]>
Commit | Line | Data |
---|---|---|
d1310b2e CM |
1 | #include <linux/bitops.h> |
2 | #include <linux/slab.h> | |
3 | #include <linux/bio.h> | |
4 | #include <linux/mm.h> | |
d1310b2e CM |
5 | #include <linux/pagemap.h> |
6 | #include <linux/page-flags.h> | |
7 | #include <linux/module.h> | |
8 | #include <linux/spinlock.h> | |
9 | #include <linux/blkdev.h> | |
10 | #include <linux/swap.h> | |
d1310b2e CM |
11 | #include <linux/writeback.h> |
12 | #include <linux/pagevec.h> | |
268bb0ce | 13 | #include <linux/prefetch.h> |
90a887c9 | 14 | #include <linux/cleancache.h> |
d1310b2e CM |
15 | #include "extent_io.h" |
16 | #include "extent_map.h" | |
2db04966 | 17 | #include "compat.h" |
902b22f3 DW |
18 | #include "ctree.h" |
19 | #include "btrfs_inode.h" | |
d1310b2e | 20 | |
d1310b2e CM |
21 | static struct kmem_cache *extent_state_cache; |
22 | static struct kmem_cache *extent_buffer_cache; | |
23 | ||
24 | static LIST_HEAD(buffers); | |
25 | static LIST_HEAD(states); | |
4bef0848 | 26 | |
b47eda86 | 27 | #define LEAK_DEBUG 0 |
3935127c | 28 | #if LEAK_DEBUG |
d397712b | 29 | static DEFINE_SPINLOCK(leak_lock); |
4bef0848 | 30 | #endif |
d1310b2e | 31 | |
d1310b2e CM |
32 | #define BUFFER_LRU_MAX 64 |
33 | ||
34 | struct tree_entry { | |
35 | u64 start; | |
36 | u64 end; | |
d1310b2e CM |
37 | struct rb_node rb_node; |
38 | }; | |
39 | ||
40 | struct extent_page_data { | |
41 | struct bio *bio; | |
42 | struct extent_io_tree *tree; | |
43 | get_extent_t *get_extent; | |
771ed689 CM |
44 | |
45 | /* tells writepage not to lock the state bits for this range | |
46 | * it still does the unlocking | |
47 | */ | |
ffbd517d CM |
48 | unsigned int extent_locked:1; |
49 | ||
50 | /* tells the submit_bio code to use a WRITE_SYNC */ | |
51 | unsigned int sync_io:1; | |
d1310b2e CM |
52 | }; |
53 | ||
54 | int __init extent_io_init(void) | |
55 | { | |
9601e3f6 CH |
56 | extent_state_cache = kmem_cache_create("extent_state", |
57 | sizeof(struct extent_state), 0, | |
58 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
59 | if (!extent_state_cache) |
60 | return -ENOMEM; | |
61 | ||
9601e3f6 CH |
62 | extent_buffer_cache = kmem_cache_create("extent_buffers", |
63 | sizeof(struct extent_buffer), 0, | |
64 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
65 | if (!extent_buffer_cache) |
66 | goto free_state_cache; | |
67 | return 0; | |
68 | ||
69 | free_state_cache: | |
70 | kmem_cache_destroy(extent_state_cache); | |
71 | return -ENOMEM; | |
72 | } | |
73 | ||
74 | void extent_io_exit(void) | |
75 | { | |
76 | struct extent_state *state; | |
2d2ae547 | 77 | struct extent_buffer *eb; |
d1310b2e CM |
78 | |
79 | while (!list_empty(&states)) { | |
2d2ae547 | 80 | state = list_entry(states.next, struct extent_state, leak_list); |
d397712b CM |
81 | printk(KERN_ERR "btrfs state leak: start %llu end %llu " |
82 | "state %lu in tree %p refs %d\n", | |
83 | (unsigned long long)state->start, | |
84 | (unsigned long long)state->end, | |
85 | state->state, state->tree, atomic_read(&state->refs)); | |
2d2ae547 | 86 | list_del(&state->leak_list); |
d1310b2e CM |
87 | kmem_cache_free(extent_state_cache, state); |
88 | ||
89 | } | |
90 | ||
2d2ae547 CM |
91 | while (!list_empty(&buffers)) { |
92 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
d397712b CM |
93 | printk(KERN_ERR "btrfs buffer leak start %llu len %lu " |
94 | "refs %d\n", (unsigned long long)eb->start, | |
95 | eb->len, atomic_read(&eb->refs)); | |
2d2ae547 CM |
96 | list_del(&eb->leak_list); |
97 | kmem_cache_free(extent_buffer_cache, eb); | |
98 | } | |
d1310b2e CM |
99 | if (extent_state_cache) |
100 | kmem_cache_destroy(extent_state_cache); | |
101 | if (extent_buffer_cache) | |
102 | kmem_cache_destroy(extent_buffer_cache); | |
103 | } | |
104 | ||
105 | void extent_io_tree_init(struct extent_io_tree *tree, | |
f993c883 | 106 | struct address_space *mapping) |
d1310b2e | 107 | { |
6bef4d31 | 108 | tree->state = RB_ROOT; |
19fe0a8b | 109 | INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC); |
d1310b2e CM |
110 | tree->ops = NULL; |
111 | tree->dirty_bytes = 0; | |
70dec807 | 112 | spin_lock_init(&tree->lock); |
6af118ce | 113 | spin_lock_init(&tree->buffer_lock); |
d1310b2e | 114 | tree->mapping = mapping; |
d1310b2e | 115 | } |
d1310b2e | 116 | |
b2950863 | 117 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
118 | { |
119 | struct extent_state *state; | |
3935127c | 120 | #if LEAK_DEBUG |
2d2ae547 | 121 | unsigned long flags; |
4bef0848 | 122 | #endif |
d1310b2e CM |
123 | |
124 | state = kmem_cache_alloc(extent_state_cache, mask); | |
2b114d1d | 125 | if (!state) |
d1310b2e CM |
126 | return state; |
127 | state->state = 0; | |
d1310b2e | 128 | state->private = 0; |
70dec807 | 129 | state->tree = NULL; |
3935127c | 130 | #if LEAK_DEBUG |
2d2ae547 CM |
131 | spin_lock_irqsave(&leak_lock, flags); |
132 | list_add(&state->leak_list, &states); | |
133 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 134 | #endif |
d1310b2e CM |
135 | atomic_set(&state->refs, 1); |
136 | init_waitqueue_head(&state->wq); | |
137 | return state; | |
138 | } | |
d1310b2e | 139 | |
4845e44f | 140 | void free_extent_state(struct extent_state *state) |
d1310b2e | 141 | { |
d1310b2e CM |
142 | if (!state) |
143 | return; | |
144 | if (atomic_dec_and_test(&state->refs)) { | |
3935127c | 145 | #if LEAK_DEBUG |
2d2ae547 | 146 | unsigned long flags; |
4bef0848 | 147 | #endif |
70dec807 | 148 | WARN_ON(state->tree); |
3935127c | 149 | #if LEAK_DEBUG |
2d2ae547 CM |
150 | spin_lock_irqsave(&leak_lock, flags); |
151 | list_del(&state->leak_list); | |
152 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 153 | #endif |
d1310b2e CM |
154 | kmem_cache_free(extent_state_cache, state); |
155 | } | |
156 | } | |
d1310b2e CM |
157 | |
158 | static struct rb_node *tree_insert(struct rb_root *root, u64 offset, | |
159 | struct rb_node *node) | |
160 | { | |
d397712b CM |
161 | struct rb_node **p = &root->rb_node; |
162 | struct rb_node *parent = NULL; | |
d1310b2e CM |
163 | struct tree_entry *entry; |
164 | ||
d397712b | 165 | while (*p) { |
d1310b2e CM |
166 | parent = *p; |
167 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
168 | ||
169 | if (offset < entry->start) | |
170 | p = &(*p)->rb_left; | |
171 | else if (offset > entry->end) | |
172 | p = &(*p)->rb_right; | |
173 | else | |
174 | return parent; | |
175 | } | |
176 | ||
177 | entry = rb_entry(node, struct tree_entry, rb_node); | |
d1310b2e CM |
178 | rb_link_node(node, parent, p); |
179 | rb_insert_color(node, root); | |
180 | return NULL; | |
181 | } | |
182 | ||
80ea96b1 | 183 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
d1310b2e CM |
184 | struct rb_node **prev_ret, |
185 | struct rb_node **next_ret) | |
186 | { | |
80ea96b1 | 187 | struct rb_root *root = &tree->state; |
d397712b | 188 | struct rb_node *n = root->rb_node; |
d1310b2e CM |
189 | struct rb_node *prev = NULL; |
190 | struct rb_node *orig_prev = NULL; | |
191 | struct tree_entry *entry; | |
192 | struct tree_entry *prev_entry = NULL; | |
193 | ||
d397712b | 194 | while (n) { |
d1310b2e CM |
195 | entry = rb_entry(n, struct tree_entry, rb_node); |
196 | prev = n; | |
197 | prev_entry = entry; | |
198 | ||
199 | if (offset < entry->start) | |
200 | n = n->rb_left; | |
201 | else if (offset > entry->end) | |
202 | n = n->rb_right; | |
d397712b | 203 | else |
d1310b2e CM |
204 | return n; |
205 | } | |
206 | ||
207 | if (prev_ret) { | |
208 | orig_prev = prev; | |
d397712b | 209 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
210 | prev = rb_next(prev); |
211 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
212 | } | |
213 | *prev_ret = prev; | |
214 | prev = orig_prev; | |
215 | } | |
216 | ||
217 | if (next_ret) { | |
218 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 219 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
220 | prev = rb_prev(prev); |
221 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
222 | } | |
223 | *next_ret = prev; | |
224 | } | |
225 | return NULL; | |
226 | } | |
227 | ||
80ea96b1 CM |
228 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
229 | u64 offset) | |
d1310b2e | 230 | { |
70dec807 | 231 | struct rb_node *prev = NULL; |
d1310b2e | 232 | struct rb_node *ret; |
70dec807 | 233 | |
80ea96b1 | 234 | ret = __etree_search(tree, offset, &prev, NULL); |
d397712b | 235 | if (!ret) |
d1310b2e CM |
236 | return prev; |
237 | return ret; | |
238 | } | |
239 | ||
9ed74f2d JB |
240 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, |
241 | struct extent_state *other) | |
242 | { | |
243 | if (tree->ops && tree->ops->merge_extent_hook) | |
244 | tree->ops->merge_extent_hook(tree->mapping->host, new, | |
245 | other); | |
246 | } | |
247 | ||
d1310b2e CM |
248 | /* |
249 | * utility function to look for merge candidates inside a given range. | |
250 | * Any extents with matching state are merged together into a single | |
251 | * extent in the tree. Extents with EXTENT_IO in their state field | |
252 | * are not merged because the end_io handlers need to be able to do | |
253 | * operations on them without sleeping (or doing allocations/splits). | |
254 | * | |
255 | * This should be called with the tree lock held. | |
256 | */ | |
1bf85046 JM |
257 | static void merge_state(struct extent_io_tree *tree, |
258 | struct extent_state *state) | |
d1310b2e CM |
259 | { |
260 | struct extent_state *other; | |
261 | struct rb_node *other_node; | |
262 | ||
5b21f2ed | 263 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
1bf85046 | 264 | return; |
d1310b2e CM |
265 | |
266 | other_node = rb_prev(&state->rb_node); | |
267 | if (other_node) { | |
268 | other = rb_entry(other_node, struct extent_state, rb_node); | |
269 | if (other->end == state->start - 1 && | |
270 | other->state == state->state) { | |
9ed74f2d | 271 | merge_cb(tree, state, other); |
d1310b2e | 272 | state->start = other->start; |
70dec807 | 273 | other->tree = NULL; |
d1310b2e CM |
274 | rb_erase(&other->rb_node, &tree->state); |
275 | free_extent_state(other); | |
276 | } | |
277 | } | |
278 | other_node = rb_next(&state->rb_node); | |
279 | if (other_node) { | |
280 | other = rb_entry(other_node, struct extent_state, rb_node); | |
281 | if (other->start == state->end + 1 && | |
282 | other->state == state->state) { | |
9ed74f2d | 283 | merge_cb(tree, state, other); |
df98b6e2 JB |
284 | state->end = other->end; |
285 | other->tree = NULL; | |
286 | rb_erase(&other->rb_node, &tree->state); | |
287 | free_extent_state(other); | |
d1310b2e CM |
288 | } |
289 | } | |
d1310b2e CM |
290 | } |
291 | ||
1bf85046 | 292 | static void set_state_cb(struct extent_io_tree *tree, |
0ca1f7ce | 293 | struct extent_state *state, int *bits) |
291d673e | 294 | { |
1bf85046 JM |
295 | if (tree->ops && tree->ops->set_bit_hook) |
296 | tree->ops->set_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
297 | } |
298 | ||
299 | static void clear_state_cb(struct extent_io_tree *tree, | |
0ca1f7ce | 300 | struct extent_state *state, int *bits) |
291d673e | 301 | { |
9ed74f2d JB |
302 | if (tree->ops && tree->ops->clear_bit_hook) |
303 | tree->ops->clear_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
304 | } |
305 | ||
3150b699 XG |
306 | static void set_state_bits(struct extent_io_tree *tree, |
307 | struct extent_state *state, int *bits); | |
308 | ||
d1310b2e CM |
309 | /* |
310 | * insert an extent_state struct into the tree. 'bits' are set on the | |
311 | * struct before it is inserted. | |
312 | * | |
313 | * This may return -EEXIST if the extent is already there, in which case the | |
314 | * state struct is freed. | |
315 | * | |
316 | * The tree lock is not taken internally. This is a utility function and | |
317 | * probably isn't what you want to call (see set/clear_extent_bit). | |
318 | */ | |
319 | static int insert_state(struct extent_io_tree *tree, | |
320 | struct extent_state *state, u64 start, u64 end, | |
0ca1f7ce | 321 | int *bits) |
d1310b2e CM |
322 | { |
323 | struct rb_node *node; | |
324 | ||
325 | if (end < start) { | |
d397712b CM |
326 | printk(KERN_ERR "btrfs end < start %llu %llu\n", |
327 | (unsigned long long)end, | |
328 | (unsigned long long)start); | |
d1310b2e CM |
329 | WARN_ON(1); |
330 | } | |
d1310b2e CM |
331 | state->start = start; |
332 | state->end = end; | |
9ed74f2d | 333 | |
3150b699 XG |
334 | set_state_bits(tree, state, bits); |
335 | ||
d1310b2e CM |
336 | node = tree_insert(&tree->state, end, &state->rb_node); |
337 | if (node) { | |
338 | struct extent_state *found; | |
339 | found = rb_entry(node, struct extent_state, rb_node); | |
d397712b CM |
340 | printk(KERN_ERR "btrfs found node %llu %llu on insert of " |
341 | "%llu %llu\n", (unsigned long long)found->start, | |
342 | (unsigned long long)found->end, | |
343 | (unsigned long long)start, (unsigned long long)end); | |
d1310b2e CM |
344 | return -EEXIST; |
345 | } | |
70dec807 | 346 | state->tree = tree; |
d1310b2e CM |
347 | merge_state(tree, state); |
348 | return 0; | |
349 | } | |
350 | ||
1bf85046 | 351 | static void split_cb(struct extent_io_tree *tree, struct extent_state *orig, |
9ed74f2d JB |
352 | u64 split) |
353 | { | |
354 | if (tree->ops && tree->ops->split_extent_hook) | |
1bf85046 | 355 | tree->ops->split_extent_hook(tree->mapping->host, orig, split); |
9ed74f2d JB |
356 | } |
357 | ||
d1310b2e CM |
358 | /* |
359 | * split a given extent state struct in two, inserting the preallocated | |
360 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
361 | * offset inside 'orig' where it should be split. | |
362 | * | |
363 | * Before calling, | |
364 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
365 | * are two extent state structs in the tree: | |
366 | * prealloc: [orig->start, split - 1] | |
367 | * orig: [ split, orig->end ] | |
368 | * | |
369 | * The tree locks are not taken by this function. They need to be held | |
370 | * by the caller. | |
371 | */ | |
372 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
373 | struct extent_state *prealloc, u64 split) | |
374 | { | |
375 | struct rb_node *node; | |
9ed74f2d JB |
376 | |
377 | split_cb(tree, orig, split); | |
378 | ||
d1310b2e CM |
379 | prealloc->start = orig->start; |
380 | prealloc->end = split - 1; | |
381 | prealloc->state = orig->state; | |
382 | orig->start = split; | |
383 | ||
384 | node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node); | |
385 | if (node) { | |
d1310b2e CM |
386 | free_extent_state(prealloc); |
387 | return -EEXIST; | |
388 | } | |
70dec807 | 389 | prealloc->tree = tree; |
d1310b2e CM |
390 | return 0; |
391 | } | |
392 | ||
393 | /* | |
394 | * utility function to clear some bits in an extent state struct. | |
395 | * it will optionally wake up any one waiting on this state (wake == 1), or | |
396 | * forcibly remove the state from the tree (delete == 1). | |
397 | * | |
398 | * If no bits are set on the state struct after clearing things, the | |
399 | * struct is freed and removed from the tree | |
400 | */ | |
401 | static int clear_state_bit(struct extent_io_tree *tree, | |
0ca1f7ce YZ |
402 | struct extent_state *state, |
403 | int *bits, int wake) | |
d1310b2e | 404 | { |
0ca1f7ce | 405 | int bits_to_clear = *bits & ~EXTENT_CTLBITS; |
32c00aff | 406 | int ret = state->state & bits_to_clear; |
d1310b2e | 407 | |
0ca1f7ce | 408 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
409 | u64 range = state->end - state->start + 1; |
410 | WARN_ON(range > tree->dirty_bytes); | |
411 | tree->dirty_bytes -= range; | |
412 | } | |
291d673e | 413 | clear_state_cb(tree, state, bits); |
32c00aff | 414 | state->state &= ~bits_to_clear; |
d1310b2e CM |
415 | if (wake) |
416 | wake_up(&state->wq); | |
0ca1f7ce | 417 | if (state->state == 0) { |
70dec807 | 418 | if (state->tree) { |
d1310b2e | 419 | rb_erase(&state->rb_node, &tree->state); |
70dec807 | 420 | state->tree = NULL; |
d1310b2e CM |
421 | free_extent_state(state); |
422 | } else { | |
423 | WARN_ON(1); | |
424 | } | |
425 | } else { | |
426 | merge_state(tree, state); | |
427 | } | |
428 | return ret; | |
429 | } | |
430 | ||
8233767a XG |
431 | static struct extent_state * |
432 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
433 | { | |
434 | if (!prealloc) | |
435 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
436 | ||
437 | return prealloc; | |
438 | } | |
439 | ||
d1310b2e CM |
440 | /* |
441 | * clear some bits on a range in the tree. This may require splitting | |
442 | * or inserting elements in the tree, so the gfp mask is used to | |
443 | * indicate which allocations or sleeping are allowed. | |
444 | * | |
445 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
446 | * the given range from the tree regardless of state (ie for truncate). | |
447 | * | |
448 | * the range [start, end] is inclusive. | |
449 | * | |
450 | * This takes the tree lock, and returns < 0 on error, > 0 if any of the | |
451 | * bits were already set, or zero if none of the bits were already set. | |
452 | */ | |
453 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
2c64c53d CM |
454 | int bits, int wake, int delete, |
455 | struct extent_state **cached_state, | |
456 | gfp_t mask) | |
d1310b2e CM |
457 | { |
458 | struct extent_state *state; | |
2c64c53d | 459 | struct extent_state *cached; |
d1310b2e | 460 | struct extent_state *prealloc = NULL; |
2c64c53d | 461 | struct rb_node *next_node; |
d1310b2e | 462 | struct rb_node *node; |
