]>
Commit | Line | Data |
---|---|---|
d1310b2e CM |
1 | #include <linux/bitops.h> |
2 | #include <linux/slab.h> | |
3 | #include <linux/bio.h> | |
4 | #include <linux/mm.h> | |
d1310b2e CM |
5 | #include <linux/pagemap.h> |
6 | #include <linux/page-flags.h> | |
7 | #include <linux/module.h> | |
8 | #include <linux/spinlock.h> | |
9 | #include <linux/blkdev.h> | |
10 | #include <linux/swap.h> | |
d1310b2e CM |
11 | #include <linux/writeback.h> |
12 | #include <linux/pagevec.h> | |
13 | #include "extent_io.h" | |
14 | #include "extent_map.h" | |
2db04966 | 15 | #include "compat.h" |
902b22f3 DW |
16 | #include "ctree.h" |
17 | #include "btrfs_inode.h" | |
d1310b2e | 18 | |
d1310b2e CM |
19 | static struct kmem_cache *extent_state_cache; |
20 | static struct kmem_cache *extent_buffer_cache; | |
21 | ||
22 | static LIST_HEAD(buffers); | |
23 | static LIST_HEAD(states); | |
4bef0848 | 24 | |
b47eda86 | 25 | #define LEAK_DEBUG 0 |
3935127c | 26 | #if LEAK_DEBUG |
d397712b | 27 | static DEFINE_SPINLOCK(leak_lock); |
4bef0848 | 28 | #endif |
d1310b2e | 29 | |
d1310b2e CM |
30 | #define BUFFER_LRU_MAX 64 |
31 | ||
32 | struct tree_entry { | |
33 | u64 start; | |
34 | u64 end; | |
d1310b2e CM |
35 | struct rb_node rb_node; |
36 | }; | |
37 | ||
38 | struct extent_page_data { | |
39 | struct bio *bio; | |
40 | struct extent_io_tree *tree; | |
41 | get_extent_t *get_extent; | |
771ed689 CM |
42 | |
43 | /* tells writepage not to lock the state bits for this range | |
44 | * it still does the unlocking | |
45 | */ | |
ffbd517d CM |
46 | unsigned int extent_locked:1; |
47 | ||
48 | /* tells the submit_bio code to use a WRITE_SYNC */ | |
49 | unsigned int sync_io:1; | |
d1310b2e CM |
50 | }; |
51 | ||
52 | int __init extent_io_init(void) | |
53 | { | |
9601e3f6 CH |
54 | extent_state_cache = kmem_cache_create("extent_state", |
55 | sizeof(struct extent_state), 0, | |
56 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
57 | if (!extent_state_cache) |
58 | return -ENOMEM; | |
59 | ||
9601e3f6 CH |
60 | extent_buffer_cache = kmem_cache_create("extent_buffers", |
61 | sizeof(struct extent_buffer), 0, | |
62 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
63 | if (!extent_buffer_cache) |
64 | goto free_state_cache; | |
65 | return 0; | |
66 | ||
67 | free_state_cache: | |
68 | kmem_cache_destroy(extent_state_cache); | |
69 | return -ENOMEM; | |
70 | } | |
71 | ||
72 | void extent_io_exit(void) | |
73 | { | |
74 | struct extent_state *state; | |
2d2ae547 | 75 | struct extent_buffer *eb; |
d1310b2e CM |
76 | |
77 | while (!list_empty(&states)) { | |
2d2ae547 | 78 | state = list_entry(states.next, struct extent_state, leak_list); |
d397712b CM |
79 | printk(KERN_ERR "btrfs state leak: start %llu end %llu " |
80 | "state %lu in tree %p refs %d\n", | |
81 | (unsigned long long)state->start, | |
82 | (unsigned long long)state->end, | |
83 | state->state, state->tree, atomic_read(&state->refs)); | |
2d2ae547 | 84 | list_del(&state->leak_list); |
d1310b2e CM |
85 | kmem_cache_free(extent_state_cache, state); |
86 | ||
87 | } | |
88 | ||
2d2ae547 CM |
89 | while (!list_empty(&buffers)) { |
90 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
d397712b CM |
91 | printk(KERN_ERR "btrfs buffer leak start %llu len %lu " |
92 | "refs %d\n", (unsigned long long)eb->start, | |
93 | eb->len, atomic_read(&eb->refs)); | |
2d2ae547 CM |
94 | list_del(&eb->leak_list); |
95 | kmem_cache_free(extent_buffer_cache, eb); | |
96 | } | |
d1310b2e CM |
97 | if (extent_state_cache) |
98 | kmem_cache_destroy(extent_state_cache); | |
99 | if (extent_buffer_cache) | |
100 | kmem_cache_destroy(extent_buffer_cache); | |
101 | } | |
102 | ||
103 | void extent_io_tree_init(struct extent_io_tree *tree, | |
104 | struct address_space *mapping, gfp_t mask) | |
105 | { | |
6bef4d31 | 106 | tree->state = RB_ROOT; |
19fe0a8b | 107 | INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC); |
d1310b2e CM |
108 | tree->ops = NULL; |
109 | tree->dirty_bytes = 0; | |
70dec807 | 110 | spin_lock_init(&tree->lock); |
6af118ce | 111 | spin_lock_init(&tree->buffer_lock); |
d1310b2e | 112 | tree->mapping = mapping; |
d1310b2e | 113 | } |
d1310b2e | 114 | |
b2950863 | 115 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
116 | { |
117 | struct extent_state *state; | |
3935127c | 118 | #if LEAK_DEBUG |
2d2ae547 | 119 | unsigned long flags; |
4bef0848 | 120 | #endif |
d1310b2e CM |
121 | |
122 | state = kmem_cache_alloc(extent_state_cache, mask); | |
2b114d1d | 123 | if (!state) |
d1310b2e CM |
124 | return state; |
125 | state->state = 0; | |
d1310b2e | 126 | state->private = 0; |
70dec807 | 127 | state->tree = NULL; |
3935127c | 128 | #if LEAK_DEBUG |
2d2ae547 CM |
129 | spin_lock_irqsave(&leak_lock, flags); |
130 | list_add(&state->leak_list, &states); | |
131 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 132 | #endif |
d1310b2e CM |
133 | atomic_set(&state->refs, 1); |
134 | init_waitqueue_head(&state->wq); | |
135 | return state; | |
136 | } | |
d1310b2e | 137 | |
4845e44f | 138 | void free_extent_state(struct extent_state *state) |
d1310b2e | 139 | { |
d1310b2e CM |
140 | if (!state) |
141 | return; | |
142 | if (atomic_dec_and_test(&state->refs)) { | |
3935127c | 143 | #if LEAK_DEBUG |
2d2ae547 | 144 | unsigned long flags; |
4bef0848 | 145 | #endif |
70dec807 | 146 | WARN_ON(state->tree); |
3935127c | 147 | #if LEAK_DEBUG |
2d2ae547 CM |
148 | spin_lock_irqsave(&leak_lock, flags); |
149 | list_del(&state->leak_list); | |
150 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 151 | #endif |
d1310b2e CM |
152 | kmem_cache_free(extent_state_cache, state); |
153 | } | |
154 | } | |
d1310b2e CM |
155 | |
156 | static struct rb_node *tree_insert(struct rb_root *root, u64 offset, | |
157 | struct rb_node *node) | |
158 | { | |
d397712b CM |
159 | struct rb_node **p = &root->rb_node; |
160 | struct rb_node *parent = NULL; | |
d1310b2e CM |
161 | struct tree_entry *entry; |
162 | ||
d397712b | 163 | while (*p) { |
d1310b2e CM |
164 | parent = *p; |
165 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
166 | ||
167 | if (offset < entry->start) | |
168 | p = &(*p)->rb_left; | |
169 | else if (offset > entry->end) | |
170 | p = &(*p)->rb_right; | |
171 | else | |
172 | return parent; | |
173 | } | |
174 | ||
175 | entry = rb_entry(node, struct tree_entry, rb_node); | |
d1310b2e CM |
176 | rb_link_node(node, parent, p); |
177 | rb_insert_color(node, root); | |
178 | return NULL; | |
179 | } | |
180 | ||
80ea96b1 | 181 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
d1310b2e CM |
182 | struct rb_node **prev_ret, |
183 | struct rb_node **next_ret) | |
184 | { | |
80ea96b1 | 185 | struct rb_root *root = &tree->state; |
d397712b | 186 | struct rb_node *n = root->rb_node; |
d1310b2e CM |
187 | struct rb_node *prev = NULL; |
188 | struct rb_node *orig_prev = NULL; | |
189 | struct tree_entry *entry; | |
190 | struct tree_entry *prev_entry = NULL; | |
191 | ||
d397712b | 192 | while (n) { |
d1310b2e CM |
193 | entry = rb_entry(n, struct tree_entry, rb_node); |
194 | prev = n; | |
195 | prev_entry = entry; | |
196 | ||
197 | if (offset < entry->start) | |
198 | n = n->rb_left; | |
199 | else if (offset > entry->end) | |
200 | n = n->rb_right; | |
d397712b | 201 | else |
d1310b2e CM |
202 | return n; |
203 | } | |
204 | ||
205 | if (prev_ret) { | |
206 | orig_prev = prev; | |
d397712b | 207 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
208 | prev = rb_next(prev); |
209 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
210 | } | |
211 | *prev_ret = prev; | |
212 | prev = orig_prev; | |
213 | } | |
214 | ||
215 | if (next_ret) { | |
216 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 217 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
218 | prev = rb_prev(prev); |
219 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
220 | } | |
221 | *next_ret = prev; | |
222 | } | |
223 | return NULL; | |
224 | } | |
225 | ||
80ea96b1 CM |
226 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
227 | u64 offset) | |
d1310b2e | 228 | { |
70dec807 | 229 | struct rb_node *prev = NULL; |
d1310b2e | 230 | struct rb_node *ret; |
70dec807 | 231 | |
80ea96b1 | 232 | ret = __etree_search(tree, offset, &prev, NULL); |
d397712b | 233 | if (!ret) |
d1310b2e CM |
234 | return prev; |
235 | return ret; | |
236 | } | |
237 | ||
9ed74f2d JB |
238 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, |
239 | struct extent_state *other) | |
240 | { | |
241 | if (tree->ops && tree->ops->merge_extent_hook) | |
242 | tree->ops->merge_extent_hook(tree->mapping->host, new, | |
243 | other); | |
244 | } | |
245 | ||
d1310b2e CM |
246 | /* |
247 | * utility function to look for merge candidates inside a given range. | |
248 | * Any extents with matching state are merged together into a single | |
249 | * extent in the tree. Extents with EXTENT_IO in their state field | |
250 | * are not merged because the end_io handlers need to be able to do | |
251 | * operations on them without sleeping (or doing allocations/splits). | |
252 | * | |
253 | * This should be called with the tree lock held. | |
254 | */ | |
255 | static int merge_state(struct extent_io_tree *tree, | |
256 | struct extent_state *state) | |
257 | { | |
258 | struct extent_state *other; | |
259 | struct rb_node *other_node; | |
260 | ||
5b21f2ed | 261 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
d1310b2e CM |
262 | return 0; |
263 | ||
264 | other_node = rb_prev(&state->rb_node); | |
265 | if (other_node) { | |
266 | other = rb_entry(other_node, struct extent_state, rb_node); | |
267 | if (other->end == state->start - 1 && | |
268 | other->state == state->state) { | |
9ed74f2d | 269 | merge_cb(tree, state, other); |
d1310b2e | 270 | state->start = other->start; |
70dec807 | 271 | other->tree = NULL; |
d1310b2e CM |
272 | rb_erase(&other->rb_node, &tree->state); |
273 | free_extent_state(other); | |
274 | } | |
275 | } | |
276 | other_node = rb_next(&state->rb_node); | |
277 | if (other_node) { | |
278 | other = rb_entry(other_node, struct extent_state, rb_node); | |
279 | if (other->start == state->end + 1 && | |
280 | other->state == state->state) { | |
9ed74f2d | 281 | merge_cb(tree, state, other); |
d1310b2e | 282 | other->start = state->start; |
70dec807 | 283 | state->tree = NULL; |
d1310b2e CM |
284 | rb_erase(&state->rb_node, &tree->state); |
285 | free_extent_state(state); | |
9ed74f2d | 286 | state = NULL; |
d1310b2e CM |
287 | } |
288 | } | |
9ed74f2d | 289 | |
d1310b2e CM |
290 | return 0; |
291 | } | |
292 | ||
9ed74f2d | 293 | static int set_state_cb(struct extent_io_tree *tree, |
0ca1f7ce | 294 | struct extent_state *state, int *bits) |
291d673e CM |
295 | { |
296 | if (tree->ops && tree->ops->set_bit_hook) { | |
9ed74f2d | 297 | return tree->ops->set_bit_hook(tree->mapping->host, |
0ca1f7ce | 298 | state, bits); |
291d673e | 299 | } |
9ed74f2d JB |
300 | |
301 | return 0; | |
291d673e CM |
302 | } |
303 | ||
304 | static void clear_state_cb(struct extent_io_tree *tree, | |
0ca1f7ce | 305 | struct extent_state *state, int *bits) |
291d673e | 306 | { |
9ed74f2d JB |
307 | if (tree->ops && tree->ops->clear_bit_hook) |
308 | tree->ops->clear_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
309 | } |
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; | |
0ca1f7ce | 326 | int bits_to_set = *bits & ~EXTENT_CTLBITS; |
9ed74f2d | 327 | int ret; |
d1310b2e CM |
328 | |
329 | if (end < start) { | |
d397712b CM |
330 | printk(KERN_ERR "btrfs end < start %llu %llu\n", |
331 | (unsigned long long)end, | |
332 | (unsigned long long)start); | |
d1310b2e CM |
333 | WARN_ON(1); |
334 | } | |
d1310b2e CM |
335 | state->start = start; |
336 | state->end = end; | |
9ed74f2d JB |
337 | ret = set_state_cb(tree, state, bits); |
338 | if (ret) | |
339 | return ret; | |
340 | ||
0ca1f7ce | 341 | if (bits_to_set & EXTENT_DIRTY) |
9ed74f2d | 342 | tree->dirty_bytes += end - start + 1; |
0ca1f7ce | 343 | state->state |= bits_to_set; |
d1310b2e CM |
344 | node = tree_insert(&tree->state, end, &state->rb_node); |
345 | if (node) { | |
346 | struct extent_state *found; | |
347 | found = rb_entry(node, struct extent_state, rb_node); | |
d397712b CM |
348 | printk(KERN_ERR "btrfs found node %llu %llu on insert of " |
349 | "%llu %llu\n", (unsigned long long)found->start, | |
350 | (unsigned long long)found->end, | |
351 | (unsigned long long)start, (unsigned long long)end); | |
d1310b2e CM |
352 | free_extent_state(state); |
353 | return -EEXIST; | |
354 | } | |
70dec807 | 355 | state->tree = tree; |
d1310b2e CM |
356 | merge_state(tree, state); |
357 | return 0; | |
358 | } | |
359 | ||
9ed74f2d JB |
360 | static int split_cb(struct extent_io_tree *tree, struct extent_state *orig, |
361 | u64 split) | |
362 | { | |
363 | if (tree->ops && tree->ops->split_extent_hook) | |
364 | return tree->ops->split_extent_hook(tree->mapping->host, | |
365 | orig, split); | |
366 | return 0; | |
367 | } | |
368 | ||
d1310b2e CM |
369 | /* |
370 | * split a given extent state struct in two, inserting the preallocated | |
371 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
372 | * offset inside 'orig' where it should be split. | |
373 | * | |
374 | * Before calling, | |
375 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
376 | * are two extent state structs in the tree: | |
377 | * prealloc: [orig->start, split - 1] | |
378 | * orig: [ split, orig->end ] | |
379 | * | |
380 | * The tree locks are not taken by this function. They need to be held | |
381 | * by the caller. | |
382 | */ | |
383 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
384 | struct extent_state *prealloc, u64 split) | |
385 | { | |
386 | struct rb_node *node; | |
9ed74f2d JB |
387 | |
388 | split_cb(tree, orig, split); | |
389 | ||
d1310b2e CM |
390 | prealloc->start = orig->start; |
391 | prealloc->end = split - 1; | |
392 | prealloc->state = orig->state; | |
393 | orig->start = split; | |
394 | ||
395 | node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node); | |
396 | if (node) { | |
d1310b2e CM |
397 | free_extent_state(prealloc); |
398 | return -EEXIST; | |
399 | } | |
70dec807 | 400 | prealloc->tree = tree; |
d1310b2e CM |
401 | return 0; |
402 | } | |
403 | ||
404 | /* | |
405 | * utility function to clear some bits in an extent state struct. | |
406 | * it will optionally wake up any one waiting on this state (wake == 1), or | |
407 | * forcibly remove the state from the tree (delete == 1). | |
408 | * | |
409 | * If no bits are set on the state struct after clearing things, the | |
410 | * struct is freed and removed from the tree | |
411 | */ | |
412 | static int clear_state_bit(struct extent_io_tree *tree, | |
0ca1f7ce YZ |
413 | struct extent_state *state, |
414 | int *bits, int wake) | |
d1310b2e | 415 | { |
0ca1f7ce | 416 | int bits_to_clear = *bits & ~EXTENT_CTLBITS; |
32c00aff | 417 | int ret = state->state & bits_to_clear; |
d1310b2e | 418 | |
0ca1f7ce | 419 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
420 | u64 range = state->end - state->start + 1; |
421 | WARN_ON(range > tree->dirty_bytes); | |
422 | tree->dirty_bytes -= range; | |
423 | } | |
291d673e | 424 | clear_state_cb(tree, state, bits); |
32c00aff | 425 | state->state &= ~bits_to_clear; |
d1310b2e CM |
426 | if (wake) |
427 | wake_up(&state->wq); | |
0ca1f7ce | 428 | if (state->state == 0) { |
70dec807 | 429 | if (state->tree) { |
d1310b2e | 430 | rb_erase(&state->rb_node, &tree->state); |
70dec807 | 431 | state->tree = NULL; |
d1310b2e CM |
432 | free_extent_state(state); |
433 | } else { | |
434 | WARN_ON(1); | |
435 | } | |
436 | } else { | |
437 | merge_state(tree, state); | |
438 | } | |
439 | return ret; | |
440 | } | |
441 | ||
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 | ||
42daec29 | 492 | if (cached && cached->tree && cached->start == start) { |
2ac55d41 JB |
493 | if (clear) |
494 | atomic_dec(&cached->refs); | |
2c64c53d | 495 | state = cached; |
42daec29 | 496 | goto hit_next; |
2c64c53d | 497 | } |
2ac55d41 JB |
498 | if (clear) |
499 | free_extent_state(cached); | |
2c64c53d | 500 | } |
d1310b2e CM |
501 | /* |
502 | * this search will find the extents that end after | |
503 | * our range starts | |
504 | */ | |
80ea96b1 | 505 | node = tree_search(tree, start); |
d1310b2e CM |
506 | if (!node) |
507 | goto out; | |
508 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 509 | hit_next: |
d1310b2e CM |
510 | if (state->start > end) |
511 | goto out; | |
512 | WARN_ON(state->end < start); | |
5c939df5 | 513 | last_end = state->end; |
d1310b2e CM |
514 | |
515 | /* | |
516 | * | ---- desired range ---- | | |
517 | * | state | or | |
518 | * | ------------- state -------------- | | |
519 | * | |
520 | * We need to split the extent we found, and may flip | |
521 | * bits on second half. | |
522 | * | |
523 | * If the extent we found extends past our range, we | |
524 | * just split and search again. It'll get split again | |
525 | * the next time though. | |
526 | * | |
527 | * If the extent we found is inside our range, we clear | |
528 | * the desired bit on it. | |
529 | */ | |
530 | ||
531 | if (state->start < start) { | |
70dec807 CM |
532 | if (!prealloc) |
533 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
d1310b2e CM |
534 | err = split_state(tree, state, prealloc, start); |
535 | BUG_ON(err == -EEXIST); | |
536 | prealloc = NULL; | |
537 | if (err) | |
538 | goto out; | |
539 | if (state->end <= end) { | |
0ca1f7ce | 540 | set |= clear_state_bit(tree, state, &bits, wake); |
5c939df5 YZ |
541 | if (last_end == (u64)-1) |
542 | goto out; | |
543 | start = last_end + 1; | |
d1310b2e CM |
544 | } |
545 | goto search_again; | |
546 | } | |
547 | /* | |
548 | * | ---- desired range ---- | | |
549 | * | state | | |
550 | * We need to split the extent, and clear the bit | |
551 | * on the first half | |
552 | */ | |
553 | if (state->start <= end && state->end > end) { | |
70dec807 CM |
554 | if (!prealloc) |
555 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
d1310b2e CM |
556 | err = split_state(tree, state, prealloc, end + 1); |
557 | BUG_ON(err == -EEXIST); | |
d1310b2e CM |
558 | if (wake) |
559 | wake_up(&state->wq); | |
