]>
Commit | Line | Data |
---|---|---|
d1310b2e CM |
1 | #include <linux/bitops.h> |
2 | #include <linux/slab.h> | |
3 | #include <linux/bio.h> | |
4 | #include <linux/mm.h> | |
5 | #include <linux/gfp.h> | |
6 | #include <linux/pagemap.h> | |
7 | #include <linux/page-flags.h> | |
8 | #include <linux/module.h> | |
9 | #include <linux/spinlock.h> | |
10 | #include <linux/blkdev.h> | |
11 | #include <linux/swap.h> | |
d1310b2e CM |
12 | #include <linux/writeback.h> |
13 | #include <linux/pagevec.h> | |
14 | #include "extent_io.h" | |
15 | #include "extent_map.h" | |
2db04966 | 16 | #include "compat.h" |
902b22f3 DW |
17 | #include "ctree.h" |
18 | #include "btrfs_inode.h" | |
d1310b2e | 19 | |
d1310b2e CM |
20 | static struct kmem_cache *extent_state_cache; |
21 | static struct kmem_cache *extent_buffer_cache; | |
22 | ||
23 | static LIST_HEAD(buffers); | |
24 | static LIST_HEAD(states); | |
4bef0848 | 25 | |
b47eda86 | 26 | #define LEAK_DEBUG 0 |
3935127c | 27 | #if LEAK_DEBUG |
d397712b | 28 | static DEFINE_SPINLOCK(leak_lock); |
4bef0848 | 29 | #endif |
d1310b2e | 30 | |
d1310b2e CM |
31 | #define BUFFER_LRU_MAX 64 |
32 | ||
33 | struct tree_entry { | |
34 | u64 start; | |
35 | u64 end; | |
d1310b2e CM |
36 | struct rb_node rb_node; |
37 | }; | |
38 | ||
39 | struct extent_page_data { | |
40 | struct bio *bio; | |
41 | struct extent_io_tree *tree; | |
42 | get_extent_t *get_extent; | |
771ed689 CM |
43 | |
44 | /* tells writepage not to lock the state bits for this range | |
45 | * it still does the unlocking | |
46 | */ | |
ffbd517d CM |
47 | unsigned int extent_locked:1; |
48 | ||
49 | /* tells the submit_bio code to use a WRITE_SYNC */ | |
50 | unsigned int sync_io:1; | |
d1310b2e CM |
51 | }; |
52 | ||
53 | int __init extent_io_init(void) | |
54 | { | |
9601e3f6 CH |
55 | extent_state_cache = kmem_cache_create("extent_state", |
56 | sizeof(struct extent_state), 0, | |
57 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
58 | if (!extent_state_cache) |
59 | return -ENOMEM; | |
60 | ||
9601e3f6 CH |
61 | extent_buffer_cache = kmem_cache_create("extent_buffers", |
62 | sizeof(struct extent_buffer), 0, | |
63 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
64 | if (!extent_buffer_cache) |
65 | goto free_state_cache; | |
66 | return 0; | |
67 | ||
68 | free_state_cache: | |
69 | kmem_cache_destroy(extent_state_cache); | |
70 | return -ENOMEM; | |
71 | } | |
72 | ||
73 | void extent_io_exit(void) | |
74 | { | |
75 | struct extent_state *state; | |
2d2ae547 | 76 | struct extent_buffer *eb; |
d1310b2e CM |
77 | |
78 | while (!list_empty(&states)) { | |
2d2ae547 | 79 | state = list_entry(states.next, struct extent_state, leak_list); |
d397712b CM |
80 | printk(KERN_ERR "btrfs state leak: start %llu end %llu " |
81 | "state %lu in tree %p refs %d\n", | |
82 | (unsigned long long)state->start, | |
83 | (unsigned long long)state->end, | |
84 | state->state, state->tree, atomic_read(&state->refs)); | |
2d2ae547 | 85 | list_del(&state->leak_list); |
d1310b2e CM |
86 | kmem_cache_free(extent_state_cache, state); |
87 | ||
88 | } | |
89 | ||
2d2ae547 CM |
90 | while (!list_empty(&buffers)) { |
91 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
d397712b CM |
92 | printk(KERN_ERR "btrfs buffer leak start %llu len %lu " |
93 | "refs %d\n", (unsigned long long)eb->start, | |
94 | eb->len, atomic_read(&eb->refs)); | |
2d2ae547 CM |
95 | list_del(&eb->leak_list); |
96 | kmem_cache_free(extent_buffer_cache, eb); | |
97 | } | |
d1310b2e CM |
98 | if (extent_state_cache) |
99 | kmem_cache_destroy(extent_state_cache); | |
100 | if (extent_buffer_cache) | |
101 | kmem_cache_destroy(extent_buffer_cache); | |
102 | } | |
103 | ||
104 | void extent_io_tree_init(struct extent_io_tree *tree, | |
105 | struct address_space *mapping, gfp_t mask) | |
106 | { | |
6bef4d31 EP |
107 | tree->state = RB_ROOT; |
108 | tree->buffer = RB_ROOT; | |
d1310b2e CM |
109 | tree->ops = NULL; |
110 | tree->dirty_bytes = 0; | |
70dec807 | 111 | spin_lock_init(&tree->lock); |
6af118ce | 112 | spin_lock_init(&tree->buffer_lock); |
d1310b2e | 113 | tree->mapping = mapping; |
d1310b2e | 114 | } |
d1310b2e | 115 | |
b2950863 | 116 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
117 | { |
118 | struct extent_state *state; | |
3935127c | 119 | #if LEAK_DEBUG |
2d2ae547 | 120 | unsigned long flags; |
4bef0848 | 121 | #endif |
d1310b2e CM |
122 | |
123 | state = kmem_cache_alloc(extent_state_cache, mask); | |
2b114d1d | 124 | if (!state) |
d1310b2e CM |
125 | return state; |
126 | state->state = 0; | |
d1310b2e | 127 | state->private = 0; |
70dec807 | 128 | state->tree = NULL; |
3935127c | 129 | #if LEAK_DEBUG |
2d2ae547 CM |
130 | spin_lock_irqsave(&leak_lock, flags); |
131 | list_add(&state->leak_list, &states); | |
132 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 133 | #endif |
d1310b2e CM |
134 | atomic_set(&state->refs, 1); |
135 | init_waitqueue_head(&state->wq); | |
136 | return state; | |
137 | } | |
d1310b2e | 138 | |
b2950863 | 139 | static void free_extent_state(struct extent_state *state) |
d1310b2e | 140 | { |
d1310b2e CM |
141 | if (!state) |
142 | return; | |
143 | if (atomic_dec_and_test(&state->refs)) { | |
3935127c | 144 | #if LEAK_DEBUG |
2d2ae547 | 145 | unsigned long flags; |
4bef0848 | 146 | #endif |
70dec807 | 147 | WARN_ON(state->tree); |
3935127c | 148 | #if LEAK_DEBUG |
2d2ae547 CM |
149 | spin_lock_irqsave(&leak_lock, flags); |
150 | list_del(&state->leak_list); | |
151 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 152 | #endif |
d1310b2e CM |
153 | kmem_cache_free(extent_state_cache, state); |
154 | } | |
155 | } | |
d1310b2e CM |
156 | |
157 | static struct rb_node *tree_insert(struct rb_root *root, u64 offset, | |
158 | struct rb_node *node) | |
159 | { | |
d397712b CM |
160 | struct rb_node **p = &root->rb_node; |
161 | struct rb_node *parent = NULL; | |
d1310b2e CM |
162 | struct tree_entry *entry; |
163 | ||
d397712b | 164 | while (*p) { |
d1310b2e CM |
165 | parent = *p; |
166 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
167 | ||
168 | if (offset < entry->start) | |
169 | p = &(*p)->rb_left; | |
170 | else if (offset > entry->end) | |
171 | p = &(*p)->rb_right; | |
172 | else | |
173 | return parent; | |
174 | } | |
175 | ||
176 | entry = rb_entry(node, struct tree_entry, rb_node); | |
d1310b2e CM |
177 | rb_link_node(node, parent, p); |
178 | rb_insert_color(node, root); | |
179 | return NULL; | |
180 | } | |
181 | ||
80ea96b1 | 182 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
d1310b2e CM |
183 | struct rb_node **prev_ret, |
184 | struct rb_node **next_ret) | |
185 | { | |
80ea96b1 | 186 | struct rb_root *root = &tree->state; |
d397712b | 187 | struct rb_node *n = root->rb_node; |
d1310b2e CM |
188 | struct rb_node *prev = NULL; |
189 | struct rb_node *orig_prev = NULL; | |
190 | struct tree_entry *entry; | |
191 | struct tree_entry *prev_entry = NULL; | |
192 | ||
d397712b | 193 | while (n) { |
d1310b2e CM |
194 | entry = rb_entry(n, struct tree_entry, rb_node); |
195 | prev = n; | |
196 | prev_entry = entry; | |
197 | ||
198 | if (offset < entry->start) | |
199 | n = n->rb_left; | |
200 | else if (offset > entry->end) | |
201 | n = n->rb_right; | |
d397712b | 202 | else |
d1310b2e CM |
203 | return n; |
204 | } | |
205 | ||
206 | if (prev_ret) { | |
207 | orig_prev = prev; | |
d397712b | 208 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
209 | prev = rb_next(prev); |
210 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
211 | } | |
212 | *prev_ret = prev; | |
213 | prev = orig_prev; | |
214 | } | |
215 | ||
216 | if (next_ret) { | |
217 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 218 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
219 | prev = rb_prev(prev); |
220 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
221 | } | |
222 | *next_ret = prev; | |
223 | } | |
224 | return NULL; | |
225 | } | |
226 | ||
80ea96b1 CM |
227 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
228 | u64 offset) | |
d1310b2e | 229 | { |
70dec807 | 230 | struct rb_node *prev = NULL; |
d1310b2e | 231 | struct rb_node *ret; |
70dec807 | 232 | |
80ea96b1 | 233 | ret = __etree_search(tree, offset, &prev, NULL); |
d397712b | 234 | if (!ret) |
d1310b2e CM |
235 | return prev; |
236 | return ret; | |
237 | } | |
238 | ||
6af118ce CM |
239 | static struct extent_buffer *buffer_tree_insert(struct extent_io_tree *tree, |
240 | u64 offset, struct rb_node *node) | |
241 | { | |
242 | struct rb_root *root = &tree->buffer; | |
d397712b CM |
243 | struct rb_node **p = &root->rb_node; |
244 | struct rb_node *parent = NULL; | |
6af118ce CM |
245 | struct extent_buffer *eb; |
246 | ||
d397712b | 247 | while (*p) { |
6af118ce CM |
248 | parent = *p; |
249 | eb = rb_entry(parent, struct extent_buffer, rb_node); | |
250 | ||
251 | if (offset < eb->start) | |
252 | p = &(*p)->rb_left; | |
253 | else if (offset > eb->start) | |
254 | p = &(*p)->rb_right; | |
255 | else | |
256 | return eb; | |
257 | } | |
258 | ||
259 | rb_link_node(node, parent, p); | |
260 | rb_insert_color(node, root); | |
261 | return NULL; | |
262 | } | |
263 | ||
264 | static struct extent_buffer *buffer_search(struct extent_io_tree *tree, | |
265 | u64 offset) | |
266 | { | |
267 | struct rb_root *root = &tree->buffer; | |
d397712b | 268 | struct rb_node *n = root->rb_node; |
6af118ce CM |
269 | struct extent_buffer *eb; |
270 | ||
d397712b | 271 | while (n) { |
6af118ce CM |
272 | eb = rb_entry(n, struct extent_buffer, rb_node); |
273 | if (offset < eb->start) | |
274 | n = n->rb_left; | |
275 | else if (offset > eb->start) | |
276 | n = n->rb_right; | |
277 | else | |
278 | return eb; | |
279 | } | |
280 | return NULL; | |
281 | } | |
282 | ||
9ed74f2d JB |
283 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, |
284 | struct extent_state *other) | |
285 | { | |
286 | if (tree->ops && tree->ops->merge_extent_hook) | |
287 | tree->ops->merge_extent_hook(tree->mapping->host, new, | |
288 | other); | |
289 | } | |
290 | ||
d1310b2e CM |
291 | /* |
292 | * utility function to look for merge candidates inside a given range. | |
293 | * Any extents with matching state are merged together into a single | |
294 | * extent in the tree. Extents with EXTENT_IO in their state field | |
295 | * are not merged because the end_io handlers need to be able to do | |
296 | * operations on them without sleeping (or doing allocations/splits). | |
297 | * | |
298 | * This should be called with the tree lock held. | |
299 | */ | |
300 | static int merge_state(struct extent_io_tree *tree, | |
301 | struct extent_state *state) | |
302 | { | |
303 | struct extent_state *other; | |
304 | struct rb_node *other_node; | |
305 | ||
5b21f2ed | 306 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
d1310b2e CM |
307 | return 0; |
308 | ||
309 | other_node = rb_prev(&state->rb_node); | |
310 | if (other_node) { | |
311 | other = rb_entry(other_node, struct extent_state, rb_node); | |
312 | if (other->end == state->start - 1 && | |
313 | other->state == state->state) { | |
9ed74f2d | 314 | merge_cb(tree, state, other); |
d1310b2e | 315 | state->start = other->start; |
70dec807 | 316 | other->tree = NULL; |
d1310b2e CM |
317 | rb_erase(&other->rb_node, &tree->state); |
318 | free_extent_state(other); | |
319 | } | |
320 | } | |
321 | other_node = rb_next(&state->rb_node); | |
322 | if (other_node) { | |
323 | other = rb_entry(other_node, struct extent_state, rb_node); | |
324 | if (other->start == state->end + 1 && | |
325 | other->state == state->state) { | |
9ed74f2d | 326 | merge_cb(tree, state, other); |
d1310b2e | 327 | other->start = state->start; |
70dec807 | 328 | state->tree = NULL; |
d1310b2e CM |
329 | rb_erase(&state->rb_node, &tree->state); |
330 | free_extent_state(state); | |
9ed74f2d | 331 | state = NULL; |
d1310b2e CM |
332 | } |
333 | } | |
9ed74f2d | 334 | |
d1310b2e CM |
335 | return 0; |
336 | } | |
337 | ||
9ed74f2d | 338 | static int set_state_cb(struct extent_io_tree *tree, |
291d673e CM |
339 | struct extent_state *state, |
340 | unsigned long bits) | |
341 | { | |
342 | if (tree->ops && tree->ops->set_bit_hook) { | |
9ed74f2d JB |
343 | return tree->ops->set_bit_hook(tree->mapping->host, |
344 | state->start, state->end, | |
345 | state->state, bits); | |
291d673e | 346 | } |
9ed74f2d JB |
347 | |
348 | return 0; | |
291d673e CM |
349 | } |
350 | ||
351 | static void clear_state_cb(struct extent_io_tree *tree, | |
352 | struct extent_state *state, | |
353 | unsigned long bits) | |
354 | { | |
9ed74f2d JB |
355 | if (tree->ops && tree->ops->clear_bit_hook) |
356 | tree->ops->clear_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
357 | } |
358 | ||
d1310b2e CM |
359 | /* |
360 | * insert an extent_state struct into the tree. 'bits' are set on the | |
361 | * struct before it is inserted. | |
362 | * | |
363 | * This may return -EEXIST if the extent is already there, in which case the | |
364 | * state struct is freed. | |
365 | * | |
366 | * The tree lock is not taken internally. This is a utility function and | |
367 | * probably isn't what you want to call (see set/clear_extent_bit). | |
368 | */ | |
369 | static int insert_state(struct extent_io_tree *tree, | |
370 | struct extent_state *state, u64 start, u64 end, | |
371 | int bits) | |
372 | { | |
373 | struct rb_node *node; | |
9ed74f2d | 374 | int ret; |
d1310b2e CM |
375 | |
376 | if (end < start) { | |
d397712b CM |
377 | printk(KERN_ERR "btrfs end < start %llu %llu\n", |
378 | (unsigned long long)end, | |
379 | (unsigned long long)start); | |
d1310b2e CM |
380 | WARN_ON(1); |
381 | } | |
d1310b2e CM |
382 | state->start = start; |
383 | state->end = end; | |
9ed74f2d JB |
384 | ret = set_state_cb(tree, state, bits); |
385 | if (ret) | |
386 | return ret; | |
387 | ||
388 | if (bits & EXTENT_DIRTY) | |
389 | tree->dirty_bytes += end - start + 1; | |
e48c465b | 390 | state->state |= bits; |
d1310b2e CM |
391 | node = tree_insert(&tree->state, end, &state->rb_node); |
392 | if (node) { | |
393 | struct extent_state *found; | |
394 | found = rb_entry(node, struct extent_state, rb_node); | |
d397712b CM |
395 | printk(KERN_ERR "btrfs found node %llu %llu on insert of " |
396 | "%llu %llu\n", (unsigned long long)found->start, | |
397 | (unsigned long long)found->end, | |
398 | (unsigned long long)start, (unsigned long long)end); | |
d1310b2e CM |
399 | free_extent_state(state); |
400 | return -EEXIST; | |
401 | } | |
70dec807 | 402 | state->tree = tree; |
d1310b2e CM |
403 | merge_state(tree, state); |
404 | return 0; | |
405 | } | |
406 | ||
9ed74f2d JB |
407 | static int split_cb(struct extent_io_tree *tree, struct extent_state *orig, |
408 | u64 split) | |
409 | { | |
410 | if (tree->ops && tree->ops->split_extent_hook) | |
411 | return tree->ops->split_extent_hook(tree->mapping->host, | |
412 | orig, split); | |
413 | return 0; | |
414 | } | |
415 | ||
d1310b2e CM |
416 | /* |
417 | * split a given extent state struct in two, inserting the preallocated | |
418 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
419 | * offset inside 'orig' where it should be split. | |
420 | * | |
421 | * Before calling, | |
422 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
423 | * are two extent state structs in the tree: | |
424 | * prealloc: [orig->start, split - 1] | |
425 | * orig: [ split, orig->end ] | |
426 | * | |
427 | * The tree locks are not taken by this function. They need to be held | |
428 | * by the caller. | |
429 | */ | |
430 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
431 | struct extent_state *prealloc, u64 split) | |
432 | { | |
433 | struct rb_node *node; | |
9ed74f2d JB |
434 | |
435 | split_cb(tree, orig, split); | |
436 | ||
d1310b2e CM |
437 | prealloc->start = orig->start; |
438 | prealloc->end = split - 1; | |
439 | prealloc->state = orig->state; | |
440 | orig->start = split; | |
441 | ||
442 | node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node); | |
443 | if (node) { | |
d1310b2e CM |
444 | free_extent_state(prealloc); |
445 | return -EEXIST; | |
446 | } | |
70dec807 | 447 | prealloc->tree = tree; |
d1310b2e CM |
448 | return 0; |
449 | } | |
450 | ||
451 | /* | |
452 | * utility function to clear some bits in an extent state struct. | |
453 | * it will optionally wake up any one waiting on this state (wake == 1), or | |
454 | * forcibly remove the state from the tree (delete == 1). | |
455 | * | |
456 | * If no bits are set on the state struct after clearing things, the | |
457 | * struct is freed and removed from the tree | |
458 | */ | |
459 | static int clear_state_bit(struct extent_io_tree *tree, | |
460 | struct extent_state *state, int bits, int wake, | |
461 | int delete) | |
462 | { | |
32c00aff JB |
463 | int bits_to_clear = bits & ~EXTENT_DO_ACCOUNTING; |
464 | int ret = state->state & bits_to_clear; | |
d1310b2e CM |
465 | |
466 | if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { | |
467 | u64 range = state->end - state->start + 1; | |
468 | WARN_ON(range > tree->dirty_bytes); | |
469 | tree->dirty_bytes -= range; | |
470 | } | |
291d673e | 471 | clear_state_cb(tree, state, bits); |
32c00aff | 472 | state->state &= ~bits_to_clear; |
d1310b2e CM |
473 | if (wake) |
474 | wake_up(&state->wq); | |
475 | if (delete || state->state == 0) { | |
70dec807 | 476 | if (state->tree) { |
ae9d1285 | 477 | clear_state_cb(tree, state, state->state); |
