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" | |
2db04966 | 16 | #include "compat.h" |
902b22f3 DW |
17 | #include "ctree.h" |
18 | #include "btrfs_inode.h" | |
4a54c8c1 | 19 | #include "volumes.h" |
21adbd5c | 20 | #include "check-integrity.h" |
0b32f4bb | 21 | #include "locking.h" |
606686ee | 22 | #include "rcu-string.h" |
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 | |
6d49ba1b | 28 | #ifdef CONFIG_BTRFS_DEBUG |
d1310b2e CM |
29 | static LIST_HEAD(buffers); |
30 | static LIST_HEAD(states); | |
4bef0848 | 31 | |
d397712b | 32 | static DEFINE_SPINLOCK(leak_lock); |
6d49ba1b ES |
33 | |
34 | static inline | |
35 | void btrfs_leak_debug_add(struct list_head *new, struct list_head *head) | |
36 | { | |
37 | unsigned long flags; | |
38 | ||
39 | spin_lock_irqsave(&leak_lock, flags); | |
40 | list_add(new, head); | |
41 | spin_unlock_irqrestore(&leak_lock, flags); | |
42 | } | |
43 | ||
44 | static inline | |
45 | void btrfs_leak_debug_del(struct list_head *entry) | |
46 | { | |
47 | unsigned long flags; | |
48 | ||
49 | spin_lock_irqsave(&leak_lock, flags); | |
50 | list_del(entry); | |
51 | spin_unlock_irqrestore(&leak_lock, flags); | |
52 | } | |
53 | ||
54 | static inline | |
55 | void btrfs_leak_debug_check(void) | |
56 | { | |
57 | struct extent_state *state; | |
58 | struct extent_buffer *eb; | |
59 | ||
60 | while (!list_empty(&states)) { | |
61 | state = list_entry(states.next, struct extent_state, leak_list); | |
62 | printk(KERN_ERR "btrfs state leak: start %llu end %llu " | |
63 | "state %lu in tree %p refs %d\n", | |
64 | (unsigned long long)state->start, | |
65 | (unsigned long long)state->end, | |
66 | state->state, state->tree, atomic_read(&state->refs)); | |
67 | list_del(&state->leak_list); | |
68 | kmem_cache_free(extent_state_cache, state); | |
69 | } | |
70 | ||
71 | while (!list_empty(&buffers)) { | |
72 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
73 | printk(KERN_ERR "btrfs buffer leak start %llu len %lu " | |
74 | "refs %d\n", (unsigned long long)eb->start, | |
75 | eb->len, atomic_read(&eb->refs)); | |
76 | list_del(&eb->leak_list); | |
77 | kmem_cache_free(extent_buffer_cache, eb); | |
78 | } | |
79 | } | |
8d599ae1 DS |
80 | |
81 | #define btrfs_debug_check_extent_io_range(inode, start, end) \ | |
82 | __btrfs_debug_check_extent_io_range(__func__, (inode), (start), (end)) | |
83 | static inline void __btrfs_debug_check_extent_io_range(const char *caller, | |
84 | struct inode *inode, u64 start, u64 end) | |
85 | { | |
86 | u64 isize = i_size_read(inode); | |
87 | ||
88 | if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) { | |
89 | printk_ratelimited(KERN_DEBUG | |
90 | "btrfs: %s: ino %llu isize %llu odd range [%llu,%llu]\n", | |
91 | caller, | |
92 | (unsigned long long)btrfs_ino(inode), | |
93 | (unsigned long long)isize, | |
94 | (unsigned long long)start, | |
95 | (unsigned long long)end); | |
96 | } | |
97 | } | |
6d49ba1b ES |
98 | #else |
99 | #define btrfs_leak_debug_add(new, head) do {} while (0) | |
100 | #define btrfs_leak_debug_del(entry) do {} while (0) | |
101 | #define btrfs_leak_debug_check() do {} while (0) | |
8d599ae1 | 102 | #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0) |
4bef0848 | 103 | #endif |
d1310b2e | 104 | |
d1310b2e CM |
105 | #define BUFFER_LRU_MAX 64 |
106 | ||
107 | struct tree_entry { | |
108 | u64 start; | |
109 | u64 end; | |
d1310b2e CM |
110 | struct rb_node rb_node; |
111 | }; | |
112 | ||
113 | struct extent_page_data { | |
114 | struct bio *bio; | |
115 | struct extent_io_tree *tree; | |
116 | get_extent_t *get_extent; | |
de0022b9 | 117 | unsigned long bio_flags; |
771ed689 CM |
118 | |
119 | /* tells writepage not to lock the state bits for this range | |
120 | * it still does the unlocking | |
121 | */ | |
ffbd517d CM |
122 | unsigned int extent_locked:1; |
123 | ||
124 | /* tells the submit_bio code to use a WRITE_SYNC */ | |
125 | unsigned int sync_io:1; | |
d1310b2e CM |
126 | }; |
127 | ||
0b32f4bb | 128 | static noinline void flush_write_bio(void *data); |
c2d904e0 JM |
129 | static inline struct btrfs_fs_info * |
130 | tree_fs_info(struct extent_io_tree *tree) | |
131 | { | |
132 | return btrfs_sb(tree->mapping->host->i_sb); | |
133 | } | |
0b32f4bb | 134 | |
d1310b2e CM |
135 | int __init extent_io_init(void) |
136 | { | |
837e1972 | 137 | extent_state_cache = kmem_cache_create("btrfs_extent_state", |
9601e3f6 CH |
138 | sizeof(struct extent_state), 0, |
139 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
140 | if (!extent_state_cache) |
141 | return -ENOMEM; | |
142 | ||
837e1972 | 143 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 CH |
144 | sizeof(struct extent_buffer), 0, |
145 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
146 | if (!extent_buffer_cache) |
147 | goto free_state_cache; | |
9be3395b CM |
148 | |
149 | btrfs_bioset = bioset_create(BIO_POOL_SIZE, | |
150 | offsetof(struct btrfs_io_bio, bio)); | |
151 | if (!btrfs_bioset) | |
152 | goto free_buffer_cache; | |
d1310b2e CM |
153 | return 0; |
154 | ||
9be3395b CM |
155 | free_buffer_cache: |
156 | kmem_cache_destroy(extent_buffer_cache); | |
157 | extent_buffer_cache = NULL; | |
158 | ||
d1310b2e CM |
159 | free_state_cache: |
160 | kmem_cache_destroy(extent_state_cache); | |
9be3395b | 161 | extent_state_cache = NULL; |
d1310b2e CM |
162 | return -ENOMEM; |
163 | } | |
164 | ||
165 | void extent_io_exit(void) | |
166 | { | |
6d49ba1b | 167 | btrfs_leak_debug_check(); |
8c0a8537 KS |
168 | |
169 | /* | |
170 | * Make sure all delayed rcu free are flushed before we | |
171 | * destroy caches. | |
172 | */ | |
173 | rcu_barrier(); | |
d1310b2e CM |
174 | if (extent_state_cache) |
175 | kmem_cache_destroy(extent_state_cache); | |
176 | if (extent_buffer_cache) | |
177 | kmem_cache_destroy(extent_buffer_cache); | |
9be3395b CM |
178 | if (btrfs_bioset) |
179 | bioset_free(btrfs_bioset); | |
d1310b2e CM |
180 | } |
181 | ||
182 | void extent_io_tree_init(struct extent_io_tree *tree, | |
f993c883 | 183 | struct address_space *mapping) |
d1310b2e | 184 | { |
6bef4d31 | 185 | tree->state = RB_ROOT; |
19fe0a8b | 186 | INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC); |
d1310b2e CM |
187 | tree->ops = NULL; |
188 | tree->dirty_bytes = 0; | |
70dec807 | 189 | spin_lock_init(&tree->lock); |
6af118ce | 190 | spin_lock_init(&tree->buffer_lock); |
d1310b2e | 191 | tree->mapping = mapping; |
d1310b2e | 192 | } |
d1310b2e | 193 | |
b2950863 | 194 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
195 | { |
196 | struct extent_state *state; | |
d1310b2e CM |
197 | |
198 | state = kmem_cache_alloc(extent_state_cache, mask); | |
2b114d1d | 199 | if (!state) |
d1310b2e CM |
200 | return state; |
201 | state->state = 0; | |
d1310b2e | 202 | state->private = 0; |
70dec807 | 203 | state->tree = NULL; |
6d49ba1b | 204 | btrfs_leak_debug_add(&state->leak_list, &states); |
d1310b2e CM |
205 | atomic_set(&state->refs, 1); |
206 | init_waitqueue_head(&state->wq); | |
143bede5 | 207 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
208 | return state; |
209 | } | |
d1310b2e | 210 | |
4845e44f | 211 | void free_extent_state(struct extent_state *state) |
d1310b2e | 212 | { |
d1310b2e CM |
213 | if (!state) |
214 | return; | |
215 | if (atomic_dec_and_test(&state->refs)) { | |
70dec807 | 216 | WARN_ON(state->tree); |
6d49ba1b | 217 | btrfs_leak_debug_del(&state->leak_list); |
143bede5 | 218 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
219 | kmem_cache_free(extent_state_cache, state); |
220 | } | |
221 | } | |
d1310b2e CM |
222 | |
223 | static struct rb_node *tree_insert(struct rb_root *root, u64 offset, | |
224 | struct rb_node *node) | |
225 | { | |
d397712b CM |
226 | struct rb_node **p = &root->rb_node; |
227 | struct rb_node *parent = NULL; | |
d1310b2e CM |
228 | struct tree_entry *entry; |
229 | ||
d397712b | 230 | while (*p) { |
d1310b2e CM |
231 | parent = *p; |
232 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
233 | ||
234 | if (offset < entry->start) | |
235 | p = &(*p)->rb_left; | |
236 | else if (offset > entry->end) | |
237 | p = &(*p)->rb_right; | |
238 | else | |
239 | return parent; | |
240 | } | |
241 | ||
d1310b2e CM |
242 | rb_link_node(node, parent, p); |
243 | rb_insert_color(node, root); | |
244 | return NULL; | |
245 | } | |
246 | ||
80ea96b1 | 247 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
d1310b2e CM |
248 | struct rb_node **prev_ret, |
249 | struct rb_node **next_ret) | |
250 | { | |
80ea96b1 | 251 | struct rb_root *root = &tree->state; |
d397712b | 252 | struct rb_node *n = root->rb_node; |
d1310b2e CM |
253 | struct rb_node *prev = NULL; |
254 | struct rb_node *orig_prev = NULL; | |
255 | struct tree_entry *entry; | |
256 | struct tree_entry *prev_entry = NULL; | |
257 | ||
d397712b | 258 | while (n) { |
d1310b2e CM |
259 | entry = rb_entry(n, struct tree_entry, rb_node); |
260 | prev = n; | |
261 | prev_entry = entry; | |
262 | ||
263 | if (offset < entry->start) | |
264 | n = n->rb_left; | |
265 | else if (offset > entry->end) | |
266 | n = n->rb_right; | |
d397712b | 267 | else |
d1310b2e CM |
268 | return n; |
269 | } | |
270 | ||
271 | if (prev_ret) { | |
272 | orig_prev = prev; | |
d397712b | 273 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
274 | prev = rb_next(prev); |
275 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
276 | } | |
277 | *prev_ret = prev; | |
278 | prev = orig_prev; | |
279 | } | |
280 | ||
281 | if (next_ret) { | |
282 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 283 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
284 | prev = rb_prev(prev); |
285 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
286 | } | |
287 | *next_ret = prev; | |
288 | } | |
289 | return NULL; | |
290 | } | |
291 | ||
80ea96b1 CM |
292 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
293 | u64 offset) | |
d1310b2e | 294 | { |
70dec807 | 295 | struct rb_node *prev = NULL; |
d1310b2e | 296 | struct rb_node *ret; |
70dec807 | 297 | |
80ea96b1 | 298 | ret = __etree_search(tree, offset, &prev, NULL); |
d397712b | 299 | if (!ret) |
d1310b2e CM |
300 | return prev; |
301 | return ret; | |
302 | } | |
303 | ||
9ed74f2d JB |
304 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, |
305 | struct extent_state *other) | |
306 | { | |
307 | if (tree->ops && tree->ops->merge_extent_hook) | |
308 | tree->ops->merge_extent_hook(tree->mapping->host, new, | |
309 | other); | |
310 | } | |
311 | ||
d1310b2e CM |
312 | /* |
313 | * utility function to look for merge candidates inside a given range. | |
314 | * Any extents with matching state are merged together into a single | |
315 | * extent in the tree. Extents with EXTENT_IO in their state field | |
316 | * are not merged because the end_io handlers need to be able to do | |
317 | * operations on them without sleeping (or doing allocations/splits). | |
318 | * | |
319 | * This should be called with the tree lock held. | |
320 | */ | |
1bf85046 JM |
321 | static void merge_state(struct extent_io_tree *tree, |
322 | struct extent_state *state) | |
d1310b2e CM |
323 | { |
324 | struct extent_state *other; | |
325 | struct rb_node *other_node; | |
326 | ||
5b21f2ed | 327 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
1bf85046 | 328 | return; |
d1310b2e CM |
329 | |
330 | other_node = rb_prev(&state->rb_node); | |
331 | if (other_node) { | |
332 | other = rb_entry(other_node, struct extent_state, rb_node); | |
333 | if (other->end == state->start - 1 && | |
334 | other->state == state->state) { | |
9ed74f2d | 335 | merge_cb(tree, state, other); |
d1310b2e | 336 | state->start = other->start; |
70dec807 | 337 | other->tree = NULL; |
d1310b2e CM |
338 | rb_erase(&other->rb_node, &tree->state); |
339 | free_extent_state(other); | |
340 | } | |
341 | } | |
342 | other_node = rb_next(&state->rb_node); | |
343 | if (other_node) { | |
344 | other = rb_entry(other_node, struct extent_state, rb_node); | |
345 | if (other->start == state->end + 1 && | |
346 | other->state == state->state) { | |
9ed74f2d | 347 | merge_cb(tree, state, other); |
df98b6e2 JB |
348 | state->end = other->end; |
349 | other->tree = NULL; | |
350 | rb_erase(&other->rb_node, &tree->state); | |
351 | free_extent_state(other); | |
d1310b2e CM |
352 | } |
353 | } | |
d1310b2e CM |
354 | } |
355 | ||
1bf85046 | 356 | static void set_state_cb(struct extent_io_tree *tree, |
41074888 | 357 | struct extent_state *state, unsigned long *bits) |
291d673e | 358 | { |
1bf85046 JM |
359 | if (tree->ops && tree->ops->set_bit_hook) |
360 | tree->ops->set_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
361 | } |
362 | ||
363 | static void clear_state_cb(struct extent_io_tree *tree, | |
41074888 | 364 | struct extent_state *state, unsigned long *bits) |
291d673e | 365 | { |
9ed74f2d JB |
366 | if (tree->ops && tree->ops->clear_bit_hook) |
367 | tree->ops->clear_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
368 | } |
369 | ||
3150b699 | 370 | static void set_state_bits(struct extent_io_tree *tree, |
41074888 | 371 | struct extent_state *state, unsigned long *bits); |
3150b699 | 372 | |
d1310b2e CM |
373 | /* |
374 | * insert an extent_state struct into the tree. 'bits' are set on the | |
375 | * struct before it is inserted. | |
376 | * | |
377 | * This may return -EEXIST if the extent is already there, in which case the | |
378 | * state struct is freed. | |
379 | * | |
380 | * The tree lock is not taken internally. This is a utility function and | |
381 | * probably isn't what you want to call (see set/clear_extent_bit). | |
382 | */ | |
383 | static int insert_state(struct extent_io_tree *tree, | |
384 | struct extent_state *state, u64 start, u64 end, | |
41074888 | 385 | unsigned long *bits) |
d1310b2e CM |
386 | { |
387 | struct rb_node *node; | |
388 | ||
31b1a2bd JL |
389 | if (end < start) |
390 | WARN(1, KERN_ERR "btrfs end < start %llu %llu\n", | |
d397712b CM |
391 | (unsigned long long)end, |
392 | (unsigned long long)start); | |
d1310b2e CM |
393 | state->start = start; |
394 | state->end = end; | |
9ed74f2d | 395 | |
3150b699 XG |
396 | set_state_bits(tree, state, bits); |
397 | ||
d1310b2e CM |
398 | node = tree_insert(&tree->state, end, &state->rb_node); |
399 | if (node) { | |
400 | struct extent_state *found; | |
401 | found = rb_entry(node, struct extent_state, rb_node); | |
d397712b CM |
402 | printk(KERN_ERR "btrfs found node %llu %llu on insert of " |
403 | "%llu %llu\n", (unsigned long long)found->start, | |
404 | (unsigned long long)found->end, | |
405 | (unsigned long long)start, (unsigned long long)end); | |
d1310b2e CM |
406 | return -EEXIST; |
407 | } | |
70dec807 | 408 | state->tree = tree; |
d1310b2e CM |
409 | merge_state(tree, state); |
410 | return 0; | |
411 | } | |
412 | ||
1bf85046 | 413 | static void split_cb(struct extent_io_tree *tree, struct extent_state *orig, |
9ed74f2d JB |
414 | u64 split) |
415 | { | |
416 | if (tree->ops && tree->ops->split_extent_hook) | |
1bf85046 | 417 | tree->ops->split_extent_hook(tree->mapping->host, orig, split); |
9ed74f2d JB |
418 | } |
419 | ||
d1310b2e CM |
420 | /* |
421 | * split a given extent state struct in two, inserting the preallocated | |
422 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
423 | * offset inside 'orig' where it should be split. | |
424 | * | |
425 | * Before calling, | |
426 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
427 | * are two extent state structs in the tree: | |
428 | * prealloc: [orig->start, split - 1] | |
429 | * orig: [ split, orig->end ] | |
430 | * | |
431 | * The tree locks are not taken by this function. They need to be held | |
432 | * by the caller. | |
433 | */ | |
434 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
435 | struct extent_state *prealloc, u64 split) | |
436 | { | |
437 | struct rb_node *node; | |
9ed74f2d JB |
438 | |
439 | split_cb(tree, orig, split); | |
440 | ||
d1310b2e CM |
441 | prealloc->start = orig->start; |
442 | prealloc->end = split - 1; | |
443 | prealloc->state = orig->state; | |
444 | orig->start = split; | |
445 | ||
446 | node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node); | |
447 | if (node) { | |
d1310b2e CM |
448 | free_extent_state(prealloc); |
449 | return -EEXIST; | |
450 | } | |
70dec807 | 451 | prealloc->tree = tree; |
d1310b2e CM |
452 | return 0; |
453 | } | |
454 | ||
cdc6a395 LZ |
455 | static struct extent_state *next_state(struct extent_state *state) |
456 | { | |
457 | struct rb_node *next = rb_next(&state->rb_node); | |
458 | if (next) | |
459 | return rb_entry(next, struct extent_state, rb_node); | |
460 | else | |
461 | return NULL; | |
462 | } | |
463 | ||
d1310b2e CM |
464 | /* |
465 | * utility function to clear some bits in an extent state struct. | |
1b303fc0 | 466 | * it will optionally wake up any one waiting on this state (wake == 1). |
d1310b2e CM |
467 | * |
468 | * If no bits are set on the state struct after clearing things, the | |
469 | * struct is freed and removed from the tree | |
470 | */ | |
cdc6a395 LZ |
471 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
472 | struct extent_state *state, | |
41074888 | 473 | unsigned long *bits, int wake) |
d1310b2e | 474 | { |
cdc6a395 | 475 | struct extent_state *next; |
41074888 | 476 | unsigned long bits_to_clear = *bits & ~EXTENT_CTLBITS; |
d1310b2e | 477 | |
0ca1f7ce | 478 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
479 | u64 range = state->end - state->start + 1; |
480 | WARN_ON(range > tree->dirty_bytes); | |
481 | tree->dirty_bytes -= range; | |
482 | } | |
291d673e | 483 | clear_state_cb(tree, state, bits); |
32c00aff | 484 | state->state &= ~bits_to_clear; |
d1310b2e CM |
485 | if (wake) |
486 | wake_up(&state->wq); | |
0ca1f7ce | 487 | if (state->state == 0) { |
cdc6a395 | 488 | next = next_state(state); |
70dec807 | 489 | if (state->tree) { |
d1310b2e | 490 | rb_erase(&state->rb_node, &tree->state); |
70dec807 | 491 | state->tree = NULL; |
d1310b2e CM |
492 | free_extent_state(state); |
493 | } else { | |
494 | WARN_ON(1); | |
495 | } | |
496 | } else { | |
497 | merge_state(tree, state); | |
cdc6a395 | 498 | next = next_state(state); |
d1310b2e | 499 | } |
cdc6a395 | 500 | return next; |
d1310b2e CM |
501 | } |
502 | ||
8233767a XG |
503 | static struct extent_state * |
504 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
505 | { | |
506 | if (!prealloc) | |
507 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
508 | ||
509 | return prealloc; | |
510 | } | |
511 | ||
48a3b636 | 512 | static void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
c2d904e0 JM |
513 | { |
514 | btrfs_panic(tree_fs_info(tree), err, "Locking error: " | |
515 | "Extent tree was modified by another " | |
516 | "thread while locked."); | |
517 | } | |
518 | ||
d1310b2e CM |
519 | /* |
520 | * clear some bits on a range in the tree. This may require splitting | |
521 | * or inserting elements in the tree, so the gfp mask is used to | |
522 | * indicate which allocations or sleeping are allowed. | |
523 | * | |
524 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
525 | * the given range from the tree regardless of state (ie for truncate). | |
526 | * | |
527 | * the range [start, end] is inclusive. | |
528 | * | |
6763af84 | 529 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e CM |
530 | */ |
531 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
41074888 | 532 | unsigned long bits, int wake, int delete, |
2c64c53d CM |
533 | struct extent_state **cached_state, |
534 | gfp_t mask) | |
d1310b2e CM |
535 | { |
536 | struct extent_state *state; | |
2c64c53d | 537 | struct extent_state *cached; |
d1310b2e CM |
538 | struct extent_state *prealloc = NULL; |
539 | struct rb_node *node; | |
5c939df5 | 540 | u64 last_end; |
d1310b2e | 541 | int err; |
2ac55d41 | 542 | int clear = 0; |
d1310b2e | 543 | |
8d599ae1 DS |
544 | btrfs_debug_check_extent_io_range(tree->mapping->host, start, end); |
545 | ||
7ee9e440 JB |
546 | if (bits & EXTENT_DELALLOC) |
547 | bits |= EXTENT_NORESERVE; | |
548 | ||
0ca1f7ce YZ |
549 | if (delete) |
550 | bits |= ~EXTENT_CTLBITS; | |
551 | bits |= EXTENT_FIRST_DELALLOC; | |
552 | ||
2ac55d41 JB |
553 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
554 | clear = 1; | |
d1310b2e CM |
555 | again: |
556 | if (!prealloc && (mask & __GFP_WAIT)) { | |
557 | prealloc = alloc_extent_state(mask); | |
558 | if (!prealloc) | |
559 | return -ENOMEM; | |
560 | } | |
561 | ||
cad321ad | 562 | spin_lock(&tree->lock); |
2c64c53d CM |
563 | if (cached_state) { |
564 | cached = *cached_state; | |
2ac55d41 JB |
565 | |
566 | if (clear) { | |
567 | *cached_state = NULL; | |
568 | cached_state = NULL; | |
569 | } | |
570 | ||
df98b6e2 JB |
571 | if (cached && cached->tree && cached->start <= start && |
572 | cached->end > start) { | |
2ac55d41 JB |
573 | if (clear) |
574 | atomic_dec(&cached->refs); | |
2c64c53d | 575 | state = cached; |
42daec29 | 576 | goto hit_next; |
2c64c53d | 577 | } |
2ac55d41 JB |
578 | if (clear) |
579 | free_extent_state(cached); | |
2c64c53d | 580 | } |
d1310b2e CM |
581 | /* |
582 | * this search will find the extents that end after | |
583 | * our range starts | |
584 | */ | |
80ea96b1 | 585 | node = tree_search(tree, start); |
d1310b2e CM |
586 | if (!node) |
587 | goto out; | |
588 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 589 | hit_next: |
d1310b2e CM |
590 | if (state->start > end) |
591 | goto out; | |
592 | WARN_ON(state->end < start); | |
5c939df5 | 593 | last_end = state->end; |
d1310b2e | 594 | |
0449314a | 595 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
596 | if (!(state->state & bits)) { |
597 | state = next_state(state); | |
0449314a | 598 | goto next; |
cdc6a395 | 599 | } |
0449314a | 600 | |
d1310b2e CM |
601 | /* |
602 | * | ---- desired range ---- | | |
603 | * | state | or | |
604 | * | ------------- state -------------- | | |
605 | * | |
606 | * We need to split the extent we found, and may flip | |
607 | * bits on second half. | |
608 | * | |
609 | * If the extent we found extends past our range, we | |
610 | * just split and search again. It'll get split again | |
611 | * the next time though. | |
612 | * | |
613 | * If the extent we found is inside our range, we clear | |
614 | * the desired bit on it. | |
615 | */ | |
616 | ||
617 | if (state->start < start) { | |
8233767a XG |
618 | prealloc = alloc_extent_state_atomic(prealloc); |
619 | BUG_ON(!prealloc); | |
d1310b2e | 620 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
621 | if (err) |
622 | extent_io_tree_panic(tree, err); | |
623 | ||
d1310b2e CM |
624 | prealloc = NULL; |
625 | if (err) | |
626 | goto out; | |
627 | if (state->end <= end) { | |
d1ac6e41 LB |
628 | state = clear_state_bit(tree, state, &bits, wake); |
629 | goto next; | |
d1310b2e CM |
630 | } |
631 | goto search_again; | |
632 | } | |
633 | /* | |
634 | * | ---- desired range ---- | | |
635 | * | state | | |
636 | * We need to split the extent, and clear the bit | |
637 | * on the first half | |
638 | */ | |
639 | if (state->start <= end && state->end > end) { | |
8233767a XG |
640 | prealloc = alloc_extent_state_atomic(prealloc); |
641 | BUG_ON(!prealloc); | |
d1310b2e | 642 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
643 | if (err) |
644 | extent_io_tree_panic(tree, err); | |
645 | ||
d1310b2e CM |
646 | if (wake) |
647 | wake_up(&state->wq); | |
42daec29 | 648 | |
6763af84 | 649 | clear_state_bit(tree, prealloc, &bits, wake); |
9ed74f2d | 650 | |
d1310b2e CM |
651 | prealloc = NULL; |
652 | goto out; | |
653 | } | |
42daec29 | 654 | |
cdc6a395 | 655 | state = clear_state_bit(tree, state, &bits, wake); |
0449314a | 656 | next: |
5c939df5 YZ |
657 | if (last_end == (u64)-1) |
658 | goto out; | |
659 | start = last_end + 1; | |
cdc6a395 | 660 | if (start <= end && state && !need_resched()) |
692e5759 | 661 | goto hit_next; |
d1310b2e CM |
662 | goto search_again; |
663 | ||
664 | out: | |
cad321ad | 665 | spin_unlock(&tree->lock); |
d1310b2e CM |
666 | if (prealloc) |
667 | free_extent_state(prealloc); | |
668 | ||
6763af84 | 669 | return 0; |
d1310b2e CM |
670 | |
671 | search_again: | |
672 | if (start > end) | |
673 | goto out; | |
