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 | ||
56 | int __init extent_io_init(void) | |
57 | { | |
9601e3f6 CH |
58 | extent_state_cache = kmem_cache_create("extent_state", |
59 | sizeof(struct extent_state), 0, | |
60 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
61 | if (!extent_state_cache) |
62 | return -ENOMEM; | |
63 | ||
9601e3f6 CH |
64 | extent_buffer_cache = kmem_cache_create("extent_buffers", |
65 | sizeof(struct extent_buffer), 0, | |
66 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
67 | if (!extent_buffer_cache) |
68 | goto free_state_cache; | |
69 | return 0; | |
70 | ||
71 | free_state_cache: | |
72 | kmem_cache_destroy(extent_state_cache); | |
73 | return -ENOMEM; | |
74 | } | |
75 | ||
76 | void extent_io_exit(void) | |
77 | { | |
78 | struct extent_state *state; | |
2d2ae547 | 79 | struct extent_buffer *eb; |
d1310b2e CM |
80 | |
81 | while (!list_empty(&states)) { | |
2d2ae547 | 82 | state = list_entry(states.next, struct extent_state, leak_list); |
d397712b CM |
83 | printk(KERN_ERR "btrfs state leak: start %llu end %llu " |
84 | "state %lu in tree %p refs %d\n", | |
85 | (unsigned long long)state->start, | |
86 | (unsigned long long)state->end, | |
87 | state->state, state->tree, atomic_read(&state->refs)); | |
2d2ae547 | 88 | list_del(&state->leak_list); |
d1310b2e CM |
89 | kmem_cache_free(extent_state_cache, state); |
90 | ||
91 | } | |
92 | ||
2d2ae547 CM |
93 | while (!list_empty(&buffers)) { |
94 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
d397712b CM |
95 | printk(KERN_ERR "btrfs buffer leak start %llu len %lu " |
96 | "refs %d\n", (unsigned long long)eb->start, | |
97 | eb->len, atomic_read(&eb->refs)); | |
2d2ae547 CM |
98 | list_del(&eb->leak_list); |
99 | kmem_cache_free(extent_buffer_cache, eb); | |
100 | } | |
d1310b2e CM |
101 | if (extent_state_cache) |
102 | kmem_cache_destroy(extent_state_cache); | |
103 | if (extent_buffer_cache) | |
104 | kmem_cache_destroy(extent_buffer_cache); | |
105 | } | |
106 | ||
107 | void extent_io_tree_init(struct extent_io_tree *tree, | |
f993c883 | 108 | struct address_space *mapping) |
d1310b2e | 109 | { |
6bef4d31 | 110 | tree->state = RB_ROOT; |
19fe0a8b | 111 | INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC); |
d1310b2e CM |
112 | tree->ops = NULL; |
113 | tree->dirty_bytes = 0; | |
70dec807 | 114 | spin_lock_init(&tree->lock); |
6af118ce | 115 | spin_lock_init(&tree->buffer_lock); |
d1310b2e | 116 | tree->mapping = mapping; |
d1310b2e | 117 | } |
d1310b2e | 118 | |
b2950863 | 119 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
120 | { |
121 | struct extent_state *state; | |
3935127c | 122 | #if LEAK_DEBUG |
2d2ae547 | 123 | unsigned long flags; |
4bef0848 | 124 | #endif |
d1310b2e CM |
125 | |
126 | state = kmem_cache_alloc(extent_state_cache, mask); | |
2b114d1d | 127 | if (!state) |
d1310b2e CM |
128 | return state; |
129 | state->state = 0; | |
d1310b2e | 130 | state->private = 0; |
70dec807 | 131 | state->tree = NULL; |
3935127c | 132 | #if LEAK_DEBUG |
2d2ae547 CM |
133 | spin_lock_irqsave(&leak_lock, flags); |
134 | list_add(&state->leak_list, &states); | |
135 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 136 | #endif |
d1310b2e CM |
137 | atomic_set(&state->refs, 1); |
138 | init_waitqueue_head(&state->wq); | |
139 | return state; | |
140 | } | |
d1310b2e | 141 | |
4845e44f | 142 | void free_extent_state(struct extent_state *state) |
d1310b2e | 143 | { |
d1310b2e CM |
144 | if (!state) |
145 | return; | |
146 | if (atomic_dec_and_test(&state->refs)) { | |
3935127c | 147 | #if LEAK_DEBUG |
2d2ae547 | 148 | unsigned long flags; |
4bef0848 | 149 | #endif |
70dec807 | 150 | WARN_ON(state->tree); |
3935127c | 151 | #if LEAK_DEBUG |
2d2ae547 CM |
152 | spin_lock_irqsave(&leak_lock, flags); |
153 | list_del(&state->leak_list); | |
154 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 155 | #endif |
d1310b2e CM |
156 | kmem_cache_free(extent_state_cache, state); |
157 | } | |
158 | } | |
d1310b2e CM |
159 | |
160 | static struct rb_node *tree_insert(struct rb_root *root, u64 offset, | |
161 | struct rb_node *node) | |
162 | { | |
d397712b CM |
163 | struct rb_node **p = &root->rb_node; |
164 | struct rb_node *parent = NULL; | |
d1310b2e CM |
165 | struct tree_entry *entry; |
166 | ||
d397712b | 167 | while (*p) { |
d1310b2e CM |
168 | parent = *p; |
169 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
170 | ||
171 | if (offset < entry->start) | |
172 | p = &(*p)->rb_left; | |
173 | else if (offset > entry->end) | |
174 | p = &(*p)->rb_right; | |
175 | else | |
176 | return parent; | |
177 | } | |
178 | ||
179 | entry = rb_entry(node, struct tree_entry, rb_node); | |
d1310b2e CM |
180 | rb_link_node(node, parent, p); |
181 | rb_insert_color(node, root); | |
182 | return NULL; | |
183 | } | |
184 | ||
80ea96b1 | 185 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
d1310b2e CM |
186 | struct rb_node **prev_ret, |
187 | struct rb_node **next_ret) | |
188 | { | |
80ea96b1 | 189 | struct rb_root *root = &tree->state; |
d397712b | 190 | struct rb_node *n = root->rb_node; |
d1310b2e CM |
191 | struct rb_node *prev = NULL; |
192 | struct rb_node *orig_prev = NULL; | |
193 | struct tree_entry *entry; | |
194 | struct tree_entry *prev_entry = NULL; | |
195 | ||
d397712b | 196 | while (n) { |
d1310b2e CM |
197 | entry = rb_entry(n, struct tree_entry, rb_node); |
198 | prev = n; | |
199 | prev_entry = entry; | |
200 | ||
201 | if (offset < entry->start) | |
202 | n = n->rb_left; | |
203 | else if (offset > entry->end) | |
204 | n = n->rb_right; | |
d397712b | 205 | else |
d1310b2e CM |
206 | return n; |
207 | } | |
208 | ||
209 | if (prev_ret) { | |
210 | orig_prev = prev; | |
d397712b | 211 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
212 | prev = rb_next(prev); |
213 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
214 | } | |
215 | *prev_ret = prev; | |
216 | prev = orig_prev; | |
217 | } | |
218 | ||
219 | if (next_ret) { | |
220 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 221 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
222 | prev = rb_prev(prev); |
223 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
224 | } | |
225 | *next_ret = prev; | |
226 | } | |
227 | return NULL; | |
228 | } | |
229 | ||
80ea96b1 CM |
230 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
231 | u64 offset) | |
d1310b2e | 232 | { |
70dec807 | 233 | struct rb_node *prev = NULL; |
d1310b2e | 234 | struct rb_node *ret; |
70dec807 | 235 | |
80ea96b1 | 236 | ret = __etree_search(tree, offset, &prev, NULL); |
d397712b | 237 | if (!ret) |
d1310b2e CM |
238 | return prev; |
239 | return ret; | |
240 | } | |
241 | ||
9ed74f2d JB |
242 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, |
243 | struct extent_state *other) | |
244 | { | |
245 | if (tree->ops && tree->ops->merge_extent_hook) | |
246 | tree->ops->merge_extent_hook(tree->mapping->host, new, | |
247 | other); | |
248 | } | |
249 | ||
d1310b2e CM |
250 | /* |
251 | * utility function to look for merge candidates inside a given range. | |
252 | * Any extents with matching state are merged together into a single | |
253 | * extent in the tree. Extents with EXTENT_IO in their state field | |
254 | * are not merged because the end_io handlers need to be able to do | |
255 | * operations on them without sleeping (or doing allocations/splits). | |
256 | * | |
257 | * This should be called with the tree lock held. | |
258 | */ | |
1bf85046 JM |
259 | static void merge_state(struct extent_io_tree *tree, |
260 | struct extent_state *state) | |
d1310b2e CM |
261 | { |
262 | struct extent_state *other; | |
263 | struct rb_node *other_node; | |
264 | ||
5b21f2ed | 265 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
1bf85046 | 266 | return; |
d1310b2e CM |
267 | |
268 | other_node = rb_prev(&state->rb_node); | |
269 | if (other_node) { | |
270 | other = rb_entry(other_node, struct extent_state, rb_node); | |
271 | if (other->end == state->start - 1 && | |
272 | other->state == state->state) { | |
9ed74f2d | 273 | merge_cb(tree, state, other); |
d1310b2e | 274 | state->start = other->start; |
70dec807 | 275 | other->tree = NULL; |
d1310b2e CM |
276 | rb_erase(&other->rb_node, &tree->state); |
277 | free_extent_state(other); | |
278 | } | |
279 | } | |
280 | other_node = rb_next(&state->rb_node); | |
281 | if (other_node) { | |
282 | other = rb_entry(other_node, struct extent_state, rb_node); | |
283 | if (other->start == state->end + 1 && | |
284 | other->state == state->state) { | |
9ed74f2d | 285 | merge_cb(tree, state, other); |
df98b6e2 JB |
286 | state->end = other->end; |
287 | other->tree = NULL; | |
288 | rb_erase(&other->rb_node, &tree->state); | |
289 | free_extent_state(other); | |
d1310b2e CM |
290 | } |
291 | } | |
d1310b2e CM |
292 | } |
293 | ||
1bf85046 | 294 | static void set_state_cb(struct extent_io_tree *tree, |
0ca1f7ce | 295 | struct extent_state *state, int *bits) |
291d673e | 296 | { |
1bf85046 JM |
297 | if (tree->ops && tree->ops->set_bit_hook) |
298 | tree->ops->set_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
299 | } |
300 | ||
301 | static void clear_state_cb(struct extent_io_tree *tree, | |
0ca1f7ce | 302 | struct extent_state *state, int *bits) |
291d673e | 303 | { |
9ed74f2d JB |
304 | if (tree->ops && tree->ops->clear_bit_hook) |
305 | tree->ops->clear_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
306 | } |
307 | ||
3150b699 XG |
308 | static void set_state_bits(struct extent_io_tree *tree, |
309 | struct extent_state *state, int *bits); | |
310 | ||
d1310b2e CM |
311 | /* |
312 | * insert an extent_state struct into the tree. 'bits' are set on the | |
313 | * struct before it is inserted. | |
314 | * | |
315 | * This may return -EEXIST if the extent is already there, in which case the | |
316 | * state struct is freed. | |
317 | * | |
318 | * The tree lock is not taken internally. This is a utility function and | |
319 | * probably isn't what you want to call (see set/clear_extent_bit). | |
320 | */ | |
321 | static int insert_state(struct extent_io_tree *tree, | |
322 | struct extent_state *state, u64 start, u64 end, | |
0ca1f7ce | 323 | int *bits) |
d1310b2e CM |
324 | { |
325 | struct rb_node *node; | |
326 | ||
327 | if (end < start) { | |
d397712b CM |
328 | printk(KERN_ERR "btrfs end < start %llu %llu\n", |
329 | (unsigned long long)end, | |
330 | (unsigned long long)start); | |
d1310b2e CM |
331 | WARN_ON(1); |
332 | } | |
d1310b2e CM |
333 | state->start = start; |
334 | state->end = end; | |
9ed74f2d | 335 | |
3150b699 XG |
336 | set_state_bits(tree, state, bits); |
337 | ||
d1310b2e CM |
338 | node = tree_insert(&tree->state, end, &state->rb_node); |
339 | if (node) { | |
340 | struct extent_state *found; | |
341 | found = rb_entry(node, struct extent_state, rb_node); | |
d397712b CM |
342 | printk(KERN_ERR "btrfs found node %llu %llu on insert of " |
343 | "%llu %llu\n", (unsigned long long)found->start, | |
344 | (unsigned long long)found->end, | |
345 | (unsigned long long)start, (unsigned long long)end); | |
d1310b2e CM |
346 | return -EEXIST; |
347 | } | |
70dec807 | 348 | state->tree = tree; |
d1310b2e CM |
349 | merge_state(tree, state); |
350 | return 0; | |
351 | } | |
352 | ||
1bf85046 | 353 | static void split_cb(struct extent_io_tree *tree, struct extent_state *orig, |
9ed74f2d JB |
354 | u64 split) |
355 | { | |
356 | if (tree->ops && tree->ops->split_extent_hook) | |
1bf85046 | 357 | tree->ops->split_extent_hook(tree->mapping->host, orig, split); |
9ed74f2d JB |
358 | } |
359 | ||
d1310b2e CM |
360 | /* |
361 | * split a given extent state struct in two, inserting the preallocated | |
362 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
363 | * offset inside 'orig' where it should be split. | |
364 | * | |
365 | * Before calling, | |
366 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
367 | * are two extent state structs in the tree: | |
368 | * prealloc: [orig->start, split - 1] | |
369 | * orig: [ split, orig->end ] | |
370 | * | |
371 | * The tree locks are not taken by this function. They need to be held | |
372 | * by the caller. | |
373 | */ | |
374 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
375 | struct extent_state *prealloc, u64 split) | |
376 | { | |
377 | struct rb_node *node; | |
9ed74f2d JB |
378 | |
379 | split_cb(tree, orig, split); | |
380 | ||
d1310b2e CM |
381 | prealloc->start = orig->start; |
382 | prealloc->end = split - 1; | |
383 | prealloc->state = orig->state; | |
384 | orig->start = split; | |
385 | ||
386 | node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node); | |
387 | if (node) { | |
d1310b2e CM |
388 | free_extent_state(prealloc); |
389 | return -EEXIST; | |
390 | } | |
70dec807 | 391 | prealloc->tree = tree; |
d1310b2e CM |
392 | return 0; |
393 | } | |
394 | ||
395 | /* | |
396 | * utility function to clear some bits in an extent state struct. | |
397 | * it will optionally wake up any one waiting on this state (wake == 1), or | |
398 | * forcibly remove the state from the tree (delete == 1). | |
399 | * | |
400 | * If no bits are set on the state struct after clearing things, the | |
401 | * struct is freed and removed from the tree | |
402 | */ | |
403 | static int clear_state_bit(struct extent_io_tree *tree, | |
0ca1f7ce YZ |
404 | struct extent_state *state, |
405 | int *bits, int wake) | |
d1310b2e | 406 | { |
0ca1f7ce | 407 | int bits_to_clear = *bits & ~EXTENT_CTLBITS; |
32c00aff | 408 | int ret = state->state & bits_to_clear; |
d1310b2e | 409 | |
0ca1f7ce | 410 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
411 | u64 range = state->end - state->start + 1; |
412 | WARN_ON(range > tree->dirty_bytes); | |
413 | tree->dirty_bytes -= range; | |
414 | } | |
291d673e | 415 | clear_state_cb(tree, state, bits); |
32c00aff | 416 | state->state &= ~bits_to_clear; |
d1310b2e CM |
417 | if (wake) |
418 | wake_up(&state->wq); | |
0ca1f7ce | 419 | if (state->state == 0) { |
70dec807 | 420 | if (state->tree) { |
d1310b2e | 421 | rb_erase(&state->rb_node, &tree->state); |
70dec807 | 422 | state->tree = NULL; |
d1310b2e CM |
423 | free_extent_state(state); |
424 | } else { | |
425 | WARN_ON(1); | |
426 | } | |
427 | } else { | |
428 | merge_state(tree, state); | |
429 | } | |
430 | return ret; | |
431 | } | |
432 | ||
8233767a XG |
433 | static struct extent_state * |
434 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
435 | { | |
436 | if (!prealloc) | |
437 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
438 | ||
439 | return prealloc; | |
440 | } | |
441 | ||
d1310b2e CM |
442 | /* |
443 | * clear some bits on a range in the tree. This may require splitting | |
444 | * or inserting elements in the tree, so the gfp mask is used to | |
445 | * indicate which allocations or sleeping are allowed. | |
446 | * | |
447 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
448 | * the given range from the tree regardless of state (ie for truncate). | |
449 | * | |
450 | * the range [start, end] is inclusive. | |
451 | * | |
452 | * This takes the tree lock, and returns < 0 on error, > 0 if any of the | |
453 | * bits were already set, or zero if none of the bits were already set. | |
454 | */ | |
455 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
2c64c53d CM |
456 | int bits, int wake, int delete, |
457 | struct extent_state **cached_state, | |
458 | gfp_t mask) | |
d1310b2e CM |
459 | { |
460 | struct extent_state *state; | |
2c64c53d | 461 | struct extent_state *cached; |
d1310b2e | 462 | struct extent_state *prealloc = NULL; |
2c64c53d | 463 | struct rb_node *next_node; |
d1310b2e | 464 | struct rb_node *node; |
5c939df5 | 465 | u64 last_end; |
d1310b2e CM |
466 | int err; |
467 | int set = 0; | |
2ac55d41 | 468 | int clear = 0; |
d1310b2e | 469 | |
0ca1f7ce YZ |
470 | if (delete) |
471 | bits |= ~EXTENT_CTLBITS; | |
472 | bits |= EXTENT_FIRST_DELALLOC; | |
473 | ||
2ac55d41 JB |
474 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
475 | clear = 1; | |
d1310b2e CM |
476 | again: |
477 | if (!prealloc && (mask & __GFP_WAIT)) { | |
478 | prealloc = alloc_extent_state(mask); | |
479 | if (!prealloc) | |
480 | return -ENOMEM; | |
481 | } | |
482 | ||
cad321ad | 483 | spin_lock(&tree->lock); |
2c64c53d CM |
484 | if (cached_state) { |
485 | cached = *cached_state; | |
2ac55d41 JB |
486 | |
487 | if (clear) { | |
488 | *cached_state = NULL; | |
489 | cached_state = NULL; | |
490 | } | |
491 | ||
df98b6e2 JB |
492 | if (cached && cached->tree && cached->start <= start && |
493 | cached->end > start) { | |
2ac55d41 JB |
494 | if (clear) |
495 | atomic_dec(&cached->refs); | |
2c64c53d | 496 | state = cached; |
42daec29 | 497 | goto hit_next; |
2c64c53d | 498 | } |
2ac55d41 JB |
499 | if (clear) |
500 | free_extent_state(cached); | |
2c64c53d | 501 | } |
d1310b2e CM |
502 | /* |
503 | * this search will find the extents that end after | |
504 | * our range starts | |
505 | */ | |
80ea96b1 | 506 | node = tree_search(tree, start); |
d1310b2e CM |
507 | if (!node) |
508 | goto out; | |
509 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 510 | hit_next: |
d1310b2e CM |
511 | if (state->start > end) |
512 | goto out; | |
513 | WARN_ON(state->end < start); | |
5c939df5 | 514 | last_end = state->end; |
d1310b2e CM |
515 | |
516 | /* | |
517 | * | ---- desired range ---- | | |
518 | * | state | or | |
519 | * | ------------- state -------------- | | |
520 | * | |
521 | * We need to split the extent we found, and may flip | |
522 | * bits on second half. | |
523 | * | |
524 | * If the extent we found extends past our range, we | |
525 | * just split and search again. It'll get split again | |
526 | * the next time though. | |
527 | * | |
528 | * If the extent we found is inside our range, we clear | |
529 | * the desired bit on it. | |
530 | */ | |
531 | ||
532 | if (state->start < start) { | |
8233767a XG |
533 | prealloc = alloc_extent_state_atomic(prealloc); |
534 | BUG_ON(!prealloc); | |
d1310b2e CM |
535 | err = split_state(tree, state, prealloc, start); |
536 | BUG_ON(err == -EEXIST); | |
537 | prealloc = NULL; | |
538 | if (err) | |
539 | goto out; | |
540 | if (state->end <= end) { | |
0ca1f7ce | 541 | set |= clear_state_bit(tree, state, &bits, wake); |
5c939df5 YZ |
542 | if (last_end == (u64)-1) |
543 | goto out; | |
544 | start = last_end + 1; | |
d1310b2e CM |
545 | } |
546 | goto search_again; | |
547 | } | |
548 | /* | |
549 | * | ---- desired range ---- | | |
550 | * | state | | |
551 | * We need to split the extent, and clear the bit | |
552 | * on the first half | |
553 | */ | |
554 | if (state->start <= end && state->end > end) { | |
8233767a XG |
555 | prealloc = alloc_extent_state_atomic(prealloc); |
556 | BUG_ON(!prealloc); | |
d1310b2e CM |
557 | err = split_state(tree, state, prealloc, end + 1); |
558 | BUG_ON(err == -EEXIST); | |
d1310b2e CM |
559 | if (wake) |
560 | wake_up(&state->wq); | |
42daec29 | 561 | |
0ca1f7ce | 562 | set |= clear_state_bit(tree, prealloc, &bits, wake); |
9ed74f2d | 563 | |
d1310b2e CM |
564 | prealloc = NULL; |
565 | goto out; | |
566 | } | |
42daec29 | 567 | |
2c64c53d CM |
568 | if (state->end < end && prealloc && !need_resched()) |
569 | next_node = rb_next(&state->rb_node); | |
570 | else | |
571 | next_node = NULL; | |
42daec29 | 572 | |
0ca1f7ce | 573 | set |= clear_state_bit(tree, state, &bits, wake); |
5c939df5 YZ |
574 | if (last_end == (u64)-1) |
575 | goto out; | |
576 | start = last_end + 1; | |
2c64c53d CM |
577 | if (start <= end && next_node) { |
578 | state = rb_entry(next_node, struct extent_state, | |
579 | rb_node); | |
580 | if (state->start == start) | |
581 | goto hit_next; | |
582 | } | |
d1310b2e CM |
583 | goto search_again; |
584 | ||
585 | out: | |
cad321ad | 586 | spin_unlock(&tree->lock); |
d1310b2e CM |
587 | if (prealloc) |
588 | free_extent_state(prealloc); | |
589 | ||
590 | return set; | |
