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" |
d1310b2e | 23 | |
d1310b2e CM |
24 | static struct kmem_cache *extent_state_cache; |
25 | static struct kmem_cache *extent_buffer_cache; | |
26 | ||
27 | static LIST_HEAD(buffers); | |
28 | static LIST_HEAD(states); | |
4bef0848 | 29 | |
b47eda86 | 30 | #define LEAK_DEBUG 0 |
3935127c | 31 | #if LEAK_DEBUG |
d397712b | 32 | static DEFINE_SPINLOCK(leak_lock); |
4bef0848 | 33 | #endif |
d1310b2e | 34 | |
d1310b2e CM |
35 | #define BUFFER_LRU_MAX 64 |
36 | ||
37 | struct tree_entry { | |
38 | u64 start; | |
39 | u64 end; | |
d1310b2e CM |
40 | struct rb_node rb_node; |
41 | }; | |
42 | ||
43 | struct extent_page_data { | |
44 | struct bio *bio; | |
45 | struct extent_io_tree *tree; | |
46 | get_extent_t *get_extent; | |
771ed689 CM |
47 | |
48 | /* tells writepage not to lock the state bits for this range | |
49 | * it still does the unlocking | |
50 | */ | |
ffbd517d CM |
51 | unsigned int extent_locked:1; |
52 | ||
53 | /* tells the submit_bio code to use a WRITE_SYNC */ | |
54 | unsigned int sync_io:1; | |
d1310b2e CM |
55 | }; |
56 | ||
0b32f4bb | 57 | static noinline void flush_write_bio(void *data); |
c2d904e0 JM |
58 | static inline struct btrfs_fs_info * |
59 | tree_fs_info(struct extent_io_tree *tree) | |
60 | { | |
61 | return btrfs_sb(tree->mapping->host->i_sb); | |
62 | } | |
0b32f4bb | 63 | |
d1310b2e CM |
64 | int __init extent_io_init(void) |
65 | { | |
9601e3f6 CH |
66 | extent_state_cache = kmem_cache_create("extent_state", |
67 | sizeof(struct extent_state), 0, | |
68 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
69 | if (!extent_state_cache) |
70 | return -ENOMEM; | |
71 | ||
9601e3f6 CH |
72 | extent_buffer_cache = kmem_cache_create("extent_buffers", |
73 | sizeof(struct extent_buffer), 0, | |
74 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
d1310b2e CM |
75 | if (!extent_buffer_cache) |
76 | goto free_state_cache; | |
77 | return 0; | |
78 | ||
79 | free_state_cache: | |
80 | kmem_cache_destroy(extent_state_cache); | |
81 | return -ENOMEM; | |
82 | } | |
83 | ||
84 | void extent_io_exit(void) | |
85 | { | |
86 | struct extent_state *state; | |
2d2ae547 | 87 | struct extent_buffer *eb; |
d1310b2e CM |
88 | |
89 | while (!list_empty(&states)) { | |
2d2ae547 | 90 | state = list_entry(states.next, struct extent_state, leak_list); |
d397712b CM |
91 | printk(KERN_ERR "btrfs state leak: start %llu end %llu " |
92 | "state %lu in tree %p refs %d\n", | |
93 | (unsigned long long)state->start, | |
94 | (unsigned long long)state->end, | |
95 | state->state, state->tree, atomic_read(&state->refs)); | |
2d2ae547 | 96 | list_del(&state->leak_list); |
d1310b2e CM |
97 | kmem_cache_free(extent_state_cache, state); |
98 | ||
99 | } | |
100 | ||
2d2ae547 CM |
101 | while (!list_empty(&buffers)) { |
102 | eb = list_entry(buffers.next, struct extent_buffer, leak_list); | |
d397712b CM |
103 | printk(KERN_ERR "btrfs buffer leak start %llu len %lu " |
104 | "refs %d\n", (unsigned long long)eb->start, | |
105 | eb->len, atomic_read(&eb->refs)); | |
2d2ae547 CM |
106 | list_del(&eb->leak_list); |
107 | kmem_cache_free(extent_buffer_cache, eb); | |
108 | } | |
d1310b2e CM |
109 | if (extent_state_cache) |
110 | kmem_cache_destroy(extent_state_cache); | |
111 | if (extent_buffer_cache) | |
112 | kmem_cache_destroy(extent_buffer_cache); | |
113 | } | |
114 | ||
115 | void extent_io_tree_init(struct extent_io_tree *tree, | |
f993c883 | 116 | struct address_space *mapping) |
d1310b2e | 117 | { |
6bef4d31 | 118 | tree->state = RB_ROOT; |
19fe0a8b | 119 | INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC); |