cad321ad | 674 | spin_unlock(&tree->lock); |
d1310b2e CM |
675 | if (mask & __GFP_WAIT) |
676 | cond_resched(); | |
677 | goto again; | |
678 | } | |
d1310b2e | 679 | |
143bede5 JM |
680 | static void wait_on_state(struct extent_io_tree *tree, |
681 | struct extent_state *state) | |
641f5219 CH |
682 | __releases(tree->lock) |
683 | __acquires(tree->lock) | |
d1310b2e CM |
684 | { |
685 | DEFINE_WAIT(wait); | |
686 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 687 | spin_unlock(&tree->lock); |
d1310b2e | 688 | schedule(); |
cad321ad | 689 | spin_lock(&tree->lock); |
d1310b2e | 690 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
691 | } |
692 | ||
693 | /* | |
694 | * waits for one or more bits to clear on a range in the state tree. | |
695 | * The range [start, end] is inclusive. | |
696 | * The tree lock is taken by this function | |
697 | */ | |
41074888 DS |
698 | static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
699 | unsigned long bits) | |
d1310b2e CM |
700 | { |
701 | struct extent_state *state; | |
702 | struct rb_node *node; | |
703 | ||
8d599ae1 DS |
704 | btrfs_debug_check_extent_io_range(tree->mapping->host, start, end); |
705 | ||
cad321ad | 706 | spin_lock(&tree->lock); |
d1310b2e CM |
707 | again: |
708 | while (1) { | |
709 | /* | |
710 | * this search will find all the extents that end after | |
711 | * our range starts | |
712 | */ | |
80ea96b1 | 713 | node = tree_search(tree, start); |
d1310b2e CM |
714 | if (!node) |
715 | break; | |
716 | ||
717 | state = rb_entry(node, struct extent_state, rb_node); | |
718 | ||
719 | if (state->start > end) | |
720 | goto out; | |
721 | ||
722 | if (state->state & bits) { | |
723 | start = state->start; | |
724 | atomic_inc(&state->refs); | |
725 | wait_on_state(tree, state); | |
726 | free_extent_state(state); | |
727 | goto again; | |
728 | } | |
729 | start = state->end + 1; | |
730 | ||
731 | if (start > end) | |
732 | break; | |
733 | ||
ded91f08 | 734 | cond_resched_lock(&tree->lock); |
d1310b2e CM |
735 | } |
736 | out: | |
cad321ad | 737 | spin_unlock(&tree->lock); |
d1310b2e | 738 | } |
d1310b2e | 739 | |
1bf85046 | 740 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 741 | struct extent_state *state, |
41074888 | 742 | unsigned long *bits) |
d1310b2e | 743 | { |
41074888 | 744 | unsigned long bits_to_set = *bits & ~EXTENT_CTLBITS; |
9ed74f2d | 745 | |
1bf85046 | 746 | set_state_cb(tree, state, bits); |
0ca1f7ce | 747 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
748 | u64 range = state->end - state->start + 1; |
749 | tree->dirty_bytes += range; | |
750 | } | |
0ca1f7ce | 751 | state->state |= bits_to_set; |
d1310b2e CM |
752 | } |
753 | ||
2c64c53d CM |
754 | static void cache_state(struct extent_state *state, |
755 | struct extent_state **cached_ptr) | |
756 | { | |
757 | if (cached_ptr && !(*cached_ptr)) { | |
758 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) { | |
759 | *cached_ptr = state; | |
760 | atomic_inc(&state->refs); | |
761 | } | |
762 | } | |
763 | } | |
764 | ||
d1310b2e | 765 | /* |
1edbb734 CM |
766 | * set some bits on a range in the tree. This may require allocations or |
767 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 768 | * |
1edbb734 CM |
769 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
770 | * part of the range already has the desired bits set. The start of the | |
771 | * existing range is returned in failed_start in this case. | |
d1310b2e | 772 | * |
1edbb734 | 773 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 774 | */ |
1edbb734 | 775 | |
3fbe5c02 JM |
776 | static int __must_check |
777 | __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
41074888 DS |
778 | unsigned long bits, unsigned long exclusive_bits, |
779 | u64 *failed_start, struct extent_state **cached_state, | |
780 | gfp_t mask) | |
d1310b2e CM |
781 | { |
782 | struct extent_state *state; | |
783 | struct extent_state *prealloc = NULL; | |
784 | struct rb_node *node; | |
d1310b2e | 785 | int err = 0; |
d1310b2e CM |
786 | u64 last_start; |
787 | u64 last_end; | |
42daec29 | 788 | |
8d599ae1 DS |
789 | btrfs_debug_check_extent_io_range(tree->mapping->host, start, end); |
790 | ||
0ca1f7ce | 791 | bits |= EXTENT_FIRST_DELALLOC; |
d1310b2e CM |
792 | again: |
793 | if (!prealloc && (mask & __GFP_WAIT)) { | |
794 | prealloc = alloc_extent_state(mask); | |
8233767a | 795 | BUG_ON(!prealloc); |
d1310b2e CM |
796 | } |
797 | ||
cad321ad | 798 | spin_lock(&tree->lock); |
9655d298 CM |
799 | if (cached_state && *cached_state) { |
800 | state = *cached_state; | |
df98b6e2 JB |
801 | if (state->start <= start && state->end > start && |
802 | state->tree) { | |
9655d298 CM |
803 | node = &state->rb_node; |
804 | goto hit_next; | |
805 | } | |
806 | } | |
d1310b2e CM |
807 | /* |
808 | * this search will find all the extents that end after | |
809 | * our range starts. | |
810 | */ | |
80ea96b1 | 811 | node = tree_search(tree, start); |
d1310b2e | 812 | if (!node) { |
8233767a XG |
813 | prealloc = alloc_extent_state_atomic(prealloc); |
814 | BUG_ON(!prealloc); | |
0ca1f7ce | 815 | err = insert_state(tree, prealloc, start, end, &bits); |
c2d904e0 JM |
816 | if (err) |
817 | extent_io_tree_panic(tree, err); | |
818 | ||
d1310b2e | 819 | prealloc = NULL; |
d1310b2e CM |
820 | goto out; |
821 | } | |
d1310b2e | 822 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 823 | hit_next: |
d1310b2e CM |
824 | last_start = state->start; |
825 | last_end = state->end; | |
826 | ||
827 | /* | |
828 | * | ---- desired range ---- | | |
829 | * | state | | |
830 | * | |
831 | * Just lock what we found and keep going | |
832 | */ | |
833 | if (state->start == start && state->end <= end) { | |
1edbb734 | 834 | if (state->state & exclusive_bits) { |
d1310b2e CM |
835 | *failed_start = state->start; |
836 | err = -EEXIST; | |
837 | goto out; | |
838 | } | |
42daec29 | 839 | |
1bf85046 | 840 | set_state_bits(tree, state, &bits); |
2c64c53d | 841 | cache_state(state, cached_state); |
d1310b2e | 842 | merge_state(tree, state); |
5c939df5 YZ |
843 | if (last_end == (u64)-1) |
844 | goto out; | |
845 | start = last_end + 1; | |
d1ac6e41 LB |
846 | state = next_state(state); |
847 | if (start < end && state && state->start == start && | |
848 | !need_resched()) | |
849 | goto hit_next; | |
d1310b2e CM |
850 | goto search_again; |
851 | } | |
852 | ||
853 | /* | |
854 | * | ---- desired range ---- | | |
855 | * | state | | |
856 | * or | |
857 | * | ------------- state -------------- | | |
858 | * | |
859 | * We need to split the extent we found, and may flip bits on | |
860 | * second half. | |
861 | * | |
862 | * If the extent we found extends past our | |
863 | * range, we just split and search again. It'll get split | |
864 | * again the next time though. | |
865 | * | |
866 | * If the extent we found is inside our range, we set the | |
867 | * desired bit on it. | |
868 | */ | |
869 | if (state->start < start) { | |
1edbb734 | 870 | if (state->state & exclusive_bits) { |
d1310b2e CM |
871 | *failed_start = start; |
872 | err = -EEXIST; | |
873 | goto out; | |
874 | } | |
8233767a XG |
875 | |
876 | prealloc = alloc_extent_state_atomic(prealloc); | |
877 | BUG_ON(!prealloc); | |
d1310b2e | 878 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
879 | if (err) |
880 | extent_io_tree_panic(tree, err); | |
881 | ||
d1310b2e CM |
882 | prealloc = NULL; |
883 | if (err) | |
884 | goto out; | |
885 | if (state->end <= end) { | |
1bf85046 | 886 | set_state_bits(tree, state, &bits); |
2c64c53d | 887 | cache_state(state, cached_state); |
d1310b2e | 888 | merge_state(tree, state); |
5c939df5 YZ |
889 | if (last_end == (u64)-1) |
890 | goto out; | |
891 | start = last_end + 1; | |
d1ac6e41 LB |
892 | state = next_state(state); |
893 | if (start < end && state && state->start == start && | |
894 | !need_resched()) | |
895 | goto hit_next; | |
d1310b2e CM |
896 | } |
897 | goto search_again; | |
898 | } | |
899 | /* | |
900 | * | ---- desired range ---- | | |
901 | * | state | or | state | | |
902 | * | |
903 | * There's a hole, we need to insert something in it and | |
904 | * ignore the extent we found. | |
905 | */ | |
906 | if (state->start > start) { | |
907 | u64 this_end; | |
908 | if (end < last_start) | |
909 | this_end = end; | |
910 | else | |
d397712b | 911 | this_end = last_start - 1; |
8233767a XG |
912 | |
913 | prealloc = alloc_extent_state_atomic(prealloc); | |
914 | BUG_ON(!prealloc); | |
c7f895a2 XG |
915 | |
916 | /* | |
917 | * Avoid to free 'prealloc' if it can be merged with | |
918 | * the later extent. | |
919 | */ | |
d1310b2e | 920 | err = insert_state(tree, prealloc, start, this_end, |
0ca1f7ce | 921 | &bits); |
c2d904e0 JM |
922 | if (err) |
923 | extent_io_tree_panic(tree, err); | |
924 | ||
9ed74f2d JB |
925 | cache_state(prealloc, cached_state); |
926 | prealloc = NULL; | |
d1310b2e CM |
927 | start = this_end + 1; |
928 | goto search_again; | |
929 | } | |
930 | /* | |
931 | * | ---- desired range ---- | | |
932 | * | state | | |
933 | * We need to split the extent, and set the bit | |
934 | * on the first half | |
935 | */ | |
936 | if (state->start <= end && state->end > end) { | |
1edbb734 | 937 | if (state->state & exclusive_bits) { |
d1310b2e CM |
938 | *failed_start = start; |
939 | err = -EEXIST; | |
940 | goto out; | |
941 | } | |
8233767a XG |
942 | |
943 | prealloc = alloc_extent_state_atomic(prealloc); | |
944 | BUG_ON(!prealloc); | |
d1310b2e | 945 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
946 | if (err) |
947 | extent_io_tree_panic(tree, err); | |
d1310b2e | 948 | |
1bf85046 | 949 | set_state_bits(tree, prealloc, &bits); |
2c64c53d | 950 | cache_state(prealloc, cached_state); |
d1310b2e CM |
951 | merge_state(tree, prealloc); |
952 | prealloc = NULL; | |
953 | goto out; | |
954 | } | |
955 | ||
956 | goto search_again; | |
957 | ||
958 | out: | |
cad321ad | 959 | spin_unlock(&tree->lock); |
d1310b2e CM |
960 | if (prealloc) |
961 | free_extent_state(prealloc); | |
962 | ||
963 | return err; | |
964 | ||
965 | search_again: | |
966 | if (start > end) | |
967 | goto out; | |
cad321ad | 968 | spin_unlock(&tree->lock); |
d1310b2e CM |
969 | if (mask & __GFP_WAIT) |
970 | cond_resched(); | |
971 | goto again; | |
972 | } | |
d1310b2e | 973 | |
41074888 DS |
974 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
975 | unsigned long bits, u64 * failed_start, | |
976 | struct extent_state **cached_state, gfp_t mask) | |
3fbe5c02 JM |
977 | { |
978 | return __set_extent_bit(tree, start, end, bits, 0, failed_start, | |
979 | cached_state, mask); | |
980 | } | |
981 | ||
982 | ||
462d6fac | 983 | /** |
10983f2e LB |
984 | * convert_extent_bit - convert all bits in a given range from one bit to |
985 | * another | |
462d6fac JB |
986 | * @tree: the io tree to search |
987 | * @start: the start offset in bytes | |
988 | * @end: the end offset in bytes (inclusive) | |
989 | * @bits: the bits to set in this range | |
990 | * @clear_bits: the bits to clear in this range | |
e6138876 | 991 | * @cached_state: state that we're going to cache |
462d6fac JB |
992 | * @mask: the allocation mask |
993 | * | |
994 | * This will go through and set bits for the given range. If any states exist | |
995 | * already in this range they are set with the given bit and cleared of the | |
996 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
997 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
998 | * boundary bits like LOCK. | |
999 | */ | |
1000 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
41074888 | 1001 | unsigned long bits, unsigned long clear_bits, |
e6138876 | 1002 | struct extent_state **cached_state, gfp_t mask) |
462d6fac JB |
1003 | { |
1004 | struct extent_state *state; | |
1005 | struct extent_state *prealloc = NULL; | |
1006 | struct rb_node *node; | |
1007 | int err = 0; | |
1008 | u64 last_start; | |
1009 | u64 last_end; | |
1010 | ||
8d599ae1 DS |
1011 | btrfs_debug_check_extent_io_range(tree->mapping->host, start, end); |
1012 | ||
462d6fac JB |
1013 | again: |
1014 | if (!prealloc && (mask & __GFP_WAIT)) { | |
1015 | prealloc = alloc_extent_state(mask); | |
1016 | if (!prealloc) | |
1017 | return -ENOMEM; | |
1018 | } | |
1019 | ||
1020 | spin_lock(&tree->lock); | |
e6138876 JB |
1021 | if (cached_state && *cached_state) { |
1022 | state = *cached_state; | |
1023 | if (state->start <= start && state->end > start && | |
1024 | state->tree) { | |
1025 | node = &state->rb_node; | |
1026 | goto hit_next; | |
1027 | } | |
1028 | } | |
1029 | ||
462d6fac JB |
1030 | /* |
1031 | * this search will find all the extents that end after | |
1032 | * our range starts. | |
1033 | */ | |
1034 | node = tree_search(tree, start); | |
1035 | if (!node) { | |
1036 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1037 | if (!prealloc) { |
1038 | err = -ENOMEM; | |
1039 | goto out; | |
1040 | } | |
462d6fac JB |
1041 | err = insert_state(tree, prealloc, start, end, &bits); |
1042 | prealloc = NULL; | |
c2d904e0 JM |
1043 | if (err) |
1044 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1045 | goto out; |
1046 | } | |
1047 | state = rb_entry(node, struct extent_state, rb_node); | |
1048 | hit_next: | |
1049 | last_start = state->start; | |
1050 | last_end = state->end; | |
1051 | ||
1052 | /* | |
1053 | * | ---- desired range ---- | | |
1054 | * | state | | |
1055 | * | |
1056 | * Just lock what we found and keep going | |
1057 | */ | |
1058 | if (state->start == start && state->end <= end) { | |
462d6fac | 1059 | set_state_bits(tree, state, &bits); |
e6138876 | 1060 | cache_state(state, cached_state); |
d1ac6e41 | 1061 | state = clear_state_bit(tree, state, &clear_bits, 0); |
462d6fac JB |
1062 | if (last_end == (u64)-1) |
1063 | goto out; | |
462d6fac | 1064 | start = last_end + 1; |
d1ac6e41 LB |
1065 | if (start < end && state && state->start == start && |
1066 | !need_resched()) | |
1067 | goto hit_next; | |
462d6fac JB |
1068 | goto search_again; |
1069 | } | |
1070 | ||
1071 | /* | |
1072 | * | ---- desired range ---- | | |
1073 | * | state | | |
1074 | * or | |
1075 | * | ------------- state -------------- | | |
1076 | * | |
1077 | * We need to split the extent we found, and may flip bits on | |
1078 | * second half. | |
1079 | * | |
1080 | * If the extent we found extends past our | |
1081 | * range, we just split and search again. It'll get split | |
1082 | * again the next time though. | |
1083 | * | |
1084 | * If the extent we found is inside our range, we set the | |
1085 | * desired bit on it. | |
1086 | */ | |
1087 | if (state->start < start) { | |
1088 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1089 | if (!prealloc) { |
1090 | err = -ENOMEM; | |
1091 | goto out; | |
1092 | } | |
462d6fac | 1093 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1094 | if (err) |
1095 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1096 | prealloc = NULL; |
1097 | if (err) | |
1098 | goto out; | |
1099 | if (state->end <= end) { | |
1100 | set_state_bits(tree, state, &bits); | |
e6138876 | 1101 | cache_state(state, cached_state); |
d1ac6e41 | 1102 | state = clear_state_bit(tree, state, &clear_bits, 0); |
462d6fac JB |
1103 | if (last_end == (u64)-1) |
1104 | goto out; | |
1105 | start = last_end + 1; | |
d1ac6e41 LB |
1106 | if (start < end && state && state->start == start && |
1107 | !need_resched()) | |
1108 | goto hit_next; | |
462d6fac JB |
1109 | } |
1110 | goto search_again; | |
1111 | } | |
1112 | /* | |
1113 | * | ---- desired range ---- | | |
1114 | * | state | or | state | | |
1115 | * | |
1116 | * There's a hole, we need to insert something in it and | |
1117 | * ignore the extent we found. | |
1118 | */ | |
1119 | if (state->start > start) { | |
1120 | u64 this_end; | |
1121 | if (end < last_start) | |
1122 | this_end = end; | |
1123 | else | |
1124 | this_end = last_start - 1; | |
1125 | ||
1126 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1127 | if (!prealloc) { |
1128 | err = -ENOMEM; | |
1129 | goto out; | |
1130 | } | |
462d6fac JB |
1131 | |
1132 | /* | |
1133 | * Avoid to free 'prealloc' if it can be merged with | |
1134 | * the later extent. | |
1135 | */ | |
1136 | err = insert_state(tree, prealloc, start, this_end, | |
1137 | &bits); | |
c2d904e0 JM |
1138 | if (err) |
1139 | extent_io_tree_panic(tree, err); | |
e6138876 | 1140 | cache_state(prealloc, cached_state); |
462d6fac JB |
1141 | prealloc = NULL; |
1142 | start = this_end + 1; | |
1143 | goto search_again; | |
1144 | } | |
1145 | /* | |
1146 | * | ---- desired range ---- | | |
1147 | * | state | | |
1148 | * We need to split the extent, and set the bit | |
1149 | * on the first half | |
1150 | */ | |
1151 | if (state->start <= end && state->end > end) { | |
1152 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1153 | if (!prealloc) { |
1154 | err = -ENOMEM; | |
1155 | goto out; | |
1156 | } | |
462d6fac JB |
1157 | |
1158 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1159 | if (err) |
1160 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1161 | |
1162 | set_state_bits(tree, prealloc, &bits); | |
e6138876 | 1163 | cache_state(prealloc, cached_state); |
462d6fac | 1164 | clear_state_bit(tree, prealloc, &clear_bits, 0); |
462d6fac JB |
1165 | prealloc = NULL; |
1166 | goto out; | |
1167 | } | |
1168 | ||
1169 | goto search_again; | |
1170 | ||
1171 | out: | |
1172 | spin_unlock(&tree->lock); | |
1173 | if (prealloc) | |
1174 | free_extent_state(prealloc); | |
1175 | ||
1176 | return err; | |
1177 | ||
1178 | search_again: | |
1179 | if (start > end) | |
1180 | goto out; | |
1181 | spin_unlock(&tree->lock); | |
1182 | if (mask & __GFP_WAIT) | |
1183 | cond_resched(); | |
1184 | goto again; | |
1185 | } | |
1186 | ||
d1310b2e CM |
1187 | /* wrappers around set/clear extent bit */ |
1188 | int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
1189 | gfp_t mask) | |
1190 | { | |
3fbe5c02 | 1191 | return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL, |
2c64c53d | 1192 | NULL, mask); |
d1310b2e | 1193 | } |
d1310b2e CM |
1194 | |
1195 | int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
41074888 | 1196 | unsigned long bits, gfp_t mask) |
d1310b2e | 1197 | { |
3fbe5c02 | 1198 | return set_extent_bit(tree, start, end, bits, NULL, |
2c64c53d | 1199 | NULL, mask); |
d1310b2e | 1200 | } |
d1310b2e CM |
1201 | |
1202 | int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
41074888 | 1203 | unsigned long bits, gfp_t mask) |
d1310b2e | 1204 | { |
2c64c53d | 1205 | return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask); |
d1310b2e | 1206 | } |
d1310b2e CM |
1207 | |
1208 | int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, | |
2ac55d41 | 1209 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
1210 | { |
1211 | return set_extent_bit(tree, start, end, | |
fee187d9 | 1212 | EXTENT_DELALLOC | EXTENT_UPTODATE, |
3fbe5c02 | 1213 | NULL, cached_state, mask); |
d1310b2e | 1214 | } |
d1310b2e | 1215 | |
9e8a4a8b LB |
1216 | int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end, |
1217 | struct extent_state **cached_state, gfp_t mask) | |
1218 | { | |
1219 | return set_extent_bit(tree, start, end, | |
1220 | EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG, | |
1221 | NULL, cached_state, mask); | |
1222 | } | |
1223 | ||
d1310b2e CM |
1224 | int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, |
1225 | gfp_t mask) | |
1226 | { | |
1227 | return clear_extent_bit(tree, start, end, | |
32c00aff | 1228 | EXTENT_DIRTY | EXTENT_DELALLOC | |
0ca1f7ce | 1229 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask); |
d1310b2e | 1230 | } |
d1310b2e CM |
1231 | |
1232 | int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, | |
1233 | gfp_t mask) | |
1234 | { | |
3fbe5c02 | 1235 | return set_extent_bit(tree, start, end, EXTENT_NEW, NULL, |
2c64c53d | 1236 | NULL, mask); |
d1310b2e | 1237 | } |
d1310b2e | 1238 | |
d1310b2e | 1239 | int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
507903b8 | 1240 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e | 1241 | { |
6b67a320 | 1242 | return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL, |
3fbe5c02 | 1243 | cached_state, mask); |
d1310b2e | 1244 | } |
d1310b2e | 1245 | |
5fd02043 JB |
1246 | int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
1247 | struct extent_state **cached_state, gfp_t mask) | |
d1310b2e | 1248 | { |
2c64c53d | 1249 | return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, |
2ac55d41 | 1250 | cached_state, mask); |
d1310b2e | 1251 | } |
d1310b2e | 1252 | |
d352ac68 CM |
1253 | /* |
1254 | * either insert or lock state struct between start and end use mask to tell | |
1255 | * us if waiting is desired. | |
1256 | */ | |
1edbb734 | 1257 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
41074888 | 1258 | unsigned long bits, struct extent_state **cached_state) |
d1310b2e CM |
1259 | { |
1260 | int err; | |
1261 | u64 failed_start; | |
1262 | while (1) { | |
3fbe5c02 JM |
1263 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits, |
1264 | EXTENT_LOCKED, &failed_start, | |
1265 | cached_state, GFP_NOFS); | |
d0082371 | 1266 | if (err == -EEXIST) { |
d1310b2e CM |
1267 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1268 | start = failed_start; | |
d0082371 | 1269 | } else |
d1310b2e | 1270 | break; |
d1310b2e CM |
1271 | WARN_ON(start > end); |
1272 | } | |
1273 | return err; | |
1274 | } | |
d1310b2e | 1275 | |
d0082371 | 1276 | int lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
1edbb734 | 1277 | { |
d0082371 | 1278 | return lock_extent_bits(tree, start, end, 0, NULL); |
1edbb734 CM |
1279 | } |
1280 | ||
d0082371 | 1281 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1282 | { |
1283 | int err; | |
1284 | u64 failed_start; | |
1285 | ||
3fbe5c02 JM |
1286 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
1287 | &failed_start, NULL, GFP_NOFS); | |
6643558d YZ |
1288 | if (err == -EEXIST) { |
1289 | if (failed_start > start) | |
1290 | clear_extent_bit(tree, start, failed_start - 1, | |
d0082371 | 1291 | EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS); |
25179201 | 1292 | return 0; |
6643558d | 1293 | } |
25179201 JB |
1294 | return 1; |
1295 | } | |
25179201 | 1296 | |
2c64c53d CM |
1297 | int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end, |
1298 | struct extent_state **cached, gfp_t mask) | |
1299 | { | |
1300 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, | |
1301 | mask); | |
1302 | } | |
1303 | ||
d0082371 | 1304 | int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e | 1305 | { |
2c64c53d | 1306 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL, |
d0082371 | 1307 | GFP_NOFS); |
d1310b2e | 1308 | } |
d1310b2e | 1309 | |
4adaa611 CM |
1310 | int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
1311 | { | |
1312 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1313 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1314 | struct page *page; | |
1315 | ||
1316 | while (index <= end_index) { | |
1317 | page = find_get_page(inode->i_mapping, index); | |
1318 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1319 | clear_page_dirty_for_io(page); | |
1320 | page_cache_release(page); | |
1321 | index++; | |
1322 | } | |
1323 | return 0; | |
1324 | } | |
1325 | ||
1326 | int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) | |
1327 | { | |
1328 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1329 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1330 | struct page *page; | |
1331 | ||
1332 | while (index <= end_index) { | |
1333 | page = find_get_page(inode->i_mapping, index); | |
1334 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1335 | account_page_redirty(page); | |
1336 | __set_page_dirty_nobuffers(page); | |
1337 | page_cache_release(page); | |
1338 | index++; | |
1339 | } | |
1340 | return 0; | |
1341 | } | |
1342 | ||
d1310b2e CM |
1343 | /* |
1344 | * helper function to set both pages and extents in the tree writeback | |
1345 | */ | |
b2950863 | 1346 | static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e CM |
1347 | { |
1348 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1349 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1350 | struct page *page; | |
