Commit | Line | Data |
---|---|---|
a52d9a80 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 "extent_map.h" | |
12 | ||
86479a04 CM |
13 | /* temporary define until extent_map moves out of btrfs */ |
14 | struct kmem_cache *btrfs_cache_create(const char *name, size_t size, | |
15 | unsigned long extra_flags, | |
16 | void (*ctor)(void *, struct kmem_cache *, | |
17 | unsigned long)); | |
18 | ||
a52d9a80 CM |
19 | static struct kmem_cache *extent_map_cache; |
20 | static struct kmem_cache *extent_state_cache; | |
6d36dcd4 | 21 | static struct kmem_cache *extent_buffer_cache; |
f510cfec | 22 | |
6d36dcd4 | 23 | static LIST_HEAD(extent_buffers); |
f510cfec CM |
24 | static LIST_HEAD(buffers); |
25 | static LIST_HEAD(states); | |
26 | ||
6d36dcd4 | 27 | static spinlock_t extent_buffers_lock; |
f510cfec | 28 | static spinlock_t state_lock = SPIN_LOCK_UNLOCKED; |
6d36dcd4 CM |
29 | static int nr_extent_buffers; |
30 | #define MAX_EXTENT_BUFFER_CACHE 128 | |
a52d9a80 CM |
31 | |
32 | struct tree_entry { | |
33 | u64 start; | |
34 | u64 end; | |
35 | int in_tree; | |
36 | struct rb_node rb_node; | |
37 | }; | |
38 | ||
a52d9a80 CM |
39 | void __init extent_map_init(void) |
40 | { | |
86479a04 | 41 | extent_map_cache = btrfs_cache_create("extent_map", |
6d36dcd4 | 42 | sizeof(struct extent_map), 0, |
a52d9a80 | 43 | NULL); |
86479a04 | 44 | extent_state_cache = btrfs_cache_create("extent_state", |
6d36dcd4 | 45 | sizeof(struct extent_state), 0, |
a52d9a80 | 46 | NULL); |
6d36dcd4 CM |
47 | extent_buffer_cache = btrfs_cache_create("extent_buffers", |
48 | sizeof(struct extent_buffer), 0, | |
49 | NULL); | |
50 | spin_lock_init(&extent_buffers_lock); | |
a52d9a80 CM |
51 | } |
52 | ||
53 | void __exit extent_map_exit(void) | |
54 | { | |
6d36dcd4 | 55 | struct extent_buffer *eb; |
f510cfec | 56 | struct extent_state *state; |
6d36dcd4 CM |
57 | |
58 | while (!list_empty(&extent_buffers)) { | |
59 | eb = list_entry(extent_buffers.next, | |
60 | struct extent_buffer, list); | |
61 | list_del(&eb->list); | |
62 | kmem_cache_free(extent_buffer_cache, eb); | |
63 | } | |
f510cfec CM |
64 | while (!list_empty(&states)) { |
65 | state = list_entry(states.next, struct extent_state, list); | |
66 | printk("state leak: start %Lu end %Lu state %lu in tree %d refs %d\n", state->start, state->end, state->state, state->in_tree, atomic_read(&state->refs)); | |
67 | list_del(&state->list); | |
68 | kmem_cache_free(extent_state_cache, state); | |
69 | ||
70 | } | |
71 | while (!list_empty(&buffers)) { | |
72 | eb = list_entry(buffers.next, | |
73 | struct extent_buffer, leak_list); | |
74 | printk("buffer leak start %Lu len %lu return %lX\n", eb->start, eb->len, eb->alloc_addr); | |
75 | list_del(&eb->leak_list); | |
76 | kmem_cache_free(extent_buffer_cache, eb); | |
77 | } | |
78 | ||
79 | ||
a52d9a80 CM |
80 | if (extent_map_cache) |
81 | kmem_cache_destroy(extent_map_cache); | |
82 | if (extent_state_cache) | |
83 | kmem_cache_destroy(extent_state_cache); | |
6d36dcd4 CM |
84 | if (extent_buffer_cache) |
85 | kmem_cache_destroy(extent_buffer_cache); | |
a52d9a80 CM |
86 | } |
87 | ||
88 | void extent_map_tree_init(struct extent_map_tree *tree, | |
89 | struct address_space *mapping, gfp_t mask) | |
90 | { | |
91 | tree->map.rb_node = NULL; | |
92 | tree->state.rb_node = NULL; | |
07157aac | 93 | tree->ops = NULL; |
a52d9a80 CM |
94 | rwlock_init(&tree->lock); |
95 | tree->mapping = mapping; | |
96 | } | |
97 | EXPORT_SYMBOL(extent_map_tree_init); | |
98 | ||
99 | struct extent_map *alloc_extent_map(gfp_t mask) | |
100 | { | |
101 | struct extent_map *em; | |
102 | em = kmem_cache_alloc(extent_map_cache, mask); | |
103 | if (!em || IS_ERR(em)) | |
104 | return em; | |
105 | em->in_tree = 0; | |
106 | atomic_set(&em->refs, 1); | |
107 | return em; | |
108 | } | |
109 | EXPORT_SYMBOL(alloc_extent_map); | |
110 | ||
111 | void free_extent_map(struct extent_map *em) | |
112 | { | |
2bf5a725 CM |
113 | if (!em) |
114 | return; | |
a52d9a80 CM |
115 | if (atomic_dec_and_test(&em->refs)) { |
116 | WARN_ON(em->in_tree); | |
117 | kmem_cache_free(extent_map_cache, em); | |
118 | } | |
119 | } | |
120 | EXPORT_SYMBOL(free_extent_map); | |
121 | ||
122 | ||
123 | struct extent_state *alloc_extent_state(gfp_t mask) | |
124 | { | |
125 | struct extent_state *state; | |
f510cfec CM |
126 | unsigned long flags; |
127 | ||
a52d9a80 CM |
128 | state = kmem_cache_alloc(extent_state_cache, mask); |
129 | if (!state || IS_ERR(state)) | |
130 | return state; | |
131 | state->state = 0; | |
132 | state->in_tree = 0; | |
07157aac | 133 | state->private = 0; |
f510cfec CM |
134 | |
135 | spin_lock_irqsave(&state_lock, flags); | |
136 | list_add(&state->list, &states); | |
137 | spin_unlock_irqrestore(&state_lock, flags); | |
138 | ||
a52d9a80 CM |
139 | atomic_set(&state->refs, 1); |
140 | init_waitqueue_head(&state->wq); | |
a52d9a80 CM |
141 | return state; |
142 | } | |
143 | EXPORT_SYMBOL(alloc_extent_state); | |
144 | ||
145 | void free_extent_state(struct extent_state *state) | |
146 | { | |
f510cfec | 147 | unsigned long flags; |
2bf5a725 CM |
148 | if (!state) |
149 | return; | |
a52d9a80 CM |
150 | if (atomic_dec_and_test(&state->refs)) { |
151 | WARN_ON(state->in_tree); | |
f510cfec CM |
152 | spin_lock_irqsave(&state_lock, flags); |
153 | list_del(&state->list); | |
154 | spin_unlock_irqrestore(&state_lock, flags); | |
a52d9a80 CM |
155 | kmem_cache_free(extent_state_cache, state); |
156 | } | |
157 | } | |
158 | EXPORT_SYMBOL(free_extent_state); | |
159 | ||
160 | static struct rb_node *tree_insert(struct rb_root *root, u64 offset, | |
161 | struct rb_node *node) | |
162 | { | |
163 | struct rb_node ** p = &root->rb_node; | |
164 | struct rb_node * parent = NULL; | |
165 | struct tree_entry *entry; | |
166 | ||
167 | while(*p) { | |
168 | parent = *p; | |
169 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
170 | ||
171 | if (offset < entry->start) | |
172 | p = &(*p)->rb_left; | |
173 | else if (offset > entry->end) | |
174 | p = &(*p)->rb_right; | |
175 | else | |
176 | return parent; | |
177 | } | |
178 | ||
179 | entry = rb_entry(node, struct tree_entry, rb_node); | |
180 | entry->in_tree = 1; | |
181 | rb_link_node(node, parent, p); | |
182 | rb_insert_color(node, root); | |
183 | return NULL; | |
184 | } | |
185 | ||
186 | static struct rb_node *__tree_search(struct rb_root *root, u64 offset, | |
187 | struct rb_node **prev_ret) | |
188 | { | |
189 | struct rb_node * n = root->rb_node; | |
190 | struct rb_node *prev = NULL; | |
191 | struct tree_entry *entry; | |
192 | struct tree_entry *prev_entry = NULL; | |
193 | ||
194 | while(n) { | |
195 | entry = rb_entry(n, struct tree_entry, rb_node); | |
196 | prev = n; | |
197 | prev_entry = entry; | |
198 | ||
199 | if (offset < entry->start) | |
200 | n = n->rb_left; | |
201 | else if (offset > entry->end) | |
202 | n = n->rb_right; | |
203 | else | |
204 | return n; | |
205 | } | |
206 | if (!prev_ret) | |
207 | return NULL; | |
208 | while(prev && offset > prev_entry->end) { | |
209 | prev = rb_next(prev); | |
210 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
211 | } | |
212 | *prev_ret = prev; | |
213 | return NULL; | |
214 | } | |
215 | ||
216 | static inline struct rb_node *tree_search(struct rb_root *root, u64 offset) | |
217 | { | |
218 | struct rb_node *prev; | |
219 | struct rb_node *ret; | |
220 | ret = __tree_search(root, offset, &prev); | |
221 | if (!ret) | |
222 | return prev; | |
223 | return ret; | |
224 | } | |
225 | ||
226 | static int tree_delete(struct rb_root *root, u64 offset) | |
227 | { | |
228 | struct rb_node *node; | |
229 | struct tree_entry *entry; | |
230 | ||
231 | node = __tree_search(root, offset, NULL); | |
232 | if (!node) | |
233 | return -ENOENT; | |
234 | entry = rb_entry(node, struct tree_entry, rb_node); | |
235 | entry->in_tree = 0; | |
236 | rb_erase(node, root); | |
237 | return 0; | |
238 | } | |
239 | ||
240 | /* | |
241 | * add_extent_mapping tries a simple backward merge with existing | |
242 | * mappings. The extent_map struct passed in will be inserted into | |
243 | * the tree directly (no copies made, just a reference taken). | |
244 | */ | |
245 | int add_extent_mapping(struct extent_map_tree *tree, | |
246 | struct extent_map *em) | |
247 | { | |
248 | int ret = 0; | |
249 | struct extent_map *prev = NULL; | |
250 | struct rb_node *rb; | |
251 | ||
252 | write_lock_irq(&tree->lock); | |
253 | rb = tree_insert(&tree->map, em->end, &em->rb_node); | |
254 | if (rb) { | |
255 | prev = rb_entry(rb, struct extent_map, rb_node); | |
256 | printk("found extent map %Lu %Lu on insert of %Lu %Lu\n", prev->start, prev->end, em->start, em->end); | |
257 | ret = -EEXIST; | |
258 | goto out; | |
259 | } | |
260 | atomic_inc(&em->refs); | |
261 | if (em->start != 0) { | |
262 | rb = rb_prev(&em->rb_node); | |
263 | if (rb) | |
264 | prev = rb_entry(rb, struct extent_map, rb_node); | |
265 | if (prev && prev->end + 1 == em->start && | |
5f39d397 CM |
266 | ((em->block_start == EXTENT_MAP_HOLE && |
267 | prev->block_start == EXTENT_MAP_HOLE) || | |
a52d9a80 CM |
268 | (em->block_start == prev->block_end + 1))) { |
269 | em->start = prev->start; | |
270 | em->block_start = prev->block_start; | |
271 | rb_erase(&prev->rb_node, &tree->map); | |
272 | prev->in_tree = 0; | |
273 | free_extent_map(prev); | |
274 | } | |
275 | } | |
276 | out: | |
277 | write_unlock_irq(&tree->lock); | |
278 | return ret; | |
279 | } | |
280 | EXPORT_SYMBOL(add_extent_mapping); | |
281 | ||
282 | /* | |
283 | * lookup_extent_mapping returns the first extent_map struct in the | |
284 | * tree that intersects the [start, end] (inclusive) range. There may | |
285 | * be additional objects in the tree that intersect, so check the object | |
286 | * returned carefully to make sure you don't need additional lookups. | |
287 | */ | |
288 | struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree, | |
289 | u64 start, u64 end) | |
290 | { | |
291 | struct extent_map *em; | |
292 | struct rb_node *rb_node; | |
293 | ||
294 | read_lock_irq(&tree->lock); | |
295 | rb_node = tree_search(&tree->map, start); | |
296 | if (!rb_node) { | |
297 | em = NULL; | |
298 | goto out; | |
299 | } | |
300 | if (IS_ERR(rb_node)) { | |
301 | em = ERR_PTR(PTR_ERR(rb_node)); | |
302 | goto out; | |
303 | } | |
304 | em = rb_entry(rb_node, struct extent_map, rb_node); | |
305 | if (em->end < start || em->start > end) { | |
306 | em = NULL; | |
307 | goto out; | |
308 | } | |
309 | atomic_inc(&em->refs); | |
310 | out: | |
311 | read_unlock_irq(&tree->lock); | |
312 | return em; | |
313 | } | |
314 | EXPORT_SYMBOL(lookup_extent_mapping); | |
315 | ||
316 | /* | |
317 | * removes an extent_map struct from the tree. No reference counts are | |
318 | * dropped, and no checks are done to see if the range is in use | |
319 | */ | |
320 | int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em) | |
321 | { | |
322 | int ret; | |
323 | ||
324 | write_lock_irq(&tree->lock); | |
325 | ret = tree_delete(&tree->map, em->end); | |
326 | write_unlock_irq(&tree->lock); | |
327 | return ret; | |
328 | } | |
329 | EXPORT_SYMBOL(remove_extent_mapping); | |
330 | ||
331 | /* | |
332 | * utility function to look for merge candidates inside a given range. | |
333 | * Any extents with matching state are merged together into a single | |
334 | * extent in the tree. Extents with EXTENT_IO in their state field | |
335 | * are not merged because the end_io handlers need to be able to do | |
