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> | |
268bb0ce | 13 | #include <linux/prefetch.h> |
90a887c9 | 14 | #include <linux/cleancache.h> |
d1310b2e CM |
15 | #include "extent_io.h" |
16 | #include "extent_map.h" | |
2db04966 | 17 | #include "compat.h" |
902b22f3 DW |
18 | #include "ctree.h" |
19 | #include "btrfs_inode.h" | |
4a54c8c1 | 20 | #include "volumes.h" |
21adbd5c | 21 | #include "check-integrity.h" |
d1310b2e | 22 | |
d1310b2e CM |
23 | static struct kmem_cache *extent_state_cache; |
24 | static struct kmem_cache *extent_buffer_cache; | |
25 | ||
26 | static LIST_HEAD(buffers); | |
27 | static LIST_HEAD(states); | |
4bef0848 | 28 | |
b47eda86 | 29 | #define LEAK_DEBUG 0 |
3935127c | 30 | #if LEAK_DEBUG |
d397712b | 31 | static DEFINE_SPINLOCK(leak_lock); |
4bef0848 | 32 | #endif |
d1310b2e | 33 | |
d1310b2e CM |
34 | #define BUFFER_LRU_MAX 64 |
35 | ||
36 | struct tree_entry { | |
37 | u64 start; | |
38 | u64 end; | |
d1310b2e CM |
39 | struct rb_node rb_node; |
40 | }; | |
41 | ||
42 | struct extent_page_data { | |
43 | struct bio *bio; | |
44 | struct extent_io_tree *tree; | |
45 | get_extent_t *get_extent; | |
771ed689 CM |
46 | |
47 | /* tells writepage not to lock the state bits for this range | |
48 | * it still does the unlocking | |
49 | */ | |
ffbd517d CM |
50 | unsigned int extent_locked:1; |
51 | ||
52 | /* tells the submit_bio code to use a WRITE_SYNC */ | |
53 | unsigned int sync_io:1; | |
d1310b2e CM |
54 | }; |
55 | ||
c2d904e0 JM |
56 | static inline struct btrfs_fs_info * |
57 | tree_fs_info(struct extent_io_tree *tree) | |
58 | { | |
59 | return btrfs_sb(tree->mapping->host->i_sb); | |
60 | } | |
61 | ||
d1310b2e CM |
62 | int __init extent_io_init(void) |
63 | { | |
9601e3f6 CH |
64 | extent_state_cache = kmem_cache_create("extent_state", |
65 | sizeof(struct extent_state), 0, | |
66 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
67 | if (!extent_state_cache) |
68 | return -ENOMEM; | |
69 | ||
9601e3f6 CH |
70 | extent_buffer_cache = kmem_cache_create("extent_buffers", |
71 | sizeof(struct extent_buffer), 0, | |
72 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
73 | if (!extent_buffer_cache) |
74 | goto free_state_cache; | |
75 | return 0; | |
76 | ||
77 | free_state_cache: | |
78 | kmem_cache_destroy(extent_state_cache); | |
79 | return -ENOMEM; | |
80 | } | |
81 | ||
82 | void extent_io_exit(void) | |
83 | { | |
84 | struct extent_state *state; | |
2d2ae547 | 85 | struct extent_buffer *eb; |
d1310b2e CM |
86 | |
87 | while (!list_empty(&states)) { | |
2d2ae547 | 88 | state = list_entry(states.next, struct extent_state, leak_list); |
d397712b CM |
89 | printk(KERN_ERR "btrfs state leak: start %llu end %llu " |
90 | "state %lu in tree %p refs %d\n", | |
91 | (unsigned long long)state->start, | |
92 | (unsigned long long)state->end, | |
93 | state->state, state->tree, atomic_read(&state->refs)); | |
2d2ae547 | 94 | list_del(&state->leak_list); |
d1310b2e CM |
95 | kmem_cache_free(extent_state_cache, state); |
96 | ||
97 | } | |
98 | ||
2d2ae547 CM |
99 | while (!list_empty(&buffers)) { |
100 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
d397712b CM |
101 | printk(KERN_ERR "btrfs buffer leak start %llu len %lu " |
102 | "refs %d\n", (unsigned long long)eb->start, | |
103 | eb->len, atomic_read(&eb->refs)); | |
2d2ae547 CM |
104 | list_del(&eb->leak_list); |
105 | kmem_cache_free(extent_buffer_cache, eb); | |
106 | } | |
d1310b2e CM |
107 | if (extent_state_cache) |
108 | kmem_cache_destroy(extent_state_cache); | |
109 | if (extent_buffer_cache) | |
110 | kmem_cache_destroy(extent_buffer_cache); | |
111 | } | |
112 | ||
113 | void extent_io_tree_init(struct extent_io_tree *tree, | |
f993c883 | 114 | struct address_space *mapping) |
d1310b2e | 115 | { |
6bef4d31 | 116 | tree->state = RB_ROOT; |
19fe0a8b | 117 | INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC); |
d1310b2e CM |
118 | tree->ops = NULL; |
119 | tree->dirty_bytes = 0; | |
70dec807 | 120 | spin_lock_init(&tree->lock); |
6af118ce | 121 | spin_lock_init(&tree->buffer_lock); |
d1310b2e | 122 | tree->mapping = mapping; |
d1310b2e | 123 | } |
d1310b2e | 124 | |
b2950863 | 125 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
126 | { |
127 | struct extent_state *state; | |
3935127c | 128 | #if LEAK_DEBUG |
2d2ae547 | 129 | unsigned long flags; |
4bef0848 | 130 | #endif |
d1310b2e CM |
131 | |
132 | state = kmem_cache_alloc(extent_state_cache, mask); | |
2b114d1d | 133 | if (!state) |
d1310b2e CM |
134 | return state; |
135 | state->state = 0; | |
d1310b2e | 136 | state->private = 0; |
70dec807 | 137 | state->tree = NULL; |
3935127c | 138 | #if LEAK_DEBUG |
2d2ae547 CM |
139 | spin_lock_irqsave(&leak_lock, flags); |
140 | list_add(&state->leak_list, &states); | |
141 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 142 | #endif |
d1310b2e CM |
143 | atomic_set(&state->refs, 1); |
144 | init_waitqueue_head(&state->wq); | |
145 | return state; | |
146 | } | |
d1310b2e | 147 | |
4845e44f | 148 | void free_extent_state(struct extent_state *state) |
d1310b2e | 149 | { |
d1310b2e CM |
150 | if (!state) |
151 | return; | |
152 | if (atomic_dec_and_test(&state->refs)) { | |
3935127c | 153 | #if LEAK_DEBUG |
2d2ae547 | 154 | unsigned long flags; |
4bef0848 | 155 | #endif |
70dec807 | 156 | WARN_ON(state->tree); |
3935127c | 157 | #if LEAK_DEBUG |
2d2ae547 CM |
158 | spin_lock_irqsave(&leak_lock, flags); |
159 | list_del(&state->leak_list); | |
160 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 161 | #endif |
d1310b2e CM |
162 | kmem_cache_free(extent_state_cache, state); |
163 | } | |
164 | } | |
d1310b2e CM |
165 | |
166 | static struct rb_node *tree_insert(struct rb_root *root, u64 offset, | |
167 | struct rb_node *node) | |
168 | { | |
d397712b CM |
169 | struct rb_node **p = &root->rb_node; |
170 | struct rb_node *parent = NULL; | |
d1310b2e CM |
171 | struct tree_entry *entry; |
172 | ||
d397712b | 173 | while (*p) { |
d1310b2e CM |
174 | parent = *p; |
175 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
176 | ||
177 | if (offset < entry->start) | |
178 | p = &(*p)->rb_left; | |
179 | else if (offset > entry->end) | |
180 | p = &(*p)->rb_right; | |
181 | else | |
182 | return parent; | |
183 | } | |
184 | ||
185 | entry = rb_entry(node, struct tree_entry, rb_node); | |
d1310b2e CM |
186 | rb_link_node(node, parent, p); |
187 | rb_insert_color(node, root); | |
188 | return NULL; | |
189 | } | |
190 | ||
80ea96b1 | 191 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
d1310b2e CM |
192 | struct rb_node **prev_ret, |
193 | struct rb_node **next_ret) | |
194 | { | |
80ea96b1 | 195 | struct rb_root *root = &tree->state; |
d397712b | 196 | struct rb_node *n = root->rb_node; |
d1310b2e CM |
197 | struct rb_node *prev = NULL; |
198 | struct rb_node *orig_prev = NULL; | |
199 | struct tree_entry *entry; | |
200 | struct tree_entry *prev_entry = NULL; | |
201 | ||
d397712b | 202 | while (n) { |
d1310b2e CM |
203 | entry = rb_entry(n, struct tree_entry, rb_node); |
204 | prev = n; | |
205 | prev_entry = entry; | |
206 | ||
207 | if (offset < entry->start) | |
208 | n = n->rb_left; | |
209 | else if (offset > entry->end) | |
210 | n = n->rb_right; | |
d397712b | 211 | else |
d1310b2e CM |
212 | return n; |
213 | } | |
214 | ||
215 | if (prev_ret) { | |
216 | orig_prev = prev; | |
d397712b | 217 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
218 | prev = rb_next(prev); |
219 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
220 | } | |
221 | *prev_ret = prev; | |
222 | prev = orig_prev; | |
223 | } | |
224 | ||
225 | if (next_ret) { | |
226 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 227 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
228 | prev = rb_prev(prev); |
229 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
230 | } | |
231 | *next_ret = prev; | |
232 | } | |
233 | return NULL; | |
234 | } | |
235 | ||
80ea96b1 CM |
236 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
237 | u64 offset) | |
d1310b2e | 238 | { |
70dec807 | 239 | struct rb_node *prev = NULL; |
d1310b2e | 240 | struct rb_node *ret; |
70dec807 | 241 | |
80ea96b1 | 242 | ret = __etree_search(tree, offset, &prev, NULL); |
d397712b | 243 | if (!ret) |
d1310b2e CM |
244 | return prev; |
245 | return ret; | |
246 | } | |
247 | ||
9ed74f2d JB |
248 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, |
249 | struct extent_state *other) | |
250 | { | |
251 | if (tree->ops && tree->ops->merge_extent_hook) | |
252 | tree->ops->merge_extent_hook(tree->mapping->host, new, | |
253 | other); | |
254 | } | |
255 | ||
d1310b2e CM |
256 | /* |
257 | * utility function to look for merge candidates inside a given range. | |
258 | * Any extents with matching state are merged together into a single | |
259 | * extent in the tree. Extents with EXTENT_IO in their state field | |
260 | * are not merged because the end_io handlers need to be able to do | |
261 | * operations on them without sleeping (or doing allocations/splits). | |
262 | * | |
263 | * This should be called with the tree lock held. | |
264 | */ | |
1bf85046 JM |
265 | static void merge_state(struct extent_io_tree *tree, |
266 | struct extent_state *state) | |
d1310b2e CM |
267 | { |
268 | struct extent_state *other; | |
269 | struct rb_node *other_node; | |
270 | ||
5b21f2ed | 271 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
1bf85046 | 272 | return; |
d1310b2e CM |
273 | |
274 | other_node = rb_prev(&state->rb_node); | |
275 | if (other_node) { | |
276 | other = rb_entry(other_node, struct extent_state, rb_node); | |
277 | if (other->end == state->start - 1 && | |
278 | other->state == state->state) { | |
9ed74f2d | 279 | merge_cb(tree, state, other); |
d1310b2e | 280 | state->start = other->start; |
70dec807 | 281 | other->tree = NULL; |
d1310b2e CM |
282 | rb_erase(&other->rb_node, &tree->state); |
283 | free_extent_state(other); | |
284 | } | |
285 | } | |
286 | other_node = rb_next(&state->rb_node); | |
287 | if (other_node) { | |
288 | other = rb_entry(other_node, struct extent_state, rb_node); | |
289 | if (other->start == state->end + 1 && | |
290 | other->state == state->state) { | |
9ed74f2d | 291 | merge_cb(tree, state, other); |
df98b6e2 JB |
292 | state->end = other->end; |
293 | other->tree = NULL; | |
294 | rb_erase(&other->rb_node, &tree->state); | |
295 | free_extent_state(other); | |
d1310b2e CM |
296 | } |
297 | } | |
d1310b2e CM |
298 | } |
299 | ||
1bf85046 | 300 | static void set_state_cb(struct extent_io_tree *tree, |
0ca1f7ce | 301 | struct extent_state *state, int *bits) |
291d673e | 302 | { |
1bf85046 JM |
303 | if (tree->ops && tree->ops->set_bit_hook) |
304 | tree->ops->set_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
305 | } |
306 | ||
307 | static void clear_state_cb(struct extent_io_tree *tree, | |
0ca1f7ce | 308 | struct extent_state *state, int *bits) |
291d673e | 309 | { |
9ed74f2d JB |
310 | if (tree->ops && tree->ops->clear_bit_hook) |
311 | tree->ops->clear_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
312 | } |
313 | ||
3150b699 XG |
314 | static void set_state_bits(struct extent_io_tree *tree, |
315 | struct extent_state *state, int *bits); | |
316 | ||
d1310b2e CM |
317 | /* |
318 | * insert an extent_state struct into the tree. 'bits' are set on the | |
319 | * struct before it is inserted. | |
320 | * | |
321 | * This may return -EEXIST if the extent is already there, in which case the | |
322 | * state struct is freed. | |
323 | * | |
324 | * The tree lock is not taken internally. This is a utility function and | |
325 | * probably isn't what you want to call (see set/clear_extent_bit). | |
326 | */ | |
327 | static int insert_state(struct extent_io_tree *tree, | |
328 | struct extent_state *state, u64 start, u64 end, | |
0ca1f7ce | 329 | int *bits) |
d1310b2e CM |
330 | { |
331 | struct rb_node *node; | |
332 | ||
333 | if (end < start) { | |
d397712b CM |
334 | printk(KERN_ERR "btrfs end < start %llu %llu\n", |
335 | (unsigned long long)end, | |
336 | (unsigned long long)start); | |
d1310b2e CM |
337 | WARN_ON(1); |
338 | } | |
d1310b2e CM |
339 | state->start = start; |
340 | state->end = end; | |
9ed74f2d | 341 | |
3150b699 XG |
342 | set_state_bits(tree, state, bits); |
343 | ||
d1310b2e CM |
344 | node = tree_insert(&tree->state, end, &state->rb_node); |
345 | if (node) { | |
346 | struct extent_state *found; | |
347 | found = rb_entry(node, struct extent_state, rb_node); | |
d397712b CM |
348 | printk(KERN_ERR "btrfs found node %llu %llu on insert of " |
349 | "%llu %llu\n", (unsigned long long)found->start, | |
350 | (unsigned long long)found->end, | |
351 | (unsigned long long)start, (unsigned long long)end); | |
d1310b2e CM |
352 | return -EEXIST; |
353 | } | |
70dec807 | 354 | state->tree = tree; |
d1310b2e CM |
355 | merge_state(tree, state); |
356 | return 0; | |
357 | } | |
358 | ||
1bf85046 | 359 | static void split_cb(struct extent_io_tree *tree, struct extent_state *orig, |
9ed74f2d JB |
360 | u64 split) |
361 | { | |
362 | if (tree->ops && tree->ops->split_extent_hook) | |
1bf85046 | 363 | tree->ops->split_extent_hook(tree->mapping->host, orig, split); |
9ed74f2d JB |
364 | } |
365 | ||
d1310b2e CM |
366 | /* |
367 | * split a given extent state struct in two, inserting the preallocated | |
368 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
369 | * offset inside 'orig' where it should be split. | |
370 | * | |
371 | * Before calling, | |
372 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
373 | * are two extent state structs in the tree: | |
374 | * prealloc: [orig->start, split - 1] | |
375 | * orig: [ split, orig->end ] | |
376 | * | |
377 | * The tree locks are not taken by this function. They need to be held | |
378 | * by the caller. | |
379 | */ | |
380 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
381 | struct extent_state *prealloc, u64 split) | |
382 | { | |
383 | struct rb_node *node; | |
9ed74f2d JB |
384 | |
385 | split_cb(tree, orig, split); | |
386 | ||
d1310b2e CM |
387 | prealloc->start = orig->start; |
388 | prealloc->end = split - 1; | |
389 | prealloc->state = orig->state; | |
390 | orig->start = split; | |
391 | ||
392 | node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node); | |
393 | if (node) { | |
d1310b2e CM |
394 | free_extent_state(prealloc); |
395 | return -EEXIST; | |
396 | } | |
70dec807 | 397 | prealloc->tree = tree; |
d1310b2e CM |
398 | return 0; |
399 | } | |
400 | ||
401 | /* | |
402 | * utility function to clear some bits in an extent state struct. | |
403 | * it will optionally wake up any one waiting on this state (wake == 1), or | |
404 | * forcibly remove the state from the tree (delete == 1). | |
405 | * | |
406 | * If no bits are set on the state struct after clearing things, the | |
407 | * struct is freed and removed from the tree | |
408 | */ | |
409 | static int clear_state_bit(struct extent_io_tree *tree, | |
0ca1f7ce YZ |
410 | struct extent_state *state, |
411 | int *bits, int wake) | |
d1310b2e | 412 | { |
0ca1f7ce | 413 | int bits_to_clear = *bits & ~EXTENT_CTLBITS; |
32c00aff | 414 | int ret = state->state & bits_to_clear; |
d1310b2e | 415 | |
0ca1f7ce | 416 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
417 | u64 range = state->end - state->start + 1; |
418 | WARN_ON(range > tree->dirty_bytes); | |
419 | tree->dirty_bytes -= range; | |
420 | } | |
291d673e | 421 | clear_state_cb(tree, state, bits); |
32c00aff | 422 | state->state &= ~bits_to_clear; |
d1310b2e CM |
423 | if (wake) |
424 | wake_up(&state->wq); | |
0ca1f7ce | 425 | if (state->state == 0) { |
70dec807 | 426 | if (state->tree) { |
d1310b2e | 427 | rb_erase(&state->rb_node, &tree->state); |
70dec807 | 428 | state->tree = NULL; |
d1310b2e CM |
429 | free_extent_state(state); |
430 | } else { | |
431 | WARN_ON(1); | |
432 | } | |
433 | } else { | |
434 | merge_state(tree, state); | |
435 | } | |
436 | return ret; | |
437 | } | |
438 | ||
8233767a XG |
439 | static struct extent_state * |
440 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
441 | { | |
442 | if (!prealloc) | |
443 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
444 | ||
445 | return prealloc; | |
446 | } | |
447 | ||
c2d904e0 JM |
448 | void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
449 | { | |
450 | btrfs_panic(tree_fs_info(tree), err, "Locking error: " | |
451 | "Extent tree was modified by another " | |
452 | "thread while locked."); | |
453 | } | |
454 | ||
d1310b2e CM |
455 | /* |
456 | * clear some bits on a range in the tree. This may require splitting | |
457 | * or inserting elements in the tree, so the gfp mask is used to | |
458 | * indicate which allocations or sleeping are allowed. | |
459 | * | |
460 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
461 | * the given range from the tree regardless of state (ie for truncate). | |
462 | * | |
463 | * the range [start, end] is inclusive. | |
464 | * | |
6763af84 | 465 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e CM |
466 | */ |
467 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
2c64c53d CM |
468 | int bits, int wake, int delete, |
469 | struct extent_state **cached_state, | |
470 | gfp_t mask) | |
d1310b2e CM |
471 | { |
472 | struct extent_state *state; | |
2c64c53d | 473 | struct extent_state *cached; |
d1310b2e | 474 | struct extent_state *prealloc = NULL; |
2c64c53d | 475 | struct rb_node *next_node; |
d1310b2e | 476 | struct rb_node *node; |
5c939df5 | 477 | u64 last_end; |
d1310b2e | 478 | int err; |
2ac55d41 | 479 | int clear = 0; |
d1310b2e | 480 | |
0ca1f7ce YZ |
481 | if (delete) |
482 | bits |= ~EXTENT_CTLBITS; | |
483 | bits |= EXTENT_FIRST_DELALLOC; | |
484 | ||
2ac55d41 JB |
485 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
486 | clear = 1; | |
d1310b2e CM |
487 | again: |
488 | if (!prealloc && (mask & __GFP_WAIT)) { | |
489 | prealloc = alloc_extent_state(mask); | |
490 | if (!prealloc) | |
491 | return -ENOMEM; | |
492 | } | |
493 | ||
cad321ad | 494 | spin_lock(&tree->lock); |
2c64c53d CM |
495 | if (cached_state) { |
496 | cached = *cached_state; | |
2ac55d41 JB |
497 | |
498 | if (clear) { | |
499 | *cached_state = NULL; | |
500 | cached_state = NULL; | |
501 | } | |
502 | ||
df98b6e2 JB |
503 | if (cached && cached->tree && cached->start <= start && |
504 | cached->end > start) { | |
2ac55d41 JB |
505 | if (clear) |
506 | atomic_dec(&cached->refs); | |
2c64c53d | 507 | state = cached; |
42daec29 | 508 | goto hit_next; |
2c64c53d | 509 | } |
2ac55d41 JB |
510 | if (clear) |
511 | free_extent_state(cached); | |
2c64c53d | 512 | } |
d1310b2e CM |
513 | /* |
514 | * this search will find the extents that end after | |
515 | * our range starts | |
516 | */ | |
80ea96b1 | 517 | node = tree_search(tree, start); |
d1310b2e CM |
518 | if (!node) |
519 | goto out; | |
520 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 521 | hit_next: |
d1310b2e CM |
522 | if (state->start > end) |
523 | goto out; | |
524 | WARN_ON(state->end < start); | |
5c939df5 | 525 | last_end = state->end; |
d1310b2e | 526 | |
0449314a LB |
527 | if (state->end < end && !need_resched()) |
528 | next_node = rb_next(&state->rb_node); | |
529 | else | |
530 | next_node = NULL; | |
531 | ||
532 | /* the state doesn't have the wanted bits, go ahead */ | |
533 | if (!(state->state & bits)) | |
534 | goto next; | |
535 | ||
d1310b2e CM |
536 | /* |
537 | * | ---- desired range ---- | | |
538 | * | state | or | |
539 | * | ------------- state -------------- | | |
540 | * | |
541 | * We need to split the extent we found, and may flip | |
542 | * bits on second half. | |
543 | * | |
544 | * If the extent we found extends past our range, we | |
545 | * just split and search again. It'll get split again | |
546 | * the next time though. | |
547 | * | |
548 | * If the extent we found is inside our range, we clear | |
549 | * the desired bit on it. | |
550 | */ | |
551 | ||
552 | if (state->start < start) { | |
8233767a XG |
553 | prealloc = alloc_extent_state_atomic(prealloc); |
554 | BUG_ON(!prealloc); | |
d1310b2e | 555 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
556 | if (err) |
557 | extent_io_tree_panic(tree, err); | |
558 | ||
d1310b2e CM |
559 | prealloc = NULL; |
560 | if (err) | |