1351 | ||
1352 | while (index <= end_index) { | |
1353 | page = find_get_page(tree->mapping, index); | |
79787eaa | 1354 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ |
d1310b2e CM |
1355 | set_page_writeback(page); |
1356 | page_cache_release(page); | |
1357 | index++; | |
1358 | } | |
d1310b2e CM |
1359 | return 0; |
1360 | } | |
d1310b2e | 1361 | |
d352ac68 CM |
1362 | /* find the first state struct with 'bits' set after 'start', and |
1363 | * return it. tree->lock must be held. NULL will returned if | |
1364 | * nothing was found after 'start' | |
1365 | */ | |
48a3b636 ES |
1366 | static struct extent_state * |
1367 | find_first_extent_bit_state(struct extent_io_tree *tree, | |
41074888 | 1368 | u64 start, unsigned long bits) |
d7fc640e CM |
1369 | { |
1370 | struct rb_node *node; | |
1371 | struct extent_state *state; | |
1372 | ||
1373 | /* | |
1374 | * this search will find all the extents that end after | |
1375 | * our range starts. | |
1376 | */ | |
1377 | node = tree_search(tree, start); | |
d397712b | 1378 | if (!node) |
d7fc640e | 1379 | goto out; |
d7fc640e | 1380 | |
d397712b | 1381 | while (1) { |
d7fc640e | 1382 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1383 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1384 | return state; |
d397712b | 1385 | |
d7fc640e CM |
1386 | node = rb_next(node); |
1387 | if (!node) | |
1388 | break; | |
1389 | } | |
1390 | out: | |
1391 | return NULL; | |
1392 | } | |
d7fc640e | 1393 | |
69261c4b XG |
1394 | /* |
1395 | * find the first offset in the io tree with 'bits' set. zero is | |
1396 | * returned if we find something, and *start_ret and *end_ret are | |
1397 | * set to reflect the state struct that was found. | |
1398 | * | |
477d7eaf | 1399 | * If nothing was found, 1 is returned. If found something, return 0. |
69261c4b XG |
1400 | */ |
1401 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
41074888 | 1402 | u64 *start_ret, u64 *end_ret, unsigned long bits, |
e6138876 | 1403 | struct extent_state **cached_state) |
69261c4b XG |
1404 | { |
1405 | struct extent_state *state; | |
e6138876 | 1406 | struct rb_node *n; |
69261c4b XG |
1407 | int ret = 1; |
1408 | ||
1409 | spin_lock(&tree->lock); | |
e6138876 JB |
1410 | if (cached_state && *cached_state) { |
1411 | state = *cached_state; | |
1412 | if (state->end == start - 1 && state->tree) { | |
1413 | n = rb_next(&state->rb_node); | |
1414 | while (n) { | |
1415 | state = rb_entry(n, struct extent_state, | |
1416 | rb_node); | |
1417 | if (state->state & bits) | |
1418 | goto got_it; | |
1419 | n = rb_next(n); | |
1420 | } | |
1421 | free_extent_state(*cached_state); | |
1422 | *cached_state = NULL; | |
1423 | goto out; | |
1424 | } | |
1425 | free_extent_state(*cached_state); | |
1426 | *cached_state = NULL; | |
1427 | } | |
1428 | ||
69261c4b | 1429 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1430 | got_it: |
69261c4b | 1431 | if (state) { |
e6138876 | 1432 | cache_state(state, cached_state); |
69261c4b XG |
1433 | *start_ret = state->start; |
1434 | *end_ret = state->end; | |
1435 | ret = 0; | |
1436 | } | |
e6138876 | 1437 | out: |
69261c4b XG |
1438 | spin_unlock(&tree->lock); |
1439 | return ret; | |
1440 | } | |
1441 | ||
d352ac68 CM |
1442 | /* |
1443 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1444 | * more than 'max_bytes'. start and end are used to return the range, | |
1445 | * | |
1446 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1447 | */ | |
c8b97818 | 1448 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1449 | u64 *start, u64 *end, u64 max_bytes, |
1450 | struct extent_state **cached_state) | |
d1310b2e CM |
1451 | { |
1452 | struct rb_node *node; | |
1453 | struct extent_state *state; | |
1454 | u64 cur_start = *start; | |
1455 | u64 found = 0; | |
1456 | u64 total_bytes = 0; | |
1457 | ||
cad321ad | 1458 | spin_lock(&tree->lock); |
c8b97818 | 1459 | |
d1310b2e CM |
1460 | /* |
1461 | * this search will find all the extents that end after | |
1462 | * our range starts. | |
1463 | */ | |
80ea96b1 | 1464 | node = tree_search(tree, cur_start); |
2b114d1d | 1465 | if (!node) { |
3b951516 CM |
1466 | if (!found) |
1467 | *end = (u64)-1; | |
d1310b2e CM |
1468 | goto out; |
1469 | } | |
1470 | ||
d397712b | 1471 | while (1) { |
d1310b2e | 1472 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1473 | if (found && (state->start != cur_start || |
1474 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1475 | goto out; |
1476 | } | |
1477 | if (!(state->state & EXTENT_DELALLOC)) { | |
1478 | if (!found) | |
1479 | *end = state->end; | |
1480 | goto out; | |
1481 | } | |
c2a128d2 | 1482 | if (!found) { |
d1310b2e | 1483 | *start = state->start; |
c2a128d2 JB |
1484 | *cached_state = state; |
1485 | atomic_inc(&state->refs); | |
1486 | } | |
d1310b2e CM |
1487 | found++; |
1488 | *end = state->end; | |
1489 | cur_start = state->end + 1; | |
1490 | node = rb_next(node); | |
1491 | if (!node) | |
1492 | break; | |
1493 | total_bytes += state->end - state->start + 1; | |
1494 | if (total_bytes >= max_bytes) | |
1495 | break; | |
1496 | } | |
1497 | out: | |
cad321ad | 1498 | spin_unlock(&tree->lock); |
d1310b2e CM |
1499 | return found; |
1500 | } | |
1501 | ||
143bede5 JM |
1502 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1503 | struct page *locked_page, | |
1504 | u64 start, u64 end) | |
c8b97818 CM |
1505 | { |
1506 | int ret; | |
1507 | struct page *pages[16]; | |
1508 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1509 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1510 | unsigned long nr_pages = end_index - index + 1; | |
1511 | int i; | |
1512 | ||
1513 | if (index == locked_page->index && end_index == index) | |
143bede5 | 1514 | return; |
c8b97818 | 1515 | |
d397712b | 1516 | while (nr_pages > 0) { |
c8b97818 | 1517 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1518 | min_t(unsigned long, nr_pages, |
1519 | ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1520 | for (i = 0; i < ret; i++) { |
1521 | if (pages[i] != locked_page) | |
1522 | unlock_page(pages[i]); | |
1523 | page_cache_release(pages[i]); | |
1524 | } | |
1525 | nr_pages -= ret; | |
1526 | index += ret; | |
1527 | cond_resched(); | |
1528 | } | |
c8b97818 CM |
1529 | } |
1530 | ||
1531 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1532 | struct page *locked_page, | |
1533 | u64 delalloc_start, | |
1534 | u64 delalloc_end) | |
1535 | { | |
1536 | unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT; | |
1537 | unsigned long start_index = index; | |
1538 | unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT; | |
1539 | unsigned long pages_locked = 0; | |
1540 | struct page *pages[16]; | |
1541 | unsigned long nrpages; | |
1542 | int ret; | |
1543 | int i; | |
1544 | ||
1545 | /* the caller is responsible for locking the start index */ | |
1546 | if (index == locked_page->index && index == end_index) | |
1547 | return 0; | |
1548 | ||
1549 | /* skip the page at the start index */ | |
1550 | nrpages = end_index - index + 1; | |
d397712b | 1551 | while (nrpages > 0) { |
c8b97818 | 1552 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1553 | min_t(unsigned long, |
1554 | nrpages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1555 | if (ret == 0) { |
1556 | ret = -EAGAIN; | |
1557 | goto done; | |
1558 | } | |
1559 | /* now we have an array of pages, lock them all */ | |
1560 | for (i = 0; i < ret; i++) { | |
1561 | /* | |
1562 | * the caller is taking responsibility for | |
1563 | * locked_page | |
1564 | */ | |
771ed689 | 1565 | if (pages[i] != locked_page) { |
c8b97818 | 1566 | lock_page(pages[i]); |
f2b1c41c CM |
1567 | if (!PageDirty(pages[i]) || |
1568 | pages[i]->mapping != inode->i_mapping) { | |
771ed689 CM |
1569 | ret = -EAGAIN; |
1570 | unlock_page(pages[i]); | |
1571 | page_cache_release(pages[i]); | |
1572 | goto done; | |
1573 | } | |
1574 | } | |
c8b97818 | 1575 | page_cache_release(pages[i]); |
771ed689 | 1576 | pages_locked++; |
c8b97818 | 1577 | } |
c8b97818 CM |
1578 | nrpages -= ret; |
1579 | index += ret; | |
1580 | cond_resched(); | |
1581 | } | |
1582 | ret = 0; | |
1583 | done: | |
1584 | if (ret && pages_locked) { | |
1585 | __unlock_for_delalloc(inode, locked_page, | |
1586 | delalloc_start, | |
1587 | ((u64)(start_index + pages_locked - 1)) << | |
1588 | PAGE_CACHE_SHIFT); | |
1589 | } | |
1590 | return ret; | |
1591 | } | |
1592 | ||
1593 | /* | |
1594 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1595 | * more than 'max_bytes'. start and end are used to return the range, | |
1596 | * | |
1597 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1598 | */ | |
1599 | static noinline u64 find_lock_delalloc_range(struct inode *inode, | |
1600 | struct extent_io_tree *tree, | |
1601 | struct page *locked_page, | |
1602 | u64 *start, u64 *end, | |
1603 | u64 max_bytes) | |
1604 | { | |
1605 | u64 delalloc_start; | |
1606 | u64 delalloc_end; | |
1607 | u64 found; | |
9655d298 | 1608 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1609 | int ret; |
1610 | int loops = 0; | |
1611 | ||
1612 | again: | |
1613 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1614 | delalloc_start = *start; | |
1615 | delalloc_end = 0; | |
1616 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1617 | max_bytes, &cached_state); |
70b99e69 | 1618 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1619 | *start = delalloc_start; |
1620 | *end = delalloc_end; | |
c2a128d2 | 1621 | free_extent_state(cached_state); |
c8b97818 CM |
1622 | return found; |
1623 | } | |
1624 | ||
70b99e69 CM |
1625 | /* |
1626 | * start comes from the offset of locked_page. We have to lock | |
1627 | * pages in order, so we can't process delalloc bytes before | |
1628 | * locked_page | |
1629 | */ | |
d397712b | 1630 | if (delalloc_start < *start) |
70b99e69 | 1631 | delalloc_start = *start; |
70b99e69 | 1632 | |
c8b97818 CM |
1633 | /* |
1634 | * make sure to limit the number of pages we try to lock down | |
1635 | * if we're looping. | |
1636 | */ | |
d397712b | 1637 | if (delalloc_end + 1 - delalloc_start > max_bytes && loops) |
771ed689 | 1638 | delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1; |
d397712b | 1639 | |
c8b97818 CM |
1640 | /* step two, lock all the pages after the page that has start */ |
1641 | ret = lock_delalloc_pages(inode, locked_page, | |
1642 | delalloc_start, delalloc_end); | |
1643 | if (ret == -EAGAIN) { | |
1644 | /* some of the pages are gone, lets avoid looping by | |
1645 | * shortening the size of the delalloc range we're searching | |
1646 | */ | |
9655d298 | 1647 | free_extent_state(cached_state); |
c8b97818 CM |
1648 | if (!loops) { |
1649 | unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1); | |
1650 | max_bytes = PAGE_CACHE_SIZE - offset; | |
1651 | loops = 1; | |
1652 | goto again; | |
1653 | } else { | |
1654 | found = 0; | |
1655 | goto out_failed; | |
1656 | } | |
1657 | } | |
79787eaa | 1658 | BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */ |
c8b97818 CM |
1659 | |
1660 | /* step three, lock the state bits for the whole range */ | |
d0082371 | 1661 | lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state); |
c8b97818 CM |
1662 | |
1663 | /* then test to make sure it is all still delalloc */ | |
1664 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1665 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1666 | if (!ret) { |
9655d298 CM |
1667 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
1668 | &cached_state, GFP_NOFS); | |
c8b97818 CM |
1669 | __unlock_for_delalloc(inode, locked_page, |
1670 | delalloc_start, delalloc_end); | |
1671 | cond_resched(); | |
1672 | goto again; | |
1673 | } | |
9655d298 | 1674 | free_extent_state(cached_state); |
c8b97818 CM |
1675 | *start = delalloc_start; |
1676 | *end = delalloc_end; | |
1677 | out_failed: | |
1678 | return found; | |
1679 | } | |
1680 | ||
c2790a2e JB |
1681 | int extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end, |
1682 | struct page *locked_page, | |
1683 | unsigned long clear_bits, | |
1684 | unsigned long page_ops) | |
c8b97818 | 1685 | { |
c2790a2e | 1686 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
c8b97818 CM |
1687 | int ret; |
1688 | struct page *pages[16]; | |
1689 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1690 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1691 | unsigned long nr_pages = end_index - index + 1; | |
1692 | int i; | |
771ed689 | 1693 | |
2c64c53d | 1694 | clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); |
c2790a2e | 1695 | if (page_ops == 0) |
771ed689 | 1696 | return 0; |
c8b97818 | 1697 | |
d397712b | 1698 | while (nr_pages > 0) { |
c8b97818 | 1699 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1700 | min_t(unsigned long, |
1701 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 | 1702 | for (i = 0; i < ret; i++) { |
8b62b72b | 1703 | |
c2790a2e | 1704 | if (page_ops & PAGE_SET_PRIVATE2) |
8b62b72b CM |
1705 | SetPagePrivate2(pages[i]); |
1706 | ||
c8b97818 CM |
1707 | if (pages[i] == locked_page) { |
1708 | page_cache_release(pages[i]); | |
1709 | continue; | |
1710 | } | |
c2790a2e | 1711 | if (page_ops & PAGE_CLEAR_DIRTY) |
c8b97818 | 1712 | clear_page_dirty_for_io(pages[i]); |
c2790a2e | 1713 | if (page_ops & PAGE_SET_WRITEBACK) |
c8b97818 | 1714 | set_page_writeback(pages[i]); |
c2790a2e | 1715 | if (page_ops & PAGE_END_WRITEBACK) |
c8b97818 | 1716 | end_page_writeback(pages[i]); |
c2790a2e | 1717 | if (page_ops & PAGE_UNLOCK) |
771ed689 | 1718 | unlock_page(pages[i]); |
c8b97818 CM |
1719 | page_cache_release(pages[i]); |
1720 | } | |
1721 | nr_pages -= ret; | |
1722 | index += ret; | |
1723 | cond_resched(); | |
1724 | } | |
1725 | return 0; | |
1726 | } | |
c8b97818 | 1727 | |
d352ac68 CM |
1728 | /* |
1729 | * count the number of bytes in the tree that have a given bit(s) | |
1730 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1731 | * cached. The total number found is returned. | |
1732 | */ | |
d1310b2e CM |
1733 | u64 count_range_bits(struct extent_io_tree *tree, |
1734 | u64 *start, u64 search_end, u64 max_bytes, | |
ec29ed5b | 1735 | unsigned long bits, int contig) |
d1310b2e CM |
1736 | { |
1737 | struct rb_node *node; | |
1738 | struct extent_state *state; | |
1739 | u64 cur_start = *start; | |
1740 | u64 total_bytes = 0; | |
ec29ed5b | 1741 | u64 last = 0; |
d1310b2e CM |
1742 | int found = 0; |
1743 | ||
1744 | if (search_end <= cur_start) { | |
d1310b2e CM |
1745 | WARN_ON(1); |
1746 | return 0; | |
1747 | } | |
1748 | ||
cad321ad | 1749 | spin_lock(&tree->lock); |
d1310b2e CM |
1750 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1751 | total_bytes = tree->dirty_bytes; | |
1752 | goto out; | |
1753 | } | |
1754 | /* | |
1755 | * this search will find all the extents that end after | |
1756 | * our range starts. | |
1757 | */ | |
80ea96b1 | 1758 | node = tree_search(tree, cur_start); |
d397712b | 1759 | if (!node) |
d1310b2e | 1760 | goto out; |
d1310b2e | 1761 | |
d397712b | 1762 | while (1) { |
d1310b2e CM |
1763 | state = rb_entry(node, struct extent_state, rb_node); |
1764 | if (state->start > search_end) | |
1765 | break; | |
ec29ed5b CM |
1766 | if (contig && found && state->start > last + 1) |
1767 | break; | |
1768 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1769 | total_bytes += min(search_end, state->end) + 1 - |
1770 | max(cur_start, state->start); | |
1771 | if (total_bytes >= max_bytes) | |
1772 | break; | |
1773 | if (!found) { | |
af60bed2 | 1774 | *start = max(cur_start, state->start); |
d1310b2e CM |
1775 | found = 1; |
1776 | } | |
ec29ed5b CM |
1777 | last = state->end; |
1778 | } else if (contig && found) { | |
1779 | break; | |
d1310b2e CM |
1780 | } |
1781 | node = rb_next(node); | |
1782 | if (!node) | |
1783 | break; | |
1784 | } | |
1785 | out: | |
cad321ad | 1786 | spin_unlock(&tree->lock); |
d1310b2e CM |
1787 | return total_bytes; |
1788 | } | |
b2950863 | 1789 | |
d352ac68 CM |
1790 | /* |
1791 | * set the private field for a given byte offset in the tree. If there isn't | |
1792 | * an extent_state there already, this does nothing. | |
1793 | */ | |
171170c1 | 1794 | static int set_state_private(struct extent_io_tree *tree, u64 start, u64 private) |
d1310b2e CM |
1795 | { |
1796 | struct rb_node *node; | |
1797 | struct extent_state *state; | |
1798 | int ret = 0; | |
1799 | ||
cad321ad | 1800 | spin_lock(&tree->lock); |
d1310b2e CM |
1801 | /* |
1802 | * this search will find all the extents that end after | |
1803 | * our range starts. | |
1804 | */ | |
80ea96b1 | 1805 | node = tree_search(tree, start); |
2b114d1d | 1806 | if (!node) { |
d1310b2e CM |
1807 | ret = -ENOENT; |
1808 | goto out; | |
1809 | } | |
1810 | state = rb_entry(node, struct extent_state, rb_node); | |
1811 | if (state->start != start) { | |
1812 | ret = -ENOENT; | |
1813 | goto out; | |
1814 | } | |
1815 | state->private = private; | |
1816 | out: | |
cad321ad | 1817 | spin_unlock(&tree->lock); |
d1310b2e CM |
1818 | return ret; |
1819 | } | |
1820 | ||
1821 | int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private) | |
1822 | { | |
1823 | struct rb_node *node; | |
1824 | struct extent_state *state; | |
1825 | int ret = 0; | |
1826 | ||
cad321ad | 1827 | spin_lock(&tree->lock); |
d1310b2e CM |
1828 | /* |
1829 | * this search will find all the extents that end after | |
1830 | * our range starts. | |
1831 | */ | |
80ea96b1 | 1832 | node = tree_search(tree, start); |
2b114d1d | 1833 | if (!node) { |
d1310b2e CM |
1834 | ret = -ENOENT; |
1835 | goto out; | |
1836 | } | |
1837 | state = rb_entry(node, struct extent_state, rb_node); | |
1838 | if (state->start != start) { | |
1839 | ret = -ENOENT; | |
1840 | goto out; | |
1841 | } | |
1842 | *private = state->private; | |
1843 | out: | |
cad321ad | 1844 | spin_unlock(&tree->lock); |
d1310b2e CM |
1845 | return ret; |
1846 | } | |
1847 | ||
1848 | /* | |
1849 | * searches a range in the state tree for a given mask. | |
70dec807 | 1850 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1851 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1852 | * range is found set. | |
1853 | */ | |
1854 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
41074888 | 1855 | unsigned long bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1856 | { |
1857 | struct extent_state *state = NULL; | |
1858 | struct rb_node *node; | |
1859 | int bitset = 0; | |
d1310b2e | 1860 | |
cad321ad | 1861 | spin_lock(&tree->lock); |
df98b6e2 JB |
1862 | if (cached && cached->tree && cached->start <= start && |
1863 | cached->end > start) | |
9655d298 CM |
1864 | node = &cached->rb_node; |
1865 | else | |
1866 | node = tree_search(tree, start); | |
d1310b2e CM |
1867 | while (node && start <= end) { |
1868 | state = rb_entry(node, struct extent_state, rb_node); | |
1869 | ||
1870 | if (filled && state->start > start) { | |
1871 | bitset = 0; | |
1872 | break; | |
1873 | } | |
1874 | ||
1875 | if (state->start > end) | |
1876 | break; | |
1877 | ||
1878 | if (state->state & bits) { | |
1879 | bitset = 1; | |
1880 | if (!filled) | |
1881 | break; | |
1882 | } else if (filled) { | |
1883 | bitset = 0; | |
1884 | break; | |
1885 | } | |
46562cec CM |
1886 | |
1887 | if (state->end == (u64)-1) | |
1888 | break; | |
1889 | ||
d1310b2e CM |
1890 | start = state->end + 1; |
1891 | if (start > end) | |
1892 | break; | |
1893 | node = rb_next(node); | |
1894 | if (!node) { | |
1895 | if (filled) | |
1896 | bitset = 0; | |
1897 | break; | |
1898 | } | |
1899 | } | |
cad321ad | 1900 | spin_unlock(&tree->lock); |
d1310b2e CM |
1901 | return bitset; |
1902 | } | |
d1310b2e CM |
1903 | |
1904 | /* | |
1905 | * helper function to set a given page up to date if all the | |
1906 | * extents in the tree for that page are up to date | |
1907 | */ | |
143bede5 | 1908 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e | 1909 | { |
4eee4fa4 | 1910 | u64 start = page_offset(page); |
d1310b2e | 1911 | u64 end = start + PAGE_CACHE_SIZE - 1; |
9655d298 | 1912 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 1913 | SetPageUptodate(page); |
d1310b2e CM |
1914 | } |
1915 | ||
4a54c8c1 JS |
1916 | /* |
1917 | * When IO fails, either with EIO or csum verification fails, we | |
1918 | * try other mirrors that might have a good copy of the data. This | |
1919 | * io_failure_record is used to record state as we go through all the | |
1920 | * mirrors. If another mirror has good data, the page is set up to date | |
1921 | * and things continue. If a good mirror can't be found, the original | |
1922 | * bio end_io callback is called to indicate things have failed. | |
1923 | */ | |
1924 | struct io_failure_record { | |
1925 | struct page *page; | |
1926 | u64 start; | |
1927 | u64 len; | |
1928 | u64 logical; | |
1929 | unsigned long bio_flags; | |
1930 | int this_mirror; | |
1931 | int failed_mirror; | |
1932 | int in_validation; | |
1933 | }; | |
1934 | ||
1935 | static int free_io_failure(struct inode *inode, struct io_failure_record *rec, | |
1936 | int did_repair) | |
1937 | { | |
1938 | int ret; | |
1939 | int err = 0; | |
1940 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
1941 | ||
1942 | set_state_private(failure_tree, rec->start, 0); | |
1943 | ret = clear_extent_bits(failure_tree, rec->start, | |
1944 | rec->start + rec->len - 1, | |
1945 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1946 | if (ret) | |
1947 | err = ret; | |
1948 | ||
53b381b3 DW |
1949 | ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start, |
1950 | rec->start + rec->len - 1, | |
1951 | EXTENT_DAMAGED, GFP_NOFS); | |
1952 | if (ret && !err) | |
1953 | err = ret; | |
4a54c8c1 JS |
1954 | |
1955 | kfree(rec); | |
1956 | return err; | |
1957 | } | |
1958 | ||
1959 | static void repair_io_failure_callback(struct bio *bio, int err) | |
1960 | { | |
1961 | complete(bio->bi_private); | |
1962 | } | |
1963 | ||
1964 | /* | |
1965 | * this bypasses the standard btrfs submit functions deliberately, as | |
1966 | * the standard behavior is to write all copies in a raid setup. here we only | |
1967 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
1968 | * submit_bio directly. | |
3ec706c8 | 1969 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
1970 | * actually prevents the read that triggered the error from finishing. |
1971 | * currently, there can be no more than two copies of every data bit. thus, | |
1972 | * exactly one rewrite is required. | |
1973 | */ | |
3ec706c8 | 1974 | int repair_io_failure(struct btrfs_fs_info *fs_info, u64 start, |
4a54c8c1 JS |
1975 | u64 length, u64 logical, struct page *page, |
1976 | int mirror_num) | |
1977 | { | |
1978 | struct bio *bio; | |
1979 | struct btrfs_device *dev; | |
1980 | DECLARE_COMPLETION_ONSTACK(compl); | |
1981 | u64 map_length = 0; | |
1982 | u64 sector; | |
1983 | struct btrfs_bio *bbio = NULL; | |
53b381b3 | 1984 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
4a54c8c1 JS |
1985 | int ret; |
1986 | ||
1987 | BUG_ON(!mirror_num); | |
1988 | ||
53b381b3 DW |
1989 | /* we can't repair anything in raid56 yet */ |
1990 | if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num)) | |
1991 | return 0; | |
1992 | ||
9be3395b | 1993 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
4a54c8c1 JS |
1994 | if (!bio) |
1995 | return -EIO; | |
1996 | bio->bi_private = &compl; | |
1997 | bio->bi_end_io = repair_io_failure_callback; | |
1998 | bio->bi_size = 0; | |
1999 | map_length = length; | |
2000 | ||
3ec706c8 | 2001 | ret = btrfs_map_block(fs_info, WRITE, logical, |
4a54c8c1 JS |
2002 | &map_length, &bbio, mirror_num); |
2003 | if (ret) { | |
2004 | bio_put(bio); | |
2005 | return -EIO; | |
2006 | } | |
2007 | BUG_ON(mirror_num != bbio->mirror_num); | |
2008 | sector = bbio->stripes[mirror_num-1].physical >> 9; | |
2009 | bio->bi_sector = sector; | |
2010 | dev = bbio->stripes[mirror_num-1].dev; | |
2011 | kfree(bbio); | |
2012 | if (!dev || !dev->bdev || !dev->writeable) { | |
2013 | bio_put(bio); | |
2014 | return -EIO; | |
2015 | } | |
2016 | bio->bi_bdev = dev->bdev; | |
4eee4fa4 | 2017 | bio_add_page(bio, page, length, start - page_offset(page)); |
