Commit | Line | Data |
---|---|---|
d1310b2e CM |
1 | #include <linux/bitops.h> |
2 | #include <linux/slab.h> | |
3 | #include <linux/bio.h> | |
4 | #include <linux/mm.h> | |
d1310b2e CM |
5 | #include <linux/pagemap.h> |
6 | #include <linux/page-flags.h> | |
7 | #include <linux/module.h> | |
8 | #include <linux/spinlock.h> | |
9 | #include <linux/blkdev.h> | |
10 | #include <linux/swap.h> | |
d1310b2e CM |
11 | #include <linux/writeback.h> |
12 | #include <linux/pagevec.h> | |
13 | #include "extent_io.h" | |
14 | #include "extent_map.h" | |
2db04966 | 15 | #include "compat.h" |
902b22f3 DW |
16 | #include "ctree.h" |
17 | #include "btrfs_inode.h" | |
d1310b2e | 18 | |
d1310b2e CM |
19 | static struct kmem_cache *extent_state_cache; |
20 | static struct kmem_cache *extent_buffer_cache; | |
21 | ||
22 | static LIST_HEAD(buffers); | |
23 | static LIST_HEAD(states); | |
4bef0848 | 24 | |
b47eda86 | 25 | #define LEAK_DEBUG 0 |
3935127c | 26 | #if LEAK_DEBUG |
d397712b | 27 | static DEFINE_SPINLOCK(leak_lock); |
4bef0848 | 28 | #endif |
d1310b2e | 29 | |
d1310b2e CM |
30 | #define BUFFER_LRU_MAX 64 |
31 | ||
32 | struct tree_entry { | |
33 | u64 start; | |
34 | u64 end; | |
d1310b2e CM |
35 | struct rb_node rb_node; |
36 | }; | |
37 | ||
38 | struct extent_page_data { | |
39 | struct bio *bio; | |
40 | struct extent_io_tree *tree; | |
41 | get_extent_t *get_extent; | |
771ed689 CM |
42 | |
43 | /* tells writepage not to lock the state bits for this range | |
44 | * it still does the unlocking | |
45 | */ | |
ffbd517d CM |
46 | unsigned int extent_locked:1; |
47 | ||
48 | /* tells the submit_bio code to use a WRITE_SYNC */ | |
49 | unsigned int sync_io:1; | |
d1310b2e CM |
50 | }; |
51 | ||
52 | int __init extent_io_init(void) | |
53 | { | |
9601e3f6 CH |
54 | extent_state_cache = kmem_cache_create("extent_state", |
55 | sizeof(struct extent_state), 0, | |
56 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
57 | if (!extent_state_cache) |
58 | return -ENOMEM; | |
59 | ||
9601e3f6 CH |
60 | extent_buffer_cache = kmem_cache_create("extent_buffers", |
61 | sizeof(struct extent_buffer), 0, | |
62 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
63 | if (!extent_buffer_cache) |
64 | goto free_state_cache; | |
65 | return 0; | |
66 | ||
67 | free_state_cache: | |
68 | kmem_cache_destroy(extent_state_cache); | |
69 | return -ENOMEM; | |
70 | } | |
71 | ||
72 | void extent_io_exit(void) | |
73 | { | |
74 | struct extent_state *state; | |
2d2ae547 | 75 | struct extent_buffer *eb; |
d1310b2e CM |
76 | |
77 | while (!list_empty(&states)) { | |
2d2ae547 | 78 | state = list_entry(states.next, struct extent_state, leak_list); |
d397712b CM |
79 | printk(KERN_ERR "btrfs state leak: start %llu end %llu " |
80 | "state %lu in tree %p refs %d\n", | |
81 | (unsigned long long)state->start, | |
82 | (unsigned long long)state->end, | |
83 | state->state, state->tree, atomic_read(&state->refs)); | |
2d2ae547 | 84 | list_del(&state->leak_list); |
d1310b2e CM |
85 | kmem_cache_free(extent_state_cache, state); |
86 | ||
87 | } | |
88 | ||
2d2ae547 CM |
89 | while (!list_empty(&buffers)) { |
90 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
d397712b CM |
91 | printk(KERN_ERR "btrfs buffer leak start %llu len %lu " |
92 | "refs %d\n", (unsigned long long)eb->start, | |
93 | eb->len, atomic_read(&eb->refs)); | |
2d2ae547 CM |
94 | list_del(&eb->leak_list); |
95 | kmem_cache_free(extent_buffer_cache, eb); | |
96 | } | |
d1310b2e CM |
97 | if (extent_state_cache) |
98 | kmem_cache_destroy(extent_state_cache); | |
99 | if (extent_buffer_cache) | |
100 | kmem_cache_destroy(extent_buffer_cache); | |
101 | } | |
102 | ||
103 | void extent_io_tree_init(struct extent_io_tree *tree, | |
104 | struct address_space *mapping, gfp_t mask) | |
105 | { | |
6bef4d31 EP |
106 | tree->state = RB_ROOT; |
107 | tree->buffer = RB_ROOT; | |
d1310b2e CM |
108 | tree->ops = NULL; |
109 | tree->dirty_bytes = 0; | |
70dec807 | 110 | spin_lock_init(&tree->lock); |
6af118ce | 111 | spin_lock_init(&tree->buffer_lock); |
d1310b2e | 112 | tree->mapping = mapping; |
d1310b2e | 113 | } |
d1310b2e | 114 | |
b2950863 | 115 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
116 | { |
117 | struct extent_state *state; | |
3935127c | 118 | #if LEAK_DEBUG |
2d2ae547 | 119 | unsigned long flags; |
4bef0848 | 120 | #endif |
d1310b2e CM |
121 | |
122 | state = kmem_cache_alloc(extent_state_cache, mask); | |
2b114d1d | 123 | if (!state) |
d1310b2e CM |
124 | return state; |
125 | state->state = 0; | |
d1310b2e | 126 | state->private = 0; |
70dec807 | 127 | state->tree = NULL; |
3935127c | 128 | #if LEAK_DEBUG |
2d2ae547 CM |
129 | spin_lock_irqsave(&leak_lock, flags); |
130 | list_add(&state->leak_list, &states); | |
131 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 132 | #endif |
d1310b2e CM |
133 | atomic_set(&state->refs, 1); |
134 | init_waitqueue_head(&state->wq); | |
135 | return state; | |
136 | } | |
d1310b2e | 137 | |
4845e44f | 138 | void free_extent_state(struct extent_state *state) |
d1310b2e | 139 | { |
d1310b2e CM |
140 | if (!state) |
141 | return; | |
142 | if (atomic_dec_and_test(&state->refs)) { | |
3935127c | 143 | #if LEAK_DEBUG |
2d2ae547 | 144 | unsigned long flags; |
4bef0848 | 145 | #endif |
70dec807 | 146 | WARN_ON(state->tree); |
3935127c | 147 | #if LEAK_DEBUG |
2d2ae547 CM |
148 | spin_lock_irqsave(&leak_lock, flags); |
149 | list_del(&state->leak_list); | |
150 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 151 | #endif |
d1310b2e CM |
152 | kmem_cache_free(extent_state_cache, state); |
153 | } | |
154 | } | |
d1310b2e CM |
155 | |
156 | static struct rb_node *tree_insert(struct rb_root *root, u64 offset, | |
157 | struct rb_node *node) | |
158 | { | |
d397712b CM |
159 | struct rb_node **p = &root->rb_node; |
160 | struct rb_node *parent = NULL; | |
d1310b2e CM |
161 | struct tree_entry *entry; |
162 | ||
d397712b | 163 | while (*p) { |
d1310b2e CM |
164 | parent = *p; |
165 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
166 | ||
167 | if (offset < entry->start) | |
168 | p = &(*p)->rb_left; | |
169 | else if (offset > entry->end) | |
170 | p = &(*p)->rb_right; | |
171 | else | |
172 | return parent; | |
173 | } | |
174 | ||
175 | entry = rb_entry(node, struct tree_entry, rb_node); | |
d1310b2e CM |
176 | rb_link_node(node, parent, p); |
177 | rb_insert_color(node, root); | |
178 | return NULL; | |
179 | } | |
180 | ||
80ea96b1 | 181 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
d1310b2e CM |
182 | struct rb_node **prev_ret, |
183 | struct rb_node **next_ret) | |
184 | { | |
80ea96b1 | 185 | struct rb_root *root = &tree->state; |
d397712b | 186 | struct rb_node *n = root->rb_node; |
d1310b2e CM |
187 | struct rb_node *prev = NULL; |
188 | struct rb_node *orig_prev = NULL; | |
189 | struct tree_entry *entry; | |
190 | struct tree_entry *prev_entry = NULL; | |
191 | ||
d397712b | 192 | while (n) { |
d1310b2e CM |
193 | entry = rb_entry(n, struct tree_entry, rb_node); |
194 | prev = n; | |
195 | prev_entry = entry; | |
196 | ||
197 | if (offset < entry->start) | |
198 | n = n->rb_left; | |
199 | else if (offset > entry->end) | |
200 | n = n->rb_right; | |
d397712b | 201 | else |
d1310b2e CM |
202 | return n; |
203 | } | |
204 | ||
205 | if (prev_ret) { | |
206 | orig_prev = prev; | |
d397712b | 207 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
208 | prev = rb_next(prev); |
209 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
210 | } | |
211 | *prev_ret = prev; | |
212 | prev = orig_prev; | |
213 | } | |
214 | ||
215 | if (next_ret) { | |
216 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 217 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
218 | prev = rb_prev(prev); |
219 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
220 | } | |
221 | *next_ret = prev; | |
222 | } | |
223 | return NULL; | |
224 | } | |
225 | ||
80ea96b1 CM |
226 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
227 | u64 offset) | |
d1310b2e | 228 | { |
70dec807 | 229 | struct rb_node *prev = NULL; |
d1310b2e | 230 | struct rb_node *ret; |
70dec807 | 231 | |
80ea96b1 | 232 | ret = __etree_search(tree, offset, &prev, NULL); |
d397712b | 233 | if (!ret) |
d1310b2e CM |
234 | return prev; |
235 | return ret; | |
236 | } | |
237 | ||
6af118ce CM |
238 | static struct extent_buffer *buffer_tree_insert(struct extent_io_tree *tree, |
239 | u64 offset, struct rb_node *node) | |
240 | { | |
241 | struct rb_root *root = &tree->buffer; | |
d397712b CM |
242 | struct rb_node **p = &root->rb_node; |
243 | struct rb_node *parent = NULL; | |
6af118ce CM |
244 | struct extent_buffer *eb; |
245 | ||
d397712b | 246 | while (*p) { |
6af118ce CM |
247 | parent = *p; |
248 | eb = rb_entry(parent, struct extent_buffer, rb_node); | |
249 | ||
250 | if (offset < eb->start) | |
251 | p = &(*p)->rb_left; | |
252 | else if (offset > eb->start) | |
253 | p = &(*p)->rb_right; | |
254 | else | |
255 | return eb; | |
256 | } | |
257 | ||
258 | rb_link_node(node, parent, p); | |
259 | rb_insert_color(node, root); | |
260 | return NULL; | |
261 | } | |
262 | ||
263 | static struct extent_buffer *buffer_search(struct extent_io_tree *tree, | |
264 | u64 offset) | |
265 | { | |
266 | struct rb_root *root = &tree->buffer; | |
d397712b | 267 | struct rb_node *n = root->rb_node; |
6af118ce CM |
268 | struct extent_buffer *eb; |
269 | ||
d397712b | 270 | while (n) { |
6af118ce CM |
271 | eb = rb_entry(n, struct extent_buffer, rb_node); |
272 | if (offset < eb->start) | |
273 | n = n->rb_left; | |
274 | else if (offset > eb->start) | |
275 | n = n->rb_right; | |
276 | else | |
277 | return eb; | |
278 | } | |
279 | return NULL; | |
280 | } | |
281 | ||
9ed74f2d JB |
282 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, |
283 | struct extent_state *other) | |
284 | { | |
285 | if (tree->ops && tree->ops->merge_extent_hook) | |
286 | tree->ops->merge_extent_hook(tree->mapping->host, new, | |
287 | other); | |
288 | } | |
289 | ||
d1310b2e CM |
290 | /* |
291 | * utility function to look for merge candidates inside a given range. | |
292 | * Any extents with matching state are merged together into a single | |
293 | * extent in the tree. Extents with EXTENT_IO in their state field | |
294 | * are not merged because the end_io handlers need to be able to do | |
295 | * operations on them without sleeping (or doing allocations/splits). | |
296 | * | |
297 | * This should be called with the tree lock held. | |
298 | */ | |
299 | static int merge_state(struct extent_io_tree *tree, | |
300 | struct extent_state *state) | |
301 | { | |
302 | struct extent_state *other; | |
303 | struct rb_node *other_node; | |
304 | ||
5b21f2ed | 305 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
d1310b2e CM |
306 | return 0; |
307 | ||
308 | other_node = rb_prev(&state->rb_node); | |
309 | if (other_node) { | |
310 | other = rb_entry(other_node, struct extent_state, rb_node); | |
311 | if (other->end == state->start - 1 && | |
312 | other->state == state->state) { | |
9ed74f2d | 313 | merge_cb(tree, state, other); |
d1310b2e | 314 | state->start = other->start; |
70dec807 | 315 | other->tree = NULL; |
d1310b2e CM |
316 | rb_erase(&other->rb_node, &tree->state); |
317 | free_extent_state(other); | |
318 | } | |
319 | } | |
320 | other_node = rb_next(&state->rb_node); | |
321 | if (other_node) { | |
322 | other = rb_entry(other_node, struct extent_state, rb_node); | |
323 | if (other->start == state->end + 1 && | |
324 | other->state == state->state) { | |
9ed74f2d | 325 | merge_cb(tree, state, other); |
d1310b2e | 326 | other->start = state->start; |
70dec807 | 327 | state->tree = NULL; |
d1310b2e CM |
328 | rb_erase(&state->rb_node, &tree->state); |
329 | free_extent_state(state); | |
9ed74f2d | 330 | state = NULL; |
d1310b2e CM |
331 | } |
332 | } | |
9ed74f2d | 333 | |
d1310b2e CM |
334 | return 0; |
335 | } | |
336 | ||
9ed74f2d | 337 | static int set_state_cb(struct extent_io_tree *tree, |
0ca1f7ce | 338 | struct extent_state *state, int *bits) |
291d673e CM |
339 | { |
340 | if (tree->ops && tree->ops->set_bit_hook) { | |
9ed74f2d | 341 | return tree->ops->set_bit_hook(tree->mapping->host, |
0ca1f7ce | 342 | state, bits); |
291d673e | 343 | } |
9ed74f2d JB |
344 | |
345 | return 0; | |
291d673e CM |
346 | } |
347 | ||
348 | static void clear_state_cb(struct extent_io_tree *tree, | |
0ca1f7ce | 349 | struct extent_state *state, int *bits) |
291d673e | 350 | { |
9ed74f2d JB |
351 | if (tree->ops && tree->ops->clear_bit_hook) |
352 | tree->ops->clear_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
353 | } |
354 | ||
d1310b2e CM |
355 | /* |
356 | * insert an extent_state struct into the tree. 'bits' are set on the | |
357 | * struct before it is inserted. | |
358 | * | |
359 | * This may return -EEXIST if the extent is already there, in which case the | |
360 | * state struct is freed. | |
361 | * | |
362 | * The tree lock is not taken internally. This is a utility function and | |
363 | * probably isn't what you want to call (see set/clear_extent_bit). | |
364 | */ | |
365 | static int insert_state(struct extent_io_tree *tree, | |
366 | struct extent_state *state, u64 start, u64 end, | |
0ca1f7ce | 367 | int *bits) |
d1310b2e CM |
368 | { |
369 | struct rb_node *node; | |
0ca1f7ce | 370 | int bits_to_set = *bits & ~EXTENT_CTLBITS; |
9ed74f2d | 371 | int ret; |
d1310b2e CM |
372 | |
373 | if (end < start) { | |
d397712b CM |
374 | printk(KERN_ERR "btrfs end < start %llu %llu\n", |
375 | (unsigned long long)end, | |
376 | (unsigned long long)start); | |
d1310b2e CM |
377 | WARN_ON(1); |
378 | } | |
d1310b2e CM |
379 | state->start = start; |
380 | state->end = end; | |
9ed74f2d JB |
381 | ret = set_state_cb(tree, state, bits); |
382 | if (ret) | |
383 | return ret; | |
384 | ||
0ca1f7ce | 385 | if (bits_to_set & EXTENT_DIRTY) |
9ed74f2d | 386 | tree->dirty_bytes += end - start + 1; |
0ca1f7ce | 387 | state->state |= bits_to_set; |
d1310b2e CM |
388 | node = tree_insert(&tree->state, end, &state->rb_node); |
389 | if (node) { | |
390 | struct extent_state *found; | |
391 | found = rb_entry(node, struct extent_state, rb_node); | |
d397712b CM |
392 | printk(KERN_ERR "btrfs found node %llu %llu on insert of " |
393 | "%llu %llu\n", (unsigned long long)found->start, | |
394 | (unsigned long long)found->end, | |
395 | (unsigned long long)start, (unsigned long long)end); | |
d1310b2e CM |
396 | free_extent_state(state); |
397 | return -EEXIST; | |
398 | } | |
70dec807 | 399 | state->tree = tree; |
d1310b2e CM |
400 | merge_state(tree, state); |
401 | return 0; | |
402 | } | |
403 | ||
9ed74f2d JB |
404 | static int split_cb(struct extent_io_tree *tree, struct extent_state *orig, |
405 | u64 split) | |
406 | { | |
407 | if (tree->ops && tree->ops->split_extent_hook) | |
408 | return tree->ops->split_extent_hook(tree->mapping->host, | |
409 | orig, split); | |
410 | return 0; | |
411 | } | |
412 | ||
d1310b2e CM |
413 | /* |
414 | * split a given extent state struct in two, inserting the preallocated | |
415 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
416 | * offset inside 'orig' where it should be split. | |
417 | * | |
418 | * Before calling, | |
419 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
420 | * are two extent state structs in the tree: | |
421 | * prealloc: [orig->start, split - 1] | |
422 | * orig: [ split, orig->end ] | |
423 | * | |
424 | * The tree locks are not taken by this function. They need to be held | |
425 | * by the caller. | |
426 | */ | |
427 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
428 | struct extent_state *prealloc, u64 split) | |
429 | { | |
430 | struct rb_node *node; | |
9ed74f2d JB |
431 | |
432 | split_cb(tree, orig, split); | |
433 | ||
d1310b2e CM |
434 | prealloc->start = orig->start; |
435 | prealloc->end = split - 1; | |
436 | prealloc->state = orig->state; | |
437 | orig->start = split; | |
438 | ||
439 | node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node); | |
440 | if (node) { | |
d1310b2e CM |
441 | free_extent_state(prealloc); |
442 | return -EEXIST; | |
443 | } | |
70dec807 | 444 | prealloc->tree = tree; |
d1310b2e CM |
445 | return 0; |
446 | } | |
447 | ||
448 | /* | |
449 | * utility function to clear some bits in an extent state struct. | |
450 | * it will optionally wake up any one waiting on this state (wake == 1), or | |
451 | * forcibly remove the state from the tree (delete == 1). | |
452 | * | |
453 | * If no bits are set on the state struct after clearing things, the | |
454 | * struct is freed and removed from the tree | |
455 | */ | |
456 | static int clear_state_bit(struct extent_io_tree *tree, | |
0ca1f7ce YZ |
457 | struct extent_state *state, |
458 | int *bits, int wake) | |
d1310b2e | 459 | { |
0ca1f7ce | 460 | int bits_to_clear = *bits & ~EXTENT_CTLBITS; |
32c00aff | 461 | int ret = state->state & bits_to_clear; |
d1310b2e | 462 | |
0ca1f7ce | 463 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
464 | u64 range = state->end - state->start + 1; |
465 | WARN_ON(range > tree->dirty_bytes); | |
466 | tree->dirty_bytes -= range; | |