561 | goto out; | |
562 | if (state->end <= end) { | |
6763af84 | 563 | clear_state_bit(tree, state, &bits, wake); |
5c939df5 YZ |
564 | if (last_end == (u64)-1) |
565 | goto out; | |
566 | start = last_end + 1; | |
d1310b2e CM |
567 | } |
568 | goto search_again; | |
569 | } | |
570 | /* | |
571 | * | ---- desired range ---- | | |
572 | * | state | | |
573 | * We need to split the extent, and clear the bit | |
574 | * on the first half | |
575 | */ | |
576 | if (state->start <= end && state->end > end) { | |
8233767a XG |
577 | prealloc = alloc_extent_state_atomic(prealloc); |
578 | BUG_ON(!prealloc); | |
d1310b2e | 579 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
580 | if (err) |
581 | extent_io_tree_panic(tree, err); | |
582 | ||
d1310b2e CM |
583 | if (wake) |
584 | wake_up(&state->wq); | |
42daec29 | 585 | |
6763af84 | 586 | clear_state_bit(tree, prealloc, &bits, wake); |
9ed74f2d | 587 | |
d1310b2e CM |
588 | prealloc = NULL; |
589 | goto out; | |
590 | } | |
42daec29 | 591 | |
6763af84 | 592 | clear_state_bit(tree, state, &bits, wake); |
0449314a | 593 | next: |
5c939df5 YZ |
594 | if (last_end == (u64)-1) |
595 | goto out; | |
596 | start = last_end + 1; | |
2c64c53d CM |
597 | if (start <= end && next_node) { |
598 | state = rb_entry(next_node, struct extent_state, | |
599 | rb_node); | |
692e5759 | 600 | goto hit_next; |
2c64c53d | 601 | } |
d1310b2e CM |
602 | goto search_again; |
603 | ||
604 | out: | |
cad321ad | 605 | spin_unlock(&tree->lock); |
d1310b2e CM |
606 | if (prealloc) |
607 | free_extent_state(prealloc); | |
608 | ||
6763af84 | 609 | return 0; |
d1310b2e CM |
610 | |
611 | search_again: | |
612 | if (start > end) | |
613 | goto out; | |
cad321ad | 614 | spin_unlock(&tree->lock); |
d1310b2e CM |
615 | if (mask & __GFP_WAIT) |
616 | cond_resched(); | |
617 | goto again; | |
618 | } | |
d1310b2e CM |
619 | |
620 | static int wait_on_state(struct extent_io_tree *tree, | |
621 | struct extent_state *state) | |
641f5219 CH |
622 | __releases(tree->lock) |
623 | __acquires(tree->lock) | |
d1310b2e CM |
624 | { |
625 | DEFINE_WAIT(wait); | |
626 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 627 | spin_unlock(&tree->lock); |
d1310b2e | 628 | schedule(); |
cad321ad | 629 | spin_lock(&tree->lock); |
d1310b2e CM |
630 | finish_wait(&state->wq, &wait); |
631 | return 0; | |
632 | } | |
633 | ||
634 | /* | |
635 | * waits for one or more bits to clear on a range in the state tree. | |
636 | * The range [start, end] is inclusive. | |
637 | * The tree lock is taken by this function | |
638 | */ | |
639 | int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits) | |
640 | { | |
641 | struct extent_state *state; | |
642 | struct rb_node *node; | |
643 | ||
cad321ad | 644 | spin_lock(&tree->lock); |
d1310b2e CM |
645 | again: |
646 | while (1) { | |
647 | /* | |
648 | * this search will find all the extents that end after | |
649 | * our range starts | |
650 | */ | |
80ea96b1 | 651 | node = tree_search(tree, start); |
d1310b2e CM |
652 | if (!node) |
653 | break; | |
654 | ||
655 | state = rb_entry(node, struct extent_state, rb_node); | |
656 | ||
657 | if (state->start > end) | |
658 | goto out; | |
659 | ||
660 | if (state->state & bits) { | |
661 | start = state->start; | |
662 | atomic_inc(&state->refs); | |
663 | wait_on_state(tree, state); | |
664 | free_extent_state(state); | |
665 | goto again; | |
666 | } | |
667 | start = state->end + 1; | |
668 | ||
669 | if (start > end) | |
670 | break; | |
671 | ||
ded91f08 | 672 | cond_resched_lock(&tree->lock); |
d1310b2e CM |
673 | } |
674 | out: | |
cad321ad | 675 | spin_unlock(&tree->lock); |
d1310b2e CM |
676 | return 0; |
677 | } | |
d1310b2e | 678 | |
1bf85046 | 679 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 680 | struct extent_state *state, |
0ca1f7ce | 681 | int *bits) |
d1310b2e | 682 | { |
0ca1f7ce | 683 | int bits_to_set = *bits & ~EXTENT_CTLBITS; |
9ed74f2d | 684 | |
1bf85046 | 685 | set_state_cb(tree, state, bits); |
0ca1f7ce | 686 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
687 | u64 range = state->end - state->start + 1; |
688 | tree->dirty_bytes += range; | |
689 | } | |
0ca1f7ce | 690 | state->state |= bits_to_set; |
d1310b2e CM |
691 | } |
692 | ||
2c64c53d CM |
693 | static void cache_state(struct extent_state *state, |
694 | struct extent_state **cached_ptr) | |
695 | { | |
696 | if (cached_ptr && !(*cached_ptr)) { | |
697 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) { | |
698 | *cached_ptr = state; | |
699 | atomic_inc(&state->refs); | |
700 | } | |
701 | } | |
702 | } | |
703 | ||
507903b8 AJ |
704 | static void uncache_state(struct extent_state **cached_ptr) |
705 | { | |
706 | if (cached_ptr && (*cached_ptr)) { | |
707 | struct extent_state *state = *cached_ptr; | |
109b36a2 CM |
708 | *cached_ptr = NULL; |
709 | free_extent_state(state); | |
507903b8 AJ |
710 | } |
711 | } | |
712 | ||
d1310b2e | 713 | /* |
1edbb734 CM |
714 | * set some bits on a range in the tree. This may require allocations or |
715 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 716 | * |
1edbb734 CM |
717 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
718 | * part of the range already has the desired bits set. The start of the | |
719 | * existing range is returned in failed_start in this case. | |
d1310b2e | 720 | * |
1edbb734 | 721 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 722 | */ |
1edbb734 | 723 | |
4845e44f CM |
724 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
725 | int bits, int exclusive_bits, u64 *failed_start, | |
726 | struct extent_state **cached_state, gfp_t mask) | |
d1310b2e CM |
727 | { |
728 | struct extent_state *state; | |
729 | struct extent_state *prealloc = NULL; | |
730 | struct rb_node *node; | |
d1310b2e | 731 | int err = 0; |
d1310b2e CM |
732 | u64 last_start; |
733 | u64 last_end; | |
42daec29 | 734 | |
0ca1f7ce | 735 | bits |= EXTENT_FIRST_DELALLOC; |
d1310b2e CM |
736 | again: |
737 | if (!prealloc && (mask & __GFP_WAIT)) { | |
738 | prealloc = alloc_extent_state(mask); | |
8233767a | 739 | BUG_ON(!prealloc); |
d1310b2e CM |
740 | } |
741 | ||
cad321ad | 742 | spin_lock(&tree->lock); |
9655d298 CM |
743 | if (cached_state && *cached_state) { |
744 | state = *cached_state; | |
df98b6e2 JB |
745 | if (state->start <= start && state->end > start && |
746 | state->tree) { | |
9655d298 CM |
747 | node = &state->rb_node; |
748 | goto hit_next; | |
749 | } | |
750 | } | |
d1310b2e CM |
751 | /* |
752 | * this search will find all the extents that end after | |
753 | * our range starts. | |
754 | */ | |
80ea96b1 | 755 | node = tree_search(tree, start); |
d1310b2e | 756 | if (!node) { |
8233767a XG |
757 | prealloc = alloc_extent_state_atomic(prealloc); |
758 | BUG_ON(!prealloc); | |
0ca1f7ce | 759 | err = insert_state(tree, prealloc, start, end, &bits); |
c2d904e0 JM |
760 | if (err) |
761 | extent_io_tree_panic(tree, err); | |
762 | ||
d1310b2e | 763 | prealloc = NULL; |
d1310b2e CM |
764 | goto out; |
765 | } | |
d1310b2e | 766 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 767 | hit_next: |
d1310b2e CM |
768 | last_start = state->start; |
769 | last_end = state->end; | |
770 | ||
771 | /* | |
772 | * | ---- desired range ---- | | |
773 | * | state | | |
774 | * | |
775 | * Just lock what we found and keep going | |
776 | */ | |
777 | if (state->start == start && state->end <= end) { | |
40431d6c | 778 | struct rb_node *next_node; |
1edbb734 | 779 | if (state->state & exclusive_bits) { |
d1310b2e CM |
780 | *failed_start = state->start; |
781 | err = -EEXIST; | |
782 | goto out; | |
783 | } | |
42daec29 | 784 | |
1bf85046 | 785 | set_state_bits(tree, state, &bits); |
9ed74f2d | 786 | |
2c64c53d | 787 | cache_state(state, cached_state); |
d1310b2e | 788 | merge_state(tree, state); |
5c939df5 YZ |
789 | if (last_end == (u64)-1) |
790 | goto out; | |
40431d6c | 791 | |
5c939df5 | 792 | start = last_end + 1; |
df98b6e2 | 793 | next_node = rb_next(&state->rb_node); |
c7f895a2 XG |
794 | if (next_node && start < end && prealloc && !need_resched()) { |
795 | state = rb_entry(next_node, struct extent_state, | |
796 | rb_node); | |
797 | if (state->start == start) | |
798 | goto hit_next; | |
40431d6c | 799 | } |
d1310b2e CM |
800 | goto search_again; |
801 | } | |
802 | ||
803 | /* | |
804 | * | ---- desired range ---- | | |
805 | * | state | | |
806 | * or | |
807 | * | ------------- state -------------- | | |
808 | * | |
809 | * We need to split the extent we found, and may flip bits on | |
810 | * second half. | |
811 | * | |
812 | * If the extent we found extends past our | |
813 | * range, we just split and search again. It'll get split | |
814 | * again the next time though. | |
815 | * | |
816 | * If the extent we found is inside our range, we set the | |
817 | * desired bit on it. | |
818 | */ | |
819 | if (state->start < start) { | |
1edbb734 | 820 | if (state->state & exclusive_bits) { |
d1310b2e CM |
821 | *failed_start = start; |
822 | err = -EEXIST; | |
823 | goto out; | |
824 | } | |
8233767a XG |
825 | |
826 | prealloc = alloc_extent_state_atomic(prealloc); | |
827 | BUG_ON(!prealloc); | |
d1310b2e | 828 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
829 | if (err) |
830 | extent_io_tree_panic(tree, err); | |
831 | ||
d1310b2e CM |
832 | prealloc = NULL; |
833 | if (err) | |
834 | goto out; | |
835 | if (state->end <= end) { | |
1bf85046 | 836 | set_state_bits(tree, state, &bits); |
2c64c53d | 837 | cache_state(state, cached_state); |
d1310b2e | 838 | merge_state(tree, state); |
5c939df5 YZ |
839 | if (last_end == (u64)-1) |
840 | goto out; | |
841 | start = last_end + 1; | |
d1310b2e CM |
842 | } |
843 | goto search_again; | |
844 | } | |
845 | /* | |
846 | * | ---- desired range ---- | | |
847 | * | state | or | state | | |
848 | * | |
849 | * There's a hole, we need to insert something in it and | |
850 | * ignore the extent we found. | |
851 | */ | |
852 | if (state->start > start) { | |
853 | u64 this_end; | |
854 | if (end < last_start) | |
855 | this_end = end; | |
856 | else | |
d397712b | 857 | this_end = last_start - 1; |
8233767a XG |
858 | |
859 | prealloc = alloc_extent_state_atomic(prealloc); | |
860 | BUG_ON(!prealloc); | |
c7f895a2 XG |
861 | |
862 | /* | |
863 | * Avoid to free 'prealloc' if it can be merged with | |
864 | * the later extent. | |
865 | */ | |
d1310b2e | 866 | err = insert_state(tree, prealloc, start, this_end, |
0ca1f7ce | 867 | &bits); |
c2d904e0 JM |
868 | if (err) |
869 | extent_io_tree_panic(tree, err); | |
870 | ||
9ed74f2d JB |
871 | cache_state(prealloc, cached_state); |
872 | prealloc = NULL; | |
d1310b2e CM |
873 | start = this_end + 1; |
874 | goto search_again; | |
875 | } | |
876 | /* | |
877 | * | ---- desired range ---- | | |
878 | * | state | | |
879 | * We need to split the extent, and set the bit | |
880 | * on the first half | |
881 | */ | |
882 | if (state->start <= end && state->end > end) { | |
1edbb734 | 883 | if (state->state & exclusive_bits) { |
d1310b2e CM |
884 | *failed_start = start; |
885 | err = -EEXIST; | |
886 | goto out; | |
887 | } | |
8233767a XG |
888 | |
889 | prealloc = alloc_extent_state_atomic(prealloc); | |
890 | BUG_ON(!prealloc); | |
d1310b2e | 891 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
892 | if (err) |
893 | extent_io_tree_panic(tree, err); | |
d1310b2e | 894 | |
1bf85046 | 895 | set_state_bits(tree, prealloc, &bits); |
2c64c53d | 896 | cache_state(prealloc, cached_state); |
d1310b2e CM |
897 | merge_state(tree, prealloc); |
898 | prealloc = NULL; | |
899 | goto out; | |
900 | } | |
901 | ||
902 | goto search_again; | |
903 | ||
904 | out: | |
cad321ad | 905 | spin_unlock(&tree->lock); |
d1310b2e CM |
906 | if (prealloc) |
907 | free_extent_state(prealloc); | |
908 | ||
909 | return err; | |
910 | ||
911 | search_again: | |
912 | if (start > end) | |
913 | goto out; | |
cad321ad | 914 | spin_unlock(&tree->lock); |
d1310b2e CM |
915 | if (mask & __GFP_WAIT) |
916 | cond_resched(); | |
917 | goto again; | |
918 | } | |
d1310b2e | 919 | |
462d6fac JB |
920 | /** |
921 | * convert_extent - convert all bits in a given range from one bit to another | |
922 | * @tree: the io tree to search | |
923 | * @start: the start offset in bytes | |
924 | * @end: the end offset in bytes (inclusive) | |
925 | * @bits: the bits to set in this range | |
926 | * @clear_bits: the bits to clear in this range | |
927 | * @mask: the allocation mask | |
928 | * | |
929 | * This will go through and set bits for the given range. If any states exist | |
930 | * already in this range they are set with the given bit and cleared of the | |
931 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
932 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
933 | * boundary bits like LOCK. | |
934 | */ | |
935 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
936 | int bits, int clear_bits, gfp_t mask) | |
937 | { | |
938 | struct extent_state *state; | |
939 | struct extent_state *prealloc = NULL; | |
940 | struct rb_node *node; | |
941 | int err = 0; | |
942 | u64 last_start; | |
943 | u64 last_end; | |
944 | ||
945 | again: | |
946 | if (!prealloc && (mask & __GFP_WAIT)) { | |
947 | prealloc = alloc_extent_state(mask); | |
948 | if (!prealloc) | |
949 | return -ENOMEM; | |
950 | } | |
951 | ||
952 | spin_lock(&tree->lock); | |
953 | /* | |
954 | * this search will find all the extents that end after | |
955 | * our range starts. | |
956 | */ | |
957 | node = tree_search(tree, start); | |
958 | if (!node) { | |
959 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
960 | if (!prealloc) { |
961 | err = -ENOMEM; | |
962 | goto out; | |
963 | } | |
462d6fac JB |
964 | err = insert_state(tree, prealloc, start, end, &bits); |
965 | prealloc = NULL; | |
c2d904e0 JM |
966 | if (err) |
967 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
968 | goto out; |
969 | } | |
970 | state = rb_entry(node, struct extent_state, rb_node); | |
971 | hit_next: | |
972 | last_start = state->start; | |
973 | last_end = state->end; | |
974 | ||
975 | /* | |
976 | * | ---- desired range ---- | | |
977 | * | state | | |
978 | * | |
979 | * Just lock what we found and keep going | |
980 | */ | |
981 | if (state->start == start && state->end <= end) { | |
982 | struct rb_node *next_node; | |
983 | ||
984 | set_state_bits(tree, state, &bits); | |
985 | clear_state_bit(tree, state, &clear_bits, 0); | |
462d6fac JB |
986 | if (last_end == (u64)-1) |
987 | goto out; | |
988 | ||
989 | start = last_end + 1; | |
990 | next_node = rb_next(&state->rb_node); | |
991 | if (next_node && start < end && prealloc && !need_resched()) { | |
992 | state = rb_entry(next_node, struct extent_state, | |
993 | rb_node); | |
994 | if (state->start == start) | |
995 | goto hit_next; | |
996 | } | |
997 | goto search_again; | |
998 | } | |
999 | ||
1000 | /* | |
1001 | * | ---- desired range ---- | | |
1002 | * | state | | |
1003 | * or | |
1004 | * | ------------- state -------------- | | |
1005 | * | |
1006 | * We need to split the extent we found, and may flip bits on | |
1007 | * second half. | |
1008 | * | |
1009 | * If the extent we found extends past our | |
1010 | * range, we just split and search again. It'll get split | |
1011 | * again the next time though. | |
1012 | * | |
1013 | * If the extent we found is inside our range, we set the | |
1014 | * desired bit on it. | |
1015 | */ | |
1016 | if (state->start < start) { | |
1017 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1018 | if (!prealloc) { |
1019 | err = -ENOMEM; | |
1020 | goto out; | |
1021 | } | |
462d6fac | 1022 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1023 | if (err) |
1024 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1025 | prealloc = NULL; |
1026 | if (err) | |
1027 | goto out; | |
1028 | if (state->end <= end) { | |
1029 | set_state_bits(tree, state, &bits); | |
1030 | clear_state_bit(tree, state, &clear_bits, 0); | |
462d6fac JB |
1031 | if (last_end == (u64)-1) |
1032 | goto out; | |
1033 | start = last_end + 1; | |
1034 | } | |
1035 | goto search_again; | |
1036 | } | |
1037 | /* | |
1038 | * | ---- desired range ---- | | |
1039 | * | state | or | state | | |
1040 | * | |
1041 | * There's a hole, we need to insert something in it and | |
1042 | * ignore the extent we found. | |
1043 | */ | |
1044 | if (state->start > start) { | |
1045 | u64 this_end; | |
1046 | if (end < last_start) | |
1047 | this_end = end; | |
1048 | else | |
1049 | this_end = last_start - 1; | |
1050 | ||
1051 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1052 | if (!prealloc) { |
1053 | err = -ENOMEM; | |
1054 | goto out; | |
1055 | } | |
462d6fac JB |
1056 | |
1057 | /* | |
1058 | * Avoid to free 'prealloc' if it can be merged with | |
1059 | * the later extent. | |
1060 | */ | |
1061 | err = insert_state(tree, prealloc, start, this_end, | |
1062 | &bits); | |
c2d904e0 JM |
1063 | if (err) |
1064 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1065 | prealloc = NULL; |
1066 | start = this_end + 1; | |
1067 | goto search_again; | |
1068 | } | |
1069 | /* | |
1070 | * | ---- desired range ---- | | |
1071 | * | state | | |
1072 | * We need to split the extent, and set the bit | |
1073 | * on the first half | |
1074 | */ | |
1075 | if (state->start <= end && state->end > end) { | |
1076 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1077 | if (!prealloc) { |
1078 | err = -ENOMEM; | |
1079 | goto out; | |
1080 | } | |
462d6fac JB |
1081 | |
1082 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1083 | if (err) |
1084 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1085 | |
1086 | set_state_bits(tree, prealloc, &bits); | |
1087 | clear_state_bit(tree, prealloc, &clear_bits, 0); | |
462d6fac JB |
1088 | prealloc = NULL; |
1089 | goto out; | |
1090 | } | |
1091 | ||
1092 | goto search_again; | |
1093 | ||
1094 | out: | |
1095 | spin_unlock(&tree->lock); | |
1096 | if (prealloc) | |
1097 | free_extent_state(prealloc); | |
1098 | ||
1099 | return err; | |
1100 | ||
1101 | search_again: | |
1102 | if (start > end) | |
1103 | goto out; | |
1104 | spin_unlock(&tree->lock); | |
1105 | if (mask & __GFP_WAIT) | |
1106 | cond_resched(); | |
1107 | goto again; | |
1108 | } | |
1109 | ||
d1310b2e CM |
1110 | /* wrappers around set/clear extent bit */ |
1111 | int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
1112 | gfp_t mask) | |
1113 | { | |
1114 | return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL, | |
2c64c53d | 1115 | NULL, mask); |
d1310b2e | 1116 | } |
d1310b2e CM |
1117 | |
1118 | int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
1119 | int bits, gfp_t mask) | |
1120 | { | |
1121 | return set_extent_bit(tree, start, end, bits, 0, NULL, | |
2c64c53d | 1122 | NULL, mask); |
d1310b2e | 1123 | } |
d1310b2e CM |
1124 | |
1125 | int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
1126 | int bits, gfp_t mask) | |
1127 | { | |
2c64c53d | 1128 | return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask); |
d1310b2e | 1129 | } |
d1310b2e CM |
1130 | |
1131 | int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, | |
2ac55d41 | 1132 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
1133 | { |
1134 | return set_extent_bit(tree, start, end, | |
fee187d9 | 1135 | EXTENT_DELALLOC | EXTENT_UPTODATE, |
2ac55d41 | 1136 | 0, NULL, cached_state, mask); |
d1310b2e | 1137 | } |
d1310b2e CM |
1138 | |
1139 | int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
1140 | gfp_t mask) | |
1141 | { | |
1142 | return clear_extent_bit(tree, start, end, | |
32c00aff | 1143 | EXTENT_DIRTY | EXTENT_DELALLOC | |
0ca1f7ce | 1144 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask); |
d1310b2e | 1145 | } |
d1310b2e CM |
1146 | |
1147 | int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, | |
1148 | gfp_t mask) | |
1149 | { | |
1150 | return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL, | |
2c64c53d | 1151 | NULL, mask); |
d1310b2e | 1152 | } |
d1310b2e | 1153 | |
d1310b2e | 1154 | int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
507903b8 | 1155 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e | 1156 | { |
507903b8 AJ |
1157 | return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, |
1158 | NULL, cached_state, mask); | |
d1310b2e | 1159 | } |
