]>
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> | |
7 | #include <linux/module.h> | |
8 | #include <linux/spinlock.h> | |
9 | #include <linux/blkdev.h> | |
10 | #include <linux/swap.h> | |
d1310b2e CM |
11 | #include <linux/writeback.h> |
12 | #include <linux/pagevec.h> | |
268bb0ce | 13 | #include <linux/prefetch.h> |
90a887c9 | 14 | #include <linux/cleancache.h> |
d1310b2e CM |
15 | #include "extent_io.h" |
16 | #include "extent_map.h" | |
2db04966 | 17 | #include "compat.h" |
902b22f3 DW |
18 | #include "ctree.h" |
19 | #include "btrfs_inode.h" | |
4a54c8c1 | 20 | #include "volumes.h" |
21adbd5c | 21 | #include "check-integrity.h" |
0b32f4bb | 22 | #include "locking.h" |
606686ee | 23 | #include "rcu-string.h" |
d1310b2e | 24 | |
d1310b2e CM |
25 | static struct kmem_cache *extent_state_cache; |
26 | static struct kmem_cache *extent_buffer_cache; | |
27 | ||
28 | static LIST_HEAD(buffers); | |
29 | static LIST_HEAD(states); | |
4bef0848 | 30 | |
b47eda86 | 31 | #define LEAK_DEBUG 0 |
3935127c | 32 | #if LEAK_DEBUG |
d397712b | 33 | static DEFINE_SPINLOCK(leak_lock); |
4bef0848 | 34 | #endif |
d1310b2e | 35 | |
d1310b2e CM |
36 | #define BUFFER_LRU_MAX 64 |
37 | ||
38 | struct tree_entry { | |
39 | u64 start; | |
40 | u64 end; | |
d1310b2e CM |
41 | struct rb_node rb_node; |
42 | }; | |
43 | ||
44 | struct extent_page_data { | |
45 | struct bio *bio; | |
46 | struct extent_io_tree *tree; | |
47 | get_extent_t *get_extent; | |
de0022b9 | 48 | unsigned long bio_flags; |
771ed689 CM |
49 | |
50 | /* tells writepage not to lock the state bits for this range | |
51 | * it still does the unlocking | |
52 | */ | |
ffbd517d CM |
53 | unsigned int extent_locked:1; |
54 | ||
55 | /* tells the submit_bio code to use a WRITE_SYNC */ | |
56 | unsigned int sync_io:1; | |
d1310b2e CM |
57 | }; |
58 | ||
0b32f4bb | 59 | static noinline void flush_write_bio(void *data); |
c2d904e0 JM |
60 | static inline struct btrfs_fs_info * |
61 | tree_fs_info(struct extent_io_tree *tree) | |
62 | { | |
63 | return btrfs_sb(tree->mapping->host->i_sb); | |
64 | } | |
0b32f4bb | 65 | |
d1310b2e CM |
66 | int __init extent_io_init(void) |
67 | { | |
837e1972 | 68 | extent_state_cache = kmem_cache_create("btrfs_extent_state", |
9601e3f6 CH |
69 | sizeof(struct extent_state), 0, |
70 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
71 | if (!extent_state_cache) |
72 | return -ENOMEM; | |
73 | ||
837e1972 | 74 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 CH |
75 | sizeof(struct extent_buffer), 0, |
76 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
77 | if (!extent_buffer_cache) |
78 | goto free_state_cache; | |
79 | return 0; | |
80 | ||
81 | free_state_cache: | |
82 | kmem_cache_destroy(extent_state_cache); | |
83 | return -ENOMEM; | |
84 | } | |
85 | ||
86 | void extent_io_exit(void) | |
87 | { | |
88 | struct extent_state *state; | |
2d2ae547 | 89 | struct extent_buffer *eb; |
d1310b2e CM |
90 | |
91 | while (!list_empty(&states)) { | |
2d2ae547 | 92 | state = list_entry(states.next, struct extent_state, leak_list); |
d397712b CM |
93 | printk(KERN_ERR "btrfs state leak: start %llu end %llu " |
94 | "state %lu in tree %p refs %d\n", | |
95 | (unsigned long long)state->start, | |
96 | (unsigned long long)state->end, | |
97 | state->state, state->tree, atomic_read(&state->refs)); | |
2d2ae547 | 98 | list_del(&state->leak_list); |
d1310b2e CM |
99 | kmem_cache_free(extent_state_cache, state); |
100 | ||
101 | } | |
102 | ||
2d2ae547 CM |
103 | while (!list_empty(&buffers)) { |
104 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
d397712b CM |
105 | printk(KERN_ERR "btrfs buffer leak start %llu len %lu " |
106 | "refs %d\n", (unsigned long long)eb->start, | |
107 | eb->len, atomic_read(&eb->refs)); | |
2d2ae547 CM |
108 | list_del(&eb->leak_list); |
109 | kmem_cache_free(extent_buffer_cache, eb); | |
110 | } | |
8c0a8537 KS |
111 | |
112 | /* | |
113 | * Make sure all delayed rcu free are flushed before we | |
114 | * destroy caches. | |
115 | */ | |
116 | rcu_barrier(); | |
d1310b2e CM |
117 | if (extent_state_cache) |
118 | kmem_cache_destroy(extent_state_cache); | |
119 | if (extent_buffer_cache) | |
120 | kmem_cache_destroy(extent_buffer_cache); | |
121 | } | |
122 | ||
123 | void extent_io_tree_init(struct extent_io_tree *tree, | |
f993c883 | 124 | struct address_space *mapping) |
d1310b2e | 125 | { |
6bef4d31 | 126 | tree->state = RB_ROOT; |
19fe0a8b | 127 | INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC); |
d1310b2e CM |
128 | tree->ops = NULL; |
129 | tree->dirty_bytes = 0; | |
70dec807 | 130 | spin_lock_init(&tree->lock); |
6af118ce | 131 | spin_lock_init(&tree->buffer_lock); |
d1310b2e | 132 | tree->mapping = mapping; |
d1310b2e | 133 | } |
d1310b2e | 134 | |
b2950863 | 135 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
136 | { |
137 | struct extent_state *state; | |
3935127c | 138 | #if LEAK_DEBUG |
2d2ae547 | 139 | unsigned long flags; |
4bef0848 | 140 | #endif |
d1310b2e CM |
141 | |
142 | state = kmem_cache_alloc(extent_state_cache, mask); | |
2b114d1d | 143 | if (!state) |
d1310b2e CM |
144 | return state; |
145 | state->state = 0; | |
d1310b2e | 146 | state->private = 0; |
70dec807 | 147 | state->tree = NULL; |
3935127c | 148 | #if LEAK_DEBUG |
2d2ae547 CM |
149 | spin_lock_irqsave(&leak_lock, flags); |
150 | list_add(&state->leak_list, &states); | |
151 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 152 | #endif |
d1310b2e CM |
153 | atomic_set(&state->refs, 1); |
154 | init_waitqueue_head(&state->wq); | |
143bede5 | 155 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
156 | return state; |
157 | } | |
d1310b2e | 158 | |
4845e44f | 159 | void free_extent_state(struct extent_state *state) |
d1310b2e | 160 | { |
d1310b2e CM |
161 | if (!state) |
162 | return; | |
163 | if (atomic_dec_and_test(&state->refs)) { | |
3935127c | 164 | #if LEAK_DEBUG |
2d2ae547 | 165 | unsigned long flags; |
4bef0848 | 166 | #endif |
70dec807 | 167 | WARN_ON(state->tree); |
3935127c | 168 | #if LEAK_DEBUG |
2d2ae547 CM |
169 | spin_lock_irqsave(&leak_lock, flags); |
170 | list_del(&state->leak_list); | |
171 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 172 | #endif |
143bede5 | 173 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
174 | kmem_cache_free(extent_state_cache, state); |
175 | } | |
176 | } | |
d1310b2e CM |
177 | |
178 | static struct rb_node *tree_insert(struct rb_root *root, u64 offset, | |
179 | struct rb_node *node) | |
180 | { | |
d397712b CM |
181 | struct rb_node **p = &root->rb_node; |
182 | struct rb_node *parent = NULL; | |
d1310b2e CM |
183 | struct tree_entry *entry; |
184 | ||
d397712b | 185 | while (*p) { |
d1310b2e CM |
186 | parent = *p; |
187 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
188 | ||
189 | if (offset < entry->start) | |
190 | p = &(*p)->rb_left; | |
191 | else if (offset > entry->end) | |
192 | p = &(*p)->rb_right; | |
193 | else | |
194 | return parent; | |
195 | } | |
196 | ||
d1310b2e CM |
197 | rb_link_node(node, parent, p); |
198 | rb_insert_color(node, root); | |
199 | return NULL; | |
200 | } | |
201 | ||
80ea96b1 | 202 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
d1310b2e CM |
203 | struct rb_node **prev_ret, |
204 | struct rb_node **next_ret) | |
205 | { | |
80ea96b1 | 206 | struct rb_root *root = &tree->state; |
d397712b | 207 | struct rb_node *n = root->rb_node; |
d1310b2e CM |
208 | struct rb_node *prev = NULL; |
209 | struct rb_node *orig_prev = NULL; | |
210 | struct tree_entry *entry; | |
211 | struct tree_entry *prev_entry = NULL; | |
212 | ||
d397712b | 213 | while (n) { |
d1310b2e CM |
214 | entry = rb_entry(n, struct tree_entry, rb_node); |
215 | prev = n; | |
216 | prev_entry = entry; | |
217 | ||
218 | if (offset < entry->start) | |
219 | n = n->rb_left; | |
220 | else if (offset > entry->end) | |
221 | n = n->rb_right; | |
d397712b | 222 | else |
d1310b2e CM |
223 | return n; |
224 | } | |
225 | ||
226 | if (prev_ret) { | |
227 | orig_prev = prev; | |
d397712b | 228 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
229 | prev = rb_next(prev); |
230 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
231 | } | |
232 | *prev_ret = prev; | |
233 | prev = orig_prev; | |
234 | } | |
235 | ||
236 | if (next_ret) { | |
237 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 238 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
239 | prev = rb_prev(prev); |
240 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
241 | } | |
242 | *next_ret = prev; | |
243 | } | |
244 | return NULL; | |
245 | } | |
246 | ||
80ea96b1 CM |
247 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
248 | u64 offset) | |
d1310b2e | 249 | { |
70dec807 | 250 | struct rb_node *prev = NULL; |
d1310b2e | 251 | struct rb_node *ret; |
70dec807 | 252 | |
80ea96b1 | 253 | ret = __etree_search(tree, offset, &prev, NULL); |
d397712b | 254 | if (!ret) |
d1310b2e CM |
255 | return prev; |
256 | return ret; | |
257 | } | |
258 | ||
9ed74f2d JB |
259 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, |
260 | struct extent_state *other) | |
261 | { | |
262 | if (tree->ops && tree->ops->merge_extent_hook) | |
263 | tree->ops->merge_extent_hook(tree->mapping->host, new, | |
264 | other); | |
265 | } | |
266 | ||
d1310b2e CM |
267 | /* |
268 | * utility function to look for merge candidates inside a given range. | |
269 | * Any extents with matching state are merged together into a single | |
270 | * extent in the tree. Extents with EXTENT_IO in their state field | |
271 | * are not merged because the end_io handlers need to be able to do | |
272 | * operations on them without sleeping (or doing allocations/splits). | |
273 | * | |
274 | * This should be called with the tree lock held. | |
275 | */ | |
1bf85046 JM |
276 | static void merge_state(struct extent_io_tree *tree, |
277 | struct extent_state *state) | |
d1310b2e CM |
278 | { |
279 | struct extent_state *other; | |
280 | struct rb_node *other_node; | |
281 | ||
5b21f2ed | 282 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
1bf85046 | 283 | return; |
d1310b2e CM |
284 | |
285 | other_node = rb_prev(&state->rb_node); | |
286 | if (other_node) { | |
287 | other = rb_entry(other_node, struct extent_state, rb_node); | |
288 | if (other->end == state->start - 1 && | |
289 | other->state == state->state) { | |
9ed74f2d | 290 | merge_cb(tree, state, other); |
d1310b2e | 291 | state->start = other->start; |
70dec807 | 292 | other->tree = NULL; |
d1310b2e CM |
293 | rb_erase(&other->rb_node, &tree->state); |
294 | free_extent_state(other); | |
295 | } | |
296 | } | |
297 | other_node = rb_next(&state->rb_node); | |
298 | if (other_node) { | |
299 | other = rb_entry(other_node, struct extent_state, rb_node); | |
300 | if (other->start == state->end + 1 && | |
301 | other->state == state->state) { | |
9ed74f2d | 302 | merge_cb(tree, state, other); |
df98b6e2 JB |
303 | state->end = other->end; |
304 | other->tree = NULL; | |
305 | rb_erase(&other->rb_node, &tree->state); | |
306 | free_extent_state(other); | |
d1310b2e CM |
307 | } |
308 | } | |
d1310b2e CM |
309 | } |
310 | ||
1bf85046 | 311 | static void set_state_cb(struct extent_io_tree *tree, |
0ca1f7ce | 312 | struct extent_state *state, int *bits) |
291d673e | 313 | { |
1bf85046 JM |
314 | if (tree->ops && tree->ops->set_bit_hook) |
315 | tree->ops->set_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
316 | } |
317 | ||
318 | static void clear_state_cb(struct extent_io_tree *tree, | |
0ca1f7ce | 319 | struct extent_state *state, int *bits) |
291d673e | 320 | { |
9ed74f2d JB |
321 | if (tree->ops && tree->ops->clear_bit_hook) |
322 | tree->ops->clear_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
323 | } |
324 | ||
3150b699 XG |
325 | static void set_state_bits(struct extent_io_tree *tree, |
326 | struct extent_state *state, int *bits); | |
327 | ||
d1310b2e CM |
328 | /* |
329 | * insert an extent_state struct into the tree. 'bits' are set on the | |
330 | * struct before it is inserted. | |
331 | * | |
332 | * This may return -EEXIST if the extent is already there, in which case the | |
333 | * state struct is freed. | |
334 | * | |
335 | * The tree lock is not taken internally. This is a utility function and | |
336 | * probably isn't what you want to call (see set/clear_extent_bit). | |
337 | */ | |
338 | static int insert_state(struct extent_io_tree *tree, | |
339 | struct extent_state *state, u64 start, u64 end, | |
0ca1f7ce | 340 | int *bits) |
d1310b2e CM |
341 | { |
342 | struct rb_node *node; | |
343 | ||
31b1a2bd JL |
344 | if (end < start) |
345 | WARN(1, KERN_ERR "btrfs end < start %llu %llu\n", | |
d397712b CM |
346 | (unsigned long long)end, |
347 | (unsigned long long)start); | |
d1310b2e CM |
348 | state->start = start; |
349 | state->end = end; | |
9ed74f2d | 350 | |
3150b699 XG |
351 | set_state_bits(tree, state, bits); |
352 | ||
d1310b2e CM |
353 | node = tree_insert(&tree->state, end, &state->rb_node); |
354 | if (node) { | |
355 | struct extent_state *found; | |
356 | found = rb_entry(node, struct extent_state, rb_node); | |
d397712b CM |
357 | printk(KERN_ERR "btrfs found node %llu %llu on insert of " |
358 | "%llu %llu\n", (unsigned long long)found->start, | |
359 | (unsigned long long)found->end, | |
360 | (unsigned long long)start, (unsigned long long)end); | |
d1310b2e CM |
361 | return -EEXIST; |
362 | } | |
70dec807 | 363 | state->tree = tree; |
d1310b2e CM |
364 | merge_state(tree, state); |
365 | return 0; | |
366 | } | |
367 | ||
1bf85046 | 368 | static void split_cb(struct extent_io_tree *tree, struct extent_state *orig, |
9ed74f2d JB |
369 | u64 split) |
370 | { | |
371 | if (tree->ops && tree->ops->split_extent_hook) | |
1bf85046 | 372 | tree->ops->split_extent_hook(tree->mapping->host, orig, split); |
9ed74f2d JB |
373 | } |
374 | ||
d1310b2e CM |
375 | /* |
376 | * split a given extent state struct in two, inserting the preallocated | |
377 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
378 | * offset inside 'orig' where it should be split. | |
379 | * | |
380 | * Before calling, | |
381 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
382 | * are two extent state structs in the tree: | |
383 | * prealloc: [orig->start, split - 1] | |
384 | * orig: [ split, orig->end ] | |
385 | * | |
386 | * The tree locks are not taken by this function. They need to be held | |
387 | * by the caller. | |
388 | */ | |
389 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
390 | struct extent_state *prealloc, u64 split) | |
391 | { | |
392 | struct rb_node *node; | |
9ed74f2d JB |
393 | |
394 | split_cb(tree, orig, split); | |
395 | ||
d1310b2e CM |
396 | prealloc->start = orig->start; |
397 | prealloc->end = split - 1; | |
398 | prealloc->state = orig->state; | |
399 | orig->start = split; | |
400 | ||
401 | node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node); | |
402 | if (node) { | |
d1310b2e CM |
403 | free_extent_state(prealloc); |
404 | return -EEXIST; | |
405 | } | |
70dec807 | 406 | prealloc->tree = tree; |
d1310b2e CM |
407 | return 0; |
408 | } | |
409 | ||
cdc6a395 LZ |
410 | static struct extent_state *next_state(struct extent_state *state) |
411 | { | |
412 | struct rb_node *next = rb_next(&state->rb_node); | |
413 | if (next) | |
414 | return rb_entry(next, struct extent_state, rb_node); | |
415 | else | |
416 | return NULL; | |
417 | } | |
418 | ||
d1310b2e CM |
419 | /* |
420 | * utility function to clear some bits in an extent state struct. | |
1b303fc0 | 421 | * it will optionally wake up any one waiting on this state (wake == 1). |
d1310b2e CM |
422 | * |
423 | * If no bits are set on the state struct after clearing things, the | |
424 | * struct is freed and removed from the tree | |
425 | */ | |
cdc6a395 LZ |
426 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
427 | struct extent_state *state, | |
428 | int *bits, int wake) | |
d1310b2e | 429 | { |
cdc6a395 | 430 | struct extent_state *next; |
0ca1f7ce | 431 | int bits_to_clear = *bits & ~EXTENT_CTLBITS; |
d1310b2e | 432 | |
0ca1f7ce | 433 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
434 | u64 range = state->end - state->start + 1; |
435 | WARN_ON(range > tree->dirty_bytes); | |
436 | tree->dirty_bytes -= range; | |
437 | } | |
291d673e | 438 | clear_state_cb(tree, state, bits); |
32c00aff | 439 | state->state &= ~bits_to_clear; |
d1310b2e CM |
440 | if (wake) |
441 | wake_up(&state->wq); | |
0ca1f7ce | 442 | if (state->state == 0) { |
cdc6a395 | 443 | next = next_state(state); |
70dec807 | 444 | if (state->tree) { |
d1310b2e | 445 | rb_erase(&state->rb_node, &tree->state); |
70dec807 | 446 | state->tree = NULL; |
d1310b2e CM |
447 | free_extent_state(state); |
448 | } else { | |
449 | WARN_ON(1); | |
450 | } | |
451 | } else { | |
452 | merge_state(tree, state); | |
cdc6a395 | 453 | next = next_state(state); |
d1310b2e | 454 | } |
cdc6a395 | 455 | return next; |
d1310b2e CM |
456 | } |
457 | ||
8233767a XG |
458 | static struct extent_state * |
459 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
460 | { | |
461 | if (!prealloc) | |
462 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
463 | ||
464 | return prealloc; | |
465 | } | |
466 | ||
c2d904e0 JM |
467 | void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
468 | { | |
469 | btrfs_panic(tree_fs_info(tree), err, "Locking error: " | |
470 | "Extent tree was modified by another " | |
471 | "thread while locked."); | |
472 | } | |
473 | ||
d1310b2e CM |
474 | /* |
475 | * clear some bits on a range in the tree. This may require splitting | |
476 | * or inserting elements in the tree, so the gfp mask is used to | |
477 | * indicate which allocations or sleeping are allowed. | |
478 | * | |
479 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
480 | * the given range from the tree regardless of state (ie for truncate). | |
481 | * | |
482 | * the range [start, end] is inclusive. | |
483 | * | |
6763af84 | 484 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e CM |
485 | */ |
486 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
2c64c53d CM |
487 | int bits, int wake, int delete, |
488 | struct extent_state **cached_state, | |
489 | gfp_t mask) | |
d1310b2e CM |
490 | { |
491 | struct extent_state *state; | |
2c64c53d | 492 | struct extent_state *cached; |
d1310b2e CM |
493 | struct extent_state *prealloc = NULL; |
494 | struct rb_node *node; | |
5c939df5 | 495 | u64 last_end; |
d1310b2e | 496 | int err; |
2ac55d41 | 497 | int clear = 0; |
d1310b2e | 498 | |
0ca1f7ce YZ |
499 | if (delete) |
500 | bits |= ~EXTENT_CTLBITS; | |
501 | bits |= EXTENT_FIRST_DELALLOC; | |
502 | ||
2ac55d41 JB |
503 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
504 | clear = 1; | |
d1310b2e CM |
505 | again: |
506 | if (!prealloc && (mask & __GFP_WAIT)) { | |
507 | prealloc = alloc_extent_state(mask); | |
508 | if (!prealloc) | |
509 | return -ENOMEM; | |
510 | } | |
511 | ||
cad321ad | 512 | spin_lock(&tree->lock); |
2c64c53d CM |
513 | if (cached_state) { |
514 | cached = *cached_state; | |
2ac55d41 JB |
515 | |
516 | if (clear) { | |
517 | *cached_state = NULL; | |
518 | cached_state = NULL; | |
519 | } | |
520 | ||
df98b6e2 JB |
521 | if (cached && cached->tree && cached->start <= start && |
522 | cached->end > start) { | |
2ac55d41 JB |
523 | if (clear) |
524 | atomic_dec(&cached->refs); | |
2c64c53d | 525 | state = cached; |
42daec29 | 526 | goto hit_next; |
2c64c53d | 527 | } |
2ac55d41 JB |
528 | if (clear) |
529 | free_extent_state(cached); | |
2c64c53d | 530 | } |
d1310b2e CM |
531 | /* |
532 | * this search will find the extents that end after | |
533 | * our range starts | |
534 | */ | |
80ea96b1 | 535 | node = tree_search(tree, start); |
d1310b2e CM |
536 | if (!node) |
537 | goto out; | |
538 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 539 | hit_next: |
d1310b2e CM |
540 | if (state->start > end) |
541 | goto out; | |
542 | WARN_ON(state->end < start); | |
5c939df5 | 543 | last_end = state->end; |
d1310b2e | 544 | |
0449314a | 545 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
546 | if (!(state->state & bits)) { |
547 | state = next_state(state); | |
0449314a | 548 | goto next; |
cdc6a395 | 549 | } |
0449314a | 550 | |
d1310b2e CM |
551 | /* |
552 | * | ---- desired range ---- | | |
553 | * | state | or | |
554 | * | ------------- state -------------- | | |
555 | * | |
556 | * We need to split the extent we found, and may flip | |
557 | * bits on second half. | |
558 | * | |
559 | * If the extent we found extends past our range, we | |
560 | * just split and search again. It'll get split again | |
561 | * the next time though. | |
562 | * | |
563 | * If the extent we found is inside our range, we clear | |
564 | * the desired bit on it. | |
565 | */ | |
566 | ||
567 | if (state->start < start) { | |
8233767a XG |
568 | prealloc = alloc_extent_state_atomic(prealloc); |
569 | BUG_ON(!prealloc); | |
d1310b2e | 570 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
571 | if (err) |
572 | extent_io_tree_panic(tree, err); | |
573 | ||
d1310b2e CM |
574 | prealloc = NULL; |
575 | if (err) | |
576 | goto out; | |
577 | if (state->end <= end) { | |
d1ac6e41 LB |
578 | state = clear_state_bit(tree, state, &bits, wake); |
579 | goto next; | |
d1310b2e CM |
580 | } |
581 | goto search_again; | |
582 | } | |
583 | /* | |
584 | * | ---- desired range ---- | | |
585 | * | state | | |
586 | * We need to split the extent, and clear the bit | |
587 | * on the first half | |
588 | */ | |
589 | if (state->start <= end && state->end > end) { | |
8233767a XG |
590 | prealloc = alloc_extent_state_atomic(prealloc); |
591 | BUG_ON(!prealloc); | |
d1310b2e | 592 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
593 | if (err) |
594 | extent_io_tree_panic(tree, err); | |
595 | ||
d1310b2e CM |
596 | if (wake) |
597 | wake_up(&state->wq); | |
42daec29 | 598 | |
6763af84 | 599 | clear_state_bit(tree, prealloc, &bits, wake); |
9ed74f2d | 600 | |
d1310b2e CM |
601 | prealloc = NULL; |
602 | goto out; | |
603 | } | |
42daec29 | 604 | |
cdc6a395 | 605 | state = clear_state_bit(tree, state, &bits, wake); |
0449314a | 606 | next: |
5c939df5 YZ |
607 | if (last_end == (u64)-1) |
608 | goto out; | |
609 | start = last_end + 1; | |
cdc6a395 | 610 | if (start <= end && state && !need_resched()) |
692e5759 | 611 | goto hit_next; |
d1310b2e CM |
612 | goto search_again; |
613 | ||
614 | out: | |
cad321ad | 615 | spin_unlock(&tree->lock); |
d1310b2e CM |
616 | if (prealloc) |
617 | free_extent_state(prealloc); | |
618 | ||
6763af84 | 619 | return 0; |
d1310b2e CM |
620 | |
621 | search_again: | |
622 | if (start > end) | |
623 | goto out; | |
cad321ad | 624 | spin_unlock(&tree->lock); |
d1310b2e CM |
625 | if (mask & __GFP_WAIT) |
626 | cond_resched(); | |
627 | goto again; | |
628 | } | |
d1310b2e | 629 | |
143bede5 JM |
630 | static void wait_on_state(struct extent_io_tree *tree, |
631 | struct extent_state *state) | |
641f5219 CH |
632 | __releases(tree->lock) |
633 | __acquires(tree->lock) | |
d1310b2e CM |
634 | { |
635 | DEFINE_WAIT(wait); | |