5c939df5 | 463 | u64 last_end; |
d1310b2e CM |
464 | int err; |
465 | int set = 0; | |
2ac55d41 | 466 | int clear = 0; |
d1310b2e | 467 | |
0ca1f7ce YZ |
468 | if (delete) |
469 | bits |= ~EXTENT_CTLBITS; | |
470 | bits |= EXTENT_FIRST_DELALLOC; | |
471 | ||
2ac55d41 JB |
472 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
473 | clear = 1; | |
d1310b2e CM |
474 | again: |
475 | if (!prealloc && (mask & __GFP_WAIT)) { | |
476 | prealloc = alloc_extent_state(mask); | |
477 | if (!prealloc) | |
478 | return -ENOMEM; | |
479 | } | |
480 | ||
cad321ad | 481 | spin_lock(&tree->lock); |
2c64c53d CM |
482 | if (cached_state) { |
483 | cached = *cached_state; | |
2ac55d41 JB |
484 | |
485 | if (clear) { | |
486 | *cached_state = NULL; | |
487 | cached_state = NULL; | |
488 | } | |
489 | ||
df98b6e2 JB |
490 | if (cached && cached->tree && cached->start <= start && |
491 | cached->end > start) { | |
2ac55d41 JB |
492 | if (clear) |
493 | atomic_dec(&cached->refs); | |
2c64c53d | 494 | state = cached; |
42daec29 | 495 | goto hit_next; |
2c64c53d | 496 | } |
2ac55d41 JB |
497 | if (clear) |
498 | free_extent_state(cached); | |
2c64c53d | 499 | } |
d1310b2e CM |
500 | /* |
501 | * this search will find the extents that end after | |
502 | * our range starts | |
503 | */ | |
80ea96b1 | 504 | node = tree_search(tree, start); |
d1310b2e CM |
505 | if (!node) |
506 | goto out; | |
507 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 508 | hit_next: |
d1310b2e CM |
509 | if (state->start > end) |
510 | goto out; | |
511 | WARN_ON(state->end < start); | |
5c939df5 | 512 | last_end = state->end; |
d1310b2e CM |
513 | |
514 | /* | |
515 | * | ---- desired range ---- | | |
516 | * | state | or | |
517 | * | ------------- state -------------- | | |
518 | * | |
519 | * We need to split the extent we found, and may flip | |
520 | * bits on second half. | |
521 | * | |
522 | * If the extent we found extends past our range, we | |
523 | * just split and search again. It'll get split again | |
524 | * the next time though. | |
525 | * | |
526 | * If the extent we found is inside our range, we clear | |
527 | * the desired bit on it. | |
528 | */ | |
529 | ||
530 | if (state->start < start) { | |
8233767a XG |
531 | prealloc = alloc_extent_state_atomic(prealloc); |
532 | BUG_ON(!prealloc); | |
d1310b2e CM |
533 | err = split_state(tree, state, prealloc, start); |
534 | BUG_ON(err == -EEXIST); | |
535 | prealloc = NULL; | |
536 | if (err) | |
537 | goto out; | |
538 | if (state->end <= end) { | |
0ca1f7ce | 539 | set |= clear_state_bit(tree, state, &bits, wake); |
5c939df5 YZ |
540 | if (last_end == (u64)-1) |
541 | goto out; | |
542 | start = last_end + 1; | |
d1310b2e CM |
543 | } |
544 | goto search_again; | |
545 | } | |
546 | /* | |
547 | * | ---- desired range ---- | | |
548 | * | state | | |
549 | * We need to split the extent, and clear the bit | |
550 | * on the first half | |
551 | */ | |
552 | if (state->start <= end && state->end > end) { | |
8233767a XG |
553 | prealloc = alloc_extent_state_atomic(prealloc); |
554 | BUG_ON(!prealloc); | |
d1310b2e CM |
555 | err = split_state(tree, state, prealloc, end + 1); |
556 | BUG_ON(err == -EEXIST); | |
d1310b2e CM |
557 | if (wake) |
558 | wake_up(&state->wq); | |
42daec29 | 559 | |
0ca1f7ce | 560 | set |= clear_state_bit(tree, prealloc, &bits, wake); |
9ed74f2d | 561 | |
d1310b2e CM |
562 | prealloc = NULL; |
563 | goto out; | |
564 | } | |
42daec29 | 565 | |
2c64c53d CM |
566 | if (state->end < end && prealloc && !need_resched()) |
567 | next_node = rb_next(&state->rb_node); | |
568 | else | |
569 | next_node = NULL; | |
42daec29 | 570 | |
0ca1f7ce | 571 | set |= clear_state_bit(tree, state, &bits, wake); |
5c939df5 YZ |
572 | if (last_end == (u64)-1) |
573 | goto out; | |
574 | start = last_end + 1; | |
2c64c53d CM |
575 | if (start <= end && next_node) { |
576 | state = rb_entry(next_node, struct extent_state, | |
577 | rb_node); | |
578 | if (state->start == start) | |
579 | goto hit_next; | |
580 | } | |
d1310b2e CM |
581 | goto search_again; |
582 | ||
583 | out: | |
cad321ad | 584 | spin_unlock(&tree->lock); |
d1310b2e CM |
585 | if (prealloc) |
586 | free_extent_state(prealloc); | |
587 | ||
588 | return set; | |
589 | ||
590 | search_again: | |
591 | if (start > end) | |
592 | goto out; | |
cad321ad | 593 | spin_unlock(&tree->lock); |
d1310b2e CM |
594 | if (mask & __GFP_WAIT) |
595 | cond_resched(); | |
596 | goto again; | |
597 | } | |
d1310b2e CM |
598 | |
599 | static int wait_on_state(struct extent_io_tree *tree, | |
600 | struct extent_state *state) | |
641f5219 CH |
601 | __releases(tree->lock) |
602 | __acquires(tree->lock) | |
d1310b2e CM |
603 | { |
604 | DEFINE_WAIT(wait); | |
605 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 606 | spin_unlock(&tree->lock); |
d1310b2e | 607 | schedule(); |
cad321ad | 608 | spin_lock(&tree->lock); |
d1310b2e CM |
609 | finish_wait(&state->wq, &wait); |
610 | return 0; | |
611 | } | |
612 | ||
613 | /* | |
614 | * waits for one or more bits to clear on a range in the state tree. | |
615 | * The range [start, end] is inclusive. | |
616 | * The tree lock is taken by this function | |
617 | */ | |
618 | int wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits) | |
619 | { | |
620 | struct extent_state *state; | |
621 | struct rb_node *node; | |
622 | ||
cad321ad | 623 | spin_lock(&tree->lock); |
d1310b2e CM |
624 | again: |
625 | while (1) { | |
626 | /* | |
627 | * this search will find all the extents that end after | |
628 | * our range starts | |
629 | */ | |
80ea96b1 | 630 | node = tree_search(tree, start); |
d1310b2e CM |
631 | if (!node) |
632 | break; | |
633 | ||
634 | state = rb_entry(node, struct extent_state, rb_node); | |
635 | ||
636 | if (state->start > end) | |
637 | goto out; | |
638 | ||
639 | if (state->state & bits) { | |
640 | start = state->start; | |
641 | atomic_inc(&state->refs); | |
642 | wait_on_state(tree, state); | |
643 | free_extent_state(state); | |
644 | goto again; | |
645 | } | |
646 | start = state->end + 1; | |
647 | ||
648 | if (start > end) | |
649 | break; | |
650 | ||
ded91f08 | 651 | cond_resched_lock(&tree->lock); |
d1310b2e CM |
652 | } |
653 | out: | |
cad321ad | 654 | spin_unlock(&tree->lock); |
d1310b2e CM |
655 | return 0; |
656 | } | |
d1310b2e | 657 | |
1bf85046 | 658 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 659 | struct extent_state *state, |
0ca1f7ce | 660 | int *bits) |
d1310b2e | 661 | { |
0ca1f7ce | 662 | int bits_to_set = *bits & ~EXTENT_CTLBITS; |
9ed74f2d | 663 | |
1bf85046 | 664 | set_state_cb(tree, state, bits); |
0ca1f7ce | 665 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
666 | u64 range = state->end - state->start + 1; |
667 | tree->dirty_bytes += range; | |
668 | } | |
0ca1f7ce | 669 | state->state |= bits_to_set; |
d1310b2e CM |
670 | } |
671 | ||
2c64c53d CM |
672 | static void cache_state(struct extent_state *state, |
673 | struct extent_state **cached_ptr) | |
674 | { | |
675 | if (cached_ptr && !(*cached_ptr)) { | |
676 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) { | |
677 | *cached_ptr = state; | |
678 | atomic_inc(&state->refs); | |
679 | } | |
680 | } | |
681 | } | |
682 | ||
507903b8 AJ |
683 | static void uncache_state(struct extent_state **cached_ptr) |
684 | { | |
685 | if (cached_ptr && (*cached_ptr)) { | |
686 | struct extent_state *state = *cached_ptr; | |
109b36a2 CM |
687 | *cached_ptr = NULL; |
688 | free_extent_state(state); | |
507903b8 AJ |
689 | } |
690 | } | |
691 | ||
d1310b2e | 692 | /* |
1edbb734 CM |
693 | * set some bits on a range in the tree. This may require allocations or |
694 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 695 | * |
1edbb734 CM |
696 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
697 | * part of the range already has the desired bits set. The start of the | |
698 | * existing range is returned in failed_start in this case. | |
d1310b2e | 699 | * |
1edbb734 | 700 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 701 | */ |
1edbb734 | 702 | |
4845e44f CM |
703 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
704 | int bits, int exclusive_bits, u64 *failed_start, | |
705 | struct extent_state **cached_state, gfp_t mask) | |
d1310b2e CM |
706 | { |
707 | struct extent_state *state; | |
708 | struct extent_state *prealloc = NULL; | |
709 | struct rb_node *node; | |
d1310b2e | 710 | int err = 0; |
d1310b2e CM |
711 | u64 last_start; |
712 | u64 last_end; | |
42daec29 | 713 | |
0ca1f7ce | 714 | bits |= EXTENT_FIRST_DELALLOC; |
d1310b2e CM |
715 | again: |
716 | if (!prealloc && (mask & __GFP_WAIT)) { | |
717 | prealloc = alloc_extent_state(mask); | |
8233767a | 718 | BUG_ON(!prealloc); |
d1310b2e CM |
719 | } |
720 | ||
cad321ad | 721 | spin_lock(&tree->lock); |
9655d298 CM |
722 | if (cached_state && *cached_state) { |
723 | state = *cached_state; | |
df98b6e2 JB |
724 | if (state->start <= start && state->end > start && |
725 | state->tree) { | |
9655d298 CM |
726 | node = &state->rb_node; |
727 | goto hit_next; | |
728 | } | |
729 | } | |
d1310b2e CM |
730 | /* |
731 | * this search will find all the extents that end after | |
732 | * our range starts. | |
733 | */ | |
80ea96b1 | 734 | node = tree_search(tree, start); |
d1310b2e | 735 | if (!node) { |
8233767a XG |
736 | prealloc = alloc_extent_state_atomic(prealloc); |
737 | BUG_ON(!prealloc); | |
0ca1f7ce | 738 | err = insert_state(tree, prealloc, start, end, &bits); |
d1310b2e CM |
739 | prealloc = NULL; |
740 | BUG_ON(err == -EEXIST); | |
741 | goto out; | |
742 | } | |
d1310b2e | 743 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 744 | hit_next: |
d1310b2e CM |
745 | last_start = state->start; |
746 | last_end = state->end; | |
747 | ||
748 | /* | |
749 | * | ---- desired range ---- | | |
750 | * | state | | |
751 | * | |
752 | * Just lock what we found and keep going | |
753 | */ | |
754 | if (state->start == start && state->end <= end) { | |
40431d6c | 755 | struct rb_node *next_node; |
1edbb734 | 756 | if (state->state & exclusive_bits) { |
d1310b2e CM |
757 | *failed_start = state->start; |
758 | err = -EEXIST; | |
759 | goto out; | |
760 | } | |
42daec29 | 761 | |
1bf85046 | 762 | set_state_bits(tree, state, &bits); |
9ed74f2d | 763 | |
2c64c53d | 764 | cache_state(state, cached_state); |
d1310b2e | 765 | merge_state(tree, state); |
5c939df5 YZ |
766 | if (last_end == (u64)-1) |
767 | goto out; | |
40431d6c | 768 | |
5c939df5 | 769 | start = last_end + 1; |
df98b6e2 | 770 | next_node = rb_next(&state->rb_node); |
c7f895a2 XG |
771 | if (next_node && start < end && prealloc && !need_resched()) { |
772 | state = rb_entry(next_node, struct extent_state, | |
773 | rb_node); | |
774 | if (state->start == start) | |
775 | goto hit_next; | |
40431d6c | 776 | } |
d1310b2e CM |
777 | goto search_again; |
778 | } | |
779 | ||
780 | /* | |
781 | * | ---- desired range ---- | | |
782 | * | state | | |
783 | * or | |
784 | * | ------------- state -------------- | | |
785 | * | |
786 | * We need to split the extent we found, and may flip bits on | |
787 | * second half. | |
788 | * | |
789 | * If the extent we found extends past our | |
790 | * range, we just split and search again. It'll get split | |
791 | * again the next time though. | |
792 | * | |
793 | * If the extent we found is inside our range, we set the | |
794 | * desired bit on it. | |
795 | */ | |
796 | if (state->start < start) { | |
1edbb734 | 797 | if (state->state & exclusive_bits) { |
d1310b2e CM |
798 | *failed_start = start; |
799 | err = -EEXIST; | |
800 | goto out; | |
801 | } | |
8233767a XG |
802 | |
803 | prealloc = alloc_extent_state_atomic(prealloc); | |
804 | BUG_ON(!prealloc); | |
d1310b2e CM |
805 | err = split_state(tree, state, prealloc, start); |
806 | BUG_ON(err == -EEXIST); | |
807 | prealloc = NULL; | |
808 | if (err) | |
809 | goto out; | |
810 | if (state->end <= end) { | |
1bf85046 | 811 | set_state_bits(tree, state, &bits); |
2c64c53d | 812 | cache_state(state, cached_state); |
d1310b2e | 813 | merge_state(tree, state); |
5c939df5 YZ |
814 | if (last_end == (u64)-1) |
815 | goto out; | |
816 | start = last_end + 1; | |
d1310b2e CM |
817 | } |
818 | goto search_again; | |
819 | } | |
820 | /* | |
821 | * | ---- desired range ---- | | |
822 | * | state | or | state | | |
823 | * | |
824 | * There's a hole, we need to insert something in it and | |
825 | * ignore the extent we found. | |
826 | */ | |
827 | if (state->start > start) { | |
828 | u64 this_end; | |
829 | if (end < last_start) | |
830 | this_end = end; | |
831 | else | |
d397712b | 832 | this_end = last_start - 1; |
8233767a XG |
833 | |
834 | prealloc = alloc_extent_state_atomic(prealloc); | |
835 | BUG_ON(!prealloc); | |
c7f895a2 XG |
836 | |
837 | /* | |
838 | * Avoid to free 'prealloc' if it can be merged with | |
839 | * the later extent. | |
840 | */ | |
d1310b2e | 841 | err = insert_state(tree, prealloc, start, this_end, |
0ca1f7ce | 842 | &bits); |
d1310b2e | 843 | BUG_ON(err == -EEXIST); |
9ed74f2d | 844 | if (err) { |
c7f895a2 | 845 | free_extent_state(prealloc); |
9ed74f2d | 846 | prealloc = NULL; |
d1310b2e | 847 | goto out; |
9ed74f2d JB |
848 | } |
849 | cache_state(prealloc, cached_state); | |
850 | prealloc = NULL; | |
d1310b2e CM |
851 | start = this_end + 1; |
852 | goto search_again; | |
853 | } | |
854 | /* | |
855 | * | ---- desired range ---- | | |
856 | * | state | | |
857 | * We need to split the extent, and set the bit | |
858 | * on the first half | |
859 | */ | |
860 | if (state->start <= end && state->end > end) { | |
1edbb734 | 861 | if (state->state & exclusive_bits) { |
d1310b2e CM |
862 | *failed_start = start; |
863 | err = -EEXIST; | |
864 | goto out; | |
865 | } | |
8233767a XG |
866 | |
867 | prealloc = alloc_extent_state_atomic(prealloc); | |
868 | BUG_ON(!prealloc); | |
d1310b2e CM |
869 | err = split_state(tree, state, prealloc, end + 1); |
870 | BUG_ON(err == -EEXIST); | |
871 | ||
1bf85046 | 872 | set_state_bits(tree, prealloc, &bits); |
2c64c53d | 873 | cache_state(prealloc, cached_state); |
d1310b2e CM |
874 | merge_state(tree, prealloc); |
875 | prealloc = NULL; | |
876 | goto out; | |
877 | } | |
878 | ||
879 | goto search_again; | |
880 | ||
881 | out: | |
cad321ad | 882 | spin_unlock(&tree->lock); |
d1310b2e CM |
883 | if (prealloc) |
884 | free_extent_state(prealloc); | |
885 | ||
886 | return err; | |
887 | ||
888 | search_again: | |
889 | if (start > end) | |
890 | goto out; | |
cad321ad | 891 | spin_unlock(&tree->lock); |
d1310b2e CM |
892 | if (mask & __GFP_WAIT) |
893 | cond_resched(); | |
894 | goto again; | |
895 | } | |
d1310b2e CM |
896 | |
897 | /* wrappers around set/clear extent bit */ | |
898 | int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
899 | gfp_t mask) | |
900 | { | |
901 | return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL, | |
2c64c53d | 902 | NULL, mask); |
d1310b2e | 903 | } |
d1310b2e CM |
904 | |
905 | int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
906 | int bits, gfp_t mask) | |
907 | { | |
908 | return set_extent_bit(tree, start, end, bits, 0, NULL, | |
2c64c53d | 909 | NULL, mask); |
d1310b2e | 910 | } |
d1310b2e CM |
911 | |
912 | int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
913 | int bits, gfp_t mask) | |
914 | { | |
2c64c53d | 915 | return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask); |
d1310b2e | 916 | } |
d1310b2e CM |
917 | |
918 | int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, | |
2ac55d41 | 919 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
920 | { |
921 | return set_extent_bit(tree, start, end, | |
40431d6c | 922 | EXTENT_DELALLOC | EXTENT_DIRTY | EXTENT_UPTODATE, |
2ac55d41 | 923 | 0, NULL, cached_state, mask); |
d1310b2e | 924 | } |
d1310b2e CM |
925 | |
926 | int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
927 | gfp_t mask) | |
928 | { | |
929 | return clear_extent_bit(tree, start, end, | |
32c00aff | 930 | EXTENT_DIRTY | EXTENT_DELALLOC | |
0ca1f7ce | 931 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask); |
d1310b2e | 932 | } |
d1310b2e CM |
933 | |
934 | int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, | |
935 | gfp_t mask) | |
936 | { | |
937 | return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL, | |
2c64c53d | 938 | NULL, mask); |
d1310b2e | 939 | } |
d1310b2e | 940 | |
d1310b2e | 941 | int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
507903b8 | 942 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e | 943 | { |
507903b8 AJ |
944 | return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, |
945 | NULL, cached_state, mask); | |
d1310b2e | 946 | } |
d1310b2e | 947 | |
d397712b | 948 | static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, |
2ac55d41 JB |
949 | u64 end, struct extent_state **cached_state, |
950 | gfp_t mask) | |
d1310b2e | 951 | { |
2c64c53d | 952 | return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, |
2ac55d41 | 953 | cached_state, mask); |
d1310b2e | 954 | } |
d1310b2e | 955 | |
d352ac68 CM |
956 | /* |
957 | * either insert or lock state struct between start and end use mask to tell | |
958 | * us if waiting is desired. | |
959 | */ | |
1edbb734 | 960 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
2c64c53d | 961 | int bits, struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
962 | { |
963 | int err; | |
964 | u64 failed_start; | |
965 | while (1) { | |