d1310b2e CM |
120 | tree->ops = NULL; |
121 | tree->dirty_bytes = 0; | |
70dec807 | 122 | spin_lock_init(&tree->lock); |
6af118ce | 123 | spin_lock_init(&tree->buffer_lock); |
d1310b2e | 124 | tree->mapping = mapping; |
d1310b2e | 125 | } |
d1310b2e | 126 | |
b2950863 | 127 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
128 | { |
129 | struct extent_state *state; | |
3935127c | 130 | #if LEAK_DEBUG |
2d2ae547 | 131 | unsigned long flags; |
4bef0848 | 132 | #endif |
d1310b2e CM |
133 | |
134 | state = kmem_cache_alloc(extent_state_cache, mask); | |
2b114d1d | 135 | if (!state) |
d1310b2e CM |
136 | return state; |
137 | state->state = 0; | |
d1310b2e | 138 | state->private = 0; |
70dec807 | 139 | state->tree = NULL; |
3935127c | 140 | #if LEAK_DEBUG |
2d2ae547 CM |
141 | spin_lock_irqsave(&leak_lock, flags); |
142 | list_add(&state->leak_list, &states); | |
143 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 144 | #endif |
d1310b2e CM |
145 | atomic_set(&state->refs, 1); |
146 | init_waitqueue_head(&state->wq); | |
143bede5 | 147 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
148 | return state; |
149 | } | |
d1310b2e | 150 | |
4845e44f | 151 | void free_extent_state(struct extent_state *state) |
d1310b2e | 152 | { |
d1310b2e CM |
153 | if (!state) |
154 | return; | |
155 | if (atomic_dec_and_test(&state->refs)) { | |
3935127c | 156 | #if LEAK_DEBUG |
2d2ae547 | 157 | unsigned long flags; |
4bef0848 | 158 | #endif |
70dec807 | 159 | WARN_ON(state->tree); |
3935127c | 160 | #if LEAK_DEBUG |
2d2ae547 CM |
161 | spin_lock_irqsave(&leak_lock, flags); |
162 | list_del(&state->leak_list); | |
163 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 164 | #endif |
143bede5 | 165 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
166 | kmem_cache_free(extent_state_cache, state); |
167 | } | |
168 | } | |
d1310b2e CM |
169 | |
170 | static struct rb_node *tree_insert(struct rb_root *root, u64 offset, | |
171 | struct rb_node *node) | |
172 | { | |
d397712b CM |
173 | struct rb_node **p = &root->rb_node; |
174 | struct rb_node *parent = NULL; | |
d1310b2e CM |
175 | struct tree_entry *entry; |
176 | ||
d397712b | 177 | while (*p) { |
d1310b2e CM |
178 | parent = *p; |
179 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
180 | ||
181 | if (offset < entry->start) | |
182 | p = &(*p)->rb_left; | |
183 | else if (offset > entry->end) | |
184 | p = &(*p)->rb_right; | |
185 | else | |
186 | return parent; | |
187 | } | |
188 | ||
d1310b2e CM |
189 | rb_link_node(node, parent, p); |
190 | rb_insert_color(node, root); | |
191 | return NULL; | |
192 | } | |
193 | ||
80ea96b1 | 194 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
d1310b2e CM |
195 | struct rb_node **prev_ret, |
196 | struct rb_node **next_ret) | |
197 | { | |
80ea96b1 | 198 | struct rb_root *root = &tree->state; |
d397712b | 199 | struct rb_node *n = root->rb_node; |
d1310b2e CM |
200 | struct rb_node *prev = NULL; |
201 | struct rb_node *orig_prev = NULL; | |
202 | struct tree_entry *entry; | |
203 | struct tree_entry *prev_entry = NULL; | |
204 | ||
d397712b | 205 | while (n) { |
d1310b2e CM |
206 | entry = rb_entry(n, struct tree_entry, rb_node); |
207 | prev = n; | |
208 | prev_entry = entry; | |
209 | ||
210 | if (offset < entry->start) | |
211 | n = n->rb_left; | |
212 | else if (offset > entry->end) | |
213 | n = n->rb_right; | |
d397712b | 214 | else |
d1310b2e CM |
215 | return n; |
216 | } | |
217 | ||
218 | if (prev_ret) { | |
219 | orig_prev = prev; | |
d397712b | 220 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
221 | prev = rb_next(prev); |
