]>
Commit | Line | Data |
---|---|---|
d1310b2e CM |
1 | #include <linux/bitops.h> |
2 | #include <linux/slab.h> | |
3 | #include <linux/bio.h> | |
4 | #include <linux/mm.h> | |
d1310b2e CM |
5 | #include <linux/pagemap.h> |
6 | #include <linux/page-flags.h> | |
7 | #include <linux/module.h> | |
8 | #include <linux/spinlock.h> | |
9 | #include <linux/blkdev.h> | |
10 | #include <linux/swap.h> | |
d1310b2e CM |
11 | #include <linux/writeback.h> |
12 | #include <linux/pagevec.h> | |
268bb0ce | 13 | #include <linux/prefetch.h> |
90a887c9 | 14 | #include <linux/cleancache.h> |
d1310b2e CM |
15 | #include "extent_io.h" |
16 | #include "extent_map.h" | |
2db04966 | 17 | #include "compat.h" |
902b22f3 DW |
18 | #include "ctree.h" |
19 | #include "btrfs_inode.h" | |
4a54c8c1 | 20 | #include "volumes.h" |
21adbd5c | 21 | #include "check-integrity.h" |
0b32f4bb | 22 | #include "locking.h" |
606686ee | 23 | #include "rcu-string.h" |
d1310b2e | 24 | |
d1310b2e CM |
25 | static struct kmem_cache *extent_state_cache; |
26 | static struct kmem_cache *extent_buffer_cache; | |
27 | ||
28 | static LIST_HEAD(buffers); | |
29 | static LIST_HEAD(states); | |
4bef0848 | 30 | |
b47eda86 | 31 | #define LEAK_DEBUG 0 |
3935127c | 32 | #if LEAK_DEBUG |
d397712b | 33 | static DEFINE_SPINLOCK(leak_lock); |
4bef0848 | 34 | #endif |
d1310b2e | 35 | |
d1310b2e CM |
36 | #define BUFFER_LRU_MAX 64 |
37 | ||
38 | struct tree_entry { | |
39 | u64 start; | |
40 | u64 end; | |
d1310b2e CM |
41 | struct rb_node rb_node; |
42 | }; | |
43 | ||
44 | struct extent_page_data { | |
45 | struct bio *bio; | |
46 | struct extent_io_tree *tree; | |
47 | get_extent_t *get_extent; | |
771ed689 CM |
48 | |
49 | /* tells writepage not to lock the state bits for this range | |
50 | * it still does the unlocking | |
51 | */ | |
ffbd517d CM |
52 | unsigned int extent_locked:1; |
53 | ||
54 | /* tells the submit_bio code to use a WRITE_SYNC */ | |
55 | unsigned int sync_io:1; | |
d1310b2e CM |
56 | }; |
57 | ||
0b32f4bb | 58 | static noinline void flush_write_bio(void *data); |
c2d904e0 JM |
59 | static inline struct btrfs_fs_info * |
60 | tree_fs_info(struct extent_io_tree *tree) | |
61 | { | |
62 | return btrfs_sb(tree->mapping->host->i_sb); | |
63 | } | |
0b32f4bb | 64 | |
d1310b2e CM |
65 | int __init extent_io_init(void) |
66 | { | |
9601e3f6 CH |
67 | extent_state_cache = kmem_cache_create("extent_state", |
68 | sizeof(struct extent_state), 0, | |
69 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
70 | if (!extent_state_cache) |
71 | return -ENOMEM; | |
72 | ||
9601e3f6 CH |
73 | extent_buffer_cache = kmem_cache_create("extent_buffers", |
74 | sizeof(struct extent_buffer), 0, | |
75 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
76 | if (!extent_buffer_cache) |
77 | goto free_state_cache; | |
78 | return 0; | |
79 | ||
80 | free_state_cache: | |
81 | kmem_cache_destroy(extent_state_cache); | |
82 | return -ENOMEM; | |
83 | } | |
84 | ||
85 | void extent_io_exit(void) | |
86 | { | |
87 | struct extent_state *state; | |
2d2ae547 | 88 | struct extent_buffer *eb; |
d1310b2e CM |
89 | |
90 | while (!list_empty(&states)) { | |
2d2ae547 | 91 | state = list_entry(states.next, struct extent_state, leak_list); |
d397712b CM |
92 | printk(KERN_ERR "btrfs state leak: start %llu end %llu " |
93 | "state %lu in tree %p refs %d\n", | |
94 | (unsigned long long)state->start, | |
95 | (unsigned long long)state->end, | |
96 | state->state, state->tree, atomic_read(&state->refs)); | |
2d2ae547 | 97 | list_del(&state->leak_list); |
d1310b2e CM |
98 | kmem_cache_free(extent_state_cache, state); |
99 | ||
100 | } | |
101 | ||
2d2ae547 CM |
102 | while (!list_empty(&buffers)) { |
103 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
d397712b CM |
104 | printk(KERN_ERR "btrfs buffer leak start %llu len %lu " |
105 | "refs %d\n", (unsigned long long)eb->start, | |
106 | eb->len, atomic_read(&eb->refs)); | |
2d2ae547 CM |
107 | list_del(&eb->leak_list); |
108 | kmem_cache_free(extent_buffer_cache, eb); | |
109 | } | |
d1310b2e CM |
110 | if (extent_state_cache) |
111 | kmem_cache_destroy(extent_state_cache); | |
112 | if (extent_buffer_cache) | |
113 | kmem_cache_destroy(extent_buffer_cache); | |
114 | } | |
115 | ||
116 | void extent_io_tree_init(struct extent_io_tree *tree, | |
f993c883 | 117 | struct address_space *mapping) |
d1310b2e | 118 | { |
6bef4d31 | 119 | tree->state = RB_ROOT; |
19fe0a8b | 120 | INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC); |
d1310b2e CM |
121 | tree->ops = NULL; |
122 | tree->dirty_bytes = 0; | |
70dec807 | 123 | spin_lock_init(&tree->lock); |
6af118ce | 124 | spin_lock_init(&tree->buffer_lock); |
d1310b2e | 125 | tree->mapping = mapping; |
d1310b2e | 126 | } |
d1310b2e | 127 | |
b2950863 | 128 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
129 | { |
130 | struct extent_state *state; | |
3935127c | 131 | #if LEAK_DEBUG |
2d2ae547 | 132 | unsigned long flags; |
4bef0848 | 133 | #endif |
d1310b2e CM |
134 | |
135 | state = kmem_cache_alloc(extent_state_cache, mask); | |
2b114d1d | 136 | if (!state) |
d1310b2e CM |
137 | return state; |
138 | state->state = 0; | |
d1310b2e | 139 | state->private = 0; |
70dec807 | 140 | state->tree = NULL; |
3935127c | 141 | #if LEAK_DEBUG |
2d2ae547 CM |
142 | spin_lock_irqsave(&leak_lock, flags); |
143 | list_add(&state->leak_list, &states); | |
144 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 145 | #endif |
d1310b2e CM |
146 | atomic_set(&state->refs, 1); |
147 | init_waitqueue_head(&state->wq); | |
143bede5 | 148 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
149 | return state; |
150 | } | |
d1310b2e | 151 | |
4845e44f | 152 | void free_extent_state(struct extent_state *state) |
d1310b2e | 153 | { |
d1310b2e CM |
154 | if (!state) |
155 | return; | |
156 | if (atomic_dec_and_test(&state->refs)) { | |
3935127c | 157 | #if LEAK_DEBUG |
2d2ae547 | 158 | unsigned long flags; |
4bef0848 | 159 | #endif |
70dec807 | 160 | WARN_ON(state->tree); |
3935127c | 161 | #if LEAK_DEBUG |
2d2ae547 CM |
162 | spin_lock_irqsave(&leak_lock, flags); |
163 | list_del(&state->leak_list); | |
164 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 165 | #endif |
143bede5 | 166 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
167 | kmem_cache_free(extent_state_cache, state); |
168 | } | |
169 | } | |
d1310b2e CM |
170 | |
171 | static struct rb_node *tree_insert(struct rb_root *root, u64 offset, | |
172 | struct rb_node *node) | |
173 | { | |
d397712b CM |
174 | struct rb_node **p = &root->rb_node; |
175 | struct rb_node *parent = NULL; | |
d1310b2e CM |
176 | struct tree_entry *entry; |
177 | ||
d397712b | 178 | while (*p) { |
d1310b2e CM |
179 | parent = *p; |
180 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
181 | ||
182 | if (offset < entry->start) | |
183 | p = &(*p)->rb_left; | |
184 | else if (offset > entry->end) | |
185 | p = &(*p)->rb_right; | |
186 | else | |
187 | return parent; | |
188 | } | |
189 | ||
d1310b2e CM |
190 | rb_link_node(node, parent, p); |
191 | rb_insert_color(node, root); | |
192 | return NULL; | |
193 | } | |
194 | ||
80ea96b1 | 195 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
d1310b2e CM |
196 | struct rb_node **prev_ret, |
197 | struct rb_node **next_ret) | |
198 | { | |
80ea96b1 | 199 | struct rb_root *root = &tree->state; |
d397712b | 200 | struct rb_node *n = root->rb_node; |
d1310b2e CM |
201 | struct rb_node *prev = NULL; |
202 | struct rb_node *orig_prev = NULL; | |
203 | struct tree_entry *entry; | |
204 | struct tree_entry *prev_entry = NULL; | |
205 | ||
d397712b | 206 | while (n) { |
d1310b2e CM |
207 | entry = rb_entry(n, struct tree_entry, rb_node); |
208 | prev = n; | |
209 | prev_entry = entry; | |
210 | ||
211 | if (offset < entry->start) | |
212 | n = n->rb_left; | |
213 | else if (offset > entry->end) | |
214 | n = n->rb_right; | |
d397712b | 215 | else |
d1310b2e CM |
216 | return n; |
217 | } | |
218 | ||
219 | if (prev_ret) { | |
220 | orig_prev = prev; | |
d397712b | 221 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
222 | prev = rb_next(prev); |
223 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
224 | } | |
225 | *prev_ret = prev; | |
226 | prev = orig_prev; | |
227 | } | |
228 | ||
229 | if (next_ret) { | |
230 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 231 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
232 | prev = rb_prev(prev); |
233 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
234 | } | |
235 | *next_ret = prev; | |
236 | } | |
237 | return NULL; | |
238 | } | |
239 | ||
80ea96b1 CM |
240 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
241 | u64 offset) | |
d1310b2e | 242 | { |
70dec807 | 243 | struct rb_node *prev = NULL; |
d1310b2e | 244 | struct rb_node *ret; |
70dec807 | 245 | |
80ea96b1 | 246 | ret = __etree_search(tree, offset, &prev, NULL); |
d397712b | 247 | if (!ret) |
d1310b2e CM |
248 | return prev; |
249 | return ret; | |
250 | } | |
251 | ||
9ed74f2d JB |
252 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, |
253 | struct extent_state *other) | |
254 | { | |
255 | if (tree->ops && tree->ops->merge_extent_hook) | |
256 | tree->ops->merge_extent_hook(tree->mapping->host, new, | |
257 | other); | |
258 | } | |
259 | ||
d1310b2e CM |
260 | /* |
261 | * utility function to look for merge candidates inside a given range. | |
262 | * Any extents with matching state are merged together into a single | |
263 | * extent in the tree. Extents with EXTENT_IO in their state field | |
264 | * are not merged because the end_io handlers need to be able to do | |
265 | * operations on them without sleeping (or doing allocations/splits). | |
266 | * | |
267 | * This should be called with the tree lock held. | |
268 | */ | |
1bf85046 JM |
269 | static void merge_state(struct extent_io_tree *tree, |
270 | struct extent_state *state) | |
d1310b2e CM |
271 | { |
272 | struct extent_state *other; | |
273 | struct rb_node *other_node; | |
274 | ||
5b21f2ed | 275 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
1bf85046 | 276 | return; |
d1310b2e CM |
277 | |
278 | other_node = rb_prev(&state->rb_node); | |
279 | if (other_node) { | |
280 | other = rb_entry(other_node, struct extent_state, rb_node); | |
281 | if (other->end == state->start - 1 && | |
282 | other->state == state->state) { | |
9ed74f2d | 283 | merge_cb(tree, state, other); |
d1310b2e | 284 | state->start = other->start; |
70dec807 | 285 | other->tree = NULL; |
d1310b2e CM |
286 | rb_erase(&other->rb_node, &tree->state); |
287 | free_extent_state(other); | |
288 | } | |
289 | } | |
290 | other_node = rb_next(&state->rb_node); | |
291 | if (other_node) { | |
292 | other = rb_entry(other_node, struct extent_state, rb_node); | |
293 | if (other->start == state->end + 1 && | |
294 | other->state == state->state) { | |
9ed74f2d | 295 | merge_cb(tree, state, other); |
df98b6e2 JB |
296 | state->end = other->end; |
297 | other->tree = NULL; | |
298 | rb_erase(&other->rb_node, &tree->state); | |
299 | free_extent_state(other); | |
d1310b2e CM |
300 | } |
301 | } | |
d1310b2e CM |
302 | } |
303 | ||
1bf85046 | 304 | static void set_state_cb(struct extent_io_tree *tree, |
0ca1f7ce | 305 | struct extent_state *state, int *bits) |
291d673e | 306 | { |
1bf85046 JM |
307 | if (tree->ops && tree->ops->set_bit_hook) |
308 | tree->ops->set_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
309 | } |
310 | ||
311 | static void clear_state_cb(struct extent_io_tree *tree, | |
0ca1f7ce | 312 | struct extent_state *state, int *bits) |
291d673e | 313 | { |
9ed74f2d JB |
314 | if (tree->ops && tree->ops->clear_bit_hook) |
315 | tree->ops->clear_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
316 | } |
317 | ||
3150b699 XG |
318 | static void set_state_bits(struct extent_io_tree *tree, |
319 | struct extent_state *state, int *bits); | |
320 | ||
d1310b2e CM |
321 | /* |
322 | * insert an extent_state struct into the tree. 'bits' are set on the | |
323 | * struct before it is inserted. | |
324 | * | |
325 | * This may return -EEXIST if the extent is already there, in which case the | |
326 | * state struct is freed. | |
327 | * | |
328 | * The tree lock is not taken internally. This is a utility function and | |
329 | * probably isn't what you want to call (see set/clear_extent_bit). | |
330 | */ | |
331 | static int insert_state(struct extent_io_tree *tree, | |
332 | struct extent_state *state, u64 start, u64 end, | |
0ca1f7ce | 333 | int *bits) |
d1310b2e CM |
334 | { |
335 | struct rb_node *node; | |
336 | ||
337 | if (end < start) { | |
d397712b CM |
338 | printk(KERN_ERR "btrfs end < start %llu %llu\n", |
339 | (unsigned long long)end, | |
340 | (unsigned long long)start); | |
d1310b2e CM |
341 | WARN_ON(1); |
342 | } | |
d1310b2e CM |
343 | state->start = start; |
344 | state->end = end; | |
9ed74f2d | 345 | |
3150b699 XG |
346 | set_state_bits(tree, state, bits); |
347 | ||
d1310b2e CM |
348 | node = tree_insert(&tree->state, end, &state->rb_node); |
349 | if (node) { | |
350 | struct extent_state *found; | |
351 | found = rb_entry(node, struct extent_state, rb_node); | |
d397712b CM |
352 | printk(KERN_ERR "btrfs found node %llu %llu on insert of " |
353 | "%llu %llu\n", (unsigned long long)found->start, | |
354 | (unsigned long long)found->end, | |
355 | (unsigned long long)start, (unsigned long long)end); | |
d1310b2e CM |
356 | return -EEXIST; |
357 | } | |
70dec807 | 358 | state->tree = tree; |
d1310b2e CM |
359 | merge_state(tree, state); |
360 | return 0; | |
361 | } | |
362 | ||
1bf85046 | 363 | static void split_cb(struct extent_io_tree *tree, struct extent_state *orig, |
9ed74f2d JB |
364 | u64 split) |
365 | { | |
366 | if (tree->ops && tree->ops->split_extent_hook) | |
1bf85046 | 367 | tree->ops->split_extent_hook(tree->mapping->host, orig, split); |
9ed74f2d JB |
368 | } |
369 | ||
d1310b2e CM |
370 | /* |
371 | * split a given extent state struct in two, inserting the preallocated | |
372 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
373 | * offset inside 'orig' where it should be split. | |
374 | * | |
375 | * Before calling, | |
376 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
377 | * are two extent state structs in the tree: | |
378 | * prealloc: [orig->start, split - 1] | |
379 | * orig: [ split, orig->end ] | |
380 | * | |
381 | * The tree locks are not taken by this function. They need to be held | |
382 | * by the caller. | |
383 | */ | |
384 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
385 | struct extent_state *prealloc, u64 split) | |
386 | { | |
387 | struct rb_node *node; | |
9ed74f2d JB |
388 | |
389 | split_cb(tree, orig, split); | |
390 | ||
d1310b2e CM |
391 | prealloc->start = orig->start; |
392 | prealloc->end = split - 1; | |
393 | prealloc->state = orig->state; | |
394 | orig->start = split; | |
395 | ||
396 | node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node); | |
397 | if (node) { | |
d1310b2e CM |
398 | free_extent_state(prealloc); |
399 | return -EEXIST; | |
400 | } | |
70dec807 | 401 | prealloc->tree = tree; |
d1310b2e CM |
402 | return 0; |
403 | } | |
404 | ||
cdc6a395 LZ |
405 | static struct extent_state *next_state(struct extent_state *state) |
406 | { | |
407 | struct rb_node *next = rb_next(&state->rb_node); | |
408 | if (next) | |
409 | return rb_entry(next, struct extent_state, rb_node); | |
410 | else | |
411 | return NULL; | |
412 | } | |
413 | ||
d1310b2e CM |
414 | /* |
415 | * utility function to clear some bits in an extent state struct. | |
1b303fc0 | 416 | * it will optionally wake up any one waiting on this state (wake == 1). |
d1310b2e CM |
417 | * |
418 | * If no bits are set on the state struct after clearing things, the | |
419 | * struct is freed and removed from the tree | |
420 | */ | |
cdc6a395 LZ |
421 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
422 | struct extent_state *state, | |
423 | int *bits, int wake) | |
d1310b2e | 424 | { |
cdc6a395 | 425 | struct extent_state *next; |
0ca1f7ce | 426 | int bits_to_clear = *bits & ~EXTENT_CTLBITS; |
d1310b2e | 427 | |
0ca1f7ce | 428 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
429 | u64 range = state->end - state->start + 1; |
430 | WARN_ON(range > tree->dirty_bytes); | |
431 | tree->dirty_bytes -= range; | |
432 | } | |
291d673e | 433 | clear_state_cb(tree, state, bits); |
32c00aff | 434 | state->state &= ~bits_to_clear; |
d1310b2e CM |
435 | if (wake) |
436 | wake_up(&state->wq); | |
0ca1f7ce | 437 | if (state->state == 0) { |
cdc6a395 | 438 | next = next_state(state); |
70dec807 | 439 | if (state->tree) { |
d1310b2e | 440 | rb_erase(&state->rb_node, &tree->state); |
70dec807 | 441 | state->tree = NULL; |
d1310b2e CM |
442 | free_extent_state(state); |
443 | } else { | |
444 | WARN_ON(1); | |
445 | } | |
446 | } else { | |
447 | merge_state(tree, state); | |
cdc6a395 | 448 | next = next_state(state); |
d1310b2e | 449 | } |
cdc6a395 | 450 | return next; |
d1310b2e CM |
451 | } |
452 | ||
8233767a XG |
453 | static struct extent_state * |
454 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
455 | { | |
456 | if (!prealloc) | |
457 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
458 | ||
459 | return prealloc; | |
460 | } | |
461 | ||
c2d904e0 JM |
462 | void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
463 | { | |
464 | btrfs_panic(tree_fs_info(tree), err, "Locking error: " | |
465 | "Extent tree was modified by another " | |
466 | "thread while locked."); | |
467 | } | |
468 | ||
d1310b2e CM |
469 | /* |
470 | * clear some bits on a range in the tree. This may require splitting | |
471 | * or inserting elements in the tree, so the gfp mask is used to | |
472 | * indicate which allocations or sleeping are allowed. | |
473 | * | |
474 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
475 | * the given range from the tree regardless of state (ie for truncate). | |
476 | * | |
477 | * the range [start, end] is inclusive. | |
478 | * | |
6763af84 | 479 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e CM |
480 | */ |
481 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
2c64c53d CM |
482 | int bits, int wake, int delete, |
483 | struct extent_state **cached_state, | |
484 | gfp_t mask) | |
d1310b2e CM |
485 | { |
486 | struct extent_state *state; | |
2c64c53d | 487 | struct extent_state *cached; |
d1310b2e CM |
488 | struct extent_state *prealloc = NULL; |
489 | struct rb_node *node; | |
5c939df5 | 490 | u64 last_end; |
d1310b2e | 491 | int err; |
2ac55d41 | 492 | int clear = 0; |
d1310b2e | 493 | |
0ca1f7ce YZ |
494 | if (delete) |
495 | bits |= ~EXTENT_CTLBITS; | |
496 | bits |= EXTENT_FIRST_DELALLOC; | |
497 | ||
2ac55d41 JB |
498 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
499 | clear = 1; | |
d1310b2e CM |
500 | again: |
501 | if (!prealloc && (mask & __GFP_WAIT)) { | |
502 | prealloc = alloc_extent_state(mask); | |
503 | if (!prealloc) | |
504 | return -ENOMEM; | |
505 | } | |
506 | ||
cad321ad | 507 | spin_lock(&tree->lock); |
2c64c53d CM |
508 | if (cached_state) { |
509 | cached = *cached_state; | |
2ac55d41 JB |
510 | |
511 | if (clear) { | |
512 | *cached_state = NULL; | |
513 | cached_state = NULL; | |
514 | } | |
515 | ||
df98b6e2 JB |
516 | if (cached && cached->tree && cached->start <= start && |
517 | cached->end > start) { | |
2ac55d41 JB |
518 | if (clear) |
519 | atomic_dec(&cached->refs); | |
2c64c53d | 520 | state = cached; |
42daec29 | 521 | goto hit_next; |
2c64c53d | 522 | } |
2ac55d41 JB |
523 | if (clear) |
524 | free_extent_state(cached); | |
2c64c53d | 525 | } |
d1310b2e CM |
526 | /* |
527 | * this search will find the extents that end after | |
528 | * our range starts | |
529 | */ | |
80ea96b1 | 530 | node = tree_search(tree, start); |
d1310b2e CM |
531 | if (!node) |
532 | goto out; | |
533 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 534 | hit_next: |
d1310b2e CM |
535 | if (state->start > end) |
536 | goto out; | |
537 | WARN_ON(state->end < start); | |
5c939df5 | 538 | last_end = state->end; |
d1310b2e | 539 | |
0449314a | 540 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
541 | if (!(state->state & bits)) { |
542 | state = next_state(state); | |
0449314a | 543 | goto next; |
cdc6a395 | 544 | } |
0449314a | 545 | |
d1310b2e CM |
546 | /* |
547 | * | ---- desired range ---- | | |
548 | * | state | or | |
549 | * | ------------- state -------------- | | |
550 | * | |
551 | * We need to split the extent we found, and may flip | |
552 | * bits on second half. | |
553 | * | |
554 | * If the extent we found extends past our range, we | |
555 | * just split and search again. It'll get split again | |
556 | * the next time though. | |
557 | * | |
558 | * If the extent we found is inside our range, we clear | |
559 | * the desired bit on it. | |
560 | */ | |
561 | ||
562 | if (state->start < start) { | |
8233767a XG |
563 | prealloc = alloc_extent_state_atomic(prealloc); |
564 | BUG_ON(!prealloc); | |
d1310b2e | 565 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
566 | if (err) |
567 | extent_io_tree_panic(tree, err); | |
568 | ||
d1310b2e CM |
569 | prealloc = NULL; |
570 | if (err) | |
571 | goto out; | |
572 | if (state->end <= end) { | |
d1ac6e41 LB |
573 | state = clear_state_bit(tree, state, &bits, wake); |
574 | goto next; | |
d1310b2e CM |
575 | } |
576 | goto search_again; | |
577 | } | |
578 | /* | |
579 | * | ---- desired range ---- | | |
580 | * | state | | |
581 | * We need to split the extent, and clear the bit | |
582 | * on the first half | |
583 | */ | |
584 | if (state->start <= end && state->end > end) { | |
8233767a XG |
585 | prealloc = alloc_extent_state_atomic(prealloc); |
586 | BUG_ON(!prealloc); | |
d1310b2e | 587 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
588 | if (err) |
589 | extent_io_tree_panic(tree, err); | |
590 | ||
d1310b2e CM |
591 | if (wake) |
592 | wake_up(&state->wq); | |
42daec29 | 593 | |
6763af84 | 594 | clear_state_bit(tree, prealloc, &bits, wake); |
9ed74f2d | 595 | |
d1310b2e CM |
596 | prealloc = NULL; |
597 | goto out; | |
598 | } | |
42daec29 | 599 | |
cdc6a395 | 600 | state = clear_state_bit(tree, state, &bits, wake); |
0449314a | 601 | next: |
5c939df5 YZ |
602 | if (last_end == (u64)-1) |
603 | goto out; | |
604 | start = last_end + 1; | |
cdc6a395 | 605 | if (start <= end && state && !need_resched()) |
692e5759 | 606 | goto hit_next; |
d1310b2e CM |
607 | goto search_again; |
608 | ||
609 | out: | |
cad321ad | 610 | spin_unlock(&tree->lock); |
d1310b2e CM |
611 | if (prealloc) |
612 | free_extent_state(prealloc); | |
613 | ||
6763af84 | 614 | return 0; |