d1310b2e | 478 | rb_erase(&state->rb_node, &tree->state); |
70dec807 | 479 | state->tree = NULL; |
d1310b2e CM |
480 | free_extent_state(state); |
481 | } else { | |
482 | WARN_ON(1); | |
483 | } | |
484 | } else { | |
485 | merge_state(tree, state); | |
486 | } | |
487 | return ret; | |
488 | } | |
489 | ||
490 | /* | |
491 | * clear some bits on a range in the tree. This may require splitting | |
492 | * or inserting elements in the tree, so the gfp mask is used to | |
493 | * indicate which allocations or sleeping are allowed. | |
494 | * | |
495 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
496 | * the given range from the tree regardless of state (ie for truncate). | |
497 | * | |
498 | * the range [start, end] is inclusive. | |
499 | * | |
500 | * This takes the tree lock, and returns < 0 on error, > 0 if any of the | |
501 | * bits were already set, or zero if none of the bits were already set. | |
502 | */ | |
503 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
2c64c53d CM |
504 | int bits, int wake, int delete, |
505 | struct extent_state **cached_state, | |
506 | gfp_t mask) | |
d1310b2e CM |
507 | { |
508 | struct extent_state *state; | |
2c64c53d | 509 | struct extent_state *cached; |
d1310b2e | 510 | struct extent_state *prealloc = NULL; |
2c64c53d | 511 | struct rb_node *next_node; |
d1310b2e | 512 | struct rb_node *node; |
5c939df5 | 513 | u64 last_end; |
d1310b2e CM |
514 | int err; |
515 | int set = 0; | |
2ac55d41 | 516 | int clear = 0; |
d1310b2e | 517 | |
2ac55d41 JB |
518 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
519 | clear = 1; | |
d1310b2e CM |
520 | again: |
521 | if (!prealloc && (mask & __GFP_WAIT)) { | |
522 | prealloc = alloc_extent_state(mask); | |
523 | if (!prealloc) | |
524 | return -ENOMEM; | |
525 | } | |
526 | ||
cad321ad | 527 | spin_lock(&tree->lock); |
2c64c53d CM |
528 | if (cached_state) { |
529 | cached = *cached_state; | |
2ac55d41 JB |
530 | |
531 | if (clear) { | |
532 | *cached_state = NULL; | |
533 | cached_state = NULL; | |
534 | } | |
535 | ||
42daec29 | 536 | if (cached && cached->tree && cached->start == start) { |
2ac55d41 JB |
537 | if (clear) |
538 | atomic_dec(&cached->refs); | |
2c64c53d | 539 | state = cached; |
42daec29 | 540 | goto hit_next; |
2c64c53d | 541 | } |
2ac55d41 JB |
542 | if (clear) |
543 | free_extent_state(cached); | |
2c64c53d | 544 | } |
d1310b2e CM |
545 | /* |
546 | * this search will find the extents that end after | |
547 | * our range starts | |
548 | */ | |
80ea96b1 | 549 | node = tree_search(tree, start); |
d1310b2e CM |
550 | if (!node) |
551 | goto out; | |
552 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 553 | hit_next: |
d1310b2e CM |
554 | if (state->start > end) |
555 | goto out; | |
556 | WARN_ON(state->end < start); | |
5c939df5 | 557 | last_end = state->end; |
d1310b2e CM |
558 | |
559 | /* | |
560 | * | ---- desired range ---- | | |
561 | * | state | or | |
562 | * | ------------- state -------------- | | |
563 | * | |
564 | * We need to split the extent we found, and may flip | |
565 | * bits on second half. | |
566 | * | |
567 | * If the extent we found extends past our range, we | |
568 | * just split and search again. It'll get split again | |
569 | * the next time though. | |
570 | * | |
571 | * If the extent we found is inside our range, we clear | |
572 | * the desired bit on it. | |
573 | */ | |
574 | ||
575 | if (state->start < start) { | |
70dec807 CM |
576 | if (!prealloc) |
577 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
d1310b2e CM |
578 | err = split_state(tree, state, prealloc, start); |
579 | BUG_ON(err == -EEXIST); | |
580 | prealloc = NULL; | |
581 | if (err) | |
582 | goto out; | |
583 | if (state->end <= end) { | |
9ed74f2d JB |
584 | set |= clear_state_bit(tree, state, bits, wake, |
585 | delete); | |
5c939df5 YZ |
586 | if (last_end == (u64)-1) |
587 | goto out; | |
588 | start = last_end + 1; | |
d1310b2e CM |
589 | } |
590 | goto search_again; | |
591 | } | |
592 | /* | |
593 | * | ---- desired range ---- | | |
594 | * | state | | |
595 | * We need to split the extent, and clear the bit | |
596 | * on the first half | |
597 | */ | |
598 | if (state->start <= end && state->end > end) { | |
70dec807 CM |
599 | if (!prealloc) |
600 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
d1310b2e CM |
601 | err = split_state(tree, state, prealloc, end + 1); |
602 | BUG_ON(err == -EEXIST); | |
d1310b2e CM |
603 | if (wake) |
604 | wake_up(&state->wq); | |
42daec29 | 605 | |
9ed74f2d JB |
606 | set |= clear_state_bit(tree, prealloc, bits, wake, delete); |
607 | ||
d1310b2e CM |
608 | prealloc = NULL; |
609 | goto out; | |
610 | } | |
42daec29 | 611 | |
2c64c53d CM |
612 | if (state->end < end && prealloc && !need_resched()) |
613 | next_node = rb_next(&state->rb_node); | |
614 | else | |
615 | next_node = NULL; | |
42daec29 | 616 | |
d1310b2e | 617 | set |= clear_state_bit(tree, state, bits, wake, delete); |
5c939df5 YZ |
618 | if (last_end == (u64)-1) |
619 | goto out; | |
620 | start = last_end + 1; | |
2c64c53d CM |
621 | if (start <= end && next_node) { |
622 | state = rb_entry(next_node, struct extent_state, | |
623 | rb_node); | |
624 | if (state->start == start) | |
625 | goto hit_next; | |
626 | } | |
d1310b2e CM |
627 | goto search_again; |
628 | ||
629 | out: | |
cad321ad | 630 | spin_unlock(&tree->lock); |
d1310b2e CM |
631 | if (prealloc) |
632 | free_extent_state(prealloc); | |
633 | ||
634 | return set; | |
635 | ||
636 | search_again: | |
637 | if (start > end) | |
638 | goto out; | |
cad321ad | 639 | spin_unlock(&tree->lock); |
d1310b2e CM |
640 | if (mask & __GFP_WAIT) |
641 | cond_resched(); | |
642 | goto again; | |
643 | } | |
d1310b2e CM |
644 | |
645 | static int wait_on_state(struct extent_io_tree *tree, | |
646 | struct extent_state *state) | |
641f5219 CH |
647 | __releases(tree->lock) |
648 | __acquires(tree->lock) | |
d1310b2e CM |
649 | { |
650 | DEFINE_WAIT(wait); | |
651 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 652 | spin_unlock(&tree->lock); |
d1310b2e | 653 | schedule(); |
cad321ad | 654 | spin_lock(&tree->lock); |
d1310b2e CM |
655 | finish_wait(&state->wq, &wait); |
656 | return 0; | |
657 | } | |
658 | ||
659 | /* | |
660 | * waits for one or more bits to clear on a range in the state tree. | |
661 | * The range [start, end] is inclusive. | |
662 | * The tree lock is taken by this function | |
663 | */ | |
664 | int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits) | |
665 | { | |
666 | struct extent_state *state; | |
667 | struct rb_node *node; | |
668 | ||
cad321ad | 669 | spin_lock(&tree->lock); |
d1310b2e CM |
670 | again: |
671 | while (1) { | |
672 | /* | |
673 | * this search will find all the extents that end after | |
674 | * our range starts | |
675 | */ | |
80ea96b1 | 676 | node = tree_search(tree, start); |
d1310b2e CM |
677 | if (!node) |
678 | break; | |
679 | ||
680 | state = rb_entry(node, struct extent_state, rb_node); | |
681 | ||
682 | if (state->start > end) | |
683 | goto out; | |
684 | ||
685 | if (state->state & bits) { | |
686 | start = state->start; | |
687 | atomic_inc(&state->refs); | |
688 | wait_on_state(tree, state); | |
689 | free_extent_state(state); | |
690 | goto again; | |
691 | } | |
692 | start = state->end + 1; | |
693 | ||
694 | if (start > end) | |
695 | break; | |
696 | ||
697 | if (need_resched()) { | |
cad321ad | 698 | spin_unlock(&tree->lock); |
d1310b2e | 699 | cond_resched(); |
cad321ad | 700 | spin_lock(&tree->lock); |
d1310b2e CM |
701 | } |
702 | } | |
703 | out: | |
cad321ad | 704 | spin_unlock(&tree->lock); |
d1310b2e CM |
705 | return 0; |
706 | } | |
d1310b2e | 707 | |
9ed74f2d | 708 | static int set_state_bits(struct extent_io_tree *tree, |
d1310b2e CM |
709 | struct extent_state *state, |
710 | int bits) | |
711 | { | |
9ed74f2d JB |
712 | int ret; |
713 | ||
714 | ret = set_state_cb(tree, state, bits); | |
715 | if (ret) | |
716 | return ret; | |
717 | ||
d1310b2e CM |
718 | if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
719 | u64 range = state->end - state->start + 1; | |
720 | tree->dirty_bytes += range; | |
721 | } | |
b0c68f8b | 722 | state->state |= bits; |
9ed74f2d JB |
723 | |
724 | return 0; | |
d1310b2e CM |
725 | } |
726 | ||
2c64c53d CM |
727 | static void cache_state(struct extent_state *state, |
728 | struct extent_state **cached_ptr) | |
729 | { | |
730 | if (cached_ptr && !(*cached_ptr)) { | |
731 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) { | |
732 | *cached_ptr = state; | |
733 | atomic_inc(&state->refs); | |
734 | } | |
735 | } | |
736 | } | |
737 | ||
d1310b2e | 738 | /* |
1edbb734 CM |
739 | * set some bits on a range in the tree. This may require allocations or |
740 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 741 | * |
1edbb734 CM |
742 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
743 | * part of the range already has the desired bits set. The start of the | |
744 | * existing range is returned in failed_start in this case. | |
d1310b2e | 745 | * |
1edbb734 | 746 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 747 | */ |
1edbb734 | 748 | |
d397712b | 749 | static int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
1edbb734 | 750 | int bits, int exclusive_bits, u64 *failed_start, |
2c64c53d | 751 | struct extent_state **cached_state, |
d397712b | 752 | gfp_t mask) |
d1310b2e CM |
753 | { |
754 | struct extent_state *state; | |
755 | struct extent_state *prealloc = NULL; | |
756 | struct rb_node *node; | |
d1310b2e | 757 | int err = 0; |
d1310b2e CM |
758 | u64 last_start; |
759 | u64 last_end; | |
42daec29 | 760 | |
d1310b2e CM |
761 | again: |
762 | if (!prealloc && (mask & __GFP_WAIT)) { | |
763 | prealloc = alloc_extent_state(mask); | |
764 | if (!prealloc) | |
765 | return -ENOMEM; | |
766 | } | |
767 | ||
cad321ad | 768 | spin_lock(&tree->lock); |
9655d298 CM |
769 | if (cached_state && *cached_state) { |
770 | state = *cached_state; | |
771 | if (state->start == start && state->tree) { | |
772 | node = &state->rb_node; | |
773 | goto hit_next; | |
774 | } | |
775 | } | |
d1310b2e CM |
776 | /* |
777 | * this search will find all the extents that end after | |
778 | * our range starts. | |
779 | */ | |
80ea96b1 | 780 | node = tree_search(tree, start); |
d1310b2e CM |
781 | if (!node) { |
782 | err = insert_state(tree, prealloc, start, end, bits); | |
783 | prealloc = NULL; | |
784 | BUG_ON(err == -EEXIST); | |
785 | goto out; | |
786 | } | |
d1310b2e | 787 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 788 | hit_next: |
d1310b2e CM |
789 | last_start = state->start; |
790 | last_end = state->end; | |
791 | ||
792 | /* | |
793 | * | ---- desired range ---- | | |
794 | * | state | | |
795 | * | |
796 | * Just lock what we found and keep going | |
797 | */ | |
798 | if (state->start == start && state->end <= end) { | |
40431d6c | 799 | struct rb_node *next_node; |
1edbb734 | 800 | if (state->state & exclusive_bits) { |
d1310b2e CM |
801 | *failed_start = state->start; |
802 | err = -EEXIST; | |
803 | goto out; | |
804 | } | |
42daec29 | 805 | |
9ed74f2d JB |
806 | err = set_state_bits(tree, state, bits); |
807 | if (err) | |
808 | goto out; | |
809 | ||
2c64c53d | 810 | cache_state(state, cached_state); |
d1310b2e | 811 | merge_state(tree, state); |
5c939df5 YZ |
812 | if (last_end == (u64)-1) |
813 | goto out; | |
40431d6c | 814 | |
5c939df5 | 815 | start = last_end + 1; |
40431d6c CM |
816 | if (start < end && prealloc && !need_resched()) { |
817 | next_node = rb_next(node); | |
818 | if (next_node) { | |
819 | state = rb_entry(next_node, struct extent_state, | |
820 | rb_node); | |
821 | if (state->start == start) | |
822 | goto hit_next; | |
823 | } | |
824 | } | |
d1310b2e CM |
825 | goto search_again; |
826 | } | |
827 | ||
828 | /* | |
829 | * | ---- desired range ---- | | |
830 | * | state | | |
831 | * or | |
832 | * | ------------- state -------------- | | |
833 | * | |
834 | * We need to split the extent we found, and may flip bits on | |
835 | * second half. | |
836 | * | |
837 | * If the extent we found extends past our | |
838 | * range, we just split and search again. It'll get split | |
839 | * again the next time though. | |
840 | * | |
841 | * If the extent we found is inside our range, we set the | |
842 | * desired bit on it. | |
843 | */ | |
844 | if (state->start < start) { | |
1edbb734 | 845 | if (state->state & exclusive_bits) { |
d1310b2e CM |
846 | *failed_start = start; |
847 | err = -EEXIST; | |
848 | goto out; | |
849 | } | |
850 | err = split_state(tree, state, prealloc, start); | |
851 | BUG_ON(err == -EEXIST); | |
852 | prealloc = NULL; | |
853 | if (err) | |
854 | goto out; | |
855 | if (state->end <= end) { | |
9ed74f2d JB |
856 | err = set_state_bits(tree, state, bits); |
857 | if (err) | |
858 | goto out; | |
2c64c53d | 859 | cache_state(state, cached_state); |
d1310b2e | 860 | merge_state(tree, state); |
5c939df5 YZ |
861 | if (last_end == (u64)-1) |
862 | goto out; | |
863 | start = last_end + 1; | |
d1310b2e CM |
864 | } |
865 | goto search_again; | |
866 | } | |
867 | /* | |
868 | * | ---- desired range ---- | | |
869 | * | state | or | state | | |
870 | * | |
871 | * There's a hole, we need to insert something in it and | |
872 | * ignore the extent we found. | |
873 | */ | |
874 | if (state->start > start) { | |
875 | u64 this_end; | |
876 | if (end < last_start) | |
877 | this_end = end; | |
878 | else | |
d397712b | 879 | this_end = last_start - 1; |
d1310b2e CM |
880 | err = insert_state(tree, prealloc, start, this_end, |
881 | bits); | |
d1310b2e | 882 | BUG_ON(err == -EEXIST); |
9ed74f2d JB |
883 | if (err) { |
884 | prealloc = NULL; | |
d1310b2e | 885 | goto out; |
9ed74f2d JB |
886 | } |
887 | cache_state(prealloc, cached_state); | |
888 | prealloc = NULL; | |
d1310b2e CM |
889 | start = this_end + 1; |
890 | goto search_again; | |
891 | } | |
892 | /* | |
893 | * | ---- desired range ---- | | |
894 | * | state | | |
895 | * We need to split the extent, and set the bit | |
896 | * on the first half | |
897 | */ | |
898 | if (state->start <= end && state->end > end) { | |
1edbb734 | 899 | if (state->state & exclusive_bits) { |
d1310b2e CM |
900 | *failed_start = start; |
901 | err = -EEXIST; | |
902 | goto out; | |
903 | } | |
904 | err = split_state(tree, state, prealloc, end + 1); | |
905 | BUG_ON(err == -EEXIST); | |
906 | ||
9ed74f2d JB |
907 | err = set_state_bits(tree, prealloc, bits); |
908 | if (err) { | |
909 | prealloc = NULL; | |
910 | goto out; | |
911 | } | |
2c64c53d | 912 | cache_state(prealloc, cached_state); |
d1310b2e CM |
913 | merge_state(tree, prealloc); |
914 | prealloc = NULL; | |
915 | goto out; | |
916 | } | |
917 | ||
918 | goto search_again; | |
919 | ||
920 | out: | |
cad321ad | 921 | spin_unlock(&tree->lock); |
d1310b2e CM |
922 | if (prealloc) |
923 | free_extent_state(prealloc); | |
924 | ||
925 | return err; | |
926 | ||
927 | search_again: | |
928 | if (start > end) | |
929 | goto out; | |
cad321ad | 930 | spin_unlock(&tree->lock); |
d1310b2e CM |
931 | if (mask & __GFP_WAIT) |
932 | cond_resched(); | |
933 | goto again; | |
934 | } | |
d1310b2e CM |
935 | |
936 | /* wrappers around set/clear extent bit */ | |
937 | int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
938 | gfp_t mask) | |
939 | { | |
940 | return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL, | |
2c64c53d | 941 | NULL, mask); |
d1310b2e | 942 | } |
d1310b2e CM |
943 | |
944 | int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
945 | int bits, gfp_t mask) | |
946 | { | |
947 | return set_extent_bit(tree, start, end, bits, 0, NULL, | |
2c64c53d | 948 | NULL, mask); |
d1310b2e | 949 | } |
d1310b2e CM |
950 | |
951 | int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
952 | int bits, gfp_t mask) | |
953 | { | |
2c64c53d | 954 | return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask); |
d1310b2e | 955 | } |
d1310b2e CM |
956 | |
957 | int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, | |
2ac55d41 | 958 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
959 | { |
960 | return set_extent_bit(tree, start, end, | |
40431d6c | 961 | EXTENT_DELALLOC | EXTENT_DIRTY | EXTENT_UPTODATE, |
2ac55d41 | 962 | 0, NULL, cached_state, mask); |
d1310b2e | 963 | } |
d1310b2e CM |
964 | |
965 | int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
966 | gfp_t mask) | |
967 | { | |
968 | return clear_extent_bit(tree, start, end, | |
32c00aff JB |
969 | EXTENT_DIRTY | EXTENT_DELALLOC | |
970 | EXTENT_DO_ACCOUNTING, 0, 0, | |
2c64c53d | 971 | NULL, mask); |
d1310b2e | 972 | } |
d1310b2e CM |
973 | |
974 | int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, | |
975 | gfp_t mask) | |
976 | { | |
977 | return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL, | |
2c64c53d | 978 | NULL, mask); |
d1310b2e | 979 | } |
d1310b2e | 980 | |
b2950863 | 981 | static int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end, |
d1310b2e CM |
982 | gfp_t mask) |
983 | { | |
2c64c53d CM |
984 | return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, |
985 | NULL, mask); | |
d1310b2e | 986 | } |
d1310b2e CM |
987 | |
988 | int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, | |
989 | gfp_t mask) | |
990 | { | |
991 | return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL, | |
2c64c53d | 992 | NULL, mask); |
d1310b2e | 993 | } |
d1310b2e | 994 | |