d1310b2e | 1160 | |
d397712b | 1161 | static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, |
2ac55d41 JB |
1162 | u64 end, struct extent_state **cached_state, |
1163 | gfp_t mask) | |
d1310b2e | 1164 | { |
2c64c53d | 1165 | return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, |
2ac55d41 | 1166 | cached_state, mask); |
d1310b2e | 1167 | } |
d1310b2e | 1168 | |
d352ac68 CM |
1169 | /* |
1170 | * either insert or lock state struct between start and end use mask to tell | |
1171 | * us if waiting is desired. | |
1172 | */ | |
1edbb734 | 1173 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
2c64c53d | 1174 | int bits, struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
1175 | { |
1176 | int err; | |
1177 | u64 failed_start; | |
1178 | while (1) { | |
1edbb734 | 1179 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits, |
2c64c53d CM |
1180 | EXTENT_LOCKED, &failed_start, |
1181 | cached_state, mask); | |
d1310b2e CM |
1182 | if (err == -EEXIST && (mask & __GFP_WAIT)) { |
1183 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); | |
1184 | start = failed_start; | |
1185 | } else { | |
1186 | break; | |
1187 | } | |
1188 | WARN_ON(start > end); | |
1189 | } | |
1190 | return err; | |
1191 | } | |
d1310b2e | 1192 | |
1edbb734 CM |
1193 | int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) |
1194 | { | |
2c64c53d | 1195 | return lock_extent_bits(tree, start, end, 0, NULL, mask); |
1edbb734 CM |
1196 | } |
1197 | ||
25179201 JB |
1198 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end, |
1199 | gfp_t mask) | |
1200 | { | |
1201 | int err; | |
1202 | u64 failed_start; | |
1203 | ||
2c64c53d CM |
1204 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
1205 | &failed_start, NULL, mask); | |
6643558d YZ |
1206 | if (err == -EEXIST) { |
1207 | if (failed_start > start) | |
1208 | clear_extent_bit(tree, start, failed_start - 1, | |
2c64c53d | 1209 | EXTENT_LOCKED, 1, 0, NULL, mask); |
25179201 | 1210 | return 0; |
6643558d | 1211 | } |
25179201 JB |
1212 | return 1; |
1213 | } | |
25179201 | 1214 | |
2c64c53d CM |
1215 | int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end, |
1216 | struct extent_state **cached, gfp_t mask) | |
1217 | { | |
1218 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, | |
1219 | mask); | |
1220 | } | |
1221 | ||
507903b8 | 1222 | int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) |
d1310b2e | 1223 | { |
2c64c53d CM |
1224 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL, |
1225 | mask); | |
d1310b2e | 1226 | } |
d1310b2e | 1227 | |
d1310b2e CM |
1228 | /* |
1229 | * helper function to set both pages and extents in the tree writeback | |
1230 | */ | |
b2950863 | 1231 | static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e CM |
1232 | { |
1233 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1234 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1235 | struct page *page; | |
1236 | ||
1237 | while (index <= end_index) { | |
1238 | page = find_get_page(tree->mapping, index); | |
1239 | BUG_ON(!page); | |
1240 | set_page_writeback(page); | |
1241 | page_cache_release(page); | |
1242 | index++; | |
1243 | } | |
d1310b2e CM |
1244 | return 0; |
1245 | } | |
d1310b2e | 1246 | |
d352ac68 CM |
1247 | /* find the first state struct with 'bits' set after 'start', and |
1248 | * return it. tree->lock must be held. NULL will returned if | |
1249 | * nothing was found after 'start' | |
1250 | */ | |
d7fc640e CM |
1251 | struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree, |
1252 | u64 start, int bits) | |
1253 | { | |
1254 | struct rb_node *node; | |
1255 | struct extent_state *state; | |
1256 | ||
1257 | /* | |
1258 | * this search will find all the extents that end after | |
1259 | * our range starts. | |
1260 | */ | |
1261 | node = tree_search(tree, start); | |
d397712b | 1262 | if (!node) |
d7fc640e | 1263 | goto out; |
d7fc640e | 1264 | |
d397712b | 1265 | while (1) { |
d7fc640e | 1266 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1267 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1268 | return state; |
d397712b | 1269 | |
d7fc640e CM |
1270 | node = rb_next(node); |
1271 | if (!node) | |
1272 | break; | |
1273 | } | |
1274 | out: | |
1275 | return NULL; | |
1276 | } | |
d7fc640e | 1277 | |
69261c4b XG |
1278 | /* |
1279 | * find the first offset in the io tree with 'bits' set. zero is | |
1280 | * returned if we find something, and *start_ret and *end_ret are | |
1281 | * set to reflect the state struct that was found. | |
1282 | * | |
1283 | * If nothing was found, 1 is returned, < 0 on error | |
1284 | */ | |
1285 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
1286 | u64 *start_ret, u64 *end_ret, int bits) | |
1287 | { | |
1288 | struct extent_state *state; | |
1289 | int ret = 1; | |
1290 | ||
1291 | spin_lock(&tree->lock); | |
1292 | state = find_first_extent_bit_state(tree, start, bits); | |
1293 | if (state) { | |
1294 | *start_ret = state->start; | |
1295 | *end_ret = state->end; | |
1296 | ret = 0; | |
1297 | } | |
1298 | spin_unlock(&tree->lock); | |
1299 | return ret; | |
1300 | } | |
1301 | ||
d352ac68 CM |
1302 | /* |
1303 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1304 | * more than 'max_bytes'. start and end are used to return the range, | |
1305 | * | |
1306 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1307 | */ | |
c8b97818 | 1308 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1309 | u64 *start, u64 *end, u64 max_bytes, |
1310 | struct extent_state **cached_state) | |
d1310b2e CM |
1311 | { |
1312 | struct rb_node *node; | |
1313 | struct extent_state *state; | |
1314 | u64 cur_start = *start; | |
1315 | u64 found = 0; | |
1316 | u64 total_bytes = 0; | |
1317 | ||
cad321ad | 1318 | spin_lock(&tree->lock); |
c8b97818 | 1319 | |
d1310b2e CM |
1320 | /* |
1321 | * this search will find all the extents that end after | |
1322 | * our range starts. | |
1323 | */ | |
80ea96b1 | 1324 | node = tree_search(tree, cur_start); |
2b114d1d | 1325 | if (!node) { |
3b951516 CM |
1326 | if (!found) |
1327 | *end = (u64)-1; | |
d1310b2e CM |
1328 | goto out; |
1329 | } | |
1330 | ||
d397712b | 1331 | while (1) { |
d1310b2e | 1332 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1333 | if (found && (state->start != cur_start || |
1334 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1335 | goto out; |
1336 | } | |
1337 | if (!(state->state & EXTENT_DELALLOC)) { | |
1338 | if (!found) | |
1339 | *end = state->end; | |
1340 | goto out; | |
1341 | } | |
c2a128d2 | 1342 | if (!found) { |
d1310b2e | 1343 | *start = state->start; |
c2a128d2 JB |
1344 | *cached_state = state; |
1345 | atomic_inc(&state->refs); | |
1346 | } | |
d1310b2e CM |
1347 | found++; |
1348 | *end = state->end; | |
1349 | cur_start = state->end + 1; | |
1350 | node = rb_next(node); | |
1351 | if (!node) | |
1352 | break; | |
1353 | total_bytes += state->end - state->start + 1; | |
1354 | if (total_bytes >= max_bytes) | |
1355 | break; | |
1356 | } | |
1357 | out: | |
cad321ad | 1358 | spin_unlock(&tree->lock); |
d1310b2e CM |
1359 | return found; |
1360 | } | |
1361 | ||
c8b97818 CM |
1362 | static noinline int __unlock_for_delalloc(struct inode *inode, |
1363 | struct page *locked_page, | |
1364 | u64 start, u64 end) | |
1365 | { | |
1366 | int ret; | |
1367 | struct page *pages[16]; | |
1368 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1369 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1370 | unsigned long nr_pages = end_index - index + 1; | |
1371 | int i; | |
1372 | ||
1373 | if (index == locked_page->index && end_index == index) | |
1374 | return 0; | |
1375 | ||
d397712b | 1376 | while (nr_pages > 0) { |
c8b97818 | 1377 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1378 | min_t(unsigned long, nr_pages, |
1379 | ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1380 | for (i = 0; i < ret; i++) { |
1381 | if (pages[i] != locked_page) | |
1382 | unlock_page(pages[i]); | |
1383 | page_cache_release(pages[i]); | |
1384 | } | |
1385 | nr_pages -= ret; | |
1386 | index += ret; | |
1387 | cond_resched(); | |
1388 | } | |
1389 | return 0; | |
1390 | } | |
1391 | ||
1392 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1393 | struct page *locked_page, | |
1394 | u64 delalloc_start, | |
1395 | u64 delalloc_end) | |
1396 | { | |
1397 | unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT; | |
1398 | unsigned long start_index = index; | |
1399 | unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT; | |
1400 | unsigned long pages_locked = 0; | |
1401 | struct page *pages[16]; | |
1402 | unsigned long nrpages; | |
1403 | int ret; | |
1404 | int i; | |
1405 | ||
1406 | /* the caller is responsible for locking the start index */ | |
1407 | if (index == locked_page->index && index == end_index) | |
1408 | return 0; | |
1409 | ||
1410 | /* skip the page at the start index */ | |
1411 | nrpages = end_index - index + 1; | |
d397712b | 1412 | while (nrpages > 0) { |
c8b97818 | 1413 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1414 | min_t(unsigned long, |
1415 | nrpages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1416 | if (ret == 0) { |
1417 | ret = -EAGAIN; | |
1418 | goto done; | |
1419 | } | |
1420 | /* now we have an array of pages, lock them all */ | |
1421 | for (i = 0; i < ret; i++) { | |
1422 | /* | |
1423 | * the caller is taking responsibility for | |
1424 | * locked_page | |
1425 | */ | |
771ed689 | 1426 | if (pages[i] != locked_page) { |
c8b97818 | 1427 | lock_page(pages[i]); |
f2b1c41c CM |
1428 | if (!PageDirty(pages[i]) || |
1429 | pages[i]->mapping != inode->i_mapping) { | |
771ed689 CM |
1430 | ret = -EAGAIN; |
1431 | unlock_page(pages[i]); | |
1432 | page_cache_release(pages[i]); | |
1433 | goto done; | |
1434 | } | |
1435 | } | |
c8b97818 | 1436 | page_cache_release(pages[i]); |
771ed689 | 1437 | pages_locked++; |
c8b97818 | 1438 | } |
c8b97818 CM |
1439 | nrpages -= ret; |
1440 | index += ret; | |
1441 | cond_resched(); | |
1442 | } | |
1443 | ret = 0; | |
1444 | done: | |
1445 | if (ret && pages_locked) { | |
1446 | __unlock_for_delalloc(inode, locked_page, | |
1447 | delalloc_start, | |
1448 | ((u64)(start_index + pages_locked - 1)) << | |
1449 | PAGE_CACHE_SHIFT); | |
1450 | } | |
1451 | return ret; | |
1452 | } | |
1453 | ||
1454 | /* | |
1455 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1456 | * more than 'max_bytes'. start and end are used to return the range, | |
1457 | * | |
1458 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1459 | */ | |
1460 | static noinline u64 find_lock_delalloc_range(struct inode *inode, | |
1461 | struct extent_io_tree *tree, | |
1462 | struct page *locked_page, | |
1463 | u64 *start, u64 *end, | |
1464 | u64 max_bytes) | |
1465 | { | |
1466 | u64 delalloc_start; | |
1467 | u64 delalloc_end; | |
1468 | u64 found; | |
9655d298 | 1469 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1470 | int ret; |
1471 | int loops = 0; | |
1472 | ||
1473 | again: | |
1474 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1475 | delalloc_start = *start; | |
1476 | delalloc_end = 0; | |
1477 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1478 | max_bytes, &cached_state); |
70b99e69 | 1479 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1480 | *start = delalloc_start; |
1481 | *end = delalloc_end; | |
c2a128d2 | 1482 | free_extent_state(cached_state); |
c8b97818 CM |
1483 | return found; |
1484 | } | |
1485 | ||
70b99e69 CM |
1486 | /* |
1487 | * start comes from the offset of locked_page. We have to lock | |
1488 | * pages in order, so we can't process delalloc bytes before | |
1489 | * locked_page | |
1490 | */ | |
d397712b | 1491 | if (delalloc_start < *start) |
70b99e69 | 1492 | delalloc_start = *start; |
70b99e69 | 1493 | |
c8b97818 CM |
1494 | /* |
1495 | * make sure to limit the number of pages we try to lock down | |
1496 | * if we're looping. | |
1497 | */ | |
d397712b | 1498 | if (delalloc_end + 1 - delalloc_start > max_bytes && loops) |
771ed689 | 1499 | delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1; |
d397712b | 1500 | |
c8b97818 CM |
1501 | /* step two, lock all the pages after the page that has start */ |
1502 | ret = lock_delalloc_pages(inode, locked_page, | |
1503 | delalloc_start, delalloc_end); | |
1504 | if (ret == -EAGAIN) { | |
1505 | /* some of the pages are gone, lets avoid looping by | |
1506 | * shortening the size of the delalloc range we're searching | |
1507 | */ | |
9655d298 | 1508 | free_extent_state(cached_state); |
c8b97818 CM |
1509 | if (!loops) { |
1510 | unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1); | |
1511 | max_bytes = PAGE_CACHE_SIZE - offset; | |
1512 | loops = 1; | |
1513 | goto again; | |
1514 | } else { | |
1515 | found = 0; | |
1516 | goto out_failed; | |
1517 | } | |
1518 | } | |
1519 | BUG_ON(ret); | |
1520 | ||
1521 | /* step three, lock the state bits for the whole range */ | |
9655d298 CM |
1522 | lock_extent_bits(tree, delalloc_start, delalloc_end, |
1523 | 0, &cached_state, GFP_NOFS); | |
c8b97818 CM |
1524 | |
1525 | /* then test to make sure it is all still delalloc */ | |
1526 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1527 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1528 | if (!ret) { |
9655d298 CM |
1529 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
1530 | &cached_state, GFP_NOFS); | |
c8b97818 CM |
1531 | __unlock_for_delalloc(inode, locked_page, |
1532 | delalloc_start, delalloc_end); | |
1533 | cond_resched(); | |
1534 | goto again; | |
1535 | } | |
9655d298 | 1536 | free_extent_state(cached_state); |
c8b97818 CM |
1537 | *start = delalloc_start; |
1538 | *end = delalloc_end; | |
1539 | out_failed: | |
1540 | return found; | |
1541 | } | |
1542 | ||
1543 | int extent_clear_unlock_delalloc(struct inode *inode, | |
1544 | struct extent_io_tree *tree, | |
1545 | u64 start, u64 end, struct page *locked_page, | |
a791e35e | 1546 | unsigned long op) |
c8b97818 CM |
1547 | { |
1548 | int ret; | |
1549 | struct page *pages[16]; | |
1550 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1551 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1552 | unsigned long nr_pages = end_index - index + 1; | |
1553 | int i; | |
771ed689 | 1554 | int clear_bits = 0; |
c8b97818 | 1555 | |
a791e35e | 1556 | if (op & EXTENT_CLEAR_UNLOCK) |
771ed689 | 1557 | clear_bits |= EXTENT_LOCKED; |
a791e35e | 1558 | if (op & EXTENT_CLEAR_DIRTY) |
c8b97818 CM |
1559 | clear_bits |= EXTENT_DIRTY; |
1560 | ||
a791e35e | 1561 | if (op & EXTENT_CLEAR_DELALLOC) |
771ed689 CM |
1562 | clear_bits |= EXTENT_DELALLOC; |
1563 | ||
2c64c53d | 1564 | clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); |
32c00aff JB |
1565 | if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | |
1566 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK | | |
1567 | EXTENT_SET_PRIVATE2))) | |
771ed689 | 1568 | return 0; |
c8b97818 | 1569 | |
d397712b | 1570 | while (nr_pages > 0) { |
c8b97818 | 1571 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1572 | min_t(unsigned long, |
1573 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 | 1574 | for (i = 0; i < ret; i++) { |
8b62b72b | 1575 | |
a791e35e | 1576 | if (op & EXTENT_SET_PRIVATE2) |
8b62b72b CM |
1577 | SetPagePrivate2(pages[i]); |
1578 | ||
c8b97818 CM |
1579 | if (pages[i] == locked_page) { |
1580 | page_cache_release(pages[i]); | |
1581 | continue; | |
1582 | } | |
a791e35e | 1583 | if (op & EXTENT_CLEAR_DIRTY) |
c8b97818 | 1584 | clear_page_dirty_for_io(pages[i]); |
a791e35e | 1585 | if (op & EXTENT_SET_WRITEBACK) |
c8b97818 | 1586 | set_page_writeback(pages[i]); |
a791e35e | 1587 | if (op & EXTENT_END_WRITEBACK) |
c8b97818 | 1588 | end_page_writeback(pages[i]); |
a791e35e | 1589 | if (op & EXTENT_CLEAR_UNLOCK_PAGE) |
771ed689 | 1590 | unlock_page(pages[i]); |
c8b97818 CM |
1591 | page_cache_release(pages[i]); |
1592 | } | |
1593 | nr_pages -= ret; | |
1594 | index += ret; | |
1595 | cond_resched(); | |
1596 | } | |
1597 | return 0; | |
1598 | } | |
c8b97818 | 1599 | |
d352ac68 CM |
1600 | /* |
1601 | * count the number of bytes in the tree that have a given bit(s) | |
1602 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1603 | * cached. The total number found is returned. | |
1604 | */ | |
d1310b2e CM |
1605 | u64 count_range_bits(struct extent_io_tree *tree, |
1606 | u64 *start, u64 search_end, u64 max_bytes, | |
ec29ed5b | 1607 | unsigned long bits, int contig) |
d1310b2e CM |
1608 | { |
1609 | struct rb_node *node; | |
1610 | struct extent_state *state; | |
1611 | u64 cur_start = *start; | |
1612 | u64 total_bytes = 0; | |
ec29ed5b | 1613 | u64 last = 0; |
d1310b2e CM |
1614 | int found = 0; |
1615 | ||
1616 | if (search_end <= cur_start) { | |
d1310b2e CM |
1617 | WARN_ON(1); |
1618 | return 0; | |
1619 | } | |
1620 | ||
cad321ad | 1621 | spin_lock(&tree->lock); |
d1310b2e CM |
1622 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1623 | total_bytes = tree->dirty_bytes; | |
1624 | goto out; | |
1625 | } | |
1626 | /* | |
1627 | * this search will find all the extents that end after | |
1628 | * our range starts. | |
1629 | */ | |
80ea96b1 | 1630 | node = tree_search(tree, cur_start); |
d397712b | 1631 | if (!node) |
d1310b2e | 1632 | goto out; |
d1310b2e | 1633 | |
d397712b | 1634 | while (1) { |
d1310b2e CM |
1635 | state = rb_entry(node, struct extent_state, rb_node); |
1636 | if (state->start > search_end) | |
1637 | break; | |
ec29ed5b CM |
1638 | if (contig && found && state->start > last + 1) |
1639 | break; | |
1640 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1641 | total_bytes += min(search_end, state->end) + 1 - |
1642 | max(cur_start, state->start); | |
1643 | if (total_bytes >= max_bytes) | |
1644 | break; | |
1645 | if (!found) { | |
af60bed2 | 1646 | *start = max(cur_start, state->start); |
d1310b2e CM |
1647 | found = 1; |
1648 | } | |
ec29ed5b CM |
1649 | last = state->end; |
1650 | } else if (contig && found) { | |
1651 | break; | |
d1310b2e CM |
1652 | } |
1653 | node = rb_next(node); | |
1654 | if (!node) | |
1655 | break; | |
1656 | } | |
1657 | out: | |
cad321ad | 1658 | spin_unlock(&tree->lock); |
d1310b2e CM |
1659 | return total_bytes; |
1660 | } | |
b2950863 | 1661 | |
d352ac68 CM |
1662 | /* |
1663 | * set the private field for a given byte offset in the tree. If there isn't | |
1664 | * an extent_state there already, this does nothing. | |
1665 | */ | |
d1310b2e CM |
1666 | int set_state_private(struct extent_io_tree *tree, u64 start, u64 private) |
1667 | { | |
1668 | struct rb_node *node; | |
1669 | struct extent_state *state; | |
1670 | int ret = 0; | |
1671 | ||
cad321ad | 1672 | spin_lock(&tree->lock); |
d1310b2e CM |
1673 | /* |
1674 | * this search will find all the extents that end after | |
1675 | * our range starts. | |
1676 | */ | |
80ea96b1 | 1677 | node = tree_search(tree, start); |
2b114d1d | 1678 | if (!node) { |
d1310b2e CM |
1679 | ret = -ENOENT; |
1680 | goto out; | |
1681 | } | |
1682 | state = rb_entry(node, struct extent_state, rb_node); | |
1683 | if (state->start != start) { | |
1684 | ret = -ENOENT; | |
1685 | goto out; | |
1686 | } | |
1687 | state->private = private; | |
1688 | out: | |
cad321ad | 1689 | spin_unlock(&tree->lock); |
d1310b2e CM |
1690 | return ret; |
1691 | } | |
1692 | ||
1693 | int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private) | |
1694 | { | |
1695 | struct rb_node *node; | |
1696 | struct extent_state *state; | |
1697 | int ret = 0; | |
1698 | ||
cad321ad | 1699 | spin_lock(&tree->lock); |
d1310b2e CM |
1700 | /* |
1701 | * this search will find all the extents that end after | |
1702 | * our range starts. | |
1703 | */ | |
80ea96b1 | 1704 | node = tree_search(tree, start); |
2b114d1d | 1705 | if (!node) { |
d1310b2e CM |
1706 | ret = -ENOENT; |
1707 | goto out; | |
1708 | } | |
1709 | state = rb_entry(node, struct extent_state, rb_node); | |
1710 | if (state->start != start) { | |
1711 | ret = -ENOENT; | |
1712 | goto out; | |
1713 | } | |
1714 | *private = state->private; | |
1715 | out: | |
cad321ad | 1716 | spin_unlock(&tree->lock); |
d1310b2e CM |
1717 | return ret; |
1718 | } | |
1719 | ||
1720 | /* | |
1721 | * searches a range in the state tree for a given mask. | |
