cdc6a395 | 701 | if (start <= end && state && !need_resched()) |
692e5759 | 702 | goto hit_next; |
d1310b2e CM |
703 | goto search_again; |
704 | ||
705 | out: | |
cad321ad | 706 | spin_unlock(&tree->lock); |
d1310b2e CM |
707 | if (prealloc) |
708 | free_extent_state(prealloc); | |
709 | ||
6763af84 | 710 | return 0; |
d1310b2e CM |
711 | |
712 | search_again: | |
713 | if (start > end) | |
714 | goto out; | |
cad321ad | 715 | spin_unlock(&tree->lock); |
d1310b2e CM |
716 | if (mask & __GFP_WAIT) |
717 | cond_resched(); | |
718 | goto again; | |
719 | } | |
d1310b2e | 720 | |
143bede5 JM |
721 | static void wait_on_state(struct extent_io_tree *tree, |
722 | struct extent_state *state) | |
641f5219 CH |
723 | __releases(tree->lock) |
724 | __acquires(tree->lock) | |
d1310b2e CM |
725 | { |
726 | DEFINE_WAIT(wait); | |
727 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 728 | spin_unlock(&tree->lock); |
d1310b2e | 729 | schedule(); |
cad321ad | 730 | spin_lock(&tree->lock); |
d1310b2e | 731 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
732 | } |
733 | ||
734 | /* | |
735 | * waits for one or more bits to clear on a range in the state tree. | |
736 | * The range [start, end] is inclusive. | |
737 | * The tree lock is taken by this function | |
738 | */ | |
41074888 DS |
739 | static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
740 | unsigned long bits) | |
d1310b2e CM |
741 | { |
742 | struct extent_state *state; | |
743 | struct rb_node *node; | |
744 | ||
a5dee37d | 745 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 746 | |
cad321ad | 747 | spin_lock(&tree->lock); |
d1310b2e CM |
748 | again: |
749 | while (1) { | |
750 | /* | |
751 | * this search will find all the extents that end after | |
752 | * our range starts | |
753 | */ | |
80ea96b1 | 754 | node = tree_search(tree, start); |
c50d3e71 | 755 | process_node: |
d1310b2e CM |
756 | if (!node) |
757 | break; | |
758 | ||
759 | state = rb_entry(node, struct extent_state, rb_node); | |
760 | ||
761 | if (state->start > end) | |
762 | goto out; | |
763 | ||
764 | if (state->state & bits) { | |
765 | start = state->start; | |
766 | atomic_inc(&state->refs); | |
767 | wait_on_state(tree, state); | |
768 | free_extent_state(state); | |
769 | goto again; | |
770 | } | |
771 | start = state->end + 1; | |
772 | ||
773 | if (start > end) | |
774 | break; | |
775 | ||
c50d3e71 FM |
776 | if (!cond_resched_lock(&tree->lock)) { |
777 | node = rb_next(node); | |
778 | goto process_node; | |
779 | } | |
d1310b2e CM |
780 | } |
781 | out: | |
cad321ad | 782 | spin_unlock(&tree->lock); |
d1310b2e | 783 | } |
d1310b2e | 784 | |
1bf85046 | 785 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 786 | struct extent_state *state, |
41074888 | 787 | unsigned long *bits) |
d1310b2e | 788 | { |
41074888 | 789 | unsigned long bits_to_set = *bits & ~EXTENT_CTLBITS; |
9ed74f2d | 790 | |
1bf85046 | 791 | set_state_cb(tree, state, bits); |
0ca1f7ce | 792 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
793 | u64 range = state->end - state->start + 1; |
794 | tree->dirty_bytes += range; | |
795 | } | |
0ca1f7ce | 796 | state->state |= bits_to_set; |
d1310b2e CM |
797 | } |
798 | ||
2c64c53d CM |
799 | static void cache_state(struct extent_state *state, |
800 | struct extent_state **cached_ptr) | |
801 | { | |
802 | if (cached_ptr && !(*cached_ptr)) { | |
803 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) { | |
804 | *cached_ptr = state; | |
805 | atomic_inc(&state->refs); | |
806 | } | |
807 | } | |
808 | } | |
809 | ||
d1310b2e | 810 | /* |
1edbb734 CM |
811 | * set some bits on a range in the tree. This may require allocations or |
812 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 813 | * |
1edbb734 CM |
814 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
815 | * part of the range already has the desired bits set. The start of the | |
816 | * existing range is returned in failed_start in this case. | |
d1310b2e | 817 | * |
1edbb734 | 818 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 819 | */ |
1edbb734 | 820 | |
3fbe5c02 JM |
821 | static int __must_check |
822 | __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
41074888 DS |
823 | unsigned long bits, unsigned long exclusive_bits, |
824 | u64 *failed_start, struct extent_state **cached_state, | |
825 | gfp_t mask) | |
d1310b2e CM |
826 | { |
827 | struct extent_state *state; | |
828 | struct extent_state *prealloc = NULL; | |
829 | struct rb_node *node; | |
12cfbad9 FDBM |
830 | struct rb_node **p; |
831 | struct rb_node *parent; | |
d1310b2e | 832 | int err = 0; |
d1310b2e CM |
833 | u64 last_start; |
834 | u64 last_end; | |
42daec29 | 835 | |
a5dee37d | 836 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 837 | |
0ca1f7ce | 838 | bits |= EXTENT_FIRST_DELALLOC; |
d1310b2e CM |
839 | again: |
840 | if (!prealloc && (mask & __GFP_WAIT)) { | |
841 | prealloc = alloc_extent_state(mask); | |
8233767a | 842 | BUG_ON(!prealloc); |
d1310b2e CM |
843 | } |
844 | ||
cad321ad | 845 | spin_lock(&tree->lock); |
9655d298 CM |
846 | if (cached_state && *cached_state) { |
847 | state = *cached_state; | |
df98b6e2 | 848 | if (state->start <= start && state->end > start && |
27a3507d | 849 | extent_state_in_tree(state)) { |
9655d298 CM |
850 | node = &state->rb_node; |
851 | goto hit_next; | |
852 | } | |
853 | } | |
d1310b2e CM |
854 | /* |
855 | * this search will find all the extents that end after | |
856 | * our range starts. | |
857 | */ | |
12cfbad9 | 858 | node = tree_search_for_insert(tree, start, &p, &parent); |
d1310b2e | 859 | if (!node) { |
8233767a XG |
860 | prealloc = alloc_extent_state_atomic(prealloc); |
861 | BUG_ON(!prealloc); | |
12cfbad9 FDBM |
862 | err = insert_state(tree, prealloc, start, end, |
863 | &p, &parent, &bits); | |
c2d904e0 JM |
864 | if (err) |
865 | extent_io_tree_panic(tree, err); | |
866 | ||
c42ac0bc | 867 | cache_state(prealloc, cached_state); |
d1310b2e | 868 | prealloc = NULL; |
d1310b2e CM |
869 | goto out; |
870 | } | |
d1310b2e | 871 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 872 | hit_next: |
d1310b2e CM |
873 | last_start = state->start; |
874 | last_end = state->end; | |
875 | ||
876 | /* | |
877 | * | ---- desired range ---- | | |
878 | * | state | | |
879 | * | |
880 | * Just lock what we found and keep going | |
881 | */ | |
882 | if (state->start == start && state->end <= end) { | |
1edbb734 | 883 | if (state->state & exclusive_bits) { |
d1310b2e CM |
884 | *failed_start = state->start; |
885 | err = -EEXIST; | |
886 | goto out; | |
887 | } | |
42daec29 | 888 | |
1bf85046 | 889 | set_state_bits(tree, state, &bits); |
2c64c53d | 890 | cache_state(state, cached_state); |
d1310b2e | 891 | merge_state(tree, state); |
5c939df5 YZ |
892 | if (last_end == (u64)-1) |
893 | goto out; | |
894 | start = last_end + 1; | |
d1ac6e41 LB |
895 | state = next_state(state); |
896 | if (start < end && state && state->start == start && | |
897 | !need_resched()) | |
898 | goto hit_next; | |
d1310b2e CM |
899 | goto search_again; |
900 | } | |
901 | ||
902 | /* | |
903 | * | ---- desired range ---- | | |
904 | * | state | | |
905 | * or | |
906 | * | ------------- state -------------- | | |
907 | * | |
908 | * We need to split the extent we found, and may flip bits on | |
909 | * second half. | |
910 | * | |
911 | * If the extent we found extends past our | |
912 | * range, we just split and search again. It'll get split | |
913 | * again the next time though. | |
914 | * | |
915 | * If the extent we found is inside our range, we set the | |
916 | * desired bit on it. | |
917 | */ | |
918 | if (state->start < start) { | |
1edbb734 | 919 | if (state->state & exclusive_bits) { |
d1310b2e CM |
920 | *failed_start = start; |
921 | err = -EEXIST; | |
922 | goto out; | |
923 | } | |
8233767a XG |
924 | |
925 | prealloc = alloc_extent_state_atomic(prealloc); | |
926 | BUG_ON(!prealloc); | |
d1310b2e | 927 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
928 | if (err) |
929 | extent_io_tree_panic(tree, err); | |
930 | ||
d1310b2e CM |
931 | prealloc = NULL; |
932 | if (err) | |
933 | goto out; | |
934 | if (state->end <= end) { | |
1bf85046 | 935 | set_state_bits(tree, state, &bits); |
2c64c53d | 936 | cache_state(state, cached_state); |
d1310b2e | 937 | merge_state(tree, state); |
5c939df5 YZ |
938 | if (last_end == (u64)-1) |
939 | goto out; | |
940 | start = last_end + 1; | |
d1ac6e41 LB |
941 | state = next_state(state); |
942 | if (start < end && state && state->start == start && | |
943 | !need_resched()) | |
944 | goto hit_next; | |
d1310b2e CM |
945 | } |
946 | goto search_again; | |
947 | } | |
948 | /* | |
949 | * | ---- desired range ---- | | |
950 | * | state | or | state | | |
951 | * | |
952 | * There's a hole, we need to insert something in it and | |
953 | * ignore the extent we found. | |
954 | */ | |
955 | if (state->start > start) { | |
956 | u64 this_end; | |
957 | if (end < last_start) | |
958 | this_end = end; | |
959 | else | |
d397712b | 960 | this_end = last_start - 1; |
8233767a XG |
961 | |
962 | prealloc = alloc_extent_state_atomic(prealloc); | |
963 | BUG_ON(!prealloc); | |
c7f895a2 XG |
964 | |
965 | /* | |
966 | * Avoid to free 'prealloc' if it can be merged with | |
967 | * the later extent. | |
968 | */ | |
d1310b2e | 969 | err = insert_state(tree, prealloc, start, this_end, |
12cfbad9 | 970 | NULL, NULL, &bits); |
c2d904e0 JM |
971 | if (err) |
972 | extent_io_tree_panic(tree, err); | |
973 | ||
9ed74f2d JB |
974 | cache_state(prealloc, cached_state); |
975 | prealloc = NULL; | |
d1310b2e CM |
976 | start = this_end + 1; |
977 | goto search_again; | |
978 | } | |
979 | /* | |
980 | * | ---- desired range ---- | | |
981 | * | state | | |
982 | * We need to split the extent, and set the bit | |
983 | * on the first half | |
984 | */ | |
985 | if (state->start <= end && state->end > end) { | |
1edbb734 | 986 | if (state->state & exclusive_bits) { |
d1310b2e CM |
987 | *failed_start = start; |
988 | err = -EEXIST; | |
989 | goto out; | |
990 | } | |
8233767a XG |
991 | |
992 | prealloc = alloc_extent_state_atomic(prealloc); | |
993 | BUG_ON(!prealloc); | |
d1310b2e | 994 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
995 | if (err) |
996 | extent_io_tree_panic(tree, err); | |
d1310b2e | 997 | |
1bf85046 | 998 | set_state_bits(tree, prealloc, &bits); |
2c64c53d | 999 | cache_state(prealloc, cached_state); |
d1310b2e CM |
1000 | merge_state(tree, prealloc); |
1001 | prealloc = NULL; | |
1002 | goto out; | |
1003 | } | |
1004 | ||
1005 | goto search_again; | |
1006 | ||
1007 | out: | |
cad321ad | 1008 | spin_unlock(&tree->lock); |
d1310b2e CM |
1009 | if (prealloc) |
1010 | free_extent_state(prealloc); | |
1011 | ||
1012 | return err; | |
1013 | ||
1014 | search_again: | |
1015 | if (start > end) | |
1016 | goto out; | |
cad321ad | 1017 | spin_unlock(&tree->lock); |
d1310b2e CM |
1018 | if (mask & __GFP_WAIT) |
1019 | cond_resched(); | |
1020 | goto again; | |
1021 | } | |
d1310b2e | 1022 | |
41074888 DS |
1023 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
1024 | unsigned long bits, u64 * failed_start, | |
1025 | struct extent_state **cached_state, gfp_t mask) | |
3fbe5c02 JM |
1026 | { |
1027 | return __set_extent_bit(tree, start, end, bits, 0, failed_start, | |
1028 | cached_state, mask); | |
1029 | } | |
1030 | ||
1031 | ||
462d6fac | 1032 | /** |
10983f2e LB |
1033 | * convert_extent_bit - convert all bits in a given range from one bit to |
1034 | * another | |
462d6fac JB |
1035 | * @tree: the io tree to search |
1036 | * @start: the start offset in bytes | |
1037 | * @end: the end offset in bytes (inclusive) | |
1038 | * @bits: the bits to set in this range | |
1039 | * @clear_bits: the bits to clear in this range | |
e6138876 | 1040 | * @cached_state: state that we're going to cache |
462d6fac JB |
1041 | * @mask: the allocation mask |
1042 | * | |
1043 | * This will go through and set bits for the given range. If any states exist | |
1044 | * already in this range they are set with the given bit and cleared of the | |
1045 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
1046 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
1047 | * boundary bits like LOCK. | |
1048 | */ | |
1049 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
41074888 | 1050 | unsigned long bits, unsigned long clear_bits, |
e6138876 | 1051 | struct extent_state **cached_state, gfp_t mask) |
462d6fac JB |
1052 | { |
1053 | struct extent_state *state; | |
1054 | struct extent_state *prealloc = NULL; | |
1055 | struct rb_node *node; | |
12cfbad9 FDBM |
1056 | struct rb_node **p; |
1057 | struct rb_node *parent; | |
462d6fac JB |
1058 | int err = 0; |
1059 | u64 last_start; | |
1060 | u64 last_end; | |
1061 | ||
a5dee37d | 1062 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 1063 | |
462d6fac JB |
1064 | again: |
1065 | if (!prealloc && (mask & __GFP_WAIT)) { | |
1066 | prealloc = alloc_extent_state(mask); | |
1067 | if (!prealloc) | |
1068 | return -ENOMEM; | |
1069 | } | |
1070 | ||
1071 | spin_lock(&tree->lock); | |
e6138876 JB |
1072 | if (cached_state && *cached_state) { |
1073 | state = *cached_state; | |
1074 | if (state->start <= start && state->end > start && | |
27a3507d | 1075 | extent_state_in_tree(state)) { |
e6138876 JB |
1076 | node = &state->rb_node; |
1077 | goto hit_next; | |
1078 | } | |
1079 | } | |
1080 | ||
462d6fac JB |
1081 | /* |
1082 | * this search will find all the extents that end after | |
1083 | * our range starts. | |
1084 | */ | |
12cfbad9 | 1085 | node = tree_search_for_insert(tree, start, &p, &parent); |
462d6fac JB |
1086 | if (!node) { |
1087 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1088 | if (!prealloc) { |
1089 | err = -ENOMEM; | |
1090 | goto out; | |
1091 | } | |
12cfbad9 FDBM |
1092 | err = insert_state(tree, prealloc, start, end, |
1093 | &p, &parent, &bits); | |
c2d904e0 JM |
1094 | if (err) |
1095 | extent_io_tree_panic(tree, err); | |
c42ac0bc FDBM |
1096 | cache_state(prealloc, cached_state); |
1097 | prealloc = NULL; | |
462d6fac JB |
1098 | goto out; |
1099 | } | |
1100 | state = rb_entry(node, struct extent_state, rb_node); | |
1101 | hit_next: | |
1102 | last_start = state->start; | |
1103 | last_end = state->end; | |
1104 | ||
1105 | /* | |
1106 | * | ---- desired range ---- | | |
1107 | * | state | | |
1108 | * | |
1109 | * Just lock what we found and keep going | |
1110 | */ | |
1111 | if (state->start == start && state->end <= end) { | |
462d6fac | 1112 | set_state_bits(tree, state, &bits); |
e6138876 | 1113 | cache_state(state, cached_state); |
d1ac6e41 | 1114 | state = clear_state_bit(tree, state, &clear_bits, 0); |
462d6fac JB |
1115 | if (last_end == (u64)-1) |
1116 | goto out; | |
462d6fac | 1117 | start = last_end + 1; |
d1ac6e41 LB |
1118 | if (start < end && state && state->start == start && |
1119 | !need_resched()) | |
1120 | goto hit_next; | |
462d6fac JB |
1121 | goto search_again; |
1122 | } | |
1123 | ||
1124 | /* | |
1125 | * | ---- desired range ---- | | |
1126 | * | state | | |
1127 | * or | |
1128 | * | ------------- state -------------- | | |
1129 | * | |
1130 | * We need to split the extent we found, and may flip bits on | |
1131 | * second half. | |
1132 | * | |
1133 | * If the extent we found extends past our | |
1134 | * range, we just split and search again. It'll get split | |
1135 | * again the next time though. | |
1136 | * | |
1137 | * If the extent we found is inside our range, we set the | |
1138 | * desired bit on it. | |
1139 | */ | |
1140 | if (state->start < start) { | |
1141 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1142 | if (!prealloc) { |
1143 | err = -ENOMEM; | |
1144 | goto out; | |
1145 | } | |
462d6fac | 1146 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1147 | if (err) |
1148 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1149 | prealloc = NULL; |
1150 | if (err) | |
1151 | goto out; | |
1152 | if (state->end <= end) { | |
1153 | set_state_bits(tree, state, &bits); | |
e6138876 | 1154 | cache_state(state, cached_state); |
d1ac6e41 | 1155 | state = clear_state_bit(tree, state, &clear_bits, 0); |
462d6fac JB |
1156 | if (last_end == (u64)-1) |
1157 | goto out; | |
1158 | start = last_end + 1; | |
d1ac6e41 LB |
1159 | if (start < end && state && state->start == start && |
1160 | !need_resched()) | |
1161 | goto hit_next; | |
462d6fac JB |
1162 | } |
1163 | goto search_again; | |
1164 | } | |
1165 | /* | |
1166 | * | ---- desired range ---- | | |
1167 | * | state | or | state | | |
1168 | * | |
1169 | * There's a hole, we need to insert something in it and | |
1170 | * ignore the extent we found. | |
1171 | */ | |
1172 | if (state->start > start) { | |
1173 | u64 this_end; | |
1174 | if (end < last_start) | |
1175 | this_end = end; | |
1176 | else | |
1177 | this_end = last_start - 1; | |
1178 | ||
1179 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1180 | if (!prealloc) { |
1181 | err = -ENOMEM; | |
1182 | goto out; | |
1183 | } | |
462d6fac JB |
1184 | |
1185 | /* | |
1186 | * Avoid to free 'prealloc' if it can be merged with | |
1187 | * the later extent. | |
1188 | */ | |
1189 | err = insert_state(tree, prealloc, start, this_end, | |
12cfbad9 | 1190 | NULL, NULL, &bits); |
c2d904e0 JM |
1191 | if (err) |
1192 | extent_io_tree_panic(tree, err); | |
e6138876 | 1193 | cache_state(prealloc, cached_state); |
462d6fac JB |
1194 | prealloc = NULL; |
1195 | start = this_end + 1; | |
1196 | goto search_again; | |
1197 | } | |
1198 | /* | |
1199 | * | ---- desired range ---- | | |
1200 | * | state | | |
1201 | * We need to split the extent, and set the bit | |
1202 | * on the first half | |
1203 | */ | |
1204 | if (state->start <= end && state->end > end) { | |
1205 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1206 | if (!prealloc) { |
1207 | err = -ENOMEM; | |
1208 | goto out; | |
1209 | } | |
462d6fac JB |
1210 | |
1211 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1212 | if (err) |
1213 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1214 | |
1215 | set_state_bits(tree, prealloc, &bits); | |
e6138876 | 1216 | cache_state(prealloc, cached_state); |
462d6fac | 1217 | clear_state_bit(tree, prealloc, &clear_bits, 0); |
462d6fac JB |
1218 | prealloc = NULL; |
1219 | goto out; | |
1220 | } | |
1221 | ||
1222 | goto search_again; | |
1223 | ||
1224 | out: | |
1225 | spin_unlock(&tree->lock); | |
1226 | if (prealloc) | |
1227 | free_extent_state(prealloc); | |
1228 | ||
1229 | return err; | |
1230 | ||
1231 | search_again: | |
1232 | if (start > end) | |
1233 | goto out; | |
1234 | spin_unlock(&tree->lock); | |
1235 | if (mask & __GFP_WAIT) | |
1236 | cond_resched(); | |
1237 | goto again; | |
1238 | } | |
1239 | ||
d1310b2e CM |
1240 | /* wrappers around set/clear extent bit */ |
1241 | int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
1242 | gfp_t mask) | |
1243 | { | |
3fbe5c02 | 1244 | return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL, |
2c64c53d | 1245 | NULL, mask); |
d1310b2e | 1246 | } |
d1310b2e CM |
1247 | |
1248 | int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
41074888 | 1249 | unsigned long bits, gfp_t mask) |
d1310b2e | 1250 | { |
3fbe5c02 | 1251 | return set_extent_bit(tree, start, end, bits, NULL, |
2c64c53d | 1252 | NULL, mask); |
d1310b2e | 1253 | } |
d1310b2e CM |
1254 | |
1255 | int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
41074888 | 1256 | unsigned long bits, gfp_t mask) |
d1310b2e | 1257 | { |
2c64c53d | 1258 | return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask); |
d1310b2e | 1259 | } |
d1310b2e CM |
1260 | |
1261 | int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, | |
2ac55d41 | 1262 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
1263 | { |
1264 | return set_extent_bit(tree, start, end, | |
fee187d9 | 1265 | EXTENT_DELALLOC | EXTENT_UPTODATE, |
3fbe5c02 | 1266 | NULL, cached_state, mask); |
d1310b2e | 1267 | } |
d1310b2e | 1268 | |
9e8a4a8b LB |
1269 | int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end, |
1270 | struct extent_state **cached_state, gfp_t mask) | |
1271 | { | |
1272 | return set_extent_bit(tree, start, end, | |
1273 | EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG, | |
1274 | NULL, cached_state, mask); | |
1275 | } | |
1276 | ||
d1310b2e CM |
1277 | int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, |
1278 | gfp_t mask) | |
1279 | { | |
1280 | return clear_extent_bit(tree, start, end, | |
32c00aff | 1281 | EXTENT_DIRTY | EXTENT_DELALLOC | |
0ca1f7ce | 1282 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask); |
d1310b2e | 1283 | } |
d1310b2e CM |
1284 | |