d1310b2e CM |
615 | |
616 | search_again: | |
617 | if (start > end) | |
618 | goto out; | |
cad321ad | 619 | spin_unlock(&tree->lock); |
d1310b2e CM |
620 | if (mask & __GFP_WAIT) |
621 | cond_resched(); | |
622 | goto again; | |
623 | } | |
d1310b2e | 624 | |
143bede5 JM |
625 | static void wait_on_state(struct extent_io_tree *tree, |
626 | struct extent_state *state) | |
641f5219 CH |
627 | __releases(tree->lock) |
628 | __acquires(tree->lock) | |
d1310b2e CM |
629 | { |
630 | DEFINE_WAIT(wait); | |
631 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 632 | spin_unlock(&tree->lock); |
d1310b2e | 633 | schedule(); |
cad321ad | 634 | spin_lock(&tree->lock); |
d1310b2e | 635 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
636 | } |
637 | ||
638 | /* | |
639 | * waits for one or more bits to clear on a range in the state tree. | |
640 | * The range [start, end] is inclusive. | |
641 | * The tree lock is taken by this function | |
642 | */ | |
143bede5 | 643 | void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits) |
d1310b2e CM |
644 | { |
645 | struct extent_state *state; | |
646 | struct rb_node *node; | |
647 | ||
cad321ad | 648 | spin_lock(&tree->lock); |
d1310b2e CM |
649 | again: |
650 | while (1) { | |
651 | /* | |
652 | * this search will find all the extents that end after | |
653 | * our range starts | |
654 | */ | |
80ea96b1 | 655 | node = tree_search(tree, start); |
d1310b2e CM |
656 | if (!node) |
657 | break; | |
658 | ||
659 | state = rb_entry(node, struct extent_state, rb_node); | |
660 | ||
661 | if (state->start > end) | |
662 | goto out; | |
663 | ||
664 | if (state->state & bits) { | |
665 | start = state->start; | |
666 | atomic_inc(&state->refs); | |
667 | wait_on_state(tree, state); | |
668 | free_extent_state(state); | |
669 | goto again; | |
670 | } | |
671 | start = state->end + 1; | |
672 | ||
673 | if (start > end) | |
674 | break; | |
675 | ||
ded91f08 | 676 | cond_resched_lock(&tree->lock); |
d1310b2e CM |
677 | } |
678 | out: | |
cad321ad | 679 | spin_unlock(&tree->lock); |
d1310b2e | 680 | } |
d1310b2e | 681 | |
1bf85046 | 682 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 683 | struct extent_state *state, |
0ca1f7ce | 684 | int *bits) |
d1310b2e | 685 | { |
0ca1f7ce | 686 | int bits_to_set = *bits & ~EXTENT_CTLBITS; |
9ed74f2d | 687 | |
1bf85046 | 688 | set_state_cb(tree, state, bits); |
0ca1f7ce | 689 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
690 | u64 range = state->end - state->start + 1; |
691 | tree->dirty_bytes += range; | |
692 | } | |
0ca1f7ce | 693 | state->state |= bits_to_set; |
d1310b2e CM |
694 | } |
695 | ||
2c64c53d CM |
696 | static void cache_state(struct extent_state *state, |
697 | struct extent_state **cached_ptr) | |
698 | { | |
699 | if (cached_ptr && !(*cached_ptr)) { | |
700 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) { | |
701 | *cached_ptr = state; | |
702 | atomic_inc(&state->refs); | |
703 | } | |
704 | } | |
705 | } | |
706 | ||
507903b8 AJ |
707 | static void uncache_state(struct extent_state **cached_ptr) |
708 | { | |
709 | if (cached_ptr && (*cached_ptr)) { | |
710 | struct extent_state *state = *cached_ptr; | |
109b36a2 CM |
711 | *cached_ptr = NULL; |
712 | free_extent_state(state); | |
507903b8 AJ |
713 | } |
714 | } | |
715 | ||
d1310b2e | 716 | /* |
1edbb734 CM |
717 | * set some bits on a range in the tree. This may require allocations or |
718 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 719 | * |
1edbb734 CM |
720 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
721 | * part of the range already has the desired bits set. The start of the | |
722 | * existing range is returned in failed_start in this case. | |
d1310b2e | 723 | * |
1edbb734 | 724 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 725 | */ |
1edbb734 | 726 | |
3fbe5c02 JM |
727 | static int __must_check |
728 | __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
729 | int bits, int exclusive_bits, u64 *failed_start, | |
730 | struct extent_state **cached_state, gfp_t mask) | |
d1310b2e CM |
731 | { |
732 | struct extent_state *state; | |
733 | struct extent_state *prealloc = NULL; | |
734 | struct rb_node *node; | |
d1310b2e | 735 | int err = 0; |
d1310b2e CM |
736 | u64 last_start; |
737 | u64 last_end; | |
42daec29 | 738 | |
0ca1f7ce | 739 | bits |= EXTENT_FIRST_DELALLOC; |
d1310b2e CM |
740 | again: |
741 | if (!prealloc && (mask & __GFP_WAIT)) { | |
742 | prealloc = alloc_extent_state(mask); | |
8233767a | 743 | BUG_ON(!prealloc); |
d1310b2e CM |
744 | } |
745 | ||
cad321ad | 746 | spin_lock(&tree->lock); |
9655d298 CM |
747 | if (cached_state && *cached_state) { |
748 | state = *cached_state; | |
df98b6e2 JB |
749 | if (state->start <= start && state->end > start && |
750 | state->tree) { | |
9655d298 CM |
751 | node = &state->rb_node; |
752 | goto hit_next; | |
753 | } | |
754 | } | |
d1310b2e CM |
755 | /* |
756 | * this search will find all the extents that end after | |
757 | * our range starts. | |
758 | */ | |
80ea96b1 | 759 | node = tree_search(tree, start); |
d1310b2e | 760 | if (!node) { |
8233767a XG |
761 | prealloc = alloc_extent_state_atomic(prealloc); |
762 | BUG_ON(!prealloc); | |
0ca1f7ce | 763 | err = insert_state(tree, prealloc, start, end, &bits); |
c2d904e0 JM |
764 | if (err) |
765 | extent_io_tree_panic(tree, err); | |
766 | ||
d1310b2e | 767 | prealloc = NULL; |
d1310b2e CM |
768 | goto out; |
769 | } | |
d1310b2e | 770 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 771 | hit_next: |
d1310b2e CM |
772 | last_start = state->start; |
773 | last_end = state->end; | |
774 | ||
775 | /* | |
776 | * | ---- desired range ---- | | |
777 | * | state | | |
778 | * | |
779 | * Just lock what we found and keep going | |
780 | */ | |
781 | if (state->start == start && state->end <= end) { | |
1edbb734 | 782 | if (state->state & exclusive_bits) { |
d1310b2e CM |
783 | *failed_start = state->start; |
784 | err = -EEXIST; | |
785 | goto out; | |
786 | } | |
42daec29 | 787 | |
1bf85046 | 788 | set_state_bits(tree, state, &bits); |
2c64c53d | 789 | cache_state(state, cached_state); |
d1310b2e | 790 | merge_state(tree, state); |
5c939df5 YZ |
791 | if (last_end == (u64)-1) |
792 | goto out; | |
793 | start = last_end + 1; | |
d1ac6e41 LB |
794 | state = next_state(state); |
795 | if (start < end && state && state->start == start && | |
796 | !need_resched()) | |
797 | goto hit_next; | |
d1310b2e CM |
798 | goto search_again; |
799 | } | |
800 | ||
801 | /* | |
802 | * | ---- desired range ---- | | |
803 | * | state | | |
804 | * or | |
805 | * | ------------- state -------------- | | |
806 | * | |
807 | * We need to split the extent we found, and may flip bits on | |
808 | * second half. | |
809 | * | |
810 | * If the extent we found extends past our | |
811 | * range, we just split and search again. It'll get split | |
812 | * again the next time though. | |
813 | * | |
814 | * If the extent we found is inside our range, we set the | |
815 | * desired bit on it. | |
816 | */ | |
817 | if (state->start < start) { | |
1edbb734 | 818 | if (state->state & exclusive_bits) { |
d1310b2e CM |
819 | *failed_start = start; |
820 | err = -EEXIST; | |
821 | goto out; | |
822 | } | |
8233767a XG |
823 | |
824 | prealloc = alloc_extent_state_atomic(prealloc); | |
825 | BUG_ON(!prealloc); | |
d1310b2e | 826 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
827 | if (err) |
828 | extent_io_tree_panic(tree, err); | |
829 | ||
d1310b2e CM |
830 | prealloc = NULL; |
831 | if (err) | |
832 | goto out; | |
833 | if (state->end <= end) { | |
1bf85046 | 834 | set_state_bits(tree, state, &bits); |
2c64c53d | 835 | cache_state(state, cached_state); |
d1310b2e | 836 | merge_state(tree, state); |
5c939df5 YZ |
837 | if (last_end == (u64)-1) |
838 | goto out; | |
839 | start = last_end + 1; | |
d1ac6e41 LB |
840 | state = next_state(state); |
841 | if (start < end && state && state->start == start && | |
842 | !need_resched()) | |
843 | goto hit_next; | |
d1310b2e CM |
844 | } |
845 | goto search_again; | |
846 | } | |
847 | /* | |
848 | * | ---- desired range ---- | | |
849 | * | state | or | state | | |
850 | * | |
851 | * There's a hole, we need to insert something in it and | |
852 | * ignore the extent we found. | |
853 | */ | |
854 | if (state->start > start) { | |
855 | u64 this_end; | |
856 | if (end < last_start) | |
857 | this_end = end; | |
858 | else | |
d397712b | 859 | this_end = last_start - 1; |
8233767a XG |
860 | |
861 | prealloc = alloc_extent_state_atomic(prealloc); | |
862 | BUG_ON(!prealloc); | |
c7f895a2 XG |
863 | |
864 | /* | |
865 | * Avoid to free 'prealloc' if it can be merged with | |
866 | * the later extent. | |
867 | */ | |
d1310b2e | 868 | err = insert_state(tree, prealloc, start, this_end, |
0ca1f7ce | 869 | &bits); |
c2d904e0 JM |
870 | if (err) |
871 | extent_io_tree_panic(tree, err); | |
872 | ||
9ed74f2d JB |
873 | cache_state(prealloc, cached_state); |
874 | prealloc = NULL; | |
d1310b2e CM |
875 | start = this_end + 1; |
876 | goto search_again; | |
877 | } | |
878 | /* | |
879 | * | ---- desired range ---- | | |
880 | * | state | | |
881 | * We need to split the extent, and set the bit | |
882 | * on the first half | |
883 | */ | |
884 | if (state->start <= end && state->end > end) { | |
1edbb734 | 885 | if (state->state & exclusive_bits) { |
d1310b2e CM |
886 | *failed_start = start; |
887 | err = -EEXIST; | |
888 | goto out; | |
889 | } | |
8233767a XG |
890 | |
891 | prealloc = alloc_extent_state_atomic(prealloc); | |
892 | BUG_ON(!prealloc); | |
d1310b2e | 893 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
894 | if (err) |
895 | extent_io_tree_panic(tree, err); | |
d1310b2e | 896 | |
1bf85046 | 897 | set_state_bits(tree, prealloc, &bits); |
2c64c53d | 898 | cache_state(prealloc, cached_state); |
d1310b2e CM |
899 | merge_state(tree, prealloc); |
900 | prealloc = NULL; | |
901 | goto out; | |
902 | } | |
903 | ||
904 | goto search_again; | |
905 | ||
906 | out: | |
cad321ad | 907 | spin_unlock(&tree->lock); |
d1310b2e CM |
908 | if (prealloc) |
909 | free_extent_state(prealloc); | |
910 | ||
911 | return err; | |
912 | ||
913 | search_again: | |
914 | if (start > end) | |
915 | goto out; | |
cad321ad | 916 | spin_unlock(&tree->lock); |
d1310b2e CM |
917 | if (mask & __GFP_WAIT) |
918 | cond_resched(); | |
919 | goto again; | |
920 | } | |
d1310b2e | 921 | |
3fbe5c02 JM |
922 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits, |
923 | u64 *failed_start, struct extent_state **cached_state, | |
924 | gfp_t mask) | |
925 | { | |
926 | return __set_extent_bit(tree, start, end, bits, 0, failed_start, | |
927 | cached_state, mask); | |
928 | } | |
929 | ||
930 | ||
462d6fac JB |
931 | /** |
932 | * convert_extent - convert all bits in a given range from one bit to another | |
933 | * @tree: the io tree to search | |
934 | * @start: the start offset in bytes | |
935 | * @end: the end offset in bytes (inclusive) | |
936 | * @bits: the bits to set in this range | |
937 | * @clear_bits: the bits to clear in this range | |
938 | * @mask: the allocation mask | |
939 | * | |
940 | * This will go through and set bits for the given range. If any states exist | |
941 | * already in this range they are set with the given bit and cleared of the | |
942 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
943 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
944 | * boundary bits like LOCK. | |
945 | */ | |
946 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
947 | int bits, int clear_bits, gfp_t mask) | |
948 | { | |
949 | struct extent_state *state; | |
950 | struct extent_state *prealloc = NULL; | |
951 | struct rb_node *node; | |
952 | int err = 0; | |
953 | u64 last_start; | |
954 | u64 last_end; | |
955 | ||
956 | again: | |
957 | if (!prealloc && (mask & __GFP_WAIT)) { | |
958 | prealloc = alloc_extent_state(mask); | |
959 | if (!prealloc) | |
960 | return -ENOMEM; | |
961 | } | |
962 | ||
963 | spin_lock(&tree->lock); | |
964 | /* | |
965 | * this search will find all the extents that end after | |
966 | * our range starts. | |
967 | */ | |
968 | node = tree_search(tree, start); | |
969 | if (!node) { | |
970 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
971 | if (!prealloc) { |
972 | err = -ENOMEM; | |
973 | goto out; | |
974 | } | |
462d6fac JB |
975 | err = insert_state(tree, prealloc, start, end, &bits); |
976 | prealloc = NULL; | |
c2d904e0 JM |
977 | if (err) |
978 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
979 | goto out; |
980 | } | |
981 | state = rb_entry(node, struct extent_state, rb_node); | |
982 | hit_next: | |
983 | last_start = state->start; | |
984 | last_end = state->end; | |
985 | ||
986 | /* | |
987 | * | ---- desired range ---- | | |
988 | * | state | | |
989 | * | |
990 | * Just lock what we found and keep going | |
991 | */ | |
992 | if (state->start == start && state->end <= end) { | |
462d6fac | 993 | set_state_bits(tree, state, &bits); |
d1ac6e41 | 994 | state = clear_state_bit(tree, state, &clear_bits, 0); |
462d6fac JB |
995 | if (last_end == (u64)-1) |
996 | goto out; | |
462d6fac | 997 | start = last_end + 1; |
d1ac6e41 LB |
998 | if (start < end && state && state->start == start && |
999 | !need_resched()) | |
1000 | goto hit_next; | |
462d6fac JB |
1001 | goto search_again; |
1002 | } | |
1003 | ||
1004 | /* | |
1005 | * | ---- desired range ---- | | |
1006 | * | state | | |
1007 | * or | |
1008 | * | ------------- state -------------- | | |
1009 | * | |
1010 | * We need to split the extent we found, and may flip bits on | |
1011 | * second half. | |
1012 | * | |
1013 | * If the extent we found extends past our | |
1014 | * range, we just split and search again. It'll get split | |
1015 | * again the next time though. | |
1016 | * | |
1017 | * If the extent we found is inside our range, we set the | |
1018 | * desired bit on it. | |
1019 | */ | |
1020 | if (state->start < start) { | |
1021 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1022 | if (!prealloc) { |
1023 | err = -ENOMEM; | |
1024 | goto out; | |
1025 | } | |
462d6fac | 1026 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1027 | if (err) |
1028 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1029 | prealloc = NULL; |
1030 | if (err) | |
1031 | goto out; | |
1032 | if (state->end <= end) { | |
1033 | set_state_bits(tree, state, &bits); | |
d1ac6e41 | 1034 | state = clear_state_bit(tree, state, &clear_bits, 0); |
462d6fac JB |
1035 | if (last_end == (u64)-1) |
1036 | goto out; | |
1037 | start = last_end + 1; | |
d1ac6e41 LB |
1038 | if (start < end && state && state->start == start && |
1039 | !need_resched()) | |
1040 | goto hit_next; | |
462d6fac JB |
1041 | } |
1042 | goto search_again; | |
1043 | } | |
1044 | /* | |
1045 | * | ---- desired range ---- | | |
1046 | * | state | or | state | | |
1047 | * | |
1048 | * There's a hole, we need to insert something in it and | |
1049 | * ignore the extent we found. | |
1050 | */ | |
1051 | if (state->start > start) { | |
1052 | u64 this_end; | |
1053 | if (end < last_start) | |
1054 | this_end = end; | |
1055 | else | |
1056 | this_end = last_start - 1; | |
1057 | ||
1058 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1059 | if (!prealloc) { |
1060 | err = -ENOMEM; | |
1061 | goto out; | |
1062 | } | |
462d6fac JB |
1063 | |
1064 | /* | |
1065 | * Avoid to free 'prealloc' if it can be merged with | |
1066 | * the later extent. | |
1067 | */ | |
1068 | err = insert_state(tree, prealloc, start, this_end, | |
1069 | &bits); | |
c2d904e0 JM |
1070 | if (err) |
1071 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1072 | prealloc = NULL; |
1073 | start = this_end + 1; | |
1074 | goto search_again; | |
1075 | } | |
1076 | /* | |
1077 | * | ---- desired range ---- | | |
1078 | * | state | | |
1079 | * We need to split the extent, and set the bit | |
1080 | * on the first half | |
1081 | */ | |
1082 | if (state->start <= end && state->end > end) { | |
1083 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1084 | if (!prealloc) { |
1085 | err = -ENOMEM; | |
1086 | goto out; | |
1087 | } | |
462d6fac JB |
1088 | |
1089 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1090 | if (err) |
1091 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1092 | |
1093 | set_state_bits(tree, prealloc, &bits); | |
1094 | clear_state_bit(tree, prealloc, &clear_bits, 0); | |
462d6fac JB |
1095 | prealloc = NULL; |
1096 | goto out; | |
1097 | } | |
1098 | ||
1099 | goto search_again; | |
1100 | ||
1101 | out: | |
1102 | spin_unlock(&tree->lock); | |
1103 | if (prealloc) | |
1104 | free_extent_state(prealloc); | |
1105 | ||
1106 | return err; | |
1107 | ||
1108 | search_again: | |
1109 | if (start > end) | |
1110 | goto out; | |
1111 | spin_unlock(&tree->lock); | |
1112 | if (mask & __GFP_WAIT) | |
1113 | cond_resched(); | |
1114 | goto again; | |
1115 | } | |
1116 | ||
d1310b2e CM |
1117 | /* wrappers around set/clear extent bit */ |
1118 | int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
1119 | gfp_t mask) | |
1120 | { | |
3fbe5c02 | 1121 | return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL, |
2c64c53d | 1122 | NULL, mask); |
d1310b2e | 1123 | } |
d1310b2e CM |
1124 | |
1125 | int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
1126 | int bits, gfp_t mask) | |
1127 | { | |
3fbe5c02 | 1128 | return set_extent_bit(tree, start, end, bits, NULL, |
2c64c53d | 1129 | NULL, mask); |
d1310b2e | 1130 | } |
d1310b2e CM |
1131 | |
1132 | int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
1133 | int bits, gfp_t mask) | |
1134 | { | |
2c64c53d | 1135 | return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask); |
d1310b2e | 1136 | } |
d1310b2e CM |
1137 | |
1138 | int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, | |
2ac55d41 | 1139 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
1140 | { |
1141 | return set_extent_bit(tree, start, end, | |
fee187d9 | 1142 | EXTENT_DELALLOC | EXTENT_UPTODATE, |
3fbe5c02 | 1143 | NULL, cached_state, mask); |
d1310b2e | 1144 | } |
d1310b2e CM |
1145 | |
1146 | int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
1147 | gfp_t mask) | |
1148 | { | |
1149 | return clear_extent_bit(tree, start, end, | |
32c00aff | 1150 | EXTENT_DIRTY | EXTENT_DELALLOC | |
0ca1f7ce | 1151 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask); |
d1310b2e | 1152 | } |
d1310b2e CM |
1153 | |
1154 | int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, | |
1155 | gfp_t mask) | |
1156 | { | |
3fbe5c02 | 1157 | return set_extent_bit(tree, start, end, EXTENT_NEW, NULL, |
2c64c53d | 1158 | NULL, mask); |
d1310b2e | 1159 | } |
d1310b2e | 1160 | |
d1310b2e | 1161 | int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
507903b8 | 1162 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e | 1163 | { |
507903b8 | 1164 | return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, |
3fbe5c02 | 1165 | cached_state, mask); |
d1310b2e | 1166 | } |
d1310b2e | 1167 | |
5fd02043 JB |
1168 | int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
1169 | struct extent_state **cached_state, gfp_t mask) | |
d1310b2e | 1170 | { |
2c64c53d | 1171 | return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, |
2ac55d41 | 1172 | cached_state, mask); |
d1310b2e | 1173 | } |
d1310b2e | 1174 | |
d352ac68 CM |
1175 | /* |
1176 | * either insert or lock state struct between start and end use mask to tell | |
1177 | * us if waiting is desired. | |
1178 | */ | |
1edbb734 | 1179 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
d0082371 | 1180 | int bits, struct extent_state **cached_state) |
d1310b2e CM |
1181 | { |
1182 | int err; | |
1183 | u64 failed_start; | |
1184 | while (1) { | |
3fbe5c02 JM |
1185 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits, |
1186 | EXTENT_LOCKED, &failed_start, | |
1187 | cached_state, GFP_NOFS); | |
d0082371 | 1188 | if (err == -EEXIST) { |
d1310b2e CM |
1189 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1190 | start = failed_start; | |
d0082371 | 1191 | } else |
d1310b2e | 1192 | break; |
d1310b2e CM |
1193 | WARN_ON(start > end); |
1194 | } | |
1195 | return err; | |
1196 | } | |
d1310b2e | 1197 | |
d0082371 | 1198 | int lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
1edbb734 | 1199 | { |
d0082371 | 1200 | return lock_extent_bits(tree, start, end, 0, NULL); |
1edbb734 CM |
1201 | } |
1202 | ||
d0082371 | 1203 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1204 | { |
1205 | int err; | |
1206 | u64 failed_start; | |
1207 | ||
3fbe5c02 JM |
1208 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
1209 | &failed_start, NULL, GFP_NOFS); | |
6643558d YZ |
1210 | if (err == -EEXIST) { |
1211 | if (failed_start > start) | |
1212 | clear_extent_bit(tree, start, failed_start - 1, | |
d0082371 | 1213 | EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS); |
25179201 | 1214 | return 0; |
6643558d | 1215 | } |
25179201 JB |
1216 | return 1; |
1217 | } | |
25179201 | 1218 | |
2c64c53d CM |
1219 | int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end, |
1220 | struct extent_state **cached, gfp_t mask) | |
1221 | { | |
1222 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, | |
1223 | mask); | |
1224 | } | |
1225 | ||
d0082371 | 1226 | int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e | 1227 | { |
2c64c53d | 1228 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL, |
d0082371 | 1229 | GFP_NOFS); |
d1310b2e | 1230 | } |
d1310b2e | 1231 | |
d1310b2e CM |
1232 | /* |
1233 | * helper function to set both pages and extents in the tree writeback | |
1234 | */ | |
b2950863 | 1235 | static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e CM |
1236 | { |
1237 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1238 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1239 | struct page *page; | |
1240 | ||
1241 | while (index <= end_index) { | |
1242 | page = find_get_page(tree->mapping, index); | |
79787eaa | 1243 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ |
d1310b2e CM |
1244 | set_page_writeback(page); |
1245 | page_cache_release(page); | |
1246 | index++; | |
1247 | } | |
d1310b2e CM |
1248 | return 0; |
1249 | } | |
d1310b2e | 1250 | |
d352ac68 CM |
1251 | /* find the first state struct with 'bits' set after 'start', and |
1252 | * return it. tree->lock must be held. NULL will returned if | |
1253 | * nothing was found after 'start' | |
1254 | */ | |
d7fc640e CM |
1255 | struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree, |
1256 | u64 start, int bits) | |
1257 | { | |
1258 | struct rb_node *node; | |
1259 | struct extent_state *state; | |