d397712b | 995 | static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, |
2ac55d41 JB |
996 | u64 end, struct extent_state **cached_state, |
997 | gfp_t mask) | |
d1310b2e | 998 | { |
2c64c53d | 999 | return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, |
2ac55d41 | 1000 | cached_state, mask); |
d1310b2e | 1001 | } |
d1310b2e | 1002 | |
d1310b2e CM |
1003 | int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
1004 | { | |
1005 | return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK); | |
1006 | } | |
d1310b2e | 1007 | |
d352ac68 CM |
1008 | /* |
1009 | * either insert or lock state struct between start and end use mask to tell | |
1010 | * us if waiting is desired. | |
1011 | */ | |
1edbb734 | 1012 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
2c64c53d | 1013 | int bits, struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
1014 | { |
1015 | int err; | |
1016 | u64 failed_start; | |
1017 | while (1) { | |
1edbb734 | 1018 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits, |
2c64c53d CM |
1019 | EXTENT_LOCKED, &failed_start, |
1020 | cached_state, mask); | |
d1310b2e CM |
1021 | if (err == -EEXIST && (mask & __GFP_WAIT)) { |
1022 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); | |
1023 | start = failed_start; | |
1024 | } else { | |
1025 | break; | |
1026 | } | |
1027 | WARN_ON(start > end); | |
1028 | } | |
1029 | return err; | |
1030 | } | |
d1310b2e | 1031 | |
1edbb734 CM |
1032 | int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) |
1033 | { | |
2c64c53d | 1034 | return lock_extent_bits(tree, start, end, 0, NULL, mask); |
1edbb734 CM |
1035 | } |
1036 | ||
25179201 JB |
1037 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end, |
1038 | gfp_t mask) | |
1039 | { | |
1040 | int err; | |
1041 | u64 failed_start; | |
1042 | ||
2c64c53d CM |
1043 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
1044 | &failed_start, NULL, mask); | |
6643558d YZ |
1045 | if (err == -EEXIST) { |
1046 | if (failed_start > start) | |
1047 | clear_extent_bit(tree, start, failed_start - 1, | |
2c64c53d | 1048 | EXTENT_LOCKED, 1, 0, NULL, mask); |
25179201 | 1049 | return 0; |
6643558d | 1050 | } |
25179201 JB |
1051 | return 1; |
1052 | } | |
25179201 | 1053 | |
2c64c53d CM |
1054 | int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end, |
1055 | struct extent_state **cached, gfp_t mask) | |
1056 | { | |
1057 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, | |
1058 | mask); | |
1059 | } | |
1060 | ||
d1310b2e CM |
1061 | int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, |
1062 | gfp_t mask) | |
1063 | { | |
2c64c53d CM |
1064 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL, |
1065 | mask); | |
d1310b2e | 1066 | } |
d1310b2e CM |
1067 | |
1068 | /* | |
1069 | * helper function to set pages and extents in the tree dirty | |
1070 | */ | |
1071 | int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end) | |
1072 | { | |
1073 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1074 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1075 | struct page *page; | |
1076 | ||
1077 | while (index <= end_index) { | |
1078 | page = find_get_page(tree->mapping, index); | |
1079 | BUG_ON(!page); | |
1080 | __set_page_dirty_nobuffers(page); | |
1081 | page_cache_release(page); | |
1082 | index++; | |
1083 | } | |
d1310b2e CM |
1084 | return 0; |
1085 | } | |
d1310b2e CM |
1086 | |
1087 | /* | |
1088 | * helper function to set both pages and extents in the tree writeback | |
1089 | */ | |
b2950863 | 1090 | static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e CM |
1091 | { |
1092 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1093 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1094 | struct page *page; | |
1095 | ||
1096 | while (index <= end_index) { | |
1097 | page = find_get_page(tree->mapping, index); | |
1098 | BUG_ON(!page); | |
1099 | set_page_writeback(page); | |
1100 | page_cache_release(page); | |
1101 | index++; | |
1102 | } | |
d1310b2e CM |
1103 | return 0; |
1104 | } | |
d1310b2e | 1105 | |
d352ac68 CM |
1106 | /* |
1107 | * find the first offset in the io tree with 'bits' set. zero is | |
1108 | * returned if we find something, and *start_ret and *end_ret are | |
1109 | * set to reflect the state struct that was found. | |
1110 | * | |
1111 | * If nothing was found, 1 is returned, < 0 on error | |
1112 | */ | |
d1310b2e CM |
1113 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, |
1114 | u64 *start_ret, u64 *end_ret, int bits) | |
1115 | { | |
1116 | struct rb_node *node; | |
1117 | struct extent_state *state; | |
1118 | int ret = 1; | |
1119 | ||
cad321ad | 1120 | spin_lock(&tree->lock); |
d1310b2e CM |
1121 | /* |
1122 | * this search will find all the extents that end after | |
1123 | * our range starts. | |
1124 | */ | |
80ea96b1 | 1125 | node = tree_search(tree, start); |
d397712b | 1126 | if (!node) |
d1310b2e | 1127 | goto out; |
d1310b2e | 1128 | |
d397712b | 1129 | while (1) { |
d1310b2e CM |
1130 | state = rb_entry(node, struct extent_state, rb_node); |
1131 | if (state->end >= start && (state->state & bits)) { | |
1132 | *start_ret = state->start; | |
1133 | *end_ret = state->end; | |
1134 | ret = 0; | |
1135 | break; | |
1136 | } | |
1137 | node = rb_next(node); | |
1138 | if (!node) | |
1139 | break; | |
1140 | } | |
1141 | out: | |
cad321ad | 1142 | spin_unlock(&tree->lock); |
d1310b2e CM |
1143 | return ret; |
1144 | } | |
d1310b2e | 1145 | |
d352ac68 CM |
1146 | /* find the first state struct with 'bits' set after 'start', and |
1147 | * return it. tree->lock must be held. NULL will returned if | |
1148 | * nothing was found after 'start' | |
1149 | */ | |
d7fc640e CM |
1150 | struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree, |
1151 | u64 start, int bits) | |
1152 | { | |
1153 | struct rb_node *node; | |
1154 | struct extent_state *state; | |
1155 | ||
1156 | /* | |
1157 | * this search will find all the extents that end after | |
1158 | * our range starts. | |
1159 | */ | |
1160 | node = tree_search(tree, start); | |
d397712b | 1161 | if (!node) |
d7fc640e | 1162 | goto out; |
d7fc640e | 1163 | |
d397712b | 1164 | while (1) { |
d7fc640e | 1165 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1166 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1167 | return state; |
d397712b | 1168 | |
d7fc640e CM |
1169 | node = rb_next(node); |
1170 | if (!node) | |
1171 | break; | |
1172 | } | |
1173 | out: | |
1174 | return NULL; | |
1175 | } | |
d7fc640e | 1176 | |
d352ac68 CM |
1177 | /* |
1178 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1179 | * more than 'max_bytes'. start and end are used to return the range, | |
1180 | * | |
1181 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1182 | */ | |
c8b97818 | 1183 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1184 | u64 *start, u64 *end, u64 max_bytes, |
1185 | struct extent_state **cached_state) | |
d1310b2e CM |
1186 | { |
1187 | struct rb_node *node; | |
1188 | struct extent_state *state; | |
1189 | u64 cur_start = *start; | |
1190 | u64 found = 0; | |
1191 | u64 total_bytes = 0; | |
1192 | ||
cad321ad | 1193 | spin_lock(&tree->lock); |
c8b97818 | 1194 | |
d1310b2e CM |
1195 | /* |
1196 | * this search will find all the extents that end after | |
1197 | * our range starts. | |
1198 | */ | |
80ea96b1 | 1199 | node = tree_search(tree, cur_start); |
2b114d1d | 1200 | if (!node) { |
3b951516 CM |
1201 | if (!found) |
1202 | *end = (u64)-1; | |
d1310b2e CM |
1203 | goto out; |
1204 | } | |
1205 | ||
d397712b | 1206 | while (1) { |
d1310b2e | 1207 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1208 | if (found && (state->start != cur_start || |
1209 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1210 | goto out; |
1211 | } | |
1212 | if (!(state->state & EXTENT_DELALLOC)) { | |
1213 | if (!found) | |
1214 | *end = state->end; | |
1215 | goto out; | |
1216 | } | |
c2a128d2 | 1217 | if (!found) { |
d1310b2e | 1218 | *start = state->start; |
c2a128d2 JB |
1219 | *cached_state = state; |
1220 | atomic_inc(&state->refs); | |
1221 | } | |
d1310b2e CM |
1222 | found++; |
1223 | *end = state->end; | |
1224 | cur_start = state->end + 1; | |
1225 | node = rb_next(node); | |
1226 | if (!node) | |
1227 | break; | |
1228 | total_bytes += state->end - state->start + 1; | |
1229 | if (total_bytes >= max_bytes) | |
1230 | break; | |
1231 | } | |
1232 | out: | |
cad321ad | 1233 | spin_unlock(&tree->lock); |
d1310b2e CM |
1234 | return found; |
1235 | } | |
1236 | ||
c8b97818 CM |
1237 | static noinline int __unlock_for_delalloc(struct inode *inode, |
1238 | struct page *locked_page, | |
1239 | u64 start, u64 end) | |
1240 | { | |
1241 | int ret; | |
1242 | struct page *pages[16]; | |
1243 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1244 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1245 | unsigned long nr_pages = end_index - index + 1; | |
1246 | int i; | |
1247 | ||
1248 | if (index == locked_page->index && end_index == index) | |
1249 | return 0; | |
1250 | ||
d397712b | 1251 | while (nr_pages > 0) { |
c8b97818 | 1252 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1253 | min_t(unsigned long, nr_pages, |
1254 | ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1255 | for (i = 0; i < ret; i++) { |
1256 | if (pages[i] != locked_page) | |
1257 | unlock_page(pages[i]); | |
1258 | page_cache_release(pages[i]); | |
1259 | } | |
1260 | nr_pages -= ret; | |
1261 | index += ret; | |
1262 | cond_resched(); | |
1263 | } | |
1264 | return 0; | |
1265 | } | |
1266 | ||
1267 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1268 | struct page *locked_page, | |
1269 | u64 delalloc_start, | |
1270 | u64 delalloc_end) | |
1271 | { | |
1272 | unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT; | |
1273 | unsigned long start_index = index; | |
1274 | unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT; | |
1275 | unsigned long pages_locked = 0; | |
1276 | struct page *pages[16]; | |
1277 | unsigned long nrpages; | |
1278 | int ret; | |
1279 | int i; | |
1280 | ||
1281 | /* the caller is responsible for locking the start index */ | |
1282 | if (index == locked_page->index && index == end_index) | |
1283 | return 0; | |
1284 | ||
1285 | /* skip the page at the start index */ | |
1286 | nrpages = end_index - index + 1; | |
d397712b | 1287 | while (nrpages > 0) { |
c8b97818 | 1288 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1289 | min_t(unsigned long, |
1290 | nrpages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1291 | if (ret == 0) { |
1292 | ret = -EAGAIN; | |
1293 | goto done; | |
1294 | } | |
1295 | /* now we have an array of pages, lock them all */ | |
1296 | for (i = 0; i < ret; i++) { | |
1297 | /* | |
1298 | * the caller is taking responsibility for | |
1299 | * locked_page | |
1300 | */ | |
771ed689 | 1301 | if (pages[i] != locked_page) { |
c8b97818 | 1302 | lock_page(pages[i]); |
f2b1c41c CM |
1303 | if (!PageDirty(pages[i]) || |
1304 | pages[i]->mapping != inode->i_mapping) { | |
771ed689 CM |
1305 | ret = -EAGAIN; |
1306 | unlock_page(pages[i]); | |
1307 | page_cache_release(pages[i]); | |
1308 | goto done; | |
1309 | } | |
1310 | } | |
c8b97818 | 1311 | page_cache_release(pages[i]); |
771ed689 | 1312 | pages_locked++; |
c8b97818 | 1313 | } |
c8b97818 CM |
1314 | nrpages -= ret; |
1315 | index += ret; | |
1316 | cond_resched(); | |
1317 | } | |
1318 | ret = 0; | |
1319 | done: | |
1320 | if (ret && pages_locked) { | |
1321 | __unlock_for_delalloc(inode, locked_page, | |
1322 | delalloc_start, | |
1323 | ((u64)(start_index + pages_locked - 1)) << | |
1324 | PAGE_CACHE_SHIFT); | |
1325 | } | |
1326 | return ret; | |
1327 | } | |
1328 | ||
1329 | /* | |
1330 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1331 | * more than 'max_bytes'. start and end are used to return the range, | |
1332 | * | |
1333 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1334 | */ | |
1335 | static noinline u64 find_lock_delalloc_range(struct inode *inode, | |
1336 | struct extent_io_tree *tree, | |
1337 | struct page *locked_page, | |
1338 | u64 *start, u64 *end, | |
1339 | u64 max_bytes) | |
1340 | { | |
1341 | u64 delalloc_start; | |
1342 | u64 delalloc_end; | |
1343 | u64 found; | |
9655d298 | 1344 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1345 | int ret; |
1346 | int loops = 0; | |
1347 | ||
1348 | again: | |
1349 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1350 | delalloc_start = *start; | |
1351 | delalloc_end = 0; | |
1352 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1353 | max_bytes, &cached_state); |
70b99e69 | 1354 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1355 | *start = delalloc_start; |
1356 | *end = delalloc_end; | |
c2a128d2 | 1357 | free_extent_state(cached_state); |
c8b97818 CM |
1358 | return found; |
1359 | } | |
1360 | ||
70b99e69 CM |
1361 | /* |
1362 | * start comes from the offset of locked_page. We have to lock | |
1363 | * pages in order, so we can't process delalloc bytes before | |
1364 | * locked_page | |
1365 | */ | |
d397712b | 1366 | if (delalloc_start < *start) |
70b99e69 | 1367 | delalloc_start = *start; |
70b99e69 | 1368 | |
c8b97818 CM |
1369 | /* |
1370 | * make sure to limit the number of pages we try to lock down | |
1371 | * if we're looping. | |
1372 | */ | |
d397712b | 1373 | if (delalloc_end + 1 - delalloc_start > max_bytes && loops) |
771ed689 | 1374 | delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1; |
d397712b | 1375 | |
c8b97818 CM |
1376 | /* step two, lock all the pages after the page that has start */ |
1377 | ret = lock_delalloc_pages(inode, locked_page, | |
1378 | delalloc_start, delalloc_end); | |
1379 | if (ret == -EAGAIN) { | |
1380 | /* some of the pages are gone, lets avoid looping by | |
1381 | * shortening the size of the delalloc range we're searching | |
1382 | */ | |
9655d298 | 1383 | free_extent_state(cached_state); |
c8b97818 CM |
1384 | if (!loops) { |
1385 | unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1); | |
1386 | max_bytes = PAGE_CACHE_SIZE - offset; | |
1387 | loops = 1; | |
1388 | goto again; | |
1389 | } else { | |
1390 | found = 0; | |
1391 | goto out_failed; | |
1392 | } | |
1393 | } | |
1394 | BUG_ON(ret); | |
1395 | ||
1396 | /* step three, lock the state bits for the whole range */ | |
9655d298 CM |
1397 | lock_extent_bits(tree, delalloc_start, delalloc_end, |
1398 | 0, &cached_state, GFP_NOFS); | |
c8b97818 CM |
1399 | |
1400 | /* then test to make sure it is all still delalloc */ | |
1401 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1402 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1403 | if (!ret) { |
9655d298 CM |
1404 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
1405 | &cached_state, GFP_NOFS); | |
c8b97818 CM |
1406 | __unlock_for_delalloc(inode, locked_page, |
1407 | delalloc_start, delalloc_end); | |
1408 | cond_resched(); | |
1409 | goto again; | |
1410 | } | |
9655d298 | 1411 | free_extent_state(cached_state); |
c8b97818 CM |
1412 | *start = delalloc_start; |
1413 | *end = delalloc_end; | |
1414 | out_failed: | |
1415 | return found; | |
1416 | } | |
1417 | ||
1418 | int extent_clear_unlock_delalloc(struct inode *inode, | |
1419 | struct extent_io_tree *tree, | |
1420 | u64 start, u64 end, struct page *locked_page, | |
a791e35e | 1421 | unsigned long op) |
c8b97818 CM |
1422 | { |
1423 | int ret; | |
1424 | struct page *pages[16]; | |
1425 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1426 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1427 | unsigned long nr_pages = end_index - index + 1; | |
1428 | int i; | |
771ed689 | 1429 | int clear_bits = 0; |
c8b97818 | 1430 | |
a791e35e | 1431 | if (op & EXTENT_CLEAR_UNLOCK) |
771ed689 | 1432 | clear_bits |= EXTENT_LOCKED; |
a791e35e | 1433 | if (op & EXTENT_CLEAR_DIRTY) |
c8b97818 CM |
1434 | clear_bits |= EXTENT_DIRTY; |
1435 | ||
a791e35e | 1436 | if (op & EXTENT_CLEAR_DELALLOC) |
771ed689 CM |
1437 | clear_bits |= EXTENT_DELALLOC; |
1438 | ||
32c00aff JB |
1439 | if (op & EXTENT_CLEAR_ACCOUNTING) |
1440 | clear_bits |= EXTENT_DO_ACCOUNTING; | |
1441 | ||
2c64c53d | 1442 | clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); |
32c00aff JB |
1443 | if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | |
1444 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK | | |
1445 | EXTENT_SET_PRIVATE2))) | |
771ed689 | 1446 | return 0; |
c8b97818 | 1447 | |
d397712b | 1448 | while (nr_pages > 0) { |
c8b97818 | 1449 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1450 | min_t(unsigned long, |
1451 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 | 1452 | for (i = 0; i < ret; i++) { |
8b62b72b | 1453 | |
a791e35e | 1454 | if (op & EXTENT_SET_PRIVATE2) |
8b62b72b CM |
1455 | SetPagePrivate2(pages[i]); |
1456 | ||
c8b97818 CM |
1457 | if (pages[i] == locked_page) { |
1458 | page_cache_release(pages[i]); | |
1459 | continue; | |
1460 | } | |
a791e35e | 1461 | if (op & EXTENT_CLEAR_DIRTY) |
c8b97818 | 1462 | clear_page_dirty_for_io(pages[i]); |
a791e35e | 1463 | if (op & EXTENT_SET_WRITEBACK) |
c8b97818 | 1464 | set_page_writeback(pages[i]); |
a791e35e | 1465 | if (op & EXTENT_END_WRITEBACK) |
c8b97818 | 1466 | end_page_writeback(pages[i]); |
a791e35e | 1467 | if (op & EXTENT_CLEAR_UNLOCK_PAGE) |
771ed689 | 1468 | unlock_page(pages[i]); |
c8b97818 CM |
1469 | page_cache_release(pages[i]); |
1470 | } | |
1471 | nr_pages -= ret; | |