1285 | int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, | |
1286 | gfp_t mask) | |
1287 | { | |
3fbe5c02 | 1288 | return set_extent_bit(tree, start, end, EXTENT_NEW, NULL, |
2c64c53d | 1289 | NULL, mask); |
d1310b2e | 1290 | } |
d1310b2e | 1291 | |
d1310b2e | 1292 | int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
507903b8 | 1293 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e | 1294 | { |
6b67a320 | 1295 | return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL, |
3fbe5c02 | 1296 | cached_state, mask); |
d1310b2e | 1297 | } |
d1310b2e | 1298 | |
5fd02043 JB |
1299 | int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
1300 | struct extent_state **cached_state, gfp_t mask) | |
d1310b2e | 1301 | { |
2c64c53d | 1302 | return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, |
2ac55d41 | 1303 | cached_state, mask); |
d1310b2e | 1304 | } |
d1310b2e | 1305 | |
d352ac68 CM |
1306 | /* |
1307 | * either insert or lock state struct between start and end use mask to tell | |
1308 | * us if waiting is desired. | |
1309 | */ | |
1edbb734 | 1310 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
41074888 | 1311 | unsigned long bits, struct extent_state **cached_state) |
d1310b2e CM |
1312 | { |
1313 | int err; | |
1314 | u64 failed_start; | |
1315 | while (1) { | |
3fbe5c02 JM |
1316 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits, |
1317 | EXTENT_LOCKED, &failed_start, | |
1318 | cached_state, GFP_NOFS); | |
d0082371 | 1319 | if (err == -EEXIST) { |
d1310b2e CM |
1320 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1321 | start = failed_start; | |
d0082371 | 1322 | } else |
d1310b2e | 1323 | break; |
d1310b2e CM |
1324 | WARN_ON(start > end); |
1325 | } | |
1326 | return err; | |
1327 | } | |
d1310b2e | 1328 | |
d0082371 | 1329 | int lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
1edbb734 | 1330 | { |
d0082371 | 1331 | return lock_extent_bits(tree, start, end, 0, NULL); |
1edbb734 CM |
1332 | } |
1333 | ||
d0082371 | 1334 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1335 | { |
1336 | int err; | |
1337 | u64 failed_start; | |
1338 | ||
3fbe5c02 JM |
1339 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
1340 | &failed_start, NULL, GFP_NOFS); | |
6643558d YZ |
1341 | if (err == -EEXIST) { |
1342 | if (failed_start > start) | |
1343 | clear_extent_bit(tree, start, failed_start - 1, | |
d0082371 | 1344 | EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS); |
25179201 | 1345 | return 0; |
6643558d | 1346 | } |
25179201 JB |
1347 | return 1; |
1348 | } | |
25179201 | 1349 | |
2c64c53d CM |
1350 | int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end, |
1351 | struct extent_state **cached, gfp_t mask) | |
1352 | { | |
1353 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, | |
1354 | mask); | |
1355 | } | |
1356 | ||
d0082371 | 1357 | int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e | 1358 | { |
2c64c53d | 1359 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL, |
d0082371 | 1360 | GFP_NOFS); |
d1310b2e | 1361 | } |
d1310b2e | 1362 | |
4adaa611 CM |
1363 | int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
1364 | { | |
1365 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1366 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1367 | struct page *page; | |
1368 | ||
1369 | while (index <= end_index) { | |
1370 | page = find_get_page(inode->i_mapping, index); | |
1371 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1372 | clear_page_dirty_for_io(page); | |
1373 | page_cache_release(page); | |
1374 | index++; | |
1375 | } | |
1376 | return 0; | |
1377 | } | |
1378 | ||
1379 | int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) | |
1380 | { | |
1381 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1382 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1383 | struct page *page; | |
1384 | ||
1385 | while (index <= end_index) { | |
1386 | page = find_get_page(inode->i_mapping, index); | |
1387 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1388 | account_page_redirty(page); | |
1389 | __set_page_dirty_nobuffers(page); | |
1390 | page_cache_release(page); | |
1391 | index++; | |
1392 | } | |
1393 | return 0; | |
1394 | } | |
1395 | ||
d1310b2e CM |
1396 | /* |
1397 | * helper function to set both pages and extents in the tree writeback | |
1398 | */ | |
b2950863 | 1399 | static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e CM |
1400 | { |
1401 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1402 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1403 | struct page *page; | |
1404 | ||
1405 | while (index <= end_index) { | |
1406 | page = find_get_page(tree->mapping, index); | |
79787eaa | 1407 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ |
d1310b2e CM |
1408 | set_page_writeback(page); |
1409 | page_cache_release(page); | |
1410 | index++; | |
1411 | } | |
d1310b2e CM |
1412 | return 0; |
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, | |
41074888 | 1421 | u64 start, unsigned long 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, | |
41074888 | 1455 | u64 *start_ret, u64 *end_ret, unsigned long 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) { |
e6138876 | 1485 | cache_state(state, cached_state); |
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]; | |
1561 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1562 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
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]); | |
1576 | page_cache_release(pages[i]); | |
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 | { | |
1589 | unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT; | |
1590 | unsigned long start_index = index; | |
1591 | unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT; | |
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]); | |
1624 | page_cache_release(pages[i]); | |
1625 | goto done; | |
1626 | } | |
1627 | } | |
c8b97818 | 1628 | page_cache_release(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)) << | |
1641 | PAGE_CACHE_SHIFT); | |
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) { |
7bf811a5 | 1701 | max_bytes = PAGE_CACHE_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 */ | |
d0082371 | 1712 | lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &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 | ||
c2790a2e JB |
1732 | int extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end, |
1733 | struct page *locked_page, | |
1734 | unsigned long clear_bits, | |
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]; | |
1740 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1741 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
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) |
771ed689 | 1747 | return 0; |
c8b97818 | 1748 | |
d397712b | 1749 | while (nr_pages > 0) { |
c8b97818 | 1750 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1751 | min_t(unsigned long, |
1752 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 | 1753 | for (i = 0; i < ret; i++) { |
8b62b72b | 1754 | |
c2790a2e | 1755 | if (page_ops & PAGE_SET_PRIVATE2) |
8b62b72b CM |
1756 | SetPagePrivate2(pages[i]); |
1757 | ||
c8b97818 CM |
1758 | if (pages[i] == locked_page) { |
1759 | page_cache_release(pages[i]); | |
1760 | continue; | |
1761 | } | |
c2790a2e | 1762 | if (page_ops & PAGE_CLEAR_DIRTY) |
c8b97818 | 1763 | clear_page_dirty_for_io(pages[i]); |
c2790a2e | 1764 | if (page_ops & PAGE_SET_WRITEBACK) |
c8b97818 | 1765 | set_page_writeback(pages[i]); |
c2790a2e | 1766 | if (page_ops & PAGE_END_WRITEBACK) |
c8b97818 | 1767 | end_page_writeback(pages[i]); |
c2790a2e | 1768 | if (page_ops & PAGE_UNLOCK) |
771ed689 | 1769 | unlock_page(pages[i]); |
c8b97818 CM |
1770 | page_cache_release(pages[i]); |
1771 | } | |
1772 | nr_pages -= ret; | |
1773 | index += ret; | |
1774 | cond_resched(); | |
1775 | } | |
1776 | return 0; | |
1777 | } | |
c8b97818 | 1778 | |
d352ac68 CM |
1779 | /* |
1780 | * count the number of bytes in the tree that have a given bit(s) | |
1781 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1782 | * cached. The total number found is returned. | |
1783 | */ | |
d1310b2e CM |
1784 | u64 count_range_bits(struct extent_io_tree *tree, |
1785 | u64 *start, u64 search_end, u64 max_bytes, | |
ec29ed5b | 1786 | unsigned long bits, int contig) |
d1310b2e CM |
1787 | { |
1788 | struct rb_node *node; | |
1789 | struct extent_state *state; | |
1790 | u64 cur_start = *start; | |
1791 | u64 total_bytes = 0; | |
ec29ed5b | 1792 | u64 last = 0; |
d1310b2e CM |
1793 | int found = 0; |
1794 | ||
fae7f21c | 1795 | if (WARN_ON(search_end <= cur_start)) |
d1310b2e | 1796 | return 0; |
d1310b2e | 1797 | |
cad321ad | 1798 | spin_lock(&tree->lock); |
d1310b2e CM |
1799 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1800 | total_bytes = tree->dirty_bytes; | |
1801 | goto out; | |
1802 | } | |
1803 | /* | |
1804 | * this search will find all the extents that end after | |
1805 | * our range starts. | |
1806 | */ | |
80ea96b1 | 1807 | node = tree_search(tree, cur_start); |
d397712b | 1808 | if (!node) |
d1310b2e | 1809 | goto out; |
d1310b2e | 1810 | |
d397712b | 1811 | while (1) { |
d1310b2e CM |
1812 | state = rb_entry(node, struct extent_state, rb_node); |
1813 | if (state->start > search_end) | |
1814 | break; | |
ec29ed5b CM |
1815 | if (contig && found && state->start > last + 1) |
1816 | break; | |
1817 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1818 | total_bytes += min(search_end, state->end) + 1 - |
1819 | max(cur_start, state->start); | |
1820 | if (total_bytes >= max_bytes) | |
1821 | break; | |
1822 | if (!found) { | |
af60bed2 | 1823 | *start = max(cur_start, state->start); |
d1310b2e CM |
1824 | found = 1; |
1825 | } | |
ec29ed5b CM |
1826 | last = state->end; |
1827 | } else if (contig && found) { | |
1828 | break; | |
d1310b2e CM |
1829 | } |
1830 | node = rb_next(node); | |
1831 | if (!node) | |
1832 | break; | |
1833 | } | |
1834 | out: | |
cad321ad | 1835 | spin_unlock(&tree->lock); |
d1310b2e CM |
1836 | return total_bytes; |
1837 | } | |
b2950863 | 1838 | |
d352ac68 CM |
1839 | /* |
1840 | * set the private field for a given byte offset in the tree. If there isn't | |
1841 | * an extent_state there already, this does nothing. | |
1842 | */ | |
171170c1 | 1843 | static int set_state_private(struct extent_io_tree *tree, u64 start, u64 private) |
d1310b2e CM |
1844 | { |
1845 | struct rb_node *node; | |
1846 | struct extent_state *state; | |
1847 | int ret = 0; | |
1848 | ||
cad321ad | 1849 | spin_lock(&tree->lock); |
d1310b2e CM |
1850 | /* |
1851 | * this search will find all the extents that end after | |
1852 | * our range starts. | |
1853 | */ | |
80ea96b1 | 1854 | node = tree_search(tree, start); |
2b114d1d | 1855 | if (!node) { |
d1310b2e CM |
1856 | ret = -ENOENT; |
1857 | goto out; | |
1858 | } | |
1859 | state = rb_entry(node, struct extent_state, rb_node); | |
1860 | if (state->start != start) { | |
1861 | ret = -ENOENT; | |
1862 | goto out; | |
1863 | } | |
1864 | state->private = private; | |
1865 | out: | |
cad321ad | 1866 | spin_unlock(&tree->lock); |
d1310b2e CM |
1867 | return ret; |
1868 | } | |
1869 | ||
1870 | int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private) | |
1871 | { | |
1872 | struct rb_node *node; | |
1873 | struct extent_state *state; | |
1874 | int ret = 0; | |
1875 | ||
cad321ad | 1876 | spin_lock(&tree->lock); |
d1310b2e CM |
1877 | /* |
1878 | * this search will find all the extents that end after | |
1879 | * our range starts. | |
1880 | */ | |
80ea96b1 | 1881 | node = tree_search(tree, start); |
2b114d1d | 1882 | if (!node) { |
d1310b2e CM |
1883 | ret = -ENOENT; |
1884 | goto out; | |
1885 | } | |
1886 | state = rb_entry(node, struct extent_state, rb_node); | |
1887 | if (state->start != start) { | |
1888 | ret = -ENOENT; | |
1889 | goto out; | |
1890 | } | |
1891 | *private = state->private; | |
1892 | out: | |
cad321ad | 1893 | spin_unlock(&tree->lock); |
d1310b2e CM |
1894 | return ret; |
1895 | } | |
1896 | ||
1897 | /* | |
1898 | * searches a range in the state tree for a given mask. | |
70dec807 | 1899 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1900 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1901 | * range is found set. | |
1902 | */ | |
1903 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
41074888 | 1904 | unsigned long bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1905 | { |
1906 | struct extent_state *state = NULL; | |
1907 | struct rb_node *node; | |
1908 | int bitset = 0; | |
d1310b2e | 1909 | |
cad321ad | 1910 | spin_lock(&tree->lock); |
27a3507d | 1911 | if (cached && extent_state_in_tree(cached) && cached->start <= start && |
df98b6e2 | 1912 | cached->end > start) |
9655d298 CM |
1913 | node = &cached->rb_node; |
1914 | else | |
1915 | node = tree_search(tree, start); | |
d1310b2e CM |
1916 | while (node && start <= end) { |
1917 | state = rb_entry(node, struct extent_state, rb_node); | |
1918 | ||
1919 | if (filled && state->start > start) { | |
1920 | bitset = 0; | |
1921 | break; | |
1922 | } | |
1923 | ||
1924 | if (state->start > end) | |
1925 | break; | |
1926 | ||
1927 | if (state->state & bits) { | |
1928 | bitset = 1; | |
1929 | if (!filled) | |
1930 | break; | |
1931 | } else if (filled) { | |
1932 | bitset = 0; | |
1933 | break; | |
1934 | } | |
46562cec CM |
1935 | |
1936 | if (state->end == (u64)-1) | |
1937 | break; | |
1938 | ||
d1310b2e CM |
1939 | start = state->end + 1; |
1940 | if (start > end) | |
1941 | break; | |
1942 | node = rb_next(node); | |
1943 | if (!node) { | |
1944 | if (filled) | |
1945 | bitset = 0; | |
1946 | break; | |
1947 | } | |
1948 | } | |
cad321ad | 1949 | spin_unlock(&tree->lock); |
d1310b2e CM |
1950 | return bitset; |
1951 | } | |
d1310b2e CM |
1952 | |
1953 | /* | |
1954 | * helper function to set a given page up to date if all the | |
1955 | * extents in the tree for that page are up to date | |
1956 | */ | |
143bede5 | 1957 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e | 1958 | { |
4eee4fa4 | 1959 | u64 start = page_offset(page); |
d1310b2e | 1960 | u64 end = start + PAGE_CACHE_SIZE - 1; |
9655d298 | 1961 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 1962 | SetPageUptodate(page); |
d1310b2e CM |
1963 | } |
1964 | ||
4a54c8c1 JS |
1965 | /* |
1966 | * When IO fails, either with EIO or csum verification fails, we | |
1967 | * try other mirrors that might have a good copy of the data. This | |
1968 | * io_failure_record is used to record state as we go through all the | |
1969 | * mirrors. If another mirror has good data, the page is set up to date | |
1970 | * and things continue. If a good mirror can't be found, the original | |
1971 | * bio end_io callback is called to indicate things have failed. | |
1972 | */ | |
1973 | struct io_failure_record { | |
1974 | struct page *page; | |
1975 | u64 start; | |
1976 | u64 len; | |
1977 | u64 logical; | |
1978 | unsigned long bio_flags; | |
1979 | int this_mirror; | |
1980 | int failed_mirror; | |
1981 | int in_validation; | |
1982 | }; | |
1983 | ||
454ff3de | 1984 | static int free_io_failure(struct inode *inode, struct io_failure_record *rec) |
4a54c8c1 JS |
1985 | { |
1986 | int ret; | |
1987 | int err = 0; | |
1988 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
1989 | ||
1990 | set_state_private(failure_tree, rec->start, 0); | |
1991 | ret = clear_extent_bits(failure_tree, rec->start, | |
1992 | rec->start + rec->len - 1, | |
1993 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1994 | if (ret) | |
1995 | err = ret; | |
1996 | ||
53b381b3 DW |
1997 | ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start, |
1998 | rec->start + rec->len - 1, | |
1999 | EXTENT_DAMAGED, GFP_NOFS); | |
2000 | if (ret && !err) | |
2001 | err = ret; | |
4a54c8c1 JS |
2002 | |
2003 | kfree(rec); | |
2004 | return err; | |
2005 | } | |
2006 | ||
4a54c8c1 JS |
2007 | /* |
2008 | * this bypasses the standard btrfs submit functions deliberately, as | |
2009 | * the standard behavior is to write all copies in a raid setup. here we only | |
2010 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
2011 | * submit_bio directly. | |
3ec706c8 | 2012 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
2013 | * actually prevents the read that triggered the error from finishing. |
2014 | * currently, there can be no more than two copies of every data bit. thus, | |
2015 | * exactly one rewrite is required. | |
2016 | */ | |
3ec706c8 | 2017 | int repair_io_failure(struct btrfs_fs_info *fs_info, u64 start, |
4a54c8c1 JS |
2018 | u64 length, u64 logical, struct page *page, |
2019 | int mirror_num) | |
2020 | { | |
2021 | struct bio *bio; | |
2022 | struct btrfs_device *dev; | |
4a54c8c1 JS |
2023 | u64 map_length = 0; |
2024 | u64 sector; | |
2025 | struct btrfs_bio *bbio = NULL; | |
53b381b3 | 2026 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
4a54c8c1 JS |
2027 | int ret; |
2028 | ||
908960c6 | 2029 | ASSERT(!(fs_info->sb->s_flags & MS_RDONLY)); |
4a54c8c1 JS |
2030 | BUG_ON(!mirror_num); |
2031 | ||
53b381b3 DW |
2032 | /* we can't repair anything in raid56 yet */ |
2033 | if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num)) | |
2034 | return 0; | |
2035 | ||
9be3395b | 2036 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
4a54c8c1 JS |
2037 | if (!bio) |
2038 | return -EIO; | |
4f024f37 | 2039 | bio->bi_iter.bi_size = 0; |
4a54c8c1 JS |
2040 | map_length = length; |
2041 | ||
3ec706c8 | 2042 | ret = btrfs_map_block(fs_info, WRITE, logical, |
4a54c8c1 JS |
2043 | &map_length, &bbio, mirror_num); |
2044 | if (ret) { | |
2045 | bio_put(bio); | |
2046 | return -EIO; | |
2047 | } | |
2048 | BUG_ON(mirror_num != bbio->mirror_num); | |
2049 | sector = bbio->stripes[mirror_num-1].physical >> 9; | |
4f024f37 | 2050 | bio->bi_iter.bi_sector = sector; |
4a54c8c1 JS |
2051 | dev = bbio->stripes[mirror_num-1].dev; |
2052 | kfree(bbio); | |
2053 | if (!dev || !dev->bdev || !dev->writeable) { | |
2054 | bio_put(bio); | |
2055 | return -EIO; | |
2056 | } | |
2057 | bio->bi_bdev = dev->bdev; | |
4eee4fa4 | 2058 | bio_add_page(bio, page, length, start - page_offset(page)); |
4a54c8c1 | 2059 | |
33879d45 | 2060 | if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) { |
4a54c8c1 JS |
2061 | /* try to remap that extent elsewhere? */ |
2062 | bio_put(bio); | |
442a4f63 | 2063 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
2064 | return -EIO; |
2065 | } | |
2066 | ||
efe120a0 FH |
2067 | printk_ratelimited_in_rcu(KERN_INFO |
2068 | "BTRFS: read error corrected: ino %lu off %llu " | |
2069 | "(dev %s sector %llu)\n", page->mapping->host->i_ino, | |
2070 | start, rcu_str_deref(dev->name), sector); | |
4a54c8c1 JS |
2071 | |
2072 | bio_put(bio); | |
2073 | return 0; | |
2074 | } | |
2075 | ||
ea466794 JB |
2076 | int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb, |
2077 | int mirror_num) | |
2078 | { | |
ea466794 JB |
2079 | u64 start = eb->start; |
2080 | unsigned long i, num_pages = num_extent_pages(eb->start, eb->len); | |
d95603b2 | 2081 | int ret = 0; |
ea466794 | 2082 | |
908960c6 ID |
2083 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
2084 | return -EROFS; | |
2085 | ||
ea466794 JB |
2086 | for (i = 0; i < num_pages; i++) { |
2087 | struct page *p = extent_buffer_page(eb, i); | |
3ec706c8 | 2088 | ret = repair_io_failure(root->fs_info, start, PAGE_CACHE_SIZE, |
ea466794 JB |
2089 | start, p, mirror_num); |
2090 | if (ret) | |
2091 | break; | |
2092 | start += PAGE_CACHE_SIZE; | |
2093 | } | |
2094 | ||
2095 | return ret; | |
2096 | } | |
2097 | ||
4a54c8c1 JS |
2098 | /* |
2099 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2100 | * to see if we need to process or clean up an io_failure_record | |
2101 | */ | |
2102 | static int clean_io_failure(u64 start, struct page *page) | |
2103 | { | |
2104 | u64 private; | |
2105 | u64 private_failure; | |
2106 | struct io_failure_record *failrec; | |
908960c6 ID |
2107 | struct inode *inode = page->mapping->host; |
2108 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
4a54c8c1 JS |
2109 | struct extent_state *state; |
2110 | int num_copies; | |
4a54c8c1 | 2111 | int ret; |
4a54c8c1 JS |
2112 | |
2113 | private = 0; | |
2114 | ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
2115 | (u64)-1, 1, EXTENT_DIRTY, 0); | |
2116 | if (!ret) | |
2117 | return 0; | |
2118 | ||
2119 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start, | |
2120 | &private_failure); | |
2121 | if (ret) | |
2122 | return 0; | |
2123 | ||
2124 | failrec = (struct io_failure_record *)(unsigned long) private_failure; | |
2125 | BUG_ON(!failrec->this_mirror); | |