336 | * operations on them without sleeping (or doing allocations/splits). | |
337 | * | |
338 | * This should be called with the tree lock held. | |
339 | */ | |
340 | static int merge_state(struct extent_map_tree *tree, | |
341 | struct extent_state *state) | |
342 | { | |
343 | struct extent_state *other; | |
344 | struct rb_node *other_node; | |
345 | ||
346 | if (state->state & EXTENT_IOBITS) | |
347 | return 0; | |
348 | ||
349 | other_node = rb_prev(&state->rb_node); | |
350 | if (other_node) { | |
351 | other = rb_entry(other_node, struct extent_state, rb_node); | |
352 | if (other->end == state->start - 1 && | |
353 | other->state == state->state) { | |
354 | state->start = other->start; | |
355 | other->in_tree = 0; | |
356 | rb_erase(&other->rb_node, &tree->state); | |
357 | free_extent_state(other); | |
358 | } | |
359 | } | |
360 | other_node = rb_next(&state->rb_node); | |
361 | if (other_node) { | |
362 | other = rb_entry(other_node, struct extent_state, rb_node); | |
363 | if (other->start == state->end + 1 && | |
364 | other->state == state->state) { | |
365 | other->start = state->start; | |
366 | state->in_tree = 0; | |
367 | rb_erase(&state->rb_node, &tree->state); | |
368 | free_extent_state(state); | |
369 | } | |
370 | } | |
371 | return 0; | |
372 | } | |
373 | ||
374 | /* | |
375 | * insert an extent_state struct into the tree. 'bits' are set on the | |
376 | * struct before it is inserted. | |
377 | * | |
378 | * This may return -EEXIST if the extent is already there, in which case the | |
379 | * state struct is freed. | |
380 | * | |
381 | * The tree lock is not taken internally. This is a utility function and | |
382 | * probably isn't what you want to call (see set/clear_extent_bit). | |
383 | */ | |
384 | static int insert_state(struct extent_map_tree *tree, | |
385 | struct extent_state *state, u64 start, u64 end, | |
386 | int bits) | |
387 | { | |
388 | struct rb_node *node; | |
389 | ||
390 | if (end < start) { | |
391 | printk("end < start %Lu %Lu\n", end, start); | |
392 | WARN_ON(1); | |
393 | } | |
394 | state->state |= bits; | |
395 | state->start = start; | |
396 | state->end = end; | |
a52d9a80 CM |
397 | node = tree_insert(&tree->state, end, &state->rb_node); |
398 | if (node) { | |
399 | struct extent_state *found; | |
400 | found = rb_entry(node, struct extent_state, rb_node); | |
b888db2b | 401 | printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end); |
a52d9a80 CM |
402 | free_extent_state(state); |
403 | return -EEXIST; | |
404 | } | |
405 | merge_state(tree, state); | |
406 | return 0; | |
407 | } | |
408 | ||
409 | /* | |
410 | * split a given extent state struct in two, inserting the preallocated | |
411 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
412 | * offset inside 'orig' where it should be split. | |
413 | * | |
414 | * Before calling, | |
415 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
416 | * are two extent state structs in the tree: | |
417 | * prealloc: [orig->start, split - 1] | |
418 | * orig: [ split, orig->end ] | |
419 | * | |
420 | * The tree locks are not taken by this function. They need to be held | |
421 | * by the caller. | |
422 | */ | |
423 | static int split_state(struct extent_map_tree *tree, struct extent_state *orig, | |
424 | struct extent_state *prealloc, u64 split) | |
425 | { | |
426 | struct rb_node *node; | |
427 | prealloc->start = orig->start; | |
428 | prealloc->end = split - 1; | |
429 | prealloc->state = orig->state; | |
430 | orig->start = split; | |
f510cfec | 431 | |
a52d9a80 CM |
432 | node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node); |
433 | if (node) { | |
434 | struct extent_state *found; | |
435 | found = rb_entry(node, struct extent_state, rb_node); | |
b888db2b | 436 | printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end); |
a52d9a80 CM |
437 | free_extent_state(prealloc); |
438 | return -EEXIST; | |
439 | } | |
440 | return 0; | |
441 | } | |
442 | ||
443 | /* | |
444 | * utility function to clear some bits in an extent state struct. | |
445 | * it will optionally wake up any one waiting on this state (wake == 1), or | |
446 | * forcibly remove the state from the tree (delete == 1). | |
447 | * | |
448 | * If no bits are set on the state struct after clearing things, the | |
449 | * struct is freed and removed from the tree | |
450 | */ | |
451 | static int clear_state_bit(struct extent_map_tree *tree, | |
452 | struct extent_state *state, int bits, int wake, | |
453 | int delete) | |
454 | { | |
455 | int ret = state->state & bits; | |
456 | state->state &= ~bits; | |
457 | if (wake) | |
458 | wake_up(&state->wq); | |
459 | if (delete || state->state == 0) { | |
460 | if (state->in_tree) { | |
461 | rb_erase(&state->rb_node, &tree->state); | |
462 | state->in_tree = 0; | |
463 | free_extent_state(state); | |
464 | } else { | |
465 | WARN_ON(1); | |
466 | } | |
467 | } else { | |
468 | merge_state(tree, state); | |
469 | } | |
470 | return ret; | |
471 | } | |
472 | ||
473 | /* | |
474 | * clear some bits on a range in the tree. This may require splitting | |
475 | * or inserting elements in the tree, so the gfp mask is used to | |
476 | * indicate which allocations or sleeping are allowed. | |
477 | * | |
478 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
479 | * the given range from the tree regardless of state (ie for truncate). | |
480 | * | |
481 | * the range [start, end] is inclusive. | |
482 | * | |
483 | * This takes the tree lock, and returns < 0 on error, > 0 if any of the | |
484 | * bits were already set, or zero if none of the bits were already set. | |
485 | */ | |
486 | int clear_extent_bit(struct extent_map_tree *tree, u64 start, u64 end, | |
487 | int bits, int wake, int delete, gfp_t mask) | |
488 | { | |
489 | struct extent_state *state; | |
490 | struct extent_state *prealloc = NULL; | |
491 | struct rb_node *node; | |
90f1c19a | 492 | unsigned long flags; |
a52d9a80 CM |
493 | int err; |
494 | int set = 0; | |
495 | ||
496 | again: | |
497 | if (!prealloc && (mask & __GFP_WAIT)) { | |
498 | prealloc = alloc_extent_state(mask); | |
499 | if (!prealloc) | |
500 | return -ENOMEM; | |
501 | } | |
502 | ||
90f1c19a | 503 | write_lock_irqsave(&tree->lock, flags); |
a52d9a80 CM |
504 | /* |
505 | * this search will find the extents that end after | |
506 | * our range starts | |
507 | */ | |
508 | node = tree_search(&tree->state, start); | |
509 | if (!node) | |
510 | goto out; | |
511 | state = rb_entry(node, struct extent_state, rb_node); | |
512 | if (state->start > end) | |
513 | goto out; | |
514 | WARN_ON(state->end < start); | |
515 | ||
516 | /* | |
517 | * | ---- desired range ---- | | |
518 | * | state | or | |
519 | * | ------------- state -------------- | | |
520 | * | |
521 | * We need to split the extent we found, and may flip | |
522 | * bits on second half. | |
523 | * | |
524 | * If the extent we found extends past our range, we | |
525 | * just split and search again. It'll get split again | |
526 | * the next time though. | |
527 | * | |
528 | * If the extent we found is inside our range, we clear | |
529 | * the desired bit on it. | |
530 | */ | |
531 | ||
532 | if (state->start < start) { | |
533 | err = split_state(tree, state, prealloc, start); | |
534 | BUG_ON(err == -EEXIST); | |
535 | prealloc = NULL; | |
536 | if (err) | |
537 | goto out; | |
538 | if (state->end <= end) { | |
539 | start = state->end + 1; | |
540 | set |= clear_state_bit(tree, state, bits, | |
541 | wake, delete); | |
542 | } else { | |
543 | start = state->start; | |
544 | } | |
545 | goto search_again; | |
546 | } | |
547 | /* | |
548 | * | ---- desired range ---- | | |
549 | * | state | | |
550 | * We need to split the extent, and clear the bit | |
551 | * on the first half | |
552 | */ | |
553 | if (state->start <= end && state->end > end) { | |
554 | err = split_state(tree, state, prealloc, end + 1); | |
555 | BUG_ON(err == -EEXIST); | |
556 | ||
557 | if (wake) | |
558 | wake_up(&state->wq); | |
559 | set |= clear_state_bit(tree, prealloc, bits, | |
560 | wake, delete); | |
561 | prealloc = NULL; | |
562 | goto out; | |
563 | } | |
564 | ||
565 | start = state->end + 1; | |
566 | set |= clear_state_bit(tree, state, bits, wake, delete); | |
567 | goto search_again; | |
568 | ||
569 | out: | |
90f1c19a | 570 | write_unlock_irqrestore(&tree->lock, flags); |
a52d9a80 CM |
571 | if (prealloc) |
572 | free_extent_state(prealloc); | |
573 | ||
574 | return set; | |
575 | ||
576 | search_again: | |
96b5179d | 577 | if (start > end) |
a52d9a80 | 578 | goto out; |
90f1c19a | 579 | write_unlock_irqrestore(&tree->lock, flags); |
a52d9a80 CM |
580 | if (mask & __GFP_WAIT) |
581 | cond_resched(); | |
582 | goto again; | |
583 | } | |
584 | EXPORT_SYMBOL(clear_extent_bit); | |
585 | ||
586 | static int wait_on_state(struct extent_map_tree *tree, | |
587 | struct extent_state *state) | |
588 | { | |
589 | DEFINE_WAIT(wait); | |
590 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
591 | read_unlock_irq(&tree->lock); | |
592 | schedule(); | |
593 | read_lock_irq(&tree->lock); | |
594 | finish_wait(&state->wq, &wait); | |
595 | return 0; | |
596 | } | |
597 | ||
598 | /* | |
599 | * waits for one or more bits to clear on a range in the state tree. | |
600 | * The range [start, end] is inclusive. | |
601 | * The tree lock is taken by this function | |
602 | */ | |
603 | int wait_extent_bit(struct extent_map_tree *tree, u64 start, u64 end, int bits) | |
604 | { | |
605 | struct extent_state *state; | |
606 | struct rb_node *node; | |
607 | ||
608 | read_lock_irq(&tree->lock); | |
609 | again: | |
610 | while (1) { | |
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 | break; | |
618 | ||
619 | state = rb_entry(node, struct extent_state, rb_node); | |
620 | ||
621 | if (state->start > end) | |
622 | goto out; | |
623 | ||
624 | if (state->state & bits) { | |
625 | start = state->start; | |
626 | atomic_inc(&state->refs); | |
627 | wait_on_state(tree, state); | |
628 | free_extent_state(state); | |
629 | goto again; | |
630 | } | |
631 | start = state->end + 1; | |
632 | ||
633 | if (start > end) | |
634 | break; | |
635 | ||
636 | if (need_resched()) { | |
637 | read_unlock_irq(&tree->lock); | |
638 | cond_resched(); | |
639 | read_lock_irq(&tree->lock); | |
640 | } | |
641 | } | |
642 | out: | |
643 | read_unlock_irq(&tree->lock); | |
644 | return 0; | |
645 | } | |
646 | EXPORT_SYMBOL(wait_extent_bit); | |
647 | ||
648 | /* | |
649 | * set some bits on a range in the tree. This may require allocations | |
650 | * or sleeping, so the gfp mask is used to indicate what is allowed. | |
651 | * | |
652 | * If 'exclusive' == 1, this will fail with -EEXIST if some part of the | |
653 | * range already has the desired bits set. The start of the existing | |
654 | * range is returned in failed_start in this case. | |
655 | * | |
656 | * [start, end] is inclusive | |
657 | * This takes the tree lock. | |
658 | */ | |
659 | int set_extent_bit(struct extent_map_tree *tree, u64 start, u64 end, int bits, | |
660 | int exclusive, u64 *failed_start, gfp_t mask) | |
661 | { | |
662 | struct extent_state *state; | |
663 | struct extent_state *prealloc = NULL; | |
664 | struct rb_node *node; | |
90f1c19a | 665 | unsigned long flags; |
a52d9a80 CM |
666 | int err = 0; |
667 | int set; | |
668 | u64 last_start; | |
669 | u64 last_end; | |
670 | again: | |
671 | if (!prealloc && (mask & __GFP_WAIT)) { | |
672 | prealloc = alloc_extent_state(mask); | |