1edbb734 | 966 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits, |
2c64c53d CM |
967 | EXTENT_LOCKED, &failed_start, |
968 | cached_state, mask); | |
d1310b2e CM |
969 | if (err == -EEXIST && (mask & __GFP_WAIT)) { |
970 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); | |
971 | start = failed_start; | |
972 | } else { | |
973 | break; | |
974 | } | |
975 | WARN_ON(start > end); | |
976 | } | |
977 | return err; | |
978 | } | |
d1310b2e | 979 | |
1edbb734 CM |
980 | int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) |
981 | { | |
2c64c53d | 982 | return lock_extent_bits(tree, start, end, 0, NULL, mask); |
1edbb734 CM |
983 | } |
984 | ||
25179201 JB |
985 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end, |
986 | gfp_t mask) | |
987 | { | |
988 | int err; | |
989 | u64 failed_start; | |
990 | ||
2c64c53d CM |
991 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
992 | &failed_start, NULL, mask); | |
6643558d YZ |
993 | if (err == -EEXIST) { |
994 | if (failed_start > start) | |
995 | clear_extent_bit(tree, start, failed_start - 1, | |
2c64c53d | 996 | EXTENT_LOCKED, 1, 0, NULL, mask); |
25179201 | 997 | return 0; |
6643558d | 998 | } |
25179201 JB |
999 | return 1; |
1000 | } | |
25179201 | 1001 | |
2c64c53d CM |
1002 | int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end, |
1003 | struct extent_state **cached, gfp_t mask) | |
1004 | { | |
1005 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, | |
1006 | mask); | |
1007 | } | |
1008 | ||
507903b8 | 1009 | int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) |
d1310b2e | 1010 | { |
2c64c53d CM |
1011 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL, |
1012 | mask); | |
d1310b2e | 1013 | } |
d1310b2e | 1014 | |
d1310b2e CM |
1015 | /* |
1016 | * helper function to set both pages and extents in the tree writeback | |
1017 | */ | |
b2950863 | 1018 | static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e CM |
1019 | { |
1020 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1021 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1022 | struct page *page; | |
1023 | ||
1024 | while (index <= end_index) { | |
1025 | page = find_get_page(tree->mapping, index); | |
1026 | BUG_ON(!page); | |
1027 | set_page_writeback(page); | |
1028 | page_cache_release(page); | |
1029 | index++; | |
1030 | } | |
d1310b2e CM |
1031 | return 0; |
1032 | } | |
d1310b2e | 1033 | |
d352ac68 CM |
1034 | /* find the first state struct with 'bits' set after 'start', and |
1035 | * return it. tree->lock must be held. NULL will returned if | |
1036 | * nothing was found after 'start' | |
1037 | */ | |
d7fc640e CM |
1038 | struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree, |
1039 | u64 start, int bits) | |
1040 | { | |
1041 | struct rb_node *node; | |
1042 | struct extent_state *state; | |
1043 | ||
1044 | /* | |
1045 | * this search will find all the extents that end after | |
1046 | * our range starts. | |
1047 | */ | |
1048 | node = tree_search(tree, start); | |
d397712b | 1049 | if (!node) |
d7fc640e | 1050 | goto out; |
d7fc640e | 1051 | |
d397712b | 1052 | while (1) { |
d7fc640e | 1053 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1054 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1055 | return state; |
d397712b | 1056 | |
d7fc640e CM |
1057 | node = rb_next(node); |
1058 | if (!node) | |
1059 | break; | |
1060 | } | |
1061 | out: | |
1062 | return NULL; | |
1063 | } | |
d7fc640e | 1064 | |
69261c4b XG |
1065 | /* |
1066 | * find the first offset in the io tree with 'bits' set. zero is | |
1067 | * returned if we find something, and *start_ret and *end_ret are | |
1068 | * set to reflect the state struct that was found. | |
1069 | * | |
1070 | * If nothing was found, 1 is returned, < 0 on error | |
1071 | */ | |
1072 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
1073 | u64 *start_ret, u64 *end_ret, int bits) | |
1074 | { | |
1075 | struct extent_state *state; | |
1076 | int ret = 1; | |
1077 | ||
1078 | spin_lock(&tree->lock); | |
1079 | state = find_first_extent_bit_state(tree, start, bits); | |
1080 | if (state) { | |
1081 | *start_ret = state->start; | |
1082 | *end_ret = state->end; | |
1083 | ret = 0; | |
1084 | } | |
1085 | spin_unlock(&tree->lock); | |
1086 | return ret; | |
1087 | } | |
1088 | ||
d352ac68 CM |
1089 | /* |
1090 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1091 | * more than 'max_bytes'. start and end are used to return the range, | |
1092 | * | |
1093 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1094 | */ | |
c8b97818 | 1095 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1096 | u64 *start, u64 *end, u64 max_bytes, |
1097 | struct extent_state **cached_state) | |
d1310b2e CM |
1098 | { |
1099 | struct rb_node *node; | |
1100 | struct extent_state *state; | |
1101 | u64 cur_start = *start; | |
1102 | u64 found = 0; | |
1103 | u64 total_bytes = 0; | |
1104 | ||
cad321ad | 1105 | spin_lock(&tree->lock); |
c8b97818 | 1106 | |
d1310b2e CM |
1107 | /* |
1108 | * this search will find all the extents that end after | |
1109 | * our range starts. | |
1110 | */ | |
80ea96b1 | 1111 | node = tree_search(tree, cur_start); |
2b114d1d | 1112 | if (!node) { |
3b951516 CM |
1113 | if (!found) |
1114 | *end = (u64)-1; | |
d1310b2e CM |
1115 | goto out; |
1116 | } | |
1117 | ||
d397712b | 1118 | while (1) { |
d1310b2e | 1119 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1120 | if (found && (state->start != cur_start || |
1121 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1122 | goto out; |
1123 | } | |
1124 | if (!(state->state & EXTENT_DELALLOC)) { | |
1125 | if (!found) | |
1126 | *end = state->end; | |
1127 | goto out; | |
1128 | } | |
c2a128d2 | 1129 | if (!found) { |
d1310b2e | 1130 | *start = state->start; |
c2a128d2 JB |
1131 | *cached_state = state; |
1132 | atomic_inc(&state->refs); | |
1133 | } | |
d1310b2e CM |
1134 | found++; |
1135 | *end = state->end; | |
1136 | cur_start = state->end + 1; | |
1137 | node = rb_next(node); | |
1138 | if (!node) | |
1139 | break; | |
1140 | total_bytes += state->end - state->start + 1; | |
1141 | if (total_bytes >= max_bytes) | |
1142 | break; | |
1143 | } | |
1144 | out: | |
cad321ad | 1145 | spin_unlock(&tree->lock); |
d1310b2e CM |
1146 | return found; |
1147 | } | |
1148 | ||
c8b97818 CM |
1149 | static noinline int __unlock_for_delalloc(struct inode *inode, |
1150 | struct page *locked_page, | |
1151 | u64 start, u64 end) | |
1152 | { | |
1153 | int ret; | |
1154 | struct page *pages[16]; | |
1155 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1156 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1157 | unsigned long nr_pages = end_index - index + 1; | |
1158 | int i; | |
1159 | ||
1160 | if (index == locked_page->index && end_index == index) | |
1161 | return 0; | |
1162 | ||
d397712b | 1163 | while (nr_pages > 0) { |
c8b97818 | 1164 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1165 | min_t(unsigned long, nr_pages, |
1166 | ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1167 | for (i = 0; i < ret; i++) { |
1168 | if (pages[i] != locked_page) | |
1169 | unlock_page(pages[i]); | |
1170 | page_cache_release(pages[i]); | |
1171 | } | |
1172 | nr_pages -= ret; | |
1173 | index += ret; | |
1174 | cond_resched(); | |
1175 | } | |
1176 | return 0; | |
1177 | } | |
1178 | ||
1179 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1180 | struct page *locked_page, | |
1181 | u64 delalloc_start, | |
1182 | u64 delalloc_end) | |
1183 | { | |
1184 | unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT; | |
1185 | unsigned long start_index = index; | |
1186 | unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT; | |
1187 | unsigned long pages_locked = 0; | |
1188 | struct page *pages[16]; | |
1189 | unsigned long nrpages; | |
1190 | int ret; | |
1191 | int i; | |
1192 | ||
1193 | /* the caller is responsible for locking the start index */ | |
1194 | if (index == locked_page->index && index == end_index) | |
1195 | return 0; | |
1196 | ||
1197 | /* skip the page at the start index */ | |
1198 | nrpages = end_index - index + 1; | |
d397712b | 1199 | while (nrpages > 0) { |
c8b97818 | 1200 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1201 | min_t(unsigned long, |
1202 | nrpages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1203 | if (ret == 0) { |
1204 | ret = -EAGAIN; | |
1205 | goto done; | |
1206 | } | |
1207 | /* now we have an array of pages, lock them all */ | |
1208 | for (i = 0; i < ret; i++) { | |
1209 | /* | |
1210 | * the caller is taking responsibility for | |
1211 | * locked_page | |
1212 | */ | |
771ed689 | 1213 | if (pages[i] != locked_page) { |
c8b97818 | 1214 | lock_page(pages[i]); |
f2b1c41c CM |
1215 | if (!PageDirty(pages[i]) || |
1216 | pages[i]->mapping != inode->i_mapping) { | |
771ed689 CM |
1217 | ret = -EAGAIN; |
1218 | unlock_page(pages[i]); | |
1219 | page_cache_release(pages[i]); | |
1220 | goto done; | |
1221 | } | |
1222 | } | |
c8b97818 | 1223 | page_cache_release(pages[i]); |
771ed689 | 1224 | pages_locked++; |
c8b97818 | 1225 | } |
c8b97818 CM |
1226 | nrpages -= ret; |
1227 | index += ret; | |
1228 | cond_resched(); | |
1229 | } | |
1230 | ret = 0; | |
1231 | done: | |
1232 | if (ret && pages_locked) { | |
1233 | __unlock_for_delalloc(inode, locked_page, | |
1234 | delalloc_start, | |
1235 | ((u64)(start_index + pages_locked - 1)) << | |
1236 | PAGE_CACHE_SHIFT); | |
1237 | } | |
1238 | return ret; | |
1239 | } | |
1240 | ||
1241 | /* | |
1242 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1243 | * more than 'max_bytes'. start and end are used to return the range, | |
1244 | * | |
1245 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1246 | */ | |
1247 | static noinline u64 find_lock_delalloc_range(struct inode *inode, | |
1248 | struct extent_io_tree *tree, | |
1249 | struct page *locked_page, | |
1250 | u64 *start, u64 *end, | |
1251 | u64 max_bytes) | |
1252 | { | |
1253 | u64 delalloc_start; | |
1254 | u64 delalloc_end; | |
1255 | u64 found; | |
9655d298 | 1256 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1257 | int ret; |
1258 | int loops = 0; | |
1259 | ||
1260 | again: | |
1261 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1262 | delalloc_start = *start; | |
1263 | delalloc_end = 0; | |
1264 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1265 | max_bytes, &cached_state); |
70b99e69 | 1266 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1267 | *start = delalloc_start; |
1268 | *end = delalloc_end; | |
c2a128d2 | 1269 | free_extent_state(cached_state); |
c8b97818 CM |
1270 | return found; |
1271 | } | |
1272 | ||
70b99e69 CM |
1273 | /* |
1274 | * start comes from the offset of locked_page. We have to lock | |
1275 | * pages in order, so we can't process delalloc bytes before | |
1276 | * locked_page | |
1277 | */ | |
d397712b | 1278 | if (delalloc_start < *start) |
70b99e69 | 1279 | delalloc_start = *start; |
70b99e69 | 1280 | |
c8b97818 CM |
1281 | /* |
1282 | * make sure to limit the number of pages we try to lock down | |
1283 | * if we're looping. | |
1284 | */ | |
d397712b | 1285 | if (delalloc_end + 1 - delalloc_start > max_bytes && loops) |
771ed689 | 1286 | delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1; |
d397712b | 1287 | |
c8b97818 CM |
1288 | /* step two, lock all the pages after the page that has start */ |
1289 | ret = lock_delalloc_pages(inode, locked_page, | |
1290 | delalloc_start, delalloc_end); | |
1291 | if (ret == -EAGAIN) { | |
1292 | /* some of the pages are gone, lets avoid looping by | |
1293 | * shortening the size of the delalloc range we're searching | |
1294 | */ | |
9655d298 | 1295 | free_extent_state(cached_state); |
c8b97818 CM |
1296 | if (!loops) { |
1297 | unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1); | |
1298 | max_bytes = PAGE_CACHE_SIZE - offset; | |
1299 | loops = 1; | |
1300 | goto again; | |
1301 | } else { | |
1302 | found = 0; | |
1303 | goto out_failed; | |
1304 | } | |
1305 | } | |
1306 | BUG_ON(ret); | |
1307 | ||
1308 | /* step three, lock the state bits for the whole range */ | |
9655d298 CM |
1309 | lock_extent_bits(tree, delalloc_start, delalloc_end, |
1310 | 0, &cached_state, GFP_NOFS); | |
c8b97818 CM |
1311 | |
1312 | /* then test to make sure it is all still delalloc */ | |
1313 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1314 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1315 | if (!ret) { |
9655d298 CM |
1316 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
1317 | &cached_state, GFP_NOFS); | |
c8b97818 CM |
1318 | __unlock_for_delalloc(inode, locked_page, |
1319 | delalloc_start, delalloc_end); | |
1320 | cond_resched(); | |
1321 | goto again; | |
1322 | } | |
9655d298 | 1323 | free_extent_state(cached_state); |
c8b97818 CM |
1324 | *start = delalloc_start; |
1325 | *end = delalloc_end; | |
1326 | out_failed: | |
1327 | return found; | |
1328 | } | |
1329 | ||
1330 | int extent_clear_unlock_delalloc(struct inode *inode, | |
1331 | struct extent_io_tree *tree, | |
1332 | u64 start, u64 end, struct page *locked_page, | |
a791e35e | 1333 | unsigned long op) |
c8b97818 CM |
1334 | { |
1335 | int ret; | |
1336 | struct page *pages[16]; | |
1337 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1338 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1339 | unsigned long nr_pages = end_index - index + 1; | |
1340 | int i; | |
771ed689 | 1341 | int clear_bits = 0; |
c8b97818 | 1342 | |
a791e35e | 1343 | if (op & EXTENT_CLEAR_UNLOCK) |
771ed689 | 1344 | clear_bits |= EXTENT_LOCKED; |
a791e35e | 1345 | if (op & EXTENT_CLEAR_DIRTY) |
c8b97818 CM |
1346 | clear_bits |= EXTENT_DIRTY; |
1347 | ||
a791e35e | 1348 | if (op & EXTENT_CLEAR_DELALLOC) |
771ed689 CM |
1349 | clear_bits |= EXTENT_DELALLOC; |
1350 | ||
2c64c53d | 1351 | clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); |
32c00aff JB |
1352 | if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | |
1353 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK | | |
1354 | EXTENT_SET_PRIVATE2))) | |
771ed689 | 1355 | return 0; |
c8b97818 | 1356 | |
d397712b | 1357 | while (nr_pages > 0) { |
c8b97818 | 1358 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1359 | min_t(unsigned long, |
1360 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 | 1361 | for (i = 0; i < ret; i++) { |
8b62b72b | 1362 | |
a791e35e | 1363 | if (op & EXTENT_SET_PRIVATE2) |
8b62b72b CM |
1364 | SetPagePrivate2(pages[i]); |
1365 | ||
c8b97818 CM |
1366 | if (pages[i] == locked_page) { |
1367 | page_cache_release(pages[i]); | |
1368 | continue; | |
1369 | } | |
a791e35e | 1370 | if (op & EXTENT_CLEAR_DIRTY) |
c8b97818 | 1371 | clear_page_dirty_for_io(pages[i]); |
a791e35e | 1372 | if (op & EXTENT_SET_WRITEBACK) |
c8b97818 | 1373 | set_page_writeback(pages[i]); |
a791e35e | 1374 | if (op & EXTENT_END_WRITEBACK) |
c8b97818 | 1375 | end_page_writeback(pages[i]); |
a791e35e | 1376 | if (op & EXTENT_CLEAR_UNLOCK_PAGE) |
771ed689 | 1377 | unlock_page(pages[i]); |
c8b97818 CM |
1378 | page_cache_release(pages[i]); |
1379 | } | |
1380 | nr_pages -= ret; | |
1381 | index += ret; | |
1382 | cond_resched(); | |
1383 | } | |
1384 | return 0; | |
1385 | } | |
c8b97818 | 1386 | |
d352ac68 CM |
1387 | /* |
1388 | * count the number of bytes in the tree that have a given bit(s) | |
1389 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1390 | * cached. The total number found is returned. | |
1391 | */ | |
d1310b2e CM |
1392 | u64 count_range_bits(struct extent_io_tree *tree, |
1393 | u64 *start, u64 search_end, u64 max_bytes, | |
ec29ed5b | 1394 | unsigned long bits, int contig) |
d1310b2e CM |
1395 | { |
1396 | struct rb_node *node; | |
1397 | struct extent_state *state; | |
1398 | u64 cur_start = *start; | |
1399 | u64 total_bytes = 0; | |
ec29ed5b | 1400 | u64 last = 0; |
d1310b2e CM |
1401 | int found = 0; |
1402 | ||
1403 | if (search_end <= cur_start) { | |
d1310b2e CM |
1404 | WARN_ON(1); |
1405 | return 0; | |
1406 | } | |
1407 | ||
cad321ad | 1408 | spin_lock(&tree->lock); |
d1310b2e CM |
1409 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1410 | total_bytes = tree->dirty_bytes; | |
1411 | goto out; | |
1412 | } | |
1413 | /* | |
1414 | * this search will find all the extents that end after | |
1415 | * our range starts. | |
1416 | */ | |
80ea96b1 | 1417 | node = tree_search(tree, cur_start); |
d397712b | 1418 | if (!node) |
d1310b2e | 1419 | goto out; |
d1310b2e | 1420 | |
d397712b | 1421 | while (1) { |
d1310b2e CM |
1422 | state = rb_entry(node, struct extent_state, rb_node); |
1423 | if (state->start > search_end) | |
1424 | break; | |
ec29ed5b CM |
1425 | if (contig && found && state->start > last + 1) |
1426 | break; | |
1427 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1428 | total_bytes += min(search_end, state->end) + 1 - |
1429 | max(cur_start, state->start); | |
1430 | if (total_bytes >= max_bytes) | |
1431 | break; | |
1432 | if (!found) { | |
af60bed2 | 1433 | *start = max(cur_start, state->start); |
d1310b2e CM |
1434 | found = 1; |
1435 | } | |
ec29ed5b CM |
1436 | last = state->end; |
1437 | } else if (contig && found) { | |
1438 | break; | |
d1310b2e CM |
1439 | } |
1440 | node = rb_next(node); | |
1441 | if (!node) | |
1442 | break; | |
1443 | } | |
1444 | out: | |
cad321ad | 1445 | spin_unlock(&tree->lock); |
d1310b2e CM |
1446 | return total_bytes; |
1447 | } | |
b2950863 | 1448 | |
d352ac68 CM |