2126 | ||
2127 | if (failrec->in_validation) { | |
2128 | /* there was no real error, just free the record */ | |
2129 | pr_debug("clean_io_failure: freeing dummy error at %llu\n", | |
2130 | failrec->start); | |
4a54c8c1 JS |
2131 | goto out; |
2132 | } | |
908960c6 ID |
2133 | if (fs_info->sb->s_flags & MS_RDONLY) |
2134 | goto out; | |
4a54c8c1 JS |
2135 | |
2136 | spin_lock(&BTRFS_I(inode)->io_tree.lock); | |
2137 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, | |
2138 | failrec->start, | |
2139 | EXTENT_LOCKED); | |
2140 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); | |
2141 | ||
883d0de4 MX |
2142 | if (state && state->start <= failrec->start && |
2143 | state->end >= failrec->start + failrec->len - 1) { | |
3ec706c8 SB |
2144 | num_copies = btrfs_num_copies(fs_info, failrec->logical, |
2145 | failrec->len); | |
4a54c8c1 | 2146 | if (num_copies > 1) { |
454ff3de MX |
2147 | repair_io_failure(fs_info, start, failrec->len, |
2148 | failrec->logical, page, | |
2149 | failrec->failed_mirror); | |
4a54c8c1 JS |
2150 | } |
2151 | } | |
2152 | ||
2153 | out: | |
454ff3de | 2154 | free_io_failure(inode, failrec); |
4a54c8c1 | 2155 | |
454ff3de | 2156 | return 0; |
4a54c8c1 JS |
2157 | } |
2158 | ||
2159 | /* | |
2160 | * this is a generic handler for readpage errors (default | |
2161 | * readpage_io_failed_hook). if other copies exist, read those and write back | |
2162 | * good data to the failed position. does not investigate in remapping the | |
2163 | * failed extent elsewhere, hoping the device will be smart enough to do this as | |
2164 | * needed | |
2165 | */ | |
2166 | ||
facc8a22 MX |
2167 | static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset, |
2168 | struct page *page, u64 start, u64 end, | |
2169 | int failed_mirror) | |
4a54c8c1 JS |
2170 | { |
2171 | struct io_failure_record *failrec = NULL; | |
2172 | u64 private; | |
2173 | struct extent_map *em; | |
2174 | struct inode *inode = page->mapping->host; | |
2175 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
2176 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2177 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
2178 | struct bio *bio; | |
facc8a22 MX |
2179 | struct btrfs_io_bio *btrfs_failed_bio; |
2180 | struct btrfs_io_bio *btrfs_bio; | |
4a54c8c1 JS |
2181 | int num_copies; |
2182 | int ret; | |
2183 | int read_mode; | |
2184 | u64 logical; | |
2185 | ||
2186 | BUG_ON(failed_bio->bi_rw & REQ_WRITE); | |
2187 | ||
2188 | ret = get_state_private(failure_tree, start, &private); | |
2189 | if (ret) { | |
2190 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2191 | if (!failrec) | |
2192 | return -ENOMEM; | |
2193 | failrec->start = start; | |
2194 | failrec->len = end - start + 1; | |
2195 | failrec->this_mirror = 0; | |
2196 | failrec->bio_flags = 0; | |
2197 | failrec->in_validation = 0; | |
2198 | ||
2199 | read_lock(&em_tree->lock); | |
2200 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2201 | if (!em) { | |
2202 | read_unlock(&em_tree->lock); | |
2203 | kfree(failrec); | |
2204 | return -EIO; | |
2205 | } | |
2206 | ||
68ba990f | 2207 | if (em->start > start || em->start + em->len <= start) { |
4a54c8c1 JS |
2208 | free_extent_map(em); |
2209 | em = NULL; | |
2210 | } | |
2211 | read_unlock(&em_tree->lock); | |
2212 | ||
7a2d6a64 | 2213 | if (!em) { |
4a54c8c1 JS |
2214 | kfree(failrec); |
2215 | return -EIO; | |
2216 | } | |
2217 | logical = start - em->start; | |
2218 | logical = em->block_start + logical; | |
2219 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2220 | logical = em->block_start; | |
2221 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2222 | extent_set_compress_type(&failrec->bio_flags, | |
2223 | em->compress_type); | |
2224 | } | |
2225 | pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, " | |
2226 | "len=%llu\n", logical, start, failrec->len); | |
2227 | failrec->logical = logical; | |
2228 | free_extent_map(em); | |
2229 | ||
2230 | /* set the bits in the private failure tree */ | |
2231 | ret = set_extent_bits(failure_tree, start, end, | |
2232 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
2233 | if (ret >= 0) | |
2234 | ret = set_state_private(failure_tree, start, | |
2235 | (u64)(unsigned long)failrec); | |
2236 | /* set the bits in the inode's tree */ | |
2237 | if (ret >= 0) | |
2238 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED, | |
2239 | GFP_NOFS); | |
2240 | if (ret < 0) { | |
2241 | kfree(failrec); | |
2242 | return ret; | |
2243 | } | |
2244 | } else { | |
2245 | failrec = (struct io_failure_record *)(unsigned long)private; | |
2246 | pr_debug("bio_readpage_error: (found) logical=%llu, " | |
2247 | "start=%llu, len=%llu, validation=%d\n", | |
2248 | failrec->logical, failrec->start, failrec->len, | |
2249 | failrec->in_validation); | |
2250 | /* | |
2251 | * when data can be on disk more than twice, add to failrec here | |
2252 | * (e.g. with a list for failed_mirror) to make | |
2253 | * clean_io_failure() clean all those errors at once. | |
2254 | */ | |
2255 | } | |
5d964051 SB |
2256 | num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info, |
2257 | failrec->logical, failrec->len); | |
4a54c8c1 JS |
2258 | if (num_copies == 1) { |
2259 | /* | |
2260 | * we only have a single copy of the data, so don't bother with | |
2261 | * all the retry and error correction code that follows. no | |
2262 | * matter what the error is, it is very likely to persist. | |
2263 | */ | |
09a7f7a2 MX |
2264 | pr_debug("bio_readpage_error: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n", |
2265 | num_copies, failrec->this_mirror, failed_mirror); | |
454ff3de | 2266 | free_io_failure(inode, failrec); |
4a54c8c1 JS |
2267 | return -EIO; |
2268 | } | |
2269 | ||
4a54c8c1 JS |
2270 | /* |
2271 | * there are two premises: | |
2272 | * a) deliver good data to the caller | |
2273 | * b) correct the bad sectors on disk | |
2274 | */ | |
2275 | if (failed_bio->bi_vcnt > 1) { | |
2276 | /* | |
2277 | * to fulfill b), we need to know the exact failing sectors, as | |
2278 | * we don't want to rewrite any more than the failed ones. thus, | |
2279 | * we need separate read requests for the failed bio | |
2280 | * | |
2281 | * if the following BUG_ON triggers, our validation request got | |
2282 | * merged. we need separate requests for our algorithm to work. | |
2283 | */ | |
2284 | BUG_ON(failrec->in_validation); | |
2285 | failrec->in_validation = 1; | |
2286 | failrec->this_mirror = failed_mirror; | |
2287 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; | |
2288 | } else { | |
2289 | /* | |
2290 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2291 | * of the failed sector and if we succeed, we have setup | |
2292 | * everything for repair_io_failure to do the rest for us. | |
2293 | */ | |
2294 | if (failrec->in_validation) { | |
2295 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2296 | failrec->in_validation = 0; | |
2297 | failrec->this_mirror = 0; | |
2298 | } | |
2299 | failrec->failed_mirror = failed_mirror; | |
2300 | failrec->this_mirror++; | |
2301 | if (failrec->this_mirror == failed_mirror) | |
2302 | failrec->this_mirror++; | |
2303 | read_mode = READ_SYNC; | |
2304 | } | |
2305 | ||
facc8a22 MX |
2306 | if (failrec->this_mirror > num_copies) { |
2307 | pr_debug("bio_readpage_error: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n", | |
4a54c8c1 | 2308 | num_copies, failrec->this_mirror, failed_mirror); |
454ff3de | 2309 | free_io_failure(inode, failrec); |
4a54c8c1 JS |
2310 | return -EIO; |
2311 | } | |
2312 | ||
9be3395b | 2313 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
e627ee7b | 2314 | if (!bio) { |
454ff3de | 2315 | free_io_failure(inode, failrec); |
e627ee7b TI |
2316 | return -EIO; |
2317 | } | |
4a54c8c1 | 2318 | bio->bi_end_io = failed_bio->bi_end_io; |
4f024f37 | 2319 | bio->bi_iter.bi_sector = failrec->logical >> 9; |
4a54c8c1 | 2320 | bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; |
4f024f37 | 2321 | bio->bi_iter.bi_size = 0; |
4a54c8c1 | 2322 | |
facc8a22 MX |
2323 | btrfs_failed_bio = btrfs_io_bio(failed_bio); |
2324 | if (btrfs_failed_bio->csum) { | |
2325 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2326 | u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); | |
2327 | ||
2328 | btrfs_bio = btrfs_io_bio(bio); | |
2329 | btrfs_bio->csum = btrfs_bio->csum_inline; | |
2330 | phy_offset >>= inode->i_sb->s_blocksize_bits; | |
2331 | phy_offset *= csum_size; | |
2332 | memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + phy_offset, | |
2333 | csum_size); | |
2334 | } | |
2335 | ||
4a54c8c1 JS |
2336 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); |
2337 | ||
2338 | pr_debug("bio_readpage_error: submitting new read[%#x] to " | |
2339 | "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode, | |
2340 | failrec->this_mirror, num_copies, failrec->in_validation); | |
2341 | ||
013bd4c3 TI |
2342 | ret = tree->ops->submit_bio_hook(inode, read_mode, bio, |
2343 | failrec->this_mirror, | |
2344 | failrec->bio_flags, 0); | |
6c387ab2 | 2345 | if (ret) { |
454ff3de | 2346 | free_io_failure(inode, failrec); |
6c387ab2 MX |
2347 | bio_put(bio); |
2348 | } | |
2349 | ||
013bd4c3 | 2350 | return ret; |
4a54c8c1 JS |
2351 | } |
2352 | ||
d1310b2e CM |
2353 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2354 | ||
87826df0 JM |
2355 | int end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
2356 | { | |
2357 | int uptodate = (err == 0); | |
2358 | struct extent_io_tree *tree; | |
3e2426bd | 2359 | int ret = 0; |
87826df0 JM |
2360 | |
2361 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
2362 | ||
2363 | if (tree->ops && tree->ops->writepage_end_io_hook) { | |
2364 | ret = tree->ops->writepage_end_io_hook(page, start, | |
2365 | end, NULL, uptodate); | |
2366 | if (ret) | |
2367 | uptodate = 0; | |
2368 | } | |
2369 | ||
87826df0 | 2370 | if (!uptodate) { |
87826df0 JM |
2371 | ClearPageUptodate(page); |
2372 | SetPageError(page); | |
5dca6eea LB |
2373 | ret = ret < 0 ? ret : -EIO; |
2374 | mapping_set_error(page->mapping, ret); | |
87826df0 JM |
2375 | } |
2376 | return 0; | |
2377 | } | |
2378 | ||
d1310b2e CM |
2379 | /* |
2380 | * after a writepage IO is done, we need to: | |
2381 | * clear the uptodate bits on error | |
2382 | * clear the writeback bits in the extent tree for this IO | |
2383 | * end_page_writeback if the page has no more pending IO | |
2384 | * | |
2385 | * Scheduling is not allowed, so the extent state tree is expected | |
2386 | * to have one and only one object corresponding to this IO. | |
2387 | */ | |
d1310b2e | 2388 | static void end_bio_extent_writepage(struct bio *bio, int err) |
d1310b2e | 2389 | { |
2c30c71b | 2390 | struct bio_vec *bvec; |
d1310b2e CM |
2391 | u64 start; |
2392 | u64 end; | |
2c30c71b | 2393 | int i; |
d1310b2e | 2394 | |
2c30c71b | 2395 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2396 | struct page *page = bvec->bv_page; |
902b22f3 | 2397 | |
17a5adcc AO |
2398 | /* We always issue full-page reads, but if some block |
2399 | * in a page fails to read, blk_update_request() will | |
2400 | * advance bv_offset and adjust bv_len to compensate. | |
2401 | * Print a warning for nonzero offsets, and an error | |
2402 | * if they don't add up to a full page. */ | |
efe120a0 FH |
2403 | if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) { |
2404 | if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE) | |
2405 | btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info, | |
2406 | "partial page write in btrfs with offset %u and length %u", | |
2407 | bvec->bv_offset, bvec->bv_len); | |
2408 | else | |
2409 | btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info, | |
2410 | "incomplete page write in btrfs with offset %u and " | |
2411 | "length %u", | |
2412 | bvec->bv_offset, bvec->bv_len); | |
2413 | } | |
d1310b2e | 2414 | |
17a5adcc AO |
2415 | start = page_offset(page); |
2416 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
d1310b2e | 2417 | |
87826df0 JM |
2418 | if (end_extent_writepage(page, err, start, end)) |
2419 | continue; | |
70dec807 | 2420 | |
17a5adcc | 2421 | end_page_writeback(page); |
2c30c71b | 2422 | } |
2b1f55b0 | 2423 | |
d1310b2e | 2424 | bio_put(bio); |
d1310b2e CM |
2425 | } |
2426 | ||
883d0de4 MX |
2427 | static void |
2428 | endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len, | |
2429 | int uptodate) | |
2430 | { | |
2431 | struct extent_state *cached = NULL; | |
2432 | u64 end = start + len - 1; | |
2433 | ||
2434 | if (uptodate && tree->track_uptodate) | |
2435 | set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC); | |
2436 | unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC); | |
2437 | } | |
2438 | ||
d1310b2e CM |
2439 | /* |
2440 | * after a readpage IO is done, we need to: | |
2441 | * clear the uptodate bits on error | |
2442 | * set the uptodate bits if things worked | |
2443 | * set the page up to date if all extents in the tree are uptodate | |
2444 | * clear the lock bit in the extent tree | |
2445 | * unlock the page if there are no other extents locked for it | |
2446 | * | |
2447 | * Scheduling is not allowed, so the extent state tree is expected | |
2448 | * to have one and only one object corresponding to this IO. | |
2449 | */ | |
d1310b2e | 2450 | static void end_bio_extent_readpage(struct bio *bio, int err) |
d1310b2e | 2451 | { |
2c30c71b | 2452 | struct bio_vec *bvec; |
d1310b2e | 2453 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); |
facc8a22 | 2454 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); |
902b22f3 | 2455 | struct extent_io_tree *tree; |
facc8a22 | 2456 | u64 offset = 0; |
d1310b2e CM |
2457 | u64 start; |
2458 | u64 end; | |
facc8a22 | 2459 | u64 len; |
883d0de4 MX |
2460 | u64 extent_start = 0; |
2461 | u64 extent_len = 0; | |
5cf1ab56 | 2462 | int mirror; |
d1310b2e | 2463 | int ret; |
2c30c71b | 2464 | int i; |
d1310b2e | 2465 | |
d20f7043 CM |
2466 | if (err) |
2467 | uptodate = 0; | |
2468 | ||
2c30c71b | 2469 | bio_for_each_segment_all(bvec, bio, i) { |
d1310b2e | 2470 | struct page *page = bvec->bv_page; |
a71754fc | 2471 | struct inode *inode = page->mapping->host; |
507903b8 | 2472 | |
be3940c0 | 2473 | pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, " |
c1dc0896 | 2474 | "mirror=%u\n", (u64)bio->bi_iter.bi_sector, err, |
9be3395b | 2475 | io_bio->mirror_num); |
a71754fc | 2476 | tree = &BTRFS_I(inode)->io_tree; |
902b22f3 | 2477 | |
17a5adcc AO |
2478 | /* We always issue full-page reads, but if some block |
2479 | * in a page fails to read, blk_update_request() will | |
2480 | * advance bv_offset and adjust bv_len to compensate. | |
2481 | * Print a warning for nonzero offsets, and an error | |
2482 | * if they don't add up to a full page. */ | |
efe120a0 FH |
2483 | if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) { |
2484 | if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE) | |
2485 | btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info, | |
2486 | "partial page read in btrfs with offset %u and length %u", | |
2487 | bvec->bv_offset, bvec->bv_len); | |
2488 | else | |
2489 | btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info, | |
2490 | "incomplete page read in btrfs with offset %u and " | |
2491 | "length %u", | |
2492 | bvec->bv_offset, bvec->bv_len); | |
2493 | } | |
d1310b2e | 2494 | |
17a5adcc AO |
2495 | start = page_offset(page); |
2496 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
facc8a22 | 2497 | len = bvec->bv_len; |
d1310b2e | 2498 | |
9be3395b | 2499 | mirror = io_bio->mirror_num; |
f2a09da9 MX |
2500 | if (likely(uptodate && tree->ops && |
2501 | tree->ops->readpage_end_io_hook)) { | |
facc8a22 MX |
2502 | ret = tree->ops->readpage_end_io_hook(io_bio, offset, |
2503 | page, start, end, | |
2504 | mirror); | |
5ee0844d | 2505 | if (ret) |
d1310b2e | 2506 | uptodate = 0; |
5ee0844d | 2507 | else |
4a54c8c1 | 2508 | clean_io_failure(start, page); |
d1310b2e | 2509 | } |
ea466794 | 2510 | |
f2a09da9 MX |
2511 | if (likely(uptodate)) |
2512 | goto readpage_ok; | |
2513 | ||
2514 | if (tree->ops && tree->ops->readpage_io_failed_hook) { | |
5cf1ab56 | 2515 | ret = tree->ops->readpage_io_failed_hook(page, mirror); |
ea466794 JB |
2516 | if (!ret && !err && |
2517 | test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
2518 | uptodate = 1; | |
f2a09da9 | 2519 | } else { |
f4a8e656 JS |
2520 | /* |
2521 | * The generic bio_readpage_error handles errors the | |
2522 | * following way: If possible, new read requests are | |
2523 | * created and submitted and will end up in | |
2524 | * end_bio_extent_readpage as well (if we're lucky, not | |
2525 | * in the !uptodate case). In that case it returns 0 and | |
2526 | * we just go on with the next page in our bio. If it | |
2527 | * can't handle the error it will return -EIO and we | |
2528 | * remain responsible for that page. | |
2529 | */ | |
facc8a22 MX |
2530 | ret = bio_readpage_error(bio, offset, page, start, end, |
2531 | mirror); | |
7e38326f | 2532 | if (ret == 0) { |
3b951516 CM |
2533 | uptodate = |
2534 | test_bit(BIO_UPTODATE, &bio->bi_flags); | |
d20f7043 CM |
2535 | if (err) |
2536 | uptodate = 0; | |
38c1c2e4 | 2537 | offset += len; |
7e38326f CM |
2538 | continue; |
2539 | } | |
2540 | } | |
f2a09da9 | 2541 | readpage_ok: |
883d0de4 | 2542 | if (likely(uptodate)) { |
a71754fc JB |
2543 | loff_t i_size = i_size_read(inode); |
2544 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
a583c026 | 2545 | unsigned off; |
a71754fc JB |
2546 | |
2547 | /* Zero out the end if this page straddles i_size */ | |
a583c026 LB |
2548 | off = i_size & (PAGE_CACHE_SIZE-1); |
2549 | if (page->index == end_index && off) | |
2550 | zero_user_segment(page, off, PAGE_CACHE_SIZE); | |
17a5adcc | 2551 | SetPageUptodate(page); |
70dec807 | 2552 | } else { |
17a5adcc AO |
2553 | ClearPageUptodate(page); |
2554 | SetPageError(page); | |
70dec807 | 2555 | } |
17a5adcc | 2556 | unlock_page(page); |
facc8a22 | 2557 | offset += len; |
883d0de4 MX |
2558 | |
2559 | if (unlikely(!uptodate)) { | |
2560 | if (extent_len) { | |
2561 | endio_readpage_release_extent(tree, | |
2562 | extent_start, | |
2563 | extent_len, 1); | |
2564 | extent_start = 0; | |
2565 | extent_len = 0; | |
2566 | } | |
2567 | endio_readpage_release_extent(tree, start, | |
2568 | end - start + 1, 0); | |
2569 | } else if (!extent_len) { | |
2570 | extent_start = start; | |
2571 | extent_len = end + 1 - start; | |
2572 | } else if (extent_start + extent_len == start) { | |
2573 | extent_len += end + 1 - start; | |
2574 | } else { | |
2575 | endio_readpage_release_extent(tree, extent_start, | |
2576 | extent_len, uptodate); | |
2577 | extent_start = start; | |
2578 | extent_len = end + 1 - start; | |
2579 | } | |
2c30c71b | 2580 | } |
d1310b2e | 2581 | |
883d0de4 MX |
2582 | if (extent_len) |
2583 | endio_readpage_release_extent(tree, extent_start, extent_len, | |
2584 | uptodate); | |
facc8a22 MX |
2585 | if (io_bio->end_io) |
2586 | io_bio->end_io(io_bio, err); | |
d1310b2e | 2587 | bio_put(bio); |
d1310b2e CM |
2588 | } |
2589 | ||
9be3395b CM |
2590 | /* |
2591 | * this allocates from the btrfs_bioset. We're returning a bio right now | |
2592 | * but you can call btrfs_io_bio for the appropriate container_of magic | |
2593 | */ | |
88f794ed MX |
2594 | struct bio * |
2595 | btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
2596 | gfp_t gfp_flags) | |
d1310b2e | 2597 | { |
facc8a22 | 2598 | struct btrfs_io_bio *btrfs_bio; |
d1310b2e CM |
2599 | struct bio *bio; |
2600 | ||
9be3395b | 2601 | bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset); |
d1310b2e CM |
2602 | |
2603 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
9be3395b CM |
2604 | while (!bio && (nr_vecs /= 2)) { |
2605 | bio = bio_alloc_bioset(gfp_flags, | |
2606 | nr_vecs, btrfs_bioset); | |
2607 | } | |
d1310b2e CM |
2608 | } |
2609 | ||
2610 | if (bio) { | |
2611 | bio->bi_bdev = bdev; | |
4f024f37 | 2612 | bio->bi_iter.bi_sector = first_sector; |
facc8a22 MX |
2613 | btrfs_bio = btrfs_io_bio(bio); |
2614 | btrfs_bio->csum = NULL; | |
2615 | btrfs_bio->csum_allocated = NULL; | |
2616 | btrfs_bio->end_io = NULL; | |
d1310b2e CM |
2617 | } |
2618 | return bio; | |
2619 | } | |
2620 | ||
9be3395b CM |
2621 | struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask) |
2622 | { | |
23ea8e5a MX |
2623 | struct btrfs_io_bio *btrfs_bio; |
2624 | struct bio *new; | |
9be3395b | 2625 | |
23ea8e5a MX |
2626 | new = bio_clone_bioset(bio, gfp_mask, btrfs_bioset); |
2627 | if (new) { | |
2628 | btrfs_bio = btrfs_io_bio(new); | |
2629 | btrfs_bio->csum = NULL; | |
