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> | |
12 | #include <linux/version.h> | |
13 | #include <linux/writeback.h> | |
14 | #include <linux/pagevec.h> | |
15 | #include "extent_io.h" | |
16 | #include "extent_map.h" | |
17 | ||
18 | /* temporary define until extent_map moves out of btrfs */ | |
19 | struct kmem_cache *btrfs_cache_create(const char *name, size_t size, | |
20 | unsigned long extra_flags, | |
21 | void (*ctor)(void *, struct kmem_cache *, | |
22 | unsigned long)); | |
23 | ||
24 | static struct kmem_cache *extent_state_cache; | |
25 | static struct kmem_cache *extent_buffer_cache; | |
26 | ||
27 | static LIST_HEAD(buffers); | |
28 | static LIST_HEAD(states); | |
29 | ||
d1310b2e CM |
30 | #define BUFFER_LRU_MAX 64 |
31 | ||
32 | struct tree_entry { | |
33 | u64 start; | |
34 | u64 end; | |
d1310b2e CM |
35 | struct rb_node rb_node; |
36 | }; | |
37 | ||
38 | struct extent_page_data { | |
39 | struct bio *bio; | |
40 | struct extent_io_tree *tree; | |
41 | get_extent_t *get_extent; | |
42 | }; | |
43 | ||
44 | int __init extent_io_init(void) | |
45 | { | |
46 | extent_state_cache = btrfs_cache_create("extent_state", | |
47 | sizeof(struct extent_state), 0, | |
48 | NULL); | |
49 | if (!extent_state_cache) | |
50 | return -ENOMEM; | |
51 | ||
52 | extent_buffer_cache = btrfs_cache_create("extent_buffers", | |
53 | sizeof(struct extent_buffer), 0, | |
54 | NULL); | |
55 | if (!extent_buffer_cache) | |
56 | goto free_state_cache; | |
57 | return 0; | |
58 | ||
59 | free_state_cache: | |
60 | kmem_cache_destroy(extent_state_cache); | |
61 | return -ENOMEM; | |
62 | } | |
63 | ||
64 | void extent_io_exit(void) | |
65 | { | |
66 | struct extent_state *state; | |
67 | ||
68 | while (!list_empty(&states)) { | |
69 | state = list_entry(states.next, struct extent_state, list); | |
70dec807 | 70 | printk("state leak: start %Lu end %Lu state %lu in tree %p refs %d\n", state->start, state->end, state->state, state->tree, atomic_read(&state->refs)); |
d1310b2e CM |
71 | list_del(&state->list); |
72 | kmem_cache_free(extent_state_cache, state); | |
73 | ||
74 | } | |
75 | ||
76 | if (extent_state_cache) | |
77 | kmem_cache_destroy(extent_state_cache); | |
78 | if (extent_buffer_cache) | |
79 | kmem_cache_destroy(extent_buffer_cache); | |
80 | } | |
81 | ||
82 | void extent_io_tree_init(struct extent_io_tree *tree, | |
83 | struct address_space *mapping, gfp_t mask) | |
84 | { | |
85 | tree->state.rb_node = NULL; | |
86 | tree->ops = NULL; | |
87 | tree->dirty_bytes = 0; | |
70dec807 | 88 | spin_lock_init(&tree->lock); |
d1310b2e CM |
89 | spin_lock_init(&tree->lru_lock); |
90 | tree->mapping = mapping; | |
91 | INIT_LIST_HEAD(&tree->buffer_lru); | |
92 | tree->lru_size = 0; | |
93 | } | |
94 | EXPORT_SYMBOL(extent_io_tree_init); | |
95 | ||
96 | void extent_io_tree_empty_lru(struct extent_io_tree *tree) | |
97 | { | |
98 | struct extent_buffer *eb; | |
99 | while(!list_empty(&tree->buffer_lru)) { | |
100 | eb = list_entry(tree->buffer_lru.next, struct extent_buffer, | |
101 | lru); | |
102 | list_del_init(&eb->lru); | |
103 | free_extent_buffer(eb); | |
104 | } | |
105 | } | |
106 | EXPORT_SYMBOL(extent_io_tree_empty_lru); | |
107 | ||
108 | struct extent_state *alloc_extent_state(gfp_t mask) | |
109 | { | |
110 | struct extent_state *state; | |
d1310b2e CM |
111 | |
112 | state = kmem_cache_alloc(extent_state_cache, mask); | |
113 | if (!state || IS_ERR(state)) | |
114 | return state; | |
115 | state->state = 0; | |
d1310b2e | 116 | state->private = 0; |
70dec807 | 117 | state->tree = NULL; |
d1310b2e CM |
118 | |
119 | atomic_set(&state->refs, 1); | |
120 | init_waitqueue_head(&state->wq); | |
121 | return state; | |
122 | } | |
123 | EXPORT_SYMBOL(alloc_extent_state); | |
124 | ||
125 | void free_extent_state(struct extent_state *state) | |
126 | { | |
d1310b2e CM |
127 | if (!state) |
128 | return; | |
129 | if (atomic_dec_and_test(&state->refs)) { | |
70dec807 | 130 | WARN_ON(state->tree); |
d1310b2e CM |
131 | kmem_cache_free(extent_state_cache, state); |
132 | } | |
133 | } | |
134 | EXPORT_SYMBOL(free_extent_state); | |
135 | ||
136 | static struct rb_node *tree_insert(struct rb_root *root, u64 offset, | |
137 | struct rb_node *node) | |
138 | { | |
139 | struct rb_node ** p = &root->rb_node; | |
140 | struct rb_node * parent = NULL; | |
141 | struct tree_entry *entry; | |
142 | ||
143 | while(*p) { | |
144 | parent = *p; | |
145 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
146 | ||
147 | if (offset < entry->start) | |
148 | p = &(*p)->rb_left; | |
149 | else if (offset > entry->end) | |
150 | p = &(*p)->rb_right; | |
151 | else | |
152 | return parent; | |
153 | } | |
154 | ||
155 | entry = rb_entry(node, struct tree_entry, rb_node); | |
d1310b2e CM |
156 | rb_link_node(node, parent, p); |
157 | rb_insert_color(node, root); | |
158 | return NULL; | |
159 | } | |
160 | ||
161 | static struct rb_node *__tree_search(struct rb_root *root, u64 offset, | |
162 | struct rb_node **prev_ret, | |
163 | struct rb_node **next_ret) | |
164 | { | |
165 | struct rb_node * n = root->rb_node; | |
166 | struct rb_node *prev = NULL; | |
167 | struct rb_node *orig_prev = NULL; | |
168 | struct tree_entry *entry; | |
169 | struct tree_entry *prev_entry = NULL; | |
170 | ||
171 | while(n) { | |
172 | entry = rb_entry(n, struct tree_entry, rb_node); | |
173 | prev = n; | |
174 | prev_entry = entry; | |
175 | ||
176 | if (offset < entry->start) | |
177 | n = n->rb_left; | |
178 | else if (offset > entry->end) | |
179 | n = n->rb_right; | |
180 | else | |
181 | return n; | |
182 | } | |
183 | ||
184 | if (prev_ret) { | |
185 | orig_prev = prev; | |
186 | while(prev && offset > prev_entry->end) { | |
187 | prev = rb_next(prev); | |
188 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
189 | } | |
190 | *prev_ret = prev; | |
191 | prev = orig_prev; | |
192 | } | |
193 | ||
194 | if (next_ret) { | |
195 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
196 | while(prev && offset < prev_entry->start) { | |
197 | prev = rb_prev(prev); | |
198 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
199 | } | |
200 | *next_ret = prev; | |
201 | } | |
202 | return NULL; | |
203 | } | |
204 | ||
205 | static inline struct rb_node *tree_search(struct rb_root *root, u64 offset) | |
206 | { | |
70dec807 | 207 | struct rb_node *prev = NULL; |
d1310b2e | 208 | struct rb_node *ret; |
70dec807 | 209 | |
d1310b2e CM |
210 | ret = __tree_search(root, offset, &prev, NULL); |
211 | if (!ret) | |
212 | return prev; | |
213 | return ret; | |
214 | } | |
215 | ||
216 | /* | |
217 | * utility function to look for merge candidates inside a given range. | |
218 | * Any extents with matching state are merged together into a single | |
219 | * extent in the tree. Extents with EXTENT_IO in their state field | |
220 | * are not merged because the end_io handlers need to be able to do | |
221 | * operations on them without sleeping (or doing allocations/splits). | |
222 | * | |
223 | * This should be called with the tree lock held. | |
224 | */ | |
225 | static int merge_state(struct extent_io_tree *tree, | |
226 | struct extent_state *state) | |
227 | { | |
228 | struct extent_state *other; | |
229 | struct rb_node *other_node; | |
230 | ||
231 | if (state->state & EXTENT_IOBITS) | |
232 | return 0; | |
233 | ||
234 | other_node = rb_prev(&state->rb_node); | |
235 | if (other_node) { | |
236 | other = rb_entry(other_node, struct extent_state, rb_node); | |
237 | if (other->end == state->start - 1 && | |
238 | other->state == state->state) { | |
239 | state->start = other->start; | |
70dec807 | 240 | other->tree = NULL; |
d1310b2e CM |
241 | rb_erase(&other->rb_node, &tree->state); |
242 | free_extent_state(other); | |
243 | } | |
244 | } | |
245 | other_node = rb_next(&state->rb_node); | |
246 | if (other_node) { | |
247 | other = rb_entry(other_node, struct extent_state, rb_node); | |
248 | if (other->start == state->end + 1 && | |
249 | other->state == state->state) { | |
250 | other->start = state->start; | |
70dec807 | 251 | state->tree = NULL; |
d1310b2e CM |
252 | rb_erase(&state->rb_node, &tree->state); |
253 | free_extent_state(state); | |
254 | } | |
255 | } | |
256 | return 0; | |
257 | } | |
258 | ||
291d673e CM |
259 | static void set_state_cb(struct extent_io_tree *tree, |
260 | struct extent_state *state, | |
261 | unsigned long bits) | |
262 | { | |
263 | if (tree->ops && tree->ops->set_bit_hook) { | |
264 | tree->ops->set_bit_hook(tree->mapping->host, state->start, | |
265 | state->end, bits); | |
266 | } | |
267 | } | |
268 | ||
269 | static void clear_state_cb(struct extent_io_tree *tree, | |
270 | struct extent_state *state, | |
271 | unsigned long bits) | |
272 | { | |
273 | if (tree->ops && tree->ops->set_bit_hook) { | |
274 | tree->ops->clear_bit_hook(tree->mapping->host, state->start, | |
275 | state->end, bits); | |
276 | } | |
277 | } | |
278 | ||
d1310b2e CM |
279 | /* |
280 | * insert an extent_state struct into the tree. 'bits' are set on the | |
281 | * struct before it is inserted. | |
282 | * | |
283 | * This may return -EEXIST if the extent is already there, in which case the | |
284 | * state struct is freed. | |
285 | * | |
286 | * The tree lock is not taken internally. This is a utility function and | |
287 | * probably isn't what you want to call (see set/clear_extent_bit). | |
288 | */ | |
289 | static int insert_state(struct extent_io_tree *tree, | |
290 | struct extent_state *state, u64 start, u64 end, | |
291 | int bits) | |
292 | { | |
293 | struct rb_node *node; | |
294 | ||
295 | if (end < start) { | |
296 | printk("end < start %Lu %Lu\n", end, start); | |
297 | WARN_ON(1); | |
298 | } | |
299 | if (bits & EXTENT_DIRTY) | |
300 | tree->dirty_bytes += end - start + 1; | |
301 | state->state |= bits; | |
302 | state->start = start; | |
303 | state->end = end; | |
291d673e | 304 | set_state_cb(tree, state, bits); |
d1310b2e CM |
305 | node = tree_insert(&tree->state, end, &state->rb_node); |
306 | if (node) { | |
307 | struct extent_state *found; | |
308 | found = rb_entry(node, struct extent_state, rb_node); | |
309 | printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end); | |
310 | free_extent_state(state); | |
311 | return -EEXIST; | |
312 | } | |
70dec807 | 313 | state->tree = tree; |
d1310b2e CM |
314 | merge_state(tree, state); |
315 | return 0; | |
316 | } | |
317 | ||
318 | /* | |
319 | * split a given extent state struct in two, inserting the preallocated | |
320 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
321 | * offset inside 'orig' where it should be split. | |
322 | * | |
323 | * Before calling, | |
324 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
325 | * are two extent state structs in the tree: | |
326 | * prealloc: [orig->start, split - 1] | |
327 | * orig: [ split, orig->end ] | |
328 | * | |
329 | * The tree locks are not taken by this function. They need to be held | |
330 | * by the caller. | |
331 | */ | |
332 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
333 | struct extent_state *prealloc, u64 split) | |
334 | { | |
335 | struct rb_node *node; | |
336 | prealloc->start = orig->start; | |
337 | prealloc->end = split - 1; | |
338 | prealloc->state = orig->state; | |
339 | orig->start = split; | |
340 | ||
341 | node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node); | |
342 | if (node) { | |
343 | struct extent_state *found; | |
344 | found = rb_entry(node, struct extent_state, rb_node); | |
345 | printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end); | |
346 | free_extent_state(prealloc); | |
347 | return -EEXIST; | |
348 | } | |
70dec807 | 349 | prealloc->tree = tree; |
d1310b2e CM |
350 | return 0; |
351 | } | |
352 | ||
353 | /* | |
354 | * utility function to clear some bits in an extent state struct. | |
355 | * it will optionally wake up any one waiting on this state (wake == 1), or | |
356 | * forcibly remove the state from the tree (delete == 1). | |
357 | * | |
358 | * If no bits are set on the state struct after clearing things, the | |
359 | * struct is freed and removed from the tree | |
360 | */ | |
361 | static int clear_state_bit(struct extent_io_tree *tree, | |
362 | struct extent_state *state, int bits, int wake, | |
363 | int delete) | |
364 | { | |
365 | int ret = state->state & bits; | |
366 | ||
367 | if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { | |
368 | u64 range = state->end - state->start + 1; | |
369 | WARN_ON(range > tree->dirty_bytes); | |
370 | tree->dirty_bytes -= range; | |
371 | } | |
372 | state->state &= ~bits; | |
291d673e | 373 | clear_state_cb(tree, state, bits); |
d1310b2e CM |
374 | if (wake) |
375 | wake_up(&state->wq); | |
376 | if (delete || state->state == 0) { | |
70dec807 | 377 | if (state->tree) { |
d1310b2e | 378 | rb_erase(&state->rb_node, &tree->state); |
70dec807 | 379 | state->tree = NULL; |
d1310b2e CM |
380 | free_extent_state(state); |
381 | } else { | |
382 | WARN_ON(1); | |
383 | } | |
384 | } else { | |
385 | merge_state(tree, state); | |
386 | } | |
387 | return ret; | |
388 | } | |
389 | ||
390 | /* | |
391 | * clear some bits on a range in the tree. This may require splitting | |
392 | * or inserting elements in the tree, so the gfp mask is used to | |
393 | * indicate which allocations or sleeping are allowed. | |
394 | * | |
395 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
396 | * the given range from the tree regardless of state (ie for truncate). | |
397 | * | |
398 | * the range [start, end] is inclusive. | |
399 | * | |
400 | * This takes the tree lock, and returns < 0 on error, > 0 if any of the | |
401 | * bits were already set, or zero if none of the bits were already set. | |
402 | */ | |
403 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
404 | int bits, int wake, int delete, gfp_t mask) | |
