]>
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> | |
5 | #include <linux/gfp.h> | |
6 | #include <linux/pagemap.h> | |
7 | #include <linux/page-flags.h> | |
8 | #include <linux/module.h> | |
9 | #include <linux/spinlock.h> | |
10 | #include <linux/blkdev.h> | |
11 | #include <linux/swap.h> | |
d1310b2e CM |
12 | #include <linux/writeback.h> |
13 | #include <linux/pagevec.h> | |
14 | #include "extent_io.h" | |
15 | #include "extent_map.h" | |
2db04966 | 16 | #include "compat.h" |
902b22f3 DW |
17 | #include "ctree.h" |
18 | #include "btrfs_inode.h" | |
d1310b2e | 19 | |
d1310b2e CM |
20 | static struct kmem_cache *extent_state_cache; |
21 | static struct kmem_cache *extent_buffer_cache; | |
22 | ||
23 | static LIST_HEAD(buffers); | |
24 | static LIST_HEAD(states); | |
4bef0848 | 25 | |
b47eda86 | 26 | #define LEAK_DEBUG 0 |
3935127c | 27 | #if LEAK_DEBUG |
d397712b | 28 | static DEFINE_SPINLOCK(leak_lock); |
4bef0848 | 29 | #endif |
d1310b2e | 30 | |
d1310b2e CM |
31 | #define BUFFER_LRU_MAX 64 |
32 | ||
33 | struct tree_entry { | |
34 | u64 start; | |
35 | u64 end; | |
d1310b2e CM |
36 | struct rb_node rb_node; |
37 | }; | |
38 | ||
39 | struct extent_page_data { | |
40 | struct bio *bio; | |
41 | struct extent_io_tree *tree; | |
42 | get_extent_t *get_extent; | |
771ed689 CM |
43 | |
44 | /* tells writepage not to lock the state bits for this range | |
45 | * it still does the unlocking | |
46 | */ | |
ffbd517d CM |
47 | unsigned int extent_locked:1; |
48 | ||
49 | /* tells the submit_bio code to use a WRITE_SYNC */ | |
50 | unsigned int sync_io:1; | |
d1310b2e CM |
51 | }; |
52 | ||
53 | int __init extent_io_init(void) | |
54 | { | |
9601e3f6 CH |
55 | extent_state_cache = kmem_cache_create("extent_state", |
56 | sizeof(struct extent_state), 0, | |
57 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
58 | if (!extent_state_cache) |
59 | return -ENOMEM; | |
60 | ||
9601e3f6 CH |
61 | extent_buffer_cache = kmem_cache_create("extent_buffers", |
62 | sizeof(struct extent_buffer), 0, | |
63 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
64 | if (!extent_buffer_cache) |
65 | goto free_state_cache; | |
66 | return 0; | |
67 | ||
68 | free_state_cache: | |
69 | kmem_cache_destroy(extent_state_cache); | |
70 | return -ENOMEM; | |
71 | } | |
72 | ||
73 | void extent_io_exit(void) | |
74 | { | |
75 | struct extent_state *state; | |
2d2ae547 | 76 | struct extent_buffer *eb; |
d1310b2e CM |
77 | |
78 | while (!list_empty(&states)) { | |
2d2ae547 | 79 | state = list_entry(states.next, struct extent_state, leak_list); |
d397712b CM |
80 | printk(KERN_ERR "btrfs state leak: start %llu end %llu " |
81 | "state %lu in tree %p refs %d\n", | |
82 | (unsigned long long)state->start, | |
83 | (unsigned long long)state->end, | |
84 | state->state, state->tree, atomic_read(&state->refs)); | |
2d2ae547 | 85 | list_del(&state->leak_list); |
d1310b2e CM |
86 | kmem_cache_free(extent_state_cache, state); |
87 | ||
88 | } | |
89 | ||
2d2ae547 CM |
90 | while (!list_empty(&buffers)) { |
91 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
d397712b CM |
92 | printk(KERN_ERR "btrfs buffer leak start %llu len %lu " |
93 | "refs %d\n", (unsigned long long)eb->start, | |
94 | eb->len, atomic_read(&eb->refs)); | |
2d2ae547 CM |
95 | list_del(&eb->leak_list); |
96 | kmem_cache_free(extent_buffer_cache, eb); | |
97 | } | |
d1310b2e CM |
98 | if (extent_state_cache) |
99 | kmem_cache_destroy(extent_state_cache); | |
100 | if (extent_buffer_cache) | |
101 | kmem_cache_destroy(extent_buffer_cache); | |
102 | } | |
103 | ||
104 | void extent_io_tree_init(struct extent_io_tree *tree, | |
105 | struct address_space *mapping, gfp_t mask) | |
106 | { | |
107 | tree->state.rb_node = NULL; | |
6af118ce | 108 | tree->buffer.rb_node = NULL; |
d1310b2e CM |
109 | tree->ops = NULL; |
110 | tree->dirty_bytes = 0; | |
70dec807 | 111 | spin_lock_init(&tree->lock); |
6af118ce | 112 | spin_lock_init(&tree->buffer_lock); |
d1310b2e | 113 | tree->mapping = mapping; |
d1310b2e | 114 | } |
d1310b2e | 115 | |
b2950863 | 116 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
117 | { |
118 | struct extent_state *state; | |
3935127c | 119 | #if LEAK_DEBUG |
2d2ae547 | 120 | unsigned long flags; |
4bef0848 | 121 | #endif |
d1310b2e CM |
122 | |
123 | state = kmem_cache_alloc(extent_state_cache, mask); | |
2b114d1d | 124 | if (!state) |
d1310b2e CM |
125 | return state; |
126 | state->state = 0; | |
d1310b2e | 127 | state->private = 0; |
70dec807 | 128 | state->tree = NULL; |
3935127c | 129 | #if LEAK_DEBUG |
2d2ae547 CM |
130 | spin_lock_irqsave(&leak_lock, flags); |
131 | list_add(&state->leak_list, &states); | |
132 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 133 | #endif |
d1310b2e CM |
134 | atomic_set(&state->refs, 1); |
135 | init_waitqueue_head(&state->wq); | |
136 | return state; | |
137 | } | |
d1310b2e | 138 | |
b2950863 | 139 | static void free_extent_state(struct extent_state *state) |
d1310b2e | 140 | { |
d1310b2e CM |
141 | if (!state) |
142 | return; | |
143 | if (atomic_dec_and_test(&state->refs)) { | |
3935127c | 144 | #if LEAK_DEBUG |
2d2ae547 | 145 | unsigned long flags; |
4bef0848 | 146 | #endif |
70dec807 | 147 | WARN_ON(state->tree); |
3935127c | 148 | #if LEAK_DEBUG |
2d2ae547 CM |
149 | spin_lock_irqsave(&leak_lock, flags); |
150 | list_del(&state->leak_list); | |
151 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 152 | #endif |
d1310b2e CM |
153 | kmem_cache_free(extent_state_cache, state); |
154 | } | |
155 | } | |
d1310b2e CM |
156 | |
157 | static struct rb_node *tree_insert(struct rb_root *root, u64 offset, | |
158 | struct rb_node *node) | |
159 | { | |
d397712b CM |
160 | struct rb_node **p = &root->rb_node; |
161 | struct rb_node *parent = NULL; | |
d1310b2e CM |
162 | struct tree_entry *entry; |
163 | ||
d397712b | 164 | while (*p) { |
d1310b2e CM |
165 | parent = *p; |
166 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
167 | ||
168 | if (offset < entry->start) | |
169 | p = &(*p)->rb_left; | |
170 | else if (offset > entry->end) | |
171 | p = &(*p)->rb_right; | |
172 | else | |
173 | return parent; | |
174 | } | |
175 | ||
176 | entry = rb_entry(node, struct tree_entry, rb_node); | |
d1310b2e CM |
177 | rb_link_node(node, parent, p); |
178 | rb_insert_color(node, root); | |
179 | return NULL; | |
180 | } | |
181 | ||
80ea96b1 | 182 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
d1310b2e CM |
183 | struct rb_node **prev_ret, |
184 | struct rb_node **next_ret) | |
185 | { | |
80ea96b1 | 186 | struct rb_root *root = &tree->state; |
d397712b | 187 | struct rb_node *n = root->rb_node; |
d1310b2e CM |
188 | struct rb_node *prev = NULL; |
189 | struct rb_node *orig_prev = NULL; | |
190 | struct tree_entry *entry; | |
191 | struct tree_entry *prev_entry = NULL; | |
192 | ||
d397712b | 193 | while (n) { |
d1310b2e CM |
194 | entry = rb_entry(n, struct tree_entry, rb_node); |
195 | prev = n; | |
196 | prev_entry = entry; | |
197 | ||
198 | if (offset < entry->start) | |
199 | n = n->rb_left; | |
200 | else if (offset > entry->end) | |
201 | n = n->rb_right; | |
d397712b | 202 | else |
d1310b2e CM |
203 | return n; |
204 | } | |
205 | ||
206 | if (prev_ret) { | |
207 | orig_prev = prev; | |
d397712b | 208 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
209 | prev = rb_next(prev); |
210 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
211 | } | |
212 | *prev_ret = prev; | |
213 | prev = orig_prev; | |
214 | } | |
215 | ||
216 | if (next_ret) { | |
217 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 218 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
219 | prev = rb_prev(prev); |
220 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
221 | } | |
222 | *next_ret = prev; | |
223 | } | |
224 | return NULL; | |
225 | } | |
226 | ||
80ea96b1 CM |
227 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
228 | u64 offset) | |
d1310b2e | 229 | { |
70dec807 | 230 | struct rb_node *prev = NULL; |
d1310b2e | 231 | struct rb_node *ret; |
70dec807 | 232 | |
80ea96b1 | 233 | ret = __etree_search(tree, offset, &prev, NULL); |
d397712b | 234 | if (!ret) |
d1310b2e CM |
235 | return prev; |
236 | return ret; | |
237 | } | |
238 | ||
6af118ce CM |
239 | static struct extent_buffer *buffer_tree_insert(struct extent_io_tree *tree, |
240 | u64 offset, struct rb_node *node) | |
241 | { | |
242 | struct rb_root *root = &tree->buffer; | |
d397712b CM |
243 | struct rb_node **p = &root->rb_node; |
244 | struct rb_node *parent = NULL; | |
6af118ce CM |
245 | struct extent_buffer *eb; |
246 | ||
d397712b | 247 | while (*p) { |
6af118ce CM |
248 | parent = *p; |
249 | eb = rb_entry(parent, struct extent_buffer, rb_node); | |
250 | ||
251 | if (offset < eb->start) | |
252 | p = &(*p)->rb_left; | |
253 | else if (offset > eb->start) | |
254 | p = &(*p)->rb_right; | |
255 | else | |
256 | return eb; | |
257 | } | |
258 | ||
259 | rb_link_node(node, parent, p); | |
260 | rb_insert_color(node, root); | |
261 | return NULL; | |
262 | } | |
263 | ||
264 | static struct extent_buffer *buffer_search(struct extent_io_tree *tree, | |
265 | u64 offset) | |
266 | { | |
267 | struct rb_root *root = &tree->buffer; | |
d397712b | 268 | struct rb_node *n = root->rb_node; |
6af118ce CM |
269 | struct extent_buffer *eb; |
270 | ||
d397712b | 271 | while (n) { |
6af118ce CM |
272 | eb = rb_entry(n, struct extent_buffer, rb_node); |
273 | if (offset < eb->start) | |
274 | n = n->rb_left; | |
275 | else if (offset > eb->start) | |
276 | n = n->rb_right; | |
277 | else | |
278 | return eb; | |
279 | } | |
280 | return NULL; | |
281 | } | |
282 | ||
d1310b2e CM |
283 | /* |
284 | * utility function to look for merge candidates inside a given range. | |
285 | * Any extents with matching state are merged together into a single | |
286 | * extent in the tree. Extents with EXTENT_IO in their state field | |
287 | * are not merged because the end_io handlers need to be able to do | |
288 | * operations on them without sleeping (or doing allocations/splits). | |
289 | * | |
290 | * This should be called with the tree lock held. | |
291 | */ | |
292 | static int merge_state(struct extent_io_tree *tree, | |
293 | struct extent_state *state) | |
294 | { | |
295 | struct extent_state *other; | |
296 | struct rb_node *other_node; | |
297 | ||
5b21f2ed | 298 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
d1310b2e CM |
299 | return 0; |
300 | ||
301 | other_node = rb_prev(&state->rb_node); | |
302 | if (other_node) { | |
303 | other = rb_entry(other_node, struct extent_state, rb_node); | |
304 | if (other->end == state->start - 1 && | |
305 | other->state == state->state) { | |
306 | state->start = other->start; | |
70dec807 | 307 | other->tree = NULL; |
d1310b2e CM |
308 | rb_erase(&other->rb_node, &tree->state); |
309 | free_extent_state(other); | |
310 | } | |
311 | } | |
312 | other_node = rb_next(&state->rb_node); | |
313 | if (other_node) { | |
314 | other = rb_entry(other_node, struct extent_state, rb_node); | |
315 | if (other->start == state->end + 1 && | |
316 | other->state == state->state) { | |
317 | other->start = state->start; | |
70dec807 | 318 | state->tree = NULL; |
d1310b2e CM |
319 | rb_erase(&state->rb_node, &tree->state); |
320 | free_extent_state(state); | |
321 | } | |
322 | } | |
323 | return 0; | |
324 | } | |
325 | ||
291d673e CM |
326 | static void set_state_cb(struct extent_io_tree *tree, |
327 | struct extent_state *state, | |
328 | unsigned long bits) | |
329 | { | |
330 | if (tree->ops && tree->ops->set_bit_hook) { | |
331 | tree->ops->set_bit_hook(tree->mapping->host, state->start, | |
b0c68f8b | 332 | state->end, state->state, bits); |
291d673e CM |
333 | } |
334 | } | |
335 | ||
336 | static void clear_state_cb(struct extent_io_tree *tree, | |
337 | struct extent_state *state, | |
338 | unsigned long bits) | |
339 | { | |
c584482b | 340 | if (tree->ops && tree->ops->clear_bit_hook) { |
291d673e | 341 | tree->ops->clear_bit_hook(tree->mapping->host, state->start, |
b0c68f8b | 342 | state->end, state->state, bits); |
291d673e CM |
343 | } |
344 | } | |
345 | ||
d1310b2e CM |
346 | /* |
347 | * insert an extent_state struct into the tree. 'bits' are set on the | |
348 | * struct before it is inserted. | |
349 | * | |
350 | * This may return -EEXIST if the extent is already there, in which case the | |
351 | * state struct is freed. | |
352 | * | |
353 | * The tree lock is not taken internally. This is a utility function and | |
354 | * probably isn't what you want to call (see set/clear_extent_bit). | |
355 | */ | |
356 | static int insert_state(struct extent_io_tree *tree, | |
357 | struct extent_state *state, u64 start, u64 end, | |
358 | int bits) | |
359 | { | |
360 | struct rb_node *node; | |
361 | ||
362 | if (end < start) { | |
d397712b CM |
363 | printk(KERN_ERR "btrfs end < start %llu %llu\n", |
364 | (unsigned long long)end, | |
365 | (unsigned long long)start); | |
d1310b2e CM |
366 | WARN_ON(1); |
367 | } | |
368 | if (bits & EXTENT_DIRTY) | |
369 | tree->dirty_bytes += end - start + 1; | |
d1310b2e CM |
370 | state->start = start; |
371 | state->end = end; | |
e48c465b CM |
372 | set_state_cb(tree, state, bits); |
373 | state->state |= bits; | |
d1310b2e CM |
374 | node = tree_insert(&tree->state, end, &state->rb_node); |
375 | if (node) { | |
376 | struct extent_state *found; | |
377 | found = rb_entry(node, struct extent_state, rb_node); | |
d397712b CM |
378 | printk(KERN_ERR "btrfs found node %llu %llu on insert of " |
379 | "%llu %llu\n", (unsigned long long)found->start, | |
380 | (unsigned long long)found->end, | |
381 | (unsigned long long)start, (unsigned long long)end); | |
d1310b2e CM |
382 | free_extent_state(state); |
383 | return -EEXIST; | |
384 | } | |
70dec807 | 385 | state->tree = tree; |
d1310b2e CM |
386 | merge_state(tree, state); |
387 | return 0; | |
388 | } | |
389 | ||
390 | /* | |
391 | * split a given extent state struct in two, inserting the preallocated | |
392 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
393 | * offset inside 'orig' where it should be split. | |
394 | * | |
395 | * Before calling, | |
396 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
397 | * are two extent state structs in the tree: | |
398 | * prealloc: [orig->start, split - 1] | |
399 | * orig: [ split, orig->end ] | |
400 | * | |
401 | * The tree locks are not taken by this function. They need to be held | |
402 | * by the caller. | |
403 | */ | |
404 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
405 | struct extent_state *prealloc, u64 split) | |
406 | { | |
407 | struct rb_node *node; | |
408 | prealloc->start = orig->start; | |
409 | prealloc->end = split - 1; | |
410 | prealloc->state = orig->state; | |
411 | orig->start = split; | |
412 | ||
413 | node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node); | |
414 | if (node) { | |
d1310b2e CM |
415 | free_extent_state(prealloc); |
416 | return -EEXIST; | |
417 | } | |
70dec807 | 418 | prealloc->tree = tree; |
d1310b2e CM |
419 | return 0; |
420 | } | |
421 | ||
422 | /* | |
423 | * utility function to clear some bits in an extent state struct. | |
424 | * it will optionally wake up any one waiting on this state (wake == 1), or | |
425 | * forcibly remove the state from the tree (delete == 1). | |
426 | * | |
427 | * If no bits are set on the state struct after clearing things, the | |
428 | * struct is freed and removed from the tree | |
429 | */ | |
430 | static int clear_state_bit(struct extent_io_tree *tree, | |
431 | struct extent_state *state, int bits, int wake, | |
432 | int delete) | |
433 | { | |
434 | int ret = state->state & bits; | |
435 | ||
436 | if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { | |
437 | u64 range = state->end - state->start + 1; | |
438 | WARN_ON(range > tree->dirty_bytes); | |
439 | tree->dirty_bytes -= range; | |
440 | } | |
291d673e | 441 | clear_state_cb(tree, state, bits); |
b0c68f8b | 442 | state->state &= ~bits; |
d1310b2e CM |
443 | if (wake) |
444 | wake_up(&state->wq); | |
445 | if (delete || state->state == 0) { | |
70dec807 | 446 | if (state->tree) { |
ae9d1285 | 447 | clear_state_cb(tree, state, state->state); |
d1310b2e | 448 | rb_erase(&state->rb_node, &tree->state); |
70dec807 | 449 | state->tree = NULL; |
d1310b2e CM |
450 | free_extent_state(state); |
451 | } else { | |
452 | WARN_ON(1); | |
453 | } | |
454 | } else { | |
455 | merge_state(tree, state); | |
456 | } | |
457 | return ret; | |
458 | } | |
459 | ||
460 | /* | |
461 | * clear some bits on a range in the tree. This may require splitting | |