21adbd5c | 2018 | btrfsic_submit_bio(WRITE_SYNC, bio); |
4a54c8c1 JS |
2019 | wait_for_completion(&compl); |
2020 | ||
2021 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) { | |
2022 | /* try to remap that extent elsewhere? */ | |
2023 | bio_put(bio); | |
442a4f63 | 2024 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
2025 | return -EIO; |
2026 | } | |
2027 | ||
d5b025d5 | 2028 | printk_ratelimited_in_rcu(KERN_INFO "btrfs read error corrected: ino %lu off %llu " |
606686ee JB |
2029 | "(dev %s sector %llu)\n", page->mapping->host->i_ino, |
2030 | start, rcu_str_deref(dev->name), sector); | |
4a54c8c1 JS |
2031 | |
2032 | bio_put(bio); | |
2033 | return 0; | |
2034 | } | |
2035 | ||
ea466794 JB |
2036 | int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb, |
2037 | int mirror_num) | |
2038 | { | |
ea466794 JB |
2039 | u64 start = eb->start; |
2040 | unsigned long i, num_pages = num_extent_pages(eb->start, eb->len); | |
d95603b2 | 2041 | int ret = 0; |
ea466794 JB |
2042 | |
2043 | for (i = 0; i < num_pages; i++) { | |
2044 | struct page *p = extent_buffer_page(eb, i); | |
3ec706c8 | 2045 | ret = repair_io_failure(root->fs_info, start, PAGE_CACHE_SIZE, |
ea466794 JB |
2046 | start, p, mirror_num); |
2047 | if (ret) | |
2048 | break; | |
2049 | start += PAGE_CACHE_SIZE; | |
2050 | } | |
2051 | ||
2052 | return ret; | |
2053 | } | |
2054 | ||
4a54c8c1 JS |
2055 | /* |
2056 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2057 | * to see if we need to process or clean up an io_failure_record | |
2058 | */ | |
2059 | static int clean_io_failure(u64 start, struct page *page) | |
2060 | { | |
2061 | u64 private; | |
2062 | u64 private_failure; | |
2063 | struct io_failure_record *failrec; | |
3ec706c8 | 2064 | struct btrfs_fs_info *fs_info; |
4a54c8c1 JS |
2065 | struct extent_state *state; |
2066 | int num_copies; | |
2067 | int did_repair = 0; | |
2068 | int ret; | |
2069 | struct inode *inode = page->mapping->host; | |
2070 | ||
2071 | private = 0; | |
2072 | ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
2073 | (u64)-1, 1, EXTENT_DIRTY, 0); | |
2074 | if (!ret) | |
2075 | return 0; | |
2076 | ||
2077 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start, | |
2078 | &private_failure); | |
2079 | if (ret) | |
2080 | return 0; | |
2081 | ||
2082 | failrec = (struct io_failure_record *)(unsigned long) private_failure; | |
2083 | BUG_ON(!failrec->this_mirror); | |
2084 | ||
2085 | if (failrec->in_validation) { | |
2086 | /* there was no real error, just free the record */ | |
2087 | pr_debug("clean_io_failure: freeing dummy error at %llu\n", | |
2088 | failrec->start); | |
2089 | did_repair = 1; | |
2090 | goto out; | |
2091 | } | |
2092 | ||
2093 | spin_lock(&BTRFS_I(inode)->io_tree.lock); | |
2094 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, | |
2095 | failrec->start, | |
2096 | EXTENT_LOCKED); | |
2097 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); | |
2098 | ||
883d0de4 MX |
2099 | if (state && state->start <= failrec->start && |
2100 | state->end >= failrec->start + failrec->len - 1) { | |
3ec706c8 SB |
2101 | fs_info = BTRFS_I(inode)->root->fs_info; |
2102 | num_copies = btrfs_num_copies(fs_info, failrec->logical, | |
2103 | failrec->len); | |
4a54c8c1 | 2104 | if (num_copies > 1) { |
3ec706c8 | 2105 | ret = repair_io_failure(fs_info, start, failrec->len, |
4a54c8c1 JS |
2106 | failrec->logical, page, |
2107 | failrec->failed_mirror); | |
2108 | did_repair = !ret; | |
2109 | } | |
53b381b3 | 2110 | ret = 0; |
4a54c8c1 JS |
2111 | } |
2112 | ||
2113 | out: | |
2114 | if (!ret) | |
2115 | ret = free_io_failure(inode, failrec, did_repair); | |
2116 | ||
2117 | return ret; | |
2118 | } | |
2119 | ||
2120 | /* | |
2121 | * this is a generic handler for readpage errors (default | |
2122 | * readpage_io_failed_hook). if other copies exist, read those and write back | |
2123 | * good data to the failed position. does not investigate in remapping the | |
2124 | * failed extent elsewhere, hoping the device will be smart enough to do this as | |
2125 | * needed | |
2126 | */ | |
2127 | ||
facc8a22 MX |
2128 | static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset, |
2129 | struct page *page, u64 start, u64 end, | |
2130 | int failed_mirror) | |
4a54c8c1 JS |
2131 | { |
2132 | struct io_failure_record *failrec = NULL; | |
2133 | u64 private; | |
2134 | struct extent_map *em; | |
2135 | struct inode *inode = page->mapping->host; | |
2136 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
2137 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2138 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
2139 | struct bio *bio; | |
facc8a22 MX |
2140 | struct btrfs_io_bio *btrfs_failed_bio; |
2141 | struct btrfs_io_bio *btrfs_bio; | |
4a54c8c1 JS |
2142 | int num_copies; |
2143 | int ret; | |
2144 | int read_mode; | |
2145 | u64 logical; | |
2146 | ||
2147 | BUG_ON(failed_bio->bi_rw & REQ_WRITE); | |
2148 | ||
2149 | ret = get_state_private(failure_tree, start, &private); | |
2150 | if (ret) { | |
2151 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2152 | if (!failrec) | |
2153 | return -ENOMEM; | |
2154 | failrec->start = start; | |
2155 | failrec->len = end - start + 1; | |
2156 | failrec->this_mirror = 0; | |
2157 | failrec->bio_flags = 0; | |
2158 | failrec->in_validation = 0; | |
2159 | ||
2160 | read_lock(&em_tree->lock); | |
2161 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2162 | if (!em) { | |
2163 | read_unlock(&em_tree->lock); | |
2164 | kfree(failrec); | |
2165 | return -EIO; | |
2166 | } | |
2167 | ||
2168 | if (em->start > start || em->start + em->len < start) { | |
2169 | free_extent_map(em); | |
2170 | em = NULL; | |
2171 | } | |
2172 | read_unlock(&em_tree->lock); | |
2173 | ||
7a2d6a64 | 2174 | if (!em) { |
4a54c8c1 JS |
2175 | kfree(failrec); |
2176 | return -EIO; | |
2177 | } | |
2178 | logical = start - em->start; | |
2179 | logical = em->block_start + logical; | |
2180 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2181 | logical = em->block_start; | |
2182 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2183 | extent_set_compress_type(&failrec->bio_flags, | |
2184 | em->compress_type); | |
2185 | } | |
2186 | pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, " | |
2187 | "len=%llu\n", logical, start, failrec->len); | |
2188 | failrec->logical = logical; | |
2189 | free_extent_map(em); | |
2190 | ||
2191 | /* set the bits in the private failure tree */ | |
2192 | ret = set_extent_bits(failure_tree, start, end, | |
2193 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
2194 | if (ret >= 0) | |
2195 | ret = set_state_private(failure_tree, start, | |
2196 | (u64)(unsigned long)failrec); | |
2197 | /* set the bits in the inode's tree */ | |
2198 | if (ret >= 0) | |
2199 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED, | |
2200 | GFP_NOFS); | |
2201 | if (ret < 0) { | |
2202 | kfree(failrec); | |
2203 | return ret; | |
2204 | } | |
2205 | } else { | |
2206 | failrec = (struct io_failure_record *)(unsigned long)private; | |
2207 | pr_debug("bio_readpage_error: (found) logical=%llu, " | |
2208 | "start=%llu, len=%llu, validation=%d\n", | |
2209 | failrec->logical, failrec->start, failrec->len, | |
2210 | failrec->in_validation); | |
2211 | /* | |
2212 | * when data can be on disk more than twice, add to failrec here | |
2213 | * (e.g. with a list for failed_mirror) to make | |
2214 | * clean_io_failure() clean all those errors at once. | |
2215 | */ | |
2216 | } | |
5d964051 SB |
2217 | num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info, |
2218 | failrec->logical, failrec->len); | |
4a54c8c1 JS |
2219 | if (num_copies == 1) { |
2220 | /* | |
2221 | * we only have a single copy of the data, so don't bother with | |
2222 | * all the retry and error correction code that follows. no | |
2223 | * matter what the error is, it is very likely to persist. | |
2224 | */ | |
09a7f7a2 MX |
2225 | pr_debug("bio_readpage_error: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n", |
2226 | num_copies, failrec->this_mirror, failed_mirror); | |
4a54c8c1 JS |
2227 | free_io_failure(inode, failrec, 0); |
2228 | return -EIO; | |
2229 | } | |
2230 | ||
4a54c8c1 JS |
2231 | /* |
2232 | * there are two premises: | |
2233 | * a) deliver good data to the caller | |
2234 | * b) correct the bad sectors on disk | |
2235 | */ | |
2236 | if (failed_bio->bi_vcnt > 1) { | |
2237 | /* | |
2238 | * to fulfill b), we need to know the exact failing sectors, as | |
2239 | * we don't want to rewrite any more than the failed ones. thus, | |
2240 | * we need separate read requests for the failed bio | |
2241 | * | |
2242 | * if the following BUG_ON triggers, our validation request got | |
2243 | * merged. we need separate requests for our algorithm to work. | |
2244 | */ | |
2245 | BUG_ON(failrec->in_validation); | |
2246 | failrec->in_validation = 1; | |
2247 | failrec->this_mirror = failed_mirror; | |
2248 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; | |
2249 | } else { | |
2250 | /* | |
2251 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2252 | * of the failed sector and if we succeed, we have setup | |
2253 | * everything for repair_io_failure to do the rest for us. | |
2254 | */ | |
2255 | if (failrec->in_validation) { | |
2256 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2257 | failrec->in_validation = 0; | |
2258 | failrec->this_mirror = 0; | |
2259 | } | |
2260 | failrec->failed_mirror = failed_mirror; | |
2261 | failrec->this_mirror++; | |
2262 | if (failrec->this_mirror == failed_mirror) | |
2263 | failrec->this_mirror++; | |
2264 | read_mode = READ_SYNC; | |
2265 | } | |
2266 | ||
facc8a22 MX |
2267 | if (failrec->this_mirror > num_copies) { |
2268 | pr_debug("bio_readpage_error: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n", | |
4a54c8c1 JS |
2269 | num_copies, failrec->this_mirror, failed_mirror); |
2270 | free_io_failure(inode, failrec, 0); | |
2271 | return -EIO; | |
2272 | } | |
2273 | ||
9be3395b | 2274 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
e627ee7b TI |
2275 | if (!bio) { |
2276 | free_io_failure(inode, failrec, 0); | |
2277 | return -EIO; | |
2278 | } | |
4a54c8c1 JS |
2279 | bio->bi_end_io = failed_bio->bi_end_io; |
2280 | bio->bi_sector = failrec->logical >> 9; | |
2281 | bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; | |
2282 | bio->bi_size = 0; | |
2283 | ||
facc8a22 MX |
2284 | btrfs_failed_bio = btrfs_io_bio(failed_bio); |
2285 | if (btrfs_failed_bio->csum) { | |
2286 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2287 | u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); | |
2288 | ||
2289 | btrfs_bio = btrfs_io_bio(bio); | |
2290 | btrfs_bio->csum = btrfs_bio->csum_inline; | |
2291 | phy_offset >>= inode->i_sb->s_blocksize_bits; | |
2292 | phy_offset *= csum_size; | |
2293 | memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + phy_offset, | |
2294 | csum_size); | |
2295 | } | |
2296 | ||
4a54c8c1 JS |
2297 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); |
2298 | ||
2299 | pr_debug("bio_readpage_error: submitting new read[%#x] to " | |
2300 | "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode, | |
2301 | failrec->this_mirror, num_copies, failrec->in_validation); | |
2302 | ||
013bd4c3 TI |
2303 | ret = tree->ops->submit_bio_hook(inode, read_mode, bio, |
2304 | failrec->this_mirror, | |
2305 | failrec->bio_flags, 0); | |
2306 | return ret; | |
4a54c8c1 JS |
2307 | } |
2308 | ||
d1310b2e CM |
2309 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2310 | ||
87826df0 JM |
2311 | int end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
2312 | { | |
2313 | int uptodate = (err == 0); | |
2314 | struct extent_io_tree *tree; | |
2315 | int ret; | |
2316 | ||
2317 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
2318 | ||
2319 | if (tree->ops && tree->ops->writepage_end_io_hook) { | |
2320 | ret = tree->ops->writepage_end_io_hook(page, start, | |
2321 | end, NULL, uptodate); | |
2322 | if (ret) | |
2323 | uptodate = 0; | |
2324 | } | |
2325 | ||
87826df0 | 2326 | if (!uptodate) { |
87826df0 JM |
2327 | ClearPageUptodate(page); |
2328 | SetPageError(page); | |
2329 | } | |
2330 | return 0; | |
2331 | } | |
2332 | ||
d1310b2e CM |
2333 | /* |
2334 | * after a writepage IO is done, we need to: | |
2335 | * clear the uptodate bits on error | |
2336 | * clear the writeback bits in the extent tree for this IO | |
2337 | * end_page_writeback if the page has no more pending IO | |
2338 | * | |
2339 | * Scheduling is not allowed, so the extent state tree is expected | |
2340 | * to have one and only one object corresponding to this IO. | |
2341 | */ | |
d1310b2e | 2342 | static void end_bio_extent_writepage(struct bio *bio, int err) |
d1310b2e | 2343 | { |
d1310b2e | 2344 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
902b22f3 | 2345 | struct extent_io_tree *tree; |
d1310b2e CM |
2346 | u64 start; |
2347 | u64 end; | |
d1310b2e | 2348 | |
d1310b2e CM |
2349 | do { |
2350 | struct page *page = bvec->bv_page; | |
902b22f3 DW |
2351 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
2352 | ||
17a5adcc AO |
2353 | /* We always issue full-page reads, but if some block |
2354 | * in a page fails to read, blk_update_request() will | |
2355 | * advance bv_offset and adjust bv_len to compensate. | |
2356 | * Print a warning for nonzero offsets, and an error | |
2357 | * if they don't add up to a full page. */ | |
2358 | if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) | |
2359 | printk("%s page write in btrfs with offset %u and length %u\n", | |
2360 | bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE | |
2361 | ? KERN_ERR "partial" : KERN_INFO "incomplete", | |
2362 | bvec->bv_offset, bvec->bv_len); | |
d1310b2e | 2363 | |
17a5adcc AO |
2364 | start = page_offset(page); |
2365 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
d1310b2e CM |
2366 | |
2367 | if (--bvec >= bio->bi_io_vec) | |
2368 | prefetchw(&bvec->bv_page->flags); | |
1259ab75 | 2369 | |
87826df0 JM |
2370 | if (end_extent_writepage(page, err, start, end)) |
2371 | continue; | |
70dec807 | 2372 | |
17a5adcc | 2373 | end_page_writeback(page); |
d1310b2e | 2374 | } while (bvec >= bio->bi_io_vec); |
2b1f55b0 | 2375 | |
d1310b2e | 2376 | bio_put(bio); |
d1310b2e CM |
2377 | } |
2378 | ||
883d0de4 MX |
2379 | static void |
2380 | endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len, | |
2381 | int uptodate) | |
2382 | { | |
2383 | struct extent_state *cached = NULL; | |
2384 | u64 end = start + len - 1; | |
2385 | ||
2386 | if (uptodate && tree->track_uptodate) | |
2387 | set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC); | |
2388 | unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC); | |
2389 | } | |
2390 | ||
d1310b2e CM |
2391 | /* |
2392 | * after a readpage IO is done, we need to: | |
2393 | * clear the uptodate bits on error | |
2394 | * set the uptodate bits if things worked | |
2395 | * set the page up to date if all extents in the tree are uptodate | |
2396 | * clear the lock bit in the extent tree | |
2397 | * unlock the page if there are no other extents locked for it | |
2398 | * | |
2399 | * Scheduling is not allowed, so the extent state tree is expected | |
2400 | * to have one and only one object corresponding to this IO. | |
2401 | */ | |
d1310b2e | 2402 | static void end_bio_extent_readpage(struct bio *bio, int err) |
d1310b2e CM |
2403 | { |
2404 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
4125bf76 CM |
2405 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
2406 | struct bio_vec *bvec = bio->bi_io_vec; | |
facc8a22 | 2407 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); |
902b22f3 | 2408 | struct extent_io_tree *tree; |
facc8a22 | 2409 | u64 offset = 0; |
d1310b2e CM |
2410 | u64 start; |
2411 | u64 end; | |
facc8a22 | 2412 | u64 len; |
883d0de4 MX |
2413 | u64 extent_start = 0; |
2414 | u64 extent_len = 0; | |
5cf1ab56 | 2415 | int mirror; |
d1310b2e CM |
2416 | int ret; |
2417 | ||
d20f7043 CM |
2418 | if (err) |
2419 | uptodate = 0; | |
2420 | ||
d1310b2e CM |
2421 | do { |
2422 | struct page *page = bvec->bv_page; | |
a71754fc | 2423 | struct inode *inode = page->mapping->host; |
507903b8 | 2424 | |
be3940c0 | 2425 | pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, " |
9be3395b CM |
2426 | "mirror=%lu\n", (u64)bio->bi_sector, err, |
2427 | io_bio->mirror_num); | |
a71754fc | 2428 | tree = &BTRFS_I(inode)->io_tree; |
902b22f3 | 2429 | |
17a5adcc AO |
2430 | /* We always issue full-page reads, but if some block |
2431 | * in a page fails to read, blk_update_request() will | |
2432 | * advance bv_offset and adjust bv_len to compensate. | |
2433 | * Print a warning for nonzero offsets, and an error | |
2434 | * if they don't add up to a full page. */ | |
2435 | if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) | |
2436 | printk("%s page read in btrfs with offset %u and length %u\n", | |
2437 | bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE | |
2438 | ? KERN_ERR "partial" : KERN_INFO "incomplete", | |
2439 | bvec->bv_offset, bvec->bv_len); | |
d1310b2e | 2440 | |
17a5adcc AO |
2441 | start = page_offset(page); |
2442 | end = start + bvec->bv_offset + bvec->bv_len - 1; | |
facc8a22 | 2443 | len = bvec->bv_len; |
d1310b2e | 2444 | |
4125bf76 | 2445 | if (++bvec <= bvec_end) |
d1310b2e CM |
2446 | prefetchw(&bvec->bv_page->flags); |
2447 | ||
9be3395b | 2448 | mirror = io_bio->mirror_num; |
f2a09da9 MX |
2449 | if (likely(uptodate && tree->ops && |
2450 | tree->ops->readpage_end_io_hook)) { | |
facc8a22 MX |
2451 | ret = tree->ops->readpage_end_io_hook(io_bio, offset, |
2452 | page, start, end, | |
2453 | mirror); | |
5ee0844d | 2454 | if (ret) |
d1310b2e | 2455 | uptodate = 0; |
5ee0844d | 2456 | else |
4a54c8c1 | 2457 | clean_io_failure(start, page); |
d1310b2e | 2458 | } |
ea466794 | 2459 | |
f2a09da9 MX |
2460 | if (likely(uptodate)) |
2461 | goto readpage_ok; | |
2462 | ||
2463 | if (tree->ops && tree->ops->readpage_io_failed_hook) { | |
5cf1ab56 | 2464 | ret = tree->ops->readpage_io_failed_hook(page, mirror); |
ea466794 JB |
2465 | if (!ret && !err && |
2466 | test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
2467 | uptodate = 1; | |
f2a09da9 | 2468 | } else { |
f4a8e656 JS |
2469 | /* |
2470 | * The generic bio_readpage_error handles errors the | |
2471 | * following way: If possible, new read requests are | |
2472 | * created and submitted and will end up in | |
2473 | * end_bio_extent_readpage as well (if we're lucky, not | |
2474 | * in the !uptodate case). In that case it returns 0 and | |
2475 | * we just go on with the next page in our bio. If it | |
2476 | * can't handle the error it will return -EIO and we | |
2477 | * remain responsible for that page. | |
2478 | */ | |
facc8a22 MX |
2479 | ret = bio_readpage_error(bio, offset, page, start, end, |
2480 | mirror); | |
7e38326f | 2481 | if (ret == 0) { |
3b951516 CM |
2482 | uptodate = |
2483 | test_bit(BIO_UPTODATE, &bio->bi_flags); | |
d20f7043 CM |
2484 | if (err) |
2485 | uptodate = 0; | |
7e38326f CM |
2486 | continue; |
2487 | } | |
2488 | } | |
f2a09da9 | 2489 | readpage_ok: |
883d0de4 | 2490 | if (likely(uptodate)) { |
a71754fc JB |
2491 | loff_t i_size = i_size_read(inode); |
2492 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
2493 | unsigned offset; | |
2494 | ||
2495 | /* Zero out the end if this page straddles i_size */ | |
2496 | offset = i_size & (PAGE_CACHE_SIZE-1); | |
2497 | if (page->index == end_index && offset) | |
2498 | zero_user_segment(page, offset, PAGE_CACHE_SIZE); | |
17a5adcc | 2499 | SetPageUptodate(page); |
70dec807 | 2500 | } else { |
17a5adcc AO |
2501 | ClearPageUptodate(page); |
2502 | SetPageError(page); | |
70dec807 | 2503 | } |
17a5adcc | 2504 | unlock_page(page); |
facc8a22 | 2505 | offset += len; |
883d0de4 MX |
2506 | |
2507 | if (unlikely(!uptodate)) { | |
2508 | if (extent_len) { | |
2509 | endio_readpage_release_extent(tree, | |
2510 | extent_start, | |
2511 | extent_len, 1); | |
2512 | extent_start = 0; | |
2513 | extent_len = 0; | |
2514 | } | |
2515 | endio_readpage_release_extent(tree, start, | |
2516 | end - start + 1, 0); | |
2517 | } else if (!extent_len) { | |
2518 | extent_start = start; | |
2519 | extent_len = end + 1 - start; | |
2520 | } else if (extent_start + extent_len == start) { | |
2521 | extent_len += end + 1 - start; | |
2522 | } else { | |
2523 | endio_readpage_release_extent(tree, extent_start, | |
2524 | extent_len, uptodate); | |
2525 | extent_start = start; | |
2526 | extent_len = end + 1 - start; | |
2527 | } | |
4125bf76 | 2528 | } while (bvec <= bvec_end); |
d1310b2e | 2529 | |
883d0de4 MX |
2530 | if (extent_len) |
2531 | endio_readpage_release_extent(tree, extent_start, extent_len, | |
2532 | uptodate); | |
facc8a22 MX |
2533 | if (io_bio->end_io) |
2534 | io_bio->end_io(io_bio, err); | |
d1310b2e | 2535 | bio_put(bio); |
d1310b2e CM |
2536 | } |
2537 | ||
9be3395b CM |
2538 | /* |
2539 | * this allocates from the btrfs_bioset. We're returning a bio right now | |
2540 | * but you can call btrfs_io_bio for the appropriate container_of magic | |
2541 | */ | |
88f794ed MX |
2542 | struct bio * |
2543 | btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
2544 | gfp_t gfp_flags) | |
d1310b2e | 2545 | { |
facc8a22 | 2546 | struct btrfs_io_bio *btrfs_bio; |
d1310b2e CM |
2547 | struct bio *bio; |
2548 | ||
9be3395b | 2549 | bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset); |
d1310b2e CM |
2550 | |
2551 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
9be3395b CM |
2552 | while (!bio && (nr_vecs /= 2)) { |
2553 | bio = bio_alloc_bioset(gfp_flags, | |
2554 | nr_vecs, btrfs_bioset); | |
2555 | } | |
d1310b2e CM |
2556 | } |
2557 | ||
2558 | if (bio) { | |
e1c4b745 | 2559 | bio->bi_size = 0; |
d1310b2e CM |
2560 | bio->bi_bdev = bdev; |
2561 | bio->bi_sector = first_sector; | |
facc8a22 MX |
2562 | btrfs_bio = btrfs_io_bio(bio); |
2563 | btrfs_bio->csum = NULL; | |
2564 | btrfs_bio->csum_allocated = NULL; | |
2565 | btrfs_bio->end_io = NULL; | |
d1310b2e CM |
2566 | } |
2567 | return bio; | |
2568 | } | |
2569 | ||
9be3395b CM |
2570 | struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask) |
2571 | { | |
2572 | return bio_clone_bioset(bio, gfp_mask, btrfs_bioset); | |
2573 | } | |
2574 | ||
2575 | ||
2576 | /* this also allocates from the btrfs_bioset */ | |
2577 | struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs) | |
2578 | { | |
facc8a22 MX |
2579 | struct btrfs_io_bio *btrfs_bio; |
2580 | struct bio *bio; | |
2581 | ||
2582 | bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset); | |
2583 | if (bio) { | |
2584 | btrfs_bio = btrfs_io_bio(bio); | |
2585 | btrfs_bio->csum = NULL; | |
2586 | btrfs_bio->csum_allocated = NULL; | |
2587 | btrfs_bio->end_io = NULL; | |
2588 | } | |
2589 | return bio; | |
9be3395b CM |
2590 | } |
2591 | ||
2592 | ||
355808c2 JM |
2593 | static int __must_check submit_one_bio(int rw, struct bio *bio, |
2594 | int mirror_num, unsigned long bio_flags) | |
d1310b2e | 2595 | { |
d1310b2e | 2596 | int ret = 0; |
70dec807 CM |
2597 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
2598 | struct page *page = bvec->bv_page; | |
2599 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 2600 | u64 start; |
70dec807 | 2601 | |
4eee4fa4 | 2602 | start = page_offset(page) + bvec->bv_offset; |
70dec807 | 2603 | |
902b22f3 | 2604 | bio->bi_private = NULL; |
d1310b2e CM |
2605 | |
2606 | bio_get(bio); | |
2607 | ||
065631f6 | 2608 | if (tree->ops && tree->ops->submit_bio_hook) |
6b82ce8d | 2609 | ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio, |
eaf25d93 | 2610 | mirror_num, bio_flags, start); |
0b86a832 | 2611 | else |
21adbd5c | 2612 | btrfsic_submit_bio(rw, bio); |
4a54c8c1 | 2613 | |
d1310b2e CM |
2614 | if (bio_flagged(bio, BIO_EOPNOTSUPP)) |
2615 | ret = -EOPNOTSUPP; | |
2616 | bio_put(bio); | |
2617 | return ret; | |
2618 | } | |
2619 | ||
64a16701 | 2620 | static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page, |
3444a972 JM |
2621 | unsigned long offset, size_t size, struct bio *bio, |
2622 | unsigned long bio_flags) | |
2623 | { | |
2624 | int ret = 0; | |