1472 | index += ret; | |
1473 | cond_resched(); | |
1474 | } | |
1475 | return 0; | |
1476 | } | |
c8b97818 | 1477 | |
d352ac68 CM |
1478 | /* |
1479 | * count the number of bytes in the tree that have a given bit(s) | |
1480 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1481 | * cached. The total number found is returned. | |
1482 | */ | |
d1310b2e CM |
1483 | u64 count_range_bits(struct extent_io_tree *tree, |
1484 | u64 *start, u64 search_end, u64 max_bytes, | |
1485 | unsigned long bits) | |
1486 | { | |
1487 | struct rb_node *node; | |
1488 | struct extent_state *state; | |
1489 | u64 cur_start = *start; | |
1490 | u64 total_bytes = 0; | |
1491 | int found = 0; | |
1492 | ||
1493 | if (search_end <= cur_start) { | |
d1310b2e CM |
1494 | WARN_ON(1); |
1495 | return 0; | |
1496 | } | |
1497 | ||
cad321ad | 1498 | spin_lock(&tree->lock); |
d1310b2e CM |
1499 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1500 | total_bytes = tree->dirty_bytes; | |
1501 | goto out; | |
1502 | } | |
1503 | /* | |
1504 | * this search will find all the extents that end after | |
1505 | * our range starts. | |
1506 | */ | |
80ea96b1 | 1507 | node = tree_search(tree, cur_start); |
d397712b | 1508 | if (!node) |
d1310b2e | 1509 | goto out; |
d1310b2e | 1510 | |
d397712b | 1511 | while (1) { |
d1310b2e CM |
1512 | state = rb_entry(node, struct extent_state, rb_node); |
1513 | if (state->start > search_end) | |
1514 | break; | |
1515 | if (state->end >= cur_start && (state->state & bits)) { | |
1516 | total_bytes += min(search_end, state->end) + 1 - | |
1517 | max(cur_start, state->start); | |
1518 | if (total_bytes >= max_bytes) | |
1519 | break; | |
1520 | if (!found) { | |
1521 | *start = state->start; | |
1522 | found = 1; | |
1523 | } | |
1524 | } | |
1525 | node = rb_next(node); | |
1526 | if (!node) | |
1527 | break; | |
1528 | } | |
1529 | out: | |
cad321ad | 1530 | spin_unlock(&tree->lock); |
d1310b2e CM |
1531 | return total_bytes; |
1532 | } | |
b2950863 | 1533 | |
d352ac68 CM |
1534 | /* |
1535 | * set the private field for a given byte offset in the tree. If there isn't | |
1536 | * an extent_state there already, this does nothing. | |
1537 | */ | |
d1310b2e CM |
1538 | int set_state_private(struct extent_io_tree *tree, u64 start, u64 private) |
1539 | { | |
1540 | struct rb_node *node; | |
1541 | struct extent_state *state; | |
1542 | int ret = 0; | |
1543 | ||
cad321ad | 1544 | spin_lock(&tree->lock); |
d1310b2e CM |
1545 | /* |
1546 | * this search will find all the extents that end after | |
1547 | * our range starts. | |
1548 | */ | |
80ea96b1 | 1549 | node = tree_search(tree, start); |
2b114d1d | 1550 | if (!node) { |
d1310b2e CM |
1551 | ret = -ENOENT; |
1552 | goto out; | |
1553 | } | |
1554 | state = rb_entry(node, struct extent_state, rb_node); | |
1555 | if (state->start != start) { | |
1556 | ret = -ENOENT; | |
1557 | goto out; | |
1558 | } | |
1559 | state->private = private; | |
1560 | out: | |
cad321ad | 1561 | spin_unlock(&tree->lock); |
d1310b2e CM |
1562 | return ret; |
1563 | } | |
1564 | ||
1565 | int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private) | |
1566 | { | |
1567 | struct rb_node *node; | |
1568 | struct extent_state *state; | |
1569 | int ret = 0; | |
1570 | ||
cad321ad | 1571 | spin_lock(&tree->lock); |
d1310b2e CM |
1572 | /* |
1573 | * this search will find all the extents that end after | |
1574 | * our range starts. | |
1575 | */ | |
80ea96b1 | 1576 | node = tree_search(tree, start); |
2b114d1d | 1577 | if (!node) { |
d1310b2e CM |
1578 | ret = -ENOENT; |
1579 | goto out; | |
1580 | } | |
1581 | state = rb_entry(node, struct extent_state, rb_node); | |
1582 | if (state->start != start) { | |
1583 | ret = -ENOENT; | |
1584 | goto out; | |
1585 | } | |
1586 | *private = state->private; | |
1587 | out: | |
cad321ad | 1588 | spin_unlock(&tree->lock); |
d1310b2e CM |
1589 | return ret; |
1590 | } | |
1591 | ||
1592 | /* | |
1593 | * searches a range in the state tree for a given mask. | |
70dec807 | 1594 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1595 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1596 | * range is found set. | |
1597 | */ | |
1598 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9655d298 | 1599 | int bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1600 | { |
1601 | struct extent_state *state = NULL; | |
1602 | struct rb_node *node; | |
1603 | int bitset = 0; | |
d1310b2e | 1604 | |
cad321ad | 1605 | spin_lock(&tree->lock); |
9655d298 CM |
1606 | if (cached && cached->tree && cached->start == start) |
1607 | node = &cached->rb_node; | |
1608 | else | |
1609 | node = tree_search(tree, start); | |
d1310b2e CM |
1610 | while (node && start <= end) { |
1611 | state = rb_entry(node, struct extent_state, rb_node); | |
1612 | ||
1613 | if (filled && state->start > start) { | |
1614 | bitset = 0; | |
1615 | break; | |
1616 | } | |
1617 | ||
1618 | if (state->start > end) | |
1619 | break; | |
1620 | ||
1621 | if (state->state & bits) { | |
1622 | bitset = 1; | |
1623 | if (!filled) | |
1624 | break; | |
1625 | } else if (filled) { | |
1626 | bitset = 0; | |
1627 | break; | |
1628 | } | |
46562cec CM |
1629 | |
1630 | if (state->end == (u64)-1) | |
1631 | break; | |
1632 | ||
d1310b2e CM |
1633 | start = state->end + 1; |
1634 | if (start > end) | |
1635 | break; | |
1636 | node = rb_next(node); | |
1637 | if (!node) { | |
1638 | if (filled) | |
1639 | bitset = 0; | |
1640 | break; | |
1641 | } | |
1642 | } | |
cad321ad | 1643 | spin_unlock(&tree->lock); |
d1310b2e CM |
1644 | return bitset; |
1645 | } | |
d1310b2e CM |
1646 | |
1647 | /* | |
1648 | * helper function to set a given page up to date if all the | |
1649 | * extents in the tree for that page are up to date | |
1650 | */ | |
1651 | static int check_page_uptodate(struct extent_io_tree *tree, | |
1652 | struct page *page) | |
1653 | { | |
1654 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1655 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
9655d298 | 1656 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e CM |
1657 | SetPageUptodate(page); |
1658 | return 0; | |
1659 | } | |
1660 | ||
1661 | /* | |
1662 | * helper function to unlock a page if all the extents in the tree | |
1663 | * for that page are unlocked | |
1664 | */ | |
1665 | static int check_page_locked(struct extent_io_tree *tree, | |
1666 | struct page *page) | |
1667 | { | |
1668 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1669 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
9655d298 | 1670 | if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) |
d1310b2e CM |
1671 | unlock_page(page); |
1672 | return 0; | |
1673 | } | |
1674 | ||
1675 | /* | |
1676 | * helper function to end page writeback if all the extents | |
1677 | * in the tree for that page are done with writeback | |
1678 | */ | |
1679 | static int check_page_writeback(struct extent_io_tree *tree, | |
1680 | struct page *page) | |
1681 | { | |
1edbb734 | 1682 | end_page_writeback(page); |
d1310b2e CM |
1683 | return 0; |
1684 | } | |
1685 | ||
1686 | /* lots and lots of room for performance fixes in the end_bio funcs */ | |
1687 | ||
1688 | /* | |
1689 | * after a writepage IO is done, we need to: | |
1690 | * clear the uptodate bits on error | |
1691 | * clear the writeback bits in the extent tree for this IO | |
1692 | * end_page_writeback if the page has no more pending IO | |
1693 | * | |
1694 | * Scheduling is not allowed, so the extent state tree is expected | |
1695 | * to have one and only one object corresponding to this IO. | |
1696 | */ | |
d1310b2e | 1697 | static void end_bio_extent_writepage(struct bio *bio, int err) |
d1310b2e | 1698 | { |
1259ab75 | 1699 | int uptodate = err == 0; |
d1310b2e | 1700 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
902b22f3 | 1701 | struct extent_io_tree *tree; |
d1310b2e CM |
1702 | u64 start; |
1703 | u64 end; | |
1704 | int whole_page; | |
1259ab75 | 1705 | int ret; |
d1310b2e | 1706 | |
d1310b2e CM |
1707 | do { |
1708 | struct page *page = bvec->bv_page; | |
902b22f3 DW |
1709 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
1710 | ||
d1310b2e CM |
1711 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
1712 | bvec->bv_offset; | |
1713 | end = start + bvec->bv_len - 1; | |
1714 | ||
1715 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
1716 | whole_page = 1; | |
1717 | else | |
1718 | whole_page = 0; | |
1719 | ||
1720 | if (--bvec >= bio->bi_io_vec) | |
1721 | prefetchw(&bvec->bv_page->flags); | |
1259ab75 CM |
1722 | if (tree->ops && tree->ops->writepage_end_io_hook) { |
1723 | ret = tree->ops->writepage_end_io_hook(page, start, | |
902b22f3 | 1724 | end, NULL, uptodate); |
1259ab75 CM |
1725 | if (ret) |
1726 | uptodate = 0; | |
1727 | } | |
1728 | ||
1729 | if (!uptodate && tree->ops && | |
1730 | tree->ops->writepage_io_failed_hook) { | |
1731 | ret = tree->ops->writepage_io_failed_hook(bio, page, | |
902b22f3 | 1732 | start, end, NULL); |
1259ab75 | 1733 | if (ret == 0) { |
1259ab75 CM |
1734 | uptodate = (err == 0); |
1735 | continue; | |
1736 | } | |
1737 | } | |
1738 | ||
d1310b2e | 1739 | if (!uptodate) { |
2ac55d41 | 1740 | clear_extent_uptodate(tree, start, end, NULL, GFP_NOFS); |
d1310b2e CM |
1741 | ClearPageUptodate(page); |
1742 | SetPageError(page); | |
1743 | } | |
70dec807 | 1744 | |
d1310b2e CM |
1745 | if (whole_page) |
1746 | end_page_writeback(page); | |
1747 | else | |
1748 | check_page_writeback(tree, page); | |
d1310b2e | 1749 | } while (bvec >= bio->bi_io_vec); |
2b1f55b0 | 1750 | |
d1310b2e | 1751 | bio_put(bio); |
d1310b2e CM |
1752 | } |
1753 | ||
1754 | /* | |
1755 | * after a readpage IO is done, we need to: | |
1756 | * clear the uptodate bits on error | |
1757 | * set the uptodate bits if things worked | |
1758 | * set the page up to date if all extents in the tree are uptodate | |
1759 | * clear the lock bit in the extent tree | |
1760 | * unlock the page if there are no other extents locked for it | |
1761 | * | |
1762 | * Scheduling is not allowed, so the extent state tree is expected | |
1763 | * to have one and only one object corresponding to this IO. | |
1764 | */ | |
d1310b2e | 1765 | static void end_bio_extent_readpage(struct bio *bio, int err) |
d1310b2e CM |
1766 | { |
1767 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
4125bf76 CM |
1768 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
1769 | struct bio_vec *bvec = bio->bi_io_vec; | |
902b22f3 | 1770 | struct extent_io_tree *tree; |
d1310b2e CM |
1771 | u64 start; |
1772 | u64 end; | |
1773 | int whole_page; | |
1774 | int ret; | |
1775 | ||
d20f7043 CM |
1776 | if (err) |
1777 | uptodate = 0; | |
1778 | ||
d1310b2e CM |
1779 | do { |
1780 | struct page *page = bvec->bv_page; | |
902b22f3 DW |
1781 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
1782 | ||
d1310b2e CM |
1783 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
1784 | bvec->bv_offset; | |
1785 | end = start + bvec->bv_len - 1; | |
1786 | ||
1787 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
1788 | whole_page = 1; | |
1789 | else | |
1790 | whole_page = 0; | |
1791 | ||
4125bf76 | 1792 | if (++bvec <= bvec_end) |
d1310b2e CM |
1793 | prefetchw(&bvec->bv_page->flags); |
1794 | ||
1795 | if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) { | |
70dec807 | 1796 | ret = tree->ops->readpage_end_io_hook(page, start, end, |
902b22f3 | 1797 | NULL); |
d1310b2e CM |
1798 | if (ret) |
1799 | uptodate = 0; | |
1800 | } | |
7e38326f CM |
1801 | if (!uptodate && tree->ops && |
1802 | tree->ops->readpage_io_failed_hook) { | |
1803 | ret = tree->ops->readpage_io_failed_hook(bio, page, | |
902b22f3 | 1804 | start, end, NULL); |
7e38326f | 1805 | if (ret == 0) { |
3b951516 CM |
1806 | uptodate = |
1807 | test_bit(BIO_UPTODATE, &bio->bi_flags); | |
d20f7043 CM |
1808 | if (err) |
1809 | uptodate = 0; | |
7e38326f CM |
1810 | continue; |
1811 | } | |
1812 | } | |
d1310b2e | 1813 | |
771ed689 | 1814 | if (uptodate) { |
902b22f3 DW |
1815 | set_extent_uptodate(tree, start, end, |
1816 | GFP_ATOMIC); | |
771ed689 | 1817 | } |
902b22f3 | 1818 | unlock_extent(tree, start, end, GFP_ATOMIC); |
d1310b2e | 1819 | |
70dec807 CM |
1820 | if (whole_page) { |
1821 | if (uptodate) { | |
1822 | SetPageUptodate(page); | |
1823 | } else { | |
1824 | ClearPageUptodate(page); | |
1825 | SetPageError(page); | |
1826 | } | |
d1310b2e | 1827 | unlock_page(page); |
70dec807 CM |
1828 | } else { |
1829 | if (uptodate) { | |
1830 | check_page_uptodate(tree, page); | |
1831 | } else { | |
1832 | ClearPageUptodate(page); | |
1833 | SetPageError(page); | |
1834 | } | |
d1310b2e | 1835 | check_page_locked(tree, page); |
70dec807 | 1836 | } |
4125bf76 | 1837 | } while (bvec <= bvec_end); |
d1310b2e CM |
1838 | |
1839 | bio_put(bio); | |
d1310b2e CM |
1840 | } |
1841 | ||
1842 | /* | |
1843 | * IO done from prepare_write is pretty simple, we just unlock | |
1844 | * the structs in the extent tree when done, and set the uptodate bits | |
1845 | * as appropriate. | |
1846 | */ | |
d1310b2e | 1847 | static void end_bio_extent_preparewrite(struct bio *bio, int err) |
d1310b2e CM |
1848 | { |
1849 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
1850 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
902b22f3 | 1851 | struct extent_io_tree *tree; |
d1310b2e CM |
1852 | u64 start; |
1853 | u64 end; | |
1854 | ||
d1310b2e CM |
1855 | do { |
1856 | struct page *page = bvec->bv_page; | |
902b22f3 DW |
1857 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
1858 | ||
d1310b2e CM |
1859 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
1860 | bvec->bv_offset; | |
1861 | end = start + bvec->bv_len - 1; | |
1862 | ||
1863 | if (--bvec >= bio->bi_io_vec) | |
1864 | prefetchw(&bvec->bv_page->flags); | |
1865 | ||
1866 | if (uptodate) { | |
1867 | set_extent_uptodate(tree, start, end, GFP_ATOMIC); | |
1868 | } else { | |
1869 | ClearPageUptodate(page); | |
1870 | SetPageError(page); | |
1871 | } | |
1872 | ||
1873 | unlock_extent(tree, start, end, GFP_ATOMIC); | |
1874 | ||
1875 | } while (bvec >= bio->bi_io_vec); | |
1876 | ||
1877 | bio_put(bio); | |
d1310b2e CM |
1878 | } |
1879 | ||
1880 | static struct bio * | |
1881 | extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
1882 | gfp_t gfp_flags) | |
1883 | { | |
1884 | struct bio *bio; | |
1885 | ||
1886 | bio = bio_alloc(gfp_flags, nr_vecs); | |
1887 | ||
1888 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
1889 | while (!bio && (nr_vecs /= 2)) | |
1890 | bio = bio_alloc(gfp_flags, nr_vecs); | |
1891 | } | |
1892 | ||
1893 | if (bio) { | |
e1c4b745 | 1894 | bio->bi_size = 0; |
d1310b2e CM |
1895 | bio->bi_bdev = bdev; |
1896 | bio->bi_sector = first_sector; | |
1897 | } | |
1898 | return bio; | |
1899 | } | |
1900 | ||
c8b97818 CM |
1901 | static int submit_one_bio(int rw, struct bio *bio, int mirror_num, |
1902 | unsigned long bio_flags) | |
d1310b2e | 1903 | { |
d1310b2e | 1904 | int ret = 0; |
70dec807 CM |
1905 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
1906 | struct page *page = bvec->bv_page; | |
1907 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 CM |
1908 | u64 start; |
1909 | u64 end; | |
1910 | ||
1911 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset; | |
1912 | end = start + bvec->bv_len - 1; | |
1913 | ||
902b22f3 | 1914 | bio->bi_private = NULL; |
d1310b2e CM |
1915 | |
1916 | bio_get(bio); | |
1917 | ||
065631f6 | 1918 | if (tree->ops && tree->ops->submit_bio_hook) |
f188591e | 1919 | tree->ops->submit_bio_hook(page->mapping->host, rw, bio, |
c8b97818 | 1920 | mirror_num, bio_flags); |
0b86a832 CM |
1921 | else |
1922 | submit_bio(rw, bio); | |
d1310b2e CM |
1923 | if (bio_flagged(bio, BIO_EOPNOTSUPP)) |
1924 | ret = -EOPNOTSUPP; | |
1925 | bio_put(bio); | |
1926 | return ret; | |
1927 | } | |
1928 | ||
1929 | static int submit_extent_page(int rw, struct extent_io_tree *tree, | |
1930 | struct page *page, sector_t sector, | |
1931 | size_t size, unsigned long offset, | |
1932 | struct block_device *bdev, | |
1933 | struct bio **bio_ret, | |
1934 | unsigned long max_pages, | |
f188591e | 1935 | bio_end_io_t end_io_func, |
c8b97818 CM |
1936 | int mirror_num, |
1937 | unsigned long prev_bio_flags, | |
1938 | unsigned long bio_flags) | |
d1310b2e CM |
1939 | { |
1940 | int ret = 0; | |
1941 | struct bio *bio; | |
1942 | int nr; | |
c8b97818 CM |
1943 | int contig = 0; |
1944 | int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED; | |
1945 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; | |
5b050f04 | 1946 | size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE); |
d1310b2e CM |
1947 | |
1948 | if (bio_ret && *bio_ret) { | |
1949 | bio = *bio_ret; | |
c8b97818 CM |
1950 | if (old_compressed) |
1951 | contig = bio->bi_sector == sector; | |
1952 | else | |
1953 | contig = bio->bi_sector + (bio->bi_size >> 9) == | |
1954 | sector; | |
1955 | ||
1956 | if (prev_bio_flags != bio_flags || !contig || | |
239b14b3 | 1957 | (tree->ops && tree->ops->merge_bio_hook && |
c8b97818 CM |
1958 | tree->ops->merge_bio_hook(page, offset, page_size, bio, |
1959 | bio_flags)) || | |
1960 | bio_add_page(bio, page, page_size, offset) < page_size) { | |
1961 | ret = submit_one_bio(rw, bio, mirror_num, | |
1962 | prev_bio_flags); | |
d1310b2e CM |
1963 | bio = NULL; |
1964 | } else { | |
1965 | return 0; | |
1966 | } | |
1967 | } | |
c8b97818 CM |
1968 | if (this_compressed) |