462 | * or inserting elements in the tree, so the gfp mask is used to | |
463 | * indicate which allocations or sleeping are allowed. | |
464 | * | |
465 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
466 | * the given range from the tree regardless of state (ie for truncate). | |
467 | * | |
468 | * the range [start, end] is inclusive. | |
469 | * | |
470 | * This takes the tree lock, and returns < 0 on error, > 0 if any of the | |
471 | * bits were already set, or zero if none of the bits were already set. | |
472 | */ | |
473 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
474 | int bits, int wake, int delete, gfp_t mask) | |
475 | { | |
476 | struct extent_state *state; | |
477 | struct extent_state *prealloc = NULL; | |
478 | struct rb_node *node; | |
5c939df5 | 479 | u64 last_end; |
d1310b2e CM |
480 | int err; |
481 | int set = 0; | |
482 | ||
483 | again: | |
484 | if (!prealloc && (mask & __GFP_WAIT)) { | |
485 | prealloc = alloc_extent_state(mask); | |
486 | if (!prealloc) | |
487 | return -ENOMEM; | |
488 | } | |
489 | ||
cad321ad | 490 | spin_lock(&tree->lock); |
d1310b2e CM |
491 | /* |
492 | * this search will find the extents that end after | |
493 | * our range starts | |
494 | */ | |
80ea96b1 | 495 | node = tree_search(tree, start); |
d1310b2e CM |
496 | if (!node) |
497 | goto out; | |
498 | state = rb_entry(node, struct extent_state, rb_node); | |
499 | if (state->start > end) | |
500 | goto out; | |
501 | WARN_ON(state->end < start); | |
5c939df5 | 502 | last_end = state->end; |
d1310b2e CM |
503 | |
504 | /* | |
505 | * | ---- desired range ---- | | |
506 | * | state | or | |
507 | * | ------------- state -------------- | | |
508 | * | |
509 | * We need to split the extent we found, and may flip | |
510 | * bits on second half. | |
511 | * | |
512 | * If the extent we found extends past our range, we | |
513 | * just split and search again. It'll get split again | |
514 | * the next time though. | |
515 | * | |
516 | * If the extent we found is inside our range, we clear | |
517 | * the desired bit on it. | |
518 | */ | |
519 | ||
520 | if (state->start < start) { | |
70dec807 CM |
521 | if (!prealloc) |
522 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
d1310b2e CM |
523 | err = split_state(tree, state, prealloc, start); |
524 | BUG_ON(err == -EEXIST); | |
525 | prealloc = NULL; | |
526 | if (err) | |
527 | goto out; | |
528 | if (state->end <= end) { | |
d1310b2e CM |
529 | set |= clear_state_bit(tree, state, bits, |
530 | wake, delete); | |
5c939df5 YZ |
531 | if (last_end == (u64)-1) |
532 | goto out; | |
533 | start = last_end + 1; | |
d1310b2e CM |
534 | } else { |
535 | start = state->start; | |
536 | } | |
537 | goto search_again; | |
538 | } | |
539 | /* | |
540 | * | ---- desired range ---- | | |
541 | * | state | | |
542 | * We need to split the extent, and clear the bit | |
543 | * on the first half | |
544 | */ | |
545 | if (state->start <= end && state->end > end) { | |
70dec807 CM |
546 | if (!prealloc) |
547 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
d1310b2e CM |
548 | err = split_state(tree, state, prealloc, end + 1); |
549 | BUG_ON(err == -EEXIST); | |
550 | ||
551 | if (wake) | |
552 | wake_up(&state->wq); | |
553 | set |= clear_state_bit(tree, prealloc, bits, | |
554 | wake, delete); | |
555 | prealloc = NULL; | |
556 | goto out; | |
557 | } | |
558 | ||
d1310b2e | 559 | set |= clear_state_bit(tree, state, bits, wake, delete); |
5c939df5 YZ |
560 | if (last_end == (u64)-1) |
561 | goto out; | |
562 | start = last_end + 1; | |
d1310b2e CM |
563 | goto search_again; |
564 | ||
565 | out: | |
cad321ad | 566 | spin_unlock(&tree->lock); |
d1310b2e CM |
567 | if (prealloc) |
568 | free_extent_state(prealloc); | |
569 | ||
570 | return set; | |
571 | ||
572 | search_again: | |
573 | if (start > end) | |
574 | goto out; | |
cad321ad | 575 | spin_unlock(&tree->lock); |
d1310b2e CM |
576 | if (mask & __GFP_WAIT) |
577 | cond_resched(); | |
578 | goto again; | |
579 | } | |
d1310b2e CM |
580 | |
581 | static int wait_on_state(struct extent_io_tree *tree, | |
582 | struct extent_state *state) | |
641f5219 CH |
583 | __releases(tree->lock) |
584 | __acquires(tree->lock) | |
d1310b2e CM |
585 | { |
586 | DEFINE_WAIT(wait); | |
587 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 588 | spin_unlock(&tree->lock); |
d1310b2e | 589 | schedule(); |
cad321ad | 590 | spin_lock(&tree->lock); |
d1310b2e CM |
591 | finish_wait(&state->wq, &wait); |
592 | return 0; | |
593 | } | |
594 | ||
595 | /* | |
596 | * waits for one or more bits to clear on a range in the state tree. | |
597 | * The range [start, end] is inclusive. | |
598 | * The tree lock is taken by this function | |
599 | */ | |
600 | int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits) | |
601 | { | |
602 | struct extent_state *state; | |
603 | struct rb_node *node; | |
604 | ||
cad321ad | 605 | spin_lock(&tree->lock); |
d1310b2e CM |
606 | again: |
607 | while (1) { | |
608 | /* | |
609 | * this search will find all the extents that end after | |
610 | * our range starts | |
611 | */ | |
80ea96b1 | 612 | node = tree_search(tree, start); |
d1310b2e CM |
613 | if (!node) |
614 | break; | |
615 | ||
616 | state = rb_entry(node, struct extent_state, rb_node); | |
617 | ||
618 | if (state->start > end) | |
619 | goto out; | |
620 | ||
621 | if (state->state & bits) { | |
622 | start = state->start; | |
623 | atomic_inc(&state->refs); | |
624 | wait_on_state(tree, state); | |
625 | free_extent_state(state); | |
626 | goto again; | |
627 | } | |
628 | start = state->end + 1; | |
629 | ||
630 | if (start > end) | |
631 | break; | |
632 | ||
633 | if (need_resched()) { | |
cad321ad | 634 | spin_unlock(&tree->lock); |
d1310b2e | 635 | cond_resched(); |
cad321ad | 636 | spin_lock(&tree->lock); |
d1310b2e CM |
637 | } |
638 | } | |
639 | out: | |
cad321ad | 640 | spin_unlock(&tree->lock); |
d1310b2e CM |
641 | return 0; |
642 | } | |
d1310b2e CM |
643 | |
644 | static void set_state_bits(struct extent_io_tree *tree, | |
645 | struct extent_state *state, | |
646 | int bits) | |
647 | { | |
648 | if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { | |
649 | u64 range = state->end - state->start + 1; | |
650 | tree->dirty_bytes += range; | |
651 | } | |
291d673e | 652 | set_state_cb(tree, state, bits); |
b0c68f8b | 653 | state->state |= bits; |
d1310b2e CM |
654 | } |
655 | ||
656 | /* | |
1edbb734 CM |
657 | * set some bits on a range in the tree. This may require allocations or |
658 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 659 | * |
1edbb734 CM |
660 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
661 | * part of the range already has the desired bits set. The start of the | |
662 | * existing range is returned in failed_start in this case. | |
d1310b2e | 663 | * |
1edbb734 | 664 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 665 | */ |
1edbb734 | 666 | |
d397712b | 667 | static int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
1edbb734 | 668 | int bits, int exclusive_bits, u64 *failed_start, |
d397712b | 669 | gfp_t mask) |
d1310b2e CM |
670 | { |
671 | struct extent_state *state; | |
672 | struct extent_state *prealloc = NULL; | |
673 | struct rb_node *node; | |
d1310b2e | 674 | int err = 0; |
d1310b2e CM |
675 | u64 last_start; |
676 | u64 last_end; | |
677 | again: | |
678 | if (!prealloc && (mask & __GFP_WAIT)) { | |
679 | prealloc = alloc_extent_state(mask); | |
680 | if (!prealloc) | |
681 | return -ENOMEM; | |
682 | } | |
683 | ||
cad321ad | 684 | spin_lock(&tree->lock); |
d1310b2e CM |
685 | /* |
686 | * this search will find all the extents that end after | |
687 | * our range starts. | |
688 | */ | |
80ea96b1 | 689 | node = tree_search(tree, start); |
d1310b2e CM |
690 | if (!node) { |
691 | err = insert_state(tree, prealloc, start, end, bits); | |
692 | prealloc = NULL; | |
693 | BUG_ON(err == -EEXIST); | |
694 | goto out; | |
695 | } | |
d1310b2e | 696 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 697 | hit_next: |
d1310b2e CM |
698 | last_start = state->start; |
699 | last_end = state->end; | |
700 | ||
701 | /* | |
702 | * | ---- desired range ---- | | |
703 | * | state | | |
704 | * | |
705 | * Just lock what we found and keep going | |
706 | */ | |
707 | if (state->start == start && state->end <= end) { | |
40431d6c | 708 | struct rb_node *next_node; |
1edbb734 | 709 | if (state->state & exclusive_bits) { |
d1310b2e CM |
710 | *failed_start = state->start; |
711 | err = -EEXIST; | |
712 | goto out; | |
713 | } | |
714 | set_state_bits(tree, state, bits); | |
d1310b2e | 715 | merge_state(tree, state); |
5c939df5 YZ |
716 | if (last_end == (u64)-1) |
717 | goto out; | |
40431d6c | 718 | |
5c939df5 | 719 | start = last_end + 1; |
40431d6c CM |
720 | if (start < end && prealloc && !need_resched()) { |
721 | next_node = rb_next(node); | |
722 | if (next_node) { | |
723 | state = rb_entry(next_node, struct extent_state, | |
724 | rb_node); | |
725 | if (state->start == start) | |
726 | goto hit_next; | |
727 | } | |
728 | } | |
d1310b2e CM |
729 | goto search_again; |
730 | } | |
731 | ||
732 | /* | |
733 | * | ---- desired range ---- | | |
734 | * | state | | |
735 | * or | |
736 | * | ------------- state -------------- | | |
737 | * | |
738 | * We need to split the extent we found, and may flip bits on | |
739 | * second half. | |
740 | * | |
741 | * If the extent we found extends past our | |
742 | * range, we just split and search again. It'll get split | |
743 | * again the next time though. | |
744 | * | |
745 | * If the extent we found is inside our range, we set the | |
746 | * desired bit on it. | |
747 | */ | |
748 | if (state->start < start) { | |
1edbb734 | 749 | if (state->state & exclusive_bits) { |
d1310b2e CM |
750 | *failed_start = start; |
751 | err = -EEXIST; | |
752 | goto out; | |
753 | } | |
754 | err = split_state(tree, state, prealloc, start); | |
755 | BUG_ON(err == -EEXIST); | |
756 | prealloc = NULL; | |
757 | if (err) | |
758 | goto out; | |
759 | if (state->end <= end) { | |
760 | set_state_bits(tree, state, bits); | |
d1310b2e | 761 | merge_state(tree, state); |
5c939df5 YZ |
762 | if (last_end == (u64)-1) |
763 | goto out; | |
764 | start = last_end + 1; | |
d1310b2e CM |
765 | } else { |
766 | start = state->start; | |
767 | } | |
768 | goto search_again; | |
769 | } | |
770 | /* | |
771 | * | ---- desired range ---- | | |
772 | * | state | or | state | | |
773 | * | |
774 | * There's a hole, we need to insert something in it and | |
775 | * ignore the extent we found. | |
776 | */ | |
777 | if (state->start > start) { | |
778 | u64 this_end; | |
779 | if (end < last_start) | |
780 | this_end = end; | |
781 | else | |
d397712b | 782 | this_end = last_start - 1; |
d1310b2e CM |
783 | err = insert_state(tree, prealloc, start, this_end, |
784 | bits); | |
785 | prealloc = NULL; | |
786 | BUG_ON(err == -EEXIST); | |
787 | if (err) | |
788 | goto out; | |
789 | start = this_end + 1; | |
790 | goto search_again; | |
791 | } | |
792 | /* | |
793 | * | ---- desired range ---- | | |
794 | * | state | | |
795 | * We need to split the extent, and set the bit | |
796 | * on the first half | |
797 | */ | |
798 | if (state->start <= end && state->end > end) { | |
1edbb734 | 799 | if (state->state & exclusive_bits) { |
d1310b2e CM |
800 | *failed_start = start; |
801 | err = -EEXIST; | |
802 | goto out; | |
803 | } | |
804 | err = split_state(tree, state, prealloc, end + 1); | |
805 | BUG_ON(err == -EEXIST); | |
806 | ||
807 | set_state_bits(tree, prealloc, bits); | |
808 | merge_state(tree, prealloc); | |
809 | prealloc = NULL; | |
810 | goto out; | |
811 | } | |
812 | ||
813 | goto search_again; | |
814 | ||
815 | out: | |
cad321ad | 816 | spin_unlock(&tree->lock); |
d1310b2e CM |
817 | if (prealloc) |
818 | free_extent_state(prealloc); | |
819 | ||
820 | return err; | |
821 | ||
822 | search_again: | |
823 | if (start > end) | |
824 | goto out; | |
cad321ad | 825 | spin_unlock(&tree->lock); |
d1310b2e CM |
826 | if (mask & __GFP_WAIT) |
827 | cond_resched(); | |
828 | goto again; | |
829 | } | |
d1310b2e CM |
830 | |
831 | /* wrappers around set/clear extent bit */ | |
832 | int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
833 | gfp_t mask) | |
834 | { | |
835 | return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL, | |
836 | mask); | |
837 | } | |
d1310b2e | 838 | |
e6dcd2dc CM |
839 | int set_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end, |
840 | gfp_t mask) | |
841 | { | |
842 | return set_extent_bit(tree, start, end, EXTENT_ORDERED, 0, NULL, mask); | |
843 | } | |
e6dcd2dc | 844 | |
d1310b2e CM |
845 | int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
846 | int bits, gfp_t mask) | |
847 | { | |
848 | return set_extent_bit(tree, start, end, bits, 0, NULL, | |
849 | mask); | |
850 | } | |
d1310b2e CM |
851 | |
852 | int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
853 | int bits, gfp_t mask) | |
854 | { | |
855 | return clear_extent_bit(tree, start, end, bits, 0, 0, mask); | |
856 | } | |
d1310b2e CM |
857 | |
858 | int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, | |
859 | gfp_t mask) | |
860 | { | |
861 | return set_extent_bit(tree, start, end, | |
40431d6c | 862 | EXTENT_DELALLOC | EXTENT_DIRTY | EXTENT_UPTODATE, |
e6dcd2dc | 863 | 0, NULL, mask); |
d1310b2e | 864 | } |
d1310b2e CM |
865 | |
866 | int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
867 | gfp_t mask) | |
868 | { | |
869 | return clear_extent_bit(tree, start, end, | |
870 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask); | |
871 | } | |
d1310b2e | 872 | |
e6dcd2dc CM |
873 | int clear_extent_ordered(struct extent_io_tree *tree, u64 start, u64 end, |
874 | gfp_t mask) | |
875 | { | |
876 | return clear_extent_bit(tree, start, end, EXTENT_ORDERED, 1, 0, mask); | |
877 | } | |
e6dcd2dc | 878 | |
d1310b2e CM |
879 | int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, |
880 | gfp_t mask) | |
881 | { | |
882 | return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL, | |
883 | mask); | |
884 | } | |
d1310b2e | 885 | |
b2950863 | 886 | static int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end, |
d1310b2e CM |
887 | gfp_t mask) |
888 | { | |
889 | return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask); | |
890 | } | |
d1310b2e CM |
891 | |
892 | int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, | |
893 | gfp_t mask) | |
894 | { | |
895 | return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL, | |
896 | mask); | |
897 | } | |
d1310b2e | 898 | |
d397712b CM |
899 | static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, |
900 | u64 end, gfp_t mask) | |
d1310b2e CM |
901 | { |
902 | return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask); | |
903 | } | |
d1310b2e | 904 | |
d1310b2e CM |
905 | int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
906 | { | |
907 | return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK); | |
908 | } | |
d1310b2e | 909 | |
d352ac68 CM |
910 | /* |
911 | * either insert or lock state struct between start and end use mask to tell | |
912 | * us if waiting is desired. | |
913 | */ | |
1edbb734 CM |
914 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
915 | int bits, gfp_t mask) | |
d1310b2e CM |
916 | { |
917 | int err; | |
918 | u64 failed_start; | |
919 | while (1) { | |
1edbb734 CM |
920 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits, |
921 | EXTENT_LOCKED, &failed_start, mask); | |
d1310b2e CM |
922 | if (err == -EEXIST && (mask & __GFP_WAIT)) { |
923 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); | |
924 | start = failed_start; | |
925 | } else { | |
926 | break; | |
927 | } | |
928 | WARN_ON(start > end); | |
929 | } | |
930 | return err; | |
931 | } | |
d1310b2e | 932 | |
1edbb734 CM |
933 | int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) |
934 | { | |
935 | return lock_extent_bits(tree, start, end, 0, mask); | |
936 | } | |
937 | ||
25179201 JB |
938 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end, |
939 | gfp_t mask) | |
940 | { | |
941 | int err; | |
942 | u64 failed_start; | |
943 | ||
944 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1, | |
945 | &failed_start, mask); | |
6643558d YZ |
946 | if (err == -EEXIST) { |
947 | if (failed_start > start) | |
948 | clear_extent_bit(tree, start, failed_start - 1, | |
949 | EXTENT_LOCKED, 1, 0, mask); | |
25179201 | 950 | return 0; |
6643558d | 951 | } |
25179201 JB |
952 | return 1; |
953 | } | |
25179201 | 954 | |
d1310b2e CM |
955 | int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, |