405 | { | |
406 | struct extent_state *state; | |
407 | struct extent_state *prealloc = NULL; | |
408 | struct rb_node *node; | |
409 | unsigned long flags; | |
410 | int err; | |
411 | int set = 0; | |
412 | ||
413 | again: | |
414 | if (!prealloc && (mask & __GFP_WAIT)) { | |
415 | prealloc = alloc_extent_state(mask); | |
416 | if (!prealloc) | |
417 | return -ENOMEM; | |
418 | } | |
419 | ||
70dec807 | 420 | spin_lock_irqsave(&tree->lock, flags); |
d1310b2e CM |
421 | /* |
422 | * this search will find the extents that end after | |
423 | * our range starts | |
424 | */ | |
425 | node = tree_search(&tree->state, start); | |
426 | if (!node) | |
427 | goto out; | |
428 | state = rb_entry(node, struct extent_state, rb_node); | |
429 | if (state->start > end) | |
430 | goto out; | |
431 | WARN_ON(state->end < start); | |
432 | ||
433 | /* | |
434 | * | ---- desired range ---- | | |
435 | * | state | or | |
436 | * | ------------- state -------------- | | |
437 | * | |
438 | * We need to split the extent we found, and may flip | |
439 | * bits on second half. | |
440 | * | |
441 | * If the extent we found extends past our range, we | |
442 | * just split and search again. It'll get split again | |
443 | * the next time though. | |
444 | * | |
445 | * If the extent we found is inside our range, we clear | |
446 | * the desired bit on it. | |
447 | */ | |
448 | ||
449 | if (state->start < start) { | |
70dec807 CM |
450 | if (!prealloc) |
451 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
d1310b2e CM |
452 | err = split_state(tree, state, prealloc, start); |
453 | BUG_ON(err == -EEXIST); | |
454 | prealloc = NULL; | |
455 | if (err) | |
456 | goto out; | |
457 | if (state->end <= end) { | |
458 | start = state->end + 1; | |
459 | set |= clear_state_bit(tree, state, bits, | |
460 | wake, delete); | |
461 | } else { | |
462 | start = state->start; | |
463 | } | |
464 | goto search_again; | |
465 | } | |
466 | /* | |
467 | * | ---- desired range ---- | | |
468 | * | state | | |
469 | * We need to split the extent, and clear the bit | |
470 | * on the first half | |
471 | */ | |
472 | if (state->start <= end && state->end > end) { | |
70dec807 CM |
473 | if (!prealloc) |
474 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
d1310b2e CM |
475 | err = split_state(tree, state, prealloc, end + 1); |
476 | BUG_ON(err == -EEXIST); | |
477 | ||
478 | if (wake) | |
479 | wake_up(&state->wq); | |
480 | set |= clear_state_bit(tree, prealloc, bits, | |
481 | wake, delete); | |
482 | prealloc = NULL; | |
483 | goto out; | |
484 | } | |
485 | ||
486 | start = state->end + 1; | |
487 | set |= clear_state_bit(tree, state, bits, wake, delete); | |
488 | goto search_again; | |
489 | ||
490 | out: | |
70dec807 | 491 | spin_unlock_irqrestore(&tree->lock, flags); |
d1310b2e CM |
492 | if (prealloc) |
493 | free_extent_state(prealloc); | |
494 | ||
495 | return set; | |
496 | ||
497 | search_again: | |
498 | if (start > end) | |
499 | goto out; | |
70dec807 | 500 | spin_unlock_irqrestore(&tree->lock, flags); |
d1310b2e CM |
501 | if (mask & __GFP_WAIT) |
502 | cond_resched(); | |
503 | goto again; | |
504 | } | |
505 | EXPORT_SYMBOL(clear_extent_bit); | |
506 | ||
507 | static int wait_on_state(struct extent_io_tree *tree, | |
508 | struct extent_state *state) | |
509 | { | |
510 | DEFINE_WAIT(wait); | |
511 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
70dec807 | 512 | spin_unlock_irq(&tree->lock); |
d1310b2e | 513 | schedule(); |
70dec807 | 514 | spin_lock_irq(&tree->lock); |
d1310b2e CM |
515 | finish_wait(&state->wq, &wait); |
516 | return 0; | |
517 | } | |
518 | ||
519 | /* | |
520 | * waits for one or more bits to clear on a range in the state tree. | |
521 | * The range [start, end] is inclusive. | |
522 | * The tree lock is taken by this function | |
523 | */ | |
524 | int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits) | |
525 | { | |
526 | struct extent_state *state; | |
527 | struct rb_node *node; | |
528 | ||
70dec807 | 529 | spin_lock_irq(&tree->lock); |
d1310b2e CM |
530 | again: |
531 | while (1) { | |
532 | /* | |
533 | * this search will find all the extents that end after | |
534 | * our range starts | |
535 | */ | |
536 | node = tree_search(&tree->state, start); | |
537 | if (!node) | |
538 | break; | |
539 | ||
540 | state = rb_entry(node, struct extent_state, rb_node); | |
541 | ||
542 | if (state->start > end) | |
543 | goto out; | |
544 | ||
545 | if (state->state & bits) { | |
546 | start = state->start; | |
547 | atomic_inc(&state->refs); | |
548 | wait_on_state(tree, state); | |
549 | free_extent_state(state); | |
550 | goto again; | |
551 | } | |
552 | start = state->end + 1; | |
553 | ||
554 | if (start > end) | |
555 | break; | |
556 | ||
557 | if (need_resched()) { | |
70dec807 | 558 | spin_unlock_irq(&tree->lock); |
d1310b2e | 559 | cond_resched(); |
70dec807 | 560 | spin_lock_irq(&tree->lock); |
d1310b2e CM |
561 | } |
562 | } | |
563 | out: | |
70dec807 | 564 | spin_unlock_irq(&tree->lock); |
d1310b2e CM |
565 | return 0; |
566 | } | |
567 | EXPORT_SYMBOL(wait_extent_bit); | |
568 | ||
569 | static void set_state_bits(struct extent_io_tree *tree, | |
570 | struct extent_state *state, | |
571 | int bits) | |
572 | { | |
573 | if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { | |
574 | u64 range = state->end - state->start + 1; | |
575 | tree->dirty_bytes += range; | |
576 | } | |
577 | state->state |= bits; | |
291d673e | 578 | set_state_cb(tree, state, bits); |
d1310b2e CM |
579 | } |
580 | ||
581 | /* | |
582 | * set some bits on a range in the tree. This may require allocations | |
583 | * or sleeping, so the gfp mask is used to indicate what is allowed. | |
584 | * | |
585 | * If 'exclusive' == 1, this will fail with -EEXIST if some part of the | |
586 | * range already has the desired bits set. The start of the existing | |
587 | * range is returned in failed_start in this case. | |
588 | * | |
589 | * [start, end] is inclusive | |
590 | * This takes the tree lock. | |
591 | */ | |
592 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits, | |
593 | int exclusive, u64 *failed_start, gfp_t mask) | |
594 | { | |
595 | struct extent_state *state; | |
596 | struct extent_state *prealloc = NULL; | |
597 | struct rb_node *node; | |
598 | unsigned long flags; | |
599 | int err = 0; | |
600 | int set; | |
601 | u64 last_start; | |
602 | u64 last_end; | |
603 | again: | |
604 | if (!prealloc && (mask & __GFP_WAIT)) { | |
605 | prealloc = alloc_extent_state(mask); | |
606 | if (!prealloc) | |
607 | return -ENOMEM; | |
608 | } | |
609 | ||
70dec807 | 610 | spin_lock_irqsave(&tree->lock, flags); |
d1310b2e CM |
611 | /* |
612 | * this search will find all the extents that end after | |
613 | * our range starts. | |
614 | */ | |
615 | node = tree_search(&tree->state, start); | |
616 | if (!node) { | |
617 | err = insert_state(tree, prealloc, start, end, bits); | |
618 | prealloc = NULL; | |
619 | BUG_ON(err == -EEXIST); | |
620 | goto out; | |
621 | } | |
622 | ||
623 | state = rb_entry(node, struct extent_state, rb_node); | |
624 | last_start = state->start; | |
625 | last_end = state->end; | |
626 | ||
627 | /* | |
628 | * | ---- desired range ---- | | |
629 | * | state | | |
630 | * | |
631 | * Just lock what we found and keep going | |
632 | */ | |
633 | if (state->start == start && state->end <= end) { | |
634 | set = state->state & bits; | |
635 | if (set && exclusive) { | |
636 | *failed_start = state->start; | |
637 | err = -EEXIST; | |
638 | goto out; | |
639 | } | |
640 | set_state_bits(tree, state, bits); | |
641 | start = state->end + 1; | |
642 | merge_state(tree, state); | |
643 | goto search_again; | |
644 | } | |
645 | ||
646 | /* | |
647 | * | ---- desired range ---- | | |
648 | * | state | | |
649 | * or | |
650 | * | ------------- state -------------- | | |
651 | * | |
652 | * We need to split the extent we found, and may flip bits on | |
653 | * second half. | |
654 | * | |
655 | * If the extent we found extends past our | |
656 | * range, we just split and search again. It'll get split | |
657 | * again the next time though. | |
658 | * | |
659 | * If the extent we found is inside our range, we set the | |
660 | * desired bit on it. | |
661 | */ | |
662 | if (state->start < start) { | |
663 | set = state->state & bits; | |
664 | if (exclusive && set) { | |
665 | *failed_start = start; | |
666 | err = -EEXIST; | |
667 | goto out; | |
668 | } | |
669 | err = split_state(tree, state, prealloc, start); | |
670 | BUG_ON(err == -EEXIST); | |
671 | prealloc = NULL; | |
672 | if (err) | |
673 | goto out; | |
674 | if (state->end <= end) { | |
675 | set_state_bits(tree, state, bits); | |
676 | start = state->end + 1; | |
677 | merge_state(tree, state); | |
678 | } else { | |
679 | start = state->start; | |
680 | } | |
681 | goto search_again; | |
682 | } | |
683 | /* | |
684 | * | ---- desired range ---- | | |
685 | * | state | or | state | | |
686 | * | |
687 | * There's a hole, we need to insert something in it and | |
688 | * ignore the extent we found. | |
689 | */ | |
690 | if (state->start > start) { | |
691 | u64 this_end; | |
692 | if (end < last_start) | |
693 | this_end = end; | |
694 | else | |
695 | this_end = last_start -1; | |
696 | err = insert_state(tree, prealloc, start, this_end, | |
697 | bits); | |
698 | prealloc = NULL; | |
699 | BUG_ON(err == -EEXIST); | |
700 | if (err) | |
701 | goto out; | |
702 | start = this_end + 1; | |
703 | goto search_again; | |
704 | } | |
705 | /* | |
706 | * | ---- desired range ---- | | |
707 | * | state | | |
708 | * We need to split the extent, and set the bit | |
709 | * on the first half | |
710 | */ | |
711 | if (state->start <= end && state->end > end) { | |
712 | set = state->state & bits; | |
713 | if (exclusive && set) { | |
714 | *failed_start = start; | |
715 | err = -EEXIST; | |
716 | goto out; | |
717 | } | |
718 | err = split_state(tree, state, prealloc, end + 1); | |
719 | BUG_ON(err == -EEXIST); | |
720 | ||
721 | set_state_bits(tree, prealloc, bits); | |
722 | merge_state(tree, prealloc); | |
723 | prealloc = NULL; | |
724 | goto out; | |
725 | } | |
726 | ||
727 | goto search_again; | |
728 | ||
729 | out: | |
70dec807 | 730 | spin_unlock_irqrestore(&tree->lock, flags); |
d1310b2e CM |
731 | if (prealloc) |
732 | free_extent_state(prealloc); | |
733 | ||
734 | return err; | |
735 | ||
736 | search_again: | |
737 | if (start > end) | |
738 | goto out; | |
70dec807 | 739 | spin_unlock_irqrestore(&tree->lock, flags); |
d1310b2e CM |
740 | if (mask & __GFP_WAIT) |
741 | cond_resched(); | |
742 | goto again; | |
743 | } | |
744 | EXPORT_SYMBOL(set_extent_bit); | |
745 | ||
746 | /* wrappers around set/clear extent bit */ | |
747 | int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
748 | gfp_t mask) | |
749 | { | |
750 | return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL, | |
751 | mask); | |
752 | } | |
753 | EXPORT_SYMBOL(set_extent_dirty); | |
754 | ||
755 | int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
756 | int bits, gfp_t mask) | |
757 | { | |
758 | return set_extent_bit(tree, start, end, bits, 0, NULL, | |
759 | mask); | |
760 | } | |
761 | EXPORT_SYMBOL(set_extent_bits); | |
762 | ||
763 | int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
764 | int bits, gfp_t mask) | |
765 | { | |
766 | return clear_extent_bit(tree, start, end, bits, 0, 0, mask); | |
767 | } | |
768 | EXPORT_SYMBOL(clear_extent_bits); | |
769 | ||
770 | int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, | |
771 | gfp_t mask) | |
772 | { | |
773 | return set_extent_bit(tree, start, end, | |
774 | EXTENT_DELALLOC | EXTENT_DIRTY, 0, NULL, | |
775 | mask); | |
776 | } | |
777 | EXPORT_SYMBOL(set_extent_delalloc); | |
778 | ||
779 | int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
780 | gfp_t mask) | |
781 | { | |
782 | return clear_extent_bit(tree, start, end, | |
783 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask); | |
784 | } | |
785 | EXPORT_SYMBOL(clear_extent_dirty); | |
786 | ||
787 | int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, | |
788 | gfp_t mask) | |
789 | { | |
790 | return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL, | |
791 | mask); | |
792 | } | |
793 | EXPORT_SYMBOL(set_extent_new); | |
794 | ||
795 | int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end, | |
796 | gfp_t mask) | |
797 | { | |
798 | return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask); | |
799 | } | |
800 | EXPORT_SYMBOL(clear_extent_new); | |
801 | ||
802 | int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, | |
803 | gfp_t mask) | |
804 | { | |
805 | return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL, | |
806 | mask); | |
807 | } | |
808 | EXPORT_SYMBOL(set_extent_uptodate); | |
809 | ||
810 | int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, | |
811 | gfp_t mask) | |
812 | { | |
813 | return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask); | |
814 | } | |
815 | EXPORT_SYMBOL(clear_extent_uptodate); | |
816 | ||
817 | int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end, | |
818 | gfp_t mask) | |
819 | { | |
820 | return set_extent_bit(tree, start, end, EXTENT_WRITEBACK, | |
821 | 0, NULL, mask); | |
822 | } | |
823 | EXPORT_SYMBOL(set_extent_writeback); | |
824 | ||
825 | int clear_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end, | |
826 | gfp_t mask) | |
827 | { | |
828 | return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask); | |
829 | } | |
830 | EXPORT_SYMBOL(clear_extent_writeback); | |
831 | ||
832 | int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end) | |
833 | { | |
834 | return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK); | |
835 | } | |
836 | EXPORT_SYMBOL(wait_on_extent_writeback); | |
837 | ||
d1310b2e CM |