2625 | if (tree->ops && tree->ops->merge_bio_hook) | |
64a16701 | 2626 | ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio, |
3444a972 JM |
2627 | bio_flags); |
2628 | BUG_ON(ret < 0); | |
2629 | return ret; | |
2630 | ||
2631 | } | |
2632 | ||
d1310b2e CM |
2633 | static int submit_extent_page(int rw, struct extent_io_tree *tree, |
2634 | struct page *page, sector_t sector, | |
2635 | size_t size, unsigned long offset, | |
2636 | struct block_device *bdev, | |
2637 | struct bio **bio_ret, | |
2638 | unsigned long max_pages, | |
f188591e | 2639 | bio_end_io_t end_io_func, |
c8b97818 CM |
2640 | int mirror_num, |
2641 | unsigned long prev_bio_flags, | |
2642 | unsigned long bio_flags) | |
d1310b2e CM |
2643 | { |
2644 | int ret = 0; | |
2645 | struct bio *bio; | |
2646 | int nr; | |
c8b97818 CM |
2647 | int contig = 0; |
2648 | int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED; | |
2649 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; | |
5b050f04 | 2650 | size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE); |
d1310b2e CM |
2651 | |
2652 | if (bio_ret && *bio_ret) { | |
2653 | bio = *bio_ret; | |
c8b97818 CM |
2654 | if (old_compressed) |
2655 | contig = bio->bi_sector == sector; | |
2656 | else | |
f73a1c7d | 2657 | contig = bio_end_sector(bio) == sector; |
c8b97818 CM |
2658 | |
2659 | if (prev_bio_flags != bio_flags || !contig || | |
64a16701 | 2660 | merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) || |
c8b97818 CM |
2661 | bio_add_page(bio, page, page_size, offset) < page_size) { |
2662 | ret = submit_one_bio(rw, bio, mirror_num, | |
2663 | prev_bio_flags); | |
79787eaa JM |
2664 | if (ret < 0) |
2665 | return ret; | |
d1310b2e CM |
2666 | bio = NULL; |
2667 | } else { | |
2668 | return 0; | |
2669 | } | |
2670 | } | |
c8b97818 CM |
2671 | if (this_compressed) |
2672 | nr = BIO_MAX_PAGES; | |
2673 | else | |
2674 | nr = bio_get_nr_vecs(bdev); | |
2675 | ||
88f794ed | 2676 | bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH); |
5df67083 TI |
2677 | if (!bio) |
2678 | return -ENOMEM; | |
70dec807 | 2679 | |
c8b97818 | 2680 | bio_add_page(bio, page, page_size, offset); |
d1310b2e CM |
2681 | bio->bi_end_io = end_io_func; |
2682 | bio->bi_private = tree; | |
70dec807 | 2683 | |
d397712b | 2684 | if (bio_ret) |
d1310b2e | 2685 | *bio_ret = bio; |
d397712b | 2686 | else |
c8b97818 | 2687 | ret = submit_one_bio(rw, bio, mirror_num, bio_flags); |
d1310b2e CM |
2688 | |
2689 | return ret; | |
2690 | } | |
2691 | ||
48a3b636 ES |
2692 | static void attach_extent_buffer_page(struct extent_buffer *eb, |
2693 | struct page *page) | |
d1310b2e CM |
2694 | { |
2695 | if (!PagePrivate(page)) { | |
2696 | SetPagePrivate(page); | |
d1310b2e | 2697 | page_cache_get(page); |
4f2de97a JB |
2698 | set_page_private(page, (unsigned long)eb); |
2699 | } else { | |
2700 | WARN_ON(page->private != (unsigned long)eb); | |
d1310b2e CM |
2701 | } |
2702 | } | |
2703 | ||
4f2de97a | 2704 | void set_page_extent_mapped(struct page *page) |
d1310b2e | 2705 | { |
4f2de97a JB |
2706 | if (!PagePrivate(page)) { |
2707 | SetPagePrivate(page); | |
2708 | page_cache_get(page); | |
2709 | set_page_private(page, EXTENT_PAGE_PRIVATE); | |
2710 | } | |
d1310b2e CM |
2711 | } |
2712 | ||
125bac01 MX |
2713 | static struct extent_map * |
2714 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
2715 | u64 start, u64 len, get_extent_t *get_extent, | |
2716 | struct extent_map **em_cached) | |
2717 | { | |
2718 | struct extent_map *em; | |
2719 | ||
2720 | if (em_cached && *em_cached) { | |
2721 | em = *em_cached; | |
2722 | if (em->in_tree && start >= em->start && | |
2723 | start < extent_map_end(em)) { | |
2724 | atomic_inc(&em->refs); | |
2725 | return em; | |
2726 | } | |
2727 | ||
2728 | free_extent_map(em); | |
2729 | *em_cached = NULL; | |
2730 | } | |
2731 | ||
2732 | em = get_extent(inode, page, pg_offset, start, len, 0); | |
2733 | if (em_cached && !IS_ERR_OR_NULL(em)) { | |
2734 | BUG_ON(*em_cached); | |
2735 | atomic_inc(&em->refs); | |
2736 | *em_cached = em; | |
2737 | } | |
2738 | return em; | |
2739 | } | |
d1310b2e CM |
2740 | /* |
2741 | * basic readpage implementation. Locked extent state structs are inserted | |
2742 | * into the tree that are removed when the IO is done (by the end_io | |
2743 | * handlers) | |
79787eaa | 2744 | * XXX JDM: This needs looking at to ensure proper page locking |
d1310b2e | 2745 | */ |
9974090b MX |
2746 | static int __do_readpage(struct extent_io_tree *tree, |
2747 | struct page *page, | |
2748 | get_extent_t *get_extent, | |
125bac01 | 2749 | struct extent_map **em_cached, |
9974090b MX |
2750 | struct bio **bio, int mirror_num, |
2751 | unsigned long *bio_flags, int rw) | |
d1310b2e CM |
2752 | { |
2753 | struct inode *inode = page->mapping->host; | |
4eee4fa4 | 2754 | u64 start = page_offset(page); |
d1310b2e CM |
2755 | u64 page_end = start + PAGE_CACHE_SIZE - 1; |
2756 | u64 end; | |
2757 | u64 cur = start; | |
2758 | u64 extent_offset; | |
2759 | u64 last_byte = i_size_read(inode); | |
2760 | u64 block_start; | |
2761 | u64 cur_end; | |
2762 | sector_t sector; | |
2763 | struct extent_map *em; | |
2764 | struct block_device *bdev; | |
2765 | int ret; | |
2766 | int nr = 0; | |
4b384318 | 2767 | int parent_locked = *bio_flags & EXTENT_BIO_PARENT_LOCKED; |
306e16ce | 2768 | size_t pg_offset = 0; |
d1310b2e | 2769 | size_t iosize; |
c8b97818 | 2770 | size_t disk_io_size; |
d1310b2e | 2771 | size_t blocksize = inode->i_sb->s_blocksize; |
4b384318 | 2772 | unsigned long this_bio_flag = *bio_flags & EXTENT_BIO_PARENT_LOCKED; |
d1310b2e CM |
2773 | |
2774 | set_page_extent_mapped(page); | |
2775 | ||
9974090b | 2776 | end = page_end; |
90a887c9 DM |
2777 | if (!PageUptodate(page)) { |
2778 | if (cleancache_get_page(page) == 0) { | |
2779 | BUG_ON(blocksize != PAGE_SIZE); | |
9974090b | 2780 | unlock_extent(tree, start, end); |
90a887c9 DM |
2781 | goto out; |
2782 | } | |
2783 | } | |
2784 | ||
c8b97818 CM |
2785 | if (page->index == last_byte >> PAGE_CACHE_SHIFT) { |
2786 | char *userpage; | |
2787 | size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1); | |
2788 | ||
2789 | if (zero_offset) { | |
2790 | iosize = PAGE_CACHE_SIZE - zero_offset; | |
7ac687d9 | 2791 | userpage = kmap_atomic(page); |
c8b97818 CM |
2792 | memset(userpage + zero_offset, 0, iosize); |
2793 | flush_dcache_page(page); | |
7ac687d9 | 2794 | kunmap_atomic(userpage); |
c8b97818 CM |
2795 | } |
2796 | } | |
d1310b2e | 2797 | while (cur <= end) { |
c8f2f24b JB |
2798 | unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1; |
2799 | ||
d1310b2e CM |
2800 | if (cur >= last_byte) { |
2801 | char *userpage; | |
507903b8 AJ |
2802 | struct extent_state *cached = NULL; |
2803 | ||
306e16ce | 2804 | iosize = PAGE_CACHE_SIZE - pg_offset; |
7ac687d9 | 2805 | userpage = kmap_atomic(page); |
306e16ce | 2806 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2807 | flush_dcache_page(page); |
7ac687d9 | 2808 | kunmap_atomic(userpage); |
d1310b2e | 2809 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 2810 | &cached, GFP_NOFS); |
4b384318 MF |
2811 | if (!parent_locked) |
2812 | unlock_extent_cached(tree, cur, | |
2813 | cur + iosize - 1, | |
2814 | &cached, GFP_NOFS); | |
d1310b2e CM |
2815 | break; |
2816 | } | |
125bac01 MX |
2817 | em = __get_extent_map(inode, page, pg_offset, cur, |
2818 | end - cur + 1, get_extent, em_cached); | |
c704005d | 2819 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 2820 | SetPageError(page); |
4b384318 MF |
2821 | if (!parent_locked) |
2822 | unlock_extent(tree, cur, end); | |
d1310b2e CM |
2823 | break; |
2824 | } | |
d1310b2e CM |
2825 | extent_offset = cur - em->start; |
2826 | BUG_ON(extent_map_end(em) <= cur); | |
2827 | BUG_ON(end < cur); | |
2828 | ||
261507a0 | 2829 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
4b384318 | 2830 | this_bio_flag |= EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
2831 | extent_set_compress_type(&this_bio_flag, |
2832 | em->compress_type); | |
2833 | } | |
c8b97818 | 2834 | |
d1310b2e CM |
2835 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
2836 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 2837 | iosize = ALIGN(iosize, blocksize); |
c8b97818 CM |
2838 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
2839 | disk_io_size = em->block_len; | |
2840 | sector = em->block_start >> 9; | |
2841 | } else { | |
2842 | sector = (em->block_start + extent_offset) >> 9; | |
2843 | disk_io_size = iosize; | |
2844 | } | |
d1310b2e CM |
2845 | bdev = em->bdev; |
2846 | block_start = em->block_start; | |
d899e052 YZ |
2847 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2848 | block_start = EXTENT_MAP_HOLE; | |
d1310b2e CM |
2849 | free_extent_map(em); |
2850 | em = NULL; | |
2851 | ||
2852 | /* we've found a hole, just zero and go on */ | |
2853 | if (block_start == EXTENT_MAP_HOLE) { | |
2854 | char *userpage; | |
507903b8 AJ |
2855 | struct extent_state *cached = NULL; |
2856 | ||
7ac687d9 | 2857 | userpage = kmap_atomic(page); |
306e16ce | 2858 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2859 | flush_dcache_page(page); |
7ac687d9 | 2860 | kunmap_atomic(userpage); |
d1310b2e CM |
2861 | |
2862 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 AJ |
2863 | &cached, GFP_NOFS); |
2864 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
2865 | &cached, GFP_NOFS); | |
d1310b2e | 2866 | cur = cur + iosize; |
306e16ce | 2867 | pg_offset += iosize; |
d1310b2e CM |
2868 | continue; |
2869 | } | |
2870 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
2871 | if (test_range_bit(tree, cur, cur_end, |
2872 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 2873 | check_page_uptodate(tree, page); |
4b384318 MF |
2874 | if (!parent_locked) |
2875 | unlock_extent(tree, cur, cur + iosize - 1); | |
d1310b2e | 2876 | cur = cur + iosize; |
306e16ce | 2877 | pg_offset += iosize; |
d1310b2e CM |
2878 | continue; |
2879 | } | |
70dec807 CM |
2880 | /* we have an inline extent but it didn't get marked up |
2881 | * to date. Error out | |
2882 | */ | |
2883 | if (block_start == EXTENT_MAP_INLINE) { | |
2884 | SetPageError(page); | |
4b384318 MF |
2885 | if (!parent_locked) |
2886 | unlock_extent(tree, cur, cur + iosize - 1); | |
70dec807 | 2887 | cur = cur + iosize; |
306e16ce | 2888 | pg_offset += iosize; |
70dec807 CM |
2889 | continue; |
2890 | } | |
d1310b2e | 2891 | |
c8f2f24b | 2892 | pnr -= page->index; |
d4c7ca86 | 2893 | ret = submit_extent_page(rw, tree, page, |
306e16ce | 2894 | sector, disk_io_size, pg_offset, |
89642229 | 2895 | bdev, bio, pnr, |
c8b97818 CM |
2896 | end_bio_extent_readpage, mirror_num, |
2897 | *bio_flags, | |
2898 | this_bio_flag); | |
c8f2f24b JB |
2899 | if (!ret) { |
2900 | nr++; | |
2901 | *bio_flags = this_bio_flag; | |
2902 | } else { | |
d1310b2e | 2903 | SetPageError(page); |
4b384318 MF |
2904 | if (!parent_locked) |
2905 | unlock_extent(tree, cur, cur + iosize - 1); | |
edd33c99 | 2906 | } |
d1310b2e | 2907 | cur = cur + iosize; |
306e16ce | 2908 | pg_offset += iosize; |
d1310b2e | 2909 | } |
90a887c9 | 2910 | out: |
d1310b2e CM |
2911 | if (!nr) { |
2912 | if (!PageError(page)) | |
2913 | SetPageUptodate(page); | |
2914 | unlock_page(page); | |
2915 | } | |
2916 | return 0; | |
2917 | } | |
2918 | ||
9974090b MX |
2919 | static inline void __do_contiguous_readpages(struct extent_io_tree *tree, |
2920 | struct page *pages[], int nr_pages, | |
2921 | u64 start, u64 end, | |
2922 | get_extent_t *get_extent, | |
125bac01 | 2923 | struct extent_map **em_cached, |
9974090b MX |
2924 | struct bio **bio, int mirror_num, |
2925 | unsigned long *bio_flags, int rw) | |
2926 | { | |
2927 | struct inode *inode; | |
2928 | struct btrfs_ordered_extent *ordered; | |
2929 | int index; | |
2930 | ||
2931 | inode = pages[0]->mapping->host; | |
2932 | while (1) { | |
2933 | lock_extent(tree, start, end); | |
2934 | ordered = btrfs_lookup_ordered_range(inode, start, | |
2935 | end - start + 1); | |
2936 | if (!ordered) | |
2937 | break; | |
2938 | unlock_extent(tree, start, end); | |
2939 | btrfs_start_ordered_extent(inode, ordered, 1); | |
2940 | btrfs_put_ordered_extent(ordered); | |
2941 | } | |
2942 | ||
2943 | for (index = 0; index < nr_pages; index++) { | |
125bac01 MX |
2944 | __do_readpage(tree, pages[index], get_extent, em_cached, bio, |
2945 | mirror_num, bio_flags, rw); | |
9974090b MX |
2946 | page_cache_release(pages[index]); |
2947 | } | |
2948 | } | |
2949 | ||
2950 | static void __extent_readpages(struct extent_io_tree *tree, | |
2951 | struct page *pages[], | |
2952 | int nr_pages, get_extent_t *get_extent, | |
125bac01 | 2953 | struct extent_map **em_cached, |
9974090b MX |
2954 | struct bio **bio, int mirror_num, |
2955 | unsigned long *bio_flags, int rw) | |
2956 | { | |
35a3621b | 2957 | u64 start = 0; |
9974090b MX |
2958 | u64 end = 0; |
2959 | u64 page_start; | |
2960 | int index; | |
35a3621b | 2961 | int first_index = 0; |
9974090b MX |
2962 | |
2963 | for (index = 0; index < nr_pages; index++) { | |
2964 | page_start = page_offset(pages[index]); | |
2965 | if (!end) { | |
2966 | start = page_start; | |
2967 | end = start + PAGE_CACHE_SIZE - 1; | |
2968 | first_index = index; | |
2969 | } else if (end + 1 == page_start) { | |
2970 | end += PAGE_CACHE_SIZE; | |
2971 | } else { | |
2972 | __do_contiguous_readpages(tree, &pages[first_index], | |
2973 | index - first_index, start, | |
125bac01 MX |
2974 | end, get_extent, em_cached, |
2975 | bio, mirror_num, bio_flags, | |
2976 | rw); | |
9974090b MX |
2977 | start = page_start; |
2978 | end = start + PAGE_CACHE_SIZE - 1; | |
2979 | first_index = index; | |
2980 | } | |
2981 | } | |
2982 | ||
2983 | if (end) | |
2984 | __do_contiguous_readpages(tree, &pages[first_index], | |
2985 | index - first_index, start, | |
125bac01 | 2986 | end, get_extent, em_cached, bio, |
9974090b MX |
2987 | mirror_num, bio_flags, rw); |
2988 | } | |
2989 | ||
2990 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
2991 | struct page *page, | |
2992 | get_extent_t *get_extent, | |
2993 | struct bio **bio, int mirror_num, | |
2994 | unsigned long *bio_flags, int rw) | |
2995 | { | |
2996 | struct inode *inode = page->mapping->host; | |
2997 | struct btrfs_ordered_extent *ordered; | |
2998 | u64 start = page_offset(page); | |
2999 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
3000 | int ret; | |
3001 | ||
3002 | while (1) { | |
3003 | lock_extent(tree, start, end); | |
3004 | ordered = btrfs_lookup_ordered_extent(inode, start); | |
3005 | if (!ordered) | |
3006 | break; | |
3007 | unlock_extent(tree, start, end); | |
3008 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3009 | btrfs_put_ordered_extent(ordered); | |
3010 | } | |
3011 | ||
125bac01 MX |
3012 | ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num, |
3013 | bio_flags, rw); | |
9974090b MX |
3014 | return ret; |
3015 | } | |
3016 | ||
d1310b2e | 3017 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, |
8ddc7d9c | 3018 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3019 | { |
3020 | struct bio *bio = NULL; | |
c8b97818 | 3021 | unsigned long bio_flags = 0; |
d1310b2e CM |
3022 | int ret; |
3023 | ||
8ddc7d9c | 3024 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, |
d4c7ca86 | 3025 | &bio_flags, READ); |
d1310b2e | 3026 | if (bio) |
8ddc7d9c | 3027 | ret = submit_one_bio(READ, bio, mirror_num, bio_flags); |
d1310b2e CM |
3028 | return ret; |
3029 | } | |
d1310b2e | 3030 | |
4b384318 MF |
3031 | int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page, |
3032 | get_extent_t *get_extent, int mirror_num) | |
3033 | { | |
3034 | struct bio *bio = NULL; | |
3035 | unsigned long bio_flags = EXTENT_BIO_PARENT_LOCKED; | |
3036 | int ret; | |
3037 | ||
3038 | ret = __do_readpage(tree, page, get_extent, NULL, &bio, mirror_num, | |
3039 | &bio_flags, READ); | |
3040 | if (bio) | |
3041 | ret = submit_one_bio(READ, bio, mirror_num, bio_flags); | |
3042 | return ret; | |
3043 | } | |
3044 | ||
11c8349b CM |
3045 | static noinline void update_nr_written(struct page *page, |
3046 | struct writeback_control *wbc, | |
3047 | unsigned long nr_written) | |
3048 | { | |
3049 | wbc->nr_to_write -= nr_written; | |
3050 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && | |
3051 | wbc->range_start == 0 && wbc->range_end == LLONG_MAX)) | |
3052 | page->mapping->writeback_index = page->index + nr_written; | |
3053 | } | |
3054 | ||
d1310b2e CM |
3055 | /* |
3056 | * the writepage semantics are similar to regular writepage. extent | |
3057 | * records are inserted to lock ranges in the tree, and as dirty areas | |
3058 | * are found, they are marked writeback. Then the lock bits are removed | |
3059 | * and the end_io handler clears the writeback ranges | |
3060 | */ | |
3061 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
3062 | void *data) | |
3063 | { | |
3064 | struct inode *inode = page->mapping->host; | |
3065 | struct extent_page_data *epd = data; | |
3066 | struct extent_io_tree *tree = epd->tree; | |
4eee4fa4 | 3067 | u64 start = page_offset(page); |
d1310b2e CM |
3068 | u64 delalloc_start; |
3069 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
3070 | u64 end; | |
3071 | u64 cur = start; | |
3072 | u64 extent_offset; | |
3073 | u64 last_byte = i_size_read(inode); | |
3074 | u64 block_start; | |
3075 | u64 iosize; | |
3076 | sector_t sector; | |
2c64c53d | 3077 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
3078 | struct extent_map *em; |
3079 | struct block_device *bdev; | |
3080 | int ret; | |
3081 | int nr = 0; | |
7f3c74fb | 3082 | size_t pg_offset = 0; |
d1310b2e CM |
3083 | size_t blocksize; |
3084 | loff_t i_size = i_size_read(inode); | |
3085 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
3086 | u64 nr_delalloc; | |
3087 | u64 delalloc_end; | |
c8b97818 CM |
3088 | int page_started; |
3089 | int compressed; | |
ffbd517d | 3090 | int write_flags; |
771ed689 | 3091 | unsigned long nr_written = 0; |
9e487107 | 3092 | bool fill_delalloc = true; |
d1310b2e | 3093 | |
ffbd517d | 3094 | if (wbc->sync_mode == WB_SYNC_ALL) |
721a9602 | 3095 | write_flags = WRITE_SYNC; |
ffbd517d CM |
3096 | else |
3097 | write_flags = WRITE; | |
3098 | ||
1abe9b8a | 3099 | trace___extent_writepage(page, inode, wbc); |
3100 | ||
d1310b2e | 3101 | WARN_ON(!PageLocked(page)); |
bf0da8c1 CM |
3102 | |
3103 | ClearPageError(page); | |
3104 | ||
7f3c74fb | 3105 | pg_offset = i_size & (PAGE_CACHE_SIZE - 1); |
211c17f5 | 3106 | if (page->index > end_index || |
7f3c74fb | 3107 | (page->index == end_index && !pg_offset)) { |
d47992f8 | 3108 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE); |
d1310b2e CM |
3109 | unlock_page(page); |
3110 | return 0; | |
3111 | } | |
3112 | ||
3113 | if (page->index == end_index) { | |
3114 | char *userpage; | |
3115 | ||
7ac687d9 | 3116 | userpage = kmap_atomic(page); |
7f3c74fb CM |
3117 | memset(userpage + pg_offset, 0, |
3118 | PAGE_CACHE_SIZE - pg_offset); | |
7ac687d9 | 3119 | kunmap_atomic(userpage); |
211c17f5 | 3120 | flush_dcache_page(page); |
d1310b2e | 3121 | } |
7f3c74fb | 3122 | pg_offset = 0; |
d1310b2e CM |
3123 | |
3124 | set_page_extent_mapped(page); | |
3125 | ||
9e487107 JB |
3126 | if (!tree->ops || !tree->ops->fill_delalloc) |
3127 | fill_delalloc = false; | |
3128 | ||
d1310b2e CM |
3129 | delalloc_start = start; |
3130 | delalloc_end = 0; | |
c8b97818 | 3131 | page_started = 0; |
9e487107 | 3132 | if (!epd->extent_locked && fill_delalloc) { |
f85d7d6c | 3133 | u64 delalloc_to_write = 0; |
11c8349b CM |
3134 | /* |
3135 | * make sure the wbc mapping index is at least updated | |
3136 | * to this page. | |
3137 | */ | |
3138 | update_nr_written(page, wbc, 0); | |
3139 | ||
d397712b | 3140 | while (delalloc_end < page_end) { |
771ed689 | 3141 | nr_delalloc = find_lock_delalloc_range(inode, tree, |
c8b97818 CM |
3142 | page, |
3143 | &delalloc_start, | |
d1310b2e CM |
3144 | &delalloc_end, |
3145 | 128 * 1024 * 1024); | |
771ed689 CM |
3146 | if (nr_delalloc == 0) { |
3147 | delalloc_start = delalloc_end + 1; | |
3148 | continue; | |
3149 | } | |
013bd4c3 TI |
3150 | ret = tree->ops->fill_delalloc(inode, page, |
3151 | delalloc_start, | |
3152 | delalloc_end, | |
3153 | &page_started, | |
3154 | &nr_written); | |
79787eaa JM |
3155 | /* File system has been set read-only */ |
3156 | if (ret) { | |
3157 | SetPageError(page); | |
3158 | goto done; | |
3159 | } | |
f85d7d6c CM |
3160 | /* |
3161 | * delalloc_end is already one less than the total | |
3162 | * length, so we don't subtract one from | |
3163 | * PAGE_CACHE_SIZE | |
3164 | */ | |
3165 | delalloc_to_write += (delalloc_end - delalloc_start + | |
3166 | PAGE_CACHE_SIZE) >> | |
3167 | PAGE_CACHE_SHIFT; | |
d1310b2e | 3168 | delalloc_start = delalloc_end + 1; |
d1310b2e | 3169 | } |
f85d7d6c CM |
3170 | if (wbc->nr_to_write < delalloc_to_write) { |
3171 | int thresh = 8192; | |
3172 | ||
3173 | if (delalloc_to_write < thresh * 2) | |
3174 | thresh = delalloc_to_write; | |
3175 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3176 | thresh); | |
3177 | } | |
c8b97818 | 3178 | |
771ed689 CM |
3179 | /* did the fill delalloc function already unlock and start |
3180 | * the IO? | |
3181 | */ | |
3182 | if (page_started) { | |
3183 | ret = 0; | |
11c8349b CM |
3184 | /* |
3185 | * we've unlocked the page, so we can't update | |
3186 | * the mapping's writeback index, just update | |
3187 | * nr_to_write. | |
3188 | */ | |
3189 | wbc->nr_to_write -= nr_written; | |
3190 | goto done_unlocked; | |
771ed689 | 3191 | } |
c8b97818 | 3192 | } |
247e743c | 3193 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
3194 | ret = tree->ops->writepage_start_hook(page, start, |
3195 | page_end); | |
87826df0 JM |
3196 | if (ret) { |
3197 | /* Fixup worker will requeue */ | |
3198 | if (ret == -EBUSY) | |
3199 | wbc->pages_skipped++; | |
3200 | else | |
3201 | redirty_page_for_writepage(wbc, page); | |
11c8349b | 3202 | update_nr_written(page, wbc, nr_written); |
247e743c | 3203 | unlock_page(page); |
771ed689 | 3204 | ret = 0; |
11c8349b | 3205 | goto done_unlocked; |
247e743c CM |
3206 | } |
3207 | } | |
3208 | ||
11c8349b CM |
3209 | /* |
3210 | * we don't want to touch the inode after unlocking the page, | |
3211 | * so we update the mapping writeback index now | |
3212 | */ | |
3213 | update_nr_written(page, wbc, nr_written + 1); | |
771ed689 | 3214 | |
d1310b2e | 3215 | end = page_end; |
d1310b2e | 3216 | if (last_byte <= start) { |
e6dcd2dc CM |
3217 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3218 | tree->ops->writepage_end_io_hook(page, start, | |
3219 | page_end, NULL, 1); | |
d1310b2e CM |
3220 | goto done; |
3221 | } | |
3222 | ||
d1310b2e CM |
3223 | blocksize = inode->i_sb->s_blocksize; |
3224 | ||
3225 | while (cur <= end) { | |
3226 | if (cur >= last_byte) { | |
e6dcd2dc CM |
3227 | if (tree->ops && tree->ops->writepage_end_io_hook) |
3228 | tree->ops->writepage_end_io_hook(page, cur, | |
3229 | page_end, NULL, 1); | |
d1310b2e CM |
3230 | break; |
3231 | } | |
7f3c74fb | 3232 | em = epd->get_extent(inode, page, pg_offset, cur, |
d1310b2e | 3233 | end - cur + 1, 1); |
c704005d | 3234 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e CM |
3235 | SetPageError(page); |
3236 | break; | |
3237 | } | |
3238 | ||
3239 | extent_offset = cur - em->start; | |
3240 | BUG_ON(extent_map_end(em) <= cur); | |
3241 | BUG_ON(end < cur); | |
3242 | iosize = min(extent_map_end(em) - cur, end - cur + 1); | |
fda2832f | 3243 | iosize = ALIGN(iosize, blocksize); |
d1310b2e CM |
3244 | sector = (em->block_start + extent_offset) >> 9; |
3245 | bdev = em->bdev; | |
3246 | block_start = em->block_start; | |
c8b97818 | 3247 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
3248 | free_extent_map(em); |
3249 | em = NULL; | |
3250 | ||
c8b97818 CM |
3251 | /* |
3252 | * compressed and inline extents are written through other | |
3253 | * paths in the FS | |
3254 | */ | |
3255 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3256 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
3257 | /* |
3258 | * end_io notification does not happen here for | |
3259 | * compressed extents | |
3260 | */ | |
3261 | if (!compressed && tree->ops && | |
