]>
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 | 896 | |
462d6fac JB |
897 | /** |
898 | * convert_extent - convert all bits in a given range from one bit to another | |
899 | * @tree: the io tree to search | |
900 | * @start: the start offset in bytes | |
901 | * @end: the end offset in bytes (inclusive) | |
902 | * @bits: the bits to set in this range | |
903 | * @clear_bits: the bits to clear in this range | |
904 | * @mask: the allocation mask | |
905 | * | |
906 | * This will go through and set bits for the given range. If any states exist | |
907 | * already in this range they are set with the given bit and cleared of the | |
908 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
909 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
910 | * boundary bits like LOCK. | |
911 | */ | |
912 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
913 | int bits, int clear_bits, gfp_t mask) | |
914 | { | |
915 | struct extent_state *state; | |
916 | struct extent_state *prealloc = NULL; | |
917 | struct rb_node *node; | |
918 | int err = 0; | |
919 | u64 last_start; | |
920 | u64 last_end; | |
921 | ||
922 | again: | |
923 | if (!prealloc && (mask & __GFP_WAIT)) { | |
924 | prealloc = alloc_extent_state(mask); | |
925 | if (!prealloc) | |
926 | return -ENOMEM; | |
927 | } | |
928 | ||
929 | spin_lock(&tree->lock); | |
930 | /* | |
931 | * this search will find all the extents that end after | |
932 | * our range starts. | |
933 | */ | |
934 | node = tree_search(tree, start); | |
935 | if (!node) { | |
936 | prealloc = alloc_extent_state_atomic(prealloc); | |
937 | if (!prealloc) | |
938 | return -ENOMEM; | |
939 | err = insert_state(tree, prealloc, start, end, &bits); | |
940 | prealloc = NULL; | |
941 | BUG_ON(err == -EEXIST); | |
942 | goto out; | |
943 | } | |
944 | state = rb_entry(node, struct extent_state, rb_node); | |
945 | hit_next: | |
946 | last_start = state->start; | |
947 | last_end = state->end; | |
948 | ||
949 | /* | |
950 | * | ---- desired range ---- | | |
951 | * | state | | |
952 | * | |
953 | * Just lock what we found and keep going | |
954 | */ | |
955 | if (state->start == start && state->end <= end) { | |
956 | struct rb_node *next_node; | |
957 | ||
958 | set_state_bits(tree, state, &bits); | |
959 | clear_state_bit(tree, state, &clear_bits, 0); | |
960 | ||
961 | merge_state(tree, state); | |
962 | if (last_end == (u64)-1) | |
963 | goto out; | |
964 | ||
965 | start = last_end + 1; | |
966 | next_node = rb_next(&state->rb_node); | |
967 | if (next_node && start < end && prealloc && !need_resched()) { | |
968 | state = rb_entry(next_node, struct extent_state, | |
969 | rb_node); | |
970 | if (state->start == start) | |
971 | goto hit_next; | |
972 | } | |
973 | goto search_again; | |
974 | } | |
975 | ||
976 | /* | |
977 | * | ---- desired range ---- | | |
978 | * | state | | |
979 | * or | |
980 | * | ------------- state -------------- | | |
981 | * | |
982 | * We need to split the extent we found, and may flip bits on | |
983 | * second half. | |
984 | * | |
985 | * If the extent we found extends past our | |
986 | * range, we just split and search again. It'll get split | |
987 | * again the next time though. | |
988 | * | |
989 | * If the extent we found is inside our range, we set the | |
990 | * desired bit on it. | |
991 | */ | |
992 | if (state->start < start) { | |
993 | prealloc = alloc_extent_state_atomic(prealloc); | |
994 | if (!prealloc) | |
995 | return -ENOMEM; | |
996 | err = split_state(tree, state, prealloc, start); | |
997 | BUG_ON(err == -EEXIST); | |
998 | prealloc = NULL; | |
999 | if (err) | |
1000 | goto out; | |
1001 | if (state->end <= end) { | |
1002 | set_state_bits(tree, state, &bits); | |
1003 | clear_state_bit(tree, state, &clear_bits, 0); | |
1004 | merge_state(tree, state); | |
1005 | if (last_end == (u64)-1) | |
1006 | goto out; | |
1007 | start = last_end + 1; | |
1008 | } | |
1009 | goto search_again; | |
1010 | } | |
1011 | /* | |
1012 | * | ---- desired range ---- | | |
1013 | * | state | or | state | | |
1014 | * | |
1015 | * There's a hole, we need to insert something in it and | |
1016 | * ignore the extent we found. | |
1017 | */ | |
1018 | if (state->start > start) { | |
1019 | u64 this_end; | |
1020 | if (end < last_start) | |
1021 | this_end = end; | |
1022 | else | |
1023 | this_end = last_start - 1; | |
1024 | ||
1025 | prealloc = alloc_extent_state_atomic(prealloc); | |
1026 | if (!prealloc) | |
1027 | return -ENOMEM; | |
1028 | ||
1029 | /* | |
1030 | * Avoid to free 'prealloc' if it can be merged with | |
1031 | * the later extent. | |
1032 | */ | |
1033 | err = insert_state(tree, prealloc, start, this_end, | |
1034 | &bits); | |
1035 | BUG_ON(err == -EEXIST); | |
1036 | if (err) { | |
1037 | free_extent_state(prealloc); | |
1038 | prealloc = NULL; | |
1039 | goto out; | |
1040 | } | |
1041 | prealloc = NULL; | |
1042 | start = this_end + 1; | |
1043 | goto search_again; | |
1044 | } | |
1045 | /* | |
1046 | * | ---- desired range ---- | | |
1047 | * | state | | |
1048 | * We need to split the extent, and set the bit | |
1049 | * on the first half | |
1050 | */ | |
1051 | if (state->start <= end && state->end > end) { | |
1052 | prealloc = alloc_extent_state_atomic(prealloc); | |
1053 | if (!prealloc) | |
1054 | return -ENOMEM; | |
1055 | ||
1056 | err = split_state(tree, state, prealloc, end + 1); | |
1057 | BUG_ON(err == -EEXIST); | |
1058 | ||
1059 | set_state_bits(tree, prealloc, &bits); | |
1060 | clear_state_bit(tree, prealloc, &clear_bits, 0); | |
1061 | ||
1062 | merge_state(tree, prealloc); | |
1063 | prealloc = NULL; | |
1064 | goto out; | |
1065 | } | |
1066 | ||
1067 | goto search_again; | |
1068 | ||
1069 | out: | |
1070 | spin_unlock(&tree->lock); | |
1071 | if (prealloc) | |
1072 | free_extent_state(prealloc); | |
1073 | ||
1074 | return err; | |
1075 | ||
1076 | search_again: | |
1077 | if (start > end) | |
1078 | goto out; | |
1079 | spin_unlock(&tree->lock); | |
1080 | if (mask & __GFP_WAIT) | |
1081 | cond_resched(); | |
1082 | goto again; | |
1083 | } | |
1084 | ||
d1310b2e CM |
1085 | /* wrappers around set/clear extent bit */ |
1086 | int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
1087 | gfp_t mask) | |
1088 | { | |
1089 | return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL, | |
2c64c53d | 1090 | NULL, mask); |
d1310b2e | 1091 | } |
d1310b2e CM |
1092 | |
1093 | int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
1094 | int bits, gfp_t mask) | |
1095 | { | |
1096 | return set_extent_bit(tree, start, end, bits, 0, NULL, | |
2c64c53d | 1097 | NULL, mask); |
d1310b2e | 1098 | } |
d1310b2e CM |
1099 | |
1100 | int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
1101 | int bits, gfp_t mask) | |
1102 | { | |
2c64c53d | 1103 | return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask); |
d1310b2e | 1104 | } |
d1310b2e CM |
1105 | |
1106 | int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, | |
2ac55d41 | 1107 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
1108 | { |
1109 | return set_extent_bit(tree, start, end, | |
fee187d9 | 1110 | EXTENT_DELALLOC | EXTENT_UPTODATE, |
2ac55d41 | 1111 | 0, NULL, cached_state, mask); |
d1310b2e | 1112 | } |
d1310b2e CM |
1113 | |
1114 | int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
1115 | gfp_t mask) | |
1116 | { | |
1117 | return clear_extent_bit(tree, start, end, | |
32c00aff | 1118 | EXTENT_DIRTY | EXTENT_DELALLOC | |
0ca1f7ce | 1119 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask); |
d1310b2e | 1120 | } |
d1310b2e CM |
1121 | |
1122 | int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, | |
1123 | gfp_t mask) | |
1124 | { | |
1125 | return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL, | |
2c64c53d | 1126 | NULL, mask); |
d1310b2e | 1127 | } |
d1310b2e | 1128 | |
d1310b2e | 1129 | int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
507903b8 | 1130 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e | 1131 | { |
507903b8 AJ |
1132 | return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, |
1133 | NULL, cached_state, mask); | |
d1310b2e | 1134 | } |
d1310b2e | 1135 | |
d397712b | 1136 | static int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, |
2ac55d41 JB |
1137 | u64 end, struct extent_state **cached_state, |
1138 | gfp_t mask) | |
d1310b2e | 1139 | { |
2c64c53d | 1140 | return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, |
2ac55d41 | 1141 | cached_state, mask); |
d1310b2e | 1142 | } |
d1310b2e | 1143 | |
d352ac68 CM |
1144 | /* |
1145 | * either insert or lock state struct between start and end use mask to tell | |
1146 | * us if waiting is desired. | |
1147 | */ | |
1edbb734 | 1148 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
2c64c53d | 1149 | int bits, struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
1150 | { |
1151 | int err; | |
1152 | u64 failed_start; | |
1153 | while (1) { | |
1edbb734 | 1154 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED | bits, |
2c64c53d CM |
1155 | EXTENT_LOCKED, &failed_start, |
1156 | cached_state, mask); | |
d1310b2e CM |
1157 | if (err == -EEXIST && (mask & __GFP_WAIT)) { |
1158 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); | |
1159 | start = failed_start; | |
1160 | } else { | |
1161 | break; | |
1162 | } | |
1163 | WARN_ON(start > end); | |
1164 | } | |
1165 | return err; | |
1166 | } | |
d1310b2e | 1167 | |
1edbb734 CM |
1168 | int lock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) |
1169 | { | |
2c64c53d | 1170 | return lock_extent_bits(tree, start, end, 0, NULL, mask); |
1edbb734 CM |
1171 | } |
1172 | ||
25179201 JB |
1173 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end, |
1174 | gfp_t mask) | |
1175 | { | |
1176 | int err; | |
1177 | u64 failed_start; | |
1178 | ||
2c64c53d CM |
1179 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
1180 | &failed_start, NULL, mask); | |
6643558d YZ |
1181 | if (err == -EEXIST) { |
1182 | if (failed_start > start) | |
1183 | clear_extent_bit(tree, start, failed_start - 1, | |
2c64c53d | 1184 | EXTENT_LOCKED, 1, 0, NULL, mask); |
25179201 | 1185 | return 0; |
6643558d | 1186 | } |
25179201 JB |
1187 | return 1; |
1188 | } | |
25179201 | 1189 | |
2c64c53d CM |
1190 | int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end, |
1191 | struct extent_state **cached, gfp_t mask) | |
1192 | { | |
1193 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, | |
1194 | mask); | |
1195 | } | |
1196 | ||
507903b8 | 1197 | int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end, gfp_t mask) |
d1310b2e | 1198 | { |
2c64c53d CM |
1199 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL, |
1200 | mask); | |
d1310b2e | 1201 | } |
d1310b2e | 1202 | |
d1310b2e CM |
1203 | /* |
1204 | * helper function to set both pages and extents in the tree writeback | |
1205 | */ | |
b2950863 | 1206 | static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e CM |
1207 | { |
1208 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1209 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1210 | struct page *page; | |
1211 | ||
1212 | while (index <= end_index) { | |
1213 | page = find_get_page(tree->mapping, index); | |
1214 | BUG_ON(!page); | |
1215 | set_page_writeback(page); | |
1216 | page_cache_release(page); | |
1217 | index++; | |
1218 | } | |
d1310b2e CM |
1219 | return 0; |
1220 | } | |
d1310b2e | 1221 | |
d352ac68 CM |
1222 | /* find the first state struct with 'bits' set after 'start', and |
1223 | * return it. tree->lock must be held. NULL will returned if | |
1224 | * nothing was found after 'start' | |
1225 | */ | |
d7fc640e CM |
1226 | struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree, |
1227 | u64 start, int bits) | |
1228 | { | |
1229 | struct rb_node *node; | |
1230 | struct extent_state *state; | |
1231 | ||
1232 | /* | |
1233 | * this search will find all the extents that end after | |
1234 | * our range starts. | |
1235 | */ | |
1236 | node = tree_search(tree, start); | |
d397712b | 1237 | if (!node) |
d7fc640e | 1238 | goto out; |
d7fc640e | 1239 | |
d397712b | 1240 | while (1) { |
d7fc640e | 1241 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1242 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1243 | return state; |
d397712b | 1244 | |
d7fc640e CM |
1245 | node = rb_next(node); |
1246 | if (!node) | |
1247 | break; | |
1248 | } | |
1249 | out: | |
1250 | return NULL; | |
1251 | } | |
d7fc640e | 1252 | |
69261c4b XG |
1253 | /* |
1254 | * find the first offset in the io tree with 'bits' set. zero is | |
1255 | * returned if we find something, and *start_ret and *end_ret are | |
1256 | * set to reflect the state struct that was found. | |
1257 | * | |
1258 | * If nothing was found, 1 is returned, < 0 on error | |
1259 | */ | |
1260 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
1261 | u64 *start_ret, u64 *end_ret, int bits) | |
1262 | { | |
1263 | struct extent_state *state; | |
1264 | int ret = 1; | |
1265 | ||
1266 | spin_lock(&tree->lock); | |
1267 | state = find_first_extent_bit_state(tree, start, bits); | |
1268 | if (state) { | |
1269 | *start_ret = state->start; | |
1270 | *end_ret = state->end; | |
1271 | ret = 0; | |
1272 | } | |
1273 | spin_unlock(&tree->lock); | |
1274 | return ret; | |
1275 | } | |
1276 | ||
d352ac68 CM |
1277 | /* |
1278 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1279 | * more than 'max_bytes'. start and end are used to return the range, | |
1280 | * | |
1281 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1282 | */ | |
c8b97818 | 1283 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1284 | u64 *start, u64 *end, u64 max_bytes, |
1285 | struct extent_state **cached_state) | |
d1310b2e CM |
1286 | { |
1287 | struct rb_node *node; | |
1288 | struct extent_state *state; | |
1289 | u64 cur_start = *start; | |
1290 | u64 found = 0; | |
1291 | u64 total_bytes = 0; | |
1292 | ||
cad321ad | 1293 | spin_lock(&tree->lock); |
c8b97818 | 1294 | |
d1310b2e CM |
1295 | /* |
1296 | * this search will find all the extents that end after | |
1297 | * our range starts. | |
1298 | */ | |
80ea96b1 | 1299 | node = tree_search(tree, cur_start); |
2b114d1d | 1300 | if (!node) { |
3b951516 CM |