1260 | ||
1261 | /* | |
1262 | * this search will find all the extents that end after | |
1263 | * our range starts. | |
1264 | */ | |
1265 | node = tree_search(tree, start); | |
d397712b | 1266 | if (!node) |
d7fc640e | 1267 | goto out; |
d7fc640e | 1268 | |
d397712b | 1269 | while (1) { |
d7fc640e | 1270 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1271 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1272 | return state; |
d397712b | 1273 | |
d7fc640e CM |
1274 | node = rb_next(node); |
1275 | if (!node) | |
1276 | break; | |
1277 | } | |
1278 | out: | |
1279 | return NULL; | |
1280 | } | |
d7fc640e | 1281 | |
69261c4b XG |
1282 | /* |
1283 | * find the first offset in the io tree with 'bits' set. zero is | |
1284 | * returned if we find something, and *start_ret and *end_ret are | |
1285 | * set to reflect the state struct that was found. | |
1286 | * | |
477d7eaf | 1287 | * If nothing was found, 1 is returned. If found something, return 0. |
69261c4b XG |
1288 | */ |
1289 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
1290 | u64 *start_ret, u64 *end_ret, int bits) | |
1291 | { | |
1292 | struct extent_state *state; | |
1293 | int ret = 1; | |
1294 | ||
1295 | spin_lock(&tree->lock); | |
1296 | state = find_first_extent_bit_state(tree, start, bits); | |
1297 | if (state) { | |
1298 | *start_ret = state->start; | |
1299 | *end_ret = state->end; | |
1300 | ret = 0; | |
1301 | } | |
1302 | spin_unlock(&tree->lock); | |
1303 | return ret; | |
1304 | } | |
1305 | ||
d352ac68 CM |
1306 | /* |
1307 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1308 | * more than 'max_bytes'. start and end are used to return the range, | |
1309 | * | |
1310 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1311 | */ | |
c8b97818 | 1312 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1313 | u64 *start, u64 *end, u64 max_bytes, |
1314 | struct extent_state **cached_state) | |
d1310b2e CM |
1315 | { |
1316 | struct rb_node *node; | |
1317 | struct extent_state *state; | |
1318 | u64 cur_start = *start; | |
1319 | u64 found = 0; | |
1320 | u64 total_bytes = 0; | |
1321 | ||
cad321ad | 1322 | spin_lock(&tree->lock); |
c8b97818 | 1323 | |
d1310b2e CM |
1324 | /* |
1325 | * this search will find all the extents that end after | |
1326 | * our range starts. | |
1327 | */ | |
80ea96b1 | 1328 | node = tree_search(tree, cur_start); |
2b114d1d | 1329 | if (!node) { |
3b951516 CM |
1330 | if (!found) |
1331 | *end = (u64)-1; | |
d1310b2e CM |
1332 | goto out; |
1333 | } | |
1334 | ||
d397712b | 1335 | while (1) { |
d1310b2e | 1336 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1337 | if (found && (state->start != cur_start || |
1338 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1339 | goto out; |
1340 | } | |
1341 | if (!(state->state & EXTENT_DELALLOC)) { | |
1342 | if (!found) | |
1343 | *end = state->end; | |
1344 | goto out; | |
1345 | } | |
c2a128d2 | 1346 | if (!found) { |
d1310b2e | 1347 | *start = state->start; |
c2a128d2 JB |
1348 | *cached_state = state; |
1349 | atomic_inc(&state->refs); | |
1350 | } | |
d1310b2e CM |
1351 | found++; |
1352 | *end = state->end; | |
1353 | cur_start = state->end + 1; | |
1354 | node = rb_next(node); | |
1355 | if (!node) | |
1356 | break; | |
1357 | total_bytes += state->end - state->start + 1; | |
1358 | if (total_bytes >= max_bytes) | |
1359 | break; | |
1360 | } | |
1361 | out: | |
cad321ad | 1362 | spin_unlock(&tree->lock); |
d1310b2e CM |
1363 | return found; |
1364 | } | |
1365 | ||
143bede5 JM |
1366 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1367 | struct page *locked_page, | |
1368 | u64 start, u64 end) | |
c8b97818 CM |
1369 | { |
1370 | int ret; | |
1371 | struct page *pages[16]; | |
1372 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1373 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1374 | unsigned long nr_pages = end_index - index + 1; | |
1375 | int i; | |
1376 | ||
1377 | if (index == locked_page->index && end_index == index) | |
143bede5 | 1378 | return; |
c8b97818 | 1379 | |
d397712b | 1380 | while (nr_pages > 0) { |
c8b97818 | 1381 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1382 | min_t(unsigned long, nr_pages, |
1383 | ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1384 | for (i = 0; i < ret; i++) { |
1385 | if (pages[i] != locked_page) | |
1386 | unlock_page(pages[i]); | |
1387 | page_cache_release(pages[i]); | |
1388 | } | |
1389 | nr_pages -= ret; | |
1390 | index += ret; | |
1391 | cond_resched(); | |
1392 | } | |
c8b97818 CM |
1393 | } |
1394 | ||
1395 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1396 | struct page *locked_page, | |
1397 | u64 delalloc_start, | |
1398 | u64 delalloc_end) | |
1399 | { | |
1400 | unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT; | |
1401 | unsigned long start_index = index; | |
1402 | unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT; | |
1403 | unsigned long pages_locked = 0; | |
1404 | struct page *pages[16]; | |
1405 | unsigned long nrpages; | |
1406 | int ret; | |
1407 | int i; | |
1408 | ||
1409 | /* the caller is responsible for locking the start index */ | |
1410 | if (index == locked_page->index && index == end_index) | |
1411 | return 0; | |
1412 | ||
1413 | /* skip the page at the start index */ | |
1414 | nrpages = end_index - index + 1; | |
d397712b | 1415 | while (nrpages > 0) { |
c8b97818 | 1416 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1417 | min_t(unsigned long, |
1418 | nrpages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1419 | if (ret == 0) { |
1420 | ret = -EAGAIN; | |
1421 | goto done; | |
1422 | } | |
1423 | /* now we have an array of pages, lock them all */ | |
1424 | for (i = 0; i < ret; i++) { | |
1425 | /* | |
1426 | * the caller is taking responsibility for | |
1427 | * locked_page | |
1428 | */ | |
771ed689 | 1429 | if (pages[i] != locked_page) { |
c8b97818 | 1430 | lock_page(pages[i]); |
f2b1c41c CM |
1431 | if (!PageDirty(pages[i]) || |
1432 | pages[i]->mapping != inode->i_mapping) { | |
771ed689 CM |
1433 | ret = -EAGAIN; |
1434 | unlock_page(pages[i]); | |
1435 | page_cache_release(pages[i]); | |
1436 | goto done; | |
1437 | } | |
1438 | } | |
c8b97818 | 1439 | page_cache_release(pages[i]); |
771ed689 | 1440 | pages_locked++; |
c8b97818 | 1441 | } |
c8b97818 CM |
1442 | nrpages -= ret; |
1443 | index += ret; | |
1444 | cond_resched(); | |
1445 | } | |
1446 | ret = 0; | |
1447 | done: | |
1448 | if (ret && pages_locked) { | |
1449 | __unlock_for_delalloc(inode, locked_page, | |
1450 | delalloc_start, | |
1451 | ((u64)(start_index + pages_locked - 1)) << | |
1452 | PAGE_CACHE_SHIFT); | |
1453 | } | |
1454 | return ret; | |
1455 | } | |
1456 | ||
1457 | /* | |
1458 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1459 | * more than 'max_bytes'. start and end are used to return the range, | |
1460 | * | |
1461 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1462 | */ | |
1463 | static noinline u64 find_lock_delalloc_range(struct inode *inode, | |
1464 | struct extent_io_tree *tree, | |
1465 | struct page *locked_page, | |
1466 | u64 *start, u64 *end, | |
1467 | u64 max_bytes) | |
1468 | { | |
1469 | u64 delalloc_start; | |
1470 | u64 delalloc_end; | |
1471 | u64 found; | |
9655d298 | 1472 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1473 | int ret; |
1474 | int loops = 0; | |
1475 | ||
1476 | again: | |
1477 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1478 | delalloc_start = *start; | |
1479 | delalloc_end = 0; | |
1480 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1481 | max_bytes, &cached_state); |
70b99e69 | 1482 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1483 | *start = delalloc_start; |
1484 | *end = delalloc_end; | |
c2a128d2 | 1485 | free_extent_state(cached_state); |
c8b97818 CM |
1486 | return found; |
1487 | } | |
1488 | ||
70b99e69 CM |
1489 | /* |
1490 | * start comes from the offset of locked_page. We have to lock | |
1491 | * pages in order, so we can't process delalloc bytes before | |
1492 | * locked_page | |
1493 | */ | |
d397712b | 1494 | if (delalloc_start < *start) |
70b99e69 | 1495 | delalloc_start = *start; |
70b99e69 | 1496 | |
c8b97818 CM |
1497 | /* |
1498 | * make sure to limit the number of pages we try to lock down | |
1499 | * if we're looping. | |
1500 | */ | |
d397712b | 1501 | if (delalloc_end + 1 - delalloc_start > max_bytes && loops) |
771ed689 | 1502 | delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1; |
d397712b | 1503 | |
c8b97818 CM |
1504 | /* step two, lock all the pages after the page that has start */ |
1505 | ret = lock_delalloc_pages(inode, locked_page, | |
1506 | delalloc_start, delalloc_end); | |
1507 | if (ret == -EAGAIN) { | |
1508 | /* some of the pages are gone, lets avoid looping by | |
1509 | * shortening the size of the delalloc range we're searching | |
1510 | */ | |
9655d298 | 1511 | free_extent_state(cached_state); |
c8b97818 CM |
1512 | if (!loops) { |
1513 | unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1); | |
1514 | max_bytes = PAGE_CACHE_SIZE - offset; | |
1515 | loops = 1; | |
1516 | goto again; | |
1517 | } else { | |
1518 | found = 0; | |
1519 | goto out_failed; | |
1520 | } | |
1521 | } | |
79787eaa | 1522 | BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */ |
c8b97818 CM |
1523 | |
1524 | /* step three, lock the state bits for the whole range */ | |
d0082371 | 1525 | lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state); |
c8b97818 CM |
1526 | |
1527 | /* then test to make sure it is all still delalloc */ | |
1528 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1529 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1530 | if (!ret) { |
9655d298 CM |
1531 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
1532 | &cached_state, GFP_NOFS); | |
c8b97818 CM |
1533 | __unlock_for_delalloc(inode, locked_page, |
1534 | delalloc_start, delalloc_end); | |
1535 | cond_resched(); | |
1536 | goto again; | |
1537 | } | |
9655d298 | 1538 | free_extent_state(cached_state); |
c8b97818 CM |
1539 | *start = delalloc_start; |
1540 | *end = delalloc_end; | |
1541 | out_failed: | |
1542 | return found; | |
1543 | } | |
1544 | ||
1545 | int extent_clear_unlock_delalloc(struct inode *inode, | |
1546 | struct extent_io_tree *tree, | |
1547 | u64 start, u64 end, struct page *locked_page, | |
a791e35e | 1548 | unsigned long op) |
c8b97818 CM |
1549 | { |
1550 | int ret; | |
1551 | struct page *pages[16]; | |
1552 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1553 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1554 | unsigned long nr_pages = end_index - index + 1; | |
1555 | int i; | |
771ed689 | 1556 | int clear_bits = 0; |
c8b97818 | 1557 | |
a791e35e | 1558 | if (op & EXTENT_CLEAR_UNLOCK) |
771ed689 | 1559 | clear_bits |= EXTENT_LOCKED; |
a791e35e | 1560 | if (op & EXTENT_CLEAR_DIRTY) |
c8b97818 CM |
1561 | clear_bits |= EXTENT_DIRTY; |
1562 | ||
a791e35e | 1563 | if (op & EXTENT_CLEAR_DELALLOC) |
771ed689 CM |
1564 | clear_bits |= EXTENT_DELALLOC; |
1565 | ||
2c64c53d | 1566 | clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); |
32c00aff JB |
1567 | if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | |
1568 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK | | |
1569 | EXTENT_SET_PRIVATE2))) | |
771ed689 | 1570 | return 0; |
c8b97818 | 1571 | |
d397712b | 1572 | while (nr_pages > 0) { |
c8b97818 | 1573 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1574 | min_t(unsigned long, |
1575 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 | 1576 | for (i = 0; i < ret; i++) { |
8b62b72b | 1577 | |
a791e35e | 1578 | if (op & EXTENT_SET_PRIVATE2) |
8b62b72b CM |
1579 | SetPagePrivate2(pages[i]); |
1580 | ||
c8b97818 CM |
1581 | if (pages[i] == locked_page) { |
1582 | page_cache_release(pages[i]); | |
1583 | continue; | |
1584 | } | |
a791e35e | 1585 | if (op & EXTENT_CLEAR_DIRTY) |
c8b97818 | 1586 | clear_page_dirty_for_io(pages[i]); |
a791e35e | 1587 | if (op & EXTENT_SET_WRITEBACK) |
c8b97818 | 1588 | set_page_writeback(pages[i]); |
a791e35e | 1589 | if (op & EXTENT_END_WRITEBACK) |
c8b97818 | 1590 | end_page_writeback(pages[i]); |
a791e35e | 1591 | if (op & EXTENT_CLEAR_UNLOCK_PAGE) |
771ed689 | 1592 | unlock_page(pages[i]); |
c8b97818 CM |
1593 | page_cache_release(pages[i]); |
1594 | } | |
1595 | nr_pages -= ret; | |
1596 | index += ret; | |
1597 | cond_resched(); | |
1598 | } | |
1599 | return 0; | |
1600 | } | |
c8b97818 | 1601 | |
d352ac68 CM |
1602 | /* |
1603 | * count the number of bytes in the tree that have a given bit(s) | |
1604 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1605 | * cached. The total number found is returned. | |
1606 | */ | |
d1310b2e CM |
1607 | u64 count_range_bits(struct extent_io_tree *tree, |
1608 | u64 *start, u64 search_end, u64 max_bytes, | |
ec29ed5b | 1609 | unsigned long bits, int contig) |
d1310b2e CM |
1610 | { |
1611 | struct rb_node *node; | |
1612 | struct extent_state *state; | |
1613 | u64 cur_start = *start; | |
1614 | u64 total_bytes = 0; | |
ec29ed5b | 1615 | u64 last = 0; |
d1310b2e CM |
1616 | int found = 0; |
1617 | ||
1618 | if (search_end <= cur_start) { | |
d1310b2e CM |
1619 | WARN_ON(1); |
1620 | return 0; | |
1621 | } | |
1622 | ||
cad321ad | 1623 | spin_lock(&tree->lock); |
d1310b2e CM |
1624 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1625 | total_bytes = tree->dirty_bytes; | |
1626 | goto out; | |
1627 | } | |
1628 | /* | |
1629 | * this search will find all the extents that end after | |
1630 | * our range starts. | |
1631 | */ | |
80ea96b1 | 1632 | node = tree_search(tree, cur_start); |
d397712b | 1633 | if (!node) |
d1310b2e | 1634 | goto out; |
d1310b2e | 1635 | |
d397712b | 1636 | while (1) { |
d1310b2e CM |
1637 | state = rb_entry(node, struct extent_state, rb_node); |
1638 | if (state->start > search_end) | |
1639 | break; | |
ec29ed5b CM |
1640 | if (contig && found && state->start > last + 1) |
1641 | break; | |
1642 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1643 | total_bytes += min(search_end, state->end) + 1 - |
1644 | max(cur_start, state->start); | |
1645 | if (total_bytes >= max_bytes) | |
1646 | break; | |
1647 | if (!found) { | |
af60bed2 | 1648 | *start = max(cur_start, state->start); |
d1310b2e CM |
1649 | found = 1; |
1650 | } | |
ec29ed5b CM |
1651 | last = state->end; |
1652 | } else if (contig && found) { | |
1653 | break; | |
d1310b2e CM |
1654 | } |
1655 | node = rb_next(node); | |
1656 | if (!node) | |
1657 | break; | |
1658 | } | |
1659 | out: | |
cad321ad | 1660 | spin_unlock(&tree->lock); |
d1310b2e CM |
1661 | return total_bytes; |
1662 | } | |
b2950863 | 1663 | |
d352ac68 CM |
1664 | /* |
1665 | * set the private field for a given byte offset in the tree. If there isn't | |
1666 | * an extent_state there already, this does nothing. | |
1667 | */ | |
d1310b2e CM |
1668 | int set_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 | state->private = private; | |
1690 | out: | |
cad321ad | 1691 | spin_unlock(&tree->lock); |
d1310b2e CM |
1692 | return ret; |
1693 | } | |
1694 | ||
1695 | int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private) | |
1696 | { | |
1697 | struct rb_node *node; | |
1698 | struct extent_state *state; | |
1699 | int ret = 0; | |
1700 | ||
cad321ad | 1701 | spin_lock(&tree->lock); |
d1310b2e CM |
1702 | /* |
1703 | * this search will find all the extents that end after | |
1704 | * our range starts. | |
1705 | */ | |
80ea96b1 | 1706 | node = tree_search(tree, start); |
2b114d1d | 1707 | if (!node) { |
d1310b2e CM |
1708 | ret = -ENOENT; |
1709 | goto out; | |
1710 | } | |
1711 | state = rb_entry(node, struct extent_state, rb_node); | |
1712 | if (state->start != start) { | |
1713 | ret = -ENOENT; | |
1714 | goto out; | |
1715 | } | |
1716 | *private = state->private; | |
1717 | out: | |
cad321ad | 1718 | spin_unlock(&tree->lock); |
d1310b2e CM |
1719 | return ret; |
1720 | } | |
1721 | ||
1722 | /* | |
1723 | * searches a range in the state tree for a given mask. | |
70dec807 | 1724 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1725 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1726 | * range is found set. | |
1727 | */ | |
1728 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9655d298 | 1729 | int bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1730 | { |
1731 | struct extent_state *state = NULL; | |
1732 | struct rb_node *node; | |
1733 | int bitset = 0; | |
d1310b2e | 1734 | |
cad321ad | 1735 | spin_lock(&tree->lock); |
df98b6e2 JB |
1736 | if (cached && cached->tree && cached->start <= start && |
1737 | cached->end > start) | |
9655d298 CM |
1738 | node = &cached->rb_node; |
1739 | else | |
1740 | node = tree_search(tree, start); | |
d1310b2e CM |
1741 | while (node && start <= end) { |
1742 | state = rb_entry(node, struct extent_state, rb_node); | |
1743 | ||
1744 | if (filled && state->start > start) { | |
1745 | bitset = 0; | |
1746 | break; | |
1747 | } | |
1748 | ||
1749 | if (state->start > end) | |
1750 | break; | |
1751 | ||
1752 | if (state->state & bits) { | |
1753 | bitset = 1; | |
1754 | if (!filled) | |
1755 | break; | |
1756 | } else if (filled) { | |
1757 | bitset = 0; | |
1758 | break; | |
1759 | } | |
46562cec CM |
1760 | |
1761 | if (state->end == (u64)-1) | |
1762 | break; | |
1763 | ||
d1310b2e CM |
1764 | start = state->end + 1; |
1765 | if (start > end) | |
1766 | break; | |
1767 | node = rb_next(node); | |
1768 | if (!node) { | |
1769 | if (filled) | |
1770 | bitset = 0; | |
1771 | break; | |
1772 | } | |
1773 | } | |
cad321ad | 1774 | spin_unlock(&tree->lock); |
d1310b2e CM |
1775 | return bitset; |
1776 | } | |
d1310b2e CM |
1777 | |
1778 | /* | |
1779 | * helper function to set a given page up to date if all the | |
1780 | * extents in the tree for that page are up to date | |
1781 | */ | |
143bede5 | 1782 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e CM |
1783 | { |
1784 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1785 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
9655d298 | 1786 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 1787 | SetPageUptodate(page); |
d1310b2e CM |
1788 | } |
1789 | ||
1790 | /* | |
1791 | * helper function to unlock a page if all the extents in the tree | |
1792 | * for that page are unlocked | |
1793 | */ | |
143bede5 | 1794 | static void check_page_locked(struct extent_io_tree *tree, struct page *page) |
d1310b2e CM |
1795 | { |
1796 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1797 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
9655d298 | 1798 | if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) |
d1310b2e | 1799 | unlock_page(page); |
d1310b2e CM |
1800 | } |
1801 | ||
1802 | /* | |
1803 | * helper function to end page writeback if all the extents | |
1804 | * in the tree for that page are done with writeback | |
1805 | */ | |
143bede5 JM |
1806 | static void check_page_writeback(struct extent_io_tree *tree, |
1807 | struct page *page) | |
d1310b2e | 1808 | { |
1edbb734 | 1809 | end_page_writeback(page); |
d1310b2e CM |
1810 | } |
1811 | ||
4a54c8c1 JS |
1812 | /* |
1813 | * When IO fails, either with EIO or csum verification fails, we | |
1814 | * try other mirrors that might have a good copy of the data. This | |
1815 | * io_failure_record is used to record state as we go through all the | |
1816 | * mirrors. If another mirror has good data, the page is set up to date | |
1817 | * and things continue. If a good mirror can't be found, the original | |
1818 | * bio end_io callback is called to indicate things have failed. | |
1819 | */ | |
1820 | struct io_failure_record { | |
1821 | struct page *page; | |
1822 | u64 start; | |
1823 | u64 len; | |
1824 | u64 logical; | |
1825 | unsigned long bio_flags; | |
1826 | int this_mirror; | |
1827 | int failed_mirror; | |
1828 | int in_validation; | |
1829 | }; | |
1830 | ||
1831 | static int free_io_failure(struct inode *inode, struct io_failure_record *rec, | |
1832 | int did_repair) | |
1833 | { | |
1834 | int ret; | |
1835 | int err = 0; | |
1836 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
1837 | ||
1838 | set_state_private(failure_tree, rec->start, 0); | |
1839 | ret = clear_extent_bits(failure_tree, rec->start, | |
1840 | rec->start + rec->len - 1, | |
1841 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1842 | if (ret) | |
1843 | err = ret; | |
1844 | ||
1845 | if (did_repair) { | |
1846 | ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start, | |
1847 | rec->start + rec->len - 1, | |
1848 | EXTENT_DAMAGED, GFP_NOFS); | |
1849 | if (ret && !err) | |
1850 | err = ret; | |
1851 | } | |
1852 | ||
1853 | kfree(rec); | |
1854 | return err; | |
1855 | } | |
1856 | ||
1857 | static void repair_io_failure_callback(struct bio *bio, int err) | |
1858 | { | |
1859 | complete(bio->bi_private); | |
1860 | } | |
1861 | ||
1862 | /* | |
1863 | * this bypasses the standard btrfs submit functions deliberately, as | |
1864 | * the standard behavior is to write all copies in a raid setup. here we only | |
1865 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
1866 | * submit_bio directly. | |
1867 | * to avoid any synchonization issues, wait for the data after writing, which | |
1868 | * actually prevents the read that triggered the error from finishing. | |
1869 | * currently, there can be no more than two copies of every data bit. thus, | |
1870 | * exactly one rewrite is required. | |
1871 | */ | |
1872 | int repair_io_failure(struct btrfs_mapping_tree *map_tree, u64 start, | |
1873 | u64 length, u64 logical, struct page *page, | |
1874 | int mirror_num) | |
1875 | { | |
1876 | struct bio *bio; | |
1877 | struct btrfs_device *dev; | |
1878 | DECLARE_COMPLETION_ONSTACK(compl); | |
1879 | u64 map_length = 0; | |
1880 | u64 sector; | |
1881 | struct btrfs_bio *bbio = NULL; | |
1882 | int ret; | |
1883 | ||
1884 | BUG_ON(!mirror_num); | |
1885 | ||
1886 | bio = bio_alloc(GFP_NOFS, 1); | |
1887 | if (!bio) | |
1888 | return -EIO; | |
1889 | bio->bi_private = &compl; | |
1890 | bio->bi_end_io = repair_io_failure_callback; | |
1891 | bio->bi_size = 0; | |
1892 | map_length = length; | |
1893 | ||
1894 | ret = btrfs_map_block(map_tree, WRITE, logical, | |
1895 | &map_length, &bbio, mirror_num); | |
1896 | if (ret) { | |
1897 | bio_put(bio); | |
1898 | return -EIO; | |
1899 | } | |
1900 | BUG_ON(mirror_num != bbio->mirror_num); | |
1901 | sector = bbio->stripes[mirror_num-1].physical >> 9; | |
1902 | bio->bi_sector = sector; | |
1903 | dev = bbio->stripes[mirror_num-1].dev; | |
1904 | kfree(bbio); | |
1905 | if (!dev || !dev->bdev || !dev->writeable) { | |
