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