636 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 637 | spin_unlock(&tree->lock); |
d1310b2e | 638 | schedule(); |
cad321ad | 639 | spin_lock(&tree->lock); |
d1310b2e | 640 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
641 | } |
642 | ||
643 | /* | |
644 | * waits for one or more bits to clear on a range in the state tree. | |
645 | * The range [start, end] is inclusive. | |
646 | * The tree lock is taken by this function | |
647 | */ | |
143bede5 | 648 | void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits) |
d1310b2e CM |
649 | { |
650 | struct extent_state *state; | |
651 | struct rb_node *node; | |
652 | ||
cad321ad | 653 | spin_lock(&tree->lock); |
d1310b2e CM |
654 | again: |
655 | while (1) { | |
656 | /* | |
657 | * this search will find all the extents that end after | |
658 | * our range starts | |
659 | */ | |
80ea96b1 | 660 | node = tree_search(tree, start); |
d1310b2e CM |
661 | if (!node) |
662 | break; | |
663 | ||
664 | state = rb_entry(node, struct extent_state, rb_node); | |
665 | ||
666 | if (state->start > end) | |
667 | goto out; | |
668 | ||
669 | if (state->state & bits) { | |
670 | start = state->start; | |
671 | atomic_inc(&state->refs); | |
672 | wait_on_state(tree, state); | |
673 | free_extent_state(state); | |
674 | goto again; | |
675 | } | |
676 | start = state->end + 1; | |
677 | ||
678 | if (start > end) | |
679 | break; | |
680 | ||
ded91f08 | 681 | cond_resched_lock(&tree->lock); |
d1310b2e CM |
682 | } |
683 | out: | |
cad321ad | 684 | spin_unlock(&tree->lock); |
d1310b2e | 685 | } |
d1310b2e | 686 | |
1bf85046 | 687 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 688 | struct extent_state *state, |
0ca1f7ce | 689 | int *bits) |
d1310b2e | 690 | { |
0ca1f7ce | 691 | int bits_to_set = *bits & ~EXTENT_CTLBITS; |
9ed74f2d | 692 | |
1bf85046 | 693 | set_state_cb(tree, state, bits); |
0ca1f7ce | 694 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
695 | u64 range = state->end - state->start + 1; |
696 | tree->dirty_bytes += range; | |
697 | } | |
0ca1f7ce | 698 | state->state |= bits_to_set; |
d1310b2e CM |
699 | } |
700 | ||
2c64c53d CM |
701 | static void cache_state(struct extent_state *state, |
702 | struct extent_state **cached_ptr) | |
703 | { | |
704 | if (cached_ptr && !(*cached_ptr)) { | |
705 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) { | |
706 | *cached_ptr = state; | |
707 | atomic_inc(&state->refs); | |
708 | } | |
709 | } | |
710 | } | |
711 | ||
507903b8 AJ |
712 | static void uncache_state(struct extent_state **cached_ptr) |
713 | { | |
714 | if (cached_ptr && (*cached_ptr)) { | |
715 | struct extent_state *state = *cached_ptr; | |
109b36a2 CM |
716 | *cached_ptr = NULL; |
717 | free_extent_state(state); | |
507903b8 AJ |
718 | } |
719 | } | |
720 | ||
d1310b2e | 721 | /* |
1edbb734 CM |
722 | * set some bits on a range in the tree. This may require allocations or |
723 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 724 | * |
1edbb734 CM |
725 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
726 | * part of the range already has the desired bits set. The start of the | |
727 | * existing range is returned in failed_start in this case. | |
d1310b2e | 728 | * |
1edbb734 | 729 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 730 | */ |
1edbb734 | 731 | |
3fbe5c02 JM |
732 | static int __must_check |
733 | __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
734 | int bits, int exclusive_bits, u64 *failed_start, | |
735 | struct extent_state **cached_state, gfp_t mask) | |
d1310b2e CM |
736 | { |
737 | struct extent_state *state; | |
738 | struct extent_state *prealloc = NULL; | |
739 | struct rb_node *node; | |
d1310b2e | 740 | int err = 0; |
d1310b2e CM |
741 | u64 last_start; |
742 | u64 last_end; | |
42daec29 | 743 | |
0ca1f7ce | 744 | bits |= EXTENT_FIRST_DELALLOC; |
d1310b2e CM |
745 | again: |
746 | if (!prealloc && (mask & __GFP_WAIT)) { | |
747 | prealloc = alloc_extent_state(mask); | |
8233767a | 748 | BUG_ON(!prealloc); |
d1310b2e CM |
749 | } |
750 | ||
cad321ad | 751 | spin_lock(&tree->lock); |
9655d298 CM |
752 | if (cached_state && *cached_state) { |
753 | state = *cached_state; | |
df98b6e2 JB |
754 | if (state->start <= start && state->end > start && |
755 | state->tree) { | |
9655d298 CM |
756 | node = &state->rb_node; |
757 | goto hit_next; | |
758 | } | |
759 | } | |
d1310b2e CM |
760 | /* |
761 | * this search will find all the extents that end after | |
762 | * our range starts. | |
763 | */ | |
80ea96b1 | 764 | node = tree_search(tree, start); |
d1310b2e | 765 | if (!node) { |
8233767a XG |
766 | prealloc = alloc_extent_state_atomic(prealloc); |
767 | BUG_ON(!prealloc); | |
0ca1f7ce | 768 | err = insert_state(tree, prealloc, start, end, &bits); |
c2d904e0 JM |
769 | if (err) |
770 | extent_io_tree_panic(tree, err); | |
771 | ||
d1310b2e | 772 | prealloc = NULL; |
d1310b2e CM |
773 | goto out; |
774 | } | |
d1310b2e | 775 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 776 | hit_next: |
d1310b2e CM |
777 | last_start = state->start; |
778 | last_end = state->end; | |
779 | ||
780 | /* | |
781 | * | ---- desired range ---- | | |
782 | * | state | | |
783 | * | |
784 | * Just lock what we found and keep going | |
785 | */ | |
786 | if (state->start == start && state->end <= end) { | |
1edbb734 | 787 | if (state->state & exclusive_bits) { |
d1310b2e CM |
788 | *failed_start = state->start; |
789 | err = -EEXIST; | |
790 | goto out; | |
791 | } | |
42daec29 | 792 | |
1bf85046 | 793 | set_state_bits(tree, state, &bits); |
2c64c53d | 794 | cache_state(state, cached_state); |
d1310b2e | 795 | merge_state(tree, state); |
5c939df5 YZ |
796 | if (last_end == (u64)-1) |
797 | goto out; | |
798 | start = last_end + 1; | |
d1ac6e41 LB |
799 | state = next_state(state); |
800 | if (start < end && state && state->start == start && | |
801 | !need_resched()) | |
802 | goto hit_next; | |
d1310b2e CM |
803 | goto search_again; |
804 | } | |
805 | ||
806 | /* | |
807 | * | ---- desired range ---- | | |
808 | * | state | | |
809 | * or | |
810 | * | ------------- state -------------- | | |
811 | * | |
812 | * We need to split the extent we found, and may flip bits on | |
813 | * second half. | |
814 | * | |
815 | * If the extent we found extends past our | |
816 | * range, we just split and search again. It'll get split | |
817 | * again the next time though. | |
818 | * | |
819 | * If the extent we found is inside our range, we set the | |
820 | * desired bit on it. | |
821 | */ | |
822 | if (state->start < start) { | |
1edbb734 | 823 | if (state->state & exclusive_bits) { |
d1310b2e CM |
824 | *failed_start = start; |
825 | err = -EEXIST; | |
826 | goto out; | |
827 | } | |
8233767a XG |
828 | |
829 | prealloc = alloc_extent_state_atomic(prealloc); | |
830 | BUG_ON(!prealloc); | |
d1310b2e | 831 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
832 | if (err) |
833 | extent_io_tree_panic(tree, err); | |
834 | ||
d1310b2e CM |
835 | prealloc = NULL; |
836 | if (err) | |
837 | goto out; | |
838 | if (state->end <= end) { | |
1bf85046 | 839 | set_state_bits(tree, state, &bits); |
2c64c53d | 840 | cache_state(state, cached_state); |
d1310b2e | 841 | merge_state(tree, state); |
5c939df5 YZ |
842 | if (last_end == (u64)-1) |
843 | goto out; | |
844 | start = last_end + 1; | |
d1ac6e41 LB |
845 | state = next_state(state); |
846 | if (start < end && state && state->start == start && | |
847 | !need_resched()) | |
848 | goto hit_next; | |
d1310b2e CM |
849 | } |
850 | goto search_again; | |
851 | } | |
852 | /* | |
853 | * | ---- desired range ---- | | |
854 | * | state | or | state | | |
855 | * | |
856 | * There's a hole, we need to insert something in it and | |
857 | * ignore the extent we found. | |
858 | */ | |
859 | if (state->start > start) { | |
860 | u64 this_end; | |
861 | if (end < last_start) | |
862 | this_end = end; | |
863 | else | |
d397712b | 864 | this_end = last_start - 1; |
8233767a XG |
865 | |
866 | prealloc = alloc_extent_state_atomic(prealloc); | |
867 | BUG_ON(!prealloc); | |
c7f895a2 XG |
868 | |
869 | /* | |
870 | * Avoid to free 'prealloc' if it can be merged with | |
871 | * the later extent. | |
872 | */ | |
d1310b2e | 873 | err = insert_state(tree, prealloc, start, this_end, |
0ca1f7ce | 874 | &bits); |
c2d904e0 JM |
875 | if (err) |
876 | extent_io_tree_panic(tree, err); | |
877 | ||
9ed74f2d JB |
878 | cache_state(prealloc, cached_state); |
879 | prealloc = NULL; | |
d1310b2e CM |
880 | start = this_end + 1; |
881 | goto search_again; | |
882 | } | |
883 | /* | |
884 | * | ---- desired range ---- | | |
885 | * | state | | |
886 | * We need to split the extent, and set the bit | |
887 | * on the first half | |
888 | */ | |
889 | if (state->start <= end && state->end > end) { | |
1edbb734 | 890 | if (state->state & exclusive_bits) { |
d1310b2e CM |
891 | *failed_start = start; |
892 | err = -EEXIST; | |
893 | goto out; | |
894 | } | |
8233767a XG |
895 | |
896 | prealloc = alloc_extent_state_atomic(prealloc); | |
897 | BUG_ON(!prealloc); | |
d1310b2e | 898 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
899 | if (err) |
900 | extent_io_tree_panic(tree, err); | |
d1310b2e | 901 | |
1bf85046 | 902 | set_state_bits(tree, prealloc, &bits); |
2c64c53d | 903 | cache_state(prealloc, cached_state); |
d1310b2e CM |
904 | merge_state(tree, prealloc); |
905 | prealloc = NULL; | |
906 | goto out; | |
907 | } | |
908 | ||
909 | goto search_again; | |
910 | ||
911 | out: | |
cad321ad | 912 | spin_unlock(&tree->lock); |
d1310b2e CM |
913 | if (prealloc) |
914 | free_extent_state(prealloc); | |
915 | ||
916 | return err; | |
917 | ||
918 | search_again: | |
919 | if (start > end) | |
920 | goto out; | |
cad321ad | 921 | spin_unlock(&tree->lock); |
d1310b2e CM |
922 | if (mask & __GFP_WAIT) |
923 | cond_resched(); | |
924 | goto again; | |
925 | } | |
d1310b2e | 926 | |
3fbe5c02 JM |
927 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits, |
928 | u64 *failed_start, struct extent_state **cached_state, | |
929 | gfp_t mask) | |
930 | { | |
931 | return __set_extent_bit(tree, start, end, bits, 0, failed_start, | |
932 | cached_state, mask); | |
933 | } | |
934 | ||
935 | ||
462d6fac | 936 | /** |
10983f2e LB |
937 | * convert_extent_bit - convert all bits in a given range from one bit to |
938 | * another | |
462d6fac JB |
939 | * @tree: the io tree to search |
940 | * @start: the start offset in bytes | |
941 | * @end: the end offset in bytes (inclusive) | |
942 | * @bits: the bits to set in this range | |
943 | * @clear_bits: the bits to clear in this range | |
e6138876 | 944 | * @cached_state: state that we're going to cache |
462d6fac JB |
945 | * @mask: the allocation mask |
946 | * | |
947 | * This will go through and set bits for the given range. If any states exist | |
948 | * already in this range they are set with the given bit and cleared of the | |
949 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
950 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
951 | * boundary bits like LOCK. | |
952 | */ | |
953 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
e6138876 JB |
954 | int bits, int clear_bits, |
955 | struct extent_state **cached_state, gfp_t mask) | |
462d6fac JB |
956 | { |
957 | struct extent_state *state; | |
958 | struct extent_state *prealloc = NULL; | |
959 | struct rb_node *node; | |
960 | int err = 0; | |
961 | u64 last_start; | |
962 | u64 last_end; | |
963 | ||
964 | again: | |
965 | if (!prealloc && (mask & __GFP_WAIT)) { | |
966 | prealloc = alloc_extent_state(mask); | |
967 | if (!prealloc) | |
968 | return -ENOMEM; | |
969 | } | |
970 | ||
971 | spin_lock(&tree->lock); | |
e6138876 JB |
972 | if (cached_state && *cached_state) { |
973 | state = *cached_state; | |
974 | if (state->start <= start && state->end > start && | |
975 | state->tree) { | |
976 | node = &state->rb_node; | |
977 | goto hit_next; | |
978 | } | |
979 | } | |
980 | ||
462d6fac JB |
981 | /* |
982 | * this search will find all the extents that end after | |
983 | * our range starts. | |
984 | */ | |
985 | node = tree_search(tree, start); | |
986 | if (!node) { | |
987 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
988 | if (!prealloc) { |
989 | err = -ENOMEM; | |
990 | goto out; | |
991 | } | |
462d6fac JB |
992 | err = insert_state(tree, prealloc, start, end, &bits); |
993 | prealloc = NULL; | |
c2d904e0 JM |
994 | if (err) |
995 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
996 | goto out; |
997 | } | |
998 | state = rb_entry(node, struct extent_state, rb_node); | |
999 | hit_next: | |
1000 | last_start = state->start; | |
1001 | last_end = state->end; | |
1002 | ||
1003 | /* | |
1004 | * | ---- desired range ---- | | |
1005 | * | state | | |
1006 | * | |
1007 | * Just lock what we found and keep going | |
1008 | */ | |
1009 | if (state->start == start && state->end <= end) { | |
462d6fac | 1010 | set_state_bits(tree, state, &bits); |
e6138876 | 1011 | cache_state(state, cached_state); |
d1ac6e41 | 1012 | state = clear_state_bit(tree, state, &clear_bits, 0); |
462d6fac JB |
1013 | if (last_end == (u64)-1) |
1014 | goto out; | |
462d6fac | 1015 | start = last_end + 1; |
d1ac6e41 LB |
1016 | if (start < end && state && state->start == start && |
1017 | !need_resched()) | |
1018 | goto hit_next; | |
462d6fac JB |
1019 | goto search_again; |
1020 | } | |
1021 | ||
1022 | /* | |
1023 | * | ---- desired range ---- | | |
1024 | * | state | | |
1025 | * or | |
1026 | * | ------------- state -------------- | | |
1027 | * | |
1028 | * We need to split the extent we found, and may flip bits on | |
1029 | * second half. | |
1030 | * | |
1031 | * If the extent we found extends past our | |
1032 | * range, we just split and search again. It'll get split | |
1033 | * again the next time though. | |
1034 | * | |
1035 | * If the extent we found is inside our range, we set the | |
1036 | * desired bit on it. | |
1037 | */ | |
1038 | if (state->start < start) { | |
1039 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1040 | if (!prealloc) { |
1041 | err = -ENOMEM; | |
1042 | goto out; | |
1043 | } | |
462d6fac | 1044 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1045 | if (err) |
1046 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1047 | prealloc = NULL; |
1048 | if (err) | |
1049 | goto out; | |
1050 | if (state->end <= end) { | |
1051 | set_state_bits(tree, state, &bits); | |
e6138876 | 1052 | cache_state(state, cached_state); |
d1ac6e41 | 1053 | state = clear_state_bit(tree, state, &clear_bits, 0); |
462d6fac JB |
1054 | if (last_end == (u64)-1) |
1055 | goto out; | |
1056 | start = last_end + 1; | |
d1ac6e41 LB |
1057 | if (start < end && state && state->start == start && |
1058 | !need_resched()) | |
1059 | goto hit_next; | |
462d6fac JB |
1060 | } |
1061 | goto search_again; | |
1062 | } | |
1063 | /* | |
1064 | * | ---- desired range ---- | | |
1065 | * | state | or | state | | |
1066 | * | |
1067 | * There's a hole, we need to insert something in it and | |
1068 | * ignore the extent we found. | |
1069 | */ | |
1070 | if (state->start > start) { | |
1071 | u64 this_end; | |
1072 | if (end < last_start) | |
1073 | this_end = end; | |
1074 | else | |
1075 | this_end = last_start - 1; | |
1076 | ||
1077 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1078 | if (!prealloc) { |
1079 | err = -ENOMEM; | |
1080 | goto out; | |
1081 | } | |
462d6fac JB |
1082 | |
1083 | /* | |
1084 | * Avoid to free 'prealloc' if it can be merged with | |
1085 | * the later extent. | |
1086 | */ | |
1087 | err = insert_state(tree, prealloc, start, this_end, | |
1088 | &bits); | |
c2d904e0 JM |
1089 | if (err) |
1090 | extent_io_tree_panic(tree, err); | |
e6138876 | 1091 | cache_state(prealloc, cached_state); |
462d6fac JB |
1092 | prealloc = NULL; |
1093 | start = this_end + 1; | |
1094 | goto search_again; | |
1095 | } | |
1096 | /* | |
1097 | * | ---- desired range ---- | | |
1098 | * | state | | |
1099 | * We need to split the extent, and set the bit | |
1100 | * on the first half | |
1101 | */ | |
1102 | if (state->start <= end && state->end > end) { | |
1103 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1104 | if (!prealloc) { |
1105 | err = -ENOMEM; | |
1106 | goto out; | |
1107 | } | |
462d6fac JB |
1108 | |
1109 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1110 | if (err) |
1111 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1112 | |
1113 | set_state_bits(tree, prealloc, &bits); | |
e6138876 | 1114 | cache_state(prealloc, cached_state); |
462d6fac | 1115 | clear_state_bit(tree, prealloc, &clear_bits, 0); |
462d6fac JB |
1116 | prealloc = NULL; |
1117 | goto out; | |
1118 | } | |
1119 | ||
1120 | goto search_again; | |
1121 | ||
1122 | out: | |
1123 | spin_unlock(&tree->lock); | |
1124 | if (prealloc) | |
1125 | free_extent_state(prealloc); | |
1126 | ||
1127 | return err; | |
1128 | ||
1129 | search_again: | |
1130 | if (start > end) | |
1131 | goto out; | |
1132 | spin_unlock(&tree->lock); | |
1133 | if (mask & __GFP_WAIT) | |
1134 | cond_resched(); | |
1135 | goto again; | |
1136 | } | |
1137 | ||
d1310b2e CM |
1138 | /* wrappers around set/clear extent bit */ |
1139 | int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
1140 | gfp_t mask) | |
1141 | { | |
3fbe5c02 | 1142 | return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL, |
2c64c53d | 1143 | NULL, mask); |
d1310b2e | 1144 | } |
d1310b2e CM |
1145 | |
1146 | int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
1147 | int bits, gfp_t mask) | |
1148 | { | |
3fbe5c02 | 1149 | return set_extent_bit(tree, start, end, bits, NULL, |
2c64c53d | 1150 | NULL, mask); |
d1310b2e | 1151 | } |
d1310b2e CM |
1152 | |
1153 | int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
1154 | int bits, gfp_t mask) | |
1155 | { | |
2c64c53d | 1156 | return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask); |
d1310b2e | 1157 | } |
d1310b2e CM |
1158 | |
1159 | int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, | |
2ac55d41 | 1160 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
1161 | { |
1162 | return set_extent_bit(tree, start, end, | |
fee187d9 | 1163 | EXTENT_DELALLOC | EXTENT_UPTODATE, |
3fbe5c02 | 1164 | NULL, cached_state, mask); |
d1310b2e | 1165 | } |
d1310b2e | 1166 | |
9e8a4a8b LB |
1167 | int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end, |
1168 | struct extent_state **cached_state, gfp_t mask) | |
1169 | { | |
1170 | return set_extent_bit(tree, start, end, | |
1171 | EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG, | |
1172 | NULL, cached_state, mask); | |
1173 | } | |
1174 | ||
d1310b2e CM |
1175 | int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, |
1176 | gfp_t mask) | |
1177 | { | |
1178 | return clear_extent_bit(tree, start, end, | |
32c00aff | 1179 | EXTENT_DIRTY | EXTENT_DELALLOC | |
0ca1f7ce | 1180 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask); |
d1310b2e | 1181 | } |
d1310b2e CM |
1182 | |
1183 | int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, | |
1184 | gfp_t mask) | |
1185 | { | |
3fbe5c02 | 1186 | return set_extent_bit(tree, start, end, EXTENT_NEW, NULL, |
2c64c53d | 1187 | NULL, mask); |
d1310b2e | 1188 | } |
d1310b2e | 1189 | |
d1310b2e | 1190 | int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
507903b8 | 1191 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e | 1192 | { |
507903b8 | 1193 | return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, |
3fbe5c02 | 1194 | cached_state, mask); |
d1310b2e | 1195 | } |
d1310b2e | 1196 | |
5fd02043 JB |
1197 | int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
1198 | struct extent_state **cached_state, gfp_t mask) | |
d1310b2e | 1199 | { |
2c64c53d | 1200 | return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, |
2ac55d41 | 1201 | cached_state, mask); |
d1310b2e | 1202 | } |
d1310b2e | 1203 | |
d352ac68 CM |
1204 | /* |
1205 | * either insert or lock state struct between start and end use mask to tell | |
1206 | * us if waiting is desired. | |
1207 | */ | |
1edbb734 | 1208 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
d0082371 | 1209 | int bits, struct extent_state **cached_state) |
d1310b2e CM |
1210 | { |
1211 | int err; | |
1212 | u64 failed_start; | |
1213 | while (1) { | |
3fbe5c02 JM |
1214 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits, |
1215 | EXTENT_LOCKED, &failed_start, | |
1216 | cached_state, GFP_NOFS); | |
d0082371 | 1217 | if (err == -EEXIST) { |
d1310b2e CM |
1218 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1219 | start = failed_start; | |
d0082371 | 1220 | } else |
d1310b2e | 1221 | break; |
d1310b2e CM |
1222 | WARN_ON(start > end); |
1223 | } | |
1224 | return err; | |
1225 | } | |
d1310b2e | 1226 | |
d0082371 | 1227 | int lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
1edbb734 | 1228 | { |
d0082371 | 1229 | return lock_extent_bits(tree, start, end, 0, NULL); |
1edbb734 CM |
1230 | } |
1231 | ||
d0082371 | 1232 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1233 | { |
1234 | int err; | |
1235 | u64 failed_start; | |
1236 | ||
3fbe5c02 JM |
1237 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
1238 | &failed_start, NULL, GFP_NOFS); | |
6643558d YZ |
1239 | if (err == -EEXIST) { |
1240 | if (failed_start > start) | |
1241 | clear_extent_bit(tree, start, failed_start - 1, | |
d0082371 | 1242 | EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS); |
25179201 | 1243 | return 0; |
6643558d | 1244 | } |
25179201 JB |
1245 | return 1; |
1246 | } | |
25179201 | 1247 | |
2c64c53d CM |
1248 | int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end, |
1249 | struct extent_state **cached, gfp_t mask) | |
1250 | { | |
1251 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, | |
1252 | mask); | |
1253 | } | |
1254 | ||
d0082371 | 1255 | int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e | 1256 | { |
2c64c53d | 1257 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL, |
d0082371 | 1258 | GFP_NOFS); |
d1310b2e | 1259 | } |
d1310b2e | 1260 | |
d1310b2e CM |
1261 | /* |
1262 | * helper function to set both pages and extents in the tree writeback | |
1263 | */ | |
b2950863 | 1264 | static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e CM |
1265 | { |
1266 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1267 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1268 | struct page *page; | |
1269 | ||
1270 | while (index <= end_index) { | |
1271 | page = find_get_page(tree->mapping, index); | |
79787eaa | 1272 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ |
d1310b2e CM |
1273 | set_page_writeback(page); |
1274 | page_cache_release(page); | |
1275 | index++; | |
1276 | } | |
d1310b2e CM |
1277 | return 0; |
1278 | } | |
d1310b2e | 1279 | |
d352ac68 CM |
1280 | /* find the first state struct with 'bits' set after 'start', and |
1281 | * return it. tree->lock must be held. NULL will returned if | |
1282 | * nothing was found after 'start' | |
1283 | */ | |
d7fc640e CM |
1284 | struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree, |