1906 | bio_put(bio); | |
1907 | return -EIO; | |
1908 | } | |
1909 | bio->bi_bdev = dev->bdev; | |
1910 | bio_add_page(bio, page, length, start-page_offset(page)); | |
21adbd5c | 1911 | btrfsic_submit_bio(WRITE_SYNC, bio); |
4a54c8c1 JS |
1912 | wait_for_completion(&compl); |
1913 | ||
1914 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) { | |
1915 | /* try to remap that extent elsewhere? */ | |
1916 | bio_put(bio); | |
442a4f63 | 1917 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
1918 | return -EIO; |
1919 | } | |
1920 | ||
d5b025d5 | 1921 | printk_ratelimited_in_rcu(KERN_INFO "btrfs read error corrected: ino %lu off %llu " |
606686ee JB |
1922 | "(dev %s sector %llu)\n", page->mapping->host->i_ino, |
1923 | start, rcu_str_deref(dev->name), sector); | |
4a54c8c1 JS |
1924 | |
1925 | bio_put(bio); | |
1926 | return 0; | |
1927 | } | |
1928 | ||
ea466794 JB |
1929 | int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb, |
1930 | int mirror_num) | |
1931 | { | |
1932 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
1933 | u64 start = eb->start; | |
1934 | unsigned long i, num_pages = num_extent_pages(eb->start, eb->len); | |
d95603b2 | 1935 | int ret = 0; |
ea466794 JB |
1936 | |
1937 | for (i = 0; i < num_pages; i++) { | |
1938 | struct page *p = extent_buffer_page(eb, i); | |
1939 | ret = repair_io_failure(map_tree, start, PAGE_CACHE_SIZE, | |
1940 | start, p, mirror_num); | |
1941 | if (ret) | |
1942 | break; | |
1943 | start += PAGE_CACHE_SIZE; | |
1944 | } | |
1945 | ||
1946 | return ret; | |
1947 | } | |
1948 | ||
4a54c8c1 JS |
1949 | /* |
1950 | * each time an IO finishes, we do a fast check in the IO failure tree | |
1951 | * to see if we need to process or clean up an io_failure_record | |
1952 | */ | |
1953 | static int clean_io_failure(u64 start, struct page *page) | |
1954 | { | |
1955 | u64 private; | |
1956 | u64 private_failure; | |
1957 | struct io_failure_record *failrec; | |
1958 | struct btrfs_mapping_tree *map_tree; | |
1959 | struct extent_state *state; | |
1960 | int num_copies; | |
1961 | int did_repair = 0; | |
1962 | int ret; | |
1963 | struct inode *inode = page->mapping->host; | |
1964 | ||
1965 | private = 0; | |
1966 | ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
1967 | (u64)-1, 1, EXTENT_DIRTY, 0); | |
1968 | if (!ret) | |
1969 | return 0; | |
1970 | ||
1971 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start, | |
1972 | &private_failure); | |
1973 | if (ret) | |
1974 | return 0; | |
1975 | ||
1976 | failrec = (struct io_failure_record *)(unsigned long) private_failure; | |
1977 | BUG_ON(!failrec->this_mirror); | |
1978 | ||
1979 | if (failrec->in_validation) { | |
1980 | /* there was no real error, just free the record */ | |
1981 | pr_debug("clean_io_failure: freeing dummy error at %llu\n", | |
1982 | failrec->start); | |
1983 | did_repair = 1; | |
1984 | goto out; | |
1985 | } | |
1986 | ||
1987 | spin_lock(&BTRFS_I(inode)->io_tree.lock); | |
1988 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, | |
1989 | failrec->start, | |
1990 | EXTENT_LOCKED); | |
1991 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); | |
1992 | ||
1993 | if (state && state->start == failrec->start) { | |
1994 | map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree; | |
1995 | num_copies = btrfs_num_copies(map_tree, failrec->logical, | |
1996 | failrec->len); | |
1997 | if (num_copies > 1) { | |
1998 | ret = repair_io_failure(map_tree, start, failrec->len, | |
1999 | failrec->logical, page, | |
2000 | failrec->failed_mirror); | |
2001 | did_repair = !ret; | |
2002 | } | |
2003 | } | |
2004 | ||
2005 | out: | |
2006 | if (!ret) | |
2007 | ret = free_io_failure(inode, failrec, did_repair); | |
2008 | ||
2009 | return ret; | |
2010 | } | |
2011 | ||
2012 | /* | |
2013 | * this is a generic handler for readpage errors (default | |
2014 | * readpage_io_failed_hook). if other copies exist, read those and write back | |
2015 | * good data to the failed position. does not investigate in remapping the | |
2016 | * failed extent elsewhere, hoping the device will be smart enough to do this as | |
2017 | * needed | |
2018 | */ | |
2019 | ||
2020 | static int bio_readpage_error(struct bio *failed_bio, struct page *page, | |
2021 | u64 start, u64 end, int failed_mirror, | |
2022 | struct extent_state *state) | |
2023 | { | |
2024 | struct io_failure_record *failrec = NULL; | |
2025 | u64 private; | |
2026 | struct extent_map *em; | |
2027 | struct inode *inode = page->mapping->host; | |
2028 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
2029 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2030 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
2031 | struct bio *bio; | |
2032 | int num_copies; | |
2033 | int ret; | |
2034 | int read_mode; | |
2035 | u64 logical; | |
2036 | ||
2037 | BUG_ON(failed_bio->bi_rw & REQ_WRITE); | |
2038 | ||
2039 | ret = get_state_private(failure_tree, start, &private); | |
2040 | if (ret) { | |
2041 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2042 | if (!failrec) | |
2043 | return -ENOMEM; | |
2044 | failrec->start = start; | |
2045 | failrec->len = end - start + 1; | |
2046 | failrec->this_mirror = 0; | |
2047 | failrec->bio_flags = 0; | |
2048 | failrec->in_validation = 0; | |
2049 | ||
2050 | read_lock(&em_tree->lock); | |
2051 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2052 | if (!em) { | |
2053 | read_unlock(&em_tree->lock); | |
2054 | kfree(failrec); | |
2055 | return -EIO; | |
2056 | } | |
2057 | ||
2058 | if (em->start > start || em->start + em->len < start) { | |
2059 | free_extent_map(em); | |
2060 | em = NULL; | |
2061 | } | |
2062 | read_unlock(&em_tree->lock); | |
2063 | ||
2064 | if (!em || IS_ERR(em)) { | |
2065 | kfree(failrec); | |
2066 | return -EIO; | |
2067 | } | |
2068 | logical = start - em->start; | |
2069 | logical = em->block_start + logical; | |
2070 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2071 | logical = em->block_start; | |
2072 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2073 | extent_set_compress_type(&failrec->bio_flags, | |
2074 | em->compress_type); | |
2075 | } | |
2076 | pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, " | |
2077 | "len=%llu\n", logical, start, failrec->len); | |
2078 | failrec->logical = logical; | |
2079 | free_extent_map(em); | |
2080 | ||
2081 | /* set the bits in the private failure tree */ | |
2082 | ret = set_extent_bits(failure_tree, start, end, | |
2083 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
2084 | if (ret >= 0) | |
2085 | ret = set_state_private(failure_tree, start, | |
2086 | (u64)(unsigned long)failrec); | |
2087 | /* set the bits in the inode's tree */ | |
2088 | if (ret >= 0) | |
2089 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED, | |
2090 | GFP_NOFS); | |
2091 | if (ret < 0) { | |
2092 | kfree(failrec); | |
2093 | return ret; | |
2094 | } | |
2095 | } else { | |
2096 | failrec = (struct io_failure_record *)(unsigned long)private; | |
2097 | pr_debug("bio_readpage_error: (found) logical=%llu, " | |
2098 | "start=%llu, len=%llu, validation=%d\n", | |
2099 | failrec->logical, failrec->start, failrec->len, | |
2100 | failrec->in_validation); | |
2101 | /* | |
2102 | * when data can be on disk more than twice, add to failrec here | |
2103 | * (e.g. with a list for failed_mirror) to make | |
2104 | * clean_io_failure() clean all those errors at once. | |
2105 | */ | |
2106 | } | |
2107 | num_copies = btrfs_num_copies( | |
2108 | &BTRFS_I(inode)->root->fs_info->mapping_tree, | |
2109 | failrec->logical, failrec->len); | |
2110 | if (num_copies == 1) { | |
2111 | /* | |
2112 | * we only have a single copy of the data, so don't bother with | |
2113 | * all the retry and error correction code that follows. no | |
2114 | * matter what the error is, it is very likely to persist. | |
2115 | */ | |
2116 | pr_debug("bio_readpage_error: cannot repair, num_copies == 1. " | |
2117 | "state=%p, num_copies=%d, next_mirror %d, " | |
2118 | "failed_mirror %d\n", state, num_copies, | |
2119 | failrec->this_mirror, failed_mirror); | |
2120 | free_io_failure(inode, failrec, 0); | |
2121 | return -EIO; | |
2122 | } | |
2123 | ||
2124 | if (!state) { | |
2125 | spin_lock(&tree->lock); | |
2126 | state = find_first_extent_bit_state(tree, failrec->start, | |
2127 | EXTENT_LOCKED); | |
2128 | if (state && state->start != failrec->start) | |
2129 | state = NULL; | |
2130 | spin_unlock(&tree->lock); | |
2131 | } | |
2132 | ||
2133 | /* | |
2134 | * there are two premises: | |
2135 | * a) deliver good data to the caller | |
2136 | * b) correct the bad sectors on disk | |
2137 | */ | |
2138 | if (failed_bio->bi_vcnt > 1) { | |
2139 | /* | |
2140 | * to fulfill b), we need to know the exact failing sectors, as | |
2141 | * we don't want to rewrite any more than the failed ones. thus, | |
2142 | * we need separate read requests for the failed bio | |
2143 | * | |
2144 | * if the following BUG_ON triggers, our validation request got | |
2145 | * merged. we need separate requests for our algorithm to work. | |
2146 | */ | |
2147 | BUG_ON(failrec->in_validation); | |
2148 | failrec->in_validation = 1; | |
2149 | failrec->this_mirror = failed_mirror; | |
2150 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; | |
2151 | } else { | |
2152 | /* | |
2153 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2154 | * of the failed sector and if we succeed, we have setup | |
2155 | * everything for repair_io_failure to do the rest for us. | |
2156 | */ | |
2157 | if (failrec->in_validation) { | |
2158 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2159 | failrec->in_validation = 0; | |
2160 | failrec->this_mirror = 0; | |
2161 | } | |
2162 | failrec->failed_mirror = failed_mirror; | |
2163 | failrec->this_mirror++; | |
2164 | if (failrec->this_mirror == failed_mirror) | |
2165 | failrec->this_mirror++; | |
2166 | read_mode = READ_SYNC; | |
2167 | } | |
2168 | ||
2169 | if (!state || failrec->this_mirror > num_copies) { | |
2170 | pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, " | |
2171 | "next_mirror %d, failed_mirror %d\n", state, | |
2172 | num_copies, failrec->this_mirror, failed_mirror); | |
2173 | free_io_failure(inode, failrec, 0); | |
2174 | return -EIO; | |
2175 | } | |
2176 | ||
2177 | bio = bio_alloc(GFP_NOFS, 1); | |
e627ee7b TI |
2178 | if (!bio) { |
2179 | free_io_failure(inode, failrec, 0); | |
2180 | return -EIO; | |
2181 | } | |
4a54c8c1 JS |
2182 | bio->bi_private = state; |
2183 | bio->bi_end_io = failed_bio->bi_end_io; | |
2184 | bio->bi_sector = failrec->logical >> 9; | |
2185 | bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; | |
2186 | bio->bi_size = 0; | |
2187 | ||
2188 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); | |
2189 | ||
2190 | pr_debug("bio_readpage_error: submitting new read[%#x] to " | |
2191 | "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode, | |
2192 | failrec->this_mirror, num_copies, failrec->in_validation); | |
2193 | ||
013bd4c3 TI |
2194 | ret = tree->ops->submit_bio_hook(inode, read_mode, bio, |
2195 | failrec->this_mirror, | |
2196 | failrec->bio_flags, 0); | |
2197 | return ret; | |
4a54c8c1 JS |
2198 | } |
2199 | ||
d1310b2e CM |
2200 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2201 | ||
87826df0 JM |
2202 | int end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
2203 | { | |
2204 | int uptodate = (err == 0); | |
2205 | struct extent_io_tree *tree; | |
2206 | int ret; | |
2207 | ||
2208 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
2209 | ||
2210 | if (tree->ops && tree->ops->writepage_end_io_hook) { | |
2211 | ret = tree->ops->writepage_end_io_hook(page, start, | |
2212 | end, NULL, uptodate); | |
2213 | if (ret) | |
2214 | uptodate = 0; | |
2215 | } | |
2216 | ||
87826df0 | 2217 | if (!uptodate) { |
87826df0 JM |
2218 | ClearPageUptodate(page); |
2219 | SetPageError(page); | |
2220 | } | |
2221 | return 0; | |
2222 | } | |
2223 | ||
d1310b2e CM |
2224 | /* |
2225 | * after a writepage IO is done, we need to: | |
2226 | * clear the uptodate bits on error | |
2227 | * clear the writeback bits in the extent tree for this IO | |
2228 | * end_page_writeback if the page has no more pending IO | |
2229 | * | |
2230 | * Scheduling is not allowed, so the extent state tree is expected | |
2231 | * to have one and only one object corresponding to this IO. | |
2232 | */ | |
d1310b2e | 2233 | static void end_bio_extent_writepage(struct bio *bio, int err) |
d1310b2e | 2234 | { |
d1310b2e | 2235 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
902b22f3 | 2236 | struct extent_io_tree *tree; |
d1310b2e CM |
2237 | u64 start; |
2238 | u64 end; | |
2239 | int whole_page; | |
2240 | ||
d1310b2e CM |
2241 | do { |
2242 | struct page *page = bvec->bv_page; | |
902b22f3 DW |
2243 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
2244 | ||
d1310b2e CM |
2245 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
2246 | bvec->bv_offset; | |
2247 | end = start + bvec->bv_len - 1; | |
2248 | ||
2249 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
2250 | whole_page = 1; | |
2251 | else | |
2252 | whole_page = 0; | |
2253 | ||
2254 | if (--bvec >= bio->bi_io_vec) | |
2255 | prefetchw(&bvec->bv_page->flags); | |
1259ab75 | 2256 | |
87826df0 JM |
2257 | if (end_extent_writepage(page, err, start, end)) |
2258 | continue; | |
70dec807 | 2259 | |
d1310b2e CM |
2260 | if (whole_page) |
2261 | end_page_writeback(page); | |
2262 | else | |
2263 | check_page_writeback(tree, page); | |
d1310b2e | 2264 | } while (bvec >= bio->bi_io_vec); |
2b1f55b0 | 2265 | |
d1310b2e | 2266 | bio_put(bio); |
d1310b2e CM |
2267 | } |
2268 | ||
2269 | /* | |
2270 | * after a readpage IO is done, we need to: | |
2271 | * clear the uptodate bits on error | |
2272 | * set the uptodate bits if things worked | |
2273 | * set the page up to date if all extents in the tree are uptodate | |
2274 | * clear the lock bit in the extent tree | |
2275 | * unlock the page if there are no other extents locked for it | |
2276 | * | |
2277 | * Scheduling is not allowed, so the extent state tree is expected | |
2278 | * to have one and only one object corresponding to this IO. | |
2279 | */ | |
d1310b2e | 2280 | static void end_bio_extent_readpage(struct bio *bio, int err) |
d1310b2e CM |
2281 | { |
2282 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
4125bf76 CM |
2283 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
2284 | struct bio_vec *bvec = bio->bi_io_vec; | |
902b22f3 | 2285 | struct extent_io_tree *tree; |
d1310b2e CM |
2286 | u64 start; |
2287 | u64 end; | |
2288 | int whole_page; | |
5cf1ab56 | 2289 | int mirror; |
d1310b2e CM |
2290 | int ret; |
2291 | ||
d20f7043 CM |
2292 | if (err) |
2293 | uptodate = 0; | |
2294 | ||
d1310b2e CM |
2295 | do { |
2296 | struct page *page = bvec->bv_page; | |
507903b8 AJ |
2297 | struct extent_state *cached = NULL; |
2298 | struct extent_state *state; | |
2299 | ||
4a54c8c1 JS |
2300 | pr_debug("end_bio_extent_readpage: bi_vcnt=%d, idx=%d, err=%d, " |
2301 | "mirror=%ld\n", bio->bi_vcnt, bio->bi_idx, err, | |
2302 | (long int)bio->bi_bdev); | |
902b22f3 DW |
2303 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
2304 | ||
d1310b2e CM |
2305 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
2306 | bvec->bv_offset; | |
2307 | end = start + bvec->bv_len - 1; | |
2308 | ||
2309 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
2310 | whole_page = 1; | |
2311 | else | |
2312 | whole_page = 0; | |
2313 | ||
4125bf76 | 2314 | if (++bvec <= bvec_end) |
d1310b2e CM |
2315 | prefetchw(&bvec->bv_page->flags); |
2316 | ||
507903b8 | 2317 | spin_lock(&tree->lock); |
0d399205 | 2318 | state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED); |
109b36a2 | 2319 | if (state && state->start == start) { |
507903b8 AJ |
2320 | /* |
2321 | * take a reference on the state, unlock will drop | |
2322 | * the ref | |
2323 | */ | |
2324 | cache_state(state, &cached); | |
2325 | } | |
2326 | spin_unlock(&tree->lock); | |
2327 | ||
5cf1ab56 | 2328 | mirror = (int)(unsigned long)bio->bi_bdev; |
d1310b2e | 2329 | if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) { |
70dec807 | 2330 | ret = tree->ops->readpage_end_io_hook(page, start, end, |
5cf1ab56 | 2331 | state, mirror); |
442a4f63 SB |
2332 | if (ret) { |
2333 | /* no IO indicated but software detected errors | |
2334 | * in the block, either checksum errors or | |
2335 | * issues with the contents */ | |
2336 | struct btrfs_root *root = | |
2337 | BTRFS_I(page->mapping->host)->root; | |
2338 | struct btrfs_device *device; | |
2339 | ||
d1310b2e | 2340 | uptodate = 0; |
442a4f63 SB |
2341 | device = btrfs_find_device_for_logical( |
2342 | root, start, mirror); | |
2343 | if (device) | |
2344 | btrfs_dev_stat_inc_and_print(device, | |
2345 | BTRFS_DEV_STAT_CORRUPTION_ERRS); | |
2346 | } else { | |
4a54c8c1 | 2347 | clean_io_failure(start, page); |
442a4f63 | 2348 | } |
d1310b2e | 2349 | } |
ea466794 | 2350 | |
ea466794 | 2351 | if (!uptodate && tree->ops && tree->ops->readpage_io_failed_hook) { |
5cf1ab56 | 2352 | ret = tree->ops->readpage_io_failed_hook(page, mirror); |
ea466794 JB |
2353 | if (!ret && !err && |
2354 | test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
2355 | uptodate = 1; | |
2356 | } else if (!uptodate) { | |
f4a8e656 JS |
2357 | /* |
2358 | * The generic bio_readpage_error handles errors the | |
2359 | * following way: If possible, new read requests are | |
2360 | * created and submitted and will end up in | |
2361 | * end_bio_extent_readpage as well (if we're lucky, not | |
2362 | * in the !uptodate case). In that case it returns 0 and | |
2363 | * we just go on with the next page in our bio. If it | |
2364 | * can't handle the error it will return -EIO and we | |
2365 | * remain responsible for that page. | |
2366 | */ | |
5cf1ab56 | 2367 | ret = bio_readpage_error(bio, page, start, end, mirror, NULL); |
7e38326f | 2368 | if (ret == 0) { |
3b951516 CM |
2369 | uptodate = |
2370 | test_bit(BIO_UPTODATE, &bio->bi_flags); | |
d20f7043 CM |
2371 | if (err) |
2372 | uptodate = 0; | |
507903b8 | 2373 | uncache_state(&cached); |
7e38326f CM |
2374 | continue; |
2375 | } | |
2376 | } | |
d1310b2e | 2377 | |
0b32f4bb | 2378 | if (uptodate && tree->track_uptodate) { |
507903b8 | 2379 | set_extent_uptodate(tree, start, end, &cached, |
902b22f3 | 2380 | GFP_ATOMIC); |
771ed689 | 2381 | } |
507903b8 | 2382 | unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC); |
d1310b2e | 2383 | |
70dec807 CM |
2384 | if (whole_page) { |
2385 | if (uptodate) { | |
2386 | SetPageUptodate(page); | |
2387 | } else { | |
2388 | ClearPageUptodate(page); | |
2389 | SetPageError(page); | |
2390 | } | |
d1310b2e | 2391 | unlock_page(page); |
70dec807 CM |
2392 | } else { |
2393 | if (uptodate) { | |
2394 | check_page_uptodate(tree, page); | |
2395 | } else { | |
2396 | ClearPageUptodate(page); | |
2397 | SetPageError(page); | |
2398 | } | |
d1310b2e | 2399 | check_page_locked(tree, page); |
70dec807 | 2400 | } |
4125bf76 | 2401 | } while (bvec <= bvec_end); |
d1310b2e CM |
2402 | |
2403 | bio_put(bio); | |
d1310b2e CM |
2404 | } |
2405 | ||
88f794ed MX |
2406 | struct bio * |
2407 | btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
2408 | gfp_t gfp_flags) | |
d1310b2e CM |
2409 | { |
2410 | struct bio *bio; | |
2411 | ||
2412 | bio = bio_alloc(gfp_flags, nr_vecs); | |
2413 | ||
2414 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
2415 | while (!bio && (nr_vecs /= 2)) | |
2416 | bio = bio_alloc(gfp_flags, nr_vecs); | |
2417 | } | |
2418 | ||
2419 | if (bio) { | |
e1c4b745 | 2420 | bio->bi_size = 0; |
d1310b2e CM |
2421 | bio->bi_bdev = bdev; |
2422 | bio->bi_sector = first_sector; | |
2423 | } | |
2424 | return bio; | |
2425 | } | |
2426 | ||
79787eaa JM |
2427 | /* |
2428 | * Since writes are async, they will only return -ENOMEM. | |
2429 | * Reads can return the full range of I/O error conditions. | |
2430 | */ | |
355808c2 JM |
2431 | static int __must_check submit_one_bio(int rw, struct bio *bio, |
2432 | int mirror_num, unsigned long bio_flags) | |
d1310b2e | 2433 | { |
d1310b2e | 2434 | int ret = 0; |
70dec807 CM |
2435 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
2436 | struct page *page = bvec->bv_page; | |
2437 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 2438 | u64 start; |
70dec807 CM |
2439 | |
2440 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset; | |
70dec807 | 2441 | |
902b22f3 | 2442 | bio->bi_private = NULL; |
d1310b2e CM |
2443 | |
2444 | bio_get(bio); | |
2445 | ||
065631f6 | 2446 | if (tree->ops && tree->ops->submit_bio_hook) |
6b82ce8d | 2447 | ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio, |
eaf25d93 | 2448 | mirror_num, bio_flags, start); |
0b86a832 | 2449 | else |
21adbd5c | 2450 | btrfsic_submit_bio(rw, bio); |
4a54c8c1 | 2451 | |
d1310b2e CM |
2452 | if (bio_flagged(bio, BIO_EOPNOTSUPP)) |
2453 | ret = -EOPNOTSUPP; | |
2454 | bio_put(bio); | |
2455 | return ret; | |
2456 | } | |
2457 | ||
3444a972 JM |
2458 | static int merge_bio(struct extent_io_tree *tree, struct page *page, |
2459 | unsigned long offset, size_t size, struct bio *bio, | |
2460 | unsigned long bio_flags) | |
2461 | { | |
2462 | int ret = 0; | |
2463 | if (tree->ops && tree->ops->merge_bio_hook) | |
2464 | ret = tree->ops->merge_bio_hook(page, offset, size, bio, | |
2465 | bio_flags); | |
2466 | BUG_ON(ret < 0); | |
2467 | return ret; | |
2468 | ||
2469 | } | |
2470 | ||
d1310b2e CM |
2471 | static int submit_extent_page(int rw, struct extent_io_tree *tree, |
2472 | struct page *page, sector_t sector, | |
2473 | size_t size, unsigned long offset, | |
2474 | struct block_device *bdev, | |
2475 | struct bio **bio_ret, | |
2476 | unsigned long max_pages, | |
f188591e | 2477 | bio_end_io_t end_io_func, |
c8b97818 CM |
2478 | int mirror_num, |
2479 | unsigned long prev_bio_flags, | |
2480 | unsigned long bio_flags) | |
d1310b2e CM |
2481 | { |
2482 | int ret = 0; | |
2483 | struct bio *bio; | |
2484 | int nr; | |
c8b97818 CM |
2485 | int contig = 0; |
2486 | int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED; | |
2487 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; | |
5b050f04 | 2488 | size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE); |
d1310b2e CM |
2489 | |
2490 | if (bio_ret && *bio_ret) { | |
2491 | bio = *bio_ret; | |
c8b97818 CM |
2492 | if (old_compressed) |
2493 | contig = bio->bi_sector == sector; | |
2494 | else | |
2495 | contig = bio->bi_sector + (bio->bi_size >> 9) == | |
2496 | sector; | |
2497 | ||
2498 | if (prev_bio_flags != bio_flags || !contig || | |