591 | ||
592 | search_again: | |
593 | if (start > end) | |
594 | goto out; | |
cad321ad | 595 | spin_unlock(&tree->lock); |
d1310b2e CM |
596 | if (mask & __GFP_WAIT) |
597 | cond_resched(); | |
598 | goto again; | |
599 | } | |
d1310b2e CM |
600 | |
601 | static int wait_on_state(struct extent_io_tree *tree, | |
602 | struct extent_state *state) | |
641f5219 CH |
603 | __releases(tree->lock) |
604 | __acquires(tree->lock) | |
d1310b2e CM |
605 | { |
606 | DEFINE_WAIT(wait); | |
607 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 608 | spin_unlock(&tree->lock); |
d1310b2e | 609 | schedule(); |
cad321ad | 610 | spin_lock(&tree->lock); |
d1310b2e CM |
611 | finish_wait(&state->wq, &wait); |
612 | return 0; | |
613 | } | |
614 | ||
615 | /* | |
616 | * waits for one or more bits to clear on a range in the state tree. | |
617 | * The range [start, end] is inclusive. | |
618 | * The tree lock is taken by this function | |
619 | */ | |
620 | int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits) | |
621 | { | |
622 | struct extent_state *state; | |
623 | struct rb_node *node; | |
624 | ||
cad321ad | 625 | spin_lock(&tree->lock); |
d1310b2e CM |
626 | again: |
627 | while (1) { | |
628 | /* | |
629 | * this search will find all the extents that end after | |
630 | * our range starts | |
631 | */ | |
80ea96b1 | 632 | node = tree_search(tree, start); |
d1310b2e CM |
633 | if (!node) |
634 | break; | |
635 | ||
636 | state = rb_entry(node, struct extent_state, rb_node); | |
637 | ||
638 | if (state->start > end) | |
639 | goto out; | |
640 | ||
641 | if (state->state & bits) { | |
642 | start = state->start; | |
643 | atomic_inc(&state->refs); | |
644 | wait_on_state(tree, state); | |
645 | free_extent_state(state); | |
646 | goto again; | |
647 | } | |
648 | start = state->end + 1; | |
649 | ||
650 | if (start > end) | |
651 | break; | |
652 | ||
ded91f08 | 653 | cond_resched_lock(&tree->lock); |
d1310b2e CM |
654 | } |
655 | out: | |
cad321ad | 656 | spin_unlock(&tree->lock); |
d1310b2e CM |
657 | return 0; |
658 | } | |
d1310b2e | 659 | |
1bf85046 | 660 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 661 | struct extent_state *state, |
0ca1f7ce | 662 | int *bits) |
d1310b2e | 663 | { |
0ca1f7ce | 664 | int bits_to_set = *bits & ~EXTENT_CTLBITS; |
9ed74f2d | 665 | |
1bf85046 | 666 | set_state_cb(tree, state, bits); |
0ca1f7ce | 667 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
668 | u64 range = state->end - state->start + 1; |
669 | tree->dirty_bytes += range; | |
670 | } | |
0ca1f7ce | 671 | state->state |= bits_to_set; |
d1310b2e CM |
672 | } |
673 | ||
2c64c53d CM |
674 | static void cache_state(struct extent_state *state, |
675 | struct extent_state **cached_ptr) | |
676 | { | |
677 | if (cached_ptr && !(*cached_ptr)) { | |
678 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) { | |
679 | *cached_ptr = state; | |
680 | atomic_inc(&state->refs); | |
681 | } | |
682 | } | |
683 | } | |
684 | ||
507903b8 AJ |
685 | static void uncache_state(struct extent_state **cached_ptr) |
686 | { | |
687 | if (cached_ptr && (*cached_ptr)) { | |
688 | struct extent_state *state = *cached_ptr; | |
109b36a2 CM |
689 | *cached_ptr = NULL; |
690 | free_extent_state(state); | |
507903b8 AJ |
691 | } |
692 | } | |
693 | ||
d1310b2e | 694 | /* |
1edbb734 CM |
695 | * set some bits on a range in the tree. This may require allocations or |
696 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 697 | * |
1edbb734 CM |
698 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
699 | * part of the range already has the desired bits set. The start of the | |
700 | * existing range is returned in failed_start in this case. | |
d1310b2e | 701 | * |
1edbb734 | 702 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 703 | */ |
1edbb734 | 704 | |
4845e44f CM |
705 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
706 | int bits, int exclusive_bits, u64 *failed_start, | |
707 | struct extent_state **cached_state, gfp_t mask) | |
d1310b2e CM |
708 | { |
709 | struct extent_state *state; | |
710 | struct extent_state *prealloc = NULL; | |
711 | struct rb_node *node; | |
d1310b2e | 712 | int err = 0; |
d1310b2e CM |
713 | u64 last_start; |
714 | u64 last_end; | |
42daec29 | 715 | |
0ca1f7ce | 716 | bits |= EXTENT_FIRST_DELALLOC; |
d1310b2e CM |
717 | again: |
718 | if (!prealloc && (mask & __GFP_WAIT)) { | |
719 | prealloc = alloc_extent_state(mask); | |
8233767a | 720 | BUG_ON(!prealloc); |
d1310b2e CM |
721 | } |
722 | ||
cad321ad | 723 | spin_lock(&tree->lock); |
9655d298 CM |
724 | if (cached_state && *cached_state) { |
725 | state = *cached_state; | |
df98b6e2 JB |
726 | if (state->start <= start && state->end > start && |
727 | state->tree) { | |
9655d298 CM |
728 | node = &state->rb_node; |
729 | goto hit_next; | |
730 | } | |
731 | } | |
d1310b2e CM |
732 | /* |
733 | * this search will find all the extents that end after | |
734 | * our range starts. | |
735 | */ | |
80ea96b1 | 736 | node = tree_search(tree, start); |
d1310b2e | 737 | if (!node) { |
8233767a XG |
738 | prealloc = alloc_extent_state_atomic(prealloc); |
739 | BUG_ON(!prealloc); | |
0ca1f7ce | 740 | err = insert_state(tree, prealloc, start, end, &bits); |
d1310b2e CM |
741 | prealloc = NULL; |
742 | BUG_ON(err == -EEXIST); | |
743 | goto out; | |
744 | } | |
d1310b2e | 745 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 746 | hit_next: |
d1310b2e CM |
747 | last_start = state->start; |
748 | last_end = state->end; | |
749 | ||
750 | /* | |
751 | * | ---- desired range ---- | | |
752 | * | state | | |
753 | * | |
754 | * Just lock what we found and keep going | |
755 | */ | |
756 | if (state->start == start && state->end <= end) { | |
40431d6c | 757 | struct rb_node *next_node; |
1edbb734 | 758 | if (state->state & exclusive_bits) { |
d1310b2e CM |
759 | *failed_start = state->start; |
760 | err = -EEXIST; | |
761 | goto out; | |
762 | } | |
42daec29 | 763 | |
1bf85046 | 764 | set_state_bits(tree, state, &bits); |
9ed74f2d | 765 | |
2c64c53d | 766 | cache_state(state, cached_state); |
d1310b2e | 767 | merge_state(tree, state); |
5c939df5 YZ |
768 | if (last_end == (u64)-1) |
769 | goto out; | |
40431d6c | 770 | |
5c939df5 | 771 | start = last_end + 1; |
df98b6e2 | 772 | next_node = rb_next(&state->rb_node); |
c7f895a2 XG |
773 | if (next_node && start < end && prealloc && !need_resched()) { |
774 | state = rb_entry(next_node, struct extent_state, | |
775 | rb_node); | |
776 | if (state->start == start) | |
777 | goto hit_next; | |
40431d6c | 778 | } |
d1310b2e CM |
779 | goto search_again; |
780 | } | |
781 | ||
782 | /* | |
783 | * | ---- desired range ---- | | |
784 | * | state | | |
785 | * or | |
786 | * | ------------- state -------------- | | |
787 | * | |
788 | * We need to split the extent we found, and may flip bits on | |
789 | * second half. | |
790 | * | |
791 | * If the extent we found extends past our | |
792 | * range, we just split and search again. It'll get split | |
793 | * again the next time though. | |
794 | * | |
795 | * If the extent we found is inside our range, we set the | |
796 | * desired bit on it. | |
797 | */ | |
798 | if (state->start < start) { | |
1edbb734 | 799 | if (state->state & exclusive_bits) { |
d1310b2e CM |
800 | *failed_start = start; |
801 | err = -EEXIST; | |
802 | goto out; | |
803 | } | |
8233767a XG |
804 | |
805 | prealloc = alloc_extent_state_atomic(prealloc); | |
806 | BUG_ON(!prealloc); | |
d1310b2e CM |
807 | err = split_state(tree, state, prealloc, start); |
808 | BUG_ON(err == -EEXIST); | |
809 | prealloc = NULL; | |
810 | if (err) | |
811 | goto out; | |
812 | if (state->end <= end) { | |
1bf85046 | 813 | set_state_bits(tree, state, &bits); |
2c64c53d | 814 | cache_state(state, cached_state); |
d1310b2e | 815 | merge_state(tree, state); |
5c939df5 YZ |
816 | if (last_end == (u64)-1) |
817 | goto out; | |
818 | start = last_end + 1; | |
d1310b2e CM |
819 | } |
820 | goto search_again; | |
821 | } | |
822 | /* | |
823 | * | ---- desired range ---- | | |
824 | * | state | or | state | | |
825 | * | |
826 | * There's a hole, we need to insert something in it and | |
827 | * ignore the extent we found. | |
828 | */ | |
829 | if (state->start > start) { | |
830 | u64 this_end; | |
831 | if (end < last_start) | |
832 | this_end = end; | |
833 | else | |
d397712b | 834 | this_end = last_start - 1; |
8233767a XG |
835 | |
836 | prealloc = alloc_extent_state_atomic(prealloc); | |
837 | BUG_ON(!prealloc); | |
c7f895a2 XG |
838 | |
839 | /* | |
840 | * Avoid to free 'prealloc' if it can be merged with | |
841 | * the later extent. | |
842 | */ | |
d1310b2e | 843 | err = insert_state(tree, prealloc, start, this_end, |
0ca1f7ce | 844 | &bits); |
d1310b2e | 845 | BUG_ON(err == -EEXIST); |
9ed74f2d | 846 | if (err) { |
c7f895a2 | 847 | free_extent_state(prealloc); |
9ed74f2d | 848 | prealloc = NULL; |
d1310b2e | 849 | goto out; |
9ed74f2d JB |
850 | } |
851 | cache_state(prealloc, cached_state); | |
852 | prealloc = NULL; | |
d1310b2e CM |
853 | start = this_end + 1; |
854 | goto search_again; | |
855 | } | |
856 | /* | |
857 | * | ---- desired range ---- | | |
858 | * | state | | |
859 | * We need to split the extent, and set the bit | |
860 | * on the first half | |
861 | */ | |
862 | if (state->start <= end && state->end > end) { | |
1edbb734 | 863 | if (state->state & exclusive_bits) { |
d1310b2e CM |
864 | *failed_start = start; |
865 | err = -EEXIST; | |
866 | goto out; | |
867 | } | |
8233767a XG |
868 | |
869 | prealloc = alloc_extent_state_atomic(prealloc); | |
870 | BUG_ON(!prealloc); | |
d1310b2e CM |
871 | err = split_state(tree, state, prealloc, end + 1); |
872 | BUG_ON(err == -EEXIST); | |
873 | ||
1bf85046 | 874 | set_state_bits(tree, prealloc, &bits); |
2c64c53d | 875 | cache_state(prealloc, cached_state); |
d1310b2e CM |
876 | merge_state(tree, prealloc); |
877 | prealloc = NULL; | |
878 | goto out; | |
879 | } | |
880 | ||
881 | goto search_again; | |
882 | ||
883 | out: | |
cad321ad | 884 | spin_unlock(&tree->lock); |
d1310b2e CM |
885 | if (prealloc) |
886 | free_extent_state(prealloc); | |
887 | ||
888 | return err; | |
889 | ||
890 | search_again: | |
891 | if (start > end) | |
892 | goto out; | |
cad321ad | 893 | spin_unlock(&tree->lock); |
d1310b2e CM |
894 | if (mask & __GFP_WAIT) |
895 | cond_resched(); | |
896 | goto again; | |
897 | } | |
d1310b2e | 898 | |
462d6fac JB |
899 | /** |
900 | * convert_extent - convert all bits in a given range from one bit to another | |
901 | * @tree: the io tree to search | |
902 | * @start: the start offset in bytes | |
903 | * @end: the end offset in bytes (inclusive) | |
904 | * @bits: the bits to set in this range | |
905 | * @clear_bits: the bits to clear in this range | |
906 | * @mask: the allocation mask | |
907 | * | |
908 | * This will go through and set bits for the given range. If any states exist | |
909 | * already in this range they are set with the given bit and cleared of the | |
910 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
911 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
912 | * boundary bits like LOCK. | |
913 | */ | |
914 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
915 | int bits, int clear_bits, gfp_t mask) | |
916 | { | |
917 | struct extent_state *state; | |
918 | struct extent_state *prealloc = NULL; | |
919 | struct rb_node *node; | |
920 | int err = 0; | |
921 | u64 last_start; | |
922 | u64 last_end; | |
923 | ||
924 | again: | |
925 | if (!prealloc && (mask & __GFP_WAIT)) { | |
926 | prealloc = alloc_extent_state(mask); | |
927 | if (!prealloc) | |
928 | return -ENOMEM; | |
929 | } | |
930 | ||
931 | spin_lock(&tree->lock); | |
932 | /* | |
933 | * this search will find all the extents that end after | |
934 | * our range starts. | |
935 | */ | |
936 | node = tree_search(tree, start); | |
937 | if (!node) { | |
938 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
939 | if (!prealloc) { |
940 | err = -ENOMEM; | |
941 | goto out; | |
942 | } | |
462d6fac JB |
943 | err = insert_state(tree, prealloc, start, end, &bits); |
944 | prealloc = NULL; | |
945 | BUG_ON(err == -EEXIST); | |
946 | goto out; | |
947 | } | |
948 | state = rb_entry(node, struct extent_state, rb_node); | |
949 | hit_next: | |
950 | last_start = state->start; | |
951 | last_end = state->end; | |
952 | ||
953 | /* | |
954 | * | ---- desired range ---- | | |
955 | * | state | | |
956 | * | |
957 | * Just lock what we found and keep going | |
958 | */ | |
959 | if (state->start == start && state->end <= end) { | |
960 | struct rb_node *next_node; | |
961 | ||
962 | set_state_bits(tree, state, &bits); | |
963 | clear_state_bit(tree, state, &clear_bits, 0); | |
964 | ||
965 | merge_state(tree, state); | |
966 | if (last_end == (u64)-1) | |
967 | goto out; | |
968 | ||
969 | start = last_end + 1; | |
970 | next_node = rb_next(&state->rb_node); | |
971 | if (next_node && start < end && prealloc && !need_resched()) { | |
972 | state = rb_entry(next_node, struct extent_state, | |
973 | rb_node); | |
974 | if (state->start == start) | |
975 | goto hit_next; | |
976 | } | |
977 | goto search_again; | |
978 | } | |
979 | ||
980 | /* | |
981 | * | ---- desired range ---- | | |
982 | * | state | | |
983 | * or | |
984 | * | ------------- state -------------- | | |
985 | * | |
986 | * We need to split the extent we found, and may flip bits on | |
987 | * second half. | |
988 | * | |
989 | * If the extent we found extends past our | |
990 | * range, we just split and search again. It'll get split | |
991 | * again the next time though. | |
992 | * | |
993 | * If the extent we found is inside our range, we set the | |
994 | * desired bit on it. | |
995 | */ | |
996 | if (state->start < start) { | |
997 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
998 | if (!prealloc) { |
999 | err = -ENOMEM; | |
1000 | goto out; | |
1001 | } | |
462d6fac JB |
1002 | err = split_state(tree, state, prealloc, start); |
1003 | BUG_ON(err == -EEXIST); | |
1004 | prealloc = NULL; | |
1005 | if (err) | |
1006 | goto out; | |
1007 | if (state->end <= end) { | |
1008 | set_state_bits(tree, state, &bits); | |
1009 | clear_state_bit(tree, state, &clear_bits, 0); | |
1010 | merge_state(tree, state); | |
1011 | if (last_end == (u64)-1) | |
1012 | goto out; | |
1013 | start = last_end + 1; | |
1014 | } | |
1015 | goto search_again; | |
1016 | } | |
1017 | /* | |
1018 | * | ---- desired range ---- | | |
1019 | * | state | or | state | | |
1020 | * | |
1021 | * There's a hole, we need to insert something in it and | |
1022 | * ignore the extent we found. | |
1023 | */ | |
1024 | if (state->start > start) { | |
1025 | u64 this_end; | |
1026 | if (end < last_start) | |
1027 | this_end = end; | |
1028 | else | |
1029 | this_end = last_start - 1; | |
1030 | ||
1031 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1032 | if (!prealloc) { |
1033 | err = -ENOMEM; | |
1034 | goto out; | |
1035 | } | |
462d6fac JB |
1036 | |
1037 | /* | |
1038 | * Avoid to free 'prealloc' if it can be merged with | |
1039 | * the later extent. | |
1040 | */ | |
1041 | err = insert_state(tree, prealloc, start, this_end, | |
1042 | &bits); | |
1043 | BUG_ON(err == -EEXIST); | |
1044 | if (err) { | |
1045 | free_extent_state(prealloc); | |
1046 | prealloc = NULL; | |
1047 | goto out; | |
1048 | } | |
1049 | prealloc = NULL; | |
1050 | start = this_end + 1; | |
1051 | goto search_again; | |
1052 | } | |
1053 | /* | |
1054 | * | ---- desired range ---- | | |
1055 | * | state | | |
1056 | * We need to split the extent, and set the bit | |
1057 | * on the first half | |
1058 | */ | |
1059 | if (state->start <= end && state->end > end) { | |
1060 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1061 | if (!prealloc) { |
1062 | err = -ENOMEM; | |
1063 | goto out; | |
1064 | } | |
462d6fac JB |
1065 | |
1066 | err = split_state(tree, state, prealloc, end + 1); | |
1067 | BUG_ON(err == -EEXIST); | |
1068 | ||
1069 | set_state_bits(tree, prealloc, &bits); | |
1070 | clear_state_bit(tree, prealloc, &clear_bits, 0); | |
1071 | ||
1072 | merge_state(tree, prealloc); | |
1073 | prealloc = NULL; | |
1074 | goto out; | |
1075 | } | |
1076 | ||
1077 | goto search_again; | |
1078 | ||
1079 | out: | |
1080 | spin_unlock(&tree->lock); | |
1081 | if (prealloc) | |
1082 | free_extent_state(prealloc); | |
1083 | ||
1084 | return err; | |
1085 | ||
1086 | search_again: | |
1087 | if (start > end) | |
1088 | goto out; | |
1089 | spin_unlock(&tree->lock); | |
1090 | if (mask & __GFP_WAIT) | |
1091 | cond_resched(); | |
1092 | goto again; | |
1093 | } | |
1094 | ||
d1310b2e CM |
1095 | /* wrappers around set/clear extent bit */ |
1096 | int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
1097 | gfp_t mask) | |
1098 | { | |
1099 | return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL, | |
2c64c53d | 1100 | NULL, mask); |
d1310b2e | 1101 | } |
d1310b2e CM |
1102 | |
1103 | int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
1104 | int bits, gfp_t mask) | |
1105 | { | |
1106 | return set_extent_bit(tree, start, end, bits, 0, NULL, | |
2c64c53d | 1107 | NULL, mask); |
d1310b2e | 1108 | } |
d1310b2e CM |
1109 | |
1110 | int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
1111 | int bits, gfp_t mask) | |
1112 | { | |
2c64c53d | 1113 | return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask); |
d1310b2e | 1114 | } |
d1310b2e CM |
1115 | |
1116 | int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, | |
2ac55d41 | 1117 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
1118 | { |
1119 | return set_extent_bit(tree, start, end, | |
fee187d9 | 1120 | EXTENT_DELALLOC | EXTENT_UPTODATE, |
2ac55d41 | 1121 | 0, NULL, cached_state, mask); |
d1310b2e | 1122 | } |
d1310b2e CM |
1123 | |
1124 | int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
1125 | gfp_t mask) | |
1126 | { | |
1127 | return clear_extent_bit(tree, start, end, | |
32c00aff | 1128 | EXTENT_DIRTY | EXTENT_DELALLOC | |
0ca1f7ce | 1129 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask); |
d1310b2e | 1130 | } |
d1310b2e CM |
1131 | |
1132 | int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, | |
1133 | gfp_t mask) | |
1134 | { | |
1135 | return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL, | |
2c64c53d | 1136 | NULL, mask); |
d1310b2e | 1137 | } |
d1310b2e | 1138 | |
d1310b2e | 1139 | int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
507903b8 | 1140 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e | 1141 | { |
507903b8 AJ |
1142 | return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, |
1143 | NULL, cached_state, mask); | |
d1310b2e | 1144 | } |
d1310b2e | 1145 | |
d397712b | 1146 | static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, |
2ac55d41 JB |
1147 | u64 end, struct extent_state **cached_state, |
1148 | gfp_t mask) | |
d1310b2e | 1149 | { |
2c64c53d | 1150 | return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, |
2ac55d41 | 1151 | cached_state, mask); |
d1310b2e | 1152 | } |
d1310b2e | 1153 | |
d352ac68 CM |
1154 | /* |
1155 | * either insert or lock state struct between start and end use mask to tell | |
1156 | * us if waiting is desired. | |
1157 | */ | |
1edbb734 | 1158 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
2c64c53d | 1159 | int bits, struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
1160 | { |
1161 | int err; | |
1162 | u64 failed_start; | |
1163 | while (1) { | |
1edbb734 | 1164 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits, |
2c64c53d CM |
1165 | EXTENT_LOCKED, &failed_start, |
1166 | cached_state, mask); | |
d1310b2e CM |
1167 | if (err == -EEXIST && (mask & __GFP_WAIT)) { |
1168 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); | |
1169 | start = failed_start; | |
1170 | } else { | |
1171 | break; | |
1172 | } | |
1173 | WARN_ON(start > end); | |
1174 | } | |
1175 | return err; | |
1176 | } | |
d1310b2e | 1177 | |