222 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
223 | } | |
224 | *prev_ret = prev; | |
225 | prev = orig_prev; | |
226 | } | |
227 | ||
228 | if (next_ret) { | |
229 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
d397712b | 230 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
231 | prev = rb_prev(prev); |
232 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
233 | } | |
234 | *next_ret = prev; | |
235 | } | |
236 | return NULL; | |
237 | } | |
238 | ||
80ea96b1 CM |
239 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
240 | u64 offset) | |
d1310b2e | 241 | { |
70dec807 | 242 | struct rb_node *prev = NULL; |
d1310b2e | 243 | struct rb_node *ret; |
70dec807 | 244 | |
80ea96b1 | 245 | ret = __etree_search(tree, offset, &prev, NULL); |
d397712b | 246 | if (!ret) |
d1310b2e CM |
247 | return prev; |
248 | return ret; | |
249 | } | |
250 | ||
9ed74f2d JB |
251 | static void merge_cb(struct extent_io_tree *tree, struct extent_state *new, |
252 | struct extent_state *other) | |
253 | { | |
254 | if (tree->ops && tree->ops->merge_extent_hook) | |
255 | tree->ops->merge_extent_hook(tree->mapping->host, new, | |
256 | other); | |
257 | } | |
258 | ||
d1310b2e CM |
259 | /* |
260 | * utility function to look for merge candidates inside a given range. | |
261 | * Any extents with matching state are merged together into a single | |
262 | * extent in the tree. Extents with EXTENT_IO in their state field | |
263 | * are not merged because the end_io handlers need to be able to do | |
264 | * operations on them without sleeping (or doing allocations/splits). | |
265 | * | |
266 | * This should be called with the tree lock held. | |
267 | */ | |
1bf85046 JM |
268 | static void merge_state(struct extent_io_tree *tree, |
269 | struct extent_state *state) | |
d1310b2e CM |
270 | { |
271 | struct extent_state *other; | |
272 | struct rb_node *other_node; | |
273 | ||
5b21f2ed | 274 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
1bf85046 | 275 | return; |
d1310b2e CM |
276 | |
277 | other_node = rb_prev(&state->rb_node); | |
278 | if (other_node) { | |
279 | other = rb_entry(other_node, struct extent_state, rb_node); | |
280 | if (other->end == state->start - 1 && | |
281 | other->state == state->state) { | |
9ed74f2d | 282 | merge_cb(tree, state, other); |
d1310b2e | 283 | state->start = other->start; |
70dec807 | 284 | other->tree = NULL; |
d1310b2e CM |
285 | rb_erase(&other->rb_node, &tree->state); |
286 | free_extent_state(other); | |
287 | } | |
288 | } | |
289 | other_node = rb_next(&state->rb_node); | |
290 | if (other_node) { | |
291 | other = rb_entry(other_node, struct extent_state, rb_node); | |
292 | if (other->start == state->end + 1 && | |
293 | other->state == state->state) { | |
9ed74f2d | 294 | merge_cb(tree, state, other); |
df98b6e2 JB |
295 | state->end = other->end; |
296 | other->tree = NULL; | |
297 | rb_erase(&other->rb_node, &tree->state); | |
298 | free_extent_state(other); | |
d1310b2e CM |
299 | } |
300 | } | |
d1310b2e CM |
301 | } |
302 | ||
1bf85046 | 303 | static void set_state_cb(struct extent_io_tree *tree, |
0ca1f7ce | 304 | struct extent_state *state, int *bits) |
291d673e | 305 | { |
1bf85046 JM |
306 | if (tree->ops && tree->ops->set_bit_hook) |
307 | tree->ops->set_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
308 | } |
309 | ||
310 | static void clear_state_cb(struct extent_io_tree *tree, | |
0ca1f7ce | 311 | struct extent_state *state, int *bits) |
291d673e | 312 | { |
9ed74f2d JB |
313 | if (tree->ops && tree->ops->clear_bit_hook) |
314 | tree->ops->clear_bit_hook(tree->mapping->host, state, bits); | |