42daec29 | 560 | |
0ca1f7ce | 561 | set |= clear_state_bit(tree, prealloc, &bits, wake); |
9ed74f2d | 562 | |
d1310b2e CM |
563 | prealloc = NULL; |
564 | goto out; | |
565 | } | |
42daec29 | 566 | |
2c64c53d CM |
567 | if (state->end < end && prealloc && !need_resched()) |
568 | next_node = rb_next(&state->rb_node); | |
569 | else | |
570 | next_node = NULL; | |
42daec29 | 571 | |
0ca1f7ce | 572 | set |= clear_state_bit(tree, state, &bits, wake); |
5c939df5 YZ |
573 | if (last_end == (u64)-1) |
574 | goto out; | |
575 | start = last_end + 1; | |
2c64c53d CM |
576 | if (start <= end && next_node) { |
577 | state = rb_entry(next_node, struct extent_state, | |
578 | rb_node); | |
579 | if (state->start == start) | |
580 | goto hit_next; | |
581 | } | |
d1310b2e CM |
582 | goto search_again; |
583 | ||
584 | out: | |
cad321ad | 585 | spin_unlock(&tree->lock); |
d1310b2e CM |
586 | if (prealloc) |
587 | free_extent_state(prealloc); | |
588 | ||
589 | return set; | |
590 | ||
591 | search_again: | |
592 | if (start > end) | |
593 | goto out; | |
cad321ad | 594 | spin_unlock(&tree->lock); |
d1310b2e CM |
595 | if (mask & __GFP_WAIT) |
596 | cond_resched(); | |
597 | goto again; | |
598 | } | |
d1310b2e CM |
599 | |
600 | static int wait_on_state(struct extent_io_tree *tree, | |
601 | struct extent_state *state) | |
641f5219 CH |
602 | __releases(tree->lock) |
603 | __acquires(tree->lock) | |
d1310b2e CM |
604 | { |
605 | DEFINE_WAIT(wait); | |
606 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 607 | spin_unlock(&tree->lock); |
d1310b2e | 608 | schedule(); |
cad321ad | 609 | spin_lock(&tree->lock); |
d1310b2e CM |
610 | finish_wait(&state->wq, &wait); |
611 | return 0; | |
612 | } | |
613 | ||
614 | /* | |
615 | * waits for one or more bits to clear on a range in the state tree. | |
616 | * The range [start, end] is inclusive. | |
617 | * The tree lock is taken by this function | |
618 | */ | |
619 | int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits) | |
620 | { | |
621 | struct extent_state *state; | |
622 | struct rb_node *node; | |
623 | ||
cad321ad | 624 | spin_lock(&tree->lock); |
d1310b2e CM |
625 | again: |
626 | while (1) { | |
627 | /* | |
628 | * this search will find all the extents that end after | |
629 | * our range starts | |
630 | */ | |
80ea96b1 | 631 | node = tree_search(tree, start); |
d1310b2e CM |
632 | if (!node) |
633 | break; | |
634 | ||
635 | state = rb_entry(node, struct extent_state, rb_node); | |
636 | ||
637 | if (state->start > end) | |
638 | goto out; | |
639 | ||
640 | if (state->state & bits) { | |
641 | start = state->start; | |
642 | atomic_inc(&state->refs); | |
643 | wait_on_state(tree, state); | |
644 | free_extent_state(state); | |
645 | goto again; | |
646 | } | |
647 | start = state->end + 1; | |
648 | ||
649 | if (start > end) | |
650 | break; | |
651 | ||
652 | if (need_resched()) { | |
cad321ad | 653 | spin_unlock(&tree->lock); |
d1310b2e | 654 | cond_resched(); |
cad321ad | 655 | spin_lock(&tree->lock); |
d1310b2e CM |
656 | } |
657 | } | |
658 | out: | |
cad321ad | 659 | spin_unlock(&tree->lock); |
d1310b2e CM |
660 | return 0; |
661 | } | |
d1310b2e | 662 | |
9ed74f2d | 663 | static int set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 664 | struct extent_state *state, |
0ca1f7ce | 665 | int *bits) |
d1310b2e | 666 | { |
9ed74f2d | 667 | int ret; |
0ca1f7ce | 668 | int bits_to_set = *bits & ~EXTENT_CTLBITS; |
9ed74f2d JB |
669 | |
670 | ret = set_state_cb(tree, state, bits); | |
671 | if (ret) | |
672 | return ret; | |
0ca1f7ce | 673 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
674 | u64 range = state->end - state->start + 1; |
675 | tree->dirty_bytes += range; | |
676 | } | |
0ca1f7ce | 677 | state->state |= bits_to_set; |
9ed74f2d JB |
678 | |
679 | return 0; | |
d1310b2e CM |
680 | } |
681 | ||
2c64c53d CM |
682 | static void cache_state(struct extent_state *state, |
683 | struct extent_state **cached_ptr) | |
684 | { | |
685 | if (cached_ptr && !(*cached_ptr)) { | |
686 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) { | |
687 | *cached_ptr = state; | |
688 | atomic_inc(&state->refs); | |
689 | } | |
690 | } | |
691 | } | |
692 | ||
507903b8 AJ |
693 | static void uncache_state(struct extent_state **cached_ptr) |
694 | { | |
695 | if (cached_ptr && (*cached_ptr)) { | |
696 | struct extent_state *state = *cached_ptr; | |
697 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) { | |
698 | *cached_ptr = NULL; | |
699 | free_extent_state(state); | |
700 | } | |
701 | } | |
702 | } | |
703 | ||
d1310b2e | 704 | /* |
1edbb734 CM |
705 | * set some bits on a range in the tree. This may require allocations or |
706 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 707 | * |
1edbb734 CM |
708 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
709 | * part of the range already has the desired bits set. The start of the | |
710 | * existing range is returned in failed_start in this case. | |
d1310b2e | 711 | * |
1edbb734 | 712 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 713 | */ |
1edbb734 | 714 | |
4845e44f CM |
715 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
716 | int bits, int exclusive_bits, u64 *failed_start, | |
717 | struct extent_state **cached_state, gfp_t mask) | |
d1310b2e CM |
718 | { |
719 | struct extent_state *state; | |
720 | struct extent_state *prealloc = NULL; | |
721 | struct rb_node *node; | |
d1310b2e | 722 | int err = 0; |
d1310b2e CM |
723 | u64 last_start; |
724 | u64 last_end; | |
42daec29 | 725 | |
0ca1f7ce | 726 | bits |= EXTENT_FIRST_DELALLOC; |
d1310b2e CM |
727 | again: |
728 | if (!prealloc && (mask & __GFP_WAIT)) { | |
729 | prealloc = alloc_extent_state(mask); | |
730 | if (!prealloc) | |
731 | return -ENOMEM; | |
732 | } | |
733 | ||
cad321ad | 734 | spin_lock(&tree->lock); |
9655d298 CM |
735 | if (cached_state && *cached_state) { |
736 | state = *cached_state; | |
737 | if (state->start == start && state->tree) { | |
738 | node = &state->rb_node; | |
739 | goto hit_next; | |
740 | } | |
741 | } | |
d1310b2e CM |
742 | /* |
743 | * this search will find all the extents that end after | |
744 | * our range starts. | |
745 | */ | |
80ea96b1 | 746 | node = tree_search(tree, start); |
d1310b2e | 747 | if (!node) { |
0ca1f7ce | 748 | err = insert_state(tree, prealloc, start, end, &bits); |
d1310b2e CM |
749 | prealloc = NULL; |
750 | BUG_ON(err == -EEXIST); | |
751 | goto out; | |
752 | } | |
d1310b2e | 753 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 754 | hit_next: |
d1310b2e CM |
755 | last_start = state->start; |
756 | last_end = state->end; | |
757 | ||
758 | /* | |
759 | * | ---- desired range ---- | | |
760 | * | state | | |
761 | * | |
762 | * Just lock what we found and keep going | |
763 | */ | |
764 | if (state->start == start && state->end <= end) { | |
40431d6c | 765 | struct rb_node *next_node; |
1edbb734 | 766 | if (state->state & exclusive_bits) { |
d1310b2e CM |
767 | *failed_start = state->start; |
768 | err = -EEXIST; | |
769 | goto out; | |
770 | } | |
42daec29 | 771 | |
0ca1f7ce | 772 | err = set_state_bits(tree, state, &bits); |
9ed74f2d JB |
773 | if (err) |
774 | goto out; | |
775 | ||
2c64c53d | 776 | cache_state(state, cached_state); |
d1310b2e | 777 | merge_state(tree, state); |
5c939df5 YZ |
778 | if (last_end == (u64)-1) |
779 | goto out; | |
40431d6c | 780 | |
5c939df5 | 781 | start = last_end + 1; |
40431d6c CM |
782 | if (start < end && prealloc && !need_resched()) { |
783 | next_node = rb_next(node); | |
784 | if (next_node) { | |
785 | state = rb_entry(next_node, struct extent_state, | |
786 | rb_node); | |
787 | if (state->start == start) | |
788 | goto hit_next; | |
789 | } | |
790 | } | |
d1310b2e CM |
791 | goto search_again; |
792 | } | |
793 | ||
794 | /* | |
795 | * | ---- desired range ---- | | |
796 | * | state | | |
797 | * or | |
798 | * | ------------- state -------------- | | |
799 | * | |
800 | * We need to split the extent we found, and may flip bits on | |
801 | * second half. | |
802 | * | |
803 | * If the extent we found extends past our | |
804 | * range, we just split and search again. It'll get split | |
805 | * again the next time though. | |
806 | * | |
807 | * If the extent we found is inside our range, we set the | |
808 | * desired bit on it. | |
809 | */ | |
810 | if (state->start < start) { | |
1edbb734 | 811 | if (state->state & exclusive_bits) { |
d1310b2e CM |
812 | *failed_start = start; |
813 | err = -EEXIST; | |
814 | goto out; | |
815 | } | |
816 | err = split_state(tree, state, prealloc, start); | |
817 | BUG_ON(err == -EEXIST); | |
818 | prealloc = NULL; | |
819 | if (err) | |
820 | goto out; | |
821 | if (state->end <= end) { | |
0ca1f7ce | 822 | err = set_state_bits(tree, state, &bits); |
9ed74f2d JB |
823 | if (err) |
824 | goto out; | |
2c64c53d | 825 | cache_state(state, cached_state); |
d1310b2e | 826 | merge_state(tree, state); |
5c939df5 YZ |
827 | if (last_end == (u64)-1) |
828 | goto out; | |
829 | start = last_end + 1; | |
d1310b2e CM |
830 | } |
831 | goto search_again; | |
832 | } | |
833 | /* | |
834 | * | ---- desired range ---- | | |
835 | * | state | or | state | | |
836 | * | |
837 | * There's a hole, we need to insert something in it and | |
838 | * ignore the extent we found. | |
839 | */ | |
840 | if (state->start > start) { | |
841 | u64 this_end; | |
842 | if (end < last_start) | |
843 | this_end = end; | |
844 | else | |
d397712b | 845 | this_end = last_start - 1; |
d1310b2e | 846 | err = insert_state(tree, prealloc, start, this_end, |
0ca1f7ce | 847 | &bits); |
d1310b2e | 848 | BUG_ON(err == -EEXIST); |
9ed74f2d JB |
849 | if (err) { |
850 | prealloc = NULL; | |
d1310b2e | 851 | goto out; |
9ed74f2d JB |
852 | } |
853 | cache_state(prealloc, cached_state); | |
854 | prealloc = NULL; | |
d1310b2e CM |
855 | start = this_end + 1; |
856 | goto search_again; | |
857 | } | |
858 | /* | |
859 | * | ---- desired range ---- | | |
860 | * | state | | |
861 | * We need to split the extent, and set the bit | |
862 | * on the first half | |
863 | */ | |
864 | if (state->start <= end && state->end > end) { | |
1edbb734 | 865 | if (state->state & exclusive_bits) { |
d1310b2e CM |
866 | *failed_start = start; |
867 | err = -EEXIST; | |
868 | goto out; | |
869 | } | |
870 | err = split_state(tree, state, prealloc, end + 1); | |
871 | BUG_ON(err == -EEXIST); | |
872 | ||
0ca1f7ce | 873 | err = set_state_bits(tree, prealloc, &bits); |
9ed74f2d JB |
874 | if (err) { |
875 | prealloc = NULL; | |
876 | goto out; | |
877 | } | |
2c64c53d | 878 | cache_state(prealloc, cached_state); |
d1310b2e CM |
879 | merge_state(tree, prealloc); |
880 | prealloc = NULL; | |
881 | goto out; | |
882 | } | |
883 | ||
884 | goto search_again; | |
885 | ||
886 | out: | |
cad321ad | 887 | spin_unlock(&tree->lock); |
d1310b2e CM |
888 | if (prealloc) |
889 | free_extent_state(prealloc); | |
890 | ||
891 | return err; | |
892 | ||
893 | search_again: | |
894 | if (start > end) | |
895 | goto out; | |
cad321ad | 896 | spin_unlock(&tree->lock); |
d1310b2e CM |
897 | if (mask & __GFP_WAIT) |
898 | cond_resched(); | |
899 | goto again; | |
900 | } | |
d1310b2e CM |
901 | |
902 | /* wrappers around set/clear extent bit */ | |
903 | int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
904 | gfp_t mask) | |
905 | { | |
906 | return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL, | |
2c64c53d | 907 | NULL, mask); |
d1310b2e | 908 | } |
d1310b2e CM |
909 | |
910 | int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
911 | int bits, gfp_t mask) | |
912 | { | |
913 | return set_extent_bit(tree, start, end, bits, 0, NULL, | |
2c64c53d | 914 | NULL, mask); |
d1310b2e | 915 | } |
d1310b2e CM |
916 | |
917 | int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
918 | int bits, gfp_t mask) | |
919 | { | |
2c64c53d | 920 | return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask); |
d1310b2e | 921 | } |
d1310b2e CM |
922 | |
923 | int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, | |
2ac55d41 | 924 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
925 | { |
926 | return set_extent_bit(tree, start, end, | |
40431d6c | 927 | EXTENT_DELALLOC | EXTENT_DIRTY | EXTENT_UPTODATE, |
2ac55d41 | 928 | 0, NULL, cached_state, mask); |
d1310b2e | 929 | } |
d1310b2e CM |
930 | |
931 | int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
932 | gfp_t mask) | |
933 | { | |
934 | return clear_extent_bit(tree, start, end, | |
32c00aff | 935 | EXTENT_DIRTY | EXTENT_DELALLOC | |
0ca1f7ce | 936 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask); |
d1310b2e | 937 | } |
d1310b2e CM |
938 | |
939 | int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, | |
940 | gfp_t mask) | |
941 | { | |
942 | return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL, | |
2c64c53d | 943 | NULL, mask); |
d1310b2e | 944 | } |
d1310b2e | 945 | |
b2950863 | 946 | static int clear_extent_new(struct extent_io_tree *tree, u64 start, u64 end, |
d1310b2e CM |
947 | gfp_t mask) |
948 | { | |
2c64c53d CM |
949 | return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, |
950 | NULL, mask); | |
d1310b2e | 951 | } |
d1310b2e CM |
952 | |
953 | int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, | |
507903b8 | 954 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e | 955 | { |
507903b8 AJ |
956 | return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, |
957 | NULL, cached_state, mask); | |
d1310b2e | 958 | } |
d1310b2e | 959 | |
d397712b | 960 | static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, |
2ac55d41 JB |
961 | u64 end, struct extent_state **cached_state, |
962 | gfp_t mask) | |
d1310b2e | 963 | { |
2c64c53d | 964 | return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, |
2ac55d41 | 965 | cached_state, mask); |
d1310b2e | 966 | } |
d1310b2e | 967 | |
d1310b2e CM |
968 | int wait_on_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
969 | { | |
970 | return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK); | |
971 | } | |
d1310b2e | 972 | |
d352ac68 CM |
973 | /* |
974 | * either insert or lock state struct between start and end use mask to tell | |
975 | * us if waiting is desired. | |
976 | */ | |
1edbb734 | 977 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
2c64c53d | 978 | int bits, struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
979 | { |
980 | int err; | |
981 | u64 failed_start; | |
982 | while (1) { | |
1edbb734 | 983 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits, |
2c64c53d CM |
984 | EXTENT_LOCKED, &failed_start, |
985 | cached_state, mask); | |
d1310b2e CM |
986 | if (err == -EEXIST && (mask & __GFP_WAIT)) { |
987 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); | |
988 | start = failed_start; | |
989 | } else { | |
990 | break; | |
991 | } | |
992 | WARN_ON(start > end); | |
993 | } | |
994 | return err; | |
995 | } | |
d1310b2e | 996 | |
1edbb734 CM |
997 | int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) |
998 | { | |
2c64c53d | 999 | return lock_extent_bits(tree, start, end, 0, NULL, mask); |
1edbb734 CM |
1000 | } |
1001 | ||
25179201 JB |
1002 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end, |
1003 | gfp_t mask) | |
1004 | { | |
1005 | int err; | |
1006 | u64 failed_start; | |
1007 | ||
2c64c53d CM |
1008 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
1009 | &failed_start, NULL, mask); | |
6643558d YZ |
1010 | if (err == -EEXIST) { |
1011 | if (failed_start > start) | |
1012 | clear_extent_bit(tree, start, failed_start - 1, | |
2c64c53d | 1013 | EXTENT_LOCKED, 1, 0, NULL, mask); |
25179201 | 1014 | return 0; |
6643558d | 1015 | } |
25179201 JB |
1016 | return 1; |
1017 | } | |
25179201 | 1018 | |
2c64c53d CM |
1019 | int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end, |
1020 | struct extent_state **cached, gfp_t mask) | |
1021 | { | |
1022 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, | |
1023 | mask); | |
1024 | } | |
1025 | ||
507903b8 | 1026 | int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) |
d1310b2e | 1027 | { |
2c64c53d CM |
1028 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL, |
1029 | mask); | |
d1310b2e | 1030 | } |
d1310b2e CM |
1031 | |
1032 | /* | |
1033 | * helper function to set pages and extents in the tree dirty | |
1034 | */ | |
1035 | int set_range_dirty(struct extent_io_tree *tree, u64 start, u64 end) | |
1036 | { | |
1037 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1038 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1039 | struct page *page; | |
1040 | ||
1041 | while (index <= end_index) { | |
1042 | page = find_get_page(tree->mapping, index); | |
1043 | BUG_ON(!page); | |
1044 | __set_page_dirty_nobuffers(page); | |
1045 | page_cache_release(page); | |
1046 | index++; | |
1047 | } | |
d1310b2e CM |
1048 | return 0; |
1049 | } | |
d1310b2e CM |
1050 | |
1051 | /* | |
1052 | * helper function to set both pages and extents in the tree writeback | |
1053 | */ | |
b2950863 | 1054 | static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e CM |
1055 | { |
1056 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1057 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1058 | struct page *page; | |
1059 | ||
1060 | while (index <= end_index) { | |
1061 | page = find_get_page(tree->mapping, index); | |
1062 | BUG_ON(!page); | |
1063 | set_page_writeback(page); | |
1064 | page_cache_release(page); | |
1065 | index++; | |
1066 | } | |
d1310b2e CM |
1067 | return 0; |
1068 | } | |
d1310b2e | 1069 | |
d352ac68 CM |
1070 | /* |
1071 | * find the first offset in the io tree with 'bits' set. zero is | |
1072 | * returned if we find something, and *start_ret and *end_ret are | |
1073 | * set to reflect the state struct that was found. | |
1074 | * | |
1075 | * If nothing was found, 1 is returned, < 0 on error | |
1076 | */ | |
d1310b2e CM |
1077 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, |
1078 | u64 *start_ret, u64 *end_ret, int bits) | |
1079 | { | |
1080 | struct rb_node *node; | |
1081 | struct extent_state *state; | |
1082 | int ret = 1; | |
1083 | ||
cad321ad | 1084 | spin_lock(&tree->lock); |
d1310b2e CM |
1085 | /* |
1086 | * this search will find all the extents that end after | |