673 | if (!prealloc) | |
674 | return -ENOMEM; | |
675 | } | |
676 | ||
90f1c19a | 677 | write_lock_irqsave(&tree->lock, flags); |
a52d9a80 CM |
678 | /* |
679 | * this search will find all the extents that end after | |
680 | * our range starts. | |
681 | */ | |
682 | node = tree_search(&tree->state, start); | |
683 | if (!node) { | |
684 | err = insert_state(tree, prealloc, start, end, bits); | |
685 | prealloc = NULL; | |
686 | BUG_ON(err == -EEXIST); | |
687 | goto out; | |
688 | } | |
689 | ||
690 | state = rb_entry(node, struct extent_state, rb_node); | |
691 | last_start = state->start; | |
692 | last_end = state->end; | |
693 | ||
694 | /* | |
695 | * | ---- desired range ---- | | |
696 | * | state | | |
697 | * | |
698 | * Just lock what we found and keep going | |
699 | */ | |
700 | if (state->start == start && state->end <= end) { | |
701 | set = state->state & bits; | |
702 | if (set && exclusive) { | |
703 | *failed_start = state->start; | |
704 | err = -EEXIST; | |
705 | goto out; | |
706 | } | |
707 | state->state |= bits; | |
708 | start = state->end + 1; | |
709 | merge_state(tree, state); | |
710 | goto search_again; | |
711 | } | |
712 | ||
713 | /* | |
714 | * | ---- desired range ---- | | |
715 | * | state | | |
716 | * or | |
717 | * | ------------- state -------------- | | |
718 | * | |
719 | * We need to split the extent we found, and may flip bits on | |
720 | * second half. | |
721 | * | |
722 | * If the extent we found extends past our | |
723 | * range, we just split and search again. It'll get split | |
724 | * again the next time though. | |
725 | * | |
726 | * If the extent we found is inside our range, we set the | |
727 | * desired bit on it. | |
728 | */ | |
729 | if (state->start < start) { | |
730 | set = state->state & bits; | |
731 | if (exclusive && set) { | |
732 | *failed_start = start; | |
733 | err = -EEXIST; | |
734 | goto out; | |
735 | } | |
736 | err = split_state(tree, state, prealloc, start); | |
737 | BUG_ON(err == -EEXIST); | |
738 | prealloc = NULL; | |
739 | if (err) | |
740 | goto out; | |
741 | if (state->end <= end) { | |
742 | state->state |= bits; | |
743 | start = state->end + 1; | |
744 | merge_state(tree, state); | |
745 | } else { | |
746 | start = state->start; | |
747 | } | |
748 | goto search_again; | |
749 | } | |
a52d9a80 CM |
750 | /* |
751 | * | ---- desired range ---- | | |
752 | * | state | or | state | | |
753 | * | |
754 | * There's a hole, we need to insert something in it and | |
755 | * ignore the extent we found. | |
756 | */ | |
757 | if (state->start > start) { | |
758 | u64 this_end; | |
759 | if (end < last_start) | |
760 | this_end = end; | |
761 | else | |
762 | this_end = last_start -1; | |
763 | err = insert_state(tree, prealloc, start, this_end, | |
764 | bits); | |
765 | prealloc = NULL; | |
766 | BUG_ON(err == -EEXIST); | |
767 | if (err) | |
768 | goto out; | |
769 | start = this_end + 1; | |
770 | goto search_again; | |
771 | } | |
a8c450b2 CM |
772 | /* |
773 | * | ---- desired range ---- | | |
774 | * | state | | |
775 | * We need to split the extent, and set the bit | |
776 | * on the first half | |
777 | */ | |
778 | if (state->start <= end && state->end > end) { | |
779 | set = state->state & bits; | |
780 | if (exclusive && set) { | |
781 | *failed_start = start; | |
782 | err = -EEXIST; | |
783 | goto out; | |
784 | } | |
785 | err = split_state(tree, state, prealloc, end + 1); | |
786 | BUG_ON(err == -EEXIST); | |
787 | ||
788 | prealloc->state |= bits; | |
789 | merge_state(tree, prealloc); | |
790 | prealloc = NULL; | |
791 | goto out; | |
792 | } | |
793 | ||
a52d9a80 CM |
794 | goto search_again; |
795 | ||
796 | out: | |
90f1c19a | 797 | write_unlock_irqrestore(&tree->lock, flags); |
a52d9a80 CM |
798 | if (prealloc) |
799 | free_extent_state(prealloc); | |
800 | ||
801 | return err; | |
802 | ||
803 | search_again: | |
804 | if (start > end) | |
805 | goto out; | |
90f1c19a | 806 | write_unlock_irqrestore(&tree->lock, flags); |
a52d9a80 CM |
807 | if (mask & __GFP_WAIT) |
808 | cond_resched(); | |
809 | goto again; | |
810 | } | |
811 | EXPORT_SYMBOL(set_extent_bit); | |
812 | ||
813 | /* wrappers around set/clear extent bit */ | |
814 | int set_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end, | |
815 | gfp_t mask) | |
816 | { | |
817 | return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL, | |
818 | mask); | |
819 | } | |
820 | EXPORT_SYMBOL(set_extent_dirty); | |
821 | ||
96b5179d CM |
822 | int set_extent_bits(struct extent_map_tree *tree, u64 start, u64 end, |
823 | int bits, gfp_t mask) | |
824 | { | |
825 | return set_extent_bit(tree, start, end, bits, 0, NULL, | |
826 | mask); | |
827 | } | |
828 | EXPORT_SYMBOL(set_extent_bits); | |
829 | ||
830 | int clear_extent_bits(struct extent_map_tree *tree, u64 start, u64 end, | |
831 | int bits, gfp_t mask) | |
832 | { | |
833 | return clear_extent_bit(tree, start, end, bits, 0, 0, mask); | |
834 | } | |
835 | EXPORT_SYMBOL(clear_extent_bits); | |
836 | ||
b888db2b CM |
837 | int set_extent_delalloc(struct extent_map_tree *tree, u64 start, u64 end, |
838 | gfp_t mask) | |
839 | { | |
840 | return set_extent_bit(tree, start, end, | |
841 | EXTENT_DELALLOC | EXTENT_DIRTY, 0, NULL, | |
842 | mask); | |
843 | } | |
844 | EXPORT_SYMBOL(set_extent_delalloc); | |
845 | ||
a52d9a80 CM |
846 | int clear_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end, |
847 | gfp_t mask) | |
848 | { | |
b888db2b CM |
849 | return clear_extent_bit(tree, start, end, |
850 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask); | |
a52d9a80 CM |
851 | } |
852 | EXPORT_SYMBOL(clear_extent_dirty); | |
853 | ||
854 | int set_extent_new(struct extent_map_tree *tree, u64 start, u64 end, | |
855 | gfp_t mask) | |
856 | { | |
857 | return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL, | |
858 | mask); | |
859 | } | |
860 | EXPORT_SYMBOL(set_extent_new); | |
861 | ||
862 | int clear_extent_new(struct extent_map_tree *tree, u64 start, u64 end, | |
863 | gfp_t mask) | |
864 | { | |
865 | return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask); | |
866 | } | |
867 | EXPORT_SYMBOL(clear_extent_new); | |
868 | ||
869 | int set_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end, | |
870 | gfp_t mask) | |
871 | { | |
872 | return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL, | |
873 | mask); | |
874 | } | |
875 | EXPORT_SYMBOL(set_extent_uptodate); | |
876 | ||
877 | int clear_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end, | |
878 | gfp_t mask) | |
879 | { | |
880 | return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask); | |
881 | } | |
882 | EXPORT_SYMBOL(clear_extent_uptodate); | |
883 | ||
884 | int set_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end, | |
885 | gfp_t mask) | |
886 | { | |
887 | return set_extent_bit(tree, start, end, EXTENT_WRITEBACK, | |
888 | 0, NULL, mask); | |
889 | } | |
890 | EXPORT_SYMBOL(set_extent_writeback); | |
891 | ||
892 | int clear_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end, | |
893 | gfp_t mask) | |
894 | { | |
895 | return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask); | |
896 | } | |
897 | EXPORT_SYMBOL(clear_extent_writeback); | |
898 | ||
899 | int wait_on_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end) | |
900 | { | |
901 | return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK); | |
902 | } | |
903 | EXPORT_SYMBOL(wait_on_extent_writeback); | |
904 | ||
905 | /* | |
906 | * locks a range in ascending order, waiting for any locked regions | |
907 | * it hits on the way. [start,end] are inclusive, and this will sleep. | |
908 | */ | |
909 | int lock_extent(struct extent_map_tree *tree, u64 start, u64 end, gfp_t mask) | |
910 | { | |
911 | int err; | |
912 | u64 failed_start; | |
913 | while (1) { | |
914 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1, | |
915 | &failed_start, mask); | |
916 | if (err == -EEXIST && (mask & __GFP_WAIT)) { | |
917 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); | |
918 | start = failed_start; | |
919 | } else { | |
920 | break; | |
921 | } | |
922 | WARN_ON(start > end); | |
923 | } | |
924 | return err; | |
925 | } | |
926 | EXPORT_SYMBOL(lock_extent); | |
927 | ||
928 | int unlock_extent(struct extent_map_tree *tree, u64 start, u64 end, | |
929 | gfp_t mask) | |
930 | { | |
931 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask); | |
932 | } | |
933 | EXPORT_SYMBOL(unlock_extent); | |
934 | ||
935 | /* | |
936 | * helper function to set pages and extents in the tree dirty | |
937 | */ | |
938 | int set_range_dirty(struct extent_map_tree *tree, u64 start, u64 end) | |
939 | { | |
940 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
941 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
942 | struct page *page; | |
943 | ||
944 | while (index <= end_index) { | |
945 | page = find_get_page(tree->mapping, index); | |
946 | BUG_ON(!page); | |
947 | __set_page_dirty_nobuffers(page); | |
948 | page_cache_release(page); | |
949 | index++; | |
950 | } | |
951 | set_extent_dirty(tree, start, end, GFP_NOFS); | |
952 | return 0; | |
953 | } | |
954 | EXPORT_SYMBOL(set_range_dirty); | |
955 | ||
956 | /* | |
957 | * helper function to set both pages and extents in the tree writeback | |
958 | */ | |
959 | int set_range_writeback(struct extent_map_tree *tree, u64 start, u64 end) | |
960 | { | |
961 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
962 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
963 | struct page *page; | |
964 | ||
965 | while (index <= end_index) { | |
966 | page = find_get_page(tree->mapping, index); | |
967 | BUG_ON(!page); | |
968 | set_page_writeback(page); | |
969 | page_cache_release(page); | |
970 | index++; | |
971 | } | |
972 | set_extent_writeback(tree, start, end, GFP_NOFS); | |
973 | return 0; | |
974 | } | |
975 | EXPORT_SYMBOL(set_range_writeback); | |
976 | ||
5f39d397 CM |
977 | int find_first_extent_bit(struct extent_map_tree *tree, u64 start, |
978 | u64 *start_ret, u64 *end_ret, int bits) | |
979 | { | |
980 | struct rb_node *node; | |
981 | struct extent_state *state; | |
982 | int ret = 1; | |
983 | ||
984 | write_lock_irq(&tree->lock); | |
985 | /* | |
986 | * this search will find all the extents that end after | |
987 | * our range starts. | |
988 | */ | |
989 | node = tree_search(&tree->state, start); | |
990 | if (!node || IS_ERR(node)) { | |
991 | goto out; | |
992 | } | |
993 | ||
994 | while(1) { | |
995 | state = rb_entry(node, struct extent_state, rb_node); | |
996 | if (state->state & bits) { | |
997 | *start_ret = state->start; | |
998 | *end_ret = state->end; | |
999 | ret = 0; | |
f510cfec | 1000 | break; |
5f39d397 CM |
1001 | } |
1002 | node = rb_next(node); | |
1003 | if (!node) | |
1004 | break; | |
1005 | } | |
1006 | out: | |
1007 | write_unlock_irq(&tree->lock); | |
1008 | return ret; | |
1009 | } | |
1010 | EXPORT_SYMBOL(find_first_extent_bit); | |
1011 | ||
b888db2b CM |
1012 | u64 find_lock_delalloc_range(struct extent_map_tree *tree, |
1013 | u64 start, u64 lock_start, u64 *end, u64 max_bytes) | |
1014 | { | |
1015 | struct rb_node *node; | |
1016 | struct extent_state *state; | |
1017 | u64 cur_start = start; | |
1018 | u64 found = 0; | |
1019 | u64 total_bytes = 0; | |
1020 | ||
1021 | write_lock_irq(&tree->lock); | |
1022 | /* | |
1023 | * this search will find all the extents that end after | |
1024 | * our range starts. | |
1025 | */ | |
1026 | search_again: | |