1449 | /* |
1450 | * set the private field for a given byte offset in the tree. If there isn't | |
1451 | * an extent_state there already, this does nothing. | |
1452 | */ | |
d1310b2e CM |
1453 | int set_state_private(struct extent_io_tree *tree, u64 start, u64 private) |
1454 | { | |
1455 | struct rb_node *node; | |
1456 | struct extent_state *state; | |
1457 | int ret = 0; | |
1458 | ||
cad321ad | 1459 | spin_lock(&tree->lock); |
d1310b2e CM |
1460 | /* |
1461 | * this search will find all the extents that end after | |
1462 | * our range starts. | |
1463 | */ | |
80ea96b1 | 1464 | node = tree_search(tree, start); |
2b114d1d | 1465 | if (!node) { |
d1310b2e CM |
1466 | ret = -ENOENT; |
1467 | goto out; | |
1468 | } | |
1469 | state = rb_entry(node, struct extent_state, rb_node); | |
1470 | if (state->start != start) { | |
1471 | ret = -ENOENT; | |
1472 | goto out; | |
1473 | } | |
1474 | state->private = private; | |
1475 | out: | |
cad321ad | 1476 | spin_unlock(&tree->lock); |
d1310b2e CM |
1477 | return ret; |
1478 | } | |
1479 | ||
1480 | int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private) | |
1481 | { | |
1482 | struct rb_node *node; | |
1483 | struct extent_state *state; | |
1484 | int ret = 0; | |
1485 | ||
cad321ad | 1486 | spin_lock(&tree->lock); |
d1310b2e CM |
1487 | /* |
1488 | * this search will find all the extents that end after | |
1489 | * our range starts. | |
1490 | */ | |
80ea96b1 | 1491 | node = tree_search(tree, start); |
2b114d1d | 1492 | if (!node) { |
d1310b2e CM |
1493 | ret = -ENOENT; |
1494 | goto out; | |
1495 | } | |
1496 | state = rb_entry(node, struct extent_state, rb_node); | |
1497 | if (state->start != start) { | |
1498 | ret = -ENOENT; | |
1499 | goto out; | |
1500 | } | |
1501 | *private = state->private; | |
1502 | out: | |
cad321ad | 1503 | spin_unlock(&tree->lock); |
d1310b2e CM |
1504 | return ret; |
1505 | } | |
1506 | ||
1507 | /* | |
1508 | * searches a range in the state tree for a given mask. | |
70dec807 | 1509 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1510 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1511 | * range is found set. | |
1512 | */ | |
1513 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9655d298 | 1514 | int bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1515 | { |
1516 | struct extent_state *state = NULL; | |
1517 | struct rb_node *node; | |
1518 | int bitset = 0; | |
d1310b2e | 1519 | |
cad321ad | 1520 | spin_lock(&tree->lock); |
df98b6e2 JB |
1521 | if (cached && cached->tree && cached->start <= start && |
1522 | cached->end > start) | |
9655d298 CM |
1523 | node = &cached->rb_node; |
1524 | else | |
1525 | node = tree_search(tree, start); | |
d1310b2e CM |
1526 | while (node && start <= end) { |
1527 | state = rb_entry(node, struct extent_state, rb_node); | |
1528 | ||
1529 | if (filled && state->start > start) { | |
1530 | bitset = 0; | |
1531 | break; | |
1532 | } | |
1533 | ||
1534 | if (state->start > end) | |
1535 | break; | |
1536 | ||
1537 | if (state->state & bits) { | |
1538 | bitset = 1; | |
1539 | if (!filled) | |
1540 | break; | |
1541 | } else if (filled) { | |
1542 | bitset = 0; | |
1543 | break; | |
1544 | } | |
46562cec CM |
1545 | |
1546 | if (state->end == (u64)-1) | |
1547 | break; | |
1548 | ||
d1310b2e CM |
1549 | start = state->end + 1; |
1550 | if (start > end) | |
1551 | break; | |
1552 | node = rb_next(node); | |
1553 | if (!node) { | |
1554 | if (filled) | |
1555 | bitset = 0; | |
1556 | break; | |
1557 | } | |
1558 | } | |
cad321ad | 1559 | spin_unlock(&tree->lock); |
d1310b2e CM |
1560 | return bitset; |
1561 | } | |
d1310b2e CM |
1562 | |
1563 | /* | |
1564 | * helper function to set a given page up to date if all the | |
1565 | * extents in the tree for that page are up to date | |
1566 | */ | |
1567 | static int check_page_uptodate(struct extent_io_tree *tree, | |
1568 | struct page *page) | |
1569 | { | |
1570 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1571 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
9655d298 | 1572 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e CM |
1573 | SetPageUptodate(page); |
1574 | return 0; | |
1575 | } | |
1576 | ||
1577 | /* | |
1578 | * helper function to unlock a page if all the extents in the tree | |
1579 | * for that page are unlocked | |
1580 | */ | |
1581 | static int check_page_locked(struct extent_io_tree *tree, | |
1582 | struct page *page) | |
1583 | { | |
1584 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1585 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
9655d298 | 1586 | if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) |
d1310b2e CM |
1587 | unlock_page(page); |
1588 | return 0; | |
1589 | } | |
1590 | ||
1591 | /* | |
1592 | * helper function to end page writeback if all the extents | |
1593 | * in the tree for that page are done with writeback | |
1594 | */ | |
1595 | static int check_page_writeback(struct extent_io_tree *tree, | |
1596 | struct page *page) | |
1597 | { | |
1edbb734 | 1598 | end_page_writeback(page); |
d1310b2e CM |
1599 | return 0; |
1600 | } | |
1601 | ||
1602 | /* lots and lots of room for performance fixes in the end_bio funcs */ | |
1603 | ||
1604 | /* | |
1605 | * after a writepage IO is done, we need to: | |
1606 | * clear the uptodate bits on error | |
1607 | * clear the writeback bits in the extent tree for this IO | |
1608 | * end_page_writeback if the page has no more pending IO | |
1609 | * | |
1610 | * Scheduling is not allowed, so the extent state tree is expected | |
1611 | * to have one and only one object corresponding to this IO. | |
1612 | */ | |
d1310b2e | 1613 | static void end_bio_extent_writepage(struct bio *bio, int err) |
d1310b2e | 1614 | { |
1259ab75 | 1615 | int uptodate = err == 0; |
d1310b2e | 1616 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
902b22f3 | 1617 | struct extent_io_tree *tree; |
d1310b2e CM |
1618 | u64 start; |
1619 | u64 end; | |
1620 | int whole_page; | |
1259ab75 | 1621 | int ret; |
d1310b2e | 1622 | |
d1310b2e CM |
1623 | do { |
1624 | struct page *page = bvec->bv_page; | |
902b22f3 DW |
1625 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
1626 | ||
d1310b2e CM |
1627 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
1628 | bvec->bv_offset; | |
1629 | end = start + bvec->bv_len - 1; | |
1630 | ||
1631 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
1632 | whole_page = 1; | |
1633 | else | |
1634 | whole_page = 0; | |
1635 | ||
1636 | if (--bvec >= bio->bi_io_vec) | |
1637 | prefetchw(&bvec->bv_page->flags); | |
1259ab75 CM |
1638 | if (tree->ops && tree->ops->writepage_end_io_hook) { |
1639 | ret = tree->ops->writepage_end_io_hook(page, start, | |
902b22f3 | 1640 | end, NULL, uptodate); |
1259ab75 CM |
1641 | if (ret) |
1642 | uptodate = 0; | |
1643 | } | |
1644 | ||
1645 | if (!uptodate && tree->ops && | |
1646 | tree->ops->writepage_io_failed_hook) { | |
1647 | ret = tree->ops->writepage_io_failed_hook(bio, page, | |
902b22f3 | 1648 | start, end, NULL); |
1259ab75 | 1649 | if (ret == 0) { |
1259ab75 CM |
1650 | uptodate = (err == 0); |
1651 | continue; | |
1652 | } | |
1653 | } | |
1654 | ||
d1310b2e | 1655 | if (!uptodate) { |
2ac55d41 | 1656 | clear_extent_uptodate(tree, start, end, NULL, GFP_NOFS); |
d1310b2e CM |
1657 | ClearPageUptodate(page); |
1658 | SetPageError(page); | |
1659 | } | |
70dec807 | 1660 | |
d1310b2e CM |
1661 | if (whole_page) |
1662 | end_page_writeback(page); | |
1663 | else | |
1664 | check_page_writeback(tree, page); | |
d1310b2e | 1665 | } while (bvec >= bio->bi_io_vec); |
2b1f55b0 | 1666 | |
d1310b2e | 1667 | bio_put(bio); |
d1310b2e CM |
1668 | } |
1669 | ||
1670 | /* | |
1671 | * after a readpage IO is done, we need to: | |
1672 | * clear the uptodate bits on error | |
1673 | * set the uptodate bits if things worked | |
1674 | * set the page up to date if all extents in the tree are uptodate | |
1675 | * clear the lock bit in the extent tree | |
1676 | * unlock the page if there are no other extents locked for it | |
1677 | * | |
1678 | * Scheduling is not allowed, so the extent state tree is expected | |
1679 | * to have one and only one object corresponding to this IO. | |
1680 | */ | |
d1310b2e | 1681 | static void end_bio_extent_readpage(struct bio *bio, int err) |
d1310b2e CM |
1682 | { |
1683 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
4125bf76 CM |
1684 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
1685 | struct bio_vec *bvec = bio->bi_io_vec; | |
902b22f3 | 1686 | struct extent_io_tree *tree; |
d1310b2e CM |
1687 | u64 start; |
1688 | u64 end; | |
1689 | int whole_page; | |
1690 | int ret; | |
1691 | ||
d20f7043 CM |
1692 | if (err) |
1693 | uptodate = 0; | |
1694 | ||
d1310b2e CM |
1695 | do { |
1696 | struct page *page = bvec->bv_page; | |
507903b8 AJ |
1697 | struct extent_state *cached = NULL; |
1698 | struct extent_state *state; | |
1699 | ||
902b22f3 DW |
1700 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
1701 | ||
d1310b2e CM |
1702 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
1703 | bvec->bv_offset; | |
1704 | end = start + bvec->bv_len - 1; | |
1705 | ||
1706 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
1707 | whole_page = 1; | |
1708 | else | |
1709 | whole_page = 0; | |
1710 | ||
4125bf76 | 1711 | if (++bvec <= bvec_end) |
d1310b2e CM |
1712 | prefetchw(&bvec->bv_page->flags); |
1713 | ||
507903b8 | 1714 | spin_lock(&tree->lock); |
0d399205 | 1715 | state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED); |
109b36a2 | 1716 | if (state && state->start == start) { |
507903b8 AJ |
1717 | /* |
1718 | * take a reference on the state, unlock will drop | |
1719 | * the ref | |
1720 | */ | |
1721 | cache_state(state, &cached); | |
1722 | } | |
1723 | spin_unlock(&tree->lock); | |
1724 | ||
d1310b2e | 1725 | if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) { |
70dec807 | 1726 | ret = tree->ops->readpage_end_io_hook(page, start, end, |
507903b8 | 1727 | state); |
d1310b2e CM |
1728 | if (ret) |
1729 | uptodate = 0; | |
1730 | } | |
7e38326f CM |
1731 | if (!uptodate && tree->ops && |
1732 | tree->ops->readpage_io_failed_hook) { | |
1733 | ret = tree->ops->readpage_io_failed_hook(bio, page, | |
902b22f3 | 1734 | start, end, NULL); |
7e38326f | 1735 | if (ret == 0) { |
3b951516 CM |
1736 | uptodate = |
1737 | test_bit(BIO_UPTODATE, &bio->bi_flags); | |
d20f7043 CM |
1738 | if (err) |
1739 | uptodate = 0; | |
507903b8 | 1740 | uncache_state(&cached); |
7e38326f CM |
1741 | continue; |
1742 | } | |
1743 | } | |
d1310b2e | 1744 | |
771ed689 | 1745 | if (uptodate) { |
507903b8 | 1746 | set_extent_uptodate(tree, start, end, &cached, |
902b22f3 | 1747 | GFP_ATOMIC); |
771ed689 | 1748 | } |
507903b8 | 1749 | unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC); |
d1310b2e | 1750 | |
70dec807 CM |
1751 | if (whole_page) { |
1752 | if (uptodate) { | |
1753 | SetPageUptodate(page); | |
1754 | } else { | |
1755 | ClearPageUptodate(page); | |
1756 | SetPageError(page); | |
1757 | } | |
d1310b2e | 1758 | unlock_page(page); |
70dec807 CM |
1759 | } else { |
1760 | if (uptodate) { | |
1761 | check_page_uptodate(tree, page); | |
1762 | } else { | |
1763 | ClearPageUptodate(page); | |
1764 | SetPageError(page); | |
1765 | } | |
d1310b2e | 1766 | check_page_locked(tree, page); |
70dec807 | 1767 | } |
4125bf76 | 1768 | } while (bvec <= bvec_end); |
d1310b2e CM |
1769 | |
1770 | bio_put(bio); | |
d1310b2e CM |
1771 | } |
1772 | ||
88f794ed MX |
1773 | struct bio * |
1774 | btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
1775 | gfp_t gfp_flags) | |
d1310b2e CM |
1776 | { |
1777 | struct bio *bio; | |
1778 | ||
1779 | bio = bio_alloc(gfp_flags, nr_vecs); | |
1780 | ||
1781 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
1782 | while (!bio && (nr_vecs /= 2)) | |
1783 | bio = bio_alloc(gfp_flags, nr_vecs); | |
1784 | } | |
1785 | ||
1786 | if (bio) { | |
e1c4b745 | 1787 | bio->bi_size = 0; |
d1310b2e CM |
1788 | bio->bi_bdev = bdev; |
1789 | bio->bi_sector = first_sector; | |
1790 | } | |
1791 | return bio; | |
1792 | } | |
1793 | ||
c8b97818 CM |
1794 | static int submit_one_bio(int rw, struct bio *bio, int mirror_num, |
1795 | unsigned long bio_flags) | |
d1310b2e | 1796 | { |
d1310b2e | 1797 | int ret = 0; |
70dec807 CM |
1798 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
1799 | struct page *page = bvec->bv_page; | |
1800 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 1801 | u64 start; |
70dec807 CM |
1802 | |
1803 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset; | |
70dec807 | 1804 | |
902b22f3 | 1805 | bio->bi_private = NULL; |
d1310b2e CM |
1806 | |
1807 | bio_get(bio); | |
1808 | ||
065631f6 | 1809 | if (tree->ops && tree->ops->submit_bio_hook) |
6b82ce8d | 1810 | ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio, |
eaf25d93 | 1811 | mirror_num, bio_flags, start); |
0b86a832 CM |
1812 | else |
1813 | submit_bio(rw, bio); | |
d1310b2e CM |
1814 | if (bio_flagged(bio, BIO_EOPNOTSUPP)) |
1815 | ret = -EOPNOTSUPP; | |
1816 | bio_put(bio); | |
1817 | return ret; | |
1818 | } | |
1819 | ||
1820 | static int submit_extent_page(int rw, struct extent_io_tree *tree, | |
1821 | struct page *page, sector_t sector, | |
1822 | size_t size, unsigned long offset, | |
1823 | struct block_device *bdev, | |
1824 | struct bio **bio_ret, | |
1825 | unsigned long max_pages, | |
f188591e | 1826 | bio_end_io_t end_io_func, |
c8b97818 CM |
1827 | int mirror_num, |
1828 | unsigned long prev_bio_flags, | |
1829 | unsigned long bio_flags) | |
d1310b2e CM |
1830 | { |
1831 | int ret = 0; | |
1832 | struct bio *bio; | |
1833 | int nr; | |
c8b97818 CM |
1834 | int contig = 0; |
1835 | int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED; | |
1836 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; | |
5b050f04 | 1837 | size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE); |
d1310b2e CM |
1838 | |
1839 | if (bio_ret && *bio_ret) { | |
1840 | bio = *bio_ret; | |
c8b97818 CM |
1841 | if (old_compressed) |
1842 | contig = bio->bi_sector == sector; | |
1843 | else | |
1844 | contig = bio->bi_sector + (bio->bi_size >> 9) == | |
1845 | sector; | |
1846 | ||
1847 | if (prev_bio_flags != bio_flags || !contig || | |
239b14b3 | 1848 | (tree->ops && tree->ops->merge_bio_hook && |
c8b97818 CM |
1849 | tree->ops->merge_bio_hook(page, offset, page_size, bio, |
1850 | bio_flags)) || | |
1851 | bio_add_page(bio, page, page_size, offset) < page_size) { | |
1852 | ret = submit_one_bio(rw, bio, mirror_num, | |
1853 | prev_bio_flags); | |
d1310b2e CM |
1854 | bio = NULL; |
1855 | } else { | |
1856 | return 0; | |
1857 | } | |
1858 | } | |
c8b97818 CM |
1859 | if (this_compressed) |
1860 | nr = BIO_MAX_PAGES; | |
1861 | else | |
1862 | nr = bio_get_nr_vecs(bdev); | |
1863 | ||
88f794ed | 1864 | bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH); |
5df67083 TI |
1865 | if (!bio) |
1866 | return -ENOMEM; | |
70dec807 | 1867 | |
c8b97818 | 1868 | bio_add_page(bio, page, page_size, offset); |
d1310b2e CM |
1869 | bio->bi_end_io = end_io_func; |
1870 | bio->bi_private = tree; | |
70dec807 | 1871 | |
d397712b | 1872 | if (bio_ret) |
d1310b2e | 1873 | *bio_ret = bio; |
d397712b | 1874 | else |
c8b97818 | 1875 | ret = submit_one_bio(rw, bio, mirror_num, bio_flags); |
d1310b2e CM |
1876 | |
1877 | return ret; | |
1878 | } | |
1879 | ||
1880 | void set_page_extent_mapped(struct page *page) | |
1881 | { | |
1882 | if (!PagePrivate(page)) { | |
1883 | SetPagePrivate(page); | |
d1310b2e | 1884 | page_cache_get(page); |
6af118ce | 1885 | set_page_private(page, EXTENT_PAGE_PRIVATE); |
d1310b2e CM |
1886 | } |
1887 | } | |
1888 | ||
b2950863 | 1889 | static void set_page_extent_head(struct page *page, unsigned long len) |
d1310b2e | 1890 | { |
eb14ab8e | 1891 | WARN_ON(!PagePrivate(page)); |
d1310b2e CM |
1892 | set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2); |
1893 | } | |
1894 | ||
1895 | /* | |
1896 | * basic readpage implementation. Locked extent state structs are inserted | |
1897 | * into the tree that are removed when the IO is done (by the end_io | |
1898 | * handlers) | |
1899 | */ | |
1900 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
1901 | struct page *page, | |
1902 | get_extent_t *get_extent, | |
c8b97818 CM |
1903 | struct bio **bio, int mirror_num, |
1904 | unsigned long *bio_flags) | |
d1310b2e CM |
1905 | { |
1906 | struct inode *inode = page->mapping->host; | |
1907 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1908 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
1909 | u64 end; | |
1910 | u64 cur = start; | |
1911 | u64 extent_offset; | |
1912 | u64 last_byte = i_size_read(inode); | |
1913 | u64 block_start; | |
1914 | u64 cur_end; | |
1915 | sector_t sector; | |
1916 | struct extent_map *em; | |
1917 | struct block_device *bdev; | |
11c65dcc | 1918 | struct btrfs_ordered_extent *ordered; |
d1310b2e CM |
1919 | int ret; |
1920 | int nr = 0; | |
306e16ce | 1921 | size_t pg_offset = 0; |
d1310b2e | 1922 | size_t iosize; |
c8b97818 | 1923 | size_t disk_io_size; |
d1310b2e | 1924 | size_t blocksize = inode->i_sb->s_blocksize; |
c8b97818 | 1925 | unsigned long this_bio_flag = 0; |
d1310b2e CM |
1926 | |
1927 | set_page_extent_mapped(page); | |
1928 | ||
90a887c9 DM |
1929 | if (!PageUptodate(page)) { |
1930 | if (cleancache_get_page(page) == 0) { | |
1931 | BUG_ON(blocksize != PAGE_SIZE); | |
1932 | goto out; | |
1933 | } | |