1301 | if (!found) |
1302 | *end = (u64)-1; | |
d1310b2e CM |
1303 | goto out; |
1304 | } | |
1305 | ||
d397712b | 1306 | while (1) { |
d1310b2e | 1307 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1308 | if (found && (state->start != cur_start || |
1309 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1310 | goto out; |
1311 | } | |
1312 | if (!(state->state & EXTENT_DELALLOC)) { | |
1313 | if (!found) | |
1314 | *end = state->end; | |
1315 | goto out; | |
1316 | } | |
c2a128d2 | 1317 | if (!found) { |
d1310b2e | 1318 | *start = state->start; |
c2a128d2 JB |
1319 | *cached_state = state; |
1320 | atomic_inc(&state->refs); | |
1321 | } | |
d1310b2e CM |
1322 | found++; |
1323 | *end = state->end; | |
1324 | cur_start = state->end + 1; | |
1325 | node = rb_next(node); | |
1326 | if (!node) | |
1327 | break; | |
1328 | total_bytes += state->end - state->start + 1; | |
1329 | if (total_bytes >= max_bytes) | |
1330 | break; | |
1331 | } | |
1332 | out: | |
cad321ad | 1333 | spin_unlock(&tree->lock); |
d1310b2e CM |
1334 | return found; |
1335 | } | |
1336 | ||
c8b97818 CM |
1337 | static noinline int __unlock_for_delalloc(struct inode *inode, |
1338 | struct page *locked_page, | |
1339 | u64 start, u64 end) | |
1340 | { | |
1341 | int ret; | |
1342 | struct page *pages[16]; | |
1343 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1344 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1345 | unsigned long nr_pages = end_index - index + 1; | |
1346 | int i; | |
1347 | ||
1348 | if (index == locked_page->index && end_index == index) | |
1349 | return 0; | |
1350 | ||
d397712b | 1351 | while (nr_pages > 0) { |
c8b97818 | 1352 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1353 | min_t(unsigned long, nr_pages, |
1354 | ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1355 | for (i = 0; i < ret; i++) { |
1356 | if (pages[i] != locked_page) | |
1357 | unlock_page(pages[i]); | |
1358 | page_cache_release(pages[i]); | |
1359 | } | |
1360 | nr_pages -= ret; | |
1361 | index += ret; | |
1362 | cond_resched(); | |
1363 | } | |
1364 | return 0; | |
1365 | } | |
1366 | ||
1367 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1368 | struct page *locked_page, | |
1369 | u64 delalloc_start, | |
1370 | u64 delalloc_end) | |
1371 | { | |
1372 | unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT; | |
1373 | unsigned long start_index = index; | |
1374 | unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT; | |
1375 | unsigned long pages_locked = 0; | |
1376 | struct page *pages[16]; | |
1377 | unsigned long nrpages; | |
1378 | int ret; | |
1379 | int i; | |
1380 | ||
1381 | /* the caller is responsible for locking the start index */ | |
1382 | if (index == locked_page->index && index == end_index) | |
1383 | return 0; | |
1384 | ||
1385 | /* skip the page at the start index */ | |
1386 | nrpages = end_index - index + 1; | |
d397712b | 1387 | while (nrpages > 0) { |
c8b97818 | 1388 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1389 | min_t(unsigned long, |
1390 | nrpages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1391 | if (ret == 0) { |
1392 | ret = -EAGAIN; | |
1393 | goto done; | |
1394 | } | |
1395 | /* now we have an array of pages, lock them all */ | |
1396 | for (i = 0; i < ret; i++) { | |
1397 | /* | |
1398 | * the caller is taking responsibility for | |
1399 | * locked_page | |
1400 | */ | |
771ed689 | 1401 | if (pages[i] != locked_page) { |
c8b97818 | 1402 | lock_page(pages[i]); |
f2b1c41c CM |
1403 | if (!PageDirty(pages[i]) || |
1404 | pages[i]->mapping != inode->i_mapping) { | |
771ed689 CM |
1405 | ret = -EAGAIN; |
1406 | unlock_page(pages[i]); | |
1407 | page_cache_release(pages[i]); | |
1408 | goto done; | |
1409 | } | |
1410 | } | |
c8b97818 | 1411 | page_cache_release(pages[i]); |
771ed689 | 1412 | pages_locked++; |
c8b97818 | 1413 | } |
c8b97818 CM |
1414 | nrpages -= ret; |
1415 | index += ret; | |
1416 | cond_resched(); | |
1417 | } | |
1418 | ret = 0; | |
1419 | done: | |
1420 | if (ret && pages_locked) { | |
1421 | __unlock_for_delalloc(inode, locked_page, | |
1422 | delalloc_start, | |
1423 | ((u64)(start_index + pages_locked - 1)) << | |
1424 | PAGE_CACHE_SHIFT); | |
1425 | } | |
1426 | return ret; | |
1427 | } | |
1428 | ||
1429 | /* | |
1430 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1431 | * more than 'max_bytes'. start and end are used to return the range, | |
1432 | * | |
1433 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1434 | */ | |
1435 | static noinline u64 find_lock_delalloc_range(struct inode *inode, | |
1436 | struct extent_io_tree *tree, | |
1437 | struct page *locked_page, | |
1438 | u64 *start, u64 *end, | |
1439 | u64 max_bytes) | |
1440 | { | |
1441 | u64 delalloc_start; | |
1442 | u64 delalloc_end; | |
1443 | u64 found; | |
9655d298 | 1444 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1445 | int ret; |
1446 | int loops = 0; | |
1447 | ||
1448 | again: | |
1449 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1450 | delalloc_start = *start; | |
1451 | delalloc_end = 0; | |
1452 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1453 | max_bytes, &cached_state); |
70b99e69 | 1454 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1455 | *start = delalloc_start; |
1456 | *end = delalloc_end; | |
c2a128d2 | 1457 | free_extent_state(cached_state); |
c8b97818 CM |
1458 | return found; |
1459 | } | |
1460 | ||
70b99e69 CM |
1461 | /* |
1462 | * start comes from the offset of locked_page. We have to lock | |
1463 | * pages in order, so we can't process delalloc bytes before | |
1464 | * locked_page | |
1465 | */ | |
d397712b | 1466 | if (delalloc_start < *start) |
70b99e69 | 1467 | delalloc_start = *start; |
70b99e69 | 1468 | |
c8b97818 CM |
1469 | /* |
1470 | * make sure to limit the number of pages we try to lock down | |
1471 | * if we're looping. | |
1472 | */ | |
d397712b | 1473 | if (delalloc_end + 1 - delalloc_start > max_bytes && loops) |
771ed689 | 1474 | delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1; |
d397712b | 1475 | |
c8b97818 CM |
1476 | /* step two, lock all the pages after the page that has start */ |
1477 | ret = lock_delalloc_pages(inode, locked_page, | |
1478 | delalloc_start, delalloc_end); | |
1479 | if (ret == -EAGAIN) { | |
1480 | /* some of the pages are gone, lets avoid looping by | |
1481 | * shortening the size of the delalloc range we're searching | |
1482 | */ | |
9655d298 | 1483 | free_extent_state(cached_state); |
c8b97818 CM |
1484 | if (!loops) { |
1485 | unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1); | |
1486 | max_bytes = PAGE_CACHE_SIZE - offset; | |
1487 | loops = 1; | |
1488 | goto again; | |
1489 | } else { | |
1490 | found = 0; | |
1491 | goto out_failed; | |
1492 | } | |
1493 | } | |
1494 | BUG_ON(ret); | |
1495 | ||
1496 | /* step three, lock the state bits for the whole range */ | |
9655d298 CM |
1497 | lock_extent_bits(tree, delalloc_start, delalloc_end, |
1498 | 0, &cached_state, GFP_NOFS); | |
c8b97818 CM |
1499 | |
1500 | /* then test to make sure it is all still delalloc */ | |
1501 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1502 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1503 | if (!ret) { |
9655d298 CM |
1504 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
1505 | &cached_state, GFP_NOFS); | |
c8b97818 CM |
1506 | __unlock_for_delalloc(inode, locked_page, |
1507 | delalloc_start, delalloc_end); | |
1508 | cond_resched(); | |
1509 | goto again; | |
1510 | } | |
9655d298 | 1511 | free_extent_state(cached_state); |
c8b97818 CM |
1512 | *start = delalloc_start; |
1513 | *end = delalloc_end; | |
1514 | out_failed: | |
1515 | return found; | |
1516 | } | |
1517 | ||
1518 | int extent_clear_unlock_delalloc(struct inode *inode, | |
1519 | struct extent_io_tree *tree, | |
1520 | u64 start, u64 end, struct page *locked_page, | |
a791e35e | 1521 | unsigned long op) |
c8b97818 CM |
1522 | { |
1523 | int ret; | |
1524 | struct page *pages[16]; | |
1525 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1526 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1527 | unsigned long nr_pages = end_index - index + 1; | |
1528 | int i; | |
771ed689 | 1529 | int clear_bits = 0; |
c8b97818 | 1530 | |
a791e35e | 1531 | if (op & EXTENT_CLEAR_UNLOCK) |
771ed689 | 1532 | clear_bits |= EXTENT_LOCKED; |
a791e35e | 1533 | if (op & EXTENT_CLEAR_DIRTY) |
c8b97818 CM |
1534 | clear_bits |= EXTENT_DIRTY; |
1535 | ||
a791e35e | 1536 | if (op & EXTENT_CLEAR_DELALLOC) |
771ed689 CM |
1537 | clear_bits |= EXTENT_DELALLOC; |
1538 | ||
2c64c53d | 1539 | clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); |
32c00aff JB |
1540 | if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | |
1541 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK | | |
1542 | EXTENT_SET_PRIVATE2))) | |
771ed689 | 1543 | return 0; |
c8b97818 | 1544 | |
d397712b | 1545 | while (nr_pages > 0) { |
c8b97818 | 1546 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1547 | min_t(unsigned long, |
1548 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 | 1549 | for (i = 0; i < ret; i++) { |
8b62b72b | 1550 | |
a791e35e | 1551 | if (op & EXTENT_SET_PRIVATE2) |
8b62b72b CM |
1552 | SetPagePrivate2(pages[i]); |
1553 | ||
c8b97818 CM |
1554 | if (pages[i] == locked_page) { |
1555 | page_cache_release(pages[i]); | |
1556 | continue; | |
1557 | } | |
a791e35e | 1558 | if (op & EXTENT_CLEAR_DIRTY) |
c8b97818 | 1559 | clear_page_dirty_for_io(pages[i]); |
a791e35e | 1560 | if (op & EXTENT_SET_WRITEBACK) |
c8b97818 | 1561 | set_page_writeback(pages[i]); |
a791e35e | 1562 | if (op & EXTENT_END_WRITEBACK) |
c8b97818 | 1563 | end_page_writeback(pages[i]); |
a791e35e | 1564 | if (op & EXTENT_CLEAR_UNLOCK_PAGE) |
771ed689 | 1565 | unlock_page(pages[i]); |
c8b97818 CM |
1566 | page_cache_release(pages[i]); |
1567 | } | |
1568 | nr_pages -= ret; | |
1569 | index += ret; | |
1570 | cond_resched(); | |
1571 | } | |
1572 | return 0; | |
1573 | } | |
c8b97818 | 1574 | |
d352ac68 CM |
1575 | /* |
1576 | * count the number of bytes in the tree that have a given bit(s) | |
1577 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1578 | * cached. The total number found is returned. | |
1579 | */ | |
d1310b2e CM |
1580 | u64 count_range_bits(struct extent_io_tree *tree, |
1581 | u64 *start, u64 search_end, u64 max_bytes, | |
ec29ed5b | 1582 | unsigned long bits, int contig) |
d1310b2e CM |
1583 | { |
1584 | struct rb_node *node; | |
1585 | struct extent_state *state; | |
1586 | u64 cur_start = *start; | |
1587 | u64 total_bytes = 0; | |
ec29ed5b | 1588 | u64 last = 0; |
d1310b2e CM |
1589 | int found = 0; |
1590 | ||
1591 | if (search_end <= cur_start) { | |
d1310b2e CM |
1592 | WARN_ON(1); |
1593 | return 0; | |
1594 | } | |
1595 | ||
cad321ad | 1596 | spin_lock(&tree->lock); |
d1310b2e CM |
1597 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1598 | total_bytes = tree->dirty_bytes; | |
1599 | goto out; | |
1600 | } | |
1601 | /* | |
1602 | * this search will find all the extents that end after | |
1603 | * our range starts. | |
1604 | */ | |
80ea96b1 | 1605 | node = tree_search(tree, cur_start); |
d397712b | 1606 | if (!node) |
d1310b2e | 1607 | goto out; |
d1310b2e | 1608 | |
d397712b | 1609 | while (1) { |
d1310b2e CM |
1610 | state = rb_entry(node, struct extent_state, rb_node); |
1611 | if (state->start > search_end) | |
1612 | break; | |
ec29ed5b CM |
1613 | if (contig && found && state->start > last + 1) |
1614 | break; | |
1615 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1616 | total_bytes += min(search_end, state->end) + 1 - |
1617 | max(cur_start, state->start); | |
1618 | if (total_bytes >= max_bytes) | |
1619 | break; | |
1620 | if (!found) { | |
af60bed2 | 1621 | *start = max(cur_start, state->start); |
d1310b2e CM |
1622 | found = 1; |
1623 | } | |
ec29ed5b CM |
1624 | last = state->end; |
1625 | } else if (contig && found) { | |
1626 | break; | |
d1310b2e CM |
1627 | } |
1628 | node = rb_next(node); | |
1629 | if (!node) | |
1630 | break; | |
1631 | } | |
1632 | out: | |
cad321ad | 1633 | spin_unlock(&tree->lock); |
d1310b2e CM |
1634 | return total_bytes; |
1635 | } | |
b2950863 | 1636 | |
d352ac68 CM |
1637 | /* |
1638 | * set the private field for a given byte offset in the tree. If there isn't | |
1639 | * an extent_state there already, this does nothing. | |
1640 | */ | |
d1310b2e CM |
1641 | int set_state_private(struct extent_io_tree *tree, u64 start, u64 private) |
1642 | { | |
1643 | struct rb_node *node; | |
1644 | struct extent_state *state; | |
1645 | int ret = 0; | |
1646 | ||
cad321ad | 1647 | spin_lock(&tree->lock); |
d1310b2e CM |
1648 | /* |
1649 | * this search will find all the extents that end after | |
1650 | * our range starts. | |
1651 | */ | |
80ea96b1 | 1652 | node = tree_search(tree, start); |
2b114d1d | 1653 | if (!node) { |
d1310b2e CM |
1654 | ret = -ENOENT; |
1655 | goto out; | |
1656 | } | |
1657 | state = rb_entry(node, struct extent_state, rb_node); | |
1658 | if (state->start != start) { | |
1659 | ret = -ENOENT; | |
1660 | goto out; | |
1661 | } | |
1662 | state->private = private; | |
1663 | out: | |
cad321ad | 1664 | spin_unlock(&tree->lock); |
d1310b2e CM |
1665 | return ret; |
1666 | } | |
1667 | ||
1668 | int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private) | |
1669 | { | |
1670 | struct rb_node *node; | |
1671 | struct extent_state *state; | |
1672 | int ret = 0; | |
1673 | ||
cad321ad | 1674 | spin_lock(&tree->lock); |
d1310b2e CM |
1675 | /* |
1676 | * this search will find all the extents that end after | |
1677 | * our range starts. | |
1678 | */ | |
80ea96b1 | 1679 | node = tree_search(tree, start); |
2b114d1d | 1680 | if (!node) { |
d1310b2e CM |
1681 | ret = -ENOENT; |
1682 | goto out; | |
1683 | } | |
1684 | state = rb_entry(node, struct extent_state, rb_node); | |
1685 | if (state->start != start) { | |
1686 | ret = -ENOENT; | |
1687 | goto out; | |
1688 | } | |
1689 | *private = state->private; | |
1690 | out: | |
cad321ad | 1691 | spin_unlock(&tree->lock); |
d1310b2e CM |
1692 | return ret; |
1693 | } | |
1694 | ||
1695 | /* | |
1696 | * searches a range in the state tree for a given mask. | |