467 | } | |
291d673e | 468 | clear_state_cb(tree, state, bits); |
32c00aff | 469 | state->state &= ~bits_to_clear; |
d1310b2e CM |
470 | if (wake) |
471 | wake_up(&state->wq); | |
0ca1f7ce | 472 | if (state->state == 0) { |
70dec807 | 473 | if (state->tree) { |
d1310b2e | 474 | rb_erase(&state->rb_node, &tree->state); |
70dec807 | 475 | state->tree = NULL; |
d1310b2e CM |
476 | free_extent_state(state); |
477 | } else { | |
478 | WARN_ON(1); | |
479 | } | |
480 | } else { | |
481 | merge_state(tree, state); | |
482 | } | |
483 | return ret; | |
484 | } | |
485 | ||
486 | /* | |
487 | * clear some bits on a range in the tree. This may require splitting | |
488 | * or inserting elements in the tree, so the gfp mask is used to | |
489 | * indicate which allocations or sleeping are allowed. | |
490 | * | |
491 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
492 | * the given range from the tree regardless of state (ie for truncate). | |
493 | * | |
494 | * the range [start, end] is inclusive. | |
495 | * | |
496 | * This takes the tree lock, and returns < 0 on error, > 0 if any of the | |
497 | * bits were already set, or zero if none of the bits were already set. | |
498 | */ | |
499 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
2c64c53d CM |
500 | int bits, int wake, int delete, |
501 | struct extent_state **cached_state, | |
502 | gfp_t mask) | |
d1310b2e CM |
503 | { |
504 | struct extent_state *state; | |
2c64c53d | 505 | struct extent_state *cached; |
d1310b2e | 506 | struct extent_state *prealloc = NULL; |
2c64c53d | 507 | struct rb_node *next_node; |
d1310b2e | 508 | struct rb_node *node; |
5c939df5 | 509 | u64 last_end; |
d1310b2e CM |
510 | int err; |
511 | int set = 0; | |
2ac55d41 | 512 | int clear = 0; |
d1310b2e | 513 | |
0ca1f7ce YZ |
514 | if (delete) |
515 | bits |= ~EXTENT_CTLBITS; | |
516 | bits |= EXTENT_FIRST_DELALLOC; | |
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) { | |
0ca1f7ce | 584 | set |= clear_state_bit(tree, state, &bits, wake); |
5c939df5 YZ |
585 | if (last_end == (u64)-1) |
586 | goto out; | |
587 | start = last_end + 1; | |
d1310b2e CM |
588 | } |
589 | goto search_again; | |
590 | } | |
591 | /* | |
592 | * | ---- desired range ---- | | |
593 | * | state | | |
594 | * We need to split the extent, and clear the bit | |
595 | * on the first half | |
596 | */ | |
597 | if (state->start <= end && state->end > end) { | |
70dec807 CM |
598 | if (!prealloc) |
599 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
d1310b2e CM |
600 | err = split_state(tree, state, prealloc, end + 1); |
601 | BUG_ON(err == -EEXIST); | |
d1310b2e CM |
602 | if (wake) |
603 | wake_up(&state->wq); | |
42daec29 | 604 | |
0ca1f7ce | 605 | set |= clear_state_bit(tree, prealloc, &bits, wake); |
9ed74f2d | 606 | |
d1310b2e CM |
607 | prealloc = NULL; |
608 | goto out; | |
609 | } | |
42daec29 | 610 | |
2c64c53d CM |
611 | if (state->end < end && prealloc && !need_resched()) |
612 | next_node = rb_next(&state->rb_node); | |
613 | else | |
614 | next_node = NULL; | |
42daec29 | 615 | |
0ca1f7ce | 616 | set |= clear_state_bit(tree, state, &bits, wake); |
5c939df5 YZ |
617 | if (last_end == (u64)-1) |
618 | goto out; | |
619 | start = last_end + 1; | |
2c64c53d CM |
620 | if (start <= end && next_node) { |
621 | state = rb_entry(next_node, struct extent_state, | |
622 | rb_node); | |
623 | if (state->start == start) | |
624 | goto hit_next; | |
625 | } | |
d1310b2e CM |
626 | goto search_again; |
627 | ||
628 | out: | |
cad321ad | 629 | spin_unlock(&tree->lock); |
d1310b2e CM |
630 | if (prealloc) |
631 | free_extent_state(prealloc); | |
632 | ||
633 | return set; | |
634 | ||
635 | search_again: | |
636 | if (start > end) | |
637 | goto out; | |
cad321ad | 638 | spin_unlock(&tree->lock); |
d1310b2e CM |
639 | if (mask & __GFP_WAIT) |
640 | cond_resched(); | |
641 | goto again; | |
642 | } | |
d1310b2e CM |
643 | |
644 | static int wait_on_state(struct extent_io_tree *tree, | |
645 | struct extent_state *state) | |
641f5219 CH |
646 | __releases(tree->lock) |
647 | __acquires(tree->lock) | |
d1310b2e CM |
648 | { |
649 | DEFINE_WAIT(wait); | |
650 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 651 | spin_unlock(&tree->lock); |
d1310b2e | 652 | schedule(); |
cad321ad | 653 | spin_lock(&tree->lock); |
d1310b2e CM |
654 | finish_wait(&state->wq, &wait); |
655 | return 0; | |
656 | } | |
657 | ||
658 | /* | |
659 | * waits for one or more bits to clear on a range in the state tree. | |
660 | * The range [start, end] is inclusive. | |
661 | * The tree lock is taken by this function | |
662 | */ | |
663 | int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits) | |
664 | { | |
665 | struct extent_state *state; | |
666 | struct rb_node *node; | |
667 | ||
cad321ad | 668 | spin_lock(&tree->lock); |
d1310b2e CM |
669 | again: |
670 | while (1) { | |
671 | /* | |
672 | * this search will find all the extents that end after | |
673 | * our range starts | |
674 | */ | |
80ea96b1 | 675 | node = tree_search(tree, start); |
d1310b2e CM |
676 | if (!node) |
677 | break; | |
678 | ||
679 | state = rb_entry(node, struct extent_state, rb_node); | |
680 | ||
681 | if (state->start > end) | |
682 | goto out; | |
683 | ||
684 | if (state->state & bits) { | |
685 | start = state->start; | |
686 | atomic_inc(&state->refs); | |
687 | wait_on_state(tree, state); | |
688 | free_extent_state(state); | |
689 | goto again; | |
690 | } | |
691 | start = state->end + 1; | |
692 | ||
693 | if (start > end) | |
694 | break; | |
695 | ||
696 | if (need_resched()) { | |
cad321ad | 697 | spin_unlock(&tree->lock); |
d1310b2e | 698 | cond_resched(); |
cad321ad | 699 | spin_lock(&tree->lock); |
d1310b2e CM |
700 | } |
701 | } | |
702 | out: | |
cad321ad | 703 | spin_unlock(&tree->lock); |
d1310b2e CM |
704 | return 0; |
705 | } | |
d1310b2e | 706 | |
9ed74f2d | 707 | static int set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 708 | struct extent_state *state, |
0ca1f7ce | 709 | int *bits) |
d1310b2e | 710 | { |
9ed74f2d | 711 | int ret; |
0ca1f7ce | 712 | int bits_to_set = *bits & ~EXTENT_CTLBITS; |
9ed74f2d JB |
713 | |
714 | ret = set_state_cb(tree, state, bits); | |
715 | if (ret) | |
716 | return ret; | |
0ca1f7ce | 717 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
718 | u64 range = state->end - state->start + 1; |
719 | tree->dirty_bytes += range; | |
720 | } | |
0ca1f7ce | 721 | state->state |= bits_to_set; |
9ed74f2d JB |
722 | |
723 | return 0; | |
d1310b2e CM |
724 | } |
725 | ||
2c64c53d CM |
726 | static void cache_state(struct extent_state *state, |
727 | struct extent_state **cached_ptr) | |
728 | { | |
729 | if (cached_ptr && !(*cached_ptr)) { | |
730 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) { | |
731 | *cached_ptr = state; | |
732 | atomic_inc(&state->refs); | |
733 | } | |
734 | } | |
735 | } | |
736 | ||
d1310b2e | 737 | /* |
1edbb734 CM |
738 | * set some bits on a range in the tree. This may require allocations or |
739 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 740 | * |
1edbb734 CM |
741 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
742 | * part of the range already has the desired bits set. The start of the | |
743 | * existing range is returned in failed_start in this case. | |
d1310b2e | 744 | * |
1edbb734 | 745 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 746 | */ |
1edbb734 | 747 | |
4845e44f CM |
748 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
749 | int bits, int exclusive_bits, u64 *failed_start, | |
750 | struct extent_state **cached_state, gfp_t mask) | |
d1310b2e CM |
751 | { |
752 | struct extent_state *state; | |
753 | struct extent_state *prealloc = NULL; | |
754 | struct rb_node *node; | |
d1310b2e | 755 | int err = 0; |
d1310b2e CM |
756 | u64 last_start; |
757 | u64 last_end; | |
42daec29 | 758 | |
0ca1f7ce | 759 | bits |= EXTENT_FIRST_DELALLOC; |
d1310b2e CM |
760 | again: |
761 | if (!prealloc && (mask & __GFP_WAIT)) { | |
762 | prealloc = alloc_extent_state(mask); | |
763 | if (!prealloc) | |
764 | return -ENOMEM; | |
765 | } | |
766 | ||
cad321ad | 767 | spin_lock(&tree->lock); |
9655d298 CM |
768 | if (cached_state && *cached_state) { |
769 | state = *cached_state; | |
770 | if (state->start == start && state->tree) { | |
771 | node = &state->rb_node; | |
772 | goto hit_next; | |
773 | } | |
774 | } | |
d1310b2e CM |
775 | /* |
776 | * this search will find all the extents that end after | |
777 | * our range starts. | |
778 | */ | |
80ea96b1 | 779 | node = tree_search(tree, start); |
d1310b2e | 780 | if (!node) { |
0ca1f7ce | 781 | err = insert_state(tree, prealloc, start, end, &bits); |
d1310b2e CM |
782 | prealloc = NULL; |
783 | BUG_ON(err == -EEXIST); | |
784 | goto out; | |
785 | } | |
d1310b2e | 786 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 787 | hit_next: |
d1310b2e CM |
788 | last_start = state->start; |
789 | last_end = state->end; | |
790 | ||
791 | /* | |
792 | * | ---- desired range ---- | | |
793 | * | state | | |
794 | * | |
795 | * Just lock what we found and keep going | |
796 | */ | |
797 | if (state->start == start && state->end <= end) { | |
40431d6c | 798 | struct rb_node *next_node; |
1edbb734 | 799 | if (state->state & exclusive_bits) { |
d1310b2e CM |
800 | *failed_start = state->start; |
801 | err = -EEXIST; | |
802 | goto out; | |
803 | } | |
42daec29 | 804 | |
0ca1f7ce | 805 | err = set_state_bits(tree, state, &bits); |
9ed74f2d JB |
806 | if (err) |
807 | goto out; | |
808 | ||
2c64c53d | 809 | cache_state(state, cached_state); |
d1310b2e | 810 | merge_state(tree, state); |
5c939df5 YZ |
811 | if (last_end == (u64)-1) |
812 | goto out; | |
40431d6c | 813 | |
5c939df5 | 814 | start = last_end + 1; |
40431d6c CM |
815 | if (start < end && prealloc && !need_resched()) { |
816 | next_node = rb_next(node); | |
817 | if (next_node) { | |
818 | state = rb_entry(next_node, struct extent_state, | |
819 | rb_node); | |
820 | if (state->start == start) | |
821 | goto hit_next; | |
822 | } | |
823 | } | |
d1310b2e CM |
824 | goto search_again; |
825 | } | |
826 | ||
827 | /* | |
828 | * | ---- desired range ---- | | |
829 | * | state | | |
830 | * or | |
831 | * | ------------- state -------------- | | |
832 | * | |
833 | * We need to split the extent we found, and may flip bits on | |
834 | * second half. | |
835 | * | |
836 | * If the extent we found extends past our | |
837 | * range, we just split and search again. It'll get split | |
838 | * again the next time though. | |
839 | * | |
840 | * If the extent we found is inside our range, we set the | |
841 | * desired bit on it. | |
842 | */ | |
843 | if (state->start < start) { | |
1edbb734 | 844 | if (state->state & exclusive_bits) { |
d1310b2e CM |
845 | *failed_start = start; |
846 | err = -EEXIST; | |
847 | goto out; | |
848 | } | |
849 | err = split_state(tree, state, prealloc, start); | |
850 | BUG_ON(err == -EEXIST); | |
851 | prealloc = NULL; | |
852 | if (err) | |
853 | goto out; | |
854 | if (state->end <= end) { | |
0ca1f7ce | 855 | err = set_state_bits(tree, state, &bits); |
9ed74f2d JB |
856 | if (err) |
857 | goto out; | |
2c64c53d | 858 | cache_state(state, cached_state); |
d1310b2e | 859 | merge_state(tree, state); |
5c939df5 YZ |
860 | if (last_end == (u64)-1) |
861 | goto out; | |
862 | start = last_end + 1; | |
d1310b2e CM |
863 | } |
864 | goto search_again; | |
865 | } | |
866 | /* | |
867 | * | ---- desired range ---- | | |
868 | * | state | or | state | | |
869 | * | |
870 | * There's a hole, we need to insert something in it and | |
871 | * ignore the extent we found. | |
872 | */ | |
873 | if (state->start > start) { | |
874 | u64 this_end; | |
875 | if (end < last_start) | |
876 | this_end = end; | |
877 | else | |
d397712b | 878 | this_end = last_start - 1; |
d1310b2e | 879 | err = insert_state(tree, prealloc, start, this_end, |
0ca1f7ce | 880 | &bits); |
d1310b2e | 881 | BUG_ON(err == -EEXIST); |
9ed74f2d JB |
882 | if (err) { |
883 | prealloc = NULL; | |
d1310b2e | 884 | goto out; |
9ed74f2d JB |
885 | } |
886 | cache_state(prealloc, cached_state); | |
887 | prealloc = NULL; | |
d1310b2e CM |
888 | start = this_end + 1; |
889 | goto search_again; | |
890 | } | |
891 | /* | |
892 | * | ---- desired range ---- | | |
893 | * | state | | |
894 | * We need to split the extent, and set the bit | |
895 | * on the first half | |
896 | */ | |
897 | if (state->start <= end && state->end > end) { | |
1edbb734 | 898 | if (state->state & exclusive_bits) { |
d1310b2e CM |
899 | *failed_start = start; |
900 | err = -EEXIST; | |
901 | goto out; | |
902 | } | |
903 | err = split_state(tree, state, prealloc, end + 1); | |
904 | BUG_ON(err == -EEXIST); | |
905 | ||
0ca1f7ce | 906 | err = set_state_bits(tree, prealloc, &bits); |
9ed74f2d JB |
907 | if (err) { |
908 | prealloc = NULL; | |
909 | goto out; | |
910 | } | |
2c64c53d | 911 | cache_state(prealloc, cached_state); |
d1310b2e CM |
912 | merge_state(tree, prealloc); |
913 | prealloc = NULL; | |
914 | goto out; | |
915 | } | |
916 | ||
917 | goto search_again; | |
918 | ||
919 | out: | |
cad321ad | 920 | spin_unlock(&tree->lock); |
d1310b2e CM |
921 | if (prealloc) |
922 | free_extent_state(prealloc); | |
923 | ||
924 | return err; | |
925 | ||
926 | search_again: | |
927 | if (start > end) | |
928 | goto out; | |
cad321ad | 929 | spin_unlock(&tree->lock); |
d1310b2e CM |
930 | if (mask & __GFP_WAIT) |
931 | cond_resched(); | |
932 | goto again; | |
933 | } | |
d1310b2e CM |
934 | |
935 | /* wrappers around set/clear extent bit */ | |
936 | int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
937 | gfp_t mask) | |
938 | { | |
939 | return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL, | |
2c64c53d | 940 | NULL, mask); |
d1310b2e | 941 | } |
d1310b2e CM |
942 | |
943 | int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
944 | int bits, gfp_t mask) | |
945 | { | |
946 | return set_extent_bit(tree, start, end, bits, 0, NULL, | |
2c64c53d | 947 | NULL, mask); |
d1310b2e | 948 | } |
d1310b2e CM |
949 | |
950 | int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
951 | int bits, gfp_t mask) | |
952 | { | |
2c64c53d | 953 | return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask); |
d1310b2e | 954 | } |
d1310b2e CM |
955 | |
956 | int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, | |
2ac55d41 | 957 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
958 | { |
959 | return set_extent_bit(tree, start, end, | |
40431d6c | 960 | EXTENT_DELALLOC | EXTENT_DIRTY | EXTENT_UPTODATE, |
2ac55d41 | 961 | 0, NULL, cached_state, mask); |
d1310b2e | 962 | } |
d1310b2e CM |
963 | |
964 | int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
965 | gfp_t mask) | |
966 | { | |
967 | return clear_extent_bit(tree, start, end, | |
32c00aff | 968 | EXTENT_DIRTY | EXTENT_DELALLOC | |
0ca1f7ce | 969 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask); |
d1310b2e | 970 | } |
d1310b2e CM |
971 | |
972 | int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, | |
973 | gfp_t mask) | |
974 | { | |
975 | return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL, | |
2c64c53d | 976 | NULL, mask); |
d1310b2e | 977 | } |
d1310b2e | 978 | |
b2950863 | 979 | static int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end, |
d1310b2e CM |
980 | gfp_t mask) |
981 | { | |
2c64c53d CM |
982 | return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, |
983 | NULL, mask); | |
d1310b2e | 984 | } |
d1310b2e CM |
985 | |
986 | int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, | |
987 | gfp_t mask) | |
988 | { | |
989 | return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL, | |
2c64c53d | 990 | NULL, mask); |
d1310b2e | 991 | } |
d1310b2e | 992 | |
d397712b | 993 | static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, |
2ac55d41 JB |
994 | u64 end, struct extent_state **cached_state, |
995 | gfp_t mask) | |
d1310b2e | 996 | { |
2c64c53d | 997 | return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, |
2ac55d41 | 998 | cached_state, mask); |
d1310b2e | 999 | } |
d1310b2e | 1000 | |
d1310b2e CM |
1001 | int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
1002 | { | |
1003 | return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK); | |
1004 | } | |
d1310b2e | 1005 | |
d352ac68 CM |
1006 | /* |
1007 | * either insert or lock state struct between start and end use mask to tell | |
1008 | * us if waiting is desired. | |
1009 | */ | |
1edbb734 | 1010 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
2c64c53d | 1011 | int bits, struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
1012 | { |
1013 | int err; | |
1014 | u64 failed_start; | |
1015 | while (1) { | |
1edbb734 | 1016 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits, |
2c64c53d CM |
1017 | EXTENT_LOCKED, &failed_start, |
1018 | cached_state, mask); | |
d1310b2e CM |
1019 | if (err == -EEXIST && (mask & __GFP_WAIT)) { |
1020 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); | |
1021 | start = failed_start; | |
1022 | } else { | |
1023 | break; | |
1024 | } | |
1025 | WARN_ON(start > end); | |
1026 | } | |