1285 | u64 start, int bits) | |
1286 | { | |
1287 | struct rb_node *node; | |
1288 | struct extent_state *state; | |
1289 | ||
1290 | /* | |
1291 | * this search will find all the extents that end after | |
1292 | * our range starts. | |
1293 | */ | |
1294 | node = tree_search(tree, start); | |
d397712b | 1295 | if (!node) |
d7fc640e | 1296 | goto out; |
d7fc640e | 1297 | |
d397712b | 1298 | while (1) { |
d7fc640e | 1299 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1300 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1301 | return state; |
d397712b | 1302 | |
d7fc640e CM |
1303 | node = rb_next(node); |
1304 | if (!node) | |
1305 | break; | |
1306 | } | |
1307 | out: | |
1308 | return NULL; | |
1309 | } | |
d7fc640e | 1310 | |
69261c4b XG |
1311 | /* |
1312 | * find the first offset in the io tree with 'bits' set. zero is | |
1313 | * returned if we find something, and *start_ret and *end_ret are | |
1314 | * set to reflect the state struct that was found. | |
1315 | * | |
477d7eaf | 1316 | * If nothing was found, 1 is returned. If found something, return 0. |
69261c4b XG |
1317 | */ |
1318 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
e6138876 JB |
1319 | u64 *start_ret, u64 *end_ret, int bits, |
1320 | struct extent_state **cached_state) | |
69261c4b XG |
1321 | { |
1322 | struct extent_state *state; | |
e6138876 | 1323 | struct rb_node *n; |
69261c4b XG |
1324 | int ret = 1; |
1325 | ||
1326 | spin_lock(&tree->lock); | |
e6138876 JB |
1327 | if (cached_state && *cached_state) { |
1328 | state = *cached_state; | |
1329 | if (state->end == start - 1 && state->tree) { | |
1330 | n = rb_next(&state->rb_node); | |
1331 | while (n) { | |
1332 | state = rb_entry(n, struct extent_state, | |
1333 | rb_node); | |
1334 | if (state->state & bits) | |
1335 | goto got_it; | |
1336 | n = rb_next(n); | |
1337 | } | |
1338 | free_extent_state(*cached_state); | |
1339 | *cached_state = NULL; | |
1340 | goto out; | |
1341 | } | |
1342 | free_extent_state(*cached_state); | |
1343 | *cached_state = NULL; | |
1344 | } | |
1345 | ||
69261c4b | 1346 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1347 | got_it: |
69261c4b | 1348 | if (state) { |
e6138876 | 1349 | cache_state(state, cached_state); |
69261c4b XG |
1350 | *start_ret = state->start; |
1351 | *end_ret = state->end; | |
1352 | ret = 0; | |
1353 | } | |
e6138876 | 1354 | out: |
69261c4b XG |
1355 | spin_unlock(&tree->lock); |
1356 | return ret; | |
1357 | } | |
1358 | ||
d352ac68 CM |
1359 | /* |
1360 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1361 | * more than 'max_bytes'. start and end are used to return the range, | |
1362 | * | |
1363 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1364 | */ | |
c8b97818 | 1365 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1366 | u64 *start, u64 *end, u64 max_bytes, |
1367 | struct extent_state **cached_state) | |
d1310b2e CM |
1368 | { |
1369 | struct rb_node *node; | |
1370 | struct extent_state *state; | |
1371 | u64 cur_start = *start; | |
1372 | u64 found = 0; | |
1373 | u64 total_bytes = 0; | |
1374 | ||
cad321ad | 1375 | spin_lock(&tree->lock); |
c8b97818 | 1376 | |
d1310b2e CM |
1377 | /* |
1378 | * this search will find all the extents that end after | |
1379 | * our range starts. | |
1380 | */ | |
80ea96b1 | 1381 | node = tree_search(tree, cur_start); |
2b114d1d | 1382 | if (!node) { |
3b951516 CM |
1383 | if (!found) |
1384 | *end = (u64)-1; | |
d1310b2e CM |
1385 | goto out; |
1386 | } | |
1387 | ||
d397712b | 1388 | while (1) { |
d1310b2e | 1389 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1390 | if (found && (state->start != cur_start || |
1391 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1392 | goto out; |
1393 | } | |
1394 | if (!(state->state & EXTENT_DELALLOC)) { | |
1395 | if (!found) | |
1396 | *end = state->end; | |
1397 | goto out; | |
1398 | } | |
c2a128d2 | 1399 | if (!found) { |
d1310b2e | 1400 | *start = state->start; |
c2a128d2 JB |
1401 | *cached_state = state; |
1402 | atomic_inc(&state->refs); | |
1403 | } | |
d1310b2e CM |
1404 | found++; |
1405 | *end = state->end; | |
1406 | cur_start = state->end + 1; | |
1407 | node = rb_next(node); | |
1408 | if (!node) | |
1409 | break; | |
1410 | total_bytes += state->end - state->start + 1; | |
1411 | if (total_bytes >= max_bytes) | |
1412 | break; | |
1413 | } | |
1414 | out: | |
cad321ad | 1415 | spin_unlock(&tree->lock); |
d1310b2e CM |
1416 | return found; |
1417 | } | |
1418 | ||
143bede5 JM |
1419 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1420 | struct page *locked_page, | |
1421 | u64 start, u64 end) | |
c8b97818 CM |
1422 | { |
1423 | int ret; | |
1424 | struct page *pages[16]; | |
1425 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1426 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1427 | unsigned long nr_pages = end_index - index + 1; | |
1428 | int i; | |
1429 | ||
1430 | if (index == locked_page->index && end_index == index) | |
143bede5 | 1431 | return; |
c8b97818 | 1432 | |
d397712b | 1433 | while (nr_pages > 0) { |
c8b97818 | 1434 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1435 | min_t(unsigned long, nr_pages, |
1436 | ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1437 | for (i = 0; i < ret; i++) { |
1438 | if (pages[i] != locked_page) | |
1439 | unlock_page(pages[i]); | |
1440 | page_cache_release(pages[i]); | |
1441 | } | |
1442 | nr_pages -= ret; | |
1443 | index += ret; | |
1444 | cond_resched(); | |
1445 | } | |
c8b97818 CM |
1446 | } |
1447 | ||
1448 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1449 | struct page *locked_page, | |
1450 | u64 delalloc_start, | |
1451 | u64 delalloc_end) | |
1452 | { | |
1453 | unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT; | |
1454 | unsigned long start_index = index; | |
1455 | unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT; | |
1456 | unsigned long pages_locked = 0; | |
1457 | struct page *pages[16]; | |
1458 | unsigned long nrpages; | |
1459 | int ret; | |
1460 | int i; | |
1461 | ||
1462 | /* the caller is responsible for locking the start index */ | |
1463 | if (index == locked_page->index && index == end_index) | |
1464 | return 0; | |
1465 | ||
1466 | /* skip the page at the start index */ | |
1467 | nrpages = end_index - index + 1; | |
d397712b | 1468 | while (nrpages > 0) { |
c8b97818 | 1469 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1470 | min_t(unsigned long, |
1471 | nrpages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1472 | if (ret == 0) { |
1473 | ret = -EAGAIN; | |
1474 | goto done; | |
1475 | } | |
1476 | /* now we have an array of pages, lock them all */ | |
1477 | for (i = 0; i < ret; i++) { | |
1478 | /* | |
1479 | * the caller is taking responsibility for | |
1480 | * locked_page | |
1481 | */ | |
771ed689 | 1482 | if (pages[i] != locked_page) { |
c8b97818 | 1483 | lock_page(pages[i]); |
f2b1c41c CM |
1484 | if (!PageDirty(pages[i]) || |
1485 | pages[i]->mapping != inode->i_mapping) { | |
771ed689 CM |
1486 | ret = -EAGAIN; |
1487 | unlock_page(pages[i]); | |
1488 | page_cache_release(pages[i]); | |
1489 | goto done; | |
1490 | } | |
1491 | } | |
c8b97818 | 1492 | page_cache_release(pages[i]); |
771ed689 | 1493 | pages_locked++; |
c8b97818 | 1494 | } |
c8b97818 CM |
1495 | nrpages -= ret; |
1496 | index += ret; | |
1497 | cond_resched(); | |
1498 | } | |
1499 | ret = 0; | |
1500 | done: | |
1501 | if (ret && pages_locked) { | |
1502 | __unlock_for_delalloc(inode, locked_page, | |
1503 | delalloc_start, | |
1504 | ((u64)(start_index + pages_locked - 1)) << | |
1505 | PAGE_CACHE_SHIFT); | |
1506 | } | |
1507 | return ret; | |
1508 | } | |
1509 | ||
1510 | /* | |
1511 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1512 | * more than 'max_bytes'. start and end are used to return the range, | |
1513 | * | |
1514 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1515 | */ | |
1516 | static noinline u64 find_lock_delalloc_range(struct inode *inode, | |
1517 | struct extent_io_tree *tree, | |
1518 | struct page *locked_page, | |
1519 | u64 *start, u64 *end, | |
1520 | u64 max_bytes) | |
1521 | { | |
1522 | u64 delalloc_start; | |
1523 | u64 delalloc_end; | |
1524 | u64 found; | |
9655d298 | 1525 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1526 | int ret; |
1527 | int loops = 0; | |
1528 | ||
1529 | again: | |
1530 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1531 | delalloc_start = *start; | |
1532 | delalloc_end = 0; | |
1533 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1534 | max_bytes, &cached_state); |
70b99e69 | 1535 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1536 | *start = delalloc_start; |
1537 | *end = delalloc_end; | |
c2a128d2 | 1538 | free_extent_state(cached_state); |
c8b97818 CM |
1539 | return found; |
1540 | } | |
1541 | ||
70b99e69 CM |
1542 | /* |
1543 | * start comes from the offset of locked_page. We have to lock | |
1544 | * pages in order, so we can't process delalloc bytes before | |
1545 | * locked_page | |
1546 | */ | |
d397712b | 1547 | if (delalloc_start < *start) |
70b99e69 | 1548 | delalloc_start = *start; |
70b99e69 | 1549 | |
c8b97818 CM |
1550 | /* |
1551 | * make sure to limit the number of pages we try to lock down | |
1552 | * if we're looping. | |
1553 | */ | |
d397712b | 1554 | if (delalloc_end + 1 - delalloc_start > max_bytes && loops) |
771ed689 | 1555 | delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1; |
d397712b | 1556 | |
c8b97818 CM |
1557 | /* step two, lock all the pages after the page that has start */ |
1558 | ret = lock_delalloc_pages(inode, locked_page, | |
1559 | delalloc_start, delalloc_end); | |
1560 | if (ret == -EAGAIN) { | |
1561 | /* some of the pages are gone, lets avoid looping by | |
1562 | * shortening the size of the delalloc range we're searching | |
1563 | */ | |
9655d298 | 1564 | free_extent_state(cached_state); |
c8b97818 CM |
1565 | if (!loops) { |
1566 | unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1); | |
1567 | max_bytes = PAGE_CACHE_SIZE - offset; | |
1568 | loops = 1; | |
1569 | goto again; | |
1570 | } else { | |
1571 | found = 0; | |
1572 | goto out_failed; | |
1573 | } | |
1574 | } | |
79787eaa | 1575 | BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */ |
c8b97818 CM |
1576 | |
1577 | /* step three, lock the state bits for the whole range */ | |
d0082371 | 1578 | lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state); |
c8b97818 CM |
1579 | |
1580 | /* then test to make sure it is all still delalloc */ | |
1581 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1582 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1583 | if (!ret) { |
9655d298 CM |
1584 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
1585 | &cached_state, GFP_NOFS); | |
c8b97818 CM |
1586 | __unlock_for_delalloc(inode, locked_page, |
1587 | delalloc_start, delalloc_end); | |
1588 | cond_resched(); | |
1589 | goto again; | |
1590 | } | |
9655d298 | 1591 | free_extent_state(cached_state); |
c8b97818 CM |
1592 | *start = delalloc_start; |
1593 | *end = delalloc_end; | |
1594 | out_failed: | |
1595 | return found; | |
1596 | } | |
1597 | ||
1598 | int extent_clear_unlock_delalloc(struct inode *inode, | |
1599 | struct extent_io_tree *tree, | |
1600 | u64 start, u64 end, struct page *locked_page, | |
a791e35e | 1601 | unsigned long op) |
c8b97818 CM |
1602 | { |
1603 | int ret; | |
1604 | struct page *pages[16]; | |
1605 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1606 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1607 | unsigned long nr_pages = end_index - index + 1; | |
1608 | int i; | |
771ed689 | 1609 | int clear_bits = 0; |
c8b97818 | 1610 | |
a791e35e | 1611 | if (op & EXTENT_CLEAR_UNLOCK) |
771ed689 | 1612 | clear_bits |= EXTENT_LOCKED; |
a791e35e | 1613 | if (op & EXTENT_CLEAR_DIRTY) |
c8b97818 CM |
1614 | clear_bits |= EXTENT_DIRTY; |
1615 | ||
a791e35e | 1616 | if (op & EXTENT_CLEAR_DELALLOC) |
771ed689 CM |
1617 | clear_bits |= EXTENT_DELALLOC; |
1618 | ||
2c64c53d | 1619 | clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); |
32c00aff JB |
1620 | if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | |
1621 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK | | |
1622 | EXTENT_SET_PRIVATE2))) | |
771ed689 | 1623 | return 0; |
c8b97818 | 1624 | |
d397712b | 1625 | while (nr_pages > 0) { |
c8b97818 | 1626 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1627 | min_t(unsigned long, |
1628 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 | 1629 | for (i = 0; i < ret; i++) { |
8b62b72b | 1630 | |
a791e35e | 1631 | if (op & EXTENT_SET_PRIVATE2) |
8b62b72b CM |
1632 | SetPagePrivate2(pages[i]); |
1633 | ||
c8b97818 CM |
1634 | if (pages[i] == locked_page) { |
1635 | page_cache_release(pages[i]); | |
1636 | continue; | |
1637 | } | |
a791e35e | 1638 | if (op & EXTENT_CLEAR_DIRTY) |
c8b97818 | 1639 | clear_page_dirty_for_io(pages[i]); |
a791e35e | 1640 | if (op & EXTENT_SET_WRITEBACK) |
c8b97818 | 1641 | set_page_writeback(pages[i]); |
a791e35e | 1642 | if (op & EXTENT_END_WRITEBACK) |
c8b97818 | 1643 | end_page_writeback(pages[i]); |
a791e35e | 1644 | if (op & EXTENT_CLEAR_UNLOCK_PAGE) |
771ed689 | 1645 | unlock_page(pages[i]); |
c8b97818 CM |
1646 | page_cache_release(pages[i]); |
1647 | } | |
1648 | nr_pages -= ret; | |
1649 | index += ret; | |
1650 | cond_resched(); | |
1651 | } | |
1652 | return 0; | |
1653 | } | |
c8b97818 | 1654 | |
d352ac68 CM |
1655 | /* |
1656 | * count the number of bytes in the tree that have a given bit(s) | |
1657 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1658 | * cached. The total number found is returned. | |
1659 | */ | |
d1310b2e CM |
1660 | u64 count_range_bits(struct extent_io_tree *tree, |
1661 | u64 *start, u64 search_end, u64 max_bytes, | |
ec29ed5b | 1662 | unsigned long bits, int contig) |
d1310b2e CM |
1663 | { |
1664 | struct rb_node *node; | |
1665 | struct extent_state *state; | |
1666 | u64 cur_start = *start; | |
1667 | u64 total_bytes = 0; | |
ec29ed5b | 1668 | u64 last = 0; |
d1310b2e CM |
1669 | int found = 0; |
1670 | ||
1671 | if (search_end <= cur_start) { | |
d1310b2e CM |
1672 | WARN_ON(1); |
1673 | return 0; | |
1674 | } | |
1675 | ||
cad321ad | 1676 | spin_lock(&tree->lock); |
d1310b2e CM |
1677 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1678 | total_bytes = tree->dirty_bytes; | |
1679 | goto out; | |
1680 | } | |
1681 | /* | |
1682 | * this search will find all the extents that end after | |
1683 | * our range starts. | |
1684 | */ | |
80ea96b1 | 1685 | node = tree_search(tree, cur_start); |
d397712b | 1686 | if (!node) |
d1310b2e | 1687 | goto out; |
d1310b2e | 1688 | |
d397712b | 1689 | while (1) { |
d1310b2e CM |
1690 | state = rb_entry(node, struct extent_state, rb_node); |
1691 | if (state->start > search_end) | |
1692 | break; | |
ec29ed5b CM |
1693 | if (contig && found && state->start > last + 1) |
1694 | break; | |
1695 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1696 | total_bytes += min(search_end, state->end) + 1 - |
1697 | max(cur_start, state->start); | |
1698 | if (total_bytes >= max_bytes) | |
1699 | break; | |
1700 | if (!found) { | |
af60bed2 | 1701 | *start = max(cur_start, state->start); |
d1310b2e CM |
1702 | found = 1; |
1703 | } | |
ec29ed5b CM |
1704 | last = state->end; |
1705 | } else if (contig && found) { | |
1706 | break; | |
d1310b2e CM |
1707 | } |
1708 | node = rb_next(node); | |
1709 | if (!node) | |
1710 | break; | |
1711 | } | |
1712 | out: | |
cad321ad | 1713 | spin_unlock(&tree->lock); |
d1310b2e CM |
1714 | return total_bytes; |
1715 | } | |
b2950863 | 1716 | |
d352ac68 CM |
1717 | /* |
1718 | * set the private field for a given byte offset in the tree. If there isn't | |
1719 | * an extent_state there already, this does nothing. | |
1720 | */ | |
d1310b2e CM |
1721 | int set_state_private(struct extent_io_tree *tree, u64 start, u64 private) |
1722 | { | |
1723 | struct rb_node *node; | |
1724 | struct extent_state *state; | |
1725 | int ret = 0; | |
1726 | ||
cad321ad | 1727 | spin_lock(&tree->lock); |
d1310b2e CM |
1728 | /* |
1729 | * this search will find all the extents that end after | |
1730 | * our range starts. | |
1731 | */ | |
80ea96b1 | 1732 | node = tree_search(tree, start); |
2b114d1d | 1733 | if (!node) { |
d1310b2e CM |
1734 | ret = -ENOENT; |
1735 | goto out; | |
1736 | } | |
1737 | state = rb_entry(node, struct extent_state, rb_node); | |
1738 | if (state->start != start) { | |
1739 | ret = -ENOENT; | |
1740 | goto out; | |
1741 | } | |
1742 | state->private = private; | |
1743 | out: | |
cad321ad | 1744 | spin_unlock(&tree->lock); |
d1310b2e CM |
1745 | return ret; |
1746 | } | |
1747 | ||
1748 | int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private) | |
1749 | { | |
1750 | struct rb_node *node; | |
1751 | struct extent_state *state; | |
1752 | int ret = 0; | |
1753 | ||
cad321ad | 1754 | spin_lock(&tree->lock); |
d1310b2e CM |
1755 | /* |
1756 | * this search will find all the extents that end after | |
1757 | * our range starts. | |
1758 | */ | |
80ea96b1 | 1759 | node = tree_search(tree, start); |
2b114d1d | 1760 | if (!node) { |
d1310b2e CM |
1761 | ret = -ENOENT; |
1762 | goto out; | |
1763 | } | |
1764 | state = rb_entry(node, struct extent_state, rb_node); | |
1765 | if (state->start != start) { | |
1766 | ret = -ENOENT; | |
1767 | goto out; | |
1768 | } | |
1769 | *private = state->private; | |
1770 | out: | |
cad321ad | 1771 | spin_unlock(&tree->lock); |
d1310b2e CM |
1772 | return ret; |
1773 | } | |
1774 | ||
1775 | /* | |
1776 | * searches a range in the state tree for a given mask. | |
70dec807 | 1777 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1778 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1779 | * range is found set. | |
1780 | */ | |
1781 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9655d298 | 1782 | int bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1783 | { |
1784 | struct extent_state *state = NULL; | |
1785 | struct rb_node *node; | |
1786 | int bitset = 0; | |
d1310b2e | 1787 | |
cad321ad | 1788 | spin_lock(&tree->lock); |
df98b6e2 JB |
1789 | if (cached && cached->tree && cached->start <= start && |
1790 | cached->end > start) | |
9655d298 CM |
1791 | node = &cached->rb_node; |
1792 | else | |
1793 | node = tree_search(tree, start); | |
d1310b2e CM |
1794 | while (node && start <= end) { |
1795 | state = rb_entry(node, struct extent_state, rb_node); | |
1796 | ||
1797 | if (filled && state->start > start) { | |
1798 | bitset = 0; | |
1799 | break; | |
1800 | } | |
1801 | ||
1802 | if (state->start > end) | |
1803 | break; | |
1804 | ||
1805 | if (state->state & bits) { | |
1806 | bitset = 1; | |
1807 | if (!filled) | |
1808 | break; | |
1809 | } else if (filled) { | |
1810 | bitset = 0; | |
1811 | break; | |
1812 | } | |
46562cec CM |
1813 | |
1814 | if (state->end == (u64)-1) | |
1815 | break; | |
1816 | ||
d1310b2e CM |
1817 | start = state->end + 1; |
1818 | if (start > end) | |
1819 | break; | |
1820 | node = rb_next(node); | |
1821 | if (!node) { | |
1822 | if (filled) | |
1823 | bitset = 0; | |
1824 | break; | |
1825 | } | |
1826 | } | |
cad321ad | 1827 | spin_unlock(&tree->lock); |
d1310b2e CM |
1828 | return bitset; |
1829 | } | |
d1310b2e CM |
1830 | |
1831 | /* | |
1832 | * helper function to set a given page up to date if all the | |
1833 | * extents in the tree for that page are up to date | |
1834 | */ | |
143bede5 | 1835 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e CM |
1836 | { |
1837 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1838 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
9655d298 | 1839 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 1840 | SetPageUptodate(page); |
d1310b2e CM |
1841 | } |
1842 | ||
1843 | /* | |
1844 | * helper function to unlock a page if all the extents in the tree | |
1845 | * for that page are unlocked | |
1846 | */ | |
143bede5 | 1847 | static void check_page_locked(struct extent_io_tree *tree, struct page *page) |
d1310b2e CM |
1848 | { |
1849 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1850 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
9655d298 | 1851 | if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) |
d1310b2e | 1852 | unlock_page(page); |
d1310b2e CM |
1853 | } |
1854 | ||
1855 | /* | |
1856 | * helper function to end page writeback if all the extents | |
1857 | * in the tree for that page are done with writeback | |
1858 | */ | |
143bede5 JM |
1859 | static void check_page_writeback(struct extent_io_tree *tree, |
1860 | struct page *page) | |
d1310b2e | 1861 | { |
1edbb734 | 1862 | end_page_writeback(page); |
d1310b2e CM |
1863 | } |
1864 | ||
4a54c8c1 JS |
1865 | /* |
1866 | * When IO fails, either with EIO or csum verification fails, we | |
1867 | * try other mirrors that might have a good copy of the data. This | |
1868 | * io_failure_record is used to record state as we go through all the | |
1869 | * mirrors. If another mirror has good data, the page is set up to date | |
1870 | * and things continue. If a good mirror can't be found, the original | |
1871 | * bio end_io callback is called to indicate things have failed. | |
1872 | */ | |
1873 | struct io_failure_record { | |
1874 | struct page *page; | |
1875 | u64 start; | |
1876 | u64 len; | |
1877 | u64 logical; | |
1878 | unsigned long bio_flags; | |
1879 | int this_mirror; | |
1880 | int failed_mirror; | |
1881 | int in_validation; | |
1882 | }; | |
1883 | ||
1884 | static int free_io_failure(struct inode *inode, struct io_failure_record *rec, | |
1885 | int did_repair) | |
1886 | { | |
1887 | int ret; | |
1888 | int err = 0; | |
1889 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
1890 | ||
1891 | set_state_private(failure_tree, rec->start, 0); | |
1892 | ret = clear_extent_bits(failure_tree, rec->start, | |
1893 | rec->start + rec->len - 1, | |
1894 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1895 | if (ret) | |
1896 | err = ret; | |
1897 | ||
1898 | if (did_repair) { | |
1899 | ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start, | |
1900 | rec->start + rec->len - 1, | |
1901 | EXTENT_DAMAGED, GFP_NOFS); | |
1902 | if (ret && !err) | |
1903 | err = ret; | |
1904 | } | |
1905 | ||
1906 | kfree(rec); | |
1907 | return err; | |
1908 | } | |
1909 | ||
1910 | static void repair_io_failure_callback(struct bio *bio, int err) | |
1911 | { | |
1912 | complete(bio->bi_private); | |
1913 | } | |
1914 | ||
1915 | /* | |
1916 | * this bypasses the standard btrfs submit functions deliberately, as | |
1917 | * the standard behavior is to write all copies in a raid setup. here we only | |
1918 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
1919 | * submit_bio directly. | |
1920 | * to avoid any synchonization issues, wait for the data after writing, which | |
1921 | * actually prevents the read that triggered the error from finishing. | |
1922 | * currently, there can be no more than two copies of every data bit. thus, | |
1923 | * exactly one rewrite is required. | |
1924 | */ | |