956 | gfp_t mask) | |
957 | { | |
958 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask); | |
959 | } | |
d1310b2e CM |
960 | |
961 | /* | |
962 | * helper function to set pages and extents in the tree dirty | |
963 | */ | |
964 | int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end) | |
965 | { | |
966 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
967 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
968 | struct page *page; | |
969 | ||
970 | while (index <= end_index) { | |
971 | page = find_get_page(tree->mapping, index); | |
972 | BUG_ON(!page); | |
973 | __set_page_dirty_nobuffers(page); | |
974 | page_cache_release(page); | |
975 | index++; | |
976 | } | |
d1310b2e CM |
977 | return 0; |
978 | } | |
d1310b2e CM |
979 | |
980 | /* | |
981 | * helper function to set both pages and extents in the tree writeback | |
982 | */ | |
b2950863 | 983 | static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e CM |
984 | { |
985 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
986 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
987 | struct page *page; | |
988 | ||
989 | while (index <= end_index) { | |
990 | page = find_get_page(tree->mapping, index); | |
991 | BUG_ON(!page); | |
992 | set_page_writeback(page); | |
993 | page_cache_release(page); | |
994 | index++; | |
995 | } | |
d1310b2e CM |
996 | return 0; |
997 | } | |
d1310b2e | 998 | |
d352ac68 CM |
999 | /* |
1000 | * find the first offset in the io tree with 'bits' set. zero is | |
1001 | * returned if we find something, and *start_ret and *end_ret are | |
1002 | * set to reflect the state struct that was found. | |
1003 | * | |
1004 | * If nothing was found, 1 is returned, < 0 on error | |
1005 | */ | |
d1310b2e CM |
1006 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, |
1007 | u64 *start_ret, u64 *end_ret, int bits) | |
1008 | { | |
1009 | struct rb_node *node; | |
1010 | struct extent_state *state; | |
1011 | int ret = 1; | |
1012 | ||
cad321ad | 1013 | spin_lock(&tree->lock); |
d1310b2e CM |
1014 | /* |
1015 | * this search will find all the extents that end after | |
1016 | * our range starts. | |
1017 | */ | |
80ea96b1 | 1018 | node = tree_search(tree, start); |
d397712b | 1019 | if (!node) |
d1310b2e | 1020 | goto out; |
d1310b2e | 1021 | |
d397712b | 1022 | while (1) { |
d1310b2e CM |
1023 | state = rb_entry(node, struct extent_state, rb_node); |
1024 | if (state->end >= start && (state->state & bits)) { | |
1025 | *start_ret = state->start; | |
1026 | *end_ret = state->end; | |
1027 | ret = 0; | |
1028 | break; | |
1029 | } | |
1030 | node = rb_next(node); | |
1031 | if (!node) | |
1032 | break; | |
1033 | } | |
1034 | out: | |
cad321ad | 1035 | spin_unlock(&tree->lock); |
d1310b2e CM |
1036 | return ret; |
1037 | } | |
d1310b2e | 1038 | |
d352ac68 CM |
1039 | /* find the first state struct with 'bits' set after 'start', and |
1040 | * return it. tree->lock must be held. NULL will returned if | |
1041 | * nothing was found after 'start' | |
1042 | */ | |
d7fc640e CM |
1043 | struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree, |
1044 | u64 start, int bits) | |
1045 | { | |
1046 | struct rb_node *node; | |
1047 | struct extent_state *state; | |
1048 | ||
1049 | /* | |
1050 | * this search will find all the extents that end after | |
1051 | * our range starts. | |
1052 | */ | |
1053 | node = tree_search(tree, start); | |
d397712b | 1054 | if (!node) |
d7fc640e | 1055 | goto out; |
d7fc640e | 1056 | |
d397712b | 1057 | while (1) { |
d7fc640e | 1058 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1059 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1060 | return state; |
d397712b | 1061 | |
d7fc640e CM |
1062 | node = rb_next(node); |
1063 | if (!node) | |
1064 | break; | |
1065 | } | |
1066 | out: | |
1067 | return NULL; | |
1068 | } | |
d7fc640e | 1069 | |
d352ac68 CM |
1070 | /* |
1071 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1072 | * more than 'max_bytes'. start and end are used to return the range, | |
1073 | * | |
1074 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1075 | */ | |
c8b97818 CM |
1076 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
1077 | u64 *start, u64 *end, u64 max_bytes) | |
d1310b2e CM |
1078 | { |
1079 | struct rb_node *node; | |
1080 | struct extent_state *state; | |
1081 | u64 cur_start = *start; | |
1082 | u64 found = 0; | |
1083 | u64 total_bytes = 0; | |
1084 | ||
cad321ad | 1085 | spin_lock(&tree->lock); |
c8b97818 | 1086 | |
d1310b2e CM |
1087 | /* |
1088 | * this search will find all the extents that end after | |
1089 | * our range starts. | |
1090 | */ | |
80ea96b1 | 1091 | node = tree_search(tree, cur_start); |
2b114d1d | 1092 | if (!node) { |
3b951516 CM |
1093 | if (!found) |
1094 | *end = (u64)-1; | |
d1310b2e CM |
1095 | goto out; |
1096 | } | |
1097 | ||
d397712b | 1098 | while (1) { |
d1310b2e | 1099 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1100 | if (found && (state->start != cur_start || |
1101 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1102 | goto out; |
1103 | } | |
1104 | if (!(state->state & EXTENT_DELALLOC)) { | |
1105 | if (!found) | |
1106 | *end = state->end; | |
1107 | goto out; | |
1108 | } | |
d1310b2e CM |
1109 | if (!found) |
1110 | *start = state->start; | |
1111 | found++; | |
1112 | *end = state->end; | |
1113 | cur_start = state->end + 1; | |
1114 | node = rb_next(node); | |
1115 | if (!node) | |
1116 | break; | |
1117 | total_bytes += state->end - state->start + 1; | |
1118 | if (total_bytes >= max_bytes) | |
1119 | break; | |
1120 | } | |
1121 | out: | |
cad321ad | 1122 | spin_unlock(&tree->lock); |
d1310b2e CM |
1123 | return found; |
1124 | } | |
1125 | ||
c8b97818 CM |
1126 | static noinline int __unlock_for_delalloc(struct inode *inode, |
1127 | struct page *locked_page, | |
1128 | u64 start, u64 end) | |
1129 | { | |
1130 | int ret; | |
1131 | struct page *pages[16]; | |
1132 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1133 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1134 | unsigned long nr_pages = end_index - index + 1; | |
1135 | int i; | |
1136 | ||
1137 | if (index == locked_page->index && end_index == index) | |
1138 | return 0; | |
1139 | ||
d397712b | 1140 | while (nr_pages > 0) { |
c8b97818 | 1141 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1142 | min_t(unsigned long, nr_pages, |
1143 | ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1144 | for (i = 0; i < ret; i++) { |
1145 | if (pages[i] != locked_page) | |
1146 | unlock_page(pages[i]); | |
1147 | page_cache_release(pages[i]); | |
1148 | } | |
1149 | nr_pages -= ret; | |
1150 | index += ret; | |
1151 | cond_resched(); | |
1152 | } | |
1153 | return 0; | |
1154 | } | |
1155 | ||
1156 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1157 | struct page *locked_page, | |
1158 | u64 delalloc_start, | |
1159 | u64 delalloc_end) | |
1160 | { | |
1161 | unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT; | |
1162 | unsigned long start_index = index; | |
1163 | unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT; | |
1164 | unsigned long pages_locked = 0; | |
1165 | struct page *pages[16]; | |
1166 | unsigned long nrpages; | |
1167 | int ret; | |
1168 | int i; | |
1169 | ||
1170 | /* the caller is responsible for locking the start index */ | |
1171 | if (index == locked_page->index && index == end_index) | |
1172 | return 0; | |
1173 | ||
1174 | /* skip the page at the start index */ | |
1175 | nrpages = end_index - index + 1; | |
d397712b | 1176 | while (nrpages > 0) { |
c8b97818 | 1177 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1178 | min_t(unsigned long, |
1179 | nrpages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1180 | if (ret == 0) { |
1181 | ret = -EAGAIN; | |
1182 | goto done; | |
1183 | } | |
1184 | /* now we have an array of pages, lock them all */ | |
1185 | for (i = 0; i < ret; i++) { | |
1186 | /* | |
1187 | * the caller is taking responsibility for | |
1188 | * locked_page | |
1189 | */ | |
771ed689 | 1190 | if (pages[i] != locked_page) { |
c8b97818 | 1191 | lock_page(pages[i]); |
f2b1c41c CM |
1192 | if (!PageDirty(pages[i]) || |
1193 | pages[i]->mapping != inode->i_mapping) { | |
771ed689 CM |
1194 | ret = -EAGAIN; |
1195 | unlock_page(pages[i]); | |
1196 | page_cache_release(pages[i]); | |
1197 | goto done; | |
1198 | } | |
1199 | } | |
c8b97818 | 1200 | page_cache_release(pages[i]); |
771ed689 | 1201 | pages_locked++; |
c8b97818 | 1202 | } |
c8b97818 CM |
1203 | nrpages -= ret; |
1204 | index += ret; | |
1205 | cond_resched(); | |
1206 | } | |
1207 | ret = 0; | |
1208 | done: | |
1209 | if (ret && pages_locked) { | |
1210 | __unlock_for_delalloc(inode, locked_page, | |
1211 | delalloc_start, | |
1212 | ((u64)(start_index + pages_locked - 1)) << | |
1213 | PAGE_CACHE_SHIFT); | |
1214 | } | |
1215 | return ret; | |
1216 | } | |
1217 | ||
1218 | /* | |
1219 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1220 | * more than 'max_bytes'. start and end are used to return the range, | |
1221 | * | |
1222 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1223 | */ | |
1224 | static noinline u64 find_lock_delalloc_range(struct inode *inode, | |
1225 | struct extent_io_tree *tree, | |
1226 | struct page *locked_page, | |
1227 | u64 *start, u64 *end, | |
1228 | u64 max_bytes) | |
1229 | { | |
1230 | u64 delalloc_start; | |
1231 | u64 delalloc_end; | |
1232 | u64 found; | |
1233 | int ret; | |
1234 | int loops = 0; | |
1235 | ||
1236 | again: | |
1237 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1238 | delalloc_start = *start; | |
1239 | delalloc_end = 0; | |
1240 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
1241 | max_bytes); | |
70b99e69 | 1242 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1243 | *start = delalloc_start; |
1244 | *end = delalloc_end; | |
1245 | return found; | |
1246 | } | |
1247 | ||
70b99e69 CM |
1248 | /* |
1249 | * start comes from the offset of locked_page. We have to lock | |
1250 | * pages in order, so we can't process delalloc bytes before | |
1251 | * locked_page | |
1252 | */ | |
d397712b | 1253 | if (delalloc_start < *start) |
70b99e69 | 1254 | delalloc_start = *start; |
70b99e69 | 1255 | |
c8b97818 CM |
1256 | /* |
1257 | * make sure to limit the number of pages we try to lock down | |
1258 | * if we're looping. | |
1259 | */ | |
d397712b | 1260 | if (delalloc_end + 1 - delalloc_start > max_bytes && loops) |
771ed689 | 1261 | delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1; |
d397712b | 1262 | |
c8b97818 CM |
1263 | /* step two, lock all the pages after the page that has start */ |
1264 | ret = lock_delalloc_pages(inode, locked_page, | |
1265 | delalloc_start, delalloc_end); | |
1266 | if (ret == -EAGAIN) { | |
1267 | /* some of the pages are gone, lets avoid looping by | |
1268 | * shortening the size of the delalloc range we're searching | |
1269 | */ | |
1270 | if (!loops) { | |
1271 | unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1); | |
1272 | max_bytes = PAGE_CACHE_SIZE - offset; | |
1273 | loops = 1; | |
1274 | goto again; | |
1275 | } else { | |
1276 | found = 0; | |
1277 | goto out_failed; | |
1278 | } | |
1279 | } | |
1280 | BUG_ON(ret); | |
1281 | ||
1282 | /* step three, lock the state bits for the whole range */ | |
1283 | lock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS); | |
1284 | ||
1285 | /* then test to make sure it is all still delalloc */ | |
1286 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
1287 | EXTENT_DELALLOC, 1); | |
1288 | if (!ret) { | |
1289 | unlock_extent(tree, delalloc_start, delalloc_end, GFP_NOFS); | |
1290 | __unlock_for_delalloc(inode, locked_page, | |
1291 | delalloc_start, delalloc_end); | |
1292 | cond_resched(); | |
1293 | goto again; | |
1294 | } | |
1295 | *start = delalloc_start; | |
1296 | *end = delalloc_end; | |
1297 | out_failed: | |
1298 | return found; | |
1299 | } | |
1300 | ||
1301 | int extent_clear_unlock_delalloc(struct inode *inode, | |
1302 | struct extent_io_tree *tree, | |
1303 | u64 start, u64 end, struct page *locked_page, | |
771ed689 CM |
1304 | int unlock_pages, |
1305 | int clear_unlock, | |
1306 | int clear_delalloc, int clear_dirty, | |
1307 | int set_writeback, | |
c8b97818 CM |
1308 | int end_writeback) |
1309 | { | |
1310 | int ret; | |
1311 | struct page *pages[16]; | |
1312 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1313 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1314 | unsigned long nr_pages = end_index - index + 1; | |
1315 | int i; | |
771ed689 | 1316 | int clear_bits = 0; |
c8b97818 | 1317 | |
771ed689 CM |
1318 | if (clear_unlock) |
1319 | clear_bits |= EXTENT_LOCKED; | |
c8b97818 CM |
1320 | if (clear_dirty) |
1321 | clear_bits |= EXTENT_DIRTY; | |
1322 | ||
771ed689 CM |
1323 | if (clear_delalloc) |
1324 | clear_bits |= EXTENT_DELALLOC; | |
1325 | ||
c8b97818 | 1326 | clear_extent_bit(tree, start, end, clear_bits, 1, 0, GFP_NOFS); |
771ed689 CM |
1327 | if (!(unlock_pages || clear_dirty || set_writeback || end_writeback)) |
1328 | return 0; | |
c8b97818 | 1329 | |
d397712b | 1330 | while (nr_pages > 0) { |
c8b97818 | 1331 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1332 | min_t(unsigned long, |
1333 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1334 | for (i = 0; i < ret; i++) { |
1335 | if (pages[i] == locked_page) { | |
1336 | page_cache_release(pages[i]); | |
1337 | continue; | |
1338 | } | |
1339 | if (clear_dirty) | |
1340 | clear_page_dirty_for_io(pages[i]); | |
1341 | if (set_writeback) | |
1342 | set_page_writeback(pages[i]); | |
1343 | if (end_writeback) | |
1344 | end_page_writeback(pages[i]); | |
771ed689 CM |
1345 | if (unlock_pages) |
1346 | unlock_page(pages[i]); | |
c8b97818 CM |
1347 | page_cache_release(pages[i]); |
1348 | } | |
1349 | nr_pages -= ret; | |
1350 | index += ret; | |
1351 | cond_resched(); | |
1352 | } | |
1353 | return 0; | |
1354 | } | |
c8b97818 | 1355 | |
d352ac68 CM |
1356 | /* |
1357 | * count the number of bytes in the tree that have a given bit(s) | |
1358 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1359 | * cached. The total number found is returned. | |
1360 | */ | |
d1310b2e CM |
1361 | u64 count_range_bits(struct extent_io_tree *tree, |
1362 | u64 *start, u64 search_end, u64 max_bytes, | |
1363 | unsigned long bits) | |
1364 | { | |
1365 | struct rb_node *node; | |
1366 | struct extent_state *state; | |
1367 | u64 cur_start = *start; | |
1368 | u64 total_bytes = 0; | |
1369 | int found = 0; | |
1370 | ||
1371 | if (search_end <= cur_start) { | |
d1310b2e CM |
1372 | WARN_ON(1); |
1373 | return 0; | |
1374 | } | |
1375 | ||
cad321ad | 1376 | spin_lock(&tree->lock); |
d1310b2e CM |
1377 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1378 | total_bytes = tree->dirty_bytes; | |
1379 | goto out; | |
1380 | } | |
1381 | /* | |
1382 | * this search will find all the extents that end after | |
1383 | * our range starts. | |
1384 | */ | |
80ea96b1 | 1385 | node = tree_search(tree, cur_start); |
d397712b | 1386 | if (!node) |
d1310b2e | 1387 | goto out; |
d1310b2e | 1388 | |
d397712b | 1389 | while (1) { |
d1310b2e CM |
1390 | state = rb_entry(node, struct extent_state, rb_node); |
1391 | if (state->start > search_end) | |
1392 | break; | |
1393 | if (state->end >= cur_start && (state->state & bits)) { | |
1394 | total_bytes += min(search_end, state->end) + 1 - | |
1395 | max(cur_start, state->start); | |
1396 | if (total_bytes >= max_bytes) | |
1397 | break; | |
1398 | if (!found) { | |
1399 | *start = state->start; | |
1400 | found = 1; | |
1401 | } | |
1402 | } | |
1403 | node = rb_next(node); | |
1404 | if (!node) | |
1405 | break; | |
1406 | } | |
1407 | out: | |
cad321ad | 1408 | spin_unlock(&tree->lock); |
d1310b2e CM |
1409 | return total_bytes; |
1410 | } | |
b2950863 | 1411 | |
d352ac68 CM |
1412 | /* |
1413 | * set the private field for a given byte offset in the tree. If there isn't | |
1414 | * an extent_state there already, this does nothing. | |
1415 | */ | |
d1310b2e CM |
1416 | int set_state_private(struct extent_io_tree *tree, u64 start, u64 private) |
1417 | { | |
1418 | struct rb_node *node; | |
1419 | struct extent_state *state; | |
1420 | int ret = 0; | |
1421 | ||
cad321ad | 1422 | spin_lock(&tree->lock); |
d1310b2e CM |
1423 | /* |
1424 | * this search will find all the extents that end after | |
1425 | * our range starts. | |
1426 | */ | |
80ea96b1 | 1427 | node = tree_search(tree, start); |
2b114d1d | 1428 | if (!node) { |
d1310b2e CM |
1429 | ret = -ENOENT; |
1430 | goto out; | |
1431 | } | |