838 | int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) |
839 | { | |
840 | int err; | |
841 | u64 failed_start; | |
842 | while (1) { | |
843 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1, | |
844 | &failed_start, mask); | |
845 | if (err == -EEXIST && (mask & __GFP_WAIT)) { | |
846 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); | |
847 | start = failed_start; | |
848 | } else { | |
849 | break; | |
850 | } | |
851 | WARN_ON(start > end); | |
852 | } | |
853 | return err; | |
854 | } | |
855 | EXPORT_SYMBOL(lock_extent); | |
856 | ||
857 | int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, | |
858 | gfp_t mask) | |
859 | { | |
860 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask); | |
861 | } | |
862 | EXPORT_SYMBOL(unlock_extent); | |
863 | ||
864 | /* | |
865 | * helper function to set pages and extents in the tree dirty | |
866 | */ | |
867 | int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end) | |
868 | { | |
869 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
870 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
871 | struct page *page; | |
872 | ||
873 | while (index <= end_index) { | |
874 | page = find_get_page(tree->mapping, index); | |
875 | BUG_ON(!page); | |
876 | __set_page_dirty_nobuffers(page); | |
877 | page_cache_release(page); | |
878 | index++; | |
879 | } | |
880 | set_extent_dirty(tree, start, end, GFP_NOFS); | |
881 | return 0; | |
882 | } | |
883 | EXPORT_SYMBOL(set_range_dirty); | |
884 | ||
885 | /* | |
886 | * helper function to set both pages and extents in the tree writeback | |
887 | */ | |
888 | int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) | |
889 | { | |
890 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
891 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
892 | struct page *page; | |
893 | ||
894 | while (index <= end_index) { | |
895 | page = find_get_page(tree->mapping, index); | |
896 | BUG_ON(!page); | |
897 | set_page_writeback(page); | |
898 | page_cache_release(page); | |
899 | index++; | |
900 | } | |
901 | set_extent_writeback(tree, start, end, GFP_NOFS); | |
902 | return 0; | |
903 | } | |
904 | EXPORT_SYMBOL(set_range_writeback); | |
905 | ||
906 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
907 | u64 *start_ret, u64 *end_ret, int bits) | |
908 | { | |
909 | struct rb_node *node; | |
910 | struct extent_state *state; | |
911 | int ret = 1; | |
912 | ||
70dec807 | 913 | spin_lock_irq(&tree->lock); |
d1310b2e CM |
914 | /* |
915 | * this search will find all the extents that end after | |
916 | * our range starts. | |
917 | */ | |
918 | node = tree_search(&tree->state, start); | |
919 | if (!node || IS_ERR(node)) { | |
920 | goto out; | |
921 | } | |
922 | ||
923 | while(1) { | |
924 | state = rb_entry(node, struct extent_state, rb_node); | |
925 | if (state->end >= start && (state->state & bits)) { | |
926 | *start_ret = state->start; | |
927 | *end_ret = state->end; | |
928 | ret = 0; | |
929 | break; | |
930 | } | |
931 | node = rb_next(node); | |
932 | if (!node) | |
933 | break; | |
934 | } | |
935 | out: | |
70dec807 | 936 | spin_unlock_irq(&tree->lock); |
d1310b2e CM |
937 | return ret; |
938 | } | |
939 | EXPORT_SYMBOL(find_first_extent_bit); | |
940 | ||
941 | u64 find_lock_delalloc_range(struct extent_io_tree *tree, | |
942 | u64 *start, u64 *end, u64 max_bytes) | |
943 | { | |
944 | struct rb_node *node; | |
945 | struct extent_state *state; | |
946 | u64 cur_start = *start; | |
947 | u64 found = 0; | |
948 | u64 total_bytes = 0; | |
949 | ||
70dec807 | 950 | spin_lock_irq(&tree->lock); |
d1310b2e CM |
951 | /* |
952 | * this search will find all the extents that end after | |
953 | * our range starts. | |
954 | */ | |
955 | search_again: | |
956 | node = tree_search(&tree->state, cur_start); | |
957 | if (!node || IS_ERR(node)) { | |
958 | *end = (u64)-1; | |
959 | goto out; | |
960 | } | |
961 | ||
962 | while(1) { | |
963 | state = rb_entry(node, struct extent_state, rb_node); | |
964 | if (found && state->start != cur_start) { | |
965 | goto out; | |
966 | } | |
967 | if (!(state->state & EXTENT_DELALLOC)) { | |
968 | if (!found) | |
969 | *end = state->end; | |
970 | goto out; | |
971 | } | |
972 | if (!found) { | |
973 | struct extent_state *prev_state; | |
974 | struct rb_node *prev_node = node; | |
975 | while(1) { | |
976 | prev_node = rb_prev(prev_node); | |
977 | if (!prev_node) | |
978 | break; | |
979 | prev_state = rb_entry(prev_node, | |
980 | struct extent_state, | |
981 | rb_node); | |
982 | if (!(prev_state->state & EXTENT_DELALLOC)) | |
983 | break; | |
984 | state = prev_state; | |
985 | node = prev_node; | |
986 | } | |
987 | } | |
988 | if (state->state & EXTENT_LOCKED) { | |
989 | DEFINE_WAIT(wait); | |
990 | atomic_inc(&state->refs); | |
991 | prepare_to_wait(&state->wq, &wait, | |
992 | TASK_UNINTERRUPTIBLE); | |
70dec807 | 993 | spin_unlock_irq(&tree->lock); |
d1310b2e | 994 | schedule(); |
70dec807 | 995 | spin_lock_irq(&tree->lock); |
d1310b2e CM |
996 | finish_wait(&state->wq, &wait); |
997 | free_extent_state(state); | |
998 | goto search_again; | |
999 | } | |
1000 | state->state |= EXTENT_LOCKED; | |
291d673e | 1001 | set_state_cb(tree, state, EXTENT_LOCKED); |
d1310b2e CM |
1002 | if (!found) |
1003 | *start = state->start; | |
1004 | found++; | |
1005 | *end = state->end; | |
1006 | cur_start = state->end + 1; | |
1007 | node = rb_next(node); | |
1008 | if (!node) | |
1009 | break; | |
1010 | total_bytes += state->end - state->start + 1; | |
1011 | if (total_bytes >= max_bytes) | |
1012 | break; | |
1013 | } | |
1014 | out: | |
70dec807 | 1015 | spin_unlock_irq(&tree->lock); |
d1310b2e CM |
1016 | return found; |
1017 | } | |
1018 | ||
1019 | u64 count_range_bits(struct extent_io_tree *tree, | |
1020 | u64 *start, u64 search_end, u64 max_bytes, | |
1021 | unsigned long bits) | |
1022 | { | |
1023 | struct rb_node *node; | |
1024 | struct extent_state *state; | |
1025 | u64 cur_start = *start; | |
1026 | u64 total_bytes = 0; | |
1027 | int found = 0; | |
1028 | ||
1029 | if (search_end <= cur_start) { | |
1030 | printk("search_end %Lu start %Lu\n", search_end, cur_start); | |
1031 | WARN_ON(1); | |
1032 | return 0; | |
1033 | } | |
1034 | ||
70dec807 | 1035 | spin_lock_irq(&tree->lock); |
d1310b2e CM |
1036 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1037 | total_bytes = tree->dirty_bytes; | |
1038 | goto out; | |
1039 | } | |
1040 | /* | |
1041 | * this search will find all the extents that end after | |
1042 | * our range starts. | |
1043 | */ | |
1044 | node = tree_search(&tree->state, cur_start); | |
1045 | if (!node || IS_ERR(node)) { | |
1046 | goto out; | |
1047 | } | |
1048 | ||
1049 | while(1) { | |
1050 | state = rb_entry(node, struct extent_state, rb_node); | |
1051 | if (state->start > search_end) | |
1052 | break; | |
1053 | if (state->end >= cur_start && (state->state & bits)) { | |
1054 | total_bytes += min(search_end, state->end) + 1 - | |
1055 | max(cur_start, state->start); | |
1056 | if (total_bytes >= max_bytes) | |
1057 | break; | |
1058 | if (!found) { | |
1059 | *start = state->start; | |
1060 | found = 1; | |
1061 | } | |
1062 | } | |
1063 | node = rb_next(node); | |
1064 | if (!node) | |
1065 | break; | |
1066 | } | |
1067 | out: | |
70dec807 | 1068 | spin_unlock_irq(&tree->lock); |
d1310b2e CM |
1069 | return total_bytes; |
1070 | } | |
1071 | /* | |
1072 | * helper function to lock both pages and extents in the tree. | |
1073 | * pages must be locked first. | |
1074 | */ | |
1075 | int lock_range(struct extent_io_tree *tree, u64 start, u64 end) | |
1076 | { | |
1077 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1078 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1079 | struct page *page; | |
1080 | int err; | |
1081 | ||
1082 | while (index <= end_index) { | |
1083 | page = grab_cache_page(tree->mapping, index); | |
1084 | if (!page) { | |
1085 | err = -ENOMEM; | |
1086 | goto failed; | |
1087 | } | |
1088 | if (IS_ERR(page)) { | |
1089 | err = PTR_ERR(page); | |
1090 | goto failed; | |
1091 | } | |
1092 | index++; | |
1093 | } | |
1094 | lock_extent(tree, start, end, GFP_NOFS); | |
1095 | return 0; | |
1096 | ||
1097 | failed: | |
1098 | /* | |
1099 | * we failed above in getting the page at 'index', so we undo here | |
1100 | * up to but not including the page at 'index' | |
1101 | */ | |
1102 | end_index = index; | |
1103 | index = start >> PAGE_CACHE_SHIFT; | |
1104 | while (index < end_index) { | |
1105 | page = find_get_page(tree->mapping, index); | |
1106 | unlock_page(page); | |
1107 | page_cache_release(page); | |
1108 | index++; | |
1109 | } | |
1110 | return err; | |
1111 | } | |
1112 | EXPORT_SYMBOL(lock_range); | |
1113 | ||
1114 | /* | |
1115 | * helper function to unlock both pages and extents in the tree. | |
1116 | */ | |
1117 | int unlock_range(struct extent_io_tree *tree, u64 start, u64 end) | |
1118 | { | |
1119 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1120 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1121 | struct page *page; | |
1122 | ||
1123 | while (index <= end_index) { | |
1124 | page = find_get_page(tree->mapping, index); | |
1125 | unlock_page(page); | |
1126 | page_cache_release(page); | |
1127 | index++; | |
1128 | } | |
1129 | unlock_extent(tree, start, end, GFP_NOFS); | |
1130 | return 0; | |
1131 | } | |
1132 | EXPORT_SYMBOL(unlock_range); | |
1133 | ||
1134 | int set_state_private(struct extent_io_tree *tree, u64 start, u64 private) | |
1135 | { | |
1136 | struct rb_node *node; | |
1137 | struct extent_state *state; | |
1138 | int ret = 0; | |
1139 | ||
70dec807 | 1140 | spin_lock_irq(&tree->lock); |
d1310b2e CM |
1141 | /* |
1142 | * this search will find all the extents that end after | |
1143 | * our range starts. | |
1144 | */ | |
1145 | node = tree_search(&tree->state, start); | |
1146 | if (!node || IS_ERR(node)) { | |
1147 | ret = -ENOENT; | |
1148 | goto out; | |
1149 | } | |
1150 | state = rb_entry(node, struct extent_state, rb_node); | |
1151 | if (state->start != start) { | |
1152 | ret = -ENOENT; | |
1153 | goto out; | |
1154 | } | |
1155 | state->private = private; | |
1156 | out: | |
70dec807 | 1157 | spin_unlock_irq(&tree->lock); |
d1310b2e CM |
1158 | return ret; |
1159 | } | |
1160 | ||
1161 | int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private) | |
1162 | { | |
1163 | struct rb_node *node; | |
1164 | struct extent_state *state; | |
1165 | int ret = 0; | |
1166 | ||
70dec807 | 1167 | spin_lock_irq(&tree->lock); |
d1310b2e CM |
1168 | /* |
1169 | * this search will find all the extents that end after | |
1170 | * our range starts. | |
1171 | */ | |
1172 | node = tree_search(&tree->state, start); | |
1173 | if (!node || IS_ERR(node)) { | |
1174 | ret = -ENOENT; | |
1175 | goto out; | |
1176 | } | |
1177 | state = rb_entry(node, struct extent_state, rb_node); | |
1178 | if (state->start != start) { | |
1179 | ret = -ENOENT; | |
1180 | goto out; | |
1181 | } | |
1182 | *private = state->private; | |
1183 | out: | |
70dec807 | 1184 | spin_unlock_irq(&tree->lock); |
d1310b2e CM |
1185 | return ret; |
1186 | } | |
1187 | ||
1188 | /* | |
1189 | * searches a range in the state tree for a given mask. | |
70dec807 | 1190 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1191 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1192 | * range is found set. | |
1193 | */ | |
1194 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
1195 | int bits, int filled) | |
1196 | { | |
1197 | struct extent_state *state = NULL; | |
1198 | struct rb_node *node; | |
1199 | int bitset = 0; | |
1200 | unsigned long flags; | |
1201 | ||
70dec807 | 1202 | spin_lock_irqsave(&tree->lock, flags); |
d1310b2e CM |
1203 | node = tree_search(&tree->state, start); |
1204 | while (node && start <= end) { | |
1205 | state = rb_entry(node, struct extent_state, rb_node); | |
1206 | ||
1207 | if (filled && state->start > start) { | |
1208 | bitset = 0; | |
1209 | break; | |
1210 | } | |
1211 | ||
1212 | if (state->start > end) | |
1213 | break; | |
1214 | ||
1215 | if (state->state & bits) { | |
1216 | bitset = 1; | |
1217 | if (!filled) | |
1218 | break; | |
1219 | } else if (filled) { | |
1220 | bitset = 0; | |
1221 | break; | |
1222 | } | |
1223 | start = state->end + 1; | |
1224 | if (start > end) | |
1225 | break; | |
1226 | node = rb_next(node); | |
1227 | if (!node) { | |
1228 | if (filled) | |
1229 | bitset = 0; | |
1230 | break; | |
1231 | } | |
1232 | } | |
70dec807 | 1233 | spin_unlock_irqrestore(&tree->lock, flags); |
d1310b2e CM |
1234 | return bitset; |
1235 | } | |
1236 | EXPORT_SYMBOL(test_range_bit); | |
1237 | ||
1238 | /* | |
1239 | * helper function to set a given page up to date if all the | |
1240 | * extents in the tree for that page are up to date | |
1241 | */ | |
1242 | static int check_page_uptodate(struct extent_io_tree *tree, | |
1243 | struct page *page) | |
1244 | { | |
1245 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1246 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
1247 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1)) | |
1248 | SetPageUptodate(page); | |
1249 | return 0; | |
1250 | } | |
1251 | ||
1252 | /* | |
1253 | * helper function to unlock a page if all the extents in the tree | |
1254 | * for that page are unlocked | |
1255 | */ | |
1256 | static int check_page_locked(struct extent_io_tree *tree, | |
1257 | struct page *page) | |
1258 | { | |
1259 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1260 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
1261 | if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0)) | |
1262 | unlock_page(page); | |
1263 | return 0; | |
1264 | } | |
1265 | ||
1266 | /* | |
1267 | * helper function to end page writeback if all the extents | |
1268 | * in the tree for that page are done with writeback | |