1027 | node = tree_search(&tree->state, cur_start); | |
1028 | if (!node || IS_ERR(node)) { | |
1029 | goto out; | |
1030 | } | |
1031 | ||
1032 | while(1) { | |
1033 | state = rb_entry(node, struct extent_state, rb_node); | |
1034 | if (state->start != cur_start) { | |
1035 | goto out; | |
1036 | } | |
1037 | if (!(state->state & EXTENT_DELALLOC)) { | |
1038 | goto out; | |
1039 | } | |
1040 | if (state->start >= lock_start) { | |
1041 | if (state->state & EXTENT_LOCKED) { | |
1042 | DEFINE_WAIT(wait); | |
1043 | atomic_inc(&state->refs); | |
1044 | write_unlock_irq(&tree->lock); | |
1045 | schedule(); | |
1046 | write_lock_irq(&tree->lock); | |
1047 | finish_wait(&state->wq, &wait); | |
1048 | free_extent_state(state); | |
1049 | goto search_again; | |
1050 | } | |
1051 | state->state |= EXTENT_LOCKED; | |
1052 | } | |
1053 | found++; | |
1054 | *end = state->end; | |
1055 | cur_start = state->end + 1; | |
1056 | node = rb_next(node); | |
1057 | if (!node) | |
1058 | break; | |
1059 | total_bytes = state->end - state->start + 1; | |
1060 | if (total_bytes >= max_bytes) | |
1061 | break; | |
1062 | } | |
1063 | out: | |
1064 | write_unlock_irq(&tree->lock); | |
1065 | return found; | |
1066 | } | |
1067 | ||
a52d9a80 CM |
1068 | /* |
1069 | * helper function to lock both pages and extents in the tree. | |
1070 | * pages must be locked first. | |
1071 | */ | |
1072 | int lock_range(struct extent_map_tree *tree, u64 start, u64 end) | |
1073 | { | |
1074 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1075 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1076 | struct page *page; | |
1077 | int err; | |
1078 | ||
1079 | while (index <= end_index) { | |
1080 | page = grab_cache_page(tree->mapping, index); | |
1081 | if (!page) { | |
1082 | err = -ENOMEM; | |
1083 | goto failed; | |
1084 | } | |
1085 | if (IS_ERR(page)) { | |
1086 | err = PTR_ERR(page); | |
1087 | goto failed; | |
1088 | } | |
1089 | index++; | |
1090 | } | |
1091 | lock_extent(tree, start, end, GFP_NOFS); | |
1092 | return 0; | |
1093 | ||
1094 | failed: | |
1095 | /* | |
1096 | * we failed above in getting the page at 'index', so we undo here | |
1097 | * up to but not including the page at 'index' | |
1098 | */ | |
1099 | end_index = index; | |
1100 | index = start >> PAGE_CACHE_SHIFT; | |
1101 | while (index < end_index) { | |
1102 | page = find_get_page(tree->mapping, index); | |
1103 | unlock_page(page); | |
1104 | page_cache_release(page); | |
1105 | index++; | |
1106 | } | |
1107 | return err; | |
1108 | } | |
1109 | EXPORT_SYMBOL(lock_range); | |
1110 | ||
1111 | /* | |
1112 | * helper function to unlock both pages and extents in the tree. | |
1113 | */ | |
1114 | int unlock_range(struct extent_map_tree *tree, u64 start, u64 end) | |
1115 | { | |
1116 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1117 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1118 | struct page *page; | |
1119 | ||
1120 | while (index <= end_index) { | |
1121 | page = find_get_page(tree->mapping, index); | |
1122 | unlock_page(page); | |
1123 | page_cache_release(page); | |
1124 | index++; | |
1125 | } | |
1126 | unlock_extent(tree, start, end, GFP_NOFS); | |
1127 | return 0; | |
1128 | } | |
1129 | EXPORT_SYMBOL(unlock_range); | |
1130 | ||
07157aac CM |
1131 | int set_state_private(struct extent_map_tree *tree, u64 start, u64 private) |
1132 | { | |
1133 | struct rb_node *node; | |
1134 | struct extent_state *state; | |
1135 | int ret = 0; | |
1136 | ||
1137 | write_lock_irq(&tree->lock); | |
1138 | /* | |
1139 | * this search will find all the extents that end after | |
1140 | * our range starts. | |
1141 | */ | |
1142 | node = tree_search(&tree->state, start); | |
1143 | if (!node || IS_ERR(node)) { | |
1144 | ret = -ENOENT; | |
1145 | goto out; | |
1146 | } | |
1147 | state = rb_entry(node, struct extent_state, rb_node); | |
1148 | if (state->start != start) { | |
1149 | ret = -ENOENT; | |
1150 | goto out; | |
1151 | } | |
1152 | state->private = private; | |
1153 | out: | |
1154 | write_unlock_irq(&tree->lock); | |
1155 | return ret; | |
07157aac CM |
1156 | } |
1157 | ||
1158 | int get_state_private(struct extent_map_tree *tree, u64 start, u64 *private) | |
1159 | { | |
1160 | struct rb_node *node; | |
1161 | struct extent_state *state; | |
1162 | int ret = 0; | |
1163 | ||
1164 | read_lock_irq(&tree->lock); | |
1165 | /* | |
1166 | * this search will find all the extents that end after | |
1167 | * our range starts. | |
1168 | */ | |
1169 | node = tree_search(&tree->state, start); | |
1170 | if (!node || IS_ERR(node)) { | |
1171 | ret = -ENOENT; | |
1172 | goto out; | |
1173 | } | |
1174 | state = rb_entry(node, struct extent_state, rb_node); | |
1175 | if (state->start != start) { | |
1176 | ret = -ENOENT; | |
1177 | goto out; | |
1178 | } | |
1179 | *private = state->private; | |
1180 | out: | |
1181 | read_unlock_irq(&tree->lock); | |
1182 | return ret; | |
1183 | } | |
1184 | ||
a52d9a80 CM |
1185 | /* |
1186 | * searches a range in the state tree for a given mask. | |
1187 | * If 'filled' == 1, this returns 1 only if ever extent in the tree | |
1188 | * has the bits set. Otherwise, 1 is returned if any bit in the | |
1189 | * range is found set. | |
1190 | */ | |
1a5bc167 CM |
1191 | int test_range_bit(struct extent_map_tree *tree, u64 start, u64 end, |
1192 | int bits, int filled) | |
a52d9a80 CM |
1193 | { |
1194 | struct extent_state *state = NULL; | |
1195 | struct rb_node *node; | |
1196 | int bitset = 0; | |
1197 | ||
1198 | read_lock_irq(&tree->lock); | |
1199 | node = tree_search(&tree->state, start); | |
1200 | while (node && start <= end) { | |
1201 | state = rb_entry(node, struct extent_state, rb_node); | |
1202 | if (state->start > end) | |
1203 | break; | |
1204 | ||
1205 | if (filled && state->start > start) { | |
1206 | bitset = 0; | |
1207 | break; | |
1208 | } | |
1209 | if (state->state & bits) { | |
1210 | bitset = 1; | |
1211 | if (!filled) | |
1212 | break; | |
1213 | } else if (filled) { | |
1214 | bitset = 0; | |
1215 | break; | |
1216 | } | |
1217 | start = state->end + 1; | |
1218 | if (start > end) | |
1219 | break; | |
1220 | node = rb_next(node); | |
1221 | } | |
1222 | read_unlock_irq(&tree->lock); | |
1223 | return bitset; | |
1224 | } | |
1a5bc167 | 1225 | EXPORT_SYMBOL(test_range_bit); |
a52d9a80 CM |
1226 | |
1227 | /* | |
1228 | * helper function to set a given page up to date if all the | |
1229 | * extents in the tree for that page are up to date | |
1230 | */ | |
1231 | static int check_page_uptodate(struct extent_map_tree *tree, | |
1232 | struct page *page) | |
1233 | { | |
1234 | u64 start = page->index << PAGE_CACHE_SHIFT; | |
1235 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
1236 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1)) | |
1237 | SetPageUptodate(page); | |
1238 | return 0; | |
1239 | } | |
1240 | ||
1241 | /* | |
1242 | * helper function to unlock a page if all the extents in the tree | |
1243 | * for that page are unlocked | |
1244 | */ | |
1245 | static int check_page_locked(struct extent_map_tree *tree, | |
1246 | struct page *page) | |
1247 | { | |
1248 | u64 start = page->index << PAGE_CACHE_SHIFT; | |
1249 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
1250 | if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0)) | |
1251 | unlock_page(page); | |
1252 | return 0; | |
1253 | } | |
1254 | ||
1255 | /* | |
1256 | * helper function to end page writeback if all the extents | |
1257 | * in the tree for that page are done with writeback | |
1258 | */ | |
1259 | static int check_page_writeback(struct extent_map_tree *tree, | |
1260 | struct page *page) | |
1261 | { | |
1262 | u64 start = page->index << PAGE_CACHE_SHIFT; | |
1263 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
1264 | if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0)) | |
1265 | end_page_writeback(page); | |
1266 | return 0; | |
1267 | } | |
1268 | ||
1269 | /* lots and lots of room for performance fixes in the end_bio funcs */ | |
1270 | ||
1271 | /* | |
1272 | * after a writepage IO is done, we need to: | |
1273 | * clear the uptodate bits on error | |
1274 | * clear the writeback bits in the extent tree for this IO | |
1275 | * end_page_writeback if the page has no more pending IO | |
1276 | * | |
1277 | * Scheduling is not allowed, so the extent state tree is expected | |
1278 | * to have one and only one object corresponding to this IO. | |
1279 | */ | |
1280 | static int end_bio_extent_writepage(struct bio *bio, | |
1281 | unsigned int bytes_done, int err) | |
1282 | { | |
1283 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
1284 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
1285 | struct extent_map_tree *tree = bio->bi_private; | |
1286 | u64 start; | |
1287 | u64 end; | |
1288 | int whole_page; | |
1289 | ||
1290 | if (bio->bi_size) | |
1291 | return 1; | |
1292 | ||
1293 | do { | |
1294 | struct page *page = bvec->bv_page; | |
1295 | start = (page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset; | |
1296 | end = start + bvec->bv_len - 1; | |
1297 | ||
1298 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
1299 | whole_page = 1; | |
1300 | else | |
1301 | whole_page = 0; | |
1302 | ||
1303 | if (--bvec >= bio->bi_io_vec) | |
1304 | prefetchw(&bvec->bv_page->flags); | |
1305 | ||
1306 | if (!uptodate) { | |
1307 | clear_extent_uptodate(tree, start, end, GFP_ATOMIC); | |
1308 | ClearPageUptodate(page); | |
1309 | SetPageError(page); | |
1310 | } | |
1311 | clear_extent_writeback(tree, start, end, GFP_ATOMIC); | |
1312 | ||
1313 | if (whole_page) | |
1314 | end_page_writeback(page); | |
1315 | else | |
1316 | check_page_writeback(tree, page); | |
0e2752a7 CH |
1317 | if (tree->ops && tree->ops->writepage_end_io_hook) |
1318 | tree->ops->writepage_end_io_hook(page, start, end); | |
a52d9a80 CM |
1319 | } while (bvec >= bio->bi_io_vec); |
1320 | ||
1321 | bio_put(bio); | |
1322 | return 0; | |
1323 | } | |
1324 | ||
1325 | /* | |
1326 | * after a readpage IO is done, we need to: | |
1327 | * clear the uptodate bits on error | |
1328 | * set the uptodate bits if things worked | |
1329 | * set the page up to date if all extents in the tree are uptodate | |
1330 | * clear the lock bit in the extent tree | |
1331 | * unlock the page if there are no other extents locked for it | |
1332 | * | |
1333 | * Scheduling is not allowed, so the extent state tree is expected | |
1334 | * to have one and only one object corresponding to this IO. | |
1335 | */ | |
1336 | static int end_bio_extent_readpage(struct bio *bio, | |
1337 | unsigned int bytes_done, int err) | |
1338 | { | |
07157aac | 1339 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); |
a52d9a80 CM |
1340 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
1341 | struct extent_map_tree *tree = bio->bi_private; | |
1342 | u64 start; | |
1343 | u64 end; | |
1344 | int whole_page; | |
07157aac | 1345 | int ret; |
a52d9a80 CM |
1346 | |
1347 | if (bio->bi_size) | |
1348 | return 1; | |
1349 | ||
1350 | do { | |
1351 | struct page *page = bvec->bv_page; | |
1352 | start = (page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset; | |
1353 | end = start + bvec->bv_len - 1; | |
1354 | ||
1355 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
1356 | whole_page = 1; | |
1357 | else | |
1358 | whole_page = 0; | |
1359 | ||
1360 | if (--bvec >= bio->bi_io_vec) | |
1361 | prefetchw(&bvec->bv_page->flags); | |
1362 | ||
07157aac CM |
1363 | if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) { |