3262 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
3263 | tree->ops->writepage_end_io_hook(page, cur, |
3264 | cur + iosize - 1, | |
3265 | NULL, 1); | |
c8b97818 CM |
3266 | else if (compressed) { |
3267 | /* we don't want to end_page_writeback on | |
3268 | * a compressed extent. this happens | |
3269 | * elsewhere | |
3270 | */ | |
3271 | nr++; | |
3272 | } | |
3273 | ||
3274 | cur += iosize; | |
7f3c74fb | 3275 | pg_offset += iosize; |
d1310b2e CM |
3276 | continue; |
3277 | } | |
d1310b2e CM |
3278 | /* leave this out until we have a page_mkwrite call */ |
3279 | if (0 && !test_range_bit(tree, cur, cur + iosize - 1, | |
9655d298 | 3280 | EXTENT_DIRTY, 0, NULL)) { |
d1310b2e | 3281 | cur = cur + iosize; |
7f3c74fb | 3282 | pg_offset += iosize; |
d1310b2e CM |
3283 | continue; |
3284 | } | |
c8b97818 | 3285 | |
d1310b2e CM |
3286 | if (tree->ops && tree->ops->writepage_io_hook) { |
3287 | ret = tree->ops->writepage_io_hook(page, cur, | |
3288 | cur + iosize - 1); | |
3289 | } else { | |
3290 | ret = 0; | |
3291 | } | |
1259ab75 | 3292 | if (ret) { |
d1310b2e | 3293 | SetPageError(page); |
1259ab75 | 3294 | } else { |
d1310b2e | 3295 | unsigned long max_nr = end_index + 1; |
7f3c74fb | 3296 | |
d1310b2e CM |
3297 | set_range_writeback(tree, cur, cur + iosize - 1); |
3298 | if (!PageWriteback(page)) { | |
d397712b CM |
3299 | printk(KERN_ERR "btrfs warning page %lu not " |
3300 | "writeback, cur %llu end %llu\n", | |
3301 | page->index, (unsigned long long)cur, | |
d1310b2e CM |
3302 | (unsigned long long)end); |
3303 | } | |
3304 | ||
ffbd517d CM |
3305 | ret = submit_extent_page(write_flags, tree, page, |
3306 | sector, iosize, pg_offset, | |
3307 | bdev, &epd->bio, max_nr, | |
c8b97818 CM |
3308 | end_bio_extent_writepage, |
3309 | 0, 0, 0); | |
d1310b2e CM |
3310 | if (ret) |
3311 | SetPageError(page); | |
3312 | } | |
3313 | cur = cur + iosize; | |
7f3c74fb | 3314 | pg_offset += iosize; |
d1310b2e CM |
3315 | nr++; |
3316 | } | |
3317 | done: | |
3318 | if (nr == 0) { | |
3319 | /* make sure the mapping tag for page dirty gets cleared */ | |
3320 | set_page_writeback(page); | |
3321 | end_page_writeback(page); | |
3322 | } | |
d1310b2e | 3323 | unlock_page(page); |
771ed689 | 3324 | |
11c8349b CM |
3325 | done_unlocked: |
3326 | ||
2c64c53d CM |
3327 | /* drop our reference on any cached states */ |
3328 | free_extent_state(cached_state); | |
d1310b2e CM |
3329 | return 0; |
3330 | } | |
3331 | ||
0b32f4bb JB |
3332 | static int eb_wait(void *word) |
3333 | { | |
3334 | io_schedule(); | |
3335 | return 0; | |
3336 | } | |
3337 | ||
fd8b2b61 | 3338 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb JB |
3339 | { |
3340 | wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait, | |
3341 | TASK_UNINTERRUPTIBLE); | |
3342 | } | |
3343 | ||
3344 | static int lock_extent_buffer_for_io(struct extent_buffer *eb, | |
3345 | struct btrfs_fs_info *fs_info, | |
3346 | struct extent_page_data *epd) | |
3347 | { | |
3348 | unsigned long i, num_pages; | |
3349 | int flush = 0; | |
3350 | int ret = 0; | |
3351 | ||
3352 | if (!btrfs_try_tree_write_lock(eb)) { | |
3353 | flush = 1; | |
3354 | flush_write_bio(epd); | |
3355 | btrfs_tree_lock(eb); | |
3356 | } | |
3357 | ||
3358 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3359 | btrfs_tree_unlock(eb); | |
3360 | if (!epd->sync_io) | |
3361 | return 0; | |
3362 | if (!flush) { | |
3363 | flush_write_bio(epd); | |
3364 | flush = 1; | |
3365 | } | |
a098d8e8 CM |
3366 | while (1) { |
3367 | wait_on_extent_buffer_writeback(eb); | |
3368 | btrfs_tree_lock(eb); | |
3369 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3370 | break; | |
0b32f4bb | 3371 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3372 | } |
3373 | } | |
3374 | ||
51561ffe JB |
3375 | /* |
3376 | * We need to do this to prevent races in people who check if the eb is | |
3377 | * under IO since we can end up having no IO bits set for a short period | |
3378 | * of time. | |
3379 | */ | |
3380 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
3381 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3382 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 3383 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 3384 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
e2d84521 MX |
3385 | __percpu_counter_add(&fs_info->dirty_metadata_bytes, |
3386 | -eb->len, | |
3387 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 3388 | ret = 1; |
51561ffe JB |
3389 | } else { |
3390 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
3391 | } |
3392 | ||
3393 | btrfs_tree_unlock(eb); | |
3394 | ||
3395 | if (!ret) | |
3396 | return ret; | |
3397 | ||
3398 | num_pages = num_extent_pages(eb->start, eb->len); | |
3399 | for (i = 0; i < num_pages; i++) { | |
3400 | struct page *p = extent_buffer_page(eb, i); | |
3401 | ||
3402 | if (!trylock_page(p)) { | |
3403 | if (!flush) { | |
3404 | flush_write_bio(epd); | |
3405 | flush = 1; | |
3406 | } | |
3407 | lock_page(p); | |
3408 | } | |
3409 | } | |
3410 | ||
3411 | return ret; | |
3412 | } | |
3413 | ||
3414 | static void end_extent_buffer_writeback(struct extent_buffer *eb) | |
3415 | { | |
3416 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
3417 | smp_mb__after_clear_bit(); | |
3418 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); | |
3419 | } | |
3420 | ||
3421 | static void end_bio_extent_buffer_writepage(struct bio *bio, int err) | |
3422 | { | |
3423 | int uptodate = err == 0; | |
3424 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
3425 | struct extent_buffer *eb; | |
3426 | int done; | |
3427 | ||
3428 | do { | |
3429 | struct page *page = bvec->bv_page; | |
3430 | ||
3431 | bvec--; | |
3432 | eb = (struct extent_buffer *)page->private; | |
3433 | BUG_ON(!eb); | |
3434 | done = atomic_dec_and_test(&eb->io_pages); | |
3435 | ||
3436 | if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) { | |
3437 | set_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
3438 | ClearPageUptodate(page); | |
3439 | SetPageError(page); | |
3440 | } | |
3441 | ||
3442 | end_page_writeback(page); | |
3443 | ||
3444 | if (!done) | |
3445 | continue; | |
3446 | ||
3447 | end_extent_buffer_writeback(eb); | |
3448 | } while (bvec >= bio->bi_io_vec); | |
3449 | ||
3450 | bio_put(bio); | |
3451 | ||
3452 | } | |
3453 | ||
3454 | static int write_one_eb(struct extent_buffer *eb, | |
3455 | struct btrfs_fs_info *fs_info, | |
3456 | struct writeback_control *wbc, | |
3457 | struct extent_page_data *epd) | |
3458 | { | |
3459 | struct block_device *bdev = fs_info->fs_devices->latest_bdev; | |
3460 | u64 offset = eb->start; | |
3461 | unsigned long i, num_pages; | |
de0022b9 | 3462 | unsigned long bio_flags = 0; |
d4c7ca86 | 3463 | int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META; |
d7dbe9e7 | 3464 | int ret = 0; |
0b32f4bb JB |
3465 | |
3466 | clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
3467 | num_pages = num_extent_pages(eb->start, eb->len); | |
3468 | atomic_set(&eb->io_pages, num_pages); | |
de0022b9 JB |
3469 | if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID) |
3470 | bio_flags = EXTENT_BIO_TREE_LOG; | |
3471 | ||
0b32f4bb JB |
3472 | for (i = 0; i < num_pages; i++) { |
3473 | struct page *p = extent_buffer_page(eb, i); | |
3474 | ||
3475 | clear_page_dirty_for_io(p); | |
3476 | set_page_writeback(p); | |
3477 | ret = submit_extent_page(rw, eb->tree, p, offset >> 9, | |
3478 | PAGE_CACHE_SIZE, 0, bdev, &epd->bio, | |
3479 | -1, end_bio_extent_buffer_writepage, | |
de0022b9 JB |
3480 | 0, epd->bio_flags, bio_flags); |
3481 | epd->bio_flags = bio_flags; | |
0b32f4bb JB |
3482 | if (ret) { |
3483 | set_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
3484 | SetPageError(p); | |
3485 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) | |
3486 | end_extent_buffer_writeback(eb); | |
3487 | ret = -EIO; | |
3488 | break; | |
3489 | } | |
3490 | offset += PAGE_CACHE_SIZE; | |
3491 | update_nr_written(p, wbc, 1); | |
3492 | unlock_page(p); | |
3493 | } | |
3494 | ||
3495 | if (unlikely(ret)) { | |
3496 | for (; i < num_pages; i++) { | |
3497 | struct page *p = extent_buffer_page(eb, i); | |
3498 | unlock_page(p); | |
3499 | } | |
3500 | } | |
3501 | ||
3502 | return ret; | |
3503 | } | |
3504 | ||
3505 | int btree_write_cache_pages(struct address_space *mapping, | |
3506 | struct writeback_control *wbc) | |
3507 | { | |
3508 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; | |
3509 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; | |
3510 | struct extent_buffer *eb, *prev_eb = NULL; | |
3511 | struct extent_page_data epd = { | |
3512 | .bio = NULL, | |
3513 | .tree = tree, | |
3514 | .extent_locked = 0, | |
3515 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
de0022b9 | 3516 | .bio_flags = 0, |
0b32f4bb JB |
3517 | }; |
3518 | int ret = 0; | |
3519 | int done = 0; | |
3520 | int nr_to_write_done = 0; | |
3521 | struct pagevec pvec; | |
3522 | int nr_pages; | |
3523 | pgoff_t index; | |
3524 | pgoff_t end; /* Inclusive */ | |
3525 | int scanned = 0; | |
3526 | int tag; | |
3527 | ||
3528 | pagevec_init(&pvec, 0); | |
3529 | if (wbc->range_cyclic) { | |
3530 | index = mapping->writeback_index; /* Start from prev offset */ | |
3531 | end = -1; | |
3532 | } else { | |
3533 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
3534 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
3535 | scanned = 1; | |
3536 | } | |
3537 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3538 | tag = PAGECACHE_TAG_TOWRITE; | |
3539 | else | |
3540 | tag = PAGECACHE_TAG_DIRTY; | |
3541 | retry: | |
3542 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3543 | tag_pages_for_writeback(mapping, index, end); | |
3544 | while (!done && !nr_to_write_done && (index <= end) && | |
3545 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, | |
3546 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
3547 | unsigned i; | |
3548 | ||
3549 | scanned = 1; | |
3550 | for (i = 0; i < nr_pages; i++) { | |
3551 | struct page *page = pvec.pages[i]; | |
3552 | ||
3553 | if (!PagePrivate(page)) | |
3554 | continue; | |
3555 | ||
3556 | if (!wbc->range_cyclic && page->index > end) { | |
3557 | done = 1; | |
3558 | break; | |
3559 | } | |
3560 | ||
b5bae261 JB |
3561 | spin_lock(&mapping->private_lock); |
3562 | if (!PagePrivate(page)) { | |
3563 | spin_unlock(&mapping->private_lock); | |
3564 | continue; | |
3565 | } | |
3566 | ||
0b32f4bb | 3567 | eb = (struct extent_buffer *)page->private; |
b5bae261 JB |
3568 | |
3569 | /* | |
3570 | * Shouldn't happen and normally this would be a BUG_ON | |
3571 | * but no sense in crashing the users box for something | |
3572 | * we can survive anyway. | |
3573 | */ | |
0b32f4bb | 3574 | if (!eb) { |
b5bae261 | 3575 | spin_unlock(&mapping->private_lock); |
0b32f4bb JB |
3576 | WARN_ON(1); |
3577 | continue; | |
3578 | } | |
3579 | ||
b5bae261 JB |
3580 | if (eb == prev_eb) { |
3581 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 3582 | continue; |
b5bae261 | 3583 | } |
0b32f4bb | 3584 | |
b5bae261 JB |
3585 | ret = atomic_inc_not_zero(&eb->refs); |
3586 | spin_unlock(&mapping->private_lock); | |
3587 | if (!ret) | |
0b32f4bb | 3588 | continue; |
0b32f4bb JB |
3589 | |
3590 | prev_eb = eb; | |
3591 | ret = lock_extent_buffer_for_io(eb, fs_info, &epd); | |
3592 | if (!ret) { | |
3593 | free_extent_buffer(eb); | |
3594 | continue; | |
3595 | } | |
3596 | ||
3597 | ret = write_one_eb(eb, fs_info, wbc, &epd); | |
3598 | if (ret) { | |
3599 | done = 1; | |
3600 | free_extent_buffer(eb); | |
3601 | break; | |
3602 | } | |
3603 | free_extent_buffer(eb); | |
3604 | ||
3605 | /* | |
3606 | * the filesystem may choose to bump up nr_to_write. | |
3607 | * We have to make sure to honor the new nr_to_write | |
3608 | * at any time | |
3609 | */ | |
3610 | nr_to_write_done = wbc->nr_to_write <= 0; | |
3611 | } | |
3612 | pagevec_release(&pvec); | |
3613 | cond_resched(); | |
3614 | } | |
3615 | if (!scanned && !done) { | |
3616 | /* | |
3617 | * We hit the last page and there is more work to be done: wrap | |
3618 | * back to the start of the file | |
3619 | */ | |
3620 | scanned = 1; | |
3621 | index = 0; | |
3622 | goto retry; | |
3623 | } | |
3624 | flush_write_bio(&epd); | |
3625 | return ret; | |
3626 | } | |
3627 | ||
d1310b2e | 3628 | /** |
4bef0848 | 3629 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
3630 | * @mapping: address space structure to write |
3631 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
3632 | * @writepage: function called for each page | |
3633 | * @data: data passed to writepage function | |
3634 | * | |
3635 | * If a page is already under I/O, write_cache_pages() skips it, even | |
3636 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
3637 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
3638 | * and msync() need to guarantee that all the data which was dirty at the time | |
3639 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
3640 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
3641 | * existing IO to complete. | |
3642 | */ | |
b2950863 | 3643 | static int extent_write_cache_pages(struct extent_io_tree *tree, |
4bef0848 CM |
3644 | struct address_space *mapping, |
3645 | struct writeback_control *wbc, | |
d2c3f4f6 CM |
3646 | writepage_t writepage, void *data, |
3647 | void (*flush_fn)(void *)) | |
d1310b2e | 3648 | { |
7fd1a3f7 | 3649 | struct inode *inode = mapping->host; |
d1310b2e CM |
3650 | int ret = 0; |
3651 | int done = 0; | |
f85d7d6c | 3652 | int nr_to_write_done = 0; |
d1310b2e CM |
3653 | struct pagevec pvec; |
3654 | int nr_pages; | |
3655 | pgoff_t index; | |
3656 | pgoff_t end; /* Inclusive */ | |
3657 | int scanned = 0; | |
f7aaa06b | 3658 | int tag; |
d1310b2e | 3659 | |
7fd1a3f7 JB |
3660 | /* |
3661 | * We have to hold onto the inode so that ordered extents can do their | |
3662 | * work when the IO finishes. The alternative to this is failing to add | |
3663 | * an ordered extent if the igrab() fails there and that is a huge pain | |
3664 | * to deal with, so instead just hold onto the inode throughout the | |
3665 | * writepages operation. If it fails here we are freeing up the inode | |
3666 | * anyway and we'd rather not waste our time writing out stuff that is | |
3667 | * going to be truncated anyway. | |
3668 | */ | |
3669 | if (!igrab(inode)) | |
3670 | return 0; | |
3671 | ||
d1310b2e CM |
3672 | pagevec_init(&pvec, 0); |
3673 | if (wbc->range_cyclic) { | |
3674 | index = mapping->writeback_index; /* Start from prev offset */ | |
3675 | end = -1; | |
3676 | } else { | |
3677 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
3678 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
3679 | scanned = 1; |
3680 | } | |
f7aaa06b JB |
3681 | if (wbc->sync_mode == WB_SYNC_ALL) |
3682 | tag = PAGECACHE_TAG_TOWRITE; | |
3683 | else | |
3684 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 3685 | retry: |
f7aaa06b JB |
3686 | if (wbc->sync_mode == WB_SYNC_ALL) |
3687 | tag_pages_for_writeback(mapping, index, end); | |
f85d7d6c | 3688 | while (!done && !nr_to_write_done && (index <= end) && |
f7aaa06b JB |
3689 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
3690 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
d1310b2e CM |
3691 | unsigned i; |
3692 | ||
3693 | scanned = 1; | |
3694 | for (i = 0; i < nr_pages; i++) { | |
3695 | struct page *page = pvec.pages[i]; | |
3696 | ||
3697 | /* | |
3698 | * At this point we hold neither mapping->tree_lock nor | |
3699 | * lock on the page itself: the page may be truncated or | |
3700 | * invalidated (changing page->mapping to NULL), or even | |
3701 | * swizzled back from swapper_space to tmpfs file | |
3702 | * mapping | |
3703 | */ | |
c8f2f24b JB |
3704 | if (!trylock_page(page)) { |
3705 | flush_fn(data); | |
3706 | lock_page(page); | |
01d658f2 | 3707 | } |
d1310b2e CM |
3708 | |
3709 | if (unlikely(page->mapping != mapping)) { | |
3710 | unlock_page(page); | |
3711 | continue; | |
3712 | } | |
3713 | ||
3714 | if (!wbc->range_cyclic && page->index > end) { | |
3715 | done = 1; | |
3716 | unlock_page(page); | |
3717 | continue; | |
3718 | } | |
3719 | ||
d2c3f4f6 | 3720 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 CM |
3721 | if (PageWriteback(page)) |
3722 | flush_fn(data); | |
d1310b2e | 3723 | wait_on_page_writeback(page); |
d2c3f4f6 | 3724 | } |
d1310b2e CM |
3725 | |
3726 | if (PageWriteback(page) || | |
3727 | !clear_page_dirty_for_io(page)) { | |
3728 | unlock_page(page); | |
3729 | continue; | |
3730 | } | |
3731 | ||
3732 | ret = (*writepage)(page, wbc, data); | |
3733 | ||
3734 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
3735 | unlock_page(page); | |
3736 | ret = 0; | |
3737 | } | |
f85d7d6c | 3738 | if (ret) |
d1310b2e | 3739 | done = 1; |
f85d7d6c CM |
3740 | |
3741 | /* | |
3742 | * the filesystem may choose to bump up nr_to_write. | |
3743 | * We have to make sure to honor the new nr_to_write | |
3744 | * at any time | |
3745 | */ | |
3746 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
3747 | } |
3748 | pagevec_release(&pvec); | |
3749 | cond_resched(); | |
3750 | } | |
3751 | if (!scanned && !done) { | |
3752 | /* | |
3753 | * We hit the last page and there is more work to be done: wrap | |
3754 | * back to the start of the file | |
3755 | */ | |
3756 | scanned = 1; | |
3757 | index = 0; | |
3758 | goto retry; | |
3759 | } | |
7fd1a3f7 | 3760 | btrfs_add_delayed_iput(inode); |
d1310b2e CM |
3761 | return ret; |
3762 | } | |
d1310b2e | 3763 | |
ffbd517d | 3764 | static void flush_epd_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 3765 | { |
d2c3f4f6 | 3766 | if (epd->bio) { |
355808c2 JM |
3767 | int rw = WRITE; |
3768 | int ret; | |
3769 | ||
ffbd517d | 3770 | if (epd->sync_io) |
355808c2 JM |
3771 | rw = WRITE_SYNC; |
3772 | ||
de0022b9 | 3773 | ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags); |
79787eaa | 3774 | BUG_ON(ret < 0); /* -ENOMEM */ |
d2c3f4f6 CM |
3775 | epd->bio = NULL; |
3776 | } | |
3777 | } | |
3778 | ||
ffbd517d CM |
3779 | static noinline void flush_write_bio(void *data) |
3780 | { | |
3781 | struct extent_page_data *epd = data; | |
3782 | flush_epd_write_bio(epd); | |
3783 | } | |
3784 | ||
d1310b2e CM |
3785 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, |
3786 | get_extent_t *get_extent, | |
3787 | struct writeback_control *wbc) | |
3788 | { | |
3789 | int ret; | |
d1310b2e CM |
3790 | struct extent_page_data epd = { |
3791 | .bio = NULL, | |
3792 | .tree = tree, | |
3793 | .get_extent = get_extent, | |
771ed689 | 3794 | .extent_locked = 0, |
ffbd517d | 3795 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
de0022b9 | 3796 | .bio_flags = 0, |
d1310b2e | 3797 | }; |
d1310b2e | 3798 | |
d1310b2e CM |
3799 | ret = __extent_writepage(page, wbc, &epd); |
3800 | ||
ffbd517d | 3801 | flush_epd_write_bio(&epd); |
d1310b2e CM |
3802 | return ret; |
3803 | } | |
d1310b2e | 3804 | |
771ed689 CM |
3805 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, |
3806 | u64 start, u64 end, get_extent_t *get_extent, | |
3807 | int mode) | |
3808 | { | |
3809 | int ret = 0; | |
3810 | struct address_space *mapping = inode->i_mapping; | |
3811 | struct page *page; | |
3812 | unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >> | |
3813 | PAGE_CACHE_SHIFT; | |
3814 | ||
3815 | struct extent_page_data epd = { | |
3816 | .bio = NULL, | |
3817 | .tree = tree, | |
3818 | .get_extent = get_extent, | |
3819 | .extent_locked = 1, | |
ffbd517d | 3820 | .sync_io = mode == WB_SYNC_ALL, |
de0022b9 | 3821 | .bio_flags = 0, |
771ed689 CM |
3822 | }; |
3823 | struct writeback_control wbc_writepages = { | |
771ed689 | 3824 | .sync_mode = mode, |
771ed689 CM |
3825 | .nr_to_write = nr_pages * 2, |
3826 | .range_start = start, | |
3827 | .range_end = end + 1, | |
3828 | }; | |
3829 | ||
d397712b | 3830 | while (start <= end) { |
771ed689 CM |
3831 | page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); |
3832 | if (clear_page_dirty_for_io(page)) | |
3833 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
3834 | else { | |
3835 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
3836 | tree->ops->writepage_end_io_hook(page, start, | |
3837 | start + PAGE_CACHE_SIZE - 1, | |
3838 | NULL, 1); | |
3839 | unlock_page(page); | |
3840 | } | |
3841 | page_cache_release(page); | |
3842 | start += PAGE_CACHE_SIZE; | |
3843 | } | |
3844 | ||
ffbd517d | 3845 | flush_epd_write_bio(&epd); |
771ed689 CM |
3846 | return ret; |
3847 | } | |
d1310b2e CM |
3848 | |
3849 | int extent_writepages(struct extent_io_tree *tree, | |
3850 | struct address_space *mapping, | |
3851 | get_extent_t *get_extent, | |
3852 | struct writeback_control *wbc) | |
3853 | { | |
3854 | int ret = 0; | |
3855 | struct extent_page_data epd = { | |
3856 | .bio = NULL, | |
3857 | .tree = tree, | |
3858 | .get_extent = get_extent, | |
771ed689 | 3859 | .extent_locked = 0, |
ffbd517d | 3860 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
de0022b9 | 3861 | .bio_flags = 0, |
d1310b2e CM |
3862 | }; |
3863 | ||
4bef0848 | 3864 | ret = extent_write_cache_pages(tree, mapping, wbc, |
d2c3f4f6 CM |
3865 | __extent_writepage, &epd, |
3866 | flush_write_bio); | |
ffbd517d | 3867 | flush_epd_write_bio(&epd); |
d1310b2e CM |
3868 | return ret; |
3869 | } | |
d1310b2e CM |
3870 | |
3871 | int extent_readpages(struct extent_io_tree *tree, | |
3872 | struct address_space *mapping, | |
3873 | struct list_head *pages, unsigned nr_pages, | |
3874 | get_extent_t get_extent) | |
3875 | { | |
3876 | struct bio *bio = NULL; | |
3877 | unsigned page_idx; | |
c8b97818 | 3878 | unsigned long bio_flags = 0; |
67c9684f LB |
3879 | struct page *pagepool[16]; |
3880 | struct page *page; | |
125bac01 | 3881 | struct extent_map *em_cached = NULL; |
67c9684f | 3882 | int nr = 0; |
d1310b2e | 3883 | |
d1310b2e | 3884 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
67c9684f | 3885 | page = list_entry(pages->prev, struct page, lru); |
d1310b2e CM |
3886 | |
3887 | prefetchw(&page->flags); | |
3888 | list_del(&page->lru); | |
67c9684f | 3889 | if (add_to_page_cache_lru(page, mapping, |
43e817a1 | 3890 | page->index, GFP_NOFS)) { |
67c9684f LB |
3891 | page_cache_release(page); |
3892 | continue; | |
d1310b2e | 3893 | } |
67c9684f LB |
3894 | |
3895 | pagepool[nr++] = page; | |
3896 | if (nr < ARRAY_SIZE(pagepool)) | |
3897 | continue; | |
125bac01 | 3898 | __extent_readpages(tree, pagepool, nr, get_extent, &em_cached, |
9974090b | 3899 | &bio, 0, &bio_flags, READ); |
67c9684f | 3900 | nr = 0; |
d1310b2e | 3901 | } |
9974090b | 3902 | if (nr) |
125bac01 | 3903 | __extent_readpages(tree, pagepool, nr, get_extent, &em_cached, |
9974090b | 3904 | &bio, 0, &bio_flags, READ); |
67c9684f | 3905 | |
125bac01 MX |
3906 | if (em_cached) |
3907 | free_extent_map(em_cached); | |
3908 | ||
d1310b2e CM |
3909 | BUG_ON(!list_empty(pages)); |
3910 | if (bio) | |
79787eaa | 3911 | return submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
3912 | return 0; |
3913 | } | |
d1310b2e CM |
3914 | |
3915 | /* | |
3916 | * basic invalidatepage code, this waits on any locked or writeback | |
3917 | * ranges corresponding to the page, and then deletes any extent state | |
3918 | * records from the tree | |
3919 | */ | |
3920 | int extent_invalidatepage(struct extent_io_tree *tree, | |
3921 | struct page *page, unsigned long offset) | |
3922 | { | |
2ac55d41 | 3923 | struct extent_state *cached_state = NULL; |
4eee4fa4 | 3924 | u64 start = page_offset(page); |
d1310b2e CM |
3925 | u64 end = start + PAGE_CACHE_SIZE - 1; |
3926 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
3927 | ||
fda2832f | 3928 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
3929 | if (start > end) |
3930 | return 0; | |
3931 | ||
d0082371 | 3932 | lock_extent_bits(tree, start, end, 0, &cached_state); |
1edbb734 | 3933 | wait_on_page_writeback(page); |
d1310b2e | 3934 | clear_extent_bit(tree, start, end, |
32c00aff JB |
3935 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
3936 | EXTENT_DO_ACCOUNTING, | |
2ac55d41 | 3937 | 1, 1, &cached_state, GFP_NOFS); |
d1310b2e CM |
3938 | return 0; |
3939 | } | |
d1310b2e | 3940 | |
7b13b7b1 CM |
3941 | /* |