1269 | */ | |
1270 | static int check_page_writeback(struct extent_io_tree *tree, | |
1271 | struct page *page) | |
1272 | { | |
1273 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1274 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
1275 | if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0)) | |
1276 | end_page_writeback(page); | |
1277 | return 0; | |
1278 | } | |
1279 | ||
1280 | /* lots and lots of room for performance fixes in the end_bio funcs */ | |
1281 | ||
1282 | /* | |
1283 | * after a writepage IO is done, we need to: | |
1284 | * clear the uptodate bits on error | |
1285 | * clear the writeback bits in the extent tree for this IO | |
1286 | * end_page_writeback if the page has no more pending IO | |
1287 | * | |
1288 | * Scheduling is not allowed, so the extent state tree is expected | |
1289 | * to have one and only one object corresponding to this IO. | |
1290 | */ | |
1291 | #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23) | |
1292 | static void end_bio_extent_writepage(struct bio *bio, int err) | |
1293 | #else | |
1294 | static int end_bio_extent_writepage(struct bio *bio, | |
1295 | unsigned int bytes_done, int err) | |
1296 | #endif | |
1297 | { | |
1298 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
1299 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
70dec807 CM |
1300 | struct extent_state *state = bio->bi_private; |
1301 | struct extent_io_tree *tree = state->tree; | |
1302 | struct rb_node *node; | |
d1310b2e CM |
1303 | u64 start; |
1304 | u64 end; | |
70dec807 | 1305 | u64 cur; |
d1310b2e | 1306 | int whole_page; |
70dec807 | 1307 | unsigned long flags; |
d1310b2e CM |
1308 | |
1309 | #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23) | |
1310 | if (bio->bi_size) | |
1311 | return 1; | |
1312 | #endif | |
d1310b2e CM |
1313 | do { |
1314 | struct page *page = bvec->bv_page; | |
1315 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + | |
1316 | bvec->bv_offset; | |
1317 | end = start + bvec->bv_len - 1; | |
1318 | ||
1319 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
1320 | whole_page = 1; | |
1321 | else | |
1322 | whole_page = 0; | |
1323 | ||
1324 | if (--bvec >= bio->bi_io_vec) | |
1325 | prefetchw(&bvec->bv_page->flags); | |
1326 | ||
1327 | if (!uptodate) { | |
1328 | clear_extent_uptodate(tree, start, end, GFP_ATOMIC); | |
1329 | ClearPageUptodate(page); | |
1330 | SetPageError(page); | |
1331 | } | |
70dec807 CM |
1332 | |
1333 | if (tree->ops && tree->ops->writepage_end_io_hook) { | |
1334 | tree->ops->writepage_end_io_hook(page, start, end, | |
1335 | state); | |
1336 | } | |
1337 | ||
1338 | /* | |
1339 | * bios can get merged in funny ways, and so we need to | |
1340 | * be careful with the state variable. We know the | |
1341 | * state won't be merged with others because it has | |
1342 | * WRITEBACK set, but we can't be sure each biovec is | |
1343 | * sequential in the file. So, if our cached state | |
1344 | * doesn't match the expected end, search the tree | |
1345 | * for the correct one. | |
1346 | */ | |
1347 | ||
1348 | spin_lock_irqsave(&tree->lock, flags); | |
1349 | if (!state || state->end != end) { | |
1350 | state = NULL; | |
1351 | node = __tree_search(&tree->state, start, NULL, NULL); | |
1352 | if (node) { | |
1353 | state = rb_entry(node, struct extent_state, | |
1354 | rb_node); | |
1355 | if (state->end != end || | |
1356 | !(state->state & EXTENT_WRITEBACK)) | |
1357 | state = NULL; | |
1358 | } | |
1359 | if (!state) { | |
1360 | spin_unlock_irqrestore(&tree->lock, flags); | |
1361 | clear_extent_writeback(tree, start, | |
1362 | end, GFP_ATOMIC); | |
1363 | goto next_io; | |
1364 | } | |
1365 | } | |
1366 | cur = end; | |
1367 | while(1) { | |
1368 | struct extent_state *clear = state; | |
1369 | cur = state->start; | |
1370 | node = rb_prev(&state->rb_node); | |
1371 | if (node) { | |
1372 | state = rb_entry(node, | |
1373 | struct extent_state, | |
1374 | rb_node); | |
1375 | } else { | |
1376 | state = NULL; | |
1377 | } | |
1378 | ||
1379 | clear_state_bit(tree, clear, EXTENT_WRITEBACK, | |
1380 | 1, 0); | |
1381 | if (cur == start) | |
1382 | break; | |
1383 | if (cur < start) { | |
1384 | WARN_ON(1); | |
1385 | break; | |
1386 | } | |
1387 | if (!node) | |
1388 | break; | |
1389 | } | |
1390 | /* before releasing the lock, make sure the next state | |
1391 | * variable has the expected bits set and corresponds | |
1392 | * to the correct offsets in the file | |
1393 | */ | |
1394 | if (state && (state->end + 1 != start || | |
1395 | !state->state & EXTENT_WRITEBACK)) { | |
1396 | state = NULL; | |
1397 | } | |
1398 | spin_unlock_irqrestore(&tree->lock, flags); | |
1399 | next_io: | |
d1310b2e CM |
1400 | |
1401 | if (whole_page) | |
1402 | end_page_writeback(page); | |
1403 | else | |
1404 | check_page_writeback(tree, page); | |
d1310b2e | 1405 | } while (bvec >= bio->bi_io_vec); |
d1310b2e CM |
1406 | bio_put(bio); |
1407 | #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23) | |
1408 | return 0; | |
1409 | #endif | |
1410 | } | |
1411 | ||
1412 | /* | |
1413 | * after a readpage IO is done, we need to: | |
1414 | * clear the uptodate bits on error | |
1415 | * set the uptodate bits if things worked | |
1416 | * set the page up to date if all extents in the tree are uptodate | |
1417 | * clear the lock bit in the extent tree | |
1418 | * unlock the page if there are no other extents locked for it | |
1419 | * | |
1420 | * Scheduling is not allowed, so the extent state tree is expected | |
1421 | * to have one and only one object corresponding to this IO. | |
1422 | */ | |
1423 | #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23) | |
1424 | static void end_bio_extent_readpage(struct bio *bio, int err) | |
1425 | #else | |
1426 | static int end_bio_extent_readpage(struct bio *bio, | |
1427 | unsigned int bytes_done, int err) | |
1428 | #endif | |
1429 | { | |
1430 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
1431 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
70dec807 CM |
1432 | struct extent_state *state = bio->bi_private; |
1433 | struct extent_io_tree *tree = state->tree; | |
1434 | struct rb_node *node; | |
d1310b2e CM |
1435 | u64 start; |
1436 | u64 end; | |
70dec807 CM |
1437 | u64 cur; |
1438 | unsigned long flags; | |
d1310b2e CM |
1439 | int whole_page; |
1440 | int ret; | |
1441 | ||
1442 | #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23) | |
1443 | if (bio->bi_size) | |
1444 | return 1; | |
1445 | #endif | |
1446 | ||
1447 | do { | |
1448 | struct page *page = bvec->bv_page; | |
1449 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + | |
1450 | bvec->bv_offset; | |
1451 | end = start + bvec->bv_len - 1; | |
1452 | ||
1453 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
1454 | whole_page = 1; | |
1455 | else | |
1456 | whole_page = 0; | |
1457 | ||
1458 | if (--bvec >= bio->bi_io_vec) | |
1459 | prefetchw(&bvec->bv_page->flags); | |
1460 | ||
1461 | if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) { | |
70dec807 CM |
1462 | ret = tree->ops->readpage_end_io_hook(page, start, end, |
1463 | state); | |
d1310b2e CM |
1464 | if (ret) |
1465 | uptodate = 0; | |
1466 | } | |
d1310b2e | 1467 | |
70dec807 CM |
1468 | spin_lock_irqsave(&tree->lock, flags); |
1469 | if (!state || state->end != end) { | |
1470 | state = NULL; | |
1471 | node = __tree_search(&tree->state, start, NULL, NULL); | |
1472 | if (node) { | |
1473 | state = rb_entry(node, struct extent_state, | |
1474 | rb_node); | |
1475 | if (state->end != end || | |
1476 | !(state->state & EXTENT_LOCKED)) | |
1477 | state = NULL; | |
1478 | } | |
1479 | if (!state) { | |
1480 | spin_unlock_irqrestore(&tree->lock, flags); | |
1481 | set_extent_uptodate(tree, start, end, | |
1482 | GFP_ATOMIC); | |
1483 | unlock_extent(tree, start, end, GFP_ATOMIC); | |
1484 | goto next_io; | |
1485 | } | |
1486 | } | |
d1310b2e | 1487 | |
70dec807 CM |
1488 | cur = end; |
1489 | while(1) { | |
1490 | struct extent_state *clear = state; | |
1491 | cur = state->start; | |
1492 | node = rb_prev(&state->rb_node); | |
1493 | if (node) { | |
1494 | state = rb_entry(node, | |
1495 | struct extent_state, | |
1496 | rb_node); | |
1497 | } else { | |
1498 | state = NULL; | |
1499 | } | |
1500 | clear->state |= EXTENT_UPTODATE; | |
291d673e | 1501 | set_state_cb(tree, clear, EXTENT_UPTODATE); |
70dec807 CM |
1502 | clear_state_bit(tree, clear, EXTENT_LOCKED, |
1503 | 1, 0); | |
1504 | if (cur == start) | |
1505 | break; | |
1506 | if (cur < start) { | |
1507 | WARN_ON(1); | |
1508 | break; | |
1509 | } | |
1510 | if (!node) | |
1511 | break; | |
1512 | } | |
1513 | /* before releasing the lock, make sure the next state | |
1514 | * variable has the expected bits set and corresponds | |
1515 | * to the correct offsets in the file | |
1516 | */ | |
1517 | if (state && (state->end + 1 != start || | |
1518 | !state->state & EXTENT_WRITEBACK)) { | |
1519 | state = NULL; | |
1520 | } | |
1521 | spin_unlock_irqrestore(&tree->lock, flags); | |
1522 | next_io: | |
1523 | if (whole_page) { | |
1524 | if (uptodate) { | |
1525 | SetPageUptodate(page); | |
1526 | } else { | |
1527 | ClearPageUptodate(page); | |
1528 | SetPageError(page); | |
1529 | } | |
d1310b2e | 1530 | unlock_page(page); |
70dec807 CM |
1531 | } else { |
1532 | if (uptodate) { | |
1533 | check_page_uptodate(tree, page); | |
1534 | } else { | |
1535 | ClearPageUptodate(page); | |
1536 | SetPageError(page); | |
1537 | } | |
d1310b2e | 1538 | check_page_locked(tree, page); |
70dec807 | 1539 | } |
d1310b2e CM |
1540 | } while (bvec >= bio->bi_io_vec); |
1541 | ||
1542 | bio_put(bio); | |
1543 | #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23) | |
1544 | return 0; | |
1545 | #endif | |
1546 | } | |
1547 | ||
1548 | /* | |
1549 | * IO done from prepare_write is pretty simple, we just unlock | |
1550 | * the structs in the extent tree when done, and set the uptodate bits | |
1551 | * as appropriate. | |
1552 | */ | |
1553 | #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23) | |
1554 | static void end_bio_extent_preparewrite(struct bio *bio, int err) | |
1555 | #else | |
1556 | static int end_bio_extent_preparewrite(struct bio *bio, | |
1557 | unsigned int bytes_done, int err) | |
1558 | #endif | |
1559 | { | |
1560 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
1561 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
70dec807 CM |
1562 | struct extent_state *state = bio->bi_private; |
1563 | struct extent_io_tree *tree = state->tree; | |
d1310b2e CM |
1564 | u64 start; |
1565 | u64 end; | |
1566 | ||
1567 | #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23) | |
1568 | if (bio->bi_size) | |
1569 | return 1; | |
1570 | #endif | |
1571 | ||
1572 | do { | |
1573 | struct page *page = bvec->bv_page; | |
1574 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + | |
1575 | bvec->bv_offset; | |
1576 | end = start + bvec->bv_len - 1; | |
1577 | ||
1578 | if (--bvec >= bio->bi_io_vec) | |
1579 | prefetchw(&bvec->bv_page->flags); | |
1580 | ||
1581 | if (uptodate) { | |
1582 | set_extent_uptodate(tree, start, end, GFP_ATOMIC); | |
1583 | } else { | |
1584 | ClearPageUptodate(page); | |
1585 | SetPageError(page); | |
1586 | } | |
1587 | ||
1588 | unlock_extent(tree, start, end, GFP_ATOMIC); | |
1589 | ||
1590 | } while (bvec >= bio->bi_io_vec); | |
1591 | ||
1592 | bio_put(bio); | |
1593 | #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23) | |
1594 | return 0; | |
1595 | #endif | |
1596 | } | |
1597 | ||
1598 | static struct bio * | |
1599 | extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
1600 | gfp_t gfp_flags) | |
1601 | { | |
1602 | struct bio *bio; | |
1603 | ||
1604 | bio = bio_alloc(gfp_flags, nr_vecs); | |
1605 | ||
1606 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
1607 | while (!bio && (nr_vecs /= 2)) | |
1608 | bio = bio_alloc(gfp_flags, nr_vecs); | |
1609 | } | |
1610 | ||
1611 | if (bio) { | |
1612 | bio->bi_bdev = bdev; | |
1613 | bio->bi_sector = first_sector; | |
1614 | } | |
1615 | return bio; | |
1616 | } | |
1617 | ||
1618 | static int submit_one_bio(int rw, struct bio *bio) | |
1619 | { | |
1620 | u64 maxsector; | |
1621 | int ret = 0; | |
70dec807 CM |
1622 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
1623 | struct page *page = bvec->bv_page; | |
1624 | struct extent_io_tree *tree = bio->bi_private; | |
1625 | struct rb_node *node; | |
1626 | struct extent_state *state; | |
1627 | u64 start; | |
1628 | u64 end; | |
1629 | ||
1630 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset; | |
1631 | end = start + bvec->bv_len - 1; | |
1632 | ||
1633 | spin_lock_irq(&tree->lock); | |
1634 | node = __tree_search(&tree->state, start, NULL, NULL); | |
1635 | BUG_ON(!node); | |
1636 | state = rb_entry(node, struct extent_state, rb_node); | |
1637 | while(state->end < end) { | |
1638 | node = rb_next(node); | |
1639 | state = rb_entry(node, struct extent_state, rb_node); | |
1640 | } | |
1641 | BUG_ON(state->end != end); | |
1642 | spin_unlock_irq(&tree->lock); | |
1643 | ||
1644 | bio->bi_private = state; | |
d1310b2e CM |
1645 | |
1646 | bio_get(bio); | |
1647 | ||
1648 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; | |
1649 | if (maxsector < bio->bi_sector) { | |
1650 | printk("sector too large max %Lu got %llu\n", maxsector, | |
1651 | (unsigned long long)bio->bi_sector); | |
1652 | WARN_ON(1); | |
1653 | } | |
1654 | ||
1655 | submit_bio(rw, bio); | |
1656 | if (bio_flagged(bio, BIO_EOPNOTSUPP)) | |
1657 | ret = -EOPNOTSUPP; | |
1658 | bio_put(bio); | |
1659 | return ret; | |
1660 | } | |
1661 | ||
1662 | static int submit_extent_page(int rw, struct extent_io_tree *tree, | |
1663 | struct page *page, sector_t sector, | |
1664 | size_t size, unsigned long offset, | |
1665 | struct block_device *bdev, | |
1666 | struct bio **bio_ret, | |