2630 | btrfs_bio->csum_allocated = NULL; | |
2631 | btrfs_bio->end_io = NULL; | |
2632 | } | |
2633 | return new; | |
2634 | } | |
9be3395b CM |
2635 | |
2636 | /* this also allocates from the btrfs_bioset */ | |
2637 | struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs) | |
2638 | { | |
facc8a22 MX |
2639 | struct btrfs_io_bio *btrfs_bio; |
2640 | struct bio *bio; | |
2641 | ||
2642 | bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset); | |
2643 | if (bio) { | |
2644 | btrfs_bio = btrfs_io_bio(bio); | |
2645 | btrfs_bio->csum = NULL; | |
2646 | btrfs_bio->csum_allocated = NULL; | |
2647 | btrfs_bio->end_io = NULL; | |
2648 | } | |
2649 | return bio; | |
9be3395b CM |
2650 | } |
2651 | ||
2652 | ||
355808c2 JM |
2653 | static int __must_check submit_one_bio(int rw, struct bio *bio, |
2654 | int mirror_num, unsigned long bio_flags) | |
d1310b2e | 2655 | { |
d1310b2e | 2656 | int ret = 0; |
70dec807 CM |
2657 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
2658 | struct page *page = bvec->bv_page; | |
2659 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 2660 | u64 start; |
70dec807 | 2661 | |
4eee4fa4 | 2662 | start = page_offset(page) + bvec->bv_offset; |
70dec807 | 2663 | |
902b22f3 | 2664 | bio->bi_private = NULL; |
d1310b2e CM |
2665 | |
2666 | bio_get(bio); | |
2667 | ||
065631f6 | 2668 | if (tree->ops && tree->ops->submit_bio_hook) |
6b82ce8d | 2669 | ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio, |
eaf25d93 | 2670 | mirror_num, bio_flags, start); |
0b86a832 | 2671 | else |
21adbd5c | 2672 | btrfsic_submit_bio(rw, bio); |
4a54c8c1 | 2673 | |
d1310b2e CM |
2674 | if (bio_flagged(bio, BIO_EOPNOTSUPP)) |
2675 | ret = -EOPNOTSUPP; | |
2676 | bio_put(bio); | |
2677 | return ret; | |
2678 | } | |
2679 | ||
64a16701 | 2680 | static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page, |
3444a972 JM |
2681 | unsigned long offset, size_t size, struct bio *bio, |
2682 | unsigned long bio_flags) | |
2683 | { | |
2684 | int ret = 0; | |
2685 | if (tree->ops && tree->ops->merge_bio_hook) | |
64a16701 | 2686 | ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio, |
3444a972 JM |
2687 | bio_flags); |
2688 | BUG_ON(ret < 0); | |
2689 | return ret; | |
2690 | ||
2691 | } | |
2692 | ||
d1310b2e CM |
2693 | static int submit_extent_page(int rw, struct extent_io_tree *tree, |
2694 | struct page *page, sector_t sector, | |
2695 | size_t size, unsigned long offset, | |
2696 | struct block_device *bdev, | |
2697 | struct bio **bio_ret, | |
2698 | unsigned long max_pages, | |
f188591e | 2699 | bio_end_io_t end_io_func, |
c8b97818 CM |
2700 | int mirror_num, |
2701 | unsigned long prev_bio_flags, | |
2702 | unsigned long bio_flags) | |
d1310b2e CM |
2703 | { |
2704 | int ret = 0; | |
2705 | struct bio *bio; | |
2706 | int nr; | |
c8b97818 CM |
2707 | int contig = 0; |
2708 | int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED; | |
2709 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; | |
5b050f04 | 2710 | size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE); |
d1310b2e CM |
2711 | |
2712 | if (bio_ret && *bio_ret) { | |
2713 | bio = *bio_ret; | |
c8b97818 | 2714 | if (old_compressed) |
4f024f37 | 2715 | contig = bio->bi_iter.bi_sector == sector; |
c8b97818 | 2716 | else |
f73a1c7d | 2717 | contig = bio_end_sector(bio) == sector; |
c8b97818 CM |
2718 | |
2719 | if (prev_bio_flags != bio_flags || !contig || | |
64a16701 | 2720 | merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) || |
c8b97818 CM |
2721 | bio_add_page(bio, page, page_size, offset) < page_size) { |
2722 | ret = submit_one_bio(rw, bio, mirror_num, | |
2723 | prev_bio_flags); | |
79787eaa JM |
2724 | if (ret < 0) |
2725 | return ret; | |
d1310b2e CM |
2726 | bio = NULL; |
2727 | } else { | |
2728 | return 0; | |
2729 | } | |
2730 | } | |
c8b97818 CM |
2731 | if (this_compressed) |
2732 | nr = BIO_MAX_PAGES; | |
2733 | else | |
2734 | nr = bio_get_nr_vecs(bdev); | |
2735 | ||
88f794ed | 2736 | bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH); |
5df67083 TI |
2737 | if (!bio) |
2738 | return -ENOMEM; | |
70dec807 | 2739 | |
c8b97818 | 2740 | bio_add_page(bio, page, page_size, offset); |
d1310b2e CM |
2741 | bio->bi_end_io = end_io_func; |
2742 | bio->bi_private = tree; | |
70dec807 | 2743 | |
d397712b | 2744 | if (bio_ret) |
d1310b2e | 2745 | *bio_ret = bio; |
d397712b | 2746 | else |
c8b97818 | 2747 | ret = submit_one_bio(rw, bio, mirror_num, bio_flags); |
d1310b2e CM |
2748 | |
2749 | return ret; | |
2750 | } | |
2751 | ||
48a3b636 ES |
2752 | static void attach_extent_buffer_page(struct extent_buffer *eb, |
2753 | struct page *page) | |
d1310b2e CM |
2754 | { |
2755 | if (!PagePrivate(page)) { | |
2756 | SetPagePrivate(page); | |
d1310b2e | 2757 | page_cache_get(page); |
4f2de97a JB |
2758 | set_page_private(page, (unsigned long)eb); |
2759 | } else { | |
2760 | WARN_ON(page->private != (unsigned long)eb); | |
d1310b2e CM |
2761 | } |
2762 | } | |
2763 | ||
4f2de97a | 2764 | void set_page_extent_mapped(struct page *page) |
d1310b2e | 2765 | { |
4f2de97a JB |
2766 | if (!PagePrivate(page)) { |
2767 | SetPagePrivate(page); | |
2768 | page_cache_get(page); | |
2769 | set_page_private(page, EXTENT_PAGE_PRIVATE); | |
2770 | } | |
d1310b2e CM |
2771 | } |
2772 | ||
125bac01 MX |
2773 | static struct extent_map * |
2774 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
2775 | u64 start, u64 len, get_extent_t *get_extent, | |
2776 | struct extent_map **em_cached) | |
2777 | { | |
2778 | struct extent_map *em; | |
2779 | ||
2780 | if (em_cached && *em_cached) { | |
2781 | em = *em_cached; | |
cbc0e928 | 2782 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 MX |
2783 | start < extent_map_end(em)) { |
2784 | atomic_inc(&em->refs); | |
2785 | return em; | |
2786 | } | |
2787 | ||
2788 | free_extent_map(em); | |
2789 | *em_cached = NULL; | |
2790 | } | |
2791 | ||
2792 | em = get_extent(inode, page, pg_offset, start, len, 0); | |
2793 | if (em_cached && !IS_ERR_OR_NULL(em)) { | |
2794 | BUG_ON(*em_cached); | |
2795 | atomic_inc(&em->refs); | |
2796 | *em_cached = em; | |
2797 | } | |
2798 | return em; | |
2799 | } | |
d1310b2e CM |
2800 | /* |
2801 | * basic readpage implementation. Locked extent state structs are inserted | |
2802 | * into the tree that are removed when the IO is done (by the end_io | |
2803 | * handlers) | |
79787eaa | 2804 | * XXX JDM: This needs looking at to ensure proper page locking |
d1310b2e | 2805 | */ |
9974090b MX |
2806 | static int __do_readpage(struct extent_io_tree *tree, |
2807 | struct page *page, | |
2808 | get_extent_t *get_extent, | |
125bac01 | 2809 | struct extent_map **em_cached, |
9974090b MX |
2810 | struct bio **bio, int mirror_num, |
2811 | unsigned long *bio_flags, int rw) | |
d1310b2e CM |
2812 | { |
2813 | struct inode *inode = page->mapping->host; | |
4eee4fa4 | 2814 | u64 start = page_offset(page); |
d1310b2e CM |
2815 | u64 page_end = start + PAGE_CACHE_SIZE - 1; |
2816 | u64 end; | |
2817 | u64 cur = start; | |
2818 | u64 extent_offset; | |
2819 | u64 last_byte = i_size_read(inode); | |
2820 | u64 block_start; | |
2821 | u64 cur_end; | |
2822 | sector_t sector; | |
2823 | struct extent_map *em; | |
2824 | struct block_device *bdev; | |
2825 | int ret; | |
2826 | int nr = 0; | |
4b384318 | 2827 | int parent_locked = *bio_flags & EXTENT_BIO_PARENT_LOCKED; |
306e16ce | 2828 | size_t pg_offset = 0; |
d1310b2e | 2829 | size_t iosize; |
c8b97818 | 2830 | size_t disk_io_size; |
d1310b2e | 2831 | size_t blocksize = inode->i_sb->s_blocksize; |
4b384318 | 2832 | unsigned long this_bio_flag = *bio_flags & EXTENT_BIO_PARENT_LOCKED; |
d1310b2e CM |
2833 | |
2834 | set_page_extent_mapped(page); | |
2835 | ||
9974090b | 2836 | end = page_end; |
90a887c9 DM |
2837 | if (!PageUptodate(page)) { |
2838 | if (cleancache_get_page(page) == 0) { | |
2839 | BUG_ON(blocksize != PAGE_SIZE); | |
9974090b | 2840 | unlock_extent(tree, start, end); |
90a887c9 DM |
2841 | goto out; |
2842 | } | |
2843 | } | |
2844 | ||
c8b97818 CM |
2845 | if (page->index == last_byte >> PAGE_CACHE_SHIFT) { |
2846 | char *userpage; | |
2847 | size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1); | |
2848 | ||
2849 | if (zero_offset) { | |
2850 | iosize = PAGE_CACHE_SIZE - zero_offset; | |
7ac687d9 | 2851 | userpage = kmap_atomic(page); |
c8b97818 CM |
2852 | memset(userpage + zero_offset, 0, iosize); |
2853 | flush_dcache_page(page); | |
7ac687d9 | 2854 | kunmap_atomic(userpage); |
c8b97818 CM |
2855 | } |
2856 | } | |
d1310b2e | 2857 | while (cur <= end) { |
c8f2f24b JB |
2858 | unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1; |
2859 | ||
d1310b2e CM |
2860 | if (cur >= last_byte) { |
2861 | char *userpage; | |
507903b8 AJ |
2862 | struct extent_state *cached = NULL; |
2863 | ||
306e16ce | 2864 | iosize = PAGE_CACHE_SIZE - pg_offset; |
7ac687d9 | 2865 | userpage = kmap_atomic(page); |
306e16ce | 2866 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2867 | flush_dcache_page(page); |
7ac687d9 | 2868 | kunmap_atomic(userpage); |
d1310b2e | 2869 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 2870 | &cached, GFP_NOFS); |
4b384318 MF |
2871 | if (!parent_locked) |
2872 | unlock_extent_cached(tree, cur, | |
2873 | cur + iosize - 1, | |
2874 | &cached, GFP_NOFS); | |
d1310b2e CM |
2875 | break; |
2876 | } | |
125bac01 MX |
2877 | em = __get_extent_map(inode, page, pg_offset, cur, |
2878 | end - cur + 1, get_extent, em_cached); | |
c704005d | 2879 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 2880 | SetPageError(page); |
4b384318 MF |
2881 | if (!parent_locked) |
2882 | unlock_extent(tree, cur, end); | |
d1310b2e CM |
2883 | break; |
2884 | } | |
d1310b2e CM |
2885 | extent_offset = cur - em->start; |
2886 | BUG_ON(extent_map_end(em) <= cur); | |
2887 | BUG_ON(end < cur); | |
2888 | ||
261507a0 | 2889 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
4b384318 | 2890 | this_bio_flag |= EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
2891 | extent_set_compress_type(&this_bio_flag, |
2892 | em->compress_type); | |
2893 | } | |
c8b97818 | 2894 | |
d1310b2e CM |
2895 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
2896 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 2897 | iosize = ALIGN(iosize, blocksize); |
c8b97818 CM |
2898 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
2899 | disk_io_size = em->block_len; | |
2900 | sector = em->block_start >> 9; | |
2901 | } else { | |
2902 | sector = (em->block_start + extent_offset) >> 9; | |
2903 | disk_io_size = iosize; | |
2904 | } | |
d1310b2e CM |
2905 | bdev = em->bdev; |
2906 | block_start = em->block_start; | |
d899e052 YZ |
2907 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2908 | block_start = EXTENT_MAP_HOLE; | |
d1310b2e CM |
2909 | free_extent_map(em); |
2910 | em = NULL; | |
2911 | ||
2912 | /* we've found a hole, just zero and go on */ | |
2913 | if (block_start == EXTENT_MAP_HOLE) { | |
2914 | char *userpage; | |
507903b8 AJ |
2915 | struct extent_state *cached = NULL; |
2916 | ||
7ac687d9 | 2917 | userpage = kmap_atomic(page); |
306e16ce | 2918 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2919 | flush_dcache_page(page); |
7ac687d9 | 2920 | kunmap_atomic(userpage); |
d1310b2e CM |
2921 | |
2922 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 AJ |
2923 | &cached, GFP_NOFS); |
2924 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
2925 | &cached, GFP_NOFS); | |
d1310b2e | 2926 | cur = cur + iosize; |
306e16ce | 2927 | pg_offset += iosize; |
d1310b2e CM |
2928 | continue; |
2929 | } | |
2930 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
2931 | if (test_range_bit(tree, cur, cur_end, |
2932 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 2933 | check_page_uptodate(tree, page); |
4b384318 MF |
2934 | if (!parent_locked) |
2935 | unlock_extent(tree, cur, cur + iosize - 1); | |
d1310b2e | 2936 | cur = cur + iosize; |
306e16ce | 2937 | pg_offset += iosize; |
d1310b2e CM |
2938 | continue; |
2939 | } | |
70dec807 CM |
2940 | /* we have an inline extent but it didn't get marked up |
2941 | * to date. Error out | |
2942 | */ | |
2943 | if (block_start == EXTENT_MAP_INLINE) { | |
2944 | SetPageError(page); | |
4b384318 MF |
2945 | if (!parent_locked) |
2946 | unlock_extent(tree, cur, cur + iosize - 1); | |
70dec807 | 2947 | cur = cur + iosize; |
306e16ce | 2948 | pg_offset += iosize; |
70dec807 CM |
2949 | continue; |
2950 | } | |
d1310b2e | 2951 | |
c8f2f24b | 2952 | pnr -= page->index; |
d4c7ca86 | 2953 | ret = submit_extent_page(rw, tree, page, |
306e16ce | 2954 | sector, disk_io_size, pg_offset, |
89642229 | 2955 | bdev, bio, pnr, |
c8b97818 CM |
2956 | end_bio_extent_readpage, mirror_num, |
2957 | *bio_flags, | |
2958 | this_bio_flag); | |
c8f2f24b JB |
2959 | if (!ret) { |
2960 | nr++; | |
2961 | *bio_flags = this_bio_flag; | |
2962 | } else { | |
d1310b2e | 2963 | SetPageError(page); |
4b384318 MF |
2964 | if (!parent_locked) |
2965 | unlock_extent(tree, cur, cur + iosize - 1); | |
edd33c99 | 2966 | } |
d1310b2e | 2967 | cur = cur + iosize; |
306e16ce | 2968 | pg_offset += iosize; |
d1310b2e | 2969 | } |
90a887c9 | 2970 | out: |
d1310b2e CM |
2971 | if (!nr) { |
2972 | if (!PageError(page)) | |
2973 | SetPageUptodate(page); | |
2974 | unlock_page(page); | |
2975 | } | |
2976 | return 0; | |
2977 | } | |
2978 | ||
9974090b MX |
2979 | static inline void __do_contiguous_readpages(struct extent_io_tree *tree, |
2980 | struct page *pages[], int nr_pages, | |
2981 | u64 start, u64 end, | |
2982 | get_extent_t *get_extent, | |
125bac01 | 2983 | struct extent_map **em_cached, |
9974090b MX |
2984 | struct bio **bio, int mirror_num, |
2985 | unsigned long *bio_flags, int rw) | |
2986 | { | |
2987 | struct inode *inode; | |
2988 | struct btrfs_ordered_extent *ordered; | |
2989 | int index; | |
2990 | ||
2991 | inode = pages[0]->mapping->host; | |
2992 | while (1) { | |
2993 | lock_extent(tree, start, end); | |
2994 | ordered = btrfs_lookup_ordered_range(inode, start, | |
2995 | end - start + 1); | |
2996 | if (!ordered) | |
2997 | break; | |
2998 | unlock_extent(tree, start, end); | |
2999 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3000 | btrfs_put_ordered_extent(ordered); | |
3001 | } | |
3002 | ||
3003 | for (index = 0; index < nr_pages; index++) { | |
125bac01 MX |
3004 | __do_readpage(tree, pages[index], get_extent, em_cached, bio, |
3005 | mirror_num, bio_flags, rw); | |
9974090b MX |
3006 | page_cache_release(pages[index]); |
3007 | } | |
3008 | } | |
3009 | ||
3010 | static void __extent_readpages(struct extent_io_tree *tree, | |
3011 | struct page *pages[], | |
3012 | int nr_pages, get_extent_t *get_extent, | |
125bac01 | 3013 | struct extent_map **em_cached, |
9974090b MX |
3014 | struct bio **bio, int mirror_num, |
3015 | unsigned long *bio_flags, int rw) | |
3016 | { | |
35a3621b | 3017 | u64 start = 0; |
9974090b MX |
3018 | u64 end = 0; |
3019 | u64 page_start; | |
3020 | int index; | |
35a3621b | 3021 | int first_index = 0; |
9974090b MX |
3022 | |
3023 | for (index = 0; index < nr_pages; index++) { | |
3024 | page_start = page_offset(pages[index]); | |
3025 | if (!end) { | |
3026 | start = page_start; | |
3027 | end = start + PAGE_CACHE_SIZE - 1; | |
3028 | first_index = index; | |
3029 | } else if (end + 1 == page_start) { | |
3030 | end += PAGE_CACHE_SIZE; | |
3031 | } else { | |
3032 | __do_contiguous_readpages(tree, &pages[first_index], | |
3033 | index - first_index, start, | |
125bac01 MX |
3034 | end, get_extent, em_cached, |
3035 | bio, mirror_num, bio_flags, | |
3036 | rw); | |
9974090b MX |
3037 | start = page_start; |
3038 | end = start + PAGE_CACHE_SIZE - 1; | |
3039 | first_index = index; | |
3040 | } | |
3041 | } | |
3042 | ||
3043 | if (end) | |
3044 | __do_contiguous_readpages(tree, &pages[first_index], | |
3045 | index - first_index, start, | |
125bac01 | 3046 | end, get_extent, em_cached, bio, |
9974090b MX |
3047 | mirror_num, bio_flags, rw); |
3048 | } | |
3049 | ||
3050 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
3051 | struct page *page, | |
3052 | get_extent_t *get_extent, | |
3053 | struct bio **bio, int mirror_num, | |
3054 | unsigned long *bio_flags, int rw) | |
3055 | { | |
3056 | struct inode *inode = page->mapping->host; | |
3057 | struct btrfs_ordered_extent *ordered; | |
3058 | u64 start = page_offset(page); | |
3059 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
3060 | int ret; | |
3061 | ||
3062 | while (1) { | |
3063 | lock_extent(tree, start, end); | |
3064 | ordered = btrfs_lookup_ordered_extent(inode, start); | |
3065 | if (!ordered) | |
3066 | break; | |
3067 | unlock_extent(tree, start, end); | |
3068 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3069 | btrfs_put_ordered_extent(ordered); | |
3070 | } | |
3071 | ||
125bac01 MX |
3072 | ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num, |
3073 | bio_flags, rw); | |
9974090b MX |
3074 | return ret; |
3075 | } | |
3076 | ||
d1310b2e | 3077 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, |
8ddc7d9c | 3078 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3079 | { |
3080 | struct bio *bio = NULL; | |
c8b97818 | 3081 | unsigned long bio_flags = 0; |
d1310b2e CM |
3082 | int ret; |
3083 | ||
8ddc7d9c | 3084 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, |
d4c7ca86 | 3085 | &bio_flags, READ); |
d1310b2e | 3086 | if (bio) |
8ddc7d9c | 3087 | ret = submit_one_bio(READ, bio, mirror_num, bio_flags); |
d1310b2e CM |
3088 | return ret; |
3089 | } | |
d1310b2e | 3090 | |
4b384318 MF |
3091 | int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page, |
3092 | get_extent_t *get_extent, int mirror_num) | |
3093 | { | |
3094 | struct bio *bio = NULL; | |
3095 | unsigned long bio_flags = EXTENT_BIO_PARENT_LOCKED; | |
3096 | int ret; | |
3097 | ||
3098 | ret = __do_readpage(tree, page, get_extent, NULL, &bio, mirror_num, | |
3099 | &bio_flags, READ); | |
3100 | if (bio) | |
3101 | ret = submit_one_bio(READ, bio, mirror_num, bio_flags); | |
3102 | return ret; | |
3103 | } | |
3104 | ||
11c8349b CM |
3105 | static noinline void update_nr_written(struct page *page, |
3106 | struct writeback_control *wbc, | |
3107 | unsigned long nr_written) | |
3108 | { | |
3109 | wbc->nr_to_write -= nr_written; | |
3110 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && | |
3111 | wbc->range_start == 0 && wbc->range_end == LLONG_MAX)) | |
3112 | page->mapping->writeback_index = page->index + nr_written; | |
3113 | } | |
3114 | ||
d1310b2e | 3115 | /* |
40f76580 CM |
3116 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
3117 | * | |
3118 | * This returns 1 if our fill_delalloc function did all the work required | |
3119 | * to write the page (copy into inline extent). In this case the IO has | |
3120 | * been started and the page is already unlocked. | |
3121 | * | |
3122 | * This returns 0 if all went well (page still locked) | |
3123 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 3124 | */ |
40f76580 CM |
3125 | static noinline_for_stack int writepage_delalloc(struct inode *inode, |
3126 | struct page *page, struct writeback_control *wbc, | |
3127 | struct extent_page_data *epd, | |
3128 | u64 delalloc_start, | |
3129 | unsigned long *nr_written) | |
3130 | { | |
3131 | struct extent_io_tree *tree = epd->tree; | |
3132 | u64 page_end = delalloc_start + PAGE_CACHE_SIZE - 1; | |
3133 | u64 nr_delalloc; | |
3134 | u64 delalloc_to_write = 0; | |
3135 | u64 delalloc_end = 0; | |
3136 | int ret; | |
3137 | int page_started = 0; | |
3138 | ||
3139 | if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc) | |
3140 | return 0; | |
3141 | ||
3142 | while (delalloc_end < page_end) { | |
3143 | nr_delalloc = find_lock_delalloc_range(inode, tree, | |
3144 | page, | |
3145 | &delalloc_start, | |
3146 | &delalloc_end, | |
3147 | 128 * 1024 * 1024); | |
3148 | if (nr_delalloc == 0) { | |
3149 | delalloc_start = delalloc_end + 1; | |
3150 | continue; | |
3151 | } | |
3152 | ret = tree->ops->fill_delalloc(inode, page, | |
3153 | delalloc_start, | |
3154 | delalloc_end, | |
3155 | &page_started, | |
3156 | nr_written); | |
3157 | /* File system has been set read-only */ | |
3158 | if (ret) { | |
3159 | SetPageError(page); | |
3160 | /* fill_delalloc should be return < 0 for error | |
3161 | * but just in case, we use > 0 here meaning the | |
3162 | * IO is started, so we don't want to return > 0 | |
3163 | * unless things are going well. | |
3164 | */ | |
3165 | ret = ret < 0 ? ret : -EIO; | |
3166 | goto done; | |
3167 | } | |
3168 | /* | |
3169 | * delalloc_end is already one less than the total | |
3170 | * length, so we don't subtract one from | |