1364 | ret = tree->ops->readpage_end_io_hook(page, start, end); | |
1365 | if (ret) | |
1366 | uptodate = 0; | |
1367 | } | |
a52d9a80 CM |
1368 | if (uptodate) { |
1369 | set_extent_uptodate(tree, start, end, GFP_ATOMIC); | |
1370 | if (whole_page) | |
1371 | SetPageUptodate(page); | |
1372 | else | |
1373 | check_page_uptodate(tree, page); | |
1374 | } else { | |
1375 | ClearPageUptodate(page); | |
1376 | SetPageError(page); | |
1377 | } | |
1378 | ||
1379 | unlock_extent(tree, start, end, GFP_ATOMIC); | |
1380 | ||
1381 | if (whole_page) | |
1382 | unlock_page(page); | |
1383 | else | |
1384 | check_page_locked(tree, page); | |
1385 | } while (bvec >= bio->bi_io_vec); | |
1386 | ||
1387 | bio_put(bio); | |
1388 | return 0; | |
1389 | } | |
1390 | ||
1391 | /* | |
1392 | * IO done from prepare_write is pretty simple, we just unlock | |
1393 | * the structs in the extent tree when done, and set the uptodate bits | |
1394 | * as appropriate. | |
1395 | */ | |
1396 | static int end_bio_extent_preparewrite(struct bio *bio, | |
1397 | unsigned int bytes_done, int err) | |
1398 | { | |
1399 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
1400 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
1401 | struct extent_map_tree *tree = bio->bi_private; | |
1402 | u64 start; | |
1403 | u64 end; | |
1404 | ||
1405 | if (bio->bi_size) | |
1406 | return 1; | |
1407 | ||
1408 | do { | |
1409 | struct page *page = bvec->bv_page; | |
1410 | start = (page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset; | |
1411 | end = start + bvec->bv_len - 1; | |
1412 | ||
1413 | if (--bvec >= bio->bi_io_vec) | |
1414 | prefetchw(&bvec->bv_page->flags); | |
1415 | ||
1416 | if (uptodate) { | |
1417 | set_extent_uptodate(tree, start, end, GFP_ATOMIC); | |
1418 | } else { | |
1419 | ClearPageUptodate(page); | |
1420 | SetPageError(page); | |
1421 | } | |
1422 | ||
1423 | unlock_extent(tree, start, end, GFP_ATOMIC); | |
1424 | ||
1425 | } while (bvec >= bio->bi_io_vec); | |
1426 | ||
1427 | bio_put(bio); | |
1428 | return 0; | |
1429 | } | |
1430 | ||
1431 | static int submit_extent_page(int rw, struct extent_map_tree *tree, | |
1432 | struct page *page, sector_t sector, | |
1433 | size_t size, unsigned long offset, | |
1434 | struct block_device *bdev, | |
1435 | bio_end_io_t end_io_func) | |
1436 | { | |
1437 | struct bio *bio; | |
1438 | int ret = 0; | |
1439 | ||
1440 | bio = bio_alloc(GFP_NOIO, 1); | |
1441 | ||
1442 | bio->bi_sector = sector; | |
1443 | bio->bi_bdev = bdev; | |
1444 | bio->bi_io_vec[0].bv_page = page; | |
1445 | bio->bi_io_vec[0].bv_len = size; | |
1446 | bio->bi_io_vec[0].bv_offset = offset; | |
1447 | ||
1448 | bio->bi_vcnt = 1; | |
1449 | bio->bi_idx = 0; | |
1450 | bio->bi_size = size; | |
1451 | ||
1452 | bio->bi_end_io = end_io_func; | |
1453 | bio->bi_private = tree; | |
1454 | ||
1455 | bio_get(bio); | |
1456 | submit_bio(rw, bio); | |
1457 | ||
1458 | if (bio_flagged(bio, BIO_EOPNOTSUPP)) | |
1459 | ret = -EOPNOTSUPP; | |
1460 | ||
1461 | bio_put(bio); | |
1462 | return ret; | |
1463 | } | |
1464 | ||
b3cfa35a CH |
1465 | void set_page_extent_mapped(struct page *page) |
1466 | { | |
1467 | if (!PagePrivate(page)) { | |
1468 | SetPagePrivate(page); | |
1469 | WARN_ON(!page->mapping->a_ops->invalidatepage); | |
1470 | set_page_private(page, 1); | |
1471 | page_cache_get(page); | |
1472 | } | |
1473 | } | |
1474 | ||
a52d9a80 CM |
1475 | /* |
1476 | * basic readpage implementation. Locked extent state structs are inserted | |
1477 | * into the tree that are removed when the IO is done (by the end_io | |
1478 | * handlers) | |
1479 | */ | |
1480 | int extent_read_full_page(struct extent_map_tree *tree, struct page *page, | |
1481 | get_extent_t *get_extent) | |
1482 | { | |
1483 | struct inode *inode = page->mapping->host; | |
1484 | u64 start = page->index << PAGE_CACHE_SHIFT; | |
1485 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
1486 | u64 end; | |
1487 | u64 cur = start; | |
1488 | u64 extent_offset; | |
1489 | u64 last_byte = i_size_read(inode); | |
1490 | u64 block_start; | |
1491 | u64 cur_end; | |
1492 | sector_t sector; | |
1493 | struct extent_map *em; | |
1494 | struct block_device *bdev; | |
1495 | int ret; | |
1496 | int nr = 0; | |
1497 | size_t page_offset = 0; | |
1498 | size_t iosize; | |
1499 | size_t blocksize = inode->i_sb->s_blocksize; | |
1500 | ||
b3cfa35a | 1501 | set_page_extent_mapped(page); |
a52d9a80 CM |
1502 | |
1503 | end = page_end; | |
1504 | lock_extent(tree, start, end, GFP_NOFS); | |
1505 | ||
1506 | while (cur <= end) { | |
1507 | if (cur >= last_byte) { | |
1508 | iosize = PAGE_CACHE_SIZE - page_offset; | |
1509 | zero_user_page(page, page_offset, iosize, KM_USER0); | |
1510 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
1511 | GFP_NOFS); | |
1512 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
1513 | break; | |
1514 | } | |
1515 | em = get_extent(inode, page, page_offset, cur, end, 0); | |
1516 | if (IS_ERR(em) || !em) { | |
1517 | SetPageError(page); | |
1518 | unlock_extent(tree, cur, end, GFP_NOFS); | |
1519 | break; | |
1520 | } | |
1521 | ||
1522 | extent_offset = cur - em->start; | |
1523 | BUG_ON(em->end < cur); | |
1524 | BUG_ON(end < cur); | |
1525 | ||
1526 | iosize = min(em->end - cur, end - cur) + 1; | |
1527 | cur_end = min(em->end, end); | |
1528 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
1529 | sector = (em->block_start + extent_offset) >> 9; | |
1530 | bdev = em->bdev; | |
1531 | block_start = em->block_start; | |
1532 | free_extent_map(em); | |
1533 | em = NULL; | |
1534 | ||
1535 | /* we've found a hole, just zero and go on */ | |
5f39d397 | 1536 | if (block_start == EXTENT_MAP_HOLE) { |
a52d9a80 CM |
1537 | zero_user_page(page, page_offset, iosize, KM_USER0); |
1538 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
1539 | GFP_NOFS); | |
1540 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
1541 | cur = cur + iosize; | |
1542 | page_offset += iosize; | |
1543 | continue; | |
1544 | } | |
1545 | /* the get_extent function already copied into the page */ | |
1546 | if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) { | |
1547 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
1548 | cur = cur + iosize; | |
1549 | page_offset += iosize; | |
1550 | continue; | |
1551 | } | |
1552 | ||
07157aac CM |
1553 | ret = 0; |
1554 | if (tree->ops && tree->ops->readpage_io_hook) { | |
1555 | ret = tree->ops->readpage_io_hook(page, cur, | |
1556 | cur + iosize - 1); | |
1557 | } | |
1558 | if (!ret) { | |
1559 | ret = submit_extent_page(READ, tree, page, | |
1560 | sector, iosize, page_offset, | |
1561 | bdev, end_bio_extent_readpage); | |
1562 | } | |
a52d9a80 CM |
1563 | if (ret) |
1564 | SetPageError(page); | |
1565 | cur = cur + iosize; | |
1566 | page_offset += iosize; | |
1567 | nr++; | |
1568 | } | |
1569 | if (!nr) { | |
1570 | if (!PageError(page)) | |
1571 | SetPageUptodate(page); | |
1572 | unlock_page(page); | |
1573 | } | |
1574 | return 0; | |
1575 | } | |
1576 | EXPORT_SYMBOL(extent_read_full_page); | |
1577 | ||
1578 | /* | |
1579 | * the writepage semantics are similar to regular writepage. extent | |
1580 | * records are inserted to lock ranges in the tree, and as dirty areas | |
1581 | * are found, they are marked writeback. Then the lock bits are removed | |
1582 | * and the end_io handler clears the writeback ranges | |
1583 | */ | |
1584 | int extent_write_full_page(struct extent_map_tree *tree, struct page *page, | |
1585 | get_extent_t *get_extent, | |
1586 | struct writeback_control *wbc) | |
1587 | { | |
1588 | struct inode *inode = page->mapping->host; | |
1589 | u64 start = page->index << PAGE_CACHE_SHIFT; | |
1590 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
1591 | u64 end; | |
1592 | u64 cur = start; | |
1593 | u64 extent_offset; | |
1594 | u64 last_byte = i_size_read(inode); | |
1595 | u64 block_start; | |
1596 | sector_t sector; | |
1597 | struct extent_map *em; | |
1598 | struct block_device *bdev; | |
1599 | int ret; | |
1600 | int nr = 0; | |
1601 | size_t page_offset = 0; | |
1602 | size_t iosize; | |
1603 | size_t blocksize; | |
1604 | loff_t i_size = i_size_read(inode); | |
1605 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
b888db2b CM |
1606 | u64 nr_delalloc; |
1607 | u64 delalloc_end; | |
a52d9a80 | 1608 | |
b888db2b | 1609 | WARN_ON(!PageLocked(page)); |
a52d9a80 CM |
1610 | if (page->index > end_index) { |
1611 | clear_extent_dirty(tree, start, page_end, GFP_NOFS); | |
1612 | unlock_page(page); | |
1613 | return 0; | |
1614 | } | |
1615 | ||
1616 | if (page->index == end_index) { | |
1617 | size_t offset = i_size & (PAGE_CACHE_SIZE - 1); | |
1618 | zero_user_page(page, offset, | |
1619 | PAGE_CACHE_SIZE - offset, KM_USER0); | |
1620 | } | |
1621 | ||
b3cfa35a | 1622 | set_page_extent_mapped(page); |
a52d9a80 | 1623 | |
a52d9a80 | 1624 | lock_extent(tree, start, page_end, GFP_NOFS); |
b888db2b CM |
1625 | nr_delalloc = find_lock_delalloc_range(tree, start, page_end + 1, |
1626 | &delalloc_end, | |
1627 | 128 * 1024 * 1024); | |
1628 | if (nr_delalloc) { | |
07157aac | 1629 | tree->ops->fill_delalloc(inode, start, delalloc_end); |
b888db2b CM |
1630 | if (delalloc_end >= page_end + 1) { |
1631 | clear_extent_bit(tree, page_end + 1, delalloc_end, | |
1632 | EXTENT_LOCKED | EXTENT_DELALLOC, | |
1633 | 1, 0, GFP_NOFS); | |
1634 | } | |
1635 | clear_extent_bit(tree, start, page_end, EXTENT_DELALLOC, | |
1636 | 0, 0, GFP_NOFS); | |
1637 | if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) { | |
1638 | printk("found delalloc bits after clear extent_bit\n"); | |
1639 | } | |
1640 | } else if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) { | |
1641 | printk("found delalloc bits after find_delalloc_range returns 0\n"); | |
1642 | } | |
1643 | ||
1644 | end = page_end; | |
1645 | if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) { | |
1646 | printk("found delalloc bits after lock_extent\n"); | |
1647 | } | |
a52d9a80 CM |
1648 | |
1649 | if (last_byte <= start) { | |
1650 | clear_extent_dirty(tree, start, page_end, GFP_NOFS); | |
1651 | goto done; | |
1652 | } | |
1653 | ||
1654 | set_extent_uptodate(tree, start, page_end, GFP_NOFS); | |
1655 | blocksize = inode->i_sb->s_blocksize; | |
1656 | ||
1657 | while (cur <= end) { | |
1658 | if (cur >= last_byte) { | |
1659 | clear_extent_dirty(tree, cur, page_end, GFP_NOFS); | |
1660 | break; | |
1661 | } | |
b888db2b | 1662 | em = get_extent(inode, page, page_offset, cur, end, 0); |
a52d9a80 CM |
1663 | if (IS_ERR(em) || !em) { |
1664 | SetPageError(page); | |
1665 | break; | |
1666 | } | |
1667 | ||
1668 | extent_offset = cur - em->start; | |
1669 | BUG_ON(em->end < cur); | |
1670 | BUG_ON(end < cur); | |
1671 | iosize = min(em->end - cur, end - cur) + 1; | |
1672 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
1673 | sector = (em->block_start + extent_offset) >> 9; | |
1674 | bdev = em->bdev; | |
1675 | block_start = em->block_start; | |
1676 | free_extent_map(em); | |
1677 | em = NULL; | |
1678 | ||
5f39d397 CM |
1679 | if (block_start == EXTENT_MAP_HOLE || |
1680 | block_start == EXTENT_MAP_INLINE) { | |
a52d9a80 CM |
1681 | clear_extent_dirty(tree, cur, |
1682 | cur + iosize - 1, GFP_NOFS); | |
1683 | cur = cur + iosize; | |
1684 | page_offset += iosize; | |
1685 | continue; | |
1686 | } | |
1687 | ||
1688 | /* leave this out until we have a page_mkwrite call */ | |
1689 | if (0 && !test_range_bit(tree, cur, cur + iosize - 1, | |
1690 | EXTENT_DIRTY, 0)) { | |