3444a972 | 2499 | merge_bio(tree, page, offset, page_size, bio, bio_flags) || |
c8b97818 CM |
2500 | bio_add_page(bio, page, page_size, offset) < page_size) { |
2501 | ret = submit_one_bio(rw, bio, mirror_num, | |
2502 | prev_bio_flags); | |
79787eaa JM |
2503 | if (ret < 0) |
2504 | return ret; | |
d1310b2e CM |
2505 | bio = NULL; |
2506 | } else { | |
2507 | return 0; | |
2508 | } | |
2509 | } | |
c8b97818 CM |
2510 | if (this_compressed) |
2511 | nr = BIO_MAX_PAGES; | |
2512 | else | |
2513 | nr = bio_get_nr_vecs(bdev); | |
2514 | ||
88f794ed | 2515 | bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH); |
5df67083 TI |
2516 | if (!bio) |
2517 | return -ENOMEM; | |
70dec807 | 2518 | |
c8b97818 | 2519 | bio_add_page(bio, page, page_size, offset); |
d1310b2e CM |
2520 | bio->bi_end_io = end_io_func; |
2521 | bio->bi_private = tree; | |
70dec807 | 2522 | |
d397712b | 2523 | if (bio_ret) |
d1310b2e | 2524 | *bio_ret = bio; |
d397712b | 2525 | else |
c8b97818 | 2526 | ret = submit_one_bio(rw, bio, mirror_num, bio_flags); |
d1310b2e CM |
2527 | |
2528 | return ret; | |
2529 | } | |
2530 | ||
4f2de97a | 2531 | void attach_extent_buffer_page(struct extent_buffer *eb, struct page *page) |
d1310b2e CM |
2532 | { |
2533 | if (!PagePrivate(page)) { | |
2534 | SetPagePrivate(page); | |
d1310b2e | 2535 | page_cache_get(page); |
4f2de97a JB |
2536 | set_page_private(page, (unsigned long)eb); |
2537 | } else { | |
2538 | WARN_ON(page->private != (unsigned long)eb); | |
d1310b2e CM |
2539 | } |
2540 | } | |
2541 | ||
4f2de97a | 2542 | void set_page_extent_mapped(struct page *page) |
d1310b2e | 2543 | { |
4f2de97a JB |
2544 | if (!PagePrivate(page)) { |
2545 | SetPagePrivate(page); | |
2546 | page_cache_get(page); | |
2547 | set_page_private(page, EXTENT_PAGE_PRIVATE); | |
2548 | } | |
d1310b2e CM |
2549 | } |
2550 | ||
2551 | /* | |
2552 | * basic readpage implementation. Locked extent state structs are inserted | |
2553 | * into the tree that are removed when the IO is done (by the end_io | |
2554 | * handlers) | |
79787eaa | 2555 | * XXX JDM: This needs looking at to ensure proper page locking |
d1310b2e CM |
2556 | */ |
2557 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
2558 | struct page *page, | |
2559 | get_extent_t *get_extent, | |
c8b97818 CM |
2560 | struct bio **bio, int mirror_num, |
2561 | unsigned long *bio_flags) | |
d1310b2e CM |
2562 | { |
2563 | struct inode *inode = page->mapping->host; | |
2564 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2565 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2566 | u64 end; | |
2567 | u64 cur = start; | |
2568 | u64 extent_offset; | |
2569 | u64 last_byte = i_size_read(inode); | |
2570 | u64 block_start; | |
2571 | u64 cur_end; | |
2572 | sector_t sector; | |
2573 | struct extent_map *em; | |
2574 | struct block_device *bdev; | |
11c65dcc | 2575 | struct btrfs_ordered_extent *ordered; |
d1310b2e CM |
2576 | int ret; |
2577 | int nr = 0; | |
306e16ce | 2578 | size_t pg_offset = 0; |
d1310b2e | 2579 | size_t iosize; |
c8b97818 | 2580 | size_t disk_io_size; |
d1310b2e | 2581 | size_t blocksize = inode->i_sb->s_blocksize; |
c8b97818 | 2582 | unsigned long this_bio_flag = 0; |
d1310b2e CM |
2583 | |
2584 | set_page_extent_mapped(page); | |
2585 | ||
90a887c9 DM |
2586 | if (!PageUptodate(page)) { |
2587 | if (cleancache_get_page(page) == 0) { | |
2588 | BUG_ON(blocksize != PAGE_SIZE); | |
2589 | goto out; | |
2590 | } | |
2591 | } | |
2592 | ||
d1310b2e | 2593 | end = page_end; |
11c65dcc | 2594 | while (1) { |
d0082371 | 2595 | lock_extent(tree, start, end); |
11c65dcc JB |
2596 | ordered = btrfs_lookup_ordered_extent(inode, start); |
2597 | if (!ordered) | |
2598 | break; | |
d0082371 | 2599 | unlock_extent(tree, start, end); |
11c65dcc JB |
2600 | btrfs_start_ordered_extent(inode, ordered, 1); |
2601 | btrfs_put_ordered_extent(ordered); | |
2602 | } | |
d1310b2e | 2603 | |
c8b97818 CM |
2604 | if (page->index == last_byte >> PAGE_CACHE_SHIFT) { |
2605 | char *userpage; | |
2606 | size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1); | |
2607 | ||
2608 | if (zero_offset) { | |
2609 | iosize = PAGE_CACHE_SIZE - zero_offset; | |
7ac687d9 | 2610 | userpage = kmap_atomic(page); |
c8b97818 CM |
2611 | memset(userpage + zero_offset, 0, iosize); |
2612 | flush_dcache_page(page); | |
7ac687d9 | 2613 | kunmap_atomic(userpage); |
c8b97818 CM |
2614 | } |
2615 | } | |
d1310b2e CM |
2616 | while (cur <= end) { |
2617 | if (cur >= last_byte) { | |
2618 | char *userpage; | |
507903b8 AJ |
2619 | struct extent_state *cached = NULL; |
2620 | ||
306e16ce | 2621 | iosize = PAGE_CACHE_SIZE - pg_offset; |
7ac687d9 | 2622 | userpage = kmap_atomic(page); |
306e16ce | 2623 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2624 | flush_dcache_page(page); |
7ac687d9 | 2625 | kunmap_atomic(userpage); |
d1310b2e | 2626 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 AJ |
2627 | &cached, GFP_NOFS); |
2628 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
2629 | &cached, GFP_NOFS); | |
d1310b2e CM |
2630 | break; |
2631 | } | |
306e16ce | 2632 | em = get_extent(inode, page, pg_offset, cur, |
d1310b2e | 2633 | end - cur + 1, 0); |
c704005d | 2634 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 2635 | SetPageError(page); |
d0082371 | 2636 | unlock_extent(tree, cur, end); |
d1310b2e CM |
2637 | break; |
2638 | } | |
d1310b2e CM |
2639 | extent_offset = cur - em->start; |
2640 | BUG_ON(extent_map_end(em) <= cur); | |
2641 | BUG_ON(end < cur); | |
2642 | ||
261507a0 | 2643 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
c8b97818 | 2644 | this_bio_flag = EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
2645 | extent_set_compress_type(&this_bio_flag, |
2646 | em->compress_type); | |
2647 | } | |
c8b97818 | 2648 | |
d1310b2e CM |
2649 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
2650 | cur_end = min(extent_map_end(em) - 1, end); | |
2651 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
c8b97818 CM |
2652 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
2653 | disk_io_size = em->block_len; | |
2654 | sector = em->block_start >> 9; | |
2655 | } else { | |
2656 | sector = (em->block_start + extent_offset) >> 9; | |
2657 | disk_io_size = iosize; | |
2658 | } | |
d1310b2e CM |
2659 | bdev = em->bdev; |
2660 | block_start = em->block_start; | |
d899e052 YZ |
2661 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2662 | block_start = EXTENT_MAP_HOLE; | |
d1310b2e CM |
2663 | free_extent_map(em); |
2664 | em = NULL; | |
2665 | ||
2666 | /* we've found a hole, just zero and go on */ | |
2667 | if (block_start == EXTENT_MAP_HOLE) { | |
2668 | char *userpage; | |
507903b8 AJ |
2669 | struct extent_state *cached = NULL; |
2670 | ||
7ac687d9 | 2671 | userpage = kmap_atomic(page); |
306e16ce | 2672 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2673 | flush_dcache_page(page); |
7ac687d9 | 2674 | kunmap_atomic(userpage); |
d1310b2e CM |
2675 | |
2676 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 AJ |
2677 | &cached, GFP_NOFS); |
2678 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
2679 | &cached, GFP_NOFS); | |
d1310b2e | 2680 | cur = cur + iosize; |
306e16ce | 2681 | pg_offset += iosize; |
d1310b2e CM |
2682 | continue; |
2683 | } | |
2684 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
2685 | if (test_range_bit(tree, cur, cur_end, |
2686 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 2687 | check_page_uptodate(tree, page); |
d0082371 | 2688 | unlock_extent(tree, cur, cur + iosize - 1); |
d1310b2e | 2689 | cur = cur + iosize; |
306e16ce | 2690 | pg_offset += iosize; |
d1310b2e CM |
2691 | continue; |
2692 | } | |
70dec807 CM |
2693 | /* we have an inline extent but it didn't get marked up |
2694 | * to date. Error out | |
2695 | */ | |
2696 | if (block_start == EXTENT_MAP_INLINE) { | |
2697 | SetPageError(page); | |
d0082371 | 2698 | unlock_extent(tree, cur, cur + iosize - 1); |
70dec807 | 2699 | cur = cur + iosize; |
306e16ce | 2700 | pg_offset += iosize; |
70dec807 CM |
2701 | continue; |
2702 | } | |
d1310b2e CM |
2703 | |
2704 | ret = 0; | |
2705 | if (tree->ops && tree->ops->readpage_io_hook) { | |
2706 | ret = tree->ops->readpage_io_hook(page, cur, | |
2707 | cur + iosize - 1); | |
2708 | } | |
2709 | if (!ret) { | |
89642229 CM |
2710 | unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1; |
2711 | pnr -= page->index; | |
d1310b2e | 2712 | ret = submit_extent_page(READ, tree, page, |
306e16ce | 2713 | sector, disk_io_size, pg_offset, |
89642229 | 2714 | bdev, bio, pnr, |
c8b97818 CM |
2715 | end_bio_extent_readpage, mirror_num, |
2716 | *bio_flags, | |
2717 | this_bio_flag); | |
79787eaa | 2718 | BUG_ON(ret == -ENOMEM); |
89642229 | 2719 | nr++; |
c8b97818 | 2720 | *bio_flags = this_bio_flag; |
d1310b2e CM |
2721 | } |
2722 | if (ret) | |
2723 | SetPageError(page); | |
2724 | cur = cur + iosize; | |
306e16ce | 2725 | pg_offset += iosize; |
d1310b2e | 2726 | } |
90a887c9 | 2727 | out: |
d1310b2e CM |
2728 | if (!nr) { |
2729 | if (!PageError(page)) | |
2730 | SetPageUptodate(page); | |
2731 | unlock_page(page); | |
2732 | } | |
2733 | return 0; | |
2734 | } | |
2735 | ||
2736 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, | |
8ddc7d9c | 2737 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
2738 | { |
2739 | struct bio *bio = NULL; | |
c8b97818 | 2740 | unsigned long bio_flags = 0; |
d1310b2e CM |
2741 | int ret; |
2742 | ||
8ddc7d9c | 2743 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, |
c8b97818 | 2744 | &bio_flags); |
d1310b2e | 2745 | if (bio) |
8ddc7d9c | 2746 | ret = submit_one_bio(READ, bio, mirror_num, bio_flags); |
d1310b2e CM |
2747 | return ret; |
2748 | } | |
d1310b2e | 2749 | |
11c8349b CM |
2750 | static noinline void update_nr_written(struct page *page, |
2751 | struct writeback_control *wbc, | |
2752 | unsigned long nr_written) | |
2753 | { | |
2754 | wbc->nr_to_write -= nr_written; | |
2755 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && | |
2756 | wbc->range_start == 0 && wbc->range_end == LLONG_MAX)) | |
2757 | page->mapping->writeback_index = page->index + nr_written; | |
2758 | } | |
2759 | ||
d1310b2e CM |
2760 | /* |
2761 | * the writepage semantics are similar to regular writepage. extent | |
2762 | * records are inserted to lock ranges in the tree, and as dirty areas | |
2763 | * are found, they are marked writeback. Then the lock bits are removed | |
2764 | * and the end_io handler clears the writeback ranges | |
2765 | */ | |
2766 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
2767 | void *data) | |
2768 | { | |
2769 | struct inode *inode = page->mapping->host; | |
2770 | struct extent_page_data *epd = data; | |
2771 | struct extent_io_tree *tree = epd->tree; | |
2772 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2773 | u64 delalloc_start; | |
2774 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2775 | u64 end; | |
2776 | u64 cur = start; | |
2777 | u64 extent_offset; | |
2778 | u64 last_byte = i_size_read(inode); | |
2779 | u64 block_start; | |
2780 | u64 iosize; | |
2781 | sector_t sector; | |
2c64c53d | 2782 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
2783 | struct extent_map *em; |
2784 | struct block_device *bdev; | |
2785 | int ret; | |
2786 | int nr = 0; | |
7f3c74fb | 2787 | size_t pg_offset = 0; |
d1310b2e CM |
2788 | size_t blocksize; |
2789 | loff_t i_size = i_size_read(inode); | |
2790 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
2791 | u64 nr_delalloc; | |
2792 | u64 delalloc_end; | |
c8b97818 CM |
2793 | int page_started; |
2794 | int compressed; | |
ffbd517d | 2795 | int write_flags; |
771ed689 | 2796 | unsigned long nr_written = 0; |
9e487107 | 2797 | bool fill_delalloc = true; |
d1310b2e | 2798 | |
ffbd517d | 2799 | if (wbc->sync_mode == WB_SYNC_ALL) |
721a9602 | 2800 | write_flags = WRITE_SYNC; |
ffbd517d CM |
2801 | else |
2802 | write_flags = WRITE; | |
2803 | ||
1abe9b8a | 2804 | trace___extent_writepage(page, inode, wbc); |
2805 | ||
d1310b2e | 2806 | WARN_ON(!PageLocked(page)); |
bf0da8c1 CM |
2807 | |
2808 | ClearPageError(page); | |
2809 | ||
7f3c74fb | 2810 | pg_offset = i_size & (PAGE_CACHE_SIZE - 1); |
211c17f5 | 2811 | if (page->index > end_index || |
7f3c74fb | 2812 | (page->index == end_index && !pg_offset)) { |
39be25cd | 2813 | page->mapping->a_ops->invalidatepage(page, 0); |
d1310b2e CM |
2814 | unlock_page(page); |
2815 | return 0; | |
2816 | } | |
2817 | ||
2818 | if (page->index == end_index) { | |
2819 | char *userpage; | |
2820 | ||
7ac687d9 | 2821 | userpage = kmap_atomic(page); |
7f3c74fb CM |
2822 | memset(userpage + pg_offset, 0, |
2823 | PAGE_CACHE_SIZE - pg_offset); | |
7ac687d9 | 2824 | kunmap_atomic(userpage); |
211c17f5 | 2825 | flush_dcache_page(page); |
d1310b2e | 2826 | } |
7f3c74fb | 2827 | pg_offset = 0; |
d1310b2e CM |
2828 | |
2829 | set_page_extent_mapped(page); | |
2830 | ||
9e487107 JB |
2831 | if (!tree->ops || !tree->ops->fill_delalloc) |
2832 | fill_delalloc = false; | |
2833 | ||
d1310b2e CM |
2834 | delalloc_start = start; |
2835 | delalloc_end = 0; | |
c8b97818 | 2836 | page_started = 0; |
9e487107 | 2837 | if (!epd->extent_locked && fill_delalloc) { |
f85d7d6c | 2838 | u64 delalloc_to_write = 0; |
11c8349b CM |
2839 | /* |
2840 | * make sure the wbc mapping index is at least updated | |
2841 | * to this page. | |
2842 | */ | |
2843 | update_nr_written(page, wbc, 0); | |
2844 | ||
d397712b | 2845 | while (delalloc_end < page_end) { |
771ed689 | 2846 | nr_delalloc = find_lock_delalloc_range(inode, tree, |
c8b97818 CM |
2847 | page, |
2848 | &delalloc_start, | |
d1310b2e CM |
2849 | &delalloc_end, |
2850 | 128 * 1024 * 1024); | |
771ed689 CM |
2851 | if (nr_delalloc == 0) { |
2852 | delalloc_start = delalloc_end + 1; | |
2853 | continue; | |
2854 | } | |
013bd4c3 TI |
2855 | ret = tree->ops->fill_delalloc(inode, page, |
2856 | delalloc_start, | |
2857 | delalloc_end, | |
2858 | &page_started, | |
2859 | &nr_written); | |
79787eaa JM |
2860 | /* File system has been set read-only */ |
2861 | if (ret) { | |
2862 | SetPageError(page); | |
2863 | goto done; | |
2864 | } | |
f85d7d6c CM |
2865 | /* |
2866 | * delalloc_end is already one less than the total | |
2867 | * length, so we don't subtract one from | |
2868 | * PAGE_CACHE_SIZE | |
2869 | */ | |
2870 | delalloc_to_write += (delalloc_end - delalloc_start + | |
2871 | PAGE_CACHE_SIZE) >> | |
2872 | PAGE_CACHE_SHIFT; | |
d1310b2e | 2873 | delalloc_start = delalloc_end + 1; |
d1310b2e | 2874 | } |
f85d7d6c CM |
2875 | if (wbc->nr_to_write < delalloc_to_write) { |
2876 | int thresh = 8192; | |
2877 | ||
2878 | if (delalloc_to_write < thresh * 2) | |
2879 | thresh = delalloc_to_write; | |
2880 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
2881 | thresh); | |
2882 | } | |
c8b97818 | 2883 | |
771ed689 CM |
2884 | /* did the fill delalloc function already unlock and start |
2885 | * the IO? | |
2886 | */ | |
2887 | if (page_started) { | |
2888 | ret = 0; | |
11c8349b CM |
2889 | /* |
2890 | * we've unlocked the page, so we can't update | |
2891 | * the mapping's writeback index, just update | |
2892 | * nr_to_write. | |
2893 | */ | |
2894 | wbc->nr_to_write -= nr_written; | |
2895 | goto done_unlocked; | |
771ed689 | 2896 | } |
c8b97818 | 2897 | } |
247e743c | 2898 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
2899 | ret = tree->ops->writepage_start_hook(page, start, |
2900 | page_end); | |
87826df0 JM |
2901 | if (ret) { |
2902 | /* Fixup worker will requeue */ | |
2903 | if (ret == -EBUSY) | |
2904 | wbc->pages_skipped++; | |
2905 | else | |
2906 | redirty_page_for_writepage(wbc, page); | |
11c8349b | 2907 | update_nr_written(page, wbc, nr_written); |
247e743c | 2908 | unlock_page(page); |
771ed689 | 2909 | ret = 0; |
11c8349b | 2910 | goto done_unlocked; |
247e743c CM |
2911 | } |
2912 | } | |
2913 | ||
11c8349b CM |
2914 | /* |
2915 | * we don't want to touch the inode after unlocking the page, | |
2916 | * so we update the mapping writeback index now | |
2917 | */ | |
2918 | update_nr_written(page, wbc, nr_written + 1); | |
771ed689 | 2919 | |
d1310b2e | 2920 | end = page_end; |
d1310b2e | 2921 | if (last_byte <= start) { |
e6dcd2dc CM |
2922 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2923 | tree->ops->writepage_end_io_hook(page, start, | |
2924 | page_end, NULL, 1); | |
d1310b2e CM |
2925 | goto done; |
2926 | } | |
2927 | ||
d1310b2e CM |
2928 | blocksize = inode->i_sb->s_blocksize; |
2929 | ||
2930 | while (cur <= end) { | |
2931 | if (cur >= last_byte) { | |
e6dcd2dc CM |
2932 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2933 | tree->ops->writepage_end_io_hook(page, cur, | |
2934 | page_end, NULL, 1); | |
d1310b2e CM |
2935 | break; |
2936 | } | |
7f3c74fb | 2937 | em = epd->get_extent(inode, page, pg_offset, cur, |
d1310b2e | 2938 | end - cur + 1, 1); |
c704005d | 2939 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e CM |
2940 | SetPageError(page); |
2941 | break; | |
2942 | } | |
2943 | ||
2944 | extent_offset = cur - em->start; | |
2945 | BUG_ON(extent_map_end(em) <= cur); | |
2946 | BUG_ON(end < cur); | |
2947 | iosize = min(extent_map_end(em) - cur, end - cur + 1); | |
2948 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
2949 | sector = (em->block_start + extent_offset) >> 9; | |
2950 | bdev = em->bdev; | |
2951 | block_start = em->block_start; | |
c8b97818 | 2952 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
2953 | free_extent_map(em); |
2954 | em = NULL; | |
2955 | ||
c8b97818 CM |
2956 | /* |
2957 | * compressed and inline extents are written through other | |
2958 | * paths in the FS | |
2959 | */ | |
2960 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 2961 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
2962 | /* |
2963 | * end_io notification does not happen here for | |
2964 | * compressed extents | |
2965 | */ | |
2966 | if (!compressed && tree->ops && | |
2967 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
2968 | tree->ops->writepage_end_io_hook(page, cur, |
2969 | cur + iosize - 1, | |
2970 | NULL, 1); | |
c8b97818 CM |
2971 | else if (compressed) { |
2972 | /* we don't want to end_page_writeback on | |
2973 | * a compressed extent. this happens | |
2974 | * elsewhere | |
2975 | */ | |
2976 | nr++; | |
2977 | } | |
2978 | ||
2979 | cur += iosize; | |
7f3c74fb | 2980 | pg_offset += iosize; |
d1310b2e CM |
2981 | continue; |
2982 | } | |
d1310b2e CM |
2983 | /* leave this out until we have a page_mkwrite call */ |
2984 | if (0 && !test_range_bit(tree, cur, cur + iosize - 1, | |
9655d298 | 2985 | EXTENT_DIRTY, 0, NULL)) { |
d1310b2e | 2986 | cur = cur + iosize; |
7f3c74fb | 2987 | pg_offset += iosize; |
d1310b2e CM |
2988 | continue; |
2989 | } | |
c8b97818 | 2990 | |
d1310b2e CM |
2991 | if (tree->ops && tree->ops->writepage_io_hook) { |
2992 | ret = tree->ops->writepage_io_hook(page, cur, | |
2993 | cur + iosize - 1); | |
2994 | } else { | |
2995 | ret = 0; | |
2996 | } | |
1259ab75 | 2997 | if (ret) { |
d1310b2e | 2998 | SetPageError(page); |
1259ab75 | 2999 | } else { |
d1310b2e | 3000 | unsigned long max_nr = end_index + 1; |
7f3c74fb | 3001 | |
d1310b2e CM |
3002 | set_range_writeback(tree, cur, cur + iosize - 1); |
3003 | if (!PageWriteback(page)) { | |
d397712b CM |
3004 | printk(KERN_ERR "btrfs warning page %lu not " |
3005 | "writeback, cur %llu end %llu\n", | |
3006 | page->index, (unsigned long long)cur, | |
d1310b2e CM |
3007 | (unsigned long long)end); |
3008 | } | |
3009 | ||
ffbd517d CM |
3010 | ret = submit_extent_page(write_flags, tree, page, |
3011 | sector, iosize, pg_offset, | |
3012 | bdev, &epd->bio, max_nr, | |
c8b97818 CM |
3013 | end_bio_extent_writepage, |
3014 | 0, 0, 0); | |
d1310b2e CM |
3015 | if (ret) |
3016 | SetPageError(page); | |
3017 | } | |
3018 | cur = cur + iosize; | |
7f3c74fb | 3019 | pg_offset += iosize; |
d1310b2e CM |
3020 | nr++; |
3021 | } | |
3022 | done: | |
3023 | if (nr == 0) { | |
3024 | /* make sure the mapping tag for page dirty gets cleared */ | |
3025 | set_page_writeback(page); | |
3026 | end_page_writeback(page); | |
3027 | } | |
d1310b2e | 3028 | unlock_page(page); |
771ed689 | 3029 | |
11c8349b CM |
3030 | done_unlocked: |
3031 | ||
2c64c53d CM |
3032 | /* drop our reference on any cached states */ |
3033 | free_extent_state(cached_state); | |
d1310b2e CM |
3034 | return 0; |
3035 | } | |
3036 | ||
0b32f4bb JB |
3037 | static int eb_wait(void *word) |
3038 | { | |
3039 | io_schedule(); | |
3040 | return 0; | |
3041 | } | |
3042 | ||
3043 | static void wait_on_extent_buffer_writeback(struct extent_buffer *eb) | |
3044 | { | |
3045 | wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait, | |
3046 | TASK_UNINTERRUPTIBLE); | |
3047 | } | |
3048 | ||
3049 | static int lock_extent_buffer_for_io(struct extent_buffer *eb, | |
3050 | struct btrfs_fs_info *fs_info, | |
3051 | struct extent_page_data *epd) | |
3052 | { | |
3053 | unsigned long i, num_pages; | |
3054 | int flush = 0; | |
3055 | int ret = 0; | |
3056 | ||
3057 | if (!btrfs_try_tree_write_lock(eb)) { | |
3058 | flush = 1; | |
3059 | flush_write_bio(epd); | |
3060 | btrfs_tree_lock(eb); | |
3061 | } | |
3062 | ||
3063 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3064 | btrfs_tree_unlock(eb); | |
3065 | if (!epd->sync_io) | |
3066 | return 0; | |
3067 | if (!flush) { | |
3068 | flush_write_bio(epd); | |
3069 | flush = 1; | |
3070 | } | |
a098d8e8 CM |
3071 | while (1) { |
3072 | wait_on_extent_buffer_writeback(eb); | |
3073 | btrfs_tree_lock(eb); | |
3074 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3075 | break; | |
0b32f4bb | 3076 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3077 | } |
3078 | } | |
3079 | ||
3080 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { | |
3081 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
3082 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); | |
3083 | spin_lock(&fs_info->delalloc_lock); | |
3084 | if (fs_info->dirty_metadata_bytes >= eb->len) | |
3085 | fs_info->dirty_metadata_bytes -= eb->len; | |
3086 | else | |
3087 | WARN_ON(1); | |
3088 | spin_unlock(&fs_info->delalloc_lock); | |
3089 | ret = 1; | |
3090 | } | |
3091 | ||
3092 | btrfs_tree_unlock(eb); | |
3093 | ||
3094 | if (!ret) | |
3095 | return ret; | |
3096 | ||
3097 | num_pages = num_extent_pages(eb->start, eb->len); | |
3098 | for (i = 0; i < num_pages; i++) { | |
3099 | struct page *p = extent_buffer_page(eb, i); | |
3100 | ||
3101 | if (!trylock_page(p)) { | |
3102 | if (!flush) { | |
3103 | flush_write_bio(epd); | |
3104 | flush = 1; | |
3105 | } | |
3106 | lock_page(p); | |