70dec807 | 1697 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1698 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1699 | * range is found set. | |
1700 | */ | |
1701 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9655d298 | 1702 | int bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1703 | { |
1704 | struct extent_state *state = NULL; | |
1705 | struct rb_node *node; | |
1706 | int bitset = 0; | |
d1310b2e | 1707 | |
cad321ad | 1708 | spin_lock(&tree->lock); |
df98b6e2 JB |
1709 | if (cached && cached->tree && cached->start <= start && |
1710 | cached->end > start) | |
9655d298 CM |
1711 | node = &cached->rb_node; |
1712 | else | |
1713 | node = tree_search(tree, start); | |
d1310b2e CM |
1714 | while (node && start <= end) { |
1715 | state = rb_entry(node, struct extent_state, rb_node); | |
1716 | ||
1717 | if (filled && state->start > start) { | |
1718 | bitset = 0; | |
1719 | break; | |
1720 | } | |
1721 | ||
1722 | if (state->start > end) | |
1723 | break; | |
1724 | ||
1725 | if (state->state & bits) { | |
1726 | bitset = 1; | |
1727 | if (!filled) | |
1728 | break; | |
1729 | } else if (filled) { | |
1730 | bitset = 0; | |
1731 | break; | |
1732 | } | |
46562cec CM |
1733 | |
1734 | if (state->end == (u64)-1) | |
1735 | break; | |
1736 | ||
d1310b2e CM |
1737 | start = state->end + 1; |
1738 | if (start > end) | |
1739 | break; | |
1740 | node = rb_next(node); | |
1741 | if (!node) { | |
1742 | if (filled) | |
1743 | bitset = 0; | |
1744 | break; | |
1745 | } | |
1746 | } | |
cad321ad | 1747 | spin_unlock(&tree->lock); |
d1310b2e CM |
1748 | return bitset; |
1749 | } | |
d1310b2e CM |
1750 | |
1751 | /* | |
1752 | * helper function to set a given page up to date if all the | |
1753 | * extents in the tree for that page are up to date | |
1754 | */ | |
1755 | static int check_page_uptodate(struct extent_io_tree *tree, | |
1756 | struct page *page) | |
1757 | { | |
1758 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1759 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
9655d298 | 1760 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e CM |
1761 | SetPageUptodate(page); |
1762 | return 0; | |
1763 | } | |
1764 | ||
1765 | /* | |
1766 | * helper function to unlock a page if all the extents in the tree | |
1767 | * for that page are unlocked | |
1768 | */ | |
1769 | static int check_page_locked(struct extent_io_tree *tree, | |
1770 | struct page *page) | |
1771 | { | |
1772 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1773 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
9655d298 | 1774 | if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) |
d1310b2e CM |
1775 | unlock_page(page); |
1776 | return 0; | |
1777 | } | |
1778 | ||
1779 | /* | |
1780 | * helper function to end page writeback if all the extents | |
1781 | * in the tree for that page are done with writeback | |
1782 | */ | |
1783 | static int check_page_writeback(struct extent_io_tree *tree, | |
1784 | struct page *page) | |
1785 | { | |
1edbb734 | 1786 | end_page_writeback(page); |
d1310b2e CM |
1787 | return 0; |
1788 | } | |
1789 | ||
1790 | /* lots and lots of room for performance fixes in the end_bio funcs */ | |
1791 | ||
1792 | /* | |
1793 | * after a writepage IO is done, we need to: | |
1794 | * clear the uptodate bits on error | |
1795 | * clear the writeback bits in the extent tree for this IO | |
1796 | * end_page_writeback if the page has no more pending IO | |
1797 | * | |
1798 | * Scheduling is not allowed, so the extent state tree is expected | |
1799 | * to have one and only one object corresponding to this IO. | |
1800 | */ | |
d1310b2e | 1801 | static void end_bio_extent_writepage(struct bio *bio, int err) |
d1310b2e | 1802 | { |
1259ab75 | 1803 | int uptodate = err == 0; |
d1310b2e | 1804 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
902b22f3 | 1805 | struct extent_io_tree *tree; |
d1310b2e CM |
1806 | u64 start; |
1807 | u64 end; | |
1808 | int whole_page; | |
1259ab75 | 1809 | int ret; |
d1310b2e | 1810 | |
d1310b2e CM |
1811 | do { |
1812 | struct page *page = bvec->bv_page; | |
902b22f3 DW |
1813 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
1814 | ||
d1310b2e CM |
1815 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
1816 | bvec->bv_offset; | |
1817 | end = start + bvec->bv_len - 1; | |
1818 | ||
1819 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
1820 | whole_page = 1; | |
1821 | else | |
1822 | whole_page = 0; | |
1823 | ||
1824 | if (--bvec >= bio->bi_io_vec) | |
1825 | prefetchw(&bvec->bv_page->flags); | |
1259ab75 CM |
1826 | if (tree->ops && tree->ops->writepage_end_io_hook) { |
1827 | ret = tree->ops->writepage_end_io_hook(page, start, | |
902b22f3 | 1828 | end, NULL, uptodate); |
1259ab75 CM |
1829 | if (ret) |
1830 | uptodate = 0; | |
1831 | } | |
1832 | ||
1833 | if (!uptodate && tree->ops && | |
1834 | tree->ops->writepage_io_failed_hook) { | |
1835 | ret = tree->ops->writepage_io_failed_hook(bio, page, | |
902b22f3 | 1836 | start, end, NULL); |
1259ab75 | 1837 | if (ret == 0) { |
1259ab75 CM |
1838 | uptodate = (err == 0); |
1839 | continue; | |
1840 | } | |
1841 | } | |
1842 | ||
d1310b2e | 1843 | if (!uptodate) { |
2ac55d41 | 1844 | clear_extent_uptodate(tree, start, end, NULL, GFP_NOFS); |
d1310b2e CM |
1845 | ClearPageUptodate(page); |
1846 | SetPageError(page); | |
1847 | } | |
70dec807 | 1848 | |
d1310b2e CM |
1849 | if (whole_page) |
1850 | end_page_writeback(page); | |
1851 | else | |
1852 | check_page_writeback(tree, page); | |
d1310b2e | 1853 | } while (bvec >= bio->bi_io_vec); |
2b1f55b0 | 1854 | |
d1310b2e | 1855 | bio_put(bio); |
d1310b2e CM |
1856 | } |
1857 | ||
1858 | /* | |
1859 | * after a readpage IO is done, we need to: | |
1860 | * clear the uptodate bits on error | |
1861 | * set the uptodate bits if things worked | |
1862 | * set the page up to date if all extents in the tree are uptodate | |
1863 | * clear the lock bit in the extent tree | |
1864 | * unlock the page if there are no other extents locked for it | |
1865 | * | |
1866 | * Scheduling is not allowed, so the extent state tree is expected | |
1867 | * to have one and only one object corresponding to this IO. | |
1868 | */ | |
d1310b2e | 1869 | static void end_bio_extent_readpage(struct bio *bio, int err) |
d1310b2e CM |
1870 | { |
1871 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
4125bf76 CM |
1872 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
1873 | struct bio_vec *bvec = bio->bi_io_vec; | |
902b22f3 | 1874 | struct extent_io_tree *tree; |
d1310b2e CM |
1875 | u64 start; |
1876 | u64 end; | |
1877 | int whole_page; | |
1878 | int ret; | |
1879 | ||
d20f7043 CM |
1880 | if (err) |
1881 | uptodate = 0; | |
1882 | ||
d1310b2e CM |
1883 | do { |
1884 | struct page *page = bvec->bv_page; | |
507903b8 AJ |
1885 | struct extent_state *cached = NULL; |
1886 | struct extent_state *state; | |
1887 | ||
902b22f3 DW |
1888 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
1889 | ||
d1310b2e CM |
1890 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
1891 | bvec->bv_offset; | |
1892 | end = start + bvec->bv_len - 1; | |
1893 | ||
1894 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
1895 | whole_page = 1; | |
1896 | else | |
1897 | whole_page = 0; | |
1898 | ||
4125bf76 | 1899 | if (++bvec <= bvec_end) |
d1310b2e CM |
1900 | prefetchw(&bvec->bv_page->flags); |
1901 | ||
507903b8 | 1902 | spin_lock(&tree->lock); |
0d399205 | 1903 | state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED); |
109b36a2 | 1904 | if (state && state->start == start) { |
507903b8 AJ |
1905 | /* |
1906 | * take a reference on the state, unlock will drop | |
1907 | * the ref | |
1908 | */ | |
1909 | cache_state(state, &cached); | |
1910 | } | |
1911 | spin_unlock(&tree->lock); | |
1912 | ||
d1310b2e | 1913 | if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) { |
70dec807 | 1914 | ret = tree->ops->readpage_end_io_hook(page, start, end, |
507903b8 | 1915 | state); |
d1310b2e CM |
1916 | if (ret) |
1917 | uptodate = 0; | |
1918 | } | |
7e38326f CM |
1919 | if (!uptodate && tree->ops && |
1920 | tree->ops->readpage_io_failed_hook) { | |
1921 | ret = tree->ops->readpage_io_failed_hook(bio, page, | |
902b22f3 | 1922 | start, end, NULL); |
7e38326f | 1923 | if (ret == 0) { |
3b951516 CM |
1924 | uptodate = |
1925 | test_bit(BIO_UPTODATE, &bio->bi_flags); | |
d20f7043 CM |
1926 | if (err) |
1927 | uptodate = 0; | |
507903b8 | 1928 | uncache_state(&cached); |
7e38326f CM |
1929 | continue; |
1930 | } | |
1931 | } | |
d1310b2e | 1932 | |
771ed689 | 1933 | if (uptodate) { |
507903b8 | 1934 | set_extent_uptodate(tree, start, end, &cached, |
902b22f3 | 1935 | GFP_ATOMIC); |
771ed689 | 1936 | } |
507903b8 | 1937 | unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC); |
d1310b2e | 1938 | |
70dec807 CM |
1939 | if (whole_page) { |
1940 | if (uptodate) { | |
1941 | SetPageUptodate(page); | |
1942 | } else { | |
1943 | ClearPageUptodate(page); | |
1944 | SetPageError(page); | |
1945 | } | |
d1310b2e | 1946 | unlock_page(page); |
70dec807 CM |
1947 | } else { |
1948 | if (uptodate) { | |
1949 | check_page_uptodate(tree, page); | |
1950 | } else { | |
1951 | ClearPageUptodate(page); | |
1952 | SetPageError(page); | |
1953 | } | |
d1310b2e | 1954 | check_page_locked(tree, page); |
70dec807 | 1955 | } |
4125bf76 | 1956 | } while (bvec <= bvec_end); |
d1310b2e CM |
1957 | |
1958 | bio_put(bio); | |
d1310b2e CM |
1959 | } |
1960 | ||
88f794ed MX |
1961 | struct bio * |
1962 | btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
1963 | gfp_t gfp_flags) | |
d1310b2e CM |
1964 | { |
1965 | struct bio *bio; | |
1966 | ||
1967 | bio = bio_alloc(gfp_flags, nr_vecs); | |
1968 | ||
1969 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
1970 | while (!bio && (nr_vecs /= 2)) | |
1971 | bio = bio_alloc(gfp_flags, nr_vecs); | |
1972 | } | |
1973 | ||
1974 | if (bio) { | |
e1c4b745 | 1975 | bio->bi_size = 0; |
d1310b2e CM |
1976 | bio->bi_bdev = bdev; |
1977 | bio->bi_sector = first_sector; | |
1978 | } | |
1979 | return bio; | |
1980 | } | |
1981 | ||
c8b97818 CM |
1982 | static int submit_one_bio(int rw, struct bio *bio, int mirror_num, |
1983 | unsigned long bio_flags) | |
d1310b2e | 1984 | { |
d1310b2e | 1985 | int ret = 0; |
70dec807 CM |
1986 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
1987 | struct page *page = bvec->bv_page; | |
1988 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 1989 | u64 start; |
70dec807 CM |
1990 | |
1991 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset; | |
70dec807 | 1992 | |
902b22f3 | 1993 | bio->bi_private = NULL; |
d1310b2e CM |
1994 | |
1995 | bio_get(bio); | |
1996 | ||
065631f6 | 1997 | if (tree->ops && tree->ops->submit_bio_hook) |
6b82ce8d | 1998 | ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio, |
eaf25d93 | 1999 | mirror_num, bio_flags, start); |
0b86a832 CM |
2000 | else |
2001 | submit_bio(rw, bio); | |
d1310b2e CM |
2002 | if (bio_flagged(bio, BIO_EOPNOTSUPP)) |
2003 | ret = -EOPNOTSUPP; | |
2004 | bio_put(bio); | |
2005 | return ret; | |
2006 | } | |
2007 | ||
2008 | static int submit_extent_page(int rw, struct extent_io_tree *tree, | |
2009 | struct page *page, sector_t sector, | |
2010 | size_t size, unsigned long offset, | |
2011 | struct block_device *bdev, | |
2012 | struct bio **bio_ret, | |
2013 | unsigned long max_pages, | |
f188591e | 2014 | bio_end_io_t end_io_func, |
c8b97818 CM |
2015 | int mirror_num, |
2016 | unsigned long prev_bio_flags, | |
2017 | unsigned long bio_flags) | |
d1310b2e CM |
2018 | { |
2019 | int ret = 0; | |
2020 | struct bio *bio; | |
2021 | int nr; | |
c8b97818 CM |
2022 | int contig = 0; |
2023 | int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED; | |
2024 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; | |
5b050f04 | 2025 | size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE); |
d1310b2e CM |
2026 | |
2027 | if (bio_ret && *bio_ret) { | |
2028 | bio = *bio_ret; | |
c8b97818 CM |
2029 | if (old_compressed) |
2030 | contig = bio->bi_sector == sector; | |
2031 | else | |
2032 | contig = bio->bi_sector + (bio->bi_size >> 9) == | |
2033 | sector; | |
2034 | ||
2035 | if (prev_bio_flags != bio_flags || !contig || | |
239b14b3 | 2036 | (tree->ops && tree->ops->merge_bio_hook && |
c8b97818 CM |
2037 | tree->ops->merge_bio_hook(page, offset, page_size, bio, |
2038 | bio_flags)) || | |
2039 | bio_add_page(bio, page, page_size, offset) < page_size) { | |
2040 | ret = submit_one_bio(rw, bio, mirror_num, | |
2041 | prev_bio_flags); | |
d1310b2e CM |
2042 | bio = NULL; |
2043 | } else { | |
2044 | return 0; | |
2045 | } | |
2046 | } | |
c8b97818 CM |
2047 | if (this_compressed) |
2048 | nr = BIO_MAX_PAGES; | |
2049 | else | |
2050 | nr = bio_get_nr_vecs(bdev); | |
2051 | ||
88f794ed | 2052 | bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH); |
5df67083 TI |
2053 | if (!bio) |
2054 | return -ENOMEM; | |
70dec807 | 2055 | |
c8b97818 | 2056 | bio_add_page(bio, page, page_size, offset); |
d1310b2e CM |
2057 | bio->bi_end_io = end_io_func; |
2058 | bio->bi_private = tree; | |
70dec807 | 2059 | |
d397712b | 2060 | if (bio_ret) |
d1310b2e | 2061 | *bio_ret = bio; |
d397712b | 2062 | else |
c8b97818 | 2063 | ret = submit_one_bio(rw, bio, mirror_num, bio_flags); |
d1310b2e CM |
2064 | |
2065 | return ret; | |
2066 | } | |
2067 | ||
2068 | void set_page_extent_mapped(struct page *page) | |
2069 | { | |
2070 | if (!PagePrivate(page)) { | |
2071 | SetPagePrivate(page); | |
d1310b2e | 2072 | page_cache_get(page); |
6af118ce | 2073 | set_page_private(page, EXTENT_PAGE_PRIVATE); |
d1310b2e CM |
2074 | } |
2075 | } | |
2076 | ||
b2950863 | 2077 | static void set_page_extent_head(struct page *page, unsigned long len) |
d1310b2e | 2078 | { |
eb14ab8e | 2079 | WARN_ON(!PagePrivate(page)); |
d1310b2e CM |
2080 | set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2); |
2081 | } | |
2082 | ||
2083 | /* | |
2084 | * basic readpage implementation. Locked extent state structs are inserted | |
2085 | * into the tree that are removed when the IO is done (by the end_io | |
2086 | * handlers) | |
2087 | */ | |
2088 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
2089 | struct page *page, | |