1925 | int repair_io_failure(struct btrfs_mapping_tree *map_tree, u64 start, | |
1926 | u64 length, u64 logical, struct page *page, | |
1927 | int mirror_num) | |
1928 | { | |
1929 | struct bio *bio; | |
1930 | struct btrfs_device *dev; | |
1931 | DECLARE_COMPLETION_ONSTACK(compl); | |
1932 | u64 map_length = 0; | |
1933 | u64 sector; | |
1934 | struct btrfs_bio *bbio = NULL; | |
1935 | int ret; | |
1936 | ||
1937 | BUG_ON(!mirror_num); | |
1938 | ||
1939 | bio = bio_alloc(GFP_NOFS, 1); | |
1940 | if (!bio) | |
1941 | return -EIO; | |
1942 | bio->bi_private = &compl; | |
1943 | bio->bi_end_io = repair_io_failure_callback; | |
1944 | bio->bi_size = 0; | |
1945 | map_length = length; | |
1946 | ||
1947 | ret = btrfs_map_block(map_tree, WRITE, logical, | |
1948 | &map_length, &bbio, mirror_num); | |
1949 | if (ret) { | |
1950 | bio_put(bio); | |
1951 | return -EIO; | |
1952 | } | |
1953 | BUG_ON(mirror_num != bbio->mirror_num); | |
1954 | sector = bbio->stripes[mirror_num-1].physical >> 9; | |
1955 | bio->bi_sector = sector; | |
1956 | dev = bbio->stripes[mirror_num-1].dev; | |
1957 | kfree(bbio); | |
1958 | if (!dev || !dev->bdev || !dev->writeable) { | |
1959 | bio_put(bio); | |
1960 | return -EIO; | |
1961 | } | |
1962 | bio->bi_bdev = dev->bdev; | |
1963 | bio_add_page(bio, page, length, start-page_offset(page)); | |
21adbd5c | 1964 | btrfsic_submit_bio(WRITE_SYNC, bio); |
4a54c8c1 JS |
1965 | wait_for_completion(&compl); |
1966 | ||
1967 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) { | |
1968 | /* try to remap that extent elsewhere? */ | |
1969 | bio_put(bio); | |
442a4f63 | 1970 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
1971 | return -EIO; |
1972 | } | |
1973 | ||
d5b025d5 | 1974 | printk_ratelimited_in_rcu(KERN_INFO "btrfs read error corrected: ino %lu off %llu " |
606686ee JB |
1975 | "(dev %s sector %llu)\n", page->mapping->host->i_ino, |
1976 | start, rcu_str_deref(dev->name), sector); | |
4a54c8c1 JS |
1977 | |
1978 | bio_put(bio); | |
1979 | return 0; | |
1980 | } | |
1981 | ||
ea466794 JB |
1982 | int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb, |
1983 | int mirror_num) | |
1984 | { | |
1985 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
1986 | u64 start = eb->start; | |
1987 | unsigned long i, num_pages = num_extent_pages(eb->start, eb->len); | |
d95603b2 | 1988 | int ret = 0; |
ea466794 JB |
1989 | |
1990 | for (i = 0; i < num_pages; i++) { | |
1991 | struct page *p = extent_buffer_page(eb, i); | |
1992 | ret = repair_io_failure(map_tree, start, PAGE_CACHE_SIZE, | |
1993 | start, p, mirror_num); | |
1994 | if (ret) | |
1995 | break; | |
1996 | start += PAGE_CACHE_SIZE; | |
1997 | } | |
1998 | ||
1999 | return ret; | |
2000 | } | |
2001 | ||
4a54c8c1 JS |
2002 | /* |
2003 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2004 | * to see if we need to process or clean up an io_failure_record | |
2005 | */ | |
2006 | static int clean_io_failure(u64 start, struct page *page) | |
2007 | { | |
2008 | u64 private; | |
2009 | u64 private_failure; | |
2010 | struct io_failure_record *failrec; | |
2011 | struct btrfs_mapping_tree *map_tree; | |
2012 | struct extent_state *state; | |
2013 | int num_copies; | |
2014 | int did_repair = 0; | |
2015 | int ret; | |
2016 | struct inode *inode = page->mapping->host; | |
2017 | ||
2018 | private = 0; | |
2019 | ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
2020 | (u64)-1, 1, EXTENT_DIRTY, 0); | |
2021 | if (!ret) | |
2022 | return 0; | |
2023 | ||
2024 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start, | |
2025 | &private_failure); | |
2026 | if (ret) | |
2027 | return 0; | |
2028 | ||
2029 | failrec = (struct io_failure_record *)(unsigned long) private_failure; | |
2030 | BUG_ON(!failrec->this_mirror); | |
2031 | ||
2032 | if (failrec->in_validation) { | |
2033 | /* there was no real error, just free the record */ | |
2034 | pr_debug("clean_io_failure: freeing dummy error at %llu\n", | |
2035 | failrec->start); | |
2036 | did_repair = 1; | |
2037 | goto out; | |
2038 | } | |
2039 | ||
2040 | spin_lock(&BTRFS_I(inode)->io_tree.lock); | |
2041 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, | |
2042 | failrec->start, | |
2043 | EXTENT_LOCKED); | |
2044 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); | |
2045 | ||
2046 | if (state && state->start == failrec->start) { | |
5d964051 SB |
2047 | num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info, |
2048 | failrec->logical, failrec->len); | |
4a54c8c1 | 2049 | if (num_copies > 1) { |
5d964051 | 2050 | map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree; |
4a54c8c1 JS |
2051 | ret = repair_io_failure(map_tree, start, failrec->len, |
2052 | failrec->logical, page, | |
2053 | failrec->failed_mirror); | |
2054 | did_repair = !ret; | |
2055 | } | |
2056 | } | |
2057 | ||
2058 | out: | |
2059 | if (!ret) | |
2060 | ret = free_io_failure(inode, failrec, did_repair); | |
2061 | ||
2062 | return ret; | |
2063 | } | |
2064 | ||
2065 | /* | |
2066 | * this is a generic handler for readpage errors (default | |
2067 | * readpage_io_failed_hook). if other copies exist, read those and write back | |
2068 | * good data to the failed position. does not investigate in remapping the | |
2069 | * failed extent elsewhere, hoping the device will be smart enough to do this as | |
2070 | * needed | |
2071 | */ | |
2072 | ||
2073 | static int bio_readpage_error(struct bio *failed_bio, struct page *page, | |
2074 | u64 start, u64 end, int failed_mirror, | |
2075 | struct extent_state *state) | |
2076 | { | |
2077 | struct io_failure_record *failrec = NULL; | |
2078 | u64 private; | |
2079 | struct extent_map *em; | |
2080 | struct inode *inode = page->mapping->host; | |
2081 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
2082 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2083 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
2084 | struct bio *bio; | |
2085 | int num_copies; | |
2086 | int ret; | |
2087 | int read_mode; | |
2088 | u64 logical; | |
2089 | ||
2090 | BUG_ON(failed_bio->bi_rw & REQ_WRITE); | |
2091 | ||
2092 | ret = get_state_private(failure_tree, start, &private); | |
2093 | if (ret) { | |
2094 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2095 | if (!failrec) | |
2096 | return -ENOMEM; | |
2097 | failrec->start = start; | |
2098 | failrec->len = end - start + 1; | |
2099 | failrec->this_mirror = 0; | |
2100 | failrec->bio_flags = 0; | |
2101 | failrec->in_validation = 0; | |
2102 | ||
2103 | read_lock(&em_tree->lock); | |
2104 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2105 | if (!em) { | |
2106 | read_unlock(&em_tree->lock); | |
2107 | kfree(failrec); | |
2108 | return -EIO; | |
2109 | } | |
2110 | ||
2111 | if (em->start > start || em->start + em->len < start) { | |
2112 | free_extent_map(em); | |
2113 | em = NULL; | |
2114 | } | |
2115 | read_unlock(&em_tree->lock); | |
2116 | ||
7a2d6a64 | 2117 | if (!em) { |
4a54c8c1 JS |
2118 | kfree(failrec); |
2119 | return -EIO; | |
2120 | } | |
2121 | logical = start - em->start; | |
2122 | logical = em->block_start + logical; | |
2123 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2124 | logical = em->block_start; | |
2125 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2126 | extent_set_compress_type(&failrec->bio_flags, | |
2127 | em->compress_type); | |
2128 | } | |
2129 | pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, " | |
2130 | "len=%llu\n", logical, start, failrec->len); | |
2131 | failrec->logical = logical; | |
2132 | free_extent_map(em); | |
2133 | ||
2134 | /* set the bits in the private failure tree */ | |
2135 | ret = set_extent_bits(failure_tree, start, end, | |
2136 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
2137 | if (ret >= 0) | |
2138 | ret = set_state_private(failure_tree, start, | |
2139 | (u64)(unsigned long)failrec); | |
2140 | /* set the bits in the inode's tree */ | |
2141 | if (ret >= 0) | |
2142 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED, | |
2143 | GFP_NOFS); | |
2144 | if (ret < 0) { | |
2145 | kfree(failrec); | |
2146 | return ret; | |
2147 | } | |
2148 | } else { | |
2149 | failrec = (struct io_failure_record *)(unsigned long)private; | |
2150 | pr_debug("bio_readpage_error: (found) logical=%llu, " | |
2151 | "start=%llu, len=%llu, validation=%d\n", | |
2152 | failrec->logical, failrec->start, failrec->len, | |
2153 | failrec->in_validation); | |
2154 | /* | |
2155 | * when data can be on disk more than twice, add to failrec here | |
2156 | * (e.g. with a list for failed_mirror) to make | |
2157 | * clean_io_failure() clean all those errors at once. | |
2158 | */ | |
2159 | } | |
5d964051 SB |
2160 | num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info, |
2161 | failrec->logical, failrec->len); | |
4a54c8c1 JS |
2162 | if (num_copies == 1) { |
2163 | /* | |
2164 | * we only have a single copy of the data, so don't bother with | |
2165 | * all the retry and error correction code that follows. no | |
2166 | * matter what the error is, it is very likely to persist. | |
2167 | */ | |
2168 | pr_debug("bio_readpage_error: cannot repair, num_copies == 1. " | |
2169 | "state=%p, num_copies=%d, next_mirror %d, " | |
2170 | "failed_mirror %d\n", state, num_copies, | |
2171 | failrec->this_mirror, failed_mirror); | |
2172 | free_io_failure(inode, failrec, 0); | |
2173 | return -EIO; | |
2174 | } | |
2175 | ||
2176 | if (!state) { | |
2177 | spin_lock(&tree->lock); | |
2178 | state = find_first_extent_bit_state(tree, failrec->start, | |
2179 | EXTENT_LOCKED); | |
2180 | if (state && state->start != failrec->start) | |
2181 | state = NULL; | |
2182 | spin_unlock(&tree->lock); | |
2183 | } | |
2184 | ||
2185 | /* | |
2186 | * there are two premises: | |
2187 | * a) deliver good data to the caller | |
2188 | * b) correct the bad sectors on disk | |
2189 | */ | |
2190 | if (failed_bio->bi_vcnt > 1) { | |
2191 | /* | |
2192 | * to fulfill b), we need to know the exact failing sectors, as | |
2193 | * we don't want to rewrite any more than the failed ones. thus, | |
2194 | * we need separate read requests for the failed bio | |
2195 | * | |
2196 | * if the following BUG_ON triggers, our validation request got | |
2197 | * merged. we need separate requests for our algorithm to work. | |
2198 | */ | |
2199 | BUG_ON(failrec->in_validation); | |
2200 | failrec->in_validation = 1; | |
2201 | failrec->this_mirror = failed_mirror; | |
2202 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; | |
2203 | } else { | |
2204 | /* | |
2205 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2206 | * of the failed sector and if we succeed, we have setup | |
2207 | * everything for repair_io_failure to do the rest for us. | |
2208 | */ | |
2209 | if (failrec->in_validation) { | |
2210 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2211 | failrec->in_validation = 0; | |
2212 | failrec->this_mirror = 0; | |
2213 | } | |
2214 | failrec->failed_mirror = failed_mirror; | |
2215 | failrec->this_mirror++; | |
2216 | if (failrec->this_mirror == failed_mirror) | |
2217 | failrec->this_mirror++; | |
2218 | read_mode = READ_SYNC; | |
2219 | } | |
2220 | ||
2221 | if (!state || failrec->this_mirror > num_copies) { | |
2222 | pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, " | |
2223 | "next_mirror %d, failed_mirror %d\n", state, | |
2224 | num_copies, failrec->this_mirror, failed_mirror); | |
2225 | free_io_failure(inode, failrec, 0); | |
2226 | return -EIO; | |
2227 | } | |
2228 | ||
2229 | bio = bio_alloc(GFP_NOFS, 1); | |
e627ee7b TI |
2230 | if (!bio) { |
2231 | free_io_failure(inode, failrec, 0); | |
2232 | return -EIO; | |
2233 | } | |
4a54c8c1 JS |
2234 | bio->bi_private = state; |
2235 | bio->bi_end_io = failed_bio->bi_end_io; | |
2236 | bio->bi_sector = failrec->logical >> 9; | |
2237 | bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; | |
2238 | bio->bi_size = 0; | |
2239 | ||
2240 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); | |
2241 | ||
2242 | pr_debug("bio_readpage_error: submitting new read[%#x] to " | |
2243 | "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode, | |
2244 | failrec->this_mirror, num_copies, failrec->in_validation); | |
2245 | ||
013bd4c3 TI |
2246 | ret = tree->ops->submit_bio_hook(inode, read_mode, bio, |
2247 | failrec->this_mirror, | |
2248 | failrec->bio_flags, 0); | |
2249 | return ret; | |
4a54c8c1 JS |
2250 | } |
2251 | ||
d1310b2e CM |
2252 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2253 | ||
87826df0 JM |
2254 | int end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
2255 | { | |
2256 | int uptodate = (err == 0); | |
2257 | struct extent_io_tree *tree; | |
2258 | int ret; | |
2259 | ||
2260 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
2261 | ||
2262 | if (tree->ops && tree->ops->writepage_end_io_hook) { | |
2263 | ret = tree->ops->writepage_end_io_hook(page, start, | |
2264 | end, NULL, uptodate); | |
2265 | if (ret) | |
2266 | uptodate = 0; | |
2267 | } | |
2268 | ||
87826df0 | 2269 | if (!uptodate) { |
87826df0 JM |
2270 | ClearPageUptodate(page); |
2271 | SetPageError(page); | |
2272 | } | |
2273 | return 0; | |
2274 | } | |
2275 | ||
d1310b2e CM |
2276 | /* |
2277 | * after a writepage IO is done, we need to: | |
2278 | * clear the uptodate bits on error | |
2279 | * clear the writeback bits in the extent tree for this IO | |
2280 | * end_page_writeback if the page has no more pending IO | |
2281 | * | |
2282 | * Scheduling is not allowed, so the extent state tree is expected | |
2283 | * to have one and only one object corresponding to this IO. | |
2284 | */ | |
d1310b2e | 2285 | static void end_bio_extent_writepage(struct bio *bio, int err) |
d1310b2e | 2286 | { |
d1310b2e | 2287 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
902b22f3 | 2288 | struct extent_io_tree *tree; |
d1310b2e CM |
2289 | u64 start; |
2290 | u64 end; | |
2291 | int whole_page; | |
2292 | ||
d1310b2e CM |
2293 | do { |
2294 | struct page *page = bvec->bv_page; | |
902b22f3 DW |
2295 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
2296 | ||
d1310b2e CM |
2297 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
2298 | bvec->bv_offset; | |
2299 | end = start + bvec->bv_len - 1; | |
2300 | ||
2301 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
2302 | whole_page = 1; | |
2303 | else | |
2304 | whole_page = 0; | |
2305 | ||
2306 | if (--bvec >= bio->bi_io_vec) | |
2307 | prefetchw(&bvec->bv_page->flags); | |
1259ab75 | 2308 | |
87826df0 JM |
2309 | if (end_extent_writepage(page, err, start, end)) |
2310 | continue; | |
70dec807 | 2311 | |
d1310b2e CM |
2312 | if (whole_page) |
2313 | end_page_writeback(page); | |
2314 | else | |
2315 | check_page_writeback(tree, page); | |
d1310b2e | 2316 | } while (bvec >= bio->bi_io_vec); |
2b1f55b0 | 2317 | |
d1310b2e | 2318 | bio_put(bio); |
d1310b2e CM |
2319 | } |
2320 | ||
2321 | /* | |
2322 | * after a readpage IO is done, we need to: | |
2323 | * clear the uptodate bits on error | |
2324 | * set the uptodate bits if things worked | |
2325 | * set the page up to date if all extents in the tree are uptodate | |
2326 | * clear the lock bit in the extent tree | |
2327 | * unlock the page if there are no other extents locked for it | |
2328 | * | |
2329 | * Scheduling is not allowed, so the extent state tree is expected | |
2330 | * to have one and only one object corresponding to this IO. | |
2331 | */ | |
d1310b2e | 2332 | static void end_bio_extent_readpage(struct bio *bio, int err) |
d1310b2e CM |
2333 | { |
2334 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
4125bf76 CM |
2335 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
2336 | struct bio_vec *bvec = bio->bi_io_vec; | |
902b22f3 | 2337 | struct extent_io_tree *tree; |
d1310b2e CM |
2338 | u64 start; |
2339 | u64 end; | |
2340 | int whole_page; | |
5cf1ab56 | 2341 | int mirror; |
d1310b2e CM |
2342 | int ret; |
2343 | ||
d20f7043 CM |
2344 | if (err) |
2345 | uptodate = 0; | |
2346 | ||
d1310b2e CM |
2347 | do { |
2348 | struct page *page = bvec->bv_page; | |
507903b8 AJ |
2349 | struct extent_state *cached = NULL; |
2350 | struct extent_state *state; | |
2351 | ||
be3940c0 KO |
2352 | pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, " |
2353 | "mirror=%ld\n", (u64)bio->bi_sector, err, | |
4a54c8c1 | 2354 | (long int)bio->bi_bdev); |
902b22f3 DW |
2355 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
2356 | ||
d1310b2e CM |
2357 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
2358 | bvec->bv_offset; | |
2359 | end = start + bvec->bv_len - 1; | |
2360 | ||
2361 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
2362 | whole_page = 1; | |
2363 | else | |
2364 | whole_page = 0; | |
2365 | ||
4125bf76 | 2366 | if (++bvec <= bvec_end) |
d1310b2e CM |
2367 | prefetchw(&bvec->bv_page->flags); |
2368 | ||
507903b8 | 2369 | spin_lock(&tree->lock); |
0d399205 | 2370 | state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED); |
109b36a2 | 2371 | if (state && state->start == start) { |
507903b8 AJ |
2372 | /* |
2373 | * take a reference on the state, unlock will drop | |
2374 | * the ref | |
2375 | */ | |
2376 | cache_state(state, &cached); | |
2377 | } | |
2378 | spin_unlock(&tree->lock); | |
2379 | ||
5cf1ab56 | 2380 | mirror = (int)(unsigned long)bio->bi_bdev; |
d1310b2e | 2381 | if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) { |
70dec807 | 2382 | ret = tree->ops->readpage_end_io_hook(page, start, end, |
5cf1ab56 | 2383 | state, mirror); |
5ee0844d | 2384 | if (ret) |
d1310b2e | 2385 | uptodate = 0; |
5ee0844d | 2386 | else |
4a54c8c1 | 2387 | clean_io_failure(start, page); |
d1310b2e | 2388 | } |
ea466794 | 2389 | |
ea466794 | 2390 | if (!uptodate && tree->ops && tree->ops->readpage_io_failed_hook) { |
5cf1ab56 | 2391 | ret = tree->ops->readpage_io_failed_hook(page, mirror); |
ea466794 JB |
2392 | if (!ret && !err && |
2393 | test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
2394 | uptodate = 1; | |
2395 | } else if (!uptodate) { | |
f4a8e656 JS |
2396 | /* |
2397 | * The generic bio_readpage_error handles errors the | |
2398 | * following way: If possible, new read requests are | |
2399 | * created and submitted and will end up in | |
2400 | * end_bio_extent_readpage as well (if we're lucky, not | |
2401 | * in the !uptodate case). In that case it returns 0 and | |
2402 | * we just go on with the next page in our bio. If it | |
2403 | * can't handle the error it will return -EIO and we | |
2404 | * remain responsible for that page. | |
2405 | */ | |
5cf1ab56 | 2406 | ret = bio_readpage_error(bio, page, start, end, mirror, NULL); |
7e38326f | 2407 | if (ret == 0) { |
3b951516 CM |
2408 | uptodate = |
2409 | test_bit(BIO_UPTODATE, &bio->bi_flags); | |
d20f7043 CM |
2410 | if (err) |
2411 | uptodate = 0; | |
507903b8 | 2412 | uncache_state(&cached); |
7e38326f CM |
2413 | continue; |
2414 | } | |
2415 | } | |
d1310b2e | 2416 | |
0b32f4bb | 2417 | if (uptodate && tree->track_uptodate) { |
507903b8 | 2418 | set_extent_uptodate(tree, start, end, &cached, |
902b22f3 | 2419 | GFP_ATOMIC); |
771ed689 | 2420 | } |
507903b8 | 2421 | unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC); |
d1310b2e | 2422 | |
70dec807 CM |
2423 | if (whole_page) { |
2424 | if (uptodate) { | |
2425 | SetPageUptodate(page); | |
2426 | } else { | |
2427 | ClearPageUptodate(page); | |
2428 | SetPageError(page); | |
2429 | } | |
d1310b2e | 2430 | unlock_page(page); |
70dec807 CM |
2431 | } else { |
2432 | if (uptodate) { | |
2433 | check_page_uptodate(tree, page); | |
2434 | } else { | |
2435 | ClearPageUptodate(page); | |
2436 | SetPageError(page); | |
2437 | } | |
d1310b2e | 2438 | check_page_locked(tree, page); |
70dec807 | 2439 | } |
4125bf76 | 2440 | } while (bvec <= bvec_end); |
d1310b2e CM |
2441 | |
2442 | bio_put(bio); | |
d1310b2e CM |
2443 | } |
2444 | ||
88f794ed MX |
2445 | struct bio * |
2446 | btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
2447 | gfp_t gfp_flags) | |
d1310b2e CM |
2448 | { |
2449 | struct bio *bio; | |
2450 | ||
2451 | bio = bio_alloc(gfp_flags, nr_vecs); | |
2452 | ||
2453 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
2454 | while (!bio && (nr_vecs /= 2)) | |
2455 | bio = bio_alloc(gfp_flags, nr_vecs); | |
2456 | } | |
2457 | ||
2458 | if (bio) { | |
e1c4b745 | 2459 | bio->bi_size = 0; |
d1310b2e CM |
2460 | bio->bi_bdev = bdev; |
2461 | bio->bi_sector = first_sector; | |
2462 | } | |
2463 | return bio; | |
2464 | } | |
2465 | ||
79787eaa JM |
2466 | /* |
2467 | * Since writes are async, they will only return -ENOMEM. | |
2468 | * Reads can return the full range of I/O error conditions. | |
2469 | */ | |
355808c2 JM |
2470 | static int __must_check submit_one_bio(int rw, struct bio *bio, |
2471 | int mirror_num, unsigned long bio_flags) | |
d1310b2e | 2472 | { |
d1310b2e | 2473 | int ret = 0; |
70dec807 CM |
2474 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
2475 | struct page *page = bvec->bv_page; | |
2476 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 2477 | u64 start; |
70dec807 CM |
2478 | |
2479 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset; | |
70dec807 | 2480 | |
902b22f3 | 2481 | bio->bi_private = NULL; |
d1310b2e CM |
2482 | |
2483 | bio_get(bio); | |
2484 | ||
065631f6 | 2485 | if (tree->ops && tree->ops->submit_bio_hook) |
6b82ce8d | 2486 | ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio, |
eaf25d93 | 2487 | mirror_num, bio_flags, start); |
0b86a832 | 2488 | else |
21adbd5c | 2489 | btrfsic_submit_bio(rw, bio); |
4a54c8c1 | 2490 | |
d1310b2e CM |
2491 | if (bio_flagged(bio, BIO_EOPNOTSUPP)) |
2492 | ret = -EOPNOTSUPP; | |
2493 | bio_put(bio); | |
2494 | return ret; | |
2495 | } | |
2496 | ||
3444a972 JM |
2497 | static int merge_bio(struct extent_io_tree *tree, struct page *page, |
2498 | unsigned long offset, size_t size, struct bio *bio, | |
2499 | unsigned long bio_flags) | |
2500 | { | |
2501 | int ret = 0; | |
2502 | if (tree->ops && tree->ops->merge_bio_hook) | |
2503 | ret = tree->ops->merge_bio_hook(page, offset, size, bio, | |
2504 | bio_flags); | |
2505 | BUG_ON(ret < 0); | |
2506 | return ret; | |
2507 | ||
2508 | } | |
2509 | ||
d1310b2e CM |
2510 | static int submit_extent_page(int rw, struct extent_io_tree *tree, |
2511 | struct page *page, sector_t sector, | |
2512 | size_t size, unsigned long offset, | |
2513 | struct block_device *bdev, | |
2514 | struct bio **bio_ret, | |
2515 | unsigned long max_pages, | |
f188591e | 2516 | bio_end_io_t end_io_func, |
c8b97818 CM |
2517 | int mirror_num, |
2518 | unsigned long prev_bio_flags, | |
2519 | unsigned long bio_flags) | |
d1310b2e CM |
2520 | { |
2521 | int ret = 0; | |
2522 | struct bio *bio; | |
2523 | int nr; | |
c8b97818 CM |
2524 | int contig = 0; |
2525 | int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED; | |
2526 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; | |
5b050f04 | 2527 | size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE); |
d1310b2e CM |
2528 | |
2529 | if (bio_ret && *bio_ret) { | |
2530 | bio = *bio_ret; | |
c8b97818 CM |
2531 | if (old_compressed) |