1027 | return err; | |
1028 | } | |
d1310b2e | 1029 | |
1edbb734 CM |
1030 | int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) |
1031 | { | |
2c64c53d | 1032 | return lock_extent_bits(tree, start, end, 0, NULL, mask); |
1edbb734 CM |
1033 | } |
1034 | ||
25179201 JB |
1035 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end, |
1036 | gfp_t mask) | |
1037 | { | |
1038 | int err; | |
1039 | u64 failed_start; | |
1040 | ||
2c64c53d CM |
1041 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
1042 | &failed_start, NULL, mask); | |
6643558d YZ |
1043 | if (err == -EEXIST) { |
1044 | if (failed_start > start) | |
1045 | clear_extent_bit(tree, start, failed_start - 1, | |
2c64c53d | 1046 | EXTENT_LOCKED, 1, 0, NULL, mask); |
25179201 | 1047 | return 0; |
6643558d | 1048 | } |
25179201 JB |
1049 | return 1; |
1050 | } | |
25179201 | 1051 | |
2c64c53d CM |
1052 | int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end, |
1053 | struct extent_state **cached, gfp_t mask) | |
1054 | { | |
1055 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, | |
1056 | mask); | |
1057 | } | |
1058 | ||
d1310b2e CM |
1059 | int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, |
1060 | gfp_t mask) | |
1061 | { | |
2c64c53d CM |
1062 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL, |
1063 | mask); | |
d1310b2e | 1064 | } |
d1310b2e CM |
1065 | |
1066 | /* | |
1067 | * helper function to set pages and extents in the tree dirty | |
1068 | */ | |
1069 | int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end) | |
1070 | { | |
1071 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1072 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1073 | struct page *page; | |
1074 | ||
1075 | while (index <= end_index) { | |
1076 | page = find_get_page(tree->mapping, index); | |
1077 | BUG_ON(!page); | |
1078 | __set_page_dirty_nobuffers(page); | |
1079 | page_cache_release(page); | |
1080 | index++; | |
1081 | } | |
d1310b2e CM |
1082 | return 0; |
1083 | } | |
d1310b2e CM |
1084 | |
1085 | /* | |
1086 | * helper function to set both pages and extents in the tree writeback | |
1087 | */ | |
b2950863 | 1088 | static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e CM |
1089 | { |
1090 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1091 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1092 | struct page *page; | |
1093 | ||
1094 | while (index <= end_index) { | |
1095 | page = find_get_page(tree->mapping, index); | |
1096 | BUG_ON(!page); | |
1097 | set_page_writeback(page); | |
1098 | page_cache_release(page); | |
1099 | index++; | |
1100 | } | |
d1310b2e CM |
1101 | return 0; |
1102 | } | |
d1310b2e | 1103 | |
d352ac68 CM |
1104 | /* |
1105 | * find the first offset in the io tree with 'bits' set. zero is | |
1106 | * returned if we find something, and *start_ret and *end_ret are | |
1107 | * set to reflect the state struct that was found. | |
1108 | * | |
1109 | * If nothing was found, 1 is returned, < 0 on error | |
1110 | */ | |
d1310b2e CM |
1111 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, |
1112 | u64 *start_ret, u64 *end_ret, int bits) | |
1113 | { | |
1114 | struct rb_node *node; | |
1115 | struct extent_state *state; | |
1116 | int ret = 1; | |
1117 | ||
cad321ad | 1118 | spin_lock(&tree->lock); |
d1310b2e CM |
1119 | /* |
1120 | * this search will find all the extents that end after | |
1121 | * our range starts. | |
1122 | */ | |
80ea96b1 | 1123 | node = tree_search(tree, start); |
d397712b | 1124 | if (!node) |
d1310b2e | 1125 | goto out; |
d1310b2e | 1126 | |
d397712b | 1127 | while (1) { |
d1310b2e CM |
1128 | state = rb_entry(node, struct extent_state, rb_node); |
1129 | if (state->end >= start && (state->state & bits)) { | |
1130 | *start_ret = state->start; | |
1131 | *end_ret = state->end; | |
1132 | ret = 0; | |
1133 | break; | |
1134 | } | |
1135 | node = rb_next(node); | |
1136 | if (!node) | |
1137 | break; | |
1138 | } | |
1139 | out: | |
cad321ad | 1140 | spin_unlock(&tree->lock); |
d1310b2e CM |
1141 | return ret; |
1142 | } | |
d1310b2e | 1143 | |
d352ac68 CM |
1144 | /* find the first state struct with 'bits' set after 'start', and |
1145 | * return it. tree->lock must be held. NULL will returned if | |
1146 | * nothing was found after 'start' | |
1147 | */ | |
d7fc640e CM |
1148 | struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree, |
1149 | u64 start, int bits) | |
1150 | { | |
1151 | struct rb_node *node; | |
1152 | struct extent_state *state; | |
1153 | ||
1154 | /* | |
1155 | * this search will find all the extents that end after | |
1156 | * our range starts. | |
1157 | */ | |
1158 | node = tree_search(tree, start); | |
d397712b | 1159 | if (!node) |
d7fc640e | 1160 | goto out; |
d7fc640e | 1161 | |
d397712b | 1162 | while (1) { |
d7fc640e | 1163 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1164 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1165 | return state; |
d397712b | 1166 | |
d7fc640e CM |
1167 | node = rb_next(node); |
1168 | if (!node) | |
1169 | break; | |
1170 | } | |
1171 | out: | |
1172 | return NULL; | |
1173 | } | |
d7fc640e | 1174 | |
d352ac68 CM |
1175 | /* |
1176 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1177 | * more than 'max_bytes'. start and end are used to return the range, | |
1178 | * | |
1179 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1180 | */ | |
c8b97818 | 1181 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1182 | u64 *start, u64 *end, u64 max_bytes, |
1183 | struct extent_state **cached_state) | |
d1310b2e CM |
1184 | { |
1185 | struct rb_node *node; | |
1186 | struct extent_state *state; | |
1187 | u64 cur_start = *start; | |
1188 | u64 found = 0; | |
1189 | u64 total_bytes = 0; | |
1190 | ||
cad321ad | 1191 | spin_lock(&tree->lock); |
c8b97818 | 1192 | |
d1310b2e CM |
1193 | /* |
1194 | * this search will find all the extents that end after | |
1195 | * our range starts. | |
1196 | */ | |
80ea96b1 | 1197 | node = tree_search(tree, cur_start); |
2b114d1d | 1198 | if (!node) { |
3b951516 CM |
1199 | if (!found) |
1200 | *end = (u64)-1; | |
d1310b2e CM |
1201 | goto out; |
1202 | } | |
1203 | ||
d397712b | 1204 | while (1) { |
d1310b2e | 1205 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1206 | if (found && (state->start != cur_start || |
1207 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1208 | goto out; |
1209 | } | |
1210 | if (!(state->state & EXTENT_DELALLOC)) { | |
1211 | if (!found) | |
1212 | *end = state->end; | |
1213 | goto out; | |
1214 | } | |
c2a128d2 | 1215 | if (!found) { |
d1310b2e | 1216 | *start = state->start; |
c2a128d2 JB |
1217 | *cached_state = state; |
1218 | atomic_inc(&state->refs); | |
1219 | } | |
d1310b2e CM |
1220 | found++; |
1221 | *end = state->end; | |
1222 | cur_start = state->end + 1; | |
1223 | node = rb_next(node); | |
1224 | if (!node) | |
1225 | break; | |
1226 | total_bytes += state->end - state->start + 1; | |
1227 | if (total_bytes >= max_bytes) | |
1228 | break; | |
1229 | } | |
1230 | out: | |
cad321ad | 1231 | spin_unlock(&tree->lock); |
d1310b2e CM |
1232 | return found; |
1233 | } | |
1234 | ||
c8b97818 CM |
1235 | static noinline int __unlock_for_delalloc(struct inode *inode, |
1236 | struct page *locked_page, | |
1237 | u64 start, u64 end) | |
1238 | { | |
1239 | int ret; | |
1240 | struct page *pages[16]; | |
1241 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1242 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1243 | unsigned long nr_pages = end_index - index + 1; | |
1244 | int i; | |
1245 | ||
1246 | if (index == locked_page->index && end_index == index) | |
1247 | return 0; | |
1248 | ||
d397712b | 1249 | while (nr_pages > 0) { |
c8b97818 | 1250 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1251 | min_t(unsigned long, nr_pages, |
1252 | ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1253 | for (i = 0; i < ret; i++) { |
1254 | if (pages[i] != locked_page) | |
1255 | unlock_page(pages[i]); | |
1256 | page_cache_release(pages[i]); | |
1257 | } | |
1258 | nr_pages -= ret; | |
1259 | index += ret; | |
1260 | cond_resched(); | |
1261 | } | |
1262 | return 0; | |
1263 | } | |
1264 | ||
1265 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1266 | struct page *locked_page, | |
1267 | u64 delalloc_start, | |
1268 | u64 delalloc_end) | |
1269 | { | |
1270 | unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT; | |
1271 | unsigned long start_index = index; | |
1272 | unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT; | |
1273 | unsigned long pages_locked = 0; | |
1274 | struct page *pages[16]; | |
1275 | unsigned long nrpages; | |
1276 | int ret; | |
1277 | int i; | |
1278 | ||
1279 | /* the caller is responsible for locking the start index */ | |
1280 | if (index == locked_page->index && index == end_index) | |
1281 | return 0; | |
1282 | ||
1283 | /* skip the page at the start index */ | |
1284 | nrpages = end_index - index + 1; | |
d397712b | 1285 | while (nrpages > 0) { |
c8b97818 | 1286 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1287 | min_t(unsigned long, |
1288 | nrpages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1289 | if (ret == 0) { |
1290 | ret = -EAGAIN; | |
1291 | goto done; | |
1292 | } | |
1293 | /* now we have an array of pages, lock them all */ | |
1294 | for (i = 0; i < ret; i++) { | |
1295 | /* | |
1296 | * the caller is taking responsibility for | |
1297 | * locked_page | |
1298 | */ | |
771ed689 | 1299 | if (pages[i] != locked_page) { |
c8b97818 | 1300 | lock_page(pages[i]); |
f2b1c41c CM |
1301 | if (!PageDirty(pages[i]) || |
1302 | pages[i]->mapping != inode->i_mapping) { | |
771ed689 CM |
1303 | ret = -EAGAIN; |
1304 | unlock_page(pages[i]); | |
1305 | page_cache_release(pages[i]); | |
1306 | goto done; | |
1307 | } | |
1308 | } | |
c8b97818 | 1309 | page_cache_release(pages[i]); |
771ed689 | 1310 | pages_locked++; |
c8b97818 | 1311 | } |
c8b97818 CM |
1312 | nrpages -= ret; |
1313 | index += ret; | |
1314 | cond_resched(); | |
1315 | } | |
1316 | ret = 0; | |
1317 | done: | |
1318 | if (ret && pages_locked) { | |
1319 | __unlock_for_delalloc(inode, locked_page, | |
1320 | delalloc_start, | |
1321 | ((u64)(start_index + pages_locked - 1)) << | |
1322 | PAGE_CACHE_SHIFT); | |
1323 | } | |
1324 | return ret; | |
1325 | } | |
1326 | ||
1327 | /* | |
1328 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1329 | * more than 'max_bytes'. start and end are used to return the range, | |
1330 | * | |
1331 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1332 | */ | |
1333 | static noinline u64 find_lock_delalloc_range(struct inode *inode, | |
1334 | struct extent_io_tree *tree, | |
1335 | struct page *locked_page, | |
1336 | u64 *start, u64 *end, | |
1337 | u64 max_bytes) | |
1338 | { | |
1339 | u64 delalloc_start; | |
1340 | u64 delalloc_end; | |
1341 | u64 found; | |
9655d298 | 1342 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1343 | int ret; |
1344 | int loops = 0; | |
1345 | ||
1346 | again: | |
1347 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1348 | delalloc_start = *start; | |
1349 | delalloc_end = 0; | |
1350 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1351 | max_bytes, &cached_state); |
70b99e69 | 1352 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1353 | *start = delalloc_start; |
1354 | *end = delalloc_end; | |
c2a128d2 | 1355 | free_extent_state(cached_state); |
c8b97818 CM |
1356 | return found; |
1357 | } | |
1358 | ||
70b99e69 CM |
1359 | /* |
1360 | * start comes from the offset of locked_page. We have to lock | |
1361 | * pages in order, so we can't process delalloc bytes before | |
1362 | * locked_page | |
1363 | */ | |
d397712b | 1364 | if (delalloc_start < *start) |
70b99e69 | 1365 | delalloc_start = *start; |
70b99e69 | 1366 | |
c8b97818 CM |
1367 | /* |
1368 | * make sure to limit the number of pages we try to lock down | |
1369 | * if we're looping. | |
1370 | */ | |
d397712b | 1371 | if (delalloc_end + 1 - delalloc_start > max_bytes && loops) |
771ed689 | 1372 | delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1; |
d397712b | 1373 | |
c8b97818 CM |
1374 | /* step two, lock all the pages after the page that has start */ |
1375 | ret = lock_delalloc_pages(inode, locked_page, | |
1376 | delalloc_start, delalloc_end); | |
1377 | if (ret == -EAGAIN) { | |
1378 | /* some of the pages are gone, lets avoid looping by | |
1379 | * shortening the size of the delalloc range we're searching | |
1380 | */ | |
9655d298 | 1381 | free_extent_state(cached_state); |
c8b97818 CM |
1382 | if (!loops) { |
1383 | unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1); | |
1384 | max_bytes = PAGE_CACHE_SIZE - offset; | |
1385 | loops = 1; | |
1386 | goto again; | |
1387 | } else { | |
1388 | found = 0; | |
1389 | goto out_failed; | |
1390 | } | |
1391 | } | |
1392 | BUG_ON(ret); | |
1393 | ||
1394 | /* step three, lock the state bits for the whole range */ | |
9655d298 CM |
1395 | lock_extent_bits(tree, delalloc_start, delalloc_end, |
1396 | 0, &cached_state, GFP_NOFS); | |
c8b97818 CM |
1397 | |
1398 | /* then test to make sure it is all still delalloc */ | |
1399 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1400 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1401 | if (!ret) { |
9655d298 CM |
1402 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
1403 | &cached_state, GFP_NOFS); | |
c8b97818 CM |
1404 | __unlock_for_delalloc(inode, locked_page, |
1405 | delalloc_start, delalloc_end); | |
1406 | cond_resched(); | |
1407 | goto again; | |
1408 | } | |
9655d298 | 1409 | free_extent_state(cached_state); |
c8b97818 CM |
1410 | *start = delalloc_start; |
1411 | *end = delalloc_end; | |
1412 | out_failed: | |
1413 | return found; | |
1414 | } | |
1415 | ||
1416 | int extent_clear_unlock_delalloc(struct inode *inode, | |
1417 | struct extent_io_tree *tree, | |
1418 | u64 start, u64 end, struct page *locked_page, | |
a791e35e | 1419 | unsigned long op) |
c8b97818 CM |
1420 | { |
1421 | int ret; | |
1422 | struct page *pages[16]; | |
1423 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1424 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1425 | unsigned long nr_pages = end_index - index + 1; | |
1426 | int i; | |
771ed689 | 1427 | int clear_bits = 0; |
c8b97818 | 1428 | |
a791e35e | 1429 | if (op & EXTENT_CLEAR_UNLOCK) |
771ed689 | 1430 | clear_bits |= EXTENT_LOCKED; |
a791e35e | 1431 | if (op & EXTENT_CLEAR_DIRTY) |
c8b97818 CM |
1432 | clear_bits |= EXTENT_DIRTY; |
1433 | ||
a791e35e | 1434 | if (op & EXTENT_CLEAR_DELALLOC) |
771ed689 CM |
1435 | clear_bits |= EXTENT_DELALLOC; |
1436 | ||
2c64c53d | 1437 | clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); |
32c00aff JB |
1438 | if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | |
1439 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK | | |
1440 | EXTENT_SET_PRIVATE2))) | |
771ed689 | 1441 | return 0; |
c8b97818 | 1442 | |
d397712b | 1443 | while (nr_pages > 0) { |
c8b97818 | 1444 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1445 | min_t(unsigned long, |
1446 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 | 1447 | for (i = 0; i < ret; i++) { |
8b62b72b | 1448 | |
a791e35e | 1449 | if (op & EXTENT_SET_PRIVATE2) |
8b62b72b CM |
1450 | SetPagePrivate2(pages[i]); |
1451 | ||
c8b97818 CM |
1452 | if (pages[i] == locked_page) { |
1453 | page_cache_release(pages[i]); | |
1454 | continue; | |
1455 | } | |
a791e35e | 1456 | if (op & EXTENT_CLEAR_DIRTY) |
c8b97818 | 1457 | clear_page_dirty_for_io(pages[i]); |
a791e35e | 1458 | if (op & EXTENT_SET_WRITEBACK) |
c8b97818 | 1459 | set_page_writeback(pages[i]); |
a791e35e | 1460 | if (op & EXTENT_END_WRITEBACK) |
c8b97818 | 1461 | end_page_writeback(pages[i]); |
a791e35e | 1462 | if (op & EXTENT_CLEAR_UNLOCK_PAGE) |
771ed689 | 1463 | unlock_page(pages[i]); |
c8b97818 CM |
1464 | page_cache_release(pages[i]); |
1465 | } | |
1466 | nr_pages -= ret; | |
1467 | index += ret; | |
1468 | cond_resched(); | |
1469 | } | |
1470 | return 0; | |
1471 | } | |
c8b97818 | 1472 | |
d352ac68 CM |
1473 | /* |
1474 | * count the number of bytes in the tree that have a given bit(s) | |
1475 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1476 | * cached. The total number found is returned. | |
1477 | */ | |
d1310b2e CM |
1478 | u64 count_range_bits(struct extent_io_tree *tree, |
1479 | u64 *start, u64 search_end, u64 max_bytes, | |
1480 | unsigned long bits) | |
1481 | { | |
1482 | struct rb_node *node; | |
1483 | struct extent_state *state; | |
1484 | u64 cur_start = *start; | |
1485 | u64 total_bytes = 0; | |
1486 | int found = 0; | |
1487 | ||
1488 | if (search_end <= cur_start) { | |
d1310b2e CM |
1489 | WARN_ON(1); |
1490 | return 0; | |
1491 | } | |
1492 | ||
cad321ad | 1493 | spin_lock(&tree->lock); |
d1310b2e CM |
1494 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1495 | total_bytes = tree->dirty_bytes; | |
1496 | goto out; | |
1497 | } | |
1498 | /* | |
1499 | * this search will find all the extents that end after | |
1500 | * our range starts. | |
1501 | */ | |
80ea96b1 | 1502 | node = tree_search(tree, cur_start); |