1691 | cur = cur + iosize; | |
1692 | page_offset += iosize; | |
1693 | continue; | |
1694 | } | |
1695 | clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS); | |
b06355f0 CH |
1696 | if (tree->ops && tree->ops->writepage_io_hook) { |
1697 | ret = tree->ops->writepage_io_hook(page, cur, | |
1698 | cur + iosize - 1); | |
1699 | } else { | |
1700 | ret = 0; | |
1701 | } | |
a52d9a80 CM |
1702 | if (ret) |
1703 | SetPageError(page); | |
07157aac CM |
1704 | else { |
1705 | set_range_writeback(tree, cur, cur + iosize - 1); | |
1706 | ret = submit_extent_page(WRITE, tree, page, sector, | |
1707 | iosize, page_offset, bdev, | |
1708 | end_bio_extent_writepage); | |
1709 | if (ret) | |
1710 | SetPageError(page); | |
1711 | } | |
a52d9a80 CM |
1712 | cur = cur + iosize; |
1713 | page_offset += iosize; | |
1714 | nr++; | |
1715 | } | |
1716 | done: | |
a52d9a80 CM |
1717 | unlock_extent(tree, start, page_end, GFP_NOFS); |
1718 | unlock_page(page); | |
1719 | return 0; | |
1720 | } | |
1721 | EXPORT_SYMBOL(extent_write_full_page); | |
1722 | ||
1723 | /* | |
1724 | * basic invalidatepage code, this waits on any locked or writeback | |
1725 | * ranges corresponding to the page, and then deletes any extent state | |
1726 | * records from the tree | |
1727 | */ | |
1728 | int extent_invalidatepage(struct extent_map_tree *tree, | |
1729 | struct page *page, unsigned long offset) | |
1730 | { | |
1731 | u64 start = (page->index << PAGE_CACHE_SHIFT); | |
1732 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
1733 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
1734 | ||
1735 | start += (offset + blocksize -1) & ~(blocksize - 1); | |
1736 | if (start > end) | |
1737 | return 0; | |
1738 | ||
1739 | lock_extent(tree, start, end, GFP_NOFS); | |
1740 | wait_on_extent_writeback(tree, start, end); | |
2bf5a725 CM |
1741 | clear_extent_bit(tree, start, end, |
1742 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC, | |
a52d9a80 CM |
1743 | 1, 1, GFP_NOFS); |
1744 | return 0; | |
1745 | } | |
1746 | EXPORT_SYMBOL(extent_invalidatepage); | |
1747 | ||
1748 | /* | |
1749 | * simple commit_write call, set_range_dirty is used to mark both | |
1750 | * the pages and the extent records as dirty | |
1751 | */ | |
1752 | int extent_commit_write(struct extent_map_tree *tree, | |
1753 | struct inode *inode, struct page *page, | |
1754 | unsigned from, unsigned to) | |
1755 | { | |
1756 | loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to; | |
1757 | ||
b3cfa35a | 1758 | set_page_extent_mapped(page); |
a52d9a80 CM |
1759 | set_page_dirty(page); |
1760 | ||
1761 | if (pos > inode->i_size) { | |
1762 | i_size_write(inode, pos); | |
1763 | mark_inode_dirty(inode); | |
1764 | } | |
1765 | return 0; | |
1766 | } | |
1767 | EXPORT_SYMBOL(extent_commit_write); | |
1768 | ||
1769 | int extent_prepare_write(struct extent_map_tree *tree, | |
1770 | struct inode *inode, struct page *page, | |
1771 | unsigned from, unsigned to, get_extent_t *get_extent) | |
1772 | { | |
1773 | u64 page_start = page->index << PAGE_CACHE_SHIFT; | |
1774 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
1775 | u64 block_start; | |
1776 | u64 orig_block_start; | |
1777 | u64 block_end; | |
1778 | u64 cur_end; | |
1779 | struct extent_map *em; | |
1780 | unsigned blocksize = 1 << inode->i_blkbits; | |
1781 | size_t page_offset = 0; | |
1782 | size_t block_off_start; | |
1783 | size_t block_off_end; | |
1784 | int err = 0; | |
1785 | int iocount = 0; | |
1786 | int ret = 0; | |
1787 | int isnew; | |
1788 | ||
b3cfa35a CH |
1789 | set_page_extent_mapped(page); |
1790 | ||
a52d9a80 CM |
1791 | block_start = (page_start + from) & ~((u64)blocksize - 1); |
1792 | block_end = (page_start + to - 1) | (blocksize - 1); | |
1793 | orig_block_start = block_start; | |
1794 | ||
1795 | lock_extent(tree, page_start, page_end, GFP_NOFS); | |
1796 | while(block_start <= block_end) { | |
1797 | em = get_extent(inode, page, page_offset, block_start, | |
1798 | block_end, 1); | |
1799 | if (IS_ERR(em) || !em) { | |
1800 | goto err; | |
1801 | } | |
1802 | cur_end = min(block_end, em->end); | |
1803 | block_off_start = block_start & (PAGE_CACHE_SIZE - 1); | |
1804 | block_off_end = block_off_start + blocksize; | |
1805 | isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS); | |
1806 | ||
1807 | if (!PageUptodate(page) && isnew && | |
1808 | (block_off_end > to || block_off_start < from)) { | |
1809 | void *kaddr; | |
1810 | ||
1811 | kaddr = kmap_atomic(page, KM_USER0); | |
1812 | if (block_off_end > to) | |
1813 | memset(kaddr + to, 0, block_off_end - to); | |
1814 | if (block_off_start < from) | |
1815 | memset(kaddr + block_off_start, 0, | |
1816 | from - block_off_start); | |
1817 | flush_dcache_page(page); | |
1818 | kunmap_atomic(kaddr, KM_USER0); | |
1819 | } | |
1820 | if (!isnew && !PageUptodate(page) && | |
1821 | (block_off_end > to || block_off_start < from) && | |
1822 | !test_range_bit(tree, block_start, cur_end, | |
1823 | EXTENT_UPTODATE, 1)) { | |
1824 | u64 sector; | |
1825 | u64 extent_offset = block_start - em->start; | |
1826 | size_t iosize; | |
1827 | sector = (em->block_start + extent_offset) >> 9; | |
1828 | iosize = (cur_end - block_start + blocksize - 1) & | |
1829 | ~((u64)blocksize - 1); | |
1830 | /* | |
1831 | * we've already got the extent locked, but we | |
1832 | * need to split the state such that our end_bio | |
1833 | * handler can clear the lock. | |
1834 | */ | |
1835 | set_extent_bit(tree, block_start, | |
1836 | block_start + iosize - 1, | |
1837 | EXTENT_LOCKED, 0, NULL, GFP_NOFS); | |
1838 | ret = submit_extent_page(READ, tree, page, | |
1839 | sector, iosize, page_offset, em->bdev, | |
1840 | end_bio_extent_preparewrite); | |
1841 | iocount++; | |
1842 | block_start = block_start + iosize; | |
1843 | } else { | |
1844 | set_extent_uptodate(tree, block_start, cur_end, | |
1845 | GFP_NOFS); | |
1846 | unlock_extent(tree, block_start, cur_end, GFP_NOFS); | |
1847 | block_start = cur_end + 1; | |
1848 | } | |
1849 | page_offset = block_start & (PAGE_CACHE_SIZE - 1); | |
1850 | free_extent_map(em); | |
1851 | } | |
1852 | if (iocount) { | |
1853 | wait_extent_bit(tree, orig_block_start, | |
1854 | block_end, EXTENT_LOCKED); | |
1855 | } | |
1856 | check_page_uptodate(tree, page); | |
1857 | err: | |
1858 | /* FIXME, zero out newly allocated blocks on error */ | |
1859 | return err; | |
1860 | } | |
1861 | EXPORT_SYMBOL(extent_prepare_write); | |
1862 | ||
1863 | /* | |
1864 | * a helper for releasepage. As long as there are no locked extents | |
1865 | * in the range corresponding to the page, both state records and extent | |
1866 | * map records are removed | |
1867 | */ | |
1868 | int try_release_extent_mapping(struct extent_map_tree *tree, struct page *page) | |
1869 | { | |
1870 | struct extent_map *em; | |
1871 | u64 start = page->index << PAGE_CACHE_SHIFT; | |
1872 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
1873 | u64 orig_start = start; | |
b888db2b | 1874 | int ret = 1; |
a52d9a80 CM |
1875 | |
1876 | while (start <= end) { | |
1877 | em = lookup_extent_mapping(tree, start, end); | |
1878 | if (!em || IS_ERR(em)) | |
1879 | break; | |
b888db2b CM |
1880 | if (!test_range_bit(tree, em->start, em->end, |
1881 | EXTENT_LOCKED, 0)) { | |
1882 | remove_extent_mapping(tree, em); | |
1883 | /* once for the rb tree */ | |
a52d9a80 | 1884 | free_extent_map(em); |
a52d9a80 | 1885 | } |
a52d9a80 | 1886 | start = em->end + 1; |
a52d9a80 CM |
1887 | /* once for us */ |
1888 | free_extent_map(em); | |
1889 | } | |
b888db2b CM |
1890 | if (test_range_bit(tree, orig_start, end, EXTENT_LOCKED, 0)) |
1891 | ret = 0; | |
1892 | else | |
1893 | clear_extent_bit(tree, orig_start, end, EXTENT_UPTODATE, | |
1894 | 1, 1, GFP_NOFS); | |
1895 | return ret; | |
a52d9a80 CM |
1896 | } |
1897 | EXPORT_SYMBOL(try_release_extent_mapping); | |
1898 | ||
d396c6f5 CH |
1899 | sector_t extent_bmap(struct address_space *mapping, sector_t iblock, |
1900 | get_extent_t *get_extent) | |
1901 | { | |
1902 | struct inode *inode = mapping->host; | |
1903 | u64 start = iblock << inode->i_blkbits; | |
1904 | u64 end = start + (1 << inode->i_blkbits) - 1; | |
1905 | struct extent_map *em; | |
1906 | ||
1907 | em = get_extent(inode, NULL, 0, start, end, 0); | |
1908 | if (!em || IS_ERR(em)) | |
1909 | return 0; | |
1910 | ||
d396c6f5 | 1911 | if (em->block_start == EXTENT_MAP_INLINE || |
5f39d397 CM |
1912 | em->block_start == EXTENT_MAP_HOLE) |
1913 | return 0; | |
d396c6f5 CH |
1914 | |
1915 | return (em->block_start + start - em->start) >> inode->i_blkbits; | |
1916 | } | |
5f39d397 | 1917 | |
6d36dcd4 CM |
1918 | static struct extent_buffer *__alloc_extent_buffer(gfp_t mask) |
1919 | { | |
1920 | struct extent_buffer *eb = NULL; | |
f510cfec | 1921 | |
6d36dcd4 CM |
1922 | spin_lock(&extent_buffers_lock); |
1923 | if (!list_empty(&extent_buffers)) { | |
1924 | eb = list_entry(extent_buffers.next, struct extent_buffer, | |
1925 | list); | |
1926 | list_del(&eb->list); | |
1927 | WARN_ON(nr_extent_buffers == 0); | |
1928 | nr_extent_buffers--; | |
1929 | } | |
1930 | spin_unlock(&extent_buffers_lock); | |
f510cfec | 1931 | |
6d36dcd4 CM |
1932 | if (eb) { |
1933 | memset(eb, 0, sizeof(*eb)); | |
f510cfec CM |
1934 | } else { |
1935 | eb = kmem_cache_zalloc(extent_buffer_cache, mask); | |
6d36dcd4 | 1936 | } |
f510cfec CM |
1937 | spin_lock(&extent_buffers_lock); |
1938 | list_add(&eb->leak_list, &buffers); | |
1939 | spin_unlock(&extent_buffers_lock); | |
1940 | ||
1941 | return eb; | |
6d36dcd4 CM |
1942 | } |
1943 | ||
1944 | static void __free_extent_buffer(struct extent_buffer *eb) | |
1945 | { | |
f510cfec CM |
1946 | |
1947 | spin_lock(&extent_buffers_lock); | |
1948 | list_del_init(&eb->leak_list); | |
1949 | spin_unlock(&extent_buffers_lock); | |
1950 | ||
6d36dcd4 CM |
1951 | if (nr_extent_buffers >= MAX_EXTENT_BUFFER_CACHE) { |
1952 | kmem_cache_free(extent_buffer_cache, eb); | |
1953 | } else { | |
1954 | spin_lock(&extent_buffers_lock); | |
1955 | list_add(&eb->list, &extent_buffers); | |
1956 | nr_extent_buffers++; | |
1957 | spin_unlock(&extent_buffers_lock); | |
1958 | } | |
1959 | } | |
1960 | ||
1961 | static inline struct page *extent_buffer_page(struct extent_buffer *eb, int i) | |
1962 | { | |
1963 | struct page *p; | |
db94535d | 1964 | |
09e71a32 CM |
1965 | if (i < EXTENT_INLINE_PAGES) |
1966 | return eb->pages[i]; | |
6d36dcd4 | 1967 | i += eb->start >> PAGE_CACHE_SHIFT; |
09e71a32 | 1968 | p = find_get_page(eb->pages[0]->mapping, i); |
6d36dcd4 CM |
1969 | page_cache_release(p); |
1970 | return p; | |
1971 | } | |
1972 | ||
db94535d CM |
1973 | static inline unsigned long num_extent_pages(u64 start, u64 len) |
1974 | { | |
1975 | return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) - | |
1976 | (start >> PAGE_CACHE_SHIFT); | |
1977 | } | |
5f39d397 CM |
1978 | struct extent_buffer *alloc_extent_buffer(struct extent_map_tree *tree, |
1979 | u64 start, unsigned long len, | |
1980 | gfp_t mask) | |
1981 | { | |
db94535d | 1982 | unsigned long num_pages = num_extent_pages(start, len); |
5f39d397 CM |
1983 | unsigned long i; |
1984 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1985 | struct extent_buffer *eb; | |
1986 | struct page *p; | |
1987 | struct address_space *mapping = tree->mapping; | |
1988 | int uptodate = 0; | |
1989 | ||
6d36dcd4 | 1990 | eb = __alloc_extent_buffer(mask); |
5f39d397 CM |
1991 | if (!eb || IS_ERR(eb)) |
1992 | return NULL; | |
1993 | ||