1087 | * our range starts. | |
1088 | */ | |
80ea96b1 | 1089 | node = tree_search(tree, start); |
d397712b | 1090 | if (!node) |
d1310b2e | 1091 | goto out; |
d1310b2e | 1092 | |
d397712b | 1093 | while (1) { |
d1310b2e CM |
1094 | state = rb_entry(node, struct extent_state, rb_node); |
1095 | if (state->end >= start && (state->state & bits)) { | |
1096 | *start_ret = state->start; | |
1097 | *end_ret = state->end; | |
1098 | ret = 0; | |
1099 | break; | |
1100 | } | |
1101 | node = rb_next(node); | |
1102 | if (!node) | |
1103 | break; | |
1104 | } | |
1105 | out: | |
cad321ad | 1106 | spin_unlock(&tree->lock); |
d1310b2e CM |
1107 | return ret; |
1108 | } | |
d1310b2e | 1109 | |
d352ac68 CM |
1110 | /* find the first state struct with 'bits' set after 'start', and |
1111 | * return it. tree->lock must be held. NULL will returned if | |
1112 | * nothing was found after 'start' | |
1113 | */ | |
d7fc640e CM |
1114 | struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree, |
1115 | u64 start, int bits) | |
1116 | { | |
1117 | struct rb_node *node; | |
1118 | struct extent_state *state; | |
1119 | ||
1120 | /* | |
1121 | * this search will find all the extents that end after | |
1122 | * our range starts. | |
1123 | */ | |
1124 | node = tree_search(tree, start); | |
d397712b | 1125 | if (!node) |
d7fc640e | 1126 | goto out; |
d7fc640e | 1127 | |
d397712b | 1128 | while (1) { |
d7fc640e | 1129 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1130 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1131 | return state; |
d397712b | 1132 | |
d7fc640e CM |
1133 | node = rb_next(node); |
1134 | if (!node) | |
1135 | break; | |
1136 | } | |
1137 | out: | |
1138 | return NULL; | |
1139 | } | |
d7fc640e | 1140 | |
d352ac68 CM |
1141 | /* |
1142 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1143 | * more than 'max_bytes'. start and end are used to return the range, | |
1144 | * | |
1145 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1146 | */ | |
c8b97818 | 1147 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1148 | u64 *start, u64 *end, u64 max_bytes, |
1149 | struct extent_state **cached_state) | |
d1310b2e CM |
1150 | { |
1151 | struct rb_node *node; | |
1152 | struct extent_state *state; | |
1153 | u64 cur_start = *start; | |
1154 | u64 found = 0; | |
1155 | u64 total_bytes = 0; | |
1156 | ||
cad321ad | 1157 | spin_lock(&tree->lock); |
c8b97818 | 1158 | |
d1310b2e CM |
1159 | /* |
1160 | * this search will find all the extents that end after | |
1161 | * our range starts. | |
1162 | */ | |
80ea96b1 | 1163 | node = tree_search(tree, cur_start); |
2b114d1d | 1164 | if (!node) { |
3b951516 CM |
1165 | if (!found) |
1166 | *end = (u64)-1; | |
d1310b2e CM |
1167 | goto out; |
1168 | } | |
1169 | ||
d397712b | 1170 | while (1) { |
d1310b2e | 1171 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1172 | if (found && (state->start != cur_start || |
1173 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1174 | goto out; |
1175 | } | |
1176 | if (!(state->state & EXTENT_DELALLOC)) { | |
1177 | if (!found) | |
1178 | *end = state->end; | |
1179 | goto out; | |
1180 | } | |
c2a128d2 | 1181 | if (!found) { |
d1310b2e | 1182 | *start = state->start; |
c2a128d2 JB |
1183 | *cached_state = state; |
1184 | atomic_inc(&state->refs); | |
1185 | } | |
d1310b2e CM |
1186 | found++; |
1187 | *end = state->end; | |
1188 | cur_start = state->end + 1; | |
1189 | node = rb_next(node); | |
1190 | if (!node) | |
1191 | break; | |
1192 | total_bytes += state->end - state->start + 1; | |
1193 | if (total_bytes >= max_bytes) | |
1194 | break; | |
1195 | } | |
1196 | out: | |
cad321ad | 1197 | spin_unlock(&tree->lock); |
d1310b2e CM |
1198 | return found; |
1199 | } | |
1200 | ||
c8b97818 CM |
1201 | static noinline int __unlock_for_delalloc(struct inode *inode, |
1202 | struct page *locked_page, | |
1203 | u64 start, u64 end) | |
1204 | { | |
1205 | int ret; | |
1206 | struct page *pages[16]; | |
1207 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1208 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1209 | unsigned long nr_pages = end_index - index + 1; | |
1210 | int i; | |
1211 | ||
1212 | if (index == locked_page->index && end_index == index) | |
1213 | return 0; | |
1214 | ||
d397712b | 1215 | while (nr_pages > 0) { |
c8b97818 | 1216 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1217 | min_t(unsigned long, nr_pages, |
1218 | ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1219 | for (i = 0; i < ret; i++) { |
1220 | if (pages[i] != locked_page) | |
1221 | unlock_page(pages[i]); | |
1222 | page_cache_release(pages[i]); | |
1223 | } | |
1224 | nr_pages -= ret; | |
1225 | index += ret; | |
1226 | cond_resched(); | |
1227 | } | |
1228 | return 0; | |
1229 | } | |
1230 | ||
1231 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1232 | struct page *locked_page, | |
1233 | u64 delalloc_start, | |
1234 | u64 delalloc_end) | |
1235 | { | |
1236 | unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT; | |
1237 | unsigned long start_index = index; | |
1238 | unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT; | |
1239 | unsigned long pages_locked = 0; | |
1240 | struct page *pages[16]; | |
1241 | unsigned long nrpages; | |
1242 | int ret; | |
1243 | int i; | |
1244 | ||
1245 | /* the caller is responsible for locking the start index */ | |
1246 | if (index == locked_page->index && index == end_index) | |
1247 | return 0; | |
1248 | ||
1249 | /* skip the page at the start index */ | |
1250 | nrpages = end_index - index + 1; | |
d397712b | 1251 | while (nrpages > 0) { |
c8b97818 | 1252 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1253 | min_t(unsigned long, |
1254 | nrpages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1255 | if (ret == 0) { |
1256 | ret = -EAGAIN; | |
1257 | goto done; | |
1258 | } | |
1259 | /* now we have an array of pages, lock them all */ | |
1260 | for (i = 0; i < ret; i++) { | |
1261 | /* | |
1262 | * the caller is taking responsibility for | |
1263 | * locked_page | |
1264 | */ | |
771ed689 | 1265 | if (pages[i] != locked_page) { |
c8b97818 | 1266 | lock_page(pages[i]); |
f2b1c41c CM |
1267 | if (!PageDirty(pages[i]) || |
1268 | pages[i]->mapping != inode->i_mapping) { | |
771ed689 CM |
1269 | ret = -EAGAIN; |
1270 | unlock_page(pages[i]); | |
1271 | page_cache_release(pages[i]); | |
1272 | goto done; | |
1273 | } | |
1274 | } | |
c8b97818 | 1275 | page_cache_release(pages[i]); |
771ed689 | 1276 | pages_locked++; |
c8b97818 | 1277 | } |
c8b97818 CM |
1278 | nrpages -= ret; |
1279 | index += ret; | |
1280 | cond_resched(); | |
1281 | } | |
1282 | ret = 0; | |
1283 | done: | |
1284 | if (ret && pages_locked) { | |
1285 | __unlock_for_delalloc(inode, locked_page, | |
1286 | delalloc_start, | |
1287 | ((u64)(start_index + pages_locked - 1)) << | |
1288 | PAGE_CACHE_SHIFT); | |
1289 | } | |
1290 | return ret; | |
1291 | } | |
1292 | ||
1293 | /* | |
1294 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1295 | * more than 'max_bytes'. start and end are used to return the range, | |
1296 | * | |
1297 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1298 | */ | |
1299 | static noinline u64 find_lock_delalloc_range(struct inode *inode, | |
1300 | struct extent_io_tree *tree, | |
1301 | struct page *locked_page, | |
1302 | u64 *start, u64 *end, | |
1303 | u64 max_bytes) | |
1304 | { | |
1305 | u64 delalloc_start; | |
1306 | u64 delalloc_end; | |
1307 | u64 found; | |
9655d298 | 1308 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1309 | int ret; |
1310 | int loops = 0; | |
1311 | ||
1312 | again: | |
1313 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1314 | delalloc_start = *start; | |
1315 | delalloc_end = 0; | |
1316 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1317 | max_bytes, &cached_state); |
70b99e69 | 1318 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1319 | *start = delalloc_start; |
1320 | *end = delalloc_end; | |
c2a128d2 | 1321 | free_extent_state(cached_state); |
c8b97818 CM |
1322 | return found; |
1323 | } | |
1324 | ||
70b99e69 CM |
1325 | /* |
1326 | * start comes from the offset of locked_page. We have to lock | |
1327 | * pages in order, so we can't process delalloc bytes before | |
1328 | * locked_page | |
1329 | */ | |
d397712b | 1330 | if (delalloc_start < *start) |
70b99e69 | 1331 | delalloc_start = *start; |
70b99e69 | 1332 | |
c8b97818 CM |
1333 | /* |
1334 | * make sure to limit the number of pages we try to lock down | |
1335 | * if we're looping. | |
1336 | */ | |
d397712b | 1337 | if (delalloc_end + 1 - delalloc_start > max_bytes && loops) |
771ed689 | 1338 | delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1; |
d397712b | 1339 | |
c8b97818 CM |
1340 | /* step two, lock all the pages after the page that has start */ |
1341 | ret = lock_delalloc_pages(inode, locked_page, | |
1342 | delalloc_start, delalloc_end); | |
1343 | if (ret == -EAGAIN) { | |
1344 | /* some of the pages are gone, lets avoid looping by | |
1345 | * shortening the size of the delalloc range we're searching | |
1346 | */ | |
9655d298 | 1347 | free_extent_state(cached_state); |
c8b97818 CM |
1348 | if (!loops) { |
1349 | unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1); | |
1350 | max_bytes = PAGE_CACHE_SIZE - offset; | |
1351 | loops = 1; | |
1352 | goto again; | |
1353 | } else { | |
1354 | found = 0; | |
1355 | goto out_failed; | |
1356 | } | |
1357 | } | |
1358 | BUG_ON(ret); | |
1359 | ||
1360 | /* step three, lock the state bits for the whole range */ | |
9655d298 CM |
1361 | lock_extent_bits(tree, delalloc_start, delalloc_end, |
1362 | 0, &cached_state, GFP_NOFS); | |
c8b97818 CM |
1363 | |
1364 | /* then test to make sure it is all still delalloc */ | |
1365 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1366 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1367 | if (!ret) { |
9655d298 CM |
1368 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
1369 | &cached_state, GFP_NOFS); | |
c8b97818 CM |
1370 | __unlock_for_delalloc(inode, locked_page, |
1371 | delalloc_start, delalloc_end); | |
1372 | cond_resched(); | |
1373 | goto again; | |
1374 | } | |
9655d298 | 1375 | free_extent_state(cached_state); |
c8b97818 CM |
1376 | *start = delalloc_start; |
1377 | *end = delalloc_end; | |
1378 | out_failed: | |
1379 | return found; | |
1380 | } | |
1381 | ||
1382 | int extent_clear_unlock_delalloc(struct inode *inode, | |
1383 | struct extent_io_tree *tree, | |
1384 | u64 start, u64 end, struct page *locked_page, | |
a791e35e | 1385 | unsigned long op) |
c8b97818 CM |
1386 | { |
1387 | int ret; | |
1388 | struct page *pages[16]; | |
1389 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1390 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1391 | unsigned long nr_pages = end_index - index + 1; | |
1392 | int i; | |
771ed689 | 1393 | int clear_bits = 0; |
c8b97818 | 1394 | |
a791e35e | 1395 | if (op & EXTENT_CLEAR_UNLOCK) |
771ed689 | 1396 | clear_bits |= EXTENT_LOCKED; |
a791e35e | 1397 | if (op & EXTENT_CLEAR_DIRTY) |
c8b97818 CM |
1398 | clear_bits |= EXTENT_DIRTY; |
1399 | ||
a791e35e | 1400 | if (op & EXTENT_CLEAR_DELALLOC) |
771ed689 CM |
1401 | clear_bits |= EXTENT_DELALLOC; |
1402 | ||
2c64c53d | 1403 | clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); |
32c00aff JB |
1404 | if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | |
1405 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK | | |
1406 | EXTENT_SET_PRIVATE2))) | |
771ed689 | 1407 | return 0; |
c8b97818 | 1408 | |
d397712b | 1409 | while (nr_pages > 0) { |
c8b97818 | 1410 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1411 | min_t(unsigned long, |
1412 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 | 1413 | for (i = 0; i < ret; i++) { |
8b62b72b | 1414 | |
a791e35e | 1415 | if (op & EXTENT_SET_PRIVATE2) |
8b62b72b CM |
1416 | SetPagePrivate2(pages[i]); |
1417 | ||
c8b97818 CM |
1418 | if (pages[i] == locked_page) { |
1419 | page_cache_release(pages[i]); | |
1420 | continue; | |
1421 | } | |
a791e35e | 1422 | if (op & EXTENT_CLEAR_DIRTY) |
c8b97818 | 1423 | clear_page_dirty_for_io(pages[i]); |
a791e35e | 1424 | if (op & EXTENT_SET_WRITEBACK) |
c8b97818 | 1425 | set_page_writeback(pages[i]); |
a791e35e | 1426 | if (op & EXTENT_END_WRITEBACK) |
c8b97818 | 1427 | end_page_writeback(pages[i]); |
a791e35e | 1428 | if (op & EXTENT_CLEAR_UNLOCK_PAGE) |
771ed689 | 1429 | unlock_page(pages[i]); |
c8b97818 CM |
1430 | page_cache_release(pages[i]); |
1431 | } | |
1432 | nr_pages -= ret; | |
1433 | index += ret; | |
1434 | cond_resched(); | |
1435 | } | |
1436 | return 0; | |
1437 | } | |
c8b97818 | 1438 | |
d352ac68 CM |
1439 | /* |
1440 | * count the number of bytes in the tree that have a given bit(s) | |
1441 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1442 | * cached. The total number found is returned. | |
1443 | */ | |
d1310b2e CM |
1444 | u64 count_range_bits(struct extent_io_tree *tree, |
1445 | u64 *start, u64 search_end, u64 max_bytes, | |
ec29ed5b | 1446 | unsigned long bits, int contig) |
d1310b2e CM |
1447 | { |
1448 | struct rb_node *node; | |
1449 | struct extent_state *state; | |
1450 | u64 cur_start = *start; | |
1451 | u64 total_bytes = 0; | |
ec29ed5b | 1452 | u64 last = 0; |
d1310b2e CM |
1453 | int found = 0; |
1454 | ||
1455 | if (search_end <= cur_start) { | |
d1310b2e CM |
1456 | WARN_ON(1); |
1457 | return 0; | |
1458 | } | |
1459 | ||
cad321ad | 1460 | spin_lock(&tree->lock); |
d1310b2e CM |
1461 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1462 | total_bytes = tree->dirty_bytes; | |
1463 | goto out; | |
1464 | } | |
1465 | /* | |
1466 | * this search will find all the extents that end after | |
1467 | * our range starts. | |
1468 | */ | |
80ea96b1 | 1469 | node = tree_search(tree, cur_start); |
d397712b | 1470 | if (!node) |
d1310b2e | 1471 | goto out; |
d1310b2e | 1472 | |
d397712b | 1473 | while (1) { |
d1310b2e CM |
1474 | state = rb_entry(node, struct extent_state, rb_node); |
1475 | if (state->start > search_end) | |
1476 | break; | |
ec29ed5b CM |
1477 | if (contig && found && state->start > last + 1) |
1478 | break; | |
1479 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1480 | total_bytes += min(search_end, state->end) + 1 - |
1481 | max(cur_start, state->start); | |
1482 | if (total_bytes >= max_bytes) | |
1483 | break; | |
1484 | if (!found) { | |
1485 | *start = state->start; | |
1486 | found = 1; | |
1487 | } | |
ec29ed5b CM |
1488 | last = state->end; |
1489 | } else if (contig && found) { | |
1490 | break; | |
d1310b2e CM |
1491 | } |
1492 | node = rb_next(node); | |
1493 | if (!node) | |
1494 | break; | |
1495 | } | |
1496 | out: | |
cad321ad | 1497 | spin_unlock(&tree->lock); |
d1310b2e CM |
1498 | return total_bytes; |
1499 | } | |
b2950863 | 1500 | |
d352ac68 CM |
1501 | /* |
1502 | * set the private field for a given byte offset in the tree. If there isn't | |
1503 | * an extent_state there already, this does nothing. | |
1504 | */ | |
d1310b2e CM |
1505 | int set_state_private(struct extent_io_tree *tree, u64 start, u64 private) |
1506 | { | |
1507 | struct rb_node *node; | |
1508 | struct extent_state *state; | |
1509 | int ret = 0; | |
1510 | ||
cad321ad | 1511 | spin_lock(&tree->lock); |
d1310b2e CM |
1512 | /* |
1513 | * this search will find all the extents that end after | |
1514 | * our range starts. | |
1515 | */ | |
80ea96b1 | 1516 | node = tree_search(tree, start); |
2b114d1d | 1517 | if (!node) { |
d1310b2e CM |
1518 | ret = -ENOENT; |
1519 | goto out; | |
1520 | } | |
1521 | state = rb_entry(node, struct extent_state, rb_node); | |
1522 | if (state->start != start) { | |
1523 | ret = -ENOENT; | |
1524 | goto out; | |
1525 | } | |
1526 | state->private = private; | |
1527 | out: | |
cad321ad | 1528 | spin_unlock(&tree->lock); |
d1310b2e CM |
1529 | return ret; |
1530 | } | |
1531 | ||
1532 | int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private) | |
1533 | { | |
1534 | struct rb_node *node; | |
1535 | struct extent_state *state; | |
1536 | int ret = 0; | |
1537 | ||
cad321ad | 1538 | spin_lock(&tree->lock); |
d1310b2e CM |
1539 | /* |
1540 | * this search will find all the extents that end after | |
1541 | * our range starts. | |
1542 | */ | |
80ea96b1 | 1543 | node = tree_search(tree, start); |
2b114d1d | 1544 | if (!node) { |
d1310b2e CM |
1545 | ret = -ENOENT; |
1546 | goto out; | |
1547 | } | |
1548 | state = rb_entry(node, struct extent_state, rb_node); | |
1549 | if (state->start != start) { | |
1550 | ret = -ENOENT; | |
1551 | goto out; | |
1552 | } | |
1553 | *private = state->private; | |
1554 | out: | |
cad321ad | 1555 | spin_unlock(&tree->lock); |
d1310b2e CM |
1556 | return ret; |
1557 | } | |
1558 | ||
1559 | /* | |
1560 | * searches a range in the state tree for a given mask. | |
70dec807 | 1561 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1562 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1563 | * range is found set. | |
1564 | */ | |
1565 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9655d298 | 1566 | int bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1567 | { |
1568 | struct extent_state *state = NULL; | |
1569 | struct rb_node *node; | |
1570 | int bitset = 0; | |
d1310b2e | 1571 | |
cad321ad | 1572 | spin_lock(&tree->lock); |
9655d298 CM |
1573 | if (cached && cached->tree && cached->start == start) |
1574 | node = &cached->rb_node; | |
1575 | else | |
1576 | node = tree_search(tree, start); | |
d1310b2e CM |
1577 | while (node && start <= end) { |
1578 | state = rb_entry(node, struct extent_state, rb_node); | |
1579 | ||
1580 | if (filled && state->start > start) { | |
1581 | bitset = 0; | |
1582 | break; | |
1583 | } | |
1584 | ||
1585 | if (state->start > end) | |
1586 | break; | |
1587 | ||
1588 | if (state->state & bits) { | |
1589 | bitset = 1; | |
1590 | if (!filled) | |
1591 | break; | |
1592 | } else if (filled) { | |
1593 | bitset = 0; | |
1594 | break; | |
1595 | } | |
46562cec CM |
1596 | |
1597 | if (state->end == (u64)-1) | |
1598 | break; | |
1599 | ||
d1310b2e CM |
1600 | start = state->end + 1; |
1601 | if (start > end) | |
1602 | break; | |
1603 | node = rb_next(node); | |