3171 | * PAGE_CACHE_SIZE | |
3172 | */ | |
3173 | delalloc_to_write += (delalloc_end - delalloc_start + | |
3174 | PAGE_CACHE_SIZE) >> | |
3175 | PAGE_CACHE_SHIFT; | |
3176 | delalloc_start = delalloc_end + 1; | |
3177 | } | |
3178 | if (wbc->nr_to_write < delalloc_to_write) { | |
3179 | int thresh = 8192; | |
3180 | ||
3181 | if (delalloc_to_write < thresh * 2) | |
3182 | thresh = delalloc_to_write; | |
3183 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3184 | thresh); | |
3185 | } | |
3186 | ||
3187 | /* did the fill delalloc function already unlock and start | |
3188 | * the IO? | |
3189 | */ | |
3190 | if (page_started) { | |
3191 | /* | |
3192 | * we've unlocked the page, so we can't update | |
3193 | * the mapping's writeback index, just update | |
3194 | * nr_to_write. | |
3195 | */ | |
3196 | wbc->nr_to_write -= *nr_written; | |
3197 | return 1; | |
3198 | } | |
3199 | ||
3200 | ret = 0; | |
3201 | ||
3202 | done: | |
3203 | return ret; | |
3204 | } | |
3205 | ||
3206 | /* | |
3207 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3208 | * and does the loop to map the page into extents and bios. | |
3209 | * | |
3210 | * We return 1 if the IO is started and the page is unlocked, | |
3211 | * 0 if all went well (page still locked) | |
3212 | * < 0 if there were errors (page still locked) | |
3213 | */ | |
3214 | static noinline_for_stack int __extent_writepage_io(struct inode *inode, | |
3215 | struct page *page, | |
3216 | struct writeback_control *wbc, | |
3217 | struct extent_page_data *epd, | |
3218 | loff_t i_size, | |
3219 | unsigned long nr_written, | |
3220 | int write_flags, int *nr_ret) | |
d1310b2e | 3221 | { |
d1310b2e | 3222 | struct extent_io_tree *tree = epd->tree; |
4eee4fa4 | 3223 | u64 start = page_offset(page); |
d1310b2e CM |
3224 | u64 page_end = start + PAGE_CACHE_SIZE - 1; |
3225 | u64 end; | |
3226 | u64 cur = start; | |
3227 | u64 extent_offset; | |
d1310b2e CM |
3228 | u64 block_start; |
3229 | u64 iosize; | |
3230 | sector_t sector; | |
2c64c53d | 3231 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
3232 | struct extent_map *em; |
3233 | struct block_device *bdev; | |
7f3c74fb | 3234 | size_t pg_offset = 0; |
d1310b2e | 3235 | size_t blocksize; |
40f76580 CM |
3236 | int ret = 0; |
3237 | int nr = 0; | |
3238 | bool compressed; | |
c8b97818 | 3239 | |
247e743c | 3240 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
3241 | ret = tree->ops->writepage_start_hook(page, start, |
3242 | page_end); | |
87826df0 JM |
3243 | if (ret) { |
3244 | /* Fixup worker will requeue */ | |
3245 | if (ret == -EBUSY) | |
3246 | wbc->pages_skipped++; | |
3247 | else | |
3248 | redirty_page_for_writepage(wbc, page); | |
40f76580 | 3249 | |
11c8349b | 3250 | update_nr_written(page, wbc, nr_written); |
247e743c | 3251 | unlock_page(page); |
40f76580 | 3252 | ret = 1; |
11c8349b | 3253 | goto done_unlocked; |
247e743c CM |
3254 | } |
3255 | } | |
3256 | ||
11c8349b CM |
3257 | /* |
3258 | * we don't want to touch the inode after unlocking the page, | |
3259 | * so we update the mapping writeback index now | |
3260 | */ | |
3261 | update_nr_written(page, wbc, nr_written + 1); | |
771ed689 | 3262 | |
d1310b2e | 3263 | end = page_end; |
40f76580 | 3264 | if (i_size <= start) { |
e6dcd2dc CM |
3265 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3266 | tree->ops->writepage_end_io_hook(page, start, | |
3267 | page_end, NULL, 1); | |
d1310b2e CM |
3268 | goto done; |
3269 | } | |
3270 | ||
d1310b2e CM |
3271 | blocksize = inode->i_sb->s_blocksize; |
3272 | ||
3273 | while (cur <= end) { | |
40f76580 CM |
3274 | u64 em_end; |
3275 | if (cur >= i_size) { | |
e6dcd2dc CM |
3276 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3277 | tree->ops->writepage_end_io_hook(page, cur, | |
3278 | page_end, NULL, 1); | |
d1310b2e CM |
3279 | break; |
3280 | } | |
7f3c74fb | 3281 | em = epd->get_extent(inode, page, pg_offset, cur, |
d1310b2e | 3282 | end - cur + 1, 1); |
c704005d | 3283 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3284 | SetPageError(page); |
61391d56 | 3285 | ret = PTR_ERR_OR_ZERO(em); |
d1310b2e CM |
3286 | break; |
3287 | } | |
3288 | ||
3289 | extent_offset = cur - em->start; | |
40f76580 CM |
3290 | em_end = extent_map_end(em); |
3291 | BUG_ON(em_end <= cur); | |
d1310b2e | 3292 | BUG_ON(end < cur); |
40f76580 | 3293 | iosize = min(em_end - cur, end - cur + 1); |
fda2832f | 3294 | iosize = ALIGN(iosize, blocksize); |
d1310b2e CM |
3295 | sector = (em->block_start + extent_offset) >> 9; |
3296 | bdev = em->bdev; | |
3297 | block_start = em->block_start; | |
c8b97818 | 3298 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
3299 | free_extent_map(em); |
3300 | em = NULL; | |
3301 | ||
c8b97818 CM |
3302 | /* |
3303 | * compressed and inline extents are written through other | |
3304 | * paths in the FS | |
3305 | */ | |
3306 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3307 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
3308 | /* |
3309 | * end_io notification does not happen here for | |
3310 | * compressed extents | |
3311 | */ | |
3312 | if (!compressed && tree->ops && | |
3313 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
3314 | tree->ops->writepage_end_io_hook(page, cur, |
3315 | cur + iosize - 1, | |
3316 | NULL, 1); | |
c8b97818 CM |
3317 | else if (compressed) { |
3318 | /* we don't want to end_page_writeback on | |
3319 | * a compressed extent. this happens | |
3320 | * elsewhere | |
3321 | */ | |
3322 | nr++; | |
3323 | } | |
3324 | ||
3325 | cur += iosize; | |
7f3c74fb | 3326 | pg_offset += iosize; |
d1310b2e CM |
3327 | continue; |
3328 | } | |
c8b97818 | 3329 | |
d1310b2e CM |
3330 | if (tree->ops && tree->ops->writepage_io_hook) { |
3331 | ret = tree->ops->writepage_io_hook(page, cur, | |
3332 | cur + iosize - 1); | |
3333 | } else { | |
3334 | ret = 0; | |
3335 | } | |
1259ab75 | 3336 | if (ret) { |
d1310b2e | 3337 | SetPageError(page); |
1259ab75 | 3338 | } else { |
40f76580 | 3339 | unsigned long max_nr = (i_size >> PAGE_CACHE_SHIFT) + 1; |
7f3c74fb | 3340 | |
d1310b2e CM |
3341 | set_range_writeback(tree, cur, cur + iosize - 1); |
3342 | if (!PageWriteback(page)) { | |
efe120a0 FH |
3343 | btrfs_err(BTRFS_I(inode)->root->fs_info, |
3344 | "page %lu not writeback, cur %llu end %llu", | |
c1c9ff7c | 3345 | page->index, cur, end); |
d1310b2e CM |
3346 | } |
3347 | ||
ffbd517d CM |
3348 | ret = submit_extent_page(write_flags, tree, page, |
3349 | sector, iosize, pg_offset, | |
3350 | bdev, &epd->bio, max_nr, | |
c8b97818 CM |
3351 | end_bio_extent_writepage, |
3352 | 0, 0, 0); | |
d1310b2e CM |
3353 | if (ret) |
3354 | SetPageError(page); | |
3355 | } | |
3356 | cur = cur + iosize; | |
7f3c74fb | 3357 | pg_offset += iosize; |
d1310b2e CM |
3358 | nr++; |
3359 | } | |
40f76580 CM |
3360 | done: |
3361 | *nr_ret = nr; | |
3362 | ||
3363 | done_unlocked: | |
3364 | ||
3365 | /* drop our reference on any cached states */ | |
3366 | free_extent_state(cached_state); | |
3367 | return ret; | |
3368 | } | |
3369 | ||
3370 | /* | |
3371 | * the writepage semantics are similar to regular writepage. extent | |
3372 | * records are inserted to lock ranges in the tree, and as dirty areas | |
3373 | * are found, they are marked writeback. Then the lock bits are removed | |
3374 | * and the end_io handler clears the writeback ranges | |
3375 | */ | |
3376 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
3377 | void *data) | |
3378 | { | |
3379 | struct inode *inode = page->mapping->host; | |
3380 | struct extent_page_data *epd = data; | |
3381 | u64 start = page_offset(page); | |
3382 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
3383 | int ret; | |
3384 | int nr = 0; | |
3385 | size_t pg_offset = 0; | |
3386 | loff_t i_size = i_size_read(inode); | |
3387 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
3388 | int write_flags; | |
3389 | unsigned long nr_written = 0; | |
3390 | ||
3391 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3392 | write_flags = WRITE_SYNC; | |
3393 | else | |
3394 | write_flags = WRITE; | |
3395 | ||
3396 | trace___extent_writepage(page, inode, wbc); | |
3397 | ||
3398 | WARN_ON(!PageLocked(page)); | |
3399 | ||
3400 | ClearPageError(page); | |
3401 | ||
3402 | pg_offset = i_size & (PAGE_CACHE_SIZE - 1); | |
3403 | if (page->index > end_index || | |
3404 | (page->index == end_index && !pg_offset)) { | |
3405 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE); | |
3406 | unlock_page(page); | |
3407 | return 0; | |
3408 | } | |
3409 | ||
3410 | if (page->index == end_index) { | |
3411 | char *userpage; | |
3412 | ||
3413 | userpage = kmap_atomic(page); | |
3414 | memset(userpage + pg_offset, 0, | |
3415 | PAGE_CACHE_SIZE - pg_offset); | |
3416 | kunmap_atomic(userpage); | |
3417 | flush_dcache_page(page); | |
3418 | } | |
3419 | ||
3420 | pg_offset = 0; | |
3421 | ||
3422 | set_page_extent_mapped(page); | |
3423 | ||
3424 | ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written); | |
3425 | if (ret == 1) | |
3426 | goto done_unlocked; | |
3427 | if (ret) | |
3428 | goto done; | |
3429 | ||
3430 | ret = __extent_writepage_io(inode, page, wbc, epd, | |
3431 | i_size, nr_written, write_flags, &nr); | |
3432 | if (ret == 1) | |
3433 | goto done_unlocked; | |
3434 | ||
d1310b2e CM |
3435 | done: |
3436 | if (nr == 0) { | |
3437 | /* make sure the mapping tag for page dirty gets cleared */ | |
3438 | set_page_writeback(page); | |
3439 | end_page_writeback(page); | |
3440 | } | |
61391d56 FM |
3441 | if (PageError(page)) { |
3442 | ret = ret < 0 ? ret : -EIO; | |
3443 | end_extent_writepage(page, ret, start, page_end); | |
3444 | } | |
d1310b2e | 3445 | unlock_page(page); |
40f76580 | 3446 | return ret; |
771ed689 | 3447 | |
11c8349b | 3448 | done_unlocked: |
d1310b2e CM |
3449 | return 0; |
3450 | } | |
3451 | ||
fd8b2b61 | 3452 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 3453 | { |
74316201 N |
3454 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
3455 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
3456 | } |
3457 | ||
0e378df1 CM |
3458 | static noinline_for_stack int |
3459 | lock_extent_buffer_for_io(struct extent_buffer *eb, | |
3460 | struct btrfs_fs_info *fs_info, | |
3461 | struct extent_page_data *epd) | |
0b32f4bb JB |
3462 | { |
3463 | unsigned long i, num_pages; | |
3464 | int flush = 0; | |
3465 | int ret = 0; | |
3466 | ||
3467 | if (!btrfs_try_tree_write_lock(eb)) { | |
3468 | flush = 1; | |
3469 | flush_write_bio(epd); | |
3470 | btrfs_tree_lock(eb); | |
3471 | } | |
3472 | ||
3473 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3474 | btrfs_tree_unlock(eb); | |
3475 | if (!epd->sync_io) | |
3476 | return 0; | |
3477 | if (!flush) { | |
3478 | flush_write_bio(epd); | |
3479 | flush = 1; | |
3480 | } | |
a098d8e8 CM |
3481 | while (1) { |
3482 | wait_on_extent_buffer_writeback(eb); | |
3483 | btrfs_tree_lock(eb); | |
3484 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3485 | break; | |
0b32f4bb | 3486 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3487 | } |
3488 | } | |
3489 | ||
51561ffe JB |
3490 | /* |
3491 | * We need to do this to prevent races in people who check if the eb is | |
3492 | * under IO since we can end up having no IO bits set for a short period | |
3493 | * of time. | |
3494 | */ | |
3495 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
3496 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3497 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 3498 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 3499 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
e2d84521 MX |
3500 | __percpu_counter_add(&fs_info->dirty_metadata_bytes, |
3501 | -eb->len, | |
3502 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 3503 | ret = 1; |
51561ffe JB |
3504 | } else { |
3505 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
3506 | } |
3507 | ||
3508 | btrfs_tree_unlock(eb); | |
3509 | ||
3510 | if (!ret) | |
3511 | return ret; | |
3512 | ||
3513 | num_pages = num_extent_pages(eb->start, eb->len); | |
3514 | for (i = 0; i < num_pages; i++) { | |
3515 | struct page *p = extent_buffer_page(eb, i); | |
3516 | ||
3517 | if (!trylock_page(p)) { | |
3518 | if (!flush) { | |
3519 | flush_write_bio(epd); | |
3520 | flush = 1; | |
3521 | } | |
3522 | lock_page(p); | |
3523 | } | |
3524 | } | |
3525 | ||
3526 | return ret; | |
3527 | } | |
3528 | ||
3529 | static void end_extent_buffer_writeback(struct extent_buffer *eb) | |
3530 | { | |
3531 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
4e857c58 | 3532 | smp_mb__after_atomic(); |
0b32f4bb JB |
3533 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); |
3534 | } | |
3535 | ||
3536 | static void end_bio_extent_buffer_writepage(struct bio *bio, int err) | |
3537 | { | |
2c30c71b | 3538 | struct bio_vec *bvec; |
0b32f4bb | 3539 | struct extent_buffer *eb; |
2c30c71b | 3540 | int i, done; |
0b32f4bb | 3541 | |
2c30c71b | 3542 | bio_for_each_segment_all(bvec, bio, i) { |
0b32f4bb JB |
3543 | struct page *page = bvec->bv_page; |
3544 | ||
0b32f4bb JB |
3545 | eb = (struct extent_buffer *)page->private; |
3546 | BUG_ON(!eb); | |
3547 | done = atomic_dec_and_test(&eb->io_pages); | |
3548 | ||
2c30c71b | 3549 | if (err || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) { |
0b32f4bb JB |
3550 | set_bit(EXTENT_BUFFER_IOERR, &eb->bflags); |
3551 | ClearPageUptodate(page); | |
3552 | SetPageError(page); | |
3553 | } | |
3554 | ||
3555 | end_page_writeback(page); | |
3556 | ||
3557 | if (!done) | |
3558 | continue; | |
3559 | ||
3560 | end_extent_buffer_writeback(eb); | |
2c30c71b | 3561 | } |
0b32f4bb JB |
3562 | |
3563 | bio_put(bio); | |
0b32f4bb JB |
3564 | } |
3565 | ||
0e378df1 | 3566 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb JB |
3567 | struct btrfs_fs_info *fs_info, |
3568 | struct writeback_control *wbc, | |
3569 | struct extent_page_data *epd) | |
3570 | { | |
3571 | struct block_device *bdev = fs_info->fs_devices->latest_bdev; | |
f28491e0 | 3572 | struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree; |
0b32f4bb JB |
3573 | u64 offset = eb->start; |
3574 | unsigned long i, num_pages; | |
de0022b9 | 3575 | unsigned long bio_flags = 0; |
d4c7ca86 | 3576 | int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META; |
d7dbe9e7 | 3577 | int ret = 0; |
0b32f4bb JB |
3578 | |
3579 | clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
3580 | num_pages = num_extent_pages(eb->start, eb->len); | |
3581 | atomic_set(&eb->io_pages, num_pages); | |
de0022b9 JB |
3582 | if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID) |
3583 | bio_flags = EXTENT_BIO_TREE_LOG; | |
3584 | ||
0b32f4bb JB |
3585 | for (i = 0; i < num_pages; i++) { |
3586 | struct page *p = extent_buffer_page(eb, i); | |
3587 | ||
3588 | clear_page_dirty_for_io(p); | |
3589 | set_page_writeback(p); | |
f28491e0 | 3590 | ret = submit_extent_page(rw, tree, p, offset >> 9, |
0b32f4bb JB |
3591 | PAGE_CACHE_SIZE, 0, bdev, &epd->bio, |
3592 | -1, end_bio_extent_buffer_writepage, | |
de0022b9 JB |
3593 | 0, epd->bio_flags, bio_flags); |
3594 | epd->bio_flags = bio_flags; | |
0b32f4bb JB |
3595 | if (ret) { |
3596 | set_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
3597 | SetPageError(p); | |
3598 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) | |
3599 | end_extent_buffer_writeback(eb); | |
3600 | ret = -EIO; | |
3601 | break; | |
3602 | } | |
3603 | offset += PAGE_CACHE_SIZE; | |
3604 | update_nr_written(p, wbc, 1); | |
3605 | unlock_page(p); | |
3606 | } | |
3607 | ||
3608 | if (unlikely(ret)) { | |
3609 | for (; i < num_pages; i++) { | |
3610 | struct page *p = extent_buffer_page(eb, i); | |
3611 | unlock_page(p); | |
3612 | } | |
3613 | } | |
3614 | ||
3615 | return ret; | |
3616 | } | |
3617 | ||
3618 | int btree_write_cache_pages(struct address_space *mapping, | |
3619 | struct writeback_control *wbc) | |
3620 | { | |
3621 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; | |
3622 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; | |
3623 | struct extent_buffer *eb, *prev_eb = NULL; | |
3624 | struct extent_page_data epd = { | |
3625 | .bio = NULL, | |
3626 | .tree = tree, | |
3627 | .extent_locked = 0, | |
3628 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
de0022b9 | 3629 | .bio_flags = 0, |
0b32f4bb JB |
3630 | }; |
3631 | int ret = 0; | |
3632 | int done = 0; | |
3633 | int nr_to_write_done = 0; | |
3634 | struct pagevec pvec; | |
3635 | int nr_pages; | |
3636 | pgoff_t index; | |
3637 | pgoff_t end; /* Inclusive */ | |
3638 | int scanned = 0; | |
3639 | int tag; | |
3640 | ||
3641 | pagevec_init(&pvec, 0); | |
3642 | if (wbc->range_cyclic) { | |
3643 | index = mapping->writeback_index; /* Start from prev offset */ | |
3644 | end = -1; | |
3645 | } else { | |
3646 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
3647 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
3648 | scanned = 1; | |
3649 | } | |
3650 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3651 | tag = PAGECACHE_TAG_TOWRITE; | |
3652 | else | |
3653 | tag = PAGECACHE_TAG_DIRTY; | |
3654 | retry: | |
3655 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3656 | tag_pages_for_writeback(mapping, index, end); | |
3657 | while (!done && !nr_to_write_done && (index <= end) && | |
3658 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, | |
3659 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
3660 | unsigned i; | |
3661 | ||
3662 | scanned = 1; | |
3663 | for (i = 0; i < nr_pages; i++) { | |
3664 | struct page *page = pvec.pages[i]; | |
3665 | ||
3666 | if (!PagePrivate(page)) | |
3667 | continue; | |
3668 | ||
3669 | if (!wbc->range_cyclic && page->index > end) { | |
3670 | done = 1; | |
3671 | break; | |
3672 | } | |
3673 | ||
b5bae261 JB |
3674 | spin_lock(&mapping->private_lock); |
3675 | if (!PagePrivate(page)) { | |
3676 | spin_unlock(&mapping->private_lock); | |
3677 | continue; | |
3678 | } | |
3679 | ||
0b32f4bb | 3680 | eb = (struct extent_buffer *)page->private; |
b5bae261 JB |
3681 | |
3682 | /* | |
3683 | * Shouldn't happen and normally this would be a BUG_ON | |
3684 | * but no sense in crashing the users box for something | |
3685 | * we can survive anyway. | |
3686 | */ | |
fae7f21c | 3687 | if (WARN_ON(!eb)) { |
b5bae261 | 3688 | spin_unlock(&mapping->private_lock); |
0b32f4bb JB |
3689 | continue; |
3690 | } | |
3691 | ||
b5bae261 JB |
3692 | if (eb == prev_eb) { |
3693 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 3694 | continue; |
b5bae261 | 3695 | } |
0b32f4bb | 3696 | |
b5bae261 JB |
3697 | ret = atomic_inc_not_zero(&eb->refs); |
3698 | spin_unlock(&mapping->private_lock); | |
3699 | if (!ret) | |
0b32f4bb | 3700 | continue; |
0b32f4bb JB |
3701 | |
3702 | prev_eb = eb; | |
3703 | ret = lock_extent_buffer_for_io(eb, fs_info, &epd); | |
3704 | if (!ret) { | |
3705 | free_extent_buffer(eb); | |
3706 | continue; | |
3707 | } | |
3708 | ||
3709 | ret = write_one_eb(eb, fs_info, wbc, &epd); | |
3710 | if (ret) { | |
3711 | done = 1; | |
3712 | free_extent_buffer(eb); | |
3713 | break; | |
3714 | } | |
3715 | free_extent_buffer(eb); | |
3716 | ||
3717 | /* | |
3718 | * the filesystem may choose to bump up nr_to_write. | |
3719 | * We have to make sure to honor the new nr_to_write | |
3720 | * at any time | |
3721 | */ | |
3722 | nr_to_write_done = wbc->nr_to_write <= 0; | |
3723 | } | |
3724 | pagevec_release(&pvec); | |
3725 | cond_resched(); | |
3726 | } | |
3727 | if (!scanned && !done) { | |
3728 | /* | |
3729 | * We hit the last page and there is more work to be done: wrap | |
3730 | * back to the start of the file | |
3731 | */ | |
3732 | scanned = 1; | |
3733 | index = 0; | |
3734 | goto retry; | |
3735 | } | |
3736 | flush_write_bio(&epd); | |
3737 | return ret; | |
3738 | } | |
3739 | ||
d1310b2e | 3740 | /** |
4bef0848 | 3741 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
3742 | * @mapping: address space structure to write |
3743 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
3744 | * @writepage: function called for each page | |
3745 | * @data: data passed to writepage function | |
3746 | * | |
3747 | * If a page is already under I/O, write_cache_pages() skips it, even | |
3748 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
3749 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
3750 | * and msync() need to guarantee that all the data which was dirty at the time | |
3751 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
3752 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
3753 | * existing IO to complete. | |
3754 | */ | |
b2950863 | 3755 | static int extent_write_cache_pages(struct extent_io_tree *tree, |
4bef0848 CM |