d397712b | 1503 | if (!node) |
d1310b2e | 1504 | goto out; |
d1310b2e | 1505 | |
d397712b | 1506 | while (1) { |
d1310b2e CM |
1507 | state = rb_entry(node, struct extent_state, rb_node); |
1508 | if (state->start > search_end) | |
1509 | break; | |
1510 | if (state->end >= cur_start && (state->state & bits)) { | |
1511 | total_bytes += min(search_end, state->end) + 1 - | |
1512 | max(cur_start, state->start); | |
1513 | if (total_bytes >= max_bytes) | |
1514 | break; | |
1515 | if (!found) { | |
1516 | *start = state->start; | |
1517 | found = 1; | |
1518 | } | |
1519 | } | |
1520 | node = rb_next(node); | |
1521 | if (!node) | |
1522 | break; | |
1523 | } | |
1524 | out: | |
cad321ad | 1525 | spin_unlock(&tree->lock); |
d1310b2e CM |
1526 | return total_bytes; |
1527 | } | |
b2950863 | 1528 | |
d352ac68 CM |
1529 | /* |
1530 | * set the private field for a given byte offset in the tree. If there isn't | |
1531 | * an extent_state there already, this does nothing. | |
1532 | */ | |
d1310b2e CM |
1533 | int set_state_private(struct extent_io_tree *tree, u64 start, u64 private) |
1534 | { | |
1535 | struct rb_node *node; | |
1536 | struct extent_state *state; | |
1537 | int ret = 0; | |
1538 | ||
cad321ad | 1539 | spin_lock(&tree->lock); |
d1310b2e CM |
1540 | /* |
1541 | * this search will find all the extents that end after | |
1542 | * our range starts. | |
1543 | */ | |
80ea96b1 | 1544 | node = tree_search(tree, start); |
2b114d1d | 1545 | if (!node) { |
d1310b2e CM |
1546 | ret = -ENOENT; |
1547 | goto out; | |
1548 | } | |
1549 | state = rb_entry(node, struct extent_state, rb_node); | |
1550 | if (state->start != start) { | |
1551 | ret = -ENOENT; | |
1552 | goto out; | |
1553 | } | |
1554 | state->private = private; | |
1555 | out: | |
cad321ad | 1556 | spin_unlock(&tree->lock); |
d1310b2e CM |
1557 | return ret; |
1558 | } | |
1559 | ||
1560 | int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private) | |
1561 | { | |
1562 | struct rb_node *node; | |
1563 | struct extent_state *state; | |
1564 | int ret = 0; | |
1565 | ||
cad321ad | 1566 | spin_lock(&tree->lock); |
d1310b2e CM |
1567 | /* |
1568 | * this search will find all the extents that end after | |
1569 | * our range starts. | |
1570 | */ | |
80ea96b1 | 1571 | node = tree_search(tree, start); |
2b114d1d | 1572 | if (!node) { |
d1310b2e CM |
1573 | ret = -ENOENT; |
1574 | goto out; | |
1575 | } | |
1576 | state = rb_entry(node, struct extent_state, rb_node); | |
1577 | if (state->start != start) { | |
1578 | ret = -ENOENT; | |
1579 | goto out; | |
1580 | } | |
1581 | *private = state->private; | |
1582 | out: | |
cad321ad | 1583 | spin_unlock(&tree->lock); |
d1310b2e CM |
1584 | return ret; |
1585 | } | |
1586 | ||
1587 | /* | |
1588 | * searches a range in the state tree for a given mask. | |
70dec807 | 1589 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1590 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1591 | * range is found set. | |
1592 | */ | |
1593 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9655d298 | 1594 | int bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1595 | { |
1596 | struct extent_state *state = NULL; | |
1597 | struct rb_node *node; | |
1598 | int bitset = 0; | |
d1310b2e | 1599 | |
cad321ad | 1600 | spin_lock(&tree->lock); |
9655d298 CM |
1601 | if (cached && cached->tree && cached->start == start) |
1602 | node = &cached->rb_node; | |
1603 | else | |
1604 | node = tree_search(tree, start); | |
d1310b2e CM |
1605 | while (node && start <= end) { |
1606 | state = rb_entry(node, struct extent_state, rb_node); | |
1607 | ||
1608 | if (filled && state->start > start) { | |
1609 | bitset = 0; | |
1610 | break; | |
1611 | } | |
1612 | ||
1613 | if (state->start > end) | |
1614 | break; | |
1615 | ||
1616 | if (state->state & bits) { | |
1617 | bitset = 1; | |
1618 | if (!filled) | |
1619 | break; | |
1620 | } else if (filled) { | |
1621 | bitset = 0; | |
1622 | break; | |
1623 | } | |
46562cec CM |
1624 | |
1625 | if (state->end == (u64)-1) | |
1626 | break; | |
1627 | ||
d1310b2e CM |
1628 | start = state->end + 1; |
1629 | if (start > end) | |
1630 | break; | |
1631 | node = rb_next(node); | |
1632 | if (!node) { | |
1633 | if (filled) | |
1634 | bitset = 0; | |
1635 | break; | |
1636 | } | |
1637 | } | |
cad321ad | 1638 | spin_unlock(&tree->lock); |
d1310b2e CM |
1639 | return bitset; |
1640 | } | |
d1310b2e CM |
1641 | |
1642 | /* | |
1643 | * helper function to set a given page up to date if all the | |
1644 | * extents in the tree for that page are up to date | |
1645 | */ | |
1646 | static int check_page_uptodate(struct extent_io_tree *tree, | |
1647 | struct page *page) | |
1648 | { | |
1649 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1650 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
9655d298 | 1651 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e CM |
1652 | SetPageUptodate(page); |
1653 | return 0; | |
1654 | } | |
1655 | ||
1656 | /* | |
1657 | * helper function to unlock a page if all the extents in the tree | |
1658 | * for that page are unlocked | |
1659 | */ | |
1660 | static int check_page_locked(struct extent_io_tree *tree, | |
1661 | struct page *page) | |
1662 | { | |
1663 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1664 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
9655d298 | 1665 | if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) |
d1310b2e CM |
1666 | unlock_page(page); |
1667 | return 0; | |
1668 | } | |
1669 | ||
1670 | /* | |
1671 | * helper function to end page writeback if all the extents | |
1672 | * in the tree for that page are done with writeback | |
1673 | */ | |
1674 | static int check_page_writeback(struct extent_io_tree *tree, | |
1675 | struct page *page) | |
1676 | { | |
1edbb734 | 1677 | end_page_writeback(page); |
d1310b2e CM |
1678 | return 0; |
1679 | } | |
1680 | ||
1681 | /* lots and lots of room for performance fixes in the end_bio funcs */ | |
1682 | ||
1683 | /* | |
1684 | * after a writepage IO is done, we need to: | |
1685 | * clear the uptodate bits on error | |
1686 | * clear the writeback bits in the extent tree for this IO | |
1687 | * end_page_writeback if the page has no more pending IO | |
1688 | * | |
1689 | * Scheduling is not allowed, so the extent state tree is expected | |
1690 | * to have one and only one object corresponding to this IO. | |
1691 | */ | |
d1310b2e | 1692 | static void end_bio_extent_writepage(struct bio *bio, int err) |
d1310b2e | 1693 | { |
1259ab75 | 1694 | int uptodate = err == 0; |
d1310b2e | 1695 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
902b22f3 | 1696 | struct extent_io_tree *tree; |
d1310b2e CM |
1697 | u64 start; |
1698 | u64 end; | |
1699 | int whole_page; | |
1259ab75 | 1700 | int ret; |
d1310b2e | 1701 | |
d1310b2e CM |
1702 | do { |
1703 | struct page *page = bvec->bv_page; | |
902b22f3 DW |
1704 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
1705 | ||
d1310b2e CM |
1706 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
1707 | bvec->bv_offset; | |
1708 | end = start + bvec->bv_len - 1; | |
1709 | ||
1710 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
1711 | whole_page = 1; | |
1712 | else | |
1713 | whole_page = 0; | |
1714 | ||
1715 | if (--bvec >= bio->bi_io_vec) | |
1716 | prefetchw(&bvec->bv_page->flags); | |
1259ab75 CM |
1717 | if (tree->ops && tree->ops->writepage_end_io_hook) { |
1718 | ret = tree->ops->writepage_end_io_hook(page, start, | |
902b22f3 | 1719 | end, NULL, uptodate); |
1259ab75 CM |
1720 | if (ret) |
1721 | uptodate = 0; | |
1722 | } | |
1723 | ||
1724 | if (!uptodate && tree->ops && | |
1725 | tree->ops->writepage_io_failed_hook) { | |
1726 | ret = tree->ops->writepage_io_failed_hook(bio, page, | |
902b22f3 | 1727 | start, end, NULL); |
1259ab75 | 1728 | if (ret == 0) { |
1259ab75 CM |
1729 | uptodate = (err == 0); |
1730 | continue; | |
1731 | } | |
1732 | } | |
1733 | ||
d1310b2e | 1734 | if (!uptodate) { |
2ac55d41 | 1735 | clear_extent_uptodate(tree, start, end, NULL, GFP_NOFS); |
d1310b2e CM |
1736 | ClearPageUptodate(page); |
1737 | SetPageError(page); | |
1738 | } | |
70dec807 | 1739 | |
d1310b2e CM |
1740 | if (whole_page) |
1741 | end_page_writeback(page); | |
1742 | else | |
1743 | check_page_writeback(tree, page); | |
d1310b2e | 1744 | } while (bvec >= bio->bi_io_vec); |
2b1f55b0 | 1745 | |
d1310b2e | 1746 | bio_put(bio); |
d1310b2e CM |
1747 | } |
1748 | ||
1749 | /* | |
1750 | * after a readpage IO is done, we need to: | |
1751 | * clear the uptodate bits on error | |
1752 | * set the uptodate bits if things worked | |
1753 | * set the page up to date if all extents in the tree are uptodate | |
1754 | * clear the lock bit in the extent tree | |
1755 | * unlock the page if there are no other extents locked for it | |
1756 | * | |
1757 | * Scheduling is not allowed, so the extent state tree is expected | |
1758 | * to have one and only one object corresponding to this IO. | |
1759 | */ | |
d1310b2e | 1760 | static void end_bio_extent_readpage(struct bio *bio, int err) |
d1310b2e CM |
1761 | { |
1762 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
4125bf76 CM |
1763 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
1764 | struct bio_vec *bvec = bio->bi_io_vec; | |
902b22f3 | 1765 | struct extent_io_tree *tree; |
d1310b2e CM |
1766 | u64 start; |
1767 | u64 end; | |
1768 | int whole_page; | |
1769 | int ret; | |
1770 | ||
d20f7043 CM |
1771 | if (err) |
1772 | uptodate = 0; | |
1773 | ||
d1310b2e CM |
1774 | do { |
1775 | struct page *page = bvec->bv_page; | |
902b22f3 DW |
1776 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
1777 | ||
d1310b2e CM |
1778 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
1779 | bvec->bv_offset; | |
1780 | end = start + bvec->bv_len - 1; | |
1781 | ||
1782 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
1783 | whole_page = 1; | |
1784 | else | |
1785 | whole_page = 0; | |
1786 | ||
4125bf76 | 1787 | if (++bvec <= bvec_end) |
d1310b2e CM |
1788 | prefetchw(&bvec->bv_page->flags); |
1789 | ||
1790 | if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) { | |
70dec807 | 1791 | ret = tree->ops->readpage_end_io_hook(page, start, end, |
902b22f3 | 1792 | NULL); |
d1310b2e CM |
1793 | if (ret) |
1794 | uptodate = 0; | |
1795 | } | |
7e38326f CM |
1796 | if (!uptodate && tree->ops && |
1797 | tree->ops->readpage_io_failed_hook) { | |
1798 | ret = tree->ops->readpage_io_failed_hook(bio, page, | |
902b22f3 | 1799 | start, end, NULL); |
7e38326f | 1800 | if (ret == 0) { |
3b951516 CM |
1801 | uptodate = |
1802 | test_bit(BIO_UPTODATE, &bio->bi_flags); | |
d20f7043 CM |
1803 | if (err) |
1804 | uptodate = 0; | |
7e38326f CM |
1805 | continue; |
1806 | } | |
1807 | } | |
d1310b2e | 1808 | |
771ed689 | 1809 | if (uptodate) { |
902b22f3 DW |
1810 | set_extent_uptodate(tree, start, end, |
1811 | GFP_ATOMIC); | |
771ed689 | 1812 | } |
902b22f3 | 1813 | unlock_extent(tree, start, end, GFP_ATOMIC); |
d1310b2e | 1814 | |
70dec807 CM |
1815 | if (whole_page) { |
1816 | if (uptodate) { | |
1817 | SetPageUptodate(page); | |
1818 | } else { | |
1819 | ClearPageUptodate(page); | |
1820 | SetPageError(page); | |
1821 | } | |
d1310b2e | 1822 | unlock_page(page); |
70dec807 CM |
1823 | } else { |
1824 | if (uptodate) { | |
1825 | check_page_uptodate(tree, page); | |
1826 | } else { | |
1827 | ClearPageUptodate(page); | |
1828 | SetPageError(page); | |
1829 | } | |
d1310b2e | 1830 | check_page_locked(tree, page); |
70dec807 | 1831 | } |
4125bf76 | 1832 | } while (bvec <= bvec_end); |
d1310b2e CM |
1833 | |
1834 | bio_put(bio); | |
d1310b2e CM |
1835 | } |
1836 | ||
1837 | /* | |
1838 | * IO done from prepare_write is pretty simple, we just unlock | |
1839 | * the structs in the extent tree when done, and set the uptodate bits | |
1840 | * as appropriate. | |
1841 | */ | |
d1310b2e | 1842 | static void end_bio_extent_preparewrite(struct bio *bio, int err) |
d1310b2e CM |
1843 | { |
1844 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
1845 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
902b22f3 | 1846 | struct extent_io_tree *tree; |
d1310b2e CM |
1847 | u64 start; |
1848 | u64 end; | |
1849 | ||
d1310b2e CM |
1850 | do { |
1851 | struct page *page = bvec->bv_page; | |
902b22f3 DW |
1852 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
1853 | ||
d1310b2e CM |
1854 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
1855 | bvec->bv_offset; | |
1856 | end = start + bvec->bv_len - 1; | |
1857 | ||
1858 | if (--bvec >= bio->bi_io_vec) | |
1859 | prefetchw(&bvec->bv_page->flags); | |
1860 | ||
1861 | if (uptodate) { | |
1862 | set_extent_uptodate(tree, start, end, GFP_ATOMIC); | |
1863 | } else { | |
1864 | ClearPageUptodate(page); | |
1865 | SetPageError(page); | |
1866 | } | |
1867 | ||
1868 | unlock_extent(tree, start, end, GFP_ATOMIC); | |
1869 | ||
1870 | } while (bvec >= bio->bi_io_vec); | |
1871 | ||
1872 | bio_put(bio); | |
d1310b2e CM |
1873 | } |
1874 | ||
1875 | static struct bio * | |
1876 | extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
1877 | gfp_t gfp_flags) | |
1878 | { | |
1879 | struct bio *bio; | |
1880 | ||
1881 | bio = bio_alloc(gfp_flags, nr_vecs); | |
1882 | ||
1883 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
1884 | while (!bio && (nr_vecs /= 2)) | |
1885 | bio = bio_alloc(gfp_flags, nr_vecs); | |
1886 | } | |
1887 | ||
1888 | if (bio) { | |
e1c4b745 | 1889 | bio->bi_size = 0; |
d1310b2e CM |
1890 | bio->bi_bdev = bdev; |
1891 | bio->bi_sector = first_sector; | |
1892 | } | |
1893 | return bio; | |
1894 | } | |
1895 | ||
c8b97818 CM |
1896 | static int submit_one_bio(int rw, struct bio *bio, int mirror_num, |
1897 | unsigned long bio_flags) | |
d1310b2e | 1898 | { |
d1310b2e | 1899 | int ret = 0; |
70dec807 CM |
1900 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
1901 | struct page *page = bvec->bv_page; | |
1902 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 CM |
1903 | u64 start; |
1904 | u64 end; | |
1905 | ||
1906 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset; | |
1907 | end = start + bvec->bv_len - 1; | |
1908 | ||
902b22f3 | 1909 | bio->bi_private = NULL; |
d1310b2e CM |
1910 | |
1911 | bio_get(bio); | |
1912 | ||
065631f6 | 1913 | if (tree->ops && tree->ops->submit_bio_hook) |
f188591e | 1914 | tree->ops->submit_bio_hook(page->mapping->host, rw, bio, |
eaf25d93 | 1915 | mirror_num, bio_flags, start); |
0b86a832 CM |
1916 | else |
1917 | submit_bio(rw, bio); | |
d1310b2e CM |
1918 | if (bio_flagged(bio, BIO_EOPNOTSUPP)) |
1919 | ret = -EOPNOTSUPP; | |
1920 | bio_put(bio); | |
1921 | return ret; | |
1922 | } | |
1923 | ||
1924 | static int submit_extent_page(int rw, struct extent_io_tree *tree, | |
1925 | struct page *page, sector_t sector, | |
1926 | size_t size, unsigned long offset, | |
1927 | struct block_device *bdev, | |
1928 | struct bio **bio_ret, | |
1929 | unsigned long max_pages, | |
f188591e | 1930 | bio_end_io_t end_io_func, |
c8b97818 CM |
1931 | int mirror_num, |
1932 | unsigned long prev_bio_flags, | |
1933 | unsigned long bio_flags) | |
d1310b2e CM |
1934 | { |
1935 | int ret = 0; | |
1936 | struct bio *bio; | |
1937 | int nr; | |
c8b97818 CM |
1938 | int contig = 0; |
1939 | int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED; | |
1940 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; | |
5b050f04 | 1941 | size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE); |
d1310b2e CM |
1942 | |
1943 | if (bio_ret && *bio_ret) { | |
1944 | bio = *bio_ret; | |
c8b97818 CM |
1945 | if (old_compressed) |
1946 | contig = bio->bi_sector == sector; | |
1947 | else | |
1948 | contig = bio->bi_sector + (bio->bi_size >> 9) == | |
1949 | sector; | |
1950 | ||
1951 | if (prev_bio_flags != bio_flags || !contig || | |
239b14b3 | 1952 | (tree->ops && tree->ops->merge_bio_hook && |
c8b97818 CM |
1953 | tree->ops->merge_bio_hook(page, offset, page_size, bio, |
1954 | bio_flags)) || | |
1955 | bio_add_page(bio, page, page_size, offset) < page_size) { | |
1956 | ret = submit_one_bio(rw, bio, mirror_num, | |
1957 | prev_bio_flags); | |
d1310b2e CM |
1958 | bio = NULL; |
1959 | } else { | |
1960 | return 0; | |
1961 | } | |
1962 | } | |
c8b97818 CM |
1963 | if (this_compressed) |
1964 | nr = BIO_MAX_PAGES; | |
1965 | else | |
1966 | nr = bio_get_nr_vecs(bdev); | |
1967 | ||
d1310b2e | 1968 | bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH); |
70dec807 | 1969 | |
c8b97818 | 1970 | bio_add_page(bio, page, page_size, offset); |
d1310b2e CM |
1971 | bio->bi_end_io = end_io_func; |
1972 | bio->bi_private = tree; | |
70dec807 | 1973 | |
d397712b | 1974 | if (bio_ret) |
d1310b2e | 1975 | *bio_ret = bio; |
d397712b | 1976 | else |
c8b97818 | 1977 | ret = submit_one_bio(rw, bio, mirror_num, bio_flags); |
d1310b2e CM |
1978 | |
1979 | return ret; | |
1980 | } | |
1981 | ||
1982 | void set_page_extent_mapped(struct page *page) | |
1983 | { | |
1984 | if (!PagePrivate(page)) { | |
1985 | SetPagePrivate(page); | |
d1310b2e | 1986 | page_cache_get(page); |
6af118ce | 1987 | set_page_private(page, EXTENT_PAGE_PRIVATE); |
d1310b2e CM |
1988 | } |