db94535d | 1994 | eb->alloc_addr = (unsigned long)__builtin_return_address(0); |
5f39d397 CM |
1995 | eb->start = start; |
1996 | eb->len = len; | |
1997 | atomic_set(&eb->refs, 1); | |
1998 | ||
1999 | for (i = 0; i < num_pages; i++, index++) { | |
2000 | p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM); | |
6d36dcd4 | 2001 | if (!p) { |
db94535d | 2002 | WARN_ON(1); |
6d36dcd4 CM |
2003 | /* make sure the free only frees the pages we've |
2004 | * grabbed a reference on | |
2005 | */ | |
2006 | eb->len = i << PAGE_CACHE_SHIFT; | |
2007 | eb->start &= ~((u64)PAGE_CACHE_SIZE - 1); | |
5f39d397 | 2008 | goto fail; |
6d36dcd4 | 2009 | } |
f510cfec | 2010 | set_page_extent_mapped(p); |
09e71a32 CM |
2011 | if (i < EXTENT_INLINE_PAGES) |
2012 | eb->pages[i] = p; | |
5f39d397 CM |
2013 | if (!PageUptodate(p)) |
2014 | uptodate = 0; | |
2015 | unlock_page(p); | |
2016 | } | |
2017 | if (uptodate) | |
2018 | eb->flags |= EXTENT_UPTODATE; | |
2019 | return eb; | |
2020 | fail: | |
2021 | free_extent_buffer(eb); | |
2022 | return NULL; | |
2023 | } | |
2024 | EXPORT_SYMBOL(alloc_extent_buffer); | |
2025 | ||
2026 | struct extent_buffer *find_extent_buffer(struct extent_map_tree *tree, | |
2027 | u64 start, unsigned long len, | |
2028 | gfp_t mask) | |
2029 | { | |
db94535d | 2030 | unsigned long num_pages = num_extent_pages(start, len); |
5f39d397 CM |
2031 | unsigned long i; |
2032 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
2033 | struct extent_buffer *eb; | |
2034 | struct page *p; | |
2035 | struct address_space *mapping = tree->mapping; | |
14048ed0 | 2036 | int uptodate = 1; |
5f39d397 | 2037 | |
6d36dcd4 | 2038 | eb = __alloc_extent_buffer(mask); |
5f39d397 CM |
2039 | if (!eb || IS_ERR(eb)) |
2040 | return NULL; | |
2041 | ||
db94535d | 2042 | eb->alloc_addr = (unsigned long)__builtin_return_address(0); |
5f39d397 CM |
2043 | eb->start = start; |
2044 | eb->len = len; | |
2045 | atomic_set(&eb->refs, 1); | |
2046 | ||
2047 | for (i = 0; i < num_pages; i++, index++) { | |
14048ed0 | 2048 | p = find_lock_page(mapping, index); |
6d36dcd4 CM |
2049 | if (!p) { |
2050 | /* make sure the free only frees the pages we've | |
2051 | * grabbed a reference on | |
2052 | */ | |
2053 | eb->len = i << PAGE_CACHE_SHIFT; | |
2054 | eb->start &= ~((u64)PAGE_CACHE_SIZE - 1); | |
5f39d397 | 2055 | goto fail; |
6d36dcd4 | 2056 | } |
f510cfec | 2057 | set_page_extent_mapped(p); |
09e71a32 CM |
2058 | if (i < EXTENT_INLINE_PAGES) |
2059 | eb->pages[i] = p; | |
14048ed0 CM |
2060 | if (!PageUptodate(p)) |
2061 | uptodate = 0; | |
2062 | unlock_page(p); | |
5f39d397 | 2063 | } |
14048ed0 CM |
2064 | if (uptodate) |
2065 | eb->flags |= EXTENT_UPTODATE; | |
5f39d397 CM |
2066 | return eb; |
2067 | fail: | |
2068 | free_extent_buffer(eb); | |
2069 | return NULL; | |
2070 | } | |
2071 | EXPORT_SYMBOL(find_extent_buffer); | |
2072 | ||
2073 | void free_extent_buffer(struct extent_buffer *eb) | |
2074 | { | |
2075 | unsigned long i; | |
2076 | unsigned long num_pages; | |
2077 | ||
2078 | if (!eb) | |
2079 | return; | |
2080 | ||
2081 | if (!atomic_dec_and_test(&eb->refs)) | |
2082 | return; | |
2083 | ||
db94535d | 2084 | num_pages = num_extent_pages(eb->start, eb->len); |
5f39d397 | 2085 | |
09e71a32 | 2086 | for (i = 0; i < num_pages; i++) { |
6d36dcd4 | 2087 | page_cache_release(extent_buffer_page(eb, i)); |
5f39d397 | 2088 | } |
6d36dcd4 | 2089 | __free_extent_buffer(eb); |
5f39d397 CM |
2090 | } |
2091 | EXPORT_SYMBOL(free_extent_buffer); | |
2092 | ||
2093 | int clear_extent_buffer_dirty(struct extent_map_tree *tree, | |
2094 | struct extent_buffer *eb) | |
2095 | { | |
2096 | int set; | |
2097 | unsigned long i; | |
2098 | unsigned long num_pages; | |
2099 | struct page *page; | |
2100 | ||
2101 | u64 start = eb->start; | |
2102 | u64 end = start + eb->len - 1; | |
2103 | ||
2104 | set = clear_extent_dirty(tree, start, end, GFP_NOFS); | |
db94535d | 2105 | num_pages = num_extent_pages(eb->start, eb->len); |
5f39d397 CM |
2106 | |
2107 | for (i = 0; i < num_pages; i++) { | |
6d36dcd4 | 2108 | page = extent_buffer_page(eb, i); |
5f39d397 CM |
2109 | lock_page(page); |
2110 | /* | |
2111 | * if we're on the last page or the first page and the | |
2112 | * block isn't aligned on a page boundary, do extra checks | |
2113 | * to make sure we don't clean page that is partially dirty | |
2114 | */ | |
2115 | if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) || | |
2116 | ((i == num_pages - 1) && | |
2117 | ((eb->start + eb->len - 1) & (PAGE_CACHE_SIZE - 1)))) { | |
2118 | start = page->index << PAGE_CACHE_SHIFT; | |
2119 | end = start + PAGE_CACHE_SIZE - 1; | |
2120 | if (test_range_bit(tree, start, end, | |
2121 | EXTENT_DIRTY, 0)) { | |
2122 | unlock_page(page); | |
2123 | continue; | |
2124 | } | |
2125 | } | |
2126 | clear_page_dirty_for_io(page); | |
2127 | unlock_page(page); | |
2128 | } | |
2129 | return 0; | |
2130 | } | |
2131 | EXPORT_SYMBOL(clear_extent_buffer_dirty); | |
2132 | ||
2133 | int wait_on_extent_buffer_writeback(struct extent_map_tree *tree, | |
2134 | struct extent_buffer *eb) | |
2135 | { | |
2136 | return wait_on_extent_writeback(tree, eb->start, | |
2137 | eb->start + eb->len - 1); | |
2138 | } | |
2139 | EXPORT_SYMBOL(wait_on_extent_buffer_writeback); | |
2140 | ||
2141 | int set_extent_buffer_dirty(struct extent_map_tree *tree, | |
2142 | struct extent_buffer *eb) | |
2143 | { | |
2144 | return set_range_dirty(tree, eb->start, eb->start + eb->len - 1); | |
2145 | } | |
2146 | EXPORT_SYMBOL(set_extent_buffer_dirty); | |
2147 | ||
2148 | int set_extent_buffer_uptodate(struct extent_map_tree *tree, | |
2149 | struct extent_buffer *eb) | |
2150 | { | |
2151 | unsigned long i; | |
2152 | struct page *page; | |
2153 | unsigned long num_pages; | |
2154 | ||
db94535d | 2155 | num_pages = num_extent_pages(eb->start, eb->len); |
5f39d397 CM |
2156 | |
2157 | set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1, | |
2158 | GFP_NOFS); | |
2159 | for (i = 0; i < num_pages; i++) { | |
6d36dcd4 | 2160 | page = extent_buffer_page(eb, i); |
5f39d397 CM |
2161 | if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) || |
2162 | ((i == num_pages - 1) && | |
2163 | ((eb->start + eb->len - 1) & (PAGE_CACHE_SIZE - 1)))) { | |
2164 | check_page_uptodate(tree, page); | |
2165 | continue; | |
2166 | } | |
2167 | SetPageUptodate(page); | |
2168 | } | |
2169 | return 0; | |
2170 | } | |
2171 | EXPORT_SYMBOL(set_extent_buffer_uptodate); | |
2172 | ||
2173 | int extent_buffer_uptodate(struct extent_map_tree *tree, | |
2174 | struct extent_buffer *eb) | |
2175 | { | |
2176 | if (eb->flags & EXTENT_UPTODATE) | |
2177 | return 1; | |
2178 | return test_range_bit(tree, eb->start, eb->start + eb->len - 1, | |
2179 | EXTENT_UPTODATE, 1); | |
2180 | } | |
2181 | EXPORT_SYMBOL(extent_buffer_uptodate); | |
2182 | ||
2183 | int read_extent_buffer_pages(struct extent_map_tree *tree, | |
2184 | struct extent_buffer *eb, int wait) | |
2185 | { | |
2186 | unsigned long i; | |
2187 | struct page *page; | |
2188 | int err; | |
2189 | int ret = 0; | |
2190 | unsigned long num_pages; | |
2191 | ||
2192 | if (eb->flags & EXTENT_UPTODATE) | |
2193 | return 0; | |
2194 | ||
14048ed0 | 2195 | if (0 && test_range_bit(tree, eb->start, eb->start + eb->len - 1, |
5f39d397 CM |
2196 | EXTENT_UPTODATE, 1)) { |
2197 | return 0; | |
2198 | } | |
2199 | ||
db94535d | 2200 | num_pages = num_extent_pages(eb->start, eb->len); |
5f39d397 | 2201 | for (i = 0; i < num_pages; i++) { |
6d36dcd4 | 2202 | page = extent_buffer_page(eb, i); |
5f39d397 CM |
2203 | if (PageUptodate(page)) { |
2204 | continue; | |
2205 | } | |
2206 | if (!wait) { | |
2207 | if (TestSetPageLocked(page)) { | |
2208 | continue; | |
2209 | } | |
2210 | } else { | |
2211 | lock_page(page); | |
2212 | } | |
2213 | if (!PageUptodate(page)) { | |
2214 | err = page->mapping->a_ops->readpage(NULL, page); | |
2215 | if (err) { | |
2216 | ret = err; | |
2217 | } | |
2218 | } else { | |
2219 | unlock_page(page); | |
2220 | } | |
2221 | } | |
2222 | ||
2223 | if (ret || !wait) { | |
2224 | return ret; | |
2225 | } | |
2226 | ||
2227 | for (i = 0; i < num_pages; i++) { | |
6d36dcd4 | 2228 | page = extent_buffer_page(eb, i); |
5f39d397 CM |
2229 | wait_on_page_locked(page); |
2230 | if (!PageUptodate(page)) { | |
2231 | ret = -EIO; | |
2232 | } | |
2233 | } | |
2234 | eb->flags |= EXTENT_UPTODATE; | |
2235 | return ret; | |
2236 | } | |
2237 | EXPORT_SYMBOL(read_extent_buffer_pages); | |
2238 | ||
2239 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
2240 | unsigned long start, | |
2241 | unsigned long len) | |
2242 | { | |
2243 | size_t cur; | |
2244 | size_t offset; | |
2245 | struct page *page; | |
2246 | char *kaddr; | |
2247 | char *dst = (char *)dstv; | |
2248 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
2249 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
14048ed0 | 2250 | unsigned long num_pages = num_extent_pages(eb->start, eb->len); |
5f39d397 CM |
2251 | |
2252 | WARN_ON(start > eb->len); | |
2253 | WARN_ON(start + len > eb->start + eb->len); | |
2254 | ||
5f39d397 CM |
2255 | offset = start & ((unsigned long)PAGE_CACHE_SIZE - 1); |
2256 | if (i == 0) | |
2257 | offset += start_offset; | |
2258 | ||
2259 | while(len > 0) { | |
6d36dcd4 | 2260 | page = extent_buffer_page(eb, i); |
14048ed0 CM |
2261 | if (!PageUptodate(page)) { |
2262 | printk("page %lu not up to date i %lu, total %lu, len %lu\n", page->index, i, num_pages, eb->len); | |
2263 | WARN_ON(1); | |
2264 | } | |
5f39d397 CM |
2265 | WARN_ON(!PageUptodate(page)); |
2266 | ||
2267 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
ae5252bd | 2268 | kaddr = kmap_atomic(page, KM_USER0); |
5f39d397 | 2269 | memcpy(dst, kaddr + offset, cur); |
ae5252bd | 2270 | kunmap_atomic(kaddr, KM_USER0); |
5f39d397 CM |
2271 | |
2272 | dst += cur; | |
2273 | len -= cur; | |
2274 | offset = 0; | |
2275 | i++; | |
5f39d397 CM |
2276 | } |
2277 | } | |
2278 | EXPORT_SYMBOL(read_extent_buffer); | |
2279 | ||
db94535d CM |
2280 | static int __map_extent_buffer(struct extent_buffer *eb, unsigned long start, |
2281 | unsigned long min_len, char **token, char **map, | |
2282 | unsigned long *map_start, | |
2283 | unsigned long *map_len, int km) | |
5f39d397 | 2284 | { |
479965d6 | 2285 | size_t offset = start & (PAGE_CACHE_SIZE - 1); |
5f39d397 | 2286 | char *kaddr; |
db94535d | 2287 | struct page *p; |
5f39d397 CM |
2288 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); |
2289 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
479965d6 CM |
2290 | unsigned long end_i = (start_offset + start + min_len) >> |
2291 | PAGE_CACHE_SHIFT; | |
2292 | ||
2293 | if (i != end_i) | |
2294 | return -EINVAL; | |
5f39d397 | 2295 | |
db94535d CM |
2296 | if (start >= eb->len) { |
2297 | printk("bad start in map eb start %Lu len %lu caller start %lu min %lu\n", eb->start, eb->len, start, min_len); | |
2298 | WARN_ON(1); | |
2299 | } | |
5f39d397 CM |
2300 | |
2301 | if (i == 0) { | |
2302 | offset = start_offset; | |
2303 | *map_start = 0; | |
2304 | } else { | |
db94535d | 2305 | offset = 0; |
479965d6 | 2306 | *map_start = (i << PAGE_CACHE_SHIFT) - start_offset; |
5f39d397 CM |
2307 | } |
2308 | ||
db94535d CM |
2309 | p = extent_buffer_page(eb, i); |
2310 | WARN_ON(!PageUptodate(p)); | |
2311 | kaddr = kmap_atomic(p, km); | |
5f39d397 CM |
2312 | *token = kaddr; |
2313 | *map = kaddr + offset; | |
2314 | *map_len = PAGE_CACHE_SIZE - offset; | |
2315 | return 0; | |
2316 | } | |
db94535d CM |
2317 | |