1432 | state = rb_entry(node, struct extent_state, rb_node); | |
1433 | if (state->start != start) { | |
1434 | ret = -ENOENT; | |
1435 | goto out; | |
1436 | } | |
1437 | state->private = private; | |
1438 | out: | |
cad321ad | 1439 | spin_unlock(&tree->lock); |
d1310b2e CM |
1440 | return ret; |
1441 | } | |
1442 | ||
1443 | int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private) | |
1444 | { | |
1445 | struct rb_node *node; | |
1446 | struct extent_state *state; | |
1447 | int ret = 0; | |
1448 | ||
cad321ad | 1449 | spin_lock(&tree->lock); |
d1310b2e CM |
1450 | /* |
1451 | * this search will find all the extents that end after | |
1452 | * our range starts. | |
1453 | */ | |
80ea96b1 | 1454 | node = tree_search(tree, start); |
2b114d1d | 1455 | if (!node) { |
d1310b2e CM |
1456 | ret = -ENOENT; |
1457 | goto out; | |
1458 | } | |
1459 | state = rb_entry(node, struct extent_state, rb_node); | |
1460 | if (state->start != start) { | |
1461 | ret = -ENOENT; | |
1462 | goto out; | |
1463 | } | |
1464 | *private = state->private; | |
1465 | out: | |
cad321ad | 1466 | spin_unlock(&tree->lock); |
d1310b2e CM |
1467 | return ret; |
1468 | } | |
1469 | ||
1470 | /* | |
1471 | * searches a range in the state tree for a given mask. | |
70dec807 | 1472 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1473 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1474 | * range is found set. | |
1475 | */ | |
1476 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
1477 | int bits, int filled) | |
1478 | { | |
1479 | struct extent_state *state = NULL; | |
1480 | struct rb_node *node; | |
1481 | int bitset = 0; | |
d1310b2e | 1482 | |
cad321ad | 1483 | spin_lock(&tree->lock); |
80ea96b1 | 1484 | node = tree_search(tree, start); |
d1310b2e CM |
1485 | while (node && start <= end) { |
1486 | state = rb_entry(node, struct extent_state, rb_node); | |
1487 | ||
1488 | if (filled && state->start > start) { | |
1489 | bitset = 0; | |
1490 | break; | |
1491 | } | |
1492 | ||
1493 | if (state->start > end) | |
1494 | break; | |
1495 | ||
1496 | if (state->state & bits) { | |
1497 | bitset = 1; | |
1498 | if (!filled) | |
1499 | break; | |
1500 | } else if (filled) { | |
1501 | bitset = 0; | |
1502 | break; | |
1503 | } | |
1504 | start = state->end + 1; | |
1505 | if (start > end) | |
1506 | break; | |
1507 | node = rb_next(node); | |
1508 | if (!node) { | |
1509 | if (filled) | |
1510 | bitset = 0; | |
1511 | break; | |
1512 | } | |
1513 | } | |
cad321ad | 1514 | spin_unlock(&tree->lock); |
d1310b2e CM |
1515 | return bitset; |
1516 | } | |
d1310b2e CM |
1517 | |
1518 | /* | |
1519 | * helper function to set a given page up to date if all the | |
1520 | * extents in the tree for that page are up to date | |
1521 | */ | |
1522 | static int check_page_uptodate(struct extent_io_tree *tree, | |
1523 | struct page *page) | |
1524 | { | |
1525 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1526 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
1527 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1)) | |
1528 | SetPageUptodate(page); | |
1529 | return 0; | |
1530 | } | |
1531 | ||
1532 | /* | |
1533 | * helper function to unlock a page if all the extents in the tree | |
1534 | * for that page are unlocked | |
1535 | */ | |
1536 | static int check_page_locked(struct extent_io_tree *tree, | |
1537 | struct page *page) | |
1538 | { | |
1539 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1540 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
1541 | if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0)) | |
1542 | unlock_page(page); | |
1543 | return 0; | |
1544 | } | |
1545 | ||
1546 | /* | |
1547 | * helper function to end page writeback if all the extents | |
1548 | * in the tree for that page are done with writeback | |
1549 | */ | |
1550 | static int check_page_writeback(struct extent_io_tree *tree, | |
1551 | struct page *page) | |
1552 | { | |
1edbb734 | 1553 | end_page_writeback(page); |
d1310b2e CM |
1554 | return 0; |
1555 | } | |
1556 | ||
1557 | /* lots and lots of room for performance fixes in the end_bio funcs */ | |
1558 | ||
1559 | /* | |
1560 | * after a writepage IO is done, we need to: | |
1561 | * clear the uptodate bits on error | |
1562 | * clear the writeback bits in the extent tree for this IO | |
1563 | * end_page_writeback if the page has no more pending IO | |
1564 | * | |
1565 | * Scheduling is not allowed, so the extent state tree is expected | |
1566 | * to have one and only one object corresponding to this IO. | |
1567 | */ | |
d1310b2e | 1568 | static void end_bio_extent_writepage(struct bio *bio, int err) |
d1310b2e | 1569 | { |
1259ab75 | 1570 | int uptodate = err == 0; |
d1310b2e | 1571 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
902b22f3 | 1572 | struct extent_io_tree *tree; |
d1310b2e CM |
1573 | u64 start; |
1574 | u64 end; | |
1575 | int whole_page; | |
1259ab75 | 1576 | int ret; |
d1310b2e | 1577 | |
d1310b2e CM |
1578 | do { |
1579 | struct page *page = bvec->bv_page; | |
902b22f3 DW |
1580 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
1581 | ||
d1310b2e CM |
1582 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
1583 | bvec->bv_offset; | |
1584 | end = start + bvec->bv_len - 1; | |
1585 | ||
1586 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
1587 | whole_page = 1; | |
1588 | else | |
1589 | whole_page = 0; | |
1590 | ||
1591 | if (--bvec >= bio->bi_io_vec) | |
1592 | prefetchw(&bvec->bv_page->flags); | |
1259ab75 CM |
1593 | if (tree->ops && tree->ops->writepage_end_io_hook) { |
1594 | ret = tree->ops->writepage_end_io_hook(page, start, | |
902b22f3 | 1595 | end, NULL, uptodate); |
1259ab75 CM |
1596 | if (ret) |
1597 | uptodate = 0; | |
1598 | } | |
1599 | ||
1600 | if (!uptodate && tree->ops && | |
1601 | tree->ops->writepage_io_failed_hook) { | |
1602 | ret = tree->ops->writepage_io_failed_hook(bio, page, | |
902b22f3 | 1603 | start, end, NULL); |
1259ab75 | 1604 | if (ret == 0) { |
1259ab75 CM |
1605 | uptodate = (err == 0); |
1606 | continue; | |
1607 | } | |
1608 | } | |
1609 | ||
d1310b2e | 1610 | if (!uptodate) { |
1edbb734 | 1611 | clear_extent_uptodate(tree, start, end, GFP_NOFS); |
d1310b2e CM |
1612 | ClearPageUptodate(page); |
1613 | SetPageError(page); | |
1614 | } | |
70dec807 | 1615 | |
d1310b2e CM |
1616 | if (whole_page) |
1617 | end_page_writeback(page); | |
1618 | else | |
1619 | check_page_writeback(tree, page); | |
d1310b2e | 1620 | } while (bvec >= bio->bi_io_vec); |
2b1f55b0 | 1621 | |
d1310b2e | 1622 | bio_put(bio); |
d1310b2e CM |
1623 | } |
1624 | ||
1625 | /* | |
1626 | * after a readpage IO is done, we need to: | |
1627 | * clear the uptodate bits on error | |
1628 | * set the uptodate bits if things worked | |
1629 | * set the page up to date if all extents in the tree are uptodate | |
1630 | * clear the lock bit in the extent tree | |
1631 | * unlock the page if there are no other extents locked for it | |
1632 | * | |
1633 | * Scheduling is not allowed, so the extent state tree is expected | |
1634 | * to have one and only one object corresponding to this IO. | |
1635 | */ | |
d1310b2e | 1636 | static void end_bio_extent_readpage(struct bio *bio, int err) |
d1310b2e CM |
1637 | { |
1638 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
1639 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
902b22f3 | 1640 | struct extent_io_tree *tree; |
d1310b2e CM |
1641 | u64 start; |
1642 | u64 end; | |
1643 | int whole_page; | |
1644 | int ret; | |
1645 | ||
d20f7043 CM |
1646 | if (err) |
1647 | uptodate = 0; | |
1648 | ||
d1310b2e CM |
1649 | do { |
1650 | struct page *page = bvec->bv_page; | |
902b22f3 DW |
1651 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
1652 | ||
d1310b2e CM |
1653 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
1654 | bvec->bv_offset; | |
1655 | end = start + bvec->bv_len - 1; | |
1656 | ||
1657 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
1658 | whole_page = 1; | |
1659 | else | |
1660 | whole_page = 0; | |
1661 | ||
1662 | if (--bvec >= bio->bi_io_vec) | |
1663 | prefetchw(&bvec->bv_page->flags); | |
1664 | ||
1665 | if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) { | |
70dec807 | 1666 | ret = tree->ops->readpage_end_io_hook(page, start, end, |
902b22f3 | 1667 | NULL); |
d1310b2e CM |
1668 | if (ret) |
1669 | uptodate = 0; | |
1670 | } | |
7e38326f CM |
1671 | if (!uptodate && tree->ops && |
1672 | tree->ops->readpage_io_failed_hook) { | |
1673 | ret = tree->ops->readpage_io_failed_hook(bio, page, | |
902b22f3 | 1674 | start, end, NULL); |
7e38326f | 1675 | if (ret == 0) { |
3b951516 CM |
1676 | uptodate = |
1677 | test_bit(BIO_UPTODATE, &bio->bi_flags); | |
d20f7043 CM |
1678 | if (err) |
1679 | uptodate = 0; | |
7e38326f CM |
1680 | continue; |
1681 | } | |
1682 | } | |
d1310b2e | 1683 | |
771ed689 | 1684 | if (uptodate) { |
902b22f3 DW |
1685 | set_extent_uptodate(tree, start, end, |
1686 | GFP_ATOMIC); | |
771ed689 | 1687 | } |
902b22f3 | 1688 | unlock_extent(tree, start, end, GFP_ATOMIC); |
d1310b2e | 1689 | |
70dec807 CM |
1690 | if (whole_page) { |
1691 | if (uptodate) { | |
1692 | SetPageUptodate(page); | |
1693 | } else { | |
1694 | ClearPageUptodate(page); | |
1695 | SetPageError(page); | |
1696 | } | |
d1310b2e | 1697 | unlock_page(page); |
70dec807 CM |
1698 | } else { |
1699 | if (uptodate) { | |
1700 | check_page_uptodate(tree, page); | |
1701 | } else { | |
1702 | ClearPageUptodate(page); | |
1703 | SetPageError(page); | |
1704 | } | |
d1310b2e | 1705 | check_page_locked(tree, page); |
70dec807 | 1706 | } |
d1310b2e CM |
1707 | } while (bvec >= bio->bi_io_vec); |
1708 | ||
1709 | bio_put(bio); | |
d1310b2e CM |
1710 | } |
1711 | ||
1712 | /* | |
1713 | * IO done from prepare_write is pretty simple, we just unlock | |
1714 | * the structs in the extent tree when done, and set the uptodate bits | |
1715 | * as appropriate. | |
1716 | */ | |
d1310b2e | 1717 | static void end_bio_extent_preparewrite(struct bio *bio, int err) |
d1310b2e CM |
1718 | { |
1719 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
1720 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
902b22f3 | 1721 | struct extent_io_tree *tree; |
d1310b2e CM |
1722 | u64 start; |
1723 | u64 end; | |
1724 | ||
d1310b2e CM |
1725 | do { |
1726 | struct page *page = bvec->bv_page; | |
902b22f3 DW |
1727 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
1728 | ||
d1310b2e CM |
1729 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
1730 | bvec->bv_offset; | |
1731 | end = start + bvec->bv_len - 1; | |
1732 | ||
1733 | if (--bvec >= bio->bi_io_vec) | |
1734 | prefetchw(&bvec->bv_page->flags); | |
1735 | ||
1736 | if (uptodate) { | |
1737 | set_extent_uptodate(tree, start, end, GFP_ATOMIC); | |
1738 | } else { | |
1739 | ClearPageUptodate(page); | |
1740 | SetPageError(page); | |
1741 | } | |
1742 | ||
1743 | unlock_extent(tree, start, end, GFP_ATOMIC); | |
1744 | ||
1745 | } while (bvec >= bio->bi_io_vec); | |
1746 | ||
1747 | bio_put(bio); | |
d1310b2e CM |
1748 | } |
1749 | ||
1750 | static struct bio * | |
1751 | extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
1752 | gfp_t gfp_flags) | |
1753 | { | |
1754 | struct bio *bio; | |
1755 | ||
1756 | bio = bio_alloc(gfp_flags, nr_vecs); | |
1757 | ||
1758 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
1759 | while (!bio && (nr_vecs /= 2)) | |
1760 | bio = bio_alloc(gfp_flags, nr_vecs); | |
1761 | } | |
1762 | ||
1763 | if (bio) { | |
e1c4b745 | 1764 | bio->bi_size = 0; |
d1310b2e CM |
1765 | bio->bi_bdev = bdev; |
1766 | bio->bi_sector = first_sector; | |
1767 | } | |
1768 | return bio; | |
1769 | } | |
1770 | ||
c8b97818 CM |
1771 | static int submit_one_bio(int rw, struct bio *bio, int mirror_num, |
1772 | unsigned long bio_flags) | |
d1310b2e | 1773 | { |
d1310b2e | 1774 | int ret = 0; |
70dec807 CM |
1775 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
1776 | struct page *page = bvec->bv_page; | |
1777 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 CM |
1778 | u64 start; |
1779 | u64 end; | |
1780 | ||
1781 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset; | |
1782 | end = start + bvec->bv_len - 1; | |
1783 | ||
902b22f3 | 1784 | bio->bi_private = NULL; |
d1310b2e CM |
1785 | |
1786 | bio_get(bio); | |
1787 | ||
065631f6 | 1788 | if (tree->ops && tree->ops->submit_bio_hook) |
f188591e | 1789 | tree->ops->submit_bio_hook(page->mapping->host, rw, bio, |
c8b97818 | 1790 | mirror_num, bio_flags); |
0b86a832 CM |
1791 | else |
1792 | submit_bio(rw, bio); | |
d1310b2e CM |
1793 | if (bio_flagged(bio, BIO_EOPNOTSUPP)) |
1794 | ret = -EOPNOTSUPP; | |
1795 | bio_put(bio); | |
1796 | return ret; | |
1797 | } | |
1798 | ||
1799 | static int submit_extent_page(int rw, struct extent_io_tree *tree, | |
1800 | struct page *page, sector_t sector, | |
1801 | size_t size, unsigned long offset, | |
1802 | struct block_device *bdev, | |
1803 | struct bio **bio_ret, | |
1804 | unsigned long max_pages, | |
f188591e | 1805 | bio_end_io_t end_io_func, |
c8b97818 CM |
1806 | int mirror_num, |
1807 | unsigned long prev_bio_flags, | |
1808 | unsigned long bio_flags) | |
d1310b2e CM |
1809 | { |
1810 | int ret = 0; | |
1811 | struct bio *bio; | |
1812 | int nr; | |
c8b97818 CM |
1813 | int contig = 0; |
1814 | int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED; | |
1815 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; | |
5b050f04 | 1816 | size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE); |
d1310b2e CM |
1817 | |
1818 | if (bio_ret && *bio_ret) { | |
1819 | bio = *bio_ret; | |
c8b97818 CM |
1820 | if (old_compressed) |
1821 | contig = bio->bi_sector == sector; | |
1822 | else | |
1823 | contig = bio->bi_sector + (bio->bi_size >> 9) == | |
1824 | sector; | |
1825 | ||
1826 | if (prev_bio_flags != bio_flags || !contig || | |
239b14b3 | 1827 | (tree->ops && tree->ops->merge_bio_hook && |
c8b97818 CM |
1828 | tree->ops->merge_bio_hook(page, offset, page_size, bio, |
1829 | bio_flags)) || | |
1830 | bio_add_page(bio, page, page_size, offset) < page_size) { | |
1831 | ret = submit_one_bio(rw, bio, mirror_num, | |
1832 | prev_bio_flags); | |
d1310b2e CM |
1833 | bio = NULL; |
1834 | } else { | |
1835 | return 0; | |
1836 | } | |
1837 | } | |
c8b97818 CM |
1838 | if (this_compressed) |
1839 | nr = BIO_MAX_PAGES; | |
1840 | else | |
1841 | nr = bio_get_nr_vecs(bdev); | |
1842 | ||
d1310b2e | 1843 | bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH); |
70dec807 | 1844 | |
c8b97818 | 1845 | bio_add_page(bio, page, page_size, offset); |
d1310b2e CM |
1846 | bio->bi_end_io = end_io_func; |
1847 | bio->bi_private = tree; | |
70dec807 | 1848 | |
d397712b | 1849 | if (bio_ret) |
d1310b2e | 1850 | *bio_ret = bio; |
d397712b | 1851 | else |
c8b97818 | 1852 | ret = submit_one_bio(rw, bio, mirror_num, bio_flags); |
d1310b2e CM |
1853 | |
1854 | return ret; | |
1855 | } | |
1856 | ||
1857 | void set_page_extent_mapped(struct page *page) | |
1858 | { | |
1859 | if (!PagePrivate(page)) { | |
1860 | SetPagePrivate(page); | |
d1310b2e | 1861 | page_cache_get(page); |
6af118ce | 1862 | set_page_private(page, EXTENT_PAGE_PRIVATE); |
d1310b2e CM |
1863 | } |
1864 | } | |
1865 | ||
b2950863 | 1866 | static void set_page_extent_head(struct page *page, unsigned long len) |
d1310b2e CM |
1867 | { |
1868 | set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2); | |
1869 | } | |
1870 | ||
1871 | /* | |
1872 | * basic readpage implementation. Locked extent state structs are inserted | |
1873 | * into the tree that are removed when the IO is done (by the end_io | |
1874 | * handlers) | |
1875 | */ | |
1876 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
1877 | struct page *page, | |
1878 | get_extent_t *get_extent, | |
c8b97818 CM |
1879 | struct bio **bio, int mirror_num, |
1880 | unsigned long *bio_flags) | |
d1310b2e CM |
1881 | { |
1882 | struct inode *inode = page->mapping->host; | |
1883 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1884 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
1885 | u64 end; | |
1886 | u64 cur = start; | |
1887 | u64 extent_offset; | |
1888 | u64 last_byte = i_size_read(inode); | |
1889 | u64 block_start; | |
1890 | u64 cur_end; | |
1891 | sector_t sector; | |
1892 | struct extent_map *em; | |
1893 | struct block_device *bdev; | |
1894 | int ret; | |
1895 | int nr = 0; | |
1896 | size_t page_offset = 0; | |
1897 | size_t iosize; | |
c8b97818 | 1898 | size_t disk_io_size; |
d1310b2e | 1899 | size_t blocksize = inode->i_sb->s_blocksize; |
c8b97818 | 1900 | unsigned long this_bio_flag = 0; |
d1310b2e CM |
1901 | |