70dec807 | 1722 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1723 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1724 | * range is found set. | |
1725 | */ | |
1726 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9655d298 | 1727 | int bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1728 | { |
1729 | struct extent_state *state = NULL; | |
1730 | struct rb_node *node; | |
1731 | int bitset = 0; | |
d1310b2e | 1732 | |
cad321ad | 1733 | spin_lock(&tree->lock); |
df98b6e2 JB |
1734 | if (cached && cached->tree && cached->start <= start && |
1735 | cached->end > start) | |
9655d298 CM |
1736 | node = &cached->rb_node; |
1737 | else | |
1738 | node = tree_search(tree, start); | |
d1310b2e CM |
1739 | while (node && start <= end) { |
1740 | state = rb_entry(node, struct extent_state, rb_node); | |
1741 | ||
1742 | if (filled && state->start > start) { | |
1743 | bitset = 0; | |
1744 | break; | |
1745 | } | |
1746 | ||
1747 | if (state->start > end) | |
1748 | break; | |
1749 | ||
1750 | if (state->state & bits) { | |
1751 | bitset = 1; | |
1752 | if (!filled) | |
1753 | break; | |
1754 | } else if (filled) { | |
1755 | bitset = 0; | |
1756 | break; | |
1757 | } | |
46562cec CM |
1758 | |
1759 | if (state->end == (u64)-1) | |
1760 | break; | |
1761 | ||
d1310b2e CM |
1762 | start = state->end + 1; |
1763 | if (start > end) | |
1764 | break; | |
1765 | node = rb_next(node); | |
1766 | if (!node) { | |
1767 | if (filled) | |
1768 | bitset = 0; | |
1769 | break; | |
1770 | } | |
1771 | } | |
cad321ad | 1772 | spin_unlock(&tree->lock); |
d1310b2e CM |
1773 | return bitset; |
1774 | } | |
d1310b2e CM |
1775 | |
1776 | /* | |
1777 | * helper function to set a given page up to date if all the | |
1778 | * extents in the tree for that page are up to date | |
1779 | */ | |
1780 | static int check_page_uptodate(struct extent_io_tree *tree, | |
1781 | struct page *page) | |
1782 | { | |
1783 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1784 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
9655d298 | 1785 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e CM |
1786 | SetPageUptodate(page); |
1787 | return 0; | |
1788 | } | |
1789 | ||
1790 | /* | |
1791 | * helper function to unlock a page if all the extents in the tree | |
1792 | * for that page are unlocked | |
1793 | */ | |
1794 | static int check_page_locked(struct extent_io_tree *tree, | |
1795 | struct page *page) | |
1796 | { | |
1797 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1798 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
9655d298 | 1799 | if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) |
d1310b2e CM |
1800 | unlock_page(page); |
1801 | return 0; | |
1802 | } | |
1803 | ||
1804 | /* | |
1805 | * helper function to end page writeback if all the extents | |
1806 | * in the tree for that page are done with writeback | |
1807 | */ | |
1808 | static int check_page_writeback(struct extent_io_tree *tree, | |
1809 | struct page *page) | |
1810 | { | |
1edbb734 | 1811 | end_page_writeback(page); |
d1310b2e CM |
1812 | return 0; |
1813 | } | |
1814 | ||
4a54c8c1 JS |
1815 | /* |
1816 | * When IO fails, either with EIO or csum verification fails, we | |
1817 | * try other mirrors that might have a good copy of the data. This | |
1818 | * io_failure_record is used to record state as we go through all the | |
1819 | * mirrors. If another mirror has good data, the page is set up to date | |
1820 | * and things continue. If a good mirror can't be found, the original | |
1821 | * bio end_io callback is called to indicate things have failed. | |
1822 | */ | |
1823 | struct io_failure_record { | |
1824 | struct page *page; | |
1825 | u64 start; | |
1826 | u64 len; | |
1827 | u64 logical; | |
1828 | unsigned long bio_flags; | |
1829 | int this_mirror; | |
1830 | int failed_mirror; | |
1831 | int in_validation; | |
1832 | }; | |
1833 | ||
1834 | static int free_io_failure(struct inode *inode, struct io_failure_record *rec, | |
1835 | int did_repair) | |
1836 | { | |
1837 | int ret; | |
1838 | int err = 0; | |
1839 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
1840 | ||
1841 | set_state_private(failure_tree, rec->start, 0); | |
1842 | ret = clear_extent_bits(failure_tree, rec->start, | |
1843 | rec->start + rec->len - 1, | |
1844 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1845 | if (ret) | |
1846 | err = ret; | |
1847 | ||
1848 | if (did_repair) { | |
1849 | ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start, | |
1850 | rec->start + rec->len - 1, | |
1851 | EXTENT_DAMAGED, GFP_NOFS); | |
1852 | if (ret && !err) | |
1853 | err = ret; | |
1854 | } | |
1855 | ||
1856 | kfree(rec); | |
1857 | return err; | |
1858 | } | |
1859 | ||
1860 | static void repair_io_failure_callback(struct bio *bio, int err) | |
1861 | { | |
1862 | complete(bio->bi_private); | |
1863 | } | |
1864 | ||
1865 | /* | |
1866 | * this bypasses the standard btrfs submit functions deliberately, as | |
1867 | * the standard behavior is to write all copies in a raid setup. here we only | |
1868 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
1869 | * submit_bio directly. | |
1870 | * to avoid any synchonization issues, wait for the data after writing, which | |
1871 | * actually prevents the read that triggered the error from finishing. | |
1872 | * currently, there can be no more than two copies of every data bit. thus, | |
1873 | * exactly one rewrite is required. | |
1874 | */ | |
1875 | int repair_io_failure(struct btrfs_mapping_tree *map_tree, u64 start, | |
1876 | u64 length, u64 logical, struct page *page, | |
1877 | int mirror_num) | |
1878 | { | |
1879 | struct bio *bio; | |
1880 | struct btrfs_device *dev; | |
1881 | DECLARE_COMPLETION_ONSTACK(compl); | |
1882 | u64 map_length = 0; | |
1883 | u64 sector; | |
1884 | struct btrfs_bio *bbio = NULL; | |
1885 | int ret; | |
1886 | ||
1887 | BUG_ON(!mirror_num); | |
1888 | ||
1889 | bio = bio_alloc(GFP_NOFS, 1); | |
1890 | if (!bio) | |
1891 | return -EIO; | |
1892 | bio->bi_private = &compl; | |
1893 | bio->bi_end_io = repair_io_failure_callback; | |
1894 | bio->bi_size = 0; | |
1895 | map_length = length; | |
1896 | ||
1897 | ret = btrfs_map_block(map_tree, WRITE, logical, | |
1898 | &map_length, &bbio, mirror_num); | |
1899 | if (ret) { | |
1900 | bio_put(bio); | |
1901 | return -EIO; | |
1902 | } | |
1903 | BUG_ON(mirror_num != bbio->mirror_num); | |
1904 | sector = bbio->stripes[mirror_num-1].physical >> 9; | |
1905 | bio->bi_sector = sector; | |
1906 | dev = bbio->stripes[mirror_num-1].dev; | |
1907 | kfree(bbio); | |
1908 | if (!dev || !dev->bdev || !dev->writeable) { | |
1909 | bio_put(bio); | |
1910 | return -EIO; | |
1911 | } | |
1912 | bio->bi_bdev = dev->bdev; | |
1913 | bio_add_page(bio, page, length, start-page_offset(page)); | |
21adbd5c | 1914 | btrfsic_submit_bio(WRITE_SYNC, bio); |
4a54c8c1 JS |
1915 | wait_for_completion(&compl); |
1916 | ||
1917 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) { | |
1918 | /* try to remap that extent elsewhere? */ | |
1919 | bio_put(bio); | |
1920 | return -EIO; | |
1921 | } | |
1922 | ||
1923 | printk(KERN_INFO "btrfs read error corrected: ino %lu off %llu (dev %s " | |
1924 | "sector %llu)\n", page->mapping->host->i_ino, start, | |
1925 | dev->name, sector); | |
1926 | ||
1927 | bio_put(bio); | |
1928 | return 0; | |
1929 | } | |
1930 | ||
1931 | /* | |
1932 | * each time an IO finishes, we do a fast check in the IO failure tree | |
1933 | * to see if we need to process or clean up an io_failure_record | |
1934 | */ | |
1935 | static int clean_io_failure(u64 start, struct page *page) | |
1936 | { | |
1937 | u64 private; | |
1938 | u64 private_failure; | |
1939 | struct io_failure_record *failrec; | |
1940 | struct btrfs_mapping_tree *map_tree; | |
1941 | struct extent_state *state; | |
1942 | int num_copies; | |
1943 | int did_repair = 0; | |
1944 | int ret; | |
1945 | struct inode *inode = page->mapping->host; | |
1946 | ||
1947 | private = 0; | |
1948 | ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
1949 | (u64)-1, 1, EXTENT_DIRTY, 0); | |
1950 | if (!ret) | |
1951 | return 0; | |
1952 | ||
1953 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start, | |
1954 | &private_failure); | |
1955 | if (ret) | |
1956 | return 0; | |
1957 | ||
1958 | failrec = (struct io_failure_record *)(unsigned long) private_failure; | |
1959 | BUG_ON(!failrec->this_mirror); | |
1960 | ||
1961 | if (failrec->in_validation) { | |
1962 | /* there was no real error, just free the record */ | |
1963 | pr_debug("clean_io_failure: freeing dummy error at %llu\n", | |
1964 | failrec->start); | |
1965 | did_repair = 1; | |
1966 | goto out; | |
1967 | } | |
1968 | ||
1969 | spin_lock(&BTRFS_I(inode)->io_tree.lock); | |
1970 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, | |
1971 | failrec->start, | |
1972 | EXTENT_LOCKED); | |
1973 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); | |
1974 | ||
1975 | if (state && state->start == failrec->start) { | |
1976 | map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree; | |
1977 | num_copies = btrfs_num_copies(map_tree, failrec->logical, | |
1978 | failrec->len); | |
1979 | if (num_copies > 1) { | |
1980 | ret = repair_io_failure(map_tree, start, failrec->len, | |
1981 | failrec->logical, page, | |
1982 | failrec->failed_mirror); | |
1983 | did_repair = !ret; | |
1984 | } | |
1985 | } | |
1986 | ||
1987 | out: | |
1988 | if (!ret) | |
1989 | ret = free_io_failure(inode, failrec, did_repair); | |
1990 | ||
1991 | return ret; | |
1992 | } | |
1993 | ||
1994 | /* | |
1995 | * this is a generic handler for readpage errors (default | |
1996 | * readpage_io_failed_hook). if other copies exist, read those and write back | |
1997 | * good data to the failed position. does not investigate in remapping the | |
1998 | * failed extent elsewhere, hoping the device will be smart enough to do this as | |
1999 | * needed | |
2000 | */ | |
2001 | ||
2002 | static int bio_readpage_error(struct bio *failed_bio, struct page *page, | |
2003 | u64 start, u64 end, int failed_mirror, | |
2004 | struct extent_state *state) | |
2005 | { | |
2006 | struct io_failure_record *failrec = NULL; | |
2007 | u64 private; | |
2008 | struct extent_map *em; | |
2009 | struct inode *inode = page->mapping->host; | |
2010 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
2011 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2012 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
2013 | struct bio *bio; | |
2014 | int num_copies; | |
2015 | int ret; | |
2016 | int read_mode; | |
2017 | u64 logical; | |
2018 | ||
2019 | BUG_ON(failed_bio->bi_rw & REQ_WRITE); | |
2020 | ||
2021 | ret = get_state_private(failure_tree, start, &private); | |
2022 | if (ret) { | |
2023 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2024 | if (!failrec) | |
2025 | return -ENOMEM; | |
2026 | failrec->start = start; | |
2027 | failrec->len = end - start + 1; | |
2028 | failrec->this_mirror = 0; | |
2029 | failrec->bio_flags = 0; | |
2030 | failrec->in_validation = 0; | |
2031 | ||
2032 | read_lock(&em_tree->lock); | |
2033 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2034 | if (!em) { | |
2035 | read_unlock(&em_tree->lock); | |
2036 | kfree(failrec); | |
2037 | return -EIO; | |
2038 | } | |
2039 | ||
2040 | if (em->start > start || em->start + em->len < start) { | |
2041 | free_extent_map(em); | |
2042 | em = NULL; | |
2043 | } | |
2044 | read_unlock(&em_tree->lock); | |
2045 | ||
2046 | if (!em || IS_ERR(em)) { | |
2047 | kfree(failrec); | |
2048 | return -EIO; | |
2049 | } | |
2050 | logical = start - em->start; | |
2051 | logical = em->block_start + logical; | |
2052 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2053 | logical = em->block_start; | |
2054 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2055 | extent_set_compress_type(&failrec->bio_flags, | |
2056 | em->compress_type); | |
2057 | } | |
2058 | pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, " | |
2059 | "len=%llu\n", logical, start, failrec->len); | |
2060 | failrec->logical = logical; | |
2061 | free_extent_map(em); | |
2062 | ||
2063 | /* set the bits in the private failure tree */ | |
2064 | ret = set_extent_bits(failure_tree, start, end, | |
2065 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
2066 | if (ret >= 0) | |
2067 | ret = set_state_private(failure_tree, start, | |
2068 | (u64)(unsigned long)failrec); | |
2069 | /* set the bits in the inode's tree */ | |
2070 | if (ret >= 0) | |
2071 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED, | |
2072 | GFP_NOFS); | |
2073 | if (ret < 0) { | |
2074 | kfree(failrec); | |
2075 | return ret; | |
2076 | } | |
2077 | } else { | |
2078 | failrec = (struct io_failure_record *)(unsigned long)private; | |
2079 | pr_debug("bio_readpage_error: (found) logical=%llu, " | |
2080 | "start=%llu, len=%llu, validation=%d\n", | |
2081 | failrec->logical, failrec->start, failrec->len, | |
2082 | failrec->in_validation); | |
2083 | /* | |
2084 | * when data can be on disk more than twice, add to failrec here | |
2085 | * (e.g. with a list for failed_mirror) to make | |
2086 | * clean_io_failure() clean all those errors at once. | |
2087 | */ | |
2088 | } | |
2089 | num_copies = btrfs_num_copies( | |
2090 | &BTRFS_I(inode)->root->fs_info->mapping_tree, | |
2091 | failrec->logical, failrec->len); | |
2092 | if (num_copies == 1) { | |
2093 | /* | |
2094 | * we only have a single copy of the data, so don't bother with | |
2095 | * all the retry and error correction code that follows. no | |
2096 | * matter what the error is, it is very likely to persist. | |
2097 | */ | |
2098 | pr_debug("bio_readpage_error: cannot repair, num_copies == 1. " | |
2099 | "state=%p, num_copies=%d, next_mirror %d, " | |
2100 | "failed_mirror %d\n", state, num_copies, | |
2101 | failrec->this_mirror, failed_mirror); | |
2102 | free_io_failure(inode, failrec, 0); | |
2103 | return -EIO; | |
2104 | } | |
2105 | ||
2106 | if (!state) { | |
2107 | spin_lock(&tree->lock); | |
2108 | state = find_first_extent_bit_state(tree, failrec->start, | |
2109 | EXTENT_LOCKED); | |
2110 | if (state && state->start != failrec->start) | |
2111 | state = NULL; | |
2112 | spin_unlock(&tree->lock); | |
2113 | } | |
2114 | ||
2115 | /* | |
2116 | * there are two premises: | |
2117 | * a) deliver good data to the caller | |
2118 | * b) correct the bad sectors on disk | |
2119 | */ | |
2120 | if (failed_bio->bi_vcnt > 1) { | |
2121 | /* | |
2122 | * to fulfill b), we need to know the exact failing sectors, as | |
2123 | * we don't want to rewrite any more than the failed ones. thus, | |
2124 | * we need separate read requests for the failed bio | |
2125 | * | |
2126 | * if the following BUG_ON triggers, our validation request got | |
2127 | * merged. we need separate requests for our algorithm to work. | |
2128 | */ | |
2129 | BUG_ON(failrec->in_validation); | |
2130 | failrec->in_validation = 1; | |
2131 | failrec->this_mirror = failed_mirror; | |
2132 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; | |
2133 | } else { | |
2134 | /* | |
2135 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2136 | * of the failed sector and if we succeed, we have setup | |
2137 | * everything for repair_io_failure to do the rest for us. | |
2138 | */ | |
2139 | if (failrec->in_validation) { | |
2140 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2141 | failrec->in_validation = 0; | |
2142 | failrec->this_mirror = 0; | |
2143 | } | |
2144 | failrec->failed_mirror = failed_mirror; | |
2145 | failrec->this_mirror++; | |
2146 | if (failrec->this_mirror == failed_mirror) | |
2147 | failrec->this_mirror++; | |
2148 | read_mode = READ_SYNC; | |
2149 | } | |
2150 | ||
2151 | if (!state || failrec->this_mirror > num_copies) { | |
2152 | pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, " | |
2153 | "next_mirror %d, failed_mirror %d\n", state, | |
2154 | num_copies, failrec->this_mirror, failed_mirror); | |
2155 | free_io_failure(inode, failrec, 0); | |
2156 | return -EIO; | |
2157 | } | |
2158 | ||
2159 | bio = bio_alloc(GFP_NOFS, 1); | |
2160 | bio->bi_private = state; | |
2161 | bio->bi_end_io = failed_bio->bi_end_io; | |
2162 | bio->bi_sector = failrec->logical >> 9; | |
2163 | bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; | |
2164 | bio->bi_size = 0; | |
2165 | ||
2166 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); | |
2167 | ||
2168 | pr_debug("bio_readpage_error: submitting new read[%#x] to " | |
2169 | "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode, | |
2170 | failrec->this_mirror, num_copies, failrec->in_validation); | |
2171 | ||
013bd4c3 TI |
2172 | ret = tree->ops->submit_bio_hook(inode, read_mode, bio, |
2173 | failrec->this_mirror, | |
2174 | failrec->bio_flags, 0); | |
2175 | return ret; | |
4a54c8c1 JS |
2176 | } |
2177 | ||
d1310b2e CM |
2178 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2179 | ||
87826df0 JM |
2180 | int end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
2181 | { | |
2182 | int uptodate = (err == 0); | |
2183 | struct extent_io_tree *tree; | |
2184 | int ret; | |
2185 | ||
2186 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
2187 | ||
2188 | if (tree->ops && tree->ops->writepage_end_io_hook) { | |
2189 | ret = tree->ops->writepage_end_io_hook(page, start, | |
2190 | end, NULL, uptodate); | |
2191 | if (ret) | |
2192 | uptodate = 0; | |
2193 | } | |
2194 | ||
2195 | if (!uptodate && tree->ops && | |
2196 | tree->ops->writepage_io_failed_hook) { | |
2197 | ret = tree->ops->writepage_io_failed_hook(NULL, page, | |
2198 | start, end, NULL); | |
2199 | /* Writeback already completed */ | |
2200 | if (ret == 0) | |
2201 | return 1; | |
2202 | } | |
2203 | ||
2204 | if (!uptodate) { | |
2205 | clear_extent_uptodate(tree, start, end, NULL, GFP_NOFS); | |
2206 | ClearPageUptodate(page); | |
2207 | SetPageError(page); | |
2208 | } | |
2209 | return 0; | |
2210 | } | |
2211 | ||
d1310b2e CM |
2212 | /* |
2213 | * after a writepage IO is done, we need to: | |
2214 | * clear the uptodate bits on error | |
2215 | * clear the writeback bits in the extent tree for this IO | |
2216 | * end_page_writeback if the page has no more pending IO | |
2217 | * | |
2218 | * Scheduling is not allowed, so the extent state tree is expected | |
2219 | * to have one and only one object corresponding to this IO. | |
2220 | */ | |
d1310b2e | 2221 | static void end_bio_extent_writepage(struct bio *bio, int err) |
d1310b2e | 2222 | { |
d1310b2e | 2223 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
902b22f3 | 2224 | struct extent_io_tree *tree; |
d1310b2e CM |
2225 | u64 start; |
2226 | u64 end; | |
2227 | int whole_page; | |
2228 | ||
d1310b2e CM |
2229 | do { |
2230 | struct page *page = bvec->bv_page; | |
902b22f3 DW |
2231 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
2232 | ||
d1310b2e CM |
2233 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
2234 | bvec->bv_offset; | |
2235 | end = start + bvec->bv_len - 1; | |
2236 | ||
2237 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
2238 | whole_page = 1; | |
2239 | else | |
2240 | whole_page = 0; | |
2241 | ||
2242 | if (--bvec >= bio->bi_io_vec) | |
2243 | prefetchw(&bvec->bv_page->flags); | |
1259ab75 | 2244 | |
87826df0 JM |
2245 | if (end_extent_writepage(page, err, start, end)) |
2246 | continue; | |
70dec807 | 2247 | |
d1310b2e CM |
2248 | if (whole_page) |
2249 | end_page_writeback(page); | |
2250 | else | |
2251 | check_page_writeback(tree, page); | |
d1310b2e | 2252 | } while (bvec >= bio->bi_io_vec); |
2b1f55b0 | 2253 | |
d1310b2e | 2254 | bio_put(bio); |
d1310b2e CM |
2255 | } |
2256 | ||
2257 | /* | |
2258 | * after a readpage IO is done, we need to: | |
2259 | * clear the uptodate bits on error | |
2260 | * set the uptodate bits if things worked | |
2261 | * set the page up to date if all extents in the tree are uptodate | |
2262 | * clear the lock bit in the extent tree | |
2263 | * unlock the page if there are no other extents locked for it | |
2264 | * | |
2265 | * Scheduling is not allowed, so the extent state tree is expected | |
2266 | * to have one and only one object corresponding to this IO. | |
2267 | */ | |
d1310b2e | 2268 | static void end_bio_extent_readpage(struct bio *bio, int err) |