1edbb734 CM |
1178 | int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) |
1179 | { | |
2c64c53d | 1180 | return lock_extent_bits(tree, start, end, 0, NULL, mask); |
1edbb734 CM |
1181 | } |
1182 | ||
25179201 JB |
1183 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end, |
1184 | gfp_t mask) | |
1185 | { | |
1186 | int err; | |
1187 | u64 failed_start; | |
1188 | ||
2c64c53d CM |
1189 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
1190 | &failed_start, NULL, mask); | |
6643558d YZ |
1191 | if (err == -EEXIST) { |
1192 | if (failed_start > start) | |
1193 | clear_extent_bit(tree, start, failed_start - 1, | |
2c64c53d | 1194 | EXTENT_LOCKED, 1, 0, NULL, mask); |
25179201 | 1195 | return 0; |
6643558d | 1196 | } |
25179201 JB |
1197 | return 1; |
1198 | } | |
25179201 | 1199 | |
2c64c53d CM |
1200 | int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end, |
1201 | struct extent_state **cached, gfp_t mask) | |
1202 | { | |
1203 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, | |
1204 | mask); | |
1205 | } | |
1206 | ||
507903b8 | 1207 | int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) |
d1310b2e | 1208 | { |
2c64c53d CM |
1209 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL, |
1210 | mask); | |
d1310b2e | 1211 | } |
d1310b2e | 1212 | |
d1310b2e CM |
1213 | /* |
1214 | * helper function to set both pages and extents in the tree writeback | |
1215 | */ | |
b2950863 | 1216 | static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e CM |
1217 | { |
1218 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1219 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1220 | struct page *page; | |
1221 | ||
1222 | while (index <= end_index) { | |
1223 | page = find_get_page(tree->mapping, index); | |
1224 | BUG_ON(!page); | |
1225 | set_page_writeback(page); | |
1226 | page_cache_release(page); | |
1227 | index++; | |
1228 | } | |
d1310b2e CM |
1229 | return 0; |
1230 | } | |
d1310b2e | 1231 | |
d352ac68 CM |
1232 | /* find the first state struct with 'bits' set after 'start', and |
1233 | * return it. tree->lock must be held. NULL will returned if | |
1234 | * nothing was found after 'start' | |
1235 | */ | |
d7fc640e CM |
1236 | struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree, |
1237 | u64 start, int bits) | |
1238 | { | |
1239 | struct rb_node *node; | |
1240 | struct extent_state *state; | |
1241 | ||
1242 | /* | |
1243 | * this search will find all the extents that end after | |
1244 | * our range starts. | |
1245 | */ | |
1246 | node = tree_search(tree, start); | |
d397712b | 1247 | if (!node) |
d7fc640e | 1248 | goto out; |
d7fc640e | 1249 | |
d397712b | 1250 | while (1) { |
d7fc640e | 1251 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1252 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1253 | return state; |
d397712b | 1254 | |
d7fc640e CM |
1255 | node = rb_next(node); |
1256 | if (!node) | |
1257 | break; | |
1258 | } | |
1259 | out: | |
1260 | return NULL; | |
1261 | } | |
d7fc640e | 1262 | |
69261c4b XG |
1263 | /* |
1264 | * find the first offset in the io tree with 'bits' set. zero is | |
1265 | * returned if we find something, and *start_ret and *end_ret are | |
1266 | * set to reflect the state struct that was found. | |
1267 | * | |
1268 | * If nothing was found, 1 is returned, < 0 on error | |
1269 | */ | |
1270 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
1271 | u64 *start_ret, u64 *end_ret, int bits) | |
1272 | { | |
1273 | struct extent_state *state; | |
1274 | int ret = 1; | |
1275 | ||
1276 | spin_lock(&tree->lock); | |
1277 | state = find_first_extent_bit_state(tree, start, bits); | |
1278 | if (state) { | |
1279 | *start_ret = state->start; | |
1280 | *end_ret = state->end; | |
1281 | ret = 0; | |
1282 | } | |
1283 | spin_unlock(&tree->lock); | |
1284 | return ret; | |
1285 | } | |
1286 | ||
d352ac68 CM |
1287 | /* |
1288 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1289 | * more than 'max_bytes'. start and end are used to return the range, | |
1290 | * | |
1291 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1292 | */ | |
c8b97818 | 1293 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1294 | u64 *start, u64 *end, u64 max_bytes, |
1295 | struct extent_state **cached_state) | |
d1310b2e CM |
1296 | { |
1297 | struct rb_node *node; | |
1298 | struct extent_state *state; | |
1299 | u64 cur_start = *start; | |
1300 | u64 found = 0; | |
1301 | u64 total_bytes = 0; | |
1302 | ||
cad321ad | 1303 | spin_lock(&tree->lock); |
c8b97818 | 1304 | |
d1310b2e CM |
1305 | /* |
1306 | * this search will find all the extents that end after | |
1307 | * our range starts. | |
1308 | */ | |
80ea96b1 | 1309 | node = tree_search(tree, cur_start); |
2b114d1d | 1310 | if (!node) { |
3b951516 CM |
1311 | if (!found) |
1312 | *end = (u64)-1; | |
d1310b2e CM |
1313 | goto out; |
1314 | } | |
1315 | ||
d397712b | 1316 | while (1) { |
d1310b2e | 1317 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1318 | if (found && (state->start != cur_start || |
1319 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1320 | goto out; |
1321 | } | |
1322 | if (!(state->state & EXTENT_DELALLOC)) { | |
1323 | if (!found) | |
1324 | *end = state->end; | |
1325 | goto out; | |
1326 | } | |
c2a128d2 | 1327 | if (!found) { |
d1310b2e | 1328 | *start = state->start; |
c2a128d2 JB |
1329 | *cached_state = state; |
1330 | atomic_inc(&state->refs); | |
1331 | } | |
d1310b2e CM |
1332 | found++; |
1333 | *end = state->end; | |
1334 | cur_start = state->end + 1; | |
1335 | node = rb_next(node); | |
1336 | if (!node) | |
1337 | break; | |
1338 | total_bytes += state->end - state->start + 1; | |
1339 | if (total_bytes >= max_bytes) | |
1340 | break; | |
1341 | } | |
1342 | out: | |
cad321ad | 1343 | spin_unlock(&tree->lock); |
d1310b2e CM |
1344 | return found; |
1345 | } | |
1346 | ||
c8b97818 CM |
1347 | static noinline int __unlock_for_delalloc(struct inode *inode, |
1348 | struct page *locked_page, | |
1349 | u64 start, u64 end) | |
1350 | { | |
1351 | int ret; | |
1352 | struct page *pages[16]; | |
1353 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1354 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1355 | unsigned long nr_pages = end_index - index + 1; | |
1356 | int i; | |
1357 | ||
1358 | if (index == locked_page->index && end_index == index) | |
1359 | return 0; | |
1360 | ||
d397712b | 1361 | while (nr_pages > 0) { |
c8b97818 | 1362 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1363 | min_t(unsigned long, nr_pages, |
1364 | ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1365 | for (i = 0; i < ret; i++) { |
1366 | if (pages[i] != locked_page) | |
1367 | unlock_page(pages[i]); | |
1368 | page_cache_release(pages[i]); | |
1369 | } | |
1370 | nr_pages -= ret; | |
1371 | index += ret; | |
1372 | cond_resched(); | |
1373 | } | |
1374 | return 0; | |
1375 | } | |
1376 | ||
1377 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1378 | struct page *locked_page, | |
1379 | u64 delalloc_start, | |
1380 | u64 delalloc_end) | |
1381 | { | |
1382 | unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT; | |
1383 | unsigned long start_index = index; | |
1384 | unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT; | |
1385 | unsigned long pages_locked = 0; | |
1386 | struct page *pages[16]; | |
1387 | unsigned long nrpages; | |
1388 | int ret; | |
1389 | int i; | |
1390 | ||
1391 | /* the caller is responsible for locking the start index */ | |
1392 | if (index == locked_page->index && index == end_index) | |
1393 | return 0; | |
1394 | ||
1395 | /* skip the page at the start index */ | |
1396 | nrpages = end_index - index + 1; | |
d397712b | 1397 | while (nrpages > 0) { |
c8b97818 | 1398 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1399 | min_t(unsigned long, |
1400 | nrpages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1401 | if (ret == 0) { |
1402 | ret = -EAGAIN; | |
1403 | goto done; | |
1404 | } | |
1405 | /* now we have an array of pages, lock them all */ | |
1406 | for (i = 0; i < ret; i++) { | |
1407 | /* | |
1408 | * the caller is taking responsibility for | |
1409 | * locked_page | |
1410 | */ | |
771ed689 | 1411 | if (pages[i] != locked_page) { |
c8b97818 | 1412 | lock_page(pages[i]); |
f2b1c41c CM |
1413 | if (!PageDirty(pages[i]) || |
1414 | pages[i]->mapping != inode->i_mapping) { | |
771ed689 CM |
1415 | ret = -EAGAIN; |
1416 | unlock_page(pages[i]); | |
1417 | page_cache_release(pages[i]); | |
1418 | goto done; | |
1419 | } | |
1420 | } | |
c8b97818 | 1421 | page_cache_release(pages[i]); |
771ed689 | 1422 | pages_locked++; |
c8b97818 | 1423 | } |
c8b97818 CM |
1424 | nrpages -= ret; |
1425 | index += ret; | |
1426 | cond_resched(); | |
1427 | } | |
1428 | ret = 0; | |
1429 | done: | |
1430 | if (ret && pages_locked) { | |
1431 | __unlock_for_delalloc(inode, locked_page, | |
1432 | delalloc_start, | |
1433 | ((u64)(start_index + pages_locked - 1)) << | |
1434 | PAGE_CACHE_SHIFT); | |
1435 | } | |
1436 | return ret; | |
1437 | } | |
1438 | ||
1439 | /* | |
1440 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1441 | * more than 'max_bytes'. start and end are used to return the range, | |
1442 | * | |
1443 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1444 | */ | |
1445 | static noinline u64 find_lock_delalloc_range(struct inode *inode, | |
1446 | struct extent_io_tree *tree, | |
1447 | struct page *locked_page, | |
1448 | u64 *start, u64 *end, | |
1449 | u64 max_bytes) | |
1450 | { | |
1451 | u64 delalloc_start; | |
1452 | u64 delalloc_end; | |
1453 | u64 found; | |
9655d298 | 1454 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1455 | int ret; |
1456 | int loops = 0; | |
1457 | ||
1458 | again: | |
1459 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1460 | delalloc_start = *start; | |
1461 | delalloc_end = 0; | |
1462 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1463 | max_bytes, &cached_state); |
70b99e69 | 1464 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1465 | *start = delalloc_start; |
1466 | *end = delalloc_end; | |
c2a128d2 | 1467 | free_extent_state(cached_state); |
c8b97818 CM |
1468 | return found; |
1469 | } | |
1470 | ||
70b99e69 CM |
1471 | /* |
1472 | * start comes from the offset of locked_page. We have to lock | |
1473 | * pages in order, so we can't process delalloc bytes before | |
1474 | * locked_page | |
1475 | */ | |
d397712b | 1476 | if (delalloc_start < *start) |
70b99e69 | 1477 | delalloc_start = *start; |
70b99e69 | 1478 | |
c8b97818 CM |
1479 | /* |
1480 | * make sure to limit the number of pages we try to lock down | |
1481 | * if we're looping. | |
1482 | */ | |
d397712b | 1483 | if (delalloc_end + 1 - delalloc_start > max_bytes && loops) |
771ed689 | 1484 | delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1; |
d397712b | 1485 | |
c8b97818 CM |
1486 | /* step two, lock all the pages after the page that has start */ |
1487 | ret = lock_delalloc_pages(inode, locked_page, | |
1488 | delalloc_start, delalloc_end); | |
1489 | if (ret == -EAGAIN) { | |
1490 | /* some of the pages are gone, lets avoid looping by | |
1491 | * shortening the size of the delalloc range we're searching | |
1492 | */ | |
9655d298 | 1493 | free_extent_state(cached_state); |
c8b97818 CM |
1494 | if (!loops) { |
1495 | unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1); | |
1496 | max_bytes = PAGE_CACHE_SIZE - offset; | |
1497 | loops = 1; | |
1498 | goto again; | |
1499 | } else { | |
1500 | found = 0; | |
1501 | goto out_failed; | |
1502 | } | |
1503 | } | |
1504 | BUG_ON(ret); | |
1505 | ||
1506 | /* step three, lock the state bits for the whole range */ | |
9655d298 CM |
1507 | lock_extent_bits(tree, delalloc_start, delalloc_end, |
1508 | 0, &cached_state, GFP_NOFS); | |
c8b97818 CM |
1509 | |
1510 | /* then test to make sure it is all still delalloc */ | |
1511 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1512 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1513 | if (!ret) { |
9655d298 CM |
1514 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
1515 | &cached_state, GFP_NOFS); | |
c8b97818 CM |
1516 | __unlock_for_delalloc(inode, locked_page, |
1517 | delalloc_start, delalloc_end); | |
1518 | cond_resched(); | |
1519 | goto again; | |
1520 | } | |
9655d298 | 1521 | free_extent_state(cached_state); |
c8b97818 CM |
1522 | *start = delalloc_start; |
1523 | *end = delalloc_end; | |
1524 | out_failed: | |
1525 | return found; | |
1526 | } | |
1527 | ||
1528 | int extent_clear_unlock_delalloc(struct inode *inode, | |
1529 | struct extent_io_tree *tree, | |
1530 | u64 start, u64 end, struct page *locked_page, | |
a791e35e | 1531 | unsigned long op) |
c8b97818 CM |
1532 | { |
1533 | int ret; | |
1534 | struct page *pages[16]; | |
1535 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1536 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1537 | unsigned long nr_pages = end_index - index + 1; | |
1538 | int i; | |
771ed689 | 1539 | int clear_bits = 0; |
c8b97818 | 1540 | |
a791e35e | 1541 | if (op & EXTENT_CLEAR_UNLOCK) |
771ed689 | 1542 | clear_bits |= EXTENT_LOCKED; |
a791e35e | 1543 | if (op & EXTENT_CLEAR_DIRTY) |
c8b97818 CM |
1544 | clear_bits |= EXTENT_DIRTY; |
1545 | ||
a791e35e | 1546 | if (op & EXTENT_CLEAR_DELALLOC) |
771ed689 CM |
1547 | clear_bits |= EXTENT_DELALLOC; |
1548 | ||
2c64c53d | 1549 | clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); |
32c00aff JB |
1550 | if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | |
1551 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK | | |
1552 | EXTENT_SET_PRIVATE2))) | |
771ed689 | 1553 | return 0; |
c8b97818 | 1554 | |
d397712b | 1555 | while (nr_pages > 0) { |
c8b97818 | 1556 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1557 | min_t(unsigned long, |
1558 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 | 1559 | for (i = 0; i < ret; i++) { |
8b62b72b | 1560 | |
a791e35e | 1561 | if (op & EXTENT_SET_PRIVATE2) |
8b62b72b CM |
1562 | SetPagePrivate2(pages[i]); |
1563 | ||
c8b97818 CM |
1564 | if (pages[i] == locked_page) { |
1565 | page_cache_release(pages[i]); | |
1566 | continue; | |
1567 | } | |
a791e35e | 1568 | if (op & EXTENT_CLEAR_DIRTY) |
c8b97818 | 1569 | clear_page_dirty_for_io(pages[i]); |
a791e35e | 1570 | if (op & EXTENT_SET_WRITEBACK) |
c8b97818 | 1571 | set_page_writeback(pages[i]); |
a791e35e | 1572 | if (op & EXTENT_END_WRITEBACK) |
c8b97818 | 1573 | end_page_writeback(pages[i]); |
a791e35e | 1574 | if (op & EXTENT_CLEAR_UNLOCK_PAGE) |
771ed689 | 1575 | unlock_page(pages[i]); |
c8b97818 CM |
1576 | page_cache_release(pages[i]); |
1577 | } | |
1578 | nr_pages -= ret; | |
1579 | index += ret; | |
1580 | cond_resched(); | |
1581 | } | |
1582 | return 0; | |
1583 | } | |
c8b97818 | 1584 | |
d352ac68 CM |
1585 | /* |
1586 | * count the number of bytes in the tree that have a given bit(s) | |
1587 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1588 | * cached. The total number found is returned. | |
1589 | */ | |
d1310b2e CM |
1590 | u64 count_range_bits(struct extent_io_tree *tree, |
1591 | u64 *start, u64 search_end, u64 max_bytes, | |
ec29ed5b | 1592 | unsigned long bits, int contig) |
d1310b2e CM |
1593 | { |
1594 | struct rb_node *node; | |
1595 | struct extent_state *state; | |
1596 | u64 cur_start = *start; | |
1597 | u64 total_bytes = 0; | |
ec29ed5b | 1598 | u64 last = 0; |
d1310b2e CM |
1599 | int found = 0; |
1600 | ||
1601 | if (search_end <= cur_start) { | |
d1310b2e CM |
1602 | WARN_ON(1); |
1603 | return 0; | |
1604 | } | |
1605 | ||
cad321ad | 1606 | spin_lock(&tree->lock); |
d1310b2e CM |
1607 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1608 | total_bytes = tree->dirty_bytes; | |
1609 | goto out; | |
1610 | } | |
1611 | /* | |
1612 | * this search will find all the extents that end after | |
1613 | * our range starts. | |
1614 | */ | |
80ea96b1 | 1615 | node = tree_search(tree, cur_start); |
d397712b | 1616 | if (!node) |
d1310b2e | 1617 | goto out; |
d1310b2e | 1618 | |
d397712b | 1619 | while (1) { |
d1310b2e CM |
1620 | state = rb_entry(node, struct extent_state, rb_node); |
1621 | if (state->start > search_end) | |
1622 | break; | |
ec29ed5b CM |
1623 | if (contig && found && state->start > last + 1) |
1624 | break; | |
1625 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1626 | total_bytes += min(search_end, state->end) + 1 - |
1627 | max(cur_start, state->start); | |
1628 | if (total_bytes >= max_bytes) | |
1629 | break; | |
1630 | if (!found) { | |
af60bed2 | 1631 | *start = max(cur_start, state->start); |
d1310b2e CM |
1632 | found = 1; |
1633 | } | |
ec29ed5b CM |
1634 | last = state->end; |
1635 | } else if (contig && found) { | |
1636 | break; | |
d1310b2e CM |
1637 | } |
1638 | node = rb_next(node); | |
1639 | if (!node) | |
1640 | break; | |
1641 | } | |
1642 | out: | |
cad321ad | 1643 | spin_unlock(&tree->lock); |
d1310b2e CM |
1644 | return total_bytes; |
1645 | } | |
b2950863 | 1646 | |
d352ac68 CM |
1647 | /* |
1648 | * set the private field for a given byte offset in the tree. If there isn't | |
1649 | * an extent_state there already, this does nothing. | |
1650 | */ | |
d1310b2e CM |
1651 | int set_state_private(struct extent_io_tree *tree, u64 start, u64 private) |
1652 | { | |
1653 | struct rb_node *node; | |
1654 | struct extent_state *state; | |
1655 | int ret = 0; | |
1656 | ||
cad321ad | 1657 | spin_lock(&tree->lock); |
d1310b2e CM |
1658 | /* |
1659 | * this search will find all the extents that end after | |
1660 | * our range starts. | |
1661 | */ | |
80ea96b1 | 1662 | node = tree_search(tree, start); |
2b114d1d | 1663 | if (!node) { |
d1310b2e CM |
1664 | ret = -ENOENT; |
1665 | goto out; | |
1666 | } | |
1667 | state = rb_entry(node, struct extent_state, rb_node); | |
1668 | if (state->start != start) { | |
1669 | ret = -ENOENT; | |
1670 | goto out; | |
1671 | } | |
1672 | state->private = private; | |
1673 | out: | |
cad321ad | 1674 | spin_unlock(&tree->lock); |
d1310b2e CM |
1675 | return ret; |
1676 | } | |
1677 | ||
1678 | int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private) | |
1679 | { | |
1680 | struct rb_node *node; | |
1681 | struct extent_state *state; | |
1682 | int ret = 0; | |
1683 | ||
cad321ad | 1684 | spin_lock(&tree->lock); |
d1310b2e CM |
1685 | /* |
1686 | * this search will find all the extents that end after | |
1687 | * our range starts. | |
1688 | */ | |
80ea96b1 | 1689 | node = tree_search(tree, start); |
2b114d1d | 1690 | if (!node) { |
d1310b2e CM |
1691 | ret = -ENOENT; |
1692 | goto out; | |
1693 | } | |
1694 | state = rb_entry(node, struct extent_state, rb_node); | |
1695 | if (state->start != start) { | |
1696 | ret = -ENOENT; | |
1697 | goto out; | |
1698 | } | |
1699 | *private = state->private; | |
1700 | out: | |
cad321ad | 1701 | spin_unlock(&tree->lock); |
d1310b2e CM |
1702 | return ret; |
1703 | } | |
1704 | ||
1705 | /* | |
1706 | * searches a range in the state tree for a given mask. | |
70dec807 | 1707 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1708 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1709 | * range is found set. | |
1710 | */ | |
1711 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9655d298 | 1712 | int bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1713 | { |
1714 | struct extent_state *state = NULL; | |
1715 | struct rb_node *node; | |
1716 | int bitset = 0; | |
d1310b2e | 1717 | |
cad321ad | 1718 | spin_lock(&tree->lock); |
df98b6e2 JB |
1719 | if (cached && cached->tree && cached->start <= start && |
1720 | cached->end > start) | |
9655d298 CM |
1721 | node = &cached->rb_node; |
1722 | else | |
1723 | node = tree_search(tree, start); | |
d1310b2e CM |
1724 | while (node && start <= end) { |
1725 | state = rb_entry(node, struct extent_state, rb_node); | |
1726 | ||
1727 | if (filled && state->start > start) { | |
1728 | bitset = 0; | |
1729 | break; | |
1730 | } | |
1731 | ||