2090 | get_extent_t *get_extent, | |
c8b97818 CM |
2091 | struct bio **bio, int mirror_num, |
2092 | unsigned long *bio_flags) | |
d1310b2e CM |
2093 | { |
2094 | struct inode *inode = page->mapping->host; | |
2095 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2096 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2097 | u64 end; | |
2098 | u64 cur = start; | |
2099 | u64 extent_offset; | |
2100 | u64 last_byte = i_size_read(inode); | |
2101 | u64 block_start; | |
2102 | u64 cur_end; | |
2103 | sector_t sector; | |
2104 | struct extent_map *em; | |
2105 | struct block_device *bdev; | |
11c65dcc | 2106 | struct btrfs_ordered_extent *ordered; |
d1310b2e CM |
2107 | int ret; |
2108 | int nr = 0; | |
306e16ce | 2109 | size_t pg_offset = 0; |
d1310b2e | 2110 | size_t iosize; |
c8b97818 | 2111 | size_t disk_io_size; |
d1310b2e | 2112 | size_t blocksize = inode->i_sb->s_blocksize; |
c8b97818 | 2113 | unsigned long this_bio_flag = 0; |
d1310b2e CM |
2114 | |
2115 | set_page_extent_mapped(page); | |
2116 | ||
90a887c9 DM |
2117 | if (!PageUptodate(page)) { |
2118 | if (cleancache_get_page(page) == 0) { | |
2119 | BUG_ON(blocksize != PAGE_SIZE); | |
2120 | goto out; | |
2121 | } | |
2122 | } | |
2123 | ||
d1310b2e | 2124 | end = page_end; |
11c65dcc JB |
2125 | while (1) { |
2126 | lock_extent(tree, start, end, GFP_NOFS); | |
2127 | ordered = btrfs_lookup_ordered_extent(inode, start); | |
2128 | if (!ordered) | |
2129 | break; | |
2130 | unlock_extent(tree, start, end, GFP_NOFS); | |
2131 | btrfs_start_ordered_extent(inode, ordered, 1); | |
2132 | btrfs_put_ordered_extent(ordered); | |
2133 | } | |
d1310b2e | 2134 | |
c8b97818 CM |
2135 | if (page->index == last_byte >> PAGE_CACHE_SHIFT) { |
2136 | char *userpage; | |
2137 | size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1); | |
2138 | ||
2139 | if (zero_offset) { | |
2140 | iosize = PAGE_CACHE_SIZE - zero_offset; | |
2141 | userpage = kmap_atomic(page, KM_USER0); | |
2142 | memset(userpage + zero_offset, 0, iosize); | |
2143 | flush_dcache_page(page); | |
2144 | kunmap_atomic(userpage, KM_USER0); | |
2145 | } | |
2146 | } | |
d1310b2e CM |
2147 | while (cur <= end) { |
2148 | if (cur >= last_byte) { | |
2149 | char *userpage; | |
507903b8 AJ |
2150 | struct extent_state *cached = NULL; |
2151 | ||
306e16ce | 2152 | iosize = PAGE_CACHE_SIZE - pg_offset; |
d1310b2e | 2153 | userpage = kmap_atomic(page, KM_USER0); |
306e16ce | 2154 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e CM |
2155 | flush_dcache_page(page); |
2156 | kunmap_atomic(userpage, KM_USER0); | |
2157 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 AJ |
2158 | &cached, GFP_NOFS); |
2159 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
2160 | &cached, GFP_NOFS); | |
d1310b2e CM |
2161 | break; |
2162 | } | |
306e16ce | 2163 | em = get_extent(inode, page, pg_offset, cur, |
d1310b2e | 2164 | end - cur + 1, 0); |
c704005d | 2165 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e CM |
2166 | SetPageError(page); |
2167 | unlock_extent(tree, cur, end, GFP_NOFS); | |
2168 | break; | |
2169 | } | |
d1310b2e CM |
2170 | extent_offset = cur - em->start; |
2171 | BUG_ON(extent_map_end(em) <= cur); | |
2172 | BUG_ON(end < cur); | |
2173 | ||
261507a0 | 2174 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
c8b97818 | 2175 | this_bio_flag = EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
2176 | extent_set_compress_type(&this_bio_flag, |
2177 | em->compress_type); | |
2178 | } | |
c8b97818 | 2179 | |
d1310b2e CM |
2180 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
2181 | cur_end = min(extent_map_end(em) - 1, end); | |
2182 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
c8b97818 CM |
2183 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
2184 | disk_io_size = em->block_len; | |
2185 | sector = em->block_start >> 9; | |
2186 | } else { | |
2187 | sector = (em->block_start + extent_offset) >> 9; | |
2188 | disk_io_size = iosize; | |
2189 | } | |
d1310b2e CM |
2190 | bdev = em->bdev; |
2191 | block_start = em->block_start; | |
d899e052 YZ |
2192 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2193 | block_start = EXTENT_MAP_HOLE; | |
d1310b2e CM |
2194 | free_extent_map(em); |
2195 | em = NULL; | |
2196 | ||
2197 | /* we've found a hole, just zero and go on */ | |
2198 | if (block_start == EXTENT_MAP_HOLE) { | |
2199 | char *userpage; | |
507903b8 AJ |
2200 | struct extent_state *cached = NULL; |
2201 | ||
d1310b2e | 2202 | userpage = kmap_atomic(page, KM_USER0); |
306e16ce | 2203 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e CM |
2204 | flush_dcache_page(page); |
2205 | kunmap_atomic(userpage, KM_USER0); | |
2206 | ||
2207 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 AJ |
2208 | &cached, GFP_NOFS); |
2209 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
2210 | &cached, GFP_NOFS); | |
d1310b2e | 2211 | cur = cur + iosize; |
306e16ce | 2212 | pg_offset += iosize; |
d1310b2e CM |
2213 | continue; |
2214 | } | |
2215 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
2216 | if (test_range_bit(tree, cur, cur_end, |
2217 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 2218 | check_page_uptodate(tree, page); |
d1310b2e CM |
2219 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); |
2220 | cur = cur + iosize; | |
306e16ce | 2221 | pg_offset += iosize; |
d1310b2e CM |
2222 | continue; |
2223 | } | |
70dec807 CM |
2224 | /* we have an inline extent but it didn't get marked up |
2225 | * to date. Error out | |
2226 | */ | |
2227 | if (block_start == EXTENT_MAP_INLINE) { | |
2228 | SetPageError(page); | |
2229 | unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); | |
2230 | cur = cur + iosize; | |
306e16ce | 2231 | pg_offset += iosize; |
70dec807 CM |
2232 | continue; |
2233 | } | |
d1310b2e CM |
2234 | |
2235 | ret = 0; | |
2236 | if (tree->ops && tree->ops->readpage_io_hook) { | |
2237 | ret = tree->ops->readpage_io_hook(page, cur, | |
2238 | cur + iosize - 1); | |
2239 | } | |
2240 | if (!ret) { | |
89642229 CM |
2241 | unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1; |
2242 | pnr -= page->index; | |
d1310b2e | 2243 | ret = submit_extent_page(READ, tree, page, |
306e16ce | 2244 | sector, disk_io_size, pg_offset, |
89642229 | 2245 | bdev, bio, pnr, |
c8b97818 CM |
2246 | end_bio_extent_readpage, mirror_num, |
2247 | *bio_flags, | |
2248 | this_bio_flag); | |
89642229 | 2249 | nr++; |
c8b97818 | 2250 | *bio_flags = this_bio_flag; |
d1310b2e CM |
2251 | } |
2252 | if (ret) | |
2253 | SetPageError(page); | |
2254 | cur = cur + iosize; | |
306e16ce | 2255 | pg_offset += iosize; |
d1310b2e | 2256 | } |
90a887c9 | 2257 | out: |
d1310b2e CM |
2258 | if (!nr) { |
2259 | if (!PageError(page)) | |
2260 | SetPageUptodate(page); | |
2261 | unlock_page(page); | |
2262 | } | |
2263 | return 0; | |
2264 | } | |
2265 | ||
2266 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, | |
2267 | get_extent_t *get_extent) | |
2268 | { | |
2269 | struct bio *bio = NULL; | |
c8b97818 | 2270 | unsigned long bio_flags = 0; |
d1310b2e CM |
2271 | int ret; |
2272 | ||
c8b97818 CM |
2273 | ret = __extent_read_full_page(tree, page, get_extent, &bio, 0, |
2274 | &bio_flags); | |
d1310b2e | 2275 | if (bio) |
6b82ce8d | 2276 | ret = submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
2277 | return ret; |
2278 | } | |
d1310b2e | 2279 | |
11c8349b CM |
2280 | static noinline void update_nr_written(struct page *page, |
2281 | struct writeback_control *wbc, | |
2282 | unsigned long nr_written) | |
2283 | { | |
2284 | wbc->nr_to_write -= nr_written; | |
2285 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && | |
2286 | wbc->range_start == 0 && wbc->range_end == LLONG_MAX)) | |
2287 | page->mapping->writeback_index = page->index + nr_written; | |
2288 | } | |
2289 | ||
d1310b2e CM |
2290 | /* |
2291 | * the writepage semantics are similar to regular writepage. extent | |
2292 | * records are inserted to lock ranges in the tree, and as dirty areas | |
2293 | * are found, they are marked writeback. Then the lock bits are removed | |
2294 | * and the end_io handler clears the writeback ranges | |
2295 | */ | |
2296 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
2297 | void *data) | |
2298 | { | |
2299 | struct inode *inode = page->mapping->host; | |
2300 | struct extent_page_data *epd = data; | |
2301 | struct extent_io_tree *tree = epd->tree; | |
2302 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2303 | u64 delalloc_start; | |
2304 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2305 | u64 end; | |
2306 | u64 cur = start; | |
2307 | u64 extent_offset; | |
2308 | u64 last_byte = i_size_read(inode); | |
2309 | u64 block_start; | |
2310 | u64 iosize; | |
2311 | sector_t sector; | |
2c64c53d | 2312 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
2313 | struct extent_map *em; |
2314 | struct block_device *bdev; | |
2315 | int ret; | |
2316 | int nr = 0; | |
7f3c74fb | 2317 | size_t pg_offset = 0; |
d1310b2e CM |
2318 | size_t blocksize; |
2319 | loff_t i_size = i_size_read(inode); | |
2320 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
2321 | u64 nr_delalloc; | |
2322 | u64 delalloc_end; | |
c8b97818 CM |
2323 | int page_started; |
2324 | int compressed; | |
ffbd517d | 2325 | int write_flags; |
771ed689 | 2326 | unsigned long nr_written = 0; |
9e487107 | 2327 | bool fill_delalloc = true; |
d1310b2e | 2328 | |
ffbd517d | 2329 | if (wbc->sync_mode == WB_SYNC_ALL) |
721a9602 | 2330 | write_flags = WRITE_SYNC; |
ffbd517d CM |
2331 | else |
2332 | write_flags = WRITE; | |
2333 | ||
1abe9b8a | 2334 | trace___extent_writepage(page, inode, wbc); |
2335 | ||
d1310b2e | 2336 | WARN_ON(!PageLocked(page)); |
7f3c74fb | 2337 | pg_offset = i_size & (PAGE_CACHE_SIZE - 1); |
211c17f5 | 2338 | if (page->index > end_index || |
7f3c74fb | 2339 | (page->index == end_index && !pg_offset)) { |
39be25cd | 2340 | page->mapping->a_ops->invalidatepage(page, 0); |
d1310b2e CM |
2341 | unlock_page(page); |
2342 | return 0; | |
2343 | } | |
2344 | ||
2345 | if (page->index == end_index) { | |
2346 | char *userpage; | |
2347 | ||
d1310b2e | 2348 | userpage = kmap_atomic(page, KM_USER0); |
7f3c74fb CM |
2349 | memset(userpage + pg_offset, 0, |
2350 | PAGE_CACHE_SIZE - pg_offset); | |
d1310b2e | 2351 | kunmap_atomic(userpage, KM_USER0); |
211c17f5 | 2352 | flush_dcache_page(page); |
d1310b2e | 2353 | } |
7f3c74fb | 2354 | pg_offset = 0; |
d1310b2e CM |
2355 | |
2356 | set_page_extent_mapped(page); | |
2357 | ||
9e487107 JB |
2358 | if (!tree->ops || !tree->ops->fill_delalloc) |
2359 | fill_delalloc = false; | |
2360 | ||
d1310b2e CM |
2361 | delalloc_start = start; |
2362 | delalloc_end = 0; | |
c8b97818 | 2363 | page_started = 0; |
9e487107 | 2364 | if (!epd->extent_locked && fill_delalloc) { |
f85d7d6c | 2365 | u64 delalloc_to_write = 0; |
11c8349b CM |
2366 | /* |
2367 | * make sure the wbc mapping index is at least updated | |
2368 | * to this page. | |
2369 | */ | |
2370 | update_nr_written(page, wbc, 0); | |
2371 | ||
d397712b | 2372 | while (delalloc_end < page_end) { |
771ed689 | 2373 | nr_delalloc = find_lock_delalloc_range(inode, tree, |
c8b97818 CM |
2374 | page, |
2375 | &delalloc_start, | |
d1310b2e CM |
2376 | &delalloc_end, |
2377 | 128 * 1024 * 1024); | |
771ed689 CM |
2378 | if (nr_delalloc == 0) { |
2379 | delalloc_start = delalloc_end + 1; | |
2380 | continue; | |
2381 | } | |
2382 | tree->ops->fill_delalloc(inode, page, delalloc_start, | |
2383 | delalloc_end, &page_started, | |
2384 | &nr_written); | |
f85d7d6c CM |
2385 | /* |
2386 | * delalloc_end is already one less than the total | |
2387 | * length, so we don't subtract one from | |
2388 | * PAGE_CACHE_SIZE | |
2389 | */ | |
2390 | delalloc_to_write += (delalloc_end - delalloc_start + | |
2391 | PAGE_CACHE_SIZE) >> | |
2392 | PAGE_CACHE_SHIFT; | |
d1310b2e | 2393 | delalloc_start = delalloc_end + 1; |
d1310b2e | 2394 | } |
f85d7d6c CM |
2395 | if (wbc->nr_to_write < delalloc_to_write) { |
2396 | int thresh = 8192; | |
2397 | ||
2398 | if (delalloc_to_write < thresh * 2) | |
2399 | thresh = delalloc_to_write; | |
2400 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
2401 | thresh); | |
2402 | } | |
c8b97818 | 2403 | |
771ed689 CM |
2404 | /* did the fill delalloc function already unlock and start |
2405 | * the IO? | |
2406 | */ | |
2407 | if (page_started) { | |
2408 | ret = 0; | |
11c8349b CM |
2409 | /* |
2410 | * we've unlocked the page, so we can't update | |
2411 | * the mapping's writeback index, just update | |
2412 | * nr_to_write. | |
2413 | */ | |
2414 | wbc->nr_to_write -= nr_written; | |
2415 | goto done_unlocked; | |
771ed689 | 2416 | } |
c8b97818 | 2417 | } |
247e743c | 2418 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
2419 | ret = tree->ops->writepage_start_hook(page, start, |
2420 | page_end); | |
247e743c | 2421 | if (ret == -EAGAIN) { |
247e743c | 2422 | redirty_page_for_writepage(wbc, page); |
11c8349b | 2423 | update_nr_written(page, wbc, nr_written); |
247e743c | 2424 | unlock_page(page); |
771ed689 | 2425 | ret = 0; |
11c8349b | 2426 | goto done_unlocked; |
247e743c CM |
2427 | } |
2428 | } | |
2429 | ||
11c8349b CM |
2430 | /* |
2431 | * we don't want to touch the inode after unlocking the page, | |
2432 | * so we update the mapping writeback index now | |
2433 | */ | |
2434 | update_nr_written(page, wbc, nr_written + 1); | |
771ed689 | 2435 | |
d1310b2e | 2436 | end = page_end; |
d1310b2e | 2437 | if (last_byte <= start) { |
e6dcd2dc CM |
2438 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2439 | tree->ops->writepage_end_io_hook(page, start, | |
2440 | page_end, NULL, 1); | |
d1310b2e CM |
2441 | goto done; |
2442 | } | |
2443 | ||
d1310b2e CM |
2444 | blocksize = inode->i_sb->s_blocksize; |
2445 | ||
2446 | while (cur <= end) { | |
2447 | if (cur >= last_byte) { | |
e6dcd2dc CM |
2448 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2449 | tree->ops->writepage_end_io_hook(page, cur, | |
2450 | page_end, NULL, 1); | |
d1310b2e CM |
2451 | break; |
2452 | } | |
7f3c74fb | 2453 | em = epd->get_extent(inode, page, pg_offset, cur, |
d1310b2e | 2454 | end - cur + 1, 1); |
c704005d | 2455 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e CM |