1604 | if (!node) { | |
1605 | if (filled) | |
1606 | bitset = 0; | |
1607 | break; | |
1608 | } | |
1609 | } | |
cad321ad | 1610 | spin_unlock(&tree->lock); |
d1310b2e CM |
1611 | return bitset; |
1612 | } | |
d1310b2e CM |
1613 | |
1614 | /* | |
1615 | * helper function to set a given page up to date if all the | |
1616 | * extents in the tree for that page are up to date | |
1617 | */ | |
1618 | static int check_page_uptodate(struct extent_io_tree *tree, | |
1619 | struct page *page) | |
1620 | { | |
1621 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1622 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
9655d298 | 1623 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e CM |
1624 | SetPageUptodate(page); |
1625 | return 0; | |
1626 | } | |
1627 | ||
1628 | /* | |
1629 | * helper function to unlock a page if all the extents in the tree | |
1630 | * for that page are unlocked | |
1631 | */ | |
1632 | static int check_page_locked(struct extent_io_tree *tree, | |
1633 | struct page *page) | |
1634 | { | |
1635 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1636 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
9655d298 | 1637 | if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) |
d1310b2e CM |
1638 | unlock_page(page); |
1639 | return 0; | |
1640 | } | |
1641 | ||
1642 | /* | |
1643 | * helper function to end page writeback if all the extents | |
1644 | * in the tree for that page are done with writeback | |
1645 | */ | |
1646 | static int check_page_writeback(struct extent_io_tree *tree, | |
1647 | struct page *page) | |
1648 | { | |
1edbb734 | 1649 | end_page_writeback(page); |
d1310b2e CM |
1650 | return 0; |
1651 | } | |
1652 | ||
1653 | /* lots and lots of room for performance fixes in the end_bio funcs */ | |
1654 | ||
1655 | /* | |
1656 | * after a writepage IO is done, we need to: | |
1657 | * clear the uptodate bits on error | |
1658 | * clear the writeback bits in the extent tree for this IO | |
1659 | * end_page_writeback if the page has no more pending IO | |
1660 | * | |
1661 | * Scheduling is not allowed, so the extent state tree is expected | |
1662 | * to have one and only one object corresponding to this IO. | |
1663 | */ | |
d1310b2e | 1664 | static void end_bio_extent_writepage(struct bio *bio, int err) |
d1310b2e | 1665 | { |
1259ab75 | 1666 | int uptodate = err == 0; |
d1310b2e | 1667 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
902b22f3 | 1668 | struct extent_io_tree *tree; |
d1310b2e CM |
1669 | u64 start; |
1670 | u64 end; | |
1671 | int whole_page; | |
1259ab75 | 1672 | int ret; |
d1310b2e | 1673 | |
d1310b2e CM |
1674 | do { |
1675 | struct page *page = bvec->bv_page; | |
902b22f3 DW |
1676 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
1677 | ||
d1310b2e CM |
1678 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
1679 | bvec->bv_offset; | |
1680 | end = start + bvec->bv_len - 1; | |
1681 | ||
1682 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
1683 | whole_page = 1; | |
1684 | else | |
1685 | whole_page = 0; | |
1686 | ||
1687 | if (--bvec >= bio->bi_io_vec) | |
1688 | prefetchw(&bvec->bv_page->flags); | |
1259ab75 CM |
1689 | if (tree->ops && tree->ops->writepage_end_io_hook) { |
1690 | ret = tree->ops->writepage_end_io_hook(page, start, | |
902b22f3 | 1691 | end, NULL, uptodate); |
1259ab75 CM |
1692 | if (ret) |
1693 | uptodate = 0; | |
1694 | } | |
1695 | ||
1696 | if (!uptodate && tree->ops && | |
1697 | tree->ops->writepage_io_failed_hook) { | |
1698 | ret = tree->ops->writepage_io_failed_hook(bio, page, | |
902b22f3 | 1699 | start, end, NULL); |
1259ab75 | 1700 | if (ret == 0) { |
1259ab75 CM |
1701 | uptodate = (err == 0); |
1702 | continue; | |
1703 | } | |
1704 | } | |
1705 | ||
d1310b2e | 1706 | if (!uptodate) { |
2ac55d41 | 1707 | clear_extent_uptodate(tree, start, end, NULL, GFP_NOFS); |
d1310b2e CM |
1708 | ClearPageUptodate(page); |
1709 | SetPageError(page); | |
1710 | } | |
70dec807 | 1711 | |
d1310b2e CM |
1712 | if (whole_page) |
1713 | end_page_writeback(page); | |
1714 | else | |
1715 | check_page_writeback(tree, page); | |
d1310b2e | 1716 | } while (bvec >= bio->bi_io_vec); |
2b1f55b0 | 1717 | |
d1310b2e | 1718 | bio_put(bio); |
d1310b2e CM |
1719 | } |
1720 | ||
1721 | /* | |
1722 | * after a readpage IO is done, we need to: | |
1723 | * clear the uptodate bits on error | |
1724 | * set the uptodate bits if things worked | |
1725 | * set the page up to date if all extents in the tree are uptodate | |
1726 | * clear the lock bit in the extent tree | |
1727 | * unlock the page if there are no other extents locked for it | |
1728 | * | |
1729 | * Scheduling is not allowed, so the extent state tree is expected | |
1730 | * to have one and only one object corresponding to this IO. | |
1731 | */ | |
d1310b2e | 1732 | static void end_bio_extent_readpage(struct bio *bio, int err) |
d1310b2e CM |
1733 | { |
1734 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
4125bf76 CM |
1735 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
1736 | struct bio_vec *bvec = bio->bi_io_vec; | |
902b22f3 | 1737 | struct extent_io_tree *tree; |
d1310b2e CM |
1738 | u64 start; |
1739 | u64 end; | |
1740 | int whole_page; | |
1741 | int ret; | |
1742 | ||
d20f7043 CM |
1743 | if (err) |
1744 | uptodate = 0; | |
1745 | ||
d1310b2e CM |
1746 | do { |
1747 | struct page *page = bvec->bv_page; | |
507903b8 AJ |
1748 | struct extent_state *cached = NULL; |
1749 | struct extent_state *state; | |
1750 | ||
902b22f3 DW |
1751 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
1752 | ||
d1310b2e CM |
1753 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
1754 | bvec->bv_offset; | |
1755 | end = start + bvec->bv_len - 1; | |
1756 | ||
1757 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
1758 | whole_page = 1; | |
1759 | else | |
1760 | whole_page = 0; | |
1761 | ||
4125bf76 | 1762 | if (++bvec <= bvec_end) |
d1310b2e CM |
1763 | prefetchw(&bvec->bv_page->flags); |
1764 | ||
507903b8 AJ |
1765 | spin_lock(&tree->lock); |
1766 | state = find_first_extent_bit_state(tree, start, 0); | |
1767 | if (state) { | |
1768 | /* | |
1769 | * take a reference on the state, unlock will drop | |
1770 | * the ref | |
1771 | */ | |
1772 | cache_state(state, &cached); | |
1773 | } | |
1774 | spin_unlock(&tree->lock); | |
1775 | ||
d1310b2e | 1776 | if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) { |
70dec807 | 1777 | ret = tree->ops->readpage_end_io_hook(page, start, end, |
507903b8 | 1778 | state); |
d1310b2e CM |
1779 | if (ret) |
1780 | uptodate = 0; | |
1781 | } | |
7e38326f CM |
1782 | if (!uptodate && tree->ops && |
1783 | tree->ops->readpage_io_failed_hook) { | |
1784 | ret = tree->ops->readpage_io_failed_hook(bio, page, | |
902b22f3 | 1785 | start, end, NULL); |
7e38326f | 1786 | if (ret == 0) { |
3b951516 CM |
1787 | uptodate = |
1788 | test_bit(BIO_UPTODATE, &bio->bi_flags); | |
d20f7043 CM |
1789 | if (err) |
1790 | uptodate = 0; | |
507903b8 | 1791 | uncache_state(&cached); |
7e38326f CM |
1792 | continue; |
1793 | } | |
1794 | } | |
d1310b2e | 1795 | |
771ed689 | 1796 | if (uptodate) { |
507903b8 | 1797 | set_extent_uptodate(tree, start, end, &cached, |
902b22f3 | 1798 | GFP_ATOMIC); |
771ed689 | 1799 | } |
507903b8 | 1800 | unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC); |
d1310b2e | 1801 | |
70dec807 CM |
1802 | if (whole_page) { |
1803 | if (uptodate) { | |
1804 | SetPageUptodate(page); | |
1805 | } else { | |
1806 | ClearPageUptodate(page); | |
1807 | SetPageError(page); | |
1808 | } | |
d1310b2e | 1809 | unlock_page(page); |
70dec807 CM |
1810 | } else { |
1811 | if (uptodate) { | |
1812 | check_page_uptodate(tree, page); | |
1813 | } else { | |
1814 | ClearPageUptodate(page); | |
1815 | SetPageError(page); | |
1816 | } | |
d1310b2e | 1817 | check_page_locked(tree, page); |
70dec807 | 1818 | } |
4125bf76 | 1819 | } while (bvec <= bvec_end); |
d1310b2e CM |
1820 | |
1821 | bio_put(bio); | |
d1310b2e CM |
1822 | } |
1823 | ||
1824 | /* | |
1825 | * IO done from prepare_write is pretty simple, we just unlock | |
1826 | * the structs in the extent tree when done, and set the uptodate bits | |
1827 | * as appropriate. | |
1828 | */ | |
d1310b2e | 1829 | static void end_bio_extent_preparewrite(struct bio *bio, int err) |
d1310b2e CM |
1830 | { |
1831 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
1832 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
902b22f3 | 1833 | struct extent_io_tree *tree; |
d1310b2e CM |
1834 | u64 start; |
1835 | u64 end; | |
1836 | ||
d1310b2e CM |
1837 | do { |
1838 | struct page *page = bvec->bv_page; | |
507903b8 | 1839 | struct extent_state *cached = NULL; |
902b22f3 DW |
1840 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
1841 | ||
d1310b2e CM |
1842 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
1843 | bvec->bv_offset; | |
1844 | end = start + bvec->bv_len - 1; | |
1845 | ||
1846 | if (--bvec >= bio->bi_io_vec) | |
1847 | prefetchw(&bvec->bv_page->flags); | |
1848 | ||
1849 | if (uptodate) { | |
507903b8 AJ |
1850 | set_extent_uptodate(tree, start, end, &cached, |
1851 | GFP_ATOMIC); | |
d1310b2e CM |
1852 | } else { |
1853 | ClearPageUptodate(page); | |
1854 | SetPageError(page); | |
1855 | } | |
1856 | ||
507903b8 | 1857 | unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC); |
d1310b2e CM |
1858 | |
1859 | } while (bvec >= bio->bi_io_vec); | |
1860 | ||
1861 | bio_put(bio); | |
d1310b2e CM |
1862 | } |
1863 | ||
88f794ed MX |
1864 | struct bio * |
1865 | btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
1866 | gfp_t gfp_flags) | |
d1310b2e CM |
1867 | { |
1868 | struct bio *bio; | |
1869 | ||
1870 | bio = bio_alloc(gfp_flags, nr_vecs); | |
1871 | ||
1872 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
1873 | while (!bio && (nr_vecs /= 2)) | |
1874 | bio = bio_alloc(gfp_flags, nr_vecs); | |
1875 | } | |
1876 | ||
1877 | if (bio) { | |
e1c4b745 | 1878 | bio->bi_size = 0; |
d1310b2e CM |
1879 | bio->bi_bdev = bdev; |
1880 | bio->bi_sector = first_sector; | |
1881 | } | |
1882 | return bio; | |
1883 | } | |
1884 | ||
c8b97818 CM |
1885 | static int submit_one_bio(int rw, struct bio *bio, int mirror_num, |
1886 | unsigned long bio_flags) | |
d1310b2e | 1887 | { |
d1310b2e | 1888 | int ret = 0; |
70dec807 CM |
1889 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
1890 | struct page *page = bvec->bv_page; | |
1891 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 1892 | u64 start; |
70dec807 CM |
1893 | |
1894 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset; | |
70dec807 | 1895 | |
902b22f3 | 1896 | bio->bi_private = NULL; |
d1310b2e CM |
1897 | |
1898 | bio_get(bio); | |
1899 | ||
065631f6 | 1900 | if (tree->ops && tree->ops->submit_bio_hook) |
6b82ce8d | 1901 | ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio, |
eaf25d93 | 1902 | mirror_num, bio_flags, start); |
0b86a832 CM |
1903 | else |
1904 | submit_bio(rw, bio); | |
d1310b2e CM |
1905 | if (bio_flagged(bio, BIO_EOPNOTSUPP)) |
1906 | ret = -EOPNOTSUPP; | |
1907 | bio_put(bio); | |
1908 | return ret; | |
1909 | } | |
1910 | ||
1911 | static int submit_extent_page(int rw, struct extent_io_tree *tree, | |
1912 | struct page *page, sector_t sector, | |
1913 | size_t size, unsigned long offset, | |
1914 | struct block_device *bdev, | |
1915 | struct bio **bio_ret, | |
1916 | unsigned long max_pages, | |
f188591e | 1917 | bio_end_io_t end_io_func, |
c8b97818 CM |
1918 | int mirror_num, |
1919 | unsigned long prev_bio_flags, | |
1920 | unsigned long bio_flags) | |
d1310b2e CM |
1921 | { |
1922 | int ret = 0; | |
1923 | struct bio *bio; | |
1924 | int nr; | |
c8b97818 CM |
1925 | int contig = 0; |
1926 | int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED; | |
1927 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; | |
5b050f04 | 1928 | size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE); |
d1310b2e CM |
1929 | |
1930 | if (bio_ret && *bio_ret) { | |
1931 | bio = *bio_ret; | |
c8b97818 CM |
1932 | if (old_compressed) |
1933 | contig = bio->bi_sector == sector; | |
1934 | else | |
1935 | contig = bio->bi_sector + (bio->bi_size >> 9) == | |
1936 | sector; | |
1937 | ||
1938 | if (prev_bio_flags != bio_flags || !contig || | |
239b14b3 | 1939 | (tree->ops && tree->ops->merge_bio_hook && |
c8b97818 CM |
1940 | tree->ops->merge_bio_hook(page, offset, page_size, bio, |
1941 | bio_flags)) || | |
1942 | bio_add_page(bio, page, page_size, offset) < page_size) { | |
1943 | ret = submit_one_bio(rw, bio, mirror_num, | |
1944 | prev_bio_flags); | |
d1310b2e CM |
1945 | bio = NULL; |
1946 | } else { | |
1947 | return 0; | |
1948 | } | |
1949 | } | |
c8b97818 CM |
1950 | if (this_compressed) |
1951 | nr = BIO_MAX_PAGES; | |
1952 | else | |
1953 | nr = bio_get_nr_vecs(bdev); | |
1954 | ||
88f794ed | 1955 | bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH); |
5df67083 TI |
1956 | if (!bio) |
1957 | return -ENOMEM; | |
70dec807 | 1958 | |
c8b97818 | 1959 | bio_add_page(bio, page, page_size, offset); |
d1310b2e CM |
1960 | bio->bi_end_io = end_io_func; |
1961 | bio->bi_private = tree; | |
70dec807 | 1962 | |
d397712b | 1963 | if (bio_ret) |
d1310b2e | 1964 | *bio_ret = bio; |
d397712b | 1965 | else |
c8b97818 | 1966 | ret = submit_one_bio(rw, bio, mirror_num, bio_flags); |
d1310b2e CM |
1967 | |
1968 | return ret; | |
1969 | } | |
1970 | ||
1971 | void set_page_extent_mapped(struct page *page) | |
1972 | { | |
1973 | if (!PagePrivate(page)) { | |
1974 | SetPagePrivate(page); | |
d1310b2e | 1975 | page_cache_get(page); |
6af118ce | 1976 | set_page_private(page, EXTENT_PAGE_PRIVATE); |
d1310b2e CM |
1977 | } |
1978 | } | |
1979 | ||
b2950863 | 1980 | static void set_page_extent_head(struct page *page, unsigned long len) |
d1310b2e | 1981 | { |
eb14ab8e | 1982 | WARN_ON(!PagePrivate(page)); |
d1310b2e CM |
1983 | set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2); |
1984 | } | |
1985 | ||
1986 | /* | |
1987 | * basic readpage implementation. Locked extent state structs are inserted | |
1988 | * into the tree that are removed when the IO is done (by the end_io | |
1989 | * handlers) | |
1990 | */ | |
1991 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
1992 | struct page *page, | |
1993 | get_extent_t *get_extent, | |
c8b97818 CM |
1994 | struct bio **bio, int mirror_num, |
1995 | unsigned long *bio_flags) | |
d1310b2e CM |
1996 | { |
1997 | struct inode *inode = page->mapping->host; | |
1998 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1999 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2000 | u64 end; | |
2001 | u64 cur = start; | |
2002 | u64 extent_offset; | |
2003 | u64 last_byte = i_size_read(inode); | |
2004 | u64 block_start; | |
2005 | u64 cur_end; | |
2006 | sector_t sector; | |
2007 | struct extent_map *em; | |
2008 | struct block_device *bdev; | |
11c65dcc | 2009 | struct btrfs_ordered_extent *ordered; |
d1310b2e CM |
2010 | int ret; |
2011 | int nr = 0; | |
2012 | size_t page_offset = 0; | |
2013 | size_t iosize; | |
c8b97818 | 2014 | size_t disk_io_size; |
d1310b2e | 2015 | size_t blocksize = inode->i_sb->s_blocksize; |
c8b97818 | 2016 | unsigned long this_bio_flag = 0; |
d1310b2e CM |
2017 | |
2018 | set_page_extent_mapped(page); | |
2019 | ||
2020 | end = page_end; | |
11c65dcc JB |
2021 | while (1) { |
2022 | lock_extent(tree, start, end, GFP_NOFS); | |
2023 | ordered = btrfs_lookup_ordered_extent(inode, start); | |
2024 | if (!ordered) | |
2025 | break; | |
2026 | unlock_extent(tree, start, end, GFP_NOFS); | |
2027 | btrfs_start_ordered_extent(inode, ordered, 1); | |
2028 | btrfs_put_ordered_extent(ordered); | |
2029 | } | |
d1310b2e | 2030 | |
c8b97818 CM |
2031 | if (page->index == last_byte >> PAGE_CACHE_SHIFT) { |
2032 | char *userpage; | |
2033 | size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1); | |
2034 | ||
2035 | if (zero_offset) { | |
2036 | iosize = PAGE_CACHE_SIZE - zero_offset; | |
2037 | userpage = kmap_atomic(page, KM_USER0); | |
2038 | memset(userpage + zero_offset, 0, iosize); | |
2039 | flush_dcache_page(page); | |
2040 | kunmap_atomic(userpage, KM_USER0); | |
2041 | } | |
2042 | } | |
d1310b2e CM |
2043 | while (cur <= end) { |
2044 | if (cur >= last_byte) { | |
2045 | char *userpage; | |
507903b8 AJ |
2046 | struct extent_state *cached = NULL; |
2047 | ||
d1310b2e CM |
2048 | iosize = PAGE_CACHE_SIZE - page_offset; |
2049 | userpage = kmap_atomic(page, KM_USER0); | |
2050 | memset(userpage + page_offset, 0, iosize); | |
2051 | flush_dcache_page(page); | |
2052 | kunmap_atomic(userpage, KM_USER0); | |
2053 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 AJ |
2054 | &cached, GFP_NOFS); |
2055 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
2056 | &cached, GFP_NOFS); | |
d1310b2e CM |
2057 | break; |
2058 | } | |
2059 | em = get_extent(inode, page, page_offset, cur, | |
2060 | end - cur + 1, 0); | |
2061 | if (IS_ERR(em) || !em) { | |
2062 | SetPageError(page); | |
2063 | unlock_extent(tree, cur, end, GFP_NOFS); | |
2064 | break; | |
2065 | } | |
d1310b2e CM |
2066 | extent_offset = cur - em->start; |
2067 | BUG_ON(extent_map_end(em) <= cur); | |
2068 | BUG_ON(end < cur); | |
2069 | ||
261507a0 | 2070 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
c8b97818 | 2071 | this_bio_flag = EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
2072 | extent_set_compress_type(&this_bio_flag, |
2073 | em->compress_type); | |
2074 | } | |
c8b97818 | 2075 | |
d1310b2e CM |
2076 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
2077 | cur_end = min(extent_map_end(em) - 1, end); | |
2078 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
c8b97818 CM |
2079 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
2080 | disk_io_size = em->block_len; | |
2081 | sector = em->block_start >> 9; | |
2082 | } else { | |
2083 | sector = (em->block_start + extent_offset) >> 9; | |
2084 | disk_io_size = iosize; | |
2085 | } | |
d1310b2e CM |
2086 | bdev = em->bdev; |
2087 | block_start = em->block_start; | |
d899e052 YZ |
2088 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2089 | block_start = EXTENT_MAP_HOLE; | |
d1310b2e CM |
2090 | free_extent_map(em); |
2091 | em = NULL; | |
2092 | ||