1989 | } | |
1990 | ||
b2950863 | 1991 | static void set_page_extent_head(struct page *page, unsigned long len) |
d1310b2e CM |
1992 | { |
1993 | set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2); | |
1994 | } | |
1995 | ||
1996 | /* | |
1997 | * basic readpage implementation. Locked extent state structs are inserted | |
1998 | * into the tree that are removed when the IO is done (by the end_io | |
1999 | * handlers) | |
2000 | */ | |
2001 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
2002 | struct page *page, | |
2003 | get_extent_t *get_extent, | |
c8b97818 CM |
2004 | struct bio **bio, int mirror_num, |
2005 | unsigned long *bio_flags) | |
d1310b2e CM |
2006 | { |
2007 | struct inode *inode = page->mapping->host; | |
2008 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2009 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2010 | u64 end; | |
2011 | u64 cur = start; | |
2012 | u64 extent_offset; | |
2013 | u64 last_byte = i_size_read(inode); | |
2014 | u64 block_start; | |
2015 | u64 cur_end; | |
2016 | sector_t sector; | |
2017 | struct extent_map *em; | |
2018 | struct block_device *bdev; | |
11c65dcc | 2019 | struct btrfs_ordered_extent *ordered; |
d1310b2e CM |
2020 | int ret; |
2021 | int nr = 0; | |
2022 | size_t page_offset = 0; | |
2023 | size_t iosize; | |
c8b97818 | 2024 | size_t disk_io_size; |
d1310b2e | 2025 | size_t blocksize = inode->i_sb->s_blocksize; |
c8b97818 | 2026 | unsigned long this_bio_flag = 0; |
d1310b2e CM |
2027 | |
2028 | set_page_extent_mapped(page); | |
2029 | ||
2030 | end = page_end; | |
11c65dcc JB |
2031 | while (1) { |
2032 | lock_extent(tree, start, end, GFP_NOFS); | |
2033 | ordered = btrfs_lookup_ordered_extent(inode, start); | |
2034 | if (!ordered) | |
2035 | break; | |
2036 | unlock_extent(tree, start, end, GFP_NOFS); | |
2037 | btrfs_start_ordered_extent(inode, ordered, 1); | |
2038 | btrfs_put_ordered_extent(ordered); | |
2039 | } | |
d1310b2e | 2040 | |
c8b97818 CM |
2041 | if (page->index == last_byte >> PAGE_CACHE_SHIFT) { |
2042 | char *userpage; | |
2043 | size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1); | |
2044 | ||
2045 | if (zero_offset) { | |
2046 | iosize = PAGE_CACHE_SIZE - zero_offset; | |
2047 | userpage = kmap_atomic(page, KM_USER0); | |
2048 | memset(userpage + zero_offset, 0, iosize); | |
2049 | flush_dcache_page(page); | |
2050 | kunmap_atomic(userpage, KM_USER0); | |
2051 | } | |
2052 | } | |
d1310b2e CM |
2053 | while (cur <= end) { |
2054 | if (cur >= last_byte) { | |
2055 | char *userpage; | |
2056 | iosize = PAGE_CACHE_SIZE - page_offset; | |
2057 | userpage = kmap_atomic(page, KM_USER0); | |
2058 | memset(userpage + page_offset, 0, iosize); | |
2059 | flush_dcache_page(page); | |
2060 | kunmap_atomic(userpage, KM_USER0); | |
2061 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
2062 | GFP_NOFS); | |
2063 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
2064 | break; | |
2065 | } | |
2066 | em = get_extent(inode, page, page_offset, cur, | |
2067 | end - cur + 1, 0); | |
2068 | if (IS_ERR(em) || !em) { | |
2069 | SetPageError(page); | |
2070 | unlock_extent(tree, cur, end, GFP_NOFS); | |
2071 | break; | |
2072 | } | |
d1310b2e CM |
2073 | extent_offset = cur - em->start; |
2074 | BUG_ON(extent_map_end(em) <= cur); | |
2075 | BUG_ON(end < cur); | |
2076 | ||
c8b97818 CM |
2077 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
2078 | this_bio_flag = EXTENT_BIO_COMPRESSED; | |
2079 | ||
d1310b2e CM |
2080 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
2081 | cur_end = min(extent_map_end(em) - 1, end); | |
2082 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
c8b97818 CM |
2083 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
2084 | disk_io_size = em->block_len; | |
2085 | sector = em->block_start >> 9; | |
2086 | } else { | |
2087 | sector = (em->block_start + extent_offset) >> 9; | |
2088 | disk_io_size = iosize; | |
2089 | } | |
d1310b2e CM |
2090 | bdev = em->bdev; |
2091 | block_start = em->block_start; | |
d899e052 YZ |
2092 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2093 | block_start = EXTENT_MAP_HOLE; | |
d1310b2e CM |
2094 | free_extent_map(em); |
2095 | em = NULL; | |
2096 | ||
2097 | /* we've found a hole, just zero and go on */ | |
2098 | if (block_start == EXTENT_MAP_HOLE) { | |
2099 | char *userpage; | |
2100 | userpage = kmap_atomic(page, KM_USER0); | |
2101 | memset(userpage + page_offset, 0, iosize); | |
2102 | flush_dcache_page(page); | |
2103 | kunmap_atomic(userpage, KM_USER0); | |
2104 | ||
2105 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
2106 | GFP_NOFS); | |
2107 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
2108 | cur = cur + iosize; | |
2109 | page_offset += iosize; | |
2110 | continue; | |
2111 | } | |
2112 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
2113 | if (test_range_bit(tree, cur, cur_end, |
2114 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 2115 | check_page_uptodate(tree, page); |
d1310b2e CM |
2116 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); |
2117 | cur = cur + iosize; | |
2118 | page_offset += iosize; | |
2119 | continue; | |
2120 | } | |
70dec807 CM |
2121 | /* we have an inline extent but it didn't get marked up |
2122 | * to date. Error out | |
2123 | */ | |
2124 | if (block_start == EXTENT_MAP_INLINE) { | |
2125 | SetPageError(page); | |
2126 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
2127 | cur = cur + iosize; | |
2128 | page_offset += iosize; | |
2129 | continue; | |
2130 | } | |
d1310b2e CM |
2131 | |
2132 | ret = 0; | |
2133 | if (tree->ops && tree->ops->readpage_io_hook) { | |
2134 | ret = tree->ops->readpage_io_hook(page, cur, | |
2135 | cur + iosize - 1); | |
2136 | } | |
2137 | if (!ret) { | |
89642229 CM |
2138 | unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1; |
2139 | pnr -= page->index; | |
d1310b2e | 2140 | ret = submit_extent_page(READ, tree, page, |
c8b97818 | 2141 | sector, disk_io_size, page_offset, |
89642229 | 2142 | bdev, bio, pnr, |
c8b97818 CM |
2143 | end_bio_extent_readpage, mirror_num, |
2144 | *bio_flags, | |
2145 | this_bio_flag); | |
89642229 | 2146 | nr++; |
c8b97818 | 2147 | *bio_flags = this_bio_flag; |
d1310b2e CM |
2148 | } |
2149 | if (ret) | |
2150 | SetPageError(page); | |
2151 | cur = cur + iosize; | |
2152 | page_offset += iosize; | |
d1310b2e CM |
2153 | } |
2154 | if (!nr) { | |
2155 | if (!PageError(page)) | |
2156 | SetPageUptodate(page); | |
2157 | unlock_page(page); | |
2158 | } | |
2159 | return 0; | |
2160 | } | |
2161 | ||
2162 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, | |
2163 | get_extent_t *get_extent) | |
2164 | { | |
2165 | struct bio *bio = NULL; | |
c8b97818 | 2166 | unsigned long bio_flags = 0; |
d1310b2e CM |
2167 | int ret; |
2168 | ||
c8b97818 CM |
2169 | ret = __extent_read_full_page(tree, page, get_extent, &bio, 0, |
2170 | &bio_flags); | |
d1310b2e | 2171 | if (bio) |
c8b97818 | 2172 | submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
2173 | return ret; |
2174 | } | |
d1310b2e | 2175 | |
11c8349b CM |
2176 | static noinline void update_nr_written(struct page *page, |
2177 | struct writeback_control *wbc, | |
2178 | unsigned long nr_written) | |
2179 | { | |
2180 | wbc->nr_to_write -= nr_written; | |
2181 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && | |
2182 | wbc->range_start == 0 && wbc->range_end == LLONG_MAX)) | |
2183 | page->mapping->writeback_index = page->index + nr_written; | |
2184 | } | |
2185 | ||
d1310b2e CM |
2186 | /* |
2187 | * the writepage semantics are similar to regular writepage. extent | |
2188 | * records are inserted to lock ranges in the tree, and as dirty areas | |
2189 | * are found, they are marked writeback. Then the lock bits are removed | |
2190 | * and the end_io handler clears the writeback ranges | |
2191 | */ | |
2192 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
2193 | void *data) | |
2194 | { | |
2195 | struct inode *inode = page->mapping->host; | |
2196 | struct extent_page_data *epd = data; | |
2197 | struct extent_io_tree *tree = epd->tree; | |
2198 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2199 | u64 delalloc_start; | |
2200 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2201 | u64 end; | |
2202 | u64 cur = start; | |
2203 | u64 extent_offset; | |
2204 | u64 last_byte = i_size_read(inode); | |
2205 | u64 block_start; | |
2206 | u64 iosize; | |
e6dcd2dc | 2207 | u64 unlock_start; |
d1310b2e | 2208 | sector_t sector; |
2c64c53d | 2209 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
2210 | struct extent_map *em; |
2211 | struct block_device *bdev; | |
2212 | int ret; | |
2213 | int nr = 0; | |
7f3c74fb | 2214 | size_t pg_offset = 0; |
d1310b2e CM |
2215 | size_t blocksize; |
2216 | loff_t i_size = i_size_read(inode); | |
2217 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
2218 | u64 nr_delalloc; | |
2219 | u64 delalloc_end; | |
c8b97818 CM |
2220 | int page_started; |
2221 | int compressed; | |
ffbd517d | 2222 | int write_flags; |
771ed689 | 2223 | unsigned long nr_written = 0; |
d1310b2e | 2224 | |
ffbd517d CM |
2225 | if (wbc->sync_mode == WB_SYNC_ALL) |
2226 | write_flags = WRITE_SYNC_PLUG; | |
2227 | else | |
2228 | write_flags = WRITE; | |
2229 | ||
d1310b2e | 2230 | WARN_ON(!PageLocked(page)); |
7f3c74fb | 2231 | pg_offset = i_size & (PAGE_CACHE_SIZE - 1); |
211c17f5 | 2232 | if (page->index > end_index || |
7f3c74fb | 2233 | (page->index == end_index && !pg_offset)) { |
39be25cd | 2234 | page->mapping->a_ops->invalidatepage(page, 0); |
d1310b2e CM |
2235 | unlock_page(page); |
2236 | return 0; | |
2237 | } | |
2238 | ||
2239 | if (page->index == end_index) { | |
2240 | char *userpage; | |
2241 | ||
d1310b2e | 2242 | userpage = kmap_atomic(page, KM_USER0); |
7f3c74fb CM |
2243 | memset(userpage + pg_offset, 0, |
2244 | PAGE_CACHE_SIZE - pg_offset); | |
d1310b2e | 2245 | kunmap_atomic(userpage, KM_USER0); |
211c17f5 | 2246 | flush_dcache_page(page); |
d1310b2e | 2247 | } |
7f3c74fb | 2248 | pg_offset = 0; |
d1310b2e CM |
2249 | |
2250 | set_page_extent_mapped(page); | |
2251 | ||
2252 | delalloc_start = start; | |
2253 | delalloc_end = 0; | |
c8b97818 | 2254 | page_started = 0; |
771ed689 | 2255 | if (!epd->extent_locked) { |
f85d7d6c | 2256 | u64 delalloc_to_write = 0; |
11c8349b CM |
2257 | /* |
2258 | * make sure the wbc mapping index is at least updated | |
2259 | * to this page. | |
2260 | */ | |
2261 | update_nr_written(page, wbc, 0); | |
2262 | ||
d397712b | 2263 | while (delalloc_end < page_end) { |
771ed689 | 2264 | nr_delalloc = find_lock_delalloc_range(inode, tree, |
c8b97818 CM |
2265 | page, |
2266 | &delalloc_start, | |
d1310b2e CM |
2267 | &delalloc_end, |
2268 | 128 * 1024 * 1024); | |
771ed689 CM |
2269 | if (nr_delalloc == 0) { |
2270 | delalloc_start = delalloc_end + 1; | |
2271 | continue; | |
2272 | } | |
2273 | tree->ops->fill_delalloc(inode, page, delalloc_start, | |
2274 | delalloc_end, &page_started, | |
2275 | &nr_written); | |
f85d7d6c CM |
2276 | /* |
2277 | * delalloc_end is already one less than the total | |
2278 | * length, so we don't subtract one from | |
2279 | * PAGE_CACHE_SIZE | |
2280 | */ | |
2281 | delalloc_to_write += (delalloc_end - delalloc_start + | |
2282 | PAGE_CACHE_SIZE) >> | |
2283 | PAGE_CACHE_SHIFT; | |
d1310b2e | 2284 | delalloc_start = delalloc_end + 1; |
d1310b2e | 2285 | } |
f85d7d6c CM |
2286 | if (wbc->nr_to_write < delalloc_to_write) { |
2287 | int thresh = 8192; | |
2288 | ||
2289 | if (delalloc_to_write < thresh * 2) | |
2290 | thresh = delalloc_to_write; | |
2291 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
2292 | thresh); | |
2293 | } | |
c8b97818 | 2294 | |
771ed689 CM |
2295 | /* did the fill delalloc function already unlock and start |
2296 | * the IO? | |
2297 | */ | |
2298 | if (page_started) { | |
2299 | ret = 0; | |
11c8349b CM |
2300 | /* |
2301 | * we've unlocked the page, so we can't update | |
2302 | * the mapping's writeback index, just update | |
2303 | * nr_to_write. | |
2304 | */ | |
2305 | wbc->nr_to_write -= nr_written; | |
2306 | goto done_unlocked; | |
771ed689 | 2307 | } |
c8b97818 | 2308 | } |
247e743c | 2309 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
2310 | ret = tree->ops->writepage_start_hook(page, start, |
2311 | page_end); | |
247e743c | 2312 | if (ret == -EAGAIN) { |
247e743c | 2313 | redirty_page_for_writepage(wbc, page); |
11c8349b | 2314 | update_nr_written(page, wbc, nr_written); |
247e743c | 2315 | unlock_page(page); |
771ed689 | 2316 | ret = 0; |
11c8349b | 2317 | goto done_unlocked; |
247e743c CM |
2318 | } |
2319 | } | |
2320 | ||
11c8349b CM |
2321 | /* |
2322 | * we don't want to touch the inode after unlocking the page, | |
2323 | * so we update the mapping writeback index now | |
2324 | */ | |
2325 | update_nr_written(page, wbc, nr_written + 1); | |
771ed689 | 2326 | |
d1310b2e | 2327 | end = page_end; |
d1310b2e | 2328 | if (last_byte <= start) { |
e6dcd2dc CM |
2329 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2330 | tree->ops->writepage_end_io_hook(page, start, | |
2331 | page_end, NULL, 1); | |
2332 | unlock_start = page_end + 1; | |
d1310b2e CM |
2333 | goto done; |
2334 | } | |
2335 | ||
d1310b2e CM |
2336 | blocksize = inode->i_sb->s_blocksize; |
2337 | ||
2338 | while (cur <= end) { | |
2339 | if (cur >= last_byte) { | |
e6dcd2dc CM |
2340 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2341 | tree->ops->writepage_end_io_hook(page, cur, | |
2342 | page_end, NULL, 1); | |
2343 | unlock_start = page_end + 1; | |
d1310b2e CM |
2344 | break; |
2345 | } | |
7f3c74fb | 2346 | em = epd->get_extent(inode, page, pg_offset, cur, |
d1310b2e CM |
2347 | end - cur + 1, 1); |
2348 | if (IS_ERR(em) || !em) { | |
2349 | SetPageError(page); | |
2350 | break; | |
2351 | } | |
2352 | ||
2353 | extent_offset = cur - em->start; | |
2354 | BUG_ON(extent_map_end(em) <= cur); | |
2355 | BUG_ON(end < cur); | |
2356 | iosize = min(extent_map_end(em) - cur, end - cur + 1); | |
2357 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
2358 | sector = (em->block_start + extent_offset) >> 9; | |
2359 | bdev = em->bdev; | |
2360 | block_start = em->block_start; | |
c8b97818 | 2361 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
2362 | free_extent_map(em); |
2363 | em = NULL; | |
2364 | ||
c8b97818 CM |
2365 | /* |
2366 | * compressed and inline extents are written through other | |
2367 | * paths in the FS | |
2368 | */ | |
2369 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 2370 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
2371 | /* |
2372 | * end_io notification does not happen here for | |
2373 | * compressed extents | |
2374 | */ | |
2375 | if (!compressed && tree->ops && | |
2376 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
2377 | tree->ops->writepage_end_io_hook(page, cur, |
2378 | cur + iosize - 1, | |
2379 | NULL, 1); | |
c8b97818 CM |
2380 | else if (compressed) { |
2381 | /* we don't want to end_page_writeback on | |
2382 | * a compressed extent. this happens | |
2383 | * elsewhere | |
2384 | */ | |
2385 | nr++; | |
2386 | } | |
2387 | ||
2388 | cur += iosize; | |
7f3c74fb | 2389 | pg_offset += iosize; |
e6dcd2dc | 2390 | unlock_start = cur; |
d1310b2e CM |
2391 | continue; |
2392 | } | |
d1310b2e CM |
2393 | /* leave this out until we have a page_mkwrite call */ |
2394 | if (0 && !test_range_bit(tree, cur, cur + iosize - 1, | |
9655d298 | 2395 | EXTENT_DIRTY, 0, NULL)) { |
d1310b2e | 2396 | cur = cur + iosize; |
7f3c74fb | 2397 | pg_offset += iosize; |
d1310b2e CM |
2398 | continue; |
2399 | } | |
c8b97818 | 2400 | |
d1310b2e CM |
2401 | if (tree->ops && tree->ops->writepage_io_hook) { |
2402 | ret = tree->ops->writepage_io_hook(page, cur, | |
2403 | cur + iosize - 1); | |
2404 | } else { | |
2405 | ret = 0; | |
2406 | } | |
1259ab75 | 2407 | if (ret) { |
d1310b2e | 2408 | SetPageError(page); |
1259ab75 | 2409 | } else { |
d1310b2e | 2410 | unsigned long max_nr = end_index + 1; |
7f3c74fb | 2411 | |
d1310b2e CM |
2412 | set_range_writeback(tree, cur, cur + iosize - 1); |
2413 | if (!PageWriteback(page)) { | |
d397712b CM |
2414 | printk(KERN_ERR "btrfs warning page %lu not " |
2415 | "writeback, cur %llu end %llu\n", | |
2416 | page->index, (unsigned long long)cur, | |
d1310b2e CM |
2417 | (unsigned long long)end); |
2418 | } | |
2419 | ||
ffbd517d CM |
2420 | ret = submit_extent_page(write_flags, tree, page, |
2421 | sector, iosize, pg_offset, | |
2422 | bdev, &epd->bio, max_nr, | |
c8b97818 CM |
2423 | end_bio_extent_writepage, |
2424 | 0, 0, 0); | |
d1310b2e CM |
2425 | if (ret) |
2426 | SetPageError(page); | |
2427 | } | |
2428 | cur = cur + iosize; | |
7f3c74fb | 2429 | pg_offset += iosize; |
d1310b2e CM |
2430 | nr++; |
2431 | } | |
2432 | done: | |
2433 | if (nr == 0) { | |
2434 | /* make sure the mapping tag for page dirty gets cleared */ | |
2435 | set_page_writeback(page); | |
2436 | end_page_writeback(page); | |
2437 | } | |
d1310b2e | 2438 | unlock_page(page); |
771ed689 | 2439 | |
11c8349b CM |
2440 | done_unlocked: |
2441 | ||
2c64c53d CM |
2442 | /* drop our reference on any cached states */ |