2456 | SetPageError(page); |
2457 | break; | |
2458 | } | |
2459 | ||
2460 | extent_offset = cur - em->start; | |
2461 | BUG_ON(extent_map_end(em) <= cur); | |
2462 | BUG_ON(end < cur); | |
2463 | iosize = min(extent_map_end(em) - cur, end - cur + 1); | |
2464 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
2465 | sector = (em->block_start + extent_offset) >> 9; | |
2466 | bdev = em->bdev; | |
2467 | block_start = em->block_start; | |
c8b97818 | 2468 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
2469 | free_extent_map(em); |
2470 | em = NULL; | |
2471 | ||
c8b97818 CM |
2472 | /* |
2473 | * compressed and inline extents are written through other | |
2474 | * paths in the FS | |
2475 | */ | |
2476 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 2477 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
2478 | /* |
2479 | * end_io notification does not happen here for | |
2480 | * compressed extents | |
2481 | */ | |
2482 | if (!compressed && tree->ops && | |
2483 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
2484 | tree->ops->writepage_end_io_hook(page, cur, |
2485 | cur + iosize - 1, | |
2486 | NULL, 1); | |
c8b97818 CM |
2487 | else if (compressed) { |
2488 | /* we don't want to end_page_writeback on | |
2489 | * a compressed extent. this happens | |
2490 | * elsewhere | |
2491 | */ | |
2492 | nr++; | |
2493 | } | |
2494 | ||
2495 | cur += iosize; | |
7f3c74fb | 2496 | pg_offset += iosize; |
d1310b2e CM |
2497 | continue; |
2498 | } | |
d1310b2e CM |
2499 | /* leave this out until we have a page_mkwrite call */ |
2500 | if (0 && !test_range_bit(tree, cur, cur + iosize - 1, | |
9655d298 | 2501 | EXTENT_DIRTY, 0, NULL)) { |
d1310b2e | 2502 | cur = cur + iosize; |
7f3c74fb | 2503 | pg_offset += iosize; |
d1310b2e CM |
2504 | continue; |
2505 | } | |
c8b97818 | 2506 | |
d1310b2e CM |
2507 | if (tree->ops && tree->ops->writepage_io_hook) { |
2508 | ret = tree->ops->writepage_io_hook(page, cur, | |
2509 | cur + iosize - 1); | |
2510 | } else { | |
2511 | ret = 0; | |
2512 | } | |
1259ab75 | 2513 | if (ret) { |
d1310b2e | 2514 | SetPageError(page); |
1259ab75 | 2515 | } else { |
d1310b2e | 2516 | unsigned long max_nr = end_index + 1; |
7f3c74fb | 2517 | |
d1310b2e CM |
2518 | set_range_writeback(tree, cur, cur + iosize - 1); |
2519 | if (!PageWriteback(page)) { | |
d397712b CM |
2520 | printk(KERN_ERR "btrfs warning page %lu not " |
2521 | "writeback, cur %llu end %llu\n", | |
2522 | page->index, (unsigned long long)cur, | |
d1310b2e CM |
2523 | (unsigned long long)end); |
2524 | } | |
2525 | ||
ffbd517d CM |
2526 | ret = submit_extent_page(write_flags, tree, page, |
2527 | sector, iosize, pg_offset, | |
2528 | bdev, &epd->bio, max_nr, | |
c8b97818 CM |
2529 | end_bio_extent_writepage, |
2530 | 0, 0, 0); | |
d1310b2e CM |
2531 | if (ret) |
2532 | SetPageError(page); | |
2533 | } | |
2534 | cur = cur + iosize; | |
7f3c74fb | 2535 | pg_offset += iosize; |
d1310b2e CM |
2536 | nr++; |
2537 | } | |
2538 | done: | |
2539 | if (nr == 0) { | |
2540 | /* make sure the mapping tag for page dirty gets cleared */ | |
2541 | set_page_writeback(page); | |
2542 | end_page_writeback(page); | |
2543 | } | |
d1310b2e | 2544 | unlock_page(page); |
771ed689 | 2545 | |
11c8349b CM |
2546 | done_unlocked: |
2547 | ||
2c64c53d CM |
2548 | /* drop our reference on any cached states */ |
2549 | free_extent_state(cached_state); | |
d1310b2e CM |
2550 | return 0; |
2551 | } | |
2552 | ||
d1310b2e | 2553 | /** |
4bef0848 | 2554 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
2555 | * @mapping: address space structure to write |
2556 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
2557 | * @writepage: function called for each page | |
2558 | * @data: data passed to writepage function | |
2559 | * | |
2560 | * If a page is already under I/O, write_cache_pages() skips it, even | |
2561 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
2562 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
2563 | * and msync() need to guarantee that all the data which was dirty at the time | |
2564 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
2565 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
2566 | * existing IO to complete. | |
2567 | */ | |
b2950863 | 2568 | static int extent_write_cache_pages(struct extent_io_tree *tree, |
4bef0848 CM |
2569 | struct address_space *mapping, |
2570 | struct writeback_control *wbc, | |
d2c3f4f6 CM |
2571 | writepage_t writepage, void *data, |
2572 | void (*flush_fn)(void *)) | |
d1310b2e | 2573 | { |
d1310b2e CM |
2574 | int ret = 0; |
2575 | int done = 0; | |
f85d7d6c | 2576 | int nr_to_write_done = 0; |
d1310b2e CM |
2577 | struct pagevec pvec; |
2578 | int nr_pages; | |
2579 | pgoff_t index; | |
2580 | pgoff_t end; /* Inclusive */ | |
2581 | int scanned = 0; | |
f7aaa06b | 2582 | int tag; |
d1310b2e | 2583 | |
d1310b2e CM |
2584 | pagevec_init(&pvec, 0); |
2585 | if (wbc->range_cyclic) { | |
2586 | index = mapping->writeback_index; /* Start from prev offset */ | |
2587 | end = -1; | |
2588 | } else { | |
2589 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
2590 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
2591 | scanned = 1; |
2592 | } | |
f7aaa06b JB |
2593 | if (wbc->sync_mode == WB_SYNC_ALL) |
2594 | tag = PAGECACHE_TAG_TOWRITE; | |
2595 | else | |
2596 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 2597 | retry: |
f7aaa06b JB |
2598 | if (wbc->sync_mode == WB_SYNC_ALL) |
2599 | tag_pages_for_writeback(mapping, index, end); | |
f85d7d6c | 2600 | while (!done && !nr_to_write_done && (index <= end) && |
f7aaa06b JB |
2601 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
2602 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
d1310b2e CM |
2603 | unsigned i; |
2604 | ||
2605 | scanned = 1; | |
2606 | for (i = 0; i < nr_pages; i++) { | |
2607 | struct page *page = pvec.pages[i]; | |
2608 | ||
2609 | /* | |
2610 | * At this point we hold neither mapping->tree_lock nor | |
2611 | * lock on the page itself: the page may be truncated or | |
2612 | * invalidated (changing page->mapping to NULL), or even | |
2613 | * swizzled back from swapper_space to tmpfs file | |
2614 | * mapping | |
2615 | */ | |
01d658f2 CM |
2616 | if (tree->ops && |
2617 | tree->ops->write_cache_pages_lock_hook) { | |
2618 | tree->ops->write_cache_pages_lock_hook(page, | |
2619 | data, flush_fn); | |
2620 | } else { | |
2621 | if (!trylock_page(page)) { | |
2622 | flush_fn(data); | |
2623 | lock_page(page); | |
2624 | } | |
2625 | } | |
d1310b2e CM |
2626 | |
2627 | if (unlikely(page->mapping != mapping)) { | |
2628 | unlock_page(page); | |
2629 | continue; | |
2630 | } | |
2631 | ||
2632 | if (!wbc->range_cyclic && page->index > end) { | |
2633 | done = 1; | |
2634 | unlock_page(page); | |
2635 | continue; | |
2636 | } | |
2637 | ||
d2c3f4f6 | 2638 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 CM |
2639 | if (PageWriteback(page)) |
2640 | flush_fn(data); | |
d1310b2e | 2641 | wait_on_page_writeback(page); |
d2c3f4f6 | 2642 | } |
d1310b2e CM |
2643 | |
2644 | if (PageWriteback(page) || | |
2645 | !clear_page_dirty_for_io(page)) { | |
2646 | unlock_page(page); | |
2647 | continue; | |
2648 | } | |
2649 | ||
2650 | ret = (*writepage)(page, wbc, data); | |
2651 | ||
2652 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
2653 | unlock_page(page); | |
2654 | ret = 0; | |
2655 | } | |
f85d7d6c | 2656 | if (ret) |
d1310b2e | 2657 | done = 1; |
f85d7d6c CM |
2658 | |
2659 | /* | |
2660 | * the filesystem may choose to bump up nr_to_write. | |
2661 | * We have to make sure to honor the new nr_to_write | |
2662 | * at any time | |
2663 | */ | |
2664 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
2665 | } |
2666 | pagevec_release(&pvec); | |
2667 | cond_resched(); | |
2668 | } | |
2669 | if (!scanned && !done) { | |
2670 | /* | |
2671 | * We hit the last page and there is more work to be done: wrap | |
2672 | * back to the start of the file | |
2673 | */ | |
2674 | scanned = 1; | |
2675 | index = 0; | |
2676 | goto retry; | |
2677 | } | |
d1310b2e CM |
2678 | return ret; |
2679 | } | |
d1310b2e | 2680 | |
ffbd517d | 2681 | static void flush_epd_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 2682 | { |
d2c3f4f6 | 2683 | if (epd->bio) { |
ffbd517d CM |
2684 | if (epd->sync_io) |
2685 | submit_one_bio(WRITE_SYNC, epd->bio, 0, 0); | |
2686 | else | |
2687 | submit_one_bio(WRITE, epd->bio, 0, 0); | |
d2c3f4f6 CM |
2688 | epd->bio = NULL; |
2689 | } | |
2690 | } | |
2691 | ||
ffbd517d CM |
2692 | static noinline void flush_write_bio(void *data) |
2693 | { | |
2694 | struct extent_page_data *epd = data; | |
2695 | flush_epd_write_bio(epd); | |
2696 | } | |
2697 | ||
d1310b2e CM |
2698 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, |
2699 | get_extent_t *get_extent, | |
2700 | struct writeback_control *wbc) | |
2701 | { | |
2702 | int ret; | |
d1310b2e CM |
2703 | struct extent_page_data epd = { |
2704 | .bio = NULL, | |
2705 | .tree = tree, | |
2706 | .get_extent = get_extent, | |
771ed689 | 2707 | .extent_locked = 0, |
ffbd517d | 2708 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e | 2709 | }; |
d1310b2e | 2710 | |
d1310b2e CM |
2711 | ret = __extent_writepage(page, wbc, &epd); |
2712 | ||
ffbd517d | 2713 | flush_epd_write_bio(&epd); |
d1310b2e CM |
2714 | return ret; |
2715 | } | |
d1310b2e | 2716 | |
771ed689 CM |
2717 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, |
2718 | u64 start, u64 end, get_extent_t *get_extent, | |
2719 | int mode) | |
2720 | { | |
2721 | int ret = 0; | |
2722 | struct address_space *mapping = inode->i_mapping; | |
2723 | struct page *page; | |
2724 | unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >> | |
2725 | PAGE_CACHE_SHIFT; | |
2726 | ||
2727 | struct extent_page_data epd = { | |
2728 | .bio = NULL, | |
2729 | .tree = tree, | |
2730 | .get_extent = get_extent, | |
2731 | .extent_locked = 1, | |
ffbd517d | 2732 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
2733 | }; |
2734 | struct writeback_control wbc_writepages = { | |
771ed689 | 2735 | .sync_mode = mode, |
771ed689 CM |
2736 | .nr_to_write = nr_pages * 2, |
2737 | .range_start = start, | |
2738 | .range_end = end + 1, | |
2739 | }; | |
2740 | ||
d397712b | 2741 | while (start <= end) { |
771ed689 CM |
2742 | page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); |
2743 | if (clear_page_dirty_for_io(page)) | |
2744 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
2745 | else { | |
2746 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
2747 | tree->ops->writepage_end_io_hook(page, start, | |
2748 | start + PAGE_CACHE_SIZE - 1, | |
2749 | NULL, 1); | |
2750 | unlock_page(page); | |
2751 | } | |
2752 | page_cache_release(page); | |
2753 | start += PAGE_CACHE_SIZE; | |
2754 | } | |
2755 | ||
ffbd517d | 2756 | flush_epd_write_bio(&epd); |
771ed689 CM |
2757 | return ret; |
2758 | } | |
d1310b2e CM |
2759 | |
2760 | int extent_writepages(struct extent_io_tree *tree, | |
2761 | struct address_space *mapping, | |
2762 | get_extent_t *get_extent, | |
2763 | struct writeback_control *wbc) | |
2764 | { | |
2765 | int ret = 0; | |
2766 | struct extent_page_data epd = { | |
2767 | .bio = NULL, | |
2768 | .tree = tree, | |
2769 | .get_extent = get_extent, | |
771ed689 | 2770 | .extent_locked = 0, |
ffbd517d | 2771 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
2772 | }; |
2773 | ||
4bef0848 | 2774 | ret = extent_write_cache_pages(tree, mapping, wbc, |
d2c3f4f6 CM |
2775 | __extent_writepage, &epd, |
2776 | flush_write_bio); | |
ffbd517d | 2777 | flush_epd_write_bio(&epd); |
d1310b2e CM |
2778 | return ret; |
2779 | } | |
d1310b2e CM |
2780 | |
2781 | int extent_readpages(struct extent_io_tree *tree, | |
2782 | struct address_space *mapping, | |
2783 | struct list_head *pages, unsigned nr_pages, | |
2784 | get_extent_t get_extent) | |
2785 | { | |
2786 | struct bio *bio = NULL; | |
2787 | unsigned page_idx; | |
c8b97818 | 2788 | unsigned long bio_flags = 0; |
d1310b2e | 2789 | |
d1310b2e CM |
2790 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
2791 | struct page *page = list_entry(pages->prev, struct page, lru); | |
2792 | ||
2793 | prefetchw(&page->flags); | |
2794 | list_del(&page->lru); | |
28ecb609 | 2795 | if (!add_to_page_cache_lru(page, mapping, |
43e817a1 | 2796 | page->index, GFP_NOFS)) { |
f188591e | 2797 | __extent_read_full_page(tree, page, get_extent, |
c8b97818 | 2798 | &bio, 0, &bio_flags); |
d1310b2e CM |
2799 | } |
2800 | page_cache_release(page); | |
2801 | } | |
d1310b2e CM |
2802 | BUG_ON(!list_empty(pages)); |
2803 | if (bio) | |
c8b97818 | 2804 | submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
2805 | return 0; |
2806 | } | |
d1310b2e CM |
2807 | |
2808 | /* | |
2809 | * basic invalidatepage code, this waits on any locked or writeback | |
2810 | * ranges corresponding to the page, and then deletes any extent state | |
2811 | * records from the tree | |
2812 | */ | |
2813 | int extent_invalidatepage(struct extent_io_tree *tree, | |
2814 | struct page *page, unsigned long offset) | |
2815 | { | |
2ac55d41 | 2816 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
2817 | u64 start = ((u64)page->index << PAGE_CACHE_SHIFT); |
2818 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
2819 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
2820 | ||
d397712b | 2821 | start += (offset + blocksize - 1) & ~(blocksize - 1); |
d1310b2e CM |
2822 | if (start > end) |
2823 | return 0; | |
2824 | ||
2ac55d41 | 2825 | lock_extent_bits(tree, start, end, 0, &cached_state, GFP_NOFS); |
1edbb734 | 2826 | wait_on_page_writeback(page); |
d1310b2e | 2827 | clear_extent_bit(tree, start, end, |
32c00aff JB |
2828 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
2829 | EXTENT_DO_ACCOUNTING, | |
2ac55d41 | 2830 | 1, 1, &cached_state, GFP_NOFS); |
d1310b2e CM |
2831 | return 0; |
2832 | } | |
d1310b2e | 2833 | |
7b13b7b1 CM |
2834 | /* |
2835 | * a helper for releasepage, this tests for areas of the page that | |
2836 | * are locked or under IO and drops the related state bits if it is safe | |
2837 | * to drop the page. | |
2838 | */ | |
2839 | int try_release_extent_state(struct extent_map_tree *map, | |