1732 | if (state->start > end) | |
1733 | break; | |
1734 | ||
1735 | if (state->state & bits) { | |
1736 | bitset = 1; | |
1737 | if (!filled) | |
1738 | break; | |
1739 | } else if (filled) { | |
1740 | bitset = 0; | |
1741 | break; | |
1742 | } | |
46562cec CM |
1743 | |
1744 | if (state->end == (u64)-1) | |
1745 | break; | |
1746 | ||
d1310b2e CM |
1747 | start = state->end + 1; |
1748 | if (start > end) | |
1749 | break; | |
1750 | node = rb_next(node); | |
1751 | if (!node) { | |
1752 | if (filled) | |
1753 | bitset = 0; | |
1754 | break; | |
1755 | } | |
1756 | } | |
cad321ad | 1757 | spin_unlock(&tree->lock); |
d1310b2e CM |
1758 | return bitset; |
1759 | } | |
d1310b2e CM |
1760 | |
1761 | /* | |
1762 | * helper function to set a given page up to date if all the | |
1763 | * extents in the tree for that page are up to date | |
1764 | */ | |
1765 | static int check_page_uptodate(struct extent_io_tree *tree, | |
1766 | struct page *page) | |
1767 | { | |
1768 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1769 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
9655d298 | 1770 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e CM |
1771 | SetPageUptodate(page); |
1772 | return 0; | |
1773 | } | |
1774 | ||
1775 | /* | |
1776 | * helper function to unlock a page if all the extents in the tree | |
1777 | * for that page are unlocked | |
1778 | */ | |
1779 | static int check_page_locked(struct extent_io_tree *tree, | |
1780 | struct page *page) | |
1781 | { | |
1782 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1783 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
9655d298 | 1784 | if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) |
d1310b2e CM |
1785 | unlock_page(page); |
1786 | return 0; | |
1787 | } | |
1788 | ||
1789 | /* | |
1790 | * helper function to end page writeback if all the extents | |
1791 | * in the tree for that page are done with writeback | |
1792 | */ | |
1793 | static int check_page_writeback(struct extent_io_tree *tree, | |
1794 | struct page *page) | |
1795 | { | |
1edbb734 | 1796 | end_page_writeback(page); |
d1310b2e CM |
1797 | return 0; |
1798 | } | |
1799 | ||
4a54c8c1 JS |
1800 | /* |
1801 | * When IO fails, either with EIO or csum verification fails, we | |
1802 | * try other mirrors that might have a good copy of the data. This | |
1803 | * io_failure_record is used to record state as we go through all the | |
1804 | * mirrors. If another mirror has good data, the page is set up to date | |
1805 | * and things continue. If a good mirror can't be found, the original | |
1806 | * bio end_io callback is called to indicate things have failed. | |
1807 | */ | |
1808 | struct io_failure_record { | |
1809 | struct page *page; | |
1810 | u64 start; | |
1811 | u64 len; | |
1812 | u64 logical; | |
1813 | unsigned long bio_flags; | |
1814 | int this_mirror; | |
1815 | int failed_mirror; | |
1816 | int in_validation; | |
1817 | }; | |
1818 | ||
1819 | static int free_io_failure(struct inode *inode, struct io_failure_record *rec, | |
1820 | int did_repair) | |
1821 | { | |
1822 | int ret; | |
1823 | int err = 0; | |
1824 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
1825 | ||
1826 | set_state_private(failure_tree, rec->start, 0); | |
1827 | ret = clear_extent_bits(failure_tree, rec->start, | |
1828 | rec->start + rec->len - 1, | |
1829 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1830 | if (ret) | |
1831 | err = ret; | |
1832 | ||
1833 | if (did_repair) { | |
1834 | ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start, | |
1835 | rec->start + rec->len - 1, | |
1836 | EXTENT_DAMAGED, GFP_NOFS); | |
1837 | if (ret && !err) | |
1838 | err = ret; | |
1839 | } | |
1840 | ||
1841 | kfree(rec); | |
1842 | return err; | |
1843 | } | |
1844 | ||
1845 | static void repair_io_failure_callback(struct bio *bio, int err) | |
1846 | { | |
1847 | complete(bio->bi_private); | |
1848 | } | |
1849 | ||
1850 | /* | |
1851 | * this bypasses the standard btrfs submit functions deliberately, as | |
1852 | * the standard behavior is to write all copies in a raid setup. here we only | |
1853 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
1854 | * submit_bio directly. | |
1855 | * to avoid any synchonization issues, wait for the data after writing, which | |
1856 | * actually prevents the read that triggered the error from finishing. | |
1857 | * currently, there can be no more than two copies of every data bit. thus, | |
1858 | * exactly one rewrite is required. | |
1859 | */ | |
1860 | int repair_io_failure(struct btrfs_mapping_tree *map_tree, u64 start, | |
1861 | u64 length, u64 logical, struct page *page, | |
1862 | int mirror_num) | |
1863 | { | |
1864 | struct bio *bio; | |
1865 | struct btrfs_device *dev; | |
1866 | DECLARE_COMPLETION_ONSTACK(compl); | |
1867 | u64 map_length = 0; | |
1868 | u64 sector; | |
1869 | struct btrfs_bio *bbio = NULL; | |
1870 | int ret; | |
1871 | ||
1872 | BUG_ON(!mirror_num); | |
1873 | ||
1874 | bio = bio_alloc(GFP_NOFS, 1); | |
1875 | if (!bio) | |
1876 | return -EIO; | |
1877 | bio->bi_private = &compl; | |
1878 | bio->bi_end_io = repair_io_failure_callback; | |
1879 | bio->bi_size = 0; | |
1880 | map_length = length; | |
1881 | ||
1882 | ret = btrfs_map_block(map_tree, WRITE, logical, | |
1883 | &map_length, &bbio, mirror_num); | |
1884 | if (ret) { | |
1885 | bio_put(bio); | |
1886 | return -EIO; | |
1887 | } | |
1888 | BUG_ON(mirror_num != bbio->mirror_num); | |
1889 | sector = bbio->stripes[mirror_num-1].physical >> 9; | |
1890 | bio->bi_sector = sector; | |
1891 | dev = bbio->stripes[mirror_num-1].dev; | |
1892 | kfree(bbio); | |
1893 | if (!dev || !dev->bdev || !dev->writeable) { | |
1894 | bio_put(bio); | |
1895 | return -EIO; | |
1896 | } | |
1897 | bio->bi_bdev = dev->bdev; | |
1898 | bio_add_page(bio, page, length, start-page_offset(page)); | |
21adbd5c | 1899 | btrfsic_submit_bio(WRITE_SYNC, bio); |
4a54c8c1 JS |
1900 | wait_for_completion(&compl); |
1901 | ||
1902 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) { | |
1903 | /* try to remap that extent elsewhere? */ | |
1904 | bio_put(bio); | |
1905 | return -EIO; | |
1906 | } | |
1907 | ||
1908 | printk(KERN_INFO "btrfs read error corrected: ino %lu off %llu (dev %s " | |
1909 | "sector %llu)\n", page->mapping->host->i_ino, start, | |
1910 | dev->name, sector); | |
1911 | ||
1912 | bio_put(bio); | |
1913 | return 0; | |
1914 | } | |
1915 | ||
1916 | /* | |
1917 | * each time an IO finishes, we do a fast check in the IO failure tree | |
1918 | * to see if we need to process or clean up an io_failure_record | |
1919 | */ | |
1920 | static int clean_io_failure(u64 start, struct page *page) | |
1921 | { | |
1922 | u64 private; | |
1923 | u64 private_failure; | |
1924 | struct io_failure_record *failrec; | |
1925 | struct btrfs_mapping_tree *map_tree; | |
1926 | struct extent_state *state; | |
1927 | int num_copies; | |
1928 | int did_repair = 0; | |
1929 | int ret; | |
1930 | struct inode *inode = page->mapping->host; | |
1931 | ||
1932 | private = 0; | |
1933 | ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
1934 | (u64)-1, 1, EXTENT_DIRTY, 0); | |
1935 | if (!ret) | |
1936 | return 0; | |
1937 | ||
1938 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start, | |
1939 | &private_failure); | |
1940 | if (ret) | |
1941 | return 0; | |
1942 | ||
1943 | failrec = (struct io_failure_record *)(unsigned long) private_failure; | |
1944 | BUG_ON(!failrec->this_mirror); | |
1945 | ||
1946 | if (failrec->in_validation) { | |
1947 | /* there was no real error, just free the record */ | |
1948 | pr_debug("clean_io_failure: freeing dummy error at %llu\n", | |
1949 | failrec->start); | |
1950 | did_repair = 1; | |
1951 | goto out; | |
1952 | } | |
1953 | ||
1954 | spin_lock(&BTRFS_I(inode)->io_tree.lock); | |
1955 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, | |
1956 | failrec->start, | |
1957 | EXTENT_LOCKED); | |
1958 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); | |
1959 | ||
1960 | if (state && state->start == failrec->start) { | |
1961 | map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree; | |
1962 | num_copies = btrfs_num_copies(map_tree, failrec->logical, | |
1963 | failrec->len); | |
1964 | if (num_copies > 1) { | |
1965 | ret = repair_io_failure(map_tree, start, failrec->len, | |
1966 | failrec->logical, page, | |
1967 | failrec->failed_mirror); | |
1968 | did_repair = !ret; | |
1969 | } | |
1970 | } | |
1971 | ||
1972 | out: | |
1973 | if (!ret) | |
1974 | ret = free_io_failure(inode, failrec, did_repair); | |
1975 | ||
1976 | return ret; | |
1977 | } | |
1978 | ||
1979 | /* | |
1980 | * this is a generic handler for readpage errors (default | |
1981 | * readpage_io_failed_hook). if other copies exist, read those and write back | |
1982 | * good data to the failed position. does not investigate in remapping the | |
1983 | * failed extent elsewhere, hoping the device will be smart enough to do this as | |
1984 | * needed | |
1985 | */ | |
1986 | ||
1987 | static int bio_readpage_error(struct bio *failed_bio, struct page *page, | |
1988 | u64 start, u64 end, int failed_mirror, | |
1989 | struct extent_state *state) | |
1990 | { | |
1991 | struct io_failure_record *failrec = NULL; | |
1992 | u64 private; | |
1993 | struct extent_map *em; | |
1994 | struct inode *inode = page->mapping->host; | |
1995 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
1996 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
1997 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
1998 | struct bio *bio; | |
1999 | int num_copies; | |
2000 | int ret; | |
2001 | int read_mode; | |
2002 | u64 logical; | |
2003 | ||
2004 | BUG_ON(failed_bio->bi_rw & REQ_WRITE); | |
2005 | ||
2006 | ret = get_state_private(failure_tree, start, &private); | |
2007 | if (ret) { | |
2008 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2009 | if (!failrec) | |
2010 | return -ENOMEM; | |
2011 | failrec->start = start; | |
2012 | failrec->len = end - start + 1; | |
2013 | failrec->this_mirror = 0; | |
2014 | failrec->bio_flags = 0; | |
2015 | failrec->in_validation = 0; | |
2016 | ||
2017 | read_lock(&em_tree->lock); | |
2018 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2019 | if (!em) { | |
2020 | read_unlock(&em_tree->lock); | |
2021 | kfree(failrec); | |
2022 | return -EIO; | |
2023 | } | |
2024 | ||
2025 | if (em->start > start || em->start + em->len < start) { | |
2026 | free_extent_map(em); | |
2027 | em = NULL; | |
2028 | } | |
2029 | read_unlock(&em_tree->lock); | |
2030 | ||
2031 | if (!em || IS_ERR(em)) { | |
2032 | kfree(failrec); | |
2033 | return -EIO; | |
2034 | } | |
2035 | logical = start - em->start; | |
2036 | logical = em->block_start + logical; | |
2037 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2038 | logical = em->block_start; | |
2039 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2040 | extent_set_compress_type(&failrec->bio_flags, | |
2041 | em->compress_type); | |
2042 | } | |
2043 | pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, " | |
2044 | "len=%llu\n", logical, start, failrec->len); | |
2045 | failrec->logical = logical; | |
2046 | free_extent_map(em); | |
2047 | ||
2048 | /* set the bits in the private failure tree */ | |
2049 | ret = set_extent_bits(failure_tree, start, end, | |
2050 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
2051 | if (ret >= 0) | |
2052 | ret = set_state_private(failure_tree, start, | |
2053 | (u64)(unsigned long)failrec); | |
2054 | /* set the bits in the inode's tree */ | |
2055 | if (ret >= 0) | |
2056 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED, | |
2057 | GFP_NOFS); | |
2058 | if (ret < 0) { | |
2059 | kfree(failrec); | |
2060 | return ret; | |
2061 | } | |
2062 | } else { | |
2063 | failrec = (struct io_failure_record *)(unsigned long)private; | |
2064 | pr_debug("bio_readpage_error: (found) logical=%llu, " | |
2065 | "start=%llu, len=%llu, validation=%d\n", | |
2066 | failrec->logical, failrec->start, failrec->len, | |
2067 | failrec->in_validation); | |
2068 | /* | |
2069 | * when data can be on disk more than twice, add to failrec here | |
2070 | * (e.g. with a list for failed_mirror) to make | |
2071 | * clean_io_failure() clean all those errors at once. | |
2072 | */ | |
2073 | } | |
2074 | num_copies = btrfs_num_copies( | |
2075 | &BTRFS_I(inode)->root->fs_info->mapping_tree, | |
2076 | failrec->logical, failrec->len); | |
2077 | if (num_copies == 1) { | |
2078 | /* | |
2079 | * we only have a single copy of the data, so don't bother with | |
2080 | * all the retry and error correction code that follows. no | |
2081 | * matter what the error is, it is very likely to persist. | |
2082 | */ | |
2083 | pr_debug("bio_readpage_error: cannot repair, num_copies == 1. " | |
2084 | "state=%p, num_copies=%d, next_mirror %d, " | |
2085 | "failed_mirror %d\n", state, num_copies, | |
2086 | failrec->this_mirror, failed_mirror); | |
2087 | free_io_failure(inode, failrec, 0); | |
2088 | return -EIO; | |
2089 | } | |
2090 | ||
2091 | if (!state) { | |
2092 | spin_lock(&tree->lock); | |
2093 | state = find_first_extent_bit_state(tree, failrec->start, | |
2094 | EXTENT_LOCKED); | |
2095 | if (state && state->start != failrec->start) | |
2096 | state = NULL; | |
2097 | spin_unlock(&tree->lock); | |
2098 | } | |
2099 | ||
2100 | /* | |
2101 | * there are two premises: | |
2102 | * a) deliver good data to the caller | |
2103 | * b) correct the bad sectors on disk | |
2104 | */ | |
2105 | if (failed_bio->bi_vcnt > 1) { | |
2106 | /* | |
2107 | * to fulfill b), we need to know the exact failing sectors, as | |
2108 | * we don't want to rewrite any more than the failed ones. thus, | |
2109 | * we need separate read requests for the failed bio | |
2110 | * | |
2111 | * if the following BUG_ON triggers, our validation request got | |
2112 | * merged. we need separate requests for our algorithm to work. | |
2113 | */ | |
2114 | BUG_ON(failrec->in_validation); | |
2115 | failrec->in_validation = 1; | |
2116 | failrec->this_mirror = failed_mirror; | |
2117 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; | |
2118 | } else { | |
2119 | /* | |
2120 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2121 | * of the failed sector and if we succeed, we have setup | |
2122 | * everything for repair_io_failure to do the rest for us. | |
2123 | */ | |
2124 | if (failrec->in_validation) { | |
2125 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2126 | failrec->in_validation = 0; | |
2127 | failrec->this_mirror = 0; | |
2128 | } | |
2129 | failrec->failed_mirror = failed_mirror; | |
2130 | failrec->this_mirror++; | |
2131 | if (failrec->this_mirror == failed_mirror) | |
2132 | failrec->this_mirror++; | |
2133 | read_mode = READ_SYNC; | |
2134 | } | |
2135 | ||
2136 | if (!state || failrec->this_mirror > num_copies) { | |
2137 | pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, " | |
2138 | "next_mirror %d, failed_mirror %d\n", state, | |
2139 | num_copies, failrec->this_mirror, failed_mirror); | |
2140 | free_io_failure(inode, failrec, 0); | |
2141 | return -EIO; | |
2142 | } | |
2143 | ||
2144 | bio = bio_alloc(GFP_NOFS, 1); | |
2145 | bio->bi_private = state; | |
2146 | bio->bi_end_io = failed_bio->bi_end_io; | |
2147 | bio->bi_sector = failrec->logical >> 9; | |
2148 | bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; | |
2149 | bio->bi_size = 0; | |
2150 | ||
2151 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); | |
2152 | ||
2153 | pr_debug("bio_readpage_error: submitting new read[%#x] to " | |
2154 | "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode, | |
2155 | failrec->this_mirror, num_copies, failrec->in_validation); | |
2156 | ||
2157 | tree->ops->submit_bio_hook(inode, read_mode, bio, failrec->this_mirror, | |
2158 | failrec->bio_flags, 0); | |
2159 | return 0; | |
2160 | } | |
2161 | ||
d1310b2e CM |
2162 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2163 | ||
2164 | /* | |
2165 | * after a writepage IO is done, we need to: | |
2166 | * clear the uptodate bits on error | |
2167 | * clear the writeback bits in the extent tree for this IO | |
2168 | * end_page_writeback if the page has no more pending IO | |
2169 | * | |
2170 | * Scheduling is not allowed, so the extent state tree is expected | |
2171 | * to have one and only one object corresponding to this IO. | |
2172 | */ | |
d1310b2e | 2173 | static void end_bio_extent_writepage(struct bio *bio, int err) |
d1310b2e | 2174 | { |
1259ab75 | 2175 | int uptodate = err == 0; |
d1310b2e | 2176 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
902b22f3 | 2177 | struct extent_io_tree *tree; |
d1310b2e CM |
2178 | u64 start; |
2179 | u64 end; | |
2180 | int whole_page; | |
1259ab75 | 2181 | int ret; |
d1310b2e | 2182 | |
d1310b2e CM |
2183 | do { |
2184 | struct page *page = bvec->bv_page; | |
902b22f3 DW |
2185 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
2186 | ||
d1310b2e CM |
2187 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
2188 | bvec->bv_offset; | |
2189 | end = start + bvec->bv_len - 1; | |
2190 | ||
2191 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
2192 | whole_page = 1; | |
2193 | else | |
2194 | whole_page = 0; | |
2195 | ||
2196 | if (--bvec >= bio->bi_io_vec) | |
2197 | prefetchw(&bvec->bv_page->flags); | |
1259ab75 CM |
2198 | if (tree->ops && tree->ops->writepage_end_io_hook) { |
2199 | ret = tree->ops->writepage_end_io_hook(page, start, | |
902b22f3 | 2200 | end, NULL, uptodate); |
1259ab75 CM |
2201 | if (ret) |
2202 | uptodate = 0; | |
2203 | } | |
2204 | ||
2205 | if (!uptodate && tree->ops && | |
2206 | tree->ops->writepage_io_failed_hook) { | |
2207 | ret = tree->ops->writepage_io_failed_hook(bio, page, | |
902b22f3 | 2208 | start, end, NULL); |
1259ab75 | 2209 | if (ret == 0) { |
1259ab75 CM |
2210 | uptodate = (err == 0); |
2211 | continue; | |
2212 | } | |
2213 | } | |
2214 | ||
d1310b2e | 2215 | if (!uptodate) { |
2ac55d41 | 2216 | clear_extent_uptodate(tree, start, end, NULL, GFP_NOFS); |
d1310b2e CM |
2217 | ClearPageUptodate(page); |
2218 | SetPageError(page); | |
2219 | } | |
70dec807 | 2220 | |
d1310b2e CM |
2221 | if (whole_page) |
2222 | end_page_writeback(page); | |
2223 | else | |
2224 | check_page_writeback(tree, page); | |
d1310b2e | 2225 | } while (bvec >= bio->bi_io_vec); |
2b1f55b0 | 2226 | |
d1310b2e | 2227 | bio_put(bio); |
d1310b2e CM |
2228 | } |
2229 | ||
2230 | /* | |
2231 | * after a readpage IO is done, we need to: | |
2232 | * clear the uptodate bits on error | |
2233 | * set the uptodate bits if things worked | |
2234 | * set the page up to date if all extents in the tree are uptodate | |
2235 | * clear the lock bit in the extent tree | |
2236 | * unlock the page if there are no other extents locked for it | |
2237 | * | |
2238 | * Scheduling is not allowed, so the extent state tree is expected | |
2239 | * to have one and only one object corresponding to this IO. | |
2240 | */ | |
d1310b2e | 2241 | static void end_bio_extent_readpage(struct bio *bio, int err) |
d1310b2e CM |
2242 | { |
2243 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
4125bf76 CM |
2244 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
2245 | struct bio_vec *bvec = bio->bi_io_vec; | |
902b22f3 | 2246 | struct extent_io_tree *tree; |
d1310b2e CM |
2247 | u64 start; |
2248 | u64 end; | |
2249 | int whole_page; | |
2250 | int ret; | |
2251 | ||
d20f7043 CM |
2252 | if (err) |
2253 | uptodate = 0; | |
2254 | ||
d1310b2e CM |
2255 | do { |
2256 | struct page *page = bvec->bv_page; | |
507903b8 AJ |
2257 | struct extent_state *cached = NULL; |
2258 | struct extent_state *state; | |
2259 | ||
4a54c8c1 JS |
2260 | pr_debug("end_bio_extent_readpage: bi_vcnt=%d, idx=%d, err=%d, " |
2261 | "mirror=%ld\n", bio->bi_vcnt, bio->bi_idx, err, | |
2262 | (long int)bio->bi_bdev); | |
902b22f3 DW |
2263 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