2532 | contig = bio->bi_sector == sector; | |
2533 | else | |
2534 | contig = bio->bi_sector + (bio->bi_size >> 9) == | |
2535 | sector; | |
2536 | ||
2537 | if (prev_bio_flags != bio_flags || !contig || | |
3444a972 | 2538 | merge_bio(tree, page, offset, page_size, bio, bio_flags) || |
c8b97818 CM |
2539 | bio_add_page(bio, page, page_size, offset) < page_size) { |
2540 | ret = submit_one_bio(rw, bio, mirror_num, | |
2541 | prev_bio_flags); | |
79787eaa JM |
2542 | if (ret < 0) |
2543 | return ret; | |
d1310b2e CM |
2544 | bio = NULL; |
2545 | } else { | |
2546 | return 0; | |
2547 | } | |
2548 | } | |
c8b97818 CM |
2549 | if (this_compressed) |
2550 | nr = BIO_MAX_PAGES; | |
2551 | else | |
2552 | nr = bio_get_nr_vecs(bdev); | |
2553 | ||
88f794ed | 2554 | bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH); |
5df67083 TI |
2555 | if (!bio) |
2556 | return -ENOMEM; | |
70dec807 | 2557 | |
c8b97818 | 2558 | bio_add_page(bio, page, page_size, offset); |
d1310b2e CM |
2559 | bio->bi_end_io = end_io_func; |
2560 | bio->bi_private = tree; | |
70dec807 | 2561 | |
d397712b | 2562 | if (bio_ret) |
d1310b2e | 2563 | *bio_ret = bio; |
d397712b | 2564 | else |
c8b97818 | 2565 | ret = submit_one_bio(rw, bio, mirror_num, bio_flags); |
d1310b2e CM |
2566 | |
2567 | return ret; | |
2568 | } | |
2569 | ||
4f2de97a | 2570 | void attach_extent_buffer_page(struct extent_buffer *eb, struct page *page) |
d1310b2e CM |
2571 | { |
2572 | if (!PagePrivate(page)) { | |
2573 | SetPagePrivate(page); | |
d1310b2e | 2574 | page_cache_get(page); |
4f2de97a JB |
2575 | set_page_private(page, (unsigned long)eb); |
2576 | } else { | |
2577 | WARN_ON(page->private != (unsigned long)eb); | |
d1310b2e CM |
2578 | } |
2579 | } | |
2580 | ||
4f2de97a | 2581 | void set_page_extent_mapped(struct page *page) |
d1310b2e | 2582 | { |
4f2de97a JB |
2583 | if (!PagePrivate(page)) { |
2584 | SetPagePrivate(page); | |
2585 | page_cache_get(page); | |
2586 | set_page_private(page, EXTENT_PAGE_PRIVATE); | |
2587 | } | |
d1310b2e CM |
2588 | } |
2589 | ||
2590 | /* | |
2591 | * basic readpage implementation. Locked extent state structs are inserted | |
2592 | * into the tree that are removed when the IO is done (by the end_io | |
2593 | * handlers) | |
79787eaa | 2594 | * XXX JDM: This needs looking at to ensure proper page locking |
d1310b2e CM |
2595 | */ |
2596 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
2597 | struct page *page, | |
2598 | get_extent_t *get_extent, | |
c8b97818 CM |
2599 | struct bio **bio, int mirror_num, |
2600 | unsigned long *bio_flags) | |
d1310b2e CM |
2601 | { |
2602 | struct inode *inode = page->mapping->host; | |
2603 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2604 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2605 | u64 end; | |
2606 | u64 cur = start; | |
2607 | u64 extent_offset; | |
2608 | u64 last_byte = i_size_read(inode); | |
2609 | u64 block_start; | |
2610 | u64 cur_end; | |
2611 | sector_t sector; | |
2612 | struct extent_map *em; | |
2613 | struct block_device *bdev; | |
11c65dcc | 2614 | struct btrfs_ordered_extent *ordered; |
d1310b2e CM |
2615 | int ret; |
2616 | int nr = 0; | |
306e16ce | 2617 | size_t pg_offset = 0; |
d1310b2e | 2618 | size_t iosize; |
c8b97818 | 2619 | size_t disk_io_size; |
d1310b2e | 2620 | size_t blocksize = inode->i_sb->s_blocksize; |
c8b97818 | 2621 | unsigned long this_bio_flag = 0; |
d1310b2e CM |
2622 | |
2623 | set_page_extent_mapped(page); | |
2624 | ||
90a887c9 DM |
2625 | if (!PageUptodate(page)) { |
2626 | if (cleancache_get_page(page) == 0) { | |
2627 | BUG_ON(blocksize != PAGE_SIZE); | |
2628 | goto out; | |
2629 | } | |
2630 | } | |
2631 | ||
d1310b2e | 2632 | end = page_end; |
11c65dcc | 2633 | while (1) { |
d0082371 | 2634 | lock_extent(tree, start, end); |
11c65dcc JB |
2635 | ordered = btrfs_lookup_ordered_extent(inode, start); |
2636 | if (!ordered) | |
2637 | break; | |
d0082371 | 2638 | unlock_extent(tree, start, end); |
11c65dcc JB |
2639 | btrfs_start_ordered_extent(inode, ordered, 1); |
2640 | btrfs_put_ordered_extent(ordered); | |
2641 | } | |
d1310b2e | 2642 | |
c8b97818 CM |
2643 | if (page->index == last_byte >> PAGE_CACHE_SHIFT) { |
2644 | char *userpage; | |
2645 | size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1); | |
2646 | ||
2647 | if (zero_offset) { | |
2648 | iosize = PAGE_CACHE_SIZE - zero_offset; | |
7ac687d9 | 2649 | userpage = kmap_atomic(page); |
c8b97818 CM |
2650 | memset(userpage + zero_offset, 0, iosize); |
2651 | flush_dcache_page(page); | |
7ac687d9 | 2652 | kunmap_atomic(userpage); |
c8b97818 CM |
2653 | } |
2654 | } | |
d1310b2e CM |
2655 | while (cur <= end) { |
2656 | if (cur >= last_byte) { | |
2657 | char *userpage; | |
507903b8 AJ |
2658 | struct extent_state *cached = NULL; |
2659 | ||
306e16ce | 2660 | iosize = PAGE_CACHE_SIZE - pg_offset; |
7ac687d9 | 2661 | userpage = kmap_atomic(page); |
306e16ce | 2662 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2663 | flush_dcache_page(page); |
7ac687d9 | 2664 | kunmap_atomic(userpage); |
d1310b2e | 2665 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 AJ |
2666 | &cached, GFP_NOFS); |
2667 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
2668 | &cached, GFP_NOFS); | |
d1310b2e CM |
2669 | break; |
2670 | } | |
306e16ce | 2671 | em = get_extent(inode, page, pg_offset, cur, |
d1310b2e | 2672 | end - cur + 1, 0); |
c704005d | 2673 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 2674 | SetPageError(page); |
d0082371 | 2675 | unlock_extent(tree, cur, end); |
d1310b2e CM |
2676 | break; |
2677 | } | |
d1310b2e CM |
2678 | extent_offset = cur - em->start; |
2679 | BUG_ON(extent_map_end(em) <= cur); | |
2680 | BUG_ON(end < cur); | |
2681 | ||
261507a0 | 2682 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
c8b97818 | 2683 | this_bio_flag = EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
2684 | extent_set_compress_type(&this_bio_flag, |
2685 | em->compress_type); | |
2686 | } | |
c8b97818 | 2687 | |
d1310b2e CM |
2688 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
2689 | cur_end = min(extent_map_end(em) - 1, end); | |
2690 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
c8b97818 CM |
2691 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
2692 | disk_io_size = em->block_len; | |
2693 | sector = em->block_start >> 9; | |
2694 | } else { | |
2695 | sector = (em->block_start + extent_offset) >> 9; | |
2696 | disk_io_size = iosize; | |
2697 | } | |
d1310b2e CM |
2698 | bdev = em->bdev; |
2699 | block_start = em->block_start; | |
d899e052 YZ |
2700 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2701 | block_start = EXTENT_MAP_HOLE; | |
d1310b2e CM |
2702 | free_extent_map(em); |
2703 | em = NULL; | |
2704 | ||
2705 | /* we've found a hole, just zero and go on */ | |
2706 | if (block_start == EXTENT_MAP_HOLE) { | |
2707 | char *userpage; | |
507903b8 AJ |
2708 | struct extent_state *cached = NULL; |
2709 | ||
7ac687d9 | 2710 | userpage = kmap_atomic(page); |
306e16ce | 2711 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2712 | flush_dcache_page(page); |
7ac687d9 | 2713 | kunmap_atomic(userpage); |
d1310b2e CM |
2714 | |
2715 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 AJ |
2716 | &cached, GFP_NOFS); |
2717 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
2718 | &cached, GFP_NOFS); | |
d1310b2e | 2719 | cur = cur + iosize; |
306e16ce | 2720 | pg_offset += iosize; |
d1310b2e CM |
2721 | continue; |
2722 | } | |
2723 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
2724 | if (test_range_bit(tree, cur, cur_end, |
2725 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 2726 | check_page_uptodate(tree, page); |
d0082371 | 2727 | unlock_extent(tree, cur, cur + iosize - 1); |
d1310b2e | 2728 | cur = cur + iosize; |
306e16ce | 2729 | pg_offset += iosize; |
d1310b2e CM |
2730 | continue; |
2731 | } | |
70dec807 CM |
2732 | /* we have an inline extent but it didn't get marked up |
2733 | * to date. Error out | |
2734 | */ | |
2735 | if (block_start == EXTENT_MAP_INLINE) { | |
2736 | SetPageError(page); | |
d0082371 | 2737 | unlock_extent(tree, cur, cur + iosize - 1); |
70dec807 | 2738 | cur = cur + iosize; |
306e16ce | 2739 | pg_offset += iosize; |
70dec807 CM |
2740 | continue; |
2741 | } | |
d1310b2e CM |
2742 | |
2743 | ret = 0; | |
2744 | if (tree->ops && tree->ops->readpage_io_hook) { | |
2745 | ret = tree->ops->readpage_io_hook(page, cur, | |
2746 | cur + iosize - 1); | |
2747 | } | |
2748 | if (!ret) { | |
89642229 CM |
2749 | unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1; |
2750 | pnr -= page->index; | |
d1310b2e | 2751 | ret = submit_extent_page(READ, tree, page, |
306e16ce | 2752 | sector, disk_io_size, pg_offset, |
89642229 | 2753 | bdev, bio, pnr, |
c8b97818 CM |
2754 | end_bio_extent_readpage, mirror_num, |
2755 | *bio_flags, | |
2756 | this_bio_flag); | |
edd33c99 JB |
2757 | if (!ret) { |
2758 | nr++; | |
2759 | *bio_flags = this_bio_flag; | |
2760 | } | |
d1310b2e | 2761 | } |
edd33c99 | 2762 | if (ret) { |
d1310b2e | 2763 | SetPageError(page); |
edd33c99 JB |
2764 | unlock_extent(tree, cur, cur + iosize - 1); |
2765 | } | |
d1310b2e | 2766 | cur = cur + iosize; |
306e16ce | 2767 | pg_offset += iosize; |
d1310b2e | 2768 | } |
90a887c9 | 2769 | out: |
d1310b2e CM |
2770 | if (!nr) { |
2771 | if (!PageError(page)) | |
2772 | SetPageUptodate(page); | |
2773 | unlock_page(page); | |
2774 | } | |
2775 | return 0; | |
2776 | } | |
2777 | ||
2778 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, | |
8ddc7d9c | 2779 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
2780 | { |
2781 | struct bio *bio = NULL; | |
c8b97818 | 2782 | unsigned long bio_flags = 0; |
d1310b2e CM |
2783 | int ret; |
2784 | ||
8ddc7d9c | 2785 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, |
c8b97818 | 2786 | &bio_flags); |
d1310b2e | 2787 | if (bio) |
8ddc7d9c | 2788 | ret = submit_one_bio(READ, bio, mirror_num, bio_flags); |
d1310b2e CM |
2789 | return ret; |
2790 | } | |
d1310b2e | 2791 | |
11c8349b CM |
2792 | static noinline void update_nr_written(struct page *page, |
2793 | struct writeback_control *wbc, | |
2794 | unsigned long nr_written) | |
2795 | { | |
2796 | wbc->nr_to_write -= nr_written; | |
2797 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && | |
2798 | wbc->range_start == 0 && wbc->range_end == LLONG_MAX)) | |
2799 | page->mapping->writeback_index = page->index + nr_written; | |
2800 | } | |
2801 | ||
d1310b2e CM |
2802 | /* |
2803 | * the writepage semantics are similar to regular writepage. extent | |
2804 | * records are inserted to lock ranges in the tree, and as dirty areas | |
2805 | * are found, they are marked writeback. Then the lock bits are removed | |
2806 | * and the end_io handler clears the writeback ranges | |
2807 | */ | |
2808 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
2809 | void *data) | |
2810 | { | |
2811 | struct inode *inode = page->mapping->host; | |
2812 | struct extent_page_data *epd = data; | |
2813 | struct extent_io_tree *tree = epd->tree; | |
2814 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2815 | u64 delalloc_start; | |
2816 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2817 | u64 end; | |
2818 | u64 cur = start; | |
2819 | u64 extent_offset; | |
2820 | u64 last_byte = i_size_read(inode); | |
2821 | u64 block_start; | |
2822 | u64 iosize; | |
2823 | sector_t sector; | |
2c64c53d | 2824 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
2825 | struct extent_map *em; |
2826 | struct block_device *bdev; | |
2827 | int ret; | |
2828 | int nr = 0; | |
7f3c74fb | 2829 | size_t pg_offset = 0; |
d1310b2e CM |
2830 | size_t blocksize; |
2831 | loff_t i_size = i_size_read(inode); | |
2832 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
2833 | u64 nr_delalloc; | |
2834 | u64 delalloc_end; | |
c8b97818 CM |
2835 | int page_started; |
2836 | int compressed; | |
ffbd517d | 2837 | int write_flags; |
771ed689 | 2838 | unsigned long nr_written = 0; |
9e487107 | 2839 | bool fill_delalloc = true; |
d1310b2e | 2840 | |
ffbd517d | 2841 | if (wbc->sync_mode == WB_SYNC_ALL) |
721a9602 | 2842 | write_flags = WRITE_SYNC; |
ffbd517d CM |
2843 | else |
2844 | write_flags = WRITE; | |
2845 | ||
1abe9b8a | 2846 | trace___extent_writepage(page, inode, wbc); |
2847 | ||
d1310b2e | 2848 | WARN_ON(!PageLocked(page)); |
bf0da8c1 CM |
2849 | |
2850 | ClearPageError(page); | |
2851 | ||
7f3c74fb | 2852 | pg_offset = i_size & (PAGE_CACHE_SIZE - 1); |
211c17f5 | 2853 | if (page->index > end_index || |
7f3c74fb | 2854 | (page->index == end_index && !pg_offset)) { |
39be25cd | 2855 | page->mapping->a_ops->invalidatepage(page, 0); |
d1310b2e CM |
2856 | unlock_page(page); |
2857 | return 0; | |
2858 | } | |
2859 | ||
2860 | if (page->index == end_index) { | |
2861 | char *userpage; | |
2862 | ||
7ac687d9 | 2863 | userpage = kmap_atomic(page); |
7f3c74fb CM |
2864 | memset(userpage + pg_offset, 0, |
2865 | PAGE_CACHE_SIZE - pg_offset); | |
7ac687d9 | 2866 | kunmap_atomic(userpage); |
211c17f5 | 2867 | flush_dcache_page(page); |
d1310b2e | 2868 | } |
7f3c74fb | 2869 | pg_offset = 0; |
d1310b2e CM |
2870 | |
2871 | set_page_extent_mapped(page); | |
2872 | ||
9e487107 JB |
2873 | if (!tree->ops || !tree->ops->fill_delalloc) |
2874 | fill_delalloc = false; | |
2875 | ||
d1310b2e CM |
2876 | delalloc_start = start; |
2877 | delalloc_end = 0; | |
c8b97818 | 2878 | page_started = 0; |
9e487107 | 2879 | if (!epd->extent_locked && fill_delalloc) { |
f85d7d6c | 2880 | u64 delalloc_to_write = 0; |
11c8349b CM |
2881 | /* |
2882 | * make sure the wbc mapping index is at least updated | |
2883 | * to this page. | |
2884 | */ | |
2885 | update_nr_written(page, wbc, 0); | |
2886 | ||
d397712b | 2887 | while (delalloc_end < page_end) { |
771ed689 | 2888 | nr_delalloc = find_lock_delalloc_range(inode, tree, |
c8b97818 CM |
2889 | page, |
2890 | &delalloc_start, | |
d1310b2e CM |
2891 | &delalloc_end, |
2892 | 128 * 1024 * 1024); | |
771ed689 CM |
2893 | if (nr_delalloc == 0) { |
2894 | delalloc_start = delalloc_end + 1; | |
2895 | continue; | |
2896 | } | |
013bd4c3 TI |
2897 | ret = tree->ops->fill_delalloc(inode, page, |
2898 | delalloc_start, | |
2899 | delalloc_end, | |
2900 | &page_started, | |
2901 | &nr_written); | |
79787eaa JM |
2902 | /* File system has been set read-only */ |
2903 | if (ret) { | |
2904 | SetPageError(page); | |
2905 | goto done; | |
2906 | } | |
f85d7d6c CM |
2907 | /* |
2908 | * delalloc_end is already one less than the total | |
2909 | * length, so we don't subtract one from | |
2910 | * PAGE_CACHE_SIZE | |
2911 | */ | |
2912 | delalloc_to_write += (delalloc_end - delalloc_start + | |
2913 | PAGE_CACHE_SIZE) >> | |
2914 | PAGE_CACHE_SHIFT; | |
d1310b2e | 2915 | delalloc_start = delalloc_end + 1; |
d1310b2e | 2916 | } |
f85d7d6c CM |
2917 | if (wbc->nr_to_write < delalloc_to_write) { |
2918 | int thresh = 8192; | |
2919 | ||
2920 | if (delalloc_to_write < thresh * 2) | |
2921 | thresh = delalloc_to_write; | |
2922 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
2923 | thresh); | |
2924 | } | |
c8b97818 | 2925 | |
771ed689 CM |
2926 | /* did the fill delalloc function already unlock and start |
2927 | * the IO? | |
2928 | */ | |
2929 | if (page_started) { | |
2930 | ret = 0; | |
11c8349b CM |
2931 | /* |
2932 | * we've unlocked the page, so we can't update | |
2933 | * the mapping's writeback index, just update | |
2934 | * nr_to_write. | |
2935 | */ | |
2936 | wbc->nr_to_write -= nr_written; | |
2937 | goto done_unlocked; | |
771ed689 | 2938 | } |
c8b97818 | 2939 | } |
247e743c | 2940 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
2941 | ret = tree->ops->writepage_start_hook(page, start, |
2942 | page_end); | |
87826df0 JM |
2943 | if (ret) { |
2944 | /* Fixup worker will requeue */ | |
2945 | if (ret == -EBUSY) | |
2946 | wbc->pages_skipped++; | |
2947 | else | |
2948 | redirty_page_for_writepage(wbc, page); | |
11c8349b | 2949 | update_nr_written(page, wbc, nr_written); |
247e743c | 2950 | unlock_page(page); |
771ed689 | 2951 | ret = 0; |
11c8349b | 2952 | goto done_unlocked; |
247e743c CM |
2953 | } |
2954 | } | |
2955 | ||
11c8349b CM |
2956 | /* |
2957 | * we don't want to touch the inode after unlocking the page, | |
2958 | * so we update the mapping writeback index now | |
2959 | */ | |
2960 | update_nr_written(page, wbc, nr_written + 1); | |
771ed689 | 2961 | |
d1310b2e | 2962 | end = page_end; |
d1310b2e | 2963 | if (last_byte <= start) { |
e6dcd2dc CM |
2964 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2965 | tree->ops->writepage_end_io_hook(page, start, | |
2966 | page_end, NULL, 1); | |
d1310b2e CM |
2967 | goto done; |
2968 | } | |
2969 | ||
d1310b2e CM |
2970 | blocksize = inode->i_sb->s_blocksize; |
2971 | ||
2972 | while (cur <= end) { | |
2973 | if (cur >= last_byte) { | |
e6dcd2dc CM |
2974 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2975 | tree->ops->writepage_end_io_hook(page, cur, | |
2976 | page_end, NULL, 1); | |
d1310b2e CM |
2977 | break; |
2978 | } | |
7f3c74fb | 2979 | em = epd->get_extent(inode, page, pg_offset, cur, |
d1310b2e | 2980 | end - cur + 1, 1); |
c704005d | 2981 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e CM |
2982 | SetPageError(page); |
2983 | break; | |
2984 | } | |
2985 | ||
2986 | extent_offset = cur - em->start; | |
2987 | BUG_ON(extent_map_end(em) <= cur); | |
2988 | BUG_ON(end < cur); | |
2989 | iosize = min(extent_map_end(em) - cur, end - cur + 1); | |
2990 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
2991 | sector = (em->block_start + extent_offset) >> 9; | |
2992 | bdev = em->bdev; | |
2993 | block_start = em->block_start; | |
c8b97818 | 2994 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
2995 | free_extent_map(em); |
2996 | em = NULL; | |
2997 | ||
c8b97818 CM |
2998 | /* |
2999 | * compressed and inline extents are written through other | |
3000 | * paths in the FS | |
3001 | */ | |
3002 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3003 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
3004 | /* |
3005 | * end_io notification does not happen here for | |
3006 | * compressed extents | |
3007 | */ | |
3008 | if (!compressed && tree->ops && | |
3009 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
3010 | tree->ops->writepage_end_io_hook(page, cur, |
3011 | cur + iosize - 1, | |
3012 | NULL, 1); | |
c8b97818 CM |
3013 | else if (compressed) { |
3014 | /* we don't want to end_page_writeback on | |
3015 | * a compressed extent. this happens | |
3016 | * elsewhere | |
3017 | */ | |
3018 | nr++; | |
3019 | } | |
3020 | ||
3021 | cur += iosize; | |
7f3c74fb | 3022 | pg_offset += iosize; |
d1310b2e CM |
3023 | continue; |
3024 | } | |
d1310b2e CM |
3025 | /* leave this out until we have a page_mkwrite call */ |
3026 | if (0 && !test_range_bit(tree, cur, cur + iosize - 1, | |
9655d298 | 3027 | EXTENT_DIRTY, 0, NULL)) { |
d1310b2e | 3028 | cur = cur + iosize; |
7f3c74fb | 3029 | pg_offset += iosize; |
d1310b2e CM |
3030 | continue; |
3031 | } | |
c8b97818 | 3032 | |
d1310b2e CM |
3033 | if (tree->ops && tree->ops->writepage_io_hook) { |
3034 | ret = tree->ops->writepage_io_hook(page, cur, | |
3035 | cur + iosize - 1); | |
3036 | } else { | |
3037 | ret = 0; | |
3038 | } | |
1259ab75 | 3039 | if (ret) { |
d1310b2e | 3040 | SetPageError(page); |
1259ab75 | 3041 | } else { |
d1310b2e | 3042 | unsigned long max_nr = end_index + 1; |
7f3c74fb | 3043 | |
d1310b2e CM |
3044 | set_range_writeback(tree, cur, cur + iosize - 1); |
3045 | if (!PageWriteback(page)) { | |
d397712b CM |
3046 | printk(KERN_ERR "btrfs warning page %lu not " |
3047 | "writeback, cur %llu end %llu\n", | |
3048 | page->index, (unsigned long long)cur, | |
d1310b2e CM |
3049 | (unsigned long long)end); |
3050 | } | |
3051 | ||
ffbd517d CM |
3052 | ret = submit_extent_page(write_flags, tree, page, |
3053 | sector, iosize, pg_offset, | |
3054 | bdev, &epd->bio, max_nr, | |
c8b97818 CM |
3055 | end_bio_extent_writepage, |
3056 | 0, 0, 0); | |
d1310b2e CM |
3057 | if (ret) |
3058 | SetPageError(page); | |
3059 | } | |
3060 | cur = cur + iosize; | |
7f3c74fb | 3061 | pg_offset += iosize; |
d1310b2e CM |
3062 | nr++; |
3063 | } | |
3064 | done: | |
3065 | if (nr == 0) { | |
3066 | /* make sure the mapping tag for page dirty gets cleared */ | |
3067 | set_page_writeback(page); | |
3068 | end_page_writeback(page); | |
3069 | } | |
d1310b2e | 3070 | unlock_page(page); |
771ed689 | 3071 | |
11c8349b CM |
3072 | done_unlocked: |
3073 | ||
2c64c53d CM |
3074 | /* drop our reference on any cached states */ |
3075 | free_extent_state(cached_state); | |
d1310b2e CM |
3076 | return 0; |
3077 | } | |
3078 | ||
0b32f4bb JB |
3079 | static int eb_wait(void *word) |
3080 | { | |
3081 | io_schedule(); | |
3082 | return 0; | |
3083 | } | |
3084 | ||
3085 | static void wait_on_extent_buffer_writeback(struct extent_buffer *eb) | |
3086 | { | |
3087 | wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait, | |
3088 | TASK_UNINTERRUPTIBLE); | |
3089 | } | |
3090 | ||
3091 | static int lock_extent_buffer_for_io(struct extent_buffer *eb, | |
3092 | struct btrfs_fs_info *fs_info, | |
3093 | struct extent_page_data *epd) | |
3094 | { | |
3095 | unsigned long i, num_pages; | |
3096 | int flush = 0; | |
3097 | int ret = 0; | |
3098 | ||
3099 | if (!btrfs_try_tree_write_lock(eb)) { | |
3100 | flush = 1; | |
3101 | flush_write_bio(epd); | |
3102 | btrfs_tree_lock(eb); | |
3103 | } | |
3104 | ||
3105 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3106 | btrfs_tree_unlock(eb); | |
3107 | if (!epd->sync_io) | |
3108 | return 0; | |
3109 | if (!flush) { | |
3110 | flush_write_bio(epd); | |
3111 | flush = 1; | |
3112 | } | |
a098d8e8 CM |
3113 | while (1) { |
3114 | wait_on_extent_buffer_writeback(eb); | |
3115 | btrfs_tree_lock(eb); | |
3116 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3117 | break; | |
0b32f4bb | 3118 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3119 | } |
3120 | } | |
3121 | ||
51561ffe JB |
3122 | /* |
3123 | * We need to do this to prevent races in people who check if the eb is | |
3124 | * under IO since we can end up having no IO bits set for a short period | |
3125 | * of time. | |
3126 | */ | |
3127 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
3128 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3129 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 3130 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
3131 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
3132 | spin_lock(&fs_info->delalloc_lock); | |
3133 | if (fs_info->dirty_metadata_bytes >= eb->len) | |
3134 | fs_info->dirty_metadata_bytes -= eb->len; | |
3135 | else | |
3136 | WARN_ON(1); | |
3137 | spin_unlock(&fs_info->delalloc_lock); | |
3138 | ret = 1; | |
51561ffe JB |
3139 | } else { |
3140 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
3141 | } |
3142 | ||
3143 | btrfs_tree_unlock(eb); | |
3144 | ||
3145 | if (!ret) | |
3146 | return ret; | |
3147 | ||
3148 | num_pages = num_extent_pages(eb->start, eb->len); | |
3149 | for (i = 0; i < num_pages; i++) { | |
3150 | struct page *p = extent_buffer_page(eb, i); | |