2443 | free_extent_state(cached_state); | |
d1310b2e CM |
2444 | return 0; |
2445 | } | |
2446 | ||
d1310b2e | 2447 | /** |
4bef0848 | 2448 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
2449 | * @mapping: address space structure to write |
2450 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
2451 | * @writepage: function called for each page | |
2452 | * @data: data passed to writepage function | |
2453 | * | |
2454 | * If a page is already under I/O, write_cache_pages() skips it, even | |
2455 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
2456 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
2457 | * and msync() need to guarantee that all the data which was dirty at the time | |
2458 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
2459 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
2460 | * existing IO to complete. | |
2461 | */ | |
b2950863 | 2462 | static int extent_write_cache_pages(struct extent_io_tree *tree, |
4bef0848 CM |
2463 | struct address_space *mapping, |
2464 | struct writeback_control *wbc, | |
d2c3f4f6 CM |
2465 | writepage_t writepage, void *data, |
2466 | void (*flush_fn)(void *)) | |
d1310b2e | 2467 | { |
d1310b2e CM |
2468 | int ret = 0; |
2469 | int done = 0; | |
f85d7d6c | 2470 | int nr_to_write_done = 0; |
d1310b2e CM |
2471 | struct pagevec pvec; |
2472 | int nr_pages; | |
2473 | pgoff_t index; | |
2474 | pgoff_t end; /* Inclusive */ | |
2475 | int scanned = 0; | |
2476 | int range_whole = 0; | |
2477 | ||
d1310b2e CM |
2478 | pagevec_init(&pvec, 0); |
2479 | if (wbc->range_cyclic) { | |
2480 | index = mapping->writeback_index; /* Start from prev offset */ | |
2481 | end = -1; | |
2482 | } else { | |
2483 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
2484 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
2485 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) | |
2486 | range_whole = 1; | |
2487 | scanned = 1; | |
2488 | } | |
2489 | retry: | |
f85d7d6c | 2490 | while (!done && !nr_to_write_done && (index <= end) && |
d1310b2e | 2491 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, |
d397712b CM |
2492 | PAGECACHE_TAG_DIRTY, min(end - index, |
2493 | (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
d1310b2e CM |
2494 | unsigned i; |
2495 | ||
2496 | scanned = 1; | |
2497 | for (i = 0; i < nr_pages; i++) { | |
2498 | struct page *page = pvec.pages[i]; | |
2499 | ||
2500 | /* | |
2501 | * At this point we hold neither mapping->tree_lock nor | |
2502 | * lock on the page itself: the page may be truncated or | |
2503 | * invalidated (changing page->mapping to NULL), or even | |
2504 | * swizzled back from swapper_space to tmpfs file | |
2505 | * mapping | |
2506 | */ | |
4bef0848 CM |
2507 | if (tree->ops && tree->ops->write_cache_pages_lock_hook) |
2508 | tree->ops->write_cache_pages_lock_hook(page); | |
2509 | else | |
2510 | lock_page(page); | |
d1310b2e CM |
2511 | |
2512 | if (unlikely(page->mapping != mapping)) { | |
2513 | unlock_page(page); | |
2514 | continue; | |
2515 | } | |
2516 | ||
2517 | if (!wbc->range_cyclic && page->index > end) { | |
2518 | done = 1; | |
2519 | unlock_page(page); | |
2520 | continue; | |
2521 | } | |
2522 | ||
d2c3f4f6 | 2523 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 CM |
2524 | if (PageWriteback(page)) |
2525 | flush_fn(data); | |
d1310b2e | 2526 | wait_on_page_writeback(page); |
d2c3f4f6 | 2527 | } |
d1310b2e CM |
2528 | |
2529 | if (PageWriteback(page) || | |
2530 | !clear_page_dirty_for_io(page)) { | |
2531 | unlock_page(page); | |
2532 | continue; | |
2533 | } | |
2534 | ||
2535 | ret = (*writepage)(page, wbc, data); | |
2536 | ||
2537 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
2538 | unlock_page(page); | |
2539 | ret = 0; | |
2540 | } | |
f85d7d6c | 2541 | if (ret) |
d1310b2e | 2542 | done = 1; |
f85d7d6c CM |
2543 | |
2544 | /* | |
2545 | * the filesystem may choose to bump up nr_to_write. | |
2546 | * We have to make sure to honor the new nr_to_write | |
2547 | * at any time | |
2548 | */ | |
2549 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
2550 | } |
2551 | pagevec_release(&pvec); | |
2552 | cond_resched(); | |
2553 | } | |
2554 | if (!scanned && !done) { | |
2555 | /* | |
2556 | * We hit the last page and there is more work to be done: wrap | |
2557 | * back to the start of the file | |
2558 | */ | |
2559 | scanned = 1; | |
2560 | index = 0; | |
2561 | goto retry; | |
2562 | } | |
d1310b2e CM |
2563 | return ret; |
2564 | } | |
d1310b2e | 2565 | |
ffbd517d | 2566 | static void flush_epd_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 2567 | { |
d2c3f4f6 | 2568 | if (epd->bio) { |
ffbd517d CM |
2569 | if (epd->sync_io) |
2570 | submit_one_bio(WRITE_SYNC, epd->bio, 0, 0); | |
2571 | else | |
2572 | submit_one_bio(WRITE, epd->bio, 0, 0); | |
d2c3f4f6 CM |
2573 | epd->bio = NULL; |
2574 | } | |
2575 | } | |
2576 | ||
ffbd517d CM |
2577 | static noinline void flush_write_bio(void *data) |
2578 | { | |
2579 | struct extent_page_data *epd = data; | |
2580 | flush_epd_write_bio(epd); | |
2581 | } | |
2582 | ||
d1310b2e CM |
2583 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, |
2584 | get_extent_t *get_extent, | |
2585 | struct writeback_control *wbc) | |
2586 | { | |
2587 | int ret; | |
2588 | struct address_space *mapping = page->mapping; | |
2589 | struct extent_page_data epd = { | |
2590 | .bio = NULL, | |
2591 | .tree = tree, | |
2592 | .get_extent = get_extent, | |
771ed689 | 2593 | .extent_locked = 0, |
ffbd517d | 2594 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
2595 | }; |
2596 | struct writeback_control wbc_writepages = { | |
d313d7a3 | 2597 | .sync_mode = wbc->sync_mode, |
d1310b2e CM |
2598 | .older_than_this = NULL, |
2599 | .nr_to_write = 64, | |
2600 | .range_start = page_offset(page) + PAGE_CACHE_SIZE, | |
2601 | .range_end = (loff_t)-1, | |
2602 | }; | |
2603 | ||
d1310b2e CM |
2604 | ret = __extent_writepage(page, wbc, &epd); |
2605 | ||
4bef0848 | 2606 | extent_write_cache_pages(tree, mapping, &wbc_writepages, |
d2c3f4f6 | 2607 | __extent_writepage, &epd, flush_write_bio); |
ffbd517d | 2608 | flush_epd_write_bio(&epd); |
d1310b2e CM |
2609 | return ret; |
2610 | } | |
d1310b2e | 2611 | |
771ed689 CM |
2612 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, |
2613 | u64 start, u64 end, get_extent_t *get_extent, | |
2614 | int mode) | |
2615 | { | |
2616 | int ret = 0; | |
2617 | struct address_space *mapping = inode->i_mapping; | |
2618 | struct page *page; | |
2619 | unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >> | |
2620 | PAGE_CACHE_SHIFT; | |
2621 | ||
2622 | struct extent_page_data epd = { | |
2623 | .bio = NULL, | |
2624 | .tree = tree, | |
2625 | .get_extent = get_extent, | |
2626 | .extent_locked = 1, | |
ffbd517d | 2627 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
2628 | }; |
2629 | struct writeback_control wbc_writepages = { | |
771ed689 CM |
2630 | .sync_mode = mode, |
2631 | .older_than_this = NULL, | |
2632 | .nr_to_write = nr_pages * 2, | |
2633 | .range_start = start, | |
2634 | .range_end = end + 1, | |
2635 | }; | |
2636 | ||
d397712b | 2637 | while (start <= end) { |
771ed689 CM |
2638 | page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); |
2639 | if (clear_page_dirty_for_io(page)) | |
2640 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
2641 | else { | |
2642 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
2643 | tree->ops->writepage_end_io_hook(page, start, | |
2644 | start + PAGE_CACHE_SIZE - 1, | |
2645 | NULL, 1); | |
2646 | unlock_page(page); | |
2647 | } | |
2648 | page_cache_release(page); | |
2649 | start += PAGE_CACHE_SIZE; | |
2650 | } | |
2651 | ||
ffbd517d | 2652 | flush_epd_write_bio(&epd); |
771ed689 CM |
2653 | return ret; |
2654 | } | |
d1310b2e CM |
2655 | |
2656 | int extent_writepages(struct extent_io_tree *tree, | |
2657 | struct address_space *mapping, | |
2658 | get_extent_t *get_extent, | |
2659 | struct writeback_control *wbc) | |
2660 | { | |
2661 | int ret = 0; | |
2662 | struct extent_page_data epd = { | |
2663 | .bio = NULL, | |
2664 | .tree = tree, | |
2665 | .get_extent = get_extent, | |
771ed689 | 2666 | .extent_locked = 0, |
ffbd517d | 2667 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
2668 | }; |
2669 | ||
4bef0848 | 2670 | ret = extent_write_cache_pages(tree, mapping, wbc, |
d2c3f4f6 CM |
2671 | __extent_writepage, &epd, |
2672 | flush_write_bio); | |
ffbd517d | 2673 | flush_epd_write_bio(&epd); |
d1310b2e CM |
2674 | return ret; |
2675 | } | |
d1310b2e CM |
2676 | |
2677 | int extent_readpages(struct extent_io_tree *tree, | |
2678 | struct address_space *mapping, | |
2679 | struct list_head *pages, unsigned nr_pages, | |
2680 | get_extent_t get_extent) | |
2681 | { | |
2682 | struct bio *bio = NULL; | |
2683 | unsigned page_idx; | |
c8b97818 | 2684 | unsigned long bio_flags = 0; |
d1310b2e | 2685 | |
d1310b2e CM |
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); | |
28ecb609 | 2691 | if (!add_to_page_cache_lru(page, mapping, |
d1310b2e | 2692 | page->index, GFP_KERNEL)) { |
f188591e | 2693 | __extent_read_full_page(tree, page, get_extent, |
c8b97818 | 2694 | &bio, 0, &bio_flags); |
d1310b2e CM |
2695 | } |
2696 | page_cache_release(page); | |
2697 | } | |
d1310b2e CM |
2698 | BUG_ON(!list_empty(pages)); |
2699 | if (bio) | |
c8b97818 | 2700 | submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
2701 | return 0; |
2702 | } | |
d1310b2e CM |
2703 | |
2704 | /* | |
2705 | * basic invalidatepage code, this waits on any locked or writeback | |
2706 | * ranges corresponding to the page, and then deletes any extent state | |
2707 | * records from the tree | |
2708 | */ | |
2709 | int extent_invalidatepage(struct extent_io_tree *tree, | |
2710 | struct page *page, unsigned long offset) | |
2711 | { | |
2ac55d41 | 2712 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
2713 | u64 start = ((u64)page->index << PAGE_CACHE_SHIFT); |
2714 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
2715 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
2716 | ||
d397712b | 2717 | start += (offset + blocksize - 1) & ~(blocksize - 1); |
d1310b2e CM |
2718 | if (start > end) |
2719 | return 0; | |
2720 | ||
2ac55d41 | 2721 | lock_extent_bits(tree, start, end, 0, &cached_state, GFP_NOFS); |
1edbb734 | 2722 | wait_on_page_writeback(page); |
d1310b2e | 2723 | clear_extent_bit(tree, start, end, |
32c00aff JB |
2724 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
2725 | EXTENT_DO_ACCOUNTING, | |
2ac55d41 | 2726 | 1, 1, &cached_state, GFP_NOFS); |
d1310b2e CM |
2727 | return 0; |
2728 | } | |
d1310b2e CM |
2729 | |
2730 | /* | |
2731 | * simple commit_write call, set_range_dirty is used to mark both | |
2732 | * the pages and the extent records as dirty | |
2733 | */ | |
2734 | int extent_commit_write(struct extent_io_tree *tree, | |
2735 | struct inode *inode, struct page *page, | |
2736 | unsigned from, unsigned to) | |
2737 | { | |
2738 | loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to; | |
2739 | ||
2740 | set_page_extent_mapped(page); | |
2741 | set_page_dirty(page); | |
2742 | ||
2743 | if (pos > inode->i_size) { | |
2744 | i_size_write(inode, pos); | |
2745 | mark_inode_dirty(inode); | |
2746 | } | |
2747 | return 0; | |
2748 | } | |
d1310b2e CM |
2749 | |
2750 | int extent_prepare_write(struct extent_io_tree *tree, | |
2751 | struct inode *inode, struct page *page, | |
2752 | unsigned from, unsigned to, get_extent_t *get_extent) | |
2753 | { | |
2754 | u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2755 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
2756 | u64 block_start; | |
2757 | u64 orig_block_start; | |
2758 | u64 block_end; | |
2759 | u64 cur_end; | |
2760 | struct extent_map *em; | |
2761 | unsigned blocksize = 1 << inode->i_blkbits; | |
2762 | size_t page_offset = 0; | |
2763 | size_t block_off_start; | |
2764 | size_t block_off_end; | |
2765 | int err = 0; | |
2766 | int iocount = 0; | |
2767 | int ret = 0; | |
2768 | int isnew; | |
2769 | ||
2770 | set_page_extent_mapped(page); | |
2771 | ||
2772 | block_start = (page_start + from) & ~((u64)blocksize - 1); | |
2773 | block_end = (page_start + to - 1) | (blocksize - 1); | |
2774 | orig_block_start = block_start; | |
2775 | ||
2776 | lock_extent(tree, page_start, page_end, GFP_NOFS); | |
d397712b | 2777 | while (block_start <= block_end) { |
d1310b2e CM |
2778 | em = get_extent(inode, page, page_offset, block_start, |
2779 | block_end - block_start + 1, 1); | |
d397712b | 2780 | if (IS_ERR(em) || !em) |
d1310b2e | 2781 | goto err; |
d397712b | 2782 | |
d1310b2e CM |
2783 | cur_end = min(block_end, extent_map_end(em) - 1); |
2784 | block_off_start = block_start & (PAGE_CACHE_SIZE - 1); | |
2785 | block_off_end = block_off_start + blocksize; | |
2786 | isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS); | |
2787 | ||
2788 | if (!PageUptodate(page) && isnew && | |
2789 | (block_off_end > to || block_off_start < from)) { | |
2790 | void *kaddr; | |
2791 | ||
2792 | kaddr = kmap_atomic(page, KM_USER0); | |
2793 | if (block_off_end > to) | |
2794 | memset(kaddr + to, 0, block_off_end - to); | |
2795 | if (block_off_start < from) | |
2796 | memset(kaddr + block_off_start, 0, | |
2797 | from - block_off_start); | |
2798 | flush_dcache_page(page); | |
2799 | kunmap_atomic(kaddr, KM_USER0); | |
2800 | } | |
2801 | if ((em->block_start != EXTENT_MAP_HOLE && | |
2802 | em->block_start != EXTENT_MAP_INLINE) && | |
2803 | !isnew && !PageUptodate(page) && | |
2804 | (block_off_end > to || block_off_start < from) && | |
2805 | !test_range_bit(tree, block_start, cur_end, | |
9655d298 | 2806 | EXTENT_UPTODATE, 1, NULL)) { |
d1310b2e CM |
2807 | u64 sector; |
2808 | u64 extent_offset = block_start - em->start; | |
2809 | size_t iosize; | |
2810 | sector = (em->block_start + extent_offset) >> 9; | |
2811 | iosize = (cur_end - block_start + blocksize) & | |
2812 | ~((u64)blocksize - 1); | |
2813 | /* | |
2814 | * we've already got the extent locked, but we | |
2815 | * need to split the state such that our end_bio | |
2816 | * handler can clear the lock. | |
2817 | */ | |
2818 | set_extent_bit(tree, block_start, | |
2819 | block_start + iosize - 1, | |
2c64c53d | 2820 | EXTENT_LOCKED, 0, NULL, NULL, GFP_NOFS); |
d1310b2e CM |
2821 | ret = submit_extent_page(READ, tree, page, |
2822 | sector, iosize, page_offset, em->bdev, | |
2823 | NULL, 1, | |
c8b97818 CM |
2824 | end_bio_extent_preparewrite, 0, |
2825 | 0, 0); | |
d1310b2e CM |
2826 | iocount++; |
2827 | block_start = block_start + iosize; | |
2828 | } else { | |
2829 | set_extent_uptodate(tree, block_start, cur_end, | |
2830 | GFP_NOFS); | |
2831 | unlock_extent(tree, block_start, cur_end, GFP_NOFS); | |
2832 | block_start = cur_end + 1; | |
2833 | } | |
2834 | page_offset = block_start & (PAGE_CACHE_SIZE - 1); | |
2835 | free_extent_map(em); | |
2836 | } | |
2837 | if (iocount) { | |
2838 | wait_extent_bit(tree, orig_block_start, | |
2839 | block_end, EXTENT_LOCKED); | |
2840 | } | |
2841 | check_page_uptodate(tree, page); | |
2842 | err: | |
2843 | /* FIXME, zero out newly allocated blocks on error */ | |
2844 | return err; | |
2845 | } | |
d1310b2e | 2846 | |
7b13b7b1 CM |
2847 | /* |
2848 | * a helper for releasepage, this tests for areas of the page that | |
2849 | * are locked or under IO and drops the related state bits if it is safe | |
2850 | * to drop the page. | |
2851 | */ | |
2852 | int try_release_extent_state(struct extent_map_tree *map, | |
2853 | struct extent_io_tree *tree, struct page *page, | |
2854 | gfp_t mask) | |
2855 | { | |
2856 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2857 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
2858 | int ret = 1; | |
2859 | ||
211f90e6 | 2860 | if (test_range_bit(tree, start, end, |
8b62b72b | 2861 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
2862 | ret = 0; |
2863 | else { | |
2864 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
2865 | mask = GFP_NOFS; | |
11ef160f CM |
2866 | /* |
2867 | * at this point we can safely clear everything except the | |
2868 | * locked bit and the nodatasum bit | |
2869 | */ | |
2870 | clear_extent_bit(tree, start, end, | |
2871 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), | |
2872 | 0, 0, NULL, mask); | |
7b13b7b1 CM |
2873 | } |
2874 | return ret; | |
2875 | } | |
7b13b7b1 | 2876 | |
d1310b2e CM |
2877 | /* |
2878 | * a helper for releasepage. As long as there are no locked extents | |
2879 | * in the range corresponding to the page, both state records and extent | |
2880 | * map records are removed | |
2881 | */ | |
2882 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
2883 | struct extent_io_tree *tree, struct page *page, |
2884 | gfp_t mask) | |
d1310b2e CM |
2885 | { |
2886 | struct extent_map *em; | |
2887 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2888 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
7b13b7b1 | 2889 | |
70dec807 CM |
2890 | if ((mask & __GFP_WAIT) && |
2891 | page->mapping->host->i_size > 16 * 1024 * 1024) { | |
39b5637f | 2892 | u64 len; |
70dec807 | 2893 | while (start <= end) { |