1902 | set_page_extent_mapped(page); | |
1903 | ||
1904 | end = page_end; | |
1905 | lock_extent(tree, start, end, GFP_NOFS); | |
1906 | ||
c8b97818 CM |
1907 | if (page->index == last_byte >> PAGE_CACHE_SHIFT) { |
1908 | char *userpage; | |
1909 | size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1); | |
1910 | ||
1911 | if (zero_offset) { | |
1912 | iosize = PAGE_CACHE_SIZE - zero_offset; | |
1913 | userpage = kmap_atomic(page, KM_USER0); | |
1914 | memset(userpage + zero_offset, 0, iosize); | |
1915 | flush_dcache_page(page); | |
1916 | kunmap_atomic(userpage, KM_USER0); | |
1917 | } | |
1918 | } | |
d1310b2e CM |
1919 | while (cur <= end) { |
1920 | if (cur >= last_byte) { | |
1921 | char *userpage; | |
1922 | iosize = PAGE_CACHE_SIZE - page_offset; | |
1923 | userpage = kmap_atomic(page, KM_USER0); | |
1924 | memset(userpage + page_offset, 0, iosize); | |
1925 | flush_dcache_page(page); | |
1926 | kunmap_atomic(userpage, KM_USER0); | |
1927 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
1928 | GFP_NOFS); | |
1929 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
1930 | break; | |
1931 | } | |
1932 | em = get_extent(inode, page, page_offset, cur, | |
1933 | end - cur + 1, 0); | |
1934 | if (IS_ERR(em) || !em) { | |
1935 | SetPageError(page); | |
1936 | unlock_extent(tree, cur, end, GFP_NOFS); | |
1937 | break; | |
1938 | } | |
d1310b2e CM |
1939 | extent_offset = cur - em->start; |
1940 | BUG_ON(extent_map_end(em) <= cur); | |
1941 | BUG_ON(end < cur); | |
1942 | ||
c8b97818 CM |
1943 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
1944 | this_bio_flag = EXTENT_BIO_COMPRESSED; | |
1945 | ||
d1310b2e CM |
1946 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
1947 | cur_end = min(extent_map_end(em) - 1, end); | |
1948 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
c8b97818 CM |
1949 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
1950 | disk_io_size = em->block_len; | |
1951 | sector = em->block_start >> 9; | |
1952 | } else { | |
1953 | sector = (em->block_start + extent_offset) >> 9; | |
1954 | disk_io_size = iosize; | |
1955 | } | |
d1310b2e CM |
1956 | bdev = em->bdev; |
1957 | block_start = em->block_start; | |
d899e052 YZ |
1958 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
1959 | block_start = EXTENT_MAP_HOLE; | |
d1310b2e CM |
1960 | free_extent_map(em); |
1961 | em = NULL; | |
1962 | ||
1963 | /* we've found a hole, just zero and go on */ | |
1964 | if (block_start == EXTENT_MAP_HOLE) { | |
1965 | char *userpage; | |
1966 | userpage = kmap_atomic(page, KM_USER0); | |
1967 | memset(userpage + page_offset, 0, iosize); | |
1968 | flush_dcache_page(page); | |
1969 | kunmap_atomic(userpage, KM_USER0); | |
1970 | ||
1971 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
1972 | GFP_NOFS); | |
1973 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
1974 | cur = cur + iosize; | |
1975 | page_offset += iosize; | |
1976 | continue; | |
1977 | } | |
1978 | /* the get_extent function already copied into the page */ | |
1979 | if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) { | |
a1b32a59 | 1980 | check_page_uptodate(tree, page); |
d1310b2e CM |
1981 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); |
1982 | cur = cur + iosize; | |
1983 | page_offset += iosize; | |
1984 | continue; | |
1985 | } | |
70dec807 CM |
1986 | /* we have an inline extent but it didn't get marked up |
1987 | * to date. Error out | |
1988 | */ | |
1989 | if (block_start == EXTENT_MAP_INLINE) { | |
1990 | SetPageError(page); | |
1991 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
1992 | cur = cur + iosize; | |
1993 | page_offset += iosize; | |
1994 | continue; | |
1995 | } | |
d1310b2e CM |
1996 | |
1997 | ret = 0; | |
1998 | if (tree->ops && tree->ops->readpage_io_hook) { | |
1999 | ret = tree->ops->readpage_io_hook(page, cur, | |
2000 | cur + iosize - 1); | |
2001 | } | |
2002 | if (!ret) { | |
89642229 CM |
2003 | unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1; |
2004 | pnr -= page->index; | |
d1310b2e | 2005 | ret = submit_extent_page(READ, tree, page, |
c8b97818 | 2006 | sector, disk_io_size, page_offset, |
89642229 | 2007 | bdev, bio, pnr, |
c8b97818 CM |
2008 | end_bio_extent_readpage, mirror_num, |
2009 | *bio_flags, | |
2010 | this_bio_flag); | |
89642229 | 2011 | nr++; |
c8b97818 | 2012 | *bio_flags = this_bio_flag; |
d1310b2e CM |
2013 | } |
2014 | if (ret) | |
2015 | SetPageError(page); | |
2016 | cur = cur + iosize; | |
2017 | page_offset += iosize; | |
d1310b2e CM |
2018 | } |
2019 | if (!nr) { | |
2020 | if (!PageError(page)) | |
2021 | SetPageUptodate(page); | |
2022 | unlock_page(page); | |
2023 | } | |
2024 | return 0; | |
2025 | } | |
2026 | ||
2027 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, | |
2028 | get_extent_t *get_extent) | |
2029 | { | |
2030 | struct bio *bio = NULL; | |
c8b97818 | 2031 | unsigned long bio_flags = 0; |
d1310b2e CM |
2032 | int ret; |
2033 | ||
c8b97818 CM |
2034 | ret = __extent_read_full_page(tree, page, get_extent, &bio, 0, |
2035 | &bio_flags); | |
d1310b2e | 2036 | if (bio) |
c8b97818 | 2037 | submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
2038 | return ret; |
2039 | } | |
d1310b2e | 2040 | |
11c8349b CM |
2041 | static noinline void update_nr_written(struct page *page, |
2042 | struct writeback_control *wbc, | |
2043 | unsigned long nr_written) | |
2044 | { | |
2045 | wbc->nr_to_write -= nr_written; | |
2046 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && | |
2047 | wbc->range_start == 0 && wbc->range_end == LLONG_MAX)) | |
2048 | page->mapping->writeback_index = page->index + nr_written; | |
2049 | } | |
2050 | ||
d1310b2e CM |
2051 | /* |
2052 | * the writepage semantics are similar to regular writepage. extent | |
2053 | * records are inserted to lock ranges in the tree, and as dirty areas | |
2054 | * are found, they are marked writeback. Then the lock bits are removed | |
2055 | * and the end_io handler clears the writeback ranges | |
2056 | */ | |
2057 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
2058 | void *data) | |
2059 | { | |
2060 | struct inode *inode = page->mapping->host; | |
2061 | struct extent_page_data *epd = data; | |
2062 | struct extent_io_tree *tree = epd->tree; | |
2063 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2064 | u64 delalloc_start; | |
2065 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2066 | u64 end; | |
2067 | u64 cur = start; | |
2068 | u64 extent_offset; | |
2069 | u64 last_byte = i_size_read(inode); | |
2070 | u64 block_start; | |
2071 | u64 iosize; | |
e6dcd2dc | 2072 | u64 unlock_start; |
d1310b2e CM |
2073 | sector_t sector; |
2074 | struct extent_map *em; | |
2075 | struct block_device *bdev; | |
2076 | int ret; | |
2077 | int nr = 0; | |
7f3c74fb | 2078 | size_t pg_offset = 0; |
d1310b2e CM |
2079 | size_t blocksize; |
2080 | loff_t i_size = i_size_read(inode); | |
2081 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
2082 | u64 nr_delalloc; | |
2083 | u64 delalloc_end; | |
c8b97818 CM |
2084 | int page_started; |
2085 | int compressed; | |
ffbd517d | 2086 | int write_flags; |
771ed689 | 2087 | unsigned long nr_written = 0; |
d1310b2e | 2088 | |
ffbd517d CM |
2089 | if (wbc->sync_mode == WB_SYNC_ALL) |
2090 | write_flags = WRITE_SYNC_PLUG; | |
2091 | else | |
2092 | write_flags = WRITE; | |
2093 | ||
d1310b2e | 2094 | WARN_ON(!PageLocked(page)); |
7f3c74fb | 2095 | pg_offset = i_size & (PAGE_CACHE_SIZE - 1); |
211c17f5 | 2096 | if (page->index > end_index || |
7f3c74fb | 2097 | (page->index == end_index && !pg_offset)) { |
39be25cd | 2098 | page->mapping->a_ops->invalidatepage(page, 0); |
d1310b2e CM |
2099 | unlock_page(page); |
2100 | return 0; | |
2101 | } | |
2102 | ||
2103 | if (page->index == end_index) { | |
2104 | char *userpage; | |
2105 | ||
d1310b2e | 2106 | userpage = kmap_atomic(page, KM_USER0); |
7f3c74fb CM |
2107 | memset(userpage + pg_offset, 0, |
2108 | PAGE_CACHE_SIZE - pg_offset); | |
d1310b2e | 2109 | kunmap_atomic(userpage, KM_USER0); |
211c17f5 | 2110 | flush_dcache_page(page); |
d1310b2e | 2111 | } |
7f3c74fb | 2112 | pg_offset = 0; |
d1310b2e CM |
2113 | |
2114 | set_page_extent_mapped(page); | |
2115 | ||
2116 | delalloc_start = start; | |
2117 | delalloc_end = 0; | |
c8b97818 | 2118 | page_started = 0; |
771ed689 | 2119 | if (!epd->extent_locked) { |
a97adc9f | 2120 | u64 delalloc_to_write; |
11c8349b CM |
2121 | /* |
2122 | * make sure the wbc mapping index is at least updated | |
2123 | * to this page. | |
2124 | */ | |
2125 | update_nr_written(page, wbc, 0); | |
2126 | ||
d397712b | 2127 | while (delalloc_end < page_end) { |
771ed689 | 2128 | nr_delalloc = find_lock_delalloc_range(inode, tree, |
c8b97818 CM |
2129 | page, |
2130 | &delalloc_start, | |
d1310b2e CM |
2131 | &delalloc_end, |
2132 | 128 * 1024 * 1024); | |
771ed689 CM |
2133 | if (nr_delalloc == 0) { |
2134 | delalloc_start = delalloc_end + 1; | |
2135 | continue; | |
2136 | } | |
2137 | tree->ops->fill_delalloc(inode, page, delalloc_start, | |
2138 | delalloc_end, &page_started, | |
2139 | &nr_written); | |
a97adc9f CM |
2140 | delalloc_to_write = (delalloc_end - |
2141 | max_t(u64, page_offset(page), | |
2142 | delalloc_start) + 1) >> | |
2143 | PAGE_CACHE_SHIFT; | |
2144 | if (wbc->nr_to_write < delalloc_to_write) { | |
2145 | wbc->nr_to_write = min_t(long, 8192, | |
2146 | delalloc_to_write); | |
2147 | } | |
d1310b2e | 2148 | delalloc_start = delalloc_end + 1; |
d1310b2e | 2149 | } |
c8b97818 | 2150 | |
771ed689 CM |
2151 | /* did the fill delalloc function already unlock and start |
2152 | * the IO? | |
2153 | */ | |
2154 | if (page_started) { | |
2155 | ret = 0; | |
11c8349b CM |
2156 | /* |
2157 | * we've unlocked the page, so we can't update | |
2158 | * the mapping's writeback index, just update | |
2159 | * nr_to_write. | |
2160 | */ | |
2161 | wbc->nr_to_write -= nr_written; | |
2162 | goto done_unlocked; | |
771ed689 | 2163 | } |
c8b97818 | 2164 | } |
d1310b2e | 2165 | lock_extent(tree, start, page_end, GFP_NOFS); |
771ed689 | 2166 | |
e6dcd2dc | 2167 | unlock_start = start; |
d1310b2e | 2168 | |
247e743c | 2169 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
2170 | ret = tree->ops->writepage_start_hook(page, start, |
2171 | page_end); | |
247e743c CM |
2172 | if (ret == -EAGAIN) { |
2173 | unlock_extent(tree, start, page_end, GFP_NOFS); | |
2174 | redirty_page_for_writepage(wbc, page); | |
11c8349b | 2175 | update_nr_written(page, wbc, nr_written); |
247e743c | 2176 | unlock_page(page); |
771ed689 | 2177 | ret = 0; |
11c8349b | 2178 | goto done_unlocked; |
247e743c CM |
2179 | } |
2180 | } | |
2181 | ||
11c8349b CM |
2182 | /* |
2183 | * we don't want to touch the inode after unlocking the page, | |
2184 | * so we update the mapping writeback index now | |
2185 | */ | |
2186 | update_nr_written(page, wbc, nr_written + 1); | |
771ed689 | 2187 | |
d1310b2e | 2188 | end = page_end; |
d397712b CM |
2189 | if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) |
2190 | printk(KERN_ERR "btrfs delalloc bits after lock_extent\n"); | |
d1310b2e CM |
2191 | |
2192 | if (last_byte <= start) { | |
1edbb734 CM |
2193 | clear_extent_bit(tree, start, page_end, |
2194 | EXTENT_LOCKED | EXTENT_DIRTY, | |
2195 | 1, 0, GFP_NOFS); | |
e6dcd2dc CM |
2196 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2197 | tree->ops->writepage_end_io_hook(page, start, | |
2198 | page_end, NULL, 1); | |
2199 | unlock_start = page_end + 1; | |
d1310b2e CM |
2200 | goto done; |
2201 | } | |
2202 | ||
d1310b2e CM |
2203 | blocksize = inode->i_sb->s_blocksize; |
2204 | ||
2205 | while (cur <= end) { | |
2206 | if (cur >= last_byte) { | |
e6dcd2dc CM |
2207 | unlock_extent(tree, unlock_start, page_end, GFP_NOFS); |
2208 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
2209 | tree->ops->writepage_end_io_hook(page, cur, | |
2210 | page_end, NULL, 1); | |
2211 | unlock_start = page_end + 1; | |
d1310b2e CM |
2212 | break; |
2213 | } | |
7f3c74fb | 2214 | em = epd->get_extent(inode, page, pg_offset, cur, |
d1310b2e CM |
2215 | end - cur + 1, 1); |
2216 | if (IS_ERR(em) || !em) { | |
2217 | SetPageError(page); | |
2218 | break; | |
2219 | } | |
2220 | ||
2221 | extent_offset = cur - em->start; | |
2222 | BUG_ON(extent_map_end(em) <= cur); | |
2223 | BUG_ON(end < cur); | |
2224 | iosize = min(extent_map_end(em) - cur, end - cur + 1); | |
2225 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
2226 | sector = (em->block_start + extent_offset) >> 9; | |
2227 | bdev = em->bdev; | |
2228 | block_start = em->block_start; | |
c8b97818 | 2229 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
2230 | free_extent_map(em); |
2231 | em = NULL; | |
2232 | ||
c8b97818 CM |
2233 | /* |
2234 | * compressed and inline extents are written through other | |
2235 | * paths in the FS | |
2236 | */ | |
2237 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 2238 | block_start == EXTENT_MAP_INLINE) { |
d397712b | 2239 | unlock_extent(tree, unlock_start, cur + iosize - 1, |
e6dcd2dc | 2240 | GFP_NOFS); |
7f3c74fb | 2241 | |
c8b97818 CM |
2242 | /* |
2243 | * end_io notification does not happen here for | |
2244 | * compressed extents | |
2245 | */ | |
2246 | if (!compressed && tree->ops && | |
2247 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
2248 | tree->ops->writepage_end_io_hook(page, cur, |
2249 | cur + iosize - 1, | |
2250 | NULL, 1); | |
c8b97818 CM |
2251 | else if (compressed) { |
2252 | /* we don't want to end_page_writeback on | |
2253 | * a compressed extent. this happens | |
2254 | * elsewhere | |
2255 | */ | |
2256 | nr++; | |
2257 | } | |
2258 | ||
2259 | cur += iosize; | |
7f3c74fb | 2260 | pg_offset += iosize; |
e6dcd2dc | 2261 | unlock_start = cur; |
d1310b2e CM |
2262 | continue; |
2263 | } | |
d1310b2e CM |
2264 | /* leave this out until we have a page_mkwrite call */ |
2265 | if (0 && !test_range_bit(tree, cur, cur + iosize - 1, | |
2266 | EXTENT_DIRTY, 0)) { | |
2267 | cur = cur + iosize; | |
7f3c74fb | 2268 | pg_offset += iosize; |
d1310b2e CM |
2269 | continue; |
2270 | } | |
c8b97818 | 2271 | |
d1310b2e CM |
2272 | if (tree->ops && tree->ops->writepage_io_hook) { |
2273 | ret = tree->ops->writepage_io_hook(page, cur, | |
2274 | cur + iosize - 1); | |
2275 | } else { | |
2276 | ret = 0; | |
2277 | } | |
1259ab75 | 2278 | if (ret) { |
d1310b2e | 2279 | SetPageError(page); |
1259ab75 | 2280 | } else { |
d1310b2e | 2281 | unsigned long max_nr = end_index + 1; |
7f3c74fb | 2282 | |
d1310b2e CM |
2283 | set_range_writeback(tree, cur, cur + iosize - 1); |
2284 | if (!PageWriteback(page)) { | |
d397712b CM |
2285 | printk(KERN_ERR "btrfs warning page %lu not " |
2286 | "writeback, cur %llu end %llu\n", | |
2287 | page->index, (unsigned long long)cur, | |
d1310b2e CM |
2288 | (unsigned long long)end); |
2289 | } | |
2290 | ||
ffbd517d CM |
2291 | ret = submit_extent_page(write_flags, tree, page, |
2292 | sector, iosize, pg_offset, | |
2293 | bdev, &epd->bio, max_nr, | |
c8b97818 CM |
2294 | end_bio_extent_writepage, |
2295 | 0, 0, 0); | |
d1310b2e CM |
2296 | if (ret) |
2297 | SetPageError(page); | |
2298 | } | |
2299 | cur = cur + iosize; | |
7f3c74fb | 2300 | pg_offset += iosize; |
d1310b2e CM |
2301 | nr++; |
2302 | } | |
2303 | done: | |
2304 | if (nr == 0) { | |
2305 | /* make sure the mapping tag for page dirty gets cleared */ | |
2306 | set_page_writeback(page); | |
2307 | end_page_writeback(page); | |
2308 | } | |
e6dcd2dc CM |
2309 | if (unlock_start <= page_end) |
2310 | unlock_extent(tree, unlock_start, page_end, GFP_NOFS); | |
d1310b2e | 2311 | unlock_page(page); |
771ed689 | 2312 | |
11c8349b CM |
2313 | done_unlocked: |
2314 | ||
d1310b2e CM |
2315 | return 0; |
2316 | } | |
2317 | ||
d1310b2e | 2318 | /** |
4bef0848 | 2319 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
2320 | * @mapping: address space structure to write |
2321 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
2322 | * @writepage: function called for each page | |
2323 | * @data: data passed to writepage function | |
2324 | * | |
2325 | * If a page is already under I/O, write_cache_pages() skips it, even | |
2326 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
2327 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
2328 | * and msync() need to guarantee that all the data which was dirty at the time | |
2329 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
2330 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
2331 | * existing IO to complete. | |
2332 | */ | |
b2950863 | 2333 | static int extent_write_cache_pages(struct extent_io_tree *tree, |
4bef0848 CM |