2093 | /* we've found a hole, just zero and go on */ | |
2094 | if (block_start == EXTENT_MAP_HOLE) { | |
2095 | char *userpage; | |
507903b8 AJ |
2096 | struct extent_state *cached = NULL; |
2097 | ||
d1310b2e CM |
2098 | userpage = kmap_atomic(page, KM_USER0); |
2099 | memset(userpage + page_offset, 0, iosize); | |
2100 | flush_dcache_page(page); | |
2101 | kunmap_atomic(userpage, KM_USER0); | |
2102 | ||
2103 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 AJ |
2104 | &cached, GFP_NOFS); |
2105 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
2106 | &cached, GFP_NOFS); | |
d1310b2e CM |
2107 | cur = cur + iosize; |
2108 | page_offset += iosize; | |
2109 | continue; | |
2110 | } | |
2111 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
2112 | if (test_range_bit(tree, cur, cur_end, |
2113 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 2114 | check_page_uptodate(tree, page); |
d1310b2e CM |
2115 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); |
2116 | cur = cur + iosize; | |
2117 | page_offset += iosize; | |
2118 | continue; | |
2119 | } | |
70dec807 CM |
2120 | /* we have an inline extent but it didn't get marked up |
2121 | * to date. Error out | |
2122 | */ | |
2123 | if (block_start == EXTENT_MAP_INLINE) { | |
2124 | SetPageError(page); | |
2125 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
2126 | cur = cur + iosize; | |
2127 | page_offset += iosize; | |
2128 | continue; | |
2129 | } | |
d1310b2e CM |
2130 | |
2131 | ret = 0; | |
2132 | if (tree->ops && tree->ops->readpage_io_hook) { | |
2133 | ret = tree->ops->readpage_io_hook(page, cur, | |
2134 | cur + iosize - 1); | |
2135 | } | |
2136 | if (!ret) { | |
89642229 CM |
2137 | unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1; |
2138 | pnr -= page->index; | |
d1310b2e | 2139 | ret = submit_extent_page(READ, tree, page, |
c8b97818 | 2140 | sector, disk_io_size, page_offset, |
89642229 | 2141 | bdev, bio, pnr, |
c8b97818 CM |
2142 | end_bio_extent_readpage, mirror_num, |
2143 | *bio_flags, | |
2144 | this_bio_flag); | |
89642229 | 2145 | nr++; |
c8b97818 | 2146 | *bio_flags = this_bio_flag; |
d1310b2e CM |
2147 | } |
2148 | if (ret) | |
2149 | SetPageError(page); | |
2150 | cur = cur + iosize; | |
2151 | page_offset += iosize; | |
d1310b2e CM |
2152 | } |
2153 | if (!nr) { | |
2154 | if (!PageError(page)) | |
2155 | SetPageUptodate(page); | |
2156 | unlock_page(page); | |
2157 | } | |
2158 | return 0; | |
2159 | } | |
2160 | ||
2161 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, | |
2162 | get_extent_t *get_extent) | |
2163 | { | |
2164 | struct bio *bio = NULL; | |
c8b97818 | 2165 | unsigned long bio_flags = 0; |
d1310b2e CM |
2166 | int ret; |
2167 | ||
c8b97818 CM |
2168 | ret = __extent_read_full_page(tree, page, get_extent, &bio, 0, |
2169 | &bio_flags); | |
d1310b2e | 2170 | if (bio) |
6b82ce8d | 2171 | ret = submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
2172 | return ret; |
2173 | } | |
d1310b2e | 2174 | |
11c8349b CM |
2175 | static noinline void update_nr_written(struct page *page, |
2176 | struct writeback_control *wbc, | |
2177 | unsigned long nr_written) | |
2178 | { | |
2179 | wbc->nr_to_write -= nr_written; | |
2180 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && | |
2181 | wbc->range_start == 0 && wbc->range_end == LLONG_MAX)) | |
2182 | page->mapping->writeback_index = page->index + nr_written; | |
2183 | } | |
2184 | ||
d1310b2e CM |
2185 | /* |
2186 | * the writepage semantics are similar to regular writepage. extent | |
2187 | * records are inserted to lock ranges in the tree, and as dirty areas | |
2188 | * are found, they are marked writeback. Then the lock bits are removed | |
2189 | * and the end_io handler clears the writeback ranges | |
2190 | */ | |
2191 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
2192 | void *data) | |
2193 | { | |
2194 | struct inode *inode = page->mapping->host; | |
2195 | struct extent_page_data *epd = data; | |
2196 | struct extent_io_tree *tree = epd->tree; | |
2197 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2198 | u64 delalloc_start; | |
2199 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2200 | u64 end; | |
2201 | u64 cur = start; | |
2202 | u64 extent_offset; | |
2203 | u64 last_byte = i_size_read(inode); | |
2204 | u64 block_start; | |
2205 | u64 iosize; | |
2206 | sector_t sector; | |
2c64c53d | 2207 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
2208 | struct extent_map *em; |
2209 | struct block_device *bdev; | |
2210 | int ret; | |
2211 | int nr = 0; | |
7f3c74fb | 2212 | size_t pg_offset = 0; |
d1310b2e CM |
2213 | size_t blocksize; |
2214 | loff_t i_size = i_size_read(inode); | |
2215 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
2216 | u64 nr_delalloc; | |
2217 | u64 delalloc_end; | |
c8b97818 CM |
2218 | int page_started; |
2219 | int compressed; | |
ffbd517d | 2220 | int write_flags; |
771ed689 | 2221 | unsigned long nr_written = 0; |
d1310b2e | 2222 | |
ffbd517d CM |
2223 | if (wbc->sync_mode == WB_SYNC_ALL) |
2224 | write_flags = WRITE_SYNC_PLUG; | |
2225 | else | |
2226 | write_flags = WRITE; | |
2227 | ||
1abe9b8a | 2228 | trace___extent_writepage(page, inode, wbc); |
2229 | ||
d1310b2e | 2230 | WARN_ON(!PageLocked(page)); |
7f3c74fb | 2231 | pg_offset = i_size & (PAGE_CACHE_SIZE - 1); |
211c17f5 | 2232 | if (page->index > end_index || |
7f3c74fb | 2233 | (page->index == end_index && !pg_offset)) { |
39be25cd | 2234 | page->mapping->a_ops->invalidatepage(page, 0); |
d1310b2e CM |
2235 | unlock_page(page); |
2236 | return 0; | |
2237 | } | |
2238 | ||
2239 | if (page->index == end_index) { | |
2240 | char *userpage; | |
2241 | ||
d1310b2e | 2242 | userpage = kmap_atomic(page, KM_USER0); |
7f3c74fb CM |
2243 | memset(userpage + pg_offset, 0, |
2244 | PAGE_CACHE_SIZE - pg_offset); | |
d1310b2e | 2245 | kunmap_atomic(userpage, KM_USER0); |
211c17f5 | 2246 | flush_dcache_page(page); |
d1310b2e | 2247 | } |
7f3c74fb | 2248 | pg_offset = 0; |
d1310b2e CM |
2249 | |
2250 | set_page_extent_mapped(page); | |
2251 | ||
2252 | delalloc_start = start; | |
2253 | delalloc_end = 0; | |
c8b97818 | 2254 | page_started = 0; |
771ed689 | 2255 | if (!epd->extent_locked) { |
f85d7d6c | 2256 | u64 delalloc_to_write = 0; |
11c8349b CM |
2257 | /* |
2258 | * make sure the wbc mapping index is at least updated | |
2259 | * to this page. | |
2260 | */ | |
2261 | update_nr_written(page, wbc, 0); | |
2262 | ||
d397712b | 2263 | while (delalloc_end < page_end) { |
771ed689 | 2264 | nr_delalloc = find_lock_delalloc_range(inode, tree, |
c8b97818 CM |
2265 | page, |
2266 | &delalloc_start, | |
d1310b2e CM |
2267 | &delalloc_end, |
2268 | 128 * 1024 * 1024); | |
771ed689 CM |
2269 | if (nr_delalloc == 0) { |
2270 | delalloc_start = delalloc_end + 1; | |
2271 | continue; | |
2272 | } | |
2273 | tree->ops->fill_delalloc(inode, page, delalloc_start, | |
2274 | delalloc_end, &page_started, | |
2275 | &nr_written); | |
f85d7d6c CM |
2276 | /* |
2277 | * delalloc_end is already one less than the total | |
2278 | * length, so we don't subtract one from | |
2279 | * PAGE_CACHE_SIZE | |
2280 | */ | |
2281 | delalloc_to_write += (delalloc_end - delalloc_start + | |
2282 | PAGE_CACHE_SIZE) >> | |
2283 | PAGE_CACHE_SHIFT; | |
d1310b2e | 2284 | delalloc_start = delalloc_end + 1; |
d1310b2e | 2285 | } |
f85d7d6c CM |
2286 | if (wbc->nr_to_write < delalloc_to_write) { |
2287 | int thresh = 8192; | |
2288 | ||
2289 | if (delalloc_to_write < thresh * 2) | |
2290 | thresh = delalloc_to_write; | |
2291 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
2292 | thresh); | |
2293 | } | |
c8b97818 | 2294 | |
771ed689 CM |
2295 | /* did the fill delalloc function already unlock and start |
2296 | * the IO? | |
2297 | */ | |
2298 | if (page_started) { | |
2299 | ret = 0; | |
11c8349b CM |
2300 | /* |
2301 | * we've unlocked the page, so we can't update | |
2302 | * the mapping's writeback index, just update | |
2303 | * nr_to_write. | |
2304 | */ | |
2305 | wbc->nr_to_write -= nr_written; | |
2306 | goto done_unlocked; | |
771ed689 | 2307 | } |
c8b97818 | 2308 | } |
247e743c | 2309 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
2310 | ret = tree->ops->writepage_start_hook(page, start, |
2311 | page_end); | |
247e743c | 2312 | if (ret == -EAGAIN) { |
247e743c | 2313 | redirty_page_for_writepage(wbc, page); |
11c8349b | 2314 | update_nr_written(page, wbc, nr_written); |
247e743c | 2315 | unlock_page(page); |
771ed689 | 2316 | ret = 0; |
11c8349b | 2317 | goto done_unlocked; |
247e743c CM |
2318 | } |
2319 | } | |
2320 | ||
11c8349b CM |
2321 | /* |
2322 | * we don't want to touch the inode after unlocking the page, | |
2323 | * so we update the mapping writeback index now | |
2324 | */ | |
2325 | update_nr_written(page, wbc, nr_written + 1); | |
771ed689 | 2326 | |
d1310b2e | 2327 | end = page_end; |
d1310b2e | 2328 | if (last_byte <= start) { |
e6dcd2dc CM |
2329 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2330 | tree->ops->writepage_end_io_hook(page, start, | |
2331 | page_end, NULL, 1); | |
d1310b2e CM |
2332 | goto done; |
2333 | } | |
2334 | ||
d1310b2e CM |
2335 | blocksize = inode->i_sb->s_blocksize; |
2336 | ||
2337 | while (cur <= end) { | |
2338 | if (cur >= last_byte) { | |
e6dcd2dc CM |
2339 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2340 | tree->ops->writepage_end_io_hook(page, cur, | |
2341 | page_end, NULL, 1); | |
d1310b2e CM |
2342 | break; |
2343 | } | |
7f3c74fb | 2344 | em = epd->get_extent(inode, page, pg_offset, cur, |
d1310b2e CM |
2345 | end - cur + 1, 1); |
2346 | if (IS_ERR(em) || !em) { | |
2347 | SetPageError(page); | |
2348 | break; | |
2349 | } | |
2350 | ||
2351 | extent_offset = cur - em->start; | |
2352 | BUG_ON(extent_map_end(em) <= cur); | |
2353 | BUG_ON(end < cur); | |
2354 | iosize = min(extent_map_end(em) - cur, end - cur + 1); | |
2355 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
2356 | sector = (em->block_start + extent_offset) >> 9; | |
2357 | bdev = em->bdev; | |
2358 | block_start = em->block_start; | |
c8b97818 | 2359 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
2360 | free_extent_map(em); |
2361 | em = NULL; | |
2362 | ||
c8b97818 CM |
2363 | /* |
2364 | * compressed and inline extents are written through other | |
2365 | * paths in the FS | |
2366 | */ | |
2367 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 2368 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
2369 | /* |
2370 | * end_io notification does not happen here for | |
2371 | * compressed extents | |
2372 | */ | |
2373 | if (!compressed && tree->ops && | |
2374 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
2375 | tree->ops->writepage_end_io_hook(page, cur, |
2376 | cur + iosize - 1, | |
2377 | NULL, 1); | |
c8b97818 CM |
2378 | else if (compressed) { |
2379 | /* we don't want to end_page_writeback on | |
2380 | * a compressed extent. this happens | |
2381 | * elsewhere | |
2382 | */ | |
2383 | nr++; | |
2384 | } | |
2385 | ||
2386 | cur += iosize; | |
7f3c74fb | 2387 | pg_offset += iosize; |
d1310b2e CM |
2388 | continue; |
2389 | } | |
d1310b2e CM |
2390 | /* leave this out until we have a page_mkwrite call */ |
2391 | if (0 && !test_range_bit(tree, cur, cur + iosize - 1, | |
9655d298 | 2392 | EXTENT_DIRTY, 0, NULL)) { |
d1310b2e | 2393 | cur = cur + iosize; |
7f3c74fb | 2394 | pg_offset += iosize; |
d1310b2e CM |
2395 | continue; |
2396 | } | |
c8b97818 | 2397 | |
d1310b2e CM |
2398 | if (tree->ops && tree->ops->writepage_io_hook) { |
2399 | ret = tree->ops->writepage_io_hook(page, cur, | |
2400 | cur + iosize - 1); | |
2401 | } else { | |
2402 | ret = 0; | |
2403 | } | |
1259ab75 | 2404 | if (ret) { |
d1310b2e | 2405 | SetPageError(page); |
1259ab75 | 2406 | } else { |
d1310b2e | 2407 | unsigned long max_nr = end_index + 1; |
7f3c74fb | 2408 | |
d1310b2e CM |
2409 | set_range_writeback(tree, cur, cur + iosize - 1); |
2410 | if (!PageWriteback(page)) { | |
d397712b CM |
2411 | printk(KERN_ERR "btrfs warning page %lu not " |
2412 | "writeback, cur %llu end %llu\n", | |
2413 | page->index, (unsigned long long)cur, | |
d1310b2e CM |
2414 | (unsigned long long)end); |
2415 | } | |
2416 | ||
ffbd517d CM |
2417 | ret = submit_extent_page(write_flags, tree, page, |
2418 | sector, iosize, pg_offset, | |
2419 | bdev, &epd->bio, max_nr, | |
c8b97818 CM |
2420 | end_bio_extent_writepage, |
2421 | 0, 0, 0); | |
d1310b2e CM |
2422 | if (ret) |
2423 | SetPageError(page); | |
2424 | } | |
2425 | cur = cur + iosize; | |
7f3c74fb | 2426 | pg_offset += iosize; |
d1310b2e CM |
2427 | nr++; |
2428 | } | |
2429 | done: | |
2430 | if (nr == 0) { | |
2431 | /* make sure the mapping tag for page dirty gets cleared */ | |
2432 | set_page_writeback(page); | |
2433 | end_page_writeback(page); | |
2434 | } | |
d1310b2e | 2435 | unlock_page(page); |
771ed689 | 2436 | |
11c8349b CM |
2437 | done_unlocked: |
2438 | ||
2c64c53d CM |
2439 | /* drop our reference on any cached states */ |
2440 | free_extent_state(cached_state); | |
d1310b2e CM |
2441 | return 0; |
2442 | } | |
2443 | ||
d1310b2e | 2444 | /** |
4bef0848 | 2445 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
2446 | * @mapping: address space structure to write |
2447 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
2448 | * @writepage: function called for each page | |
2449 | * @data: data passed to writepage function | |
2450 | * | |
2451 | * If a page is already under I/O, write_cache_pages() skips it, even | |
2452 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
2453 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
2454 | * and msync() need to guarantee that all the data which was dirty at the time | |
2455 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
2456 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
2457 | * existing IO to complete. | |
2458 | */ | |
b2950863 | 2459 | static int extent_write_cache_pages(struct extent_io_tree *tree, |
4bef0848 CM |
2460 | struct address_space *mapping, |
2461 | struct writeback_control *wbc, | |
d2c3f4f6 CM |
2462 | writepage_t writepage, void *data, |
2463 | void (*flush_fn)(void *)) | |
d1310b2e | 2464 | { |
d1310b2e CM |
2465 | int ret = 0; |
2466 | int done = 0; | |
f85d7d6c | 2467 | int nr_to_write_done = 0; |
d1310b2e CM |
2468 | struct pagevec pvec; |
2469 | int nr_pages; | |
2470 | pgoff_t index; | |
2471 | pgoff_t end; /* Inclusive */ | |
2472 | int scanned = 0; | |
d1310b2e | 2473 | |
d1310b2e CM |
2474 | pagevec_init(&pvec, 0); |
2475 | if (wbc->range_cyclic) { | |
2476 | index = mapping->writeback_index; /* Start from prev offset */ | |
2477 | end = -1; | |
2478 | } else { | |
2479 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
2480 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
2481 | scanned = 1; |
2482 | } | |
2483 | retry: | |
f85d7d6c | 2484 | while (!done && !nr_to_write_done && (index <= end) && |
d1310b2e | 2485 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, |
d397712b CM |
2486 | PAGECACHE_TAG_DIRTY, min(end - index, |
2487 | (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
d1310b2e CM |
2488 | unsigned i; |
2489 | ||
2490 | scanned = 1; | |
2491 | for (i = 0; i < nr_pages; i++) { | |
2492 | struct page *page = pvec.pages[i]; | |
2493 | ||
2494 | /* | |
2495 | * At this point we hold neither mapping->tree_lock nor | |
2496 | * lock on the page itself: the page may be truncated or | |
2497 | * invalidated (changing page->mapping to NULL), or even | |
2498 | * swizzled back from swapper_space to tmpfs file | |
2499 | * mapping | |
2500 | */ | |
4bef0848 CM |
2501 | if (tree->ops && tree->ops->write_cache_pages_lock_hook) |
2502 | tree->ops->write_cache_pages_lock_hook(page); | |
2503 | else | |
2504 | lock_page(page); | |
d1310b2e CM |
2505 | |
2506 | if (unlikely(page->mapping != mapping)) { | |
2507 | unlock_page(page); | |
2508 | continue; | |
2509 | } | |
2510 | ||
2511 | if (!wbc->range_cyclic && page->index > end) { | |
2512 | done = 1; | |
2513 | unlock_page(page); | |
2514 | continue; | |
2515 | } | |
2516 | ||
d2c3f4f6 | 2517 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 CM |
2518 | if (PageWriteback(page)) |
2519 | flush_fn(data); | |
d1310b2e | 2520 | wait_on_page_writeback(page); |
d2c3f4f6 | 2521 | } |
d1310b2e CM |
2522 | |
2523 | if (PageWriteback(page) || | |
2524 | !clear_page_dirty_for_io(page)) { | |
2525 | unlock_page(page); | |
2526 | continue; | |
2527 | } | |
2528 | ||
2529 | ret = (*writepage)(page, wbc, data); | |
2530 | ||
2531 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
2532 | unlock_page(page); | |
2533 | ret = 0; | |
2534 | } | |
f85d7d6c | 2535 | if (ret) |
d1310b2e | 2536 | done = 1; |
f85d7d6c CM |
2537 | |
2538 | /* | |
2539 | * the filesystem may choose to bump up nr_to_write. | |
2540 | * We have to make sure to honor the new nr_to_write | |
2541 | * at any time | |
2542 | */ | |
2543 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
2544 | } |
2545 | pagevec_release(&pvec); | |
2546 | cond_resched(); | |
2547 | } | |
2548 | if (!scanned && !done) { | |
2549 | /* | |
2550 | * We hit the last page and there is more work to be done: wrap | |
2551 | * back to the start of the file | |
2552 | */ | |
2553 | scanned = 1; | |
2554 | index = 0; | |
2555 | goto retry; | |
2556 | } | |
d1310b2e CM |
2557 | return ret; |
2558 | } | |
d1310b2e | 2559 | |
ffbd517d | 2560 | static void flush_epd_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 2561 | { |
d2c3f4f6 | 2562 | if (epd->bio) { |
ffbd517d CM |
2563 | if (epd->sync_io) |
2564 | submit_one_bio(WRITE_SYNC, epd->bio, 0, 0); | |
2565 | else | |
2566 | submit_one_bio(WRITE, epd->bio, 0, 0); | |
d2c3f4f6 CM |
2567 | epd->bio = NULL; |
2568 | } | |
2569 | } | |
2570 | ||
ffbd517d CM |
2571 | static noinline void flush_write_bio(void *data) |
2572 | { | |
2573 | struct extent_page_data *epd = data; | |
2574 | flush_epd_write_bio(epd); | |
2575 | } | |
2576 | ||
d1310b2e CM |
2577 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, |
2578 | get_extent_t *get_extent, | |
2579 | struct writeback_control *wbc) | |
2580 | { | |
2581 | int ret; | |
2582 | struct address_space *mapping = page->mapping; | |
2583 | struct extent_page_data epd = { | |
2584 | .bio = NULL, | |