3756 | struct address_space *mapping, |
3757 | struct writeback_control *wbc, | |
d2c3f4f6 CM |
3758 | writepage_t writepage, void *data, |
3759 | void (*flush_fn)(void *)) | |
d1310b2e | 3760 | { |
7fd1a3f7 | 3761 | struct inode *inode = mapping->host; |
d1310b2e CM |
3762 | int ret = 0; |
3763 | int done = 0; | |
61391d56 | 3764 | int err = 0; |
f85d7d6c | 3765 | int nr_to_write_done = 0; |
d1310b2e CM |
3766 | struct pagevec pvec; |
3767 | int nr_pages; | |
3768 | pgoff_t index; | |
3769 | pgoff_t end; /* Inclusive */ | |
3770 | int scanned = 0; | |
f7aaa06b | 3771 | int tag; |
d1310b2e | 3772 | |
7fd1a3f7 JB |
3773 | /* |
3774 | * We have to hold onto the inode so that ordered extents can do their | |
3775 | * work when the IO finishes. The alternative to this is failing to add | |
3776 | * an ordered extent if the igrab() fails there and that is a huge pain | |
3777 | * to deal with, so instead just hold onto the inode throughout the | |
3778 | * writepages operation. If it fails here we are freeing up the inode | |
3779 | * anyway and we'd rather not waste our time writing out stuff that is | |
3780 | * going to be truncated anyway. | |
3781 | */ | |
3782 | if (!igrab(inode)) | |
3783 | return 0; | |
3784 | ||
d1310b2e CM |
3785 | pagevec_init(&pvec, 0); |
3786 | if (wbc->range_cyclic) { | |
3787 | index = mapping->writeback_index; /* Start from prev offset */ | |
3788 | end = -1; | |
3789 | } else { | |
3790 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
3791 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
3792 | scanned = 1; |
3793 | } | |
f7aaa06b JB |
3794 | if (wbc->sync_mode == WB_SYNC_ALL) |
3795 | tag = PAGECACHE_TAG_TOWRITE; | |
3796 | else | |
3797 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 3798 | retry: |
f7aaa06b JB |
3799 | if (wbc->sync_mode == WB_SYNC_ALL) |
3800 | tag_pages_for_writeback(mapping, index, end); | |
f85d7d6c | 3801 | while (!done && !nr_to_write_done && (index <= end) && |
f7aaa06b JB |
3802 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
3803 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
d1310b2e CM |
3804 | unsigned i; |
3805 | ||
3806 | scanned = 1; | |
3807 | for (i = 0; i < nr_pages; i++) { | |
3808 | struct page *page = pvec.pages[i]; | |
3809 | ||
3810 | /* | |
3811 | * At this point we hold neither mapping->tree_lock nor | |
3812 | * lock on the page itself: the page may be truncated or | |
3813 | * invalidated (changing page->mapping to NULL), or even | |
3814 | * swizzled back from swapper_space to tmpfs file | |
3815 | * mapping | |
3816 | */ | |
c8f2f24b JB |
3817 | if (!trylock_page(page)) { |
3818 | flush_fn(data); | |
3819 | lock_page(page); | |
01d658f2 | 3820 | } |
d1310b2e CM |
3821 | |
3822 | if (unlikely(page->mapping != mapping)) { | |
3823 | unlock_page(page); | |
3824 | continue; | |
3825 | } | |
3826 | ||
3827 | if (!wbc->range_cyclic && page->index > end) { | |
3828 | done = 1; | |
3829 | unlock_page(page); | |
3830 | continue; | |
3831 | } | |
3832 | ||
d2c3f4f6 | 3833 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 CM |
3834 | if (PageWriteback(page)) |
3835 | flush_fn(data); | |
d1310b2e | 3836 | wait_on_page_writeback(page); |
d2c3f4f6 | 3837 | } |
d1310b2e CM |
3838 | |
3839 | if (PageWriteback(page) || | |
3840 | !clear_page_dirty_for_io(page)) { | |
3841 | unlock_page(page); | |
3842 | continue; | |
3843 | } | |
3844 | ||
3845 | ret = (*writepage)(page, wbc, data); | |
3846 | ||
3847 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
3848 | unlock_page(page); | |
3849 | ret = 0; | |
3850 | } | |
61391d56 FM |
3851 | if (!err && ret < 0) |
3852 | err = ret; | |
f85d7d6c CM |
3853 | |
3854 | /* | |
3855 | * the filesystem may choose to bump up nr_to_write. | |
3856 | * We have to make sure to honor the new nr_to_write | |
3857 | * at any time | |
3858 | */ | |
3859 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
3860 | } |
3861 | pagevec_release(&pvec); | |
3862 | cond_resched(); | |
3863 | } | |
61391d56 | 3864 | if (!scanned && !done && !err) { |
d1310b2e CM |
3865 | /* |
3866 | * We hit the last page and there is more work to be done: wrap | |
3867 | * back to the start of the file | |
3868 | */ | |
3869 | scanned = 1; | |
3870 | index = 0; | |
3871 | goto retry; | |
3872 | } | |
7fd1a3f7 | 3873 | btrfs_add_delayed_iput(inode); |
61391d56 | 3874 | return err; |
d1310b2e | 3875 | } |
d1310b2e | 3876 | |
ffbd517d | 3877 | static void flush_epd_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 3878 | { |
d2c3f4f6 | 3879 | if (epd->bio) { |
355808c2 JM |
3880 | int rw = WRITE; |
3881 | int ret; | |
3882 | ||
ffbd517d | 3883 | if (epd->sync_io) |
355808c2 JM |
3884 | rw = WRITE_SYNC; |
3885 | ||
de0022b9 | 3886 | ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags); |
79787eaa | 3887 | BUG_ON(ret < 0); /* -ENOMEM */ |
d2c3f4f6 CM |
3888 | epd->bio = NULL; |
3889 | } | |
3890 | } | |
3891 | ||
ffbd517d CM |
3892 | static noinline void flush_write_bio(void *data) |
3893 | { | |
3894 | struct extent_page_data *epd = data; | |
3895 | flush_epd_write_bio(epd); | |
3896 | } | |
3897 | ||
d1310b2e CM |
3898 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, |
3899 | get_extent_t *get_extent, | |
3900 | struct writeback_control *wbc) | |
3901 | { | |
3902 | int ret; | |
d1310b2e CM |
3903 | struct extent_page_data epd = { |
3904 | .bio = NULL, | |
3905 | .tree = tree, | |
3906 | .get_extent = get_extent, | |
771ed689 | 3907 | .extent_locked = 0, |
ffbd517d | 3908 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
de0022b9 | 3909 | .bio_flags = 0, |
d1310b2e | 3910 | }; |
d1310b2e | 3911 | |
d1310b2e CM |
3912 | ret = __extent_writepage(page, wbc, &epd); |
3913 | ||
ffbd517d | 3914 | flush_epd_write_bio(&epd); |
d1310b2e CM |
3915 | return ret; |
3916 | } | |
d1310b2e | 3917 | |
771ed689 CM |
3918 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, |
3919 | u64 start, u64 end, get_extent_t *get_extent, | |
3920 | int mode) | |
3921 | { | |
3922 | int ret = 0; | |
3923 | struct address_space *mapping = inode->i_mapping; | |
3924 | struct page *page; | |
3925 | unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >> | |
3926 | PAGE_CACHE_SHIFT; | |
3927 | ||
3928 | struct extent_page_data epd = { | |
3929 | .bio = NULL, | |
3930 | .tree = tree, | |
3931 | .get_extent = get_extent, | |
3932 | .extent_locked = 1, | |
ffbd517d | 3933 | .sync_io = mode == WB_SYNC_ALL, |
de0022b9 | 3934 | .bio_flags = 0, |
771ed689 CM |
3935 | }; |
3936 | struct writeback_control wbc_writepages = { | |
771ed689 | 3937 | .sync_mode = mode, |
771ed689 CM |
3938 | .nr_to_write = nr_pages * 2, |
3939 | .range_start = start, | |
3940 | .range_end = end + 1, | |
3941 | }; | |
3942 | ||
d397712b | 3943 | while (start <= end) { |
771ed689 CM |
3944 | page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); |
3945 | if (clear_page_dirty_for_io(page)) | |
3946 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
3947 | else { | |
3948 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
3949 | tree->ops->writepage_end_io_hook(page, start, | |
3950 | start + PAGE_CACHE_SIZE - 1, | |
3951 | NULL, 1); | |
3952 | unlock_page(page); | |
3953 | } | |
3954 | page_cache_release(page); | |
3955 | start += PAGE_CACHE_SIZE; | |
3956 | } | |
3957 | ||
ffbd517d | 3958 | flush_epd_write_bio(&epd); |
771ed689 CM |
3959 | return ret; |
3960 | } | |
d1310b2e CM |
3961 | |
3962 | int extent_writepages(struct extent_io_tree *tree, | |
3963 | struct address_space *mapping, | |
3964 | get_extent_t *get_extent, | |
3965 | struct writeback_control *wbc) | |
3966 | { | |
3967 | int ret = 0; | |
3968 | struct extent_page_data epd = { | |
3969 | .bio = NULL, | |
3970 | .tree = tree, | |
3971 | .get_extent = get_extent, | |
771ed689 | 3972 | .extent_locked = 0, |
ffbd517d | 3973 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
de0022b9 | 3974 | .bio_flags = 0, |
d1310b2e CM |
3975 | }; |
3976 | ||
4bef0848 | 3977 | ret = extent_write_cache_pages(tree, mapping, wbc, |
d2c3f4f6 CM |
3978 | __extent_writepage, &epd, |
3979 | flush_write_bio); | |
ffbd517d | 3980 | flush_epd_write_bio(&epd); |
d1310b2e CM |
3981 | return ret; |
3982 | } | |
d1310b2e CM |
3983 | |
3984 | int extent_readpages(struct extent_io_tree *tree, | |
3985 | struct address_space *mapping, | |
3986 | struct list_head *pages, unsigned nr_pages, | |
3987 | get_extent_t get_extent) | |
3988 | { | |
3989 | struct bio *bio = NULL; | |
3990 | unsigned page_idx; | |
c8b97818 | 3991 | unsigned long bio_flags = 0; |
67c9684f LB |
3992 | struct page *pagepool[16]; |
3993 | struct page *page; | |
125bac01 | 3994 | struct extent_map *em_cached = NULL; |
67c9684f | 3995 | int nr = 0; |
d1310b2e | 3996 | |
d1310b2e | 3997 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
67c9684f | 3998 | page = list_entry(pages->prev, struct page, lru); |
d1310b2e CM |
3999 | |
4000 | prefetchw(&page->flags); | |
4001 | list_del(&page->lru); | |
67c9684f | 4002 | if (add_to_page_cache_lru(page, mapping, |
43e817a1 | 4003 | page->index, GFP_NOFS)) { |
67c9684f LB |
4004 | page_cache_release(page); |
4005 | continue; | |
d1310b2e | 4006 | } |
67c9684f LB |
4007 | |
4008 | pagepool[nr++] = page; | |
4009 | if (nr < ARRAY_SIZE(pagepool)) | |
4010 | continue; | |
125bac01 | 4011 | __extent_readpages(tree, pagepool, nr, get_extent, &em_cached, |
9974090b | 4012 | &bio, 0, &bio_flags, READ); |
67c9684f | 4013 | nr = 0; |
d1310b2e | 4014 | } |
9974090b | 4015 | if (nr) |
125bac01 | 4016 | __extent_readpages(tree, pagepool, nr, get_extent, &em_cached, |
9974090b | 4017 | &bio, 0, &bio_flags, READ); |
67c9684f | 4018 | |
125bac01 MX |
4019 | if (em_cached) |
4020 | free_extent_map(em_cached); | |
4021 | ||
d1310b2e CM |
4022 | BUG_ON(!list_empty(pages)); |
4023 | if (bio) | |
79787eaa | 4024 | return submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
4025 | return 0; |
4026 | } | |
d1310b2e CM |
4027 | |
4028 | /* | |
4029 | * basic invalidatepage code, this waits on any locked or writeback | |
4030 | * ranges corresponding to the page, and then deletes any extent state | |
4031 | * records from the tree | |
4032 | */ | |
4033 | int extent_invalidatepage(struct extent_io_tree *tree, | |
4034 | struct page *page, unsigned long offset) | |
4035 | { | |
2ac55d41 | 4036 | struct extent_state *cached_state = NULL; |
4eee4fa4 | 4037 | u64 start = page_offset(page); |
d1310b2e CM |
4038 | u64 end = start + PAGE_CACHE_SIZE - 1; |
4039 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
4040 | ||
fda2832f | 4041 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
4042 | if (start > end) |
4043 | return 0; | |
4044 | ||
d0082371 | 4045 | lock_extent_bits(tree, start, end, 0, &cached_state); |
1edbb734 | 4046 | wait_on_page_writeback(page); |
d1310b2e | 4047 | clear_extent_bit(tree, start, end, |
32c00aff JB |
4048 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
4049 | EXTENT_DO_ACCOUNTING, | |
2ac55d41 | 4050 | 1, 1, &cached_state, GFP_NOFS); |
d1310b2e CM |
4051 | return 0; |
4052 | } | |
d1310b2e | 4053 | |
7b13b7b1 CM |
4054 | /* |
4055 | * a helper for releasepage, this tests for areas of the page that | |
4056 | * are locked or under IO and drops the related state bits if it is safe | |
4057 | * to drop the page. | |
4058 | */ | |
48a3b636 ES |
4059 | static int try_release_extent_state(struct extent_map_tree *map, |
4060 | struct extent_io_tree *tree, | |
4061 | struct page *page, gfp_t mask) | |
7b13b7b1 | 4062 | { |
4eee4fa4 | 4063 | u64 start = page_offset(page); |
7b13b7b1 CM |
4064 | u64 end = start + PAGE_CACHE_SIZE - 1; |
4065 | int ret = 1; | |
4066 | ||
211f90e6 | 4067 | if (test_range_bit(tree, start, end, |
8b62b72b | 4068 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
4069 | ret = 0; |
4070 | else { | |
4071 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
4072 | mask = GFP_NOFS; | |
11ef160f CM |
4073 | /* |
4074 | * at this point we can safely clear everything except the | |
4075 | * locked bit and the nodatasum bit | |
4076 | */ | |
e3f24cc5 | 4077 | ret = clear_extent_bit(tree, start, end, |
11ef160f CM |
4078 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
4079 | 0, 0, NULL, mask); | |
e3f24cc5 CM |
4080 | |
4081 | /* if clear_extent_bit failed for enomem reasons, | |
4082 | * we can't allow the release to continue. | |
4083 | */ | |
4084 | if (ret < 0) | |
4085 | ret = 0; | |
4086 | else | |
4087 | ret = 1; | |
7b13b7b1 CM |
4088 | } |
4089 | return ret; | |
4090 | } | |
7b13b7b1 | 4091 | |
d1310b2e CM |
4092 | /* |
4093 | * a helper for releasepage. As long as there are no locked extents | |
4094 | * in the range corresponding to the page, both state records and extent | |
4095 | * map records are removed | |
4096 | */ | |
4097 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
4098 | struct extent_io_tree *tree, struct page *page, |
4099 | gfp_t mask) | |
d1310b2e CM |
4100 | { |
4101 | struct extent_map *em; | |
4eee4fa4 | 4102 | u64 start = page_offset(page); |
d1310b2e | 4103 | u64 end = start + PAGE_CACHE_SIZE - 1; |
7b13b7b1 | 4104 | |
70dec807 CM |
4105 | if ((mask & __GFP_WAIT) && |
4106 | page->mapping->host->i_size > 16 * 1024 * 1024) { | |
39b5637f | 4107 | u64 len; |
70dec807 | 4108 | while (start <= end) { |
39b5637f | 4109 | len = end - start + 1; |
890871be | 4110 | write_lock(&map->lock); |
39b5637f | 4111 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 4112 | if (!em) { |
890871be | 4113 | write_unlock(&map->lock); |
70dec807 CM |
4114 | break; |
4115 | } | |
7f3c74fb CM |
4116 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
4117 | em->start != start) { | |
890871be | 4118 | write_unlock(&map->lock); |
70dec807 CM |
4119 | free_extent_map(em); |
4120 | break; | |
4121 | } | |
4122 | if (!test_range_bit(tree, em->start, | |
4123 | extent_map_end(em) - 1, | |
8b62b72b | 4124 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 4125 | 0, NULL)) { |
70dec807 CM |
4126 | remove_extent_mapping(map, em); |
4127 | /* once for the rb tree */ | |
4128 | free_extent_map(em); | |
4129 | } | |
4130 | start = extent_map_end(em); | |
890871be | 4131 | write_unlock(&map->lock); |
70dec807 CM |
4132 | |
4133 | /* once for us */ | |
d1310b2e CM |
4134 | free_extent_map(em); |
4135 | } | |
d1310b2e | 4136 | } |
7b13b7b1 | 4137 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 4138 | } |
d1310b2e | 4139 | |
ec29ed5b CM |
4140 | /* |
4141 | * helper function for fiemap, which doesn't want to see any holes. | |
4142 | * This maps until we find something past 'last' | |
4143 | */ | |
4144 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
4145 | u64 offset, | |
4146 | u64 last, | |
4147 | get_extent_t *get_extent) | |
4148 | { | |
4149 | u64 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
4150 | struct extent_map *em; | |
4151 | u64 len; | |
4152 | ||
4153 | if (offset >= last) | |
4154 | return NULL; | |
4155 | ||
67871254 | 4156 | while (1) { |
ec29ed5b CM |
4157 | len = last - offset; |
4158 | if (len == 0) | |
4159 | break; | |
fda2832f | 4160 | len = ALIGN(len, sectorsize); |
ec29ed5b | 4161 | em = get_extent(inode, NULL, 0, offset, len, 0); |
c704005d | 4162 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
4163 | return em; |
4164 | ||
4165 | /* if this isn't a hole return it */ | |
4166 | if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) && | |
4167 | em->block_start != EXTENT_MAP_HOLE) { | |
4168 | return em; | |
4169 | } | |
4170 | ||
4171 | /* this is a hole, advance to the next extent */ | |
4172 | offset = extent_map_end(em); | |
4173 | free_extent_map(em); | |
4174 | if (offset >= last) | |
4175 | break; | |
4176 | } | |
4177 | return NULL; | |
4178 | } | |
4179 | ||
1506fcc8 YS |
4180 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4181 | __u64 start, __u64 len, get_extent_t *get_extent) | |
4182 | { | |
975f84fe | 4183 | int ret = 0; |
1506fcc8 YS |
4184 | u64 off = start; |
4185 | u64 max = start + len; | |
4186 | u32 flags = 0; | |
975f84fe JB |
4187 | u32 found_type; |
4188 | u64 last; | |
ec29ed5b | 4189 | u64 last_for_get_extent = 0; |
1506fcc8 | 4190 | u64 disko = 0; |
ec29ed5b | 4191 | u64 isize = i_size_read(inode); |
975f84fe | 4192 | struct btrfs_key found_key; |
1506fcc8 | 4193 | struct extent_map *em = NULL; |
2ac55d41 | 4194 | struct extent_state *cached_state = NULL; |
975f84fe | 4195 | struct btrfs_path *path; |
dc046b10 | 4196 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1506fcc8 | 4197 | int end = 0; |
ec29ed5b CM |
4198 | u64 em_start = 0; |
4199 | u64 em_len = 0; | |
4200 | u64 em_end = 0; | |
1506fcc8 YS |
4201 | |
4202 | if (len == 0) | |
4203 | return -EINVAL; | |
4204 | ||
975f84fe JB |
4205 | path = btrfs_alloc_path(); |
4206 | if (!path) | |
4207 | return -ENOMEM; | |
4208 | path->leave_spinning = 1; | |
4209 | ||
2c91943b QW |
4210 | start = round_down(start, BTRFS_I(inode)->root->sectorsize); |
4211 | len = round_up(max, BTRFS_I(inode)->root->sectorsize) - start; | |
4d479cf0 | 4212 | |
ec29ed5b CM |
4213 | /* |
4214 | * lookup the last file extent. We're not using i_size here | |
4215 | * because there might be preallocation past i_size | |
4216 | */ | |
dc046b10 JB |
4217 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1, |
4218 | 0); | |
975f84fe JB |
4219 | if (ret < 0) { |
4220 | btrfs_free_path(path); | |
4221 | return ret; | |
4222 | } | |
4223 | WARN_ON(!ret); | |
4224 | path->slots[0]--; | |
975f84fe | 4225 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
962a298f | 4226 | found_type = found_key.type; |
975f84fe | 4227 | |
ec29ed5b | 4228 | /* No extents, but there might be delalloc bits */ |
33345d01 | 4229 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 4230 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
4231 | /* have to trust i_size as the end */ |
4232 | last = (u64)-1; | |
4233 | last_for_get_extent = isize; | |
4234 | } else { | |
4235 | /* | |
4236 | * remember the start of the last extent. There are a | |
4237 | * bunch of different factors that go into the length of the | |
4238 | * extent, so its much less complex to remember where it started | |
4239 | */ | |
4240 | last = found_key.offset; | |
4241 | last_for_get_extent = last + 1; | |
975f84fe | 4242 | } |
fe09e16c | 4243 | btrfs_release_path(path); |
975f84fe | 4244 | |
ec29ed5b CM |
4245 | /* |
4246 | * we might have some extents allocated but more delalloc past those | |
4247 | * extents. so, we trust isize unless the start of the last extent is | |
4248 | * beyond isize | |
4249 | */ | |
4250 | if (last < isize) { | |
4251 | last = (u64)-1; | |
4252 | last_for_get_extent = isize; | |
4253 | } | |
4254 | ||
a52f4cd2 | 4255 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0, |
d0082371 | 4256 | &cached_state); |
ec29ed5b | 4257 | |
4d479cf0 | 4258 | em = get_extent_skip_holes(inode, start, last_for_get_extent, |
ec29ed5b | 4259 | get_extent); |
1506fcc8 YS |
4260 | if (!em) |
4261 | goto out; | |
4262 | if (IS_ERR(em)) { | |
4263 | ret = PTR_ERR(em); | |
4264 | goto out; | |
4265 | } | |
975f84fe | 4266 | |
1506fcc8 | 4267 | while (!end) { |
b76bb701 | 4268 | u64 offset_in_extent = 0; |
ea8efc74 CM |
4269 | |
4270 | /* break if the extent we found is outside the range */ | |
4271 | if (em->start >= max || extent_map_end(em) < off) | |
4272 | break; | |
4273 | ||
4274 | /* | |
4275 | * get_extent may return an extent that starts before our | |
4276 | * requested range. We have to make sure the ranges | |
4277 | * we return to fiemap always move forward and don't | |
4278 | * overlap, so adjust the offsets here | |
4279 | */ | |
4280 | em_start = max(em->start, off); | |
1506fcc8 | 4281 | |
ea8efc74 CM |
4282 | /* |
4283 | * record the offset from the start of the extent | |
b76bb701 JB |
4284 | * for adjusting the disk offset below. Only do this if the |
4285 | * extent isn't compressed since our in ram offset may be past | |
4286 | * what we have actually allocated on disk. | |
ea8efc74 | 4287 | */ |
b76bb701 JB |
4288 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
4289 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 4290 | em_end = extent_map_end(em); |
ea8efc74 | 4291 | em_len = em_end - em_start; |
1506fcc8 YS |
4292 | disko = 0; |
4293 | flags = 0; | |
4294 | ||
ea8efc74 CM |
4295 | /* |
4296 | * bump off for our next call to get_extent | |
4297 | */ | |
4298 | off = extent_map_end(em); | |
4299 | if (off >= max) | |
4300 | end = 1; | |
4301 | ||
93dbfad7 | 4302 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
4303 | end = 1; |
4304 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 4305 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
4306 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