2840 | struct extent_io_tree *tree, struct page *page, | |
2841 | gfp_t mask) | |
2842 | { | |
2843 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2844 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
2845 | int ret = 1; | |
2846 | ||
211f90e6 | 2847 | if (test_range_bit(tree, start, end, |
8b62b72b | 2848 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
2849 | ret = 0; |
2850 | else { | |
2851 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
2852 | mask = GFP_NOFS; | |
11ef160f CM |
2853 | /* |
2854 | * at this point we can safely clear everything except the | |
2855 | * locked bit and the nodatasum bit | |
2856 | */ | |
e3f24cc5 | 2857 | ret = clear_extent_bit(tree, start, end, |
11ef160f CM |
2858 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
2859 | 0, 0, NULL, mask); | |
e3f24cc5 CM |
2860 | |
2861 | /* if clear_extent_bit failed for enomem reasons, | |
2862 | * we can't allow the release to continue. | |
2863 | */ | |
2864 | if (ret < 0) | |
2865 | ret = 0; | |
2866 | else | |
2867 | ret = 1; | |
7b13b7b1 CM |
2868 | } |
2869 | return ret; | |
2870 | } | |
7b13b7b1 | 2871 | |
d1310b2e CM |
2872 | /* |
2873 | * a helper for releasepage. As long as there are no locked extents | |
2874 | * in the range corresponding to the page, both state records and extent | |
2875 | * map records are removed | |
2876 | */ | |
2877 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
2878 | struct extent_io_tree *tree, struct page *page, |
2879 | gfp_t mask) | |
d1310b2e CM |
2880 | { |
2881 | struct extent_map *em; | |
2882 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2883 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
7b13b7b1 | 2884 | |
70dec807 CM |
2885 | if ((mask & __GFP_WAIT) && |
2886 | page->mapping->host->i_size > 16 * 1024 * 1024) { | |
39b5637f | 2887 | u64 len; |
70dec807 | 2888 | while (start <= end) { |
39b5637f | 2889 | len = end - start + 1; |
890871be | 2890 | write_lock(&map->lock); |
39b5637f | 2891 | em = lookup_extent_mapping(map, start, len); |
c704005d | 2892 | if (IS_ERR_OR_NULL(em)) { |
890871be | 2893 | write_unlock(&map->lock); |
70dec807 CM |
2894 | break; |
2895 | } | |
7f3c74fb CM |
2896 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
2897 | em->start != start) { | |
890871be | 2898 | write_unlock(&map->lock); |
70dec807 CM |
2899 | free_extent_map(em); |
2900 | break; | |
2901 | } | |
2902 | if (!test_range_bit(tree, em->start, | |
2903 | extent_map_end(em) - 1, | |
8b62b72b | 2904 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 2905 | 0, NULL)) { |
70dec807 CM |
2906 | remove_extent_mapping(map, em); |
2907 | /* once for the rb tree */ | |
2908 | free_extent_map(em); | |
2909 | } | |
2910 | start = extent_map_end(em); | |
890871be | 2911 | write_unlock(&map->lock); |
70dec807 CM |
2912 | |
2913 | /* once for us */ | |
d1310b2e CM |
2914 | free_extent_map(em); |
2915 | } | |
d1310b2e | 2916 | } |
7b13b7b1 | 2917 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 2918 | } |
d1310b2e | 2919 | |
ec29ed5b CM |
2920 | /* |
2921 | * helper function for fiemap, which doesn't want to see any holes. | |
2922 | * This maps until we find something past 'last' | |
2923 | */ | |
2924 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
2925 | u64 offset, | |
2926 | u64 last, | |
2927 | get_extent_t *get_extent) | |
2928 | { | |
2929 | u64 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
2930 | struct extent_map *em; | |
2931 | u64 len; | |
2932 | ||
2933 | if (offset >= last) | |
2934 | return NULL; | |
2935 | ||
2936 | while(1) { | |
2937 | len = last - offset; | |
2938 | if (len == 0) | |
2939 | break; | |
2940 | len = (len + sectorsize - 1) & ~(sectorsize - 1); | |
2941 | em = get_extent(inode, NULL, 0, offset, len, 0); | |
c704005d | 2942 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
2943 | return em; |
2944 | ||
2945 | /* if this isn't a hole return it */ | |
2946 | if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) && | |
2947 | em->block_start != EXTENT_MAP_HOLE) { | |
2948 | return em; | |
2949 | } | |
2950 | ||
2951 | /* this is a hole, advance to the next extent */ | |
2952 | offset = extent_map_end(em); | |
2953 | free_extent_map(em); | |
2954 | if (offset >= last) | |
2955 | break; | |
2956 | } | |
2957 | return NULL; | |
2958 | } | |
2959 | ||
1506fcc8 YS |
2960 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
2961 | __u64 start, __u64 len, get_extent_t *get_extent) | |
2962 | { | |
975f84fe | 2963 | int ret = 0; |
1506fcc8 YS |
2964 | u64 off = start; |
2965 | u64 max = start + len; | |
2966 | u32 flags = 0; | |
975f84fe JB |
2967 | u32 found_type; |
2968 | u64 last; | |
ec29ed5b | 2969 | u64 last_for_get_extent = 0; |
1506fcc8 | 2970 | u64 disko = 0; |
ec29ed5b | 2971 | u64 isize = i_size_read(inode); |
975f84fe | 2972 | struct btrfs_key found_key; |
1506fcc8 | 2973 | struct extent_map *em = NULL; |
2ac55d41 | 2974 | struct extent_state *cached_state = NULL; |
975f84fe JB |
2975 | struct btrfs_path *path; |
2976 | struct btrfs_file_extent_item *item; | |
1506fcc8 | 2977 | int end = 0; |
ec29ed5b CM |
2978 | u64 em_start = 0; |
2979 | u64 em_len = 0; | |
2980 | u64 em_end = 0; | |
1506fcc8 | 2981 | unsigned long emflags; |
1506fcc8 YS |
2982 | |
2983 | if (len == 0) | |
2984 | return -EINVAL; | |
2985 | ||
975f84fe JB |
2986 | path = btrfs_alloc_path(); |
2987 | if (!path) | |
2988 | return -ENOMEM; | |
2989 | path->leave_spinning = 1; | |
2990 | ||
ec29ed5b CM |
2991 | /* |
2992 | * lookup the last file extent. We're not using i_size here | |
2993 | * because there might be preallocation past i_size | |
2994 | */ | |
975f84fe | 2995 | ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root, |
33345d01 | 2996 | path, btrfs_ino(inode), -1, 0); |
975f84fe JB |
2997 | if (ret < 0) { |
2998 | btrfs_free_path(path); | |
2999 | return ret; | |
3000 | } | |
3001 | WARN_ON(!ret); | |
3002 | path->slots[0]--; | |
3003 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3004 | struct btrfs_file_extent_item); | |
3005 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
3006 | found_type = btrfs_key_type(&found_key); | |
3007 | ||
ec29ed5b | 3008 | /* No extents, but there might be delalloc bits */ |
33345d01 | 3009 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 3010 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
3011 | /* have to trust i_size as the end */ |
3012 | last = (u64)-1; | |
3013 | last_for_get_extent = isize; | |
3014 | } else { | |
3015 | /* | |
3016 | * remember the start of the last extent. There are a | |
3017 | * bunch of different factors that go into the length of the | |
3018 | * extent, so its much less complex to remember where it started | |
3019 | */ | |
3020 | last = found_key.offset; | |
3021 | last_for_get_extent = last + 1; | |
975f84fe | 3022 | } |
975f84fe JB |
3023 | btrfs_free_path(path); |
3024 | ||
ec29ed5b CM |
3025 | /* |
3026 | * we might have some extents allocated but more delalloc past those | |
3027 | * extents. so, we trust isize unless the start of the last extent is | |
3028 | * beyond isize | |
3029 | */ | |
3030 | if (last < isize) { | |
3031 | last = (u64)-1; | |
3032 | last_for_get_extent = isize; | |
3033 | } | |
3034 | ||
2ac55d41 JB |
3035 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0, |
3036 | &cached_state, GFP_NOFS); | |
ec29ed5b CM |
3037 | |
3038 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
3039 | get_extent); | |
1506fcc8 YS |
3040 | if (!em) |
3041 | goto out; | |
3042 | if (IS_ERR(em)) { | |
3043 | ret = PTR_ERR(em); | |
3044 | goto out; | |
3045 | } | |
975f84fe | 3046 | |
1506fcc8 | 3047 | while (!end) { |
ea8efc74 CM |
3048 | u64 offset_in_extent; |
3049 | ||
3050 | /* break if the extent we found is outside the range */ | |
3051 | if (em->start >= max || extent_map_end(em) < off) | |
3052 | break; | |
3053 | ||
3054 | /* | |
3055 | * get_extent may return an extent that starts before our | |
3056 | * requested range. We have to make sure the ranges | |
3057 | * we return to fiemap always move forward and don't | |
3058 | * overlap, so adjust the offsets here | |
3059 | */ | |
3060 | em_start = max(em->start, off); | |
1506fcc8 | 3061 | |
ea8efc74 CM |
3062 | /* |
3063 | * record the offset from the start of the extent | |
3064 | * for adjusting the disk offset below | |
3065 | */ | |
3066 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 3067 | em_end = extent_map_end(em); |
ea8efc74 | 3068 | em_len = em_end - em_start; |
ec29ed5b | 3069 | emflags = em->flags; |
1506fcc8 YS |
3070 | disko = 0; |
3071 | flags = 0; | |
3072 | ||
ea8efc74 CM |
3073 | /* |
3074 | * bump off for our next call to get_extent | |
3075 | */ | |
3076 | off = extent_map_end(em); | |
3077 | if (off >= max) | |
3078 | end = 1; | |
3079 | ||
93dbfad7 | 3080 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
3081 | end = 1; |
3082 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 3083 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
3084 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
3085 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 3086 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
3087 | flags |= (FIEMAP_EXTENT_DELALLOC | |
3088 | FIEMAP_EXTENT_UNKNOWN); | |
93dbfad7 | 3089 | } else { |
ea8efc74 | 3090 | disko = em->block_start + offset_in_extent; |
1506fcc8 YS |
3091 | } |
3092 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
3093 | flags |= FIEMAP_EXTENT_ENCODED; | |
3094 | ||
1506fcc8 YS |
3095 | free_extent_map(em); |
3096 | em = NULL; | |
ec29ed5b CM |
3097 | if ((em_start >= last) || em_len == (u64)-1 || |
3098 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
3099 | flags |= FIEMAP_EXTENT_LAST; |
3100 | end = 1; | |
3101 | } | |
3102 | ||
ec29ed5b CM |
3103 | /* now scan forward to see if this is really the last extent. */ |
3104 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
3105 | get_extent); | |
3106 | if (IS_ERR(em)) { | |
3107 | ret = PTR_ERR(em); | |
3108 | goto out; | |
3109 | } | |
3110 | if (!em) { | |
975f84fe JB |
3111 | flags |= FIEMAP_EXTENT_LAST; |
3112 | end = 1; | |
3113 | } | |
ec29ed5b CM |
3114 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, |
3115 | em_len, flags); | |
3116 | if (ret) | |
3117 | goto out_free; | |
1506fcc8 YS |
3118 | } |
3119 | out_free: | |
3120 | free_extent_map(em); | |
3121 | out: | |
2ac55d41 JB |
3122 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len, |
3123 | &cached_state, GFP_NOFS); | |
1506fcc8 YS |
3124 | return ret; |
3125 | } | |
3126 | ||
d1310b2e CM |
3127 | static inline struct page *extent_buffer_page(struct extent_buffer *eb, |
3128 | unsigned long i) | |
3129 | { | |
3130 | struct page *p; | |
3131 | struct address_space *mapping; | |
3132 | ||
3133 | if (i == 0) | |
3134 | return eb->first_page; | |
3135 | i += eb->start >> PAGE_CACHE_SHIFT; | |
3136 | mapping = eb->first_page->mapping; | |
33958dc6 CM |
3137 | if (!mapping) |
3138 | return NULL; | |
0ee0fda0 SW |
3139 | |
3140 | /* | |
3141 | * extent_buffer_page is only called after pinning the page | |
3142 | * by increasing the reference count. So we know the page must | |
3143 | * be in the radix tree. | |
3144 | */ | |
0ee0fda0 | 3145 | rcu_read_lock(); |
d1310b2e | 3146 | p = radix_tree_lookup(&mapping->page_tree, i); |
0ee0fda0 | 3147 | rcu_read_unlock(); |
2b1f55b0 | 3148 | |
d1310b2e CM |
3149 | return p; |
3150 | } | |
3151 | ||
6af118ce | 3152 | static inline unsigned long num_extent_pages(u64 start, u64 len) |
728131d8 | 3153 | { |
6af118ce CM |
3154 | return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) - |
3155 | (start >> PAGE_CACHE_SHIFT); | |
728131d8 CM |
3156 | } |
3157 | ||
d1310b2e CM |
3158 | static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree, |
3159 | u64 start, | |
3160 | unsigned long len, | |
3161 | gfp_t mask) | |
3162 | { | |
3163 | struct extent_buffer *eb = NULL; | |
3935127c | 3164 | #if LEAK_DEBUG |
2d2ae547 | 3165 | unsigned long flags; |
4bef0848 | 3166 | #endif |
d1310b2e | 3167 | |
d1310b2e | 3168 | eb = kmem_cache_zalloc(extent_buffer_cache, mask); |
91ca338d TI |
3169 | if (eb == NULL) |
3170 | return NULL; | |
d1310b2e CM |
3171 | eb->start = start; |
3172 | eb->len = len; | |
bd681513 CM |
3173 | rwlock_init(&eb->lock); |
3174 | atomic_set(&eb->write_locks, 0); | |
3175 | atomic_set(&eb->read_locks, 0); | |
3176 | atomic_set(&eb->blocking_readers, 0); | |
3177 | atomic_set(&eb->blocking_writers, 0); | |
3178 | atomic_set(&eb->spinning_readers, 0); | |
3179 | atomic_set(&eb->spinning_writers, 0); | |
3180 | init_waitqueue_head(&eb->write_lock_wq); | |
3181 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 3182 | |
3935127c | 3183 | #if LEAK_DEBUG |
2d2ae547 CM |
3184 | spin_lock_irqsave(&leak_lock, flags); |
3185 | list_add(&eb->leak_list, &buffers); | |
3186 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 3187 | #endif |
d1310b2e CM |
3188 | atomic_set(&eb->refs, 1); |
3189 | ||
3190 | return eb; | |
3191 | } | |
3192 | ||
3193 | static void __free_extent_buffer(struct extent_buffer *eb) | |
3194 | { | |
3935127c | 3195 | #if LEAK_DEBUG |
2d2ae547 CM |
3196 | unsigned long flags; |
3197 | spin_lock_irqsave(&leak_lock, flags); | |
3198 | list_del(&eb->leak_list); | |
3199 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 3200 | #endif |
d1310b2e CM |
3201 | kmem_cache_free(extent_buffer_cache, eb); |
3202 | } | |
3203 | ||
897ca6e9 MX |
3204 | /* |
3205 | * Helper for releasing extent buffer page. | |
3206 | */ | |
3207 | static void btrfs_release_extent_buffer_page(struct extent_buffer *eb, | |
3208 | unsigned long start_idx) | |
3209 | { | |
3210 | unsigned long index; | |
3211 | struct page *page; | |
3212 | ||
3213 | if (!eb->first_page) | |
3214 | return; | |
3215 | ||
3216 | index = num_extent_pages(eb->start, eb->len); | |
3217 | if (start_idx >= index) | |
3218 | return; | |
3219 | ||
3220 | do { | |
3221 | index--; | |
3222 | page = extent_buffer_page(eb, index); | |
3223 | if (page) | |
3224 | page_cache_release(page); | |
3225 | } while (index != start_idx); | |
3226 | } | |
3227 | ||
3228 | /* | |
3229 | * Helper for releasing the extent buffer. | |
3230 | */ | |
3231 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