2264 | ||
d1310b2e CM |
2265 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
2266 | bvec->bv_offset; | |
2267 | end = start + bvec->bv_len - 1; | |
2268 | ||
2269 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
2270 | whole_page = 1; | |
2271 | else | |
2272 | whole_page = 0; | |
2273 | ||
4125bf76 | 2274 | if (++bvec <= bvec_end) |
d1310b2e CM |
2275 | prefetchw(&bvec->bv_page->flags); |
2276 | ||
507903b8 | 2277 | spin_lock(&tree->lock); |
0d399205 | 2278 | state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED); |
109b36a2 | 2279 | if (state && state->start == start) { |
507903b8 AJ |
2280 | /* |
2281 | * take a reference on the state, unlock will drop | |
2282 | * the ref | |
2283 | */ | |
2284 | cache_state(state, &cached); | |
2285 | } | |
2286 | spin_unlock(&tree->lock); | |
2287 | ||
d1310b2e | 2288 | if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) { |
70dec807 | 2289 | ret = tree->ops->readpage_end_io_hook(page, start, end, |
507903b8 | 2290 | state); |
d1310b2e CM |
2291 | if (ret) |
2292 | uptodate = 0; | |
4a54c8c1 JS |
2293 | else |
2294 | clean_io_failure(start, page); | |
d1310b2e | 2295 | } |
4a54c8c1 | 2296 | if (!uptodate) { |
32240a91 JS |
2297 | int failed_mirror; |
2298 | failed_mirror = (int)(unsigned long)bio->bi_bdev; | |
f4a8e656 JS |
2299 | /* |
2300 | * The generic bio_readpage_error handles errors the | |
2301 | * following way: If possible, new read requests are | |
2302 | * created and submitted and will end up in | |
2303 | * end_bio_extent_readpage as well (if we're lucky, not | |
2304 | * in the !uptodate case). In that case it returns 0 and | |
2305 | * we just go on with the next page in our bio. If it | |
2306 | * can't handle the error it will return -EIO and we | |
2307 | * remain responsible for that page. | |
2308 | */ | |
2309 | ret = bio_readpage_error(bio, page, start, end, | |
2310 | failed_mirror, NULL); | |
7e38326f | 2311 | if (ret == 0) { |
f4a8e656 | 2312 | error_handled: |
3b951516 CM |
2313 | uptodate = |
2314 | test_bit(BIO_UPTODATE, &bio->bi_flags); | |
d20f7043 CM |
2315 | if (err) |
2316 | uptodate = 0; | |
507903b8 | 2317 | uncache_state(&cached); |
7e38326f CM |
2318 | continue; |
2319 | } | |
f4a8e656 JS |
2320 | if (tree->ops && tree->ops->readpage_io_failed_hook) { |
2321 | ret = tree->ops->readpage_io_failed_hook( | |
2322 | bio, page, start, end, | |
2323 | failed_mirror, state); | |
2324 | if (ret == 0) | |
2325 | goto error_handled; | |
2326 | } | |
7e38326f | 2327 | } |
d1310b2e | 2328 | |
771ed689 | 2329 | if (uptodate) { |
507903b8 | 2330 | set_extent_uptodate(tree, start, end, &cached, |
902b22f3 | 2331 | GFP_ATOMIC); |
771ed689 | 2332 | } |
507903b8 | 2333 | unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC); |
d1310b2e | 2334 | |
70dec807 CM |
2335 | if (whole_page) { |
2336 | if (uptodate) { | |
2337 | SetPageUptodate(page); | |
2338 | } else { | |
2339 | ClearPageUptodate(page); | |
2340 | SetPageError(page); | |
2341 | } | |
d1310b2e | 2342 | unlock_page(page); |
70dec807 CM |
2343 | } else { |
2344 | if (uptodate) { | |
2345 | check_page_uptodate(tree, page); | |
2346 | } else { | |
2347 | ClearPageUptodate(page); | |
2348 | SetPageError(page); | |
2349 | } | |
d1310b2e | 2350 | check_page_locked(tree, page); |
70dec807 | 2351 | } |
4125bf76 | 2352 | } while (bvec <= bvec_end); |
d1310b2e CM |
2353 | |
2354 | bio_put(bio); | |
d1310b2e CM |
2355 | } |
2356 | ||
88f794ed MX |
2357 | struct bio * |
2358 | btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
2359 | gfp_t gfp_flags) | |
d1310b2e CM |
2360 | { |
2361 | struct bio *bio; | |
2362 | ||
2363 | bio = bio_alloc(gfp_flags, nr_vecs); | |
2364 | ||
2365 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
2366 | while (!bio && (nr_vecs /= 2)) | |
2367 | bio = bio_alloc(gfp_flags, nr_vecs); | |
2368 | } | |
2369 | ||
2370 | if (bio) { | |
e1c4b745 | 2371 | bio->bi_size = 0; |
d1310b2e CM |
2372 | bio->bi_bdev = bdev; |
2373 | bio->bi_sector = first_sector; | |
2374 | } | |
2375 | return bio; | |
2376 | } | |
2377 | ||
c8b97818 CM |
2378 | static int submit_one_bio(int rw, struct bio *bio, int mirror_num, |
2379 | unsigned long bio_flags) | |
d1310b2e | 2380 | { |
d1310b2e | 2381 | int ret = 0; |
70dec807 CM |
2382 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
2383 | struct page *page = bvec->bv_page; | |
2384 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 2385 | u64 start; |
70dec807 CM |
2386 | |
2387 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset; | |
70dec807 | 2388 | |
902b22f3 | 2389 | bio->bi_private = NULL; |
d1310b2e CM |
2390 | |
2391 | bio_get(bio); | |
2392 | ||
065631f6 | 2393 | if (tree->ops && tree->ops->submit_bio_hook) |
6b82ce8d | 2394 | ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio, |
eaf25d93 | 2395 | mirror_num, bio_flags, start); |
0b86a832 | 2396 | else |
21adbd5c | 2397 | btrfsic_submit_bio(rw, bio); |
4a54c8c1 | 2398 | |
d1310b2e CM |
2399 | if (bio_flagged(bio, BIO_EOPNOTSUPP)) |
2400 | ret = -EOPNOTSUPP; | |
2401 | bio_put(bio); | |
2402 | return ret; | |
2403 | } | |
2404 | ||
2405 | static int submit_extent_page(int rw, struct extent_io_tree *tree, | |
2406 | struct page *page, sector_t sector, | |
2407 | size_t size, unsigned long offset, | |
2408 | struct block_device *bdev, | |
2409 | struct bio **bio_ret, | |
2410 | unsigned long max_pages, | |
f188591e | 2411 | bio_end_io_t end_io_func, |
c8b97818 CM |
2412 | int mirror_num, |
2413 | unsigned long prev_bio_flags, | |
2414 | unsigned long bio_flags) | |
d1310b2e CM |
2415 | { |
2416 | int ret = 0; | |
2417 | struct bio *bio; | |
2418 | int nr; | |
c8b97818 CM |
2419 | int contig = 0; |
2420 | int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED; | |
2421 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; | |
5b050f04 | 2422 | size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE); |
d1310b2e CM |
2423 | |
2424 | if (bio_ret && *bio_ret) { | |
2425 | bio = *bio_ret; | |
c8b97818 CM |
2426 | if (old_compressed) |
2427 | contig = bio->bi_sector == sector; | |
2428 | else | |
2429 | contig = bio->bi_sector + (bio->bi_size >> 9) == | |
2430 | sector; | |
2431 | ||
2432 | if (prev_bio_flags != bio_flags || !contig || | |
239b14b3 | 2433 | (tree->ops && tree->ops->merge_bio_hook && |
c8b97818 CM |
2434 | tree->ops->merge_bio_hook(page, offset, page_size, bio, |
2435 | bio_flags)) || | |
2436 | bio_add_page(bio, page, page_size, offset) < page_size) { | |
2437 | ret = submit_one_bio(rw, bio, mirror_num, | |
2438 | prev_bio_flags); | |
d1310b2e CM |
2439 | bio = NULL; |
2440 | } else { | |
2441 | return 0; | |
2442 | } | |
2443 | } | |
c8b97818 CM |
2444 | if (this_compressed) |
2445 | nr = BIO_MAX_PAGES; | |
2446 | else | |
2447 | nr = bio_get_nr_vecs(bdev); | |
2448 | ||
88f794ed | 2449 | bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH); |
5df67083 TI |
2450 | if (!bio) |
2451 | return -ENOMEM; | |
70dec807 | 2452 | |
c8b97818 | 2453 | bio_add_page(bio, page, page_size, offset); |
d1310b2e CM |
2454 | bio->bi_end_io = end_io_func; |
2455 | bio->bi_private = tree; | |
70dec807 | 2456 | |
d397712b | 2457 | if (bio_ret) |
d1310b2e | 2458 | *bio_ret = bio; |
d397712b | 2459 | else |
c8b97818 | 2460 | ret = submit_one_bio(rw, bio, mirror_num, bio_flags); |
d1310b2e CM |
2461 | |
2462 | return ret; | |
2463 | } | |
2464 | ||
2465 | void set_page_extent_mapped(struct page *page) | |
2466 | { | |
2467 | if (!PagePrivate(page)) { | |
2468 | SetPagePrivate(page); | |
d1310b2e | 2469 | page_cache_get(page); |
6af118ce | 2470 | set_page_private(page, EXTENT_PAGE_PRIVATE); |
d1310b2e CM |
2471 | } |
2472 | } | |
2473 | ||
b2950863 | 2474 | static void set_page_extent_head(struct page *page, unsigned long len) |
d1310b2e | 2475 | { |
eb14ab8e | 2476 | WARN_ON(!PagePrivate(page)); |
d1310b2e CM |
2477 | set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2); |
2478 | } | |
2479 | ||
2480 | /* | |
2481 | * basic readpage implementation. Locked extent state structs are inserted | |
2482 | * into the tree that are removed when the IO is done (by the end_io | |
2483 | * handlers) | |
2484 | */ | |
2485 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
2486 | struct page *page, | |
2487 | get_extent_t *get_extent, | |
c8b97818 CM |
2488 | struct bio **bio, int mirror_num, |
2489 | unsigned long *bio_flags) | |
d1310b2e CM |
2490 | { |
2491 | struct inode *inode = page->mapping->host; | |
2492 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2493 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2494 | u64 end; | |
2495 | u64 cur = start; | |
2496 | u64 extent_offset; | |
2497 | u64 last_byte = i_size_read(inode); | |
2498 | u64 block_start; | |
2499 | u64 cur_end; | |
2500 | sector_t sector; | |
2501 | struct extent_map *em; | |
2502 | struct block_device *bdev; | |
11c65dcc | 2503 | struct btrfs_ordered_extent *ordered; |
d1310b2e CM |
2504 | int ret; |
2505 | int nr = 0; | |
306e16ce | 2506 | size_t pg_offset = 0; |
d1310b2e | 2507 | size_t iosize; |
c8b97818 | 2508 | size_t disk_io_size; |
d1310b2e | 2509 | size_t blocksize = inode->i_sb->s_blocksize; |
c8b97818 | 2510 | unsigned long this_bio_flag = 0; |
d1310b2e CM |
2511 | |
2512 | set_page_extent_mapped(page); | |
2513 | ||
90a887c9 DM |
2514 | if (!PageUptodate(page)) { |
2515 | if (cleancache_get_page(page) == 0) { | |
2516 | BUG_ON(blocksize != PAGE_SIZE); | |
2517 | goto out; | |
2518 | } | |
2519 | } | |
2520 | ||
d1310b2e | 2521 | end = page_end; |
11c65dcc JB |
2522 | while (1) { |
2523 | lock_extent(tree, start, end, GFP_NOFS); | |
2524 | ordered = btrfs_lookup_ordered_extent(inode, start); | |
2525 | if (!ordered) | |
2526 | break; | |
2527 | unlock_extent(tree, start, end, GFP_NOFS); | |
2528 | btrfs_start_ordered_extent(inode, ordered, 1); | |
2529 | btrfs_put_ordered_extent(ordered); | |
2530 | } | |
d1310b2e | 2531 | |
c8b97818 CM |
2532 | if (page->index == last_byte >> PAGE_CACHE_SHIFT) { |
2533 | char *userpage; | |
2534 | size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1); | |
2535 | ||
2536 | if (zero_offset) { | |
2537 | iosize = PAGE_CACHE_SIZE - zero_offset; | |
2538 | userpage = kmap_atomic(page, KM_USER0); | |
2539 | memset(userpage + zero_offset, 0, iosize); | |
2540 | flush_dcache_page(page); | |
2541 | kunmap_atomic(userpage, KM_USER0); | |
2542 | } | |
2543 | } | |
d1310b2e CM |
2544 | while (cur <= end) { |
2545 | if (cur >= last_byte) { | |
2546 | char *userpage; | |
507903b8 AJ |
2547 | struct extent_state *cached = NULL; |
2548 | ||
306e16ce | 2549 | iosize = PAGE_CACHE_SIZE - pg_offset; |
d1310b2e | 2550 | userpage = kmap_atomic(page, KM_USER0); |
306e16ce | 2551 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e CM |
2552 | flush_dcache_page(page); |
2553 | kunmap_atomic(userpage, KM_USER0); | |
2554 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 AJ |
2555 | &cached, GFP_NOFS); |
2556 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
2557 | &cached, GFP_NOFS); | |
d1310b2e CM |
2558 | break; |
2559 | } | |
306e16ce | 2560 | em = get_extent(inode, page, pg_offset, cur, |
d1310b2e | 2561 | end - cur + 1, 0); |
c704005d | 2562 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e CM |
2563 | SetPageError(page); |
2564 | unlock_extent(tree, cur, end, GFP_NOFS); | |
2565 | break; | |
2566 | } | |
d1310b2e CM |
2567 | extent_offset = cur - em->start; |
2568 | BUG_ON(extent_map_end(em) <= cur); | |
2569 | BUG_ON(end < cur); | |
2570 | ||
261507a0 | 2571 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
c8b97818 | 2572 | this_bio_flag = EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
2573 | extent_set_compress_type(&this_bio_flag, |
2574 | em->compress_type); | |
2575 | } | |
c8b97818 | 2576 | |
d1310b2e CM |
2577 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
2578 | cur_end = min(extent_map_end(em) - 1, end); | |
2579 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
c8b97818 CM |
2580 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
2581 | disk_io_size = em->block_len; | |
2582 | sector = em->block_start >> 9; | |
2583 | } else { | |
2584 | sector = (em->block_start + extent_offset) >> 9; | |
2585 | disk_io_size = iosize; | |
2586 | } | |
d1310b2e CM |
2587 | bdev = em->bdev; |
2588 | block_start = em->block_start; | |
d899e052 YZ |
2589 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2590 | block_start = EXTENT_MAP_HOLE; | |
d1310b2e CM |
2591 | free_extent_map(em); |
2592 | em = NULL; | |
2593 | ||
2594 | /* we've found a hole, just zero and go on */ | |
2595 | if (block_start == EXTENT_MAP_HOLE) { | |
2596 | char *userpage; | |
507903b8 AJ |
2597 | struct extent_state *cached = NULL; |
2598 | ||
d1310b2e | 2599 | userpage = kmap_atomic(page, KM_USER0); |
306e16ce | 2600 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e CM |
2601 | flush_dcache_page(page); |
2602 | kunmap_atomic(userpage, KM_USER0); | |
2603 | ||
2604 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 AJ |
2605 | &cached, GFP_NOFS); |
2606 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
2607 | &cached, GFP_NOFS); | |
d1310b2e | 2608 | cur = cur + iosize; |
306e16ce | 2609 | pg_offset += iosize; |
d1310b2e CM |
2610 | continue; |
2611 | } | |
2612 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
2613 | if (test_range_bit(tree, cur, cur_end, |
2614 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 2615 | check_page_uptodate(tree, page); |
d1310b2e CM |
2616 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); |
2617 | cur = cur + iosize; | |
306e16ce | 2618 | pg_offset += iosize; |
d1310b2e CM |
2619 | continue; |
2620 | } | |
70dec807 CM |
2621 | /* we have an inline extent but it didn't get marked up |
2622 | * to date. Error out | |
2623 | */ | |
2624 | if (block_start == EXTENT_MAP_INLINE) { | |
2625 | SetPageError(page); | |
2626 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
2627 | cur = cur + iosize; | |
306e16ce | 2628 | pg_offset += iosize; |
70dec807 CM |
2629 | continue; |
2630 | } | |
d1310b2e CM |
2631 | |
2632 | ret = 0; | |
2633 | if (tree->ops && tree->ops->readpage_io_hook) { | |
2634 | ret = tree->ops->readpage_io_hook(page, cur, | |
2635 | cur + iosize - 1); | |
2636 | } | |
2637 | if (!ret) { | |
89642229 CM |
2638 | unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1; |
2639 | pnr -= page->index; | |
d1310b2e | 2640 | ret = submit_extent_page(READ, tree, page, |
306e16ce | 2641 | sector, disk_io_size, pg_offset, |
89642229 | 2642 | bdev, bio, pnr, |
c8b97818 CM |
2643 | end_bio_extent_readpage, mirror_num, |
2644 | *bio_flags, | |
2645 | this_bio_flag); | |
89642229 | 2646 | nr++; |
c8b97818 | 2647 | *bio_flags = this_bio_flag; |
d1310b2e CM |
2648 | } |
2649 | if (ret) | |
2650 | SetPageError(page); | |
2651 | cur = cur + iosize; | |
306e16ce | 2652 | pg_offset += iosize; |
d1310b2e | 2653 | } |
90a887c9 | 2654 | out: |
d1310b2e CM |
2655 | if (!nr) { |
2656 | if (!PageError(page)) | |
2657 | SetPageUptodate(page); | |
2658 | unlock_page(page); | |
2659 | } | |
2660 | return 0; | |
2661 | } | |
2662 | ||
2663 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, | |
8ddc7d9c | 2664 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
2665 | { |
2666 | struct bio *bio = NULL; | |
c8b97818 | 2667 | unsigned long bio_flags = 0; |
d1310b2e CM |
2668 | int ret; |
2669 | ||
8ddc7d9c | 2670 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, |
c8b97818 | 2671 | &bio_flags); |
d1310b2e | 2672 | if (bio) |
8ddc7d9c | 2673 | ret = submit_one_bio(READ, bio, mirror_num, bio_flags); |
d1310b2e CM |
2674 | return ret; |
2675 | } | |
d1310b2e | 2676 | |
11c8349b CM |
2677 | static noinline void update_nr_written(struct page *page, |
2678 | struct writeback_control *wbc, | |
2679 | unsigned long nr_written) | |
2680 | { | |
2681 | wbc->nr_to_write -= nr_written; | |
2682 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && | |
2683 | wbc->range_start == 0 && wbc->range_end == LLONG_MAX)) | |
2684 | page->mapping->writeback_index = page->index + nr_written; | |
2685 | } | |
2686 | ||
d1310b2e CM |
2687 | /* |
2688 | * the writepage semantics are similar to regular writepage. extent | |
2689 | * records are inserted to lock ranges in the tree, and as dirty areas | |
2690 | * are found, they are marked writeback. Then the lock bits are removed | |
2691 | * and the end_io handler clears the writeback ranges | |
2692 | */ | |
2693 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
2694 | void *data) | |
2695 | { | |
2696 | struct inode *inode = page->mapping->host; | |
2697 | struct extent_page_data *epd = data; | |
2698 | struct extent_io_tree *tree = epd->tree; | |
2699 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2700 | u64 delalloc_start; | |
2701 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2702 | u64 end; | |
2703 | u64 cur = start; | |
2704 | u64 extent_offset; | |
2705 | u64 last_byte = i_size_read(inode); | |
2706 | u64 block_start; | |
2707 | u64 iosize; | |
2708 | sector_t sector; | |
2c64c53d | 2709 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
2710 | struct extent_map *em; |
2711 | struct block_device *bdev; | |
2712 | int ret; | |
2713 | int nr = 0; | |
7f3c74fb | 2714 | size_t pg_offset = 0; |
d1310b2e CM |
2715 | size_t blocksize; |
2716 | loff_t i_size = i_size_read(inode); | |
2717 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
2718 | u64 nr_delalloc; | |
2719 | u64 delalloc_end; | |
c8b97818 CM |
2720 | int page_started; |
2721 | int compressed; | |
ffbd517d | 2722 | int write_flags; |
771ed689 | 2723 | unsigned long nr_written = 0; |
9e487107 | 2724 | bool fill_delalloc = true; |
d1310b2e | 2725 | |
ffbd517d | 2726 | if (wbc->sync_mode == WB_SYNC_ALL) |
721a9602 | 2727 | write_flags = WRITE_SYNC; |
ffbd517d CM |
2728 | else |
2729 | write_flags = WRITE; | |
2730 | ||
1abe9b8a | 2731 | trace___extent_writepage(page, inode, wbc); |
2732 | ||
d1310b2e | 2733 | WARN_ON(!PageLocked(page)); |
bf0da8c1 CM |
2734 | |
2735 | ClearPageError(page); | |
2736 | ||
7f3c74fb | 2737 | pg_offset = i_size & (PAGE_CACHE_SIZE - 1); |
211c17f5 | 2738 | if (page->index > end_index || |
7f3c74fb | 2739 | (page->index == end_index && !pg_offset)) { |
39be25cd | 2740 | page->mapping->a_ops->invalidatepage(page, 0); |
d1310b2e CM |
2741 | unlock_page(page); |
2742 | return 0; | |
2743 | } | |
2744 | ||
2745 | if (page->index == end_index) { | |
2746 | char *userpage; | |
2747 | ||
d1310b2e | 2748 | userpage = kmap_atomic(page, KM_USER0); |
7f3c74fb CM |
2749 | memset(userpage + pg_offset, 0, |
2750 | PAGE_CACHE_SIZE - pg_offset); | |
d1310b2e | 2751 | kunmap_atomic(userpage, KM_USER0); |
211c17f5 | 2752 | flush_dcache_page(page); |
d1310b2e | 2753 | } |
7f3c74fb | 2754 | pg_offset = 0; |
d1310b2e CM |
2755 | |
2756 | set_page_extent_mapped(page); | |
2757 | ||
9e487107 JB |
2758 | if (!tree->ops || !tree->ops->fill_delalloc) |
2759 | fill_delalloc = false; | |
2760 | ||
d1310b2e CM |
2761 | delalloc_start = start; |
2762 | delalloc_end = 0; | |
c8b97818 | 2763 | page_started = 0; |
9e487107 | 2764 | if (!epd->extent_locked && fill_delalloc) { |
f85d7d6c | 2765 | u64 delalloc_to_write = 0; |
11c8349b CM |
2766 | /* |
2767 | * make sure the wbc mapping index is at least updated | |
2768 | * to this page. | |
2769 | */ | |
2770 | update_nr_written(page, wbc, 0); | |
2771 | ||
d397712b | 2772 | while (delalloc_end < page_end) { |
771ed689 | 2773 | nr_delalloc = find_lock_delalloc_range(inode, tree, |
c8b97818 CM |
2774 | page, |
2775 | &delalloc_start, | |
d1310b2e CM |
2776 | &delalloc_end, |
2777 | 128 * 1024 * 1024); | |
771ed689 CM |
2778 | if (nr_delalloc == 0) { |
2779 | delalloc_start = delalloc_end + 1; | |
2780 | continue; | |
2781 | } | |
2782 | tree->ops->fill_delalloc(inode, page, delalloc_start, | |
2783 | delalloc_end, &page_started, | |
2784 | &nr_written); | |
f85d7d6c CM |
2785 | /* |
2786 | * delalloc_end is already one less than the total | |