1969 | nr = BIO_MAX_PAGES; | |
1970 | else | |
1971 | nr = bio_get_nr_vecs(bdev); | |
1972 | ||
d1310b2e | 1973 | bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH); |
70dec807 | 1974 | |
c8b97818 | 1975 | bio_add_page(bio, page, page_size, offset); |
d1310b2e CM |
1976 | bio->bi_end_io = end_io_func; |
1977 | bio->bi_private = tree; | |
70dec807 | 1978 | |
d397712b | 1979 | if (bio_ret) |
d1310b2e | 1980 | *bio_ret = bio; |
d397712b | 1981 | else |
c8b97818 | 1982 | ret = submit_one_bio(rw, bio, mirror_num, bio_flags); |
d1310b2e CM |
1983 | |
1984 | return ret; | |
1985 | } | |
1986 | ||
1987 | void set_page_extent_mapped(struct page *page) | |
1988 | { | |
1989 | if (!PagePrivate(page)) { | |
1990 | SetPagePrivate(page); | |
d1310b2e | 1991 | page_cache_get(page); |
6af118ce | 1992 | set_page_private(page, EXTENT_PAGE_PRIVATE); |
d1310b2e CM |
1993 | } |
1994 | } | |
1995 | ||
b2950863 | 1996 | static void set_page_extent_head(struct page *page, unsigned long len) |
d1310b2e CM |
1997 | { |
1998 | set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2); | |
1999 | } | |
2000 | ||
2001 | /* | |
2002 | * basic readpage implementation. Locked extent state structs are inserted | |
2003 | * into the tree that are removed when the IO is done (by the end_io | |
2004 | * handlers) | |
2005 | */ | |
2006 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
2007 | struct page *page, | |
2008 | get_extent_t *get_extent, | |
c8b97818 CM |
2009 | struct bio **bio, int mirror_num, |
2010 | unsigned long *bio_flags) | |
d1310b2e CM |
2011 | { |
2012 | struct inode *inode = page->mapping->host; | |
2013 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2014 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2015 | u64 end; | |
2016 | u64 cur = start; | |
2017 | u64 extent_offset; | |
2018 | u64 last_byte = i_size_read(inode); | |
2019 | u64 block_start; | |
2020 | u64 cur_end; | |
2021 | sector_t sector; | |
2022 | struct extent_map *em; | |
2023 | struct block_device *bdev; | |
2024 | int ret; | |
2025 | int nr = 0; | |
2026 | size_t page_offset = 0; | |
2027 | size_t iosize; | |
c8b97818 | 2028 | size_t disk_io_size; |
d1310b2e | 2029 | size_t blocksize = inode->i_sb->s_blocksize; |
c8b97818 | 2030 | unsigned long this_bio_flag = 0; |
d1310b2e CM |
2031 | |
2032 | set_page_extent_mapped(page); | |
2033 | ||
2034 | end = page_end; | |
2035 | lock_extent(tree, start, end, GFP_NOFS); | |
2036 | ||
c8b97818 CM |
2037 | if (page->index == last_byte >> PAGE_CACHE_SHIFT) { |
2038 | char *userpage; | |
2039 | size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1); | |
2040 | ||
2041 | if (zero_offset) { | |
2042 | iosize = PAGE_CACHE_SIZE - zero_offset; | |
2043 | userpage = kmap_atomic(page, KM_USER0); | |
2044 | memset(userpage + zero_offset, 0, iosize); | |
2045 | flush_dcache_page(page); | |
2046 | kunmap_atomic(userpage, KM_USER0); | |
2047 | } | |
2048 | } | |
d1310b2e CM |
2049 | while (cur <= end) { |
2050 | if (cur >= last_byte) { | |
2051 | char *userpage; | |
2052 | iosize = PAGE_CACHE_SIZE - page_offset; | |
2053 | userpage = kmap_atomic(page, KM_USER0); | |
2054 | memset(userpage + page_offset, 0, iosize); | |
2055 | flush_dcache_page(page); | |
2056 | kunmap_atomic(userpage, KM_USER0); | |
2057 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
2058 | GFP_NOFS); | |
2059 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
2060 | break; | |
2061 | } | |
2062 | em = get_extent(inode, page, page_offset, cur, | |
2063 | end - cur + 1, 0); | |
2064 | if (IS_ERR(em) || !em) { | |
2065 | SetPageError(page); | |
2066 | unlock_extent(tree, cur, end, GFP_NOFS); | |
2067 | break; | |
2068 | } | |
d1310b2e CM |
2069 | extent_offset = cur - em->start; |
2070 | BUG_ON(extent_map_end(em) <= cur); | |
2071 | BUG_ON(end < cur); | |
2072 | ||
c8b97818 CM |
2073 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
2074 | this_bio_flag = EXTENT_BIO_COMPRESSED; | |
2075 | ||
d1310b2e CM |
2076 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
2077 | cur_end = min(extent_map_end(em) - 1, end); | |
2078 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
c8b97818 CM |
2079 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
2080 | disk_io_size = em->block_len; | |
2081 | sector = em->block_start >> 9; | |
2082 | } else { | |
2083 | sector = (em->block_start + extent_offset) >> 9; | |
2084 | disk_io_size = iosize; | |
2085 | } | |
d1310b2e CM |
2086 | bdev = em->bdev; |
2087 | block_start = em->block_start; | |
d899e052 YZ |
2088 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2089 | block_start = EXTENT_MAP_HOLE; | |
d1310b2e CM |
2090 | free_extent_map(em); |
2091 | em = NULL; | |
2092 | ||
2093 | /* we've found a hole, just zero and go on */ | |
2094 | if (block_start == EXTENT_MAP_HOLE) { | |
2095 | char *userpage; | |
2096 | userpage = kmap_atomic(page, KM_USER0); | |
2097 | memset(userpage + page_offset, 0, iosize); | |
2098 | flush_dcache_page(page); | |
2099 | kunmap_atomic(userpage, KM_USER0); | |
2100 | ||
2101 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
2102 | GFP_NOFS); | |
2103 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
2104 | cur = cur + iosize; | |
2105 | page_offset += iosize; | |
2106 | continue; | |
2107 | } | |
2108 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
2109 | if (test_range_bit(tree, cur, cur_end, |
2110 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 2111 | check_page_uptodate(tree, page); |
d1310b2e CM |
2112 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); |
2113 | cur = cur + iosize; | |
2114 | page_offset += iosize; | |
2115 | continue; | |
2116 | } | |
70dec807 CM |
2117 | /* we have an inline extent but it didn't get marked up |
2118 | * to date. Error out | |
2119 | */ | |
2120 | if (block_start == EXTENT_MAP_INLINE) { | |
2121 | SetPageError(page); | |
2122 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
2123 | cur = cur + iosize; | |
2124 | page_offset += iosize; | |
2125 | continue; | |
2126 | } | |
d1310b2e CM |
2127 | |
2128 | ret = 0; | |
2129 | if (tree->ops && tree->ops->readpage_io_hook) { | |
2130 | ret = tree->ops->readpage_io_hook(page, cur, | |
2131 | cur + iosize - 1); | |
2132 | } | |
2133 | if (!ret) { | |
89642229 CM |
2134 | unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1; |
2135 | pnr -= page->index; | |
d1310b2e | 2136 | ret = submit_extent_page(READ, tree, page, |
c8b97818 | 2137 | sector, disk_io_size, page_offset, |
89642229 | 2138 | bdev, bio, pnr, |
c8b97818 CM |
2139 | end_bio_extent_readpage, mirror_num, |
2140 | *bio_flags, | |
2141 | this_bio_flag); | |
89642229 | 2142 | nr++; |
c8b97818 | 2143 | *bio_flags = this_bio_flag; |
d1310b2e CM |
2144 | } |
2145 | if (ret) | |
2146 | SetPageError(page); | |
2147 | cur = cur + iosize; | |
2148 | page_offset += iosize; | |
d1310b2e CM |
2149 | } |
2150 | if (!nr) { | |
2151 | if (!PageError(page)) | |
2152 | SetPageUptodate(page); | |
2153 | unlock_page(page); | |
2154 | } | |
2155 | return 0; | |
2156 | } | |
2157 | ||
2158 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, | |
2159 | get_extent_t *get_extent) | |
2160 | { | |
2161 | struct bio *bio = NULL; | |
c8b97818 | 2162 | unsigned long bio_flags = 0; |
d1310b2e CM |
2163 | int ret; |
2164 | ||
c8b97818 CM |
2165 | ret = __extent_read_full_page(tree, page, get_extent, &bio, 0, |
2166 | &bio_flags); | |
d1310b2e | 2167 | if (bio) |
c8b97818 | 2168 | submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
2169 | return ret; |
2170 | } | |
d1310b2e | 2171 | |
11c8349b CM |
2172 | static noinline void update_nr_written(struct page *page, |
2173 | struct writeback_control *wbc, | |
2174 | unsigned long nr_written) | |
2175 | { | |
2176 | wbc->nr_to_write -= nr_written; | |
2177 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && | |
2178 | wbc->range_start == 0 && wbc->range_end == LLONG_MAX)) | |
2179 | page->mapping->writeback_index = page->index + nr_written; | |
2180 | } | |
2181 | ||
d1310b2e CM |
2182 | /* |
2183 | * the writepage semantics are similar to regular writepage. extent | |
2184 | * records are inserted to lock ranges in the tree, and as dirty areas | |
2185 | * are found, they are marked writeback. Then the lock bits are removed | |
2186 | * and the end_io handler clears the writeback ranges | |
2187 | */ | |
2188 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
2189 | void *data) | |
2190 | { | |
2191 | struct inode *inode = page->mapping->host; | |
2192 | struct extent_page_data *epd = data; | |
2193 | struct extent_io_tree *tree = epd->tree; | |
2194 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2195 | u64 delalloc_start; | |
2196 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2197 | u64 end; | |
2198 | u64 cur = start; | |
2199 | u64 extent_offset; | |
2200 | u64 last_byte = i_size_read(inode); | |
2201 | u64 block_start; | |
2202 | u64 iosize; | |
e6dcd2dc | 2203 | u64 unlock_start; |
d1310b2e | 2204 | sector_t sector; |
2c64c53d | 2205 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
2206 | struct extent_map *em; |
2207 | struct block_device *bdev; | |
2208 | int ret; | |
2209 | int nr = 0; | |
7f3c74fb | 2210 | size_t pg_offset = 0; |
d1310b2e CM |
2211 | size_t blocksize; |
2212 | loff_t i_size = i_size_read(inode); | |
2213 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
2214 | u64 nr_delalloc; | |
2215 | u64 delalloc_end; | |
c8b97818 CM |
2216 | int page_started; |
2217 | int compressed; | |
ffbd517d | 2218 | int write_flags; |
771ed689 | 2219 | unsigned long nr_written = 0; |
d1310b2e | 2220 | |
ffbd517d CM |
2221 | if (wbc->sync_mode == WB_SYNC_ALL) |
2222 | write_flags = WRITE_SYNC_PLUG; | |
2223 | else | |
2224 | write_flags = WRITE; | |
2225 | ||
d1310b2e | 2226 | WARN_ON(!PageLocked(page)); |
7f3c74fb | 2227 | pg_offset = i_size & (PAGE_CACHE_SIZE - 1); |
211c17f5 | 2228 | if (page->index > end_index || |
7f3c74fb | 2229 | (page->index == end_index && !pg_offset)) { |
39be25cd | 2230 | page->mapping->a_ops->invalidatepage(page, 0); |
d1310b2e CM |
2231 | unlock_page(page); |
2232 | return 0; | |
2233 | } | |
2234 | ||
2235 | if (page->index == end_index) { | |
2236 | char *userpage; | |
2237 | ||
d1310b2e | 2238 | userpage = kmap_atomic(page, KM_USER0); |
7f3c74fb CM |
2239 | memset(userpage + pg_offset, 0, |
2240 | PAGE_CACHE_SIZE - pg_offset); | |
d1310b2e | 2241 | kunmap_atomic(userpage, KM_USER0); |
211c17f5 | 2242 | flush_dcache_page(page); |
d1310b2e | 2243 | } |
7f3c74fb | 2244 | pg_offset = 0; |
d1310b2e CM |
2245 | |
2246 | set_page_extent_mapped(page); | |
2247 | ||
2248 | delalloc_start = start; | |
2249 | delalloc_end = 0; | |
c8b97818 | 2250 | page_started = 0; |
771ed689 | 2251 | if (!epd->extent_locked) { |
f85d7d6c | 2252 | u64 delalloc_to_write = 0; |
11c8349b CM |
2253 | /* |
2254 | * make sure the wbc mapping index is at least updated | |
2255 | * to this page. | |
2256 | */ | |
2257 | update_nr_written(page, wbc, 0); | |
2258 | ||
d397712b | 2259 | while (delalloc_end < page_end) { |
771ed689 | 2260 | nr_delalloc = find_lock_delalloc_range(inode, tree, |
c8b97818 CM |
2261 | page, |
2262 | &delalloc_start, | |
d1310b2e CM |
2263 | &delalloc_end, |
2264 | 128 * 1024 * 1024); | |
771ed689 CM |
2265 | if (nr_delalloc == 0) { |
2266 | delalloc_start = delalloc_end + 1; | |
2267 | continue; | |
2268 | } | |
2269 | tree->ops->fill_delalloc(inode, page, delalloc_start, | |
2270 | delalloc_end, &page_started, | |
2271 | &nr_written); | |
f85d7d6c CM |
2272 | /* |
2273 | * delalloc_end is already one less than the total | |
2274 | * length, so we don't subtract one from | |
2275 | * PAGE_CACHE_SIZE | |
2276 | */ | |
2277 | delalloc_to_write += (delalloc_end - delalloc_start + | |
2278 | PAGE_CACHE_SIZE) >> | |
2279 | PAGE_CACHE_SHIFT; | |
d1310b2e | 2280 | delalloc_start = delalloc_end + 1; |
d1310b2e | 2281 | } |
f85d7d6c CM |
2282 | if (wbc->nr_to_write < delalloc_to_write) { |
2283 | int thresh = 8192; | |
2284 | ||
2285 | if (delalloc_to_write < thresh * 2) | |
2286 | thresh = delalloc_to_write; | |
2287 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
2288 | thresh); | |
2289 | } | |
c8b97818 | 2290 | |
771ed689 CM |
2291 | /* did the fill delalloc function already unlock and start |
2292 | * the IO? | |
2293 | */ | |
2294 | if (page_started) { | |
2295 | ret = 0; | |
11c8349b CM |
2296 | /* |
2297 | * we've unlocked the page, so we can't update | |
2298 | * the mapping's writeback index, just update | |
2299 | * nr_to_write. | |
2300 | */ | |
2301 | wbc->nr_to_write -= nr_written; | |
2302 | goto done_unlocked; | |
771ed689 | 2303 | } |
c8b97818 | 2304 | } |
247e743c | 2305 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
2306 | ret = tree->ops->writepage_start_hook(page, start, |
2307 | page_end); | |
247e743c | 2308 | if (ret == -EAGAIN) { |
247e743c | 2309 | redirty_page_for_writepage(wbc, page); |
11c8349b | 2310 | update_nr_written(page, wbc, nr_written); |
247e743c | 2311 | unlock_page(page); |
771ed689 | 2312 | ret = 0; |
11c8349b | 2313 | goto done_unlocked; |
247e743c CM |
2314 | } |
2315 | } | |
2316 | ||
11c8349b CM |
2317 | /* |
2318 | * we don't want to touch the inode after unlocking the page, | |
2319 | * so we update the mapping writeback index now | |
2320 | */ | |
2321 | update_nr_written(page, wbc, nr_written + 1); | |
771ed689 | 2322 | |
d1310b2e | 2323 | end = page_end; |
d1310b2e | 2324 | if (last_byte <= start) { |
e6dcd2dc CM |
2325 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2326 | tree->ops->writepage_end_io_hook(page, start, | |
2327 | page_end, NULL, 1); | |
2328 | unlock_start = page_end + 1; | |
d1310b2e CM |
2329 | goto done; |
2330 | } | |
2331 | ||
d1310b2e CM |
2332 | blocksize = inode->i_sb->s_blocksize; |
2333 | ||
2334 | while (cur <= end) { | |
2335 | if (cur >= last_byte) { | |
e6dcd2dc CM |
2336 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2337 | tree->ops->writepage_end_io_hook(page, cur, | |
2338 | page_end, NULL, 1); | |
2339 | unlock_start = page_end + 1; | |
d1310b2e CM |
2340 | break; |
2341 | } | |
7f3c74fb | 2342 | em = epd->get_extent(inode, page, pg_offset, cur, |
d1310b2e CM |
2343 | end - cur + 1, 1); |
2344 | if (IS_ERR(em) || !em) { | |
2345 | SetPageError(page); | |
2346 | break; | |
2347 | } | |
2348 | ||
2349 | extent_offset = cur - em->start; | |
2350 | BUG_ON(extent_map_end(em) <= cur); | |
2351 | BUG_ON(end < cur); | |
2352 | iosize = min(extent_map_end(em) - cur, end - cur + 1); | |
2353 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
2354 | sector = (em->block_start + extent_offset) >> 9; | |
2355 | bdev = em->bdev; | |
2356 | block_start = em->block_start; | |
c8b97818 | 2357 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
2358 | free_extent_map(em); |
2359 | em = NULL; | |
2360 | ||
c8b97818 CM |
2361 | /* |
2362 | * compressed and inline extents are written through other | |
2363 | * paths in the FS | |
2364 | */ | |
2365 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 2366 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
2367 | /* |
2368 | * end_io notification does not happen here for | |
2369 | * compressed extents | |
2370 | */ | |
2371 | if (!compressed && tree->ops && | |
2372 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
2373 | tree->ops->writepage_end_io_hook(page, cur, |
2374 | cur + iosize - 1, | |
2375 | NULL, 1); | |
c8b97818 CM |
2376 | else if (compressed) { |
2377 | /* we don't want to end_page_writeback on | |
2378 | * a compressed extent. this happens | |
2379 | * elsewhere | |
2380 | */ | |
2381 | nr++; | |
2382 | } | |
2383 | ||
2384 | cur += iosize; | |
7f3c74fb | 2385 | pg_offset += iosize; |
e6dcd2dc | 2386 | unlock_start = cur; |
d1310b2e CM |
2387 | continue; |
2388 | } | |
d1310b2e CM |
2389 | /* leave this out until we have a page_mkwrite call */ |
2390 | if (0 && !test_range_bit(tree, cur, cur + iosize - 1, | |
9655d298 | 2391 | EXTENT_DIRTY, 0, NULL)) { |
d1310b2e | 2392 | cur = cur + iosize; |
7f3c74fb | 2393 | pg_offset += iosize; |
d1310b2e CM |
2394 | continue; |
2395 | } | |
c8b97818 | 2396 | |
d1310b2e CM |
2397 | if (tree->ops && tree->ops->writepage_io_hook) { |
2398 | ret = tree->ops->writepage_io_hook(page, cur, | |
2399 | cur + iosize - 1); | |
2400 | } else { | |
2401 | ret = 0; | |
2402 | } | |
1259ab75 | 2403 | if (ret) { |
d1310b2e | 2404 | SetPageError(page); |
1259ab75 | 2405 | } else { |
d1310b2e | 2406 | unsigned long max_nr = end_index + 1; |
7f3c74fb | 2407 | |
d1310b2e CM |
2408 | set_range_writeback(tree, cur, cur + iosize - 1); |
2409 | if (!PageWriteback(page)) { | |
d397712b CM |
2410 | printk(KERN_ERR "btrfs warning page %lu not " |
2411 | "writeback, cur %llu end %llu\n", | |
2412 | page->index, (unsigned long long)cur, | |
d1310b2e CM |
2413 | (unsigned long long)end); |
2414 | } | |
2415 | ||
ffbd517d CM |
2416 | ret = submit_extent_page(write_flags, tree, page, |
2417 | sector, iosize, pg_offset, | |
2418 | bdev, &epd->bio, max_nr, | |
c8b97818 CM |
2419 | end_bio_extent_writepage, |
2420 | 0, 0, 0); | |
d1310b2e CM |
2421 | if (ret) |
2422 | SetPageError(page); | |
2423 | } | |
2424 | cur = cur + iosize; | |
7f3c74fb | 2425 | pg_offset += iosize; |
d1310b2e CM |
2426 | nr++; |
2427 | } | |
2428 | done: | |
2429 | if (nr == 0) { | |
2430 | /* make sure the mapping tag for page dirty gets cleared */ | |
2431 | set_page_writeback(page); | |
2432 | end_page_writeback(page); | |