3107 | } | |
3108 | } | |
3109 | ||
3110 | return ret; | |
3111 | } | |
3112 | ||
3113 | static void end_extent_buffer_writeback(struct extent_buffer *eb) | |
3114 | { | |
3115 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
3116 | smp_mb__after_clear_bit(); | |
3117 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); | |
3118 | } | |
3119 | ||
3120 | static void end_bio_extent_buffer_writepage(struct bio *bio, int err) | |
3121 | { | |
3122 | int uptodate = err == 0; | |
3123 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
3124 | struct extent_buffer *eb; | |
3125 | int done; | |
3126 | ||
3127 | do { | |
3128 | struct page *page = bvec->bv_page; | |
3129 | ||
3130 | bvec--; | |
3131 | eb = (struct extent_buffer *)page->private; | |
3132 | BUG_ON(!eb); | |
3133 | done = atomic_dec_and_test(&eb->io_pages); | |
3134 | ||
3135 | if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) { | |
3136 | set_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
3137 | ClearPageUptodate(page); | |
3138 | SetPageError(page); | |
3139 | } | |
3140 | ||
3141 | end_page_writeback(page); | |
3142 | ||
3143 | if (!done) | |
3144 | continue; | |
3145 | ||
3146 | end_extent_buffer_writeback(eb); | |
3147 | } while (bvec >= bio->bi_io_vec); | |
3148 | ||
3149 | bio_put(bio); | |
3150 | ||
3151 | } | |
3152 | ||
3153 | static int write_one_eb(struct extent_buffer *eb, | |
3154 | struct btrfs_fs_info *fs_info, | |
3155 | struct writeback_control *wbc, | |
3156 | struct extent_page_data *epd) | |
3157 | { | |
3158 | struct block_device *bdev = fs_info->fs_devices->latest_bdev; | |
3159 | u64 offset = eb->start; | |
3160 | unsigned long i, num_pages; | |
3161 | int rw = (epd->sync_io ? WRITE_SYNC : WRITE); | |
d7dbe9e7 | 3162 | int ret = 0; |
0b32f4bb JB |
3163 | |
3164 | clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
3165 | num_pages = num_extent_pages(eb->start, eb->len); | |
3166 | atomic_set(&eb->io_pages, num_pages); | |
3167 | for (i = 0; i < num_pages; i++) { | |
3168 | struct page *p = extent_buffer_page(eb, i); | |
3169 | ||
3170 | clear_page_dirty_for_io(p); | |
3171 | set_page_writeback(p); | |
3172 | ret = submit_extent_page(rw, eb->tree, p, offset >> 9, | |
3173 | PAGE_CACHE_SIZE, 0, bdev, &epd->bio, | |
3174 | -1, end_bio_extent_buffer_writepage, | |
3175 | 0, 0, 0); | |
3176 | if (ret) { | |
3177 | set_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
3178 | SetPageError(p); | |
3179 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) | |
3180 | end_extent_buffer_writeback(eb); | |
3181 | ret = -EIO; | |
3182 | break; | |
3183 | } | |
3184 | offset += PAGE_CACHE_SIZE; | |
3185 | update_nr_written(p, wbc, 1); | |
3186 | unlock_page(p); | |
3187 | } | |
3188 | ||
3189 | if (unlikely(ret)) { | |
3190 | for (; i < num_pages; i++) { | |
3191 | struct page *p = extent_buffer_page(eb, i); | |
3192 | unlock_page(p); | |
3193 | } | |
3194 | } | |
3195 | ||
3196 | return ret; | |
3197 | } | |
3198 | ||
3199 | int btree_write_cache_pages(struct address_space *mapping, | |
3200 | struct writeback_control *wbc) | |
3201 | { | |
3202 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; | |
3203 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; | |
3204 | struct extent_buffer *eb, *prev_eb = NULL; | |
3205 | struct extent_page_data epd = { | |
3206 | .bio = NULL, | |
3207 | .tree = tree, | |
3208 | .extent_locked = 0, | |
3209 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
3210 | }; | |
3211 | int ret = 0; | |
3212 | int done = 0; | |
3213 | int nr_to_write_done = 0; | |
3214 | struct pagevec pvec; | |
3215 | int nr_pages; | |
3216 | pgoff_t index; | |
3217 | pgoff_t end; /* Inclusive */ | |
3218 | int scanned = 0; | |
3219 | int tag; | |
3220 | ||
3221 | pagevec_init(&pvec, 0); | |
3222 | if (wbc->range_cyclic) { | |
3223 | index = mapping->writeback_index; /* Start from prev offset */ | |
3224 | end = -1; | |
3225 | } else { | |
3226 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
3227 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
3228 | scanned = 1; | |
3229 | } | |
3230 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3231 | tag = PAGECACHE_TAG_TOWRITE; | |
3232 | else | |
3233 | tag = PAGECACHE_TAG_DIRTY; | |
3234 | retry: | |
3235 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3236 | tag_pages_for_writeback(mapping, index, end); | |
3237 | while (!done && !nr_to_write_done && (index <= end) && | |
3238 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, | |
3239 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
3240 | unsigned i; | |
3241 | ||
3242 | scanned = 1; | |
3243 | for (i = 0; i < nr_pages; i++) { | |
3244 | struct page *page = pvec.pages[i]; | |
3245 | ||
3246 | if (!PagePrivate(page)) | |
3247 | continue; | |
3248 | ||
3249 | if (!wbc->range_cyclic && page->index > end) { | |
3250 | done = 1; | |
3251 | break; | |
3252 | } | |
3253 | ||
3254 | eb = (struct extent_buffer *)page->private; | |
3255 | if (!eb) { | |
3256 | WARN_ON(1); | |
3257 | continue; | |
3258 | } | |
3259 | ||
3260 | if (eb == prev_eb) | |
3261 | continue; | |
3262 | ||
3263 | if (!atomic_inc_not_zero(&eb->refs)) { | |
3264 | WARN_ON(1); | |
3265 | continue; | |
3266 | } | |
3267 | ||
3268 | prev_eb = eb; | |
3269 | ret = lock_extent_buffer_for_io(eb, fs_info, &epd); | |
3270 | if (!ret) { | |
3271 | free_extent_buffer(eb); | |
3272 | continue; | |
3273 | } | |
3274 | ||
3275 | ret = write_one_eb(eb, fs_info, wbc, &epd); | |
3276 | if (ret) { | |
3277 | done = 1; | |
3278 | free_extent_buffer(eb); | |
3279 | break; | |
3280 | } | |
3281 | free_extent_buffer(eb); | |
3282 | ||
3283 | /* | |
3284 | * the filesystem may choose to bump up nr_to_write. | |
3285 | * We have to make sure to honor the new nr_to_write | |
3286 | * at any time | |
3287 | */ | |
3288 | nr_to_write_done = wbc->nr_to_write <= 0; | |
3289 | } | |
3290 | pagevec_release(&pvec); | |
3291 | cond_resched(); | |
3292 | } | |
3293 | if (!scanned && !done) { | |
3294 | /* | |
3295 | * We hit the last page and there is more work to be done: wrap | |
3296 | * back to the start of the file | |
3297 | */ | |
3298 | scanned = 1; | |
3299 | index = 0; | |
3300 | goto retry; | |
3301 | } | |
3302 | flush_write_bio(&epd); | |
3303 | return ret; | |
3304 | } | |
3305 | ||
d1310b2e | 3306 | /** |
4bef0848 | 3307 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
3308 | * @mapping: address space structure to write |
3309 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
3310 | * @writepage: function called for each page | |
3311 | * @data: data passed to writepage function | |
3312 | * | |
3313 | * If a page is already under I/O, write_cache_pages() skips it, even | |
3314 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
3315 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
3316 | * and msync() need to guarantee that all the data which was dirty at the time | |
3317 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
3318 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
3319 | * existing IO to complete. | |
3320 | */ | |
b2950863 | 3321 | static int extent_write_cache_pages(struct extent_io_tree *tree, |
4bef0848 CM |
3322 | struct address_space *mapping, |
3323 | struct writeback_control *wbc, | |
d2c3f4f6 CM |
3324 | writepage_t writepage, void *data, |
3325 | void (*flush_fn)(void *)) | |
d1310b2e | 3326 | { |
7fd1a3f7 | 3327 | struct inode *inode = mapping->host; |
d1310b2e CM |
3328 | int ret = 0; |
3329 | int done = 0; | |
f85d7d6c | 3330 | int nr_to_write_done = 0; |
d1310b2e CM |
3331 | struct pagevec pvec; |
3332 | int nr_pages; | |
3333 | pgoff_t index; | |
3334 | pgoff_t end; /* Inclusive */ | |
3335 | int scanned = 0; | |
f7aaa06b | 3336 | int tag; |
d1310b2e | 3337 | |
7fd1a3f7 JB |
3338 | /* |
3339 | * We have to hold onto the inode so that ordered extents can do their | |
3340 | * work when the IO finishes. The alternative to this is failing to add | |
3341 | * an ordered extent if the igrab() fails there and that is a huge pain | |
3342 | * to deal with, so instead just hold onto the inode throughout the | |
3343 | * writepages operation. If it fails here we are freeing up the inode | |
3344 | * anyway and we'd rather not waste our time writing out stuff that is | |
3345 | * going to be truncated anyway. | |
3346 | */ | |
3347 | if (!igrab(inode)) | |
3348 | return 0; | |
3349 | ||
d1310b2e CM |
3350 | pagevec_init(&pvec, 0); |
3351 | if (wbc->range_cyclic) { | |
3352 | index = mapping->writeback_index; /* Start from prev offset */ | |
3353 | end = -1; | |
3354 | } else { | |
3355 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
3356 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
3357 | scanned = 1; |
3358 | } | |
f7aaa06b JB |
3359 | if (wbc->sync_mode == WB_SYNC_ALL) |
3360 | tag = PAGECACHE_TAG_TOWRITE; | |
3361 | else | |
3362 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 3363 | retry: |
f7aaa06b JB |
3364 | if (wbc->sync_mode == WB_SYNC_ALL) |
3365 | tag_pages_for_writeback(mapping, index, end); | |
f85d7d6c | 3366 | while (!done && !nr_to_write_done && (index <= end) && |
f7aaa06b JB |
3367 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
3368 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
d1310b2e CM |
3369 | unsigned i; |
3370 | ||
3371 | scanned = 1; | |
3372 | for (i = 0; i < nr_pages; i++) { | |
3373 | struct page *page = pvec.pages[i]; | |
3374 | ||
3375 | /* | |
3376 | * At this point we hold neither mapping->tree_lock nor | |
3377 | * lock on the page itself: the page may be truncated or | |
3378 | * invalidated (changing page->mapping to NULL), or even | |
3379 | * swizzled back from swapper_space to tmpfs file | |
3380 | * mapping | |
3381 | */ | |
01d658f2 CM |
3382 | if (tree->ops && |
3383 | tree->ops->write_cache_pages_lock_hook) { | |
3384 | tree->ops->write_cache_pages_lock_hook(page, | |
3385 | data, flush_fn); | |
3386 | } else { | |
3387 | if (!trylock_page(page)) { | |
3388 | flush_fn(data); | |
3389 | lock_page(page); | |
3390 | } | |
3391 | } | |
d1310b2e CM |
3392 | |
3393 | if (unlikely(page->mapping != mapping)) { | |
3394 | unlock_page(page); | |
3395 | continue; | |
3396 | } | |
3397 | ||
3398 | if (!wbc->range_cyclic && page->index > end) { | |
3399 | done = 1; | |
3400 | unlock_page(page); | |
3401 | continue; | |
3402 | } | |
3403 | ||
d2c3f4f6 | 3404 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 CM |
3405 | if (PageWriteback(page)) |
3406 | flush_fn(data); | |
d1310b2e | 3407 | wait_on_page_writeback(page); |
d2c3f4f6 | 3408 | } |
d1310b2e CM |
3409 | |
3410 | if (PageWriteback(page) || | |
3411 | !clear_page_dirty_for_io(page)) { | |
3412 | unlock_page(page); | |
3413 | continue; | |
3414 | } | |
3415 | ||
3416 | ret = (*writepage)(page, wbc, data); | |
3417 | ||
3418 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
3419 | unlock_page(page); | |
3420 | ret = 0; | |
3421 | } | |
f85d7d6c | 3422 | if (ret) |
d1310b2e | 3423 | done = 1; |
f85d7d6c CM |
3424 | |
3425 | /* | |
3426 | * the filesystem may choose to bump up nr_to_write. | |
3427 | * We have to make sure to honor the new nr_to_write | |
3428 | * at any time | |
3429 | */ | |
3430 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
3431 | } |
3432 | pagevec_release(&pvec); | |
3433 | cond_resched(); | |
3434 | } | |
3435 | if (!scanned && !done) { | |
3436 | /* | |
3437 | * We hit the last page and there is more work to be done: wrap | |
3438 | * back to the start of the file | |
3439 | */ | |
3440 | scanned = 1; | |
3441 | index = 0; | |
3442 | goto retry; | |
3443 | } | |
7fd1a3f7 | 3444 | btrfs_add_delayed_iput(inode); |
d1310b2e CM |
3445 | return ret; |
3446 | } | |
d1310b2e | 3447 | |
ffbd517d | 3448 | static void flush_epd_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 3449 | { |
d2c3f4f6 | 3450 | if (epd->bio) { |
355808c2 JM |
3451 | int rw = WRITE; |
3452 | int ret; | |
3453 | ||
ffbd517d | 3454 | if (epd->sync_io) |
355808c2 JM |
3455 | rw = WRITE_SYNC; |
3456 | ||
3457 | ret = submit_one_bio(rw, epd->bio, 0, 0); | |
79787eaa | 3458 | BUG_ON(ret < 0); /* -ENOMEM */ |
d2c3f4f6 CM |
3459 | epd->bio = NULL; |
3460 | } | |
3461 | } | |
3462 | ||
ffbd517d CM |
3463 | static noinline void flush_write_bio(void *data) |
3464 | { | |
3465 | struct extent_page_data *epd = data; | |
3466 | flush_epd_write_bio(epd); | |
3467 | } | |
3468 | ||
d1310b2e CM |
3469 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, |
3470 | get_extent_t *get_extent, | |
3471 | struct writeback_control *wbc) | |
3472 | { | |
3473 | int ret; | |
d1310b2e CM |
3474 | struct extent_page_data epd = { |
3475 | .bio = NULL, | |
3476 | .tree = tree, | |
3477 | .get_extent = get_extent, | |
771ed689 | 3478 | .extent_locked = 0, |
ffbd517d | 3479 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e | 3480 | }; |
d1310b2e | 3481 | |
d1310b2e CM |
3482 | ret = __extent_writepage(page, wbc, &epd); |
3483 | ||
ffbd517d | 3484 | flush_epd_write_bio(&epd); |
d1310b2e CM |
3485 | return ret; |
3486 | } | |
d1310b2e | 3487 | |
771ed689 CM |
3488 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, |
3489 | u64 start, u64 end, get_extent_t *get_extent, | |
3490 | int mode) | |
3491 | { | |
3492 | int ret = 0; | |
3493 | struct address_space *mapping = inode->i_mapping; | |
3494 | struct page *page; | |
3495 | unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >> | |
3496 | PAGE_CACHE_SHIFT; | |
3497 | ||
3498 | struct extent_page_data epd = { | |
3499 | .bio = NULL, | |
3500 | .tree = tree, | |
3501 | .get_extent = get_extent, | |
3502 | .extent_locked = 1, | |
ffbd517d | 3503 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
3504 | }; |
3505 | struct writeback_control wbc_writepages = { | |
771ed689 | 3506 | .sync_mode = mode, |
771ed689 CM |
3507 | .nr_to_write = nr_pages * 2, |
3508 | .range_start = start, | |
3509 | .range_end = end + 1, | |
3510 | }; | |
3511 | ||
d397712b | 3512 | while (start <= end) { |
771ed689 CM |
3513 | page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); |
3514 | if (clear_page_dirty_for_io(page)) | |
3515 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
3516 | else { | |
3517 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
3518 | tree->ops->writepage_end_io_hook(page, start, | |
3519 | start + PAGE_CACHE_SIZE - 1, | |
3520 | NULL, 1); | |
3521 | unlock_page(page); | |
3522 | } | |
3523 | page_cache_release(page); | |
3524 | start += PAGE_CACHE_SIZE; | |
3525 | } | |
3526 | ||
ffbd517d | 3527 | flush_epd_write_bio(&epd); |
771ed689 CM |
3528 | return ret; |
3529 | } | |
d1310b2e CM |
3530 | |
3531 | int extent_writepages(struct extent_io_tree *tree, | |
3532 | struct address_space *mapping, | |
3533 | get_extent_t *get_extent, | |
3534 | struct writeback_control *wbc) | |
3535 | { | |
3536 | int ret = 0; | |
3537 | struct extent_page_data epd = { | |
3538 | .bio = NULL, | |
3539 | .tree = tree, | |
3540 | .get_extent = get_extent, | |
771ed689 | 3541 | .extent_locked = 0, |
ffbd517d | 3542 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
3543 | }; |
3544 | ||
4bef0848 | 3545 | ret = extent_write_cache_pages(tree, mapping, wbc, |
d2c3f4f6 CM |
3546 | __extent_writepage, &epd, |
3547 | flush_write_bio); | |
ffbd517d | 3548 | flush_epd_write_bio(&epd); |
d1310b2e CM |
3549 | return ret; |
3550 | } | |
d1310b2e CM |
3551 | |
3552 | int extent_readpages(struct extent_io_tree *tree, | |
3553 | struct address_space *mapping, | |
3554 | struct list_head *pages, unsigned nr_pages, | |
3555 | get_extent_t get_extent) | |
3556 | { | |
3557 | struct bio *bio = NULL; | |
3558 | unsigned page_idx; | |
c8b97818 | 3559 | unsigned long bio_flags = 0; |
d1310b2e | 3560 | |
d1310b2e CM |
3561 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
3562 | struct page *page = list_entry(pages->prev, struct page, lru); | |
3563 | ||
3564 | prefetchw(&page->flags); | |
3565 | list_del(&page->lru); | |
28ecb609 | 3566 | if (!add_to_page_cache_lru(page, mapping, |
43e817a1 | 3567 | page->index, GFP_NOFS)) { |
f188591e | 3568 | __extent_read_full_page(tree, page, get_extent, |
c8b97818 | 3569 | &bio, 0, &bio_flags); |
d1310b2e CM |
3570 | } |
3571 | page_cache_release(page); | |
3572 | } | |
d1310b2e CM |
3573 | BUG_ON(!list_empty(pages)); |
3574 | if (bio) | |
79787eaa | 3575 | return submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
3576 | return 0; |
3577 | } | |
d1310b2e CM |
3578 | |
3579 | /* | |
3580 | * basic invalidatepage code, this waits on any locked or writeback | |
3581 | * ranges corresponding to the page, and then deletes any extent state | |
3582 | * records from the tree | |
3583 | */ | |
3584 | int extent_invalidatepage(struct extent_io_tree *tree, | |
3585 | struct page *page, unsigned long offset) | |
3586 | { | |
2ac55d41 | 3587 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
3588 | u64 start = ((u64)page->index << PAGE_CACHE_SHIFT); |
3589 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
3590 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
3591 | ||
d397712b | 3592 | start += (offset + blocksize - 1) & ~(blocksize - 1); |
d1310b2e CM |
3593 | if (start > end) |
3594 | return 0; | |
3595 | ||
d0082371 | 3596 | lock_extent_bits(tree, start, end, 0, &cached_state); |
1edbb734 | 3597 | wait_on_page_writeback(page); |
d1310b2e | 3598 | clear_extent_bit(tree, start, end, |
32c00aff JB |
3599 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
3600 | EXTENT_DO_ACCOUNTING, | |
2ac55d41 | 3601 | 1, 1, &cached_state, GFP_NOFS); |
d1310b2e CM |
3602 | return 0; |
3603 | } | |
d1310b2e | 3604 | |
7b13b7b1 CM |
3605 | /* |
3606 | * a helper for releasepage, this tests for areas of the page that | |
3607 | * are locked or under IO and drops the related state bits if it is safe | |
3608 | * to drop the page. | |
3609 | */ | |
3610 | int try_release_extent_state(struct extent_map_tree *map, | |
3611 | struct extent_io_tree *tree, struct page *page, | |
3612 | gfp_t mask) | |
3613 | { | |
3614 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
3615 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
3616 | int ret = 1; | |
3617 | ||
211f90e6 | 3618 | if (test_range_bit(tree, start, end, |
8b62b72b | 3619 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
3620 | ret = 0; |
3621 | else { | |
3622 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
3623 | mask = GFP_NOFS; | |
11ef160f CM |
3624 | /* |
3625 | * at this point we can safely clear everything except the | |
3626 | * locked bit and the nodatasum bit | |
3627 | */ | |
e3f24cc5 | 3628 | ret = clear_extent_bit(tree, start, end, |
11ef160f CM |
3629 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
3630 | 0, 0, NULL, mask); | |
e3f24cc5 CM |
3631 | |
3632 | /* if clear_extent_bit failed for enomem reasons, | |
3633 | * we can't allow the release to continue. | |
3634 | */ | |
3635 | if (ret < 0) | |
3636 | ret = 0; | |
3637 | else | |
3638 | ret = 1; | |
7b13b7b1 CM |
3639 | } |
3640 | return ret; | |
3641 | } | |
7b13b7b1 | 3642 | |
d1310b2e CM |
3643 | /* |
3644 | * a helper for releasepage. As long as there are no locked extents | |
3645 | * in the range corresponding to the page, both state records and extent | |
3646 | * map records are removed | |
3647 | */ | |
3648 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
3649 | struct extent_io_tree *tree, struct page *page, |
3650 | gfp_t mask) | |
d1310b2e CM |
3651 | { |
3652 | struct extent_map *em; | |
3653 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
3654 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
7b13b7b1 | 3655 | |
70dec807 CM |
3656 | if ((mask & __GFP_WAIT) && |
3657 | page->mapping->host->i_size > 16 * 1024 * 1024) { | |
39b5637f | 3658 | u64 len; |
70dec807 | 3659 | while (start <= end) { |
39b5637f | 3660 | len = end - start + 1; |
890871be | 3661 | write_lock(&map->lock); |
39b5637f | 3662 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 3663 | if (!em) { |
890871be | 3664 | write_unlock(&map->lock); |
70dec807 CM |
3665 | break; |
3666 | } | |
7f3c74fb CM |
3667 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
3668 | em->start != start) { | |
890871be | 3669 | write_unlock(&map->lock); |
70dec807 CM |
3670 | free_extent_map(em); |
3671 | break; | |
3672 | } | |
3673 | if (!test_range_bit(tree, em->start, | |
3674 | extent_map_end(em) - 1, | |
8b62b72b | 3675 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 3676 | 0, NULL)) { |
70dec807 CM |
3677 | remove_extent_mapping(map, em); |
3678 | /* once for the rb tree */ | |
3679 | free_extent_map(em); | |
3680 | } | |
3681 | start = extent_map_end(em); | |
890871be | 3682 | write_unlock(&map->lock); |
70dec807 CM |
3683 | |
3684 | /* once for us */ | |
d1310b2e CM |
3685 | free_extent_map(em); |
3686 | } | |
d1310b2e | 3687 | } |
7b13b7b1 | 3688 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 3689 | } |
d1310b2e | 3690 | |
ec29ed5b CM |
3691 | /* |
3692 | * helper function for fiemap, which doesn't want to see any holes. | |
3693 | * This maps until we find something past 'last' | |
3694 | */ | |
3695 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
3696 | u64 offset, | |
3697 | u64 last, | |
3698 | get_extent_t *get_extent) | |
3699 | { | |
3700 | u64 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
3701 | struct extent_map *em; | |
3702 | u64 len; | |
3703 | ||
3704 | if (offset >= last) | |
3705 | return NULL; | |
3706 | ||
3707 | while(1) { | |
3708 | len = last - offset; | |
3709 | if (len == 0) | |
3710 | break; | |
3711 | len = (len + sectorsize - 1) & ~(sectorsize - 1); | |
3712 | em = get_extent(inode, NULL, 0, offset, len, 0); | |
c704005d | 3713 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
3714 | return em; |
3715 | ||
3716 | /* if this isn't a hole return it */ | |
3717 | if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) && | |
3718 | em->block_start != EXTENT_MAP_HOLE) { | |
3719 | return em; | |
3720 | } | |
3721 | ||
3722 | /* this is a hole, advance to the next extent */ | |
3723 | offset = extent_map_end(em); | |
3724 | free_extent_map(em); | |
3725 | if (offset >= last) | |
3726 | break; | |
3727 | } | |
3728 | return NULL; | |
3729 | } | |
3730 | ||
1506fcc8 YS |
3731 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
3732 | __u64 start, __u64 len, get_extent_t *get_extent) | |
3733 | { | |
975f84fe | 3734 | int ret = 0; |
1506fcc8 YS |
3735 | u64 off = start; |
3736 | u64 max = start + len; | |
3737 | u32 flags = 0; | |
975f84fe JB |
3738 | u32 found_type; |
3739 | u64 last; | |
ec29ed5b | 3740 | u64 last_for_get_extent = 0; |
1506fcc8 | 3741 | u64 disko = 0; |
ec29ed5b | 3742 | u64 isize = i_size_read(inode); |