2787 | * length, so we don't subtract one from | |
2788 | * PAGE_CACHE_SIZE | |
2789 | */ | |
2790 | delalloc_to_write += (delalloc_end - delalloc_start + | |
2791 | PAGE_CACHE_SIZE) >> | |
2792 | PAGE_CACHE_SHIFT; | |
d1310b2e | 2793 | delalloc_start = delalloc_end + 1; |
d1310b2e | 2794 | } |
f85d7d6c CM |
2795 | if (wbc->nr_to_write < delalloc_to_write) { |
2796 | int thresh = 8192; | |
2797 | ||
2798 | if (delalloc_to_write < thresh * 2) | |
2799 | thresh = delalloc_to_write; | |
2800 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
2801 | thresh); | |
2802 | } | |
c8b97818 | 2803 | |
771ed689 CM |
2804 | /* did the fill delalloc function already unlock and start |
2805 | * the IO? | |
2806 | */ | |
2807 | if (page_started) { | |
2808 | ret = 0; | |
11c8349b CM |
2809 | /* |
2810 | * we've unlocked the page, so we can't update | |
2811 | * the mapping's writeback index, just update | |
2812 | * nr_to_write. | |
2813 | */ | |
2814 | wbc->nr_to_write -= nr_written; | |
2815 | goto done_unlocked; | |
771ed689 | 2816 | } |
c8b97818 | 2817 | } |
247e743c | 2818 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
2819 | ret = tree->ops->writepage_start_hook(page, start, |
2820 | page_end); | |
247e743c | 2821 | if (ret == -EAGAIN) { |
247e743c | 2822 | redirty_page_for_writepage(wbc, page); |
11c8349b | 2823 | update_nr_written(page, wbc, nr_written); |
247e743c | 2824 | unlock_page(page); |
771ed689 | 2825 | ret = 0; |
11c8349b | 2826 | goto done_unlocked; |
247e743c CM |
2827 | } |
2828 | } | |
2829 | ||
11c8349b CM |
2830 | /* |
2831 | * we don't want to touch the inode after unlocking the page, | |
2832 | * so we update the mapping writeback index now | |
2833 | */ | |
2834 | update_nr_written(page, wbc, nr_written + 1); | |
771ed689 | 2835 | |
d1310b2e | 2836 | end = page_end; |
d1310b2e | 2837 | if (last_byte <= start) { |
e6dcd2dc CM |
2838 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2839 | tree->ops->writepage_end_io_hook(page, start, | |
2840 | page_end, NULL, 1); | |
d1310b2e CM |
2841 | goto done; |
2842 | } | |
2843 | ||
d1310b2e CM |
2844 | blocksize = inode->i_sb->s_blocksize; |
2845 | ||
2846 | while (cur <= end) { | |
2847 | if (cur >= last_byte) { | |
e6dcd2dc CM |
2848 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2849 | tree->ops->writepage_end_io_hook(page, cur, | |
2850 | page_end, NULL, 1); | |
d1310b2e CM |
2851 | break; |
2852 | } | |
7f3c74fb | 2853 | em = epd->get_extent(inode, page, pg_offset, cur, |
d1310b2e | 2854 | end - cur + 1, 1); |
c704005d | 2855 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e CM |
2856 | SetPageError(page); |
2857 | break; | |
2858 | } | |
2859 | ||
2860 | extent_offset = cur - em->start; | |
2861 | BUG_ON(extent_map_end(em) <= cur); | |
2862 | BUG_ON(end < cur); | |
2863 | iosize = min(extent_map_end(em) - cur, end - cur + 1); | |
2864 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
2865 | sector = (em->block_start + extent_offset) >> 9; | |
2866 | bdev = em->bdev; | |
2867 | block_start = em->block_start; | |
c8b97818 | 2868 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
2869 | free_extent_map(em); |
2870 | em = NULL; | |
2871 | ||
c8b97818 CM |
2872 | /* |
2873 | * compressed and inline extents are written through other | |
2874 | * paths in the FS | |
2875 | */ | |
2876 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 2877 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
2878 | /* |
2879 | * end_io notification does not happen here for | |
2880 | * compressed extents | |
2881 | */ | |
2882 | if (!compressed && tree->ops && | |
2883 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
2884 | tree->ops->writepage_end_io_hook(page, cur, |
2885 | cur + iosize - 1, | |
2886 | NULL, 1); | |
c8b97818 CM |
2887 | else if (compressed) { |
2888 | /* we don't want to end_page_writeback on | |
2889 | * a compressed extent. this happens | |
2890 | * elsewhere | |
2891 | */ | |
2892 | nr++; | |
2893 | } | |
2894 | ||
2895 | cur += iosize; | |
7f3c74fb | 2896 | pg_offset += iosize; |
d1310b2e CM |
2897 | continue; |
2898 | } | |
d1310b2e CM |
2899 | /* leave this out until we have a page_mkwrite call */ |
2900 | if (0 && !test_range_bit(tree, cur, cur + iosize - 1, | |
9655d298 | 2901 | EXTENT_DIRTY, 0, NULL)) { |
d1310b2e | 2902 | cur = cur + iosize; |
7f3c74fb | 2903 | pg_offset += iosize; |
d1310b2e CM |
2904 | continue; |
2905 | } | |
c8b97818 | 2906 | |
d1310b2e CM |
2907 | if (tree->ops && tree->ops->writepage_io_hook) { |
2908 | ret = tree->ops->writepage_io_hook(page, cur, | |
2909 | cur + iosize - 1); | |
2910 | } else { | |
2911 | ret = 0; | |
2912 | } | |
1259ab75 | 2913 | if (ret) { |
d1310b2e | 2914 | SetPageError(page); |
1259ab75 | 2915 | } else { |
d1310b2e | 2916 | unsigned long max_nr = end_index + 1; |
7f3c74fb | 2917 | |
d1310b2e CM |
2918 | set_range_writeback(tree, cur, cur + iosize - 1); |
2919 | if (!PageWriteback(page)) { | |
d397712b CM |
2920 | printk(KERN_ERR "btrfs warning page %lu not " |
2921 | "writeback, cur %llu end %llu\n", | |
2922 | page->index, (unsigned long long)cur, | |
d1310b2e CM |
2923 | (unsigned long long)end); |
2924 | } | |
2925 | ||
ffbd517d CM |
2926 | ret = submit_extent_page(write_flags, tree, page, |
2927 | sector, iosize, pg_offset, | |
2928 | bdev, &epd->bio, max_nr, | |
c8b97818 CM |
2929 | end_bio_extent_writepage, |
2930 | 0, 0, 0); | |
d1310b2e CM |
2931 | if (ret) |
2932 | SetPageError(page); | |
2933 | } | |
2934 | cur = cur + iosize; | |
7f3c74fb | 2935 | pg_offset += iosize; |
d1310b2e CM |
2936 | nr++; |
2937 | } | |
2938 | done: | |
2939 | if (nr == 0) { | |
2940 | /* make sure the mapping tag for page dirty gets cleared */ | |
2941 | set_page_writeback(page); | |
2942 | end_page_writeback(page); | |
2943 | } | |
d1310b2e | 2944 | unlock_page(page); |
771ed689 | 2945 | |
11c8349b CM |
2946 | done_unlocked: |
2947 | ||
2c64c53d CM |
2948 | /* drop our reference on any cached states */ |
2949 | free_extent_state(cached_state); | |
d1310b2e CM |
2950 | return 0; |
2951 | } | |
2952 | ||
d1310b2e | 2953 | /** |
4bef0848 | 2954 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
2955 | * @mapping: address space structure to write |
2956 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
2957 | * @writepage: function called for each page | |
2958 | * @data: data passed to writepage function | |
2959 | * | |
2960 | * If a page is already under I/O, write_cache_pages() skips it, even | |
2961 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
2962 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
2963 | * and msync() need to guarantee that all the data which was dirty at the time | |
2964 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
2965 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
2966 | * existing IO to complete. | |
2967 | */ | |
b2950863 | 2968 | static int extent_write_cache_pages(struct extent_io_tree *tree, |
4bef0848 CM |
2969 | struct address_space *mapping, |
2970 | struct writeback_control *wbc, | |
d2c3f4f6 CM |
2971 | writepage_t writepage, void *data, |
2972 | void (*flush_fn)(void *)) | |
d1310b2e | 2973 | { |
d1310b2e CM |
2974 | int ret = 0; |
2975 | int done = 0; | |
f85d7d6c | 2976 | int nr_to_write_done = 0; |
d1310b2e CM |
2977 | struct pagevec pvec; |
2978 | int nr_pages; | |
2979 | pgoff_t index; | |
2980 | pgoff_t end; /* Inclusive */ | |
2981 | int scanned = 0; | |
f7aaa06b | 2982 | int tag; |
d1310b2e | 2983 | |
d1310b2e CM |
2984 | pagevec_init(&pvec, 0); |
2985 | if (wbc->range_cyclic) { | |
2986 | index = mapping->writeback_index; /* Start from prev offset */ | |
2987 | end = -1; | |
2988 | } else { | |
2989 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
2990 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
2991 | scanned = 1; |
2992 | } | |
f7aaa06b JB |
2993 | if (wbc->sync_mode == WB_SYNC_ALL) |
2994 | tag = PAGECACHE_TAG_TOWRITE; | |
2995 | else | |
2996 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 2997 | retry: |
f7aaa06b JB |
2998 | if (wbc->sync_mode == WB_SYNC_ALL) |
2999 | tag_pages_for_writeback(mapping, index, end); | |
f85d7d6c | 3000 | while (!done && !nr_to_write_done && (index <= end) && |
f7aaa06b JB |
3001 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
3002 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
d1310b2e CM |
3003 | unsigned i; |
3004 | ||
3005 | scanned = 1; | |
3006 | for (i = 0; i < nr_pages; i++) { | |
3007 | struct page *page = pvec.pages[i]; | |
3008 | ||
3009 | /* | |
3010 | * At this point we hold neither mapping->tree_lock nor | |
3011 | * lock on the page itself: the page may be truncated or | |
3012 | * invalidated (changing page->mapping to NULL), or even | |
3013 | * swizzled back from swapper_space to tmpfs file | |
3014 | * mapping | |
3015 | */ | |
01d658f2 CM |
3016 | if (tree->ops && |
3017 | tree->ops->write_cache_pages_lock_hook) { | |
3018 | tree->ops->write_cache_pages_lock_hook(page, | |
3019 | data, flush_fn); | |
3020 | } else { | |
3021 | if (!trylock_page(page)) { | |
3022 | flush_fn(data); | |
3023 | lock_page(page); | |
3024 | } | |
3025 | } | |
d1310b2e CM |
3026 | |
3027 | if (unlikely(page->mapping != mapping)) { | |
3028 | unlock_page(page); | |
3029 | continue; | |
3030 | } | |
3031 | ||
3032 | if (!wbc->range_cyclic && page->index > end) { | |
3033 | done = 1; | |
3034 | unlock_page(page); | |
3035 | continue; | |
3036 | } | |
3037 | ||
d2c3f4f6 | 3038 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 CM |
3039 | if (PageWriteback(page)) |
3040 | flush_fn(data); | |
d1310b2e | 3041 | wait_on_page_writeback(page); |
d2c3f4f6 | 3042 | } |
d1310b2e CM |
3043 | |
3044 | if (PageWriteback(page) || | |
3045 | !clear_page_dirty_for_io(page)) { | |
3046 | unlock_page(page); | |
3047 | continue; | |
3048 | } | |
3049 | ||
3050 | ret = (*writepage)(page, wbc, data); | |
3051 | ||
3052 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
3053 | unlock_page(page); | |
3054 | ret = 0; | |
3055 | } | |
f85d7d6c | 3056 | if (ret) |
d1310b2e | 3057 | done = 1; |
f85d7d6c CM |
3058 | |
3059 | /* | |
3060 | * the filesystem may choose to bump up nr_to_write. | |
3061 | * We have to make sure to honor the new nr_to_write | |
3062 | * at any time | |
3063 | */ | |
3064 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
3065 | } |
3066 | pagevec_release(&pvec); | |
3067 | cond_resched(); | |
3068 | } | |
3069 | if (!scanned && !done) { | |
3070 | /* | |
3071 | * We hit the last page and there is more work to be done: wrap | |
3072 | * back to the start of the file | |
3073 | */ | |
3074 | scanned = 1; | |
3075 | index = 0; | |
3076 | goto retry; | |
3077 | } | |
d1310b2e CM |
3078 | return ret; |
3079 | } | |
d1310b2e | 3080 | |
ffbd517d | 3081 | static void flush_epd_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 3082 | { |
d2c3f4f6 | 3083 | if (epd->bio) { |
ffbd517d CM |
3084 | if (epd->sync_io) |
3085 | submit_one_bio(WRITE_SYNC, epd->bio, 0, 0); | |
3086 | else | |
3087 | submit_one_bio(WRITE, epd->bio, 0, 0); | |
d2c3f4f6 CM |
3088 | epd->bio = NULL; |
3089 | } | |
3090 | } | |
3091 | ||
ffbd517d CM |
3092 | static noinline void flush_write_bio(void *data) |
3093 | { | |
3094 | struct extent_page_data *epd = data; | |
3095 | flush_epd_write_bio(epd); | |
3096 | } | |
3097 | ||
d1310b2e CM |
3098 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, |
3099 | get_extent_t *get_extent, | |
3100 | struct writeback_control *wbc) | |
3101 | { | |
3102 | int ret; | |
d1310b2e CM |
3103 | struct extent_page_data epd = { |
3104 | .bio = NULL, | |
3105 | .tree = tree, | |
3106 | .get_extent = get_extent, | |
771ed689 | 3107 | .extent_locked = 0, |
ffbd517d | 3108 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e | 3109 | }; |
d1310b2e | 3110 | |
d1310b2e CM |
3111 | ret = __extent_writepage(page, wbc, &epd); |
3112 | ||
ffbd517d | 3113 | flush_epd_write_bio(&epd); |
d1310b2e CM |
3114 | return ret; |
3115 | } | |
d1310b2e | 3116 | |
771ed689 CM |
3117 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, |
3118 | u64 start, u64 end, get_extent_t *get_extent, | |
3119 | int mode) | |
3120 | { | |
3121 | int ret = 0; | |
3122 | struct address_space *mapping = inode->i_mapping; | |
3123 | struct page *page; | |
3124 | unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >> | |
3125 | PAGE_CACHE_SHIFT; | |
3126 | ||
3127 | struct extent_page_data epd = { | |
3128 | .bio = NULL, | |
3129 | .tree = tree, | |
3130 | .get_extent = get_extent, | |
3131 | .extent_locked = 1, | |
ffbd517d | 3132 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
3133 | }; |
3134 | struct writeback_control wbc_writepages = { | |
771ed689 | 3135 | .sync_mode = mode, |
771ed689 CM |
3136 | .nr_to_write = nr_pages * 2, |
3137 | .range_start = start, | |
3138 | .range_end = end + 1, | |
3139 | }; | |
3140 | ||
d397712b | 3141 | while (start <= end) { |
771ed689 CM |
3142 | page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); |
3143 | if (clear_page_dirty_for_io(page)) | |
3144 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
3145 | else { | |
3146 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
3147 | tree->ops->writepage_end_io_hook(page, start, | |
3148 | start + PAGE_CACHE_SIZE - 1, | |
3149 | NULL, 1); | |
3150 | unlock_page(page); | |
3151 | } | |
3152 | page_cache_release(page); | |
3153 | start += PAGE_CACHE_SIZE; | |
3154 | } | |
3155 | ||
ffbd517d | 3156 | flush_epd_write_bio(&epd); |
771ed689 CM |
3157 | return ret; |
3158 | } | |
d1310b2e CM |
3159 | |
3160 | int extent_writepages(struct extent_io_tree *tree, | |
3161 | struct address_space *mapping, | |
3162 | get_extent_t *get_extent, | |
3163 | struct writeback_control *wbc) | |
3164 | { | |
3165 | int ret = 0; | |
3166 | struct extent_page_data epd = { | |
3167 | .bio = NULL, | |
3168 | .tree = tree, | |
3169 | .get_extent = get_extent, | |
771ed689 | 3170 | .extent_locked = 0, |
ffbd517d | 3171 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
3172 | }; |
3173 | ||
4bef0848 | 3174 | ret = extent_write_cache_pages(tree, mapping, wbc, |
d2c3f4f6 CM |
3175 | __extent_writepage, &epd, |
3176 | flush_write_bio); | |
ffbd517d | 3177 | flush_epd_write_bio(&epd); |
d1310b2e CM |
3178 | return ret; |
3179 | } | |
d1310b2e CM |
3180 | |
3181 | int extent_readpages(struct extent_io_tree *tree, | |
3182 | struct address_space *mapping, | |
3183 | struct list_head *pages, unsigned nr_pages, | |
3184 | get_extent_t get_extent) | |
3185 | { | |
3186 | struct bio *bio = NULL; | |
3187 | unsigned page_idx; | |
c8b97818 | 3188 | unsigned long bio_flags = 0; |
d1310b2e | 3189 | |
d1310b2e CM |
3190 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
3191 | struct page *page = list_entry(pages->prev, struct page, lru); | |
3192 | ||
3193 | prefetchw(&page->flags); | |
3194 | list_del(&page->lru); | |
28ecb609 | 3195 | if (!add_to_page_cache_lru(page, mapping, |
43e817a1 | 3196 | page->index, GFP_NOFS)) { |
f188591e | 3197 | __extent_read_full_page(tree, page, get_extent, |
c8b97818 | 3198 | &bio, 0, &bio_flags); |
d1310b2e CM |
3199 | } |
3200 | page_cache_release(page); | |
3201 | } | |
d1310b2e CM |
3202 | BUG_ON(!list_empty(pages)); |
3203 | if (bio) | |
c8b97818 | 3204 | submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
3205 | return 0; |
3206 | } | |
d1310b2e CM |
3207 | |
3208 | /* | |
3209 | * basic invalidatepage code, this waits on any locked or writeback | |
3210 | * ranges corresponding to the page, and then deletes any extent state | |
3211 | * records from the tree | |
3212 | */ | |
3213 | int extent_invalidatepage(struct extent_io_tree *tree, | |
3214 | struct page *page, unsigned long offset) | |
3215 | { | |
2ac55d41 | 3216 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
3217 | u64 start = ((u64)page->index << PAGE_CACHE_SHIFT); |
3218 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
3219 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
3220 | ||
d397712b | 3221 | start += (offset + blocksize - 1) & ~(blocksize - 1); |
d1310b2e CM |
3222 | if (start > end) |
3223 | return 0; | |
3224 | ||
2ac55d41 | 3225 | lock_extent_bits(tree, start, end, 0, &cached_state, GFP_NOFS); |
1edbb734 | 3226 | wait_on_page_writeback(page); |
d1310b2e | 3227 | clear_extent_bit(tree, start, end, |
32c00aff JB |
3228 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
3229 | EXTENT_DO_ACCOUNTING, | |
2ac55d41 | 3230 | 1, 1, &cached_state, GFP_NOFS); |
d1310b2e CM |
3231 | return 0; |
3232 | } | |
d1310b2e | 3233 | |
7b13b7b1 CM |
3234 | /* |
3235 | * a helper for releasepage, this tests for areas of the page that | |
3236 | * are locked or under IO and drops the related state bits if it is safe | |
3237 | * to drop the page. | |
3238 | */ | |
3239 | int try_release_extent_state(struct extent_map_tree *map, | |
3240 | struct extent_io_tree *tree, struct page *page, | |
3241 | gfp_t mask) | |
3242 | { | |
3243 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
3244 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
3245 | int ret = 1; | |
3246 | ||
211f90e6 | 3247 | if (test_range_bit(tree, start, end, |
8b62b72b | 3248 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
3249 | ret = 0; |
3250 | else { | |
3251 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
3252 | mask = GFP_NOFS; | |
11ef160f CM |
3253 | /* |
3254 | * at this point we can safely clear everything except the | |
3255 | * locked bit and the nodatasum bit | |
3256 | */ | |
e3f24cc5 | 3257 | ret = clear_extent_bit(tree, start, end, |
11ef160f CM |
3258 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
3259 | 0, 0, NULL, mask); | |
e3f24cc5 CM |
3260 | |
3261 | /* if clear_extent_bit failed for enomem reasons, | |
3262 | * we can't allow the release to continue. | |
3263 | */ | |
3264 | if (ret < 0) | |
3265 | ret = 0; | |
3266 | else | |
3267 | ret = 1; | |
7b13b7b1 CM |
3268 | } |
3269 | return ret; | |
3270 | } | |
7b13b7b1 | 3271 | |
d1310b2e CM |
3272 | /* |
3273 | * a helper for releasepage. As long as there are no locked extents | |
3274 | * in the range corresponding to the page, both state records and extent | |
3275 | * map records are removed | |
3276 | */ | |
3277 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
3278 | struct extent_io_tree *tree, struct page *page, |
3279 | gfp_t mask) | |
d1310b2e CM |
3280 | { |
3281 | struct extent_map *em; | |
3282 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
3283 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
7b13b7b1 | 3284 | |
70dec807 CM |
3285 | if ((mask & __GFP_WAIT) && |
3286 | page->mapping->host->i_size > 16 * 1024 * 1024) { | |
39b5637f | 3287 | u64 len; |
70dec807 | 3288 | while (start <= end) { |
39b5637f | 3289 | len = end - start + 1; |
890871be | 3290 | write_lock(&map->lock); |
39b5637f | 3291 | em = lookup_extent_mapping(map, start, len); |
c704005d | 3292 | if (IS_ERR_OR_NULL(em)) { |
890871be | 3293 | write_unlock(&map->lock); |
70dec807 CM |
3294 | break; |
3295 | } | |
7f3c74fb CM |
3296 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
3297 | em->start != start) { | |
890871be | 3298 | write_unlock(&map->lock); |
70dec807 CM |
3299 | free_extent_map(em); |
3300 | break; | |
3301 | } | |
3302 | if (!test_range_bit(tree, em->start, | |
3303 | extent_map_end(em) - 1, | |
8b62b72b | 3304 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 3305 | 0, NULL)) { |
70dec807 CM |
3306 | remove_extent_mapping(map, em); |
3307 | /* once for the rb tree */ | |
3308 | free_extent_map(em); | |
3309 | } | |
3310 | start = extent_map_end(em); | |
890871be | 3311 | write_unlock(&map->lock); |
70dec807 CM |
3312 | |
3313 | /* once for us */ | |
d1310b2e CM |
3314 | free_extent_map(em); |
3315 | } | |
d1310b2e | 3316 | } |
7b13b7b1 | 3317 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 3318 | } |
d1310b2e | 3319 | |
ec29ed5b CM |
3320 | /* |