d1310b2e CM |
2269 | { |
2270 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
4125bf76 CM |
2271 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
2272 | struct bio_vec *bvec = bio->bi_io_vec; | |
902b22f3 | 2273 | struct extent_io_tree *tree; |
d1310b2e CM |
2274 | u64 start; |
2275 | u64 end; | |
2276 | int whole_page; | |
2277 | int ret; | |
2278 | ||
d20f7043 CM |
2279 | if (err) |
2280 | uptodate = 0; | |
2281 | ||
d1310b2e CM |
2282 | do { |
2283 | struct page *page = bvec->bv_page; | |
507903b8 AJ |
2284 | struct extent_state *cached = NULL; |
2285 | struct extent_state *state; | |
2286 | ||
4a54c8c1 JS |
2287 | pr_debug("end_bio_extent_readpage: bi_vcnt=%d, idx=%d, err=%d, " |
2288 | "mirror=%ld\n", bio->bi_vcnt, bio->bi_idx, err, | |
2289 | (long int)bio->bi_bdev); | |
902b22f3 DW |
2290 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
2291 | ||
d1310b2e CM |
2292 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
2293 | bvec->bv_offset; | |
2294 | end = start + bvec->bv_len - 1; | |
2295 | ||
2296 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
2297 | whole_page = 1; | |
2298 | else | |
2299 | whole_page = 0; | |
2300 | ||
4125bf76 | 2301 | if (++bvec <= bvec_end) |
d1310b2e CM |
2302 | prefetchw(&bvec->bv_page->flags); |
2303 | ||
507903b8 | 2304 | spin_lock(&tree->lock); |
0d399205 | 2305 | state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED); |
109b36a2 | 2306 | if (state && state->start == start) { |
507903b8 AJ |
2307 | /* |
2308 | * take a reference on the state, unlock will drop | |
2309 | * the ref | |
2310 | */ | |
2311 | cache_state(state, &cached); | |
2312 | } | |
2313 | spin_unlock(&tree->lock); | |
2314 | ||
d1310b2e | 2315 | if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) { |
70dec807 | 2316 | ret = tree->ops->readpage_end_io_hook(page, start, end, |
507903b8 | 2317 | state); |
d1310b2e CM |
2318 | if (ret) |
2319 | uptodate = 0; | |
4a54c8c1 JS |
2320 | else |
2321 | clean_io_failure(start, page); | |
d1310b2e | 2322 | } |
4a54c8c1 | 2323 | if (!uptodate) { |
32240a91 JS |
2324 | int failed_mirror; |
2325 | failed_mirror = (int)(unsigned long)bio->bi_bdev; | |
f4a8e656 JS |
2326 | /* |
2327 | * The generic bio_readpage_error handles errors the | |
2328 | * following way: If possible, new read requests are | |
2329 | * created and submitted and will end up in | |
2330 | * end_bio_extent_readpage as well (if we're lucky, not | |
2331 | * in the !uptodate case). In that case it returns 0 and | |
2332 | * we just go on with the next page in our bio. If it | |
2333 | * can't handle the error it will return -EIO and we | |
2334 | * remain responsible for that page. | |
2335 | */ | |
2336 | ret = bio_readpage_error(bio, page, start, end, | |
2337 | failed_mirror, NULL); | |
7e38326f | 2338 | if (ret == 0) { |
f4a8e656 | 2339 | error_handled: |
3b951516 CM |
2340 | uptodate = |
2341 | test_bit(BIO_UPTODATE, &bio->bi_flags); | |
d20f7043 CM |
2342 | if (err) |
2343 | uptodate = 0; | |
507903b8 | 2344 | uncache_state(&cached); |
7e38326f CM |
2345 | continue; |
2346 | } | |
f4a8e656 JS |
2347 | if (tree->ops && tree->ops->readpage_io_failed_hook) { |
2348 | ret = tree->ops->readpage_io_failed_hook( | |
2349 | bio, page, start, end, | |
2350 | failed_mirror, state); | |
2351 | if (ret == 0) | |
2352 | goto error_handled; | |
2353 | } | |
7e38326f | 2354 | } |
d1310b2e | 2355 | |
771ed689 | 2356 | if (uptodate) { |
507903b8 | 2357 | set_extent_uptodate(tree, start, end, &cached, |
902b22f3 | 2358 | GFP_ATOMIC); |
771ed689 | 2359 | } |
507903b8 | 2360 | unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC); |
d1310b2e | 2361 | |
70dec807 CM |
2362 | if (whole_page) { |
2363 | if (uptodate) { | |
2364 | SetPageUptodate(page); | |
2365 | } else { | |
2366 | ClearPageUptodate(page); | |
2367 | SetPageError(page); | |
2368 | } | |
d1310b2e | 2369 | unlock_page(page); |
70dec807 CM |
2370 | } else { |
2371 | if (uptodate) { | |
2372 | check_page_uptodate(tree, page); | |
2373 | } else { | |
2374 | ClearPageUptodate(page); | |
2375 | SetPageError(page); | |
2376 | } | |
d1310b2e | 2377 | check_page_locked(tree, page); |
70dec807 | 2378 | } |
4125bf76 | 2379 | } while (bvec <= bvec_end); |
d1310b2e CM |
2380 | |
2381 | bio_put(bio); | |
d1310b2e CM |
2382 | } |
2383 | ||
88f794ed MX |
2384 | struct bio * |
2385 | btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
2386 | gfp_t gfp_flags) | |
d1310b2e CM |
2387 | { |
2388 | struct bio *bio; | |
2389 | ||
2390 | bio = bio_alloc(gfp_flags, nr_vecs); | |
2391 | ||
2392 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
2393 | while (!bio && (nr_vecs /= 2)) | |
2394 | bio = bio_alloc(gfp_flags, nr_vecs); | |
2395 | } | |
2396 | ||
2397 | if (bio) { | |
e1c4b745 | 2398 | bio->bi_size = 0; |
d1310b2e CM |
2399 | bio->bi_bdev = bdev; |
2400 | bio->bi_sector = first_sector; | |
2401 | } | |
2402 | return bio; | |
2403 | } | |
2404 | ||
c8b97818 CM |
2405 | static int submit_one_bio(int rw, struct bio *bio, int mirror_num, |
2406 | unsigned long bio_flags) | |
d1310b2e | 2407 | { |
d1310b2e | 2408 | int ret = 0; |
70dec807 CM |
2409 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
2410 | struct page *page = bvec->bv_page; | |
2411 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 2412 | u64 start; |
70dec807 CM |
2413 | |
2414 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset; | |
70dec807 | 2415 | |
902b22f3 | 2416 | bio->bi_private = NULL; |
d1310b2e CM |
2417 | |
2418 | bio_get(bio); | |
2419 | ||
065631f6 | 2420 | if (tree->ops && tree->ops->submit_bio_hook) |
6b82ce8d | 2421 | ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio, |
eaf25d93 | 2422 | mirror_num, bio_flags, start); |
0b86a832 | 2423 | else |
21adbd5c | 2424 | btrfsic_submit_bio(rw, bio); |
4a54c8c1 | 2425 | |
d1310b2e CM |
2426 | if (bio_flagged(bio, BIO_EOPNOTSUPP)) |
2427 | ret = -EOPNOTSUPP; | |
2428 | bio_put(bio); | |
2429 | return ret; | |
2430 | } | |
2431 | ||
3444a972 JM |
2432 | static int merge_bio(struct extent_io_tree *tree, struct page *page, |
2433 | unsigned long offset, size_t size, struct bio *bio, | |
2434 | unsigned long bio_flags) | |
2435 | { | |
2436 | int ret = 0; | |
2437 | if (tree->ops && tree->ops->merge_bio_hook) | |
2438 | ret = tree->ops->merge_bio_hook(page, offset, size, bio, | |
2439 | bio_flags); | |
2440 | BUG_ON(ret < 0); | |
2441 | return ret; | |
2442 | ||
2443 | } | |
2444 | ||
d1310b2e CM |
2445 | static int submit_extent_page(int rw, struct extent_io_tree *tree, |
2446 | struct page *page, sector_t sector, | |
2447 | size_t size, unsigned long offset, | |
2448 | struct block_device *bdev, | |
2449 | struct bio **bio_ret, | |
2450 | unsigned long max_pages, | |
f188591e | 2451 | bio_end_io_t end_io_func, |
c8b97818 CM |
2452 | int mirror_num, |
2453 | unsigned long prev_bio_flags, | |
2454 | unsigned long bio_flags) | |
d1310b2e CM |
2455 | { |
2456 | int ret = 0; | |
2457 | struct bio *bio; | |
2458 | int nr; | |
c8b97818 CM |
2459 | int contig = 0; |
2460 | int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED; | |
2461 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; | |
5b050f04 | 2462 | size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE); |
d1310b2e CM |
2463 | |
2464 | if (bio_ret && *bio_ret) { | |
2465 | bio = *bio_ret; | |
c8b97818 CM |
2466 | if (old_compressed) |
2467 | contig = bio->bi_sector == sector; | |
2468 | else | |
2469 | contig = bio->bi_sector + (bio->bi_size >> 9) == | |
2470 | sector; | |
2471 | ||
2472 | if (prev_bio_flags != bio_flags || !contig || | |
3444a972 | 2473 | merge_bio(tree, page, offset, page_size, bio, bio_flags) || |
c8b97818 CM |
2474 | bio_add_page(bio, page, page_size, offset) < page_size) { |
2475 | ret = submit_one_bio(rw, bio, mirror_num, | |
2476 | prev_bio_flags); | |
d1310b2e CM |
2477 | bio = NULL; |
2478 | } else { | |
2479 | return 0; | |
2480 | } | |
2481 | } | |
c8b97818 CM |
2482 | if (this_compressed) |
2483 | nr = BIO_MAX_PAGES; | |
2484 | else | |
2485 | nr = bio_get_nr_vecs(bdev); | |
2486 | ||
88f794ed | 2487 | bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH); |
5df67083 TI |
2488 | if (!bio) |
2489 | return -ENOMEM; | |
70dec807 | 2490 | |
c8b97818 | 2491 | bio_add_page(bio, page, page_size, offset); |
d1310b2e CM |
2492 | bio->bi_end_io = end_io_func; |
2493 | bio->bi_private = tree; | |
70dec807 | 2494 | |
d397712b | 2495 | if (bio_ret) |
d1310b2e | 2496 | *bio_ret = bio; |
d397712b | 2497 | else |
c8b97818 | 2498 | ret = submit_one_bio(rw, bio, mirror_num, bio_flags); |
d1310b2e CM |
2499 | |
2500 | return ret; | |
2501 | } | |
2502 | ||
2503 | void set_page_extent_mapped(struct page *page) | |
2504 | { | |
2505 | if (!PagePrivate(page)) { | |
2506 | SetPagePrivate(page); | |
d1310b2e | 2507 | page_cache_get(page); |
6af118ce | 2508 | set_page_private(page, EXTENT_PAGE_PRIVATE); |
d1310b2e CM |
2509 | } |
2510 | } | |
2511 | ||
b2950863 | 2512 | static void set_page_extent_head(struct page *page, unsigned long len) |
d1310b2e | 2513 | { |
eb14ab8e | 2514 | WARN_ON(!PagePrivate(page)); |
d1310b2e CM |
2515 | set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2); |
2516 | } | |
2517 | ||
2518 | /* | |
2519 | * basic readpage implementation. Locked extent state structs are inserted | |
2520 | * into the tree that are removed when the IO is done (by the end_io | |
2521 | * handlers) | |
2522 | */ | |
2523 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
2524 | struct page *page, | |
2525 | get_extent_t *get_extent, | |
c8b97818 CM |
2526 | struct bio **bio, int mirror_num, |
2527 | unsigned long *bio_flags) | |
d1310b2e CM |
2528 | { |
2529 | struct inode *inode = page->mapping->host; | |
2530 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2531 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2532 | u64 end; | |
2533 | u64 cur = start; | |
2534 | u64 extent_offset; | |
2535 | u64 last_byte = i_size_read(inode); | |
2536 | u64 block_start; | |
2537 | u64 cur_end; | |
2538 | sector_t sector; | |
2539 | struct extent_map *em; | |
2540 | struct block_device *bdev; | |
11c65dcc | 2541 | struct btrfs_ordered_extent *ordered; |
d1310b2e CM |
2542 | int ret; |
2543 | int nr = 0; | |
306e16ce | 2544 | size_t pg_offset = 0; |
d1310b2e | 2545 | size_t iosize; |
c8b97818 | 2546 | size_t disk_io_size; |
d1310b2e | 2547 | size_t blocksize = inode->i_sb->s_blocksize; |
c8b97818 | 2548 | unsigned long this_bio_flag = 0; |
d1310b2e CM |
2549 | |
2550 | set_page_extent_mapped(page); | |
2551 | ||
90a887c9 DM |
2552 | if (!PageUptodate(page)) { |
2553 | if (cleancache_get_page(page) == 0) { | |
2554 | BUG_ON(blocksize != PAGE_SIZE); | |
2555 | goto out; | |
2556 | } | |
2557 | } | |
2558 | ||
d1310b2e | 2559 | end = page_end; |
11c65dcc JB |
2560 | while (1) { |
2561 | lock_extent(tree, start, end, GFP_NOFS); | |
2562 | ordered = btrfs_lookup_ordered_extent(inode, start); | |
2563 | if (!ordered) | |
2564 | break; | |
2565 | unlock_extent(tree, start, end, GFP_NOFS); | |
2566 | btrfs_start_ordered_extent(inode, ordered, 1); | |
2567 | btrfs_put_ordered_extent(ordered); | |
2568 | } | |
d1310b2e | 2569 | |
c8b97818 CM |
2570 | if (page->index == last_byte >> PAGE_CACHE_SHIFT) { |
2571 | char *userpage; | |
2572 | size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1); | |
2573 | ||
2574 | if (zero_offset) { | |
2575 | iosize = PAGE_CACHE_SIZE - zero_offset; | |
2576 | userpage = kmap_atomic(page, KM_USER0); | |
2577 | memset(userpage + zero_offset, 0, iosize); | |
2578 | flush_dcache_page(page); | |
2579 | kunmap_atomic(userpage, KM_USER0); | |
2580 | } | |
2581 | } | |
d1310b2e CM |
2582 | while (cur <= end) { |
2583 | if (cur >= last_byte) { | |
2584 | char *userpage; | |
507903b8 AJ |
2585 | struct extent_state *cached = NULL; |
2586 | ||
306e16ce | 2587 | iosize = PAGE_CACHE_SIZE - pg_offset; |
d1310b2e | 2588 | userpage = kmap_atomic(page, KM_USER0); |
306e16ce | 2589 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e CM |
2590 | flush_dcache_page(page); |
2591 | kunmap_atomic(userpage, KM_USER0); | |
2592 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 AJ |
2593 | &cached, GFP_NOFS); |
2594 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
2595 | &cached, GFP_NOFS); | |
d1310b2e CM |
2596 | break; |
2597 | } | |
306e16ce | 2598 | em = get_extent(inode, page, pg_offset, cur, |
d1310b2e | 2599 | end - cur + 1, 0); |
c704005d | 2600 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e CM |
2601 | SetPageError(page); |
2602 | unlock_extent(tree, cur, end, GFP_NOFS); | |
2603 | break; | |
2604 | } | |
d1310b2e CM |
2605 | extent_offset = cur - em->start; |
2606 | BUG_ON(extent_map_end(em) <= cur); | |
2607 | BUG_ON(end < cur); | |
2608 | ||
261507a0 | 2609 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
c8b97818 | 2610 | this_bio_flag = EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
2611 | extent_set_compress_type(&this_bio_flag, |
2612 | em->compress_type); | |
2613 | } | |
c8b97818 | 2614 | |
d1310b2e CM |
2615 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
2616 | cur_end = min(extent_map_end(em) - 1, end); | |
2617 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
c8b97818 CM |
2618 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
2619 | disk_io_size = em->block_len; | |
2620 | sector = em->block_start >> 9; | |
2621 | } else { | |
2622 | sector = (em->block_start + extent_offset) >> 9; | |
2623 | disk_io_size = iosize; | |
2624 | } | |
d1310b2e CM |
2625 | bdev = em->bdev; |
2626 | block_start = em->block_start; | |
d899e052 YZ |
2627 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2628 | block_start = EXTENT_MAP_HOLE; | |
d1310b2e CM |
2629 | free_extent_map(em); |
2630 | em = NULL; | |
2631 | ||
2632 | /* we've found a hole, just zero and go on */ | |
2633 | if (block_start == EXTENT_MAP_HOLE) { | |
2634 | char *userpage; | |
507903b8 AJ |
2635 | struct extent_state *cached = NULL; |
2636 | ||
d1310b2e | 2637 | userpage = kmap_atomic(page, KM_USER0); |
306e16ce | 2638 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e CM |
2639 | flush_dcache_page(page); |
2640 | kunmap_atomic(userpage, KM_USER0); | |
2641 | ||
2642 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 AJ |
2643 | &cached, GFP_NOFS); |
2644 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
2645 | &cached, GFP_NOFS); | |
d1310b2e | 2646 | cur = cur + iosize; |
306e16ce | 2647 | pg_offset += iosize; |
d1310b2e CM |
2648 | continue; |
2649 | } | |
2650 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
2651 | if (test_range_bit(tree, cur, cur_end, |
2652 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 2653 | check_page_uptodate(tree, page); |
d1310b2e CM |
2654 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); |
2655 | cur = cur + iosize; | |
306e16ce | 2656 | pg_offset += iosize; |
d1310b2e CM |
2657 | continue; |
2658 | } | |
70dec807 CM |
2659 | /* we have an inline extent but it didn't get marked up |
2660 | * to date. Error out | |
2661 | */ | |
2662 | if (block_start == EXTENT_MAP_INLINE) { | |
2663 | SetPageError(page); | |
2664 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
2665 | cur = cur + iosize; | |
306e16ce | 2666 | pg_offset += iosize; |
70dec807 CM |
2667 | continue; |
2668 | } | |
d1310b2e CM |
2669 | |
2670 | ret = 0; | |
2671 | if (tree->ops && tree->ops->readpage_io_hook) { | |
2672 | ret = tree->ops->readpage_io_hook(page, cur, | |
2673 | cur + iosize - 1); | |
2674 | } | |
2675 | if (!ret) { | |
89642229 CM |
2676 | unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1; |
2677 | pnr -= page->index; | |
d1310b2e | 2678 | ret = submit_extent_page(READ, tree, page, |
306e16ce | 2679 | sector, disk_io_size, pg_offset, |
89642229 | 2680 | bdev, bio, pnr, |
c8b97818 CM |
2681 | end_bio_extent_readpage, mirror_num, |
2682 | *bio_flags, | |
2683 | this_bio_flag); | |
89642229 | 2684 | nr++; |
c8b97818 | 2685 | *bio_flags = this_bio_flag; |
d1310b2e CM |
2686 | } |
2687 | if (ret) | |
2688 | SetPageError(page); | |
2689 | cur = cur + iosize; | |
306e16ce | 2690 | pg_offset += iosize; |
d1310b2e | 2691 | } |
90a887c9 | 2692 | out: |
d1310b2e CM |
2693 | if (!nr) { |
2694 | if (!PageError(page)) | |
2695 | SetPageUptodate(page); | |
2696 | unlock_page(page); | |
2697 | } | |
2698 | return 0; | |
2699 | } | |
2700 | ||
2701 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, | |
8ddc7d9c | 2702 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
2703 | { |
2704 | struct bio *bio = NULL; | |
c8b97818 | 2705 | unsigned long bio_flags = 0; |
d1310b2e CM |
2706 | int ret; |
2707 | ||
8ddc7d9c | 2708 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, |
c8b97818 | 2709 | &bio_flags); |
d1310b2e | 2710 | if (bio) |
8ddc7d9c | 2711 | ret = submit_one_bio(READ, bio, mirror_num, bio_flags); |
d1310b2e CM |
2712 | return ret; |
2713 | } | |
d1310b2e | 2714 | |
11c8349b CM |
2715 | static noinline void update_nr_written(struct page *page, |
2716 | struct writeback_control *wbc, | |
2717 | unsigned long nr_written) | |
2718 | { | |
2719 | wbc->nr_to_write -= nr_written; | |
2720 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && | |
2721 | wbc->range_start == 0 && wbc->range_end == LLONG_MAX)) | |
2722 | page->mapping->writeback_index = page->index + nr_written; | |
2723 | } | |
2724 | ||
d1310b2e CM |
2725 | /* |
2726 | * the writepage semantics are similar to regular writepage. extent | |
2727 | * records are inserted to lock ranges in the tree, and as dirty areas | |
2728 | * are found, they are marked writeback. Then the lock bits are removed | |
2729 | * and the end_io handler clears the writeback ranges | |
2730 | */ | |
2731 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
2732 | void *data) | |
2733 | { | |
2734 | struct inode *inode = page->mapping->host; | |
2735 | struct extent_page_data *epd = data; | |
2736 | struct extent_io_tree *tree = epd->tree; | |
2737 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2738 | u64 delalloc_start; | |
2739 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2740 | u64 end; | |
2741 | u64 cur = start; | |
2742 | u64 extent_offset; | |
2743 | u64 last_byte = i_size_read(inode); | |
2744 | u64 block_start; | |
2745 | u64 iosize; | |
2746 | sector_t sector; | |
2c64c53d | 2747 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
2748 | struct extent_map *em; |
2749 | struct block_device *bdev; | |
2750 | int ret; | |
2751 | int nr = 0; | |
7f3c74fb | 2752 | size_t pg_offset = 0; |
d1310b2e CM |
2753 | size_t blocksize; |
2754 | loff_t i_size = i_size_read(inode); | |
2755 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
2756 | u64 nr_delalloc; | |
2757 | u64 delalloc_end; | |
c8b97818 CM |
2758 | int page_started; |
2759 | int compressed; | |
ffbd517d | 2760 | int write_flags; |
771ed689 | 2761 | unsigned long nr_written = 0; |
9e487107 | 2762 | bool fill_delalloc = true; |
d1310b2e | 2763 | |
ffbd517d | 2764 | if (wbc->sync_mode == WB_SYNC_ALL) |
721a9602 | 2765 | write_flags = WRITE_SYNC; |
ffbd517d CM |
2766 | else |
2767 | write_flags = WRITE; | |
2768 | ||
1abe9b8a | 2769 | trace___extent_writepage(page, inode, wbc); |
2770 | ||
d1310b2e | 2771 | WARN_ON(!PageLocked(page)); |
bf0da8c1 CM |
2772 | |
2773 | ClearPageError(page); | |
2774 | ||
7f3c74fb | 2775 | pg_offset = i_size & (PAGE_CACHE_SIZE - 1); |
211c17f5 | 2776 | if (page->index > end_index || |
7f3c74fb | 2777 | (page->index == end_index && !pg_offset)) { |
39be25cd | 2778 | page->mapping->a_ops->invalidatepage(page, 0); |
d1310b2e CM |
2779 | unlock_page(page); |
2780 | return 0; | |
2781 | } | |
2782 | ||
2783 | if (page->index == end_index) { | |
2784 | char *userpage; | |
2785 | ||
d1310b2e | 2786 | userpage = kmap_atomic(page, KM_USER0); |
7f3c74fb CM |
2787 | memset(userpage + pg_offset, 0, |
2788 | PAGE_CACHE_SIZE - pg_offset); | |
d1310b2e | 2789 | kunmap_atomic(userpage, KM_USER0); |
211c17f5 | 2790 | flush_dcache_page(page); |
d1310b2e | 2791 | } |
7f3c74fb | 2792 | pg_offset = 0; |
d1310b2e CM |
2793 | |
2794 | set_page_extent_mapped(page); | |
2795 | ||
9e487107 JB |
2796 | if (!tree->ops || !tree->ops->fill_delalloc) |
2797 | fill_delalloc = false; | |
2798 | ||
d1310b2e CM |
2799 | delalloc_start = start; |
2800 | delalloc_end = 0; | |