4307 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 4308 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
4309 | flags |= (FIEMAP_EXTENT_DELALLOC | |
4310 | FIEMAP_EXTENT_UNKNOWN); | |
dc046b10 JB |
4311 | } else if (fieinfo->fi_extents_max) { |
4312 | u64 bytenr = em->block_start - | |
4313 | (em->start - em->orig_start); | |
fe09e16c | 4314 | |
ea8efc74 | 4315 | disko = em->block_start + offset_in_extent; |
fe09e16c LB |
4316 | |
4317 | /* | |
4318 | * As btrfs supports shared space, this information | |
4319 | * can be exported to userspace tools via | |
dc046b10 JB |
4320 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 |
4321 | * then we're just getting a count and we can skip the | |
4322 | * lookup stuff. | |
fe09e16c | 4323 | */ |
dc046b10 JB |
4324 | ret = btrfs_check_shared(NULL, root->fs_info, |
4325 | root->objectid, | |
4326 | btrfs_ino(inode), bytenr); | |
4327 | if (ret < 0) | |
fe09e16c | 4328 | goto out_free; |
dc046b10 | 4329 | if (ret) |
fe09e16c | 4330 | flags |= FIEMAP_EXTENT_SHARED; |
dc046b10 | 4331 | ret = 0; |
1506fcc8 YS |
4332 | } |
4333 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
4334 | flags |= FIEMAP_EXTENT_ENCODED; | |
4335 | ||
1506fcc8 YS |
4336 | free_extent_map(em); |
4337 | em = NULL; | |
ec29ed5b CM |
4338 | if ((em_start >= last) || em_len == (u64)-1 || |
4339 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
4340 | flags |= FIEMAP_EXTENT_LAST; |
4341 | end = 1; | |
4342 | } | |
4343 | ||
ec29ed5b CM |
4344 | /* now scan forward to see if this is really the last extent. */ |
4345 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
4346 | get_extent); | |
4347 | if (IS_ERR(em)) { | |
4348 | ret = PTR_ERR(em); | |
4349 | goto out; | |
4350 | } | |
4351 | if (!em) { | |
975f84fe JB |
4352 | flags |= FIEMAP_EXTENT_LAST; |
4353 | end = 1; | |
4354 | } | |
ec29ed5b CM |
4355 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, |
4356 | em_len, flags); | |
4357 | if (ret) | |
4358 | goto out_free; | |
1506fcc8 YS |
4359 | } |
4360 | out_free: | |
4361 | free_extent_map(em); | |
4362 | out: | |
fe09e16c | 4363 | btrfs_free_path(path); |
a52f4cd2 | 4364 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
2ac55d41 | 4365 | &cached_state, GFP_NOFS); |
1506fcc8 YS |
4366 | return ret; |
4367 | } | |
4368 | ||
727011e0 CM |
4369 | static void __free_extent_buffer(struct extent_buffer *eb) |
4370 | { | |
6d49ba1b | 4371 | btrfs_leak_debug_del(&eb->leak_list); |
727011e0 CM |
4372 | kmem_cache_free(extent_buffer_cache, eb); |
4373 | } | |
4374 | ||
a26e8c9f | 4375 | int extent_buffer_under_io(struct extent_buffer *eb) |
db7f3436 JB |
4376 | { |
4377 | return (atomic_read(&eb->io_pages) || | |
4378 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4379 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4380 | } | |
4381 | ||
4382 | /* | |
4383 | * Helper for releasing extent buffer page. | |
4384 | */ | |
4385 | static void btrfs_release_extent_buffer_page(struct extent_buffer *eb, | |
4386 | unsigned long start_idx) | |
4387 | { | |
4388 | unsigned long index; | |
4389 | unsigned long num_pages; | |
4390 | struct page *page; | |
4391 | int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4392 | ||
4393 | BUG_ON(extent_buffer_under_io(eb)); | |
4394 | ||
4395 | num_pages = num_extent_pages(eb->start, eb->len); | |
4396 | index = start_idx + num_pages; | |
4397 | if (start_idx >= index) | |
4398 | return; | |
4399 | ||
4400 | do { | |
4401 | index--; | |
4402 | page = extent_buffer_page(eb, index); | |
4403 | if (page && mapped) { | |
4404 | spin_lock(&page->mapping->private_lock); | |
4405 | /* | |
4406 | * We do this since we'll remove the pages after we've | |
4407 | * removed the eb from the radix tree, so we could race | |
4408 | * and have this page now attached to the new eb. So | |
4409 | * only clear page_private if it's still connected to | |
4410 | * this eb. | |
4411 | */ | |
4412 | if (PagePrivate(page) && | |
4413 | page->private == (unsigned long)eb) { | |
4414 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4415 | BUG_ON(PageDirty(page)); | |
4416 | BUG_ON(PageWriteback(page)); | |
4417 | /* | |
4418 | * We need to make sure we haven't be attached | |
4419 | * to a new eb. | |
4420 | */ | |
4421 | ClearPagePrivate(page); | |
4422 | set_page_private(page, 0); | |
4423 | /* One for the page private */ | |
4424 | page_cache_release(page); | |
4425 | } | |
4426 | spin_unlock(&page->mapping->private_lock); | |
4427 | ||
4428 | } | |
4429 | if (page) { | |
4430 | /* One for when we alloced the page */ | |
4431 | page_cache_release(page); | |
4432 | } | |
4433 | } while (index != start_idx); | |
4434 | } | |
4435 | ||
4436 | /* | |
4437 | * Helper for releasing the extent buffer. | |
4438 | */ | |
4439 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4440 | { | |
4441 | btrfs_release_extent_buffer_page(eb, 0); | |
4442 | __free_extent_buffer(eb); | |
4443 | } | |
4444 | ||
f28491e0 JB |
4445 | static struct extent_buffer * |
4446 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
4447 | unsigned long len, gfp_t mask) | |
d1310b2e CM |
4448 | { |
4449 | struct extent_buffer *eb = NULL; | |
4450 | ||
d1310b2e | 4451 | eb = kmem_cache_zalloc(extent_buffer_cache, mask); |
91ca338d TI |
4452 | if (eb == NULL) |
4453 | return NULL; | |
d1310b2e CM |
4454 | eb->start = start; |
4455 | eb->len = len; | |
f28491e0 | 4456 | eb->fs_info = fs_info; |
815a51c7 | 4457 | eb->bflags = 0; |
bd681513 CM |
4458 | rwlock_init(&eb->lock); |
4459 | atomic_set(&eb->write_locks, 0); | |
4460 | atomic_set(&eb->read_locks, 0); | |
4461 | atomic_set(&eb->blocking_readers, 0); | |
4462 | atomic_set(&eb->blocking_writers, 0); | |
4463 | atomic_set(&eb->spinning_readers, 0); | |
4464 | atomic_set(&eb->spinning_writers, 0); | |
5b25f70f | 4465 | eb->lock_nested = 0; |
bd681513 CM |
4466 | init_waitqueue_head(&eb->write_lock_wq); |
4467 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 4468 | |
6d49ba1b ES |
4469 | btrfs_leak_debug_add(&eb->leak_list, &buffers); |
4470 | ||
3083ee2e | 4471 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 4472 | atomic_set(&eb->refs, 1); |
0b32f4bb | 4473 | atomic_set(&eb->io_pages, 0); |
727011e0 | 4474 | |
b8dae313 DS |
4475 | /* |
4476 | * Sanity checks, currently the maximum is 64k covered by 16x 4k pages | |
4477 | */ | |
4478 | BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE | |
4479 | > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
4480 | BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
d1310b2e CM |
4481 | |
4482 | return eb; | |
4483 | } | |
4484 | ||
815a51c7 JS |
4485 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) |
4486 | { | |
4487 | unsigned long i; | |
4488 | struct page *p; | |
4489 | struct extent_buffer *new; | |
4490 | unsigned long num_pages = num_extent_pages(src->start, src->len); | |
4491 | ||
9ec72677 | 4492 | new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_NOFS); |
815a51c7 JS |
4493 | if (new == NULL) |
4494 | return NULL; | |
4495 | ||
4496 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4497 | p = alloc_page(GFP_NOFS); |
db7f3436 JB |
4498 | if (!p) { |
4499 | btrfs_release_extent_buffer(new); | |
4500 | return NULL; | |
4501 | } | |
815a51c7 JS |
4502 | attach_extent_buffer_page(new, p); |
4503 | WARN_ON(PageDirty(p)); | |
4504 | SetPageUptodate(p); | |
4505 | new->pages[i] = p; | |
4506 | } | |
4507 | ||
4508 | copy_extent_buffer(new, src, 0, 0, src->len); | |
4509 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); | |
4510 | set_bit(EXTENT_BUFFER_DUMMY, &new->bflags); | |
4511 | ||
4512 | return new; | |
4513 | } | |
4514 | ||
4515 | struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len) | |
4516 | { | |
4517 | struct extent_buffer *eb; | |
4518 | unsigned long num_pages = num_extent_pages(0, len); | |
4519 | unsigned long i; | |
4520 | ||
9ec72677 | 4521 | eb = __alloc_extent_buffer(NULL, start, len, GFP_NOFS); |
815a51c7 JS |
4522 | if (!eb) |
4523 | return NULL; | |
4524 | ||
4525 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4526 | eb->pages[i] = alloc_page(GFP_NOFS); |
815a51c7 JS |
4527 | if (!eb->pages[i]) |
4528 | goto err; | |
4529 | } | |
4530 | set_extent_buffer_uptodate(eb); | |
4531 | btrfs_set_header_nritems(eb, 0); | |
4532 | set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4533 | ||
4534 | return eb; | |
4535 | err: | |
84167d19 SB |
4536 | for (; i > 0; i--) |
4537 | __free_page(eb->pages[i - 1]); | |
815a51c7 JS |
4538 | __free_extent_buffer(eb); |
4539 | return NULL; | |
4540 | } | |
4541 | ||
0b32f4bb JB |
4542 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
4543 | { | |
242e18c7 | 4544 | int refs; |
0b32f4bb JB |
4545 | /* the ref bit is tricky. We have to make sure it is set |
4546 | * if we have the buffer dirty. Otherwise the | |
4547 | * code to free a buffer can end up dropping a dirty | |
4548 | * page | |
4549 | * | |
4550 | * Once the ref bit is set, it won't go away while the | |
4551 | * buffer is dirty or in writeback, and it also won't | |
4552 | * go away while we have the reference count on the | |
4553 | * eb bumped. | |
4554 | * | |
4555 | * We can't just set the ref bit without bumping the | |
4556 | * ref on the eb because free_extent_buffer might | |
4557 | * see the ref bit and try to clear it. If this happens | |
4558 | * free_extent_buffer might end up dropping our original | |
4559 | * ref by mistake and freeing the page before we are able | |
4560 | * to add one more ref. | |
4561 | * | |
4562 | * So bump the ref count first, then set the bit. If someone | |
4563 | * beat us to it, drop the ref we added. | |
4564 | */ | |
242e18c7 CM |
4565 | refs = atomic_read(&eb->refs); |
4566 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
4567 | return; | |
4568 | ||
594831c4 JB |
4569 | spin_lock(&eb->refs_lock); |
4570 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 4571 | atomic_inc(&eb->refs); |
594831c4 | 4572 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
4573 | } |
4574 | ||
2457aec6 MG |
4575 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
4576 | struct page *accessed) | |
5df4235e JB |
4577 | { |
4578 | unsigned long num_pages, i; | |
4579 | ||
0b32f4bb JB |
4580 | check_buffer_tree_ref(eb); |
4581 | ||
5df4235e JB |
4582 | num_pages = num_extent_pages(eb->start, eb->len); |
4583 | for (i = 0; i < num_pages; i++) { | |
4584 | struct page *p = extent_buffer_page(eb, i); | |
2457aec6 MG |
4585 | if (p != accessed) |
4586 | mark_page_accessed(p); | |
5df4235e JB |
4587 | } |
4588 | } | |
4589 | ||
f28491e0 JB |
4590 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
4591 | u64 start) | |
452c75c3 CS |
4592 | { |
4593 | struct extent_buffer *eb; | |
4594 | ||
4595 | rcu_read_lock(); | |
f28491e0 JB |
4596 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
4597 | start >> PAGE_CACHE_SHIFT); | |
452c75c3 CS |
4598 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
4599 | rcu_read_unlock(); | |
2457aec6 | 4600 | mark_extent_buffer_accessed(eb, NULL); |
452c75c3 CS |
4601 | return eb; |
4602 | } | |
4603 | rcu_read_unlock(); | |
4604 | ||
4605 | return NULL; | |
4606 | } | |
4607 | ||
faa2dbf0 JB |
4608 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
4609 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
4610 | u64 start, unsigned long len) | |
4611 | { | |
4612 | struct extent_buffer *eb, *exists = NULL; | |
4613 | int ret; | |
4614 | ||
4615 | eb = find_extent_buffer(fs_info, start); | |
4616 | if (eb) | |
4617 | return eb; | |
4618 | eb = alloc_dummy_extent_buffer(start, len); | |
4619 | if (!eb) | |
4620 | return NULL; | |
4621 | eb->fs_info = fs_info; | |
4622 | again: | |
4623 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); | |
4624 | if (ret) | |
4625 | goto free_eb; | |
4626 | spin_lock(&fs_info->buffer_lock); | |
4627 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
4628 | start >> PAGE_CACHE_SHIFT, eb); | |
4629 | spin_unlock(&fs_info->buffer_lock); | |
4630 | radix_tree_preload_end(); | |
4631 | if (ret == -EEXIST) { | |
4632 | exists = find_extent_buffer(fs_info, start); | |
4633 | if (exists) | |
4634 | goto free_eb; | |
4635 | else | |
4636 | goto again; | |
4637 | } | |
4638 | check_buffer_tree_ref(eb); | |
4639 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
4640 | ||
4641 | /* | |
4642 | * We will free dummy extent buffer's if they come into | |
4643 | * free_extent_buffer with a ref count of 2, but if we are using this we | |
4644 | * want the buffers to stay in memory until we're done with them, so | |
4645 | * bump the ref count again. | |
4646 | */ | |
4647 | atomic_inc(&eb->refs); | |
4648 | return eb; | |
4649 | free_eb: | |
4650 | btrfs_release_extent_buffer(eb); | |
4651 | return exists; | |
4652 | } | |
4653 | #endif | |
4654 | ||
f28491e0 | 4655 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
727011e0 | 4656 | u64 start, unsigned long len) |
d1310b2e CM |
4657 | { |
4658 | unsigned long num_pages = num_extent_pages(start, len); | |
4659 | unsigned long i; | |
4660 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
4661 | struct extent_buffer *eb; | |
6af118ce | 4662 | struct extent_buffer *exists = NULL; |
d1310b2e | 4663 | struct page *p; |
f28491e0 | 4664 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
d1310b2e | 4665 | int uptodate = 1; |
19fe0a8b | 4666 | int ret; |
d1310b2e | 4667 | |
f28491e0 | 4668 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 4669 | if (eb) |
6af118ce | 4670 | return eb; |
6af118ce | 4671 | |
f28491e0 | 4672 | eb = __alloc_extent_buffer(fs_info, start, len, GFP_NOFS); |
2b114d1d | 4673 | if (!eb) |
d1310b2e CM |
4674 | return NULL; |
4675 | ||
727011e0 | 4676 | for (i = 0; i < num_pages; i++, index++) { |
a6591715 | 4677 | p = find_or_create_page(mapping, index, GFP_NOFS); |
4804b382 | 4678 | if (!p) |
6af118ce | 4679 | goto free_eb; |
4f2de97a JB |
4680 | |
4681 | spin_lock(&mapping->private_lock); | |
4682 | if (PagePrivate(p)) { | |
4683 | /* | |
4684 | * We could have already allocated an eb for this page | |
4685 | * and attached one so lets see if we can get a ref on | |
4686 | * the existing eb, and if we can we know it's good and | |
4687 | * we can just return that one, else we know we can just | |
4688 | * overwrite page->private. | |
4689 | */ | |
4690 | exists = (struct extent_buffer *)p->private; | |
4691 | if (atomic_inc_not_zero(&exists->refs)) { | |
4692 | spin_unlock(&mapping->private_lock); | |
4693 | unlock_page(p); | |
17de39ac | 4694 | page_cache_release(p); |
2457aec6 | 4695 | mark_extent_buffer_accessed(exists, p); |
4f2de97a JB |
4696 | goto free_eb; |
4697 | } | |
4698 | ||
0b32f4bb | 4699 | /* |
4f2de97a JB |
4700 | * Do this so attach doesn't complain and we need to |
4701 | * drop the ref the old guy had. | |
4702 | */ | |
4703 | ClearPagePrivate(p); | |
0b32f4bb | 4704 | WARN_ON(PageDirty(p)); |
4f2de97a | 4705 | page_cache_release(p); |
d1310b2e | 4706 | } |
4f2de97a JB |
4707 | attach_extent_buffer_page(eb, p); |
4708 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 4709 | WARN_ON(PageDirty(p)); |
727011e0 | 4710 | eb->pages[i] = p; |
d1310b2e CM |
4711 | if (!PageUptodate(p)) |
4712 | uptodate = 0; | |
eb14ab8e CM |
4713 | |
4714 | /* | |
4715 | * see below about how we avoid a nasty race with release page | |
4716 | * and why we unlock later | |
4717 | */ | |
d1310b2e CM |
4718 | } |
4719 | if (uptodate) | |
b4ce94de | 4720 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 4721 | again: |
19fe0a8b MX |
4722 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); |
4723 | if (ret) | |
4724 | goto free_eb; | |
4725 | ||
f28491e0 JB |
4726 | spin_lock(&fs_info->buffer_lock); |
4727 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
4728 | start >> PAGE_CACHE_SHIFT, eb); | |
4729 | spin_unlock(&fs_info->buffer_lock); | |
452c75c3 | 4730 | radix_tree_preload_end(); |
19fe0a8b | 4731 | if (ret == -EEXIST) { |
f28491e0 | 4732 | exists = find_extent_buffer(fs_info, start); |
452c75c3 CS |
4733 | if (exists) |
4734 | goto free_eb; | |
4735 | else | |
115391d2 | 4736 | goto again; |
6af118ce | 4737 | } |
6af118ce | 4738 | /* add one reference for the tree */ |
0b32f4bb | 4739 | check_buffer_tree_ref(eb); |
34b41ace | 4740 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
4741 | |
4742 | /* | |
4743 | * there is a race where release page may have | |
4744 | * tried to find this extent buffer in the radix | |
4745 | * but failed. It will tell the VM it is safe to | |
4746 | * reclaim the, and it will clear the page private bit. | |
4747 | * We must make sure to set the page private bit properly | |
4748 | * after the extent buffer is in the radix tree so | |
4749 | * it doesn't get lost | |
4750 | */ | |
727011e0 CM |
4751 | SetPageChecked(eb->pages[0]); |
4752 | for (i = 1; i < num_pages; i++) { | |
4753 | p = extent_buffer_page(eb, i); | |
727011e0 CM |
4754 | ClearPageChecked(p); |
4755 | unlock_page(p); | |
4756 | } | |
4757 | unlock_page(eb->pages[0]); | |
d1310b2e CM |
4758 | return eb; |
4759 | ||
6af118ce | 4760 | free_eb: |
727011e0 CM |
4761 | for (i = 0; i < num_pages; i++) { |
4762 | if (eb->pages[i]) | |
4763 | unlock_page(eb->pages[i]); | |
4764 | } | |
eb14ab8e | 4765 | |
17de39ac | 4766 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
897ca6e9 | 4767 | btrfs_release_extent_buffer(eb); |
6af118ce | 4768 | return exists; |
d1310b2e | 4769 | } |
d1310b2e | 4770 | |
3083ee2e JB |
4771 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
4772 | { | |
4773 | struct extent_buffer *eb = | |
4774 | container_of(head, struct extent_buffer, rcu_head); | |
4775 | ||
4776 | __free_extent_buffer(eb); | |
4777 | } | |
4778 | ||
3083ee2e | 4779 | /* Expects to have eb->eb_lock already held */ |
f7a52a40 | 4780 | static int release_extent_buffer(struct extent_buffer *eb) |
3083ee2e JB |
4781 | { |
4782 | WARN_ON(atomic_read(&eb->refs) == 0); | |
4783 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 4784 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 4785 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 4786 | |
815a51c7 | 4787 | spin_unlock(&eb->refs_lock); |
3083ee2e | 4788 | |
f28491e0 JB |
4789 | spin_lock(&fs_info->buffer_lock); |
4790 | radix_tree_delete(&fs_info->buffer_radix, | |
815a51c7 | 4791 | eb->start >> PAGE_CACHE_SHIFT); |
f28491e0 | 4792 | spin_unlock(&fs_info->buffer_lock); |
34b41ace JB |
4793 | } else { |
4794 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 4795 | } |
3083ee2e JB |
4796 | |
4797 | /* Should be safe to release our pages at this point */ | |
4798 | btrfs_release_extent_buffer_page(eb, 0); | |
3083ee2e | 4799 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 4800 | return 1; |
3083ee2e JB |
4801 | } |
4802 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
4803 | |
4804 | return 0; | |
3083ee2e JB |
4805 | } |
4806 | ||
d1310b2e CM |
4807 | void free_extent_buffer(struct extent_buffer *eb) |
4808 | { | |
242e18c7 CM |
4809 | int refs; |
4810 | int old; | |
d1310b2e CM |
4811 | if (!eb) |
4812 | return; | |
4813 | ||
242e18c7 CM |
4814 | while (1) { |
4815 | refs = atomic_read(&eb->refs); | |
4816 | if (refs <= 3) | |
4817 | break; | |
4818 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
4819 | if (old == refs) | |
4820 | return; | |
4821 | } | |
4822 | ||
3083ee2e | 4823 | spin_lock(&eb->refs_lock); |
815a51c7 JS |
4824 | if (atomic_read(&eb->refs) == 2 && |
4825 | test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) | |
4826 | atomic_dec(&eb->refs); | |
4827 | ||
3083ee2e JB |
4828 | if (atomic_read(&eb->refs) == 2 && |
4829 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 4830 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
4831 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
4832 | atomic_dec(&eb->refs); | |
4833 | ||
4834 | /* | |
4835 | * I know this is terrible, but it's temporary until we stop tracking | |
4836 | * the uptodate bits and such for the extent buffers. | |
4837 | */ | |
f7a52a40 | 4838 | release_extent_buffer(eb); |
3083ee2e JB |
4839 | } |
4840 | ||
4841 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
4842 | { | |
4843 | if (!eb) | |
d1310b2e CM |
4844 | return; |
4845 | ||
3083ee2e JB |
4846 | spin_lock(&eb->refs_lock); |
4847 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
4848 | ||
0b32f4bb | 4849 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
4850 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
4851 | atomic_dec(&eb->refs); | |
f7a52a40 | 4852 | release_extent_buffer(eb); |
d1310b2e | 4853 | } |
d1310b2e | 4854 | |
1d4284bd | 4855 | void clear_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 4856 | { |