1934 | } | |
1935 | ||
d1310b2e | 1936 | end = page_end; |
11c65dcc JB |
1937 | while (1) { |
1938 | lock_extent(tree, start, end, GFP_NOFS); | |
1939 | ordered = btrfs_lookup_ordered_extent(inode, start); | |
1940 | if (!ordered) | |
1941 | break; | |
1942 | unlock_extent(tree, start, end, GFP_NOFS); | |
1943 | btrfs_start_ordered_extent(inode, ordered, 1); | |
1944 | btrfs_put_ordered_extent(ordered); | |
1945 | } | |
d1310b2e | 1946 | |
c8b97818 CM |
1947 | if (page->index == last_byte >> PAGE_CACHE_SHIFT) { |
1948 | char *userpage; | |
1949 | size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1); | |
1950 | ||
1951 | if (zero_offset) { | |
1952 | iosize = PAGE_CACHE_SIZE - zero_offset; | |
1953 | userpage = kmap_atomic(page, KM_USER0); | |
1954 | memset(userpage + zero_offset, 0, iosize); | |
1955 | flush_dcache_page(page); | |
1956 | kunmap_atomic(userpage, KM_USER0); | |
1957 | } | |
1958 | } | |
d1310b2e CM |
1959 | while (cur <= end) { |
1960 | if (cur >= last_byte) { | |
1961 | char *userpage; | |
507903b8 AJ |
1962 | struct extent_state *cached = NULL; |
1963 | ||
306e16ce | 1964 | iosize = PAGE_CACHE_SIZE - pg_offset; |
d1310b2e | 1965 | userpage = kmap_atomic(page, KM_USER0); |
306e16ce | 1966 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e CM |
1967 | flush_dcache_page(page); |
1968 | kunmap_atomic(userpage, KM_USER0); | |
1969 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 AJ |
1970 | &cached, GFP_NOFS); |
1971 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
1972 | &cached, GFP_NOFS); | |
d1310b2e CM |
1973 | break; |
1974 | } | |
306e16ce | 1975 | em = get_extent(inode, page, pg_offset, cur, |
d1310b2e | 1976 | end - cur + 1, 0); |
c704005d | 1977 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e CM |
1978 | SetPageError(page); |
1979 | unlock_extent(tree, cur, end, GFP_NOFS); | |
1980 | break; | |
1981 | } | |
d1310b2e CM |
1982 | extent_offset = cur - em->start; |
1983 | BUG_ON(extent_map_end(em) <= cur); | |
1984 | BUG_ON(end < cur); | |
1985 | ||
261507a0 | 1986 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
c8b97818 | 1987 | this_bio_flag = EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
1988 | extent_set_compress_type(&this_bio_flag, |
1989 | em->compress_type); | |
1990 | } | |
c8b97818 | 1991 | |
d1310b2e CM |
1992 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
1993 | cur_end = min(extent_map_end(em) - 1, end); | |
1994 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
c8b97818 CM |
1995 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
1996 | disk_io_size = em->block_len; | |
1997 | sector = em->block_start >> 9; | |
1998 | } else { | |
1999 | sector = (em->block_start + extent_offset) >> 9; | |
2000 | disk_io_size = iosize; | |
2001 | } | |
d1310b2e CM |
2002 | bdev = em->bdev; |
2003 | block_start = em->block_start; | |
d899e052 YZ |
2004 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2005 | block_start = EXTENT_MAP_HOLE; | |
d1310b2e CM |
2006 | free_extent_map(em); |
2007 | em = NULL; | |
2008 | ||
2009 | /* we've found a hole, just zero and go on */ | |
2010 | if (block_start == EXTENT_MAP_HOLE) { | |
2011 | char *userpage; | |
507903b8 AJ |
2012 | struct extent_state *cached = NULL; |
2013 | ||
d1310b2e | 2014 | userpage = kmap_atomic(page, KM_USER0); |
306e16ce | 2015 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e CM |
2016 | flush_dcache_page(page); |
2017 | kunmap_atomic(userpage, KM_USER0); | |
2018 | ||
2019 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 AJ |
2020 | &cached, GFP_NOFS); |
2021 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
2022 | &cached, GFP_NOFS); | |
d1310b2e | 2023 | cur = cur + iosize; |
306e16ce | 2024 | pg_offset += iosize; |
d1310b2e CM |
2025 | continue; |
2026 | } | |
2027 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
2028 | if (test_range_bit(tree, cur, cur_end, |
2029 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 2030 | check_page_uptodate(tree, page); |
d1310b2e CM |
2031 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); |
2032 | cur = cur + iosize; | |
306e16ce | 2033 | pg_offset += iosize; |
d1310b2e CM |
2034 | continue; |
2035 | } | |
70dec807 CM |
2036 | /* we have an inline extent but it didn't get marked up |
2037 | * to date. Error out | |
2038 | */ | |
2039 | if (block_start == EXTENT_MAP_INLINE) { | |
2040 | SetPageError(page); | |
2041 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
2042 | cur = cur + iosize; | |
306e16ce | 2043 | pg_offset += iosize; |
70dec807 CM |
2044 | continue; |
2045 | } | |
d1310b2e CM |
2046 | |
2047 | ret = 0; | |
2048 | if (tree->ops && tree->ops->readpage_io_hook) { | |
2049 | ret = tree->ops->readpage_io_hook(page, cur, | |
2050 | cur + iosize - 1); | |
2051 | } | |
2052 | if (!ret) { | |
89642229 CM |
2053 | unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1; |
2054 | pnr -= page->index; | |
d1310b2e | 2055 | ret = submit_extent_page(READ, tree, page, |
306e16ce | 2056 | sector, disk_io_size, pg_offset, |
89642229 | 2057 | bdev, bio, pnr, |
c8b97818 CM |
2058 | end_bio_extent_readpage, mirror_num, |
2059 | *bio_flags, | |
2060 | this_bio_flag); | |
89642229 | 2061 | nr++; |
c8b97818 | 2062 | *bio_flags = this_bio_flag; |
d1310b2e CM |
2063 | } |
2064 | if (ret) | |
2065 | SetPageError(page); | |
2066 | cur = cur + iosize; | |
306e16ce | 2067 | pg_offset += iosize; |
d1310b2e | 2068 | } |
90a887c9 | 2069 | out: |
d1310b2e CM |
2070 | if (!nr) { |
2071 | if (!PageError(page)) | |
2072 | SetPageUptodate(page); | |
2073 | unlock_page(page); | |
2074 | } | |
2075 | return 0; | |
2076 | } | |
2077 | ||
2078 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, | |
2079 | get_extent_t *get_extent) | |
2080 | { | |
2081 | struct bio *bio = NULL; | |
c8b97818 | 2082 | unsigned long bio_flags = 0; |
d1310b2e CM |
2083 | int ret; |
2084 | ||
c8b97818 CM |
2085 | ret = __extent_read_full_page(tree, page, get_extent, &bio, 0, |
2086 | &bio_flags); | |
d1310b2e | 2087 | if (bio) |
6b82ce8d | 2088 | ret = submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
2089 | return ret; |
2090 | } | |
d1310b2e | 2091 | |
11c8349b CM |
2092 | static noinline void update_nr_written(struct page *page, |
2093 | struct writeback_control *wbc, | |
2094 | unsigned long nr_written) | |
2095 | { | |
2096 | wbc->nr_to_write -= nr_written; | |
2097 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && | |
2098 | wbc->range_start == 0 && wbc->range_end == LLONG_MAX)) | |
2099 | page->mapping->writeback_index = page->index + nr_written; | |
2100 | } | |
2101 | ||
d1310b2e CM |
2102 | /* |
2103 | * the writepage semantics are similar to regular writepage. extent | |
2104 | * records are inserted to lock ranges in the tree, and as dirty areas | |
2105 | * are found, they are marked writeback. Then the lock bits are removed | |
2106 | * and the end_io handler clears the writeback ranges | |
2107 | */ | |
2108 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
2109 | void *data) | |
2110 | { | |
2111 | struct inode *inode = page->mapping->host; | |
2112 | struct extent_page_data *epd = data; | |
2113 | struct extent_io_tree *tree = epd->tree; | |
2114 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2115 | u64 delalloc_start; | |
2116 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2117 | u64 end; | |
2118 | u64 cur = start; | |
2119 | u64 extent_offset; | |
2120 | u64 last_byte = i_size_read(inode); | |
2121 | u64 block_start; | |
2122 | u64 iosize; | |
2123 | sector_t sector; | |
2c64c53d | 2124 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
2125 | struct extent_map *em; |
2126 | struct block_device *bdev; | |
2127 | int ret; | |
2128 | int nr = 0; | |
7f3c74fb | 2129 | size_t pg_offset = 0; |
d1310b2e CM |
2130 | size_t blocksize; |
2131 | loff_t i_size = i_size_read(inode); | |
2132 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
2133 | u64 nr_delalloc; | |
2134 | u64 delalloc_end; | |
c8b97818 CM |
2135 | int page_started; |
2136 | int compressed; | |
ffbd517d | 2137 | int write_flags; |
771ed689 | 2138 | unsigned long nr_written = 0; |
9e487107 | 2139 | bool fill_delalloc = true; |
d1310b2e | 2140 | |
ffbd517d | 2141 | if (wbc->sync_mode == WB_SYNC_ALL) |
721a9602 | 2142 | write_flags = WRITE_SYNC; |
ffbd517d CM |
2143 | else |
2144 | write_flags = WRITE; | |
2145 | ||
1abe9b8a | 2146 | trace___extent_writepage(page, inode, wbc); |
2147 | ||
d1310b2e | 2148 | WARN_ON(!PageLocked(page)); |
7f3c74fb | 2149 | pg_offset = i_size & (PAGE_CACHE_SIZE - 1); |
211c17f5 | 2150 | if (page->index > end_index || |
7f3c74fb | 2151 | (page->index == end_index && !pg_offset)) { |
39be25cd | 2152 | page->mapping->a_ops->invalidatepage(page, 0); |
d1310b2e CM |
2153 | unlock_page(page); |
2154 | return 0; | |
2155 | } | |
2156 | ||
2157 | if (page->index == end_index) { | |
2158 | char *userpage; | |
2159 | ||
d1310b2e | 2160 | userpage = kmap_atomic(page, KM_USER0); |
7f3c74fb CM |
2161 | memset(userpage + pg_offset, 0, |
2162 | PAGE_CACHE_SIZE - pg_offset); | |
d1310b2e | 2163 | kunmap_atomic(userpage, KM_USER0); |
211c17f5 | 2164 | flush_dcache_page(page); |
d1310b2e | 2165 | } |
7f3c74fb | 2166 | pg_offset = 0; |
d1310b2e CM |
2167 | |
2168 | set_page_extent_mapped(page); | |
2169 | ||
9e487107 JB |
2170 | if (!tree->ops || !tree->ops->fill_delalloc) |
2171 | fill_delalloc = false; | |
2172 | ||
d1310b2e CM |
2173 | delalloc_start = start; |
2174 | delalloc_end = 0; | |
c8b97818 | 2175 | page_started = 0; |
9e487107 | 2176 | if (!epd->extent_locked && fill_delalloc) { |
f85d7d6c | 2177 | u64 delalloc_to_write = 0; |
11c8349b CM |
2178 | /* |
2179 | * make sure the wbc mapping index is at least updated | |
2180 | * to this page. | |
2181 | */ | |
2182 | update_nr_written(page, wbc, 0); | |
2183 | ||
d397712b | 2184 | while (delalloc_end < page_end) { |
771ed689 | 2185 | nr_delalloc = find_lock_delalloc_range(inode, tree, |
c8b97818 CM |
2186 | page, |
2187 | &delalloc_start, | |
d1310b2e CM |
2188 | &delalloc_end, |
2189 | 128 * 1024 * 1024); | |
771ed689 CM |
2190 | if (nr_delalloc == 0) { |
2191 | delalloc_start = delalloc_end + 1; | |
2192 | continue; | |
2193 | } | |
2194 | tree->ops->fill_delalloc(inode, page, delalloc_start, | |
2195 | delalloc_end, &page_started, | |
2196 | &nr_written); | |
f85d7d6c CM |
2197 | /* |
2198 | * delalloc_end is already one less than the total | |
2199 | * length, so we don't subtract one from | |
2200 | * PAGE_CACHE_SIZE | |
2201 | */ | |
2202 | delalloc_to_write += (delalloc_end - delalloc_start + | |
2203 | PAGE_CACHE_SIZE) >> | |
2204 | PAGE_CACHE_SHIFT; | |
d1310b2e | 2205 | delalloc_start = delalloc_end + 1; |
d1310b2e | 2206 | } |
f85d7d6c CM |
2207 | if (wbc->nr_to_write < delalloc_to_write) { |
2208 | int thresh = 8192; | |
2209 | ||
2210 | if (delalloc_to_write < thresh * 2) | |
2211 | thresh = delalloc_to_write; | |
2212 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
2213 | thresh); | |
2214 | } | |
c8b97818 | 2215 | |
771ed689 CM |
2216 | /* did the fill delalloc function already unlock and start |
2217 | * the IO? | |
2218 | */ | |
2219 | if (page_started) { | |
2220 | ret = 0; | |
11c8349b CM |
2221 | /* |
2222 | * we've unlocked the page, so we can't update | |
2223 | * the mapping's writeback index, just update | |
2224 | * nr_to_write. | |
2225 | */ | |
2226 | wbc->nr_to_write -= nr_written; | |
2227 | goto done_unlocked; | |
771ed689 | 2228 | } |
c8b97818 | 2229 | } |
247e743c | 2230 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
2231 | ret = tree->ops->writepage_start_hook(page, start, |
2232 | page_end); | |
247e743c | 2233 | if (ret == -EAGAIN) { |
247e743c | 2234 | redirty_page_for_writepage(wbc, page); |
11c8349b | 2235 | update_nr_written(page, wbc, nr_written); |
247e743c | 2236 | unlock_page(page); |
771ed689 | 2237 | ret = 0; |
11c8349b | 2238 | goto done_unlocked; |
247e743c CM |
2239 | } |
2240 | } | |
2241 | ||
11c8349b CM |
2242 | /* |
2243 | * we don't want to touch the inode after unlocking the page, | |
2244 | * so we update the mapping writeback index now | |
2245 | */ | |
2246 | update_nr_written(page, wbc, nr_written + 1); | |
771ed689 | 2247 | |
d1310b2e | 2248 | end = page_end; |
d1310b2e | 2249 | if (last_byte <= start) { |
e6dcd2dc CM |
2250 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2251 | tree->ops->writepage_end_io_hook(page, start, | |
2252 | page_end, NULL, 1); | |
d1310b2e CM |
2253 | goto done; |
2254 | } | |
2255 | ||
d1310b2e CM |
2256 | blocksize = inode->i_sb->s_blocksize; |
2257 | ||
2258 | while (cur <= end) { | |
2259 | if (cur >= last_byte) { | |
e6dcd2dc CM |
2260 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2261 | tree->ops->writepage_end_io_hook(page, cur, | |
2262 | page_end, NULL, 1); | |
d1310b2e CM |
2263 | break; |
2264 | } | |
7f3c74fb | 2265 | em = epd->get_extent(inode, page, pg_offset, cur, |
d1310b2e | 2266 | end - cur + 1, 1); |
c704005d | 2267 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e CM |
2268 | SetPageError(page); |
2269 | break; | |
2270 | } | |
2271 | ||
2272 | extent_offset = cur - em->start; | |
2273 | BUG_ON(extent_map_end(em) <= cur); | |
2274 | BUG_ON(end < cur); | |
2275 | iosize = min(extent_map_end(em) - cur, end - cur + 1); | |
2276 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
2277 | sector = (em->block_start + extent_offset) >> 9; | |
2278 | bdev = em->bdev; | |
2279 | block_start = em->block_start; | |
c8b97818 | 2280 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
2281 | free_extent_map(em); |
2282 | em = NULL; | |
2283 | ||
c8b97818 CM |
2284 | /* |
2285 | * compressed and inline extents are written through other | |
2286 | * paths in the FS | |
2287 | */ | |
2288 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 2289 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
2290 | /* |
2291 | * end_io notification does not happen here for | |
2292 | * compressed extents | |
2293 | */ | |
2294 | if (!compressed && tree->ops && | |
2295 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
2296 | tree->ops->writepage_end_io_hook(page, cur, |
2297 | cur + iosize - 1, | |
2298 | NULL, 1); | |
c8b97818 CM |
2299 | else if (compressed) { |
2300 | /* we don't want to end_page_writeback on | |
2301 | * a compressed extent. this happens | |
2302 | * elsewhere | |
2303 | */ | |
2304 | nr++; | |
2305 | } | |
2306 | ||
2307 | cur += iosize; | |
7f3c74fb | 2308 | pg_offset += iosize; |
d1310b2e CM |
2309 | continue; |
2310 | } | |
d1310b2e CM |
2311 | /* leave this out until we have a page_mkwrite call */ |
2312 | if (0 && !test_range_bit(tree, cur, cur + iosize - 1, | |
9655d298 | 2313 | EXTENT_DIRTY, 0, NULL)) { |
d1310b2e | 2314 | cur = cur + iosize; |
7f3c74fb | 2315 | pg_offset += iosize; |
d1310b2e CM |
2316 | continue; |
2317 | } | |
c8b97818 | 2318 | |
d1310b2e CM |
2319 | if (tree->ops && tree->ops->writepage_io_hook) { |
2320 | ret = tree->ops->writepage_io_hook(page, cur, | |
2321 | cur + iosize - 1); | |
2322 | } else { | |
2323 | ret = 0; | |
2324 | } | |
1259ab75 | 2325 | if (ret) { |
d1310b2e | 2326 | SetPageError(page); |
1259ab75 | 2327 | } else { |
d1310b2e | 2328 | unsigned long max_nr = end_index + 1; |
7f3c74fb | 2329 | |
d1310b2e CM |
2330 | set_range_writeback(tree, cur, cur + iosize - 1); |
2331 | if (!PageWriteback(page)) { | |
d397712b CM |
2332 | printk(KERN_ERR "btrfs warning page %lu not " |
2333 | "writeback, cur %llu end %llu\n", | |
2334 | page->index, (unsigned long long)cur, | |
d1310b2e CM |
2335 | (unsigned long long)end); |
2336 | } | |
2337 | ||
ffbd517d CM |
2338 | ret = submit_extent_page(write_flags, tree, page, |
2339 | sector, iosize, pg_offset, | |
2340 | bdev, &epd->bio, max_nr, | |
c8b97818 CM |
2341 | end_bio_extent_writepage, |
2342 | 0, 0, 0); | |
d1310b2e CM |
2343 | if (ret) |
2344 | SetPageError(page); | |
2345 | } | |
2346 | cur = cur + iosize; | |
7f3c74fb | 2347 | pg_offset += iosize; |
d1310b2e CM |
2348 | nr++; |
2349 | } | |
2350 | done: | |
2351 | if (nr == 0) { | |
2352 | /* make sure the mapping tag for page dirty gets cleared */ | |
2353 | set_page_writeback(page); | |
2354 | end_page_writeback(page); | |
2355 | } | |
d1310b2e | 2356 | unlock_page(page); |
771ed689 | 2357 | |
11c8349b CM |
2358 | done_unlocked: |
2359 | ||
2c64c53d CM |
2360 | /* drop our reference on any cached states */ |
2361 | free_extent_state(cached_state); | |
d1310b2e CM |
2362 | return 0; |
2363 | } | |
2364 | ||
d1310b2e | 2365 | /** |
4bef0848 | 2366 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
2367 | * @mapping: address space structure to write |
2368 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
2369 | * @writepage: function called for each page | |
2370 | * @data: data passed to writepage function | |
2371 | * | |
2372 | * If a page is already under I/O, write_cache_pages() skips it, even | |
2373 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
2374 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
2375 | * and msync() need to guarantee that all the data which was dirty at the time | |
2376 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
2377 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
2378 | * existing IO to complete. | |
2379 | */ | |
b2950863 | 2380 | static int extent_write_cache_pages(struct extent_io_tree *tree, |
4bef0848 CM |
2381 | struct address_space *mapping, |
2382 | struct writeback_control *wbc, | |