c8b97818 | 2801 | page_started = 0; |
9e487107 | 2802 | if (!epd->extent_locked && fill_delalloc) { |
f85d7d6c | 2803 | u64 delalloc_to_write = 0; |
11c8349b CM |
2804 | /* |
2805 | * make sure the wbc mapping index is at least updated | |
2806 | * to this page. | |
2807 | */ | |
2808 | update_nr_written(page, wbc, 0); | |
2809 | ||
d397712b | 2810 | while (delalloc_end < page_end) { |
771ed689 | 2811 | nr_delalloc = find_lock_delalloc_range(inode, tree, |
c8b97818 CM |
2812 | page, |
2813 | &delalloc_start, | |
d1310b2e CM |
2814 | &delalloc_end, |
2815 | 128 * 1024 * 1024); | |
771ed689 CM |
2816 | if (nr_delalloc == 0) { |
2817 | delalloc_start = delalloc_end + 1; | |
2818 | continue; | |
2819 | } | |
013bd4c3 TI |
2820 | ret = tree->ops->fill_delalloc(inode, page, |
2821 | delalloc_start, | |
2822 | delalloc_end, | |
2823 | &page_started, | |
2824 | &nr_written); | |
2825 | BUG_ON(ret); | |
f85d7d6c CM |
2826 | /* |
2827 | * delalloc_end is already one less than the total | |
2828 | * length, so we don't subtract one from | |
2829 | * PAGE_CACHE_SIZE | |
2830 | */ | |
2831 | delalloc_to_write += (delalloc_end - delalloc_start + | |
2832 | PAGE_CACHE_SIZE) >> | |
2833 | PAGE_CACHE_SHIFT; | |
d1310b2e | 2834 | delalloc_start = delalloc_end + 1; |
d1310b2e | 2835 | } |
f85d7d6c CM |
2836 | if (wbc->nr_to_write < delalloc_to_write) { |
2837 | int thresh = 8192; | |
2838 | ||
2839 | if (delalloc_to_write < thresh * 2) | |
2840 | thresh = delalloc_to_write; | |
2841 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
2842 | thresh); | |
2843 | } | |
c8b97818 | 2844 | |
771ed689 CM |
2845 | /* did the fill delalloc function already unlock and start |
2846 | * the IO? | |
2847 | */ | |
2848 | if (page_started) { | |
2849 | ret = 0; | |
11c8349b CM |
2850 | /* |
2851 | * we've unlocked the page, so we can't update | |
2852 | * the mapping's writeback index, just update | |
2853 | * nr_to_write. | |
2854 | */ | |
2855 | wbc->nr_to_write -= nr_written; | |
2856 | goto done_unlocked; | |
771ed689 | 2857 | } |
c8b97818 | 2858 | } |
247e743c | 2859 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
2860 | ret = tree->ops->writepage_start_hook(page, start, |
2861 | page_end); | |
87826df0 JM |
2862 | if (ret) { |
2863 | /* Fixup worker will requeue */ | |
2864 | if (ret == -EBUSY) | |
2865 | wbc->pages_skipped++; | |
2866 | else | |
2867 | redirty_page_for_writepage(wbc, page); | |
11c8349b | 2868 | update_nr_written(page, wbc, nr_written); |
247e743c | 2869 | unlock_page(page); |
771ed689 | 2870 | ret = 0; |
11c8349b | 2871 | goto done_unlocked; |
247e743c CM |
2872 | } |
2873 | } | |
2874 | ||
11c8349b CM |
2875 | /* |
2876 | * we don't want to touch the inode after unlocking the page, | |
2877 | * so we update the mapping writeback index now | |
2878 | */ | |
2879 | update_nr_written(page, wbc, nr_written + 1); | |
771ed689 | 2880 | |
d1310b2e | 2881 | end = page_end; |
d1310b2e | 2882 | if (last_byte <= start) { |
e6dcd2dc CM |
2883 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2884 | tree->ops->writepage_end_io_hook(page, start, | |
2885 | page_end, NULL, 1); | |
d1310b2e CM |
2886 | goto done; |
2887 | } | |
2888 | ||
d1310b2e CM |
2889 | blocksize = inode->i_sb->s_blocksize; |
2890 | ||
2891 | while (cur <= end) { | |
2892 | if (cur >= last_byte) { | |
e6dcd2dc CM |
2893 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2894 | tree->ops->writepage_end_io_hook(page, cur, | |
2895 | page_end, NULL, 1); | |
d1310b2e CM |
2896 | break; |
2897 | } | |
7f3c74fb | 2898 | em = epd->get_extent(inode, page, pg_offset, cur, |
d1310b2e | 2899 | end - cur + 1, 1); |
c704005d | 2900 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e CM |
2901 | SetPageError(page); |
2902 | break; | |
2903 | } | |
2904 | ||
2905 | extent_offset = cur - em->start; | |
2906 | BUG_ON(extent_map_end(em) <= cur); | |
2907 | BUG_ON(end < cur); | |
2908 | iosize = min(extent_map_end(em) - cur, end - cur + 1); | |
2909 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
2910 | sector = (em->block_start + extent_offset) >> 9; | |
2911 | bdev = em->bdev; | |
2912 | block_start = em->block_start; | |
c8b97818 | 2913 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
2914 | free_extent_map(em); |
2915 | em = NULL; | |
2916 | ||
c8b97818 CM |
2917 | /* |
2918 | * compressed and inline extents are written through other | |
2919 | * paths in the FS | |
2920 | */ | |
2921 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 2922 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
2923 | /* |
2924 | * end_io notification does not happen here for | |
2925 | * compressed extents | |
2926 | */ | |
2927 | if (!compressed && tree->ops && | |
2928 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
2929 | tree->ops->writepage_end_io_hook(page, cur, |
2930 | cur + iosize - 1, | |
2931 | NULL, 1); | |
c8b97818 CM |
2932 | else if (compressed) { |
2933 | /* we don't want to end_page_writeback on | |
2934 | * a compressed extent. this happens | |
2935 | * elsewhere | |
2936 | */ | |
2937 | nr++; | |
2938 | } | |
2939 | ||
2940 | cur += iosize; | |
7f3c74fb | 2941 | pg_offset += iosize; |
d1310b2e CM |
2942 | continue; |
2943 | } | |
d1310b2e CM |
2944 | /* leave this out until we have a page_mkwrite call */ |
2945 | if (0 && !test_range_bit(tree, cur, cur + iosize - 1, | |
9655d298 | 2946 | EXTENT_DIRTY, 0, NULL)) { |
d1310b2e | 2947 | cur = cur + iosize; |
7f3c74fb | 2948 | pg_offset += iosize; |
d1310b2e CM |
2949 | continue; |
2950 | } | |
c8b97818 | 2951 | |
d1310b2e CM |
2952 | if (tree->ops && tree->ops->writepage_io_hook) { |
2953 | ret = tree->ops->writepage_io_hook(page, cur, | |
2954 | cur + iosize - 1); | |
2955 | } else { | |
2956 | ret = 0; | |
2957 | } | |
1259ab75 | 2958 | if (ret) { |
d1310b2e | 2959 | SetPageError(page); |
1259ab75 | 2960 | } else { |
d1310b2e | 2961 | unsigned long max_nr = end_index + 1; |
7f3c74fb | 2962 | |
d1310b2e CM |
2963 | set_range_writeback(tree, cur, cur + iosize - 1); |
2964 | if (!PageWriteback(page)) { | |
d397712b CM |
2965 | printk(KERN_ERR "btrfs warning page %lu not " |
2966 | "writeback, cur %llu end %llu\n", | |
2967 | page->index, (unsigned long long)cur, | |
d1310b2e CM |
2968 | (unsigned long long)end); |
2969 | } | |
2970 | ||
ffbd517d CM |
2971 | ret = submit_extent_page(write_flags, tree, page, |
2972 | sector, iosize, pg_offset, | |
2973 | bdev, &epd->bio, max_nr, | |
c8b97818 CM |
2974 | end_bio_extent_writepage, |
2975 | 0, 0, 0); | |
d1310b2e CM |
2976 | if (ret) |
2977 | SetPageError(page); | |
2978 | } | |
2979 | cur = cur + iosize; | |
7f3c74fb | 2980 | pg_offset += iosize; |
d1310b2e CM |
2981 | nr++; |
2982 | } | |
2983 | done: | |
2984 | if (nr == 0) { | |
2985 | /* make sure the mapping tag for page dirty gets cleared */ | |
2986 | set_page_writeback(page); | |
2987 | end_page_writeback(page); | |
2988 | } | |
d1310b2e | 2989 | unlock_page(page); |
771ed689 | 2990 | |
11c8349b CM |
2991 | done_unlocked: |
2992 | ||
2c64c53d CM |
2993 | /* drop our reference on any cached states */ |
2994 | free_extent_state(cached_state); | |
d1310b2e CM |
2995 | return 0; |
2996 | } | |
2997 | ||
d1310b2e | 2998 | /** |
4bef0848 | 2999 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
3000 | * @mapping: address space structure to write |
3001 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
3002 | * @writepage: function called for each page | |
3003 | * @data: data passed to writepage function | |
3004 | * | |
3005 | * If a page is already under I/O, write_cache_pages() skips it, even | |
3006 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
3007 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
3008 | * and msync() need to guarantee that all the data which was dirty at the time | |
3009 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
3010 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
3011 | * existing IO to complete. | |
3012 | */ | |
b2950863 | 3013 | static int extent_write_cache_pages(struct extent_io_tree *tree, |
4bef0848 CM |
3014 | struct address_space *mapping, |
3015 | struct writeback_control *wbc, | |
d2c3f4f6 CM |
3016 | writepage_t writepage, void *data, |
3017 | void (*flush_fn)(void *)) | |
d1310b2e | 3018 | { |
d1310b2e CM |
3019 | int ret = 0; |
3020 | int done = 0; | |
f85d7d6c | 3021 | int nr_to_write_done = 0; |
d1310b2e CM |
3022 | struct pagevec pvec; |
3023 | int nr_pages; | |
3024 | pgoff_t index; | |
3025 | pgoff_t end; /* Inclusive */ | |
3026 | int scanned = 0; | |
f7aaa06b | 3027 | int tag; |
d1310b2e | 3028 | |
d1310b2e CM |
3029 | pagevec_init(&pvec, 0); |
3030 | if (wbc->range_cyclic) { | |
3031 | index = mapping->writeback_index; /* Start from prev offset */ | |
3032 | end = -1; | |
3033 | } else { | |
3034 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
3035 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
3036 | scanned = 1; |
3037 | } | |
f7aaa06b JB |
3038 | if (wbc->sync_mode == WB_SYNC_ALL) |
3039 | tag = PAGECACHE_TAG_TOWRITE; | |
3040 | else | |
3041 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 3042 | retry: |
f7aaa06b JB |
3043 | if (wbc->sync_mode == WB_SYNC_ALL) |
3044 | tag_pages_for_writeback(mapping, index, end); | |
f85d7d6c | 3045 | while (!done && !nr_to_write_done && (index <= end) && |
f7aaa06b JB |
3046 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
3047 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
d1310b2e CM |
3048 | unsigned i; |
3049 | ||
3050 | scanned = 1; | |
3051 | for (i = 0; i < nr_pages; i++) { | |
3052 | struct page *page = pvec.pages[i]; | |
3053 | ||
3054 | /* | |
3055 | * At this point we hold neither mapping->tree_lock nor | |
3056 | * lock on the page itself: the page may be truncated or | |
3057 | * invalidated (changing page->mapping to NULL), or even | |
3058 | * swizzled back from swapper_space to tmpfs file | |
3059 | * mapping | |
3060 | */ | |
01d658f2 CM |
3061 | if (tree->ops && |
3062 | tree->ops->write_cache_pages_lock_hook) { | |
3063 | tree->ops->write_cache_pages_lock_hook(page, | |
3064 | data, flush_fn); | |
3065 | } else { | |
3066 | if (!trylock_page(page)) { | |
3067 | flush_fn(data); | |
3068 | lock_page(page); | |
3069 | } | |
3070 | } | |
d1310b2e CM |
3071 | |
3072 | if (unlikely(page->mapping != mapping)) { | |
3073 | unlock_page(page); | |
3074 | continue; | |
3075 | } | |
3076 | ||
3077 | if (!wbc->range_cyclic && page->index > end) { | |
3078 | done = 1; | |
3079 | unlock_page(page); | |
3080 | continue; | |
3081 | } | |
3082 | ||
d2c3f4f6 | 3083 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 CM |
3084 | if (PageWriteback(page)) |
3085 | flush_fn(data); | |
d1310b2e | 3086 | wait_on_page_writeback(page); |
d2c3f4f6 | 3087 | } |
d1310b2e CM |
3088 | |
3089 | if (PageWriteback(page) || | |
3090 | !clear_page_dirty_for_io(page)) { | |
3091 | unlock_page(page); | |
3092 | continue; | |
3093 | } | |
3094 | ||
3095 | ret = (*writepage)(page, wbc, data); | |
3096 | ||
3097 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
3098 | unlock_page(page); | |
3099 | ret = 0; | |
3100 | } | |
f85d7d6c | 3101 | if (ret) |
d1310b2e | 3102 | done = 1; |
f85d7d6c CM |
3103 | |
3104 | /* | |
3105 | * the filesystem may choose to bump up nr_to_write. | |
3106 | * We have to make sure to honor the new nr_to_write | |
3107 | * at any time | |
3108 | */ | |
3109 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
3110 | } |
3111 | pagevec_release(&pvec); | |
3112 | cond_resched(); | |
3113 | } | |
3114 | if (!scanned && !done) { | |
3115 | /* | |
3116 | * We hit the last page and there is more work to be done: wrap | |
3117 | * back to the start of the file | |
3118 | */ | |
3119 | scanned = 1; | |
3120 | index = 0; | |
3121 | goto retry; | |
3122 | } | |
d1310b2e CM |
3123 | return ret; |
3124 | } | |
d1310b2e | 3125 | |
ffbd517d | 3126 | static void flush_epd_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 3127 | { |
d2c3f4f6 | 3128 | if (epd->bio) { |
ffbd517d CM |
3129 | if (epd->sync_io) |
3130 | submit_one_bio(WRITE_SYNC, epd->bio, 0, 0); | |
3131 | else | |
3132 | submit_one_bio(WRITE, epd->bio, 0, 0); | |
d2c3f4f6 CM |
3133 | epd->bio = NULL; |
3134 | } | |
3135 | } | |
3136 | ||
ffbd517d CM |
3137 | static noinline void flush_write_bio(void *data) |
3138 | { | |
3139 | struct extent_page_data *epd = data; | |
3140 | flush_epd_write_bio(epd); | |
3141 | } | |
3142 | ||
d1310b2e CM |
3143 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, |
3144 | get_extent_t *get_extent, | |
3145 | struct writeback_control *wbc) | |
3146 | { | |
3147 | int ret; | |
d1310b2e CM |
3148 | struct extent_page_data epd = { |
3149 | .bio = NULL, | |
3150 | .tree = tree, | |
3151 | .get_extent = get_extent, | |
771ed689 | 3152 | .extent_locked = 0, |
ffbd517d | 3153 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e | 3154 | }; |
d1310b2e | 3155 | |
d1310b2e CM |
3156 | ret = __extent_writepage(page, wbc, &epd); |
3157 | ||
ffbd517d | 3158 | flush_epd_write_bio(&epd); |
d1310b2e CM |
3159 | return ret; |
3160 | } | |
d1310b2e | 3161 | |
771ed689 CM |
3162 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, |
3163 | u64 start, u64 end, get_extent_t *get_extent, | |
3164 | int mode) | |
3165 | { | |
3166 | int ret = 0; | |
3167 | struct address_space *mapping = inode->i_mapping; | |
3168 | struct page *page; | |
3169 | unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >> | |
3170 | PAGE_CACHE_SHIFT; | |
3171 | ||
3172 | struct extent_page_data epd = { | |
3173 | .bio = NULL, | |
3174 | .tree = tree, | |
3175 | .get_extent = get_extent, | |
3176 | .extent_locked = 1, | |
ffbd517d | 3177 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
3178 | }; |
3179 | struct writeback_control wbc_writepages = { | |
771ed689 | 3180 | .sync_mode = mode, |
771ed689 CM |
3181 | .nr_to_write = nr_pages * 2, |
3182 | .range_start = start, | |
3183 | .range_end = end + 1, | |
3184 | }; | |
3185 | ||
d397712b | 3186 | while (start <= end) { |
771ed689 CM |
3187 | page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); |
3188 | if (clear_page_dirty_for_io(page)) | |
3189 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
3190 | else { | |
3191 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
3192 | tree->ops->writepage_end_io_hook(page, start, | |
3193 | start + PAGE_CACHE_SIZE - 1, | |
3194 | NULL, 1); | |
3195 | unlock_page(page); | |
3196 | } | |
3197 | page_cache_release(page); | |
3198 | start += PAGE_CACHE_SIZE; | |
3199 | } | |
3200 | ||
ffbd517d | 3201 | flush_epd_write_bio(&epd); |
771ed689 CM |
3202 | return ret; |
3203 | } | |
d1310b2e CM |
3204 | |
3205 | int extent_writepages(struct extent_io_tree *tree, | |
3206 | struct address_space *mapping, | |
3207 | get_extent_t *get_extent, | |
3208 | struct writeback_control *wbc) | |
3209 | { | |
3210 | int ret = 0; | |
3211 | struct extent_page_data epd = { | |
3212 | .bio = NULL, | |
3213 | .tree = tree, | |
3214 | .get_extent = get_extent, | |
771ed689 | 3215 | .extent_locked = 0, |
ffbd517d | 3216 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
3217 | }; |
3218 | ||
4bef0848 | 3219 | ret = extent_write_cache_pages(tree, mapping, wbc, |
d2c3f4f6 CM |
3220 | __extent_writepage, &epd, |
3221 | flush_write_bio); | |
ffbd517d | 3222 | flush_epd_write_bio(&epd); |
d1310b2e CM |
3223 | return ret; |
3224 | } | |
d1310b2e CM |
3225 | |
3226 | int extent_readpages(struct extent_io_tree *tree, | |
3227 | struct address_space *mapping, | |
3228 | struct list_head *pages, unsigned nr_pages, | |
3229 | get_extent_t get_extent) | |
3230 | { | |
3231 | struct bio *bio = NULL; | |
3232 | unsigned page_idx; | |
c8b97818 | 3233 | unsigned long bio_flags = 0; |
d1310b2e | 3234 | |
d1310b2e CM |
3235 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
3236 | struct page *page = list_entry(pages->prev, struct page, lru); | |
3237 | ||
3238 | prefetchw(&page->flags); | |
3239 | list_del(&page->lru); | |
28ecb609 | 3240 | if (!add_to_page_cache_lru(page, mapping, |
43e817a1 | 3241 | page->index, GFP_NOFS)) { |
f188591e | 3242 | __extent_read_full_page(tree, page, get_extent, |
c8b97818 | 3243 | &bio, 0, &bio_flags); |
d1310b2e CM |
3244 | } |
3245 | page_cache_release(page); | |
3246 | } | |
d1310b2e CM |
3247 | BUG_ON(!list_empty(pages)); |
3248 | if (bio) | |
c8b97818 | 3249 | submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
3250 | return 0; |
3251 | } | |
d1310b2e CM |
3252 | |
3253 | /* | |
3254 | * basic invalidatepage code, this waits on any locked or writeback | |
3255 | * ranges corresponding to the page, and then deletes any extent state | |
3256 | * records from the tree | |
3257 | */ | |
3258 | int extent_invalidatepage(struct extent_io_tree *tree, | |
3259 | struct page *page, unsigned long offset) | |
3260 | { | |
2ac55d41 | 3261 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
3262 | u64 start = ((u64)page->index << PAGE_CACHE_SHIFT); |
3263 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
3264 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
3265 | ||
d397712b | 3266 | start += (offset + blocksize - 1) & ~(blocksize - 1); |
d1310b2e CM |
3267 | if (start > end) |
3268 | return 0; | |
3269 | ||
2ac55d41 | 3270 | lock_extent_bits(tree, start, end, 0, &cached_state, GFP_NOFS); |
1edbb734 | 3271 | wait_on_page_writeback(page); |
d1310b2e | 3272 | clear_extent_bit(tree, start, end, |
32c00aff JB |
3273 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
3274 | EXTENT_DO_ACCOUNTING, | |
2ac55d41 | 3275 | 1, 1, &cached_state, GFP_NOFS); |
d1310b2e CM |
3276 | return 0; |
3277 | } | |
d1310b2e | 3278 | |
7b13b7b1 CM |
3279 | /* |
3280 | * a helper for releasepage, this tests for areas of the page that | |
3281 | * are locked or under IO and drops the related state bits if it is safe | |
3282 | * to drop the page. | |
3283 | */ | |
3284 | int try_release_extent_state(struct extent_map_tree *map, | |
3285 | struct extent_io_tree *tree, struct page *page, | |
3286 | gfp_t mask) | |
3287 | { | |
3288 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
3289 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
3290 | int ret = 1; | |
3291 | ||
211f90e6 | 3292 | if (test_range_bit(tree, start, end, |
8b62b72b | 3293 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
3294 | ret = 0; |
3295 | else { | |
3296 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
3297 | mask = GFP_NOFS; | |
11ef160f CM |
3298 | /* |
3299 | * at this point we can safely clear everything except the | |
3300 | * locked bit and the nodatasum bit | |
3301 | */ | |
e3f24cc5 | 3302 | ret = clear_extent_bit(tree, start, end, |
11ef160f CM |
3303 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
3304 | 0, 0, NULL, mask); | |
e3f24cc5 CM |
3305 | |
3306 | /* if clear_extent_bit failed for enomem reasons, | |
3307 | * we can't allow the release to continue. | |
3308 | */ | |
3309 | if (ret < 0) | |
3310 | ret = 0; | |
3311 | else | |
3312 | ret = 1; | |
7b13b7b1 CM |
3313 | } |
3314 | return ret; | |
3315 | } | |
7b13b7b1 | 3316 | |
d1310b2e CM |
3317 | /* |
3318 | * a helper for releasepage. As long as there are no locked extents | |
3319 | * in the range corresponding to the page, both state records and extent | |
3320 | * map records are removed | |
3321 | */ | |
3322 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
3323 | struct extent_io_tree *tree, struct page *page, |
3324 | gfp_t mask) | |
d1310b2e CM |
3325 | { |
3326 | struct extent_map *em; | |
3327 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
3328 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
7b13b7b1 | 3329 | |
70dec807 CM |
3330 | if ((mask & __GFP_WAIT) && |
3331 | page->mapping->host->i_size > 16 * 1024 * 1024) { | |
39b5637f | 3332 | u64 len; |
70dec807 | 3333 | while (start <= end) { |
39b5637f | 3334 | len = end - start + 1; |
890871be | 3335 | write_lock(&map->lock); |
39b5637f | 3336 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 3337 | if (!em) { |
890871be | 3338 | write_unlock(&map->lock); |
70dec807 CM |
3339 | break; |
3340 | } | |
7f3c74fb CM |
3341 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
3342 | em->start != start) { | |
890871be | 3343 | write_unlock(&map->lock); |
70dec807 CM |
3344 | free_extent_map(em); |
3345 | break; | |
3346 | } | |