975f84fe | 3743 | struct btrfs_key found_key; |
1506fcc8 | 3744 | struct extent_map *em = NULL; |
2ac55d41 | 3745 | struct extent_state *cached_state = NULL; |
975f84fe JB |
3746 | struct btrfs_path *path; |
3747 | struct btrfs_file_extent_item *item; | |
1506fcc8 | 3748 | int end = 0; |
ec29ed5b CM |
3749 | u64 em_start = 0; |
3750 | u64 em_len = 0; | |
3751 | u64 em_end = 0; | |
1506fcc8 | 3752 | unsigned long emflags; |
1506fcc8 YS |
3753 | |
3754 | if (len == 0) | |
3755 | return -EINVAL; | |
3756 | ||
975f84fe JB |
3757 | path = btrfs_alloc_path(); |
3758 | if (!path) | |
3759 | return -ENOMEM; | |
3760 | path->leave_spinning = 1; | |
3761 | ||
4d479cf0 JB |
3762 | start = ALIGN(start, BTRFS_I(inode)->root->sectorsize); |
3763 | len = ALIGN(len, BTRFS_I(inode)->root->sectorsize); | |
3764 | ||
ec29ed5b CM |
3765 | /* |
3766 | * lookup the last file extent. We're not using i_size here | |
3767 | * because there might be preallocation past i_size | |
3768 | */ | |
975f84fe | 3769 | ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root, |
33345d01 | 3770 | path, btrfs_ino(inode), -1, 0); |
975f84fe JB |
3771 | if (ret < 0) { |
3772 | btrfs_free_path(path); | |
3773 | return ret; | |
3774 | } | |
3775 | WARN_ON(!ret); | |
3776 | path->slots[0]--; | |
3777 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3778 | struct btrfs_file_extent_item); | |
3779 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
3780 | found_type = btrfs_key_type(&found_key); | |
3781 | ||
ec29ed5b | 3782 | /* No extents, but there might be delalloc bits */ |
33345d01 | 3783 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 3784 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
3785 | /* have to trust i_size as the end */ |
3786 | last = (u64)-1; | |
3787 | last_for_get_extent = isize; | |
3788 | } else { | |
3789 | /* | |
3790 | * remember the start of the last extent. There are a | |
3791 | * bunch of different factors that go into the length of the | |
3792 | * extent, so its much less complex to remember where it started | |
3793 | */ | |
3794 | last = found_key.offset; | |
3795 | last_for_get_extent = last + 1; | |
975f84fe | 3796 | } |
975f84fe JB |
3797 | btrfs_free_path(path); |
3798 | ||
ec29ed5b CM |
3799 | /* |
3800 | * we might have some extents allocated but more delalloc past those | |
3801 | * extents. so, we trust isize unless the start of the last extent is | |
3802 | * beyond isize | |
3803 | */ | |
3804 | if (last < isize) { | |
3805 | last = (u64)-1; | |
3806 | last_for_get_extent = isize; | |
3807 | } | |
3808 | ||
2ac55d41 | 3809 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0, |
d0082371 | 3810 | &cached_state); |
ec29ed5b | 3811 | |
4d479cf0 | 3812 | em = get_extent_skip_holes(inode, start, last_for_get_extent, |
ec29ed5b | 3813 | get_extent); |
1506fcc8 YS |
3814 | if (!em) |
3815 | goto out; | |
3816 | if (IS_ERR(em)) { | |
3817 | ret = PTR_ERR(em); | |
3818 | goto out; | |
3819 | } | |
975f84fe | 3820 | |
1506fcc8 | 3821 | while (!end) { |
ea8efc74 CM |
3822 | u64 offset_in_extent; |
3823 | ||
3824 | /* break if the extent we found is outside the range */ | |
3825 | if (em->start >= max || extent_map_end(em) < off) | |
3826 | break; | |
3827 | ||
3828 | /* | |
3829 | * get_extent may return an extent that starts before our | |
3830 | * requested range. We have to make sure the ranges | |
3831 | * we return to fiemap always move forward and don't | |
3832 | * overlap, so adjust the offsets here | |
3833 | */ | |
3834 | em_start = max(em->start, off); | |
1506fcc8 | 3835 | |
ea8efc74 CM |
3836 | /* |
3837 | * record the offset from the start of the extent | |
3838 | * for adjusting the disk offset below | |
3839 | */ | |
3840 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 3841 | em_end = extent_map_end(em); |
ea8efc74 | 3842 | em_len = em_end - em_start; |
ec29ed5b | 3843 | emflags = em->flags; |
1506fcc8 YS |
3844 | disko = 0; |
3845 | flags = 0; | |
3846 | ||
ea8efc74 CM |
3847 | /* |
3848 | * bump off for our next call to get_extent | |
3849 | */ | |
3850 | off = extent_map_end(em); | |
3851 | if (off >= max) | |
3852 | end = 1; | |
3853 | ||
93dbfad7 | 3854 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
3855 | end = 1; |
3856 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 3857 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
3858 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
3859 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 3860 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
3861 | flags |= (FIEMAP_EXTENT_DELALLOC | |
3862 | FIEMAP_EXTENT_UNKNOWN); | |
93dbfad7 | 3863 | } else { |
ea8efc74 | 3864 | disko = em->block_start + offset_in_extent; |
1506fcc8 YS |
3865 | } |
3866 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
3867 | flags |= FIEMAP_EXTENT_ENCODED; | |
3868 | ||
1506fcc8 YS |
3869 | free_extent_map(em); |
3870 | em = NULL; | |
ec29ed5b CM |
3871 | if ((em_start >= last) || em_len == (u64)-1 || |
3872 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
3873 | flags |= FIEMAP_EXTENT_LAST; |
3874 | end = 1; | |
3875 | } | |
3876 | ||
ec29ed5b CM |
3877 | /* now scan forward to see if this is really the last extent. */ |
3878 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
3879 | get_extent); | |
3880 | if (IS_ERR(em)) { | |
3881 | ret = PTR_ERR(em); | |
3882 | goto out; | |
3883 | } | |
3884 | if (!em) { | |
975f84fe JB |
3885 | flags |= FIEMAP_EXTENT_LAST; |
3886 | end = 1; | |
3887 | } | |
ec29ed5b CM |
3888 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, |
3889 | em_len, flags); | |
3890 | if (ret) | |
3891 | goto out_free; | |
1506fcc8 YS |
3892 | } |
3893 | out_free: | |
3894 | free_extent_map(em); | |
3895 | out: | |
2ac55d41 JB |
3896 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len, |
3897 | &cached_state, GFP_NOFS); | |
1506fcc8 YS |
3898 | return ret; |
3899 | } | |
3900 | ||
4a54c8c1 | 3901 | inline struct page *extent_buffer_page(struct extent_buffer *eb, |
d1310b2e CM |
3902 | unsigned long i) |
3903 | { | |
727011e0 | 3904 | return eb->pages[i]; |
d1310b2e CM |
3905 | } |
3906 | ||
4a54c8c1 | 3907 | inline unsigned long num_extent_pages(u64 start, u64 len) |
728131d8 | 3908 | { |
6af118ce CM |
3909 | return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) - |
3910 | (start >> PAGE_CACHE_SHIFT); | |
728131d8 CM |
3911 | } |
3912 | ||
727011e0 CM |
3913 | static void __free_extent_buffer(struct extent_buffer *eb) |
3914 | { | |
3915 | #if LEAK_DEBUG | |
3916 | unsigned long flags; | |
3917 | spin_lock_irqsave(&leak_lock, flags); | |
3918 | list_del(&eb->leak_list); | |
3919 | spin_unlock_irqrestore(&leak_lock, flags); | |
3920 | #endif | |
3921 | if (eb->pages && eb->pages != eb->inline_pages) | |
3922 | kfree(eb->pages); | |
3923 | kmem_cache_free(extent_buffer_cache, eb); | |
3924 | } | |
3925 | ||
d1310b2e CM |
3926 | static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree, |
3927 | u64 start, | |
3928 | unsigned long len, | |
3929 | gfp_t mask) | |
3930 | { | |
3931 | struct extent_buffer *eb = NULL; | |
3935127c | 3932 | #if LEAK_DEBUG |
2d2ae547 | 3933 | unsigned long flags; |
4bef0848 | 3934 | #endif |
d1310b2e | 3935 | |
d1310b2e | 3936 | eb = kmem_cache_zalloc(extent_buffer_cache, mask); |
91ca338d TI |
3937 | if (eb == NULL) |
3938 | return NULL; | |
d1310b2e CM |
3939 | eb->start = start; |
3940 | eb->len = len; | |
4f2de97a | 3941 | eb->tree = tree; |
815a51c7 | 3942 | eb->bflags = 0; |
bd681513 CM |
3943 | rwlock_init(&eb->lock); |
3944 | atomic_set(&eb->write_locks, 0); | |
3945 | atomic_set(&eb->read_locks, 0); | |
3946 | atomic_set(&eb->blocking_readers, 0); | |
3947 | atomic_set(&eb->blocking_writers, 0); | |
3948 | atomic_set(&eb->spinning_readers, 0); | |
3949 | atomic_set(&eb->spinning_writers, 0); | |
5b25f70f | 3950 | eb->lock_nested = 0; |
bd681513 CM |
3951 | init_waitqueue_head(&eb->write_lock_wq); |
3952 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 3953 | |
3935127c | 3954 | #if LEAK_DEBUG |
2d2ae547 CM |
3955 | spin_lock_irqsave(&leak_lock, flags); |
3956 | list_add(&eb->leak_list, &buffers); | |
3957 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 3958 | #endif |
3083ee2e | 3959 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 3960 | atomic_set(&eb->refs, 1); |
0b32f4bb | 3961 | atomic_set(&eb->io_pages, 0); |
727011e0 CM |
3962 | |
3963 | if (len > MAX_INLINE_EXTENT_BUFFER_SIZE) { | |
3964 | struct page **pages; | |
3965 | int num_pages = (len + PAGE_CACHE_SIZE - 1) >> | |
3966 | PAGE_CACHE_SHIFT; | |
3967 | pages = kzalloc(num_pages, mask); | |
3968 | if (!pages) { | |
3969 | __free_extent_buffer(eb); | |
3970 | return NULL; | |
3971 | } | |
3972 | eb->pages = pages; | |
3973 | } else { | |
3974 | eb->pages = eb->inline_pages; | |
3975 | } | |
d1310b2e CM |
3976 | |
3977 | return eb; | |
3978 | } | |
3979 | ||
815a51c7 JS |
3980 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) |
3981 | { | |
3982 | unsigned long i; | |
3983 | struct page *p; | |
3984 | struct extent_buffer *new; | |
3985 | unsigned long num_pages = num_extent_pages(src->start, src->len); | |
3986 | ||
3987 | new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_ATOMIC); | |
3988 | if (new == NULL) | |
3989 | return NULL; | |
3990 | ||
3991 | for (i = 0; i < num_pages; i++) { | |
3992 | p = alloc_page(GFP_ATOMIC); | |
3993 | BUG_ON(!p); | |
3994 | attach_extent_buffer_page(new, p); | |
3995 | WARN_ON(PageDirty(p)); | |
3996 | SetPageUptodate(p); | |
3997 | new->pages[i] = p; | |
3998 | } | |
3999 | ||
4000 | copy_extent_buffer(new, src, 0, 0, src->len); | |
4001 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); | |
4002 | set_bit(EXTENT_BUFFER_DUMMY, &new->bflags); | |
4003 | ||
4004 | return new; | |
4005 | } | |
4006 | ||
4007 | struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len) | |
4008 | { | |
4009 | struct extent_buffer *eb; | |
4010 | unsigned long num_pages = num_extent_pages(0, len); | |
4011 | unsigned long i; | |
4012 | ||
4013 | eb = __alloc_extent_buffer(NULL, start, len, GFP_ATOMIC); | |
4014 | if (!eb) | |
4015 | return NULL; | |
4016 | ||
4017 | for (i = 0; i < num_pages; i++) { | |
4018 | eb->pages[i] = alloc_page(GFP_ATOMIC); | |
4019 | if (!eb->pages[i]) | |
4020 | goto err; | |
4021 | } | |
4022 | set_extent_buffer_uptodate(eb); | |
4023 | btrfs_set_header_nritems(eb, 0); | |
4024 | set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4025 | ||
4026 | return eb; | |
4027 | err: | |
4028 | for (i--; i > 0; i--) | |
4029 | __free_page(eb->pages[i]); | |
4030 | __free_extent_buffer(eb); | |
4031 | return NULL; | |
4032 | } | |
4033 | ||
0b32f4bb | 4034 | static int extent_buffer_under_io(struct extent_buffer *eb) |
d1310b2e | 4035 | { |
0b32f4bb JB |
4036 | return (atomic_read(&eb->io_pages) || |
4037 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4038 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
d1310b2e CM |
4039 | } |
4040 | ||
897ca6e9 MX |
4041 | /* |
4042 | * Helper for releasing extent buffer page. | |
4043 | */ | |
4044 | static void btrfs_release_extent_buffer_page(struct extent_buffer *eb, | |
4045 | unsigned long start_idx) | |
4046 | { | |
4047 | unsigned long index; | |
39bab87b | 4048 | unsigned long num_pages; |
897ca6e9 | 4049 | struct page *page; |
815a51c7 | 4050 | int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); |
897ca6e9 | 4051 | |
0b32f4bb | 4052 | BUG_ON(extent_buffer_under_io(eb)); |
897ca6e9 | 4053 | |
39bab87b WSH |
4054 | num_pages = num_extent_pages(eb->start, eb->len); |
4055 | index = start_idx + num_pages; | |
897ca6e9 MX |
4056 | if (start_idx >= index) |
4057 | return; | |
4058 | ||
4059 | do { | |
4060 | index--; | |
4061 | page = extent_buffer_page(eb, index); | |
815a51c7 | 4062 | if (page && mapped) { |
4f2de97a JB |
4063 | spin_lock(&page->mapping->private_lock); |
4064 | /* | |
4065 | * We do this since we'll remove the pages after we've | |
4066 | * removed the eb from the radix tree, so we could race | |
4067 | * and have this page now attached to the new eb. So | |
4068 | * only clear page_private if it's still connected to | |
4069 | * this eb. | |
4070 | */ | |
4071 | if (PagePrivate(page) && | |
4072 | page->private == (unsigned long)eb) { | |
0b32f4bb | 4073 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); |
3083ee2e JB |
4074 | BUG_ON(PageDirty(page)); |
4075 | BUG_ON(PageWriteback(page)); | |
4f2de97a JB |
4076 | /* |
4077 | * We need to make sure we haven't be attached | |
4078 | * to a new eb. | |
4079 | */ | |
4080 | ClearPagePrivate(page); | |
4081 | set_page_private(page, 0); | |
4082 | /* One for the page private */ | |
4083 | page_cache_release(page); | |
4084 | } | |
4085 | spin_unlock(&page->mapping->private_lock); | |
4086 | ||
815a51c7 JS |
4087 | } |
4088 | if (page) { | |
4f2de97a | 4089 | /* One for when we alloced the page */ |
897ca6e9 | 4090 | page_cache_release(page); |
4f2de97a | 4091 | } |
897ca6e9 MX |
4092 | } while (index != start_idx); |
4093 | } | |
4094 | ||
4095 | /* | |
4096 | * Helper for releasing the extent buffer. | |
4097 | */ | |
4098 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4099 | { | |
4100 | btrfs_release_extent_buffer_page(eb, 0); | |
4101 | __free_extent_buffer(eb); | |
4102 | } | |
4103 | ||
0b32f4bb JB |
4104 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
4105 | { | |
4106 | /* the ref bit is tricky. We have to make sure it is set | |
4107 | * if we have the buffer dirty. Otherwise the | |
4108 | * code to free a buffer can end up dropping a dirty | |
4109 | * page | |
4110 | * | |
4111 | * Once the ref bit is set, it won't go away while the | |
4112 | * buffer is dirty or in writeback, and it also won't | |
4113 | * go away while we have the reference count on the | |
4114 | * eb bumped. | |
4115 | * | |
4116 | * We can't just set the ref bit without bumping the | |
4117 | * ref on the eb because free_extent_buffer might | |
4118 | * see the ref bit and try to clear it. If this happens | |
4119 | * free_extent_buffer might end up dropping our original | |
4120 | * ref by mistake and freeing the page before we are able | |
4121 | * to add one more ref. | |
4122 | * | |
4123 | * So bump the ref count first, then set the bit. If someone | |
4124 | * beat us to it, drop the ref we added. | |
4125 | */ | |
594831c4 JB |
4126 | spin_lock(&eb->refs_lock); |
4127 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 4128 | atomic_inc(&eb->refs); |
594831c4 | 4129 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
4130 | } |
4131 | ||
5df4235e JB |
4132 | static void mark_extent_buffer_accessed(struct extent_buffer *eb) |
4133 | { | |
4134 | unsigned long num_pages, i; | |
4135 | ||
0b32f4bb JB |
4136 | check_buffer_tree_ref(eb); |
4137 | ||
5df4235e JB |
4138 | num_pages = num_extent_pages(eb->start, eb->len); |
4139 | for (i = 0; i < num_pages; i++) { | |
4140 | struct page *p = extent_buffer_page(eb, i); | |
4141 | mark_page_accessed(p); | |
4142 | } | |
4143 | } | |
4144 | ||
d1310b2e | 4145 | struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree, |
727011e0 | 4146 | u64 start, unsigned long len) |
d1310b2e CM |
4147 | { |
4148 | unsigned long num_pages = num_extent_pages(start, len); | |
4149 | unsigned long i; | |
4150 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
4151 | struct extent_buffer *eb; | |
6af118ce | 4152 | struct extent_buffer *exists = NULL; |
d1310b2e CM |
4153 | struct page *p; |
4154 | struct address_space *mapping = tree->mapping; | |
4155 | int uptodate = 1; | |
19fe0a8b | 4156 | int ret; |
d1310b2e | 4157 | |
19fe0a8b MX |
4158 | rcu_read_lock(); |
4159 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); | |
4160 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
4161 | rcu_read_unlock(); | |
5df4235e | 4162 | mark_extent_buffer_accessed(eb); |
6af118ce CM |
4163 | return eb; |
4164 | } | |
19fe0a8b | 4165 | rcu_read_unlock(); |
6af118ce | 4166 | |
ba144192 | 4167 | eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS); |
2b114d1d | 4168 | if (!eb) |
d1310b2e CM |
4169 | return NULL; |
4170 | ||
727011e0 | 4171 | for (i = 0; i < num_pages; i++, index++) { |
a6591715 | 4172 | p = find_or_create_page(mapping, index, GFP_NOFS); |
d1310b2e CM |
4173 | if (!p) { |
4174 | WARN_ON(1); | |
6af118ce | 4175 | goto free_eb; |
d1310b2e | 4176 | } |
4f2de97a JB |
4177 | |
4178 | spin_lock(&mapping->private_lock); | |
4179 | if (PagePrivate(p)) { | |
4180 | /* | |
4181 | * We could have already allocated an eb for this page | |
4182 | * and attached one so lets see if we can get a ref on | |
4183 | * the existing eb, and if we can we know it's good and | |
4184 | * we can just return that one, else we know we can just | |
4185 | * overwrite page->private. | |
4186 | */ | |
4187 | exists = (struct extent_buffer *)p->private; | |
4188 | if (atomic_inc_not_zero(&exists->refs)) { | |
4189 | spin_unlock(&mapping->private_lock); | |
4190 | unlock_page(p); | |
17de39ac | 4191 | page_cache_release(p); |
5df4235e | 4192 | mark_extent_buffer_accessed(exists); |
4f2de97a JB |
4193 | goto free_eb; |
4194 | } | |
4195 | ||
0b32f4bb | 4196 | /* |
4f2de97a JB |
4197 | * Do this so attach doesn't complain and we need to |
4198 | * drop the ref the old guy had. | |
4199 | */ | |
4200 | ClearPagePrivate(p); | |
0b32f4bb | 4201 | WARN_ON(PageDirty(p)); |
4f2de97a | 4202 | page_cache_release(p); |
d1310b2e | 4203 | } |
4f2de97a JB |
4204 | attach_extent_buffer_page(eb, p); |
4205 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 4206 | WARN_ON(PageDirty(p)); |
d1310b2e | 4207 | mark_page_accessed(p); |
727011e0 | 4208 | eb->pages[i] = p; |
d1310b2e CM |
4209 | if (!PageUptodate(p)) |
4210 | uptodate = 0; | |
eb14ab8e CM |
4211 | |
4212 | /* | |
4213 | * see below about how we avoid a nasty race with release page | |
4214 | * and why we unlock later | |
4215 | */ | |
d1310b2e CM |
4216 | } |
4217 | if (uptodate) | |
b4ce94de | 4218 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 4219 | again: |
19fe0a8b MX |
4220 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); |
4221 | if (ret) | |
4222 | goto free_eb; | |
4223 | ||
6af118ce | 4224 | spin_lock(&tree->buffer_lock); |
19fe0a8b MX |
4225 | ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb); |
4226 | if (ret == -EEXIST) { | |
4227 | exists = radix_tree_lookup(&tree->buffer, | |
4228 | start >> PAGE_CACHE_SHIFT); | |
115391d2 JB |
4229 | if (!atomic_inc_not_zero(&exists->refs)) { |
4230 | spin_unlock(&tree->buffer_lock); | |
4231 | radix_tree_preload_end(); | |
115391d2 JB |
4232 | exists = NULL; |
4233 | goto again; | |
4234 | } | |
6af118ce | 4235 | spin_unlock(&tree->buffer_lock); |
19fe0a8b | 4236 | radix_tree_preload_end(); |
5df4235e | 4237 | mark_extent_buffer_accessed(exists); |
6af118ce CM |
4238 | goto free_eb; |
4239 | } | |
6af118ce | 4240 | /* add one reference for the tree */ |
0b32f4bb | 4241 | check_buffer_tree_ref(eb); |
f044ba78 | 4242 | spin_unlock(&tree->buffer_lock); |
19fe0a8b | 4243 | radix_tree_preload_end(); |
eb14ab8e CM |
4244 | |
4245 | /* | |
4246 | * there is a race where release page may have | |
4247 | * tried to find this extent buffer in the radix | |
4248 | * but failed. It will tell the VM it is safe to | |
4249 | * reclaim the, and it will clear the page private bit. | |
4250 | * We must make sure to set the page private bit properly | |
4251 | * after the extent buffer is in the radix tree so | |
4252 | * it doesn't get lost | |
4253 | */ | |
727011e0 CM |
4254 | SetPageChecked(eb->pages[0]); |
4255 | for (i = 1; i < num_pages; i++) { | |
4256 | p = extent_buffer_page(eb, i); | |
727011e0 CM |
4257 | ClearPageChecked(p); |
4258 | unlock_page(p); | |
4259 | } | |
4260 | unlock_page(eb->pages[0]); | |
d1310b2e CM |
4261 | return eb; |
4262 | ||
6af118ce | 4263 | free_eb: |
727011e0 CM |
4264 | for (i = 0; i < num_pages; i++) { |
4265 | if (eb->pages[i]) | |
4266 | unlock_page(eb->pages[i]); | |
4267 | } | |
eb14ab8e | 4268 | |
17de39ac | 4269 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
897ca6e9 | 4270 | btrfs_release_extent_buffer(eb); |
6af118ce | 4271 | return exists; |
d1310b2e | 4272 | } |
d1310b2e CM |
4273 | |
4274 | struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree, | |
f09d1f60 | 4275 | u64 start, unsigned long len) |
d1310b2e | 4276 | { |
d1310b2e | 4277 | struct extent_buffer *eb; |
d1310b2e | 4278 | |
19fe0a8b MX |
4279 | rcu_read_lock(); |
4280 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); | |
4281 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
4282 | rcu_read_unlock(); | |
5df4235e | 4283 | mark_extent_buffer_accessed(eb); |
19fe0a8b MX |
4284 | return eb; |
4285 | } | |
4286 | rcu_read_unlock(); | |
0f9dd46c | 4287 | |
19fe0a8b | 4288 | return NULL; |
d1310b2e | 4289 | } |
d1310b2e | 4290 | |
3083ee2e JB |
4291 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
4292 | { | |
4293 | struct extent_buffer *eb = | |
4294 | container_of(head, struct extent_buffer, rcu_head); | |
4295 | ||
4296 | __free_extent_buffer(eb); | |
4297 | } | |
4298 | ||
3083ee2e | 4299 | /* Expects to have eb->eb_lock already held */ |
e64860aa | 4300 | static int release_extent_buffer(struct extent_buffer *eb, gfp_t mask) |
3083ee2e JB |
4301 | { |
4302 | WARN_ON(atomic_read(&eb->refs) == 0); | |
4303 | if (atomic_dec_and_test(&eb->refs)) { | |
815a51c7 JS |
4304 | if (test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) { |
4305 | spin_unlock(&eb->refs_lock); | |
4306 | } else { | |
4307 | struct extent_io_tree *tree = eb->tree; | |
3083ee2e | 4308 | |
815a51c7 | 4309 | spin_unlock(&eb->refs_lock); |
3083ee2e | 4310 | |
815a51c7 JS |
4311 | spin_lock(&tree->buffer_lock); |
4312 | radix_tree_delete(&tree->buffer, | |
4313 | eb->start >> PAGE_CACHE_SHIFT); | |
4314 | spin_unlock(&tree->buffer_lock); | |
4315 | } | |
3083ee2e JB |
4316 | |
4317 | /* Should be safe to release our pages at this point */ | |
4318 | btrfs_release_extent_buffer_page(eb, 0); | |
4319 | ||
4320 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); | |
e64860aa | 4321 | return 1; |
3083ee2e JB |
4322 | } |
4323 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
4324 | |
4325 | return 0; | |
3083ee2e JB |
4326 | } |
4327 | ||
d1310b2e CM |
4328 | void free_extent_buffer(struct extent_buffer *eb) |
4329 | { | |
d1310b2e CM |
4330 | if (!eb) |
4331 | return; | |
4332 | ||
3083ee2e | 4333 | spin_lock(&eb->refs_lock); |
815a51c7 JS |
4334 | if (atomic_read(&eb->refs) == 2 && |
4335 | test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) | |
4336 | atomic_dec(&eb->refs); | |
4337 | ||