39b5637f | 2894 | len = end - start + 1; |
890871be | 2895 | write_lock(&map->lock); |
39b5637f | 2896 | em = lookup_extent_mapping(map, start, len); |
70dec807 | 2897 | if (!em || IS_ERR(em)) { |
890871be | 2898 | write_unlock(&map->lock); |
70dec807 CM |
2899 | break; |
2900 | } | |
7f3c74fb CM |
2901 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
2902 | em->start != start) { | |
890871be | 2903 | write_unlock(&map->lock); |
70dec807 CM |
2904 | free_extent_map(em); |
2905 | break; | |
2906 | } | |
2907 | if (!test_range_bit(tree, em->start, | |
2908 | extent_map_end(em) - 1, | |
8b62b72b | 2909 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 2910 | 0, NULL)) { |
70dec807 CM |
2911 | remove_extent_mapping(map, em); |
2912 | /* once for the rb tree */ | |
2913 | free_extent_map(em); | |
2914 | } | |
2915 | start = extent_map_end(em); | |
890871be | 2916 | write_unlock(&map->lock); |
70dec807 CM |
2917 | |
2918 | /* once for us */ | |
d1310b2e CM |
2919 | free_extent_map(em); |
2920 | } | |
d1310b2e | 2921 | } |
7b13b7b1 | 2922 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 2923 | } |
d1310b2e CM |
2924 | |
2925 | sector_t extent_bmap(struct address_space *mapping, sector_t iblock, | |
2926 | get_extent_t *get_extent) | |
2927 | { | |
2928 | struct inode *inode = mapping->host; | |
2ac55d41 | 2929 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
2930 | u64 start = iblock << inode->i_blkbits; |
2931 | sector_t sector = 0; | |
d899e052 | 2932 | size_t blksize = (1 << inode->i_blkbits); |
d1310b2e CM |
2933 | struct extent_map *em; |
2934 | ||
2ac55d41 JB |
2935 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + blksize - 1, |
2936 | 0, &cached_state, GFP_NOFS); | |
d899e052 | 2937 | em = get_extent(inode, NULL, 0, start, blksize, 0); |
2ac55d41 JB |
2938 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, |
2939 | start + blksize - 1, &cached_state, GFP_NOFS); | |
d1310b2e CM |
2940 | if (!em || IS_ERR(em)) |
2941 | return 0; | |
2942 | ||
d899e052 | 2943 | if (em->block_start > EXTENT_MAP_LAST_BYTE) |
d1310b2e CM |
2944 | goto out; |
2945 | ||
2946 | sector = (em->block_start + start - em->start) >> inode->i_blkbits; | |
d1310b2e CM |
2947 | out: |
2948 | free_extent_map(em); | |
2949 | return sector; | |
2950 | } | |
2951 | ||
1506fcc8 YS |
2952 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
2953 | __u64 start, __u64 len, get_extent_t *get_extent) | |
2954 | { | |
2955 | int ret; | |
2956 | u64 off = start; | |
2957 | u64 max = start + len; | |
2958 | u32 flags = 0; | |
2959 | u64 disko = 0; | |
2960 | struct extent_map *em = NULL; | |
2ac55d41 | 2961 | struct extent_state *cached_state = NULL; |
1506fcc8 YS |
2962 | int end = 0; |
2963 | u64 em_start = 0, em_len = 0; | |
2964 | unsigned long emflags; | |
2965 | ret = 0; | |
2966 | ||
2967 | if (len == 0) | |
2968 | return -EINVAL; | |
2969 | ||
2ac55d41 JB |
2970 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0, |
2971 | &cached_state, GFP_NOFS); | |
1506fcc8 YS |
2972 | em = get_extent(inode, NULL, 0, off, max - off, 0); |
2973 | if (!em) | |
2974 | goto out; | |
2975 | if (IS_ERR(em)) { | |
2976 | ret = PTR_ERR(em); | |
2977 | goto out; | |
2978 | } | |
2979 | while (!end) { | |
2980 | off = em->start + em->len; | |
2981 | if (off >= max) | |
2982 | end = 1; | |
2983 | ||
2984 | em_start = em->start; | |
2985 | em_len = em->len; | |
2986 | ||
2987 | disko = 0; | |
2988 | flags = 0; | |
2989 | ||
93dbfad7 | 2990 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
2991 | end = 1; |
2992 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 2993 | } else if (em->block_start == EXTENT_MAP_HOLE) { |
1506fcc8 | 2994 | flags |= FIEMAP_EXTENT_UNWRITTEN; |
93dbfad7 | 2995 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
2996 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
2997 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 2998 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
2999 | flags |= (FIEMAP_EXTENT_DELALLOC | |
3000 | FIEMAP_EXTENT_UNKNOWN); | |
93dbfad7 | 3001 | } else { |
1506fcc8 | 3002 | disko = em->block_start; |
1506fcc8 YS |
3003 | } |
3004 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
3005 | flags |= FIEMAP_EXTENT_ENCODED; | |
3006 | ||
3007 | emflags = em->flags; | |
3008 | free_extent_map(em); | |
3009 | em = NULL; | |
3010 | ||
3011 | if (!end) { | |
3012 | em = get_extent(inode, NULL, 0, off, max - off, 0); | |
3013 | if (!em) | |
3014 | goto out; | |
3015 | if (IS_ERR(em)) { | |
3016 | ret = PTR_ERR(em); | |
3017 | goto out; | |
3018 | } | |
3019 | emflags = em->flags; | |
3020 | } | |
3021 | if (test_bit(EXTENT_FLAG_VACANCY, &emflags)) { | |
3022 | flags |= FIEMAP_EXTENT_LAST; | |
3023 | end = 1; | |
3024 | } | |
3025 | ||
3026 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, | |
3027 | em_len, flags); | |
3028 | if (ret) | |
3029 | goto out_free; | |
3030 | } | |
3031 | out_free: | |
3032 | free_extent_map(em); | |
3033 | out: | |
2ac55d41 JB |
3034 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len, |
3035 | &cached_state, GFP_NOFS); | |
1506fcc8 YS |
3036 | return ret; |
3037 | } | |
3038 | ||
d1310b2e CM |
3039 | static inline struct page *extent_buffer_page(struct extent_buffer *eb, |
3040 | unsigned long i) | |
3041 | { | |
3042 | struct page *p; | |
3043 | struct address_space *mapping; | |
3044 | ||
3045 | if (i == 0) | |
3046 | return eb->first_page; | |
3047 | i += eb->start >> PAGE_CACHE_SHIFT; | |
3048 | mapping = eb->first_page->mapping; | |
33958dc6 CM |
3049 | if (!mapping) |
3050 | return NULL; | |
0ee0fda0 SW |
3051 | |
3052 | /* | |
3053 | * extent_buffer_page is only called after pinning the page | |
3054 | * by increasing the reference count. So we know the page must | |
3055 | * be in the radix tree. | |
3056 | */ | |
0ee0fda0 | 3057 | rcu_read_lock(); |
d1310b2e | 3058 | p = radix_tree_lookup(&mapping->page_tree, i); |
0ee0fda0 | 3059 | rcu_read_unlock(); |
2b1f55b0 | 3060 | |
d1310b2e CM |
3061 | return p; |
3062 | } | |
3063 | ||
6af118ce | 3064 | static inline unsigned long num_extent_pages(u64 start, u64 len) |
728131d8 | 3065 | { |
6af118ce CM |
3066 | return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) - |
3067 | (start >> PAGE_CACHE_SHIFT); | |
728131d8 CM |
3068 | } |
3069 | ||
d1310b2e CM |
3070 | static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree, |
3071 | u64 start, | |
3072 | unsigned long len, | |
3073 | gfp_t mask) | |
3074 | { | |
3075 | struct extent_buffer *eb = NULL; | |
3935127c | 3076 | #if LEAK_DEBUG |
2d2ae547 | 3077 | unsigned long flags; |
4bef0848 | 3078 | #endif |
d1310b2e | 3079 | |
d1310b2e | 3080 | eb = kmem_cache_zalloc(extent_buffer_cache, mask); |
d1310b2e CM |
3081 | eb->start = start; |
3082 | eb->len = len; | |
b4ce94de CM |
3083 | spin_lock_init(&eb->lock); |
3084 | init_waitqueue_head(&eb->lock_wq); | |
3085 | ||
3935127c | 3086 | #if LEAK_DEBUG |
2d2ae547 CM |
3087 | spin_lock_irqsave(&leak_lock, flags); |
3088 | list_add(&eb->leak_list, &buffers); | |
3089 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 3090 | #endif |
d1310b2e CM |
3091 | atomic_set(&eb->refs, 1); |
3092 | ||
3093 | return eb; | |
3094 | } | |
3095 | ||
3096 | static void __free_extent_buffer(struct extent_buffer *eb) | |
3097 | { | |
3935127c | 3098 | #if LEAK_DEBUG |
2d2ae547 CM |
3099 | unsigned long flags; |
3100 | spin_lock_irqsave(&leak_lock, flags); | |
3101 | list_del(&eb->leak_list); | |
3102 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 3103 | #endif |
d1310b2e CM |
3104 | kmem_cache_free(extent_buffer_cache, eb); |
3105 | } | |
3106 | ||
3107 | struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree, | |
3108 | u64 start, unsigned long len, | |
3109 | struct page *page0, | |
3110 | gfp_t mask) | |
3111 | { | |
3112 | unsigned long num_pages = num_extent_pages(start, len); | |
3113 | unsigned long i; | |
3114 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
3115 | struct extent_buffer *eb; | |
6af118ce | 3116 | struct extent_buffer *exists = NULL; |
d1310b2e CM |
3117 | struct page *p; |
3118 | struct address_space *mapping = tree->mapping; | |
3119 | int uptodate = 1; | |
3120 | ||
6af118ce CM |
3121 | spin_lock(&tree->buffer_lock); |
3122 | eb = buffer_search(tree, start); | |
3123 | if (eb) { | |
3124 | atomic_inc(&eb->refs); | |
3125 | spin_unlock(&tree->buffer_lock); | |
0f9dd46c | 3126 | mark_page_accessed(eb->first_page); |
6af118ce CM |
3127 | return eb; |
3128 | } | |
3129 | spin_unlock(&tree->buffer_lock); | |
3130 | ||
d1310b2e | 3131 | eb = __alloc_extent_buffer(tree, start, len, mask); |
2b114d1d | 3132 | if (!eb) |
d1310b2e CM |
3133 | return NULL; |
3134 | ||
d1310b2e CM |
3135 | if (page0) { |
3136 | eb->first_page = page0; | |
3137 | i = 1; | |
3138 | index++; | |
3139 | page_cache_get(page0); | |
3140 | mark_page_accessed(page0); | |
3141 | set_page_extent_mapped(page0); | |
d1310b2e | 3142 | set_page_extent_head(page0, len); |
f188591e | 3143 | uptodate = PageUptodate(page0); |
d1310b2e CM |
3144 | } else { |
3145 | i = 0; | |
3146 | } | |
3147 | for (; i < num_pages; i++, index++) { | |
3148 | p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM); | |
3149 | if (!p) { | |
3150 | WARN_ON(1); | |
6af118ce | 3151 | goto free_eb; |
d1310b2e CM |
3152 | } |
3153 | set_page_extent_mapped(p); | |
3154 | mark_page_accessed(p); | |
3155 | if (i == 0) { | |
3156 | eb->first_page = p; | |
3157 | set_page_extent_head(p, len); | |
3158 | } else { | |
3159 | set_page_private(p, EXTENT_PAGE_PRIVATE); | |
3160 | } | |
3161 | if (!PageUptodate(p)) | |
3162 | uptodate = 0; | |
3163 | unlock_page(p); | |
3164 | } | |
3165 | if (uptodate) | |
b4ce94de | 3166 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 3167 | |
6af118ce CM |
3168 | spin_lock(&tree->buffer_lock); |
3169 | exists = buffer_tree_insert(tree, start, &eb->rb_node); | |
3170 | if (exists) { | |
3171 | /* add one reference for the caller */ | |
3172 | atomic_inc(&exists->refs); | |
3173 | spin_unlock(&tree->buffer_lock); | |
3174 | goto free_eb; | |
3175 | } | |
6af118ce CM |
3176 | /* add one reference for the tree */ |
3177 | atomic_inc(&eb->refs); | |
f044ba78 | 3178 | spin_unlock(&tree->buffer_lock); |
d1310b2e CM |
3179 | return eb; |
3180 | ||
6af118ce | 3181 | free_eb: |
d1310b2e | 3182 | if (!atomic_dec_and_test(&eb->refs)) |
6af118ce CM |
3183 | return exists; |
3184 | for (index = 1; index < i; index++) | |
d1310b2e | 3185 | page_cache_release(extent_buffer_page(eb, index)); |
6af118ce | 3186 | page_cache_release(extent_buffer_page(eb, 0)); |
d1310b2e | 3187 | __free_extent_buffer(eb); |
6af118ce | 3188 | return exists; |
d1310b2e | 3189 | } |
d1310b2e CM |
3190 | |
3191 | struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree, | |
3192 | u64 start, unsigned long len, | |
3193 | gfp_t mask) | |
3194 | { | |
d1310b2e | 3195 | struct extent_buffer *eb; |
d1310b2e | 3196 | |
6af118ce CM |
3197 | spin_lock(&tree->buffer_lock); |
3198 | eb = buffer_search(tree, start); | |
3199 | if (eb) | |
3200 | atomic_inc(&eb->refs); | |
3201 | spin_unlock(&tree->buffer_lock); | |
d1310b2e | 3202 | |
0f9dd46c JB |
3203 | if (eb) |
3204 | mark_page_accessed(eb->first_page); | |
3205 | ||
d1310b2e | 3206 | return eb; |
d1310b2e | 3207 | } |
d1310b2e CM |
3208 | |
3209 | void free_extent_buffer(struct extent_buffer *eb) | |
3210 | { | |
d1310b2e CM |
3211 | if (!eb) |
3212 | return; | |
3213 | ||
3214 | if (!atomic_dec_and_test(&eb->refs)) | |
3215 | return; | |
3216 | ||
6af118ce | 3217 | WARN_ON(1); |
d1310b2e | 3218 | } |
d1310b2e CM |
3219 | |
3220 | int clear_extent_buffer_dirty(struct extent_io_tree *tree, | |
3221 | struct extent_buffer *eb) | |
3222 | { | |
d1310b2e CM |
3223 | unsigned long i; |
3224 | unsigned long num_pages; | |
3225 | struct page *page; | |
3226 | ||
d1310b2e CM |
3227 | num_pages = num_extent_pages(eb->start, eb->len); |
3228 | ||
3229 | for (i = 0; i < num_pages; i++) { | |
3230 | page = extent_buffer_page(eb, i); | |
b9473439 | 3231 | if (!PageDirty(page)) |
d2c3f4f6 CM |
3232 | continue; |
3233 | ||
a61e6f29 | 3234 | lock_page(page); |
d1310b2e CM |
3235 | if (i == 0) |
3236 | set_page_extent_head(page, eb->len); | |
3237 | else | |
3238 | set_page_private(page, EXTENT_PAGE_PRIVATE); | |
3239 | ||
d1310b2e | 3240 | clear_page_dirty_for_io(page); |
0ee0fda0 | 3241 | spin_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
3242 | if (!PageDirty(page)) { |
3243 | radix_tree_tag_clear(&page->mapping->page_tree, | |
3244 | page_index(page), | |
3245 | PAGECACHE_TAG_DIRTY); | |
3246 | } | |
0ee0fda0 | 3247 | spin_unlock_irq(&page->mapping->tree_lock); |
a61e6f29 | 3248 | unlock_page(page); |
d1310b2e CM |
3249 | } |
3250 | return 0; | |
3251 | } | |
d1310b2e CM |
3252 | |
3253 | int wait_on_extent_buffer_writeback(struct extent_io_tree *tree, | |
3254 | struct extent_buffer *eb) | |
3255 | { | |
3256 | return wait_on_extent_writeback(tree, eb->start, | |
3257 | eb->start + eb->len - 1); | |
3258 | } | |
d1310b2e CM |
3259 | |
3260 | int set_extent_buffer_dirty(struct extent_io_tree *tree, | |
3261 | struct extent_buffer *eb) | |
3262 | { | |
3263 | unsigned long i; | |
3264 | unsigned long num_pages; | |
b9473439 | 3265 | int was_dirty = 0; |
d1310b2e | 3266 | |
b9473439 | 3267 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
d1310b2e | 3268 | num_pages = num_extent_pages(eb->start, eb->len); |
b9473439 | 3269 | for (i = 0; i < num_pages; i++) |
d1310b2e | 3270 | __set_page_dirty_nobuffers(extent_buffer_page(eb, i)); |
b9473439 | 3271 | return was_dirty; |
d1310b2e | 3272 | } |
d1310b2e | 3273 | |
1259ab75 | 3274 | int clear_extent_buffer_uptodate(struct extent_io_tree *tree, |
2ac55d41 JB |
3275 | struct extent_buffer *eb, |
3276 | struct extent_state **cached_state) | |
1259ab75 CM |
3277 | { |
3278 | unsigned long i; | |
3279 | struct page *page; | |
3280 | unsigned long num_pages; | |
3281 | ||
3282 | num_pages = num_extent_pages(eb->start, eb->len); | |
b4ce94de | 3283 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
1259ab75 CM |
3284 | |
3285 | clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1, | |
2ac55d41 | 3286 | cached_state, GFP_NOFS); |
1259ab75 CM |
3287 | for (i = 0; i < num_pages; i++) { |
3288 | page = extent_buffer_page(eb, i); | |
33958dc6 CM |
3289 | if (page) |
3290 | ClearPageUptodate(page); | |
1259ab75 CM |
3291 | } |
3292 | return 0; | |
3293 | } | |
3294 | ||
d1310b2e CM |
3295 | int set_extent_buffer_uptodate(struct extent_io_tree *tree, |
3296 | struct extent_buffer *eb) | |
3297 | { | |
3298 | unsigned long i; | |
3299 | struct page *page; | |
3300 | unsigned long num_pages; | |
3301 | ||
3302 | num_pages = num_extent_pages(eb->start, eb->len); | |
3303 | ||
3304 | set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1, | |
3305 | GFP_NOFS); | |
3306 | for (i = 0; i < num_pages; i++) { | |
3307 | page = extent_buffer_page(eb, i); | |
3308 | if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) || | |
3309 | ((i == num_pages - 1) && | |
3310 | ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) { | |
3311 | check_page_uptodate(tree, page); | |
3312 | continue; | |
3313 | } | |
3314 | SetPageUptodate(page); | |
3315 | } | |
3316 | return 0; | |
3317 | } | |
d1310b2e | 3318 | |
ce9adaa5 CM |
3319 | int extent_range_uptodate(struct extent_io_tree *tree, |
3320 | u64 start, u64 end) | |
3321 | { | |
3322 | struct page *page; | |
3323 | int ret; | |
3324 | int pg_uptodate = 1; | |
3325 | int uptodate; | |
3326 | unsigned long index; | |
3327 | ||
9655d298 | 3328 | ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL); |
ce9adaa5 CM |
3329 | if (ret) |
3330 | return 1; | |
d397712b | 3331 | while (start <= end) { |
ce9adaa5 CM |
3332 | index = start >> PAGE_CACHE_SHIFT; |
3333 | page = find_get_page(tree->mapping, index); | |
3334 | uptodate = PageUptodate(page); | |
3335 | page_cache_release(page); | |
3336 | if (!uptodate) { | |
3337 | pg_uptodate = 0; | |
3338 | break; | |
3339 | } | |
3340 | start += PAGE_CACHE_SIZE; | |
3341 | } | |
3342 | return pg_uptodate; | |
3343 | } | |
3344 | ||
d1310b2e | 3345 | int extent_buffer_uptodate(struct extent_io_tree *tree, |
2ac55d41 JB |
3346 | struct extent_buffer *eb, |
3347 | struct extent_state *cached_state) | |
d1310b2e | 3348 | { |
728131d8 | 3349 | int ret = 0; |
ce9adaa5 CM |
3350 | unsigned long num_pages; |
3351 | unsigned long i; | |
728131d8 CM |
3352 | struct page *page; |
3353 | int pg_uptodate = 1; | |
3354 | ||
b4ce94de | 3355 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
4235298e | 3356 | return 1; |
728131d8 | 3357 | |
4235298e | 3358 | ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1, |
2ac55d41 | 3359 | EXTENT_UPTODATE, 1, cached_state); |
4235298e CM |
3360 | if (ret) |
3361 | return ret; | |
728131d8 CM |
3362 | |
3363 | num_pages = num_extent_pages(eb->start, eb->len); | |
3364 | for (i = 0; i < num_pages; i++) { | |
3365 | page = extent_buffer_page(eb, i); | |
3366 | if (!PageUptodate(page)) { | |
3367 | pg_uptodate = 0; | |
3368 | break; | |
3369 | } | |
3370 | } | |
4235298e | 3371 | return pg_uptodate; |