3942 | * a helper for releasepage, this tests for areas of the page that | |
3943 | * are locked or under IO and drops the related state bits if it is safe | |
3944 | * to drop the page. | |
3945 | */ | |
48a3b636 ES |
3946 | static int try_release_extent_state(struct extent_map_tree *map, |
3947 | struct extent_io_tree *tree, | |
3948 | struct page *page, gfp_t mask) | |
7b13b7b1 | 3949 | { |
4eee4fa4 | 3950 | u64 start = page_offset(page); |
7b13b7b1 CM |
3951 | u64 end = start + PAGE_CACHE_SIZE - 1; |
3952 | int ret = 1; | |
3953 | ||
211f90e6 | 3954 | if (test_range_bit(tree, start, end, |
8b62b72b | 3955 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
3956 | ret = 0; |
3957 | else { | |
3958 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
3959 | mask = GFP_NOFS; | |
11ef160f CM |
3960 | /* |
3961 | * at this point we can safely clear everything except the | |
3962 | * locked bit and the nodatasum bit | |
3963 | */ | |
e3f24cc5 | 3964 | ret = clear_extent_bit(tree, start, end, |
11ef160f CM |
3965 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
3966 | 0, 0, NULL, mask); | |
e3f24cc5 CM |
3967 | |
3968 | /* if clear_extent_bit failed for enomem reasons, | |
3969 | * we can't allow the release to continue. | |
3970 | */ | |
3971 | if (ret < 0) | |
3972 | ret = 0; | |
3973 | else | |
3974 | ret = 1; | |
7b13b7b1 CM |
3975 | } |
3976 | return ret; | |
3977 | } | |
7b13b7b1 | 3978 | |
d1310b2e CM |
3979 | /* |
3980 | * a helper for releasepage. As long as there are no locked extents | |
3981 | * in the range corresponding to the page, both state records and extent | |
3982 | * map records are removed | |
3983 | */ | |
3984 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
3985 | struct extent_io_tree *tree, struct page *page, |
3986 | gfp_t mask) | |
d1310b2e CM |
3987 | { |
3988 | struct extent_map *em; | |
4eee4fa4 | 3989 | u64 start = page_offset(page); |
d1310b2e | 3990 | u64 end = start + PAGE_CACHE_SIZE - 1; |
7b13b7b1 | 3991 | |
70dec807 CM |
3992 | if ((mask & __GFP_WAIT) && |
3993 | page->mapping->host->i_size > 16 * 1024 * 1024) { | |
39b5637f | 3994 | u64 len; |
70dec807 | 3995 | while (start <= end) { |
39b5637f | 3996 | len = end - start + 1; |
890871be | 3997 | write_lock(&map->lock); |
39b5637f | 3998 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 3999 | if (!em) { |
890871be | 4000 | write_unlock(&map->lock); |
70dec807 CM |
4001 | break; |
4002 | } | |
7f3c74fb CM |
4003 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
4004 | em->start != start) { | |
890871be | 4005 | write_unlock(&map->lock); |
70dec807 CM |
4006 | free_extent_map(em); |
4007 | break; | |
4008 | } | |
4009 | if (!test_range_bit(tree, em->start, | |
4010 | extent_map_end(em) - 1, | |
8b62b72b | 4011 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 4012 | 0, NULL)) { |
70dec807 CM |
4013 | remove_extent_mapping(map, em); |
4014 | /* once for the rb tree */ | |
4015 | free_extent_map(em); | |
4016 | } | |
4017 | start = extent_map_end(em); | |
890871be | 4018 | write_unlock(&map->lock); |
70dec807 CM |
4019 | |
4020 | /* once for us */ | |
d1310b2e CM |
4021 | free_extent_map(em); |
4022 | } | |
d1310b2e | 4023 | } |
7b13b7b1 | 4024 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 4025 | } |
d1310b2e | 4026 | |
ec29ed5b CM |
4027 | /* |
4028 | * helper function for fiemap, which doesn't want to see any holes. | |
4029 | * This maps until we find something past 'last' | |
4030 | */ | |
4031 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
4032 | u64 offset, | |
4033 | u64 last, | |
4034 | get_extent_t *get_extent) | |
4035 | { | |
4036 | u64 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
4037 | struct extent_map *em; | |
4038 | u64 len; | |
4039 | ||
4040 | if (offset >= last) | |
4041 | return NULL; | |
4042 | ||
4043 | while(1) { | |
4044 | len = last - offset; | |
4045 | if (len == 0) | |
4046 | break; | |
fda2832f | 4047 | len = ALIGN(len, sectorsize); |
ec29ed5b | 4048 | em = get_extent(inode, NULL, 0, offset, len, 0); |
c704005d | 4049 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
4050 | return em; |
4051 | ||
4052 | /* if this isn't a hole return it */ | |
4053 | if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) && | |
4054 | em->block_start != EXTENT_MAP_HOLE) { | |
4055 | return em; | |
4056 | } | |
4057 | ||
4058 | /* this is a hole, advance to the next extent */ | |
4059 | offset = extent_map_end(em); | |
4060 | free_extent_map(em); | |
4061 | if (offset >= last) | |
4062 | break; | |
4063 | } | |
4064 | return NULL; | |
4065 | } | |
4066 | ||
1506fcc8 YS |
4067 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4068 | __u64 start, __u64 len, get_extent_t *get_extent) | |
4069 | { | |
975f84fe | 4070 | int ret = 0; |
1506fcc8 YS |
4071 | u64 off = start; |
4072 | u64 max = start + len; | |
4073 | u32 flags = 0; | |
975f84fe JB |
4074 | u32 found_type; |
4075 | u64 last; | |
ec29ed5b | 4076 | u64 last_for_get_extent = 0; |
1506fcc8 | 4077 | u64 disko = 0; |
ec29ed5b | 4078 | u64 isize = i_size_read(inode); |
975f84fe | 4079 | struct btrfs_key found_key; |
1506fcc8 | 4080 | struct extent_map *em = NULL; |
2ac55d41 | 4081 | struct extent_state *cached_state = NULL; |
975f84fe JB |
4082 | struct btrfs_path *path; |
4083 | struct btrfs_file_extent_item *item; | |
1506fcc8 | 4084 | int end = 0; |
ec29ed5b CM |
4085 | u64 em_start = 0; |
4086 | u64 em_len = 0; | |
4087 | u64 em_end = 0; | |
1506fcc8 | 4088 | unsigned long emflags; |
1506fcc8 YS |
4089 | |
4090 | if (len == 0) | |
4091 | return -EINVAL; | |
4092 | ||
975f84fe JB |
4093 | path = btrfs_alloc_path(); |
4094 | if (!path) | |
4095 | return -ENOMEM; | |
4096 | path->leave_spinning = 1; | |
4097 | ||
4d479cf0 JB |
4098 | start = ALIGN(start, BTRFS_I(inode)->root->sectorsize); |
4099 | len = ALIGN(len, BTRFS_I(inode)->root->sectorsize); | |
4100 | ||
ec29ed5b CM |
4101 | /* |
4102 | * lookup the last file extent. We're not using i_size here | |
4103 | * because there might be preallocation past i_size | |
4104 | */ | |
975f84fe | 4105 | ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root, |
33345d01 | 4106 | path, btrfs_ino(inode), -1, 0); |
975f84fe JB |
4107 | if (ret < 0) { |
4108 | btrfs_free_path(path); | |
4109 | return ret; | |
4110 | } | |
4111 | WARN_ON(!ret); | |
4112 | path->slots[0]--; | |
4113 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4114 | struct btrfs_file_extent_item); | |
4115 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
4116 | found_type = btrfs_key_type(&found_key); | |
4117 | ||
ec29ed5b | 4118 | /* No extents, but there might be delalloc bits */ |
33345d01 | 4119 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 4120 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
4121 | /* have to trust i_size as the end */ |
4122 | last = (u64)-1; | |
4123 | last_for_get_extent = isize; | |
4124 | } else { | |
4125 | /* | |
4126 | * remember the start of the last extent. There are a | |
4127 | * bunch of different factors that go into the length of the | |
4128 | * extent, so its much less complex to remember where it started | |
4129 | */ | |
4130 | last = found_key.offset; | |
4131 | last_for_get_extent = last + 1; | |
975f84fe | 4132 | } |
975f84fe JB |
4133 | btrfs_free_path(path); |
4134 | ||
ec29ed5b CM |
4135 | /* |
4136 | * we might have some extents allocated but more delalloc past those | |
4137 | * extents. so, we trust isize unless the start of the last extent is | |
4138 | * beyond isize | |
4139 | */ | |
4140 | if (last < isize) { | |
4141 | last = (u64)-1; | |
4142 | last_for_get_extent = isize; | |
4143 | } | |
4144 | ||
a52f4cd2 | 4145 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0, |
d0082371 | 4146 | &cached_state); |
ec29ed5b | 4147 | |
4d479cf0 | 4148 | em = get_extent_skip_holes(inode, start, last_for_get_extent, |
ec29ed5b | 4149 | get_extent); |
1506fcc8 YS |
4150 | if (!em) |
4151 | goto out; | |
4152 | if (IS_ERR(em)) { | |
4153 | ret = PTR_ERR(em); | |
4154 | goto out; | |
4155 | } | |
975f84fe | 4156 | |
1506fcc8 | 4157 | while (!end) { |
b76bb701 | 4158 | u64 offset_in_extent = 0; |
ea8efc74 CM |
4159 | |
4160 | /* break if the extent we found is outside the range */ | |
4161 | if (em->start >= max || extent_map_end(em) < off) | |
4162 | break; | |
4163 | ||
4164 | /* | |
4165 | * get_extent may return an extent that starts before our | |
4166 | * requested range. We have to make sure the ranges | |
4167 | * we return to fiemap always move forward and don't | |
4168 | * overlap, so adjust the offsets here | |
4169 | */ | |
4170 | em_start = max(em->start, off); | |
1506fcc8 | 4171 | |
ea8efc74 CM |
4172 | /* |
4173 | * record the offset from the start of the extent | |
b76bb701 JB |
4174 | * for adjusting the disk offset below. Only do this if the |
4175 | * extent isn't compressed since our in ram offset may be past | |
4176 | * what we have actually allocated on disk. | |
ea8efc74 | 4177 | */ |
b76bb701 JB |
4178 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
4179 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 4180 | em_end = extent_map_end(em); |
ea8efc74 | 4181 | em_len = em_end - em_start; |
ec29ed5b | 4182 | emflags = em->flags; |
1506fcc8 YS |
4183 | disko = 0; |
4184 | flags = 0; | |
4185 | ||
ea8efc74 CM |
4186 | /* |
4187 | * bump off for our next call to get_extent | |
4188 | */ | |
4189 | off = extent_map_end(em); | |
4190 | if (off >= max) | |
4191 | end = 1; | |
4192 | ||
93dbfad7 | 4193 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
4194 | end = 1; |
4195 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 4196 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
4197 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
4198 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 4199 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
4200 | flags |= (FIEMAP_EXTENT_DELALLOC | |
4201 | FIEMAP_EXTENT_UNKNOWN); | |
93dbfad7 | 4202 | } else { |
ea8efc74 | 4203 | disko = em->block_start + offset_in_extent; |
1506fcc8 YS |
4204 | } |
4205 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
4206 | flags |= FIEMAP_EXTENT_ENCODED; | |
4207 | ||
1506fcc8 YS |
4208 | free_extent_map(em); |
4209 | em = NULL; | |
ec29ed5b CM |
4210 | if ((em_start >= last) || em_len == (u64)-1 || |
4211 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
4212 | flags |= FIEMAP_EXTENT_LAST; |
4213 | end = 1; | |
4214 | } | |
4215 | ||
ec29ed5b CM |
4216 | /* now scan forward to see if this is really the last extent. */ |
4217 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
4218 | get_extent); | |
4219 | if (IS_ERR(em)) { | |
4220 | ret = PTR_ERR(em); | |
4221 | goto out; | |
4222 | } | |
4223 | if (!em) { | |
975f84fe JB |
4224 | flags |= FIEMAP_EXTENT_LAST; |
4225 | end = 1; | |
4226 | } | |
ec29ed5b CM |
4227 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, |
4228 | em_len, flags); | |
4229 | if (ret) | |
4230 | goto out_free; | |
1506fcc8 YS |
4231 | } |
4232 | out_free: | |
4233 | free_extent_map(em); | |
4234 | out: | |
a52f4cd2 | 4235 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
2ac55d41 | 4236 | &cached_state, GFP_NOFS); |
1506fcc8 YS |
4237 | return ret; |
4238 | } | |
4239 | ||
727011e0 CM |
4240 | static void __free_extent_buffer(struct extent_buffer *eb) |
4241 | { | |
6d49ba1b | 4242 | btrfs_leak_debug_del(&eb->leak_list); |
727011e0 CM |
4243 | kmem_cache_free(extent_buffer_cache, eb); |
4244 | } | |
4245 | ||
db7f3436 JB |
4246 | static int extent_buffer_under_io(struct extent_buffer *eb) |
4247 | { | |
4248 | return (atomic_read(&eb->io_pages) || | |
4249 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4250 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4251 | } | |
4252 | ||
4253 | /* | |
4254 | * Helper for releasing extent buffer page. | |
4255 | */ | |
4256 | static void btrfs_release_extent_buffer_page(struct extent_buffer *eb, | |
4257 | unsigned long start_idx) | |
4258 | { | |
4259 | unsigned long index; | |
4260 | unsigned long num_pages; | |
4261 | struct page *page; | |
4262 | int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4263 | ||
4264 | BUG_ON(extent_buffer_under_io(eb)); | |
4265 | ||
4266 | num_pages = num_extent_pages(eb->start, eb->len); | |
4267 | index = start_idx + num_pages; | |
4268 | if (start_idx >= index) | |
4269 | return; | |
4270 | ||
4271 | do { | |
4272 | index--; | |
4273 | page = extent_buffer_page(eb, index); | |
4274 | if (page && mapped) { | |
4275 | spin_lock(&page->mapping->private_lock); | |
4276 | /* | |
4277 | * We do this since we'll remove the pages after we've | |
4278 | * removed the eb from the radix tree, so we could race | |
4279 | * and have this page now attached to the new eb. So | |
4280 | * only clear page_private if it's still connected to | |
4281 | * this eb. | |
4282 | */ | |
4283 | if (PagePrivate(page) && | |
4284 | page->private == (unsigned long)eb) { | |
4285 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
4286 | BUG_ON(PageDirty(page)); | |
4287 | BUG_ON(PageWriteback(page)); | |
4288 | /* | |
4289 | * We need to make sure we haven't be attached | |
4290 | * to a new eb. | |
4291 | */ | |
4292 | ClearPagePrivate(page); | |
4293 | set_page_private(page, 0); | |
4294 | /* One for the page private */ | |
4295 | page_cache_release(page); | |
4296 | } | |
4297 | spin_unlock(&page->mapping->private_lock); | |
4298 | ||
4299 | } | |
4300 | if (page) { | |
4301 | /* One for when we alloced the page */ | |
4302 | page_cache_release(page); | |
4303 | } | |
4304 | } while (index != start_idx); | |
4305 | } | |
4306 | ||
4307 | /* | |
4308 | * Helper for releasing the extent buffer. | |
4309 | */ | |
4310 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4311 | { | |
4312 | btrfs_release_extent_buffer_page(eb, 0); | |
4313 | __free_extent_buffer(eb); | |
4314 | } | |
4315 | ||
d1310b2e CM |
4316 | static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree, |
4317 | u64 start, | |
4318 | unsigned long len, | |
4319 | gfp_t mask) | |
4320 | { | |
4321 | struct extent_buffer *eb = NULL; | |
4322 | ||
d1310b2e | 4323 | eb = kmem_cache_zalloc(extent_buffer_cache, mask); |
91ca338d TI |
4324 | if (eb == NULL) |
4325 | return NULL; | |
d1310b2e CM |
4326 | eb->start = start; |
4327 | eb->len = len; | |
4f2de97a | 4328 | eb->tree = tree; |
815a51c7 | 4329 | eb->bflags = 0; |
bd681513 CM |
4330 | rwlock_init(&eb->lock); |
4331 | atomic_set(&eb->write_locks, 0); | |
4332 | atomic_set(&eb->read_locks, 0); | |
4333 | atomic_set(&eb->blocking_readers, 0); | |
4334 | atomic_set(&eb->blocking_writers, 0); | |
4335 | atomic_set(&eb->spinning_readers, 0); | |
4336 | atomic_set(&eb->spinning_writers, 0); | |
5b25f70f | 4337 | eb->lock_nested = 0; |
bd681513 CM |
4338 | init_waitqueue_head(&eb->write_lock_wq); |
4339 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 4340 | |
6d49ba1b ES |
4341 | btrfs_leak_debug_add(&eb->leak_list, &buffers); |
4342 | ||
3083ee2e | 4343 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 4344 | atomic_set(&eb->refs, 1); |
0b32f4bb | 4345 | atomic_set(&eb->io_pages, 0); |
727011e0 | 4346 | |
b8dae313 DS |
4347 | /* |
4348 | * Sanity checks, currently the maximum is 64k covered by 16x 4k pages | |
4349 | */ | |
4350 | BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE | |
4351 | > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
4352 | BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE); | |
d1310b2e CM |
4353 | |
4354 | return eb; | |
4355 | } | |
4356 | ||
815a51c7 JS |
4357 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) |
4358 | { | |
4359 | unsigned long i; | |
4360 | struct page *p; | |
4361 | struct extent_buffer *new; | |
4362 | unsigned long num_pages = num_extent_pages(src->start, src->len); | |
4363 | ||
9ec72677 | 4364 | new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_NOFS); |
815a51c7 JS |
4365 | if (new == NULL) |
4366 | return NULL; | |
4367 | ||
4368 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4369 | p = alloc_page(GFP_NOFS); |
db7f3436 JB |
4370 | if (!p) { |
4371 | btrfs_release_extent_buffer(new); | |
4372 | return NULL; | |
4373 | } | |
815a51c7 JS |
4374 | attach_extent_buffer_page(new, p); |
4375 | WARN_ON(PageDirty(p)); | |
4376 | SetPageUptodate(p); | |
4377 | new->pages[i] = p; | |
4378 | } | |
4379 | ||
4380 | copy_extent_buffer(new, src, 0, 0, src->len); | |
4381 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); | |
4382 | set_bit(EXTENT_BUFFER_DUMMY, &new->bflags); | |
4383 | ||
4384 | return new; | |
4385 | } | |
4386 | ||
4387 | struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len) | |
4388 | { | |
4389 | struct extent_buffer *eb; | |
4390 | unsigned long num_pages = num_extent_pages(0, len); | |
4391 | unsigned long i; | |
4392 | ||
9ec72677 | 4393 | eb = __alloc_extent_buffer(NULL, start, len, GFP_NOFS); |
815a51c7 JS |
4394 | if (!eb) |
4395 | return NULL; | |
4396 | ||
4397 | for (i = 0; i < num_pages; i++) { | |
9ec72677 | 4398 | eb->pages[i] = alloc_page(GFP_NOFS); |
815a51c7 JS |
4399 | if (!eb->pages[i]) |
4400 | goto err; | |
4401 | } | |
4402 | set_extent_buffer_uptodate(eb); | |
4403 | btrfs_set_header_nritems(eb, 0); | |
4404 | set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4405 | ||
4406 | return eb; | |
4407 | err: | |
84167d19 SB |
4408 | for (; i > 0; i--) |
4409 | __free_page(eb->pages[i - 1]); | |
815a51c7 JS |
4410 | __free_extent_buffer(eb); |
4411 | return NULL; | |
4412 | } | |
4413 | ||
0b32f4bb JB |
4414 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
4415 | { | |
242e18c7 | 4416 | int refs; |
0b32f4bb JB |
4417 | /* the ref bit is tricky. We have to make sure it is set |
4418 | * if we have the buffer dirty. Otherwise the | |
4419 | * code to free a buffer can end up dropping a dirty | |
4420 | * page | |
4421 | * | |
4422 | * Once the ref bit is set, it won't go away while the | |
4423 | * buffer is dirty or in writeback, and it also won't | |
4424 | * go away while we have the reference count on the | |
4425 | * eb bumped. | |
4426 | * | |
4427 | * We can't just set the ref bit without bumping the | |
4428 | * ref on the eb because free_extent_buffer might | |
4429 | * see the ref bit and try to clear it. If this happens | |
4430 | * free_extent_buffer might end up dropping our original | |
4431 | * ref by mistake and freeing the page before we are able | |
4432 | * to add one more ref. | |
4433 | * | |
4434 | * So bump the ref count first, then set the bit. If someone | |
4435 | * beat us to it, drop the ref we added. | |
4436 | */ | |
242e18c7 CM |
4437 | refs = atomic_read(&eb->refs); |
4438 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
4439 | return; | |
4440 | ||
594831c4 JB |
4441 | spin_lock(&eb->refs_lock); |
4442 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 4443 | atomic_inc(&eb->refs); |
594831c4 | 4444 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
4445 | } |
4446 | ||
5df4235e JB |
4447 | static void mark_extent_buffer_accessed(struct extent_buffer *eb) |
4448 | { | |
4449 | unsigned long num_pages, i; | |
4450 | ||
0b32f4bb JB |
4451 | check_buffer_tree_ref(eb); |
4452 | ||
5df4235e JB |
4453 | num_pages = num_extent_pages(eb->start, eb->len); |
4454 | for (i = 0; i < num_pages; i++) { | |
4455 | struct page *p = extent_buffer_page(eb, i); | |
4456 | mark_page_accessed(p); | |
4457 | } | |
4458 | } | |
4459 | ||
d1310b2e | 4460 | struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree, |
727011e0 | 4461 | u64 start, unsigned long len) |
d1310b2e CM |
4462 | { |
4463 | unsigned long num_pages = num_extent_pages(start, len); | |
4464 | unsigned long i; | |
4465 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
4466 | struct extent_buffer *eb; | |
6af118ce | 4467 | struct extent_buffer *exists = NULL; |
d1310b2e CM |
4468 | struct page *p; |
4469 | struct address_space *mapping = tree->mapping; | |
4470 | int uptodate = 1; | |
19fe0a8b | 4471 | int ret; |
d1310b2e | 4472 | |
19fe0a8b MX |
4473 | rcu_read_lock(); |
4474 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); | |
4475 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
4476 | rcu_read_unlock(); | |
5df4235e | 4477 | mark_extent_buffer_accessed(eb); |
6af118ce CM |
4478 | return eb; |
4479 | } | |
19fe0a8b | 4480 | rcu_read_unlock(); |
6af118ce | 4481 | |
ba144192 | 4482 | eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS); |
2b114d1d | 4483 | if (!eb) |
d1310b2e CM |
4484 | return NULL; |
4485 | ||
727011e0 | 4486 | for (i = 0; i < num_pages; i++, index++) { |
a6591715 | 4487 | p = find_or_create_page(mapping, index, GFP_NOFS); |
4804b382 | 4488 | if (!p) |
6af118ce | 4489 | goto free_eb; |
4f2de97a JB |
4490 | |
4491 | spin_lock(&mapping->private_lock); | |
4492 | if (PagePrivate(p)) { | |
4493 | /* | |
4494 | * We could have already allocated an eb for this page | |
4495 | * and attached one so lets see if we can get a ref on | |
4496 | * the existing eb, and if we can we know it's good and | |
4497 | * we can just return that one, else we know we can just | |
4498 | * overwrite page->private. | |
4499 | */ | |
4500 | exists = (struct extent_buffer *)p->private; | |
4501 | if (atomic_inc_not_zero(&exists->refs)) { | |
4502 | spin_unlock(&mapping->private_lock); | |
4503 | unlock_page(p); | |
17de39ac | 4504 | page_cache_release(p); |
5df4235e | 4505 | mark_extent_buffer_accessed(exists); |
4f2de97a JB |
4506 | goto free_eb; |
4507 | } | |
4508 | ||
0b32f4bb | 4509 | /* |
4f2de97a JB |
4510 | * Do this so attach doesn't complain and we need to |
4511 | * drop the ref the old guy had. | |
4512 | */ | |
4513 | ClearPagePrivate(p); | |
0b32f4bb | 4514 | WARN_ON(PageDirty(p)); |
4f2de97a | 4515 | page_cache_release(p); |
d1310b2e | 4516 | } |
4f2de97a JB |
4517 | attach_extent_buffer_page(eb, p); |
4518 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 4519 | WARN_ON(PageDirty(p)); |
d1310b2e | 4520 | mark_page_accessed(p); |
727011e0 | 4521 | eb->pages[i] = p; |
d1310b2e CM |
4522 | if (!PageUptodate(p)) |
4523 | uptodate = 0; | |
eb14ab8e CM |
4524 | |
4525 | /* | |
4526 | * see below about how we avoid a nasty race with release page | |
4527 | * and why we unlock later | |
4528 | */ | |
d1310b2e CM |
4529 | } |
4530 | if (uptodate) | |
b4ce94de | 4531 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 4532 | again: |
19fe0a8b MX |
4533 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); |
4534 | if (ret) | |
4535 | goto free_eb; | |
4536 | ||
6af118ce | 4537 | spin_lock(&tree->buffer_lock); |
19fe0a8b MX |
4538 | ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb); |
4539 | if (ret == -EEXIST) { | |
4540 | exists = radix_tree_lookup(&tree->buffer, | |
4541 | start >> PAGE_CACHE_SHIFT); | |
115391d2 JB |
4542 | if (!atomic_inc_not_zero(&exists->refs)) { |
4543 | spin_unlock(&tree->buffer_lock); | |
4544 | radix_tree_preload_end(); | |
115391d2 JB |
4545 | exists = NULL; |
4546 | goto again; | |
4547 | } | |
6af118ce | 4548 | spin_unlock(&tree->buffer_lock); |
19fe0a8b | 4549 | radix_tree_preload_end(); |
5df4235e | 4550 | mark_extent_buffer_accessed(exists); |
6af118ce CM |
4551 | goto free_eb; |
4552 | } | |
6af118ce | 4553 | /* add one reference for the tree */ |
0b32f4bb | 4554 | check_buffer_tree_ref(eb); |
f044ba78 | 4555 | spin_unlock(&tree->buffer_lock); |
19fe0a8b | 4556 | radix_tree_preload_end(); |
eb14ab8e CM |
4557 | |
4558 | /* | |
4559 | * there is a race where release page may have | |
4560 | * tried to find this extent buffer in the radix | |
4561 | * but failed. It will tell the VM it is safe to | |
4562 | * reclaim the, and it will clear the page private bit. | |
4563 | * We must make sure to set the page private bit properly | |
4564 | * after the extent buffer is in the radix tree so | |
4565 | * it doesn't get lost | |
4566 | */ | |
727011e0 CM |
4567 | SetPageChecked(eb->pages[0]); |
4568 | for (i = 1; i < num_pages; i++) { | |
4569 | p = extent_buffer_page(eb, i); | |
727011e0 CM |
4570 | ClearPageChecked(p); |
4571 | unlock_page(p); | |
4572 | } | |
4573 | unlock_page(eb->pages[0]); | |
d1310b2e CM |
4574 | return eb; |
4575 | ||
6af118ce | 4576 | free_eb: |
727011e0 CM |
4577 | for (i = 0; i < num_pages; i++) { |