3321 | * helper function for fiemap, which doesn't want to see any holes. | |
3322 | * This maps until we find something past 'last' | |
3323 | */ | |
3324 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
3325 | u64 offset, | |
3326 | u64 last, | |
3327 | get_extent_t *get_extent) | |
3328 | { | |
3329 | u64 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
3330 | struct extent_map *em; | |
3331 | u64 len; | |
3332 | ||
3333 | if (offset >= last) | |
3334 | return NULL; | |
3335 | ||
3336 | while(1) { | |
3337 | len = last - offset; | |
3338 | if (len == 0) | |
3339 | break; | |
3340 | len = (len + sectorsize - 1) & ~(sectorsize - 1); | |
3341 | em = get_extent(inode, NULL, 0, offset, len, 0); | |
c704005d | 3342 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
3343 | return em; |
3344 | ||
3345 | /* if this isn't a hole return it */ | |
3346 | if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) && | |
3347 | em->block_start != EXTENT_MAP_HOLE) { | |
3348 | return em; | |
3349 | } | |
3350 | ||
3351 | /* this is a hole, advance to the next extent */ | |
3352 | offset = extent_map_end(em); | |
3353 | free_extent_map(em); | |
3354 | if (offset >= last) | |
3355 | break; | |
3356 | } | |
3357 | return NULL; | |
3358 | } | |
3359 | ||
1506fcc8 YS |
3360 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
3361 | __u64 start, __u64 len, get_extent_t *get_extent) | |
3362 | { | |
975f84fe | 3363 | int ret = 0; |
1506fcc8 YS |
3364 | u64 off = start; |
3365 | u64 max = start + len; | |
3366 | u32 flags = 0; | |
975f84fe JB |
3367 | u32 found_type; |
3368 | u64 last; | |
ec29ed5b | 3369 | u64 last_for_get_extent = 0; |
1506fcc8 | 3370 | u64 disko = 0; |
ec29ed5b | 3371 | u64 isize = i_size_read(inode); |
975f84fe | 3372 | struct btrfs_key found_key; |
1506fcc8 | 3373 | struct extent_map *em = NULL; |
2ac55d41 | 3374 | struct extent_state *cached_state = NULL; |
975f84fe JB |
3375 | struct btrfs_path *path; |
3376 | struct btrfs_file_extent_item *item; | |
1506fcc8 | 3377 | int end = 0; |
ec29ed5b CM |
3378 | u64 em_start = 0; |
3379 | u64 em_len = 0; | |
3380 | u64 em_end = 0; | |
1506fcc8 | 3381 | unsigned long emflags; |
1506fcc8 YS |
3382 | |
3383 | if (len == 0) | |
3384 | return -EINVAL; | |
3385 | ||
975f84fe JB |
3386 | path = btrfs_alloc_path(); |
3387 | if (!path) | |
3388 | return -ENOMEM; | |
3389 | path->leave_spinning = 1; | |
3390 | ||
4d479cf0 JB |
3391 | start = ALIGN(start, BTRFS_I(inode)->root->sectorsize); |
3392 | len = ALIGN(len, BTRFS_I(inode)->root->sectorsize); | |
3393 | ||
ec29ed5b CM |
3394 | /* |
3395 | * lookup the last file extent. We're not using i_size here | |
3396 | * because there might be preallocation past i_size | |
3397 | */ | |
975f84fe | 3398 | ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root, |
33345d01 | 3399 | path, btrfs_ino(inode), -1, 0); |
975f84fe JB |
3400 | if (ret < 0) { |
3401 | btrfs_free_path(path); | |
3402 | return ret; | |
3403 | } | |
3404 | WARN_ON(!ret); | |
3405 | path->slots[0]--; | |
3406 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3407 | struct btrfs_file_extent_item); | |
3408 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
3409 | found_type = btrfs_key_type(&found_key); | |
3410 | ||
ec29ed5b | 3411 | /* No extents, but there might be delalloc bits */ |
33345d01 | 3412 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 3413 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
3414 | /* have to trust i_size as the end */ |
3415 | last = (u64)-1; | |
3416 | last_for_get_extent = isize; | |
3417 | } else { | |
3418 | /* | |
3419 | * remember the start of the last extent. There are a | |
3420 | * bunch of different factors that go into the length of the | |
3421 | * extent, so its much less complex to remember where it started | |
3422 | */ | |
3423 | last = found_key.offset; | |
3424 | last_for_get_extent = last + 1; | |
975f84fe | 3425 | } |
975f84fe JB |
3426 | btrfs_free_path(path); |
3427 | ||
ec29ed5b CM |
3428 | /* |
3429 | * we might have some extents allocated but more delalloc past those | |
3430 | * extents. so, we trust isize unless the start of the last extent is | |
3431 | * beyond isize | |
3432 | */ | |
3433 | if (last < isize) { | |
3434 | last = (u64)-1; | |
3435 | last_for_get_extent = isize; | |
3436 | } | |
3437 | ||
2ac55d41 JB |
3438 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0, |
3439 | &cached_state, GFP_NOFS); | |
ec29ed5b | 3440 | |
4d479cf0 | 3441 | em = get_extent_skip_holes(inode, start, last_for_get_extent, |
ec29ed5b | 3442 | get_extent); |
1506fcc8 YS |
3443 | if (!em) |
3444 | goto out; | |
3445 | if (IS_ERR(em)) { | |
3446 | ret = PTR_ERR(em); | |
3447 | goto out; | |
3448 | } | |
975f84fe | 3449 | |
1506fcc8 | 3450 | while (!end) { |
ea8efc74 CM |
3451 | u64 offset_in_extent; |
3452 | ||
3453 | /* break if the extent we found is outside the range */ | |
3454 | if (em->start >= max || extent_map_end(em) < off) | |
3455 | break; | |
3456 | ||
3457 | /* | |
3458 | * get_extent may return an extent that starts before our | |
3459 | * requested range. We have to make sure the ranges | |
3460 | * we return to fiemap always move forward and don't | |
3461 | * overlap, so adjust the offsets here | |
3462 | */ | |
3463 | em_start = max(em->start, off); | |
1506fcc8 | 3464 | |
ea8efc74 CM |
3465 | /* |
3466 | * record the offset from the start of the extent | |
3467 | * for adjusting the disk offset below | |
3468 | */ | |
3469 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 3470 | em_end = extent_map_end(em); |
ea8efc74 | 3471 | em_len = em_end - em_start; |
ec29ed5b | 3472 | emflags = em->flags; |
1506fcc8 YS |
3473 | disko = 0; |
3474 | flags = 0; | |
3475 | ||
ea8efc74 CM |
3476 | /* |
3477 | * bump off for our next call to get_extent | |
3478 | */ | |
3479 | off = extent_map_end(em); | |
3480 | if (off >= max) | |
3481 | end = 1; | |
3482 | ||
93dbfad7 | 3483 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
3484 | end = 1; |
3485 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 3486 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
3487 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
3488 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 3489 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
3490 | flags |= (FIEMAP_EXTENT_DELALLOC | |
3491 | FIEMAP_EXTENT_UNKNOWN); | |
93dbfad7 | 3492 | } else { |
ea8efc74 | 3493 | disko = em->block_start + offset_in_extent; |
1506fcc8 YS |
3494 | } |
3495 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
3496 | flags |= FIEMAP_EXTENT_ENCODED; | |
3497 | ||
1506fcc8 YS |
3498 | free_extent_map(em); |
3499 | em = NULL; | |
ec29ed5b CM |
3500 | if ((em_start >= last) || em_len == (u64)-1 || |
3501 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
3502 | flags |= FIEMAP_EXTENT_LAST; |
3503 | end = 1; | |
3504 | } | |
3505 | ||
ec29ed5b CM |
3506 | /* now scan forward to see if this is really the last extent. */ |
3507 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
3508 | get_extent); | |
3509 | if (IS_ERR(em)) { | |
3510 | ret = PTR_ERR(em); | |
3511 | goto out; | |
3512 | } | |
3513 | if (!em) { | |
975f84fe JB |
3514 | flags |= FIEMAP_EXTENT_LAST; |
3515 | end = 1; | |
3516 | } | |
ec29ed5b CM |
3517 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, |
3518 | em_len, flags); | |
3519 | if (ret) | |
3520 | goto out_free; | |
1506fcc8 YS |
3521 | } |
3522 | out_free: | |
3523 | free_extent_map(em); | |
3524 | out: | |
2ac55d41 JB |
3525 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len, |
3526 | &cached_state, GFP_NOFS); | |
1506fcc8 YS |
3527 | return ret; |
3528 | } | |
3529 | ||
4a54c8c1 | 3530 | inline struct page *extent_buffer_page(struct extent_buffer *eb, |
d1310b2e CM |
3531 | unsigned long i) |
3532 | { | |
3533 | struct page *p; | |
3534 | struct address_space *mapping; | |
3535 | ||
3536 | if (i == 0) | |
3537 | return eb->first_page; | |
3538 | i += eb->start >> PAGE_CACHE_SHIFT; | |
3539 | mapping = eb->first_page->mapping; | |
33958dc6 CM |
3540 | if (!mapping) |
3541 | return NULL; | |
0ee0fda0 SW |
3542 | |
3543 | /* | |
3544 | * extent_buffer_page is only called after pinning the page | |
3545 | * by increasing the reference count. So we know the page must | |
3546 | * be in the radix tree. | |
3547 | */ | |
0ee0fda0 | 3548 | rcu_read_lock(); |
d1310b2e | 3549 | p = radix_tree_lookup(&mapping->page_tree, i); |
0ee0fda0 | 3550 | rcu_read_unlock(); |
2b1f55b0 | 3551 | |
d1310b2e CM |
3552 | return p; |
3553 | } | |
3554 | ||
4a54c8c1 | 3555 | inline unsigned long num_extent_pages(u64 start, u64 len) |
728131d8 | 3556 | { |
6af118ce CM |
3557 | return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) - |
3558 | (start >> PAGE_CACHE_SHIFT); | |
728131d8 CM |
3559 | } |
3560 | ||
d1310b2e CM |
3561 | static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree, |
3562 | u64 start, | |
3563 | unsigned long len, | |
3564 | gfp_t mask) | |
3565 | { | |
3566 | struct extent_buffer *eb = NULL; | |
3935127c | 3567 | #if LEAK_DEBUG |
2d2ae547 | 3568 | unsigned long flags; |
4bef0848 | 3569 | #endif |
d1310b2e | 3570 | |
d1310b2e | 3571 | eb = kmem_cache_zalloc(extent_buffer_cache, mask); |
91ca338d TI |
3572 | if (eb == NULL) |
3573 | return NULL; | |
d1310b2e CM |
3574 | eb->start = start; |
3575 | eb->len = len; | |
bd681513 CM |
3576 | rwlock_init(&eb->lock); |
3577 | atomic_set(&eb->write_locks, 0); | |
3578 | atomic_set(&eb->read_locks, 0); | |
3579 | atomic_set(&eb->blocking_readers, 0); | |
3580 | atomic_set(&eb->blocking_writers, 0); | |
3581 | atomic_set(&eb->spinning_readers, 0); | |
3582 | atomic_set(&eb->spinning_writers, 0); | |
5b25f70f | 3583 | eb->lock_nested = 0; |
bd681513 CM |
3584 | init_waitqueue_head(&eb->write_lock_wq); |
3585 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 3586 | |
3935127c | 3587 | #if LEAK_DEBUG |
2d2ae547 CM |
3588 | spin_lock_irqsave(&leak_lock, flags); |
3589 | list_add(&eb->leak_list, &buffers); | |
3590 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 3591 | #endif |
d1310b2e CM |
3592 | atomic_set(&eb->refs, 1); |
3593 | ||
3594 | return eb; | |
3595 | } | |
3596 | ||
3597 | static void __free_extent_buffer(struct extent_buffer *eb) | |
3598 | { | |
3935127c | 3599 | #if LEAK_DEBUG |
2d2ae547 CM |
3600 | unsigned long flags; |
3601 | spin_lock_irqsave(&leak_lock, flags); | |
3602 | list_del(&eb->leak_list); | |
3603 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 3604 | #endif |
d1310b2e CM |
3605 | kmem_cache_free(extent_buffer_cache, eb); |
3606 | } | |
3607 | ||
897ca6e9 MX |
3608 | /* |
3609 | * Helper for releasing extent buffer page. | |
3610 | */ | |
3611 | static void btrfs_release_extent_buffer_page(struct extent_buffer *eb, | |
3612 | unsigned long start_idx) | |
3613 | { | |
3614 | unsigned long index; | |
3615 | struct page *page; | |
3616 | ||
3617 | if (!eb->first_page) | |
3618 | return; | |
3619 | ||
3620 | index = num_extent_pages(eb->start, eb->len); | |
3621 | if (start_idx >= index) | |
3622 | return; | |
3623 | ||
3624 | do { | |
3625 | index--; | |
3626 | page = extent_buffer_page(eb, index); | |
3627 | if (page) | |
3628 | page_cache_release(page); | |
3629 | } while (index != start_idx); | |
3630 | } | |
3631 | ||
3632 | /* | |
3633 | * Helper for releasing the extent buffer. | |
3634 | */ | |
3635 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
3636 | { | |
3637 | btrfs_release_extent_buffer_page(eb, 0); | |
3638 | __free_extent_buffer(eb); | |
3639 | } | |
3640 | ||
d1310b2e CM |
3641 | struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree, |
3642 | u64 start, unsigned long len, | |
ba144192 | 3643 | struct page *page0) |
d1310b2e CM |
3644 | { |
3645 | unsigned long num_pages = num_extent_pages(start, len); | |
3646 | unsigned long i; | |
3647 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
3648 | struct extent_buffer *eb; | |
6af118ce | 3649 | struct extent_buffer *exists = NULL; |
d1310b2e CM |
3650 | struct page *p; |
3651 | struct address_space *mapping = tree->mapping; | |
3652 | int uptodate = 1; | |
19fe0a8b | 3653 | int ret; |
d1310b2e | 3654 | |
19fe0a8b MX |
3655 | rcu_read_lock(); |
3656 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); | |
3657 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
3658 | rcu_read_unlock(); | |
0f9dd46c | 3659 | mark_page_accessed(eb->first_page); |
6af118ce CM |
3660 | return eb; |
3661 | } | |
19fe0a8b | 3662 | rcu_read_unlock(); |
6af118ce | 3663 | |
ba144192 | 3664 | eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS); |
2b114d1d | 3665 | if (!eb) |
d1310b2e CM |
3666 | return NULL; |
3667 | ||
d1310b2e CM |
3668 | if (page0) { |
3669 | eb->first_page = page0; | |
3670 | i = 1; | |
3671 | index++; | |
3672 | page_cache_get(page0); | |
3673 | mark_page_accessed(page0); | |
3674 | set_page_extent_mapped(page0); | |
d1310b2e | 3675 | set_page_extent_head(page0, len); |
f188591e | 3676 | uptodate = PageUptodate(page0); |
d1310b2e CM |
3677 | } else { |
3678 | i = 0; | |
3679 | } | |
3680 | for (; i < num_pages; i++, index++) { | |
a6591715 | 3681 | p = find_or_create_page(mapping, index, GFP_NOFS); |
d1310b2e CM |
3682 | if (!p) { |
3683 | WARN_ON(1); | |
6af118ce | 3684 | goto free_eb; |
d1310b2e CM |
3685 | } |
3686 | set_page_extent_mapped(p); | |
3687 | mark_page_accessed(p); | |
3688 | if (i == 0) { | |
3689 | eb->first_page = p; | |
3690 | set_page_extent_head(p, len); | |
3691 | } else { | |
3692 | set_page_private(p, EXTENT_PAGE_PRIVATE); | |
3693 | } | |
3694 | if (!PageUptodate(p)) | |
3695 | uptodate = 0; | |
eb14ab8e CM |
3696 | |
3697 | /* | |
3698 | * see below about how we avoid a nasty race with release page | |
3699 | * and why we unlock later | |
3700 | */ | |
3701 | if (i != 0) | |
3702 | unlock_page(p); | |
d1310b2e CM |
3703 | } |
3704 | if (uptodate) | |
b4ce94de | 3705 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 3706 | |
19fe0a8b MX |
3707 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); |
3708 | if (ret) | |
3709 | goto free_eb; | |
3710 | ||
6af118ce | 3711 | spin_lock(&tree->buffer_lock); |
19fe0a8b MX |
3712 | ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb); |
3713 | if (ret == -EEXIST) { | |
3714 | exists = radix_tree_lookup(&tree->buffer, | |
3715 | start >> PAGE_CACHE_SHIFT); | |
6af118ce CM |
3716 | /* add one reference for the caller */ |
3717 | atomic_inc(&exists->refs); | |
3718 | spin_unlock(&tree->buffer_lock); | |
19fe0a8b | 3719 | radix_tree_preload_end(); |
6af118ce CM |
3720 | goto free_eb; |
3721 | } | |
6af118ce CM |
3722 | /* add one reference for the tree */ |
3723 | atomic_inc(&eb->refs); | |
f044ba78 | 3724 | spin_unlock(&tree->buffer_lock); |
19fe0a8b | 3725 | radix_tree_preload_end(); |
eb14ab8e CM |
3726 | |
3727 | /* | |
3728 | * there is a race where release page may have | |
3729 | * tried to find this extent buffer in the radix | |
3730 | * but failed. It will tell the VM it is safe to | |
3731 | * reclaim the, and it will clear the page private bit. | |
3732 | * We must make sure to set the page private bit properly | |
3733 | * after the extent buffer is in the radix tree so | |
3734 | * it doesn't get lost | |
3735 | */ | |
3736 | set_page_extent_mapped(eb->first_page); | |
3737 | set_page_extent_head(eb->first_page, eb->len); | |
3738 | if (!page0) | |
3739 | unlock_page(eb->first_page); | |
d1310b2e CM |
3740 | return eb; |
3741 | ||
6af118ce | 3742 | free_eb: |
eb14ab8e CM |
3743 | if (eb->first_page && !page0) |
3744 | unlock_page(eb->first_page); | |
3745 | ||
d1310b2e | 3746 | if (!atomic_dec_and_test(&eb->refs)) |
6af118ce | 3747 | return exists; |
897ca6e9 | 3748 | btrfs_release_extent_buffer(eb); |
6af118ce | 3749 | return exists; |
d1310b2e | 3750 | } |
d1310b2e CM |
3751 | |
3752 | struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree, | |
f09d1f60 | 3753 | u64 start, unsigned long len) |
d1310b2e | 3754 | { |
d1310b2e | 3755 | struct extent_buffer *eb; |
d1310b2e | 3756 | |
19fe0a8b MX |
3757 | rcu_read_lock(); |
3758 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); | |
3759 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
3760 | rcu_read_unlock(); | |
0f9dd46c | 3761 | mark_page_accessed(eb->first_page); |
19fe0a8b MX |
3762 | return eb; |
3763 | } | |
3764 | rcu_read_unlock(); | |
0f9dd46c | 3765 | |
19fe0a8b | 3766 | return NULL; |
d1310b2e | 3767 | } |
d1310b2e CM |
3768 | |
3769 | void free_extent_buffer(struct extent_buffer *eb) | |
3770 | { | |
d1310b2e CM |
3771 | if (!eb) |
3772 | return; | |
3773 | ||
3774 | if (!atomic_dec_and_test(&eb->refs)) | |
3775 | return; | |
3776 | ||
6af118ce | 3777 | WARN_ON(1); |
d1310b2e | 3778 | } |
d1310b2e CM |
3779 | |
3780 | int clear_extent_buffer_dirty(struct extent_io_tree *tree, | |
3781 | struct extent_buffer *eb) | |
3782 | { | |
d1310b2e CM |
3783 | unsigned long i; |
3784 | unsigned long num_pages; | |
3785 | struct page *page; | |
3786 | ||
d1310b2e CM |
3787 | num_pages = num_extent_pages(eb->start, eb->len); |
3788 | ||
3789 | for (i = 0; i < num_pages; i++) { | |
3790 | page = extent_buffer_page(eb, i); | |
b9473439 | 3791 | if (!PageDirty(page)) |
d2c3f4f6 CM |
3792 | continue; |
3793 | ||
a61e6f29 | 3794 | lock_page(page); |
eb14ab8e CM |
3795 | WARN_ON(!PagePrivate(page)); |
3796 | ||
3797 | set_page_extent_mapped(page); | |
d1310b2e CM |
3798 | if (i == 0) |
3799 | set_page_extent_head(page, eb->len); | |
d1310b2e | 3800 | |
d1310b2e | 3801 | clear_page_dirty_for_io(page); |
0ee0fda0 | 3802 | spin_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
3803 | if (!PageDirty(page)) { |
3804 | radix_tree_tag_clear(&page->mapping->page_tree, | |
3805 | page_index(page), | |
3806 | PAGECACHE_TAG_DIRTY); | |
3807 | } | |
0ee0fda0 | 3808 | spin_unlock_irq(&page->mapping->tree_lock); |
bf0da8c1 | 3809 | ClearPageError(page); |
a61e6f29 | 3810 | unlock_page(page); |
d1310b2e CM |
3811 | } |
3812 | return 0; | |
3813 | } | |
d1310b2e | 3814 | |
d1310b2e CM |
3815 | int set_extent_buffer_dirty(struct extent_io_tree *tree, |
3816 | struct extent_buffer *eb) | |
3817 | { | |
3818 | unsigned long i; | |
3819 | unsigned long num_pages; | |
b9473439 | 3820 | int was_dirty = 0; |
d1310b2e | 3821 | |
b9473439 | 3822 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
d1310b2e | 3823 | num_pages = num_extent_pages(eb->start, eb->len); |
b9473439 | 3824 | for (i = 0; i < num_pages; i++) |
d1310b2e | 3825 | __set_page_dirty_nobuffers(extent_buffer_page(eb, i)); |
b9473439 | 3826 | return was_dirty; |
d1310b2e | 3827 | } |
d1310b2e | 3828 | |
19b6caf4 CM |
3829 | static int __eb_straddles_pages(u64 start, u64 len) |
3830 | { | |
3831 | if (len < PAGE_CACHE_SIZE) | |
3832 | return 1; | |
3833 | if (start & (PAGE_CACHE_SIZE - 1)) | |
3834 | return 1; | |
3835 | if ((start + len) & (PAGE_CACHE_SIZE - 1)) | |
3836 | return 1; | |
3837 | return 0; | |
3838 | } | |
3839 | ||
3840 | static int eb_straddles_pages(struct extent_buffer *eb) | |
3841 | { | |
3842 | return __eb_straddles_pages(eb->start, eb->len); | |
3843 | } | |
3844 | ||
1259ab75 | 3845 | int clear_extent_buffer_uptodate(struct extent_io_tree *tree, |
2ac55d41 JB |
3846 | struct extent_buffer *eb, |
3847 | struct extent_state **cached_state) | |
1259ab75 CM |
3848 | { |
3849 | unsigned long i; | |
3850 | struct page *page; | |
3851 | unsigned long num_pages; | |
3852 | ||
3853 | num_pages = num_extent_pages(eb->start, eb->len); | |
b4ce94de | 3854 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
1259ab75 | 3855 | |
19b6caf4 CM |
3856 | if (eb_straddles_pages(eb)) { |
3857 | clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1, | |
3858 | cached_state, GFP_NOFS); | |
3859 | } | |
1259ab75 CM |
3860 | for (i = 0; i < num_pages; i++) { |
3861 | page = extent_buffer_page(eb, i); | |
33958dc6 CM |
3862 | if (page) |
3863 | ClearPageUptodate(page); | |
1259ab75 CM |
3864 | } |
3865 | return 0; | |
3866 | } | |
3867 | ||
d1310b2e CM |