2585 | .tree = tree, | |
2586 | .get_extent = get_extent, | |
771ed689 | 2587 | .extent_locked = 0, |
ffbd517d | 2588 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
2589 | }; |
2590 | struct writeback_control wbc_writepages = { | |
d313d7a3 | 2591 | .sync_mode = wbc->sync_mode, |
d1310b2e CM |
2592 | .older_than_this = NULL, |
2593 | .nr_to_write = 64, | |
2594 | .range_start = page_offset(page) + PAGE_CACHE_SIZE, | |
2595 | .range_end = (loff_t)-1, | |
2596 | }; | |
2597 | ||
d1310b2e CM |
2598 | ret = __extent_writepage(page, wbc, &epd); |
2599 | ||
4bef0848 | 2600 | extent_write_cache_pages(tree, mapping, &wbc_writepages, |
d2c3f4f6 | 2601 | __extent_writepage, &epd, flush_write_bio); |
ffbd517d | 2602 | flush_epd_write_bio(&epd); |
d1310b2e CM |
2603 | return ret; |
2604 | } | |
d1310b2e | 2605 | |
771ed689 CM |
2606 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, |
2607 | u64 start, u64 end, get_extent_t *get_extent, | |
2608 | int mode) | |
2609 | { | |
2610 | int ret = 0; | |
2611 | struct address_space *mapping = inode->i_mapping; | |
2612 | struct page *page; | |
2613 | unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >> | |
2614 | PAGE_CACHE_SHIFT; | |
2615 | ||
2616 | struct extent_page_data epd = { | |
2617 | .bio = NULL, | |
2618 | .tree = tree, | |
2619 | .get_extent = get_extent, | |
2620 | .extent_locked = 1, | |
ffbd517d | 2621 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
2622 | }; |
2623 | struct writeback_control wbc_writepages = { | |
771ed689 CM |
2624 | .sync_mode = mode, |
2625 | .older_than_this = NULL, | |
2626 | .nr_to_write = nr_pages * 2, | |
2627 | .range_start = start, | |
2628 | .range_end = end + 1, | |
2629 | }; | |
2630 | ||
d397712b | 2631 | while (start <= end) { |
771ed689 CM |
2632 | page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); |
2633 | if (clear_page_dirty_for_io(page)) | |
2634 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
2635 | else { | |
2636 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
2637 | tree->ops->writepage_end_io_hook(page, start, | |
2638 | start + PAGE_CACHE_SIZE - 1, | |
2639 | NULL, 1); | |
2640 | unlock_page(page); | |
2641 | } | |
2642 | page_cache_release(page); | |
2643 | start += PAGE_CACHE_SIZE; | |
2644 | } | |
2645 | ||
ffbd517d | 2646 | flush_epd_write_bio(&epd); |
771ed689 CM |
2647 | return ret; |
2648 | } | |
d1310b2e CM |
2649 | |
2650 | int extent_writepages(struct extent_io_tree *tree, | |
2651 | struct address_space *mapping, | |
2652 | get_extent_t *get_extent, | |
2653 | struct writeback_control *wbc) | |
2654 | { | |
2655 | int ret = 0; | |
2656 | struct extent_page_data epd = { | |
2657 | .bio = NULL, | |
2658 | .tree = tree, | |
2659 | .get_extent = get_extent, | |
771ed689 | 2660 | .extent_locked = 0, |
ffbd517d | 2661 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
2662 | }; |
2663 | ||
4bef0848 | 2664 | ret = extent_write_cache_pages(tree, mapping, wbc, |
d2c3f4f6 CM |
2665 | __extent_writepage, &epd, |
2666 | flush_write_bio); | |
ffbd517d | 2667 | flush_epd_write_bio(&epd); |
d1310b2e CM |
2668 | return ret; |
2669 | } | |
d1310b2e CM |
2670 | |
2671 | int extent_readpages(struct extent_io_tree *tree, | |
2672 | struct address_space *mapping, | |
2673 | struct list_head *pages, unsigned nr_pages, | |
2674 | get_extent_t get_extent) | |
2675 | { | |
2676 | struct bio *bio = NULL; | |
2677 | unsigned page_idx; | |
c8b97818 | 2678 | unsigned long bio_flags = 0; |
d1310b2e | 2679 | |
d1310b2e CM |
2680 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
2681 | struct page *page = list_entry(pages->prev, struct page, lru); | |
2682 | ||
2683 | prefetchw(&page->flags); | |
2684 | list_del(&page->lru); | |
28ecb609 | 2685 | if (!add_to_page_cache_lru(page, mapping, |
d1310b2e | 2686 | page->index, GFP_KERNEL)) { |
f188591e | 2687 | __extent_read_full_page(tree, page, get_extent, |
c8b97818 | 2688 | &bio, 0, &bio_flags); |
d1310b2e CM |
2689 | } |
2690 | page_cache_release(page); | |
2691 | } | |
d1310b2e CM |
2692 | BUG_ON(!list_empty(pages)); |
2693 | if (bio) | |
c8b97818 | 2694 | submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
2695 | return 0; |
2696 | } | |
d1310b2e CM |
2697 | |
2698 | /* | |
2699 | * basic invalidatepage code, this waits on any locked or writeback | |
2700 | * ranges corresponding to the page, and then deletes any extent state | |
2701 | * records from the tree | |
2702 | */ | |
2703 | int extent_invalidatepage(struct extent_io_tree *tree, | |
2704 | struct page *page, unsigned long offset) | |
2705 | { | |
2ac55d41 | 2706 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
2707 | u64 start = ((u64)page->index << PAGE_CACHE_SHIFT); |
2708 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
2709 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
2710 | ||
d397712b | 2711 | start += (offset + blocksize - 1) & ~(blocksize - 1); |
d1310b2e CM |
2712 | if (start > end) |
2713 | return 0; | |
2714 | ||
2ac55d41 | 2715 | lock_extent_bits(tree, start, end, 0, &cached_state, GFP_NOFS); |
1edbb734 | 2716 | wait_on_page_writeback(page); |
d1310b2e | 2717 | clear_extent_bit(tree, start, end, |
32c00aff JB |
2718 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
2719 | EXTENT_DO_ACCOUNTING, | |
2ac55d41 | 2720 | 1, 1, &cached_state, GFP_NOFS); |
d1310b2e CM |
2721 | return 0; |
2722 | } | |
d1310b2e CM |
2723 | |
2724 | /* | |
2725 | * simple commit_write call, set_range_dirty is used to mark both | |
2726 | * the pages and the extent records as dirty | |
2727 | */ | |
2728 | int extent_commit_write(struct extent_io_tree *tree, | |
2729 | struct inode *inode, struct page *page, | |
2730 | unsigned from, unsigned to) | |
2731 | { | |
2732 | loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to; | |
2733 | ||
2734 | set_page_extent_mapped(page); | |
2735 | set_page_dirty(page); | |
2736 | ||
2737 | if (pos > inode->i_size) { | |
2738 | i_size_write(inode, pos); | |
2739 | mark_inode_dirty(inode); | |
2740 | } | |
2741 | return 0; | |
2742 | } | |
d1310b2e CM |
2743 | |
2744 | int extent_prepare_write(struct extent_io_tree *tree, | |
2745 | struct inode *inode, struct page *page, | |
2746 | unsigned from, unsigned to, get_extent_t *get_extent) | |
2747 | { | |
2748 | u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2749 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
2750 | u64 block_start; | |
2751 | u64 orig_block_start; | |
2752 | u64 block_end; | |
2753 | u64 cur_end; | |
2754 | struct extent_map *em; | |
2755 | unsigned blocksize = 1 << inode->i_blkbits; | |
2756 | size_t page_offset = 0; | |
2757 | size_t block_off_start; | |
2758 | size_t block_off_end; | |
2759 | int err = 0; | |
2760 | int iocount = 0; | |
2761 | int ret = 0; | |
2762 | int isnew; | |
2763 | ||
2764 | set_page_extent_mapped(page); | |
2765 | ||
2766 | block_start = (page_start + from) & ~((u64)blocksize - 1); | |
2767 | block_end = (page_start + to - 1) | (blocksize - 1); | |
2768 | orig_block_start = block_start; | |
2769 | ||
2770 | lock_extent(tree, page_start, page_end, GFP_NOFS); | |
d397712b | 2771 | while (block_start <= block_end) { |
d1310b2e CM |
2772 | em = get_extent(inode, page, page_offset, block_start, |
2773 | block_end - block_start + 1, 1); | |
d397712b | 2774 | if (IS_ERR(em) || !em) |
d1310b2e | 2775 | goto err; |
d397712b | 2776 | |
d1310b2e CM |
2777 | cur_end = min(block_end, extent_map_end(em) - 1); |
2778 | block_off_start = block_start & (PAGE_CACHE_SIZE - 1); | |
2779 | block_off_end = block_off_start + blocksize; | |
2780 | isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS); | |
2781 | ||
2782 | if (!PageUptodate(page) && isnew && | |
2783 | (block_off_end > to || block_off_start < from)) { | |
2784 | void *kaddr; | |
2785 | ||
2786 | kaddr = kmap_atomic(page, KM_USER0); | |
2787 | if (block_off_end > to) | |
2788 | memset(kaddr + to, 0, block_off_end - to); | |
2789 | if (block_off_start < from) | |
2790 | memset(kaddr + block_off_start, 0, | |
2791 | from - block_off_start); | |
2792 | flush_dcache_page(page); | |
2793 | kunmap_atomic(kaddr, KM_USER0); | |
2794 | } | |
2795 | if ((em->block_start != EXTENT_MAP_HOLE && | |
2796 | em->block_start != EXTENT_MAP_INLINE) && | |
2797 | !isnew && !PageUptodate(page) && | |
2798 | (block_off_end > to || block_off_start < from) && | |
2799 | !test_range_bit(tree, block_start, cur_end, | |
9655d298 | 2800 | EXTENT_UPTODATE, 1, NULL)) { |
d1310b2e CM |
2801 | u64 sector; |
2802 | u64 extent_offset = block_start - em->start; | |
2803 | size_t iosize; | |
2804 | sector = (em->block_start + extent_offset) >> 9; | |
2805 | iosize = (cur_end - block_start + blocksize) & | |
2806 | ~((u64)blocksize - 1); | |
2807 | /* | |
2808 | * we've already got the extent locked, but we | |
2809 | * need to split the state such that our end_bio | |
2810 | * handler can clear the lock. | |
2811 | */ | |
2812 | set_extent_bit(tree, block_start, | |
2813 | block_start + iosize - 1, | |
2c64c53d | 2814 | EXTENT_LOCKED, 0, NULL, NULL, GFP_NOFS); |
d1310b2e CM |
2815 | ret = submit_extent_page(READ, tree, page, |
2816 | sector, iosize, page_offset, em->bdev, | |
2817 | NULL, 1, | |
c8b97818 CM |
2818 | end_bio_extent_preparewrite, 0, |
2819 | 0, 0); | |
411fc6bc AK |
2820 | if (ret && !err) |
2821 | err = ret; | |
d1310b2e CM |
2822 | iocount++; |
2823 | block_start = block_start + iosize; | |
2824 | } else { | |
507903b8 AJ |
2825 | struct extent_state *cached = NULL; |
2826 | ||
2827 | set_extent_uptodate(tree, block_start, cur_end, &cached, | |
d1310b2e | 2828 | GFP_NOFS); |
507903b8 AJ |
2829 | unlock_extent_cached(tree, block_start, cur_end, |
2830 | &cached, GFP_NOFS); | |
d1310b2e CM |
2831 | block_start = cur_end + 1; |
2832 | } | |
2833 | page_offset = block_start & (PAGE_CACHE_SIZE - 1); | |
2834 | free_extent_map(em); | |
2835 | } | |
2836 | if (iocount) { | |
2837 | wait_extent_bit(tree, orig_block_start, | |
2838 | block_end, EXTENT_LOCKED); | |
2839 | } | |
2840 | check_page_uptodate(tree, page); | |
2841 | err: | |
2842 | /* FIXME, zero out newly allocated blocks on error */ | |
2843 | return err; | |
2844 | } | |
d1310b2e | 2845 | |
7b13b7b1 CM |
2846 | /* |
2847 | * a helper for releasepage, this tests for areas of the page that | |
2848 | * are locked or under IO and drops the related state bits if it is safe | |
2849 | * to drop the page. | |
2850 | */ | |
2851 | int try_release_extent_state(struct extent_map_tree *map, | |
2852 | struct extent_io_tree *tree, struct page *page, | |
2853 | gfp_t mask) | |
2854 | { | |
2855 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2856 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
2857 | int ret = 1; | |
2858 | ||
211f90e6 | 2859 | if (test_range_bit(tree, start, end, |
8b62b72b | 2860 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
2861 | ret = 0; |
2862 | else { | |
2863 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
2864 | mask = GFP_NOFS; | |
11ef160f CM |
2865 | /* |
2866 | * at this point we can safely clear everything except the | |
2867 | * locked bit and the nodatasum bit | |
2868 | */ | |
e3f24cc5 | 2869 | ret = clear_extent_bit(tree, start, end, |
11ef160f CM |
2870 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
2871 | 0, 0, NULL, mask); | |
e3f24cc5 CM |
2872 | |
2873 | /* if clear_extent_bit failed for enomem reasons, | |
2874 | * we can't allow the release to continue. | |
2875 | */ | |
2876 | if (ret < 0) | |
2877 | ret = 0; | |
2878 | else | |
2879 | ret = 1; | |
7b13b7b1 CM |
2880 | } |
2881 | return ret; | |
2882 | } | |
7b13b7b1 | 2883 | |
d1310b2e CM |
2884 | /* |
2885 | * a helper for releasepage. As long as there are no locked extents | |
2886 | * in the range corresponding to the page, both state records and extent | |
2887 | * map records are removed | |
2888 | */ | |
2889 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
2890 | struct extent_io_tree *tree, struct page *page, |
2891 | gfp_t mask) | |
d1310b2e CM |
2892 | { |
2893 | struct extent_map *em; | |
2894 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2895 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
7b13b7b1 | 2896 | |
70dec807 CM |
2897 | if ((mask & __GFP_WAIT) && |
2898 | page->mapping->host->i_size > 16 * 1024 * 1024) { | |
39b5637f | 2899 | u64 len; |
70dec807 | 2900 | while (start <= end) { |
39b5637f | 2901 | len = end - start + 1; |
890871be | 2902 | write_lock(&map->lock); |
39b5637f | 2903 | em = lookup_extent_mapping(map, start, len); |
70dec807 | 2904 | if (!em || IS_ERR(em)) { |
890871be | 2905 | write_unlock(&map->lock); |
70dec807 CM |
2906 | break; |
2907 | } | |
7f3c74fb CM |
2908 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
2909 | em->start != start) { | |
890871be | 2910 | write_unlock(&map->lock); |
70dec807 CM |
2911 | free_extent_map(em); |
2912 | break; | |
2913 | } | |
2914 | if (!test_range_bit(tree, em->start, | |
2915 | extent_map_end(em) - 1, | |
8b62b72b | 2916 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 2917 | 0, NULL)) { |
70dec807 CM |
2918 | remove_extent_mapping(map, em); |
2919 | /* once for the rb tree */ | |
2920 | free_extent_map(em); | |
2921 | } | |
2922 | start = extent_map_end(em); | |
890871be | 2923 | write_unlock(&map->lock); |
70dec807 CM |
2924 | |
2925 | /* once for us */ | |
d1310b2e CM |
2926 | free_extent_map(em); |
2927 | } | |
d1310b2e | 2928 | } |
7b13b7b1 | 2929 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 2930 | } |
d1310b2e CM |
2931 | |
2932 | sector_t extent_bmap(struct address_space *mapping, sector_t iblock, | |
2933 | get_extent_t *get_extent) | |
2934 | { | |
2935 | struct inode *inode = mapping->host; | |
2ac55d41 | 2936 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
2937 | u64 start = iblock << inode->i_blkbits; |
2938 | sector_t sector = 0; | |
d899e052 | 2939 | size_t blksize = (1 << inode->i_blkbits); |
d1310b2e CM |
2940 | struct extent_map *em; |
2941 | ||
2ac55d41 JB |
2942 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + blksize - 1, |
2943 | 0, &cached_state, GFP_NOFS); | |
d899e052 | 2944 | em = get_extent(inode, NULL, 0, start, blksize, 0); |
2ac55d41 JB |
2945 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, |
2946 | start + blksize - 1, &cached_state, GFP_NOFS); | |
d1310b2e CM |
2947 | if (!em || IS_ERR(em)) |
2948 | return 0; | |
2949 | ||
d899e052 | 2950 | if (em->block_start > EXTENT_MAP_LAST_BYTE) |
d1310b2e CM |
2951 | goto out; |
2952 | ||
2953 | sector = (em->block_start + start - em->start) >> inode->i_blkbits; | |
d1310b2e CM |
2954 | out: |
2955 | free_extent_map(em); | |
2956 | return sector; | |
2957 | } | |
2958 | ||
ec29ed5b CM |
2959 | /* |
2960 | * helper function for fiemap, which doesn't want to see any holes. | |
2961 | * This maps until we find something past 'last' | |
2962 | */ | |
2963 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
2964 | u64 offset, | |
2965 | u64 last, | |
2966 | get_extent_t *get_extent) | |
2967 | { | |
2968 | u64 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
2969 | struct extent_map *em; | |
2970 | u64 len; | |
2971 | ||
2972 | if (offset >= last) | |
2973 | return NULL; | |
2974 | ||
2975 | while(1) { | |
2976 | len = last - offset; | |
2977 | if (len == 0) | |
2978 | break; | |
2979 | len = (len + sectorsize - 1) & ~(sectorsize - 1); | |
2980 | em = get_extent(inode, NULL, 0, offset, len, 0); | |
2981 | if (!em || IS_ERR(em)) | |
2982 | return em; | |
2983 | ||
2984 | /* if this isn't a hole return it */ | |
2985 | if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) && | |
2986 | em->block_start != EXTENT_MAP_HOLE) { | |
2987 | return em; | |
2988 | } | |
2989 | ||
2990 | /* this is a hole, advance to the next extent */ | |
2991 | offset = extent_map_end(em); | |
2992 | free_extent_map(em); | |
2993 | if (offset >= last) | |
2994 | break; | |
2995 | } | |
2996 | return NULL; | |
2997 | } | |
2998 | ||
1506fcc8 YS |
2999 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
3000 | __u64 start, __u64 len, get_extent_t *get_extent) | |
3001 | { | |
975f84fe | 3002 | int ret = 0; |
1506fcc8 YS |
3003 | u64 off = start; |
3004 | u64 max = start + len; | |
3005 | u32 flags = 0; | |
975f84fe JB |
3006 | u32 found_type; |
3007 | u64 last; | |
ec29ed5b | 3008 | u64 last_for_get_extent = 0; |
1506fcc8 | 3009 | u64 disko = 0; |
ec29ed5b | 3010 | u64 isize = i_size_read(inode); |
975f84fe | 3011 | struct btrfs_key found_key; |
1506fcc8 | 3012 | struct extent_map *em = NULL; |
2ac55d41 | 3013 | struct extent_state *cached_state = NULL; |
975f84fe JB |
3014 | struct btrfs_path *path; |
3015 | struct btrfs_file_extent_item *item; | |
1506fcc8 | 3016 | int end = 0; |
ec29ed5b CM |
3017 | u64 em_start = 0; |
3018 | u64 em_len = 0; | |
3019 | u64 em_end = 0; | |
1506fcc8 | 3020 | unsigned long emflags; |
1506fcc8 YS |
3021 | |
3022 | if (len == 0) | |
3023 | return -EINVAL; | |
3024 | ||
975f84fe JB |
3025 | path = btrfs_alloc_path(); |
3026 | if (!path) | |
3027 | return -ENOMEM; | |
3028 | path->leave_spinning = 1; | |
3029 | ||
ec29ed5b CM |
3030 | /* |
3031 | * lookup the last file extent. We're not using i_size here | |
3032 | * because there might be preallocation past i_size | |
3033 | */ | |
975f84fe JB |
3034 | ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root, |
3035 | path, inode->i_ino, -1, 0); | |
3036 | if (ret < 0) { | |
3037 | btrfs_free_path(path); | |
3038 | return ret; | |
3039 | } | |
3040 | WARN_ON(!ret); | |
3041 | path->slots[0]--; | |
3042 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3043 | struct btrfs_file_extent_item); | |
3044 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
3045 | found_type = btrfs_key_type(&found_key); | |
3046 | ||
ec29ed5b | 3047 | /* No extents, but there might be delalloc bits */ |
975f84fe JB |
3048 | if (found_key.objectid != inode->i_ino || |
3049 | found_type != BTRFS_EXTENT_DATA_KEY) { | |
ec29ed5b CM |
3050 | /* have to trust i_size as the end */ |
3051 | last = (u64)-1; | |
3052 | last_for_get_extent = isize; | |
3053 | } else { | |
3054 | /* | |
3055 | * remember the start of the last extent. There are a | |
3056 | * bunch of different factors that go into the length of the | |