1667 | unsigned long max_pages, | |
1668 | bio_end_io_t end_io_func) | |
1669 | { | |
1670 | int ret = 0; | |
1671 | struct bio *bio; | |
1672 | int nr; | |
1673 | ||
1674 | if (bio_ret && *bio_ret) { | |
1675 | bio = *bio_ret; | |
1676 | if (bio->bi_sector + (bio->bi_size >> 9) != sector || | |
1677 | bio_add_page(bio, page, size, offset) < size) { | |
1678 | ret = submit_one_bio(rw, bio); | |
1679 | bio = NULL; | |
1680 | } else { | |
1681 | return 0; | |
1682 | } | |
1683 | } | |
1684 | nr = min_t(int, max_pages, bio_get_nr_vecs(bdev)); | |
1685 | bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH); | |
1686 | if (!bio) { | |
1687 | printk("failed to allocate bio nr %d\n", nr); | |
1688 | } | |
70dec807 CM |
1689 | |
1690 | ||
d1310b2e CM |
1691 | bio_add_page(bio, page, size, offset); |
1692 | bio->bi_end_io = end_io_func; | |
1693 | bio->bi_private = tree; | |
70dec807 | 1694 | |
d1310b2e CM |
1695 | if (bio_ret) { |
1696 | *bio_ret = bio; | |
1697 | } else { | |
1698 | ret = submit_one_bio(rw, bio); | |
1699 | } | |
1700 | ||
1701 | return ret; | |
1702 | } | |
1703 | ||
1704 | void set_page_extent_mapped(struct page *page) | |
1705 | { | |
1706 | if (!PagePrivate(page)) { | |
1707 | SetPagePrivate(page); | |
1708 | WARN_ON(!page->mapping->a_ops->invalidatepage); | |
1709 | set_page_private(page, EXTENT_PAGE_PRIVATE); | |
1710 | page_cache_get(page); | |
1711 | } | |
1712 | } | |
1713 | ||
1714 | void set_page_extent_head(struct page *page, unsigned long len) | |
1715 | { | |
1716 | set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2); | |
1717 | } | |
1718 | ||
1719 | /* | |
1720 | * basic readpage implementation. Locked extent state structs are inserted | |
1721 | * into the tree that are removed when the IO is done (by the end_io | |
1722 | * handlers) | |
1723 | */ | |
1724 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
1725 | struct page *page, | |
1726 | get_extent_t *get_extent, | |
1727 | struct bio **bio) | |
1728 | { | |
1729 | struct inode *inode = page->mapping->host; | |
1730 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1731 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
1732 | u64 end; | |
1733 | u64 cur = start; | |
1734 | u64 extent_offset; | |
1735 | u64 last_byte = i_size_read(inode); | |
1736 | u64 block_start; | |
1737 | u64 cur_end; | |
1738 | sector_t sector; | |
1739 | struct extent_map *em; | |
1740 | struct block_device *bdev; | |
1741 | int ret; | |
1742 | int nr = 0; | |
1743 | size_t page_offset = 0; | |
1744 | size_t iosize; | |
1745 | size_t blocksize = inode->i_sb->s_blocksize; | |
1746 | ||
1747 | set_page_extent_mapped(page); | |
1748 | ||
1749 | end = page_end; | |
1750 | lock_extent(tree, start, end, GFP_NOFS); | |
1751 | ||
1752 | while (cur <= end) { | |
1753 | if (cur >= last_byte) { | |
1754 | char *userpage; | |
1755 | iosize = PAGE_CACHE_SIZE - page_offset; | |
1756 | userpage = kmap_atomic(page, KM_USER0); | |
1757 | memset(userpage + page_offset, 0, iosize); | |
1758 | flush_dcache_page(page); | |
1759 | kunmap_atomic(userpage, KM_USER0); | |
1760 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
1761 | GFP_NOFS); | |
1762 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
1763 | break; | |
1764 | } | |
1765 | em = get_extent(inode, page, page_offset, cur, | |
1766 | end - cur + 1, 0); | |
1767 | if (IS_ERR(em) || !em) { | |
1768 | SetPageError(page); | |
1769 | unlock_extent(tree, cur, end, GFP_NOFS); | |
1770 | break; | |
1771 | } | |
1772 | ||
1773 | extent_offset = cur - em->start; | |
1774 | BUG_ON(extent_map_end(em) <= cur); | |
1775 | BUG_ON(end < cur); | |
1776 | ||
1777 | iosize = min(extent_map_end(em) - cur, end - cur + 1); | |
1778 | cur_end = min(extent_map_end(em) - 1, end); | |
1779 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
1780 | sector = (em->block_start + extent_offset) >> 9; | |
1781 | bdev = em->bdev; | |
1782 | block_start = em->block_start; | |
1783 | free_extent_map(em); | |
1784 | em = NULL; | |
1785 | ||
1786 | /* we've found a hole, just zero and go on */ | |
1787 | if (block_start == EXTENT_MAP_HOLE) { | |
1788 | char *userpage; | |
1789 | userpage = kmap_atomic(page, KM_USER0); | |
1790 | memset(userpage + page_offset, 0, iosize); | |
1791 | flush_dcache_page(page); | |
1792 | kunmap_atomic(userpage, KM_USER0); | |
1793 | ||
1794 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
1795 | GFP_NOFS); | |
1796 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
1797 | cur = cur + iosize; | |
1798 | page_offset += iosize; | |
1799 | continue; | |
1800 | } | |
1801 | /* the get_extent function already copied into the page */ | |
1802 | if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) { | |
1803 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
1804 | cur = cur + iosize; | |
1805 | page_offset += iosize; | |
1806 | continue; | |
1807 | } | |
70dec807 CM |
1808 | /* we have an inline extent but it didn't get marked up |
1809 | * to date. Error out | |
1810 | */ | |
1811 | if (block_start == EXTENT_MAP_INLINE) { | |
1812 | SetPageError(page); | |
1813 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
1814 | cur = cur + iosize; | |
1815 | page_offset += iosize; | |
1816 | continue; | |
1817 | } | |
d1310b2e CM |
1818 | |
1819 | ret = 0; | |
1820 | if (tree->ops && tree->ops->readpage_io_hook) { | |
1821 | ret = tree->ops->readpage_io_hook(page, cur, | |
1822 | cur + iosize - 1); | |
1823 | } | |
1824 | if (!ret) { | |
1825 | unsigned long nr = (last_byte >> PAGE_CACHE_SHIFT) + 1; | |
1826 | nr -= page->index; | |
1827 | ret = submit_extent_page(READ, tree, page, | |
1828 | sector, iosize, page_offset, | |
1829 | bdev, bio, nr, | |
1830 | end_bio_extent_readpage); | |
1831 | } | |
1832 | if (ret) | |
1833 | SetPageError(page); | |
1834 | cur = cur + iosize; | |
1835 | page_offset += iosize; | |
1836 | nr++; | |
1837 | } | |
1838 | if (!nr) { | |
1839 | if (!PageError(page)) | |
1840 | SetPageUptodate(page); | |
1841 | unlock_page(page); | |
1842 | } | |
1843 | return 0; | |
1844 | } | |
1845 | ||
1846 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, | |
1847 | get_extent_t *get_extent) | |
1848 | { | |
1849 | struct bio *bio = NULL; | |
1850 | int ret; | |
1851 | ||
1852 | ret = __extent_read_full_page(tree, page, get_extent, &bio); | |
1853 | if (bio) | |
1854 | submit_one_bio(READ, bio); | |
1855 | return ret; | |
1856 | } | |
1857 | EXPORT_SYMBOL(extent_read_full_page); | |
1858 | ||
1859 | /* | |
1860 | * the writepage semantics are similar to regular writepage. extent | |
1861 | * records are inserted to lock ranges in the tree, and as dirty areas | |
1862 | * are found, they are marked writeback. Then the lock bits are removed | |
1863 | * and the end_io handler clears the writeback ranges | |
1864 | */ | |
1865 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
1866 | void *data) | |
1867 | { | |
1868 | struct inode *inode = page->mapping->host; | |
1869 | struct extent_page_data *epd = data; | |
1870 | struct extent_io_tree *tree = epd->tree; | |
1871 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1872 | u64 delalloc_start; | |
1873 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
1874 | u64 end; | |
1875 | u64 cur = start; | |
1876 | u64 extent_offset; | |
1877 | u64 last_byte = i_size_read(inode); | |
1878 | u64 block_start; | |
1879 | u64 iosize; | |
1880 | sector_t sector; | |
1881 | struct extent_map *em; | |
1882 | struct block_device *bdev; | |
1883 | int ret; | |
1884 | int nr = 0; | |
1885 | size_t page_offset = 0; | |
1886 | size_t blocksize; | |
1887 | loff_t i_size = i_size_read(inode); | |
1888 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
1889 | u64 nr_delalloc; | |
1890 | u64 delalloc_end; | |
1891 | ||
1892 | WARN_ON(!PageLocked(page)); | |
1893 | if (page->index > end_index) { | |
1894 | clear_extent_dirty(tree, start, page_end, GFP_NOFS); | |
1895 | unlock_page(page); | |
1896 | return 0; | |
1897 | } | |
1898 | ||
1899 | if (page->index == end_index) { | |
1900 | char *userpage; | |
1901 | ||
1902 | size_t offset = i_size & (PAGE_CACHE_SIZE - 1); | |
1903 | ||
1904 | userpage = kmap_atomic(page, KM_USER0); | |
1905 | memset(userpage + offset, 0, PAGE_CACHE_SIZE - offset); | |
1906 | flush_dcache_page(page); | |
1907 | kunmap_atomic(userpage, KM_USER0); | |
1908 | } | |
1909 | ||
1910 | set_page_extent_mapped(page); | |
1911 | ||
1912 | delalloc_start = start; | |
1913 | delalloc_end = 0; | |
1914 | while(delalloc_end < page_end) { | |
1915 | nr_delalloc = find_lock_delalloc_range(tree, &delalloc_start, | |
1916 | &delalloc_end, | |
1917 | 128 * 1024 * 1024); | |
1918 | if (nr_delalloc == 0) { | |
1919 | delalloc_start = delalloc_end + 1; | |
1920 | continue; | |
1921 | } | |
1922 | tree->ops->fill_delalloc(inode, delalloc_start, | |
1923 | delalloc_end); | |
1924 | clear_extent_bit(tree, delalloc_start, | |
1925 | delalloc_end, | |
1926 | EXTENT_LOCKED | EXTENT_DELALLOC, | |
1927 | 1, 0, GFP_NOFS); | |
1928 | delalloc_start = delalloc_end + 1; | |
1929 | } | |
1930 | lock_extent(tree, start, page_end, GFP_NOFS); | |
1931 | ||
1932 | end = page_end; | |
1933 | if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) { | |
1934 | printk("found delalloc bits after lock_extent\n"); | |
1935 | } | |
1936 | ||
1937 | if (last_byte <= start) { | |
1938 | clear_extent_dirty(tree, start, page_end, GFP_NOFS); | |
1939 | goto done; | |
1940 | } | |
1941 | ||
1942 | set_extent_uptodate(tree, start, page_end, GFP_NOFS); | |
1943 | blocksize = inode->i_sb->s_blocksize; | |
1944 | ||
1945 | while (cur <= end) { | |
1946 | if (cur >= last_byte) { | |
1947 | clear_extent_dirty(tree, cur, page_end, GFP_NOFS); | |
1948 | break; | |
1949 | } | |
1950 | em = epd->get_extent(inode, page, page_offset, cur, | |
1951 | end - cur + 1, 1); | |
1952 | if (IS_ERR(em) || !em) { | |
1953 | SetPageError(page); | |
1954 | break; | |
1955 | } | |
1956 | ||
1957 | extent_offset = cur - em->start; | |
1958 | BUG_ON(extent_map_end(em) <= cur); | |
1959 | BUG_ON(end < cur); | |
1960 | iosize = min(extent_map_end(em) - cur, end - cur + 1); | |
1961 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
1962 | sector = (em->block_start + extent_offset) >> 9; | |
1963 | bdev = em->bdev; | |
1964 | block_start = em->block_start; | |
1965 | free_extent_map(em); | |
1966 | em = NULL; | |
1967 | ||
1968 | if (block_start == EXTENT_MAP_HOLE || | |
1969 | block_start == EXTENT_MAP_INLINE) { | |
1970 | clear_extent_dirty(tree, cur, | |
1971 | cur + iosize - 1, GFP_NOFS); | |
1972 | cur = cur + iosize; | |
1973 | page_offset += iosize; | |
1974 | continue; | |
1975 | } | |
1976 | ||
1977 | /* leave this out until we have a page_mkwrite call */ | |
1978 | if (0 && !test_range_bit(tree, cur, cur + iosize - 1, | |
1979 | EXTENT_DIRTY, 0)) { | |
1980 | cur = cur + iosize; | |
1981 | page_offset += iosize; | |
1982 | continue; | |
1983 | } | |
1984 | clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS); | |
1985 | if (tree->ops && tree->ops->writepage_io_hook) { | |
1986 | ret = tree->ops->writepage_io_hook(page, cur, | |
1987 | cur + iosize - 1); | |
1988 | } else { | |
1989 | ret = 0; | |
1990 | } | |
1991 | if (ret) | |
1992 | SetPageError(page); | |
1993 | else { | |
1994 | unsigned long max_nr = end_index + 1; | |
1995 | set_range_writeback(tree, cur, cur + iosize - 1); | |
1996 | if (!PageWriteback(page)) { | |
1997 | printk("warning page %lu not writeback, " | |
1998 | "cur %llu end %llu\n", page->index, | |
1999 | (unsigned long long)cur, | |
2000 | (unsigned long long)end); | |
2001 | } | |
2002 | ||
2003 | ret = submit_extent_page(WRITE, tree, page, sector, | |
2004 | iosize, page_offset, bdev, | |
2005 | &epd->bio, max_nr, | |
2006 | end_bio_extent_writepage); | |
2007 | if (ret) | |
2008 | SetPageError(page); | |
2009 | } | |
2010 | cur = cur + iosize; | |
2011 | page_offset += iosize; | |
2012 | nr++; | |
2013 | } | |
2014 | done: | |
2015 | if (nr == 0) { | |
2016 | /* make sure the mapping tag for page dirty gets cleared */ | |
2017 | set_page_writeback(page); | |
2018 | end_page_writeback(page); | |
2019 | } | |
2020 | unlock_extent(tree, start, page_end, GFP_NOFS); | |
2021 | unlock_page(page); | |
2022 | return 0; | |
2023 | } | |
2024 | ||
2025 | #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18) | |
2026 | ||
2027 | /* Taken directly from 2.6.23 for 2.6.18 back port */ | |
2028 | typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc, | |
2029 | void *data); | |
2030 | ||
2031 | /** | |
2032 | * write_cache_pages - walk the list of dirty pages of the given address space | |
2033 | * and write all of them. | |
2034 | * @mapping: address space structure to write | |
2035 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
2036 | * @writepage: function called for each page | |
2037 | * @data: data passed to writepage function | |
2038 | * | |
2039 | * If a page is already under I/O, write_cache_pages() skips it, even | |
2040 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
2041 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
2042 | * and msync() need to guarantee that all the data which was dirty at the time | |
2043 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
2044 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
2045 | * existing IO to complete. | |
2046 | */ | |
2047 | static int write_cache_pages(struct address_space *mapping, | |
2048 | struct writeback_control *wbc, writepage_t writepage, | |
2049 | void *data) | |
2050 | { | |
2051 | struct backing_dev_info *bdi = mapping->backing_dev_info; | |
2052 | int ret = 0; | |
2053 | int done = 0; | |
2054 | struct pagevec pvec; | |
2055 | int nr_pages; | |
2056 | pgoff_t index; | |
2057 | pgoff_t end; /* Inclusive */ | |
2058 | int scanned = 0; | |
2059 | int range_whole = 0; | |
2060 | ||
2061 | if (wbc->nonblocking && bdi_write_congested(bdi)) { | |