2334 | struct address_space *mapping, |
2335 | struct writeback_control *wbc, | |
d2c3f4f6 CM |
2336 | writepage_t writepage, void *data, |
2337 | void (*flush_fn)(void *)) | |
d1310b2e | 2338 | { |
d1310b2e CM |
2339 | int ret = 0; |
2340 | int done = 0; | |
2341 | struct pagevec pvec; | |
2342 | int nr_pages; | |
2343 | pgoff_t index; | |
2344 | pgoff_t end; /* Inclusive */ | |
2345 | int scanned = 0; | |
2346 | int range_whole = 0; | |
2347 | ||
d1310b2e CM |
2348 | pagevec_init(&pvec, 0); |
2349 | if (wbc->range_cyclic) { | |
2350 | index = mapping->writeback_index; /* Start from prev offset */ | |
2351 | end = -1; | |
2352 | } else { | |
2353 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
2354 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
2355 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) | |
2356 | range_whole = 1; | |
2357 | scanned = 1; | |
2358 | } | |
2359 | retry: | |
2360 | while (!done && (index <= end) && | |
2361 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
d397712b CM |
2362 | PAGECACHE_TAG_DIRTY, min(end - index, |
2363 | (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
d1310b2e CM |
2364 | unsigned i; |
2365 | ||
2366 | scanned = 1; | |
2367 | for (i = 0; i < nr_pages; i++) { | |
2368 | struct page *page = pvec.pages[i]; | |
2369 | ||
2370 | /* | |
2371 | * At this point we hold neither mapping->tree_lock nor | |
2372 | * lock on the page itself: the page may be truncated or | |
2373 | * invalidated (changing page->mapping to NULL), or even | |
2374 | * swizzled back from swapper_space to tmpfs file | |
2375 | * mapping | |
2376 | */ | |
4bef0848 CM |
2377 | if (tree->ops && tree->ops->write_cache_pages_lock_hook) |
2378 | tree->ops->write_cache_pages_lock_hook(page); | |
2379 | else | |
2380 | lock_page(page); | |
d1310b2e CM |
2381 | |
2382 | if (unlikely(page->mapping != mapping)) { | |
2383 | unlock_page(page); | |
2384 | continue; | |
2385 | } | |
2386 | ||
2387 | if (!wbc->range_cyclic && page->index > end) { | |
2388 | done = 1; | |
2389 | unlock_page(page); | |
2390 | continue; | |
2391 | } | |
2392 | ||
d2c3f4f6 | 2393 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 CM |
2394 | if (PageWriteback(page)) |
2395 | flush_fn(data); | |
d1310b2e | 2396 | wait_on_page_writeback(page); |
d2c3f4f6 | 2397 | } |
d1310b2e CM |
2398 | |
2399 | if (PageWriteback(page) || | |
2400 | !clear_page_dirty_for_io(page)) { | |
2401 | unlock_page(page); | |
2402 | continue; | |
2403 | } | |
2404 | ||
2405 | ret = (*writepage)(page, wbc, data); | |
2406 | ||
2407 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
2408 | unlock_page(page); | |
2409 | ret = 0; | |
2410 | } | |
771ed689 | 2411 | if (ret || wbc->nr_to_write <= 0) |
d1310b2e | 2412 | done = 1; |
d1310b2e CM |
2413 | } |
2414 | pagevec_release(&pvec); | |
2415 | cond_resched(); | |
2416 | } | |
2417 | if (!scanned && !done) { | |
2418 | /* | |
2419 | * We hit the last page and there is more work to be done: wrap | |
2420 | * back to the start of the file | |
2421 | */ | |
2422 | scanned = 1; | |
2423 | index = 0; | |
2424 | goto retry; | |
2425 | } | |
d1310b2e CM |
2426 | return ret; |
2427 | } | |
d1310b2e | 2428 | |
ffbd517d | 2429 | static void flush_epd_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 2430 | { |
d2c3f4f6 | 2431 | if (epd->bio) { |
ffbd517d CM |
2432 | if (epd->sync_io) |
2433 | submit_one_bio(WRITE_SYNC, epd->bio, 0, 0); | |
2434 | else | |
2435 | submit_one_bio(WRITE, epd->bio, 0, 0); | |
d2c3f4f6 CM |
2436 | epd->bio = NULL; |
2437 | } | |
2438 | } | |
2439 | ||
ffbd517d CM |
2440 | static noinline void flush_write_bio(void *data) |
2441 | { | |
2442 | struct extent_page_data *epd = data; | |
2443 | flush_epd_write_bio(epd); | |
2444 | } | |
2445 | ||
d1310b2e CM |
2446 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, |
2447 | get_extent_t *get_extent, | |
2448 | struct writeback_control *wbc) | |
2449 | { | |
2450 | int ret; | |
2451 | struct address_space *mapping = page->mapping; | |
2452 | struct extent_page_data epd = { | |
2453 | .bio = NULL, | |
2454 | .tree = tree, | |
2455 | .get_extent = get_extent, | |
771ed689 | 2456 | .extent_locked = 0, |
ffbd517d | 2457 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
2458 | }; |
2459 | struct writeback_control wbc_writepages = { | |
2460 | .bdi = wbc->bdi, | |
d313d7a3 | 2461 | .sync_mode = wbc->sync_mode, |
d1310b2e CM |
2462 | .older_than_this = NULL, |
2463 | .nr_to_write = 64, | |
2464 | .range_start = page_offset(page) + PAGE_CACHE_SIZE, | |
2465 | .range_end = (loff_t)-1, | |
2466 | }; | |
2467 | ||
d1310b2e CM |
2468 | ret = __extent_writepage(page, wbc, &epd); |
2469 | ||
4bef0848 | 2470 | extent_write_cache_pages(tree, mapping, &wbc_writepages, |
d2c3f4f6 | 2471 | __extent_writepage, &epd, flush_write_bio); |
ffbd517d | 2472 | flush_epd_write_bio(&epd); |
d1310b2e CM |
2473 | return ret; |
2474 | } | |
d1310b2e | 2475 | |
771ed689 CM |
2476 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, |
2477 | u64 start, u64 end, get_extent_t *get_extent, | |
2478 | int mode) | |
2479 | { | |
2480 | int ret = 0; | |
2481 | struct address_space *mapping = inode->i_mapping; | |
2482 | struct page *page; | |
2483 | unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >> | |
2484 | PAGE_CACHE_SHIFT; | |
2485 | ||
2486 | struct extent_page_data epd = { | |
2487 | .bio = NULL, | |
2488 | .tree = tree, | |
2489 | .get_extent = get_extent, | |
2490 | .extent_locked = 1, | |
ffbd517d | 2491 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
2492 | }; |
2493 | struct writeback_control wbc_writepages = { | |
2494 | .bdi = inode->i_mapping->backing_dev_info, | |
2495 | .sync_mode = mode, | |
2496 | .older_than_this = NULL, | |
2497 | .nr_to_write = nr_pages * 2, | |
2498 | .range_start = start, | |
2499 | .range_end = end + 1, | |
2500 | }; | |
2501 | ||
d397712b | 2502 | while (start <= end) { |
771ed689 CM |
2503 | page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); |
2504 | if (clear_page_dirty_for_io(page)) | |
2505 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
2506 | else { | |
2507 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
2508 | tree->ops->writepage_end_io_hook(page, start, | |
2509 | start + PAGE_CACHE_SIZE - 1, | |
2510 | NULL, 1); | |
2511 | unlock_page(page); | |
2512 | } | |
2513 | page_cache_release(page); | |
2514 | start += PAGE_CACHE_SIZE; | |
2515 | } | |
2516 | ||
ffbd517d | 2517 | flush_epd_write_bio(&epd); |
771ed689 CM |
2518 | return ret; |
2519 | } | |
d1310b2e CM |
2520 | |
2521 | int extent_writepages(struct extent_io_tree *tree, | |
2522 | struct address_space *mapping, | |
2523 | get_extent_t *get_extent, | |
2524 | struct writeback_control *wbc) | |
2525 | { | |
2526 | int ret = 0; | |
2527 | struct extent_page_data epd = { | |
2528 | .bio = NULL, | |
2529 | .tree = tree, | |
2530 | .get_extent = get_extent, | |
771ed689 | 2531 | .extent_locked = 0, |
ffbd517d | 2532 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
2533 | }; |
2534 | ||
4bef0848 | 2535 | ret = extent_write_cache_pages(tree, mapping, wbc, |
d2c3f4f6 CM |
2536 | __extent_writepage, &epd, |
2537 | flush_write_bio); | |
ffbd517d | 2538 | flush_epd_write_bio(&epd); |
d1310b2e CM |
2539 | return ret; |
2540 | } | |
d1310b2e CM |
2541 | |
2542 | int extent_readpages(struct extent_io_tree *tree, | |
2543 | struct address_space *mapping, | |
2544 | struct list_head *pages, unsigned nr_pages, | |
2545 | get_extent_t get_extent) | |
2546 | { | |
2547 | struct bio *bio = NULL; | |
2548 | unsigned page_idx; | |
2549 | struct pagevec pvec; | |
c8b97818 | 2550 | unsigned long bio_flags = 0; |
d1310b2e CM |
2551 | |
2552 | pagevec_init(&pvec, 0); | |
2553 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { | |
2554 | struct page *page = list_entry(pages->prev, struct page, lru); | |
2555 | ||
2556 | prefetchw(&page->flags); | |
2557 | list_del(&page->lru); | |
2558 | /* | |
2559 | * what we want to do here is call add_to_page_cache_lru, | |
2560 | * but that isn't exported, so we reproduce it here | |
2561 | */ | |
2562 | if (!add_to_page_cache(page, mapping, | |
2563 | page->index, GFP_KERNEL)) { | |
2564 | ||
2565 | /* open coding of lru_cache_add, also not exported */ | |
2566 | page_cache_get(page); | |
2567 | if (!pagevec_add(&pvec, page)) | |
15916de8 | 2568 | __pagevec_lru_add_file(&pvec); |
f188591e | 2569 | __extent_read_full_page(tree, page, get_extent, |
c8b97818 | 2570 | &bio, 0, &bio_flags); |
d1310b2e CM |
2571 | } |
2572 | page_cache_release(page); | |
2573 | } | |
2574 | if (pagevec_count(&pvec)) | |
15916de8 | 2575 | __pagevec_lru_add_file(&pvec); |
d1310b2e CM |
2576 | BUG_ON(!list_empty(pages)); |
2577 | if (bio) | |
c8b97818 | 2578 | submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
2579 | return 0; |
2580 | } | |
d1310b2e CM |
2581 | |
2582 | /* | |
2583 | * basic invalidatepage code, this waits on any locked or writeback | |
2584 | * ranges corresponding to the page, and then deletes any extent state | |
2585 | * records from the tree | |
2586 | */ | |
2587 | int extent_invalidatepage(struct extent_io_tree *tree, | |
2588 | struct page *page, unsigned long offset) | |
2589 | { | |
2590 | u64 start = ((u64)page->index << PAGE_CACHE_SHIFT); | |
2591 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
2592 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
2593 | ||
d397712b | 2594 | start += (offset + blocksize - 1) & ~(blocksize - 1); |
d1310b2e CM |
2595 | if (start > end) |
2596 | return 0; | |
2597 | ||
2598 | lock_extent(tree, start, end, GFP_NOFS); | |
1edbb734 | 2599 | wait_on_page_writeback(page); |
d1310b2e CM |
2600 | clear_extent_bit(tree, start, end, |
2601 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC, | |
2602 | 1, 1, GFP_NOFS); | |
2603 | return 0; | |
2604 | } | |
d1310b2e CM |
2605 | |
2606 | /* | |
2607 | * simple commit_write call, set_range_dirty is used to mark both | |
2608 | * the pages and the extent records as dirty | |
2609 | */ | |
2610 | int extent_commit_write(struct extent_io_tree *tree, | |
2611 | struct inode *inode, struct page *page, | |
2612 | unsigned from, unsigned to) | |
2613 | { | |
2614 | loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to; | |
2615 | ||
2616 | set_page_extent_mapped(page); | |
2617 | set_page_dirty(page); | |
2618 | ||
2619 | if (pos > inode->i_size) { | |
2620 | i_size_write(inode, pos); | |
2621 | mark_inode_dirty(inode); | |
2622 | } | |
2623 | return 0; | |
2624 | } | |
d1310b2e CM |
2625 | |
2626 | int extent_prepare_write(struct extent_io_tree *tree, | |
2627 | struct inode *inode, struct page *page, | |
2628 | unsigned from, unsigned to, get_extent_t *get_extent) | |
2629 | { | |
2630 | u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2631 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
2632 | u64 block_start; | |
2633 | u64 orig_block_start; | |
2634 | u64 block_end; | |
2635 | u64 cur_end; | |
2636 | struct extent_map *em; | |
2637 | unsigned blocksize = 1 << inode->i_blkbits; | |
2638 | size_t page_offset = 0; | |
2639 | size_t block_off_start; | |
2640 | size_t block_off_end; | |
2641 | int err = 0; | |
2642 | int iocount = 0; | |
2643 | int ret = 0; | |
2644 | int isnew; | |
2645 | ||
2646 | set_page_extent_mapped(page); | |
2647 | ||
2648 | block_start = (page_start + from) & ~((u64)blocksize - 1); | |
2649 | block_end = (page_start + to - 1) | (blocksize - 1); | |
2650 | orig_block_start = block_start; | |
2651 | ||
2652 | lock_extent(tree, page_start, page_end, GFP_NOFS); | |
d397712b | 2653 | while (block_start <= block_end) { |
d1310b2e CM |
2654 | em = get_extent(inode, page, page_offset, block_start, |
2655 | block_end - block_start + 1, 1); | |
d397712b | 2656 | if (IS_ERR(em) || !em) |
d1310b2e | 2657 | goto err; |
d397712b | 2658 | |
d1310b2e CM |
2659 | cur_end = min(block_end, extent_map_end(em) - 1); |
2660 | block_off_start = block_start & (PAGE_CACHE_SIZE - 1); | |
2661 | block_off_end = block_off_start + blocksize; | |
2662 | isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS); | |
2663 | ||
2664 | if (!PageUptodate(page) && isnew && | |
2665 | (block_off_end > to || block_off_start < from)) { | |
2666 | void *kaddr; | |
2667 | ||
2668 | kaddr = kmap_atomic(page, KM_USER0); | |
2669 | if (block_off_end > to) | |
2670 | memset(kaddr + to, 0, block_off_end - to); | |
2671 | if (block_off_start < from) | |
2672 | memset(kaddr + block_off_start, 0, | |
2673 | from - block_off_start); | |
2674 | flush_dcache_page(page); | |
2675 | kunmap_atomic(kaddr, KM_USER0); | |
2676 | } | |
2677 | if ((em->block_start != EXTENT_MAP_HOLE && | |
2678 | em->block_start != EXTENT_MAP_INLINE) && | |
2679 | !isnew && !PageUptodate(page) && | |
2680 | (block_off_end > to || block_off_start < from) && | |
2681 | !test_range_bit(tree, block_start, cur_end, | |
2682 | EXTENT_UPTODATE, 1)) { | |
2683 | u64 sector; | |
2684 | u64 extent_offset = block_start - em->start; | |
2685 | size_t iosize; | |
2686 | sector = (em->block_start + extent_offset) >> 9; | |
2687 | iosize = (cur_end - block_start + blocksize) & | |
2688 | ~((u64)blocksize - 1); | |
2689 | /* | |
2690 | * we've already got the extent locked, but we | |
2691 | * need to split the state such that our end_bio | |
2692 | * handler can clear the lock. | |
2693 | */ | |
2694 | set_extent_bit(tree, block_start, | |
2695 | block_start + iosize - 1, | |
2696 | EXTENT_LOCKED, 0, NULL, GFP_NOFS); | |
2697 | ret = submit_extent_page(READ, tree, page, | |
2698 | sector, iosize, page_offset, em->bdev, | |
2699 | NULL, 1, | |
c8b97818 CM |
2700 | end_bio_extent_preparewrite, 0, |
2701 | 0, 0); | |
d1310b2e CM |
2702 | iocount++; |
2703 | block_start = block_start + iosize; | |
2704 | } else { | |
2705 | set_extent_uptodate(tree, block_start, cur_end, | |
2706 | GFP_NOFS); | |
2707 | unlock_extent(tree, block_start, cur_end, GFP_NOFS); | |
2708 | block_start = cur_end + 1; | |
2709 | } | |
2710 | page_offset = block_start & (PAGE_CACHE_SIZE - 1); | |
2711 | free_extent_map(em); | |
2712 | } | |
2713 | if (iocount) { | |
2714 | wait_extent_bit(tree, orig_block_start, | |
2715 | block_end, EXTENT_LOCKED); | |
2716 | } | |
2717 | check_page_uptodate(tree, page); | |
2718 | err: | |
2719 | /* FIXME, zero out newly allocated blocks on error */ | |
2720 | return err; | |
2721 | } | |
d1310b2e | 2722 | |
7b13b7b1 CM |
2723 | /* |
2724 | * a helper for releasepage, this tests for areas of the page that | |
2725 | * are locked or under IO and drops the related state bits if it is safe | |
2726 | * to drop the page. | |
2727 | */ | |
2728 | int try_release_extent_state(struct extent_map_tree *map, | |
2729 | struct extent_io_tree *tree, struct page *page, | |
2730 | gfp_t mask) | |
2731 | { | |
2732 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2733 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
2734 | int ret = 1; | |
2735 | ||
211f90e6 CM |
2736 | if (test_range_bit(tree, start, end, |
2737 | EXTENT_IOBITS | EXTENT_ORDERED, 0)) | |
7b13b7b1 CM |
2738 | ret = 0; |
2739 | else { | |
2740 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
2741 | mask = GFP_NOFS; | |
2742 | clear_extent_bit(tree, start, end, EXTENT_UPTODATE, | |
2743 | 1, 1, mask); | |
2744 | } | |
2745 | return ret; | |
2746 | } | |
7b13b7b1 | 2747 | |
d1310b2e CM |
2748 | /* |
2749 | * a helper for releasepage. As long as there are no locked extents | |
2750 | * in the range corresponding to the page, both state records and extent | |
2751 | * map records are removed | |
2752 | */ | |
2753 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
2754 | struct extent_io_tree *tree, struct page *page, |
2755 | gfp_t mask) | |
d1310b2e CM |
2756 | { |
2757 | struct extent_map *em; | |
2758 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2759 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
7b13b7b1 | 2760 | |
70dec807 CM |
2761 | if ((mask & __GFP_WAIT) && |
2762 | page->mapping->host->i_size > 16 * 1024 * 1024) { | |
39b5637f | 2763 | u64 len; |
70dec807 | 2764 | while (start <= end) { |
39b5637f | 2765 | len = end - start + 1; |
890871be | 2766 | write_lock(&map->lock); |
39b5637f | 2767 | em = lookup_extent_mapping(map, start, len); |
70dec807 | 2768 | if (!em || IS_ERR(em)) { |
890871be | 2769 | write_unlock(&map->lock); |
70dec807 CM |
2770 | break; |
2771 | } | |
7f3c74fb CM |
2772 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
2773 | em->start != start) { | |
890871be | 2774 | write_unlock(&map->lock); |
70dec807 CM |
2775 | free_extent_map(em); |
2776 | break; | |
2777 | } | |
2778 | if (!test_range_bit(tree, em->start, | |