3083ee2e JB |
4338 | if (atomic_read(&eb->refs) == 2 && |
4339 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 4340 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
4341 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
4342 | atomic_dec(&eb->refs); | |
4343 | ||
4344 | /* | |
4345 | * I know this is terrible, but it's temporary until we stop tracking | |
4346 | * the uptodate bits and such for the extent buffers. | |
4347 | */ | |
4348 | release_extent_buffer(eb, GFP_ATOMIC); | |
4349 | } | |
4350 | ||
4351 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
4352 | { | |
4353 | if (!eb) | |
d1310b2e CM |
4354 | return; |
4355 | ||
3083ee2e JB |
4356 | spin_lock(&eb->refs_lock); |
4357 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
4358 | ||
0b32f4bb | 4359 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
4360 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
4361 | atomic_dec(&eb->refs); | |
4362 | release_extent_buffer(eb, GFP_NOFS); | |
d1310b2e | 4363 | } |
d1310b2e | 4364 | |
1d4284bd | 4365 | void clear_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 4366 | { |
d1310b2e CM |
4367 | unsigned long i; |
4368 | unsigned long num_pages; | |
4369 | struct page *page; | |
4370 | ||
d1310b2e CM |
4371 | num_pages = num_extent_pages(eb->start, eb->len); |
4372 | ||
4373 | for (i = 0; i < num_pages; i++) { | |
4374 | page = extent_buffer_page(eb, i); | |
b9473439 | 4375 | if (!PageDirty(page)) |
d2c3f4f6 CM |
4376 | continue; |
4377 | ||
a61e6f29 | 4378 | lock_page(page); |
eb14ab8e CM |
4379 | WARN_ON(!PagePrivate(page)); |
4380 | ||
d1310b2e | 4381 | clear_page_dirty_for_io(page); |
0ee0fda0 | 4382 | spin_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
4383 | if (!PageDirty(page)) { |
4384 | radix_tree_tag_clear(&page->mapping->page_tree, | |
4385 | page_index(page), | |
4386 | PAGECACHE_TAG_DIRTY); | |
4387 | } | |
0ee0fda0 | 4388 | spin_unlock_irq(&page->mapping->tree_lock); |
bf0da8c1 | 4389 | ClearPageError(page); |
a61e6f29 | 4390 | unlock_page(page); |
d1310b2e | 4391 | } |
0b32f4bb | 4392 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 4393 | } |
d1310b2e | 4394 | |
0b32f4bb | 4395 | int set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e CM |
4396 | { |
4397 | unsigned long i; | |
4398 | unsigned long num_pages; | |
b9473439 | 4399 | int was_dirty = 0; |
d1310b2e | 4400 | |
0b32f4bb JB |
4401 | check_buffer_tree_ref(eb); |
4402 | ||
b9473439 | 4403 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 4404 | |
d1310b2e | 4405 | num_pages = num_extent_pages(eb->start, eb->len); |
3083ee2e | 4406 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
4407 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
4408 | ||
b9473439 | 4409 | for (i = 0; i < num_pages; i++) |
0b32f4bb | 4410 | set_page_dirty(extent_buffer_page(eb, i)); |
b9473439 | 4411 | return was_dirty; |
d1310b2e | 4412 | } |
d1310b2e | 4413 | |
0b32f4bb | 4414 | static int range_straddles_pages(u64 start, u64 len) |
19b6caf4 CM |
4415 | { |
4416 | if (len < PAGE_CACHE_SIZE) | |
4417 | return 1; | |
4418 | if (start & (PAGE_CACHE_SIZE - 1)) | |
4419 | return 1; | |
4420 | if ((start + len) & (PAGE_CACHE_SIZE - 1)) | |
4421 | return 1; | |
4422 | return 0; | |
4423 | } | |
4424 | ||
0b32f4bb | 4425 | int clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 CM |
4426 | { |
4427 | unsigned long i; | |
4428 | struct page *page; | |
4429 | unsigned long num_pages; | |
4430 | ||
b4ce94de | 4431 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
0b32f4bb | 4432 | num_pages = num_extent_pages(eb->start, eb->len); |
1259ab75 CM |
4433 | for (i = 0; i < num_pages; i++) { |
4434 | page = extent_buffer_page(eb, i); | |
33958dc6 CM |
4435 | if (page) |
4436 | ClearPageUptodate(page); | |
1259ab75 CM |
4437 | } |
4438 | return 0; | |
4439 | } | |
4440 | ||
0b32f4bb | 4441 | int set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e CM |
4442 | { |
4443 | unsigned long i; | |
4444 | struct page *page; | |
4445 | unsigned long num_pages; | |
4446 | ||
0b32f4bb | 4447 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 4448 | num_pages = num_extent_pages(eb->start, eb->len); |
d1310b2e CM |
4449 | for (i = 0; i < num_pages; i++) { |
4450 | page = extent_buffer_page(eb, i); | |
d1310b2e CM |
4451 | SetPageUptodate(page); |
4452 | } | |
4453 | return 0; | |
4454 | } | |
d1310b2e | 4455 | |
ce9adaa5 CM |
4456 | int extent_range_uptodate(struct extent_io_tree *tree, |
4457 | u64 start, u64 end) | |
4458 | { | |
4459 | struct page *page; | |
4460 | int ret; | |
4461 | int pg_uptodate = 1; | |
4462 | int uptodate; | |
4463 | unsigned long index; | |
4464 | ||
0b32f4bb | 4465 | if (range_straddles_pages(start, end - start + 1)) { |
19b6caf4 CM |
4466 | ret = test_range_bit(tree, start, end, |
4467 | EXTENT_UPTODATE, 1, NULL); | |
4468 | if (ret) | |
4469 | return 1; | |
4470 | } | |
d397712b | 4471 | while (start <= end) { |
ce9adaa5 CM |
4472 | index = start >> PAGE_CACHE_SHIFT; |
4473 | page = find_get_page(tree->mapping, index); | |
8bedd51b MH |
4474 | if (!page) |
4475 | return 1; | |
ce9adaa5 CM |
4476 | uptodate = PageUptodate(page); |
4477 | page_cache_release(page); | |
4478 | if (!uptodate) { | |
4479 | pg_uptodate = 0; | |
4480 | break; | |
4481 | } | |
4482 | start += PAGE_CACHE_SIZE; | |
4483 | } | |
4484 | return pg_uptodate; | |
4485 | } | |
4486 | ||
0b32f4bb | 4487 | int extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 4488 | { |
0b32f4bb | 4489 | return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 4490 | } |
d1310b2e CM |
4491 | |
4492 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
bb82ab88 | 4493 | struct extent_buffer *eb, u64 start, int wait, |
f188591e | 4494 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
4495 | { |
4496 | unsigned long i; | |
4497 | unsigned long start_i; | |
4498 | struct page *page; | |
4499 | int err; | |
4500 | int ret = 0; | |
ce9adaa5 CM |
4501 | int locked_pages = 0; |
4502 | int all_uptodate = 1; | |
d1310b2e | 4503 | unsigned long num_pages; |
727011e0 | 4504 | unsigned long num_reads = 0; |
a86c12c7 | 4505 | struct bio *bio = NULL; |
c8b97818 | 4506 | unsigned long bio_flags = 0; |
a86c12c7 | 4507 | |
b4ce94de | 4508 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
4509 | return 0; |
4510 | ||
d1310b2e CM |
4511 | if (start) { |
4512 | WARN_ON(start < eb->start); | |
4513 | start_i = (start >> PAGE_CACHE_SHIFT) - | |
4514 | (eb->start >> PAGE_CACHE_SHIFT); | |
4515 | } else { | |
4516 | start_i = 0; | |
4517 | } | |
4518 | ||
4519 | num_pages = num_extent_pages(eb->start, eb->len); | |
4520 | for (i = start_i; i < num_pages; i++) { | |
4521 | page = extent_buffer_page(eb, i); | |
bb82ab88 | 4522 | if (wait == WAIT_NONE) { |
2db04966 | 4523 | if (!trylock_page(page)) |
ce9adaa5 | 4524 | goto unlock_exit; |
d1310b2e CM |
4525 | } else { |
4526 | lock_page(page); | |
4527 | } | |
ce9adaa5 | 4528 | locked_pages++; |
727011e0 CM |
4529 | if (!PageUptodate(page)) { |
4530 | num_reads++; | |
ce9adaa5 | 4531 | all_uptodate = 0; |
727011e0 | 4532 | } |
ce9adaa5 CM |
4533 | } |
4534 | if (all_uptodate) { | |
4535 | if (start_i == 0) | |
b4ce94de | 4536 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
4537 | goto unlock_exit; |
4538 | } | |
4539 | ||
ea466794 | 4540 | clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags); |
5cf1ab56 | 4541 | eb->read_mirror = 0; |
0b32f4bb | 4542 | atomic_set(&eb->io_pages, num_reads); |
ce9adaa5 CM |
4543 | for (i = start_i; i < num_pages; i++) { |
4544 | page = extent_buffer_page(eb, i); | |
ce9adaa5 | 4545 | if (!PageUptodate(page)) { |
f188591e | 4546 | ClearPageError(page); |
a86c12c7 | 4547 | err = __extent_read_full_page(tree, page, |
f188591e | 4548 | get_extent, &bio, |
c8b97818 | 4549 | mirror_num, &bio_flags); |
d397712b | 4550 | if (err) |
d1310b2e | 4551 | ret = err; |
d1310b2e CM |
4552 | } else { |
4553 | unlock_page(page); | |
4554 | } | |
4555 | } | |
4556 | ||
355808c2 JM |
4557 | if (bio) { |
4558 | err = submit_one_bio(READ, bio, mirror_num, bio_flags); | |
79787eaa JM |
4559 | if (err) |
4560 | return err; | |
355808c2 | 4561 | } |
a86c12c7 | 4562 | |
bb82ab88 | 4563 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 4564 | return ret; |
d397712b | 4565 | |
d1310b2e CM |
4566 | for (i = start_i; i < num_pages; i++) { |
4567 | page = extent_buffer_page(eb, i); | |
4568 | wait_on_page_locked(page); | |
d397712b | 4569 | if (!PageUptodate(page)) |
d1310b2e | 4570 | ret = -EIO; |
d1310b2e | 4571 | } |
d397712b | 4572 | |
d1310b2e | 4573 | return ret; |
ce9adaa5 CM |
4574 | |
4575 | unlock_exit: | |
4576 | i = start_i; | |
d397712b | 4577 | while (locked_pages > 0) { |
ce9adaa5 CM |
4578 | page = extent_buffer_page(eb, i); |
4579 | i++; | |
4580 | unlock_page(page); | |
4581 | locked_pages--; | |
4582 | } | |
4583 | return ret; | |
d1310b2e | 4584 | } |
d1310b2e CM |
4585 | |
4586 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
4587 | unsigned long start, | |
4588 | unsigned long len) | |
4589 | { | |
4590 | size_t cur; | |
4591 | size_t offset; | |
4592 | struct page *page; | |
4593 | char *kaddr; | |
4594 | char *dst = (char *)dstv; | |
4595 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4596 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
4597 | |
4598 | WARN_ON(start > eb->len); | |
4599 | WARN_ON(start + len > eb->start + eb->len); | |
4600 | ||
4601 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4602 | ||
d397712b | 4603 | while (len > 0) { |
d1310b2e | 4604 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
4605 | |
4606 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
a6591715 | 4607 | kaddr = page_address(page); |
d1310b2e | 4608 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
4609 | |
4610 | dst += cur; | |
4611 | len -= cur; | |
4612 | offset = 0; | |
4613 | i++; | |
4614 | } | |
4615 | } | |
d1310b2e CM |
4616 | |
4617 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
a6591715 | 4618 | unsigned long min_len, char **map, |
d1310b2e | 4619 | unsigned long *map_start, |
a6591715 | 4620 | unsigned long *map_len) |
d1310b2e CM |
4621 | { |
4622 | size_t offset = start & (PAGE_CACHE_SIZE - 1); | |
4623 | char *kaddr; | |
4624 | struct page *p; | |
4625 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4626 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4627 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
4628 | PAGE_CACHE_SHIFT; | |
4629 | ||
4630 | if (i != end_i) | |
4631 | return -EINVAL; | |
4632 | ||
4633 | if (i == 0) { | |
4634 | offset = start_offset; | |
4635 | *map_start = 0; | |
4636 | } else { | |
4637 | offset = 0; | |
4638 | *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; | |
4639 | } | |
d397712b | 4640 | |
d1310b2e | 4641 | if (start + min_len > eb->len) { |
d397712b CM |
4642 | printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, " |
4643 | "wanted %lu %lu\n", (unsigned long long)eb->start, | |
4644 | eb->len, start, min_len); | |
d1310b2e | 4645 | WARN_ON(1); |
85026533 | 4646 | return -EINVAL; |
d1310b2e CM |
4647 | } |
4648 | ||
4649 | p = extent_buffer_page(eb, i); | |
a6591715 | 4650 | kaddr = page_address(p); |
d1310b2e CM |
4651 | *map = kaddr + offset; |
4652 | *map_len = PAGE_CACHE_SIZE - offset; | |
4653 | return 0; | |
4654 | } | |
d1310b2e | 4655 | |
d1310b2e CM |
4656 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, |
4657 | unsigned long start, | |
4658 | unsigned long len) | |
4659 | { | |
4660 | size_t cur; | |
4661 | size_t offset; | |
4662 | struct page *page; | |
4663 | char *kaddr; | |
4664 | char *ptr = (char *)ptrv; | |
4665 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4666 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4667 | int ret = 0; | |
4668 | ||
4669 | WARN_ON(start > eb->len); | |
4670 | WARN_ON(start + len > eb->start + eb->len); | |
4671 | ||
4672 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4673 | ||
d397712b | 4674 | while (len > 0) { |
d1310b2e | 4675 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
4676 | |
4677 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
4678 | ||
a6591715 | 4679 | kaddr = page_address(page); |
d1310b2e | 4680 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
4681 | if (ret) |
4682 | break; | |
4683 | ||
4684 | ptr += cur; | |
4685 | len -= cur; | |
4686 | offset = 0; | |
4687 | i++; | |
4688 | } | |
4689 | return ret; | |
4690 | } | |
d1310b2e CM |
4691 | |
4692 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
4693 | unsigned long start, unsigned long len) | |
4694 | { | |
4695 | size_t cur; | |
4696 | size_t offset; | |
4697 | struct page *page; | |
4698 | char *kaddr; | |
4699 | char *src = (char *)srcv; | |
4700 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4701 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4702 | ||
4703 | WARN_ON(start > eb->len); | |
4704 | WARN_ON(start + len > eb->start + eb->len); | |
4705 | ||
4706 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4707 | ||
d397712b | 4708 | while (len > 0) { |
d1310b2e CM |
4709 | page = extent_buffer_page(eb, i); |
4710 | WARN_ON(!PageUptodate(page)); | |
4711 | ||
4712 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 4713 | kaddr = page_address(page); |
d1310b2e | 4714 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
4715 | |
4716 | src += cur; | |
4717 | len -= cur; | |
4718 | offset = 0; | |
4719 | i++; | |
4720 | } | |
4721 | } | |
d1310b2e CM |
4722 | |
4723 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
4724 | unsigned long start, unsigned long len) | |
4725 | { | |
4726 | size_t cur; | |
4727 | size_t offset; | |
4728 | struct page *page; | |
4729 | char *kaddr; | |
4730 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4731 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4732 | ||
4733 | WARN_ON(start > eb->len); | |
4734 | WARN_ON(start + len > eb->start + eb->len); | |
4735 | ||
4736 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4737 | ||
d397712b | 4738 | while (len > 0) { |
d1310b2e CM |
4739 | page = extent_buffer_page(eb, i); |
4740 | WARN_ON(!PageUptodate(page)); | |
4741 | ||
4742 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 4743 | kaddr = page_address(page); |
d1310b2e | 4744 | memset(kaddr + offset, c, cur); |
d1310b2e CM |
4745 | |
4746 | len -= cur; | |
4747 | offset = 0; | |
4748 | i++; | |
4749 | } | |
4750 | } | |
d1310b2e CM |
4751 | |
4752 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
4753 | unsigned long dst_offset, unsigned long src_offset, | |
4754 | unsigned long len) | |
4755 | { | |
4756 | u64 dst_len = dst->len; | |
4757 | size_t cur; | |
4758 | size_t offset; | |
4759 | struct page *page; | |
4760 | char *kaddr; | |
4761 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4762 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
4763 | ||
4764 | WARN_ON(src->len != dst_len); | |
4765 | ||
4766 | offset = (start_offset + dst_offset) & | |
4767 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4768 | ||
d397712b | 4769 | while (len > 0) { |
d1310b2e CM |
4770 | page = extent_buffer_page(dst, i); |
4771 | WARN_ON(!PageUptodate(page)); | |
4772 | ||
4773 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
4774 | ||
a6591715 | 4775 | kaddr = page_address(page); |
d1310b2e | 4776 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
4777 | |
4778 | src_offset += cur; | |
4779 | len -= cur; | |
4780 | offset = 0; | |
4781 | i++; | |
4782 | } | |
4783 | } | |
d1310b2e CM |
4784 | |
4785 | static void move_pages(struct page *dst_page, struct page *src_page, | |
4786 | unsigned long dst_off, unsigned long src_off, | |
4787 | unsigned long len) | |
4788 | { | |
a6591715 | 4789 | char *dst_kaddr = page_address(dst_page); |
d1310b2e CM |
4790 | if (dst_page == src_page) { |
4791 | memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len); | |
4792 | } else { | |
a6591715 | 4793 | char *src_kaddr = page_address(src_page); |
d1310b2e CM |
4794 | char *p = dst_kaddr + dst_off + len; |
4795 | char *s = src_kaddr + src_off + len; | |
4796 | ||
4797 | while (len--) | |
4798 | *--p = *--s; | |
d1310b2e | 4799 | } |
d1310b2e CM |
4800 | } |
4801 | ||
3387206f ST |
4802 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
4803 | { | |
4804 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
4805 | return distance < len; | |
4806 | } | |
4807 | ||
d1310b2e CM |
4808 | static void copy_pages(struct page *dst_page, struct page *src_page, |
4809 | unsigned long dst_off, unsigned long src_off, | |
4810 | unsigned long len) | |
4811 | { | |
a6591715 | 4812 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 4813 | char *src_kaddr; |
727011e0 | 4814 | int must_memmove = 0; |
d1310b2e | 4815 | |
3387206f | 4816 | if (dst_page != src_page) { |
a6591715 | 4817 | src_kaddr = page_address(src_page); |
3387206f | 4818 | } else { |
d1310b2e | 4819 | src_kaddr = dst_kaddr; |
727011e0 CM |
4820 | if (areas_overlap(src_off, dst_off, len)) |
4821 | must_memmove = 1; | |
3387206f | 4822 | } |
d1310b2e | 4823 | |
727011e0 CM |
4824 | if (must_memmove) |
4825 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
4826 | else | |
4827 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
4828 | } |
4829 | ||
4830 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
4831 | unsigned long src_offset, unsigned long len) | |
4832 | { | |
4833 | size_t cur; | |
4834 | size_t dst_off_in_page; | |
4835 | size_t src_off_in_page; | |
4836 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4837 | unsigned long dst_i; | |
4838 | unsigned long src_i; | |
4839 | ||
4840 | if (src_offset + len > dst->len) { | |
d397712b CM |
4841 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
4842 | "len %lu dst len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
4843 | BUG_ON(1); |
4844 | } | |
4845 | if (dst_offset + len > dst->len) { | |
d397712b CM |
4846 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
4847 | "len %lu dst len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
4848 | BUG_ON(1); |
4849 | } | |
4850 | ||
d397712b | 4851 | while (len > 0) { |
d1310b2e CM |
4852 | dst_off_in_page = (start_offset + dst_offset) & |
4853 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4854 | src_off_in_page = (start_offset + src_offset) & | |
4855 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4856 | ||
4857 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
4858 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
4859 | ||
4860 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
4861 | src_off_in_page)); | |
4862 | cur = min_t(unsigned long, cur, | |
4863 | (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); | |
4864 | ||
4865 | copy_pages(extent_buffer_page(dst, dst_i), | |
4866 | extent_buffer_page(dst, src_i), | |
4867 | dst_off_in_page, src_off_in_page, cur); | |
4868 | ||
4869 | src_offset += cur; | |
4870 | dst_offset += cur; | |
4871 | len -= cur; | |
4872 | } | |
4873 | } | |
d1310b2e CM |
4874 | |
4875 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
4876 | unsigned long src_offset, unsigned long len) | |
4877 | { | |
4878 | size_t cur; | |
4879 | size_t dst_off_in_page; | |
4880 | size_t src_off_in_page; | |
4881 | unsigned long dst_end = dst_offset + len - 1; | |
4882 | unsigned long src_end = src_offset + len - 1; | |
4883 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4884 | unsigned long dst_i; | |
4885 | unsigned long src_i; | |
4886 | ||
4887 | if (src_offset + len > dst->len) { | |
d397712b CM |
4888 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
4889 | "len %lu len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
4890 | BUG_ON(1); |
4891 | } | |
4892 | if (dst_offset + len > dst->len) { | |
d397712b CM |
4893 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
4894 | "len %lu len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
4895 | BUG_ON(1); |
4896 | } | |
727011e0 | 4897 | if (dst_offset < src_offset) { |
d1310b2e CM |
4898 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
4899 | return; | |
4900 | } | |
d397712b | 4901 | while (len > 0) { |
d1310b2e CM |
4902 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; |
4903 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
4904 | ||
4905 | dst_off_in_page = (start_offset + dst_end) & | |
4906 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4907 | src_off_in_page = (start_offset + src_end) & | |
4908 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4909 | ||
4910 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
4911 | cur = min(cur, dst_off_in_page + 1); | |
4912 | move_pages(extent_buffer_page(dst, dst_i), | |
4913 | extent_buffer_page(dst, src_i), | |
4914 | dst_off_in_page - cur + 1, | |
4915 | src_off_in_page - cur + 1, cur); | |
4916 | ||
4917 | dst_end -= cur; | |
4918 | src_end -= cur; | |
4919 | len -= cur; | |
4920 | } | |
4921 | } | |
6af118ce | 4922 | |
3083ee2e | 4923 | int try_release_extent_buffer(struct page *page, gfp_t mask) |
19fe0a8b | 4924 | { |
6af118ce | 4925 | struct extent_buffer *eb; |
6af118ce | 4926 | |
3083ee2e JB |
4927 | /* |
4928 | * We need to make sure noboody is attaching this page to an eb right | |
4929 | * now. | |
4930 | */ | |
4931 | spin_lock(&page->mapping->private_lock); | |
4932 | if (!PagePrivate(page)) { | |
4933 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 4934 | return 1; |
45f49bce | 4935 | } |
6af118ce | 4936 | |
3083ee2e JB |
4937 | eb = (struct extent_buffer *)page->private; |
4938 | BUG_ON(!eb); | |
19fe0a8b MX |
4939 | |
4940 | /* | |
3083ee2e JB |
4941 | * This is a little awful but should be ok, we need to make sure that |
4942 | * the eb doesn't disappear out from under us while we're looking at | |
4943 | * this page. | |
19fe0a8b | 4944 | */ |
3083ee2e | 4945 | spin_lock(&eb->refs_lock); |
0b32f4bb | 4946 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
4947 | spin_unlock(&eb->refs_lock); |
4948 | spin_unlock(&page->mapping->private_lock); | |
4949 | return 0; | |
b9473439 | 4950 | } |
3083ee2e | 4951 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 4952 | |
3083ee2e JB |
4953 | if ((mask & GFP_NOFS) == GFP_NOFS) |
4954 | mask = GFP_NOFS; | |
19fe0a8b | 4955 | |
19fe0a8b | 4956 | /* |
3083ee2e JB |
4957 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
4958 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 4959 | */ |
3083ee2e JB |
4960 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
4961 | spin_unlock(&eb->refs_lock); | |
4962 | return 0; | |
b9473439 | 4963 | } |
19fe0a8b | 4964 | |
e64860aa | 4965 | return release_extent_buffer(eb, mask); |
6af118ce | 4966 | } |