2062 | wbc->encountered_congestion = 1; | |
2063 | return 0; | |
2064 | } | |
2065 | ||
2066 | pagevec_init(&pvec, 0); | |
2067 | if (wbc->range_cyclic) { | |
2068 | index = mapping->writeback_index; /* Start from prev offset */ | |
2069 | end = -1; | |
2070 | } else { | |
2071 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
2072 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
2073 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) | |
2074 | range_whole = 1; | |
2075 | scanned = 1; | |
2076 | } | |
2077 | retry: | |
2078 | while (!done && (index <= end) && | |
2079 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
2080 | PAGECACHE_TAG_DIRTY, | |
2081 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
2082 | unsigned i; | |
2083 | ||
2084 | scanned = 1; | |
2085 | for (i = 0; i < nr_pages; i++) { | |
2086 | struct page *page = pvec.pages[i]; | |
2087 | ||
2088 | /* | |
2089 | * At this point we hold neither mapping->tree_lock nor | |
2090 | * lock on the page itself: the page may be truncated or | |
2091 | * invalidated (changing page->mapping to NULL), or even | |
2092 | * swizzled back from swapper_space to tmpfs file | |
2093 | * mapping | |
2094 | */ | |
2095 | lock_page(page); | |
2096 | ||
2097 | if (unlikely(page->mapping != mapping)) { | |
2098 | unlock_page(page); | |
2099 | continue; | |
2100 | } | |
2101 | ||
2102 | if (!wbc->range_cyclic && page->index > end) { | |
2103 | done = 1; | |
2104 | unlock_page(page); | |
2105 | continue; | |
2106 | } | |
2107 | ||
2108 | if (wbc->sync_mode != WB_SYNC_NONE) | |
2109 | wait_on_page_writeback(page); | |
2110 | ||
2111 | if (PageWriteback(page) || | |
2112 | !clear_page_dirty_for_io(page)) { | |
2113 | unlock_page(page); | |
2114 | continue; | |
2115 | } | |
2116 | ||
2117 | ret = (*writepage)(page, wbc, data); | |
2118 | ||
2119 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
2120 | unlock_page(page); | |
2121 | ret = 0; | |
2122 | } | |
2123 | if (ret || (--(wbc->nr_to_write) <= 0)) | |
2124 | done = 1; | |
2125 | if (wbc->nonblocking && bdi_write_congested(bdi)) { | |
2126 | wbc->encountered_congestion = 1; | |
2127 | done = 1; | |
2128 | } | |
2129 | } | |
2130 | pagevec_release(&pvec); | |
2131 | cond_resched(); | |
2132 | } | |
2133 | if (!scanned && !done) { | |
2134 | /* | |
2135 | * We hit the last page and there is more work to be done: wrap | |
2136 | * back to the start of the file | |
2137 | */ | |
2138 | scanned = 1; | |
2139 | index = 0; | |
2140 | goto retry; | |
2141 | } | |
2142 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) | |
2143 | mapping->writeback_index = index; | |
2144 | return ret; | |
2145 | } | |
2146 | #endif | |
2147 | ||
2148 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, | |
2149 | get_extent_t *get_extent, | |
2150 | struct writeback_control *wbc) | |
2151 | { | |
2152 | int ret; | |
2153 | struct address_space *mapping = page->mapping; | |
2154 | struct extent_page_data epd = { | |
2155 | .bio = NULL, | |
2156 | .tree = tree, | |
2157 | .get_extent = get_extent, | |
2158 | }; | |
2159 | struct writeback_control wbc_writepages = { | |
2160 | .bdi = wbc->bdi, | |
2161 | .sync_mode = WB_SYNC_NONE, | |
2162 | .older_than_this = NULL, | |
2163 | .nr_to_write = 64, | |
2164 | .range_start = page_offset(page) + PAGE_CACHE_SIZE, | |
2165 | .range_end = (loff_t)-1, | |
2166 | }; | |
2167 | ||
2168 | ||
2169 | ret = __extent_writepage(page, wbc, &epd); | |
2170 | ||
2171 | write_cache_pages(mapping, &wbc_writepages, __extent_writepage, &epd); | |
2172 | if (epd.bio) { | |
2173 | submit_one_bio(WRITE, epd.bio); | |
2174 | } | |
2175 | return ret; | |
2176 | } | |
2177 | EXPORT_SYMBOL(extent_write_full_page); | |
2178 | ||
2179 | ||
2180 | int extent_writepages(struct extent_io_tree *tree, | |
2181 | struct address_space *mapping, | |
2182 | get_extent_t *get_extent, | |
2183 | struct writeback_control *wbc) | |
2184 | { | |
2185 | int ret = 0; | |
2186 | struct extent_page_data epd = { | |
2187 | .bio = NULL, | |
2188 | .tree = tree, | |
2189 | .get_extent = get_extent, | |
2190 | }; | |
2191 | ||
2192 | ret = write_cache_pages(mapping, wbc, __extent_writepage, &epd); | |
2193 | if (epd.bio) { | |
2194 | submit_one_bio(WRITE, epd.bio); | |
2195 | } | |
2196 | return ret; | |
2197 | } | |
2198 | EXPORT_SYMBOL(extent_writepages); | |
2199 | ||
2200 | int extent_readpages(struct extent_io_tree *tree, | |
2201 | struct address_space *mapping, | |
2202 | struct list_head *pages, unsigned nr_pages, | |
2203 | get_extent_t get_extent) | |
2204 | { | |
2205 | struct bio *bio = NULL; | |
2206 | unsigned page_idx; | |
2207 | struct pagevec pvec; | |
2208 | ||
2209 | pagevec_init(&pvec, 0); | |
2210 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { | |
2211 | struct page *page = list_entry(pages->prev, struct page, lru); | |
2212 | ||
2213 | prefetchw(&page->flags); | |
2214 | list_del(&page->lru); | |
2215 | /* | |
2216 | * what we want to do here is call add_to_page_cache_lru, | |
2217 | * but that isn't exported, so we reproduce it here | |
2218 | */ | |
2219 | if (!add_to_page_cache(page, mapping, | |
2220 | page->index, GFP_KERNEL)) { | |
2221 | ||
2222 | /* open coding of lru_cache_add, also not exported */ | |
2223 | page_cache_get(page); | |
2224 | if (!pagevec_add(&pvec, page)) | |
2225 | __pagevec_lru_add(&pvec); | |
2226 | __extent_read_full_page(tree, page, get_extent, &bio); | |
2227 | } | |
2228 | page_cache_release(page); | |
2229 | } | |
2230 | if (pagevec_count(&pvec)) | |
2231 | __pagevec_lru_add(&pvec); | |
2232 | BUG_ON(!list_empty(pages)); | |
2233 | if (bio) | |
2234 | submit_one_bio(READ, bio); | |
2235 | return 0; | |
2236 | } | |
2237 | EXPORT_SYMBOL(extent_readpages); | |
2238 | ||
2239 | /* | |
2240 | * basic invalidatepage code, this waits on any locked or writeback | |
2241 | * ranges corresponding to the page, and then deletes any extent state | |
2242 | * records from the tree | |
2243 | */ | |
2244 | int extent_invalidatepage(struct extent_io_tree *tree, | |
2245 | struct page *page, unsigned long offset) | |
2246 | { | |
2247 | u64 start = ((u64)page->index << PAGE_CACHE_SHIFT); | |
2248 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
2249 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
2250 | ||
2251 | start += (offset + blocksize -1) & ~(blocksize - 1); | |
2252 | if (start > end) | |
2253 | return 0; | |
2254 | ||
2255 | lock_extent(tree, start, end, GFP_NOFS); | |
2256 | wait_on_extent_writeback(tree, start, end); | |
2257 | clear_extent_bit(tree, start, end, | |
2258 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC, | |
2259 | 1, 1, GFP_NOFS); | |
2260 | return 0; | |
2261 | } | |
2262 | EXPORT_SYMBOL(extent_invalidatepage); | |
2263 | ||
2264 | /* | |
2265 | * simple commit_write call, set_range_dirty is used to mark both | |
2266 | * the pages and the extent records as dirty | |
2267 | */ | |
2268 | int extent_commit_write(struct extent_io_tree *tree, | |
2269 | struct inode *inode, struct page *page, | |
2270 | unsigned from, unsigned to) | |
2271 | { | |
2272 | loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to; | |
2273 | ||
2274 | set_page_extent_mapped(page); | |
2275 | set_page_dirty(page); | |
2276 | ||
2277 | if (pos > inode->i_size) { | |
2278 | i_size_write(inode, pos); | |
2279 | mark_inode_dirty(inode); | |
2280 | } | |
2281 | return 0; | |
2282 | } | |
2283 | EXPORT_SYMBOL(extent_commit_write); | |
2284 | ||
2285 | int extent_prepare_write(struct extent_io_tree *tree, | |
2286 | struct inode *inode, struct page *page, | |
2287 | unsigned from, unsigned to, get_extent_t *get_extent) | |
2288 | { | |
2289 | u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2290 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
2291 | u64 block_start; | |
2292 | u64 orig_block_start; | |
2293 | u64 block_end; | |
2294 | u64 cur_end; | |
2295 | struct extent_map *em; | |
2296 | unsigned blocksize = 1 << inode->i_blkbits; | |
2297 | size_t page_offset = 0; | |
2298 | size_t block_off_start; | |
2299 | size_t block_off_end; | |
2300 | int err = 0; | |
2301 | int iocount = 0; | |
2302 | int ret = 0; | |
2303 | int isnew; | |
2304 | ||
2305 | set_page_extent_mapped(page); | |
2306 | ||
2307 | block_start = (page_start + from) & ~((u64)blocksize - 1); | |
2308 | block_end = (page_start + to - 1) | (blocksize - 1); | |
2309 | orig_block_start = block_start; | |
2310 | ||
2311 | lock_extent(tree, page_start, page_end, GFP_NOFS); | |
2312 | while(block_start <= block_end) { | |
2313 | em = get_extent(inode, page, page_offset, block_start, | |
2314 | block_end - block_start + 1, 1); | |
2315 | if (IS_ERR(em) || !em) { | |
2316 | goto err; | |
2317 | } | |
2318 | cur_end = min(block_end, extent_map_end(em) - 1); | |
2319 | block_off_start = block_start & (PAGE_CACHE_SIZE - 1); | |
2320 | block_off_end = block_off_start + blocksize; | |
2321 | isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS); | |
2322 | ||
2323 | if (!PageUptodate(page) && isnew && | |
2324 | (block_off_end > to || block_off_start < from)) { | |
2325 | void *kaddr; | |
2326 | ||
2327 | kaddr = kmap_atomic(page, KM_USER0); | |
2328 | if (block_off_end > to) | |
2329 | memset(kaddr + to, 0, block_off_end - to); | |
2330 | if (block_off_start < from) | |
2331 | memset(kaddr + block_off_start, 0, | |
2332 | from - block_off_start); | |
2333 | flush_dcache_page(page); | |
2334 | kunmap_atomic(kaddr, KM_USER0); | |
2335 | } | |
2336 | if ((em->block_start != EXTENT_MAP_HOLE && | |
2337 | em->block_start != EXTENT_MAP_INLINE) && | |
2338 | !isnew && !PageUptodate(page) && | |
2339 | (block_off_end > to || block_off_start < from) && | |
2340 | !test_range_bit(tree, block_start, cur_end, | |
2341 | EXTENT_UPTODATE, 1)) { | |
2342 | u64 sector; | |
2343 | u64 extent_offset = block_start - em->start; | |
2344 | size_t iosize; | |
2345 | sector = (em->block_start + extent_offset) >> 9; | |
2346 | iosize = (cur_end - block_start + blocksize) & | |
2347 | ~((u64)blocksize - 1); | |
2348 | /* | |
2349 | * we've already got the extent locked, but we | |
2350 | * need to split the state such that our end_bio | |
2351 | * handler can clear the lock. | |
2352 | */ | |
2353 | set_extent_bit(tree, block_start, | |
2354 | block_start + iosize - 1, | |
2355 | EXTENT_LOCKED, 0, NULL, GFP_NOFS); | |
2356 | ret = submit_extent_page(READ, tree, page, | |
2357 | sector, iosize, page_offset, em->bdev, | |
2358 | NULL, 1, | |
2359 | end_bio_extent_preparewrite); | |
2360 | iocount++; | |
2361 | block_start = block_start + iosize; | |
2362 | } else { | |
2363 | set_extent_uptodate(tree, block_start, cur_end, | |
2364 | GFP_NOFS); | |
2365 | unlock_extent(tree, block_start, cur_end, GFP_NOFS); | |
2366 | block_start = cur_end + 1; | |
2367 | } | |
2368 | page_offset = block_start & (PAGE_CACHE_SIZE - 1); | |
2369 | free_extent_map(em); | |
2370 | } | |
2371 | if (iocount) { | |
2372 | wait_extent_bit(tree, orig_block_start, | |
2373 | block_end, EXTENT_LOCKED); | |
2374 | } | |
2375 | check_page_uptodate(tree, page); | |
2376 | err: | |
2377 | /* FIXME, zero out newly allocated blocks on error */ | |
2378 | return err; | |
2379 | } | |
2380 | EXPORT_SYMBOL(extent_prepare_write); | |
2381 | ||
2382 | /* | |
2383 | * a helper for releasepage. As long as there are no locked extents | |
2384 | * in the range corresponding to the page, both state records and extent | |
2385 | * map records are removed | |
2386 | */ | |
2387 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
2388 | struct extent_io_tree *tree, struct page *page, |
2389 | gfp_t mask) | |
d1310b2e CM |
2390 | { |
2391 | struct extent_map *em; | |
2392 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2393 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
2394 | u64 orig_start = start; | |
2395 | int ret = 1; | |
2396 | ||
70dec807 CM |
2397 | if ((mask & __GFP_WAIT) && |
2398 | page->mapping->host->i_size > 16 * 1024 * 1024) { | |
2399 | while (start <= end) { | |
2400 | spin_lock(&map->lock); | |
2401 | em = lookup_extent_mapping(map, start, end); | |
2402 | if (!em || IS_ERR(em)) { | |
2403 | spin_unlock(&map->lock); | |
2404 | break; | |
2405 | } | |
2406 | if (em->start != start) { | |
2407 | spin_unlock(&map->lock); | |
2408 | free_extent_map(em); | |
2409 | break; | |
2410 | } | |
2411 | if (!test_range_bit(tree, em->start, | |
2412 | extent_map_end(em) - 1, | |
2413 | EXTENT_LOCKED, 0)) { | |
2414 | remove_extent_mapping(map, em); | |
2415 | /* once for the rb tree */ | |
2416 | free_extent_map(em); | |
2417 | } | |
2418 | start = extent_map_end(em); | |
d1310b2e | 2419 | spin_unlock(&map->lock); |
70dec807 CM |
2420 | |
2421 | /* once for us */ | |
d1310b2e CM |
2422 | free_extent_map(em); |
2423 | } | |
d1310b2e | 2424 | } |
70dec807 | 2425 | if (test_range_bit(tree, orig_start, end, EXTENT_IOBITS, 0)) |
d1310b2e | 2426 | ret = 0; |
70dec807 CM |
2427 | else { |
2428 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
2429 | mask = GFP_NOFS; | |
d1310b2e | 2430 | clear_extent_bit(tree, orig_start, end, EXTENT_UPTODATE, |
70dec807 CM |
2431 | 1, 1, mask); |
2432 | } | |
d1310b2e CM |
2433 | return ret; |
2434 | } | |
2435 | EXPORT_SYMBOL(try_release_extent_mapping); | |
2436 | ||
2437 | sector_t extent_bmap(struct address_space *mapping, sector_t iblock, | |
2438 | get_extent_t *get_extent) | |
2439 | { | |
2440 | struct inode *inode = mapping->host; | |
2441 | u64 start = iblock << inode->i_blkbits; | |
2442 | sector_t sector = 0; | |
2443 | struct extent_map *em; | |
2444 | ||
2445 | em = get_extent(inode, NULL, 0, start, (1 << inode->i_blkbits), 0); | |
2446 | if (!em || IS_ERR(em)) | |
2447 | return 0; | |
2448 | ||
2449 | if (em->block_start == EXTENT_MAP_INLINE || | |
2450 | em->block_start == EXTENT_MAP_HOLE) | |
2451 | goto out; | |
2452 | ||