3868 | int set_extent_buffer_uptodate(struct extent_io_tree *tree, |
3869 | struct extent_buffer *eb) | |
3870 | { | |
3871 | unsigned long i; | |
3872 | struct page *page; | |
3873 | unsigned long num_pages; | |
3874 | ||
3875 | num_pages = num_extent_pages(eb->start, eb->len); | |
3876 | ||
19b6caf4 CM |
3877 | if (eb_straddles_pages(eb)) { |
3878 | set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1, | |
3879 | NULL, GFP_NOFS); | |
3880 | } | |
d1310b2e CM |
3881 | for (i = 0; i < num_pages; i++) { |
3882 | page = extent_buffer_page(eb, i); | |
3883 | if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) || | |
3884 | ((i == num_pages - 1) && | |
3885 | ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) { | |
3886 | check_page_uptodate(tree, page); | |
3887 | continue; | |
3888 | } | |
3889 | SetPageUptodate(page); | |
3890 | } | |
3891 | return 0; | |
3892 | } | |
d1310b2e | 3893 | |
ce9adaa5 CM |
3894 | int extent_range_uptodate(struct extent_io_tree *tree, |
3895 | u64 start, u64 end) | |
3896 | { | |
3897 | struct page *page; | |
3898 | int ret; | |
3899 | int pg_uptodate = 1; | |
3900 | int uptodate; | |
3901 | unsigned long index; | |
3902 | ||
19b6caf4 CM |
3903 | if (__eb_straddles_pages(start, end - start + 1)) { |
3904 | ret = test_range_bit(tree, start, end, | |
3905 | EXTENT_UPTODATE, 1, NULL); | |
3906 | if (ret) | |
3907 | return 1; | |
3908 | } | |
d397712b | 3909 | while (start <= end) { |
ce9adaa5 CM |
3910 | index = start >> PAGE_CACHE_SHIFT; |
3911 | page = find_get_page(tree->mapping, index); | |
3912 | uptodate = PageUptodate(page); | |
3913 | page_cache_release(page); | |
3914 | if (!uptodate) { | |
3915 | pg_uptodate = 0; | |
3916 | break; | |
3917 | } | |
3918 | start += PAGE_CACHE_SIZE; | |
3919 | } | |
3920 | return pg_uptodate; | |
3921 | } | |
3922 | ||
d1310b2e | 3923 | int extent_buffer_uptodate(struct extent_io_tree *tree, |
2ac55d41 JB |
3924 | struct extent_buffer *eb, |
3925 | struct extent_state *cached_state) | |
d1310b2e | 3926 | { |
728131d8 | 3927 | int ret = 0; |
ce9adaa5 CM |
3928 | unsigned long num_pages; |
3929 | unsigned long i; | |
728131d8 CM |
3930 | struct page *page; |
3931 | int pg_uptodate = 1; | |
3932 | ||
b4ce94de | 3933 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
4235298e | 3934 | return 1; |
728131d8 | 3935 | |
19b6caf4 CM |
3936 | if (eb_straddles_pages(eb)) { |
3937 | ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1, | |
3938 | EXTENT_UPTODATE, 1, cached_state); | |
3939 | if (ret) | |
3940 | return ret; | |
3941 | } | |
728131d8 CM |
3942 | |
3943 | num_pages = num_extent_pages(eb->start, eb->len); | |
3944 | for (i = 0; i < num_pages; i++) { | |
3945 | page = extent_buffer_page(eb, i); | |
3946 | if (!PageUptodate(page)) { | |
3947 | pg_uptodate = 0; | |
3948 | break; | |
3949 | } | |
3950 | } | |
4235298e | 3951 | return pg_uptodate; |
d1310b2e | 3952 | } |
d1310b2e CM |
3953 | |
3954 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
bb82ab88 | 3955 | struct extent_buffer *eb, u64 start, int wait, |
f188591e | 3956 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3957 | { |
3958 | unsigned long i; | |
3959 | unsigned long start_i; | |
3960 | struct page *page; | |
3961 | int err; | |
3962 | int ret = 0; | |
ce9adaa5 CM |
3963 | int locked_pages = 0; |
3964 | int all_uptodate = 1; | |
3965 | int inc_all_pages = 0; | |
d1310b2e | 3966 | unsigned long num_pages; |
a86c12c7 | 3967 | struct bio *bio = NULL; |
c8b97818 | 3968 | unsigned long bio_flags = 0; |
a86c12c7 | 3969 | |
b4ce94de | 3970 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
3971 | return 0; |
3972 | ||
19b6caf4 CM |
3973 | if (eb_straddles_pages(eb)) { |
3974 | if (test_range_bit(tree, eb->start, eb->start + eb->len - 1, | |
3975 | EXTENT_UPTODATE, 1, NULL)) { | |
3976 | return 0; | |
3977 | } | |
d1310b2e CM |
3978 | } |
3979 | ||
3980 | if (start) { | |
3981 | WARN_ON(start < eb->start); | |
3982 | start_i = (start >> PAGE_CACHE_SHIFT) - | |
3983 | (eb->start >> PAGE_CACHE_SHIFT); | |
3984 | } else { | |
3985 | start_i = 0; | |
3986 | } | |
3987 | ||
3988 | num_pages = num_extent_pages(eb->start, eb->len); | |
3989 | for (i = start_i; i < num_pages; i++) { | |
3990 | page = extent_buffer_page(eb, i); | |
bb82ab88 | 3991 | if (wait == WAIT_NONE) { |
2db04966 | 3992 | if (!trylock_page(page)) |
ce9adaa5 | 3993 | goto unlock_exit; |
d1310b2e CM |
3994 | } else { |
3995 | lock_page(page); | |
3996 | } | |
ce9adaa5 | 3997 | locked_pages++; |
d397712b | 3998 | if (!PageUptodate(page)) |
ce9adaa5 | 3999 | all_uptodate = 0; |
ce9adaa5 CM |
4000 | } |
4001 | if (all_uptodate) { | |
4002 | if (start_i == 0) | |
b4ce94de | 4003 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
4004 | goto unlock_exit; |
4005 | } | |
4006 | ||
4007 | for (i = start_i; i < num_pages; i++) { | |
4008 | page = extent_buffer_page(eb, i); | |
eb14ab8e CM |
4009 | |
4010 | WARN_ON(!PagePrivate(page)); | |
4011 | ||
4012 | set_page_extent_mapped(page); | |
4013 | if (i == 0) | |
4014 | set_page_extent_head(page, eb->len); | |
4015 | ||
ce9adaa5 CM |
4016 | if (inc_all_pages) |
4017 | page_cache_get(page); | |
4018 | if (!PageUptodate(page)) { | |
4019 | if (start_i == 0) | |
4020 | inc_all_pages = 1; | |
f188591e | 4021 | ClearPageError(page); |
a86c12c7 | 4022 | err = __extent_read_full_page(tree, page, |
f188591e | 4023 | get_extent, &bio, |
c8b97818 | 4024 | mirror_num, &bio_flags); |
d397712b | 4025 | if (err) |
d1310b2e | 4026 | ret = err; |
d1310b2e CM |
4027 | } else { |
4028 | unlock_page(page); | |
4029 | } | |
4030 | } | |
4031 | ||
a86c12c7 | 4032 | if (bio) |
c8b97818 | 4033 | submit_one_bio(READ, bio, mirror_num, bio_flags); |
a86c12c7 | 4034 | |
bb82ab88 | 4035 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 4036 | return ret; |
d397712b | 4037 | |
d1310b2e CM |
4038 | for (i = start_i; i < num_pages; i++) { |
4039 | page = extent_buffer_page(eb, i); | |
4040 | wait_on_page_locked(page); | |
d397712b | 4041 | if (!PageUptodate(page)) |
d1310b2e | 4042 | ret = -EIO; |
d1310b2e | 4043 | } |
d397712b | 4044 | |
d1310b2e | 4045 | if (!ret) |
b4ce94de | 4046 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 4047 | return ret; |
ce9adaa5 CM |
4048 | |
4049 | unlock_exit: | |
4050 | i = start_i; | |
d397712b | 4051 | while (locked_pages > 0) { |
ce9adaa5 CM |
4052 | page = extent_buffer_page(eb, i); |
4053 | i++; | |
4054 | unlock_page(page); | |
4055 | locked_pages--; | |
4056 | } | |
4057 | return ret; | |
d1310b2e | 4058 | } |
d1310b2e CM |
4059 | |
4060 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
4061 | unsigned long start, | |
4062 | unsigned long len) | |
4063 | { | |
4064 | size_t cur; | |
4065 | size_t offset; | |
4066 | struct page *page; | |
4067 | char *kaddr; | |
4068 | char *dst = (char *)dstv; | |
4069 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4070 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
4071 | |
4072 | WARN_ON(start > eb->len); | |
4073 | WARN_ON(start + len > eb->start + eb->len); | |
4074 | ||
4075 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4076 | ||
d397712b | 4077 | while (len > 0) { |
d1310b2e | 4078 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
4079 | |
4080 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
a6591715 | 4081 | kaddr = page_address(page); |
d1310b2e | 4082 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
4083 | |
4084 | dst += cur; | |
4085 | len -= cur; | |
4086 | offset = 0; | |
4087 | i++; | |
4088 | } | |
4089 | } | |
d1310b2e CM |
4090 | |
4091 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
a6591715 | 4092 | unsigned long min_len, char **map, |
d1310b2e | 4093 | unsigned long *map_start, |
a6591715 | 4094 | unsigned long *map_len) |
d1310b2e CM |
4095 | { |
4096 | size_t offset = start & (PAGE_CACHE_SIZE - 1); | |
4097 | char *kaddr; | |
4098 | struct page *p; | |
4099 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4100 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4101 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
4102 | PAGE_CACHE_SHIFT; | |
4103 | ||
4104 | if (i != end_i) | |
4105 | return -EINVAL; | |
4106 | ||
4107 | if (i == 0) { | |
4108 | offset = start_offset; | |
4109 | *map_start = 0; | |
4110 | } else { | |
4111 | offset = 0; | |
4112 | *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; | |
4113 | } | |
d397712b | 4114 | |
d1310b2e | 4115 | if (start + min_len > eb->len) { |
d397712b CM |
4116 | printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, " |
4117 | "wanted %lu %lu\n", (unsigned long long)eb->start, | |
4118 | eb->len, start, min_len); | |
d1310b2e | 4119 | WARN_ON(1); |
85026533 | 4120 | return -EINVAL; |
d1310b2e CM |
4121 | } |
4122 | ||
4123 | p = extent_buffer_page(eb, i); | |
a6591715 | 4124 | kaddr = page_address(p); |
d1310b2e CM |
4125 | *map = kaddr + offset; |
4126 | *map_len = PAGE_CACHE_SIZE - offset; | |
4127 | return 0; | |
4128 | } | |
d1310b2e | 4129 | |
d1310b2e CM |
4130 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, |
4131 | unsigned long start, | |
4132 | unsigned long len) | |
4133 | { | |
4134 | size_t cur; | |
4135 | size_t offset; | |
4136 | struct page *page; | |
4137 | char *kaddr; | |
4138 | char *ptr = (char *)ptrv; | |
4139 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4140 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4141 | int ret = 0; | |
4142 | ||
4143 | WARN_ON(start > eb->len); | |
4144 | WARN_ON(start + len > eb->start + eb->len); | |
4145 | ||
4146 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4147 | ||
d397712b | 4148 | while (len > 0) { |
d1310b2e | 4149 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
4150 | |
4151 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
4152 | ||
a6591715 | 4153 | kaddr = page_address(page); |
d1310b2e | 4154 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
4155 | if (ret) |
4156 | break; | |
4157 | ||
4158 | ptr += cur; | |
4159 | len -= cur; | |
4160 | offset = 0; | |
4161 | i++; | |
4162 | } | |
4163 | return ret; | |
4164 | } | |
d1310b2e CM |
4165 | |
4166 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
4167 | unsigned long start, unsigned long len) | |
4168 | { | |
4169 | size_t cur; | |
4170 | size_t offset; | |
4171 | struct page *page; | |
4172 | char *kaddr; | |
4173 | char *src = (char *)srcv; | |
4174 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4175 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4176 | ||
4177 | WARN_ON(start > eb->len); | |
4178 | WARN_ON(start + len > eb->start + eb->len); | |
4179 | ||
4180 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4181 | ||
d397712b | 4182 | while (len > 0) { |
d1310b2e CM |
4183 | page = extent_buffer_page(eb, i); |
4184 | WARN_ON(!PageUptodate(page)); | |
4185 | ||
4186 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 4187 | kaddr = page_address(page); |
d1310b2e | 4188 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
4189 | |
4190 | src += cur; | |
4191 | len -= cur; | |
4192 | offset = 0; | |
4193 | i++; | |
4194 | } | |
4195 | } | |
d1310b2e CM |
4196 | |
4197 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
4198 | unsigned long start, unsigned long len) | |
4199 | { | |
4200 | size_t cur; | |
4201 | size_t offset; | |
4202 | struct page *page; | |
4203 | char *kaddr; | |
4204 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4205 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4206 | ||
4207 | WARN_ON(start > eb->len); | |
4208 | WARN_ON(start + len > eb->start + eb->len); | |
4209 | ||
4210 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4211 | ||
d397712b | 4212 | while (len > 0) { |
d1310b2e CM |
4213 | page = extent_buffer_page(eb, i); |
4214 | WARN_ON(!PageUptodate(page)); | |
4215 | ||
4216 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 4217 | kaddr = page_address(page); |
d1310b2e | 4218 | memset(kaddr + offset, c, cur); |
d1310b2e CM |
4219 | |
4220 | len -= cur; | |
4221 | offset = 0; | |
4222 | i++; | |
4223 | } | |
4224 | } | |
d1310b2e CM |
4225 | |
4226 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
4227 | unsigned long dst_offset, unsigned long src_offset, | |
4228 | unsigned long len) | |
4229 | { | |
4230 | u64 dst_len = dst->len; | |
4231 | size_t cur; | |
4232 | size_t offset; | |
4233 | struct page *page; | |
4234 | char *kaddr; | |
4235 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4236 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
4237 | ||
4238 | WARN_ON(src->len != dst_len); | |
4239 | ||
4240 | offset = (start_offset + dst_offset) & | |
4241 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4242 | ||
d397712b | 4243 | while (len > 0) { |
d1310b2e CM |
4244 | page = extent_buffer_page(dst, i); |
4245 | WARN_ON(!PageUptodate(page)); | |
4246 | ||
4247 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
4248 | ||
a6591715 | 4249 | kaddr = page_address(page); |
d1310b2e | 4250 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
4251 | |
4252 | src_offset += cur; | |
4253 | len -= cur; | |
4254 | offset = 0; | |
4255 | i++; | |
4256 | } | |
4257 | } | |
d1310b2e CM |
4258 | |
4259 | static void move_pages(struct page *dst_page, struct page *src_page, | |
4260 | unsigned long dst_off, unsigned long src_off, | |
4261 | unsigned long len) | |
4262 | { | |
a6591715 | 4263 | char *dst_kaddr = page_address(dst_page); |
d1310b2e CM |
4264 | if (dst_page == src_page) { |
4265 | memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len); | |
4266 | } else { | |
a6591715 | 4267 | char *src_kaddr = page_address(src_page); |
d1310b2e CM |
4268 | char *p = dst_kaddr + dst_off + len; |
4269 | char *s = src_kaddr + src_off + len; | |
4270 | ||
4271 | while (len--) | |
4272 | *--p = *--s; | |
d1310b2e | 4273 | } |
d1310b2e CM |
4274 | } |
4275 | ||
3387206f ST |
4276 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
4277 | { | |
4278 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
4279 | return distance < len; | |
4280 | } | |
4281 | ||
d1310b2e CM |
4282 | static void copy_pages(struct page *dst_page, struct page *src_page, |
4283 | unsigned long dst_off, unsigned long src_off, | |
4284 | unsigned long len) | |
4285 | { | |
a6591715 | 4286 | char *dst_kaddr = page_address(dst_page); |
d1310b2e CM |
4287 | char *src_kaddr; |
4288 | ||
3387206f | 4289 | if (dst_page != src_page) { |
a6591715 | 4290 | src_kaddr = page_address(src_page); |
3387206f | 4291 | } else { |
d1310b2e | 4292 | src_kaddr = dst_kaddr; |
3387206f ST |
4293 | BUG_ON(areas_overlap(src_off, dst_off, len)); |
4294 | } | |
d1310b2e CM |
4295 | |
4296 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
4297 | } |
4298 | ||
4299 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
4300 | unsigned long src_offset, unsigned long len) | |
4301 | { | |
4302 | size_t cur; | |
4303 | size_t dst_off_in_page; | |
4304 | size_t src_off_in_page; | |
4305 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4306 | unsigned long dst_i; | |
4307 | unsigned long src_i; | |
4308 | ||
4309 | if (src_offset + len > dst->len) { | |
d397712b CM |
4310 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
4311 | "len %lu dst len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
4312 | BUG_ON(1); |
4313 | } | |
4314 | if (dst_offset + len > dst->len) { | |
d397712b CM |
4315 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
4316 | "len %lu dst len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
4317 | BUG_ON(1); |
4318 | } | |
4319 | ||
d397712b | 4320 | while (len > 0) { |
d1310b2e CM |
4321 | dst_off_in_page = (start_offset + dst_offset) & |
4322 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4323 | src_off_in_page = (start_offset + src_offset) & | |
4324 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4325 | ||
4326 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
4327 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
4328 | ||
4329 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
4330 | src_off_in_page)); | |
4331 | cur = min_t(unsigned long, cur, | |
4332 | (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); | |
4333 | ||
4334 | copy_pages(extent_buffer_page(dst, dst_i), | |
4335 | extent_buffer_page(dst, src_i), | |
4336 | dst_off_in_page, src_off_in_page, cur); | |
4337 | ||
4338 | src_offset += cur; | |
4339 | dst_offset += cur; | |
4340 | len -= cur; | |
4341 | } | |
4342 | } | |
d1310b2e CM |
4343 | |
4344 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
4345 | unsigned long src_offset, unsigned long len) | |
4346 | { | |
4347 | size_t cur; | |
4348 | size_t dst_off_in_page; | |
4349 | size_t src_off_in_page; | |
4350 | unsigned long dst_end = dst_offset + len - 1; | |
4351 | unsigned long src_end = src_offset + len - 1; | |
4352 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4353 | unsigned long dst_i; | |
4354 | unsigned long src_i; | |
4355 | ||
4356 | if (src_offset + len > dst->len) { | |
d397712b CM |
4357 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
4358 | "len %lu len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
4359 | BUG_ON(1); |
4360 | } | |
4361 | if (dst_offset + len > dst->len) { | |
d397712b CM |
4362 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
4363 | "len %lu len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
4364 | BUG_ON(1); |
4365 | } | |
3387206f | 4366 | if (!areas_overlap(src_offset, dst_offset, len)) { |
d1310b2e CM |
4367 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
4368 | return; | |
4369 | } | |
d397712b | 4370 | while (len > 0) { |
d1310b2e CM |
4371 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; |
4372 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
4373 | ||
4374 | dst_off_in_page = (start_offset + dst_end) & | |
4375 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4376 | src_off_in_page = (start_offset + src_end) & | |
4377 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4378 | ||
4379 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
4380 | cur = min(cur, dst_off_in_page + 1); | |
4381 | move_pages(extent_buffer_page(dst, dst_i), | |
4382 | extent_buffer_page(dst, src_i), | |
4383 | dst_off_in_page - cur + 1, | |
4384 | src_off_in_page - cur + 1, cur); | |
4385 | ||
4386 | dst_end -= cur; | |
4387 | src_end -= cur; | |
4388 | len -= cur; | |
4389 | } | |
4390 | } | |
6af118ce | 4391 | |
19fe0a8b MX |
4392 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
4393 | { | |
4394 | struct extent_buffer *eb = | |
4395 | container_of(head, struct extent_buffer, rcu_head); | |
4396 | ||
4397 | btrfs_release_extent_buffer(eb); | |
4398 | } | |
4399 | ||
6af118ce CM |
4400 | int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page) |
4401 | { | |
4402 | u64 start = page_offset(page); | |
4403 | struct extent_buffer *eb; | |
4404 | int ret = 1; | |
6af118ce CM |
4405 | |
4406 | spin_lock(&tree->buffer_lock); | |
19fe0a8b | 4407 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); |
45f49bce CM |
4408 | if (!eb) { |
4409 | spin_unlock(&tree->buffer_lock); | |
4410 | return ret; | |
4411 | } | |
6af118ce | 4412 | |
19fe0a8b | 4413 | if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
6af118ce CM |
4414 | ret = 0; |
4415 | goto out; | |
4416 | } | |
19fe0a8b MX |
4417 | |
4418 | /* | |
4419 | * set @eb->refs to 0 if it is already 1, and then release the @eb. | |
4420 | * Or go back. | |
4421 | */ | |
4422 | if (atomic_cmpxchg(&eb->refs, 1, 0) != 1) { | |
b9473439 CM |
4423 | ret = 0; |
4424 | goto out; | |
4425 | } | |
897ca6e9 | 4426 | |
19fe0a8b | 4427 | radix_tree_delete(&tree->buffer, start >> PAGE_CACHE_SHIFT); |
6af118ce CM |
4428 | out: |
4429 | spin_unlock(&tree->buffer_lock); | |
19fe0a8b MX |
4430 | |
4431 | /* at this point we can safely release the extent buffer */ | |
4432 | if (atomic_read(&eb->refs) == 0) | |
4433 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); | |
6af118ce CM |
4434 | return ret; |
4435 | } |