d1310b2e | 3372 | } |
d1310b2e CM |
3373 | |
3374 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
3375 | struct extent_buffer *eb, | |
a86c12c7 | 3376 | u64 start, int wait, |
f188591e | 3377 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3378 | { |
3379 | unsigned long i; | |
3380 | unsigned long start_i; | |
3381 | struct page *page; | |
3382 | int err; | |
3383 | int ret = 0; | |
ce9adaa5 CM |
3384 | int locked_pages = 0; |
3385 | int all_uptodate = 1; | |
3386 | int inc_all_pages = 0; | |
d1310b2e | 3387 | unsigned long num_pages; |
a86c12c7 | 3388 | struct bio *bio = NULL; |
c8b97818 | 3389 | unsigned long bio_flags = 0; |
a86c12c7 | 3390 | |
b4ce94de | 3391 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
3392 | return 0; |
3393 | ||
ce9adaa5 | 3394 | if (test_range_bit(tree, eb->start, eb->start + eb->len - 1, |
9655d298 | 3395 | EXTENT_UPTODATE, 1, NULL)) { |
d1310b2e CM |
3396 | return 0; |
3397 | } | |
3398 | ||
3399 | if (start) { | |
3400 | WARN_ON(start < eb->start); | |
3401 | start_i = (start >> PAGE_CACHE_SHIFT) - | |
3402 | (eb->start >> PAGE_CACHE_SHIFT); | |
3403 | } else { | |
3404 | start_i = 0; | |
3405 | } | |
3406 | ||
3407 | num_pages = num_extent_pages(eb->start, eb->len); | |
3408 | for (i = start_i; i < num_pages; i++) { | |
3409 | page = extent_buffer_page(eb, i); | |
d1310b2e | 3410 | if (!wait) { |
2db04966 | 3411 | if (!trylock_page(page)) |
ce9adaa5 | 3412 | goto unlock_exit; |
d1310b2e CM |
3413 | } else { |
3414 | lock_page(page); | |
3415 | } | |
ce9adaa5 | 3416 | locked_pages++; |
d397712b | 3417 | if (!PageUptodate(page)) |
ce9adaa5 | 3418 | all_uptodate = 0; |
ce9adaa5 CM |
3419 | } |
3420 | if (all_uptodate) { | |
3421 | if (start_i == 0) | |
b4ce94de | 3422 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
3423 | goto unlock_exit; |
3424 | } | |
3425 | ||
3426 | for (i = start_i; i < num_pages; i++) { | |
3427 | page = extent_buffer_page(eb, i); | |
3428 | if (inc_all_pages) | |
3429 | page_cache_get(page); | |
3430 | if (!PageUptodate(page)) { | |
3431 | if (start_i == 0) | |
3432 | inc_all_pages = 1; | |
f188591e | 3433 | ClearPageError(page); |
a86c12c7 | 3434 | err = __extent_read_full_page(tree, page, |
f188591e | 3435 | get_extent, &bio, |
c8b97818 | 3436 | mirror_num, &bio_flags); |
d397712b | 3437 | if (err) |
d1310b2e | 3438 | ret = err; |
d1310b2e CM |
3439 | } else { |
3440 | unlock_page(page); | |
3441 | } | |
3442 | } | |
3443 | ||
a86c12c7 | 3444 | if (bio) |
c8b97818 | 3445 | submit_one_bio(READ, bio, mirror_num, bio_flags); |
a86c12c7 | 3446 | |
d397712b | 3447 | if (ret || !wait) |
d1310b2e | 3448 | return ret; |
d397712b | 3449 | |
d1310b2e CM |
3450 | for (i = start_i; i < num_pages; i++) { |
3451 | page = extent_buffer_page(eb, i); | |
3452 | wait_on_page_locked(page); | |
d397712b | 3453 | if (!PageUptodate(page)) |
d1310b2e | 3454 | ret = -EIO; |
d1310b2e | 3455 | } |
d397712b | 3456 | |
d1310b2e | 3457 | if (!ret) |
b4ce94de | 3458 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 3459 | return ret; |
ce9adaa5 CM |
3460 | |
3461 | unlock_exit: | |
3462 | i = start_i; | |
d397712b | 3463 | while (locked_pages > 0) { |
ce9adaa5 CM |
3464 | page = extent_buffer_page(eb, i); |
3465 | i++; | |
3466 | unlock_page(page); | |
3467 | locked_pages--; | |
3468 | } | |
3469 | return ret; | |
d1310b2e | 3470 | } |
d1310b2e CM |
3471 | |
3472 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
3473 | unsigned long start, | |
3474 | unsigned long len) | |
3475 | { | |
3476 | size_t cur; | |
3477 | size_t offset; | |
3478 | struct page *page; | |
3479 | char *kaddr; | |
3480 | char *dst = (char *)dstv; | |
3481 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3482 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
3483 | |
3484 | WARN_ON(start > eb->len); | |
3485 | WARN_ON(start + len > eb->start + eb->len); | |
3486 | ||
3487 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3488 | ||
d397712b | 3489 | while (len > 0) { |
d1310b2e | 3490 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
3491 | |
3492 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
3493 | kaddr = kmap_atomic(page, KM_USER1); | |
3494 | memcpy(dst, kaddr + offset, cur); | |
3495 | kunmap_atomic(kaddr, KM_USER1); | |
3496 | ||
3497 | dst += cur; | |
3498 | len -= cur; | |
3499 | offset = 0; | |
3500 | i++; | |
3501 | } | |
3502 | } | |
d1310b2e CM |
3503 | |
3504 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
3505 | unsigned long min_len, char **token, char **map, | |
3506 | unsigned long *map_start, | |
3507 | unsigned long *map_len, int km) | |
3508 | { | |
3509 | size_t offset = start & (PAGE_CACHE_SIZE - 1); | |
3510 | char *kaddr; | |
3511 | struct page *p; | |
3512 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3513 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3514 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
3515 | PAGE_CACHE_SHIFT; | |
3516 | ||
3517 | if (i != end_i) | |
3518 | return -EINVAL; | |
3519 | ||
3520 | if (i == 0) { | |
3521 | offset = start_offset; | |
3522 | *map_start = 0; | |
3523 | } else { | |
3524 | offset = 0; | |
3525 | *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; | |
3526 | } | |
d397712b | 3527 | |
d1310b2e | 3528 | if (start + min_len > eb->len) { |
d397712b CM |
3529 | printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, " |
3530 | "wanted %lu %lu\n", (unsigned long long)eb->start, | |
3531 | eb->len, start, min_len); | |
d1310b2e CM |
3532 | WARN_ON(1); |
3533 | } | |
3534 | ||
3535 | p = extent_buffer_page(eb, i); | |
d1310b2e CM |
3536 | kaddr = kmap_atomic(p, km); |
3537 | *token = kaddr; | |
3538 | *map = kaddr + offset; | |
3539 | *map_len = PAGE_CACHE_SIZE - offset; | |
3540 | return 0; | |
3541 | } | |
d1310b2e CM |
3542 | |
3543 | int map_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
3544 | unsigned long min_len, | |
3545 | char **token, char **map, | |
3546 | unsigned long *map_start, | |
3547 | unsigned long *map_len, int km) | |
3548 | { | |
3549 | int err; | |
3550 | int save = 0; | |
3551 | if (eb->map_token) { | |
3552 | unmap_extent_buffer(eb, eb->map_token, km); | |
3553 | eb->map_token = NULL; | |
3554 | save = 1; | |
3555 | } | |
3556 | err = map_private_extent_buffer(eb, start, min_len, token, map, | |
3557 | map_start, map_len, km); | |
3558 | if (!err && save) { | |
3559 | eb->map_token = *token; | |
3560 | eb->kaddr = *map; | |
3561 | eb->map_start = *map_start; | |
3562 | eb->map_len = *map_len; | |
3563 | } | |
3564 | return err; | |
3565 | } | |
d1310b2e CM |
3566 | |
3567 | void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km) | |
3568 | { | |
3569 | kunmap_atomic(token, km); | |
3570 | } | |
d1310b2e CM |
3571 | |
3572 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, | |
3573 | unsigned long start, | |
3574 | unsigned long len) | |
3575 | { | |
3576 | size_t cur; | |
3577 | size_t offset; | |
3578 | struct page *page; | |
3579 | char *kaddr; | |
3580 | char *ptr = (char *)ptrv; | |
3581 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3582 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3583 | int ret = 0; | |
3584 | ||
3585 | WARN_ON(start > eb->len); | |
3586 | WARN_ON(start + len > eb->start + eb->len); | |
3587 | ||
3588 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3589 | ||
d397712b | 3590 | while (len > 0) { |
d1310b2e | 3591 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
3592 | |
3593 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
3594 | ||
3595 | kaddr = kmap_atomic(page, KM_USER0); | |
3596 | ret = memcmp(ptr, kaddr + offset, cur); | |
3597 | kunmap_atomic(kaddr, KM_USER0); | |
3598 | if (ret) | |
3599 | break; | |
3600 | ||
3601 | ptr += cur; | |
3602 | len -= cur; | |
3603 | offset = 0; | |
3604 | i++; | |
3605 | } | |
3606 | return ret; | |
3607 | } | |
d1310b2e CM |
3608 | |
3609 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
3610 | unsigned long start, unsigned long len) | |
3611 | { | |
3612 | size_t cur; | |
3613 | size_t offset; | |
3614 | struct page *page; | |
3615 | char *kaddr; | |
3616 | char *src = (char *)srcv; | |
3617 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3618 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3619 | ||
3620 | WARN_ON(start > eb->len); | |
3621 | WARN_ON(start + len > eb->start + eb->len); | |
3622 | ||
3623 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3624 | ||
d397712b | 3625 | while (len > 0) { |
d1310b2e CM |
3626 | page = extent_buffer_page(eb, i); |
3627 | WARN_ON(!PageUptodate(page)); | |
3628 | ||
3629 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
3630 | kaddr = kmap_atomic(page, KM_USER1); | |
3631 | memcpy(kaddr + offset, src, cur); | |
3632 | kunmap_atomic(kaddr, KM_USER1); | |
3633 | ||
3634 | src += cur; | |
3635 | len -= cur; | |
3636 | offset = 0; | |
3637 | i++; | |
3638 | } | |
3639 | } | |
d1310b2e CM |
3640 | |
3641 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
3642 | unsigned long start, unsigned long len) | |
3643 | { | |
3644 | size_t cur; | |
3645 | size_t offset; | |
3646 | struct page *page; | |
3647 | char *kaddr; | |
3648 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3649 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3650 | ||
3651 | WARN_ON(start > eb->len); | |
3652 | WARN_ON(start + len > eb->start + eb->len); | |
3653 | ||
3654 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3655 | ||
d397712b | 3656 | while (len > 0) { |
d1310b2e CM |
3657 | page = extent_buffer_page(eb, i); |
3658 | WARN_ON(!PageUptodate(page)); | |
3659 | ||
3660 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
3661 | kaddr = kmap_atomic(page, KM_USER0); | |
3662 | memset(kaddr + offset, c, cur); | |
3663 | kunmap_atomic(kaddr, KM_USER0); | |
3664 | ||
3665 | len -= cur; | |
3666 | offset = 0; | |
3667 | i++; | |
3668 | } | |
3669 | } | |
d1310b2e CM |
3670 | |
3671 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
3672 | unsigned long dst_offset, unsigned long src_offset, | |
3673 | unsigned long len) | |
3674 | { | |
3675 | u64 dst_len = dst->len; | |
3676 | size_t cur; | |
3677 | size_t offset; | |
3678 | struct page *page; | |
3679 | char *kaddr; | |
3680 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3681 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
3682 | ||
3683 | WARN_ON(src->len != dst_len); | |
3684 | ||
3685 | offset = (start_offset + dst_offset) & | |
3686 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3687 | ||
d397712b | 3688 | while (len > 0) { |
d1310b2e CM |
3689 | page = extent_buffer_page(dst, i); |
3690 | WARN_ON(!PageUptodate(page)); | |
3691 | ||
3692 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
3693 | ||
3694 | kaddr = kmap_atomic(page, KM_USER0); | |
3695 | read_extent_buffer(src, kaddr + offset, src_offset, cur); | |
3696 | kunmap_atomic(kaddr, KM_USER0); | |
3697 | ||
3698 | src_offset += cur; | |
3699 | len -= cur; | |
3700 | offset = 0; | |
3701 | i++; | |
3702 | } | |
3703 | } | |
d1310b2e CM |
3704 | |
3705 | static void move_pages(struct page *dst_page, struct page *src_page, | |
3706 | unsigned long dst_off, unsigned long src_off, | |
3707 | unsigned long len) | |
3708 | { | |
3709 | char *dst_kaddr = kmap_atomic(dst_page, KM_USER0); | |
3710 | if (dst_page == src_page) { | |
3711 | memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len); | |
3712 | } else { | |
3713 | char *src_kaddr = kmap_atomic(src_page, KM_USER1); | |
3714 | char *p = dst_kaddr + dst_off + len; | |
3715 | char *s = src_kaddr + src_off + len; | |
3716 | ||
3717 | while (len--) | |
3718 | *--p = *--s; | |
3719 | ||
3720 | kunmap_atomic(src_kaddr, KM_USER1); | |
3721 | } | |
3722 | kunmap_atomic(dst_kaddr, KM_USER0); | |
3723 | } | |
3724 | ||
3725 | static void copy_pages(struct page *dst_page, struct page *src_page, | |
3726 | unsigned long dst_off, unsigned long src_off, | |
3727 | unsigned long len) | |
3728 | { | |
3729 | char *dst_kaddr = kmap_atomic(dst_page, KM_USER0); | |
3730 | char *src_kaddr; | |
3731 | ||
3732 | if (dst_page != src_page) | |
3733 | src_kaddr = kmap_atomic(src_page, KM_USER1); | |
3734 | else | |
3735 | src_kaddr = dst_kaddr; | |
3736 | ||
3737 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
3738 | kunmap_atomic(dst_kaddr, KM_USER0); | |
3739 | if (dst_page != src_page) | |
3740 | kunmap_atomic(src_kaddr, KM_USER1); | |
3741 | } | |
3742 | ||
3743 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
3744 | unsigned long src_offset, unsigned long len) | |
3745 | { | |
3746 | size_t cur; | |
3747 | size_t dst_off_in_page; | |
3748 | size_t src_off_in_page; | |
3749 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3750 | unsigned long dst_i; | |
3751 | unsigned long src_i; | |
3752 | ||
3753 | if (src_offset + len > dst->len) { | |
d397712b CM |
3754 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
3755 | "len %lu dst len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
3756 | BUG_ON(1); |
3757 | } | |
3758 | if (dst_offset + len > dst->len) { | |
d397712b CM |
3759 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
3760 | "len %lu dst len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
3761 | BUG_ON(1); |
3762 | } | |
3763 | ||
d397712b | 3764 | while (len > 0) { |
d1310b2e CM |
3765 | dst_off_in_page = (start_offset + dst_offset) & |
3766 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3767 | src_off_in_page = (start_offset + src_offset) & | |
3768 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3769 | ||
3770 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
3771 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
3772 | ||
3773 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
3774 | src_off_in_page)); | |
3775 | cur = min_t(unsigned long, cur, | |
3776 | (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); | |
3777 | ||
3778 | copy_pages(extent_buffer_page(dst, dst_i), | |
3779 | extent_buffer_page(dst, src_i), | |
3780 | dst_off_in_page, src_off_in_page, cur); | |
3781 | ||
3782 | src_offset += cur; | |
3783 | dst_offset += cur; | |
3784 | len -= cur; | |
3785 | } | |
3786 | } | |
d1310b2e CM |
3787 | |
3788 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
3789 | unsigned long src_offset, unsigned long len) | |
3790 | { | |
3791 | size_t cur; | |
3792 | size_t dst_off_in_page; | |
3793 | size_t src_off_in_page; | |
3794 | unsigned long dst_end = dst_offset + len - 1; | |
3795 | unsigned long src_end = src_offset + len - 1; | |
3796 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3797 | unsigned long dst_i; | |
3798 | unsigned long src_i; | |
3799 | ||
3800 | if (src_offset + len > dst->len) { | |
d397712b CM |
3801 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
3802 | "len %lu len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
3803 | BUG_ON(1); |
3804 | } | |
3805 | if (dst_offset + len > dst->len) { | |
d397712b CM |
3806 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
3807 | "len %lu len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
3808 | BUG_ON(1); |
3809 | } | |
3810 | if (dst_offset < src_offset) { | |
3811 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); | |
3812 | return; | |
3813 | } | |
d397712b | 3814 | while (len > 0) { |
d1310b2e CM |
3815 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; |
3816 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
3817 | ||
3818 | dst_off_in_page = (start_offset + dst_end) & | |
3819 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3820 | src_off_in_page = (start_offset + src_end) & | |
3821 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3822 | ||
3823 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
3824 | cur = min(cur, dst_off_in_page + 1); | |
3825 | move_pages(extent_buffer_page(dst, dst_i), | |
3826 | extent_buffer_page(dst, src_i), | |
3827 | dst_off_in_page - cur + 1, | |
3828 | src_off_in_page - cur + 1, cur); | |
3829 | ||
3830 | dst_end -= cur; | |
3831 | src_end -= cur; | |
3832 | len -= cur; | |
3833 | } | |
3834 | } | |
6af118ce CM |
3835 | |
3836 | int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page) | |
3837 | { | |
3838 | u64 start = page_offset(page); | |
3839 | struct extent_buffer *eb; | |
3840 | int ret = 1; | |
3841 | unsigned long i; | |
3842 | unsigned long num_pages; | |
3843 | ||
3844 | spin_lock(&tree->buffer_lock); | |
3845 | eb = buffer_search(tree, start); | |
3846 | if (!eb) | |
3847 | goto out; | |
3848 | ||
3849 | if (atomic_read(&eb->refs) > 1) { | |
3850 | ret = 0; | |
3851 | goto out; | |
3852 | } | |
b9473439 CM |
3853 | if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
3854 | ret = 0; | |
3855 | goto out; | |
3856 | } | |
6af118ce CM |
3857 | /* at this point we can safely release the extent buffer */ |
3858 | num_pages = num_extent_pages(eb->start, eb->len); | |
b214107e CH |
3859 | for (i = 0; i < num_pages; i++) |
3860 | page_cache_release(extent_buffer_page(eb, i)); | |
6af118ce CM |
3861 | rb_erase(&eb->rb_node, &tree->buffer); |
3862 | __free_extent_buffer(eb); | |
3863 | out: | |
3864 | spin_unlock(&tree->buffer_lock); | |
3865 | return ret; | |
3866 | } |