3151 | ||
3152 | if (!trylock_page(p)) { | |
3153 | if (!flush) { | |
3154 | flush_write_bio(epd); | |
3155 | flush = 1; | |
3156 | } | |
3157 | lock_page(p); | |
3158 | } | |
3159 | } | |
3160 | ||
3161 | return ret; | |
3162 | } | |
3163 | ||
3164 | static void end_extent_buffer_writeback(struct extent_buffer *eb) | |
3165 | { | |
3166 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
3167 | smp_mb__after_clear_bit(); | |
3168 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); | |
3169 | } | |
3170 | ||
3171 | static void end_bio_extent_buffer_writepage(struct bio *bio, int err) | |
3172 | { | |
3173 | int uptodate = err == 0; | |
3174 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
3175 | struct extent_buffer *eb; | |
3176 | int done; | |
3177 | ||
3178 | do { | |
3179 | struct page *page = bvec->bv_page; | |
3180 | ||
3181 | bvec--; | |
3182 | eb = (struct extent_buffer *)page->private; | |
3183 | BUG_ON(!eb); | |
3184 | done = atomic_dec_and_test(&eb->io_pages); | |
3185 | ||
3186 | if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) { | |
3187 | set_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
3188 | ClearPageUptodate(page); | |
3189 | SetPageError(page); | |
3190 | } | |
3191 | ||
3192 | end_page_writeback(page); | |
3193 | ||
3194 | if (!done) | |
3195 | continue; | |
3196 | ||
3197 | end_extent_buffer_writeback(eb); | |
3198 | } while (bvec >= bio->bi_io_vec); | |
3199 | ||
3200 | bio_put(bio); | |
3201 | ||
3202 | } | |
3203 | ||
3204 | static int write_one_eb(struct extent_buffer *eb, | |
3205 | struct btrfs_fs_info *fs_info, | |
3206 | struct writeback_control *wbc, | |
3207 | struct extent_page_data *epd) | |
3208 | { | |
3209 | struct block_device *bdev = fs_info->fs_devices->latest_bdev; | |
3210 | u64 offset = eb->start; | |
3211 | unsigned long i, num_pages; | |
de0022b9 | 3212 | unsigned long bio_flags = 0; |
0b32f4bb | 3213 | int rw = (epd->sync_io ? WRITE_SYNC : WRITE); |
d7dbe9e7 | 3214 | int ret = 0; |
0b32f4bb JB |
3215 | |
3216 | clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
3217 | num_pages = num_extent_pages(eb->start, eb->len); | |
3218 | atomic_set(&eb->io_pages, num_pages); | |
de0022b9 JB |
3219 | if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID) |
3220 | bio_flags = EXTENT_BIO_TREE_LOG; | |
3221 | ||
0b32f4bb JB |
3222 | for (i = 0; i < num_pages; i++) { |
3223 | struct page *p = extent_buffer_page(eb, i); | |
3224 | ||
3225 | clear_page_dirty_for_io(p); | |
3226 | set_page_writeback(p); | |
3227 | ret = submit_extent_page(rw, eb->tree, p, offset >> 9, | |
3228 | PAGE_CACHE_SIZE, 0, bdev, &epd->bio, | |
3229 | -1, end_bio_extent_buffer_writepage, | |
de0022b9 JB |
3230 | 0, epd->bio_flags, bio_flags); |
3231 | epd->bio_flags = bio_flags; | |
0b32f4bb JB |
3232 | if (ret) { |
3233 | set_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
3234 | SetPageError(p); | |
3235 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) | |
3236 | end_extent_buffer_writeback(eb); | |
3237 | ret = -EIO; | |
3238 | break; | |
3239 | } | |
3240 | offset += PAGE_CACHE_SIZE; | |
3241 | update_nr_written(p, wbc, 1); | |
3242 | unlock_page(p); | |
3243 | } | |
3244 | ||
3245 | if (unlikely(ret)) { | |
3246 | for (; i < num_pages; i++) { | |
3247 | struct page *p = extent_buffer_page(eb, i); | |
3248 | unlock_page(p); | |
3249 | } | |
3250 | } | |
3251 | ||
3252 | return ret; | |
3253 | } | |
3254 | ||
3255 | int btree_write_cache_pages(struct address_space *mapping, | |
3256 | struct writeback_control *wbc) | |
3257 | { | |
3258 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; | |
3259 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; | |
3260 | struct extent_buffer *eb, *prev_eb = NULL; | |
3261 | struct extent_page_data epd = { | |
3262 | .bio = NULL, | |
3263 | .tree = tree, | |
3264 | .extent_locked = 0, | |
3265 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
de0022b9 | 3266 | .bio_flags = 0, |
0b32f4bb JB |
3267 | }; |
3268 | int ret = 0; | |
3269 | int done = 0; | |
3270 | int nr_to_write_done = 0; | |
3271 | struct pagevec pvec; | |
3272 | int nr_pages; | |
3273 | pgoff_t index; | |
3274 | pgoff_t end; /* Inclusive */ | |
3275 | int scanned = 0; | |
3276 | int tag; | |
3277 | ||
3278 | pagevec_init(&pvec, 0); | |
3279 | if (wbc->range_cyclic) { | |
3280 | index = mapping->writeback_index; /* Start from prev offset */ | |
3281 | end = -1; | |
3282 | } else { | |
3283 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
3284 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
3285 | scanned = 1; | |
3286 | } | |
3287 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3288 | tag = PAGECACHE_TAG_TOWRITE; | |
3289 | else | |
3290 | tag = PAGECACHE_TAG_DIRTY; | |
3291 | retry: | |
3292 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3293 | tag_pages_for_writeback(mapping, index, end); | |
3294 | while (!done && !nr_to_write_done && (index <= end) && | |
3295 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, | |
3296 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
3297 | unsigned i; | |
3298 | ||
3299 | scanned = 1; | |
3300 | for (i = 0; i < nr_pages; i++) { | |
3301 | struct page *page = pvec.pages[i]; | |
3302 | ||
3303 | if (!PagePrivate(page)) | |
3304 | continue; | |
3305 | ||
3306 | if (!wbc->range_cyclic && page->index > end) { | |
3307 | done = 1; | |
3308 | break; | |
3309 | } | |
3310 | ||
b5bae261 JB |
3311 | spin_lock(&mapping->private_lock); |
3312 | if (!PagePrivate(page)) { | |
3313 | spin_unlock(&mapping->private_lock); | |
3314 | continue; | |
3315 | } | |
3316 | ||
0b32f4bb | 3317 | eb = (struct extent_buffer *)page->private; |
b5bae261 JB |
3318 | |
3319 | /* | |
3320 | * Shouldn't happen and normally this would be a BUG_ON | |
3321 | * but no sense in crashing the users box for something | |
3322 | * we can survive anyway. | |
3323 | */ | |
0b32f4bb | 3324 | if (!eb) { |
b5bae261 | 3325 | spin_unlock(&mapping->private_lock); |
0b32f4bb JB |
3326 | WARN_ON(1); |
3327 | continue; | |
3328 | } | |
3329 | ||
b5bae261 JB |
3330 | if (eb == prev_eb) { |
3331 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 3332 | continue; |
b5bae261 | 3333 | } |
0b32f4bb | 3334 | |
b5bae261 JB |
3335 | ret = atomic_inc_not_zero(&eb->refs); |
3336 | spin_unlock(&mapping->private_lock); | |
3337 | if (!ret) | |
0b32f4bb | 3338 | continue; |
0b32f4bb JB |
3339 | |
3340 | prev_eb = eb; | |
3341 | ret = lock_extent_buffer_for_io(eb, fs_info, &epd); | |
3342 | if (!ret) { | |
3343 | free_extent_buffer(eb); | |
3344 | continue; | |
3345 | } | |
3346 | ||
3347 | ret = write_one_eb(eb, fs_info, wbc, &epd); | |
3348 | if (ret) { | |
3349 | done = 1; | |
3350 | free_extent_buffer(eb); | |
3351 | break; | |
3352 | } | |
3353 | free_extent_buffer(eb); | |
3354 | ||
3355 | /* | |
3356 | * the filesystem may choose to bump up nr_to_write. | |
3357 | * We have to make sure to honor the new nr_to_write | |
3358 | * at any time | |
3359 | */ | |
3360 | nr_to_write_done = wbc->nr_to_write <= 0; | |
3361 | } | |
3362 | pagevec_release(&pvec); | |
3363 | cond_resched(); | |
3364 | } | |
3365 | if (!scanned && !done) { | |
3366 | /* | |
3367 | * We hit the last page and there is more work to be done: wrap | |
3368 | * back to the start of the file | |
3369 | */ | |
3370 | scanned = 1; | |
3371 | index = 0; | |
3372 | goto retry; | |
3373 | } | |
3374 | flush_write_bio(&epd); | |
3375 | return ret; | |
3376 | } | |
3377 | ||
d1310b2e | 3378 | /** |
4bef0848 | 3379 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
3380 | * @mapping: address space structure to write |
3381 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
3382 | * @writepage: function called for each page | |
3383 | * @data: data passed to writepage function | |
3384 | * | |
3385 | * If a page is already under I/O, write_cache_pages() skips it, even | |
3386 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
3387 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
3388 | * and msync() need to guarantee that all the data which was dirty at the time | |
3389 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
3390 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
3391 | * existing IO to complete. | |
3392 | */ | |
b2950863 | 3393 | static int extent_write_cache_pages(struct extent_io_tree *tree, |
4bef0848 CM |
3394 | struct address_space *mapping, |
3395 | struct writeback_control *wbc, | |
d2c3f4f6 CM |
3396 | writepage_t writepage, void *data, |
3397 | void (*flush_fn)(void *)) | |
d1310b2e | 3398 | { |
7fd1a3f7 | 3399 | struct inode *inode = mapping->host; |
d1310b2e CM |
3400 | int ret = 0; |
3401 | int done = 0; | |
f85d7d6c | 3402 | int nr_to_write_done = 0; |
d1310b2e CM |
3403 | struct pagevec pvec; |
3404 | int nr_pages; | |
3405 | pgoff_t index; | |
3406 | pgoff_t end; /* Inclusive */ | |
3407 | int scanned = 0; | |
f7aaa06b | 3408 | int tag; |
d1310b2e | 3409 | |
7fd1a3f7 JB |
3410 | /* |
3411 | * We have to hold onto the inode so that ordered extents can do their | |
3412 | * work when the IO finishes. The alternative to this is failing to add | |
3413 | * an ordered extent if the igrab() fails there and that is a huge pain | |
3414 | * to deal with, so instead just hold onto the inode throughout the | |
3415 | * writepages operation. If it fails here we are freeing up the inode | |
3416 | * anyway and we'd rather not waste our time writing out stuff that is | |
3417 | * going to be truncated anyway. | |
3418 | */ | |
3419 | if (!igrab(inode)) | |
3420 | return 0; | |
3421 | ||
d1310b2e CM |
3422 | pagevec_init(&pvec, 0); |
3423 | if (wbc->range_cyclic) { | |
3424 | index = mapping->writeback_index; /* Start from prev offset */ | |
3425 | end = -1; | |
3426 | } else { | |
3427 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
3428 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
3429 | scanned = 1; |
3430 | } | |
f7aaa06b JB |
3431 | if (wbc->sync_mode == WB_SYNC_ALL) |
3432 | tag = PAGECACHE_TAG_TOWRITE; | |
3433 | else | |
3434 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 3435 | retry: |
f7aaa06b JB |
3436 | if (wbc->sync_mode == WB_SYNC_ALL) |
3437 | tag_pages_for_writeback(mapping, index, end); | |
f85d7d6c | 3438 | while (!done && !nr_to_write_done && (index <= end) && |
f7aaa06b JB |
3439 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
3440 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
d1310b2e CM |
3441 | unsigned i; |
3442 | ||
3443 | scanned = 1; | |
3444 | for (i = 0; i < nr_pages; i++) { | |
3445 | struct page *page = pvec.pages[i]; | |
3446 | ||
3447 | /* | |
3448 | * At this point we hold neither mapping->tree_lock nor | |
3449 | * lock on the page itself: the page may be truncated or | |
3450 | * invalidated (changing page->mapping to NULL), or even | |
3451 | * swizzled back from swapper_space to tmpfs file | |
3452 | * mapping | |
3453 | */ | |
01d658f2 CM |
3454 | if (tree->ops && |
3455 | tree->ops->write_cache_pages_lock_hook) { | |
3456 | tree->ops->write_cache_pages_lock_hook(page, | |
3457 | data, flush_fn); | |
3458 | } else { | |
3459 | if (!trylock_page(page)) { | |
3460 | flush_fn(data); | |
3461 | lock_page(page); | |
3462 | } | |
3463 | } | |
d1310b2e CM |
3464 | |
3465 | if (unlikely(page->mapping != mapping)) { | |
3466 | unlock_page(page); | |
3467 | continue; | |
3468 | } | |
3469 | ||
3470 | if (!wbc->range_cyclic && page->index > end) { | |
3471 | done = 1; | |
3472 | unlock_page(page); | |
3473 | continue; | |
3474 | } | |
3475 | ||
d2c3f4f6 | 3476 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 CM |
3477 | if (PageWriteback(page)) |
3478 | flush_fn(data); | |
d1310b2e | 3479 | wait_on_page_writeback(page); |
d2c3f4f6 | 3480 | } |
d1310b2e CM |
3481 | |
3482 | if (PageWriteback(page) || | |
3483 | !clear_page_dirty_for_io(page)) { | |
3484 | unlock_page(page); | |
3485 | continue; | |
3486 | } | |
3487 | ||
3488 | ret = (*writepage)(page, wbc, data); | |
3489 | ||
3490 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
3491 | unlock_page(page); | |
3492 | ret = 0; | |
3493 | } | |
f85d7d6c | 3494 | if (ret) |
d1310b2e | 3495 | done = 1; |
f85d7d6c CM |
3496 | |
3497 | /* | |
3498 | * the filesystem may choose to bump up nr_to_write. | |
3499 | * We have to make sure to honor the new nr_to_write | |
3500 | * at any time | |
3501 | */ | |
3502 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
3503 | } |
3504 | pagevec_release(&pvec); | |
3505 | cond_resched(); | |
3506 | } | |
3507 | if (!scanned && !done) { | |
3508 | /* | |
3509 | * We hit the last page and there is more work to be done: wrap | |
3510 | * back to the start of the file | |
3511 | */ | |
3512 | scanned = 1; | |
3513 | index = 0; | |
3514 | goto retry; | |
3515 | } | |
7fd1a3f7 | 3516 | btrfs_add_delayed_iput(inode); |
d1310b2e CM |
3517 | return ret; |
3518 | } | |
d1310b2e | 3519 | |
ffbd517d | 3520 | static void flush_epd_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 3521 | { |
d2c3f4f6 | 3522 | if (epd->bio) { |
355808c2 JM |
3523 | int rw = WRITE; |
3524 | int ret; | |
3525 | ||
ffbd517d | 3526 | if (epd->sync_io) |
355808c2 JM |
3527 | rw = WRITE_SYNC; |
3528 | ||
de0022b9 | 3529 | ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags); |
79787eaa | 3530 | BUG_ON(ret < 0); /* -ENOMEM */ |
d2c3f4f6 CM |
3531 | epd->bio = NULL; |
3532 | } | |
3533 | } | |
3534 | ||
ffbd517d CM |
3535 | static noinline void flush_write_bio(void *data) |
3536 | { | |
3537 | struct extent_page_data *epd = data; | |
3538 | flush_epd_write_bio(epd); | |
3539 | } | |
3540 | ||
d1310b2e CM |
3541 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, |
3542 | get_extent_t *get_extent, | |
3543 | struct writeback_control *wbc) | |
3544 | { | |
3545 | int ret; | |
d1310b2e CM |
3546 | struct extent_page_data epd = { |
3547 | .bio = NULL, | |
3548 | .tree = tree, | |
3549 | .get_extent = get_extent, | |
771ed689 | 3550 | .extent_locked = 0, |
ffbd517d | 3551 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
de0022b9 | 3552 | .bio_flags = 0, |
d1310b2e | 3553 | }; |
d1310b2e | 3554 | |
d1310b2e CM |
3555 | ret = __extent_writepage(page, wbc, &epd); |
3556 | ||
ffbd517d | 3557 | flush_epd_write_bio(&epd); |
d1310b2e CM |
3558 | return ret; |
3559 | } | |
d1310b2e | 3560 | |
771ed689 CM |
3561 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, |
3562 | u64 start, u64 end, get_extent_t *get_extent, | |
3563 | int mode) | |
3564 | { | |
3565 | int ret = 0; | |
3566 | struct address_space *mapping = inode->i_mapping; | |
3567 | struct page *page; | |
3568 | unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >> | |
3569 | PAGE_CACHE_SHIFT; | |
3570 | ||
3571 | struct extent_page_data epd = { | |
3572 | .bio = NULL, | |
3573 | .tree = tree, | |
3574 | .get_extent = get_extent, | |
3575 | .extent_locked = 1, | |
ffbd517d | 3576 | .sync_io = mode == WB_SYNC_ALL, |
de0022b9 | 3577 | .bio_flags = 0, |
771ed689 CM |
3578 | }; |
3579 | struct writeback_control wbc_writepages = { | |
771ed689 | 3580 | .sync_mode = mode, |
771ed689 CM |
3581 | .nr_to_write = nr_pages * 2, |
3582 | .range_start = start, | |
3583 | .range_end = end + 1, | |
3584 | }; | |
3585 | ||
d397712b | 3586 | while (start <= end) { |
771ed689 CM |
3587 | page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); |
3588 | if (clear_page_dirty_for_io(page)) | |
3589 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
3590 | else { | |
3591 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
3592 | tree->ops->writepage_end_io_hook(page, start, | |
3593 | start + PAGE_CACHE_SIZE - 1, | |
3594 | NULL, 1); | |
3595 | unlock_page(page); | |
3596 | } | |
3597 | page_cache_release(page); | |
3598 | start += PAGE_CACHE_SIZE; | |
3599 | } | |
3600 | ||
ffbd517d | 3601 | flush_epd_write_bio(&epd); |
771ed689 CM |
3602 | return ret; |
3603 | } | |
d1310b2e CM |
3604 | |
3605 | int extent_writepages(struct extent_io_tree *tree, | |
3606 | struct address_space *mapping, | |
3607 | get_extent_t *get_extent, | |
3608 | struct writeback_control *wbc) | |
3609 | { | |
3610 | int ret = 0; | |
3611 | struct extent_page_data epd = { | |
3612 | .bio = NULL, | |
3613 | .tree = tree, | |
3614 | .get_extent = get_extent, | |
771ed689 | 3615 | .extent_locked = 0, |
ffbd517d | 3616 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
de0022b9 | 3617 | .bio_flags = 0, |
d1310b2e CM |
3618 | }; |
3619 | ||
4bef0848 | 3620 | ret = extent_write_cache_pages(tree, mapping, wbc, |
d2c3f4f6 CM |
3621 | __extent_writepage, &epd, |
3622 | flush_write_bio); | |
ffbd517d | 3623 | flush_epd_write_bio(&epd); |
d1310b2e CM |
3624 | return ret; |
3625 | } | |
d1310b2e CM |
3626 | |
3627 | int extent_readpages(struct extent_io_tree *tree, | |
3628 | struct address_space *mapping, | |
3629 | struct list_head *pages, unsigned nr_pages, | |
3630 | get_extent_t get_extent) | |
3631 | { | |
3632 | struct bio *bio = NULL; | |
3633 | unsigned page_idx; | |
c8b97818 | 3634 | unsigned long bio_flags = 0; |
67c9684f LB |
3635 | struct page *pagepool[16]; |
3636 | struct page *page; | |
3637 | int i = 0; | |
3638 | int nr = 0; | |
d1310b2e | 3639 | |
d1310b2e | 3640 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
67c9684f | 3641 | page = list_entry(pages->prev, struct page, lru); |
d1310b2e CM |
3642 | |
3643 | prefetchw(&page->flags); | |
3644 | list_del(&page->lru); | |
67c9684f | 3645 | if (add_to_page_cache_lru(page, mapping, |
43e817a1 | 3646 | page->index, GFP_NOFS)) { |
67c9684f LB |
3647 | page_cache_release(page); |
3648 | continue; | |
d1310b2e | 3649 | } |
67c9684f LB |
3650 | |
3651 | pagepool[nr++] = page; | |
3652 | if (nr < ARRAY_SIZE(pagepool)) | |
3653 | continue; | |
3654 | for (i = 0; i < nr; i++) { | |
3655 | __extent_read_full_page(tree, pagepool[i], get_extent, | |
3656 | &bio, 0, &bio_flags); | |
3657 | page_cache_release(pagepool[i]); | |
3658 | } | |
3659 | nr = 0; | |
d1310b2e | 3660 | } |
67c9684f LB |
3661 | for (i = 0; i < nr; i++) { |
3662 | __extent_read_full_page(tree, pagepool[i], get_extent, | |
3663 | &bio, 0, &bio_flags); | |
3664 | page_cache_release(pagepool[i]); | |
d1310b2e | 3665 | } |
67c9684f | 3666 | |
d1310b2e CM |
3667 | BUG_ON(!list_empty(pages)); |
3668 | if (bio) | |
79787eaa | 3669 | return submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
3670 | return 0; |
3671 | } | |
d1310b2e CM |
3672 | |
3673 | /* | |
3674 | * basic invalidatepage code, this waits on any locked or writeback | |
3675 | * ranges corresponding to the page, and then deletes any extent state | |
3676 | * records from the tree | |
3677 | */ | |
3678 | int extent_invalidatepage(struct extent_io_tree *tree, | |
3679 | struct page *page, unsigned long offset) | |
3680 | { | |
2ac55d41 | 3681 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
3682 | u64 start = ((u64)page->index << PAGE_CACHE_SHIFT); |
3683 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
3684 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
3685 | ||
d397712b | 3686 | start += (offset + blocksize - 1) & ~(blocksize - 1); |
d1310b2e CM |
3687 | if (start > end) |
3688 | return 0; | |
3689 | ||
d0082371 | 3690 | lock_extent_bits(tree, start, end, 0, &cached_state); |
1edbb734 | 3691 | wait_on_page_writeback(page); |
d1310b2e | 3692 | clear_extent_bit(tree, start, end, |
32c00aff JB |
3693 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
3694 | EXTENT_DO_ACCOUNTING, | |
2ac55d41 | 3695 | 1, 1, &cached_state, GFP_NOFS); |
d1310b2e CM |
3696 | return 0; |
3697 | } | |
d1310b2e | 3698 | |
7b13b7b1 CM |
3699 | /* |
3700 | * a helper for releasepage, this tests for areas of the page that | |
3701 | * are locked or under IO and drops the related state bits if it is safe | |
3702 | * to drop the page. | |
3703 | */ | |
3704 | int try_release_extent_state(struct extent_map_tree *map, | |
3705 | struct extent_io_tree *tree, struct page *page, | |
3706 | gfp_t mask) | |
3707 | { | |
3708 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
3709 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
3710 | int ret = 1; | |
3711 | ||
211f90e6 | 3712 | if (test_range_bit(tree, start, end, |
8b62b72b | 3713 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
3714 | ret = 0; |
3715 | else { | |
3716 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
3717 | mask = GFP_NOFS; | |
11ef160f CM |
3718 | /* |
3719 | * at this point we can safely clear everything except the | |
3720 | * locked bit and the nodatasum bit | |
3721 | */ | |
e3f24cc5 | 3722 | ret = clear_extent_bit(tree, start, end, |
11ef160f CM |
3723 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
3724 | 0, 0, NULL, mask); | |
e3f24cc5 CM |
3725 | |
3726 | /* if clear_extent_bit failed for enomem reasons, | |
3727 | * we can't allow the release to continue. | |
3728 | */ | |
3729 | if (ret < 0) | |
3730 | ret = 0; | |
3731 | else | |
3732 | ret = 1; | |
7b13b7b1 CM |
3733 | } |
3734 | return ret; | |
3735 | } | |
7b13b7b1 | 3736 | |
d1310b2e CM |
3737 | /* |
3738 | * a helper for releasepage. As long as there are no locked extents | |
3739 | * in the range corresponding to the page, both state records and extent | |
3740 | * map records are removed | |
3741 | */ | |
3742 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
3743 | struct extent_io_tree *tree, struct page *page, |
3744 | gfp_t mask) | |
d1310b2e CM |
3745 | { |
3746 | struct extent_map *em; | |
3747 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
3748 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
7b13b7b1 | 3749 | |
70dec807 CM |
3750 | if ((mask & __GFP_WAIT) && |
3751 | page->mapping->host->i_size > 16 * 1024 * 1024) { | |
39b5637f | 3752 | u64 len; |
70dec807 | 3753 | while (start <= end) { |
39b5637f | 3754 | len = end - start + 1; |
890871be | 3755 | write_lock(&map->lock); |
39b5637f | 3756 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 3757 | if (!em) { |
890871be | 3758 | write_unlock(&map->lock); |
70dec807 CM |
3759 | break; |
3760 | } | |
7f3c74fb CM |
3761 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
3762 | em->start != start) { | |
890871be | 3763 | write_unlock(&map->lock); |
70dec807 CM |
3764 | free_extent_map(em); |
3765 | break; | |
3766 | } | |
3767 | if (!test_range_bit(tree, em->start, | |
3768 | extent_map_end(em) - 1, | |
8b62b72b | 3769 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 3770 | 0, NULL)) { |
70dec807 CM |
3771 | remove_extent_mapping(map, em); |
3772 | /* once for the rb tree */ | |
3773 | free_extent_map(em); | |
3774 | } | |
3775 | start = extent_map_end(em); | |
890871be | 3776 | write_unlock(&map->lock); |
70dec807 CM |
3777 | |
3778 | /* once for us */ | |
d1310b2e CM |
3779 | free_extent_map(em); |
3780 | } | |
d1310b2e | 3781 | } |
7b13b7b1 | 3782 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 3783 | } |
d1310b2e | 3784 | |
ec29ed5b CM |
3785 | /* |
3786 | * helper function for fiemap, which doesn't want to see any holes. | |
3787 | * This maps until we find something past 'last' | |
3788 | */ | |
3789 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
3790 | u64 offset, | |