d2c3f4f6 CM |
2383 | writepage_t writepage, void *data, |
2384 | void (*flush_fn)(void *)) | |
d1310b2e | 2385 | { |
d1310b2e CM |
2386 | int ret = 0; |
2387 | int done = 0; | |
f85d7d6c | 2388 | int nr_to_write_done = 0; |
d1310b2e CM |
2389 | struct pagevec pvec; |
2390 | int nr_pages; | |
2391 | pgoff_t index; | |
2392 | pgoff_t end; /* Inclusive */ | |
2393 | int scanned = 0; | |
f7aaa06b | 2394 | int tag; |
d1310b2e | 2395 | |
d1310b2e CM |
2396 | pagevec_init(&pvec, 0); |
2397 | if (wbc->range_cyclic) { | |
2398 | index = mapping->writeback_index; /* Start from prev offset */ | |
2399 | end = -1; | |
2400 | } else { | |
2401 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
2402 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
2403 | scanned = 1; |
2404 | } | |
f7aaa06b JB |
2405 | if (wbc->sync_mode == WB_SYNC_ALL) |
2406 | tag = PAGECACHE_TAG_TOWRITE; | |
2407 | else | |
2408 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 2409 | retry: |
f7aaa06b JB |
2410 | if (wbc->sync_mode == WB_SYNC_ALL) |
2411 | tag_pages_for_writeback(mapping, index, end); | |
f85d7d6c | 2412 | while (!done && !nr_to_write_done && (index <= end) && |
f7aaa06b JB |
2413 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
2414 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
d1310b2e CM |
2415 | unsigned i; |
2416 | ||
2417 | scanned = 1; | |
2418 | for (i = 0; i < nr_pages; i++) { | |
2419 | struct page *page = pvec.pages[i]; | |
2420 | ||
2421 | /* | |
2422 | * At this point we hold neither mapping->tree_lock nor | |
2423 | * lock on the page itself: the page may be truncated or | |
2424 | * invalidated (changing page->mapping to NULL), or even | |
2425 | * swizzled back from swapper_space to tmpfs file | |
2426 | * mapping | |
2427 | */ | |
4bef0848 CM |
2428 | if (tree->ops && tree->ops->write_cache_pages_lock_hook) |
2429 | tree->ops->write_cache_pages_lock_hook(page); | |
2430 | else | |
2431 | lock_page(page); | |
d1310b2e CM |
2432 | |
2433 | if (unlikely(page->mapping != mapping)) { | |
2434 | unlock_page(page); | |
2435 | continue; | |
2436 | } | |
2437 | ||
2438 | if (!wbc->range_cyclic && page->index > end) { | |
2439 | done = 1; | |
2440 | unlock_page(page); | |
2441 | continue; | |
2442 | } | |
2443 | ||
d2c3f4f6 | 2444 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 CM |
2445 | if (PageWriteback(page)) |
2446 | flush_fn(data); | |
d1310b2e | 2447 | wait_on_page_writeback(page); |
d2c3f4f6 | 2448 | } |
d1310b2e CM |
2449 | |
2450 | if (PageWriteback(page) || | |
2451 | !clear_page_dirty_for_io(page)) { | |
2452 | unlock_page(page); | |
2453 | continue; | |
2454 | } | |
2455 | ||
2456 | ret = (*writepage)(page, wbc, data); | |
2457 | ||
2458 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
2459 | unlock_page(page); | |
2460 | ret = 0; | |
2461 | } | |
f85d7d6c | 2462 | if (ret) |
d1310b2e | 2463 | done = 1; |
f85d7d6c CM |
2464 | |
2465 | /* | |
2466 | * the filesystem may choose to bump up nr_to_write. | |
2467 | * We have to make sure to honor the new nr_to_write | |
2468 | * at any time | |
2469 | */ | |
2470 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
2471 | } |
2472 | pagevec_release(&pvec); | |
2473 | cond_resched(); | |
2474 | } | |
2475 | if (!scanned && !done) { | |
2476 | /* | |
2477 | * We hit the last page and there is more work to be done: wrap | |
2478 | * back to the start of the file | |
2479 | */ | |
2480 | scanned = 1; | |
2481 | index = 0; | |
2482 | goto retry; | |
2483 | } | |
d1310b2e CM |
2484 | return ret; |
2485 | } | |
d1310b2e | 2486 | |
ffbd517d | 2487 | static void flush_epd_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 2488 | { |
d2c3f4f6 | 2489 | if (epd->bio) { |
ffbd517d CM |
2490 | if (epd->sync_io) |
2491 | submit_one_bio(WRITE_SYNC, epd->bio, 0, 0); | |
2492 | else | |
2493 | submit_one_bio(WRITE, epd->bio, 0, 0); | |
d2c3f4f6 CM |
2494 | epd->bio = NULL; |
2495 | } | |
2496 | } | |
2497 | ||
ffbd517d CM |
2498 | static noinline void flush_write_bio(void *data) |
2499 | { | |
2500 | struct extent_page_data *epd = data; | |
2501 | flush_epd_write_bio(epd); | |
2502 | } | |
2503 | ||
d1310b2e CM |
2504 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, |
2505 | get_extent_t *get_extent, | |
2506 | struct writeback_control *wbc) | |
2507 | { | |
2508 | int ret; | |
d1310b2e CM |
2509 | struct extent_page_data epd = { |
2510 | .bio = NULL, | |
2511 | .tree = tree, | |
2512 | .get_extent = get_extent, | |
771ed689 | 2513 | .extent_locked = 0, |
ffbd517d | 2514 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e | 2515 | }; |
d1310b2e | 2516 | |
d1310b2e CM |
2517 | ret = __extent_writepage(page, wbc, &epd); |
2518 | ||
ffbd517d | 2519 | flush_epd_write_bio(&epd); |
d1310b2e CM |
2520 | return ret; |
2521 | } | |
d1310b2e | 2522 | |
771ed689 CM |
2523 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, |
2524 | u64 start, u64 end, get_extent_t *get_extent, | |
2525 | int mode) | |
2526 | { | |
2527 | int ret = 0; | |
2528 | struct address_space *mapping = inode->i_mapping; | |
2529 | struct page *page; | |
2530 | unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >> | |
2531 | PAGE_CACHE_SHIFT; | |
2532 | ||
2533 | struct extent_page_data epd = { | |
2534 | .bio = NULL, | |
2535 | .tree = tree, | |
2536 | .get_extent = get_extent, | |
2537 | .extent_locked = 1, | |
ffbd517d | 2538 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
2539 | }; |
2540 | struct writeback_control wbc_writepages = { | |
771ed689 | 2541 | .sync_mode = mode, |
771ed689 CM |
2542 | .nr_to_write = nr_pages * 2, |
2543 | .range_start = start, | |
2544 | .range_end = end + 1, | |
2545 | }; | |
2546 | ||
d397712b | 2547 | while (start <= end) { |
771ed689 CM |
2548 | page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); |
2549 | if (clear_page_dirty_for_io(page)) | |
2550 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
2551 | else { | |
2552 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
2553 | tree->ops->writepage_end_io_hook(page, start, | |
2554 | start + PAGE_CACHE_SIZE - 1, | |
2555 | NULL, 1); | |
2556 | unlock_page(page); | |
2557 | } | |
2558 | page_cache_release(page); | |
2559 | start += PAGE_CACHE_SIZE; | |
2560 | } | |
2561 | ||
ffbd517d | 2562 | flush_epd_write_bio(&epd); |
771ed689 CM |
2563 | return ret; |
2564 | } | |
d1310b2e CM |
2565 | |
2566 | int extent_writepages(struct extent_io_tree *tree, | |
2567 | struct address_space *mapping, | |
2568 | get_extent_t *get_extent, | |
2569 | struct writeback_control *wbc) | |
2570 | { | |
2571 | int ret = 0; | |
2572 | struct extent_page_data epd = { | |
2573 | .bio = NULL, | |
2574 | .tree = tree, | |
2575 | .get_extent = get_extent, | |
771ed689 | 2576 | .extent_locked = 0, |
ffbd517d | 2577 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
2578 | }; |
2579 | ||
4bef0848 | 2580 | ret = extent_write_cache_pages(tree, mapping, wbc, |
d2c3f4f6 CM |
2581 | __extent_writepage, &epd, |
2582 | flush_write_bio); | |
ffbd517d | 2583 | flush_epd_write_bio(&epd); |
d1310b2e CM |
2584 | return ret; |
2585 | } | |
d1310b2e CM |
2586 | |
2587 | int extent_readpages(struct extent_io_tree *tree, | |
2588 | struct address_space *mapping, | |
2589 | struct list_head *pages, unsigned nr_pages, | |
2590 | get_extent_t get_extent) | |
2591 | { | |
2592 | struct bio *bio = NULL; | |
2593 | unsigned page_idx; | |
c8b97818 | 2594 | unsigned long bio_flags = 0; |
d1310b2e | 2595 | |
d1310b2e CM |
2596 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
2597 | struct page *page = list_entry(pages->prev, struct page, lru); | |
2598 | ||
2599 | prefetchw(&page->flags); | |
2600 | list_del(&page->lru); | |
28ecb609 | 2601 | if (!add_to_page_cache_lru(page, mapping, |
43e817a1 | 2602 | page->index, GFP_NOFS)) { |
f188591e | 2603 | __extent_read_full_page(tree, page, get_extent, |
c8b97818 | 2604 | &bio, 0, &bio_flags); |
d1310b2e CM |
2605 | } |
2606 | page_cache_release(page); | |
2607 | } | |
d1310b2e CM |
2608 | BUG_ON(!list_empty(pages)); |
2609 | if (bio) | |
c8b97818 | 2610 | submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
2611 | return 0; |
2612 | } | |
d1310b2e CM |
2613 | |
2614 | /* | |
2615 | * basic invalidatepage code, this waits on any locked or writeback | |
2616 | * ranges corresponding to the page, and then deletes any extent state | |
2617 | * records from the tree | |
2618 | */ | |
2619 | int extent_invalidatepage(struct extent_io_tree *tree, | |
2620 | struct page *page, unsigned long offset) | |
2621 | { | |
2ac55d41 | 2622 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
2623 | u64 start = ((u64)page->index << PAGE_CACHE_SHIFT); |
2624 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
2625 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
2626 | ||
d397712b | 2627 | start += (offset + blocksize - 1) & ~(blocksize - 1); |
d1310b2e CM |
2628 | if (start > end) |
2629 | return 0; | |
2630 | ||
2ac55d41 | 2631 | lock_extent_bits(tree, start, end, 0, &cached_state, GFP_NOFS); |
1edbb734 | 2632 | wait_on_page_writeback(page); |
d1310b2e | 2633 | clear_extent_bit(tree, start, end, |
32c00aff JB |
2634 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
2635 | EXTENT_DO_ACCOUNTING, | |
2ac55d41 | 2636 | 1, 1, &cached_state, GFP_NOFS); |
d1310b2e CM |
2637 | return 0; |
2638 | } | |
d1310b2e | 2639 | |
7b13b7b1 CM |
2640 | /* |
2641 | * a helper for releasepage, this tests for areas of the page that | |
2642 | * are locked or under IO and drops the related state bits if it is safe | |
2643 | * to drop the page. | |
2644 | */ | |
2645 | int try_release_extent_state(struct extent_map_tree *map, | |
2646 | struct extent_io_tree *tree, struct page *page, | |
2647 | gfp_t mask) | |
2648 | { | |
2649 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2650 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
2651 | int ret = 1; | |
2652 | ||
211f90e6 | 2653 | if (test_range_bit(tree, start, end, |
8b62b72b | 2654 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
2655 | ret = 0; |
2656 | else { | |
2657 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
2658 | mask = GFP_NOFS; | |
11ef160f CM |
2659 | /* |
2660 | * at this point we can safely clear everything except the | |
2661 | * locked bit and the nodatasum bit | |
2662 | */ | |
e3f24cc5 | 2663 | ret = clear_extent_bit(tree, start, end, |
11ef160f CM |
2664 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
2665 | 0, 0, NULL, mask); | |
e3f24cc5 CM |
2666 | |
2667 | /* if clear_extent_bit failed for enomem reasons, | |
2668 | * we can't allow the release to continue. | |
2669 | */ | |
2670 | if (ret < 0) | |
2671 | ret = 0; | |
2672 | else | |
2673 | ret = 1; | |
7b13b7b1 CM |
2674 | } |
2675 | return ret; | |
2676 | } | |
7b13b7b1 | 2677 | |
d1310b2e CM |
2678 | /* |
2679 | * a helper for releasepage. As long as there are no locked extents | |
2680 | * in the range corresponding to the page, both state records and extent | |
2681 | * map records are removed | |
2682 | */ | |
2683 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
2684 | struct extent_io_tree *tree, struct page *page, |
2685 | gfp_t mask) | |
d1310b2e CM |
2686 | { |
2687 | struct extent_map *em; | |
2688 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2689 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
7b13b7b1 | 2690 | |
70dec807 CM |
2691 | if ((mask & __GFP_WAIT) && |
2692 | page->mapping->host->i_size > 16 * 1024 * 1024) { | |
39b5637f | 2693 | u64 len; |
70dec807 | 2694 | while (start <= end) { |
39b5637f | 2695 | len = end - start + 1; |
890871be | 2696 | write_lock(&map->lock); |
39b5637f | 2697 | em = lookup_extent_mapping(map, start, len); |
c704005d | 2698 | if (IS_ERR_OR_NULL(em)) { |
890871be | 2699 | write_unlock(&map->lock); |
70dec807 CM |
2700 | break; |
2701 | } | |
7f3c74fb CM |
2702 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
2703 | em->start != start) { | |
890871be | 2704 | write_unlock(&map->lock); |
70dec807 CM |
2705 | free_extent_map(em); |
2706 | break; | |
2707 | } | |
2708 | if (!test_range_bit(tree, em->start, | |
2709 | extent_map_end(em) - 1, | |
8b62b72b | 2710 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 2711 | 0, NULL)) { |
70dec807 CM |
2712 | remove_extent_mapping(map, em); |
2713 | /* once for the rb tree */ | |
2714 | free_extent_map(em); | |
2715 | } | |
2716 | start = extent_map_end(em); | |
890871be | 2717 | write_unlock(&map->lock); |
70dec807 CM |
2718 | |
2719 | /* once for us */ | |
d1310b2e CM |
2720 | free_extent_map(em); |
2721 | } | |
d1310b2e | 2722 | } |
7b13b7b1 | 2723 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 2724 | } |
d1310b2e | 2725 | |
ec29ed5b CM |
2726 | /* |
2727 | * helper function for fiemap, which doesn't want to see any holes. | |
2728 | * This maps until we find something past 'last' | |
2729 | */ | |
2730 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
2731 | u64 offset, | |
2732 | u64 last, | |
2733 | get_extent_t *get_extent) | |
2734 | { | |
2735 | u64 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
2736 | struct extent_map *em; | |
2737 | u64 len; | |
2738 | ||
2739 | if (offset >= last) | |
2740 | return NULL; | |
2741 | ||
2742 | while(1) { | |
2743 | len = last - offset; | |
2744 | if (len == 0) | |
2745 | break; | |
2746 | len = (len + sectorsize - 1) & ~(sectorsize - 1); | |
2747 | em = get_extent(inode, NULL, 0, offset, len, 0); | |
c704005d | 2748 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
2749 | return em; |
2750 | ||
2751 | /* if this isn't a hole return it */ | |
2752 | if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) && | |
2753 | em->block_start != EXTENT_MAP_HOLE) { | |
2754 | return em; | |
2755 | } | |
2756 | ||
2757 | /* this is a hole, advance to the next extent */ | |
2758 | offset = extent_map_end(em); | |
2759 | free_extent_map(em); | |
2760 | if (offset >= last) | |
2761 | break; | |
2762 | } | |
2763 | return NULL; | |
2764 | } | |
2765 | ||
1506fcc8 YS |
2766 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
2767 | __u64 start, __u64 len, get_extent_t *get_extent) | |
2768 | { | |
975f84fe | 2769 | int ret = 0; |
1506fcc8 YS |
2770 | u64 off = start; |
2771 | u64 max = start + len; | |
2772 | u32 flags = 0; | |
975f84fe JB |
2773 | u32 found_type; |
2774 | u64 last; | |
ec29ed5b | 2775 | u64 last_for_get_extent = 0; |
1506fcc8 | 2776 | u64 disko = 0; |
ec29ed5b | 2777 | u64 isize = i_size_read(inode); |
975f84fe | 2778 | struct btrfs_key found_key; |
1506fcc8 | 2779 | struct extent_map *em = NULL; |
2ac55d41 | 2780 | struct extent_state *cached_state = NULL; |
975f84fe JB |
2781 | struct btrfs_path *path; |
2782 | struct btrfs_file_extent_item *item; | |
1506fcc8 | 2783 | int end = 0; |
ec29ed5b CM |
2784 | u64 em_start = 0; |
2785 | u64 em_len = 0; | |
2786 | u64 em_end = 0; | |
1506fcc8 | 2787 | unsigned long emflags; |
1506fcc8 YS |
2788 | |
2789 | if (len == 0) | |
2790 | return -EINVAL; | |
2791 | ||
975f84fe JB |
2792 | path = btrfs_alloc_path(); |
2793 | if (!path) | |
2794 | return -ENOMEM; | |
2795 | path->leave_spinning = 1; | |
2796 | ||
ec29ed5b CM |
2797 | /* |
2798 | * lookup the last file extent. We're not using i_size here | |
2799 | * because there might be preallocation past i_size | |
2800 | */ | |
975f84fe | 2801 | ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root, |
33345d01 | 2802 | path, btrfs_ino(inode), -1, 0); |
975f84fe JB |
2803 | if (ret < 0) { |
2804 | btrfs_free_path(path); | |
2805 | return ret; | |
2806 | } | |
2807 | WARN_ON(!ret); | |
2808 | path->slots[0]--; | |
2809 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2810 | struct btrfs_file_extent_item); | |
2811 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
2812 | found_type = btrfs_key_type(&found_key); | |
2813 | ||
ec29ed5b | 2814 | /* No extents, but there might be delalloc bits */ |
33345d01 | 2815 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 2816 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
2817 | /* have to trust i_size as the end */ |
2818 | last = (u64)-1; | |
2819 | last_for_get_extent = isize; | |
2820 | } else { | |
2821 | /* | |
2822 | * remember the start of the last extent. There are a | |
2823 | * bunch of different factors that go into the length of the | |
2824 | * extent, so its much less complex to remember where it started | |
2825 | */ | |
2826 | last = found_key.offset; | |
2827 | last_for_get_extent = last + 1; | |
975f84fe | 2828 | } |
975f84fe JB |
2829 | btrfs_free_path(path); |
2830 | ||
ec29ed5b CM |
2831 | /* |
2832 | * we might have some extents allocated but more delalloc past those | |
2833 | * extents. so, we trust isize unless the start of the last extent is | |
2834 | * beyond isize | |
2835 | */ | |
2836 | if (last < isize) { | |
2837 | last = (u64)-1; | |
2838 | last_for_get_extent = isize; | |
2839 | } | |
2840 | ||
2ac55d41 JB |
2841 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0, |
2842 | &cached_state, GFP_NOFS); | |
ec29ed5b CM |
2843 | |
2844 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
2845 | get_extent); | |
1506fcc8 YS |
2846 | if (!em) |
2847 | goto out; | |
2848 | if (IS_ERR(em)) { | |
2849 | ret = PTR_ERR(em); | |
2850 | goto out; | |
2851 | } | |
975f84fe | 2852 | |
1506fcc8 | 2853 | while (!end) { |
ea8efc74 CM |
2854 | u64 offset_in_extent; |
2855 | ||
2856 | /* break if the extent we found is outside the range */ | |