291d673e CM |
315 | } |
316 | ||
3150b699 XG |
317 | static void set_state_bits(struct extent_io_tree *tree, |
318 | struct extent_state *state, int *bits); | |
319 | ||
d1310b2e CM |
320 | /* |
321 | * insert an extent_state struct into the tree. 'bits' are set on the | |
322 | * struct before it is inserted. | |
323 | * | |
324 | * This may return -EEXIST if the extent is already there, in which case the | |
325 | * state struct is freed. | |
326 | * | |
327 | * The tree lock is not taken internally. This is a utility function and | |
328 | * probably isn't what you want to call (see set/clear_extent_bit). | |
329 | */ | |
330 | static int insert_state(struct extent_io_tree *tree, | |
331 | struct extent_state *state, u64 start, u64 end, | |
0ca1f7ce | 332 | int *bits) |
d1310b2e CM |
333 | { |
334 | struct rb_node *node; | |
335 | ||
336 | if (end < start) { | |
d397712b CM |
337 | printk(KERN_ERR "btrfs end < start %llu %llu\n", |
338 | (unsigned long long)end, | |
339 | (unsigned long long)start); | |
d1310b2e CM |
340 | WARN_ON(1); |
341 | } | |
d1310b2e CM |
342 | state->start = start; |
343 | state->end = end; | |
9ed74f2d | 344 | |
3150b699 XG |
345 | set_state_bits(tree, state, bits); |
346 | ||
d1310b2e CM |
347 | node = tree_insert(&tree->state, end, &state->rb_node); |
348 | if (node) { | |
349 | struct extent_state *found; | |
350 | found = rb_entry(node, struct extent_state, rb_node); | |
d397712b CM |
351 | printk(KERN_ERR "btrfs found node %llu %llu on insert of " |
352 | "%llu %llu\n", (unsigned long long)found->start, | |
353 | (unsigned long long)found->end, | |
354 | (unsigned long long)start, (unsigned long long)end); | |
d1310b2e CM |
355 | return -EEXIST; |
356 | } | |
70dec807 | 357 | state->tree = tree; |
d1310b2e CM |
358 | merge_state(tree, state); |
359 | return 0; | |
360 | } | |
361 | ||
1bf85046 | 362 | static void split_cb(struct extent_io_tree *tree, struct extent_state *orig, |
9ed74f2d JB |
363 | u64 split) |
364 | { | |
365 | if (tree->ops && tree->ops->split_extent_hook) | |
1bf85046 | 366 | tree->ops->split_extent_hook(tree->mapping->host, orig, split); |
9ed74f2d JB |
367 | } |
368 | ||
d1310b2e CM |
369 | /* |
370 | * split a given extent state struct in two, inserting the preallocated | |
371 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
372 | * offset inside 'orig' where it should be split. | |
373 | * | |
374 | * Before calling, | |
375 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
376 | * are two extent state structs in the tree: | |
377 | * prealloc: [orig->start, split - 1] | |
378 | * orig: [ split, orig->end ] | |
379 | * | |
380 | * The tree locks are not taken by this function. They need to be held | |
381 | * by the caller. | |
382 | */ | |
383 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
384 | struct extent_state *prealloc, u64 split) | |
385 | { | |
386 | struct rb_node *node; | |
9ed74f2d JB |
387 | |
388 | split_cb(tree, orig, split); | |
389 | ||
d1310b2e CM |
390 | prealloc->start = orig->start; |
391 | prealloc->end = split - 1; | |
392 | prealloc->state = orig->state; | |
393 | orig->start = split; | |
394 | ||
395 | node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node); | |
396 | if (node) { | |
d1310b2e CM |
397 | free_extent_state(prealloc); |
398 | return -EEXIST; | |
399 | } | |
70dec807 | 400 | prealloc->tree = tree; |
d1310b2e CM |
401 | return 0; |
402 | } | |
403 | ||
cdc6a395 LZ |
404 | static struct extent_state *next_state(struct extent_state *state) |
405 | { | |
406 | struct rb_node *next = rb_next(&state->rb_node); | |