2779 | extent_map_end(em) - 1, | |
c8b97818 CM |
2780 | EXTENT_LOCKED | EXTENT_WRITEBACK | |
2781 | EXTENT_ORDERED, | |
2782 | 0)) { | |
70dec807 CM |
2783 | remove_extent_mapping(map, em); |
2784 | /* once for the rb tree */ | |
2785 | free_extent_map(em); | |
2786 | } | |
2787 | start = extent_map_end(em); | |
890871be | 2788 | write_unlock(&map->lock); |
70dec807 CM |
2789 | |
2790 | /* once for us */ | |
d1310b2e CM |
2791 | free_extent_map(em); |
2792 | } | |
d1310b2e | 2793 | } |
7b13b7b1 | 2794 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 2795 | } |
d1310b2e CM |
2796 | |
2797 | sector_t extent_bmap(struct address_space *mapping, sector_t iblock, | |
2798 | get_extent_t *get_extent) | |
2799 | { | |
2800 | struct inode *inode = mapping->host; | |
2801 | u64 start = iblock << inode->i_blkbits; | |
2802 | sector_t sector = 0; | |
d899e052 | 2803 | size_t blksize = (1 << inode->i_blkbits); |
d1310b2e CM |
2804 | struct extent_map *em; |
2805 | ||
d899e052 YZ |
2806 | lock_extent(&BTRFS_I(inode)->io_tree, start, start + blksize - 1, |
2807 | GFP_NOFS); | |
2808 | em = get_extent(inode, NULL, 0, start, blksize, 0); | |
2809 | unlock_extent(&BTRFS_I(inode)->io_tree, start, start + blksize - 1, | |
2810 | GFP_NOFS); | |
d1310b2e CM |
2811 | if (!em || IS_ERR(em)) |
2812 | return 0; | |
2813 | ||
d899e052 | 2814 | if (em->block_start > EXTENT_MAP_LAST_BYTE) |
d1310b2e CM |
2815 | goto out; |
2816 | ||
2817 | sector = (em->block_start + start - em->start) >> inode->i_blkbits; | |
d1310b2e CM |
2818 | out: |
2819 | free_extent_map(em); | |
2820 | return sector; | |
2821 | } | |
2822 | ||
1506fcc8 YS |
2823 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
2824 | __u64 start, __u64 len, get_extent_t *get_extent) | |
2825 | { | |
2826 | int ret; | |
2827 | u64 off = start; | |
2828 | u64 max = start + len; | |
2829 | u32 flags = 0; | |
2830 | u64 disko = 0; | |
2831 | struct extent_map *em = NULL; | |
2832 | int end = 0; | |
2833 | u64 em_start = 0, em_len = 0; | |
2834 | unsigned long emflags; | |
2835 | ret = 0; | |
2836 | ||
2837 | if (len == 0) | |
2838 | return -EINVAL; | |
2839 | ||
2840 | lock_extent(&BTRFS_I(inode)->io_tree, start, start + len, | |
2841 | GFP_NOFS); | |
2842 | em = get_extent(inode, NULL, 0, off, max - off, 0); | |
2843 | if (!em) | |
2844 | goto out; | |
2845 | if (IS_ERR(em)) { | |
2846 | ret = PTR_ERR(em); | |
2847 | goto out; | |
2848 | } | |
2849 | while (!end) { | |
2850 | off = em->start + em->len; | |
2851 | if (off >= max) | |
2852 | end = 1; | |
2853 | ||
2854 | em_start = em->start; | |
2855 | em_len = em->len; | |
2856 | ||
2857 | disko = 0; | |
2858 | flags = 0; | |
2859 | ||
93dbfad7 | 2860 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
2861 | end = 1; |
2862 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 2863 | } else if (em->block_start == EXTENT_MAP_HOLE) { |
1506fcc8 | 2864 | flags |= FIEMAP_EXTENT_UNWRITTEN; |
93dbfad7 | 2865 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
2866 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
2867 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 2868 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
2869 | flags |= (FIEMAP_EXTENT_DELALLOC | |
2870 | FIEMAP_EXTENT_UNKNOWN); | |
93dbfad7 | 2871 | } else { |
1506fcc8 | 2872 | disko = em->block_start; |
1506fcc8 YS |
2873 | } |
2874 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
2875 | flags |= FIEMAP_EXTENT_ENCODED; | |
2876 | ||
2877 | emflags = em->flags; | |
2878 | free_extent_map(em); | |
2879 | em = NULL; | |
2880 | ||
2881 | if (!end) { | |
2882 | em = get_extent(inode, NULL, 0, off, max - off, 0); | |
2883 | if (!em) | |
2884 | goto out; | |
2885 | if (IS_ERR(em)) { | |
2886 | ret = PTR_ERR(em); | |
2887 | goto out; | |
2888 | } | |
2889 | emflags = em->flags; | |
2890 | } | |
2891 | if (test_bit(EXTENT_FLAG_VACANCY, &emflags)) { | |
2892 | flags |= FIEMAP_EXTENT_LAST; | |
2893 | end = 1; | |
2894 | } | |
2895 | ||
2896 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, | |
2897 | em_len, flags); | |
2898 | if (ret) | |
2899 | goto out_free; | |
2900 | } | |
2901 | out_free: | |
2902 | free_extent_map(em); | |
2903 | out: | |
2904 | unlock_extent(&BTRFS_I(inode)->io_tree, start, start + len, | |
2905 | GFP_NOFS); | |
2906 | return ret; | |
2907 | } | |
2908 | ||
d1310b2e CM |
2909 | static inline struct page *extent_buffer_page(struct extent_buffer *eb, |
2910 | unsigned long i) | |
2911 | { | |
2912 | struct page *p; | |
2913 | struct address_space *mapping; | |
2914 | ||
2915 | if (i == 0) | |
2916 | return eb->first_page; | |
2917 | i += eb->start >> PAGE_CACHE_SHIFT; | |
2918 | mapping = eb->first_page->mapping; | |
33958dc6 CM |
2919 | if (!mapping) |
2920 | return NULL; | |
0ee0fda0 SW |
2921 | |
2922 | /* | |
2923 | * extent_buffer_page is only called after pinning the page | |
2924 | * by increasing the reference count. So we know the page must | |
2925 | * be in the radix tree. | |
2926 | */ | |
0ee0fda0 | 2927 | rcu_read_lock(); |
d1310b2e | 2928 | p = radix_tree_lookup(&mapping->page_tree, i); |
0ee0fda0 | 2929 | rcu_read_unlock(); |
2b1f55b0 | 2930 | |
d1310b2e CM |
2931 | return p; |
2932 | } | |
2933 | ||
6af118ce | 2934 | static inline unsigned long num_extent_pages(u64 start, u64 len) |
728131d8 | 2935 | { |
6af118ce CM |
2936 | return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) - |
2937 | (start >> PAGE_CACHE_SHIFT); | |
728131d8 CM |
2938 | } |
2939 | ||
d1310b2e CM |
2940 | static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree, |
2941 | u64 start, | |
2942 | unsigned long len, | |
2943 | gfp_t mask) | |
2944 | { | |
2945 | struct extent_buffer *eb = NULL; | |
3935127c | 2946 | #if LEAK_DEBUG |
2d2ae547 | 2947 | unsigned long flags; |
4bef0848 | 2948 | #endif |
d1310b2e | 2949 | |
d1310b2e | 2950 | eb = kmem_cache_zalloc(extent_buffer_cache, mask); |
d1310b2e CM |
2951 | eb->start = start; |
2952 | eb->len = len; | |
b4ce94de CM |
2953 | spin_lock_init(&eb->lock); |
2954 | init_waitqueue_head(&eb->lock_wq); | |
2955 | ||
3935127c | 2956 | #if LEAK_DEBUG |
2d2ae547 CM |
2957 | spin_lock_irqsave(&leak_lock, flags); |
2958 | list_add(&eb->leak_list, &buffers); | |
2959 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 2960 | #endif |
d1310b2e CM |
2961 | atomic_set(&eb->refs, 1); |
2962 | ||
2963 | return eb; | |
2964 | } | |
2965 | ||
2966 | static void __free_extent_buffer(struct extent_buffer *eb) | |
2967 | { | |
3935127c | 2968 | #if LEAK_DEBUG |
2d2ae547 CM |
2969 | unsigned long flags; |
2970 | spin_lock_irqsave(&leak_lock, flags); | |
2971 | list_del(&eb->leak_list); | |
2972 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 2973 | #endif |
d1310b2e CM |
2974 | kmem_cache_free(extent_buffer_cache, eb); |
2975 | } | |
2976 | ||
2977 | struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree, | |
2978 | u64 start, unsigned long len, | |
2979 | struct page *page0, | |
2980 | gfp_t mask) | |
2981 | { | |
2982 | unsigned long num_pages = num_extent_pages(start, len); | |
2983 | unsigned long i; | |
2984 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
2985 | struct extent_buffer *eb; | |
6af118ce | 2986 | struct extent_buffer *exists = NULL; |
d1310b2e CM |
2987 | struct page *p; |
2988 | struct address_space *mapping = tree->mapping; | |
2989 | int uptodate = 1; | |
2990 | ||
6af118ce CM |
2991 | spin_lock(&tree->buffer_lock); |
2992 | eb = buffer_search(tree, start); | |
2993 | if (eb) { | |
2994 | atomic_inc(&eb->refs); | |
2995 | spin_unlock(&tree->buffer_lock); | |
0f9dd46c | 2996 | mark_page_accessed(eb->first_page); |
6af118ce CM |
2997 | return eb; |
2998 | } | |
2999 | spin_unlock(&tree->buffer_lock); | |
3000 | ||
d1310b2e | 3001 | eb = __alloc_extent_buffer(tree, start, len, mask); |
2b114d1d | 3002 | if (!eb) |
d1310b2e CM |
3003 | return NULL; |
3004 | ||
d1310b2e CM |
3005 | if (page0) { |
3006 | eb->first_page = page0; | |
3007 | i = 1; | |
3008 | index++; | |
3009 | page_cache_get(page0); | |
3010 | mark_page_accessed(page0); | |
3011 | set_page_extent_mapped(page0); | |
d1310b2e | 3012 | set_page_extent_head(page0, len); |
f188591e | 3013 | uptodate = PageUptodate(page0); |
d1310b2e CM |
3014 | } else { |
3015 | i = 0; | |
3016 | } | |
3017 | for (; i < num_pages; i++, index++) { | |
3018 | p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM); | |
3019 | if (!p) { | |
3020 | WARN_ON(1); | |
6af118ce | 3021 | goto free_eb; |
d1310b2e CM |
3022 | } |
3023 | set_page_extent_mapped(p); | |
3024 | mark_page_accessed(p); | |
3025 | if (i == 0) { | |
3026 | eb->first_page = p; | |
3027 | set_page_extent_head(p, len); | |
3028 | } else { | |
3029 | set_page_private(p, EXTENT_PAGE_PRIVATE); | |
3030 | } | |
3031 | if (!PageUptodate(p)) | |
3032 | uptodate = 0; | |
3033 | unlock_page(p); | |
3034 | } | |
3035 | if (uptodate) | |
b4ce94de | 3036 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 3037 | |
6af118ce CM |
3038 | spin_lock(&tree->buffer_lock); |
3039 | exists = buffer_tree_insert(tree, start, &eb->rb_node); | |
3040 | if (exists) { | |
3041 | /* add one reference for the caller */ | |
3042 | atomic_inc(&exists->refs); | |
3043 | spin_unlock(&tree->buffer_lock); | |
3044 | goto free_eb; | |
3045 | } | |
3046 | spin_unlock(&tree->buffer_lock); | |
3047 | ||
3048 | /* add one reference for the tree */ | |
3049 | atomic_inc(&eb->refs); | |
d1310b2e CM |
3050 | return eb; |
3051 | ||
6af118ce | 3052 | free_eb: |
d1310b2e | 3053 | if (!atomic_dec_and_test(&eb->refs)) |
6af118ce CM |
3054 | return exists; |
3055 | for (index = 1; index < i; index++) | |
d1310b2e | 3056 | page_cache_release(extent_buffer_page(eb, index)); |
6af118ce | 3057 | page_cache_release(extent_buffer_page(eb, 0)); |
d1310b2e | 3058 | __free_extent_buffer(eb); |
6af118ce | 3059 | return exists; |
d1310b2e | 3060 | } |
d1310b2e CM |
3061 | |
3062 | struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree, | |
3063 | u64 start, unsigned long len, | |
3064 | gfp_t mask) | |
3065 | { | |
d1310b2e | 3066 | struct extent_buffer *eb; |
d1310b2e | 3067 | |
6af118ce CM |
3068 | spin_lock(&tree->buffer_lock); |
3069 | eb = buffer_search(tree, start); | |
3070 | if (eb) | |
3071 | atomic_inc(&eb->refs); | |
3072 | spin_unlock(&tree->buffer_lock); | |
d1310b2e | 3073 | |
0f9dd46c JB |
3074 | if (eb) |
3075 | mark_page_accessed(eb->first_page); | |
3076 | ||
d1310b2e | 3077 | return eb; |
d1310b2e | 3078 | } |
d1310b2e CM |
3079 | |
3080 | void free_extent_buffer(struct extent_buffer *eb) | |
3081 | { | |
d1310b2e CM |
3082 | if (!eb) |
3083 | return; | |
3084 | ||
3085 | if (!atomic_dec_and_test(&eb->refs)) | |
3086 | return; | |
3087 | ||
6af118ce | 3088 | WARN_ON(1); |
d1310b2e | 3089 | } |
d1310b2e CM |
3090 | |
3091 | int clear_extent_buffer_dirty(struct extent_io_tree *tree, | |
3092 | struct extent_buffer *eb) | |
3093 | { | |
d1310b2e CM |
3094 | unsigned long i; |
3095 | unsigned long num_pages; | |
3096 | struct page *page; | |
3097 | ||
d1310b2e CM |
3098 | num_pages = num_extent_pages(eb->start, eb->len); |
3099 | ||
3100 | for (i = 0; i < num_pages; i++) { | |
3101 | page = extent_buffer_page(eb, i); | |
b9473439 | 3102 | if (!PageDirty(page)) |
d2c3f4f6 CM |
3103 | continue; |
3104 | ||
a61e6f29 | 3105 | lock_page(page); |
d1310b2e CM |
3106 | if (i == 0) |
3107 | set_page_extent_head(page, eb->len); | |
3108 | else | |
3109 | set_page_private(page, EXTENT_PAGE_PRIVATE); | |
3110 | ||
d1310b2e | 3111 | clear_page_dirty_for_io(page); |
0ee0fda0 | 3112 | spin_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
3113 | if (!PageDirty(page)) { |
3114 | radix_tree_tag_clear(&page->mapping->page_tree, | |
3115 | page_index(page), | |
3116 | PAGECACHE_TAG_DIRTY); | |
3117 | } | |
0ee0fda0 | 3118 | spin_unlock_irq(&page->mapping->tree_lock); |
a61e6f29 | 3119 | unlock_page(page); |
d1310b2e CM |
3120 | } |
3121 | return 0; | |
3122 | } | |
d1310b2e CM |
3123 | |
3124 | int wait_on_extent_buffer_writeback(struct extent_io_tree *tree, | |
3125 | struct extent_buffer *eb) | |
3126 | { | |
3127 | return wait_on_extent_writeback(tree, eb->start, | |
3128 | eb->start + eb->len - 1); | |
3129 | } | |
d1310b2e CM |
3130 | |
3131 | int set_extent_buffer_dirty(struct extent_io_tree *tree, | |
3132 | struct extent_buffer *eb) | |
3133 | { | |
3134 | unsigned long i; | |
3135 | unsigned long num_pages; | |
b9473439 | 3136 | int was_dirty = 0; |
d1310b2e | 3137 | |
b9473439 | 3138 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
d1310b2e | 3139 | num_pages = num_extent_pages(eb->start, eb->len); |
b9473439 | 3140 | for (i = 0; i < num_pages; i++) |
d1310b2e | 3141 | __set_page_dirty_nobuffers(extent_buffer_page(eb, i)); |
b9473439 | 3142 | return was_dirty; |
d1310b2e | 3143 | } |
d1310b2e | 3144 | |
1259ab75 CM |
3145 | int clear_extent_buffer_uptodate(struct extent_io_tree *tree, |
3146 | struct extent_buffer *eb) | |
3147 | { | |
3148 | unsigned long i; | |
3149 | struct page *page; | |
3150 | unsigned long num_pages; | |
3151 | ||
3152 | num_pages = num_extent_pages(eb->start, eb->len); | |
b4ce94de | 3153 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
1259ab75 CM |
3154 | |
3155 | clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1, | |
3156 | GFP_NOFS); | |
3157 | for (i = 0; i < num_pages; i++) { | |
3158 | page = extent_buffer_page(eb, i); | |
33958dc6 CM |
3159 | if (page) |
3160 | ClearPageUptodate(page); | |
1259ab75 CM |
3161 | } |
3162 | return 0; | |
3163 | } | |
3164 | ||
d1310b2e CM |
3165 | int set_extent_buffer_uptodate(struct extent_io_tree *tree, |
3166 | struct extent_buffer *eb) | |
3167 | { | |
3168 | unsigned long i; | |
3169 | struct page *page; | |
3170 | unsigned long num_pages; | |
3171 | ||
3172 | num_pages = num_extent_pages(eb->start, eb->len); | |
3173 | ||
3174 | set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1, | |
3175 | GFP_NOFS); | |
3176 | for (i = 0; i < num_pages; i++) { | |
3177 | page = extent_buffer_page(eb, i); | |
3178 | if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) || | |
3179 | ((i == num_pages - 1) && | |
3180 | ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) { | |
3181 | check_page_uptodate(tree, page); | |
3182 | continue; | |
3183 | } | |
3184 | SetPageUptodate(page); | |
3185 | } | |
3186 | return 0; | |
3187 | } | |
d1310b2e | 3188 | |
ce9adaa5 CM |
3189 | int extent_range_uptodate(struct extent_io_tree *tree, |
3190 | u64 start, u64 end) | |
3191 | { | |
3192 | struct page *page; | |
3193 | int ret; | |
3194 | int pg_uptodate = 1; | |
3195 | int uptodate; | |
3196 | unsigned long index; | |
3197 | ||
3198 | ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1); | |
3199 | if (ret) | |
3200 | return 1; | |
d397712b | 3201 | while (start <= end) { |
ce9adaa5 CM |
3202 | index = start >> PAGE_CACHE_SHIFT; |
3203 | page = find_get_page(tree->mapping, index); | |
3204 | uptodate = PageUptodate(page); | |
3205 | page_cache_release(page); | |
3206 | if (!uptodate) { | |
3207 | pg_uptodate = 0; | |
3208 | break; | |
3209 | } | |
3210 | start += PAGE_CACHE_SIZE; | |
3211 | } | |
3212 | return pg_uptodate; | |
3213 | } | |
3214 | ||
d1310b2e | 3215 | int extent_buffer_uptodate(struct extent_io_tree *tree, |
ce9adaa5 | 3216 | struct extent_buffer *eb) |
d1310b2e | 3217 | { |
728131d8 | 3218 | int ret = 0; |
ce9adaa5 CM |
3219 | unsigned long num_pages; |
3220 | unsigned long i; | |
728131d8 CM |
3221 | struct page *page; |
3222 | int pg_uptodate = 1; | |
3223 | ||
b4ce94de | 3224 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
4235298e | 3225 | return 1; |
728131d8 | 3226 | |
4235298e | 3227 | ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1, |
d1310b2e | 3228 | EXTENT_UPTODATE, 1); |
4235298e CM |
3229 | if (ret) |
3230 | return ret; | |
728131d8 CM |
3231 | |
3232 | num_pages = num_extent_pages(eb->start, eb->len); | |
3233 | for (i = 0; i < num_pages; i++) { | |
3234 | page = extent_buffer_page(eb, i); | |
3235 | if (!PageUptodate(page)) { | |
3236 | pg_uptodate = 0; | |
3237 | break; | |
3238 | } | |
3239 | } | |
4235298e | 3240 | return pg_uptodate; |
d1310b2e | 3241 | } |
d1310b2e CM |
3242 | |
3243 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
3244 | struct extent_buffer *eb, | |
a86c12c7 | 3245 | u64 start, int wait, |
f188591e | 3246 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3247 | { |
3248 | unsigned long i; | |
3249 | unsigned long start_i; | |
3250 | struct page *page; | |
3251 | int err; | |
3252 | int ret = 0; | |
ce9adaa5 CM |
3253 | int locked_pages = 0; |
3254 | int all_uptodate = 1; | |
3255 | int inc_all_pages = 0; | |