3057 | * extent, so its much less complex to remember where it started | |
3058 | */ | |
3059 | last = found_key.offset; | |
3060 | last_for_get_extent = last + 1; | |
975f84fe | 3061 | } |
975f84fe JB |
3062 | btrfs_free_path(path); |
3063 | ||
ec29ed5b CM |
3064 | /* |
3065 | * we might have some extents allocated but more delalloc past those | |
3066 | * extents. so, we trust isize unless the start of the last extent is | |
3067 | * beyond isize | |
3068 | */ | |
3069 | if (last < isize) { | |
3070 | last = (u64)-1; | |
3071 | last_for_get_extent = isize; | |
3072 | } | |
3073 | ||
2ac55d41 JB |
3074 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0, |
3075 | &cached_state, GFP_NOFS); | |
ec29ed5b CM |
3076 | |
3077 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
3078 | get_extent); | |
1506fcc8 YS |
3079 | if (!em) |
3080 | goto out; | |
3081 | if (IS_ERR(em)) { | |
3082 | ret = PTR_ERR(em); | |
3083 | goto out; | |
3084 | } | |
975f84fe | 3085 | |
1506fcc8 | 3086 | while (!end) { |
ea8efc74 CM |
3087 | u64 offset_in_extent; |
3088 | ||
3089 | /* break if the extent we found is outside the range */ | |
3090 | if (em->start >= max || extent_map_end(em) < off) | |
3091 | break; | |
3092 | ||
3093 | /* | |
3094 | * get_extent may return an extent that starts before our | |
3095 | * requested range. We have to make sure the ranges | |
3096 | * we return to fiemap always move forward and don't | |
3097 | * overlap, so adjust the offsets here | |
3098 | */ | |
3099 | em_start = max(em->start, off); | |
1506fcc8 | 3100 | |
ea8efc74 CM |
3101 | /* |
3102 | * record the offset from the start of the extent | |
3103 | * for adjusting the disk offset below | |
3104 | */ | |
3105 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 3106 | em_end = extent_map_end(em); |
ea8efc74 | 3107 | em_len = em_end - em_start; |
ec29ed5b | 3108 | emflags = em->flags; |
1506fcc8 YS |
3109 | disko = 0; |
3110 | flags = 0; | |
3111 | ||
ea8efc74 CM |
3112 | /* |
3113 | * bump off for our next call to get_extent | |
3114 | */ | |
3115 | off = extent_map_end(em); | |
3116 | if (off >= max) | |
3117 | end = 1; | |
3118 | ||
93dbfad7 | 3119 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
3120 | end = 1; |
3121 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 3122 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
3123 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
3124 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 3125 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
3126 | flags |= (FIEMAP_EXTENT_DELALLOC | |
3127 | FIEMAP_EXTENT_UNKNOWN); | |
93dbfad7 | 3128 | } else { |
ea8efc74 | 3129 | disko = em->block_start + offset_in_extent; |
1506fcc8 YS |
3130 | } |
3131 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
3132 | flags |= FIEMAP_EXTENT_ENCODED; | |
3133 | ||
1506fcc8 YS |
3134 | free_extent_map(em); |
3135 | em = NULL; | |
ec29ed5b CM |
3136 | if ((em_start >= last) || em_len == (u64)-1 || |
3137 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
3138 | flags |= FIEMAP_EXTENT_LAST; |
3139 | end = 1; | |
3140 | } | |
3141 | ||
ec29ed5b CM |
3142 | /* now scan forward to see if this is really the last extent. */ |
3143 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
3144 | get_extent); | |
3145 | if (IS_ERR(em)) { | |
3146 | ret = PTR_ERR(em); | |
3147 | goto out; | |
3148 | } | |
3149 | if (!em) { | |
975f84fe JB |
3150 | flags |= FIEMAP_EXTENT_LAST; |
3151 | end = 1; | |
3152 | } | |
ec29ed5b CM |
3153 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, |
3154 | em_len, flags); | |
3155 | if (ret) | |
3156 | goto out_free; | |
1506fcc8 YS |
3157 | } |
3158 | out_free: | |
3159 | free_extent_map(em); | |
3160 | out: | |
2ac55d41 JB |
3161 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len, |
3162 | &cached_state, GFP_NOFS); | |
1506fcc8 YS |
3163 | return ret; |
3164 | } | |
3165 | ||
d1310b2e CM |
3166 | static inline struct page *extent_buffer_page(struct extent_buffer *eb, |
3167 | unsigned long i) | |
3168 | { | |
3169 | struct page *p; | |
3170 | struct address_space *mapping; | |
3171 | ||
3172 | if (i == 0) | |
3173 | return eb->first_page; | |
3174 | i += eb->start >> PAGE_CACHE_SHIFT; | |
3175 | mapping = eb->first_page->mapping; | |
33958dc6 CM |
3176 | if (!mapping) |
3177 | return NULL; | |
0ee0fda0 SW |
3178 | |
3179 | /* | |
3180 | * extent_buffer_page is only called after pinning the page | |
3181 | * by increasing the reference count. So we know the page must | |
3182 | * be in the radix tree. | |
3183 | */ | |
0ee0fda0 | 3184 | rcu_read_lock(); |
d1310b2e | 3185 | p = radix_tree_lookup(&mapping->page_tree, i); |
0ee0fda0 | 3186 | rcu_read_unlock(); |
2b1f55b0 | 3187 | |
d1310b2e CM |
3188 | return p; |
3189 | } | |
3190 | ||
6af118ce | 3191 | static inline unsigned long num_extent_pages(u64 start, u64 len) |
728131d8 | 3192 | { |
6af118ce CM |
3193 | return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) - |
3194 | (start >> PAGE_CACHE_SHIFT); | |
728131d8 CM |
3195 | } |
3196 | ||
d1310b2e CM |
3197 | static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree, |
3198 | u64 start, | |
3199 | unsigned long len, | |
3200 | gfp_t mask) | |
3201 | { | |
3202 | struct extent_buffer *eb = NULL; | |
3935127c | 3203 | #if LEAK_DEBUG |
2d2ae547 | 3204 | unsigned long flags; |
4bef0848 | 3205 | #endif |
d1310b2e | 3206 | |
d1310b2e | 3207 | eb = kmem_cache_zalloc(extent_buffer_cache, mask); |
91ca338d TI |
3208 | if (eb == NULL) |
3209 | return NULL; | |
d1310b2e CM |
3210 | eb->start = start; |
3211 | eb->len = len; | |
b4ce94de CM |
3212 | spin_lock_init(&eb->lock); |
3213 | init_waitqueue_head(&eb->lock_wq); | |
3214 | ||
3935127c | 3215 | #if LEAK_DEBUG |
2d2ae547 CM |
3216 | spin_lock_irqsave(&leak_lock, flags); |
3217 | list_add(&eb->leak_list, &buffers); | |
3218 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 3219 | #endif |
d1310b2e CM |
3220 | atomic_set(&eb->refs, 1); |
3221 | ||
3222 | return eb; | |
3223 | } | |
3224 | ||
3225 | static void __free_extent_buffer(struct extent_buffer *eb) | |
3226 | { | |
3935127c | 3227 | #if LEAK_DEBUG |
2d2ae547 CM |
3228 | unsigned long flags; |
3229 | spin_lock_irqsave(&leak_lock, flags); | |
3230 | list_del(&eb->leak_list); | |
3231 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 3232 | #endif |
d1310b2e CM |
3233 | kmem_cache_free(extent_buffer_cache, eb); |
3234 | } | |
3235 | ||
897ca6e9 MX |
3236 | /* |
3237 | * Helper for releasing extent buffer page. | |
3238 | */ | |
3239 | static void btrfs_release_extent_buffer_page(struct extent_buffer *eb, | |
3240 | unsigned long start_idx) | |
3241 | { | |
3242 | unsigned long index; | |
3243 | struct page *page; | |
3244 | ||
3245 | if (!eb->first_page) | |
3246 | return; | |
3247 | ||
3248 | index = num_extent_pages(eb->start, eb->len); | |
3249 | if (start_idx >= index) | |
3250 | return; | |
3251 | ||
3252 | do { | |
3253 | index--; | |
3254 | page = extent_buffer_page(eb, index); | |
3255 | if (page) | |
3256 | page_cache_release(page); | |
3257 | } while (index != start_idx); | |
3258 | } | |
3259 | ||
3260 | /* | |
3261 | * Helper for releasing the extent buffer. | |
3262 | */ | |
3263 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
3264 | { | |
3265 | btrfs_release_extent_buffer_page(eb, 0); | |
3266 | __free_extent_buffer(eb); | |
3267 | } | |
3268 | ||
d1310b2e CM |
3269 | struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree, |
3270 | u64 start, unsigned long len, | |
3271 | struct page *page0, | |
3272 | gfp_t mask) | |
3273 | { | |
3274 | unsigned long num_pages = num_extent_pages(start, len); | |
3275 | unsigned long i; | |
3276 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
3277 | struct extent_buffer *eb; | |
6af118ce | 3278 | struct extent_buffer *exists = NULL; |
d1310b2e CM |
3279 | struct page *p; |
3280 | struct address_space *mapping = tree->mapping; | |
3281 | int uptodate = 1; | |
19fe0a8b | 3282 | int ret; |
d1310b2e | 3283 | |
19fe0a8b MX |
3284 | rcu_read_lock(); |
3285 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); | |
3286 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
3287 | rcu_read_unlock(); | |
0f9dd46c | 3288 | mark_page_accessed(eb->first_page); |
6af118ce CM |
3289 | return eb; |
3290 | } | |
19fe0a8b | 3291 | rcu_read_unlock(); |
6af118ce | 3292 | |
d1310b2e | 3293 | eb = __alloc_extent_buffer(tree, start, len, mask); |
2b114d1d | 3294 | if (!eb) |
d1310b2e CM |
3295 | return NULL; |
3296 | ||
d1310b2e CM |
3297 | if (page0) { |
3298 | eb->first_page = page0; | |
3299 | i = 1; | |
3300 | index++; | |
3301 | page_cache_get(page0); | |
3302 | mark_page_accessed(page0); | |
3303 | set_page_extent_mapped(page0); | |
d1310b2e | 3304 | set_page_extent_head(page0, len); |
f188591e | 3305 | uptodate = PageUptodate(page0); |
d1310b2e CM |
3306 | } else { |
3307 | i = 0; | |
3308 | } | |
3309 | for (; i < num_pages; i++, index++) { | |
3310 | p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM); | |
3311 | if (!p) { | |
3312 | WARN_ON(1); | |
6af118ce | 3313 | goto free_eb; |
d1310b2e CM |
3314 | } |
3315 | set_page_extent_mapped(p); | |
3316 | mark_page_accessed(p); | |
3317 | if (i == 0) { | |
3318 | eb->first_page = p; | |
3319 | set_page_extent_head(p, len); | |
3320 | } else { | |
3321 | set_page_private(p, EXTENT_PAGE_PRIVATE); | |
3322 | } | |
3323 | if (!PageUptodate(p)) | |
3324 | uptodate = 0; | |
eb14ab8e CM |
3325 | |
3326 | /* | |
3327 | * see below about how we avoid a nasty race with release page | |
3328 | * and why we unlock later | |
3329 | */ | |
3330 | if (i != 0) | |
3331 | unlock_page(p); | |
d1310b2e CM |
3332 | } |
3333 | if (uptodate) | |
b4ce94de | 3334 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 3335 | |
19fe0a8b MX |
3336 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); |
3337 | if (ret) | |
3338 | goto free_eb; | |
3339 | ||
6af118ce | 3340 | spin_lock(&tree->buffer_lock); |
19fe0a8b MX |
3341 | ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb); |
3342 | if (ret == -EEXIST) { | |
3343 | exists = radix_tree_lookup(&tree->buffer, | |
3344 | start >> PAGE_CACHE_SHIFT); | |
6af118ce CM |
3345 | /* add one reference for the caller */ |
3346 | atomic_inc(&exists->refs); | |
3347 | spin_unlock(&tree->buffer_lock); | |
19fe0a8b | 3348 | radix_tree_preload_end(); |
6af118ce CM |
3349 | goto free_eb; |
3350 | } | |
6af118ce CM |
3351 | /* add one reference for the tree */ |
3352 | atomic_inc(&eb->refs); | |
f044ba78 | 3353 | spin_unlock(&tree->buffer_lock); |
19fe0a8b | 3354 | radix_tree_preload_end(); |
eb14ab8e CM |
3355 | |
3356 | /* | |
3357 | * there is a race where release page may have | |
3358 | * tried to find this extent buffer in the radix | |
3359 | * but failed. It will tell the VM it is safe to | |
3360 | * reclaim the, and it will clear the page private bit. | |
3361 | * We must make sure to set the page private bit properly | |
3362 | * after the extent buffer is in the radix tree so | |
3363 | * it doesn't get lost | |
3364 | */ | |
3365 | set_page_extent_mapped(eb->first_page); | |
3366 | set_page_extent_head(eb->first_page, eb->len); | |
3367 | if (!page0) | |
3368 | unlock_page(eb->first_page); | |
d1310b2e CM |
3369 | return eb; |
3370 | ||
6af118ce | 3371 | free_eb: |
eb14ab8e CM |
3372 | if (eb->first_page && !page0) |
3373 | unlock_page(eb->first_page); | |
3374 | ||
d1310b2e | 3375 | if (!atomic_dec_and_test(&eb->refs)) |
6af118ce | 3376 | return exists; |
897ca6e9 | 3377 | btrfs_release_extent_buffer(eb); |
6af118ce | 3378 | return exists; |
d1310b2e | 3379 | } |
d1310b2e CM |
3380 | |
3381 | struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree, | |
3382 | u64 start, unsigned long len, | |
3383 | gfp_t mask) | |
3384 | { | |
d1310b2e | 3385 | struct extent_buffer *eb; |
d1310b2e | 3386 | |
19fe0a8b MX |
3387 | rcu_read_lock(); |
3388 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); | |
3389 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
3390 | rcu_read_unlock(); | |
0f9dd46c | 3391 | mark_page_accessed(eb->first_page); |
19fe0a8b MX |
3392 | return eb; |
3393 | } | |
3394 | rcu_read_unlock(); | |
0f9dd46c | 3395 | |
19fe0a8b | 3396 | return NULL; |
d1310b2e | 3397 | } |
d1310b2e CM |
3398 | |
3399 | void free_extent_buffer(struct extent_buffer *eb) | |
3400 | { | |
d1310b2e CM |
3401 | if (!eb) |
3402 | return; | |
3403 | ||
3404 | if (!atomic_dec_and_test(&eb->refs)) | |
3405 | return; | |
3406 | ||
6af118ce | 3407 | WARN_ON(1); |
d1310b2e | 3408 | } |
d1310b2e CM |
3409 | |
3410 | int clear_extent_buffer_dirty(struct extent_io_tree *tree, | |
3411 | struct extent_buffer *eb) | |
3412 | { | |
d1310b2e CM |
3413 | unsigned long i; |
3414 | unsigned long num_pages; | |
3415 | struct page *page; | |
3416 | ||
d1310b2e CM |
3417 | num_pages = num_extent_pages(eb->start, eb->len); |
3418 | ||
3419 | for (i = 0; i < num_pages; i++) { | |
3420 | page = extent_buffer_page(eb, i); | |
b9473439 | 3421 | if (!PageDirty(page)) |
d2c3f4f6 CM |
3422 | continue; |
3423 | ||
a61e6f29 | 3424 | lock_page(page); |
eb14ab8e CM |
3425 | WARN_ON(!PagePrivate(page)); |
3426 | ||
3427 | set_page_extent_mapped(page); | |
d1310b2e CM |
3428 | if (i == 0) |
3429 | set_page_extent_head(page, eb->len); | |
d1310b2e | 3430 | |
d1310b2e | 3431 | clear_page_dirty_for_io(page); |
0ee0fda0 | 3432 | spin_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
3433 | if (!PageDirty(page)) { |
3434 | radix_tree_tag_clear(&page->mapping->page_tree, | |
3435 | page_index(page), | |
3436 | PAGECACHE_TAG_DIRTY); | |
3437 | } | |
0ee0fda0 | 3438 | spin_unlock_irq(&page->mapping->tree_lock); |
a61e6f29 | 3439 | unlock_page(page); |
d1310b2e CM |
3440 | } |
3441 | return 0; | |
3442 | } | |
d1310b2e CM |
3443 | |
3444 | int wait_on_extent_buffer_writeback(struct extent_io_tree *tree, | |
3445 | struct extent_buffer *eb) | |
3446 | { | |
3447 | return wait_on_extent_writeback(tree, eb->start, | |
3448 | eb->start + eb->len - 1); | |
3449 | } | |
d1310b2e CM |
3450 | |
3451 | int set_extent_buffer_dirty(struct extent_io_tree *tree, | |
3452 | struct extent_buffer *eb) | |
3453 | { | |
3454 | unsigned long i; | |
3455 | unsigned long num_pages; | |
b9473439 | 3456 | int was_dirty = 0; |
d1310b2e | 3457 | |
b9473439 | 3458 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
d1310b2e | 3459 | num_pages = num_extent_pages(eb->start, eb->len); |
b9473439 | 3460 | for (i = 0; i < num_pages; i++) |
d1310b2e | 3461 | __set_page_dirty_nobuffers(extent_buffer_page(eb, i)); |
b9473439 | 3462 | return was_dirty; |
d1310b2e | 3463 | } |
d1310b2e | 3464 | |
1259ab75 | 3465 | int clear_extent_buffer_uptodate(struct extent_io_tree *tree, |
2ac55d41 JB |
3466 | struct extent_buffer *eb, |
3467 | struct extent_state **cached_state) | |
1259ab75 CM |
3468 | { |
3469 | unsigned long i; | |
3470 | struct page *page; | |
3471 | unsigned long num_pages; | |
3472 | ||
3473 | num_pages = num_extent_pages(eb->start, eb->len); | |
b4ce94de | 3474 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
1259ab75 CM |
3475 | |
3476 | clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1, | |
2ac55d41 | 3477 | cached_state, GFP_NOFS); |
1259ab75 CM |
3478 | for (i = 0; i < num_pages; i++) { |
3479 | page = extent_buffer_page(eb, i); | |
33958dc6 CM |
3480 | if (page) |
3481 | ClearPageUptodate(page); | |
1259ab75 CM |
3482 | } |
3483 | return 0; | |
3484 | } | |
3485 | ||
d1310b2e CM |
3486 | int set_extent_buffer_uptodate(struct extent_io_tree *tree, |
3487 | struct extent_buffer *eb) | |
3488 | { | |
3489 | unsigned long i; | |
3490 | struct page *page; | |
3491 | unsigned long num_pages; | |
3492 | ||
3493 | num_pages = num_extent_pages(eb->start, eb->len); | |
3494 | ||
3495 | set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1, | |
507903b8 | 3496 | NULL, GFP_NOFS); |
d1310b2e CM |
3497 | for (i = 0; i < num_pages; i++) { |
3498 | page = extent_buffer_page(eb, i); | |
3499 | if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) || | |
3500 | ((i == num_pages - 1) && | |
3501 | ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) { | |
3502 | check_page_uptodate(tree, page); | |
3503 | continue; | |
3504 | } | |
3505 | SetPageUptodate(page); | |
3506 | } | |
3507 | return 0; | |
3508 | } | |
d1310b2e | 3509 | |
ce9adaa5 CM |
3510 | int extent_range_uptodate(struct extent_io_tree *tree, |
3511 | u64 start, u64 end) | |
3512 | { | |
3513 | struct page *page; | |
3514 | int ret; | |
3515 | int pg_uptodate = 1; | |
3516 | int uptodate; | |
3517 | unsigned long index; | |
3518 | ||
9655d298 | 3519 | ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL); |
ce9adaa5 CM |
3520 | if (ret) |
3521 | return 1; | |
d397712b | 3522 | while (start <= end) { |
ce9adaa5 CM |
3523 | index = start >> PAGE_CACHE_SHIFT; |
3524 | page = find_get_page(tree->mapping, index); | |
3525 | uptodate = PageUptodate(page); | |
3526 | page_cache_release(page); | |
3527 | if (!uptodate) { | |
3528 | pg_uptodate = 0; | |
3529 | break; | |
3530 | } | |
3531 | start += PAGE_CACHE_SIZE; | |
3532 | } | |
3533 | return pg_uptodate; | |
3534 | } | |
3535 | ||
d1310b2e | 3536 | int extent_buffer_uptodate(struct extent_io_tree *tree, |
2ac55d41 JB |
3537 | struct extent_buffer *eb, |
3538 | struct extent_state *cached_state) | |
d1310b2e | 3539 | { |
728131d8 | 3540 | int ret = 0; |
ce9adaa5 CM |
3541 | unsigned long num_pages; |
3542 | unsigned long i; | |
728131d8 CM |
3543 | struct page *page; |
3544 | int pg_uptodate = 1; | |
3545 | ||
b4ce94de | 3546 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
4235298e | 3547 | return 1; |
728131d8 | 3548 | |
4235298e | 3549 | ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1, |
2ac55d41 | 3550 | EXTENT_UPTODATE, 1, cached_state); |
4235298e CM |
3551 | if (ret) |
3552 | return ret; | |
728131d8 CM |
3553 | |
3554 | num_pages = num_extent_pages(eb->start, eb->len); | |
3555 | for (i = 0; i < num_pages; i++) { | |
3556 | page = extent_buffer_page(eb, i); | |
3557 | if (!PageUptodate(page)) { | |
3558 | pg_uptodate = 0; | |
3559 | break; | |
3560 | } | |
3561 | } | |
4235298e | 3562 | return pg_uptodate; |
d1310b2e | 3563 | } |
d1310b2e CM |
3564 | |