3232 | { | |
3233 | btrfs_release_extent_buffer_page(eb, 0); | |
3234 | __free_extent_buffer(eb); | |
3235 | } | |
3236 | ||
d1310b2e CM |
3237 | struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree, |
3238 | u64 start, unsigned long len, | |
ba144192 | 3239 | struct page *page0) |
d1310b2e CM |
3240 | { |
3241 | unsigned long num_pages = num_extent_pages(start, len); | |
3242 | unsigned long i; | |
3243 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
3244 | struct extent_buffer *eb; | |
6af118ce | 3245 | struct extent_buffer *exists = NULL; |
d1310b2e CM |
3246 | struct page *p; |
3247 | struct address_space *mapping = tree->mapping; | |
3248 | int uptodate = 1; | |
19fe0a8b | 3249 | int ret; |
d1310b2e | 3250 | |
19fe0a8b MX |
3251 | rcu_read_lock(); |
3252 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); | |
3253 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
3254 | rcu_read_unlock(); | |
0f9dd46c | 3255 | mark_page_accessed(eb->first_page); |
6af118ce CM |
3256 | return eb; |
3257 | } | |
19fe0a8b | 3258 | rcu_read_unlock(); |
6af118ce | 3259 | |
ba144192 | 3260 | eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS); |
2b114d1d | 3261 | if (!eb) |
d1310b2e CM |
3262 | return NULL; |
3263 | ||
d1310b2e CM |
3264 | if (page0) { |
3265 | eb->first_page = page0; | |
3266 | i = 1; | |
3267 | index++; | |
3268 | page_cache_get(page0); | |
3269 | mark_page_accessed(page0); | |
3270 | set_page_extent_mapped(page0); | |
d1310b2e | 3271 | set_page_extent_head(page0, len); |
f188591e | 3272 | uptodate = PageUptodate(page0); |
d1310b2e CM |
3273 | } else { |
3274 | i = 0; | |
3275 | } | |
3276 | for (; i < num_pages; i++, index++) { | |
a6591715 | 3277 | p = find_or_create_page(mapping, index, GFP_NOFS); |
d1310b2e CM |
3278 | if (!p) { |
3279 | WARN_ON(1); | |
6af118ce | 3280 | goto free_eb; |
d1310b2e CM |
3281 | } |
3282 | set_page_extent_mapped(p); | |
3283 | mark_page_accessed(p); | |
3284 | if (i == 0) { | |
3285 | eb->first_page = p; | |
3286 | set_page_extent_head(p, len); | |
3287 | } else { | |
3288 | set_page_private(p, EXTENT_PAGE_PRIVATE); | |
3289 | } | |
3290 | if (!PageUptodate(p)) | |
3291 | uptodate = 0; | |
eb14ab8e CM |
3292 | |
3293 | /* | |
3294 | * see below about how we avoid a nasty race with release page | |
3295 | * and why we unlock later | |
3296 | */ | |
3297 | if (i != 0) | |
3298 | unlock_page(p); | |
d1310b2e CM |
3299 | } |
3300 | if (uptodate) | |
b4ce94de | 3301 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 3302 | |
19fe0a8b MX |
3303 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); |
3304 | if (ret) | |
3305 | goto free_eb; | |
3306 | ||
6af118ce | 3307 | spin_lock(&tree->buffer_lock); |
19fe0a8b MX |
3308 | ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb); |
3309 | if (ret == -EEXIST) { | |
3310 | exists = radix_tree_lookup(&tree->buffer, | |
3311 | start >> PAGE_CACHE_SHIFT); | |
6af118ce CM |
3312 | /* add one reference for the caller */ |
3313 | atomic_inc(&exists->refs); | |
3314 | spin_unlock(&tree->buffer_lock); | |
19fe0a8b | 3315 | radix_tree_preload_end(); |
6af118ce CM |
3316 | goto free_eb; |
3317 | } | |
6af118ce CM |
3318 | /* add one reference for the tree */ |
3319 | atomic_inc(&eb->refs); | |
f044ba78 | 3320 | spin_unlock(&tree->buffer_lock); |
19fe0a8b | 3321 | radix_tree_preload_end(); |
eb14ab8e CM |
3322 | |
3323 | /* | |
3324 | * there is a race where release page may have | |
3325 | * tried to find this extent buffer in the radix | |
3326 | * but failed. It will tell the VM it is safe to | |
3327 | * reclaim the, and it will clear the page private bit. | |
3328 | * We must make sure to set the page private bit properly | |
3329 | * after the extent buffer is in the radix tree so | |
3330 | * it doesn't get lost | |
3331 | */ | |
3332 | set_page_extent_mapped(eb->first_page); | |
3333 | set_page_extent_head(eb->first_page, eb->len); | |
3334 | if (!page0) | |
3335 | unlock_page(eb->first_page); | |
d1310b2e CM |
3336 | return eb; |
3337 | ||
6af118ce | 3338 | free_eb: |
eb14ab8e CM |
3339 | if (eb->first_page && !page0) |
3340 | unlock_page(eb->first_page); | |
3341 | ||
d1310b2e | 3342 | if (!atomic_dec_and_test(&eb->refs)) |
6af118ce | 3343 | return exists; |
897ca6e9 | 3344 | btrfs_release_extent_buffer(eb); |
6af118ce | 3345 | return exists; |
d1310b2e | 3346 | } |
d1310b2e CM |
3347 | |
3348 | struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree, | |
f09d1f60 | 3349 | u64 start, unsigned long len) |
d1310b2e | 3350 | { |
d1310b2e | 3351 | struct extent_buffer *eb; |
d1310b2e | 3352 | |
19fe0a8b MX |
3353 | rcu_read_lock(); |
3354 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); | |
3355 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
3356 | rcu_read_unlock(); | |
0f9dd46c | 3357 | mark_page_accessed(eb->first_page); |
19fe0a8b MX |
3358 | return eb; |
3359 | } | |
3360 | rcu_read_unlock(); | |
0f9dd46c | 3361 | |
19fe0a8b | 3362 | return NULL; |
d1310b2e | 3363 | } |
d1310b2e CM |
3364 | |
3365 | void free_extent_buffer(struct extent_buffer *eb) | |
3366 | { | |
d1310b2e CM |
3367 | if (!eb) |
3368 | return; | |
3369 | ||
3370 | if (!atomic_dec_and_test(&eb->refs)) | |
3371 | return; | |
3372 | ||
6af118ce | 3373 | WARN_ON(1); |
d1310b2e | 3374 | } |
d1310b2e CM |
3375 | |
3376 | int clear_extent_buffer_dirty(struct extent_io_tree *tree, | |
3377 | struct extent_buffer *eb) | |
3378 | { | |
d1310b2e CM |
3379 | unsigned long i; |
3380 | unsigned long num_pages; | |
3381 | struct page *page; | |
3382 | ||
d1310b2e CM |
3383 | num_pages = num_extent_pages(eb->start, eb->len); |
3384 | ||
3385 | for (i = 0; i < num_pages; i++) { | |
3386 | page = extent_buffer_page(eb, i); | |
b9473439 | 3387 | if (!PageDirty(page)) |
d2c3f4f6 CM |
3388 | continue; |
3389 | ||
a61e6f29 | 3390 | lock_page(page); |
eb14ab8e CM |
3391 | WARN_ON(!PagePrivate(page)); |
3392 | ||
3393 | set_page_extent_mapped(page); | |
d1310b2e CM |
3394 | if (i == 0) |
3395 | set_page_extent_head(page, eb->len); | |
d1310b2e | 3396 | |
d1310b2e | 3397 | clear_page_dirty_for_io(page); |
0ee0fda0 | 3398 | spin_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
3399 | if (!PageDirty(page)) { |
3400 | radix_tree_tag_clear(&page->mapping->page_tree, | |
3401 | page_index(page), | |
3402 | PAGECACHE_TAG_DIRTY); | |
3403 | } | |
0ee0fda0 | 3404 | spin_unlock_irq(&page->mapping->tree_lock); |
a61e6f29 | 3405 | unlock_page(page); |
d1310b2e CM |
3406 | } |
3407 | return 0; | |
3408 | } | |
d1310b2e | 3409 | |
d1310b2e CM |
3410 | int set_extent_buffer_dirty(struct extent_io_tree *tree, |
3411 | struct extent_buffer *eb) | |
3412 | { | |
3413 | unsigned long i; | |
3414 | unsigned long num_pages; | |
b9473439 | 3415 | int was_dirty = 0; |
d1310b2e | 3416 | |
b9473439 | 3417 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
d1310b2e | 3418 | num_pages = num_extent_pages(eb->start, eb->len); |
b9473439 | 3419 | for (i = 0; i < num_pages; i++) |
d1310b2e | 3420 | __set_page_dirty_nobuffers(extent_buffer_page(eb, i)); |
b9473439 | 3421 | return was_dirty; |
d1310b2e | 3422 | } |
d1310b2e | 3423 | |
19b6caf4 CM |
3424 | static int __eb_straddles_pages(u64 start, u64 len) |
3425 | { | |
3426 | if (len < PAGE_CACHE_SIZE) | |
3427 | return 1; | |
3428 | if (start & (PAGE_CACHE_SIZE - 1)) | |
3429 | return 1; | |
3430 | if ((start + len) & (PAGE_CACHE_SIZE - 1)) | |
3431 | return 1; | |
3432 | return 0; | |
3433 | } | |
3434 | ||
3435 | static int eb_straddles_pages(struct extent_buffer *eb) | |
3436 | { | |
3437 | return __eb_straddles_pages(eb->start, eb->len); | |
3438 | } | |
3439 | ||
1259ab75 | 3440 | int clear_extent_buffer_uptodate(struct extent_io_tree *tree, |
2ac55d41 JB |
3441 | struct extent_buffer *eb, |
3442 | struct extent_state **cached_state) | |
1259ab75 CM |
3443 | { |
3444 | unsigned long i; | |
3445 | struct page *page; | |
3446 | unsigned long num_pages; | |
3447 | ||
3448 | num_pages = num_extent_pages(eb->start, eb->len); | |
b4ce94de | 3449 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
1259ab75 | 3450 | |
19b6caf4 CM |
3451 | if (eb_straddles_pages(eb)) { |
3452 | clear_extent_uptodate(tree, eb->start, eb->start + eb->len - 1, | |
3453 | cached_state, GFP_NOFS); | |
3454 | } | |
1259ab75 CM |
3455 | for (i = 0; i < num_pages; i++) { |
3456 | page = extent_buffer_page(eb, i); | |
33958dc6 CM |
3457 | if (page) |
3458 | ClearPageUptodate(page); | |
1259ab75 CM |
3459 | } |
3460 | return 0; | |
3461 | } | |
3462 | ||
d1310b2e CM |
3463 | int set_extent_buffer_uptodate(struct extent_io_tree *tree, |
3464 | struct extent_buffer *eb) | |
3465 | { | |
3466 | unsigned long i; | |
3467 | struct page *page; | |
3468 | unsigned long num_pages; | |
3469 | ||
3470 | num_pages = num_extent_pages(eb->start, eb->len); | |
3471 | ||
19b6caf4 CM |
3472 | if (eb_straddles_pages(eb)) { |
3473 | set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1, | |
3474 | NULL, GFP_NOFS); | |
3475 | } | |
d1310b2e CM |
3476 | for (i = 0; i < num_pages; i++) { |
3477 | page = extent_buffer_page(eb, i); | |
3478 | if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) || | |
3479 | ((i == num_pages - 1) && | |
3480 | ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) { | |
3481 | check_page_uptodate(tree, page); | |
3482 | continue; | |
3483 | } | |
3484 | SetPageUptodate(page); | |
3485 | } | |
3486 | return 0; | |
3487 | } | |
d1310b2e | 3488 | |
ce9adaa5 CM |
3489 | int extent_range_uptodate(struct extent_io_tree *tree, |
3490 | u64 start, u64 end) | |
3491 | { | |
3492 | struct page *page; | |
3493 | int ret; | |
3494 | int pg_uptodate = 1; | |
3495 | int uptodate; | |
3496 | unsigned long index; | |
3497 | ||
19b6caf4 CM |
3498 | if (__eb_straddles_pages(start, end - start + 1)) { |
3499 | ret = test_range_bit(tree, start, end, | |
3500 | EXTENT_UPTODATE, 1, NULL); | |
3501 | if (ret) | |
3502 | return 1; | |
3503 | } | |
d397712b | 3504 | while (start <= end) { |
ce9adaa5 CM |
3505 | index = start >> PAGE_CACHE_SHIFT; |
3506 | page = find_get_page(tree->mapping, index); | |
3507 | uptodate = PageUptodate(page); | |
3508 | page_cache_release(page); | |
3509 | if (!uptodate) { | |
3510 | pg_uptodate = 0; | |
3511 | break; | |
3512 | } | |
3513 | start += PAGE_CACHE_SIZE; | |
3514 | } | |
3515 | return pg_uptodate; | |
3516 | } | |
3517 | ||
d1310b2e | 3518 | int extent_buffer_uptodate(struct extent_io_tree *tree, |
2ac55d41 JB |
3519 | struct extent_buffer *eb, |
3520 | struct extent_state *cached_state) | |
d1310b2e | 3521 | { |
728131d8 | 3522 | int ret = 0; |
ce9adaa5 CM |
3523 | unsigned long num_pages; |
3524 | unsigned long i; | |
728131d8 CM |
3525 | struct page *page; |
3526 | int pg_uptodate = 1; | |
3527 | ||
b4ce94de | 3528 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
4235298e | 3529 | return 1; |
728131d8 | 3530 | |
19b6caf4 CM |
3531 | if (eb_straddles_pages(eb)) { |
3532 | ret = test_range_bit(tree, eb->start, eb->start + eb->len - 1, | |
3533 | EXTENT_UPTODATE, 1, cached_state); | |
3534 | if (ret) | |
3535 | return ret; | |
3536 | } | |
728131d8 CM |
3537 | |
3538 | num_pages = num_extent_pages(eb->start, eb->len); | |
3539 | for (i = 0; i < num_pages; i++) { | |
3540 | page = extent_buffer_page(eb, i); | |
3541 | if (!PageUptodate(page)) { | |
3542 | pg_uptodate = 0; | |
3543 | break; | |
3544 | } | |
3545 | } | |
4235298e | 3546 | return pg_uptodate; |
d1310b2e | 3547 | } |
d1310b2e CM |
3548 | |
3549 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
3550 | struct extent_buffer *eb, | |
a86c12c7 | 3551 | u64 start, int wait, |
f188591e | 3552 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
3553 | { |
3554 | unsigned long i; | |
3555 | unsigned long start_i; | |
3556 | struct page *page; | |
3557 | int err; | |
3558 | int ret = 0; | |
ce9adaa5 CM |
3559 | int locked_pages = 0; |
3560 | int all_uptodate = 1; | |
3561 | int inc_all_pages = 0; | |
d1310b2e | 3562 | unsigned long num_pages; |
a86c12c7 | 3563 | struct bio *bio = NULL; |
c8b97818 | 3564 | unsigned long bio_flags = 0; |
a86c12c7 | 3565 | |
b4ce94de | 3566 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
3567 | return 0; |
3568 | ||
19b6caf4 CM |
3569 | if (eb_straddles_pages(eb)) { |
3570 | if (test_range_bit(tree, eb->start, eb->start + eb->len - 1, | |
3571 | EXTENT_UPTODATE, 1, NULL)) { | |
3572 | return 0; | |
3573 | } | |
d1310b2e CM |
3574 | } |
3575 | ||
3576 | if (start) { | |
3577 | WARN_ON(start < eb->start); | |
3578 | start_i = (start >> PAGE_CACHE_SHIFT) - | |
3579 | (eb->start >> PAGE_CACHE_SHIFT); | |
3580 | } else { | |
3581 | start_i = 0; | |
3582 | } | |
3583 | ||
3584 | num_pages = num_extent_pages(eb->start, eb->len); | |
3585 | for (i = start_i; i < num_pages; i++) { | |
3586 | page = extent_buffer_page(eb, i); | |
d1310b2e | 3587 | if (!wait) { |
2db04966 | 3588 | if (!trylock_page(page)) |
ce9adaa5 | 3589 | goto unlock_exit; |
d1310b2e CM |
3590 | } else { |
3591 | lock_page(page); | |
3592 | } | |
ce9adaa5 | 3593 | locked_pages++; |
d397712b | 3594 | if (!PageUptodate(page)) |
ce9adaa5 | 3595 | all_uptodate = 0; |
ce9adaa5 CM |
3596 | } |
3597 | if (all_uptodate) { | |
3598 | if (start_i == 0) | |
b4ce94de | 3599 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
3600 | goto unlock_exit; |
3601 | } | |
3602 | ||
3603 | for (i = start_i; i < num_pages; i++) { | |
3604 | page = extent_buffer_page(eb, i); | |
eb14ab8e CM |
3605 | |
3606 | WARN_ON(!PagePrivate(page)); | |
3607 | ||
3608 | set_page_extent_mapped(page); | |
3609 | if (i == 0) | |
3610 | set_page_extent_head(page, eb->len); | |
3611 | ||
ce9adaa5 CM |
3612 | if (inc_all_pages) |
3613 | page_cache_get(page); | |
3614 | if (!PageUptodate(page)) { | |
3615 | if (start_i == 0) | |
3616 | inc_all_pages = 1; | |
f188591e | 3617 | ClearPageError(page); |
a86c12c7 | 3618 | err = __extent_read_full_page(tree, page, |
f188591e | 3619 | get_extent, &bio, |
c8b97818 | 3620 | mirror_num, &bio_flags); |
d397712b | 3621 | if (err) |
d1310b2e | 3622 | ret = err; |
d1310b2e CM |
3623 | } else { |
3624 | unlock_page(page); | |
3625 | } | |
3626 | } | |
3627 | ||
a86c12c7 | 3628 | if (bio) |
c8b97818 | 3629 | submit_one_bio(READ, bio, mirror_num, bio_flags); |