2857 | if (em->start >= max || extent_map_end(em) < off) | |
2858 | break; | |
2859 | ||
2860 | /* | |
2861 | * get_extent may return an extent that starts before our | |
2862 | * requested range. We have to make sure the ranges | |
2863 | * we return to fiemap always move forward and don't | |
2864 | * overlap, so adjust the offsets here | |
2865 | */ | |
2866 | em_start = max(em->start, off); | |
1506fcc8 | 2867 | |
ea8efc74 CM |
2868 | /* |
2869 | * record the offset from the start of the extent | |
2870 | * for adjusting the disk offset below | |
2871 | */ | |
2872 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 2873 | em_end = extent_map_end(em); |
ea8efc74 | 2874 | em_len = em_end - em_start; |
ec29ed5b | 2875 | emflags = em->flags; |
1506fcc8 YS |
2876 | disko = 0; |
2877 | flags = 0; | |
2878 | ||
ea8efc74 CM |
2879 | /* |
2880 | * bump off for our next call to get_extent | |
2881 | */ | |
2882 | off = extent_map_end(em); | |
2883 | if (off >= max) | |
2884 | end = 1; | |
2885 | ||
93dbfad7 | 2886 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
2887 | end = 1; |
2888 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 2889 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
2890 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
2891 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 2892 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
2893 | flags |= (FIEMAP_EXTENT_DELALLOC | |
2894 | FIEMAP_EXTENT_UNKNOWN); | |
93dbfad7 | 2895 | } else { |
ea8efc74 | 2896 | disko = em->block_start + offset_in_extent; |
1506fcc8 YS |
2897 | } |
2898 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
2899 | flags |= FIEMAP_EXTENT_ENCODED; | |
2900 | ||
1506fcc8 YS |
2901 | free_extent_map(em); |
2902 | em = NULL; | |
ec29ed5b CM |
2903 | if ((em_start >= last) || em_len == (u64)-1 || |
2904 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
2905 | flags |= FIEMAP_EXTENT_LAST; |
2906 | end = 1; | |
2907 | } | |
2908 | ||
ec29ed5b CM |
2909 | /* now scan forward to see if this is really the last extent. */ |
2910 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
2911 | get_extent); | |
2912 | if (IS_ERR(em)) { | |
2913 | ret = PTR_ERR(em); | |
2914 | goto out; | |
2915 | } | |
2916 | if (!em) { | |
975f84fe JB |
2917 | flags |= FIEMAP_EXTENT_LAST; |
2918 | end = 1; | |
2919 | } | |
ec29ed5b CM |
2920 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, |
2921 | em_len, flags); | |
2922 | if (ret) | |
2923 | goto out_free; | |
1506fcc8 YS |
2924 | } |
2925 | out_free: | |
2926 | free_extent_map(em); | |
2927 | out: | |
2ac55d41 JB |
2928 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len, |
2929 | &cached_state, GFP_NOFS); | |
1506fcc8 YS |
2930 | return ret; |
2931 | } | |
2932 | ||
d1310b2e CM |
2933 | static inline struct page *extent_buffer_page(struct extent_buffer *eb, |
2934 | unsigned long i) | |
2935 | { | |
2936 | struct page *p; | |
2937 | struct address_space *mapping; | |
2938 | ||
2939 | if (i == 0) | |
2940 | return eb->first_page; | |
2941 | i += eb->start >> PAGE_CACHE_SHIFT; | |
2942 | mapping = eb->first_page->mapping; | |
33958dc6 CM |
2943 | if (!mapping) |
2944 | return NULL; | |
0ee0fda0 SW |
2945 | |
2946 | /* | |
2947 | * extent_buffer_page is only called after pinning the page | |
2948 | * by increasing the reference count. So we know the page must | |
2949 | * be in the radix tree. | |
2950 | */ | |
0ee0fda0 | 2951 | rcu_read_lock(); |
d1310b2e | 2952 | p = radix_tree_lookup(&mapping->page_tree, i); |
0ee0fda0 | 2953 | rcu_read_unlock(); |
2b1f55b0 | 2954 | |
d1310b2e CM |
2955 | return p; |
2956 | } | |
2957 | ||
6af118ce | 2958 | static inline unsigned long num_extent_pages(u64 start, u64 len) |
728131d8 | 2959 | { |
6af118ce CM |
2960 | return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) - |
2961 | (start >> PAGE_CACHE_SHIFT); | |
728131d8 CM |
2962 | } |
2963 | ||
d1310b2e CM |
2964 | static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree, |
2965 | u64 start, | |
2966 | unsigned long len, | |
2967 | gfp_t mask) | |
2968 | { | |
2969 | struct extent_buffer *eb = NULL; | |
3935127c | 2970 | #if LEAK_DEBUG |
2d2ae547 | 2971 | unsigned long flags; |
4bef0848 | 2972 | #endif |
d1310b2e | 2973 | |
d1310b2e | 2974 | eb = kmem_cache_zalloc(extent_buffer_cache, mask); |
91ca338d TI |
2975 | if (eb == NULL) |
2976 | return NULL; | |
d1310b2e CM |
2977 | eb->start = start; |
2978 | eb->len = len; | |
bd681513 CM |
2979 | rwlock_init(&eb->lock); |
2980 | atomic_set(&eb->write_locks, 0); | |
2981 | atomic_set(&eb->read_locks, 0); | |
2982 | atomic_set(&eb->blocking_readers, 0); | |
2983 | atomic_set(&eb->blocking_writers, 0); | |
2984 | atomic_set(&eb->spinning_readers, 0); | |
2985 | atomic_set(&eb->spinning_writers, 0); | |
2986 | init_waitqueue_head(&eb->write_lock_wq); | |
2987 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 2988 | |
3935127c | 2989 | #if LEAK_DEBUG |
2d2ae547 CM |
2990 | spin_lock_irqsave(&leak_lock, flags); |
2991 | list_add(&eb->leak_list, &buffers); | |
2992 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 2993 | #endif |
d1310b2e CM |
2994 | atomic_set(&eb->refs, 1); |
2995 | ||
2996 | return eb; | |
2997 | } | |
2998 | ||
2999 | static void __free_extent_buffer(struct extent_buffer *eb) | |
3000 | { | |
3935127c | 3001 | #if LEAK_DEBUG |
2d2ae547 CM |
3002 | unsigned long flags; |
3003 | spin_lock_irqsave(&leak_lock, flags); | |
3004 | list_del(&eb->leak_list); | |
3005 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 3006 | #endif |
d1310b2e CM |
3007 | kmem_cache_free(extent_buffer_cache, eb); |
3008 | } | |
3009 | ||
897ca6e9 MX |
3010 | /* |
3011 | * Helper for releasing extent buffer page. | |
3012 | */ | |
3013 | static void btrfs_release_extent_buffer_page(struct extent_buffer *eb, | |
3014 | unsigned long start_idx) | |
3015 | { | |
3016 | unsigned long index; | |
3017 | struct page *page; | |
3018 | ||
3019 | if (!eb->first_page) | |
3020 | return; | |
3021 | ||
3022 | index = num_extent_pages(eb->start, eb->len); | |
3023 | if (start_idx >= index) | |
3024 | return; | |
3025 | ||
3026 | do { | |
3027 | index--; | |
3028 | page = extent_buffer_page(eb, index); | |
3029 | if (page) | |
3030 | page_cache_release(page); | |
3031 | } while (index != start_idx); | |
3032 | } | |
3033 | ||
3034 | /* | |
3035 | * Helper for releasing the extent buffer. | |
3036 | */ | |
3037 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
3038 | { | |
3039 | btrfs_release_extent_buffer_page(eb, 0); | |
3040 | __free_extent_buffer(eb); | |
3041 | } | |
3042 | ||
d1310b2e CM |
3043 | struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree, |
3044 | u64 start, unsigned long len, | |
ba144192 | 3045 | struct page *page0) |
d1310b2e CM |
3046 | { |
3047 | unsigned long num_pages = num_extent_pages(start, len); | |
3048 | unsigned long i; | |
3049 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
3050 | struct extent_buffer *eb; | |
6af118ce | 3051 | struct extent_buffer *exists = NULL; |
d1310b2e CM |
3052 | struct page *p; |
3053 | struct address_space *mapping = tree->mapping; | |
3054 | int uptodate = 1; | |
19fe0a8b | 3055 | int ret; |
d1310b2e | 3056 | |
19fe0a8b MX |
3057 | rcu_read_lock(); |
3058 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); | |
3059 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
3060 | rcu_read_unlock(); | |
0f9dd46c | 3061 | mark_page_accessed(eb->first_page); |
6af118ce CM |
3062 | return eb; |
3063 | } | |
19fe0a8b | 3064 | rcu_read_unlock(); |
6af118ce | 3065 | |
ba144192 | 3066 | eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS); |
2b114d1d | 3067 | if (!eb) |
d1310b2e CM |
3068 | return NULL; |
3069 | ||
d1310b2e CM |
3070 | if (page0) { |
3071 | eb->first_page = page0; | |
3072 | i = 1; | |
3073 | index++; | |
3074 | page_cache_get(page0); | |
3075 | mark_page_accessed(page0); | |
3076 | set_page_extent_mapped(page0); | |
d1310b2e | 3077 | set_page_extent_head(page0, len); |
f188591e | 3078 | uptodate = PageUptodate(page0); |
d1310b2e CM |
3079 | } else { |
3080 | i = 0; | |
3081 | } | |
3082 | for (; i < num_pages; i++, index++) { | |
a6591715 | 3083 | p = find_or_create_page(mapping, index, GFP_NOFS); |
d1310b2e CM |
3084 | if (!p) { |
3085 | WARN_ON(1); | |
6af118ce | 3086 | goto free_eb; |
d1310b2e CM |
3087 | } |
3088 | set_page_extent_mapped(p); | |
3089 | mark_page_accessed(p); | |
3090 | if (i == 0) { | |
3091 | eb->first_page = p; | |
3092 | set_page_extent_head(p, len); | |
3093 | } else { | |
3094 | set_page_private(p, EXTENT_PAGE_PRIVATE); | |
3095 | } | |
3096 | if (!PageUptodate(p)) | |
3097 | uptodate = 0; | |
eb14ab8e CM |
3098 | |
3099 | /* | |
3100 | * see below about how we avoid a nasty race with release page | |
3101 | * and why we unlock later | |
3102 | */ | |
3103 | if (i != 0) | |
3104 | unlock_page(p); | |
d1310b2e CM |
3105 | } |
3106 | if (uptodate) | |
b4ce94de | 3107 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 3108 | |
19fe0a8b MX |
3109 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); |
3110 | if (ret) | |
3111 | goto free_eb; | |
3112 | ||
6af118ce | 3113 | spin_lock(&tree->buffer_lock); |
19fe0a8b MX |
3114 | ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb); |
3115 | if (ret == -EEXIST) { | |
3116 | exists = radix_tree_lookup(&tree->buffer, | |
3117 | start >> PAGE_CACHE_SHIFT); | |
6af118ce CM |
3118 | /* add one reference for the caller */ |
3119 | atomic_inc(&exists->refs); | |
3120 | spin_unlock(&tree->buffer_lock); | |
19fe0a8b | 3121 | radix_tree_preload_end(); |
6af118ce CM |
3122 | goto free_eb; |
3123 | } | |
6af118ce CM |
3124 | /* add one reference for the tree */ |
3125 | atomic_inc(&eb->refs); | |
f044ba78 | 3126 | spin_unlock(&tree->buffer_lock); |
19fe0a8b | 3127 | radix_tree_preload_end(); |
eb14ab8e CM |
3128 | |
3129 | /* | |
3130 | * there is a race where release page may have | |
3131 | * tried to find this extent buffer in the radix | |
3132 | * but failed. It will tell the VM it is safe to | |
3133 | * reclaim the, and it will clear the page private bit. | |
3134 | * We must make sure to set the page private bit properly | |
3135 | * after the extent buffer is in the radix tree so | |
3136 | * it doesn't get lost | |
3137 | */ | |
3138 | set_page_extent_mapped(eb->first_page); | |
3139 | set_page_extent_head(eb->first_page, eb->len); | |
3140 | if (!page0) | |
3141 | unlock_page(eb->first_page); | |
d1310b2e CM |
3142 | return eb; |
3143 | ||
6af118ce | 3144 | free_eb: |
eb14ab8e CM |
3145 | if (eb->first_page && !page0) |
3146 | unlock_page(eb->first_page); | |
3147 | ||
d1310b2e | 3148 | if (!atomic_dec_and_test(&eb->refs)) |
6af118ce | 3149 | return exists; |
897ca6e9 | 3150 | btrfs_release_extent_buffer(eb); |
6af118ce | 3151 | return exists; |
d1310b2e | 3152 | } |
d1310b2e CM |
3153 | |
3154 | struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree, | |
f09d1f60 | 3155 | u64 start, unsigned long len) |
d1310b2e | 3156 | { |
d1310b2e | 3157 | struct extent_buffer *eb; |
d1310b2e | 3158 | |
19fe0a8b MX |
3159 | rcu_read_lock(); |
3160 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); | |
3161 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
3162 | rcu_read_unlock(); | |
0f9dd46c | 3163 | mark_page_accessed(eb->first_page); |
19fe0a8b MX |
3164 | return eb; |
3165 | } | |
3166 | rcu_read_unlock(); | |
0f9dd46c | 3167 | |
19fe0a8b | 3168 | return NULL; |
d1310b2e | 3169 | } |
d1310b2e CM |
3170 | |
3171 | void free_extent_buffer(struct extent_buffer *eb) | |
3172 | { | |
d1310b2e CM |
3173 | if (!eb) |
3174 | return; | |
3175 | ||
3176 | if (!atomic_dec_and_test(&eb->refs)) | |
3177 | return; | |
3178 | ||
6af118ce | 3179 | WARN_ON(1); |
d1310b2e | 3180 | } |
d1310b2e CM |
3181 | |
3182 | int clear_extent_buffer_dirty(struct extent_io_tree *tree, | |
3183 | struct extent_buffer *eb) | |
3184 | { | |
d1310b2e CM |
3185 | unsigned long i; |
3186 | unsigned long num_pages; | |
3187 | struct page *page; | |
3188 | ||
d1310b2e CM |
3189 | num_pages = num_extent_pages(eb->start, eb->len); |
3190 | ||
3191 | for (i = 0; i < num_pages; i++) { | |
3192 | page = extent_buffer_page(eb, i); | |
b9473439 | 3193 | if (!PageDirty(page)) |
d2c3f4f6 CM |
3194 | continue; |
3195 | ||
a61e6f29 | 3196 | lock_page(page); |
eb14ab8e CM |
3197 | WARN_ON(!PagePrivate(page)); |
3198 | ||
3199 | set_page_extent_mapped(page); | |
d1310b2e CM |
3200 | if (i == 0) |
3201 | set_page_extent_head(page, eb->len); | |
d1310b2e | 3202 | |
d1310b2e | 3203 | clear_page_dirty_for_io(page); |
0ee0fda0 | 3204 | spin_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
3205 | if (!PageDirty(page)) { |
3206 | radix_tree_tag_clear(&page->mapping->page_tree, | |
3207 | page_index(page), | |
3208 | PAGECACHE_TAG_DIRTY); | |
3209 | } | |
0ee0fda0 | 3210 | spin_unlock_irq(&page->mapping->tree_lock); |
a61e6f29 | 3211 | unlock_page(page); |
d1310b2e CM |
3212 | } |
3213 | return 0; | |
3214 | } | |
d1310b2e | 3215 | |
d1310b2e CM |
3216 | int set_extent_buffer_dirty(struct extent_io_tree *tree, |
3217 | struct extent_buffer *eb) | |
3218 | { | |
3219 | unsigned long i; | |
3220 | unsigned long num_pages; | |
b9473439 | 3221 | int was_dirty = 0; |
d1310b2e | 3222 | |
b9473439 | 3223 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
d1310b2e | 3224 | num_pages = num_extent_pages(eb->start, eb->len); |
b9473439 | 3225 | for (i = 0; i < num_pages; i++) |
d1310b2e | 3226 | __set_page_dirty_nobuffers(extent_buffer_page(eb, i)); |
b9473439 | 3227 | return was_dirty; |
d1310b2e | 3228 | } |
d1310b2e | 3229 | |
19b6caf4 CM |
3230 | static int __eb_straddles_pages(u64 start, u64 len) |
3231 | { | |
3232 | if (len < PAGE_CACHE_SIZE) | |
3233 | return 1; | |
3234 | if (start & (PAGE_CACHE_SIZE - 1)) | |
3235 | return 1; | |
3236 | if ((start + len) & (PAGE_CACHE_SIZE - 1)) | |
3237 | return 1; | |
3238 | return 0; | |
3239 | } | |
3240 | ||
3241 | static int eb_straddles_pages(struct extent_buffer *eb) | |
3242 | { | |
3243 | return __eb_straddles_pages(eb->start, eb->len); | |
3244 | } | |
3245 | ||
1259ab75 | 3246 | int clear_extent_buffer_uptodate(struct extent_io_tree *tree, |
2ac55d41 JB |
3247 | struct extent_buffer *eb, |
3248 | struct extent_state **cached_state) | |
1259ab75 CM |
3249 | { |
3250 | unsigned long i; | |
3251 | struct page *page; | |
3252 | unsigned long num_pages; | |
3253 | ||
3254 | num_pages = num_extent_pages(eb->start, eb->len); | |
b4ce94de | 3255 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
1259ab75 | 3256 | |
19b6caf4 CM |
3257 | if (eb_straddles_pages(eb)) { |
3258 | clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1, | |
3259 | cached_state, GFP_NOFS); | |
3260 | } | |
1259ab75 CM |
3261 | for (i = 0; i < num_pages; i++) { |
3262 | page = extent_buffer_page(eb, i); | |
33958dc6 CM |
3263 | if (page) |
3264 | ClearPageUptodate(page); | |
1259ab75 CM |
3265 | } |
3266 | return 0; | |
3267 | } | |
3268 | ||
d1310b2e CM |
3269 | int set_extent_buffer_uptodate(struct extent_io_tree *tree, |
3270 | struct extent_buffer *eb) | |
3271 | { | |
3272 | unsigned long i; | |
3273 | struct page *page; | |
3274 | unsigned long num_pages; | |
3275 | ||
3276 | num_pages = num_extent_pages(eb->start, eb->len); | |
3277 | ||
19b6caf4 CM |
3278 | if (eb_straddles_pages(eb)) { |
3279 | set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1, | |
3280 | NULL, GFP_NOFS); | |
3281 | } | |
d1310b2e CM |
3282 | for (i = 0; i < num_pages; i++) { |
3283 | page = extent_buffer_page(eb, i); | |
3284 | if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) || | |
3285 | ((i == num_pages - 1) && | |
3286 | ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) { | |
3287 | check_page_uptodate(tree, page); | |
3288 | continue; | |
3289 | } | |
3290 | SetPageUptodate(page); | |
3291 | } | |
3292 | return 0; | |
3293 | } | |
d1310b2e | 3294 | |
ce9adaa5 CM |
3295 | int extent_range_uptodate(struct extent_io_tree *tree, |
3296 | u64 start, u64 end) | |
3297 | { | |
3298 | struct page *page; | |
3299 | int ret; | |
3300 | int pg_uptodate = 1; | |
3301 | int uptodate; | |
3302 | unsigned long index; | |
3303 | ||
19b6caf4 CM |
3304 | if (__eb_straddles_pages(start, end - start + 1)) { |
3305 | ret = test_range_bit(tree, start, end, | |
3306 | EXTENT_UPTODATE, 1, NULL); | |
3307 | if (ret) | |
3308 | return 1; | |
3309 | } | |
d397712b | 3310 | while (start <= end) { |
ce9adaa5 CM |
3311 | index = start >> PAGE_CACHE_SHIFT; |
3312 | page = find_get_page(tree->mapping, index); | |
3313 | uptodate = PageUptodate(page); | |
3314 | page_cache_release(page); | |
3315 | if (!uptodate) { | |
3316 | pg_uptodate = 0; | |
3317 | break; | |
3318 | } | |
3319 | start += PAGE_CACHE_SIZE; | |
3320 | } | |
3321 | return pg_uptodate; | |
3322 | } | |
3323 | ||
d1310b2e | 3324 | int extent_buffer_uptodate(struct extent_io_tree *tree, |
2ac55d41 JB |
3325 | struct extent_buffer *eb, |
3326 | struct extent_state *cached_state) | |
d1310b2e | 3327 | { |
728131d8 | 3328 | int ret = 0; |
ce9adaa5 CM |
3329 | unsigned long num_pages; |
3330 | unsigned long i; | |
728131d8 CM |
3331 | struct page *page; |
3332 | int pg_uptodate = 1; | |
3333 | ||
b4ce94de | 3334 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
4235298e | 3335 | return 1; |
728131d8 | 3336 | |
19b6caf4 CM |
3337 | if (eb_straddles_pages(eb)) { |
3338 | ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1, | |