3347 | if (!test_range_bit(tree, em->start, | |
3348 | extent_map_end(em) - 1, | |
8b62b72b | 3349 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 3350 | 0, NULL)) { |
70dec807 CM |
3351 | remove_extent_mapping(map, em); |
3352 | /* once for the rb tree */ | |
3353 | free_extent_map(em); | |
3354 | } | |
3355 | start = extent_map_end(em); | |
890871be | 3356 | write_unlock(&map->lock); |
70dec807 CM |
3357 | |
3358 | /* once for us */ | |
d1310b2e CM |
3359 | free_extent_map(em); |
3360 | } | |
d1310b2e | 3361 | } |
7b13b7b1 | 3362 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 3363 | } |
d1310b2e | 3364 | |
ec29ed5b CM |
3365 | /* |
3366 | * helper function for fiemap, which doesn't want to see any holes. | |
3367 | * This maps until we find something past 'last' | |
3368 | */ | |
3369 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
3370 | u64 offset, | |
3371 | u64 last, | |
3372 | get_extent_t *get_extent) | |
3373 | { | |
3374 | u64 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
3375 | struct extent_map *em; | |
3376 | u64 len; | |
3377 | ||
3378 | if (offset >= last) | |
3379 | return NULL; | |
3380 | ||
3381 | while(1) { | |
3382 | len = last - offset; | |
3383 | if (len == 0) | |
3384 | break; | |
3385 | len = (len + sectorsize - 1) & ~(sectorsize - 1); | |
3386 | em = get_extent(inode, NULL, 0, offset, len, 0); | |
c704005d | 3387 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
3388 | return em; |
3389 | ||
3390 | /* if this isn't a hole return it */ | |
3391 | if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) && | |
3392 | em->block_start != EXTENT_MAP_HOLE) { | |
3393 | return em; | |
3394 | } | |
3395 | ||
3396 | /* this is a hole, advance to the next extent */ | |
3397 | offset = extent_map_end(em); | |
3398 | free_extent_map(em); | |
3399 | if (offset >= last) | |
3400 | break; | |
3401 | } | |
3402 | return NULL; | |
3403 | } | |
3404 | ||
1506fcc8 YS |
3405 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
3406 | __u64 start, __u64 len, get_extent_t *get_extent) | |
3407 | { | |
975f84fe | 3408 | int ret = 0; |
1506fcc8 YS |
3409 | u64 off = start; |
3410 | u64 max = start + len; | |
3411 | u32 flags = 0; | |
975f84fe JB |
3412 | u32 found_type; |
3413 | u64 last; | |
ec29ed5b | 3414 | u64 last_for_get_extent = 0; |
1506fcc8 | 3415 | u64 disko = 0; |
ec29ed5b | 3416 | u64 isize = i_size_read(inode); |
975f84fe | 3417 | struct btrfs_key found_key; |
1506fcc8 | 3418 | struct extent_map *em = NULL; |
2ac55d41 | 3419 | struct extent_state *cached_state = NULL; |
975f84fe JB |
3420 | struct btrfs_path *path; |
3421 | struct btrfs_file_extent_item *item; | |
1506fcc8 | 3422 | int end = 0; |
ec29ed5b CM |
3423 | u64 em_start = 0; |
3424 | u64 em_len = 0; | |
3425 | u64 em_end = 0; | |
1506fcc8 | 3426 | unsigned long emflags; |
1506fcc8 YS |
3427 | |
3428 | if (len == 0) | |
3429 | return -EINVAL; | |
3430 | ||
975f84fe JB |
3431 | path = btrfs_alloc_path(); |
3432 | if (!path) | |
3433 | return -ENOMEM; | |
3434 | path->leave_spinning = 1; | |
3435 | ||
4d479cf0 JB |
3436 | start = ALIGN(start, BTRFS_I(inode)->root->sectorsize); |
3437 | len = ALIGN(len, BTRFS_I(inode)->root->sectorsize); | |
3438 | ||
ec29ed5b CM |
3439 | /* |
3440 | * lookup the last file extent. We're not using i_size here | |
3441 | * because there might be preallocation past i_size | |
3442 | */ | |
975f84fe | 3443 | ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root, |
33345d01 | 3444 | path, btrfs_ino(inode), -1, 0); |
975f84fe JB |
3445 | if (ret < 0) { |
3446 | btrfs_free_path(path); | |
3447 | return ret; | |
3448 | } | |
3449 | WARN_ON(!ret); | |
3450 | path->slots[0]--; | |
3451 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3452 | struct btrfs_file_extent_item); | |
3453 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
3454 | found_type = btrfs_key_type(&found_key); | |
3455 | ||
ec29ed5b | 3456 | /* No extents, but there might be delalloc bits */ |
33345d01 | 3457 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 3458 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
3459 | /* have to trust i_size as the end */ |
3460 | last = (u64)-1; | |
3461 | last_for_get_extent = isize; | |
3462 | } else { | |
3463 | /* | |
3464 | * remember the start of the last extent. There are a | |
3465 | * bunch of different factors that go into the length of the | |
3466 | * extent, so its much less complex to remember where it started | |
3467 | */ | |
3468 | last = found_key.offset; | |
3469 | last_for_get_extent = last + 1; | |
975f84fe | 3470 | } |
975f84fe JB |
3471 | btrfs_free_path(path); |
3472 | ||
ec29ed5b CM |
3473 | /* |
3474 | * we might have some extents allocated but more delalloc past those | |
3475 | * extents. so, we trust isize unless the start of the last extent is | |
3476 | * beyond isize | |
3477 | */ | |
3478 | if (last < isize) { | |
3479 | last = (u64)-1; | |
3480 | last_for_get_extent = isize; | |
3481 | } | |
3482 | ||
2ac55d41 JB |
3483 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0, |
3484 | &cached_state, GFP_NOFS); | |
ec29ed5b | 3485 | |
4d479cf0 | 3486 | em = get_extent_skip_holes(inode, start, last_for_get_extent, |
ec29ed5b | 3487 | get_extent); |
1506fcc8 YS |
3488 | if (!em) |
3489 | goto out; | |
3490 | if (IS_ERR(em)) { | |
3491 | ret = PTR_ERR(em); | |
3492 | goto out; | |
3493 | } | |
975f84fe | 3494 | |
1506fcc8 | 3495 | while (!end) { |
ea8efc74 CM |
3496 | u64 offset_in_extent; |
3497 | ||
3498 | /* break if the extent we found is outside the range */ | |
3499 | if (em->start >= max || extent_map_end(em) < off) | |
3500 | break; | |
3501 | ||
3502 | /* | |
3503 | * get_extent may return an extent that starts before our | |
3504 | * requested range. We have to make sure the ranges | |
3505 | * we return to fiemap always move forward and don't | |
3506 | * overlap, so adjust the offsets here | |
3507 | */ | |
3508 | em_start = max(em->start, off); | |
1506fcc8 | 3509 | |
ea8efc74 CM |
3510 | /* |
3511 | * record the offset from the start of the extent | |
3512 | * for adjusting the disk offset below | |
3513 | */ | |
3514 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 3515 | em_end = extent_map_end(em); |
ea8efc74 | 3516 | em_len = em_end - em_start; |
ec29ed5b | 3517 | emflags = em->flags; |
1506fcc8 YS |
3518 | disko = 0; |
3519 | flags = 0; | |
3520 | ||
ea8efc74 CM |
3521 | /* |
3522 | * bump off for our next call to get_extent | |
3523 | */ | |
3524 | off = extent_map_end(em); | |
3525 | if (off >= max) | |
3526 | end = 1; | |
3527 | ||
93dbfad7 | 3528 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
3529 | end = 1; |
3530 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 3531 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
3532 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
3533 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 3534 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
3535 | flags |= (FIEMAP_EXTENT_DELALLOC | |
3536 | FIEMAP_EXTENT_UNKNOWN); | |
93dbfad7 | 3537 | } else { |
ea8efc74 | 3538 | disko = em->block_start + offset_in_extent; |
1506fcc8 YS |
3539 | } |
3540 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
3541 | flags |= FIEMAP_EXTENT_ENCODED; | |
3542 | ||
1506fcc8 YS |
3543 | free_extent_map(em); |
3544 | em = NULL; | |
ec29ed5b CM |
3545 | if ((em_start >= last) || em_len == (u64)-1 || |
3546 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
3547 | flags |= FIEMAP_EXTENT_LAST; |
3548 | end = 1; | |
3549 | } | |
3550 | ||
ec29ed5b CM |
3551 | /* now scan forward to see if this is really the last extent. */ |
3552 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
3553 | get_extent); | |
3554 | if (IS_ERR(em)) { | |
3555 | ret = PTR_ERR(em); | |
3556 | goto out; | |
3557 | } | |
3558 | if (!em) { | |
975f84fe JB |
3559 | flags |= FIEMAP_EXTENT_LAST; |
3560 | end = 1; | |
3561 | } | |
ec29ed5b CM |
3562 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, |
3563 | em_len, flags); | |
3564 | if (ret) | |
3565 | goto out_free; | |
1506fcc8 YS |
3566 | } |
3567 | out_free: | |
3568 | free_extent_map(em); | |
3569 | out: | |
2ac55d41 JB |
3570 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len, |
3571 | &cached_state, GFP_NOFS); | |
1506fcc8 YS |
3572 | return ret; |
3573 | } | |
3574 | ||
4a54c8c1 | 3575 | inline struct page *extent_buffer_page(struct extent_buffer *eb, |
d1310b2e CM |
3576 | unsigned long i) |
3577 | { | |
3578 | struct page *p; | |
3579 | struct address_space *mapping; | |
3580 | ||
3581 | if (i == 0) | |
3582 | return eb->first_page; | |
3583 | i += eb->start >> PAGE_CACHE_SHIFT; | |
3584 | mapping = eb->first_page->mapping; | |
33958dc6 CM |
3585 | if (!mapping) |
3586 | return NULL; | |
0ee0fda0 SW |
3587 | |
3588 | /* | |
3589 | * extent_buffer_page is only called after pinning the page | |
3590 | * by increasing the reference count. So we know the page must | |
3591 | * be in the radix tree. | |
3592 | */ | |
0ee0fda0 | 3593 | rcu_read_lock(); |
d1310b2e | 3594 | p = radix_tree_lookup(&mapping->page_tree, i); |
0ee0fda0 | 3595 | rcu_read_unlock(); |
2b1f55b0 | 3596 | |
d1310b2e CM |
3597 | return p; |
3598 | } | |
3599 | ||
4a54c8c1 | 3600 | inline unsigned long num_extent_pages(u64 start, u64 len) |
728131d8 | 3601 | { |
6af118ce CM |
3602 | return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) - |
3603 | (start >> PAGE_CACHE_SHIFT); | |
728131d8 CM |
3604 | } |
3605 | ||
d1310b2e CM |
3606 | static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree, |
3607 | u64 start, | |
3608 | unsigned long len, | |
3609 | gfp_t mask) | |
3610 | { | |
3611 | struct extent_buffer *eb = NULL; | |
3935127c | 3612 | #if LEAK_DEBUG |
2d2ae547 | 3613 | unsigned long flags; |
4bef0848 | 3614 | #endif |
d1310b2e | 3615 | |
d1310b2e | 3616 | eb = kmem_cache_zalloc(extent_buffer_cache, mask); |
91ca338d TI |
3617 | if (eb == NULL) |
3618 | return NULL; | |
d1310b2e CM |
3619 | eb->start = start; |
3620 | eb->len = len; | |
bd681513 CM |
3621 | rwlock_init(&eb->lock); |
3622 | atomic_set(&eb->write_locks, 0); | |
3623 | atomic_set(&eb->read_locks, 0); | |
3624 | atomic_set(&eb->blocking_readers, 0); | |
3625 | atomic_set(&eb->blocking_writers, 0); | |
3626 | atomic_set(&eb->spinning_readers, 0); | |
3627 | atomic_set(&eb->spinning_writers, 0); | |
5b25f70f | 3628 | eb->lock_nested = 0; |
bd681513 CM |
3629 | init_waitqueue_head(&eb->write_lock_wq); |
3630 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 3631 | |
3935127c | 3632 | #if LEAK_DEBUG |
2d2ae547 CM |
3633 | spin_lock_irqsave(&leak_lock, flags); |
3634 | list_add(&eb->leak_list, &buffers); | |
3635 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 3636 | #endif |
d1310b2e CM |
3637 | atomic_set(&eb->refs, 1); |
3638 | ||
3639 | return eb; | |
3640 | } | |
3641 | ||
3642 | static void __free_extent_buffer(struct extent_buffer *eb) | |
3643 | { | |
3935127c | 3644 | #if LEAK_DEBUG |
2d2ae547 CM |
3645 | unsigned long flags; |
3646 | spin_lock_irqsave(&leak_lock, flags); | |
3647 | list_del(&eb->leak_list); | |
3648 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 3649 | #endif |
d1310b2e CM |
3650 | kmem_cache_free(extent_buffer_cache, eb); |
3651 | } | |
3652 | ||
897ca6e9 MX |
3653 | /* |
3654 | * Helper for releasing extent buffer page. | |
3655 | */ | |
3656 | static void btrfs_release_extent_buffer_page(struct extent_buffer *eb, | |
3657 | unsigned long start_idx) | |
3658 | { | |
3659 | unsigned long index; | |
3660 | struct page *page; | |
3661 | ||
3662 | if (!eb->first_page) | |
3663 | return; | |
3664 | ||
3665 | index = num_extent_pages(eb->start, eb->len); | |
3666 | if (start_idx >= index) | |
3667 | return; | |
3668 | ||
3669 | do { | |
3670 | index--; | |
3671 | page = extent_buffer_page(eb, index); | |
3672 | if (page) | |
3673 | page_cache_release(page); | |
3674 | } while (index != start_idx); | |
3675 | } | |
3676 | ||
3677 | /* | |
3678 | * Helper for releasing the extent buffer. | |
3679 | */ | |
3680 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
3681 | { | |
3682 | btrfs_release_extent_buffer_page(eb, 0); | |
3683 | __free_extent_buffer(eb); | |
3684 | } | |
3685 | ||
d1310b2e CM |
3686 | struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree, |
3687 | u64 start, unsigned long len, | |
ba144192 | 3688 | struct page *page0) |
d1310b2e CM |
3689 | { |
3690 | unsigned long num_pages = num_extent_pages(start, len); | |
3691 | unsigned long i; | |
3692 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
3693 | struct extent_buffer *eb; | |
6af118ce | 3694 | struct extent_buffer *exists = NULL; |
d1310b2e CM |
3695 | struct page *p; |
3696 | struct address_space *mapping = tree->mapping; | |
3697 | int uptodate = 1; | |
19fe0a8b | 3698 | int ret; |
d1310b2e | 3699 | |
19fe0a8b MX |
3700 | rcu_read_lock(); |
3701 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); | |
3702 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
3703 | rcu_read_unlock(); | |
0f9dd46c | 3704 | mark_page_accessed(eb->first_page); |
6af118ce CM |
3705 | return eb; |
3706 | } | |
19fe0a8b | 3707 | rcu_read_unlock(); |
6af118ce | 3708 | |
ba144192 | 3709 | eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS); |
2b114d1d | 3710 | if (!eb) |
d1310b2e CM |
3711 | return NULL; |
3712 | ||
d1310b2e CM |
3713 | if (page0) { |
3714 | eb->first_page = page0; | |
3715 | i = 1; | |
3716 | index++; | |
3717 | page_cache_get(page0); | |
3718 | mark_page_accessed(page0); | |
3719 | set_page_extent_mapped(page0); | |
d1310b2e | 3720 | set_page_extent_head(page0, len); |
f188591e | 3721 | uptodate = PageUptodate(page0); |
d1310b2e CM |
3722 | } else { |
3723 | i = 0; | |
3724 | } | |
3725 | for (; i < num_pages; i++, index++) { | |
a6591715 | 3726 | p = find_or_create_page(mapping, index, GFP_NOFS); |
d1310b2e CM |
3727 | if (!p) { |
3728 | WARN_ON(1); | |
6af118ce | 3729 | goto free_eb; |
d1310b2e CM |
3730 | } |
3731 | set_page_extent_mapped(p); | |
3732 | mark_page_accessed(p); | |
3733 | if (i == 0) { | |
3734 | eb->first_page = p; | |
3735 | set_page_extent_head(p, len); | |
3736 | } else { | |
3737 | set_page_private(p, EXTENT_PAGE_PRIVATE); | |
3738 | } | |
3739 | if (!PageUptodate(p)) | |
3740 | uptodate = 0; | |
eb14ab8e CM |
3741 | |
3742 | /* | |
3743 | * see below about how we avoid a nasty race with release page | |
3744 | * and why we unlock later | |
3745 | */ | |
3746 | if (i != 0) | |
3747 | unlock_page(p); | |
d1310b2e CM |
3748 | } |
3749 | if (uptodate) | |
b4ce94de | 3750 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 3751 | |
19fe0a8b MX |
3752 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); |
3753 | if (ret) | |
3754 | goto free_eb; | |
3755 | ||
6af118ce | 3756 | spin_lock(&tree->buffer_lock); |
19fe0a8b MX |
3757 | ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb); |
3758 | if (ret == -EEXIST) { | |
3759 | exists = radix_tree_lookup(&tree->buffer, | |
3760 | start >> PAGE_CACHE_SHIFT); | |
6af118ce CM |
3761 | /* add one reference for the caller */ |
3762 | atomic_inc(&exists->refs); | |
3763 | spin_unlock(&tree->buffer_lock); | |
19fe0a8b | 3764 | radix_tree_preload_end(); |
6af118ce CM |
3765 | goto free_eb; |
3766 | } | |
6af118ce CM |
3767 | /* add one reference for the tree */ |
3768 | atomic_inc(&eb->refs); | |
f044ba78 | 3769 | spin_unlock(&tree->buffer_lock); |
19fe0a8b | 3770 | radix_tree_preload_end(); |
eb14ab8e CM |
3771 | |
3772 | /* | |
3773 | * there is a race where release page may have | |
3774 | * tried to find this extent buffer in the radix | |
3775 | * but failed. It will tell the VM it is safe to | |
3776 | * reclaim the, and it will clear the page private bit. | |
3777 | * We must make sure to set the page private bit properly | |
3778 | * after the extent buffer is in the radix tree so | |
3779 | * it doesn't get lost | |
3780 | */ | |
3781 | set_page_extent_mapped(eb->first_page); | |
3782 | set_page_extent_head(eb->first_page, eb->len); | |
3783 | if (!page0) | |
3784 | unlock_page(eb->first_page); | |
d1310b2e CM |
3785 | return eb; |
3786 | ||
6af118ce | 3787 | free_eb: |
eb14ab8e CM |
3788 | if (eb->first_page && !page0) |
3789 | unlock_page(eb->first_page); | |
3790 | ||
d1310b2e | 3791 | if (!atomic_dec_and_test(&eb->refs)) |
6af118ce | 3792 | return exists; |
897ca6e9 | 3793 | btrfs_release_extent_buffer(eb); |
6af118ce | 3794 | return exists; |
d1310b2e | 3795 | } |
d1310b2e CM |
3796 | |
3797 | struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree, | |
f09d1f60 | 3798 | u64 start, unsigned long len) |
d1310b2e | 3799 | { |
d1310b2e | 3800 | struct extent_buffer *eb; |
d1310b2e | 3801 | |
19fe0a8b MX |
3802 | rcu_read_lock(); |
3803 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); | |
3804 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
3805 | rcu_read_unlock(); | |
0f9dd46c | 3806 | mark_page_accessed(eb->first_page); |
19fe0a8b MX |
3807 | return eb; |
3808 | } | |
3809 | rcu_read_unlock(); | |
0f9dd46c | 3810 | |
19fe0a8b | 3811 | return NULL; |
d1310b2e | 3812 | } |
d1310b2e CM |
3813 | |
3814 | void free_extent_buffer(struct extent_buffer *eb) | |
3815 | { | |
d1310b2e CM |
3816 | if (!eb) |
3817 | return; | |
3818 | ||
3819 | if (!atomic_dec_and_test(&eb->refs)) | |
3820 | return; | |
3821 | ||
6af118ce | 3822 | WARN_ON(1); |
d1310b2e | 3823 | } |
d1310b2e CM |
3824 | |
3825 | int clear_extent_buffer_dirty(struct extent_io_tree *tree, | |
3826 | struct extent_buffer *eb) | |
3827 | { | |
d1310b2e CM |
3828 | unsigned long i; |
3829 | unsigned long num_pages; | |
3830 | struct page *page; | |
3831 | ||
d1310b2e CM |
3832 | num_pages = num_extent_pages(eb->start, eb->len); |
3833 | ||
3834 | for (i = 0; i < num_pages; i++) { | |
3835 | page = extent_buffer_page(eb, i); | |
b9473439 | 3836 | if (!PageDirty(page)) |
d2c3f4f6 CM |
3837 | continue; |
3838 | ||
a61e6f29 | 3839 | lock_page(page); |
eb14ab8e CM |
3840 | WARN_ON(!PagePrivate(page)); |
3841 | ||
3842 | set_page_extent_mapped(page); | |
d1310b2e CM |
3843 | if (i == 0) |
3844 | set_page_extent_head(page, eb->len); | |
d1310b2e | 3845 | |
d1310b2e | 3846 | clear_page_dirty_for_io(page); |
0ee0fda0 | 3847 | spin_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
3848 | if (!PageDirty(page)) { |
3849 | radix_tree_tag_clear(&page->mapping->page_tree, | |
3850 | page_index(page), | |
3851 | PAGECACHE_TAG_DIRTY); | |
3852 | } | |
0ee0fda0 | 3853 | spin_unlock_irq(&page->mapping->tree_lock); |
bf0da8c1 | 3854 | ClearPageError(page); |
a61e6f29 | 3855 | unlock_page(page); |
d1310b2e CM |
3856 | } |
3857 | return 0; | |
3858 | } | |
d1310b2e | 3859 | |
d1310b2e CM |
3860 | int set_extent_buffer_dirty(struct extent_io_tree *tree, |
3861 | struct extent_buffer *eb) | |
3862 | { | |
3863 | unsigned long i; | |
3864 | unsigned long num_pages; | |
b9473439 | 3865 | int was_dirty = 0; |
d1310b2e | 3866 | |
b9473439 | 3867 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
d1310b2e | 3868 | num_pages = num_extent_pages(eb->start, eb->len); |
b9473439 | 3869 | for (i = 0; i < num_pages; i++) |
d1310b2e | 3870 | __set_page_dirty_nobuffers(extent_buffer_page(eb, i)); |
b9473439 | 3871 | return was_dirty; |
d1310b2e | 3872 | } |
d1310b2e | 3873 | |
19b6caf4 CM |
3874 | static int __eb_straddles_pages(u64 start, u64 len) |
3875 | { | |
3876 | if (len < PAGE_CACHE_SIZE) | |
3877 | return 1; | |
3878 | if (start & (PAGE_CACHE_SIZE - 1)) | |
3879 | return 1; | |
3880 | if ((start + len) & (PAGE_CACHE_SIZE - 1)) | |
3881 | return 1; | |
3882 | return 0; | |
3883 | } | |
3884 | ||
3885 | static int eb_straddles_pages(struct extent_buffer *eb) | |
3886 | { | |
3887 | return __eb_straddles_pages(eb->start, eb->len); | |
3888 | } | |
3889 | ||
1259ab75 | 3890 | int clear_extent_buffer_uptodate(struct extent_io_tree *tree, |
2ac55d41 JB |
3891 | struct extent_buffer *eb, |
3892 | struct extent_state **cached_state) | |
1259ab75 CM |
3893 | { |
3894 | unsigned long i; | |
3895 | struct page *page; | |
3896 | unsigned long num_pages; | |
3897 | ||
3898 | num_pages = num_extent_pages(eb->start, eb->len); | |
b4ce94de | 3899 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
1259ab75 | 3900 | |
50653190 CM |
3901 | clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1, |
3902 | cached_state, GFP_NOFS); | |