2433 | } | |
d1310b2e | 2434 | unlock_page(page); |
771ed689 | 2435 | |
11c8349b CM |
2436 | done_unlocked: |
2437 | ||
2c64c53d CM |
2438 | /* drop our reference on any cached states */ |
2439 | free_extent_state(cached_state); | |
d1310b2e CM |
2440 | return 0; |
2441 | } | |
2442 | ||
d1310b2e | 2443 | /** |
4bef0848 | 2444 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
2445 | * @mapping: address space structure to write |
2446 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
2447 | * @writepage: function called for each page | |
2448 | * @data: data passed to writepage function | |
2449 | * | |
2450 | * If a page is already under I/O, write_cache_pages() skips it, even | |
2451 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
2452 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
2453 | * and msync() need to guarantee that all the data which was dirty at the time | |
2454 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
2455 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
2456 | * existing IO to complete. | |
2457 | */ | |
b2950863 | 2458 | static int extent_write_cache_pages(struct extent_io_tree *tree, |
4bef0848 CM |
2459 | struct address_space *mapping, |
2460 | struct writeback_control *wbc, | |
d2c3f4f6 CM |
2461 | writepage_t writepage, void *data, |
2462 | void (*flush_fn)(void *)) | |
d1310b2e | 2463 | { |
d1310b2e CM |
2464 | int ret = 0; |
2465 | int done = 0; | |
f85d7d6c | 2466 | int nr_to_write_done = 0; |
d1310b2e CM |
2467 | struct pagevec pvec; |
2468 | int nr_pages; | |
2469 | pgoff_t index; | |
2470 | pgoff_t end; /* Inclusive */ | |
2471 | int scanned = 0; | |
2472 | int range_whole = 0; | |
2473 | ||
d1310b2e CM |
2474 | pagevec_init(&pvec, 0); |
2475 | if (wbc->range_cyclic) { | |
2476 | index = mapping->writeback_index; /* Start from prev offset */ | |
2477 | end = -1; | |
2478 | } else { | |
2479 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
2480 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
2481 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) | |
2482 | range_whole = 1; | |
2483 | scanned = 1; | |
2484 | } | |
2485 | retry: | |
f85d7d6c | 2486 | while (!done && !nr_to_write_done && (index <= end) && |
d1310b2e | 2487 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, |
d397712b CM |
2488 | PAGECACHE_TAG_DIRTY, min(end - index, |
2489 | (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
d1310b2e CM |
2490 | unsigned i; |
2491 | ||
2492 | scanned = 1; | |
2493 | for (i = 0; i < nr_pages; i++) { | |
2494 | struct page *page = pvec.pages[i]; | |
2495 | ||
2496 | /* | |
2497 | * At this point we hold neither mapping->tree_lock nor | |
2498 | * lock on the page itself: the page may be truncated or | |
2499 | * invalidated (changing page->mapping to NULL), or even | |
2500 | * swizzled back from swapper_space to tmpfs file | |
2501 | * mapping | |
2502 | */ | |
4bef0848 CM |
2503 | if (tree->ops && tree->ops->write_cache_pages_lock_hook) |
2504 | tree->ops->write_cache_pages_lock_hook(page); | |
2505 | else | |
2506 | lock_page(page); | |
d1310b2e CM |
2507 | |
2508 | if (unlikely(page->mapping != mapping)) { | |
2509 | unlock_page(page); | |
2510 | continue; | |
2511 | } | |
2512 | ||
2513 | if (!wbc->range_cyclic && page->index > end) { | |
2514 | done = 1; | |
2515 | unlock_page(page); | |
2516 | continue; | |
2517 | } | |
2518 | ||
d2c3f4f6 | 2519 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 CM |
2520 | if (PageWriteback(page)) |
2521 | flush_fn(data); | |
d1310b2e | 2522 | wait_on_page_writeback(page); |
d2c3f4f6 | 2523 | } |
d1310b2e CM |
2524 | |
2525 | if (PageWriteback(page) || | |
2526 | !clear_page_dirty_for_io(page)) { | |
2527 | unlock_page(page); | |
2528 | continue; | |
2529 | } | |
2530 | ||
2531 | ret = (*writepage)(page, wbc, data); | |
2532 | ||
2533 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
2534 | unlock_page(page); | |
2535 | ret = 0; | |
2536 | } | |
f85d7d6c | 2537 | if (ret) |
d1310b2e | 2538 | done = 1; |
f85d7d6c CM |
2539 | |
2540 | /* | |
2541 | * the filesystem may choose to bump up nr_to_write. | |
2542 | * We have to make sure to honor the new nr_to_write | |
2543 | * at any time | |
2544 | */ | |
2545 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
2546 | } |
2547 | pagevec_release(&pvec); | |
2548 | cond_resched(); | |
2549 | } | |
2550 | if (!scanned && !done) { | |
2551 | /* | |
2552 | * We hit the last page and there is more work to be done: wrap | |
2553 | * back to the start of the file | |
2554 | */ | |
2555 | scanned = 1; | |
2556 | index = 0; | |
2557 | goto retry; | |
2558 | } | |
d1310b2e CM |
2559 | return ret; |
2560 | } | |
d1310b2e | 2561 | |
ffbd517d | 2562 | static void flush_epd_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 2563 | { |
d2c3f4f6 | 2564 | if (epd->bio) { |
ffbd517d CM |
2565 | if (epd->sync_io) |
2566 | submit_one_bio(WRITE_SYNC, epd->bio, 0, 0); | |
2567 | else | |
2568 | submit_one_bio(WRITE, epd->bio, 0, 0); | |
d2c3f4f6 CM |
2569 | epd->bio = NULL; |
2570 | } | |
2571 | } | |
2572 | ||
ffbd517d CM |
2573 | static noinline void flush_write_bio(void *data) |
2574 | { | |
2575 | struct extent_page_data *epd = data; | |
2576 | flush_epd_write_bio(epd); | |
2577 | } | |
2578 | ||
d1310b2e CM |
2579 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, |
2580 | get_extent_t *get_extent, | |
2581 | struct writeback_control *wbc) | |
2582 | { | |
2583 | int ret; | |
2584 | struct address_space *mapping = page->mapping; | |
2585 | struct extent_page_data epd = { | |
2586 | .bio = NULL, | |
2587 | .tree = tree, | |
2588 | .get_extent = get_extent, | |
771ed689 | 2589 | .extent_locked = 0, |
ffbd517d | 2590 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
2591 | }; |
2592 | struct writeback_control wbc_writepages = { | |
2593 | .bdi = wbc->bdi, | |
d313d7a3 | 2594 | .sync_mode = wbc->sync_mode, |
d1310b2e CM |
2595 | .older_than_this = NULL, |
2596 | .nr_to_write = 64, | |
2597 | .range_start = page_offset(page) + PAGE_CACHE_SIZE, | |
2598 | .range_end = (loff_t)-1, | |
2599 | }; | |
2600 | ||
d1310b2e CM |
2601 | ret = __extent_writepage(page, wbc, &epd); |
2602 | ||
4bef0848 | 2603 | extent_write_cache_pages(tree, mapping, &wbc_writepages, |
d2c3f4f6 | 2604 | __extent_writepage, &epd, flush_write_bio); |
ffbd517d | 2605 | flush_epd_write_bio(&epd); |
d1310b2e CM |
2606 | return ret; |
2607 | } | |
d1310b2e | 2608 | |
771ed689 CM |
2609 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, |
2610 | u64 start, u64 end, get_extent_t *get_extent, | |
2611 | int mode) | |
2612 | { | |
2613 | int ret = 0; | |
2614 | struct address_space *mapping = inode->i_mapping; | |
2615 | struct page *page; | |
2616 | unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >> | |
2617 | PAGE_CACHE_SHIFT; | |
2618 | ||
2619 | struct extent_page_data epd = { | |
2620 | .bio = NULL, | |
2621 | .tree = tree, | |
2622 | .get_extent = get_extent, | |
2623 | .extent_locked = 1, | |
ffbd517d | 2624 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
2625 | }; |
2626 | struct writeback_control wbc_writepages = { | |
2627 | .bdi = inode->i_mapping->backing_dev_info, | |
2628 | .sync_mode = mode, | |
2629 | .older_than_this = NULL, | |
2630 | .nr_to_write = nr_pages * 2, | |
2631 | .range_start = start, | |
2632 | .range_end = end + 1, | |
2633 | }; | |
2634 | ||
d397712b | 2635 | while (start <= end) { |
771ed689 CM |
2636 | page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); |
2637 | if (clear_page_dirty_for_io(page)) | |
2638 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
2639 | else { | |
2640 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
2641 | tree->ops->writepage_end_io_hook(page, start, | |
2642 | start + PAGE_CACHE_SIZE - 1, | |
2643 | NULL, 1); | |
2644 | unlock_page(page); | |
2645 | } | |
2646 | page_cache_release(page); | |
2647 | start += PAGE_CACHE_SIZE; | |
2648 | } | |
2649 | ||
ffbd517d | 2650 | flush_epd_write_bio(&epd); |
771ed689 CM |
2651 | return ret; |
2652 | } | |
d1310b2e CM |
2653 | |
2654 | int extent_writepages(struct extent_io_tree *tree, | |
2655 | struct address_space *mapping, | |
2656 | get_extent_t *get_extent, | |
2657 | struct writeback_control *wbc) | |
2658 | { | |
2659 | int ret = 0; | |
2660 | struct extent_page_data epd = { | |
2661 | .bio = NULL, | |
2662 | .tree = tree, | |
2663 | .get_extent = get_extent, | |
771ed689 | 2664 | .extent_locked = 0, |
ffbd517d | 2665 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
2666 | }; |
2667 | ||
4bef0848 | 2668 | ret = extent_write_cache_pages(tree, mapping, wbc, |
d2c3f4f6 CM |
2669 | __extent_writepage, &epd, |
2670 | flush_write_bio); | |
ffbd517d | 2671 | flush_epd_write_bio(&epd); |
d1310b2e CM |
2672 | return ret; |
2673 | } | |
d1310b2e CM |
2674 | |
2675 | int extent_readpages(struct extent_io_tree *tree, | |
2676 | struct address_space *mapping, | |
2677 | struct list_head *pages, unsigned nr_pages, | |
2678 | get_extent_t get_extent) | |
2679 | { | |
2680 | struct bio *bio = NULL; | |
2681 | unsigned page_idx; | |
2682 | struct pagevec pvec; | |
c8b97818 | 2683 | unsigned long bio_flags = 0; |
d1310b2e CM |
2684 | |
2685 | pagevec_init(&pvec, 0); | |
2686 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { | |
2687 | struct page *page = list_entry(pages->prev, struct page, lru); | |
2688 | ||
2689 | prefetchw(&page->flags); | |
2690 | list_del(&page->lru); | |
2691 | /* | |
2692 | * what we want to do here is call add_to_page_cache_lru, | |
2693 | * but that isn't exported, so we reproduce it here | |
2694 | */ | |
2695 | if (!add_to_page_cache(page, mapping, | |
2696 | page->index, GFP_KERNEL)) { | |
2697 | ||
2698 | /* open coding of lru_cache_add, also not exported */ | |
2699 | page_cache_get(page); | |
2700 | if (!pagevec_add(&pvec, page)) | |
15916de8 | 2701 | __pagevec_lru_add_file(&pvec); |
f188591e | 2702 | __extent_read_full_page(tree, page, get_extent, |
c8b97818 | 2703 | &bio, 0, &bio_flags); |
d1310b2e CM |
2704 | } |
2705 | page_cache_release(page); | |
2706 | } | |
2707 | if (pagevec_count(&pvec)) | |
15916de8 | 2708 | __pagevec_lru_add_file(&pvec); |
d1310b2e CM |
2709 | BUG_ON(!list_empty(pages)); |
2710 | if (bio) | |
c8b97818 | 2711 | submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
2712 | return 0; |
2713 | } | |
d1310b2e CM |
2714 | |
2715 | /* | |
2716 | * basic invalidatepage code, this waits on any locked or writeback | |
2717 | * ranges corresponding to the page, and then deletes any extent state | |
2718 | * records from the tree | |
2719 | */ | |
2720 | int extent_invalidatepage(struct extent_io_tree *tree, | |
2721 | struct page *page, unsigned long offset) | |
2722 | { | |
2ac55d41 | 2723 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
2724 | u64 start = ((u64)page->index << PAGE_CACHE_SHIFT); |
2725 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
2726 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
2727 | ||
d397712b | 2728 | start += (offset + blocksize - 1) & ~(blocksize - 1); |
d1310b2e CM |
2729 | if (start > end) |
2730 | return 0; | |
2731 | ||
2ac55d41 | 2732 | lock_extent_bits(tree, start, end, 0, &cached_state, GFP_NOFS); |
1edbb734 | 2733 | wait_on_page_writeback(page); |
d1310b2e | 2734 | clear_extent_bit(tree, start, end, |
32c00aff JB |
2735 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
2736 | EXTENT_DO_ACCOUNTING, | |
2ac55d41 | 2737 | 1, 1, &cached_state, GFP_NOFS); |
d1310b2e CM |
2738 | return 0; |
2739 | } | |
d1310b2e CM |
2740 | |
2741 | /* | |
2742 | * simple commit_write call, set_range_dirty is used to mark both | |
2743 | * the pages and the extent records as dirty | |
2744 | */ | |
2745 | int extent_commit_write(struct extent_io_tree *tree, | |
2746 | struct inode *inode, struct page *page, | |
2747 | unsigned from, unsigned to) | |
2748 | { | |
2749 | loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to; | |
2750 | ||
2751 | set_page_extent_mapped(page); | |
2752 | set_page_dirty(page); | |
2753 | ||
2754 | if (pos > inode->i_size) { | |
2755 | i_size_write(inode, pos); | |
2756 | mark_inode_dirty(inode); | |
2757 | } | |
2758 | return 0; | |
2759 | } | |
d1310b2e CM |
2760 | |
2761 | int extent_prepare_write(struct extent_io_tree *tree, | |
2762 | struct inode *inode, struct page *page, | |
2763 | unsigned from, unsigned to, get_extent_t *get_extent) | |
2764 | { | |
2765 | u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2766 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
2767 | u64 block_start; | |
2768 | u64 orig_block_start; | |
2769 | u64 block_end; | |
2770 | u64 cur_end; | |
2771 | struct extent_map *em; | |
2772 | unsigned blocksize = 1 << inode->i_blkbits; | |
2773 | size_t page_offset = 0; | |
2774 | size_t block_off_start; | |
2775 | size_t block_off_end; | |
2776 | int err = 0; | |
2777 | int iocount = 0; | |
2778 | int ret = 0; | |
2779 | int isnew; | |
2780 | ||
2781 | set_page_extent_mapped(page); | |
2782 | ||
2783 | block_start = (page_start + from) & ~((u64)blocksize - 1); | |
2784 | block_end = (page_start + to - 1) | (blocksize - 1); | |
2785 | orig_block_start = block_start; | |
2786 | ||
2787 | lock_extent(tree, page_start, page_end, GFP_NOFS); | |
d397712b | 2788 | while (block_start <= block_end) { |
d1310b2e CM |
2789 | em = get_extent(inode, page, page_offset, block_start, |
2790 | block_end - block_start + 1, 1); | |
d397712b | 2791 | if (IS_ERR(em) || !em) |
d1310b2e | 2792 | goto err; |
d397712b | 2793 | |
d1310b2e CM |
2794 | cur_end = min(block_end, extent_map_end(em) - 1); |
2795 | block_off_start = block_start & (PAGE_CACHE_SIZE - 1); | |
2796 | block_off_end = block_off_start + blocksize; | |
2797 | isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS); | |
2798 | ||
2799 | if (!PageUptodate(page) && isnew && | |
2800 | (block_off_end > to || block_off_start < from)) { | |
2801 | void *kaddr; | |
2802 | ||
2803 | kaddr = kmap_atomic(page, KM_USER0); | |
2804 | if (block_off_end > to) | |
2805 | memset(kaddr + to, 0, block_off_end - to); | |
2806 | if (block_off_start < from) | |
2807 | memset(kaddr + block_off_start, 0, | |
2808 | from - block_off_start); | |
2809 | flush_dcache_page(page); | |
2810 | kunmap_atomic(kaddr, KM_USER0); | |
2811 | } | |
2812 | if ((em->block_start != EXTENT_MAP_HOLE && | |
2813 | em->block_start != EXTENT_MAP_INLINE) && | |
2814 | !isnew && !PageUptodate(page) && | |
2815 | (block_off_end > to || block_off_start < from) && | |
2816 | !test_range_bit(tree, block_start, cur_end, | |
9655d298 | 2817 | EXTENT_UPTODATE, 1, NULL)) { |
d1310b2e CM |
2818 | u64 sector; |
2819 | u64 extent_offset = block_start - em->start; | |
2820 | size_t iosize; | |
2821 | sector = (em->block_start + extent_offset) >> 9; | |
2822 | iosize = (cur_end - block_start + blocksize) & | |
2823 | ~((u64)blocksize - 1); | |
2824 | /* | |
2825 | * we've already got the extent locked, but we | |
2826 | * need to split the state such that our end_bio | |
2827 | * handler can clear the lock. | |
2828 | */ | |
2829 | set_extent_bit(tree, block_start, | |
2830 | block_start + iosize - 1, | |
2c64c53d | 2831 | EXTENT_LOCKED, 0, NULL, NULL, GFP_NOFS); |
d1310b2e CM |
2832 | ret = submit_extent_page(READ, tree, page, |
2833 | sector, iosize, page_offset, em->bdev, | |
2834 | NULL, 1, | |
c8b97818 CM |
2835 | end_bio_extent_preparewrite, 0, |
2836 | 0, 0); | |
d1310b2e CM |
2837 | iocount++; |
2838 | block_start = block_start + iosize; | |
2839 | } else { | |
2840 | set_extent_uptodate(tree, block_start, cur_end, | |
2841 | GFP_NOFS); | |
2842 | unlock_extent(tree, block_start, cur_end, GFP_NOFS); | |
2843 | block_start = cur_end + 1; | |
2844 | } | |
2845 | page_offset = block_start & (PAGE_CACHE_SIZE - 1); | |
2846 | free_extent_map(em); | |
2847 | } | |
2848 | if (iocount) { | |
2849 | wait_extent_bit(tree, orig_block_start, | |
2850 | block_end, EXTENT_LOCKED); | |
2851 | } | |
2852 | check_page_uptodate(tree, page); | |
2853 | err: | |
2854 | /* FIXME, zero out newly allocated blocks on error */ | |
2855 | return err; | |
2856 | } | |
d1310b2e | 2857 | |
7b13b7b1 CM |
2858 | /* |
2859 | * a helper for releasepage, this tests for areas of the page that | |
2860 | * are locked or under IO and drops the related state bits if it is safe | |
2861 | * to drop the page. | |
2862 | */ | |
2863 | int try_release_extent_state(struct extent_map_tree *map, | |
2864 | struct extent_io_tree *tree, struct page *page, | |
2865 | gfp_t mask) | |
2866 | { | |
2867 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2868 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
2869 | int ret = 1; | |
2870 | ||
211f90e6 | 2871 | if (test_range_bit(tree, start, end, |
8b62b72b | 2872 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
2873 | ret = 0; |
2874 | else { | |
2875 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
2876 | mask = GFP_NOFS; | |
11ef160f CM |
2877 | /* |
2878 | * at this point we can safely clear everything except the | |
2879 | * locked bit and the nodatasum bit | |
2880 | */ | |
2881 | clear_extent_bit(tree, start, end, | |
2882 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), | |
2883 | 0, 0, NULL, mask); | |
7b13b7b1 CM |
2884 | } |
2885 | return ret; | |
2886 | } | |
7b13b7b1 | 2887 | |
d1310b2e CM |
2888 | /* |
2889 | * a helper for releasepage. As long as there are no locked extents | |
2890 | * in the range corresponding to the page, both state records and extent | |