3339 | EXTENT_UPTODATE, 1, cached_state); | |
3340 | if (ret) | |
3341 | return ret; | |
3342 | } | |
728131d8 CM |
3343 | |
3344 | num_pages = num_extent_pages(eb->start, eb->len); | |
3345 | for (i = 0; i < num_pages; i++) { | |
3346 | page = extent_buffer_page(eb, i); | |
3347 | if (!PageUptodate(page)) { | |
3348 | pg_uptodate = 0; | |
3349 | break; | |
3350 | } | |
3351 | } | |
4235298e | 3352 | return pg_uptodate; |
d1310b2e | 3353 | } |
d1310b2e CM |
3354 | |
3355 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
3356 | struct extent_buffer *eb, | |
a86c12c7 | 3357 | u64 start, int wait, |
f188591e | 3358 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3359 | { |
3360 | unsigned long i; | |
3361 | unsigned long start_i; | |
3362 | struct page *page; | |
3363 | int err; | |
3364 | int ret = 0; | |
ce9adaa5 CM |
3365 | int locked_pages = 0; |
3366 | int all_uptodate = 1; | |
3367 | int inc_all_pages = 0; | |
d1310b2e | 3368 | unsigned long num_pages; |
a86c12c7 | 3369 | struct bio *bio = NULL; |
c8b97818 | 3370 | unsigned long bio_flags = 0; |
a86c12c7 | 3371 | |
b4ce94de | 3372 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
3373 | return 0; |
3374 | ||
19b6caf4 CM |
3375 | if (eb_straddles_pages(eb)) { |
3376 | if (test_range_bit(tree, eb->start, eb->start + eb->len - 1, | |
3377 | EXTENT_UPTODATE, 1, NULL)) { | |
3378 | return 0; | |
3379 | } | |
d1310b2e CM |
3380 | } |
3381 | ||
3382 | if (start) { | |
3383 | WARN_ON(start < eb->start); | |
3384 | start_i = (start >> PAGE_CACHE_SHIFT) - | |
3385 | (eb->start >> PAGE_CACHE_SHIFT); | |
3386 | } else { | |
3387 | start_i = 0; | |
3388 | } | |
3389 | ||
3390 | num_pages = num_extent_pages(eb->start, eb->len); | |
3391 | for (i = start_i; i < num_pages; i++) { | |
3392 | page = extent_buffer_page(eb, i); | |
d1310b2e | 3393 | if (!wait) { |
2db04966 | 3394 | if (!trylock_page(page)) |
ce9adaa5 | 3395 | goto unlock_exit; |
d1310b2e CM |
3396 | } else { |
3397 | lock_page(page); | |
3398 | } | |
ce9adaa5 | 3399 | locked_pages++; |
d397712b | 3400 | if (!PageUptodate(page)) |
ce9adaa5 | 3401 | all_uptodate = 0; |
ce9adaa5 CM |
3402 | } |
3403 | if (all_uptodate) { | |
3404 | if (start_i == 0) | |
b4ce94de | 3405 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
3406 | goto unlock_exit; |
3407 | } | |
3408 | ||
3409 | for (i = start_i; i < num_pages; i++) { | |
3410 | page = extent_buffer_page(eb, i); | |
eb14ab8e CM |
3411 | |
3412 | WARN_ON(!PagePrivate(page)); | |
3413 | ||
3414 | set_page_extent_mapped(page); | |
3415 | if (i == 0) | |
3416 | set_page_extent_head(page, eb->len); | |
3417 | ||
ce9adaa5 CM |
3418 | if (inc_all_pages) |
3419 | page_cache_get(page); | |
3420 | if (!PageUptodate(page)) { | |
3421 | if (start_i == 0) | |
3422 | inc_all_pages = 1; | |
f188591e | 3423 | ClearPageError(page); |
a86c12c7 | 3424 | err = __extent_read_full_page(tree, page, |
f188591e | 3425 | get_extent, &bio, |
c8b97818 | 3426 | mirror_num, &bio_flags); |
d397712b | 3427 | if (err) |
d1310b2e | 3428 | ret = err; |
d1310b2e CM |
3429 | } else { |
3430 | unlock_page(page); | |
3431 | } | |
3432 | } | |
3433 | ||
a86c12c7 | 3434 | if (bio) |
c8b97818 | 3435 | submit_one_bio(READ, bio, mirror_num, bio_flags); |
a86c12c7 | 3436 | |
d397712b | 3437 | if (ret || !wait) |
d1310b2e | 3438 | return ret; |
d397712b | 3439 | |
d1310b2e CM |
3440 | for (i = start_i; i < num_pages; i++) { |
3441 | page = extent_buffer_page(eb, i); | |
3442 | wait_on_page_locked(page); | |
d397712b | 3443 | if (!PageUptodate(page)) |
d1310b2e | 3444 | ret = -EIO; |
d1310b2e | 3445 | } |
d397712b | 3446 | |
d1310b2e | 3447 | if (!ret) |
b4ce94de | 3448 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 3449 | return ret; |
ce9adaa5 CM |
3450 | |
3451 | unlock_exit: | |
3452 | i = start_i; | |
d397712b | 3453 | while (locked_pages > 0) { |
ce9adaa5 CM |
3454 | page = extent_buffer_page(eb, i); |
3455 | i++; | |
3456 | unlock_page(page); | |
3457 | locked_pages--; | |
3458 | } | |
3459 | return ret; | |
d1310b2e | 3460 | } |
d1310b2e CM |
3461 | |
3462 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
3463 | unsigned long start, | |
3464 | unsigned long len) | |
3465 | { | |
3466 | size_t cur; | |
3467 | size_t offset; | |
3468 | struct page *page; | |
3469 | char *kaddr; | |
3470 | char *dst = (char *)dstv; | |
3471 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3472 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
3473 | |
3474 | WARN_ON(start > eb->len); | |
3475 | WARN_ON(start + len > eb->start + eb->len); | |
3476 | ||
3477 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3478 | ||
d397712b | 3479 | while (len > 0) { |
d1310b2e | 3480 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
3481 | |
3482 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
a6591715 | 3483 | kaddr = page_address(page); |
d1310b2e | 3484 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
3485 | |
3486 | dst += cur; | |
3487 | len -= cur; | |
3488 | offset = 0; | |
3489 | i++; | |
3490 | } | |
3491 | } | |
d1310b2e CM |
3492 | |
3493 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
a6591715 | 3494 | unsigned long min_len, char **map, |
d1310b2e | 3495 | unsigned long *map_start, |
a6591715 | 3496 | unsigned long *map_len) |
d1310b2e CM |
3497 | { |
3498 | size_t offset = start & (PAGE_CACHE_SIZE - 1); | |
3499 | char *kaddr; | |
3500 | struct page *p; | |
3501 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3502 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3503 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
3504 | PAGE_CACHE_SHIFT; | |
3505 | ||
3506 | if (i != end_i) | |
3507 | return -EINVAL; | |
3508 | ||
3509 | if (i == 0) { | |
3510 | offset = start_offset; | |
3511 | *map_start = 0; | |
3512 | } else { | |
3513 | offset = 0; | |
3514 | *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; | |
3515 | } | |
d397712b | 3516 | |
d1310b2e | 3517 | if (start + min_len > eb->len) { |
d397712b CM |
3518 | printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, " |
3519 | "wanted %lu %lu\n", (unsigned long long)eb->start, | |
3520 | eb->len, start, min_len); | |
d1310b2e | 3521 | WARN_ON(1); |
85026533 | 3522 | return -EINVAL; |
d1310b2e CM |
3523 | } |
3524 | ||
3525 | p = extent_buffer_page(eb, i); | |
a6591715 | 3526 | kaddr = page_address(p); |
d1310b2e CM |
3527 | *map = kaddr + offset; |
3528 | *map_len = PAGE_CACHE_SIZE - offset; | |
3529 | return 0; | |
3530 | } | |
d1310b2e | 3531 | |
d1310b2e CM |
3532 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, |
3533 | unsigned long start, | |
3534 | unsigned long len) | |
3535 | { | |
3536 | size_t cur; | |
3537 | size_t offset; | |
3538 | struct page *page; | |
3539 | char *kaddr; | |
3540 | char *ptr = (char *)ptrv; | |
3541 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3542 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3543 | int ret = 0; | |
3544 | ||
3545 | WARN_ON(start > eb->len); | |
3546 | WARN_ON(start + len > eb->start + eb->len); | |
3547 | ||
3548 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3549 | ||
d397712b | 3550 | while (len > 0) { |
d1310b2e | 3551 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
3552 | |
3553 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
3554 | ||
a6591715 | 3555 | kaddr = page_address(page); |
d1310b2e | 3556 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
3557 | if (ret) |
3558 | break; | |
3559 | ||
3560 | ptr += cur; | |
3561 | len -= cur; | |
3562 | offset = 0; | |
3563 | i++; | |
3564 | } | |
3565 | return ret; | |
3566 | } | |
d1310b2e CM |
3567 | |
3568 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
3569 | unsigned long start, unsigned long len) | |
3570 | { | |
3571 | size_t cur; | |
3572 | size_t offset; | |
3573 | struct page *page; | |
3574 | char *kaddr; | |
3575 | char *src = (char *)srcv; | |
3576 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3577 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3578 | ||
3579 | WARN_ON(start > eb->len); | |
3580 | WARN_ON(start + len > eb->start + eb->len); | |
3581 | ||
3582 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3583 | ||
d397712b | 3584 | while (len > 0) { |
d1310b2e CM |
3585 | page = extent_buffer_page(eb, i); |
3586 | WARN_ON(!PageUptodate(page)); | |
3587 | ||
3588 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 3589 | kaddr = page_address(page); |
d1310b2e | 3590 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
3591 | |
3592 | src += cur; | |
3593 | len -= cur; | |
3594 | offset = 0; | |
3595 | i++; | |
3596 | } | |
3597 | } | |
d1310b2e CM |
3598 | |
3599 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
3600 | unsigned long start, unsigned long len) | |
3601 | { | |
3602 | size_t cur; | |
3603 | size_t offset; | |
3604 | struct page *page; | |
3605 | char *kaddr; | |
3606 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3607 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3608 | ||
3609 | WARN_ON(start > eb->len); | |
3610 | WARN_ON(start + len > eb->start + eb->len); | |
3611 | ||
3612 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3613 | ||
d397712b | 3614 | while (len > 0) { |
d1310b2e CM |
3615 | page = extent_buffer_page(eb, i); |
3616 | WARN_ON(!PageUptodate(page)); | |
3617 | ||
3618 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 3619 | kaddr = page_address(page); |
d1310b2e | 3620 | memset(kaddr + offset, c, cur); |
d1310b2e CM |
3621 | |
3622 | len -= cur; | |
3623 | offset = 0; | |
3624 | i++; | |
3625 | } | |
3626 | } | |
d1310b2e CM |
3627 | |
3628 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
3629 | unsigned long dst_offset, unsigned long src_offset, | |
3630 | unsigned long len) | |
3631 | { | |
3632 | u64 dst_len = dst->len; | |
3633 | size_t cur; | |
3634 | size_t offset; | |
3635 | struct page *page; | |
3636 | char *kaddr; | |
3637 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3638 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
3639 | ||
3640 | WARN_ON(src->len != dst_len); | |
3641 | ||
3642 | offset = (start_offset + dst_offset) & | |
3643 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3644 | ||
d397712b | 3645 | while (len > 0) { |
d1310b2e CM |
3646 | page = extent_buffer_page(dst, i); |
3647 | WARN_ON(!PageUptodate(page)); | |
3648 | ||
3649 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
3650 | ||
a6591715 | 3651 | kaddr = page_address(page); |
d1310b2e | 3652 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
3653 | |
3654 | src_offset += cur; | |
3655 | len -= cur; | |
3656 | offset = 0; | |
3657 | i++; | |
3658 | } | |
3659 | } | |
d1310b2e CM |
3660 | |
3661 | static void move_pages(struct page *dst_page, struct page *src_page, | |
3662 | unsigned long dst_off, unsigned long src_off, | |
3663 | unsigned long len) | |
3664 | { | |
a6591715 | 3665 | char *dst_kaddr = page_address(dst_page); |
d1310b2e CM |
3666 | if (dst_page == src_page) { |
3667 | memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len); | |
3668 | } else { | |
a6591715 | 3669 | char *src_kaddr = page_address(src_page); |
d1310b2e CM |
3670 | char *p = dst_kaddr + dst_off + len; |
3671 | char *s = src_kaddr + src_off + len; | |
3672 | ||
3673 | while (len--) | |
3674 | *--p = *--s; | |
d1310b2e | 3675 | } |
d1310b2e CM |
3676 | } |
3677 | ||
3387206f ST |
3678 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
3679 | { | |
3680 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
3681 | return distance < len; | |
3682 | } | |
3683 | ||
d1310b2e CM |
3684 | static void copy_pages(struct page *dst_page, struct page *src_page, |
3685 | unsigned long dst_off, unsigned long src_off, | |
3686 | unsigned long len) | |
3687 | { | |
a6591715 | 3688 | char *dst_kaddr = page_address(dst_page); |
d1310b2e CM |
3689 | char *src_kaddr; |
3690 | ||
3387206f | 3691 | if (dst_page != src_page) { |
a6591715 | 3692 | src_kaddr = page_address(src_page); |
3387206f | 3693 | } else { |
d1310b2e | 3694 | src_kaddr = dst_kaddr; |
3387206f ST |
3695 | BUG_ON(areas_overlap(src_off, dst_off, len)); |
3696 | } | |
d1310b2e CM |
3697 | |
3698 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
3699 | } |
3700 | ||
3701 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
3702 | unsigned long src_offset, unsigned long len) | |
3703 | { | |
3704 | size_t cur; | |
3705 | size_t dst_off_in_page; | |
3706 | size_t src_off_in_page; | |
3707 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3708 | unsigned long dst_i; | |
3709 | unsigned long src_i; | |
3710 | ||
3711 | if (src_offset + len > dst->len) { | |
d397712b CM |
3712 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
3713 | "len %lu dst len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
3714 | BUG_ON(1); |
3715 | } | |
3716 | if (dst_offset + len > dst->len) { | |
d397712b CM |
3717 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
3718 | "len %lu dst len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
3719 | BUG_ON(1); |
3720 | } | |
3721 | ||
d397712b | 3722 | while (len > 0) { |
d1310b2e CM |
3723 | dst_off_in_page = (start_offset + dst_offset) & |
3724 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3725 | src_off_in_page = (start_offset + src_offset) & | |
3726 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3727 | ||
3728 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
3729 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
3730 | ||
3731 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
3732 | src_off_in_page)); | |
3733 | cur = min_t(unsigned long, cur, | |
3734 | (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); | |
3735 | ||
3736 | copy_pages(extent_buffer_page(dst, dst_i), | |
3737 | extent_buffer_page(dst, src_i), | |
3738 | dst_off_in_page, src_off_in_page, cur); | |
3739 | ||
3740 | src_offset += cur; | |
3741 | dst_offset += cur; | |
3742 | len -= cur; | |
3743 | } | |
3744 | } | |
d1310b2e CM |
3745 | |
3746 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
3747 | unsigned long src_offset, unsigned long len) | |
3748 | { | |
3749 | size_t cur; | |
3750 | size_t dst_off_in_page; | |
3751 | size_t src_off_in_page; | |
3752 | unsigned long dst_end = dst_offset + len - 1; | |
3753 | unsigned long src_end = src_offset + len - 1; | |
3754 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3755 | unsigned long dst_i; | |
3756 | unsigned long src_i; | |
3757 | ||
3758 | if (src_offset + len > dst->len) { | |
d397712b CM |
3759 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
3760 | "len %lu len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
3761 | BUG_ON(1); |
3762 | } | |
3763 | if (dst_offset + len > dst->len) { | |
d397712b CM |
3764 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
3765 | "len %lu len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
3766 | BUG_ON(1); |
3767 | } | |
3387206f | 3768 | if (!areas_overlap(src_offset, dst_offset, len)) { |
d1310b2e CM |
3769 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
3770 | return; | |
3771 | } | |
d397712b | 3772 | while (len > 0) { |
d1310b2e CM |
3773 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; |
3774 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
3775 | ||
3776 | dst_off_in_page = (start_offset + dst_end) & | |
3777 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3778 | src_off_in_page = (start_offset + src_end) & | |
3779 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3780 | ||
3781 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
3782 | cur = min(cur, dst_off_in_page + 1); | |
3783 | move_pages(extent_buffer_page(dst, dst_i), | |
3784 | extent_buffer_page(dst, src_i), | |
3785 | dst_off_in_page - cur + 1, | |
3786 | src_off_in_page - cur + 1, cur); | |
3787 | ||
3788 | dst_end -= cur; | |
3789 | src_end -= cur; | |
3790 | len -= cur; | |
3791 | } | |
3792 | } | |
6af118ce | 3793 | |
19fe0a8b MX |
3794 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
3795 | { | |
3796 | struct extent_buffer *eb = | |
3797 | container_of(head, struct extent_buffer, rcu_head); | |
3798 | ||
3799 | btrfs_release_extent_buffer(eb); | |
3800 | } | |
3801 | ||
6af118ce CM |
3802 | int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page) |
3803 | { | |
3804 | u64 start = page_offset(page); | |
3805 | struct extent_buffer *eb; | |
3806 | int ret = 1; | |
6af118ce CM |
3807 | |
3808 | spin_lock(&tree->buffer_lock); | |
19fe0a8b | 3809 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); |
45f49bce CM |
3810 | if (!eb) { |
3811 | spin_unlock(&tree->buffer_lock); | |
3812 | return ret; | |
3813 | } | |
6af118ce | 3814 | |
19fe0a8b | 3815 | if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
6af118ce CM |
3816 | ret = 0; |
3817 | goto out; | |
3818 | } | |
19fe0a8b MX |
3819 | |
3820 | /* | |
3821 | * set @eb->refs to 0 if it is already 1, and then release the @eb. | |
3822 | * Or go back. | |
3823 | */ | |
3824 | if (atomic_cmpxchg(&eb->refs, 1, 0) != 1) { | |
b9473439 CM |
3825 | ret = 0; |
3826 | goto out; | |
3827 | } | |
897ca6e9 | 3828 | |
19fe0a8b | 3829 | radix_tree_delete(&tree->buffer, start >> PAGE_CACHE_SHIFT); |
6af118ce CM |
3830 | out: |
3831 | spin_unlock(&tree->buffer_lock); | |
19fe0a8b MX |
3832 | |
3833 | /* at this point we can safely release the extent buffer */ | |
3834 | if (atomic_read(&eb->refs) == 0) | |
3835 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); | |
6af118ce CM |
3836 | return ret; |
3837 | } |