d1310b2e CM |
4857 | unsigned long i; |
4858 | unsigned long num_pages; | |
4859 | struct page *page; | |
4860 | ||
d1310b2e CM |
4861 | num_pages = num_extent_pages(eb->start, eb->len); |
4862 | ||
4863 | for (i = 0; i < num_pages; i++) { | |
4864 | page = extent_buffer_page(eb, i); | |
b9473439 | 4865 | if (!PageDirty(page)) |
d2c3f4f6 CM |
4866 | continue; |
4867 | ||
a61e6f29 | 4868 | lock_page(page); |
eb14ab8e CM |
4869 | WARN_ON(!PagePrivate(page)); |
4870 | ||
d1310b2e | 4871 | clear_page_dirty_for_io(page); |
0ee0fda0 | 4872 | spin_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
4873 | if (!PageDirty(page)) { |
4874 | radix_tree_tag_clear(&page->mapping->page_tree, | |
4875 | page_index(page), | |
4876 | PAGECACHE_TAG_DIRTY); | |
4877 | } | |
0ee0fda0 | 4878 | spin_unlock_irq(&page->mapping->tree_lock); |
bf0da8c1 | 4879 | ClearPageError(page); |
a61e6f29 | 4880 | unlock_page(page); |
d1310b2e | 4881 | } |
0b32f4bb | 4882 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 4883 | } |
d1310b2e | 4884 | |
0b32f4bb | 4885 | int set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e CM |
4886 | { |
4887 | unsigned long i; | |
4888 | unsigned long num_pages; | |
b9473439 | 4889 | int was_dirty = 0; |
d1310b2e | 4890 | |
0b32f4bb JB |
4891 | check_buffer_tree_ref(eb); |
4892 | ||
b9473439 | 4893 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 4894 | |
d1310b2e | 4895 | num_pages = num_extent_pages(eb->start, eb->len); |
3083ee2e | 4896 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
4897 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
4898 | ||
b9473439 | 4899 | for (i = 0; i < num_pages; i++) |
0b32f4bb | 4900 | set_page_dirty(extent_buffer_page(eb, i)); |
b9473439 | 4901 | return was_dirty; |
d1310b2e | 4902 | } |
d1310b2e | 4903 | |
0b32f4bb | 4904 | int clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 CM |
4905 | { |
4906 | unsigned long i; | |
4907 | struct page *page; | |
4908 | unsigned long num_pages; | |
4909 | ||
b4ce94de | 4910 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
0b32f4bb | 4911 | num_pages = num_extent_pages(eb->start, eb->len); |
1259ab75 CM |
4912 | for (i = 0; i < num_pages; i++) { |
4913 | page = extent_buffer_page(eb, i); | |
33958dc6 CM |
4914 | if (page) |
4915 | ClearPageUptodate(page); | |
1259ab75 CM |
4916 | } |
4917 | return 0; | |
4918 | } | |
4919 | ||
0b32f4bb | 4920 | int set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e CM |
4921 | { |
4922 | unsigned long i; | |
4923 | struct page *page; | |
4924 | unsigned long num_pages; | |
4925 | ||
0b32f4bb | 4926 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 4927 | num_pages = num_extent_pages(eb->start, eb->len); |
d1310b2e CM |
4928 | for (i = 0; i < num_pages; i++) { |
4929 | page = extent_buffer_page(eb, i); | |
d1310b2e CM |
4930 | SetPageUptodate(page); |
4931 | } | |
4932 | return 0; | |
4933 | } | |
d1310b2e | 4934 | |
0b32f4bb | 4935 | int extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 4936 | { |
0b32f4bb | 4937 | return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 4938 | } |
d1310b2e CM |
4939 | |
4940 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
bb82ab88 | 4941 | struct extent_buffer *eb, u64 start, int wait, |
f188591e | 4942 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
4943 | { |
4944 | unsigned long i; | |
4945 | unsigned long start_i; | |
4946 | struct page *page; | |
4947 | int err; | |
4948 | int ret = 0; | |
ce9adaa5 CM |
4949 | int locked_pages = 0; |
4950 | int all_uptodate = 1; | |
d1310b2e | 4951 | unsigned long num_pages; |
727011e0 | 4952 | unsigned long num_reads = 0; |
a86c12c7 | 4953 | struct bio *bio = NULL; |
c8b97818 | 4954 | unsigned long bio_flags = 0; |
a86c12c7 | 4955 | |
b4ce94de | 4956 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
4957 | return 0; |
4958 | ||
d1310b2e CM |
4959 | if (start) { |
4960 | WARN_ON(start < eb->start); | |
4961 | start_i = (start >> PAGE_CACHE_SHIFT) - | |
4962 | (eb->start >> PAGE_CACHE_SHIFT); | |
4963 | } else { | |
4964 | start_i = 0; | |
4965 | } | |
4966 | ||
4967 | num_pages = num_extent_pages(eb->start, eb->len); | |
4968 | for (i = start_i; i < num_pages; i++) { | |
4969 | page = extent_buffer_page(eb, i); | |
bb82ab88 | 4970 | if (wait == WAIT_NONE) { |
2db04966 | 4971 | if (!trylock_page(page)) |
ce9adaa5 | 4972 | goto unlock_exit; |
d1310b2e CM |
4973 | } else { |
4974 | lock_page(page); | |
4975 | } | |
ce9adaa5 | 4976 | locked_pages++; |
727011e0 CM |
4977 | if (!PageUptodate(page)) { |
4978 | num_reads++; | |
ce9adaa5 | 4979 | all_uptodate = 0; |
727011e0 | 4980 | } |
ce9adaa5 CM |
4981 | } |
4982 | if (all_uptodate) { | |
4983 | if (start_i == 0) | |
b4ce94de | 4984 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
4985 | goto unlock_exit; |
4986 | } | |
4987 | ||
ea466794 | 4988 | clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags); |
5cf1ab56 | 4989 | eb->read_mirror = 0; |
0b32f4bb | 4990 | atomic_set(&eb->io_pages, num_reads); |
ce9adaa5 CM |
4991 | for (i = start_i; i < num_pages; i++) { |
4992 | page = extent_buffer_page(eb, i); | |
ce9adaa5 | 4993 | if (!PageUptodate(page)) { |
f188591e | 4994 | ClearPageError(page); |
a86c12c7 | 4995 | err = __extent_read_full_page(tree, page, |
f188591e | 4996 | get_extent, &bio, |
d4c7ca86 JB |
4997 | mirror_num, &bio_flags, |
4998 | READ | REQ_META); | |
d397712b | 4999 | if (err) |
d1310b2e | 5000 | ret = err; |
d1310b2e CM |
5001 | } else { |
5002 | unlock_page(page); | |
5003 | } | |
5004 | } | |
5005 | ||
355808c2 | 5006 | if (bio) { |
d4c7ca86 JB |
5007 | err = submit_one_bio(READ | REQ_META, bio, mirror_num, |
5008 | bio_flags); | |
79787eaa JM |
5009 | if (err) |
5010 | return err; | |
355808c2 | 5011 | } |
a86c12c7 | 5012 | |
bb82ab88 | 5013 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 5014 | return ret; |
d397712b | 5015 | |
d1310b2e CM |
5016 | for (i = start_i; i < num_pages; i++) { |
5017 | page = extent_buffer_page(eb, i); | |
5018 | wait_on_page_locked(page); | |
d397712b | 5019 | if (!PageUptodate(page)) |
d1310b2e | 5020 | ret = -EIO; |
d1310b2e | 5021 | } |
d397712b | 5022 | |
d1310b2e | 5023 | return ret; |
ce9adaa5 CM |
5024 | |
5025 | unlock_exit: | |
5026 | i = start_i; | |
d397712b | 5027 | while (locked_pages > 0) { |
ce9adaa5 CM |
5028 | page = extent_buffer_page(eb, i); |
5029 | i++; | |
5030 | unlock_page(page); | |
5031 | locked_pages--; | |
5032 | } | |
5033 | return ret; | |
d1310b2e | 5034 | } |
d1310b2e CM |
5035 | |
5036 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
5037 | unsigned long start, | |
5038 | unsigned long len) | |
5039 | { | |
5040 | size_t cur; | |
5041 | size_t offset; | |
5042 | struct page *page; | |
5043 | char *kaddr; | |
5044 | char *dst = (char *)dstv; | |
5045 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5046 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
5047 | |
5048 | WARN_ON(start > eb->len); | |
5049 | WARN_ON(start + len > eb->start + eb->len); | |
5050 | ||
778746b5 | 5051 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5052 | |
d397712b | 5053 | while (len > 0) { |
d1310b2e | 5054 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
5055 | |
5056 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
a6591715 | 5057 | kaddr = page_address(page); |
d1310b2e | 5058 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
5059 | |
5060 | dst += cur; | |
5061 | len -= cur; | |
5062 | offset = 0; | |
5063 | i++; | |
5064 | } | |
5065 | } | |
d1310b2e | 5066 | |
550ac1d8 GH |
5067 | int read_extent_buffer_to_user(struct extent_buffer *eb, void __user *dstv, |
5068 | unsigned long start, | |
5069 | unsigned long len) | |
5070 | { | |
5071 | size_t cur; | |
5072 | size_t offset; | |
5073 | struct page *page; | |
5074 | char *kaddr; | |
5075 | char __user *dst = (char __user *)dstv; | |
5076 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5077 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5078 | int ret = 0; | |
5079 | ||
5080 | WARN_ON(start > eb->len); | |
5081 | WARN_ON(start + len > eb->start + eb->len); | |
5082 | ||
5083 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); | |
5084 | ||
5085 | while (len > 0) { | |
5086 | page = extent_buffer_page(eb, i); | |
5087 | ||
5088 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
5089 | kaddr = page_address(page); | |
5090 | if (copy_to_user(dst, kaddr + offset, cur)) { | |
5091 | ret = -EFAULT; | |
5092 | break; | |
5093 | } | |
5094 | ||
5095 | dst += cur; | |
5096 | len -= cur; | |
5097 | offset = 0; | |
5098 | i++; | |
5099 | } | |
5100 | ||
5101 | return ret; | |
5102 | } | |
5103 | ||
d1310b2e | 5104 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, |
a6591715 | 5105 | unsigned long min_len, char **map, |
d1310b2e | 5106 | unsigned long *map_start, |
a6591715 | 5107 | unsigned long *map_len) |
d1310b2e CM |
5108 | { |
5109 | size_t offset = start & (PAGE_CACHE_SIZE - 1); | |
5110 | char *kaddr; | |
5111 | struct page *p; | |
5112 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5113 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5114 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
5115 | PAGE_CACHE_SHIFT; | |
5116 | ||
5117 | if (i != end_i) | |
5118 | return -EINVAL; | |
5119 | ||
5120 | if (i == 0) { | |
5121 | offset = start_offset; | |
5122 | *map_start = 0; | |
5123 | } else { | |
5124 | offset = 0; | |
5125 | *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; | |
5126 | } | |
d397712b | 5127 | |
d1310b2e | 5128 | if (start + min_len > eb->len) { |
31b1a2bd | 5129 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, " |
c1c9ff7c GU |
5130 | "wanted %lu %lu\n", |
5131 | eb->start, eb->len, start, min_len); | |
85026533 | 5132 | return -EINVAL; |
d1310b2e CM |
5133 | } |
5134 | ||
5135 | p = extent_buffer_page(eb, i); | |
a6591715 | 5136 | kaddr = page_address(p); |
d1310b2e CM |
5137 | *map = kaddr + offset; |
5138 | *map_len = PAGE_CACHE_SIZE - offset; | |
5139 | return 0; | |
5140 | } | |
d1310b2e | 5141 | |
d1310b2e CM |
5142 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, |
5143 | unsigned long start, | |
5144 | unsigned long len) | |
5145 | { | |
5146 | size_t cur; | |
5147 | size_t offset; | |
5148 | struct page *page; | |
5149 | char *kaddr; | |
5150 | char *ptr = (char *)ptrv; | |
5151 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5152 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5153 | int ret = 0; | |
5154 | ||
5155 | WARN_ON(start > eb->len); | |
5156 | WARN_ON(start + len > eb->start + eb->len); | |
5157 | ||
778746b5 | 5158 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5159 | |
d397712b | 5160 | while (len > 0) { |
d1310b2e | 5161 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
5162 | |
5163 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
5164 | ||
a6591715 | 5165 | kaddr = page_address(page); |
d1310b2e | 5166 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
5167 | if (ret) |
5168 | break; | |
5169 | ||
5170 | ptr += cur; | |
5171 | len -= cur; | |
5172 | offset = 0; | |
5173 | i++; | |
5174 | } | |
5175 | return ret; | |
5176 | } | |
d1310b2e CM |
5177 | |
5178 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
5179 | unsigned long start, unsigned long len) | |
5180 | { | |
5181 | size_t cur; | |
5182 | size_t offset; | |
5183 | struct page *page; | |
5184 | char *kaddr; | |
5185 | char *src = (char *)srcv; | |
5186 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5187 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5188 | ||
5189 | WARN_ON(start > eb->len); | |
5190 | WARN_ON(start + len > eb->start + eb->len); | |
5191 | ||
778746b5 | 5192 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5193 | |
d397712b | 5194 | while (len > 0) { |
d1310b2e CM |
5195 | page = extent_buffer_page(eb, i); |
5196 | WARN_ON(!PageUptodate(page)); | |
5197 | ||
5198 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 5199 | kaddr = page_address(page); |
d1310b2e | 5200 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
5201 | |
5202 | src += cur; | |
5203 | len -= cur; | |
5204 | offset = 0; | |
5205 | i++; | |
5206 | } | |
5207 | } | |
d1310b2e CM |
5208 | |
5209 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
5210 | unsigned long start, unsigned long len) | |
5211 | { | |
5212 | size_t cur; | |
5213 | size_t offset; | |
5214 | struct page *page; | |
5215 | char *kaddr; | |
5216 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5217 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5218 | ||
5219 | WARN_ON(start > eb->len); | |
5220 | WARN_ON(start + len > eb->start + eb->len); | |
5221 | ||
778746b5 | 5222 | offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5223 | |
d397712b | 5224 | while (len > 0) { |
d1310b2e CM |
5225 | page = extent_buffer_page(eb, i); |
5226 | WARN_ON(!PageUptodate(page)); | |
5227 | ||
5228 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 5229 | kaddr = page_address(page); |
d1310b2e | 5230 | memset(kaddr + offset, c, cur); |
d1310b2e CM |
5231 | |
5232 | len -= cur; | |
5233 | offset = 0; | |
5234 | i++; | |
5235 | } | |
5236 | } | |
d1310b2e CM |
5237 | |
5238 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
5239 | unsigned long dst_offset, unsigned long src_offset, | |
5240 | unsigned long len) | |
5241 | { | |
5242 | u64 dst_len = dst->len; | |
5243 | size_t cur; | |
5244 | size_t offset; | |
5245 | struct page *page; | |
5246 | char *kaddr; | |
5247 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5248 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
5249 | ||
5250 | WARN_ON(src->len != dst_len); | |
5251 | ||
5252 | offset = (start_offset + dst_offset) & | |
778746b5 | 5253 | (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5254 | |
d397712b | 5255 | while (len > 0) { |
d1310b2e CM |
5256 | page = extent_buffer_page(dst, i); |
5257 | WARN_ON(!PageUptodate(page)); | |
5258 | ||
5259 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
5260 | ||
a6591715 | 5261 | kaddr = page_address(page); |
d1310b2e | 5262 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
5263 | |
5264 | src_offset += cur; | |
5265 | len -= cur; | |
5266 | offset = 0; | |
5267 | i++; | |
5268 | } | |
5269 | } | |
d1310b2e | 5270 | |
3387206f ST |
5271 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
5272 | { | |
5273 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
5274 | return distance < len; | |
5275 | } | |
5276 | ||
d1310b2e CM |
5277 | static void copy_pages(struct page *dst_page, struct page *src_page, |
5278 | unsigned long dst_off, unsigned long src_off, | |
5279 | unsigned long len) | |
5280 | { | |
a6591715 | 5281 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 5282 | char *src_kaddr; |
727011e0 | 5283 | int must_memmove = 0; |
d1310b2e | 5284 | |
3387206f | 5285 | if (dst_page != src_page) { |
a6591715 | 5286 | src_kaddr = page_address(src_page); |
3387206f | 5287 | } else { |
d1310b2e | 5288 | src_kaddr = dst_kaddr; |
727011e0 CM |
5289 | if (areas_overlap(src_off, dst_off, len)) |
5290 | must_memmove = 1; | |
3387206f | 5291 | } |
d1310b2e | 5292 | |
727011e0 CM |
5293 | if (must_memmove) |
5294 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
5295 | else | |
5296 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
5297 | } |
5298 | ||
5299 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5300 | unsigned long src_offset, unsigned long len) | |
5301 | { | |
5302 | size_t cur; | |
5303 | size_t dst_off_in_page; | |
5304 | size_t src_off_in_page; | |
5305 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5306 | unsigned long dst_i; | |
5307 | unsigned long src_i; | |
5308 | ||
5309 | if (src_offset + len > dst->len) { | |
efe120a0 | 5310 | printk(KERN_ERR "BTRFS: memmove bogus src_offset %lu move " |
d397712b | 5311 | "len %lu dst len %lu\n", src_offset, len, dst->len); |
d1310b2e CM |
5312 | BUG_ON(1); |
5313 | } | |
5314 | if (dst_offset + len > dst->len) { | |
efe120a0 | 5315 | printk(KERN_ERR "BTRFS: memmove bogus dst_offset %lu move " |
d397712b | 5316 | "len %lu dst len %lu\n", dst_offset, len, dst->len); |
d1310b2e CM |
5317 | BUG_ON(1); |
5318 | } | |
5319 | ||
d397712b | 5320 | while (len > 0) { |
d1310b2e | 5321 | dst_off_in_page = (start_offset + dst_offset) & |
778746b5 | 5322 | (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5323 | src_off_in_page = (start_offset + src_offset) & |
778746b5 | 5324 | (PAGE_CACHE_SIZE - 1); |
d1310b2e CM |
5325 | |
5326 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
5327 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
5328 | ||
5329 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
5330 | src_off_in_page)); | |
5331 | cur = min_t(unsigned long, cur, | |
5332 | (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); | |
5333 | ||
5334 | copy_pages(extent_buffer_page(dst, dst_i), | |
5335 | extent_buffer_page(dst, src_i), | |
5336 | dst_off_in_page, src_off_in_page, cur); | |
5337 | ||
5338 | src_offset += cur; | |
5339 | dst_offset += cur; | |
5340 | len -= cur; | |
5341 | } | |
5342 | } | |
d1310b2e CM |
5343 | |
5344 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5345 | unsigned long src_offset, unsigned long len) | |
5346 | { | |
5347 | size_t cur; | |
5348 | size_t dst_off_in_page; | |
5349 | size_t src_off_in_page; | |
5350 | unsigned long dst_end = dst_offset + len - 1; | |
5351 | unsigned long src_end = src_offset + len - 1; | |
5352 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5353 | unsigned long dst_i; | |
5354 | unsigned long src_i; | |
5355 | ||
5356 | if (src_offset + len > dst->len) { | |
efe120a0 | 5357 | printk(KERN_ERR "BTRFS: memmove bogus src_offset %lu move " |
d397712b | 5358 | "len %lu len %lu\n", src_offset, len, dst->len); |
d1310b2e CM |
5359 | BUG_ON(1); |
5360 | } | |
5361 | if (dst_offset + len > dst->len) { | |
efe120a0 | 5362 | printk(KERN_ERR "BTRFS: memmove bogus dst_offset %lu move " |
d397712b | 5363 | "len %lu len %lu\n", dst_offset, len, dst->len); |
d1310b2e CM |
5364 | BUG_ON(1); |
5365 | } | |
727011e0 | 5366 | if (dst_offset < src_offset) { |
d1310b2e CM |
5367 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
5368 | return; | |
5369 | } | |
d397712b | 5370 | while (len > 0) { |
d1310b2e CM |
5371 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; |
5372 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
5373 | ||
5374 | dst_off_in_page = (start_offset + dst_end) & | |
778746b5 | 5375 | (PAGE_CACHE_SIZE - 1); |
d1310b2e | 5376 | src_off_in_page = (start_offset + src_end) & |
778746b5 | 5377 | (PAGE_CACHE_SIZE - 1); |
d1310b2e CM |
5378 | |
5379 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
5380 | cur = min(cur, dst_off_in_page + 1); | |
1877e1a7 | 5381 | copy_pages(extent_buffer_page(dst, dst_i), |
d1310b2e CM |
5382 | extent_buffer_page(dst, src_i), |
5383 | dst_off_in_page - cur + 1, | |
5384 | src_off_in_page - cur + 1, cur); | |
5385 | ||
5386 | dst_end -= cur; | |
5387 | src_end -= cur; | |
5388 | len -= cur; | |
5389 | } | |
5390 | } | |
6af118ce | 5391 | |
f7a52a40 | 5392 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 5393 | { |
6af118ce | 5394 | struct extent_buffer *eb; |
6af118ce | 5395 | |
3083ee2e JB |
5396 | /* |
5397 | * We need to make sure noboody is attaching this page to an eb right | |
5398 | * now. | |
5399 | */ | |
5400 | spin_lock(&page->mapping->private_lock); | |
5401 | if (!PagePrivate(page)) { | |
5402 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 5403 | return 1; |
45f49bce | 5404 | } |
6af118ce | 5405 | |
3083ee2e JB |
5406 | eb = (struct extent_buffer *)page->private; |
5407 | BUG_ON(!eb); | |
19fe0a8b MX |
5408 | |
5409 | /* | |
3083ee2e JB |
5410 | * This is a little awful but should be ok, we need to make sure that |
5411 | * the eb doesn't disappear out from under us while we're looking at | |
5412 | * this page. | |
19fe0a8b | 5413 | */ |
3083ee2e | 5414 | spin_lock(&eb->refs_lock); |
0b32f4bb | 5415 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
5416 | spin_unlock(&eb->refs_lock); |
5417 | spin_unlock(&page->mapping->private_lock); | |
5418 | return 0; | |
b9473439 | 5419 | } |
3083ee2e | 5420 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 5421 | |
19fe0a8b | 5422 | /* |
3083ee2e JB |
5423 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
5424 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 5425 | */ |
3083ee2e JB |
5426 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
5427 | spin_unlock(&eb->refs_lock); | |
5428 | return 0; | |
b9473439 | 5429 | } |
19fe0a8b | 5430 | |
f7a52a40 | 5431 | return release_extent_buffer(eb); |
6af118ce | 5432 | } |