2453 | sector = (em->block_start + start - em->start) >> inode->i_blkbits; | |
d1310b2e CM |
2454 | out: |
2455 | free_extent_map(em); | |
2456 | return sector; | |
2457 | } | |
2458 | ||
2459 | static int add_lru(struct extent_io_tree *tree, struct extent_buffer *eb) | |
2460 | { | |
2461 | if (list_empty(&eb->lru)) { | |
2462 | extent_buffer_get(eb); | |
2463 | list_add(&eb->lru, &tree->buffer_lru); | |
2464 | tree->lru_size++; | |
2465 | if (tree->lru_size >= BUFFER_LRU_MAX) { | |
2466 | struct extent_buffer *rm; | |
2467 | rm = list_entry(tree->buffer_lru.prev, | |
2468 | struct extent_buffer, lru); | |
2469 | tree->lru_size--; | |
2470 | list_del_init(&rm->lru); | |
2471 | free_extent_buffer(rm); | |
2472 | } | |
2473 | } else | |
2474 | list_move(&eb->lru, &tree->buffer_lru); | |
2475 | return 0; | |
2476 | } | |
2477 | static struct extent_buffer *find_lru(struct extent_io_tree *tree, | |
2478 | u64 start, unsigned long len) | |
2479 | { | |
2480 | struct list_head *lru = &tree->buffer_lru; | |
2481 | struct list_head *cur = lru->next; | |
2482 | struct extent_buffer *eb; | |
2483 | ||
2484 | if (list_empty(lru)) | |
2485 | return NULL; | |
2486 | ||
2487 | do { | |
2488 | eb = list_entry(cur, struct extent_buffer, lru); | |
2489 | if (eb->start == start && eb->len == len) { | |
2490 | extent_buffer_get(eb); | |
2491 | return eb; | |
2492 | } | |
2493 | cur = cur->next; | |
2494 | } while (cur != lru); | |
2495 | return NULL; | |
2496 | } | |
2497 | ||
2498 | static inline unsigned long num_extent_pages(u64 start, u64 len) | |
2499 | { | |
2500 | return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) - | |
2501 | (start >> PAGE_CACHE_SHIFT); | |
2502 | } | |
2503 | ||
2504 | static inline struct page *extent_buffer_page(struct extent_buffer *eb, | |
2505 | unsigned long i) | |
2506 | { | |
2507 | struct page *p; | |
2508 | struct address_space *mapping; | |
2509 | ||
2510 | if (i == 0) | |
2511 | return eb->first_page; | |
2512 | i += eb->start >> PAGE_CACHE_SHIFT; | |
2513 | mapping = eb->first_page->mapping; | |
2514 | read_lock_irq(&mapping->tree_lock); | |
2515 | p = radix_tree_lookup(&mapping->page_tree, i); | |
2516 | read_unlock_irq(&mapping->tree_lock); | |
2517 | return p; | |
2518 | } | |
2519 | ||
2520 | static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree, | |
2521 | u64 start, | |
2522 | unsigned long len, | |
2523 | gfp_t mask) | |
2524 | { | |
2525 | struct extent_buffer *eb = NULL; | |
2526 | ||
2527 | spin_lock(&tree->lru_lock); | |
2528 | eb = find_lru(tree, start, len); | |
2529 | spin_unlock(&tree->lru_lock); | |
2530 | if (eb) { | |
2531 | return eb; | |
2532 | } | |
2533 | ||
2534 | eb = kmem_cache_zalloc(extent_buffer_cache, mask); | |
2535 | INIT_LIST_HEAD(&eb->lru); | |
2536 | eb->start = start; | |
2537 | eb->len = len; | |
2538 | atomic_set(&eb->refs, 1); | |
2539 | ||
2540 | return eb; | |
2541 | } | |
2542 | ||
2543 | static void __free_extent_buffer(struct extent_buffer *eb) | |
2544 | { | |
2545 | kmem_cache_free(extent_buffer_cache, eb); | |
2546 | } | |
2547 | ||
2548 | struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree, | |
2549 | u64 start, unsigned long len, | |
2550 | struct page *page0, | |
2551 | gfp_t mask) | |
2552 | { | |
2553 | unsigned long num_pages = num_extent_pages(start, len); | |
2554 | unsigned long i; | |
2555 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
2556 | struct extent_buffer *eb; | |
2557 | struct page *p; | |
2558 | struct address_space *mapping = tree->mapping; | |
2559 | int uptodate = 1; | |
2560 | ||
2561 | eb = __alloc_extent_buffer(tree, start, len, mask); | |
2562 | if (!eb || IS_ERR(eb)) | |
2563 | return NULL; | |
2564 | ||
2565 | if (eb->flags & EXTENT_BUFFER_FILLED) | |
2566 | goto lru_add; | |
2567 | ||
2568 | if (page0) { | |
2569 | eb->first_page = page0; | |
2570 | i = 1; | |
2571 | index++; | |
2572 | page_cache_get(page0); | |
2573 | mark_page_accessed(page0); | |
2574 | set_page_extent_mapped(page0); | |
2575 | WARN_ON(!PageUptodate(page0)); | |
2576 | set_page_extent_head(page0, len); | |
2577 | } else { | |
2578 | i = 0; | |
2579 | } | |
2580 | for (; i < num_pages; i++, index++) { | |
2581 | p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM); | |
2582 | if (!p) { | |
2583 | WARN_ON(1); | |
2584 | goto fail; | |
2585 | } | |
2586 | set_page_extent_mapped(p); | |
2587 | mark_page_accessed(p); | |
2588 | if (i == 0) { | |
2589 | eb->first_page = p; | |
2590 | set_page_extent_head(p, len); | |
2591 | } else { | |
2592 | set_page_private(p, EXTENT_PAGE_PRIVATE); | |
2593 | } | |
2594 | if (!PageUptodate(p)) | |
2595 | uptodate = 0; | |
2596 | unlock_page(p); | |
2597 | } | |
2598 | if (uptodate) | |
2599 | eb->flags |= EXTENT_UPTODATE; | |
2600 | eb->flags |= EXTENT_BUFFER_FILLED; | |
2601 | ||
2602 | lru_add: | |
2603 | spin_lock(&tree->lru_lock); | |
2604 | add_lru(tree, eb); | |
2605 | spin_unlock(&tree->lru_lock); | |
2606 | return eb; | |
2607 | ||
2608 | fail: | |
2609 | spin_lock(&tree->lru_lock); | |
2610 | list_del_init(&eb->lru); | |
2611 | spin_unlock(&tree->lru_lock); | |
2612 | if (!atomic_dec_and_test(&eb->refs)) | |
2613 | return NULL; | |
2614 | for (index = 1; index < i; index++) { | |
2615 | page_cache_release(extent_buffer_page(eb, index)); | |
2616 | } | |
2617 | if (i > 0) | |
2618 | page_cache_release(extent_buffer_page(eb, 0)); | |
2619 | __free_extent_buffer(eb); | |
2620 | return NULL; | |
2621 | } | |
2622 | EXPORT_SYMBOL(alloc_extent_buffer); | |
2623 | ||
2624 | struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree, | |
2625 | u64 start, unsigned long len, | |
2626 | gfp_t mask) | |
2627 | { | |
2628 | unsigned long num_pages = num_extent_pages(start, len); | |
2629 | unsigned long i; | |
2630 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
2631 | struct extent_buffer *eb; | |
2632 | struct page *p; | |
2633 | struct address_space *mapping = tree->mapping; | |
2634 | int uptodate = 1; | |
2635 | ||
2636 | eb = __alloc_extent_buffer(tree, start, len, mask); | |
2637 | if (!eb || IS_ERR(eb)) | |
2638 | return NULL; | |
2639 | ||
2640 | if (eb->flags & EXTENT_BUFFER_FILLED) | |
2641 | goto lru_add; | |
2642 | ||
2643 | for (i = 0; i < num_pages; i++, index++) { | |
2644 | p = find_lock_page(mapping, index); | |
2645 | if (!p) { | |
2646 | goto fail; | |
2647 | } | |
2648 | set_page_extent_mapped(p); | |
2649 | mark_page_accessed(p); | |
2650 | ||
2651 | if (i == 0) { | |
2652 | eb->first_page = p; | |
2653 | set_page_extent_head(p, len); | |
2654 | } else { | |
2655 | set_page_private(p, EXTENT_PAGE_PRIVATE); | |
2656 | } | |
2657 | ||
2658 | if (!PageUptodate(p)) | |
2659 | uptodate = 0; | |
2660 | unlock_page(p); | |
2661 | } | |
2662 | if (uptodate) | |
2663 | eb->flags |= EXTENT_UPTODATE; | |
2664 | eb->flags |= EXTENT_BUFFER_FILLED; | |
2665 | ||
2666 | lru_add: | |
2667 | spin_lock(&tree->lru_lock); | |
2668 | add_lru(tree, eb); | |
2669 | spin_unlock(&tree->lru_lock); | |
2670 | return eb; | |
2671 | fail: | |
2672 | spin_lock(&tree->lru_lock); | |
2673 | list_del_init(&eb->lru); | |
2674 | spin_unlock(&tree->lru_lock); | |
2675 | if (!atomic_dec_and_test(&eb->refs)) | |
2676 | return NULL; | |
2677 | for (index = 1; index < i; index++) { | |
2678 | page_cache_release(extent_buffer_page(eb, index)); | |
2679 | } | |
2680 | if (i > 0) | |
2681 | page_cache_release(extent_buffer_page(eb, 0)); | |
2682 | __free_extent_buffer(eb); | |
2683 | return NULL; | |
2684 | } | |
2685 | EXPORT_SYMBOL(find_extent_buffer); | |
2686 | ||
2687 | void free_extent_buffer(struct extent_buffer *eb) | |
2688 | { | |
2689 | unsigned long i; | |
2690 | unsigned long num_pages; | |
2691 | ||
2692 | if (!eb) | |
2693 | return; | |
2694 | ||
2695 | if (!atomic_dec_and_test(&eb->refs)) | |
2696 | return; | |
2697 | ||
2698 | WARN_ON(!list_empty(&eb->lru)); | |
2699 | num_pages = num_extent_pages(eb->start, eb->len); | |
2700 | ||
2701 | for (i = 1; i < num_pages; i++) { | |
2702 | page_cache_release(extent_buffer_page(eb, i)); | |
2703 | } | |
2704 | page_cache_release(extent_buffer_page(eb, 0)); | |
2705 | __free_extent_buffer(eb); | |
2706 | } | |
2707 | EXPORT_SYMBOL(free_extent_buffer); | |
2708 | ||
2709 | int clear_extent_buffer_dirty(struct extent_io_tree *tree, | |
2710 | struct extent_buffer *eb) | |
2711 | { | |
2712 | int set; | |
2713 | unsigned long i; | |
2714 | unsigned long num_pages; | |
2715 | struct page *page; | |
2716 | ||
2717 | u64 start = eb->start; | |
2718 | u64 end = start + eb->len - 1; | |
2719 | ||
2720 | set = clear_extent_dirty(tree, start, end, GFP_NOFS); | |
2721 | num_pages = num_extent_pages(eb->start, eb->len); | |
2722 | ||
2723 | for (i = 0; i < num_pages; i++) { | |
2724 | page = extent_buffer_page(eb, i); | |
2725 | lock_page(page); | |
2726 | if (i == 0) | |
2727 | set_page_extent_head(page, eb->len); | |
2728 | else | |
2729 | set_page_private(page, EXTENT_PAGE_PRIVATE); | |
2730 | ||
2731 | /* | |
2732 | * if we're on the last page or the first page and the | |
2733 | * block isn't aligned on a page boundary, do extra checks | |
2734 | * to make sure we don't clean page that is partially dirty | |
2735 | */ | |
2736 | if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) || | |
2737 | ((i == num_pages - 1) && | |
2738 | ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) { | |
2739 | start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2740 | end = start + PAGE_CACHE_SIZE - 1; | |
2741 | if (test_range_bit(tree, start, end, | |
2742 | EXTENT_DIRTY, 0)) { | |
2743 | unlock_page(page); | |
2744 | continue; | |
2745 | } | |
2746 | } | |
2747 | clear_page_dirty_for_io(page); | |
70dec807 | 2748 | read_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
2749 | if (!PageDirty(page)) { |
2750 | radix_tree_tag_clear(&page->mapping->page_tree, | |
2751 | page_index(page), | |
2752 | PAGECACHE_TAG_DIRTY); | |
2753 | } | |
70dec807 | 2754 | read_unlock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
2755 | unlock_page(page); |
2756 | } | |
2757 | return 0; | |
2758 | } | |
2759 | EXPORT_SYMBOL(clear_extent_buffer_dirty); | |
2760 | ||
2761 | int wait_on_extent_buffer_writeback(struct extent_io_tree *tree, | |
2762 | struct extent_buffer *eb) | |
2763 | { | |
2764 | return wait_on_extent_writeback(tree, eb->start, | |
2765 | eb->start + eb->len - 1); | |
2766 | } | |
2767 | EXPORT_SYMBOL(wait_on_extent_buffer_writeback); | |
2768 | ||
2769 | int set_extent_buffer_dirty(struct extent_io_tree *tree, | |
2770 | struct extent_buffer *eb) | |
2771 | { | |
2772 | unsigned long i; | |
2773 | unsigned long num_pages; | |
2774 | ||
2775 | num_pages = num_extent_pages(eb->start, eb->len); | |
2776 | for (i = 0; i < num_pages; i++) { | |
2777 | struct page *page = extent_buffer_page(eb, i); | |
2778 | /* writepage may need to do something special for the | |
2779 | * first page, we have to make sure page->private is | |
2780 | * properly set. releasepage may drop page->private | |
2781 | * on us if the page isn't already dirty. | |
2782 | */ | |
2783 | if (i == 0) { | |
2784 | lock_page(page); | |
2785 | set_page_extent_head(page, eb->len); | |
2786 | } else if (PagePrivate(page) && | |
2787 | page->private != EXTENT_PAGE_PRIVATE) { | |
2788 | lock_page(page); | |
2789 | set_page_extent_mapped(page); | |
2790 | unlock_page(page); | |
2791 | } | |
2792 | __set_page_dirty_nobuffers(extent_buffer_page(eb, i)); | |
2793 | if (i == 0) | |
2794 | unlock_page(page); | |
2795 | } | |
2796 | return set_extent_dirty(tree, eb->start, | |
2797 | eb->start + eb->len - 1, GFP_NOFS); | |
2798 | } | |
2799 | EXPORT_SYMBOL(set_extent_buffer_dirty); | |
2800 | ||
2801 | int set_extent_buffer_uptodate(struct extent_io_tree *tree, | |
2802 | struct extent_buffer *eb) | |
2803 | { | |
2804 | unsigned long i; | |
2805 | struct page *page; | |
2806 | unsigned long num_pages; | |
2807 | ||
2808 | num_pages = num_extent_pages(eb->start, eb->len); | |
2809 | ||
2810 | set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1, | |
2811 | GFP_NOFS); | |
2812 | for (i = 0; i < num_pages; i++) { | |
2813 | page = extent_buffer_page(eb, i); | |
2814 | if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) || | |
2815 | ((i == num_pages - 1) && | |
2816 | ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) { | |
2817 | check_page_uptodate(tree, page); | |
2818 | continue; | |
2819 | } | |
2820 | SetPageUptodate(page); | |
2821 | } | |
2822 | return 0; | |
2823 | } | |
2824 | EXPORT_SYMBOL(set_extent_buffer_uptodate); | |
2825 | ||
2826 | int extent_buffer_uptodate(struct extent_io_tree *tree, | |
2827 | struct extent_buffer *eb) | |
2828 | { | |
2829 | if (eb->flags & EXTENT_UPTODATE) | |
2830 | return 1; | |
2831 | return test_range_bit(tree, eb->start, eb->start + eb->len - 1, | |
2832 | EXTENT_UPTODATE, 1); | |
2833 | } | |
2834 | EXPORT_SYMBOL(extent_buffer_uptodate); | |
2835 | ||
2836 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
2837 | struct extent_buffer *eb, | |
2838 | u64 start, | |
2839 | int wait) | |
2840 | { | |
2841 | unsigned long i; | |
2842 | unsigned long start_i; | |
2843 | struct page *page; | |
2844 | int err; | |
2845 | int ret = 0; | |
2846 | unsigned long num_pages; | |
2847 | ||
2848 | if (eb->flags & EXTENT_UPTODATE) | |
2849 | return 0; | |
2850 | ||
2851 | if (0 && test_range_bit(tree, eb->start, eb->start + eb->len - 1, | |
2852 | EXTENT_UPTODATE, 1)) { | |
2853 | return 0; | |
2854 | } | |
2855 | ||
2856 | if (start) { | |
2857 | WARN_ON(start < eb->start); | |
2858 | start_i = (start >> PAGE_CACHE_SHIFT) - | |
2859 | (eb->start >> PAGE_CACHE_SHIFT); | |
2860 | } else { | |
2861 | start_i = 0; | |
2862 | } | |
2863 | ||
2864 | num_pages = num_extent_pages(eb->start, eb->len); | |
2865 | for (i = start_i; i < num_pages; i++) { | |
2866 | page = extent_buffer_page(eb, i); | |