2891 | * map records are removed | |
2892 | */ | |
2893 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
2894 | struct extent_io_tree *tree, struct page *page, |
2895 | gfp_t mask) | |
d1310b2e CM |
2896 | { |
2897 | struct extent_map *em; | |
2898 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2899 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
7b13b7b1 | 2900 | |
70dec807 CM |
2901 | if ((mask & __GFP_WAIT) && |
2902 | page->mapping->host->i_size > 16 * 1024 * 1024) { | |
39b5637f | 2903 | u64 len; |
70dec807 | 2904 | while (start <= end) { |
39b5637f | 2905 | len = end - start + 1; |
890871be | 2906 | write_lock(&map->lock); |
39b5637f | 2907 | em = lookup_extent_mapping(map, start, len); |
70dec807 | 2908 | if (!em || IS_ERR(em)) { |
890871be | 2909 | write_unlock(&map->lock); |
70dec807 CM |
2910 | break; |
2911 | } | |
7f3c74fb CM |
2912 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
2913 | em->start != start) { | |
890871be | 2914 | write_unlock(&map->lock); |
70dec807 CM |
2915 | free_extent_map(em); |
2916 | break; | |
2917 | } | |
2918 | if (!test_range_bit(tree, em->start, | |
2919 | extent_map_end(em) - 1, | |
8b62b72b | 2920 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 2921 | 0, NULL)) { |
70dec807 CM |
2922 | remove_extent_mapping(map, em); |
2923 | /* once for the rb tree */ | |
2924 | free_extent_map(em); | |
2925 | } | |
2926 | start = extent_map_end(em); | |
890871be | 2927 | write_unlock(&map->lock); |
70dec807 CM |
2928 | |
2929 | /* once for us */ | |
d1310b2e CM |
2930 | free_extent_map(em); |
2931 | } | |
d1310b2e | 2932 | } |
7b13b7b1 | 2933 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 2934 | } |
d1310b2e CM |
2935 | |
2936 | sector_t extent_bmap(struct address_space *mapping, sector_t iblock, | |
2937 | get_extent_t *get_extent) | |
2938 | { | |
2939 | struct inode *inode = mapping->host; | |
2ac55d41 | 2940 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
2941 | u64 start = iblock << inode->i_blkbits; |
2942 | sector_t sector = 0; | |
d899e052 | 2943 | size_t blksize = (1 << inode->i_blkbits); |
d1310b2e CM |
2944 | struct extent_map *em; |
2945 | ||
2ac55d41 JB |
2946 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + blksize - 1, |
2947 | 0, &cached_state, GFP_NOFS); | |
d899e052 | 2948 | em = get_extent(inode, NULL, 0, start, blksize, 0); |
2ac55d41 JB |
2949 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, |
2950 | start + blksize - 1, &cached_state, GFP_NOFS); | |
d1310b2e CM |
2951 | if (!em || IS_ERR(em)) |
2952 | return 0; | |
2953 | ||
d899e052 | 2954 | if (em->block_start > EXTENT_MAP_LAST_BYTE) |
d1310b2e CM |
2955 | goto out; |
2956 | ||
2957 | sector = (em->block_start + start - em->start) >> inode->i_blkbits; | |
d1310b2e CM |
2958 | out: |
2959 | free_extent_map(em); | |
2960 | return sector; | |
2961 | } | |
2962 | ||
1506fcc8 YS |
2963 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
2964 | __u64 start, __u64 len, get_extent_t *get_extent) | |
2965 | { | |
2966 | int ret; | |
2967 | u64 off = start; | |
2968 | u64 max = start + len; | |
2969 | u32 flags = 0; | |
2970 | u64 disko = 0; | |
2971 | struct extent_map *em = NULL; | |
2ac55d41 | 2972 | struct extent_state *cached_state = NULL; |
1506fcc8 YS |
2973 | int end = 0; |
2974 | u64 em_start = 0, em_len = 0; | |
2975 | unsigned long emflags; | |
2976 | ret = 0; | |
2977 | ||
2978 | if (len == 0) | |
2979 | return -EINVAL; | |
2980 | ||
2ac55d41 JB |
2981 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0, |
2982 | &cached_state, GFP_NOFS); | |
1506fcc8 YS |
2983 | em = get_extent(inode, NULL, 0, off, max - off, 0); |
2984 | if (!em) | |
2985 | goto out; | |
2986 | if (IS_ERR(em)) { | |
2987 | ret = PTR_ERR(em); | |
2988 | goto out; | |
2989 | } | |
2990 | while (!end) { | |
2991 | off = em->start + em->len; | |
2992 | if (off >= max) | |
2993 | end = 1; | |
2994 | ||
2995 | em_start = em->start; | |
2996 | em_len = em->len; | |
2997 | ||
2998 | disko = 0; | |
2999 | flags = 0; | |
3000 | ||
93dbfad7 | 3001 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
3002 | end = 1; |
3003 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 3004 | } else if (em->block_start == EXTENT_MAP_HOLE) { |
1506fcc8 | 3005 | flags |= FIEMAP_EXTENT_UNWRITTEN; |
93dbfad7 | 3006 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
3007 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
3008 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 3009 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
3010 | flags |= (FIEMAP_EXTENT_DELALLOC | |
3011 | FIEMAP_EXTENT_UNKNOWN); | |
93dbfad7 | 3012 | } else { |
1506fcc8 | 3013 | disko = em->block_start; |
1506fcc8 YS |
3014 | } |
3015 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
3016 | flags |= FIEMAP_EXTENT_ENCODED; | |
3017 | ||
3018 | emflags = em->flags; | |
3019 | free_extent_map(em); | |
3020 | em = NULL; | |
3021 | ||
3022 | if (!end) { | |
3023 | em = get_extent(inode, NULL, 0, off, max - off, 0); | |
3024 | if (!em) | |
3025 | goto out; | |
3026 | if (IS_ERR(em)) { | |
3027 | ret = PTR_ERR(em); | |
3028 | goto out; | |
3029 | } | |
3030 | emflags = em->flags; | |
3031 | } | |
3032 | if (test_bit(EXTENT_FLAG_VACANCY, &emflags)) { | |
3033 | flags |= FIEMAP_EXTENT_LAST; | |
3034 | end = 1; | |
3035 | } | |
3036 | ||
3037 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, | |
3038 | em_len, flags); | |
3039 | if (ret) | |
3040 | goto out_free; | |
3041 | } | |
3042 | out_free: | |
3043 | free_extent_map(em); | |
3044 | out: | |
2ac55d41 JB |
3045 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len, |
3046 | &cached_state, GFP_NOFS); | |
1506fcc8 YS |
3047 | return ret; |
3048 | } | |
3049 | ||
d1310b2e CM |
3050 | static inline struct page *extent_buffer_page(struct extent_buffer *eb, |
3051 | unsigned long i) | |
3052 | { | |
3053 | struct page *p; | |
3054 | struct address_space *mapping; | |
3055 | ||
3056 | if (i == 0) | |
3057 | return eb->first_page; | |
3058 | i += eb->start >> PAGE_CACHE_SHIFT; | |
3059 | mapping = eb->first_page->mapping; | |
33958dc6 CM |
3060 | if (!mapping) |
3061 | return NULL; | |
0ee0fda0 SW |
3062 | |
3063 | /* | |
3064 | * extent_buffer_page is only called after pinning the page | |
3065 | * by increasing the reference count. So we know the page must | |
3066 | * be in the radix tree. | |
3067 | */ | |
0ee0fda0 | 3068 | rcu_read_lock(); |
d1310b2e | 3069 | p = radix_tree_lookup(&mapping->page_tree, i); |
0ee0fda0 | 3070 | rcu_read_unlock(); |
2b1f55b0 | 3071 | |
d1310b2e CM |
3072 | return p; |
3073 | } | |
3074 | ||
6af118ce | 3075 | static inline unsigned long num_extent_pages(u64 start, u64 len) |
728131d8 | 3076 | { |
6af118ce CM |
3077 | return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) - |
3078 | (start >> PAGE_CACHE_SHIFT); | |
728131d8 CM |
3079 | } |
3080 | ||
d1310b2e CM |
3081 | static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree, |
3082 | u64 start, | |
3083 | unsigned long len, | |
3084 | gfp_t mask) | |
3085 | { | |
3086 | struct extent_buffer *eb = NULL; | |
3935127c | 3087 | #if LEAK_DEBUG |
2d2ae547 | 3088 | unsigned long flags; |
4bef0848 | 3089 | #endif |
d1310b2e | 3090 | |
d1310b2e | 3091 | eb = kmem_cache_zalloc(extent_buffer_cache, mask); |
d1310b2e CM |
3092 | eb->start = start; |
3093 | eb->len = len; | |
b4ce94de CM |
3094 | spin_lock_init(&eb->lock); |
3095 | init_waitqueue_head(&eb->lock_wq); | |
3096 | ||
3935127c | 3097 | #if LEAK_DEBUG |
2d2ae547 CM |
3098 | spin_lock_irqsave(&leak_lock, flags); |
3099 | list_add(&eb->leak_list, &buffers); | |
3100 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 3101 | #endif |
d1310b2e CM |
3102 | atomic_set(&eb->refs, 1); |
3103 | ||
3104 | return eb; | |
3105 | } | |
3106 | ||
3107 | static void __free_extent_buffer(struct extent_buffer *eb) | |
3108 | { | |
3935127c | 3109 | #if LEAK_DEBUG |
2d2ae547 CM |
3110 | unsigned long flags; |
3111 | spin_lock_irqsave(&leak_lock, flags); | |
3112 | list_del(&eb->leak_list); | |
3113 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 3114 | #endif |
d1310b2e CM |
3115 | kmem_cache_free(extent_buffer_cache, eb); |
3116 | } | |
3117 | ||
3118 | struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree, | |
3119 | u64 start, unsigned long len, | |
3120 | struct page *page0, | |
3121 | gfp_t mask) | |
3122 | { | |
3123 | unsigned long num_pages = num_extent_pages(start, len); | |
3124 | unsigned long i; | |
3125 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
3126 | struct extent_buffer *eb; | |
6af118ce | 3127 | struct extent_buffer *exists = NULL; |
d1310b2e CM |
3128 | struct page *p; |
3129 | struct address_space *mapping = tree->mapping; | |
3130 | int uptodate = 1; | |
3131 | ||
6af118ce CM |
3132 | spin_lock(&tree->buffer_lock); |
3133 | eb = buffer_search(tree, start); | |
3134 | if (eb) { | |
3135 | atomic_inc(&eb->refs); | |
3136 | spin_unlock(&tree->buffer_lock); | |
0f9dd46c | 3137 | mark_page_accessed(eb->first_page); |
6af118ce CM |
3138 | return eb; |
3139 | } | |
3140 | spin_unlock(&tree->buffer_lock); | |
3141 | ||
d1310b2e | 3142 | eb = __alloc_extent_buffer(tree, start, len, mask); |
2b114d1d | 3143 | if (!eb) |
d1310b2e CM |
3144 | return NULL; |
3145 | ||
d1310b2e CM |
3146 | if (page0) { |
3147 | eb->first_page = page0; | |
3148 | i = 1; | |
3149 | index++; | |
3150 | page_cache_get(page0); | |
3151 | mark_page_accessed(page0); | |
3152 | set_page_extent_mapped(page0); | |
d1310b2e | 3153 | set_page_extent_head(page0, len); |
f188591e | 3154 | uptodate = PageUptodate(page0); |
d1310b2e CM |
3155 | } else { |
3156 | i = 0; | |
3157 | } | |
3158 | for (; i < num_pages; i++, index++) { | |
3159 | p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM); | |
3160 | if (!p) { | |
3161 | WARN_ON(1); | |
6af118ce | 3162 | goto free_eb; |
d1310b2e CM |
3163 | } |
3164 | set_page_extent_mapped(p); | |
3165 | mark_page_accessed(p); | |
3166 | if (i == 0) { | |
3167 | eb->first_page = p; | |
3168 | set_page_extent_head(p, len); | |
3169 | } else { | |
3170 | set_page_private(p, EXTENT_PAGE_PRIVATE); | |
3171 | } | |
3172 | if (!PageUptodate(p)) | |
3173 | uptodate = 0; | |
3174 | unlock_page(p); | |
3175 | } | |
3176 | if (uptodate) | |
b4ce94de | 3177 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 3178 | |
6af118ce CM |
3179 | spin_lock(&tree->buffer_lock); |
3180 | exists = buffer_tree_insert(tree, start, &eb->rb_node); | |
3181 | if (exists) { | |
3182 | /* add one reference for the caller */ | |
3183 | atomic_inc(&exists->refs); | |
3184 | spin_unlock(&tree->buffer_lock); | |
3185 | goto free_eb; | |
3186 | } | |
6af118ce CM |
3187 | /* add one reference for the tree */ |
3188 | atomic_inc(&eb->refs); | |
f044ba78 | 3189 | spin_unlock(&tree->buffer_lock); |
d1310b2e CM |
3190 | return eb; |
3191 | ||
6af118ce | 3192 | free_eb: |
d1310b2e | 3193 | if (!atomic_dec_and_test(&eb->refs)) |
6af118ce CM |
3194 | return exists; |
3195 | for (index = 1; index < i; index++) | |
d1310b2e | 3196 | page_cache_release(extent_buffer_page(eb, index)); |
6af118ce | 3197 | page_cache_release(extent_buffer_page(eb, 0)); |
d1310b2e | 3198 | __free_extent_buffer(eb); |
6af118ce | 3199 | return exists; |
d1310b2e | 3200 | } |
d1310b2e CM |
3201 | |
3202 | struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree, | |
3203 | u64 start, unsigned long len, | |
3204 | gfp_t mask) | |
3205 | { | |
d1310b2e | 3206 | struct extent_buffer *eb; |
d1310b2e | 3207 | |
6af118ce CM |
3208 | spin_lock(&tree->buffer_lock); |
3209 | eb = buffer_search(tree, start); | |
3210 | if (eb) | |
3211 | atomic_inc(&eb->refs); | |
3212 | spin_unlock(&tree->buffer_lock); | |
d1310b2e | 3213 | |
0f9dd46c JB |
3214 | if (eb) |
3215 | mark_page_accessed(eb->first_page); | |
3216 | ||
d1310b2e | 3217 | return eb; |
d1310b2e | 3218 | } |
d1310b2e CM |
3219 | |
3220 | void free_extent_buffer(struct extent_buffer *eb) | |
3221 | { | |
d1310b2e CM |
3222 | if (!eb) |
3223 | return; | |
3224 | ||
3225 | if (!atomic_dec_and_test(&eb->refs)) | |
3226 | return; | |
3227 | ||
6af118ce | 3228 | WARN_ON(1); |
d1310b2e | 3229 | } |
d1310b2e CM |
3230 | |
3231 | int clear_extent_buffer_dirty(struct extent_io_tree *tree, | |
3232 | struct extent_buffer *eb) | |
3233 | { | |
d1310b2e CM |
3234 | unsigned long i; |
3235 | unsigned long num_pages; | |
3236 | struct page *page; | |
3237 | ||
d1310b2e CM |
3238 | num_pages = num_extent_pages(eb->start, eb->len); |
3239 | ||
3240 | for (i = 0; i < num_pages; i++) { | |
3241 | page = extent_buffer_page(eb, i); | |
b9473439 | 3242 | if (!PageDirty(page)) |
d2c3f4f6 CM |
3243 | continue; |
3244 | ||
a61e6f29 | 3245 | lock_page(page); |
d1310b2e CM |
3246 | if (i == 0) |
3247 | set_page_extent_head(page, eb->len); | |
3248 | else | |
3249 | set_page_private(page, EXTENT_PAGE_PRIVATE); | |
3250 | ||
d1310b2e | 3251 | clear_page_dirty_for_io(page); |
0ee0fda0 | 3252 | spin_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
3253 | if (!PageDirty(page)) { |
3254 | radix_tree_tag_clear(&page->mapping->page_tree, | |
3255 | page_index(page), | |
3256 | PAGECACHE_TAG_DIRTY); | |
3257 | } | |
0ee0fda0 | 3258 | spin_unlock_irq(&page->mapping->tree_lock); |
a61e6f29 | 3259 | unlock_page(page); |
d1310b2e CM |
3260 | } |
3261 | return 0; | |
3262 | } | |
d1310b2e CM |
3263 | |
3264 | int wait_on_extent_buffer_writeback(struct extent_io_tree *tree, | |
3265 | struct extent_buffer *eb) | |
3266 | { | |
3267 | return wait_on_extent_writeback(tree, eb->start, | |
3268 | eb->start + eb->len - 1); | |
3269 | } | |
d1310b2e CM |
3270 | |
3271 | int set_extent_buffer_dirty(struct extent_io_tree *tree, | |
3272 | struct extent_buffer *eb) | |
3273 | { | |
3274 | unsigned long i; | |
3275 | unsigned long num_pages; | |
b9473439 | 3276 | int was_dirty = 0; |
d1310b2e | 3277 | |
b9473439 | 3278 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
d1310b2e | 3279 | num_pages = num_extent_pages(eb->start, eb->len); |
b9473439 | 3280 | for (i = 0; i < num_pages; i++) |
d1310b2e | 3281 | __set_page_dirty_nobuffers(extent_buffer_page(eb, i)); |
b9473439 | 3282 | return was_dirty; |
d1310b2e | 3283 | } |
d1310b2e | 3284 | |
1259ab75 | 3285 | int clear_extent_buffer_uptodate(struct extent_io_tree *tree, |
2ac55d41 JB |
3286 | struct extent_buffer *eb, |
3287 | struct extent_state **cached_state) | |
1259ab75 CM |
3288 | { |
3289 | unsigned long i; | |
3290 | struct page *page; | |
3291 | unsigned long num_pages; | |
3292 | ||
3293 | num_pages = num_extent_pages(eb->start, eb->len); | |
b4ce94de | 3294 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
1259ab75 CM |
3295 | |
3296 | clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1, | |
2ac55d41 | 3297 | cached_state, GFP_NOFS); |
1259ab75 CM |
3298 | for (i = 0; i < num_pages; i++) { |
3299 | page = extent_buffer_page(eb, i); | |
33958dc6 CM |
3300 | if (page) |
3301 | ClearPageUptodate(page); | |
1259ab75 CM |
3302 | } |
3303 | return 0; | |
3304 | } | |
3305 | ||
d1310b2e CM |
3306 | int set_extent_buffer_uptodate(struct extent_io_tree *tree, |
3307 | struct extent_buffer *eb) | |
3308 | { | |
3309 | unsigned long i; | |
3310 | struct page *page; | |
3311 | unsigned long num_pages; | |
3312 | ||
3313 | num_pages = num_extent_pages(eb->start, eb->len); | |
3314 | ||
3315 | set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1, | |
3316 | GFP_NOFS); | |
3317 | for (i = 0; i < num_pages; i++) { | |
3318 | page = extent_buffer_page(eb, i); | |
3319 | if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) || | |
3320 | ((i == num_pages - 1) && | |
3321 | ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) { | |
3322 | check_page_uptodate(tree, page); | |
3323 | continue; | |
3324 | } | |
3325 | SetPageUptodate(page); | |
3326 | } | |
3327 | return 0; | |
3328 | } | |
d1310b2e | 3329 | |
ce9adaa5 CM |
3330 | int extent_range_uptodate(struct extent_io_tree *tree, |
3331 | u64 start, u64 end) | |
3332 | { | |
3333 | struct page *page; | |
3334 | int ret; | |
3335 | int pg_uptodate = 1; | |
3336 | int uptodate; | |
3337 | unsigned long index; | |
3338 | ||
9655d298 | 3339 | ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL); |
ce9adaa5 CM |
3340 | if (ret) |
3341 | return 1; | |
d397712b | 3342 | while (start <= end) { |
ce9adaa5 CM |
3343 | index = start >> PAGE_CACHE_SHIFT; |
3344 | page = find_get_page(tree->mapping, index); | |
3345 | uptodate = PageUptodate(page); | |
3346 | page_cache_release(page); | |
3347 | if (!uptodate) { | |
3348 | pg_uptodate = 0; | |
3349 | break; | |
3350 | } | |
3351 | start += PAGE_CACHE_SIZE; | |
3352 | } | |
3353 | return pg_uptodate; | |
3354 | } | |
3355 | ||
d1310b2e | 3356 | int extent_buffer_uptodate(struct extent_io_tree *tree, |
2ac55d41 JB |
3357 | struct extent_buffer *eb, |
3358 | struct extent_state *cached_state) | |
d1310b2e | 3359 | { |
728131d8 | 3360 | int ret = 0; |
ce9adaa5 CM |
3361 | unsigned long num_pages; |
3362 | unsigned long i; | |
728131d8 CM |
3363 | struct page *page; |
3364 | int pg_uptodate = 1; | |
3365 | ||
b4ce94de | 3366 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
4235298e | 3367 | return 1; |
728131d8 | 3368 | |
4235298e | 3369 | ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1, |
2ac55d41 | 3370 | EXTENT_UPTODATE, 1, cached_state); |
4235298e CM |
3371 | if (ret) |
3372 | return ret; | |
728131d8 CM |
3373 | |