3791 | u64 last, | |
3792 | get_extent_t *get_extent) | |
3793 | { | |
3794 | u64 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
3795 | struct extent_map *em; | |
3796 | u64 len; | |
3797 | ||
3798 | if (offset >= last) | |
3799 | return NULL; | |
3800 | ||
3801 | while(1) { | |
3802 | len = last - offset; | |
3803 | if (len == 0) | |
3804 | break; | |
3805 | len = (len + sectorsize - 1) & ~(sectorsize - 1); | |
3806 | em = get_extent(inode, NULL, 0, offset, len, 0); | |
c704005d | 3807 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
3808 | return em; |
3809 | ||
3810 | /* if this isn't a hole return it */ | |
3811 | if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) && | |
3812 | em->block_start != EXTENT_MAP_HOLE) { | |
3813 | return em; | |
3814 | } | |
3815 | ||
3816 | /* this is a hole, advance to the next extent */ | |
3817 | offset = extent_map_end(em); | |
3818 | free_extent_map(em); | |
3819 | if (offset >= last) | |
3820 | break; | |
3821 | } | |
3822 | return NULL; | |
3823 | } | |
3824 | ||
1506fcc8 YS |
3825 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
3826 | __u64 start, __u64 len, get_extent_t *get_extent) | |
3827 | { | |
975f84fe | 3828 | int ret = 0; |
1506fcc8 YS |
3829 | u64 off = start; |
3830 | u64 max = start + len; | |
3831 | u32 flags = 0; | |
975f84fe JB |
3832 | u32 found_type; |
3833 | u64 last; | |
ec29ed5b | 3834 | u64 last_for_get_extent = 0; |
1506fcc8 | 3835 | u64 disko = 0; |
ec29ed5b | 3836 | u64 isize = i_size_read(inode); |
975f84fe | 3837 | struct btrfs_key found_key; |
1506fcc8 | 3838 | struct extent_map *em = NULL; |
2ac55d41 | 3839 | struct extent_state *cached_state = NULL; |
975f84fe JB |
3840 | struct btrfs_path *path; |
3841 | struct btrfs_file_extent_item *item; | |
1506fcc8 | 3842 | int end = 0; |
ec29ed5b CM |
3843 | u64 em_start = 0; |
3844 | u64 em_len = 0; | |
3845 | u64 em_end = 0; | |
1506fcc8 | 3846 | unsigned long emflags; |
1506fcc8 YS |
3847 | |
3848 | if (len == 0) | |
3849 | return -EINVAL; | |
3850 | ||
975f84fe JB |
3851 | path = btrfs_alloc_path(); |
3852 | if (!path) | |
3853 | return -ENOMEM; | |
3854 | path->leave_spinning = 1; | |
3855 | ||
4d479cf0 JB |
3856 | start = ALIGN(start, BTRFS_I(inode)->root->sectorsize); |
3857 | len = ALIGN(len, BTRFS_I(inode)->root->sectorsize); | |
3858 | ||
ec29ed5b CM |
3859 | /* |
3860 | * lookup the last file extent. We're not using i_size here | |
3861 | * because there might be preallocation past i_size | |
3862 | */ | |
975f84fe | 3863 | ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root, |
33345d01 | 3864 | path, btrfs_ino(inode), -1, 0); |
975f84fe JB |
3865 | if (ret < 0) { |
3866 | btrfs_free_path(path); | |
3867 | return ret; | |
3868 | } | |
3869 | WARN_ON(!ret); | |
3870 | path->slots[0]--; | |
3871 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3872 | struct btrfs_file_extent_item); | |
3873 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
3874 | found_type = btrfs_key_type(&found_key); | |
3875 | ||
ec29ed5b | 3876 | /* No extents, but there might be delalloc bits */ |
33345d01 | 3877 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 3878 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
3879 | /* have to trust i_size as the end */ |
3880 | last = (u64)-1; | |
3881 | last_for_get_extent = isize; | |
3882 | } else { | |
3883 | /* | |
3884 | * remember the start of the last extent. There are a | |
3885 | * bunch of different factors that go into the length of the | |
3886 | * extent, so its much less complex to remember where it started | |
3887 | */ | |
3888 | last = found_key.offset; | |
3889 | last_for_get_extent = last + 1; | |
975f84fe | 3890 | } |
975f84fe JB |
3891 | btrfs_free_path(path); |
3892 | ||
ec29ed5b CM |
3893 | /* |
3894 | * we might have some extents allocated but more delalloc past those | |
3895 | * extents. so, we trust isize unless the start of the last extent is | |
3896 | * beyond isize | |
3897 | */ | |
3898 | if (last < isize) { | |
3899 | last = (u64)-1; | |
3900 | last_for_get_extent = isize; | |
3901 | } | |
3902 | ||
2ac55d41 | 3903 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0, |
d0082371 | 3904 | &cached_state); |
ec29ed5b | 3905 | |
4d479cf0 | 3906 | em = get_extent_skip_holes(inode, start, last_for_get_extent, |
ec29ed5b | 3907 | get_extent); |
1506fcc8 YS |
3908 | if (!em) |
3909 | goto out; | |
3910 | if (IS_ERR(em)) { | |
3911 | ret = PTR_ERR(em); | |
3912 | goto out; | |
3913 | } | |
975f84fe | 3914 | |
1506fcc8 | 3915 | while (!end) { |
ea8efc74 CM |
3916 | u64 offset_in_extent; |
3917 | ||
3918 | /* break if the extent we found is outside the range */ | |
3919 | if (em->start >= max || extent_map_end(em) < off) | |
3920 | break; | |
3921 | ||
3922 | /* | |
3923 | * get_extent may return an extent that starts before our | |
3924 | * requested range. We have to make sure the ranges | |
3925 | * we return to fiemap always move forward and don't | |
3926 | * overlap, so adjust the offsets here | |
3927 | */ | |
3928 | em_start = max(em->start, off); | |
1506fcc8 | 3929 | |
ea8efc74 CM |
3930 | /* |
3931 | * record the offset from the start of the extent | |
3932 | * for adjusting the disk offset below | |
3933 | */ | |
3934 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 3935 | em_end = extent_map_end(em); |
ea8efc74 | 3936 | em_len = em_end - em_start; |
ec29ed5b | 3937 | emflags = em->flags; |
1506fcc8 YS |
3938 | disko = 0; |
3939 | flags = 0; | |
3940 | ||
ea8efc74 CM |
3941 | /* |
3942 | * bump off for our next call to get_extent | |
3943 | */ | |
3944 | off = extent_map_end(em); | |
3945 | if (off >= max) | |
3946 | end = 1; | |
3947 | ||
93dbfad7 | 3948 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
3949 | end = 1; |
3950 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 3951 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
3952 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
3953 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 3954 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
3955 | flags |= (FIEMAP_EXTENT_DELALLOC | |
3956 | FIEMAP_EXTENT_UNKNOWN); | |
93dbfad7 | 3957 | } else { |
ea8efc74 | 3958 | disko = em->block_start + offset_in_extent; |
1506fcc8 YS |
3959 | } |
3960 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
3961 | flags |= FIEMAP_EXTENT_ENCODED; | |
3962 | ||
1506fcc8 YS |
3963 | free_extent_map(em); |
3964 | em = NULL; | |
ec29ed5b CM |
3965 | if ((em_start >= last) || em_len == (u64)-1 || |
3966 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
3967 | flags |= FIEMAP_EXTENT_LAST; |
3968 | end = 1; | |
3969 | } | |
3970 | ||
ec29ed5b CM |
3971 | /* now scan forward to see if this is really the last extent. */ |
3972 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
3973 | get_extent); | |
3974 | if (IS_ERR(em)) { | |
3975 | ret = PTR_ERR(em); | |
3976 | goto out; | |
3977 | } | |
3978 | if (!em) { | |
975f84fe JB |
3979 | flags |= FIEMAP_EXTENT_LAST; |
3980 | end = 1; | |
3981 | } | |
ec29ed5b CM |
3982 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, |
3983 | em_len, flags); | |
3984 | if (ret) | |
3985 | goto out_free; | |
1506fcc8 YS |
3986 | } |
3987 | out_free: | |
3988 | free_extent_map(em); | |
3989 | out: | |
2ac55d41 JB |
3990 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len, |
3991 | &cached_state, GFP_NOFS); | |
1506fcc8 YS |
3992 | return ret; |
3993 | } | |
3994 | ||
727011e0 CM |
3995 | static void __free_extent_buffer(struct extent_buffer *eb) |
3996 | { | |
3997 | #if LEAK_DEBUG | |
3998 | unsigned long flags; | |
3999 | spin_lock_irqsave(&leak_lock, flags); | |
4000 | list_del(&eb->leak_list); | |
4001 | spin_unlock_irqrestore(&leak_lock, flags); | |
4002 | #endif | |
4003 | if (eb->pages && eb->pages != eb->inline_pages) | |
4004 | kfree(eb->pages); | |
4005 | kmem_cache_free(extent_buffer_cache, eb); | |
4006 | } | |
4007 | ||
d1310b2e CM |
4008 | static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree, |
4009 | u64 start, | |
4010 | unsigned long len, | |
4011 | gfp_t mask) | |
4012 | { | |
4013 | struct extent_buffer *eb = NULL; | |
3935127c | 4014 | #if LEAK_DEBUG |
2d2ae547 | 4015 | unsigned long flags; |
4bef0848 | 4016 | #endif |
d1310b2e | 4017 | |
d1310b2e | 4018 | eb = kmem_cache_zalloc(extent_buffer_cache, mask); |
91ca338d TI |
4019 | if (eb == NULL) |
4020 | return NULL; | |
d1310b2e CM |
4021 | eb->start = start; |
4022 | eb->len = len; | |
4f2de97a | 4023 | eb->tree = tree; |
815a51c7 | 4024 | eb->bflags = 0; |
bd681513 CM |
4025 | rwlock_init(&eb->lock); |
4026 | atomic_set(&eb->write_locks, 0); | |
4027 | atomic_set(&eb->read_locks, 0); | |
4028 | atomic_set(&eb->blocking_readers, 0); | |
4029 | atomic_set(&eb->blocking_writers, 0); | |
4030 | atomic_set(&eb->spinning_readers, 0); | |
4031 | atomic_set(&eb->spinning_writers, 0); | |
5b25f70f | 4032 | eb->lock_nested = 0; |
bd681513 CM |
4033 | init_waitqueue_head(&eb->write_lock_wq); |
4034 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 4035 | |
3935127c | 4036 | #if LEAK_DEBUG |
2d2ae547 CM |
4037 | spin_lock_irqsave(&leak_lock, flags); |
4038 | list_add(&eb->leak_list, &buffers); | |
4039 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 4040 | #endif |
3083ee2e | 4041 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 4042 | atomic_set(&eb->refs, 1); |
0b32f4bb | 4043 | atomic_set(&eb->io_pages, 0); |
727011e0 CM |
4044 | |
4045 | if (len > MAX_INLINE_EXTENT_BUFFER_SIZE) { | |
4046 | struct page **pages; | |
4047 | int num_pages = (len + PAGE_CACHE_SIZE - 1) >> | |
4048 | PAGE_CACHE_SHIFT; | |
4049 | pages = kzalloc(num_pages, mask); | |
4050 | if (!pages) { | |
4051 | __free_extent_buffer(eb); | |
4052 | return NULL; | |
4053 | } | |
4054 | eb->pages = pages; | |
4055 | } else { | |
4056 | eb->pages = eb->inline_pages; | |
4057 | } | |
d1310b2e CM |
4058 | |
4059 | return eb; | |
4060 | } | |
4061 | ||
815a51c7 JS |
4062 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) |
4063 | { | |
4064 | unsigned long i; | |
4065 | struct page *p; | |
4066 | struct extent_buffer *new; | |
4067 | unsigned long num_pages = num_extent_pages(src->start, src->len); | |
4068 | ||
4069 | new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_ATOMIC); | |
4070 | if (new == NULL) | |
4071 | return NULL; | |
4072 | ||
4073 | for (i = 0; i < num_pages; i++) { | |
4074 | p = alloc_page(GFP_ATOMIC); | |
4075 | BUG_ON(!p); | |
4076 | attach_extent_buffer_page(new, p); | |
4077 | WARN_ON(PageDirty(p)); | |
4078 | SetPageUptodate(p); | |
4079 | new->pages[i] = p; | |
4080 | } | |
4081 | ||
4082 | copy_extent_buffer(new, src, 0, 0, src->len); | |
4083 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); | |
4084 | set_bit(EXTENT_BUFFER_DUMMY, &new->bflags); | |
4085 | ||
4086 | return new; | |
4087 | } | |
4088 | ||
4089 | struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len) | |
4090 | { | |
4091 | struct extent_buffer *eb; | |
4092 | unsigned long num_pages = num_extent_pages(0, len); | |
4093 | unsigned long i; | |
4094 | ||
4095 | eb = __alloc_extent_buffer(NULL, start, len, GFP_ATOMIC); | |
4096 | if (!eb) | |
4097 | return NULL; | |
4098 | ||
4099 | for (i = 0; i < num_pages; i++) { | |
4100 | eb->pages[i] = alloc_page(GFP_ATOMIC); | |
4101 | if (!eb->pages[i]) | |
4102 | goto err; | |
4103 | } | |
4104 | set_extent_buffer_uptodate(eb); | |
4105 | btrfs_set_header_nritems(eb, 0); | |
4106 | set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4107 | ||
4108 | return eb; | |
4109 | err: | |
84167d19 SB |
4110 | for (; i > 0; i--) |
4111 | __free_page(eb->pages[i - 1]); | |
815a51c7 JS |
4112 | __free_extent_buffer(eb); |
4113 | return NULL; | |
4114 | } | |
4115 | ||
0b32f4bb | 4116 | static int extent_buffer_under_io(struct extent_buffer *eb) |
d1310b2e | 4117 | { |
0b32f4bb JB |
4118 | return (atomic_read(&eb->io_pages) || |
4119 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4120 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
d1310b2e CM |
4121 | } |
4122 | ||
897ca6e9 MX |
4123 | /* |
4124 | * Helper for releasing extent buffer page. | |
4125 | */ | |
4126 | static void btrfs_release_extent_buffer_page(struct extent_buffer *eb, | |
4127 | unsigned long start_idx) | |
4128 | { | |
4129 | unsigned long index; | |
39bab87b | 4130 | unsigned long num_pages; |
897ca6e9 | 4131 | struct page *page; |
815a51c7 | 4132 | int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); |
897ca6e9 | 4133 | |
0b32f4bb | 4134 | BUG_ON(extent_buffer_under_io(eb)); |
897ca6e9 | 4135 | |
39bab87b WSH |
4136 | num_pages = num_extent_pages(eb->start, eb->len); |
4137 | index = start_idx + num_pages; | |
897ca6e9 MX |
4138 | if (start_idx >= index) |
4139 | return; | |
4140 | ||
4141 | do { | |
4142 | index--; | |
4143 | page = extent_buffer_page(eb, index); | |
815a51c7 | 4144 | if (page && mapped) { |
4f2de97a JB |
4145 | spin_lock(&page->mapping->private_lock); |
4146 | /* | |
4147 | * We do this since we'll remove the pages after we've | |
4148 | * removed the eb from the radix tree, so we could race | |
4149 | * and have this page now attached to the new eb. So | |
4150 | * only clear page_private if it's still connected to | |
4151 | * this eb. | |
4152 | */ | |
4153 | if (PagePrivate(page) && | |
4154 | page->private == (unsigned long)eb) { | |
0b32f4bb | 4155 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); |
3083ee2e JB |
4156 | BUG_ON(PageDirty(page)); |
4157 | BUG_ON(PageWriteback(page)); | |
4f2de97a JB |
4158 | /* |
4159 | * We need to make sure we haven't be attached | |
4160 | * to a new eb. | |
4161 | */ | |
4162 | ClearPagePrivate(page); | |
4163 | set_page_private(page, 0); | |
4164 | /* One for the page private */ | |
4165 | page_cache_release(page); | |
4166 | } | |
4167 | spin_unlock(&page->mapping->private_lock); | |
4168 | ||
815a51c7 JS |
4169 | } |
4170 | if (page) { | |
4f2de97a | 4171 | /* One for when we alloced the page */ |
897ca6e9 | 4172 | page_cache_release(page); |
4f2de97a | 4173 | } |
897ca6e9 MX |
4174 | } while (index != start_idx); |
4175 | } | |
4176 | ||
4177 | /* | |
4178 | * Helper for releasing the extent buffer. | |
4179 | */ | |
4180 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4181 | { | |
4182 | btrfs_release_extent_buffer_page(eb, 0); | |
4183 | __free_extent_buffer(eb); | |
4184 | } | |
4185 | ||
0b32f4bb JB |
4186 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
4187 | { | |
4188 | /* the ref bit is tricky. We have to make sure it is set | |
4189 | * if we have the buffer dirty. Otherwise the | |
4190 | * code to free a buffer can end up dropping a dirty | |
4191 | * page | |
4192 | * | |
4193 | * Once the ref bit is set, it won't go away while the | |
4194 | * buffer is dirty or in writeback, and it also won't | |
4195 | * go away while we have the reference count on the | |
4196 | * eb bumped. | |
4197 | * | |
4198 | * We can't just set the ref bit without bumping the | |
4199 | * ref on the eb because free_extent_buffer might | |
4200 | * see the ref bit and try to clear it. If this happens | |
4201 | * free_extent_buffer might end up dropping our original | |
4202 | * ref by mistake and freeing the page before we are able | |
4203 | * to add one more ref. | |
4204 | * | |
4205 | * So bump the ref count first, then set the bit. If someone | |
4206 | * beat us to it, drop the ref we added. | |
4207 | */ | |
594831c4 JB |
4208 | spin_lock(&eb->refs_lock); |
4209 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 4210 | atomic_inc(&eb->refs); |
594831c4 | 4211 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
4212 | } |
4213 | ||
5df4235e JB |
4214 | static void mark_extent_buffer_accessed(struct extent_buffer *eb) |
4215 | { | |
4216 | unsigned long num_pages, i; | |
4217 | ||
0b32f4bb JB |
4218 | check_buffer_tree_ref(eb); |
4219 | ||
5df4235e JB |
4220 | num_pages = num_extent_pages(eb->start, eb->len); |
4221 | for (i = 0; i < num_pages; i++) { | |
4222 | struct page *p = extent_buffer_page(eb, i); | |
4223 | mark_page_accessed(p); | |
4224 | } | |
4225 | } | |
4226 | ||
d1310b2e | 4227 | struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree, |
727011e0 | 4228 | u64 start, unsigned long len) |
d1310b2e CM |
4229 | { |
4230 | unsigned long num_pages = num_extent_pages(start, len); | |
4231 | unsigned long i; | |
4232 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
4233 | struct extent_buffer *eb; | |
6af118ce | 4234 | struct extent_buffer *exists = NULL; |
d1310b2e CM |
4235 | struct page *p; |
4236 | struct address_space *mapping = tree->mapping; | |
4237 | int uptodate = 1; | |
19fe0a8b | 4238 | int ret; |
d1310b2e | 4239 | |
19fe0a8b MX |
4240 | rcu_read_lock(); |
4241 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); | |
4242 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
4243 | rcu_read_unlock(); | |
5df4235e | 4244 | mark_extent_buffer_accessed(eb); |
6af118ce CM |
4245 | return eb; |
4246 | } | |
19fe0a8b | 4247 | rcu_read_unlock(); |
6af118ce | 4248 | |
ba144192 | 4249 | eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS); |
2b114d1d | 4250 | if (!eb) |
d1310b2e CM |
4251 | return NULL; |
4252 | ||
727011e0 | 4253 | for (i = 0; i < num_pages; i++, index++) { |
a6591715 | 4254 | p = find_or_create_page(mapping, index, GFP_NOFS); |
4804b382 | 4255 | if (!p) |
6af118ce | 4256 | goto free_eb; |
4f2de97a JB |
4257 | |
4258 | spin_lock(&mapping->private_lock); | |
4259 | if (PagePrivate(p)) { | |
4260 | /* | |
4261 | * We could have already allocated an eb for this page | |
4262 | * and attached one so lets see if we can get a ref on | |
4263 | * the existing eb, and if we can we know it's good and | |
4264 | * we can just return that one, else we know we can just | |
4265 | * overwrite page->private. | |
4266 | */ | |
4267 | exists = (struct extent_buffer *)p->private; | |
4268 | if (atomic_inc_not_zero(&exists->refs)) { | |
4269 | spin_unlock(&mapping->private_lock); | |
4270 | unlock_page(p); | |
17de39ac | 4271 | page_cache_release(p); |
5df4235e | 4272 | mark_extent_buffer_accessed(exists); |
4f2de97a JB |
4273 | goto free_eb; |
4274 | } | |
4275 | ||
0b32f4bb | 4276 | /* |
4f2de97a JB |
4277 | * Do this so attach doesn't complain and we need to |
4278 | * drop the ref the old guy had. | |
4279 | */ | |
4280 | ClearPagePrivate(p); | |
0b32f4bb | 4281 | WARN_ON(PageDirty(p)); |
4f2de97a | 4282 | page_cache_release(p); |
d1310b2e | 4283 | } |
4f2de97a JB |
4284 | attach_extent_buffer_page(eb, p); |
4285 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 4286 | WARN_ON(PageDirty(p)); |
d1310b2e | 4287 | mark_page_accessed(p); |
727011e0 | 4288 | eb->pages[i] = p; |
d1310b2e CM |
4289 | if (!PageUptodate(p)) |
4290 | uptodate = 0; | |
eb14ab8e CM |
4291 | |
4292 | /* | |
4293 | * see below about how we avoid a nasty race with release page | |
4294 | * and why we unlock later | |
4295 | */ | |
d1310b2e CM |
4296 | } |
4297 | if (uptodate) | |
b4ce94de | 4298 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 4299 | again: |
19fe0a8b MX |
4300 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); |
4301 | if (ret) | |
4302 | goto free_eb; | |
4303 | ||
6af118ce | 4304 | spin_lock(&tree->buffer_lock); |
19fe0a8b MX |
4305 | ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb); |
4306 | if (ret == -EEXIST) { | |
4307 | exists = radix_tree_lookup(&tree->buffer, | |
4308 | start >> PAGE_CACHE_SHIFT); | |
115391d2 JB |
4309 | if (!atomic_inc_not_zero(&exists->refs)) { |
4310 | spin_unlock(&tree->buffer_lock); | |
4311 | radix_tree_preload_end(); | |
115391d2 JB |
4312 | exists = NULL; |
4313 | goto again; | |
4314 | } | |
6af118ce | 4315 | spin_unlock(&tree->buffer_lock); |
19fe0a8b | 4316 | radix_tree_preload_end(); |
5df4235e | 4317 | mark_extent_buffer_accessed(exists); |
6af118ce CM |
4318 | goto free_eb; |
4319 | } | |
6af118ce | 4320 | /* add one reference for the tree */ |
0b32f4bb | 4321 | check_buffer_tree_ref(eb); |
f044ba78 | 4322 | spin_unlock(&tree->buffer_lock); |
19fe0a8b | 4323 | radix_tree_preload_end(); |
eb14ab8e CM |
4324 | |
4325 | /* | |
4326 | * there is a race where release page may have | |
4327 | * tried to find this extent buffer in the radix | |
4328 | * but failed. It will tell the VM it is safe to | |
4329 | * reclaim the, and it will clear the page private bit. | |
4330 | * We must make sure to set the page private bit properly | |
4331 | * after the extent buffer is in the radix tree so | |
4332 | * it doesn't get lost | |
4333 | */ | |
727011e0 CM |
4334 | SetPageChecked(eb->pages[0]); |
4335 | for (i = 1; i < num_pages; i++) { | |
4336 | p = extent_buffer_page(eb, i); | |
727011e0 CM |
4337 | ClearPageChecked(p); |
4338 | unlock_page(p); | |
4339 | } | |
4340 | unlock_page(eb->pages[0]); | |
d1310b2e CM |
4341 | return eb; |
4342 | ||
6af118ce | 4343 | free_eb: |
727011e0 CM |
4344 | for (i = 0; i < num_pages; i++) { |
4345 | if (eb->pages[i]) | |
4346 | unlock_page(eb->pages[i]); | |
4347 | } | |
eb14ab8e | 4348 | |
17de39ac | 4349 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
897ca6e9 | 4350 | btrfs_release_extent_buffer(eb); |
6af118ce | 4351 | return exists; |
d1310b2e | 4352 | } |
d1310b2e CM |
4353 | |
4354 | struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree, | |
f09d1f60 | 4355 | u64 start, unsigned long len) |
d1310b2e | 4356 | { |
d1310b2e | 4357 | struct extent_buffer *eb; |
d1310b2e | 4358 | |
19fe0a8b MX |
4359 | rcu_read_lock(); |
4360 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); | |
4361 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
4362 | rcu_read_unlock(); | |
5df4235e | 4363 | mark_extent_buffer_accessed(eb); |
19fe0a8b MX |
4364 | return eb; |
4365 | } | |
4366 | rcu_read_unlock(); | |
0f9dd46c | 4367 | |
19fe0a8b | 4368 | return NULL; |
d1310b2e | 4369 | } |
d1310b2e | 4370 | |
3083ee2e JB |
4371 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
4372 | { | |
4373 | struct extent_buffer *eb = | |
4374 | container_of(head, struct extent_buffer, rcu_head); | |
4375 | ||
4376 | __free_extent_buffer(eb); | |
4377 | } | |
4378 | ||
3083ee2e | 4379 | /* Expects to have eb->eb_lock already held */ |
e64860aa | 4380 | static int release_extent_buffer(struct extent_buffer *eb, gfp_t mask) |
3083ee2e JB |
4381 | { |
4382 | WARN_ON(atomic_read(&eb->refs) == 0); | |
4383 | if (atomic_dec_and_test(&eb->refs)) { | |
815a51c7 JS |
4384 | if (test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) { |
4385 | spin_unlock(&eb->refs_lock); | |
4386 | } else { | |
4387 | struct extent_io_tree *tree = eb->tree; | |
3083ee2e | 4388 | |
815a51c7 | 4389 | spin_unlock(&eb->refs_lock); |
3083ee2e | 4390 | |
815a51c7 JS |
4391 | spin_lock(&tree->buffer_lock); |
4392 | radix_tree_delete(&tree->buffer, | |
4393 | eb->start >> PAGE_CACHE_SHIFT); | |
4394 | spin_unlock(&tree->buffer_lock); | |
4395 | } | |
3083ee2e JB |
4396 | |
4397 | /* Should be safe to release our pages at this point */ | |
4398 | btrfs_release_extent_buffer_page(eb, 0); | |
3083ee2e | 4399 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 4400 | return 1; |
3083ee2e JB |