2318 | int map_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
2319 | unsigned long min_len, | |
2320 | char **token, char **map, | |
2321 | unsigned long *map_start, | |
2322 | unsigned long *map_len, int km) | |
2323 | { | |
2324 | int err; | |
2325 | int save = 0; | |
2326 | if (eb->map_token) { | |
2327 | if (start >= eb->map_start && | |
2328 | start + min_len <= eb->map_start + eb->map_len) { | |
2329 | *token = eb->map_token; | |
2330 | *map = eb->kaddr; | |
2331 | *map_start = eb->map_start; | |
2332 | *map_len = eb->map_len; | |
2333 | return 0; | |
2334 | } | |
2335 | unmap_extent_buffer(eb, eb->map_token, km); | |
2336 | eb->map_token = NULL; | |
2337 | save = 1; | |
2338 | } | |
2339 | err = __map_extent_buffer(eb, start, min_len, token, map, | |
2340 | map_start, map_len, km); | |
2341 | if (!err && save) { | |
2342 | eb->map_token = *token; | |
2343 | eb->kaddr = *map; | |
2344 | eb->map_start = *map_start; | |
2345 | eb->map_len = *map_len; | |
2346 | } | |
2347 | return err; | |
2348 | } | |
5f39d397 CM |
2349 | EXPORT_SYMBOL(map_extent_buffer); |
2350 | ||
2351 | void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km) | |
2352 | { | |
ae5252bd | 2353 | kunmap_atomic(token, km); |
5f39d397 CM |
2354 | } |
2355 | EXPORT_SYMBOL(unmap_extent_buffer); | |
2356 | ||
2357 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, | |
2358 | unsigned long start, | |
2359 | unsigned long len) | |
2360 | { | |
2361 | size_t cur; | |
2362 | size_t offset; | |
2363 | struct page *page; | |
2364 | char *kaddr; | |
2365 | char *ptr = (char *)ptrv; | |
2366 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
2367 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
2368 | int ret = 0; | |
2369 | ||
2370 | WARN_ON(start > eb->len); | |
2371 | WARN_ON(start + len > eb->start + eb->len); | |
2372 | ||
5f39d397 CM |
2373 | offset = start & ((unsigned long)PAGE_CACHE_SIZE - 1); |
2374 | if (i == 0) | |
2375 | offset += start_offset; | |
2376 | ||
2377 | while(len > 0) { | |
6d36dcd4 | 2378 | page = extent_buffer_page(eb, i); |
5f39d397 CM |
2379 | WARN_ON(!PageUptodate(page)); |
2380 | ||
2381 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
2382 | ||
ae5252bd | 2383 | kaddr = kmap_atomic(page, KM_USER0); |
5f39d397 | 2384 | ret = memcmp(ptr, kaddr + offset, cur); |
ae5252bd | 2385 | kunmap_atomic(kaddr, KM_USER0); |
5f39d397 CM |
2386 | if (ret) |
2387 | break; | |
2388 | ||
2389 | ptr += cur; | |
2390 | len -= cur; | |
2391 | offset = 0; | |
2392 | i++; | |
5f39d397 CM |
2393 | } |
2394 | return ret; | |
2395 | } | |
2396 | EXPORT_SYMBOL(memcmp_extent_buffer); | |
2397 | ||
2398 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
2399 | unsigned long start, unsigned long len) | |
2400 | { | |
2401 | size_t cur; | |
2402 | size_t offset; | |
2403 | struct page *page; | |
2404 | char *kaddr; | |
2405 | char *src = (char *)srcv; | |
2406 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
2407 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
2408 | ||
2409 | WARN_ON(start > eb->len); | |
2410 | WARN_ON(start + len > eb->start + eb->len); | |
2411 | ||
5f39d397 CM |
2412 | offset = start & ((unsigned long)PAGE_CACHE_SIZE - 1); |
2413 | if (i == 0) | |
2414 | offset += start_offset; | |
2415 | ||
2416 | while(len > 0) { | |
6d36dcd4 | 2417 | page = extent_buffer_page(eb, i); |
5f39d397 CM |
2418 | WARN_ON(!PageUptodate(page)); |
2419 | ||
2420 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
ae5252bd | 2421 | kaddr = kmap_atomic(page, KM_USER0); |
5f39d397 | 2422 | memcpy(kaddr + offset, src, cur); |
ae5252bd | 2423 | kunmap_atomic(kaddr, KM_USER0); |
5f39d397 CM |
2424 | |
2425 | src += cur; | |
2426 | len -= cur; | |
2427 | offset = 0; | |
2428 | i++; | |
5f39d397 CM |
2429 | } |
2430 | } | |
2431 | EXPORT_SYMBOL(write_extent_buffer); | |
2432 | ||
2433 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
2434 | unsigned long start, unsigned long len) | |
2435 | { | |
2436 | size_t cur; | |
2437 | size_t offset; | |
2438 | struct page *page; | |
2439 | char *kaddr; | |
2440 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
2441 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
2442 | ||
2443 | WARN_ON(start > eb->len); | |
2444 | WARN_ON(start + len > eb->start + eb->len); | |
2445 | ||
5f39d397 CM |
2446 | offset = start & ((unsigned long)PAGE_CACHE_SIZE - 1); |
2447 | if (i == 0) | |
2448 | offset += start_offset; | |
2449 | ||
2450 | while(len > 0) { | |
6d36dcd4 | 2451 | page = extent_buffer_page(eb, i); |
5f39d397 CM |
2452 | WARN_ON(!PageUptodate(page)); |
2453 | ||
2454 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
ae5252bd | 2455 | kaddr = kmap_atomic(page, KM_USER0); |
5f39d397 | 2456 | memset(kaddr + offset, c, cur); |
ae5252bd | 2457 | kunmap_atomic(kaddr, KM_USER0); |
5f39d397 CM |
2458 | |
2459 | len -= cur; | |
2460 | offset = 0; | |
2461 | i++; | |
5f39d397 CM |
2462 | } |
2463 | } | |
2464 | EXPORT_SYMBOL(memset_extent_buffer); | |
2465 | ||
2466 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
2467 | unsigned long dst_offset, unsigned long src_offset, | |
2468 | unsigned long len) | |
2469 | { | |
2470 | u64 dst_len = dst->len; | |
2471 | size_t cur; | |
2472 | size_t offset; | |
2473 | struct page *page; | |
2474 | char *kaddr; | |
2475 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
2476 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
2477 | ||
2478 | WARN_ON(src->len != dst_len); | |
2479 | ||
2480 | offset = dst_offset & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
2481 | if (i == 0) | |
2482 | offset += start_offset; | |
2483 | ||
2484 | while(len > 0) { | |
6d36dcd4 | 2485 | page = extent_buffer_page(dst, i); |
5f39d397 CM |
2486 | WARN_ON(!PageUptodate(page)); |
2487 | ||
2488 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
2489 | ||
ae5252bd | 2490 | kaddr = kmap_atomic(page, KM_USER1); |
5f39d397 | 2491 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
ae5252bd | 2492 | kunmap_atomic(kaddr, KM_USER1); |
5f39d397 CM |
2493 | |
2494 | src_offset += cur; | |
2495 | len -= cur; | |
2496 | offset = 0; | |
2497 | i++; | |
2498 | } | |
2499 | } | |
2500 | EXPORT_SYMBOL(copy_extent_buffer); | |
2501 | ||
2502 | static void move_pages(struct page *dst_page, struct page *src_page, | |
2503 | unsigned long dst_off, unsigned long src_off, | |
2504 | unsigned long len) | |
2505 | { | |
ae5252bd | 2506 | char *dst_kaddr = kmap_atomic(dst_page, KM_USER0); |
5f39d397 CM |
2507 | if (dst_page == src_page) { |
2508 | memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len); | |
2509 | } else { | |
ae5252bd | 2510 | char *src_kaddr = kmap_atomic(src_page, KM_USER1); |
5f39d397 CM |
2511 | char *p = dst_kaddr + dst_off + len; |
2512 | char *s = src_kaddr + src_off + len; | |
2513 | ||
2514 | while (len--) | |
2515 | *--p = *--s; | |
2516 | ||
ae5252bd | 2517 | kunmap_atomic(src_kaddr, KM_USER1); |
5f39d397 | 2518 | } |
ae5252bd | 2519 | kunmap_atomic(dst_kaddr, KM_USER0); |
5f39d397 CM |
2520 | } |
2521 | ||
2522 | static void copy_pages(struct page *dst_page, struct page *src_page, | |
2523 | unsigned long dst_off, unsigned long src_off, | |
2524 | unsigned long len) | |
2525 | { | |
ae5252bd | 2526 | char *dst_kaddr = kmap_atomic(dst_page, KM_USER0); |
5f39d397 CM |
2527 | char *src_kaddr; |
2528 | ||
2529 | if (dst_page != src_page) | |
ae5252bd | 2530 | src_kaddr = kmap_atomic(src_page, KM_USER1); |
5f39d397 CM |
2531 | else |
2532 | src_kaddr = dst_kaddr; | |
2533 | ||
2534 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
5f39d397 CM |
2535 | kunmap_atomic(dst_kaddr, KM_USER0); |
2536 | if (dst_page != src_page) | |
2537 | kunmap_atomic(src_kaddr, KM_USER1); | |
5f39d397 CM |
2538 | } |
2539 | ||
2540 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
2541 | unsigned long src_offset, unsigned long len) | |
2542 | { | |
2543 | size_t cur; | |
2544 | size_t dst_off_in_page; | |
2545 | size_t src_off_in_page; | |
2546 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
2547 | unsigned long dst_i; | |
2548 | unsigned long src_i; | |
2549 | ||
2550 | if (src_offset + len > dst->len) { | |
2551 | printk("memmove bogus src_offset %lu move len %lu len %lu\n", | |
2552 | src_offset, len, dst->len); | |
2553 | BUG_ON(1); | |
2554 | } | |
2555 | if (dst_offset + len > dst->len) { | |
2556 | printk("memmove bogus dst_offset %lu move len %lu len %lu\n", | |
2557 | dst_offset, len, dst->len); | |
2558 | BUG_ON(1); | |
2559 | } | |
2560 | ||
2561 | while(len > 0) { | |
2562 | dst_off_in_page = dst_offset & | |
2563 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
2564 | src_off_in_page = src_offset & | |
2565 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
2566 | ||
2567 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
2568 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
2569 | ||
2570 | if (src_i == 0) | |
2571 | src_off_in_page += start_offset; | |
2572 | if (dst_i == 0) | |
2573 | dst_off_in_page += start_offset; | |
2574 | ||
2575 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
2576 | src_off_in_page)); | |
2577 | cur = min(cur, (unsigned long)(PAGE_CACHE_SIZE - | |
2578 | dst_off_in_page)); | |
2579 | ||
6d36dcd4 CM |
2580 | copy_pages(extent_buffer_page(dst, dst_i), |
2581 | extent_buffer_page(dst, src_i), | |
5f39d397 CM |
2582 | dst_off_in_page, src_off_in_page, cur); |
2583 | ||
2584 | src_offset += cur; | |
2585 | dst_offset += cur; | |
2586 | len -= cur; | |
2587 | } | |
2588 | } | |
2589 | EXPORT_SYMBOL(memcpy_extent_buffer); | |
2590 | ||
2591 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
2592 | unsigned long src_offset, unsigned long len) | |
2593 | { | |
2594 | size_t cur; | |
2595 | size_t dst_off_in_page; | |
2596 | size_t src_off_in_page; | |
2597 | unsigned long dst_end = dst_offset + len - 1; | |
2598 | unsigned long src_end = src_offset + len - 1; | |
2599 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
2600 | unsigned long dst_i; | |
2601 | unsigned long src_i; | |
2602 | ||
2603 | if (src_offset + len > dst->len) { | |
2604 | printk("memmove bogus src_offset %lu move len %lu len %lu\n", | |
2605 | src_offset, len, dst->len); | |
2606 | BUG_ON(1); | |
2607 | } | |
2608 | if (dst_offset + len > dst->len) { | |
2609 | printk("memmove bogus dst_offset %lu move len %lu len %lu\n", | |
2610 | dst_offset, len, dst->len); | |
2611 | BUG_ON(1); | |
2612 | } | |
2613 | if (dst_offset < src_offset) { | |
2614 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); | |
2615 | return; | |
2616 | } | |
2617 | while(len > 0) { | |
2618 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; | |
2619 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
2620 | ||
2621 | dst_off_in_page = dst_end & | |
2622 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
2623 | src_off_in_page = src_end & | |
2624 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
5f39d397 CM |
2625 | if (src_i == 0) |
2626 | src_off_in_page += start_offset; | |
2627 | if (dst_i == 0) | |
2628 | dst_off_in_page += start_offset; | |
2629 | ||
2630 | cur = min(len, src_off_in_page + 1); | |
2631 | cur = min(cur, dst_off_in_page + 1); | |
6d36dcd4 CM |
2632 | move_pages(extent_buffer_page(dst, dst_i), |
2633 | extent_buffer_page(dst, src_i), | |
5f39d397 CM |
2634 | dst_off_in_page - cur + 1, |
2635 | src_off_in_page - cur + 1, cur); | |
2636 | ||
db94535d CM |
2637 | dst_end -= cur; |
2638 | src_end -= cur; | |
5f39d397 CM |
2639 | len -= cur; |
2640 | } | |
2641 | } | |
2642 | EXPORT_SYMBOL(memmove_extent_buffer); |