a86c12c7 | 3630 | |
d397712b | 3631 | if (ret || !wait) |
d1310b2e | 3632 | return ret; |
d397712b | 3633 | |
d1310b2e CM |
3634 | for (i = start_i; i < num_pages; i++) { |
3635 | page = extent_buffer_page(eb, i); | |
3636 | wait_on_page_locked(page); | |
d397712b | 3637 | if (!PageUptodate(page)) |
d1310b2e | 3638 | ret = -EIO; |
d1310b2e | 3639 | } |
d397712b | 3640 | |
d1310b2e | 3641 | if (!ret) |
b4ce94de | 3642 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 3643 | return ret; |
ce9adaa5 CM |
3644 | |
3645 | unlock_exit: | |
3646 | i = start_i; | |
d397712b | 3647 | while (locked_pages > 0) { |
ce9adaa5 CM |
3648 | page = extent_buffer_page(eb, i); |
3649 | i++; | |
3650 | unlock_page(page); | |
3651 | locked_pages--; | |
3652 | } | |
3653 | return ret; | |
d1310b2e | 3654 | } |
d1310b2e CM |
3655 | |
3656 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
3657 | unsigned long start, | |
3658 | unsigned long len) | |
3659 | { | |
3660 | size_t cur; | |
3661 | size_t offset; | |
3662 | struct page *page; | |
3663 | char *kaddr; | |
3664 | char *dst = (char *)dstv; | |
3665 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3666 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
3667 | |
3668 | WARN_ON(start > eb->len); | |
3669 | WARN_ON(start + len > eb->start + eb->len); | |
3670 | ||
3671 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3672 | ||
d397712b | 3673 | while (len > 0) { |
d1310b2e | 3674 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
3675 | |
3676 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
a6591715 | 3677 | kaddr = page_address(page); |
d1310b2e | 3678 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
3679 | |
3680 | dst += cur; | |
3681 | len -= cur; | |
3682 | offset = 0; | |
3683 | i++; | |
3684 | } | |
3685 | } | |
d1310b2e CM |
3686 | |
3687 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
a6591715 | 3688 | unsigned long min_len, char **map, |
d1310b2e | 3689 | unsigned long *map_start, |
a6591715 | 3690 | unsigned long *map_len) |
d1310b2e CM |
3691 | { |
3692 | size_t offset = start & (PAGE_CACHE_SIZE - 1); | |
3693 | char *kaddr; | |
3694 | struct page *p; | |
3695 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3696 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3697 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
3698 | PAGE_CACHE_SHIFT; | |
3699 | ||
3700 | if (i != end_i) | |
3701 | return -EINVAL; | |
3702 | ||
3703 | if (i == 0) { | |
3704 | offset = start_offset; | |
3705 | *map_start = 0; | |
3706 | } else { | |
3707 | offset = 0; | |
3708 | *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; | |
3709 | } | |
d397712b | 3710 | |
d1310b2e | 3711 | if (start + min_len > eb->len) { |
d397712b CM |
3712 | printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, " |
3713 | "wanted %lu %lu\n", (unsigned long long)eb->start, | |
3714 | eb->len, start, min_len); | |
d1310b2e | 3715 | WARN_ON(1); |
85026533 | 3716 | return -EINVAL; |
d1310b2e CM |
3717 | } |
3718 | ||
3719 | p = extent_buffer_page(eb, i); | |
a6591715 | 3720 | kaddr = page_address(p); |
d1310b2e CM |
3721 | *map = kaddr + offset; |
3722 | *map_len = PAGE_CACHE_SIZE - offset; | |
3723 | return 0; | |
3724 | } | |
d1310b2e | 3725 | |
d1310b2e CM |
3726 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, |
3727 | unsigned long start, | |
3728 | unsigned long len) | |
3729 | { | |
3730 | size_t cur; | |
3731 | size_t offset; | |
3732 | struct page *page; | |
3733 | char *kaddr; | |
3734 | char *ptr = (char *)ptrv; | |
3735 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3736 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3737 | int ret = 0; | |
3738 | ||
3739 | WARN_ON(start > eb->len); | |
3740 | WARN_ON(start + len > eb->start + eb->len); | |
3741 | ||
3742 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3743 | ||
d397712b | 3744 | while (len > 0) { |
d1310b2e | 3745 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
3746 | |
3747 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
3748 | ||
a6591715 | 3749 | kaddr = page_address(page); |
d1310b2e | 3750 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
3751 | if (ret) |
3752 | break; | |
3753 | ||
3754 | ptr += cur; | |
3755 | len -= cur; | |
3756 | offset = 0; | |
3757 | i++; | |
3758 | } | |
3759 | return ret; | |
3760 | } | |
d1310b2e CM |
3761 | |
3762 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
3763 | unsigned long start, unsigned long len) | |
3764 | { | |
3765 | size_t cur; | |
3766 | size_t offset; | |
3767 | struct page *page; | |
3768 | char *kaddr; | |
3769 | char *src = (char *)srcv; | |
3770 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3771 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3772 | ||
3773 | WARN_ON(start > eb->len); | |
3774 | WARN_ON(start + len > eb->start + eb->len); | |
3775 | ||
3776 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3777 | ||
d397712b | 3778 | while (len > 0) { |
d1310b2e CM |
3779 | page = extent_buffer_page(eb, i); |
3780 | WARN_ON(!PageUptodate(page)); | |
3781 | ||
3782 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 3783 | kaddr = page_address(page); |
d1310b2e | 3784 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
3785 | |
3786 | src += cur; | |
3787 | len -= cur; | |
3788 | offset = 0; | |
3789 | i++; | |
3790 | } | |
3791 | } | |
d1310b2e CM |
3792 | |
3793 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
3794 | unsigned long start, unsigned long len) | |
3795 | { | |
3796 | size_t cur; | |
3797 | size_t offset; | |
3798 | struct page *page; | |
3799 | char *kaddr; | |
3800 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3801 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
3802 | ||
3803 | WARN_ON(start > eb->len); | |
3804 | WARN_ON(start + len > eb->start + eb->len); | |
3805 | ||
3806 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3807 | ||
d397712b | 3808 | while (len > 0) { |
d1310b2e CM |
3809 | page = extent_buffer_page(eb, i); |
3810 | WARN_ON(!PageUptodate(page)); | |
3811 | ||
3812 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 3813 | kaddr = page_address(page); |
d1310b2e | 3814 | memset(kaddr + offset, c, cur); |
d1310b2e CM |
3815 | |
3816 | len -= cur; | |
3817 | offset = 0; | |
3818 | i++; | |
3819 | } | |
3820 | } | |
d1310b2e CM |
3821 | |
3822 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
3823 | unsigned long dst_offset, unsigned long src_offset, | |
3824 | unsigned long len) | |
3825 | { | |
3826 | u64 dst_len = dst->len; | |
3827 | size_t cur; | |
3828 | size_t offset; | |
3829 | struct page *page; | |
3830 | char *kaddr; | |
3831 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3832 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
3833 | ||
3834 | WARN_ON(src->len != dst_len); | |
3835 | ||
3836 | offset = (start_offset + dst_offset) & | |
3837 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3838 | ||
d397712b | 3839 | while (len > 0) { |
d1310b2e CM |
3840 | page = extent_buffer_page(dst, i); |
3841 | WARN_ON(!PageUptodate(page)); | |
3842 | ||
3843 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
3844 | ||
a6591715 | 3845 | kaddr = page_address(page); |
d1310b2e | 3846 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
3847 | |
3848 | src_offset += cur; | |
3849 | len -= cur; | |
3850 | offset = 0; | |
3851 | i++; | |
3852 | } | |
3853 | } | |
d1310b2e CM |
3854 | |
3855 | static void move_pages(struct page *dst_page, struct page *src_page, | |
3856 | unsigned long dst_off, unsigned long src_off, | |
3857 | unsigned long len) | |
3858 | { | |
a6591715 | 3859 | char *dst_kaddr = page_address(dst_page); |
d1310b2e CM |
3860 | if (dst_page == src_page) { |
3861 | memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len); | |
3862 | } else { | |
a6591715 | 3863 | char *src_kaddr = page_address(src_page); |
d1310b2e CM |
3864 | char *p = dst_kaddr + dst_off + len; |
3865 | char *s = src_kaddr + src_off + len; | |
3866 | ||
3867 | while (len--) | |
3868 | *--p = *--s; | |
d1310b2e | 3869 | } |
d1310b2e CM |
3870 | } |
3871 | ||
3387206f ST |
3872 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
3873 | { | |
3874 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
3875 | return distance < len; | |
3876 | } | |
3877 | ||
d1310b2e CM |
3878 | static void copy_pages(struct page *dst_page, struct page *src_page, |
3879 | unsigned long dst_off, unsigned long src_off, | |
3880 | unsigned long len) | |
3881 | { | |
a6591715 | 3882 | char *dst_kaddr = page_address(dst_page); |
d1310b2e CM |
3883 | char *src_kaddr; |
3884 | ||
3387206f | 3885 | if (dst_page != src_page) { |
a6591715 | 3886 | src_kaddr = page_address(src_page); |
3387206f | 3887 | } else { |
d1310b2e | 3888 | src_kaddr = dst_kaddr; |
3387206f ST |
3889 | BUG_ON(areas_overlap(src_off, dst_off, len)); |
3890 | } | |
d1310b2e CM |
3891 | |
3892 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
3893 | } |
3894 | ||
3895 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
3896 | unsigned long src_offset, unsigned long len) | |
3897 | { | |
3898 | size_t cur; | |
3899 | size_t dst_off_in_page; | |
3900 | size_t src_off_in_page; | |
3901 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3902 | unsigned long dst_i; | |
3903 | unsigned long src_i; | |
3904 | ||
3905 | if (src_offset + len > dst->len) { | |
d397712b CM |
3906 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
3907 | "len %lu dst len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
3908 | BUG_ON(1); |
3909 | } | |
3910 | if (dst_offset + len > dst->len) { | |
d397712b CM |
3911 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
3912 | "len %lu dst len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
3913 | BUG_ON(1); |
3914 | } | |
3915 | ||
d397712b | 3916 | while (len > 0) { |
d1310b2e CM |
3917 | dst_off_in_page = (start_offset + dst_offset) & |
3918 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3919 | src_off_in_page = (start_offset + src_offset) & | |
3920 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3921 | ||
3922 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
3923 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
3924 | ||
3925 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
3926 | src_off_in_page)); | |
3927 | cur = min_t(unsigned long, cur, | |
3928 | (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); | |
3929 | ||
3930 | copy_pages(extent_buffer_page(dst, dst_i), | |
3931 | extent_buffer_page(dst, src_i), | |
3932 | dst_off_in_page, src_off_in_page, cur); | |
3933 | ||
3934 | src_offset += cur; | |
3935 | dst_offset += cur; | |
3936 | len -= cur; | |
3937 | } | |
3938 | } | |
d1310b2e CM |
3939 | |
3940 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
3941 | unsigned long src_offset, unsigned long len) | |
3942 | { | |
3943 | size_t cur; | |
3944 | size_t dst_off_in_page; | |
3945 | size_t src_off_in_page; | |
3946 | unsigned long dst_end = dst_offset + len - 1; | |
3947 | unsigned long src_end = src_offset + len - 1; | |
3948 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
3949 | unsigned long dst_i; | |
3950 | unsigned long src_i; | |
3951 | ||
3952 | if (src_offset + len > dst->len) { | |
d397712b CM |
3953 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
3954 | "len %lu len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
3955 | BUG_ON(1); |
3956 | } | |
3957 | if (dst_offset + len > dst->len) { | |
d397712b CM |
3958 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
3959 | "len %lu len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
3960 | BUG_ON(1); |
3961 | } | |
3387206f | 3962 | if (!areas_overlap(src_offset, dst_offset, len)) { |
d1310b2e CM |
3963 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
3964 | return; | |
3965 | } | |
d397712b | 3966 | while (len > 0) { |
d1310b2e CM |
3967 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; |
3968 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
3969 | ||
3970 | dst_off_in_page = (start_offset + dst_end) & | |
3971 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3972 | src_off_in_page = (start_offset + src_end) & | |
3973 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
3974 | ||
3975 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
3976 | cur = min(cur, dst_off_in_page + 1); | |
3977 | move_pages(extent_buffer_page(dst, dst_i), | |
3978 | extent_buffer_page(dst, src_i), | |
3979 | dst_off_in_page - cur + 1, | |
3980 | src_off_in_page - cur + 1, cur); | |
3981 | ||
3982 | dst_end -= cur; | |
3983 | src_end -= cur; | |
3984 | len -= cur; | |
3985 | } | |
3986 | } | |
6af118ce | 3987 | |
19fe0a8b MX |
3988 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
3989 | { | |
3990 | struct extent_buffer *eb = | |
3991 | container_of(head, struct extent_buffer, rcu_head); | |
3992 | ||
3993 | btrfs_release_extent_buffer(eb); | |
3994 | } | |
3995 | ||
6af118ce CM |
3996 | int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page) |
3997 | { | |
3998 | u64 start = page_offset(page); | |
3999 | struct extent_buffer *eb; | |
4000 | int ret = 1; | |
6af118ce CM |
4001 | |
4002 | spin_lock(&tree->buffer_lock); | |
19fe0a8b | 4003 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); |
45f49bce CM |
4004 | if (!eb) { |
4005 | spin_unlock(&tree->buffer_lock); | |
4006 | return ret; | |
4007 | } | |
6af118ce | 4008 | |
19fe0a8b | 4009 | if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
6af118ce CM |
4010 | ret = 0; |
4011 | goto out; | |
4012 | } | |
19fe0a8b MX |
4013 | |
4014 | /* | |
4015 | * set @eb->refs to 0 if it is already 1, and then release the @eb. | |
4016 | * Or go back. | |
4017 | */ | |
4018 | if (atomic_cmpxchg(&eb->refs, 1, 0) != 1) { | |
b9473439 CM |
4019 | ret = 0; |
4020 | goto out; | |
4021 | } | |
897ca6e9 | 4022 | |
19fe0a8b | 4023 | radix_tree_delete(&tree->buffer, start >> PAGE_CACHE_SHIFT); |
6af118ce CM |
4024 | out: |
4025 | spin_unlock(&tree->buffer_lock); | |
19fe0a8b MX |
4026 | |
4027 | /* at this point we can safely release the extent buffer */ | |
4028 | if (atomic_read(&eb->refs) == 0) | |
4029 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); | |
6af118ce CM |
4030 | return ret; |
4031 | } |