3565 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
3566 | struct extent_buffer *eb, | |
a86c12c7 | 3567 | u64 start, int wait, |
f188591e | 3568 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3569 | { |
3570 | unsigned long i; | |
3571 | unsigned long start_i; | |
3572 | struct page *page; | |
3573 | int err; | |
3574 | int ret = 0; | |
ce9adaa5 CM |
3575 | int locked_pages = 0; |
3576 | int all_uptodate = 1; | |
3577 | int inc_all_pages = 0; | |
d1310b2e | 3578 | unsigned long num_pages; |
a86c12c7 | 3579 | struct bio *bio = NULL; |
c8b97818 | 3580 | unsigned long bio_flags = 0; |
a86c12c7 | 3581 | |
b4ce94de | 3582 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
3583 | return 0; |
3584 | ||
ce9adaa5 | 3585 | if (test_range_bit(tree, eb->start, eb->start + eb->len - 1, |
9655d298 | 3586 | EXTENT_UPTODATE, 1, NULL)) { |
d1310b2e CM |
3587 | return 0; |
3588 | } | |
3589 | ||
3590 | if (start) { | |
3591 | WARN_ON(start < eb->start); | |
3592 | start_i = (start >> PAGE_CACHE_SHIFT) - | |
3593 | (eb->start >> PAGE_CACHE_SHIFT); | |
3594 | } else { | |
3595 | start_i = 0; | |
3596 | } | |
3597 | ||
3598 | num_pages = num_extent_pages(eb->start, eb->len); | |
3599 | for (i = start_i; i < num_pages; i++) { | |
3600 | page = extent_buffer_page(eb, i); | |
d1310b2e | 3601 | if (!wait) { |
2db04966 | 3602 | if (!trylock_page(page)) |
ce9adaa5 | 3603 | goto unlock_exit; |
d1310b2e CM |
3604 | } else { |
3605 | lock_page(page); | |
3606 | } | |
ce9adaa5 | 3607 | locked_pages++; |
d397712b | 3608 | if (!PageUptodate(page)) |
ce9adaa5 | 3609 | all_uptodate = 0; |
ce9adaa5 CM |
3610 | } |
3611 | if (all_uptodate) { | |
3612 | if (start_i == 0) | |
b4ce94de | 3613 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
3614 | goto unlock_exit; |
3615 | } | |
3616 | ||
3617 | for (i = start_i; i < num_pages; i++) { | |
3618 | page = extent_buffer_page(eb, i); | |
eb14ab8e CM |
3619 | |
3620 | WARN_ON(!PagePrivate(page)); | |
3621 | ||
3622 | set_page_extent_mapped(page); | |
3623 | if (i == 0) | |
3624 | set_page_extent_head(page, eb->len); | |
3625 | ||
ce9adaa5 CM |
3626 | if (inc_all_pages) |
3627 | page_cache_get(page); | |
3628 | if (!PageUptodate(page)) { | |
3629 | if (start_i == 0) | |
3630 | inc_all_pages = 1; | |
f188591e | 3631 | ClearPageError(page); |
a86c12c7 | 3632 | err = __extent_read_full_page(tree, page, |
f188591e | 3633 | get_extent, &bio, |
c8b97818 | 3634 | mirror_num, &bio_flags); |
d397712b | 3635 | if (err) |
d1310b2e | 3636 | ret = err; |
d1310b2e CM |
3637 | } else { |
3638 | unlock_page(page); | |
3639 | } | |
3640 | } | |
3641 | ||
a86c12c7 | 3642 | if (bio) |
c8b97818 | 3643 | submit_one_bio(READ, bio, mirror_num, bio_flags); |
a86c12c7 | 3644 | |
d397712b | 3645 | if (ret || !wait) |
d1310b2e | 3646 | return ret; |
d397712b | 3647 | |
d1310b2e CM |
3648 | for (i = start_i; i < num_pages; i++) { |
3649 | page = extent_buffer_page(eb, i); | |
3650 | wait_on_page_locked(page); | |
d397712b | 3651 | if (!PageUptodate(page)) |
d1310b2e | 3652 | ret = -EIO; |
d1310b2e | 3653 | } |
d397712b | 3654 | |
d1310b2e | 3655 | if (!ret) |
b4ce94de | 3656 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 3657 | return ret; |
ce9adaa5 CM |
3658 | |
3659 | unlock_exit: | |
3660 | i = start_i; | |
d397712b | 3661 | while (locked_pages > 0) { |
ce9adaa5 CM |
3662 | page = extent_buffer_page(eb, i); |
3663 | i++; | |
3664 | unlock_page(page); | |
3665 | locked_pages--; | |
3666 | } | |
3667 | return ret; | |
d1310b2e | 3668 | } |
d1310b2e CM |
3669 | |
3670 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
3671 | unsigned long start, | |
3672 | unsigned long len) | |
3673 | { | |
3674 | size_t cur; | |
3675 | size_t offset; | |
3676 | struct page *page; | |
3677 | char *kaddr; | |
3678 | char *dst = (char *)dstv; | |
3679 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3680 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
3681 | |
3682 | WARN_ON(start > eb->len); | |
3683 | WARN_ON(start + len > eb->start + eb->len); | |
3684 | ||
3685 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3686 | ||
d397712b | 3687 | while (len > 0) { |
d1310b2e | 3688 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
3689 | |
3690 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
3691 | kaddr = kmap_atomic(page, KM_USER1); | |
3692 | memcpy(dst, kaddr + offset, cur); | |
3693 | kunmap_atomic(kaddr, KM_USER1); | |
3694 | ||
3695 | dst += cur; | |
3696 | len -= cur; | |
3697 | offset = 0; | |
3698 | i++; | |
3699 | } | |
3700 | } | |
d1310b2e CM |
3701 | |
3702 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
3703 | unsigned long min_len, char **token, char **map, | |
3704 | unsigned long *map_start, | |
3705 | unsigned long *map_len, int km) | |
3706 | { | |
3707 | size_t offset = start & (PAGE_CACHE_SIZE - 1); | |
3708 | char *kaddr; | |
3709 | struct page *p; | |
3710 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3711 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3712 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
3713 | PAGE_CACHE_SHIFT; | |
3714 | ||
3715 | if (i != end_i) | |
3716 | return -EINVAL; | |
3717 | ||
3718 | if (i == 0) { | |
3719 | offset = start_offset; | |
3720 | *map_start = 0; | |
3721 | } else { | |
3722 | offset = 0; | |
3723 | *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; | |
3724 | } | |
d397712b | 3725 | |
d1310b2e | 3726 | if (start + min_len > eb->len) { |
d397712b CM |
3727 | printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, " |
3728 | "wanted %lu %lu\n", (unsigned long long)eb->start, | |
3729 | eb->len, start, min_len); | |
d1310b2e | 3730 | WARN_ON(1); |
85026533 | 3731 | return -EINVAL; |
d1310b2e CM |
3732 | } |
3733 | ||
3734 | p = extent_buffer_page(eb, i); | |
d1310b2e CM |
3735 | kaddr = kmap_atomic(p, km); |
3736 | *token = kaddr; | |
3737 | *map = kaddr + offset; | |
3738 | *map_len = PAGE_CACHE_SIZE - offset; | |
3739 | return 0; | |
3740 | } | |
d1310b2e CM |
3741 | |
3742 | int map_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
3743 | unsigned long min_len, | |
3744 | char **token, char **map, | |
3745 | unsigned long *map_start, | |
3746 | unsigned long *map_len, int km) | |
3747 | { | |
3748 | int err; | |
3749 | int save = 0; | |
3750 | if (eb->map_token) { | |
3751 | unmap_extent_buffer(eb, eb->map_token, km); | |
3752 | eb->map_token = NULL; | |
3753 | save = 1; | |
3754 | } | |
3755 | err = map_private_extent_buffer(eb, start, min_len, token, map, | |
3756 | map_start, map_len, km); | |
3757 | if (!err && save) { | |
3758 | eb->map_token = *token; | |
3759 | eb->kaddr = *map; | |
3760 | eb->map_start = *map_start; | |
3761 | eb->map_len = *map_len; | |
3762 | } | |
3763 | return err; | |
3764 | } | |
d1310b2e CM |
3765 | |
3766 | void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km) | |
3767 | { | |
3768 | kunmap_atomic(token, km); | |
3769 | } | |
d1310b2e CM |
3770 | |
3771 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, | |
3772 | unsigned long start, | |
3773 | unsigned long len) | |
3774 | { | |
3775 | size_t cur; | |
3776 | size_t offset; | |
3777 | struct page *page; | |
3778 | char *kaddr; | |
3779 | char *ptr = (char *)ptrv; | |
3780 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3781 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3782 | int ret = 0; | |
3783 | ||
3784 | WARN_ON(start > eb->len); | |
3785 | WARN_ON(start + len > eb->start + eb->len); | |
3786 | ||
3787 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3788 | ||
d397712b | 3789 | while (len > 0) { |
d1310b2e | 3790 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
3791 | |
3792 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
3793 | ||
3794 | kaddr = kmap_atomic(page, KM_USER0); | |
3795 | ret = memcmp(ptr, kaddr + offset, cur); | |
3796 | kunmap_atomic(kaddr, KM_USER0); | |
3797 | if (ret) | |
3798 | break; | |
3799 | ||
3800 | ptr += cur; | |
3801 | len -= cur; | |
3802 | offset = 0; | |
3803 | i++; | |
3804 | } | |
3805 | return ret; | |
3806 | } | |
d1310b2e CM |
3807 | |
3808 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
3809 | unsigned long start, unsigned long len) | |
3810 | { | |
3811 | size_t cur; | |
3812 | size_t offset; | |
3813 | struct page *page; | |
3814 | char *kaddr; | |
3815 | char *src = (char *)srcv; | |
3816 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3817 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3818 | ||
3819 | WARN_ON(start > eb->len); | |
3820 | WARN_ON(start + len > eb->start + eb->len); | |
3821 | ||
3822 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3823 | ||
d397712b | 3824 | while (len > 0) { |
d1310b2e CM |
3825 | page = extent_buffer_page(eb, i); |
3826 | WARN_ON(!PageUptodate(page)); | |
3827 | ||
3828 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
3829 | kaddr = kmap_atomic(page, KM_USER1); | |
3830 | memcpy(kaddr + offset, src, cur); | |
3831 | kunmap_atomic(kaddr, KM_USER1); | |
3832 | ||
3833 | src += cur; | |
3834 | len -= cur; | |
3835 | offset = 0; | |
3836 | i++; | |
3837 | } | |
3838 | } | |
d1310b2e CM |
3839 | |
3840 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
3841 | unsigned long start, unsigned long len) | |
3842 | { | |
3843 | size_t cur; | |
3844 | size_t offset; | |
3845 | struct page *page; | |
3846 | char *kaddr; | |
3847 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3848 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3849 | ||
3850 | WARN_ON(start > eb->len); | |
3851 | WARN_ON(start + len > eb->start + eb->len); | |
3852 | ||
3853 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3854 | ||
d397712b | 3855 | while (len > 0) { |
d1310b2e CM |
3856 | page = extent_buffer_page(eb, i); |
3857 | WARN_ON(!PageUptodate(page)); | |
3858 | ||
3859 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
3860 | kaddr = kmap_atomic(page, KM_USER0); | |
3861 | memset(kaddr + offset, c, cur); | |
3862 | kunmap_atomic(kaddr, KM_USER0); | |
3863 | ||
3864 | len -= cur; | |
3865 | offset = 0; | |
3866 | i++; | |
3867 | } | |
3868 | } | |
d1310b2e CM |
3869 | |
3870 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
3871 | unsigned long dst_offset, unsigned long src_offset, | |
3872 | unsigned long len) | |
3873 | { | |
3874 | u64 dst_len = dst->len; | |
3875 | size_t cur; | |
3876 | size_t offset; | |
3877 | struct page *page; | |
3878 | char *kaddr; | |
3879 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3880 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
3881 | ||
3882 | WARN_ON(src->len != dst_len); | |
3883 | ||
3884 | offset = (start_offset + dst_offset) & | |
3885 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3886 | ||
d397712b | 3887 | while (len > 0) { |
d1310b2e CM |
3888 | page = extent_buffer_page(dst, i); |
3889 | WARN_ON(!PageUptodate(page)); | |
3890 | ||
3891 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
3892 | ||
3893 | kaddr = kmap_atomic(page, KM_USER0); | |
3894 | read_extent_buffer(src, kaddr + offset, src_offset, cur); | |
3895 | kunmap_atomic(kaddr, KM_USER0); | |
3896 | ||
3897 | src_offset += cur; | |
3898 | len -= cur; | |
3899 | offset = 0; | |
3900 | i++; | |
3901 | } | |
3902 | } | |
d1310b2e CM |
3903 | |
3904 | static void move_pages(struct page *dst_page, struct page *src_page, | |
3905 | unsigned long dst_off, unsigned long src_off, | |
3906 | unsigned long len) | |
3907 | { | |
3908 | char *dst_kaddr = kmap_atomic(dst_page, KM_USER0); | |
3909 | if (dst_page == src_page) { | |
3910 | memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len); | |
3911 | } else { | |
3912 | char *src_kaddr = kmap_atomic(src_page, KM_USER1); | |
3913 | char *p = dst_kaddr + dst_off + len; | |
3914 | char *s = src_kaddr + src_off + len; | |
3915 | ||
3916 | while (len--) | |
3917 | *--p = *--s; | |
3918 | ||
3919 | kunmap_atomic(src_kaddr, KM_USER1); | |
3920 | } | |
3921 | kunmap_atomic(dst_kaddr, KM_USER0); | |
3922 | } | |
3923 | ||
3387206f ST |
3924 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
3925 | { | |
3926 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
3927 | return distance < len; | |
3928 | } | |
3929 | ||
d1310b2e CM |
3930 | static void copy_pages(struct page *dst_page, struct page *src_page, |
3931 | unsigned long dst_off, unsigned long src_off, | |
3932 | unsigned long len) | |
3933 | { | |
3934 | char *dst_kaddr = kmap_atomic(dst_page, KM_USER0); | |
3935 | char *src_kaddr; | |
3936 | ||
3387206f | 3937 | if (dst_page != src_page) { |
d1310b2e | 3938 | src_kaddr = kmap_atomic(src_page, KM_USER1); |
3387206f | 3939 | } else { |
d1310b2e | 3940 | src_kaddr = dst_kaddr; |
3387206f ST |
3941 | BUG_ON(areas_overlap(src_off, dst_off, len)); |
3942 | } | |
d1310b2e CM |
3943 | |
3944 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
3945 | kunmap_atomic(dst_kaddr, KM_USER0); | |
3946 | if (dst_page != src_page) | |
3947 | kunmap_atomic(src_kaddr, KM_USER1); | |
3948 | } | |
3949 | ||
3950 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
3951 | unsigned long src_offset, unsigned long len) | |
3952 | { | |
3953 | size_t cur; | |
3954 | size_t dst_off_in_page; | |
3955 | size_t src_off_in_page; | |
3956 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3957 | unsigned long dst_i; | |
3958 | unsigned long src_i; | |
3959 | ||
3960 | if (src_offset + len > dst->len) { | |
d397712b CM |
3961 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
3962 | "len %lu dst len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
3963 | BUG_ON(1); |
3964 | } | |
3965 | if (dst_offset + len > dst->len) { | |
d397712b CM |
3966 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
3967 | "len %lu dst len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
3968 | BUG_ON(1); |
3969 | } | |
3970 | ||
d397712b | 3971 | while (len > 0) { |
d1310b2e CM |
3972 | dst_off_in_page = (start_offset + dst_offset) & |
3973 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3974 | src_off_in_page = (start_offset + src_offset) & | |
3975 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3976 | ||
3977 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
3978 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
3979 | ||
3980 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
3981 | src_off_in_page)); | |
3982 | cur = min_t(unsigned long, cur, | |
3983 | (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); | |
3984 | ||
3985 | copy_pages(extent_buffer_page(dst, dst_i), | |
3986 | extent_buffer_page(dst, src_i), | |
3987 | dst_off_in_page, src_off_in_page, cur); | |
3988 | ||
3989 | src_offset += cur; | |
3990 | dst_offset += cur; | |
3991 | len -= cur; | |
3992 | } | |
3993 | } | |
d1310b2e CM |
3994 | |
3995 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
3996 | unsigned long src_offset, unsigned long len) | |
3997 | { | |
3998 | size_t cur; | |
3999 | size_t dst_off_in_page; | |
4000 | size_t src_off_in_page; | |
4001 | unsigned long dst_end = dst_offset + len - 1; | |
4002 | unsigned long src_end = src_offset + len - 1; | |
4003 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4004 | unsigned long dst_i; | |
4005 | unsigned long src_i; | |
4006 | ||
4007 | if (src_offset + len > dst->len) { | |
d397712b CM |
4008 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
4009 | "len %lu len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
4010 | BUG_ON(1); |
4011 | } | |
4012 | if (dst_offset + len > dst->len) { | |
d397712b CM |
4013 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
4014 | "len %lu len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
4015 | BUG_ON(1); |
4016 | } | |
3387206f | 4017 | if (!areas_overlap(src_offset, dst_offset, len)) { |
d1310b2e CM |
4018 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
4019 | return; | |
4020 | } | |
d397712b | 4021 | while (len > 0) { |
d1310b2e CM |
4022 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; |
4023 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
4024 | ||
4025 | dst_off_in_page = (start_offset + dst_end) & | |
4026 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4027 | src_off_in_page = (start_offset + src_end) & | |
4028 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4029 | ||
4030 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
4031 | cur = min(cur, dst_off_in_page + 1); | |
4032 | move_pages(extent_buffer_page(dst, dst_i), | |
4033 | extent_buffer_page(dst, src_i), | |
4034 | dst_off_in_page - cur + 1, | |
4035 | src_off_in_page - cur + 1, cur); | |
4036 | ||
4037 | dst_end -= cur; | |
4038 | src_end -= cur; | |
4039 | len -= cur; | |
4040 | } | |
4041 | } | |
6af118ce | 4042 | |
19fe0a8b MX |
4043 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
4044 | { | |
4045 | struct extent_buffer *eb = | |
4046 | container_of(head, struct extent_buffer, rcu_head); | |
4047 | ||
4048 | btrfs_release_extent_buffer(eb); | |
4049 | } | |
4050 | ||
6af118ce CM |
4051 | int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page) |
4052 | { | |
4053 | u64 start = page_offset(page); | |
4054 | struct extent_buffer *eb; | |
4055 | int ret = 1; | |
6af118ce CM |
4056 | |
4057 | spin_lock(&tree->buffer_lock); | |
19fe0a8b | 4058 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); |
45f49bce CM |
4059 | if (!eb) { |
4060 | spin_unlock(&tree->buffer_lock); | |
4061 | return ret; | |
4062 | } | |
6af118ce | 4063 | |
19fe0a8b | 4064 | if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
6af118ce CM |
4065 | ret = 0; |
4066 | goto out; | |
4067 | } | |
19fe0a8b MX |
4068 | |
4069 | /* | |
4070 | * set @eb->refs to 0 if it is already 1, and then release the @eb. | |
4071 | * Or go back. | |
4072 | */ | |
4073 | if (atomic_cmpxchg(&eb->refs, 1, 0) != 1) { | |
b9473439 CM |
4074 | ret = 0; |
4075 | goto out; | |
4076 | } | |
897ca6e9 | 4077 | |
19fe0a8b | 4078 | radix_tree_delete(&tree->buffer, start >> PAGE_CACHE_SHIFT); |
6af118ce CM |
4079 | out: |
4080 | spin_unlock(&tree->buffer_lock); | |
19fe0a8b MX |
4081 | |
4082 | /* at this point we can safely release the extent buffer */ | |
4083 | if (atomic_read(&eb->refs) == 0) | |
4084 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); | |
6af118ce CM |
4085 | return ret; |
4086 | } |