3374 | num_pages = num_extent_pages(eb->start, eb->len); | |
3375 | for (i = 0; i < num_pages; i++) { | |
3376 | page = extent_buffer_page(eb, i); | |
3377 | if (!PageUptodate(page)) { | |
3378 | pg_uptodate = 0; | |
3379 | break; | |
3380 | } | |
3381 | } | |
4235298e | 3382 | return pg_uptodate; |
d1310b2e | 3383 | } |
d1310b2e CM |
3384 | |
3385 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
3386 | struct extent_buffer *eb, | |
a86c12c7 | 3387 | u64 start, int wait, |
f188591e | 3388 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3389 | { |
3390 | unsigned long i; | |
3391 | unsigned long start_i; | |
3392 | struct page *page; | |
3393 | int err; | |
3394 | int ret = 0; | |
ce9adaa5 CM |
3395 | int locked_pages = 0; |
3396 | int all_uptodate = 1; | |
3397 | int inc_all_pages = 0; | |
d1310b2e | 3398 | unsigned long num_pages; |
a86c12c7 | 3399 | struct bio *bio = NULL; |
c8b97818 | 3400 | unsigned long bio_flags = 0; |
a86c12c7 | 3401 | |
b4ce94de | 3402 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
3403 | return 0; |
3404 | ||
ce9adaa5 | 3405 | if (test_range_bit(tree, eb->start, eb->start + eb->len - 1, |
9655d298 | 3406 | EXTENT_UPTODATE, 1, NULL)) { |
d1310b2e CM |
3407 | return 0; |
3408 | } | |
3409 | ||
3410 | if (start) { | |
3411 | WARN_ON(start < eb->start); | |
3412 | start_i = (start >> PAGE_CACHE_SHIFT) - | |
3413 | (eb->start >> PAGE_CACHE_SHIFT); | |
3414 | } else { | |
3415 | start_i = 0; | |
3416 | } | |
3417 | ||
3418 | num_pages = num_extent_pages(eb->start, eb->len); | |
3419 | for (i = start_i; i < num_pages; i++) { | |
3420 | page = extent_buffer_page(eb, i); | |
d1310b2e | 3421 | if (!wait) { |
2db04966 | 3422 | if (!trylock_page(page)) |
ce9adaa5 | 3423 | goto unlock_exit; |
d1310b2e CM |
3424 | } else { |
3425 | lock_page(page); | |
3426 | } | |
ce9adaa5 | 3427 | locked_pages++; |
d397712b | 3428 | if (!PageUptodate(page)) |
ce9adaa5 | 3429 | all_uptodate = 0; |
ce9adaa5 CM |
3430 | } |
3431 | if (all_uptodate) { | |
3432 | if (start_i == 0) | |
b4ce94de | 3433 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
3434 | goto unlock_exit; |
3435 | } | |
3436 | ||
3437 | for (i = start_i; i < num_pages; i++) { | |
3438 | page = extent_buffer_page(eb, i); | |
3439 | if (inc_all_pages) | |
3440 | page_cache_get(page); | |
3441 | if (!PageUptodate(page)) { | |
3442 | if (start_i == 0) | |
3443 | inc_all_pages = 1; | |
f188591e | 3444 | ClearPageError(page); |
a86c12c7 | 3445 | err = __extent_read_full_page(tree, page, |
f188591e | 3446 | get_extent, &bio, |
c8b97818 | 3447 | mirror_num, &bio_flags); |
d397712b | 3448 | if (err) |
d1310b2e | 3449 | ret = err; |
d1310b2e CM |
3450 | } else { |
3451 | unlock_page(page); | |
3452 | } | |
3453 | } | |
3454 | ||
a86c12c7 | 3455 | if (bio) |
c8b97818 | 3456 | submit_one_bio(READ, bio, mirror_num, bio_flags); |
a86c12c7 | 3457 | |
d397712b | 3458 | if (ret || !wait) |
d1310b2e | 3459 | return ret; |
d397712b | 3460 | |
d1310b2e CM |
3461 | for (i = start_i; i < num_pages; i++) { |
3462 | page = extent_buffer_page(eb, i); | |
3463 | wait_on_page_locked(page); | |
d397712b | 3464 | if (!PageUptodate(page)) |
d1310b2e | 3465 | ret = -EIO; |
d1310b2e | 3466 | } |
d397712b | 3467 | |
d1310b2e | 3468 | if (!ret) |
b4ce94de | 3469 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 3470 | return ret; |
ce9adaa5 CM |
3471 | |
3472 | unlock_exit: | |
3473 | i = start_i; | |
d397712b | 3474 | while (locked_pages > 0) { |
ce9adaa5 CM |
3475 | page = extent_buffer_page(eb, i); |
3476 | i++; | |
3477 | unlock_page(page); | |
3478 | locked_pages--; | |
3479 | } | |
3480 | return ret; | |
d1310b2e | 3481 | } |
d1310b2e CM |
3482 | |
3483 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
3484 | unsigned long start, | |
3485 | unsigned long len) | |
3486 | { | |
3487 | size_t cur; | |
3488 | size_t offset; | |
3489 | struct page *page; | |
3490 | char *kaddr; | |
3491 | char *dst = (char *)dstv; | |
3492 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3493 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
3494 | |
3495 | WARN_ON(start > eb->len); | |
3496 | WARN_ON(start + len > eb->start + eb->len); | |
3497 | ||
3498 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3499 | ||
d397712b | 3500 | while (len > 0) { |
d1310b2e | 3501 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
3502 | |
3503 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
3504 | kaddr = kmap_atomic(page, KM_USER1); | |
3505 | memcpy(dst, kaddr + offset, cur); | |
3506 | kunmap_atomic(kaddr, KM_USER1); | |
3507 | ||
3508 | dst += cur; | |
3509 | len -= cur; | |
3510 | offset = 0; | |
3511 | i++; | |
3512 | } | |
3513 | } | |
d1310b2e CM |
3514 | |
3515 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
3516 | unsigned long min_len, char **token, char **map, | |
3517 | unsigned long *map_start, | |
3518 | unsigned long *map_len, int km) | |
3519 | { | |
3520 | size_t offset = start & (PAGE_CACHE_SIZE - 1); | |
3521 | char *kaddr; | |
3522 | struct page *p; | |
3523 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3524 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3525 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
3526 | PAGE_CACHE_SHIFT; | |
3527 | ||
3528 | if (i != end_i) | |
3529 | return -EINVAL; | |
3530 | ||
3531 | if (i == 0) { | |
3532 | offset = start_offset; | |
3533 | *map_start = 0; | |
3534 | } else { | |
3535 | offset = 0; | |
3536 | *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; | |
3537 | } | |
d397712b | 3538 | |
d1310b2e | 3539 | if (start + min_len > eb->len) { |
d397712b CM |
3540 | printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, " |
3541 | "wanted %lu %lu\n", (unsigned long long)eb->start, | |
3542 | eb->len, start, min_len); | |
d1310b2e CM |
3543 | WARN_ON(1); |
3544 | } | |
3545 | ||
3546 | p = extent_buffer_page(eb, i); | |
d1310b2e CM |
3547 | kaddr = kmap_atomic(p, km); |
3548 | *token = kaddr; | |
3549 | *map = kaddr + offset; | |
3550 | *map_len = PAGE_CACHE_SIZE - offset; | |
3551 | return 0; | |
3552 | } | |
d1310b2e CM |
3553 | |
3554 | int map_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
3555 | unsigned long min_len, | |
3556 | char **token, char **map, | |
3557 | unsigned long *map_start, | |
3558 | unsigned long *map_len, int km) | |
3559 | { | |
3560 | int err; | |
3561 | int save = 0; | |
3562 | if (eb->map_token) { | |
3563 | unmap_extent_buffer(eb, eb->map_token, km); | |
3564 | eb->map_token = NULL; | |
3565 | save = 1; | |
3566 | } | |
3567 | err = map_private_extent_buffer(eb, start, min_len, token, map, | |
3568 | map_start, map_len, km); | |
3569 | if (!err && save) { | |
3570 | eb->map_token = *token; | |
3571 | eb->kaddr = *map; | |
3572 | eb->map_start = *map_start; | |
3573 | eb->map_len = *map_len; | |
3574 | } | |
3575 | return err; | |
3576 | } | |
d1310b2e CM |
3577 | |
3578 | void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km) | |
3579 | { | |
3580 | kunmap_atomic(token, km); | |
3581 | } | |
d1310b2e CM |
3582 | |
3583 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, | |
3584 | unsigned long start, | |
3585 | unsigned long len) | |
3586 | { | |
3587 | size_t cur; | |
3588 | size_t offset; | |
3589 | struct page *page; | |
3590 | char *kaddr; | |
3591 | char *ptr = (char *)ptrv; | |
3592 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3593 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3594 | int ret = 0; | |
3595 | ||
3596 | WARN_ON(start > eb->len); | |
3597 | WARN_ON(start + len > eb->start + eb->len); | |
3598 | ||
3599 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3600 | ||
d397712b | 3601 | while (len > 0) { |
d1310b2e | 3602 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
3603 | |
3604 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
3605 | ||
3606 | kaddr = kmap_atomic(page, KM_USER0); | |
3607 | ret = memcmp(ptr, kaddr + offset, cur); | |
3608 | kunmap_atomic(kaddr, KM_USER0); | |
3609 | if (ret) | |
3610 | break; | |
3611 | ||
3612 | ptr += cur; | |
3613 | len -= cur; | |
3614 | offset = 0; | |
3615 | i++; | |
3616 | } | |
3617 | return ret; | |
3618 | } | |
d1310b2e CM |
3619 | |
3620 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
3621 | unsigned long start, unsigned long len) | |
3622 | { | |
3623 | size_t cur; | |
3624 | size_t offset; | |
3625 | struct page *page; | |
3626 | char *kaddr; | |
3627 | char *src = (char *)srcv; | |
3628 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3629 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3630 | ||
3631 | WARN_ON(start > eb->len); | |
3632 | WARN_ON(start + len > eb->start + eb->len); | |
3633 | ||
3634 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3635 | ||
d397712b | 3636 | while (len > 0) { |
d1310b2e CM |
3637 | page = extent_buffer_page(eb, i); |
3638 | WARN_ON(!PageUptodate(page)); | |
3639 | ||
3640 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
3641 | kaddr = kmap_atomic(page, KM_USER1); | |
3642 | memcpy(kaddr + offset, src, cur); | |
3643 | kunmap_atomic(kaddr, KM_USER1); | |
3644 | ||
3645 | src += cur; | |
3646 | len -= cur; | |
3647 | offset = 0; | |
3648 | i++; | |
3649 | } | |
3650 | } | |
d1310b2e CM |
3651 | |
3652 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
3653 | unsigned long start, unsigned long len) | |
3654 | { | |
3655 | size_t cur; | |
3656 | size_t offset; | |
3657 | struct page *page; | |
3658 | char *kaddr; | |
3659 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3660 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3661 | ||
3662 | WARN_ON(start > eb->len); | |
3663 | WARN_ON(start + len > eb->start + eb->len); | |
3664 | ||
3665 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3666 | ||
d397712b | 3667 | while (len > 0) { |
d1310b2e CM |
3668 | page = extent_buffer_page(eb, i); |
3669 | WARN_ON(!PageUptodate(page)); | |
3670 | ||
3671 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
3672 | kaddr = kmap_atomic(page, KM_USER0); | |
3673 | memset(kaddr + offset, c, cur); | |
3674 | kunmap_atomic(kaddr, KM_USER0); | |
3675 | ||
3676 | len -= cur; | |
3677 | offset = 0; | |
3678 | i++; | |
3679 | } | |
3680 | } | |
d1310b2e CM |
3681 | |
3682 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
3683 | unsigned long dst_offset, unsigned long src_offset, | |
3684 | unsigned long len) | |
3685 | { | |
3686 | u64 dst_len = dst->len; | |
3687 | size_t cur; | |
3688 | size_t offset; | |
3689 | struct page *page; | |
3690 | char *kaddr; | |
3691 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3692 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
3693 | ||
3694 | WARN_ON(src->len != dst_len); | |
3695 | ||
3696 | offset = (start_offset + dst_offset) & | |
3697 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3698 | ||
d397712b | 3699 | while (len > 0) { |
d1310b2e CM |
3700 | page = extent_buffer_page(dst, i); |
3701 | WARN_ON(!PageUptodate(page)); | |
3702 | ||
3703 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
3704 | ||
3705 | kaddr = kmap_atomic(page, KM_USER0); | |
3706 | read_extent_buffer(src, kaddr + offset, src_offset, cur); | |
3707 | kunmap_atomic(kaddr, KM_USER0); | |
3708 | ||
3709 | src_offset += cur; | |
3710 | len -= cur; | |
3711 | offset = 0; | |
3712 | i++; | |
3713 | } | |
3714 | } | |
d1310b2e CM |
3715 | |
3716 | static void move_pages(struct page *dst_page, struct page *src_page, | |
3717 | unsigned long dst_off, unsigned long src_off, | |
3718 | unsigned long len) | |
3719 | { | |
3720 | char *dst_kaddr = kmap_atomic(dst_page, KM_USER0); | |
3721 | if (dst_page == src_page) { | |
3722 | memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len); | |
3723 | } else { | |
3724 | char *src_kaddr = kmap_atomic(src_page, KM_USER1); | |
3725 | char *p = dst_kaddr + dst_off + len; | |
3726 | char *s = src_kaddr + src_off + len; | |
3727 | ||
3728 | while (len--) | |
3729 | *--p = *--s; | |
3730 | ||
3731 | kunmap_atomic(src_kaddr, KM_USER1); | |
3732 | } | |
3733 | kunmap_atomic(dst_kaddr, KM_USER0); | |
3734 | } | |
3735 | ||
3736 | static void copy_pages(struct page *dst_page, struct page *src_page, | |
3737 | unsigned long dst_off, unsigned long src_off, | |
3738 | unsigned long len) | |
3739 | { | |
3740 | char *dst_kaddr = kmap_atomic(dst_page, KM_USER0); | |
3741 | char *src_kaddr; | |
3742 | ||
3743 | if (dst_page != src_page) | |
3744 | src_kaddr = kmap_atomic(src_page, KM_USER1); | |
3745 | else | |
3746 | src_kaddr = dst_kaddr; | |
3747 | ||
3748 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
3749 | kunmap_atomic(dst_kaddr, KM_USER0); | |
3750 | if (dst_page != src_page) | |
3751 | kunmap_atomic(src_kaddr, KM_USER1); | |
3752 | } | |
3753 | ||
3754 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
3755 | unsigned long src_offset, unsigned long len) | |
3756 | { | |
3757 | size_t cur; | |
3758 | size_t dst_off_in_page; | |
3759 | size_t src_off_in_page; | |
3760 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3761 | unsigned long dst_i; | |
3762 | unsigned long src_i; | |
3763 | ||
3764 | if (src_offset + len > dst->len) { | |
d397712b CM |
3765 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
3766 | "len %lu dst len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
3767 | BUG_ON(1); |
3768 | } | |
3769 | if (dst_offset + len > dst->len) { | |
d397712b CM |
3770 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
3771 | "len %lu dst len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
3772 | BUG_ON(1); |
3773 | } | |
3774 | ||
d397712b | 3775 | while (len > 0) { |
d1310b2e CM |
3776 | dst_off_in_page = (start_offset + dst_offset) & |
3777 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3778 | src_off_in_page = (start_offset + src_offset) & | |
3779 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3780 | ||
3781 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
3782 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
3783 | ||
3784 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
3785 | src_off_in_page)); | |
3786 | cur = min_t(unsigned long, cur, | |
3787 | (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); | |
3788 | ||
3789 | copy_pages(extent_buffer_page(dst, dst_i), | |
3790 | extent_buffer_page(dst, src_i), | |
3791 | dst_off_in_page, src_off_in_page, cur); | |
3792 | ||
3793 | src_offset += cur; | |
3794 | dst_offset += cur; | |
3795 | len -= cur; | |
3796 | } | |
3797 | } | |
d1310b2e CM |
3798 | |
3799 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
3800 | unsigned long src_offset, unsigned long len) | |
3801 | { | |
3802 | size_t cur; | |
3803 | size_t dst_off_in_page; | |
3804 | size_t src_off_in_page; | |
3805 | unsigned long dst_end = dst_offset + len - 1; | |
3806 | unsigned long src_end = src_offset + len - 1; | |
3807 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3808 | unsigned long dst_i; | |
3809 | unsigned long src_i; | |
3810 | ||
3811 | if (src_offset + len > dst->len) { | |
d397712b CM |
3812 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
3813 | "len %lu len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
3814 | BUG_ON(1); |
3815 | } | |
3816 | if (dst_offset + len > dst->len) { | |
d397712b CM |
3817 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
3818 | "len %lu len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
3819 | BUG_ON(1); |
3820 | } | |
3821 | if (dst_offset < src_offset) { | |
3822 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); | |
3823 | return; | |
3824 | } | |
d397712b | 3825 | while (len > 0) { |
d1310b2e CM |
3826 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; |
3827 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
3828 | ||
3829 | dst_off_in_page = (start_offset + dst_end) & | |
3830 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3831 | src_off_in_page = (start_offset + src_end) & | |
3832 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3833 | ||
3834 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
3835 | cur = min(cur, dst_off_in_page + 1); | |
3836 | move_pages(extent_buffer_page(dst, dst_i), | |
3837 | extent_buffer_page(dst, src_i), | |
3838 | dst_off_in_page - cur + 1, | |
3839 | src_off_in_page - cur + 1, cur); | |
3840 | ||
3841 | dst_end -= cur; | |
3842 | src_end -= cur; | |
3843 | len -= cur; | |
3844 | } | |
3845 | } | |
6af118ce CM |
3846 | |
3847 | int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page) | |
3848 | { | |
3849 | u64 start = page_offset(page); | |
3850 | struct extent_buffer *eb; | |
3851 | int ret = 1; | |
3852 | unsigned long i; | |
3853 | unsigned long num_pages; | |
3854 | ||
3855 | spin_lock(&tree->buffer_lock); | |
3856 | eb = buffer_search(tree, start); | |
3857 | if (!eb) | |
3858 | goto out; | |
3859 | ||
3860 | if (atomic_read(&eb->refs) > 1) { | |
3861 | ret = 0; | |
3862 | goto out; | |
3863 | } | |
b9473439 CM |
3864 | if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3865 | ret = 0; | |
3866 | goto out; | |
3867 | } | |
6af118ce CM |
3868 | /* at this point we can safely release the extent buffer */ |
3869 | num_pages = num_extent_pages(eb->start, eb->len); | |
b214107e CH |
3870 | for (i = 0; i < num_pages; i++) |
3871 | page_cache_release(extent_buffer_page(eb, i)); | |
6af118ce CM |
3872 | rb_erase(&eb->rb_node, &tree->buffer); |
3873 | __free_extent_buffer(eb); | |
3874 | out: | |
3875 | spin_unlock(&tree->buffer_lock); | |
3876 | return ret; | |
3877 | } |