4401 | } |
4402 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
4403 | |
4404 | return 0; | |
3083ee2e JB |
4405 | } |
4406 | ||
d1310b2e CM |
4407 | void free_extent_buffer(struct extent_buffer *eb) |
4408 | { | |
d1310b2e CM |
4409 | if (!eb) |
4410 | return; | |
4411 | ||
3083ee2e | 4412 | spin_lock(&eb->refs_lock); |
815a51c7 JS |
4413 | if (atomic_read(&eb->refs) == 2 && |
4414 | test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) | |
4415 | atomic_dec(&eb->refs); | |
4416 | ||
3083ee2e JB |
4417 | if (atomic_read(&eb->refs) == 2 && |
4418 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 4419 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
4420 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
4421 | atomic_dec(&eb->refs); | |
4422 | ||
4423 | /* | |
4424 | * I know this is terrible, but it's temporary until we stop tracking | |
4425 | * the uptodate bits and such for the extent buffers. | |
4426 | */ | |
4427 | release_extent_buffer(eb, GFP_ATOMIC); | |
4428 | } | |
4429 | ||
4430 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
4431 | { | |
4432 | if (!eb) | |
d1310b2e CM |
4433 | return; |
4434 | ||
3083ee2e JB |
4435 | spin_lock(&eb->refs_lock); |
4436 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
4437 | ||
0b32f4bb | 4438 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
4439 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
4440 | atomic_dec(&eb->refs); | |
4441 | release_extent_buffer(eb, GFP_NOFS); | |
d1310b2e | 4442 | } |
d1310b2e | 4443 | |
1d4284bd | 4444 | void clear_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 4445 | { |
d1310b2e CM |
4446 | unsigned long i; |
4447 | unsigned long num_pages; | |
4448 | struct page *page; | |
4449 | ||
d1310b2e CM |
4450 | num_pages = num_extent_pages(eb->start, eb->len); |
4451 | ||
4452 | for (i = 0; i < num_pages; i++) { | |
4453 | page = extent_buffer_page(eb, i); | |
b9473439 | 4454 | if (!PageDirty(page)) |
d2c3f4f6 CM |
4455 | continue; |
4456 | ||
a61e6f29 | 4457 | lock_page(page); |
eb14ab8e CM |
4458 | WARN_ON(!PagePrivate(page)); |
4459 | ||
d1310b2e | 4460 | clear_page_dirty_for_io(page); |
0ee0fda0 | 4461 | spin_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
4462 | if (!PageDirty(page)) { |
4463 | radix_tree_tag_clear(&page->mapping->page_tree, | |
4464 | page_index(page), | |
4465 | PAGECACHE_TAG_DIRTY); | |
4466 | } | |
0ee0fda0 | 4467 | spin_unlock_irq(&page->mapping->tree_lock); |
bf0da8c1 | 4468 | ClearPageError(page); |
a61e6f29 | 4469 | unlock_page(page); |
d1310b2e | 4470 | } |
0b32f4bb | 4471 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 4472 | } |
d1310b2e | 4473 | |
0b32f4bb | 4474 | int set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e CM |
4475 | { |
4476 | unsigned long i; | |
4477 | unsigned long num_pages; | |
b9473439 | 4478 | int was_dirty = 0; |
d1310b2e | 4479 | |
0b32f4bb JB |
4480 | check_buffer_tree_ref(eb); |
4481 | ||
b9473439 | 4482 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 4483 | |
d1310b2e | 4484 | num_pages = num_extent_pages(eb->start, eb->len); |
3083ee2e | 4485 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
4486 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
4487 | ||
b9473439 | 4488 | for (i = 0; i < num_pages; i++) |
0b32f4bb | 4489 | set_page_dirty(extent_buffer_page(eb, i)); |
b9473439 | 4490 | return was_dirty; |
d1310b2e | 4491 | } |
d1310b2e | 4492 | |
0b32f4bb | 4493 | static int range_straddles_pages(u64 start, u64 len) |
19b6caf4 CM |
4494 | { |
4495 | if (len < PAGE_CACHE_SIZE) | |
4496 | return 1; | |
4497 | if (start & (PAGE_CACHE_SIZE - 1)) | |
4498 | return 1; | |
4499 | if ((start + len) & (PAGE_CACHE_SIZE - 1)) | |
4500 | return 1; | |
4501 | return 0; | |
4502 | } | |
4503 | ||
0b32f4bb | 4504 | int clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 CM |
4505 | { |
4506 | unsigned long i; | |
4507 | struct page *page; | |
4508 | unsigned long num_pages; | |
4509 | ||
b4ce94de | 4510 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
0b32f4bb | 4511 | num_pages = num_extent_pages(eb->start, eb->len); |
1259ab75 CM |
4512 | for (i = 0; i < num_pages; i++) { |
4513 | page = extent_buffer_page(eb, i); | |
33958dc6 CM |
4514 | if (page) |
4515 | ClearPageUptodate(page); | |
1259ab75 CM |
4516 | } |
4517 | return 0; | |
4518 | } | |
4519 | ||
0b32f4bb | 4520 | int set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e CM |
4521 | { |
4522 | unsigned long i; | |
4523 | struct page *page; | |
4524 | unsigned long num_pages; | |
4525 | ||
0b32f4bb | 4526 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 4527 | num_pages = num_extent_pages(eb->start, eb->len); |
d1310b2e CM |
4528 | for (i = 0; i < num_pages; i++) { |
4529 | page = extent_buffer_page(eb, i); | |
d1310b2e CM |
4530 | SetPageUptodate(page); |
4531 | } | |
4532 | return 0; | |
4533 | } | |
d1310b2e | 4534 | |
ce9adaa5 CM |
4535 | int extent_range_uptodate(struct extent_io_tree *tree, |
4536 | u64 start, u64 end) | |
4537 | { | |
4538 | struct page *page; | |
4539 | int ret; | |
4540 | int pg_uptodate = 1; | |
4541 | int uptodate; | |
4542 | unsigned long index; | |
4543 | ||
0b32f4bb | 4544 | if (range_straddles_pages(start, end - start + 1)) { |
19b6caf4 CM |
4545 | ret = test_range_bit(tree, start, end, |
4546 | EXTENT_UPTODATE, 1, NULL); | |
4547 | if (ret) | |
4548 | return 1; | |
4549 | } | |
d397712b | 4550 | while (start <= end) { |
ce9adaa5 CM |
4551 | index = start >> PAGE_CACHE_SHIFT; |
4552 | page = find_get_page(tree->mapping, index); | |
8bedd51b MH |
4553 | if (!page) |
4554 | return 1; | |
ce9adaa5 CM |
4555 | uptodate = PageUptodate(page); |
4556 | page_cache_release(page); | |
4557 | if (!uptodate) { | |
4558 | pg_uptodate = 0; | |
4559 | break; | |
4560 | } | |
4561 | start += PAGE_CACHE_SIZE; | |
4562 | } | |
4563 | return pg_uptodate; | |
4564 | } | |
4565 | ||
0b32f4bb | 4566 | int extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 4567 | { |
0b32f4bb | 4568 | return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 4569 | } |
d1310b2e CM |
4570 | |
4571 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
bb82ab88 | 4572 | struct extent_buffer *eb, u64 start, int wait, |
f188591e | 4573 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
4574 | { |
4575 | unsigned long i; | |
4576 | unsigned long start_i; | |
4577 | struct page *page; | |
4578 | int err; | |
4579 | int ret = 0; | |
ce9adaa5 CM |
4580 | int locked_pages = 0; |
4581 | int all_uptodate = 1; | |
d1310b2e | 4582 | unsigned long num_pages; |
727011e0 | 4583 | unsigned long num_reads = 0; |
a86c12c7 | 4584 | struct bio *bio = NULL; |
c8b97818 | 4585 | unsigned long bio_flags = 0; |
a86c12c7 | 4586 | |
b4ce94de | 4587 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
4588 | return 0; |
4589 | ||
d1310b2e CM |
4590 | if (start) { |
4591 | WARN_ON(start < eb->start); | |
4592 | start_i = (start >> PAGE_CACHE_SHIFT) - | |
4593 | (eb->start >> PAGE_CACHE_SHIFT); | |
4594 | } else { | |
4595 | start_i = 0; | |
4596 | } | |
4597 | ||
4598 | num_pages = num_extent_pages(eb->start, eb->len); | |
4599 | for (i = start_i; i < num_pages; i++) { | |
4600 | page = extent_buffer_page(eb, i); | |
bb82ab88 | 4601 | if (wait == WAIT_NONE) { |
2db04966 | 4602 | if (!trylock_page(page)) |
ce9adaa5 | 4603 | goto unlock_exit; |
d1310b2e CM |
4604 | } else { |
4605 | lock_page(page); | |
4606 | } | |
ce9adaa5 | 4607 | locked_pages++; |
727011e0 CM |
4608 | if (!PageUptodate(page)) { |
4609 | num_reads++; | |
ce9adaa5 | 4610 | all_uptodate = 0; |
727011e0 | 4611 | } |
ce9adaa5 CM |
4612 | } |
4613 | if (all_uptodate) { | |
4614 | if (start_i == 0) | |
b4ce94de | 4615 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
4616 | goto unlock_exit; |
4617 | } | |
4618 | ||
ea466794 | 4619 | clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags); |
5cf1ab56 | 4620 | eb->read_mirror = 0; |
0b32f4bb | 4621 | atomic_set(&eb->io_pages, num_reads); |
ce9adaa5 CM |
4622 | for (i = start_i; i < num_pages; i++) { |
4623 | page = extent_buffer_page(eb, i); | |
ce9adaa5 | 4624 | if (!PageUptodate(page)) { |
f188591e | 4625 | ClearPageError(page); |
a86c12c7 | 4626 | err = __extent_read_full_page(tree, page, |
f188591e | 4627 | get_extent, &bio, |
c8b97818 | 4628 | mirror_num, &bio_flags); |
d397712b | 4629 | if (err) |
d1310b2e | 4630 | ret = err; |
d1310b2e CM |
4631 | } else { |
4632 | unlock_page(page); | |
4633 | } | |
4634 | } | |
4635 | ||
355808c2 JM |
4636 | if (bio) { |
4637 | err = submit_one_bio(READ, bio, mirror_num, bio_flags); | |
79787eaa JM |
4638 | if (err) |
4639 | return err; | |
355808c2 | 4640 | } |
a86c12c7 | 4641 | |
bb82ab88 | 4642 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 4643 | return ret; |
d397712b | 4644 | |
d1310b2e CM |
4645 | for (i = start_i; i < num_pages; i++) { |
4646 | page = extent_buffer_page(eb, i); | |
4647 | wait_on_page_locked(page); | |
d397712b | 4648 | if (!PageUptodate(page)) |
d1310b2e | 4649 | ret = -EIO; |
d1310b2e | 4650 | } |
d397712b | 4651 | |
d1310b2e | 4652 | return ret; |
ce9adaa5 CM |
4653 | |
4654 | unlock_exit: | |
4655 | i = start_i; | |
d397712b | 4656 | while (locked_pages > 0) { |
ce9adaa5 CM |
4657 | page = extent_buffer_page(eb, i); |
4658 | i++; | |
4659 | unlock_page(page); | |
4660 | locked_pages--; | |
4661 | } | |
4662 | return ret; | |
d1310b2e | 4663 | } |
d1310b2e CM |
4664 | |
4665 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
4666 | unsigned long start, | |
4667 | unsigned long len) | |
4668 | { | |
4669 | size_t cur; | |
4670 | size_t offset; | |
4671 | struct page *page; | |
4672 | char *kaddr; | |
4673 | char *dst = (char *)dstv; | |
4674 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4675 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
4676 | |
4677 | WARN_ON(start > eb->len); | |
4678 | WARN_ON(start + len > eb->start + eb->len); | |
4679 | ||
4680 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4681 | ||
d397712b | 4682 | while (len > 0) { |
d1310b2e | 4683 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
4684 | |
4685 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
a6591715 | 4686 | kaddr = page_address(page); |
d1310b2e | 4687 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
4688 | |
4689 | dst += cur; | |
4690 | len -= cur; | |
4691 | offset = 0; | |
4692 | i++; | |
4693 | } | |
4694 | } | |
d1310b2e CM |
4695 | |
4696 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
a6591715 | 4697 | unsigned long min_len, char **map, |
d1310b2e | 4698 | unsigned long *map_start, |
a6591715 | 4699 | unsigned long *map_len) |
d1310b2e CM |
4700 | { |
4701 | size_t offset = start & (PAGE_CACHE_SIZE - 1); | |
4702 | char *kaddr; | |
4703 | struct page *p; | |
4704 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4705 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4706 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
4707 | PAGE_CACHE_SHIFT; | |
4708 | ||
4709 | if (i != end_i) | |
4710 | return -EINVAL; | |
4711 | ||
4712 | if (i == 0) { | |
4713 | offset = start_offset; | |
4714 | *map_start = 0; | |
4715 | } else { | |
4716 | offset = 0; | |
4717 | *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; | |
4718 | } | |
d397712b | 4719 | |
d1310b2e | 4720 | if (start + min_len > eb->len) { |
31b1a2bd | 4721 | WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, " |
d397712b CM |
4722 | "wanted %lu %lu\n", (unsigned long long)eb->start, |
4723 | eb->len, start, min_len); | |
85026533 | 4724 | return -EINVAL; |
d1310b2e CM |
4725 | } |
4726 | ||
4727 | p = extent_buffer_page(eb, i); | |
a6591715 | 4728 | kaddr = page_address(p); |
d1310b2e CM |
4729 | *map = kaddr + offset; |
4730 | *map_len = PAGE_CACHE_SIZE - offset; | |
4731 | return 0; | |
4732 | } | |
d1310b2e | 4733 | |
d1310b2e CM |
4734 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, |
4735 | unsigned long start, | |
4736 | unsigned long len) | |
4737 | { | |
4738 | size_t cur; | |
4739 | size_t offset; | |
4740 | struct page *page; | |
4741 | char *kaddr; | |
4742 | char *ptr = (char *)ptrv; | |
4743 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4744 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4745 | int ret = 0; | |
4746 | ||
4747 | WARN_ON(start > eb->len); | |
4748 | WARN_ON(start + len > eb->start + eb->len); | |
4749 | ||
4750 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4751 | ||
d397712b | 4752 | while (len > 0) { |
d1310b2e | 4753 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
4754 | |
4755 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
4756 | ||
a6591715 | 4757 | kaddr = page_address(page); |
d1310b2e | 4758 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
4759 | if (ret) |
4760 | break; | |
4761 | ||
4762 | ptr += cur; | |
4763 | len -= cur; | |
4764 | offset = 0; | |
4765 | i++; | |
4766 | } | |
4767 | return ret; | |
4768 | } | |
d1310b2e CM |
4769 | |
4770 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
4771 | unsigned long start, unsigned long len) | |
4772 | { | |
4773 | size_t cur; | |
4774 | size_t offset; | |
4775 | struct page *page; | |
4776 | char *kaddr; | |
4777 | char *src = (char *)srcv; | |
4778 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4779 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4780 | ||
4781 | WARN_ON(start > eb->len); | |
4782 | WARN_ON(start + len > eb->start + eb->len); | |
4783 | ||
4784 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4785 | ||
d397712b | 4786 | while (len > 0) { |
d1310b2e CM |
4787 | page = extent_buffer_page(eb, i); |
4788 | WARN_ON(!PageUptodate(page)); | |
4789 | ||
4790 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 4791 | kaddr = page_address(page); |
d1310b2e | 4792 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
4793 | |
4794 | src += cur; | |
4795 | len -= cur; | |
4796 | offset = 0; | |
4797 | i++; | |
4798 | } | |
4799 | } | |
d1310b2e CM |
4800 | |
4801 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
4802 | unsigned long start, unsigned long len) | |
4803 | { | |
4804 | size_t cur; | |
4805 | size_t offset; | |
4806 | struct page *page; | |
4807 | char *kaddr; | |
4808 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4809 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4810 | ||
4811 | WARN_ON(start > eb->len); | |
4812 | WARN_ON(start + len > eb->start + eb->len); | |
4813 | ||
4814 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4815 | ||
d397712b | 4816 | while (len > 0) { |
d1310b2e CM |
4817 | page = extent_buffer_page(eb, i); |
4818 | WARN_ON(!PageUptodate(page)); | |
4819 | ||
4820 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 4821 | kaddr = page_address(page); |
d1310b2e | 4822 | memset(kaddr + offset, c, cur); |
d1310b2e CM |
4823 | |
4824 | len -= cur; | |
4825 | offset = 0; | |
4826 | i++; | |
4827 | } | |
4828 | } | |
d1310b2e CM |
4829 | |
4830 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
4831 | unsigned long dst_offset, unsigned long src_offset, | |
4832 | unsigned long len) | |
4833 | { | |
4834 | u64 dst_len = dst->len; | |
4835 | size_t cur; | |
4836 | size_t offset; | |
4837 | struct page *page; | |
4838 | char *kaddr; | |
4839 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4840 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
4841 | ||
4842 | WARN_ON(src->len != dst_len); | |
4843 | ||
4844 | offset = (start_offset + dst_offset) & | |
4845 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4846 | ||
d397712b | 4847 | while (len > 0) { |
d1310b2e CM |
4848 | page = extent_buffer_page(dst, i); |
4849 | WARN_ON(!PageUptodate(page)); | |
4850 | ||
4851 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
4852 | ||
a6591715 | 4853 | kaddr = page_address(page); |
d1310b2e | 4854 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
4855 | |
4856 | src_offset += cur; | |
4857 | len -= cur; | |
4858 | offset = 0; | |
4859 | i++; | |
4860 | } | |
4861 | } | |
d1310b2e CM |
4862 | |
4863 | static void move_pages(struct page *dst_page, struct page *src_page, | |
4864 | unsigned long dst_off, unsigned long src_off, | |
4865 | unsigned long len) | |
4866 | { | |
a6591715 | 4867 | char *dst_kaddr = page_address(dst_page); |
d1310b2e CM |
4868 | if (dst_page == src_page) { |
4869 | memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len); | |
4870 | } else { | |
a6591715 | 4871 | char *src_kaddr = page_address(src_page); |
d1310b2e CM |
4872 | char *p = dst_kaddr + dst_off + len; |
4873 | char *s = src_kaddr + src_off + len; | |
4874 | ||
4875 | while (len--) | |
4876 | *--p = *--s; | |
d1310b2e | 4877 | } |
d1310b2e CM |
4878 | } |
4879 | ||
3387206f ST |
4880 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
4881 | { | |
4882 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
4883 | return distance < len; | |
4884 | } | |
4885 | ||
d1310b2e CM |
4886 | static void copy_pages(struct page *dst_page, struct page *src_page, |
4887 | unsigned long dst_off, unsigned long src_off, | |
4888 | unsigned long len) | |
4889 | { | |
a6591715 | 4890 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 4891 | char *src_kaddr; |
727011e0 | 4892 | int must_memmove = 0; |
d1310b2e | 4893 | |
3387206f | 4894 | if (dst_page != src_page) { |
a6591715 | 4895 | src_kaddr = page_address(src_page); |
3387206f | 4896 | } else { |
d1310b2e | 4897 | src_kaddr = dst_kaddr; |
727011e0 CM |
4898 | if (areas_overlap(src_off, dst_off, len)) |
4899 | must_memmove = 1; | |
3387206f | 4900 | } |
d1310b2e | 4901 | |
727011e0 CM |
4902 | if (must_memmove) |
4903 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
4904 | else | |
4905 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
4906 | } |
4907 | ||
4908 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
4909 | unsigned long src_offset, unsigned long len) | |
4910 | { | |
4911 | size_t cur; | |
4912 | size_t dst_off_in_page; | |
4913 | size_t src_off_in_page; | |
4914 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4915 | unsigned long dst_i; | |
4916 | unsigned long src_i; | |
4917 | ||
4918 | if (src_offset + len > dst->len) { | |
d397712b CM |
4919 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
4920 | "len %lu dst len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
4921 | BUG_ON(1); |
4922 | } | |
4923 | if (dst_offset + len > dst->len) { | |
d397712b CM |
4924 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
4925 | "len %lu dst len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
4926 | BUG_ON(1); |
4927 | } | |
4928 | ||
d397712b | 4929 | while (len > 0) { |
d1310b2e CM |
4930 | dst_off_in_page = (start_offset + dst_offset) & |
4931 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4932 | src_off_in_page = (start_offset + src_offset) & | |
4933 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4934 | ||
4935 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
4936 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
4937 | ||
4938 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
4939 | src_off_in_page)); | |
4940 | cur = min_t(unsigned long, cur, | |
4941 | (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); | |
4942 | ||
4943 | copy_pages(extent_buffer_page(dst, dst_i), | |
4944 | extent_buffer_page(dst, src_i), | |
4945 | dst_off_in_page, src_off_in_page, cur); | |
4946 | ||
4947 | src_offset += cur; | |
4948 | dst_offset += cur; | |
4949 | len -= cur; | |
4950 | } | |
4951 | } | |
d1310b2e CM |
4952 | |
4953 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
4954 | unsigned long src_offset, unsigned long len) | |
4955 | { | |
4956 | size_t cur; | |
4957 | size_t dst_off_in_page; | |
4958 | size_t src_off_in_page; | |
4959 | unsigned long dst_end = dst_offset + len - 1; | |
4960 | unsigned long src_end = src_offset + len - 1; | |
4961 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4962 | unsigned long dst_i; | |
4963 | unsigned long src_i; | |
4964 | ||
4965 | if (src_offset + len > dst->len) { | |
d397712b CM |
4966 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
4967 | "len %lu len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
4968 | BUG_ON(1); |
4969 | } | |
4970 | if (dst_offset + len > dst->len) { | |
d397712b CM |
4971 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
4972 | "len %lu len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
4973 | BUG_ON(1); |
4974 | } | |
727011e0 | 4975 | if (dst_offset < src_offset) { |
d1310b2e CM |
4976 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
4977 | return; | |
4978 | } | |
d397712b | 4979 | while (len > 0) { |
d1310b2e CM |
4980 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; |
4981 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
4982 | ||
4983 | dst_off_in_page = (start_offset + dst_end) & | |
4984 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4985 | src_off_in_page = (start_offset + src_end) & | |
4986 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4987 | ||
4988 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
4989 | cur = min(cur, dst_off_in_page + 1); | |
4990 | move_pages(extent_buffer_page(dst, dst_i), | |
4991 | extent_buffer_page(dst, src_i), | |
4992 | dst_off_in_page - cur + 1, | |
4993 | src_off_in_page - cur + 1, cur); | |
4994 | ||
4995 | dst_end -= cur; | |
4996 | src_end -= cur; | |
4997 | len -= cur; | |
4998 | } | |
4999 | } | |
6af118ce | 5000 | |
3083ee2e | 5001 | int try_release_extent_buffer(struct page *page, gfp_t mask) |
19fe0a8b | 5002 | { |
6af118ce | 5003 | struct extent_buffer *eb; |
6af118ce | 5004 | |
3083ee2e JB |
5005 | /* |
5006 | * We need to make sure noboody is attaching this page to an eb right | |
5007 | * now. | |
5008 | */ | |
5009 | spin_lock(&page->mapping->private_lock); | |
5010 | if (!PagePrivate(page)) { | |
5011 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 5012 | return 1; |
45f49bce | 5013 | } |
6af118ce | 5014 | |
3083ee2e JB |
5015 | eb = (struct extent_buffer *)page->private; |
5016 | BUG_ON(!eb); | |
19fe0a8b MX |
5017 | |
5018 | /* | |
3083ee2e JB |
5019 | * This is a little awful but should be ok, we need to make sure that |
5020 | * the eb doesn't disappear out from under us while we're looking at | |
5021 | * this page. | |
19fe0a8b | 5022 | */ |
3083ee2e | 5023 | spin_lock(&eb->refs_lock); |
0b32f4bb | 5024 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
5025 | spin_unlock(&eb->refs_lock); |
5026 | spin_unlock(&page->mapping->private_lock); | |
5027 | return 0; | |
b9473439 | 5028 | } |
3083ee2e | 5029 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 5030 | |
3083ee2e JB |
5031 | if ((mask & GFP_NOFS) == GFP_NOFS) |
5032 | mask = GFP_NOFS; | |
19fe0a8b | 5033 | |
19fe0a8b | 5034 | /* |
3083ee2e JB |
5035 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
5036 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 5037 | */ |
3083ee2e JB |
5038 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
5039 | spin_unlock(&eb->refs_lock); | |
5040 | return 0; | |
b9473439 | 5041 | } |
19fe0a8b | 5042 | |
e64860aa | 5043 | return release_extent_buffer(eb, mask); |
6af118ce | 5044 | } |