4578 | if (eb->pages[i]) | |
4579 | unlock_page(eb->pages[i]); | |
4580 | } | |
eb14ab8e | 4581 | |
17de39ac | 4582 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
897ca6e9 | 4583 | btrfs_release_extent_buffer(eb); |
6af118ce | 4584 | return exists; |
d1310b2e | 4585 | } |
d1310b2e CM |
4586 | |
4587 | struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree, | |
f09d1f60 | 4588 | u64 start, unsigned long len) |
d1310b2e | 4589 | { |
d1310b2e | 4590 | struct extent_buffer *eb; |
d1310b2e | 4591 | |
19fe0a8b MX |
4592 | rcu_read_lock(); |
4593 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); | |
4594 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
4595 | rcu_read_unlock(); | |
5df4235e | 4596 | mark_extent_buffer_accessed(eb); |
19fe0a8b MX |
4597 | return eb; |
4598 | } | |
4599 | rcu_read_unlock(); | |
0f9dd46c | 4600 | |
19fe0a8b | 4601 | return NULL; |
d1310b2e | 4602 | } |
d1310b2e | 4603 | |
3083ee2e JB |
4604 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
4605 | { | |
4606 | struct extent_buffer *eb = | |
4607 | container_of(head, struct extent_buffer, rcu_head); | |
4608 | ||
4609 | __free_extent_buffer(eb); | |
4610 | } | |
4611 | ||
3083ee2e | 4612 | /* Expects to have eb->eb_lock already held */ |
f7a52a40 | 4613 | static int release_extent_buffer(struct extent_buffer *eb) |
3083ee2e JB |
4614 | { |
4615 | WARN_ON(atomic_read(&eb->refs) == 0); | |
4616 | if (atomic_dec_and_test(&eb->refs)) { | |
815a51c7 JS |
4617 | if (test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) { |
4618 | spin_unlock(&eb->refs_lock); | |
4619 | } else { | |
4620 | struct extent_io_tree *tree = eb->tree; | |
3083ee2e | 4621 | |
815a51c7 | 4622 | spin_unlock(&eb->refs_lock); |
3083ee2e | 4623 | |
815a51c7 JS |
4624 | spin_lock(&tree->buffer_lock); |
4625 | radix_tree_delete(&tree->buffer, | |
4626 | eb->start >> PAGE_CACHE_SHIFT); | |
4627 | spin_unlock(&tree->buffer_lock); | |
4628 | } | |
3083ee2e JB |
4629 | |
4630 | /* Should be safe to release our pages at this point */ | |
4631 | btrfs_release_extent_buffer_page(eb, 0); | |
3083ee2e | 4632 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 4633 | return 1; |
3083ee2e JB |
4634 | } |
4635 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
4636 | |
4637 | return 0; | |
3083ee2e JB |
4638 | } |
4639 | ||
d1310b2e CM |
4640 | void free_extent_buffer(struct extent_buffer *eb) |
4641 | { | |
242e18c7 CM |
4642 | int refs; |
4643 | int old; | |
d1310b2e CM |
4644 | if (!eb) |
4645 | return; | |
4646 | ||
242e18c7 CM |
4647 | while (1) { |
4648 | refs = atomic_read(&eb->refs); | |
4649 | if (refs <= 3) | |
4650 | break; | |
4651 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
4652 | if (old == refs) | |
4653 | return; | |
4654 | } | |
4655 | ||
3083ee2e | 4656 | spin_lock(&eb->refs_lock); |
815a51c7 JS |
4657 | if (atomic_read(&eb->refs) == 2 && |
4658 | test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) | |
4659 | atomic_dec(&eb->refs); | |
4660 | ||
3083ee2e JB |
4661 | if (atomic_read(&eb->refs) == 2 && |
4662 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 4663 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
4664 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
4665 | atomic_dec(&eb->refs); | |
4666 | ||
4667 | /* | |
4668 | * I know this is terrible, but it's temporary until we stop tracking | |
4669 | * the uptodate bits and such for the extent buffers. | |
4670 | */ | |
f7a52a40 | 4671 | release_extent_buffer(eb); |
3083ee2e JB |
4672 | } |
4673 | ||
4674 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
4675 | { | |
4676 | if (!eb) | |
d1310b2e CM |
4677 | return; |
4678 | ||
3083ee2e JB |
4679 | spin_lock(&eb->refs_lock); |
4680 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
4681 | ||
0b32f4bb | 4682 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
4683 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
4684 | atomic_dec(&eb->refs); | |
f7a52a40 | 4685 | release_extent_buffer(eb); |
d1310b2e | 4686 | } |
d1310b2e | 4687 | |
1d4284bd | 4688 | void clear_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 4689 | { |
d1310b2e CM |
4690 | unsigned long i; |
4691 | unsigned long num_pages; | |
4692 | struct page *page; | |
4693 | ||
d1310b2e CM |
4694 | num_pages = num_extent_pages(eb->start, eb->len); |
4695 | ||
4696 | for (i = 0; i < num_pages; i++) { | |
4697 | page = extent_buffer_page(eb, i); | |
b9473439 | 4698 | if (!PageDirty(page)) |
d2c3f4f6 CM |
4699 | continue; |
4700 | ||
a61e6f29 | 4701 | lock_page(page); |
eb14ab8e CM |
4702 | WARN_ON(!PagePrivate(page)); |
4703 | ||
d1310b2e | 4704 | clear_page_dirty_for_io(page); |
0ee0fda0 | 4705 | spin_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
4706 | if (!PageDirty(page)) { |
4707 | radix_tree_tag_clear(&page->mapping->page_tree, | |
4708 | page_index(page), | |
4709 | PAGECACHE_TAG_DIRTY); | |
4710 | } | |
0ee0fda0 | 4711 | spin_unlock_irq(&page->mapping->tree_lock); |
bf0da8c1 | 4712 | ClearPageError(page); |
a61e6f29 | 4713 | unlock_page(page); |
d1310b2e | 4714 | } |
0b32f4bb | 4715 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 4716 | } |
d1310b2e | 4717 | |
0b32f4bb | 4718 | int set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e CM |
4719 | { |
4720 | unsigned long i; | |
4721 | unsigned long num_pages; | |
b9473439 | 4722 | int was_dirty = 0; |
d1310b2e | 4723 | |
0b32f4bb JB |
4724 | check_buffer_tree_ref(eb); |
4725 | ||
b9473439 | 4726 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 4727 | |
d1310b2e | 4728 | num_pages = num_extent_pages(eb->start, eb->len); |
3083ee2e | 4729 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
4730 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
4731 | ||
b9473439 | 4732 | for (i = 0; i < num_pages; i++) |
0b32f4bb | 4733 | set_page_dirty(extent_buffer_page(eb, i)); |
b9473439 | 4734 | return was_dirty; |
d1310b2e | 4735 | } |
d1310b2e | 4736 | |
0b32f4bb | 4737 | int clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 CM |
4738 | { |
4739 | unsigned long i; | |
4740 | struct page *page; | |
4741 | unsigned long num_pages; | |
4742 | ||
b4ce94de | 4743 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
0b32f4bb | 4744 | num_pages = num_extent_pages(eb->start, eb->len); |
1259ab75 CM |
4745 | for (i = 0; i < num_pages; i++) { |
4746 | page = extent_buffer_page(eb, i); | |
33958dc6 CM |
4747 | if (page) |
4748 | ClearPageUptodate(page); | |
1259ab75 CM |
4749 | } |
4750 | return 0; | |
4751 | } | |
4752 | ||
0b32f4bb | 4753 | int set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e CM |
4754 | { |
4755 | unsigned long i; | |
4756 | struct page *page; | |
4757 | unsigned long num_pages; | |
4758 | ||
0b32f4bb | 4759 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 4760 | num_pages = num_extent_pages(eb->start, eb->len); |
d1310b2e CM |
4761 | for (i = 0; i < num_pages; i++) { |
4762 | page = extent_buffer_page(eb, i); | |
d1310b2e CM |
4763 | SetPageUptodate(page); |
4764 | } | |
4765 | return 0; | |
4766 | } | |
d1310b2e | 4767 | |
0b32f4bb | 4768 | int extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 4769 | { |
0b32f4bb | 4770 | return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 4771 | } |
d1310b2e CM |
4772 | |
4773 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
bb82ab88 | 4774 | struct extent_buffer *eb, u64 start, int wait, |
f188591e | 4775 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
4776 | { |
4777 | unsigned long i; | |
4778 | unsigned long start_i; | |
4779 | struct page *page; | |
4780 | int err; | |
4781 | int ret = 0; | |
ce9adaa5 CM |
4782 | int locked_pages = 0; |
4783 | int all_uptodate = 1; | |
d1310b2e | 4784 | unsigned long num_pages; |
727011e0 | 4785 | unsigned long num_reads = 0; |
a86c12c7 | 4786 | struct bio *bio = NULL; |
c8b97818 | 4787 | unsigned long bio_flags = 0; |
a86c12c7 | 4788 | |
b4ce94de | 4789 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
4790 | return 0; |
4791 | ||
d1310b2e CM |
4792 | if (start) { |
4793 | WARN_ON(start < eb->start); | |
4794 | start_i = (start >> PAGE_CACHE_SHIFT) - | |
4795 | (eb->start >> PAGE_CACHE_SHIFT); | |
4796 | } else { | |
4797 | start_i = 0; | |
4798 | } | |
4799 | ||
4800 | num_pages = num_extent_pages(eb->start, eb->len); | |
4801 | for (i = start_i; i < num_pages; i++) { | |
4802 | page = extent_buffer_page(eb, i); | |
bb82ab88 | 4803 | if (wait == WAIT_NONE) { |
2db04966 | 4804 | if (!trylock_page(page)) |
ce9adaa5 | 4805 | goto unlock_exit; |
d1310b2e CM |
4806 | } else { |
4807 | lock_page(page); | |
4808 | } | |
ce9adaa5 | 4809 | locked_pages++; |
727011e0 CM |
4810 | if (!PageUptodate(page)) { |
4811 | num_reads++; | |
ce9adaa5 | 4812 | all_uptodate = 0; |
727011e0 | 4813 | } |
ce9adaa5 CM |
4814 | } |
4815 | if (all_uptodate) { | |
4816 | if (start_i == 0) | |
b4ce94de | 4817 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
4818 | goto unlock_exit; |
4819 | } | |
4820 | ||
ea466794 | 4821 | clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags); |
5cf1ab56 | 4822 | eb->read_mirror = 0; |
0b32f4bb | 4823 | atomic_set(&eb->io_pages, num_reads); |
ce9adaa5 CM |
4824 | for (i = start_i; i < num_pages; i++) { |
4825 | page = extent_buffer_page(eb, i); | |
ce9adaa5 | 4826 | if (!PageUptodate(page)) { |
f188591e | 4827 | ClearPageError(page); |
a86c12c7 | 4828 | err = __extent_read_full_page(tree, page, |
f188591e | 4829 | get_extent, &bio, |
d4c7ca86 JB |
4830 | mirror_num, &bio_flags, |
4831 | READ | REQ_META); | |
d397712b | 4832 | if (err) |
d1310b2e | 4833 | ret = err; |
d1310b2e CM |
4834 | } else { |
4835 | unlock_page(page); | |
4836 | } | |
4837 | } | |
4838 | ||
355808c2 | 4839 | if (bio) { |
d4c7ca86 JB |
4840 | err = submit_one_bio(READ | REQ_META, bio, mirror_num, |
4841 | bio_flags); | |
79787eaa JM |
4842 | if (err) |
4843 | return err; | |
355808c2 | 4844 | } |
a86c12c7 | 4845 | |
bb82ab88 | 4846 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 4847 | return ret; |
d397712b | 4848 | |
d1310b2e CM |
4849 | for (i = start_i; i < num_pages; i++) { |
4850 | page = extent_buffer_page(eb, i); | |
4851 | wait_on_page_locked(page); | |
d397712b | 4852 | if (!PageUptodate(page)) |
d1310b2e | 4853 | ret = -EIO; |
d1310b2e | 4854 | } |
d397712b | 4855 | |
d1310b2e | 4856 | return ret; |
ce9adaa5 CM |
4857 | |
4858 | unlock_exit: | |
4859 | i = start_i; | |
d397712b | 4860 | while (locked_pages > 0) { |
ce9adaa5 CM |
4861 | page = extent_buffer_page(eb, i); |
4862 | i++; | |
4863 | unlock_page(page); | |
4864 | locked_pages--; | |
4865 | } | |
4866 | return ret; | |
d1310b2e | 4867 | } |
d1310b2e CM |
4868 | |
4869 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
4870 | unsigned long start, | |
4871 | unsigned long len) | |
4872 | { | |
4873 | size_t cur; | |
4874 | size_t offset; | |
4875 | struct page *page; | |
4876 | char *kaddr; | |
4877 | char *dst = (char *)dstv; | |
4878 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4879 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
4880 | |
4881 | WARN_ON(start > eb->len); | |
4882 | WARN_ON(start + len > eb->start + eb->len); | |
4883 | ||
4884 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4885 | ||
d397712b | 4886 | while (len > 0) { |
d1310b2e | 4887 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
4888 | |
4889 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
a6591715 | 4890 | kaddr = page_address(page); |
d1310b2e | 4891 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
4892 | |
4893 | dst += cur; | |
4894 | len -= cur; | |
4895 | offset = 0; | |
4896 | i++; | |
4897 | } | |
4898 | } | |
d1310b2e CM |
4899 | |
4900 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
a6591715 | 4901 | unsigned long min_len, char **map, |
d1310b2e | 4902 | unsigned long *map_start, |
a6591715 | 4903 | unsigned long *map_len) |
d1310b2e CM |
4904 | { |
4905 | size_t offset = start & (PAGE_CACHE_SIZE - 1); | |
4906 | char *kaddr; | |
4907 | struct page *p; | |
4908 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4909 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4910 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
4911 | PAGE_CACHE_SHIFT; | |
4912 | ||
4913 | if (i != end_i) | |
4914 | return -EINVAL; | |
4915 | ||
4916 | if (i == 0) { | |
4917 | offset = start_offset; | |
4918 | *map_start = 0; | |
4919 | } else { | |
4920 | offset = 0; | |
4921 | *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; | |
4922 | } | |
d397712b | 4923 | |
d1310b2e | 4924 | if (start + min_len > eb->len) { |
31b1a2bd | 4925 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, " |
d397712b CM |
4926 | "wanted %lu %lu\n", (unsigned long long)eb->start, |
4927 | eb->len, start, min_len); | |
85026533 | 4928 | return -EINVAL; |
d1310b2e CM |
4929 | } |
4930 | ||
4931 | p = extent_buffer_page(eb, i); | |
a6591715 | 4932 | kaddr = page_address(p); |
d1310b2e CM |
4933 | *map = kaddr + offset; |
4934 | *map_len = PAGE_CACHE_SIZE - offset; | |
4935 | return 0; | |
4936 | } | |
d1310b2e | 4937 | |
d1310b2e CM |
4938 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, |
4939 | unsigned long start, | |
4940 | unsigned long len) | |
4941 | { | |
4942 | size_t cur; | |
4943 | size_t offset; | |
4944 | struct page *page; | |
4945 | char *kaddr; | |
4946 | char *ptr = (char *)ptrv; | |
4947 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4948 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4949 | int ret = 0; | |
4950 | ||
4951 | WARN_ON(start > eb->len); | |
4952 | WARN_ON(start + len > eb->start + eb->len); | |
4953 | ||
4954 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4955 | ||
d397712b | 4956 | while (len > 0) { |
d1310b2e | 4957 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
4958 | |
4959 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
4960 | ||
a6591715 | 4961 | kaddr = page_address(page); |
d1310b2e | 4962 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
4963 | if (ret) |
4964 | break; | |
4965 | ||
4966 | ptr += cur; | |
4967 | len -= cur; | |
4968 | offset = 0; | |
4969 | i++; | |
4970 | } | |
4971 | return ret; | |
4972 | } | |
d1310b2e CM |
4973 | |
4974 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
4975 | unsigned long start, unsigned long len) | |
4976 | { | |
4977 | size_t cur; | |
4978 | size_t offset; | |
4979 | struct page *page; | |
4980 | char *kaddr; | |
4981 | char *src = (char *)srcv; | |
4982 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4983 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4984 | ||
4985 | WARN_ON(start > eb->len); | |
4986 | WARN_ON(start + len > eb->start + eb->len); | |
4987 | ||
4988 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4989 | ||
d397712b | 4990 | while (len > 0) { |
d1310b2e CM |
4991 | page = extent_buffer_page(eb, i); |
4992 | WARN_ON(!PageUptodate(page)); | |
4993 | ||
4994 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 4995 | kaddr = page_address(page); |
d1310b2e | 4996 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
4997 | |
4998 | src += cur; | |
4999 | len -= cur; | |
5000 | offset = 0; | |
5001 | i++; | |
5002 | } | |
5003 | } | |
d1310b2e CM |
5004 | |
5005 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
5006 | unsigned long start, unsigned long len) | |
5007 | { | |
5008 | size_t cur; | |
5009 | size_t offset; | |
5010 | struct page *page; | |
5011 | char *kaddr; | |
5012 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5013 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
5014 | ||
5015 | WARN_ON(start > eb->len); | |
5016 | WARN_ON(start + len > eb->start + eb->len); | |
5017 | ||
5018 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
5019 | ||
d397712b | 5020 | while (len > 0) { |
d1310b2e CM |
5021 | page = extent_buffer_page(eb, i); |
5022 | WARN_ON(!PageUptodate(page)); | |
5023 | ||
5024 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 5025 | kaddr = page_address(page); |
d1310b2e | 5026 | memset(kaddr + offset, c, cur); |
d1310b2e CM |
5027 | |
5028 | len -= cur; | |
5029 | offset = 0; | |
5030 | i++; | |
5031 | } | |
5032 | } | |
d1310b2e CM |
5033 | |
5034 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
5035 | unsigned long dst_offset, unsigned long src_offset, | |
5036 | unsigned long len) | |
5037 | { | |
5038 | u64 dst_len = dst->len; | |
5039 | size_t cur; | |
5040 | size_t offset; | |
5041 | struct page *page; | |
5042 | char *kaddr; | |
5043 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5044 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
5045 | ||
5046 | WARN_ON(src->len != dst_len); | |
5047 | ||
5048 | offset = (start_offset + dst_offset) & | |
5049 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
5050 | ||
d397712b | 5051 | while (len > 0) { |
d1310b2e CM |
5052 | page = extent_buffer_page(dst, i); |
5053 | WARN_ON(!PageUptodate(page)); | |
5054 | ||
5055 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
5056 | ||
a6591715 | 5057 | kaddr = page_address(page); |
d1310b2e | 5058 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
5059 | |
5060 | src_offset += cur; | |
5061 | len -= cur; | |
5062 | offset = 0; | |
5063 | i++; | |
5064 | } | |
5065 | } | |
d1310b2e CM |
5066 | |
5067 | static void move_pages(struct page *dst_page, struct page *src_page, | |
5068 | unsigned long dst_off, unsigned long src_off, | |
5069 | unsigned long len) | |
5070 | { | |
a6591715 | 5071 | char *dst_kaddr = page_address(dst_page); |
d1310b2e CM |
5072 | if (dst_page == src_page) { |
5073 | memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len); | |
5074 | } else { | |
a6591715 | 5075 | char *src_kaddr = page_address(src_page); |
d1310b2e CM |
5076 | char *p = dst_kaddr + dst_off + len; |
5077 | char *s = src_kaddr + src_off + len; | |
5078 | ||
5079 | while (len--) | |
5080 | *--p = *--s; | |
d1310b2e | 5081 | } |
d1310b2e CM |
5082 | } |
5083 | ||
3387206f ST |
5084 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
5085 | { | |
5086 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
5087 | return distance < len; | |
5088 | } | |
5089 | ||
d1310b2e CM |
5090 | static void copy_pages(struct page *dst_page, struct page *src_page, |
5091 | unsigned long dst_off, unsigned long src_off, | |
5092 | unsigned long len) | |
5093 | { | |
a6591715 | 5094 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 5095 | char *src_kaddr; |
727011e0 | 5096 | int must_memmove = 0; |
d1310b2e | 5097 | |
3387206f | 5098 | if (dst_page != src_page) { |
a6591715 | 5099 | src_kaddr = page_address(src_page); |
3387206f | 5100 | } else { |
d1310b2e | 5101 | src_kaddr = dst_kaddr; |
727011e0 CM |
5102 | if (areas_overlap(src_off, dst_off, len)) |
5103 | must_memmove = 1; | |
3387206f | 5104 | } |
d1310b2e | 5105 | |
727011e0 CM |
5106 | if (must_memmove) |
5107 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
5108 | else | |
5109 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
5110 | } |
5111 | ||
5112 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5113 | unsigned long src_offset, unsigned long len) | |
5114 | { | |
5115 | size_t cur; | |
5116 | size_t dst_off_in_page; | |
5117 | size_t src_off_in_page; | |
5118 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5119 | unsigned long dst_i; | |
5120 | unsigned long src_i; | |
5121 | ||
5122 | if (src_offset + len > dst->len) { | |
d397712b CM |
5123 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
5124 | "len %lu dst len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
5125 | BUG_ON(1); |
5126 | } | |
5127 | if (dst_offset + len > dst->len) { | |
d397712b CM |
5128 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
5129 | "len %lu dst len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
5130 | BUG_ON(1); |
5131 | } | |
5132 | ||
d397712b | 5133 | while (len > 0) { |
d1310b2e CM |
5134 | dst_off_in_page = (start_offset + dst_offset) & |
5135 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
5136 | src_off_in_page = (start_offset + src_offset) & | |
5137 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
5138 | ||
5139 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
5140 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
5141 | ||
5142 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
5143 | src_off_in_page)); | |
5144 | cur = min_t(unsigned long, cur, | |
5145 | (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); | |
5146 | ||
5147 | copy_pages(extent_buffer_page(dst, dst_i), | |
5148 | extent_buffer_page(dst, src_i), | |
5149 | dst_off_in_page, src_off_in_page, cur); | |
5150 | ||
5151 | src_offset += cur; | |
5152 | dst_offset += cur; | |
5153 | len -= cur; | |
5154 | } | |
5155 | } | |
d1310b2e CM |
5156 | |
5157 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
5158 | unsigned long src_offset, unsigned long len) | |
5159 | { | |
5160 | size_t cur; | |
5161 | size_t dst_off_in_page; | |
5162 | size_t src_off_in_page; | |
5163 | unsigned long dst_end = dst_offset + len - 1; | |
5164 | unsigned long src_end = src_offset + len - 1; | |
5165 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
5166 | unsigned long dst_i; | |
5167 | unsigned long src_i; | |
5168 | ||
5169 | if (src_offset + len > dst->len) { | |
d397712b CM |
5170 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
5171 | "len %lu len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
5172 | BUG_ON(1); |
5173 | } | |
5174 | if (dst_offset + len > dst->len) { | |
d397712b CM |
5175 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
5176 | "len %lu len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
5177 | BUG_ON(1); |
5178 | } | |
727011e0 | 5179 | if (dst_offset < src_offset) { |
d1310b2e CM |
5180 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
5181 | return; | |
5182 | } | |
d397712b | 5183 | while (len > 0) { |
d1310b2e CM |
5184 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; |
5185 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
5186 | ||
5187 | dst_off_in_page = (start_offset + dst_end) & | |
5188 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
5189 | src_off_in_page = (start_offset + src_end) & | |
5190 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
5191 | ||
5192 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
5193 | cur = min(cur, dst_off_in_page + 1); | |
5194 | move_pages(extent_buffer_page(dst, dst_i), | |
5195 | extent_buffer_page(dst, src_i), | |
5196 | dst_off_in_page - cur + 1, | |
5197 | src_off_in_page - cur + 1, cur); | |
5198 | ||
5199 | dst_end -= cur; | |
5200 | src_end -= cur; | |
5201 | len -= cur; | |
5202 | } | |
5203 | } | |
6af118ce | 5204 | |
f7a52a40 | 5205 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 5206 | { |
6af118ce | 5207 | struct extent_buffer *eb; |
6af118ce | 5208 | |
3083ee2e JB |
5209 | /* |
5210 | * We need to make sure noboody is attaching this page to an eb right | |
5211 | * now. | |
5212 | */ | |
5213 | spin_lock(&page->mapping->private_lock); | |
5214 | if (!PagePrivate(page)) { | |
5215 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 5216 | return 1; |
45f49bce | 5217 | } |
6af118ce | 5218 | |
3083ee2e JB |
5219 | eb = (struct extent_buffer *)page->private; |
5220 | BUG_ON(!eb); | |
19fe0a8b MX |
5221 | |
5222 | /* | |
3083ee2e JB |
5223 | * This is a little awful but should be ok, we need to make sure that |
5224 | * the eb doesn't disappear out from under us while we're looking at | |
5225 | * this page. | |
19fe0a8b | 5226 | */ |
3083ee2e | 5227 | spin_lock(&eb->refs_lock); |
0b32f4bb | 5228 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
5229 | spin_unlock(&eb->refs_lock); |
5230 | spin_unlock(&page->mapping->private_lock); | |
5231 | return 0; | |
b9473439 | 5232 | } |
3083ee2e | 5233 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 5234 | |
19fe0a8b | 5235 | /* |
3083ee2e JB |
5236 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
5237 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 5238 | */ |
3083ee2e JB |
5239 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
5240 | spin_unlock(&eb->refs_lock); | |
5241 | return 0; | |
b9473439 | 5242 | } |
19fe0a8b | 5243 | |
f7a52a40 | 5244 | return release_extent_buffer(eb); |
6af118ce | 5245 | } |