d1310b2e | 3256 | unsigned long num_pages; |
a86c12c7 | 3257 | struct bio *bio = NULL; |
c8b97818 | 3258 | unsigned long bio_flags = 0; |
a86c12c7 | 3259 | |
b4ce94de | 3260 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
3261 | return 0; |
3262 | ||
ce9adaa5 | 3263 | if (test_range_bit(tree, eb->start, eb->start + eb->len - 1, |
d1310b2e CM |
3264 | EXTENT_UPTODATE, 1)) { |
3265 | return 0; | |
3266 | } | |
3267 | ||
3268 | if (start) { | |
3269 | WARN_ON(start < eb->start); | |
3270 | start_i = (start >> PAGE_CACHE_SHIFT) - | |
3271 | (eb->start >> PAGE_CACHE_SHIFT); | |
3272 | } else { | |
3273 | start_i = 0; | |
3274 | } | |
3275 | ||
3276 | num_pages = num_extent_pages(eb->start, eb->len); | |
3277 | for (i = start_i; i < num_pages; i++) { | |
3278 | page = extent_buffer_page(eb, i); | |
d1310b2e | 3279 | if (!wait) { |
2db04966 | 3280 | if (!trylock_page(page)) |
ce9adaa5 | 3281 | goto unlock_exit; |
d1310b2e CM |
3282 | } else { |
3283 | lock_page(page); | |
3284 | } | |
ce9adaa5 | 3285 | locked_pages++; |
d397712b | 3286 | if (!PageUptodate(page)) |
ce9adaa5 | 3287 | all_uptodate = 0; |
ce9adaa5 CM |
3288 | } |
3289 | if (all_uptodate) { | |
3290 | if (start_i == 0) | |
b4ce94de | 3291 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
3292 | goto unlock_exit; |
3293 | } | |
3294 | ||
3295 | for (i = start_i; i < num_pages; i++) { | |
3296 | page = extent_buffer_page(eb, i); | |
3297 | if (inc_all_pages) | |
3298 | page_cache_get(page); | |
3299 | if (!PageUptodate(page)) { | |
3300 | if (start_i == 0) | |
3301 | inc_all_pages = 1; | |
f188591e | 3302 | ClearPageError(page); |
a86c12c7 | 3303 | err = __extent_read_full_page(tree, page, |
f188591e | 3304 | get_extent, &bio, |
c8b97818 | 3305 | mirror_num, &bio_flags); |
d397712b | 3306 | if (err) |
d1310b2e | 3307 | ret = err; |
d1310b2e CM |
3308 | } else { |
3309 | unlock_page(page); | |
3310 | } | |
3311 | } | |
3312 | ||
a86c12c7 | 3313 | if (bio) |
c8b97818 | 3314 | submit_one_bio(READ, bio, mirror_num, bio_flags); |
a86c12c7 | 3315 | |
d397712b | 3316 | if (ret || !wait) |
d1310b2e | 3317 | return ret; |
d397712b | 3318 | |
d1310b2e CM |
3319 | for (i = start_i; i < num_pages; i++) { |
3320 | page = extent_buffer_page(eb, i); | |
3321 | wait_on_page_locked(page); | |
d397712b | 3322 | if (!PageUptodate(page)) |
d1310b2e | 3323 | ret = -EIO; |
d1310b2e | 3324 | } |
d397712b | 3325 | |
d1310b2e | 3326 | if (!ret) |
b4ce94de | 3327 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 3328 | return ret; |
ce9adaa5 CM |
3329 | |
3330 | unlock_exit: | |
3331 | i = start_i; | |
d397712b | 3332 | while (locked_pages > 0) { |
ce9adaa5 CM |
3333 | page = extent_buffer_page(eb, i); |
3334 | i++; | |
3335 | unlock_page(page); | |
3336 | locked_pages--; | |
3337 | } | |
3338 | return ret; | |
d1310b2e | 3339 | } |
d1310b2e CM |
3340 | |
3341 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
3342 | unsigned long start, | |
3343 | unsigned long len) | |
3344 | { | |
3345 | size_t cur; | |
3346 | size_t offset; | |
3347 | struct page *page; | |
3348 | char *kaddr; | |
3349 | char *dst = (char *)dstv; | |
3350 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3351 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
3352 | |
3353 | WARN_ON(start > eb->len); | |
3354 | WARN_ON(start + len > eb->start + eb->len); | |
3355 | ||
3356 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3357 | ||
d397712b | 3358 | while (len > 0) { |
d1310b2e | 3359 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
3360 | |
3361 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
3362 | kaddr = kmap_atomic(page, KM_USER1); | |
3363 | memcpy(dst, kaddr + offset, cur); | |
3364 | kunmap_atomic(kaddr, KM_USER1); | |
3365 | ||
3366 | dst += cur; | |
3367 | len -= cur; | |
3368 | offset = 0; | |
3369 | i++; | |
3370 | } | |
3371 | } | |
d1310b2e CM |
3372 | |
3373 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
3374 | unsigned long min_len, char **token, char **map, | |
3375 | unsigned long *map_start, | |
3376 | unsigned long *map_len, int km) | |
3377 | { | |
3378 | size_t offset = start & (PAGE_CACHE_SIZE - 1); | |
3379 | char *kaddr; | |
3380 | struct page *p; | |
3381 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3382 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3383 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
3384 | PAGE_CACHE_SHIFT; | |
3385 | ||
3386 | if (i != end_i) | |
3387 | return -EINVAL; | |
3388 | ||
3389 | if (i == 0) { | |
3390 | offset = start_offset; | |
3391 | *map_start = 0; | |
3392 | } else { | |
3393 | offset = 0; | |
3394 | *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; | |
3395 | } | |
d397712b | 3396 | |
d1310b2e | 3397 | if (start + min_len > eb->len) { |
d397712b CM |
3398 | printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, " |
3399 | "wanted %lu %lu\n", (unsigned long long)eb->start, | |
3400 | eb->len, start, min_len); | |
d1310b2e CM |
3401 | WARN_ON(1); |
3402 | } | |
3403 | ||
3404 | p = extent_buffer_page(eb, i); | |
d1310b2e CM |
3405 | kaddr = kmap_atomic(p, km); |
3406 | *token = kaddr; | |
3407 | *map = kaddr + offset; | |
3408 | *map_len = PAGE_CACHE_SIZE - offset; | |
3409 | return 0; | |
3410 | } | |
d1310b2e CM |
3411 | |
3412 | int map_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
3413 | unsigned long min_len, | |
3414 | char **token, char **map, | |
3415 | unsigned long *map_start, | |
3416 | unsigned long *map_len, int km) | |
3417 | { | |
3418 | int err; | |
3419 | int save = 0; | |
3420 | if (eb->map_token) { | |
3421 | unmap_extent_buffer(eb, eb->map_token, km); | |
3422 | eb->map_token = NULL; | |
3423 | save = 1; | |
3424 | } | |
3425 | err = map_private_extent_buffer(eb, start, min_len, token, map, | |
3426 | map_start, map_len, km); | |
3427 | if (!err && save) { | |
3428 | eb->map_token = *token; | |
3429 | eb->kaddr = *map; | |
3430 | eb->map_start = *map_start; | |
3431 | eb->map_len = *map_len; | |
3432 | } | |
3433 | return err; | |
3434 | } | |
d1310b2e CM |
3435 | |
3436 | void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km) | |
3437 | { | |
3438 | kunmap_atomic(token, km); | |
3439 | } | |
d1310b2e CM |
3440 | |
3441 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, | |
3442 | unsigned long start, | |
3443 | unsigned long len) | |
3444 | { | |
3445 | size_t cur; | |
3446 | size_t offset; | |
3447 | struct page *page; | |
3448 | char *kaddr; | |
3449 | char *ptr = (char *)ptrv; | |
3450 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3451 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3452 | int ret = 0; | |
3453 | ||
3454 | WARN_ON(start > eb->len); | |
3455 | WARN_ON(start + len > eb->start + eb->len); | |
3456 | ||
3457 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3458 | ||
d397712b | 3459 | while (len > 0) { |
d1310b2e | 3460 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
3461 | |
3462 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
3463 | ||
3464 | kaddr = kmap_atomic(page, KM_USER0); | |
3465 | ret = memcmp(ptr, kaddr + offset, cur); | |
3466 | kunmap_atomic(kaddr, KM_USER0); | |
3467 | if (ret) | |
3468 | break; | |
3469 | ||
3470 | ptr += cur; | |
3471 | len -= cur; | |
3472 | offset = 0; | |
3473 | i++; | |
3474 | } | |
3475 | return ret; | |
3476 | } | |
d1310b2e CM |
3477 | |
3478 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
3479 | unsigned long start, unsigned long len) | |
3480 | { | |
3481 | size_t cur; | |
3482 | size_t offset; | |
3483 | struct page *page; | |
3484 | char *kaddr; | |
3485 | char *src = (char *)srcv; | |
3486 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3487 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3488 | ||
3489 | WARN_ON(start > eb->len); | |
3490 | WARN_ON(start + len > eb->start + eb->len); | |
3491 | ||
3492 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3493 | ||
d397712b | 3494 | while (len > 0) { |
d1310b2e CM |
3495 | page = extent_buffer_page(eb, i); |
3496 | WARN_ON(!PageUptodate(page)); | |
3497 | ||
3498 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
3499 | kaddr = kmap_atomic(page, KM_USER1); | |
3500 | memcpy(kaddr + offset, src, cur); | |
3501 | kunmap_atomic(kaddr, KM_USER1); | |
3502 | ||
3503 | src += cur; | |
3504 | len -= cur; | |
3505 | offset = 0; | |
3506 | i++; | |
3507 | } | |
3508 | } | |
d1310b2e CM |
3509 | |
3510 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
3511 | unsigned long start, unsigned long len) | |
3512 | { | |
3513 | size_t cur; | |
3514 | size_t offset; | |
3515 | struct page *page; | |
3516 | char *kaddr; | |
3517 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3518 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3519 | ||
3520 | WARN_ON(start > eb->len); | |
3521 | WARN_ON(start + len > eb->start + eb->len); | |
3522 | ||
3523 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3524 | ||
d397712b | 3525 | while (len > 0) { |
d1310b2e CM |
3526 | page = extent_buffer_page(eb, i); |
3527 | WARN_ON(!PageUptodate(page)); | |
3528 | ||
3529 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
3530 | kaddr = kmap_atomic(page, KM_USER0); | |
3531 | memset(kaddr + offset, c, cur); | |
3532 | kunmap_atomic(kaddr, KM_USER0); | |
3533 | ||
3534 | len -= cur; | |
3535 | offset = 0; | |
3536 | i++; | |
3537 | } | |
3538 | } | |
d1310b2e CM |
3539 | |
3540 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
3541 | unsigned long dst_offset, unsigned long src_offset, | |
3542 | unsigned long len) | |
3543 | { | |
3544 | u64 dst_len = dst->len; | |
3545 | size_t cur; | |
3546 | size_t offset; | |
3547 | struct page *page; | |
3548 | char *kaddr; | |
3549 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3550 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
3551 | ||
3552 | WARN_ON(src->len != dst_len); | |
3553 | ||
3554 | offset = (start_offset + dst_offset) & | |
3555 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3556 | ||
d397712b | 3557 | while (len > 0) { |
d1310b2e CM |
3558 | page = extent_buffer_page(dst, i); |
3559 | WARN_ON(!PageUptodate(page)); | |
3560 | ||
3561 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
3562 | ||
3563 | kaddr = kmap_atomic(page, KM_USER0); | |
3564 | read_extent_buffer(src, kaddr + offset, src_offset, cur); | |
3565 | kunmap_atomic(kaddr, KM_USER0); | |
3566 | ||
3567 | src_offset += cur; | |
3568 | len -= cur; | |
3569 | offset = 0; | |
3570 | i++; | |
3571 | } | |
3572 | } | |
d1310b2e CM |
3573 | |
3574 | static void move_pages(struct page *dst_page, struct page *src_page, | |
3575 | unsigned long dst_off, unsigned long src_off, | |
3576 | unsigned long len) | |
3577 | { | |
3578 | char *dst_kaddr = kmap_atomic(dst_page, KM_USER0); | |
3579 | if (dst_page == src_page) { | |
3580 | memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len); | |
3581 | } else { | |
3582 | char *src_kaddr = kmap_atomic(src_page, KM_USER1); | |
3583 | char *p = dst_kaddr + dst_off + len; | |
3584 | char *s = src_kaddr + src_off + len; | |
3585 | ||
3586 | while (len--) | |
3587 | *--p = *--s; | |
3588 | ||
3589 | kunmap_atomic(src_kaddr, KM_USER1); | |
3590 | } | |
3591 | kunmap_atomic(dst_kaddr, KM_USER0); | |
3592 | } | |
3593 | ||
3594 | static void copy_pages(struct page *dst_page, struct page *src_page, | |
3595 | unsigned long dst_off, unsigned long src_off, | |
3596 | unsigned long len) | |
3597 | { | |
3598 | char *dst_kaddr = kmap_atomic(dst_page, KM_USER0); | |
3599 | char *src_kaddr; | |
3600 | ||
3601 | if (dst_page != src_page) | |
3602 | src_kaddr = kmap_atomic(src_page, KM_USER1); | |
3603 | else | |
3604 | src_kaddr = dst_kaddr; | |
3605 | ||
3606 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
3607 | kunmap_atomic(dst_kaddr, KM_USER0); | |
3608 | if (dst_page != src_page) | |
3609 | kunmap_atomic(src_kaddr, KM_USER1); | |
3610 | } | |
3611 | ||
3612 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
3613 | unsigned long src_offset, unsigned long len) | |
3614 | { | |
3615 | size_t cur; | |
3616 | size_t dst_off_in_page; | |
3617 | size_t src_off_in_page; | |
3618 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3619 | unsigned long dst_i; | |
3620 | unsigned long src_i; | |
3621 | ||
3622 | if (src_offset + len > dst->len) { | |
d397712b CM |
3623 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
3624 | "len %lu dst len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
3625 | BUG_ON(1); |
3626 | } | |
3627 | if (dst_offset + len > dst->len) { | |
d397712b CM |
3628 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
3629 | "len %lu dst len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
3630 | BUG_ON(1); |
3631 | } | |
3632 | ||
d397712b | 3633 | while (len > 0) { |
d1310b2e CM |
3634 | dst_off_in_page = (start_offset + dst_offset) & |
3635 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3636 | src_off_in_page = (start_offset + src_offset) & | |
3637 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3638 | ||
3639 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
3640 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
3641 | ||
3642 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
3643 | src_off_in_page)); | |
3644 | cur = min_t(unsigned long, cur, | |
3645 | (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); | |
3646 | ||
3647 | copy_pages(extent_buffer_page(dst, dst_i), | |
3648 | extent_buffer_page(dst, src_i), | |
3649 | dst_off_in_page, src_off_in_page, cur); | |
3650 | ||
3651 | src_offset += cur; | |
3652 | dst_offset += cur; | |
3653 | len -= cur; | |
3654 | } | |
3655 | } | |
d1310b2e CM |
3656 | |
3657 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
3658 | unsigned long src_offset, unsigned long len) | |
3659 | { | |
3660 | size_t cur; | |
3661 | size_t dst_off_in_page; | |
3662 | size_t src_off_in_page; | |
3663 | unsigned long dst_end = dst_offset + len - 1; | |
3664 | unsigned long src_end = src_offset + len - 1; | |
3665 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3666 | unsigned long dst_i; | |
3667 | unsigned long src_i; | |
3668 | ||
3669 | if (src_offset + len > dst->len) { | |
d397712b CM |
3670 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
3671 | "len %lu len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
3672 | BUG_ON(1); |
3673 | } | |
3674 | if (dst_offset + len > dst->len) { | |
d397712b CM |
3675 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
3676 | "len %lu len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
3677 | BUG_ON(1); |
3678 | } | |
3679 | if (dst_offset < src_offset) { | |
3680 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); | |
3681 | return; | |
3682 | } | |
d397712b | 3683 | while (len > 0) { |
d1310b2e CM |
3684 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; |
3685 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
3686 | ||
3687 | dst_off_in_page = (start_offset + dst_end) & | |
3688 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3689 | src_off_in_page = (start_offset + src_end) & | |
3690 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3691 | ||
3692 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
3693 | cur = min(cur, dst_off_in_page + 1); | |
3694 | move_pages(extent_buffer_page(dst, dst_i), | |
3695 | extent_buffer_page(dst, src_i), | |
3696 | dst_off_in_page - cur + 1, | |
3697 | src_off_in_page - cur + 1, cur); | |
3698 | ||
3699 | dst_end -= cur; | |
3700 | src_end -= cur; | |
3701 | len -= cur; | |
3702 | } | |
3703 | } | |
6af118ce CM |
3704 | |
3705 | int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page) | |
3706 | { | |
3707 | u64 start = page_offset(page); | |
3708 | struct extent_buffer *eb; | |
3709 | int ret = 1; | |
3710 | unsigned long i; | |
3711 | unsigned long num_pages; | |
3712 | ||
3713 | spin_lock(&tree->buffer_lock); | |
3714 | eb = buffer_search(tree, start); | |
3715 | if (!eb) | |
3716 | goto out; | |
3717 | ||
3718 | if (atomic_read(&eb->refs) > 1) { | |
3719 | ret = 0; | |
3720 | goto out; | |
3721 | } | |
b9473439 CM |
3722 | if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3723 | ret = 0; | |
3724 | goto out; | |
3725 | } | |
6af118ce CM |
3726 | /* at this point we can safely release the extent buffer */ |
3727 | num_pages = num_extent_pages(eb->start, eb->len); | |
b214107e CH |
3728 | for (i = 0; i < num_pages; i++) |
3729 | page_cache_release(extent_buffer_page(eb, i)); | |
6af118ce CM |
3730 | rb_erase(&eb->rb_node, &tree->buffer); |
3731 | __free_extent_buffer(eb); | |
3732 | out: | |
3733 | spin_unlock(&tree->buffer_lock); | |
3734 | return ret; | |
3735 | } |