2867 | if (PageUptodate(page)) { | |
2868 | continue; | |
2869 | } | |
2870 | if (!wait) { | |
2871 | if (TestSetPageLocked(page)) { | |
2872 | continue; | |
2873 | } | |
2874 | } else { | |
2875 | lock_page(page); | |
2876 | } | |
2877 | if (!PageUptodate(page)) { | |
2878 | err = page->mapping->a_ops->readpage(NULL, page); | |
2879 | if (err) { | |
2880 | ret = err; | |
2881 | } | |
2882 | } else { | |
2883 | unlock_page(page); | |
2884 | } | |
2885 | } | |
2886 | ||
2887 | if (ret || !wait) { | |
2888 | return ret; | |
2889 | } | |
d1310b2e CM |
2890 | for (i = start_i; i < num_pages; i++) { |
2891 | page = extent_buffer_page(eb, i); | |
2892 | wait_on_page_locked(page); | |
2893 | if (!PageUptodate(page)) { | |
2894 | ret = -EIO; | |
2895 | } | |
2896 | } | |
2897 | if (!ret) | |
2898 | eb->flags |= EXTENT_UPTODATE; | |
2899 | return ret; | |
2900 | } | |
2901 | EXPORT_SYMBOL(read_extent_buffer_pages); | |
2902 | ||
2903 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
2904 | unsigned long start, | |
2905 | unsigned long len) | |
2906 | { | |
2907 | size_t cur; | |
2908 | size_t offset; | |
2909 | struct page *page; | |
2910 | char *kaddr; | |
2911 | char *dst = (char *)dstv; | |
2912 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
2913 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
2914 | unsigned long num_pages = num_extent_pages(eb->start, eb->len); | |
2915 | ||
2916 | WARN_ON(start > eb->len); | |
2917 | WARN_ON(start + len > eb->start + eb->len); | |
2918 | ||
2919 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
2920 | ||
2921 | while(len > 0) { | |
2922 | page = extent_buffer_page(eb, i); | |
2923 | if (!PageUptodate(page)) { | |
2924 | printk("page %lu not up to date i %lu, total %lu, len %lu\n", page->index, i, num_pages, eb->len); | |
2925 | WARN_ON(1); | |
2926 | } | |
2927 | WARN_ON(!PageUptodate(page)); | |
2928 | ||
2929 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
2930 | kaddr = kmap_atomic(page, KM_USER1); | |
2931 | memcpy(dst, kaddr + offset, cur); | |
2932 | kunmap_atomic(kaddr, KM_USER1); | |
2933 | ||
2934 | dst += cur; | |
2935 | len -= cur; | |
2936 | offset = 0; | |
2937 | i++; | |
2938 | } | |
2939 | } | |
2940 | EXPORT_SYMBOL(read_extent_buffer); | |
2941 | ||
2942 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
2943 | unsigned long min_len, char **token, char **map, | |
2944 | unsigned long *map_start, | |
2945 | unsigned long *map_len, int km) | |
2946 | { | |
2947 | size_t offset = start & (PAGE_CACHE_SIZE - 1); | |
2948 | char *kaddr; | |
2949 | struct page *p; | |
2950 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
2951 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
2952 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
2953 | PAGE_CACHE_SHIFT; | |
2954 | ||
2955 | if (i != end_i) | |
2956 | return -EINVAL; | |
2957 | ||
2958 | if (i == 0) { | |
2959 | offset = start_offset; | |
2960 | *map_start = 0; | |
2961 | } else { | |
2962 | offset = 0; | |
2963 | *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; | |
2964 | } | |
2965 | if (start + min_len > eb->len) { | |
2966 | printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len, start, min_len); | |
2967 | WARN_ON(1); | |
2968 | } | |
2969 | ||
2970 | p = extent_buffer_page(eb, i); | |
2971 | WARN_ON(!PageUptodate(p)); | |
2972 | kaddr = kmap_atomic(p, km); | |
2973 | *token = kaddr; | |
2974 | *map = kaddr + offset; | |
2975 | *map_len = PAGE_CACHE_SIZE - offset; | |
2976 | return 0; | |
2977 | } | |
2978 | EXPORT_SYMBOL(map_private_extent_buffer); | |
2979 | ||
2980 | int map_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
2981 | unsigned long min_len, | |
2982 | char **token, char **map, | |
2983 | unsigned long *map_start, | |
2984 | unsigned long *map_len, int km) | |
2985 | { | |
2986 | int err; | |
2987 | int save = 0; | |
2988 | if (eb->map_token) { | |
2989 | unmap_extent_buffer(eb, eb->map_token, km); | |
2990 | eb->map_token = NULL; | |
2991 | save = 1; | |
2992 | } | |
2993 | err = map_private_extent_buffer(eb, start, min_len, token, map, | |
2994 | map_start, map_len, km); | |
2995 | if (!err && save) { | |
2996 | eb->map_token = *token; | |
2997 | eb->kaddr = *map; | |
2998 | eb->map_start = *map_start; | |
2999 | eb->map_len = *map_len; | |
3000 | } | |
3001 | return err; | |
3002 | } | |
3003 | EXPORT_SYMBOL(map_extent_buffer); | |
3004 | ||
3005 | void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km) | |
3006 | { | |
3007 | kunmap_atomic(token, km); | |
3008 | } | |
3009 | EXPORT_SYMBOL(unmap_extent_buffer); | |
3010 | ||
3011 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, | |
3012 | unsigned long start, | |
3013 | unsigned long len) | |
3014 | { | |
3015 | size_t cur; | |
3016 | size_t offset; | |
3017 | struct page *page; | |
3018 | char *kaddr; | |
3019 | char *ptr = (char *)ptrv; | |
3020 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3021 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3022 | int ret = 0; | |
3023 | ||
3024 | WARN_ON(start > eb->len); | |
3025 | WARN_ON(start + len > eb->start + eb->len); | |
3026 | ||
3027 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3028 | ||
3029 | while(len > 0) { | |
3030 | page = extent_buffer_page(eb, i); | |
3031 | WARN_ON(!PageUptodate(page)); | |
3032 | ||
3033 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
3034 | ||
3035 | kaddr = kmap_atomic(page, KM_USER0); | |
3036 | ret = memcmp(ptr, kaddr + offset, cur); | |
3037 | kunmap_atomic(kaddr, KM_USER0); | |
3038 | if (ret) | |
3039 | break; | |
3040 | ||
3041 | ptr += cur; | |
3042 | len -= cur; | |
3043 | offset = 0; | |
3044 | i++; | |
3045 | } | |
3046 | return ret; | |
3047 | } | |
3048 | EXPORT_SYMBOL(memcmp_extent_buffer); | |
3049 | ||
3050 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
3051 | unsigned long start, unsigned long len) | |
3052 | { | |
3053 | size_t cur; | |
3054 | size_t offset; | |
3055 | struct page *page; | |
3056 | char *kaddr; | |
3057 | char *src = (char *)srcv; | |
3058 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3059 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3060 | ||
3061 | WARN_ON(start > eb->len); | |
3062 | WARN_ON(start + len > eb->start + eb->len); | |
3063 | ||
3064 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3065 | ||
3066 | while(len > 0) { | |
3067 | page = extent_buffer_page(eb, i); | |
3068 | WARN_ON(!PageUptodate(page)); | |
3069 | ||
3070 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
3071 | kaddr = kmap_atomic(page, KM_USER1); | |
3072 | memcpy(kaddr + offset, src, cur); | |
3073 | kunmap_atomic(kaddr, KM_USER1); | |
3074 | ||
3075 | src += cur; | |
3076 | len -= cur; | |
3077 | offset = 0; | |
3078 | i++; | |
3079 | } | |
3080 | } | |
3081 | EXPORT_SYMBOL(write_extent_buffer); | |
3082 | ||
3083 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
3084 | unsigned long start, unsigned long len) | |
3085 | { | |
3086 | size_t cur; | |
3087 | size_t offset; | |
3088 | struct page *page; | |
3089 | char *kaddr; | |
3090 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3091 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3092 | ||
3093 | WARN_ON(start > eb->len); | |
3094 | WARN_ON(start + len > eb->start + eb->len); | |
3095 | ||
3096 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3097 | ||
3098 | while(len > 0) { | |
3099 | page = extent_buffer_page(eb, i); | |
3100 | WARN_ON(!PageUptodate(page)); | |
3101 | ||
3102 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
3103 | kaddr = kmap_atomic(page, KM_USER0); | |
3104 | memset(kaddr + offset, c, cur); | |
3105 | kunmap_atomic(kaddr, KM_USER0); | |
3106 | ||
3107 | len -= cur; | |
3108 | offset = 0; | |
3109 | i++; | |
3110 | } | |
3111 | } | |
3112 | EXPORT_SYMBOL(memset_extent_buffer); | |
3113 | ||
3114 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
3115 | unsigned long dst_offset, unsigned long src_offset, | |
3116 | unsigned long len) | |
3117 | { | |
3118 | u64 dst_len = dst->len; | |
3119 | size_t cur; | |
3120 | size_t offset; | |
3121 | struct page *page; | |
3122 | char *kaddr; | |
3123 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3124 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
3125 | ||
3126 | WARN_ON(src->len != dst_len); | |
3127 | ||
3128 | offset = (start_offset + dst_offset) & | |
3129 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3130 | ||
3131 | while(len > 0) { | |
3132 | page = extent_buffer_page(dst, i); | |
3133 | WARN_ON(!PageUptodate(page)); | |
3134 | ||
3135 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
3136 | ||
3137 | kaddr = kmap_atomic(page, KM_USER0); | |
3138 | read_extent_buffer(src, kaddr + offset, src_offset, cur); | |
3139 | kunmap_atomic(kaddr, KM_USER0); | |
3140 | ||
3141 | src_offset += cur; | |
3142 | len -= cur; | |
3143 | offset = 0; | |
3144 | i++; | |
3145 | } | |
3146 | } | |
3147 | EXPORT_SYMBOL(copy_extent_buffer); | |
3148 | ||
3149 | static void move_pages(struct page *dst_page, struct page *src_page, | |
3150 | unsigned long dst_off, unsigned long src_off, | |
3151 | unsigned long len) | |
3152 | { | |
3153 | char *dst_kaddr = kmap_atomic(dst_page, KM_USER0); | |
3154 | if (dst_page == src_page) { | |
3155 | memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len); | |
3156 | } else { | |
3157 | char *src_kaddr = kmap_atomic(src_page, KM_USER1); | |
3158 | char *p = dst_kaddr + dst_off + len; | |
3159 | char *s = src_kaddr + src_off + len; | |
3160 | ||
3161 | while (len--) | |
3162 | *--p = *--s; | |
3163 | ||
3164 | kunmap_atomic(src_kaddr, KM_USER1); | |
3165 | } | |
3166 | kunmap_atomic(dst_kaddr, KM_USER0); | |
3167 | } | |
3168 | ||
3169 | static void copy_pages(struct page *dst_page, struct page *src_page, | |
3170 | unsigned long dst_off, unsigned long src_off, | |
3171 | unsigned long len) | |
3172 | { | |
3173 | char *dst_kaddr = kmap_atomic(dst_page, KM_USER0); | |
3174 | char *src_kaddr; | |
3175 | ||
3176 | if (dst_page != src_page) | |
3177 | src_kaddr = kmap_atomic(src_page, KM_USER1); | |
3178 | else | |
3179 | src_kaddr = dst_kaddr; | |
3180 | ||
3181 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
3182 | kunmap_atomic(dst_kaddr, KM_USER0); | |
3183 | if (dst_page != src_page) | |
3184 | kunmap_atomic(src_kaddr, KM_USER1); | |
3185 | } | |
3186 | ||
3187 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
3188 | unsigned long src_offset, unsigned long len) | |
3189 | { | |
3190 | size_t cur; | |
3191 | size_t dst_off_in_page; | |
3192 | size_t src_off_in_page; | |
3193 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3194 | unsigned long dst_i; | |
3195 | unsigned long src_i; | |
3196 | ||
3197 | if (src_offset + len > dst->len) { | |
3198 | printk("memmove bogus src_offset %lu move len %lu len %lu\n", | |
3199 | src_offset, len, dst->len); | |
3200 | BUG_ON(1); | |
3201 | } | |
3202 | if (dst_offset + len > dst->len) { | |
3203 | printk("memmove bogus dst_offset %lu move len %lu len %lu\n", | |
3204 | dst_offset, len, dst->len); | |
3205 | BUG_ON(1); | |
3206 | } | |
3207 | ||
3208 | while(len > 0) { | |
3209 | dst_off_in_page = (start_offset + dst_offset) & | |
3210 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3211 | src_off_in_page = (start_offset + src_offset) & | |
3212 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3213 | ||
3214 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
3215 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
3216 | ||
3217 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
3218 | src_off_in_page)); | |
3219 | cur = min_t(unsigned long, cur, | |
3220 | (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); | |
3221 | ||
3222 | copy_pages(extent_buffer_page(dst, dst_i), | |
3223 | extent_buffer_page(dst, src_i), | |
3224 | dst_off_in_page, src_off_in_page, cur); | |
3225 | ||
3226 | src_offset += cur; | |
3227 | dst_offset += cur; | |
3228 | len -= cur; | |
3229 | } | |
3230 | } | |
3231 | EXPORT_SYMBOL(memcpy_extent_buffer); | |
3232 | ||
3233 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
3234 | unsigned long src_offset, unsigned long len) | |
3235 | { | |
3236 | size_t cur; | |
3237 | size_t dst_off_in_page; | |
3238 | size_t src_off_in_page; | |
3239 | unsigned long dst_end = dst_offset + len - 1; | |
3240 | unsigned long src_end = src_offset + len - 1; | |
3241 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3242 | unsigned long dst_i; | |
3243 | unsigned long src_i; | |
3244 | ||
3245 | if (src_offset + len > dst->len) { | |
3246 | printk("memmove bogus src_offset %lu move len %lu len %lu\n", | |
3247 | src_offset, len, dst->len); | |
3248 | BUG_ON(1); | |
3249 | } | |
3250 | if (dst_offset + len > dst->len) { | |
3251 | printk("memmove bogus dst_offset %lu move len %lu len %lu\n", | |
3252 | dst_offset, len, dst->len); | |
3253 | BUG_ON(1); | |
3254 | } | |
3255 | if (dst_offset < src_offset) { | |
3256 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); | |
3257 | return; | |
3258 | } | |
3259 | while(len > 0) { | |
3260 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; | |
3261 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
3262 | ||
3263 | dst_off_in_page = (start_offset + dst_end) & | |
3264 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3265 | src_off_in_page = (start_offset + src_end) & | |
3266 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3267 | ||
3268 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
3269 | cur = min(cur, dst_off_in_page + 1); | |
3270 | move_pages(extent_buffer_page(dst, dst_i), | |
3271 | extent_buffer_page(dst, src_i), | |
3272 | dst_off_in_page - cur + 1, | |
3273 | src_off_in_page - cur + 1, cur); | |
3274 | ||
3275 | dst_end -= cur; | |
3276 | src_end -= cur; | |
3277 | len -= cur; | |
3278 | } | |
3279 | } | |
3280 | EXPORT_SYMBOL(memmove_extent_buffer); |