3903 | ||
1259ab75 CM |
3904 | for (i = 0; i < num_pages; i++) { |
3905 | page = extent_buffer_page(eb, i); | |
33958dc6 CM |
3906 | if (page) |
3907 | ClearPageUptodate(page); | |
1259ab75 CM |
3908 | } |
3909 | return 0; | |
3910 | } | |
3911 | ||
d1310b2e CM |
3912 | int set_extent_buffer_uptodate(struct extent_io_tree *tree, |
3913 | struct extent_buffer *eb) | |
3914 | { | |
3915 | unsigned long i; | |
3916 | struct page *page; | |
3917 | unsigned long num_pages; | |
3918 | ||
3919 | num_pages = num_extent_pages(eb->start, eb->len); | |
3920 | ||
19b6caf4 CM |
3921 | if (eb_straddles_pages(eb)) { |
3922 | set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1, | |
3923 | NULL, GFP_NOFS); | |
3924 | } | |
d1310b2e CM |
3925 | for (i = 0; i < num_pages; i++) { |
3926 | page = extent_buffer_page(eb, i); | |
3927 | if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) || | |
3928 | ((i == num_pages - 1) && | |
3929 | ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) { | |
3930 | check_page_uptodate(tree, page); | |
3931 | continue; | |
3932 | } | |
3933 | SetPageUptodate(page); | |
3934 | } | |
3935 | return 0; | |
3936 | } | |
d1310b2e | 3937 | |
ce9adaa5 CM |
3938 | int extent_range_uptodate(struct extent_io_tree *tree, |
3939 | u64 start, u64 end) | |
3940 | { | |
3941 | struct page *page; | |
3942 | int ret; | |
3943 | int pg_uptodate = 1; | |
3944 | int uptodate; | |
3945 | unsigned long index; | |
3946 | ||
19b6caf4 CM |
3947 | if (__eb_straddles_pages(start, end - start + 1)) { |
3948 | ret = test_range_bit(tree, start, end, | |
3949 | EXTENT_UPTODATE, 1, NULL); | |
3950 | if (ret) | |
3951 | return 1; | |
3952 | } | |
d397712b | 3953 | while (start <= end) { |
ce9adaa5 CM |
3954 | index = start >> PAGE_CACHE_SHIFT; |
3955 | page = find_get_page(tree->mapping, index); | |
8bedd51b MH |
3956 | if (!page) |
3957 | return 1; | |
ce9adaa5 CM |
3958 | uptodate = PageUptodate(page); |
3959 | page_cache_release(page); | |
3960 | if (!uptodate) { | |
3961 | pg_uptodate = 0; | |
3962 | break; | |
3963 | } | |
3964 | start += PAGE_CACHE_SIZE; | |
3965 | } | |
3966 | return pg_uptodate; | |
3967 | } | |
3968 | ||
d1310b2e | 3969 | int extent_buffer_uptodate(struct extent_io_tree *tree, |
2ac55d41 JB |
3970 | struct extent_buffer *eb, |
3971 | struct extent_state *cached_state) | |
d1310b2e | 3972 | { |
728131d8 | 3973 | int ret = 0; |
ce9adaa5 CM |
3974 | unsigned long num_pages; |
3975 | unsigned long i; | |
728131d8 CM |
3976 | struct page *page; |
3977 | int pg_uptodate = 1; | |
3978 | ||
b4ce94de | 3979 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
4235298e | 3980 | return 1; |
728131d8 | 3981 | |
19b6caf4 CM |
3982 | if (eb_straddles_pages(eb)) { |
3983 | ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1, | |
3984 | EXTENT_UPTODATE, 1, cached_state); | |
3985 | if (ret) | |
3986 | return ret; | |
3987 | } | |
728131d8 CM |
3988 | |
3989 | num_pages = num_extent_pages(eb->start, eb->len); | |
3990 | for (i = 0; i < num_pages; i++) { | |
3991 | page = extent_buffer_page(eb, i); | |
3992 | if (!PageUptodate(page)) { | |
3993 | pg_uptodate = 0; | |
3994 | break; | |
3995 | } | |
3996 | } | |
4235298e | 3997 | return pg_uptodate; |
d1310b2e | 3998 | } |
d1310b2e CM |
3999 | |
4000 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
bb82ab88 | 4001 | struct extent_buffer *eb, u64 start, int wait, |
f188591e | 4002 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
4003 | { |
4004 | unsigned long i; | |
4005 | unsigned long start_i; | |
4006 | struct page *page; | |
4007 | int err; | |
4008 | int ret = 0; | |
ce9adaa5 CM |
4009 | int locked_pages = 0; |
4010 | int all_uptodate = 1; | |
4011 | int inc_all_pages = 0; | |
d1310b2e | 4012 | unsigned long num_pages; |
a86c12c7 | 4013 | struct bio *bio = NULL; |
c8b97818 | 4014 | unsigned long bio_flags = 0; |
a86c12c7 | 4015 | |
b4ce94de | 4016 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
4017 | return 0; |
4018 | ||
19b6caf4 CM |
4019 | if (eb_straddles_pages(eb)) { |
4020 | if (test_range_bit(tree, eb->start, eb->start + eb->len - 1, | |
4021 | EXTENT_UPTODATE, 1, NULL)) { | |
4022 | return 0; | |
4023 | } | |
d1310b2e CM |
4024 | } |
4025 | ||
4026 | if (start) { | |
4027 | WARN_ON(start < eb->start); | |
4028 | start_i = (start >> PAGE_CACHE_SHIFT) - | |
4029 | (eb->start >> PAGE_CACHE_SHIFT); | |
4030 | } else { | |
4031 | start_i = 0; | |
4032 | } | |
4033 | ||
4034 | num_pages = num_extent_pages(eb->start, eb->len); | |
4035 | for (i = start_i; i < num_pages; i++) { | |
4036 | page = extent_buffer_page(eb, i); | |
bb82ab88 | 4037 | if (wait == WAIT_NONE) { |
2db04966 | 4038 | if (!trylock_page(page)) |
ce9adaa5 | 4039 | goto unlock_exit; |
d1310b2e CM |
4040 | } else { |
4041 | lock_page(page); | |
4042 | } | |
ce9adaa5 | 4043 | locked_pages++; |
d397712b | 4044 | if (!PageUptodate(page)) |
ce9adaa5 | 4045 | all_uptodate = 0; |
ce9adaa5 CM |
4046 | } |
4047 | if (all_uptodate) { | |
4048 | if (start_i == 0) | |
b4ce94de | 4049 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
4050 | goto unlock_exit; |
4051 | } | |
4052 | ||
4053 | for (i = start_i; i < num_pages; i++) { | |
4054 | page = extent_buffer_page(eb, i); | |
eb14ab8e CM |
4055 | |
4056 | WARN_ON(!PagePrivate(page)); | |
4057 | ||
4058 | set_page_extent_mapped(page); | |
4059 | if (i == 0) | |
4060 | set_page_extent_head(page, eb->len); | |
4061 | ||
ce9adaa5 CM |
4062 | if (inc_all_pages) |
4063 | page_cache_get(page); | |
4064 | if (!PageUptodate(page)) { | |
4065 | if (start_i == 0) | |
4066 | inc_all_pages = 1; | |
f188591e | 4067 | ClearPageError(page); |
a86c12c7 | 4068 | err = __extent_read_full_page(tree, page, |
f188591e | 4069 | get_extent, &bio, |
c8b97818 | 4070 | mirror_num, &bio_flags); |
d397712b | 4071 | if (err) |
d1310b2e | 4072 | ret = err; |
d1310b2e CM |
4073 | } else { |
4074 | unlock_page(page); | |
4075 | } | |
4076 | } | |
4077 | ||
a86c12c7 | 4078 | if (bio) |
c8b97818 | 4079 | submit_one_bio(READ, bio, mirror_num, bio_flags); |
a86c12c7 | 4080 | |
bb82ab88 | 4081 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 4082 | return ret; |
d397712b | 4083 | |
d1310b2e CM |
4084 | for (i = start_i; i < num_pages; i++) { |
4085 | page = extent_buffer_page(eb, i); | |
4086 | wait_on_page_locked(page); | |
d397712b | 4087 | if (!PageUptodate(page)) |
d1310b2e | 4088 | ret = -EIO; |
d1310b2e | 4089 | } |
d397712b | 4090 | |
d1310b2e | 4091 | if (!ret) |
b4ce94de | 4092 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 4093 | return ret; |
ce9adaa5 CM |
4094 | |
4095 | unlock_exit: | |
4096 | i = start_i; | |
d397712b | 4097 | while (locked_pages > 0) { |
ce9adaa5 CM |
4098 | page = extent_buffer_page(eb, i); |
4099 | i++; | |
4100 | unlock_page(page); | |
4101 | locked_pages--; | |
4102 | } | |
4103 | return ret; | |
d1310b2e | 4104 | } |
d1310b2e CM |
4105 | |
4106 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
4107 | unsigned long start, | |
4108 | unsigned long len) | |
4109 | { | |
4110 | size_t cur; | |
4111 | size_t offset; | |
4112 | struct page *page; | |
4113 | char *kaddr; | |
4114 | char *dst = (char *)dstv; | |
4115 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4116 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
4117 | |
4118 | WARN_ON(start > eb->len); | |
4119 | WARN_ON(start + len > eb->start + eb->len); | |
4120 | ||
4121 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4122 | ||
d397712b | 4123 | while (len > 0) { |
d1310b2e | 4124 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
4125 | |
4126 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
a6591715 | 4127 | kaddr = page_address(page); |
d1310b2e | 4128 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
4129 | |
4130 | dst += cur; | |
4131 | len -= cur; | |
4132 | offset = 0; | |
4133 | i++; | |
4134 | } | |
4135 | } | |
d1310b2e CM |
4136 | |
4137 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
a6591715 | 4138 | unsigned long min_len, char **map, |
d1310b2e | 4139 | unsigned long *map_start, |
a6591715 | 4140 | unsigned long *map_len) |
d1310b2e CM |
4141 | { |
4142 | size_t offset = start & (PAGE_CACHE_SIZE - 1); | |
4143 | char *kaddr; | |
4144 | struct page *p; | |
4145 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4146 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4147 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
4148 | PAGE_CACHE_SHIFT; | |
4149 | ||
4150 | if (i != end_i) | |
4151 | return -EINVAL; | |
4152 | ||
4153 | if (i == 0) { | |
4154 | offset = start_offset; | |
4155 | *map_start = 0; | |
4156 | } else { | |
4157 | offset = 0; | |
4158 | *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; | |
4159 | } | |
d397712b | 4160 | |
d1310b2e | 4161 | if (start + min_len > eb->len) { |
d397712b CM |
4162 | printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, " |
4163 | "wanted %lu %lu\n", (unsigned long long)eb->start, | |
4164 | eb->len, start, min_len); | |
d1310b2e | 4165 | WARN_ON(1); |
85026533 | 4166 | return -EINVAL; |
d1310b2e CM |
4167 | } |
4168 | ||
4169 | p = extent_buffer_page(eb, i); | |
a6591715 | 4170 | kaddr = page_address(p); |
d1310b2e CM |
4171 | *map = kaddr + offset; |
4172 | *map_len = PAGE_CACHE_SIZE - offset; | |
4173 | return 0; | |
4174 | } | |
d1310b2e | 4175 | |
d1310b2e CM |
4176 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, |
4177 | unsigned long start, | |
4178 | unsigned long len) | |
4179 | { | |
4180 | size_t cur; | |
4181 | size_t offset; | |
4182 | struct page *page; | |
4183 | char *kaddr; | |
4184 | char *ptr = (char *)ptrv; | |
4185 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4186 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4187 | int ret = 0; | |
4188 | ||
4189 | WARN_ON(start > eb->len); | |
4190 | WARN_ON(start + len > eb->start + eb->len); | |
4191 | ||
4192 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4193 | ||
d397712b | 4194 | while (len > 0) { |
d1310b2e | 4195 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
4196 | |
4197 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
4198 | ||
a6591715 | 4199 | kaddr = page_address(page); |
d1310b2e | 4200 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
4201 | if (ret) |
4202 | break; | |
4203 | ||
4204 | ptr += cur; | |
4205 | len -= cur; | |
4206 | offset = 0; | |
4207 | i++; | |
4208 | } | |
4209 | return ret; | |
4210 | } | |
d1310b2e CM |
4211 | |
4212 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
4213 | unsigned long start, unsigned long len) | |
4214 | { | |
4215 | size_t cur; | |
4216 | size_t offset; | |
4217 | struct page *page; | |
4218 | char *kaddr; | |
4219 | char *src = (char *)srcv; | |
4220 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4221 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4222 | ||
4223 | WARN_ON(start > eb->len); | |
4224 | WARN_ON(start + len > eb->start + eb->len); | |
4225 | ||
4226 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4227 | ||
d397712b | 4228 | while (len > 0) { |
d1310b2e CM |
4229 | page = extent_buffer_page(eb, i); |
4230 | WARN_ON(!PageUptodate(page)); | |
4231 | ||
4232 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 4233 | kaddr = page_address(page); |
d1310b2e | 4234 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
4235 | |
4236 | src += cur; | |
4237 | len -= cur; | |
4238 | offset = 0; | |
4239 | i++; | |
4240 | } | |
4241 | } | |
d1310b2e CM |
4242 | |
4243 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
4244 | unsigned long start, unsigned long len) | |
4245 | { | |
4246 | size_t cur; | |
4247 | size_t offset; | |
4248 | struct page *page; | |
4249 | char *kaddr; | |
4250 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4251 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4252 | ||
4253 | WARN_ON(start > eb->len); | |
4254 | WARN_ON(start + len > eb->start + eb->len); | |
4255 | ||
4256 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4257 | ||
d397712b | 4258 | while (len > 0) { |
d1310b2e CM |
4259 | page = extent_buffer_page(eb, i); |
4260 | WARN_ON(!PageUptodate(page)); | |
4261 | ||
4262 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 4263 | kaddr = page_address(page); |
d1310b2e | 4264 | memset(kaddr + offset, c, cur); |
d1310b2e CM |
4265 | |
4266 | len -= cur; | |
4267 | offset = 0; | |
4268 | i++; | |
4269 | } | |
4270 | } | |
d1310b2e CM |
4271 | |
4272 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
4273 | unsigned long dst_offset, unsigned long src_offset, | |
4274 | unsigned long len) | |
4275 | { | |
4276 | u64 dst_len = dst->len; | |
4277 | size_t cur; | |
4278 | size_t offset; | |
4279 | struct page *page; | |
4280 | char *kaddr; | |
4281 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4282 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
4283 | ||
4284 | WARN_ON(src->len != dst_len); | |
4285 | ||
4286 | offset = (start_offset + dst_offset) & | |
4287 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4288 | ||
d397712b | 4289 | while (len > 0) { |
d1310b2e CM |
4290 | page = extent_buffer_page(dst, i); |
4291 | WARN_ON(!PageUptodate(page)); | |
4292 | ||
4293 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
4294 | ||
a6591715 | 4295 | kaddr = page_address(page); |
d1310b2e | 4296 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
4297 | |
4298 | src_offset += cur; | |
4299 | len -= cur; | |
4300 | offset = 0; | |
4301 | i++; | |
4302 | } | |
4303 | } | |
d1310b2e CM |
4304 | |
4305 | static void move_pages(struct page *dst_page, struct page *src_page, | |
4306 | unsigned long dst_off, unsigned long src_off, | |
4307 | unsigned long len) | |
4308 | { | |
a6591715 | 4309 | char *dst_kaddr = page_address(dst_page); |
d1310b2e CM |
4310 | if (dst_page == src_page) { |
4311 | memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len); | |
4312 | } else { | |
a6591715 | 4313 | char *src_kaddr = page_address(src_page); |
d1310b2e CM |
4314 | char *p = dst_kaddr + dst_off + len; |
4315 | char *s = src_kaddr + src_off + len; | |
4316 | ||
4317 | while (len--) | |
4318 | *--p = *--s; | |
d1310b2e | 4319 | } |
d1310b2e CM |
4320 | } |
4321 | ||
3387206f ST |
4322 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
4323 | { | |
4324 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
4325 | return distance < len; | |
4326 | } | |
4327 | ||
d1310b2e CM |
4328 | static void copy_pages(struct page *dst_page, struct page *src_page, |
4329 | unsigned long dst_off, unsigned long src_off, | |
4330 | unsigned long len) | |
4331 | { | |
a6591715 | 4332 | char *dst_kaddr = page_address(dst_page); |
d1310b2e CM |
4333 | char *src_kaddr; |
4334 | ||
3387206f | 4335 | if (dst_page != src_page) { |
a6591715 | 4336 | src_kaddr = page_address(src_page); |
3387206f | 4337 | } else { |
d1310b2e | 4338 | src_kaddr = dst_kaddr; |
3387206f ST |
4339 | BUG_ON(areas_overlap(src_off, dst_off, len)); |
4340 | } | |
d1310b2e CM |
4341 | |
4342 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
4343 | } |
4344 | ||
4345 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
4346 | unsigned long src_offset, unsigned long len) | |
4347 | { | |
4348 | size_t cur; | |
4349 | size_t dst_off_in_page; | |
4350 | size_t src_off_in_page; | |
4351 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4352 | unsigned long dst_i; | |
4353 | unsigned long src_i; | |
4354 | ||
4355 | if (src_offset + len > dst->len) { | |
d397712b CM |
4356 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
4357 | "len %lu dst len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
4358 | BUG_ON(1); |
4359 | } | |
4360 | if (dst_offset + len > dst->len) { | |
d397712b CM |
4361 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
4362 | "len %lu dst len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
4363 | BUG_ON(1); |
4364 | } | |
4365 | ||
d397712b | 4366 | while (len > 0) { |
d1310b2e CM |
4367 | dst_off_in_page = (start_offset + dst_offset) & |
4368 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4369 | src_off_in_page = (start_offset + src_offset) & | |
4370 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4371 | ||
4372 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
4373 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
4374 | ||
4375 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
4376 | src_off_in_page)); | |
4377 | cur = min_t(unsigned long, cur, | |
4378 | (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); | |
4379 | ||
4380 | copy_pages(extent_buffer_page(dst, dst_i), | |
4381 | extent_buffer_page(dst, src_i), | |
4382 | dst_off_in_page, src_off_in_page, cur); | |
4383 | ||
4384 | src_offset += cur; | |
4385 | dst_offset += cur; | |
4386 | len -= cur; | |
4387 | } | |
4388 | } | |
d1310b2e CM |
4389 | |
4390 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
4391 | unsigned long src_offset, unsigned long len) | |
4392 | { | |
4393 | size_t cur; | |
4394 | size_t dst_off_in_page; | |
4395 | size_t src_off_in_page; | |
4396 | unsigned long dst_end = dst_offset + len - 1; | |
4397 | unsigned long src_end = src_offset + len - 1; | |
4398 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4399 | unsigned long dst_i; | |
4400 | unsigned long src_i; | |
4401 | ||
4402 | if (src_offset + len > dst->len) { | |
d397712b CM |
4403 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
4404 | "len %lu len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
4405 | BUG_ON(1); |
4406 | } | |
4407 | if (dst_offset + len > dst->len) { | |
d397712b CM |
4408 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
4409 | "len %lu len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
4410 | BUG_ON(1); |
4411 | } | |
3387206f | 4412 | if (!areas_overlap(src_offset, dst_offset, len)) { |
d1310b2e CM |
4413 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
4414 | return; | |
4415 | } | |
d397712b | 4416 | while (len > 0) { |
d1310b2e CM |
4417 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; |
4418 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
4419 | ||
4420 | dst_off_in_page = (start_offset + dst_end) & | |
4421 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4422 | src_off_in_page = (start_offset + src_end) & | |
4423 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4424 | ||
4425 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
4426 | cur = min(cur, dst_off_in_page + 1); | |
4427 | move_pages(extent_buffer_page(dst, dst_i), | |
4428 | extent_buffer_page(dst, src_i), | |
4429 | dst_off_in_page - cur + 1, | |
4430 | src_off_in_page - cur + 1, cur); | |
4431 | ||
4432 | dst_end -= cur; | |
4433 | src_end -= cur; | |
4434 | len -= cur; | |
4435 | } | |
4436 | } | |
6af118ce | 4437 | |
19fe0a8b MX |
4438 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
4439 | { | |
4440 | struct extent_buffer *eb = | |
4441 | container_of(head, struct extent_buffer, rcu_head); | |
4442 | ||
4443 | btrfs_release_extent_buffer(eb); | |
4444 | } | |
4445 | ||
6af118ce CM |
4446 | int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page) |
4447 | { | |
4448 | u64 start = page_offset(page); | |
4449 | struct extent_buffer *eb; | |
4450 | int ret = 1; | |
6af118ce CM |
4451 | |
4452 | spin_lock(&tree->buffer_lock); | |
19fe0a8b | 4453 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); |
45f49bce CM |
4454 | if (!eb) { |
4455 | spin_unlock(&tree->buffer_lock); | |
4456 | return ret; | |
4457 | } | |
6af118ce | 4458 | |
19fe0a8b | 4459 | if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
6af118ce CM |
4460 | ret = 0; |
4461 | goto out; | |
4462 | } | |
19fe0a8b MX |
4463 | |
4464 | /* | |
4465 | * set @eb->refs to 0 if it is already 1, and then release the @eb. | |
4466 | * Or go back. | |
4467 | */ | |
4468 | if (atomic_cmpxchg(&eb->refs, 1, 0) != 1) { | |
b9473439 CM |
4469 | ret = 0; |
4470 | goto out; | |
4471 | } | |
897ca6e9 | 4472 | |
19fe0a8b | 4473 | radix_tree_delete(&tree->buffer, start >> PAGE_CACHE_SHIFT); |
6af118ce CM |
4474 | out: |
4475 | spin_unlock(&tree->buffer_lock); | |
19fe0a8b MX |
4476 | |
4477 | /* at this point we can safely release the extent buffer */ | |
4478 | if (atomic_read(&eb->refs) == 0) | |
4479 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); | |
6af118ce CM |
4480 | return ret; |
4481 | } |