407 | if (next) | |
408 | return rb_entry(next, struct extent_state, rb_node); | |
409 | else | |
410 | return NULL; | |
411 | } | |
412 | ||
d1310b2e CM |
413 | /* |
414 | * utility function to clear some bits in an extent state struct. | |
1b303fc0 | 415 | * it will optionally wake up any one waiting on this state (wake == 1). |
d1310b2e CM |
416 | * |
417 | * If no bits are set on the state struct after clearing things, the | |
418 | * struct is freed and removed from the tree | |
419 | */ | |
cdc6a395 LZ |
420 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
421 | struct extent_state *state, | |
422 | int *bits, int wake) | |
d1310b2e | 423 | { |
cdc6a395 | 424 | struct extent_state *next; |
0ca1f7ce | 425 | int bits_to_clear = *bits & ~EXTENT_CTLBITS; |
d1310b2e | 426 | |
0ca1f7ce | 427 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
428 | u64 range = state->end - state->start + 1; |
429 | WARN_ON(range > tree->dirty_bytes); | |
430 | tree->dirty_bytes -= range; | |
431 | } | |
291d673e | 432 | clear_state_cb(tree, state, bits); |
32c00aff | 433 | state->state &= ~bits_to_clear; |
d1310b2e CM |
434 | if (wake) |
435 | wake_up(&state->wq); | |
0ca1f7ce | 436 | if (state->state == 0) { |
cdc6a395 | 437 | next = next_state(state); |
70dec807 | 438 | if (state->tree) { |
d1310b2e | 439 | rb_erase(&state->rb_node, &tree->state); |
70dec807 | 440 | state->tree = NULL; |
d1310b2e CM |
441 | free_extent_state(state); |
442 | } else { | |
443 | WARN_ON(1); | |
444 | } | |
445 | } else { | |
446 | merge_state(tree, state); | |
cdc6a395 | 447 | next = next_state(state); |
d1310b2e | 448 | } |
cdc6a395 | 449 | return next; |
d1310b2e CM |
450 | } |
451 | ||
8233767a XG |
452 | static struct extent_state * |
453 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
454 | { | |
455 | if (!prealloc) | |
456 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
457 | ||
458 | return prealloc; | |
459 | } | |
460 | ||
c2d904e0 JM |
461 | void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
462 | { | |
463 | btrfs_panic(tree_fs_info(tree), err, "Locking error: " | |
464 | "Extent tree was modified by another " | |
465 | "thread while locked."); | |
466 | } | |
467 | ||
d1310b2e CM |
468 | /* |
469 | * clear some bits on a range in the tree. This may require splitting | |
470 | * or inserting elements in the tree, so the gfp mask is used to | |
471 | * indicate which allocations or sleeping are allowed. | |
472 | * | |
473 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
474 | * the given range from the tree regardless of state (ie for truncate). | |
475 | * | |
476 | * the range [start, end] is inclusive. | |
477 | * | |
6763af84 | 478 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e CM |
479 | */ |
480 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
2c64c53d CM |
481 | int bits, int wake, int delete, |
482 | struct extent_state **cached_state, | |
483 | gfp_t mask) | |
d1310b2e CM |
484 | { |
485 | struct extent_state *state; | |
2c64c53d | 486 | struct extent_state *cached; |
d1310b2e CM |
487 | struct extent_state *prealloc = NULL; |
488 | struct rb_node *node; | |
5c939df5 | 489 | u64 last_end; |
d1310b2e | 490 | int err; |
2ac55d41 | 491 | int clear = 0; |
d1310b2e | 492 | |
0ca1f7ce YZ |
493 | if (delete) |
494 | bits |= ~EXTENT_CTLBITS; | |
495 | bits |= EXTENT_FIRST_DELALLOC; | |
496 | ||
2ac55d41 JB |
497 | if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY)) |
498 | clear = 1; | |
d1310b2e CM |
499 | again: |
500 | if (!prealloc && (mask & __GFP_WAIT)) { | |
501 | prealloc = alloc_extent_state(mask); | |
502 | if (!prealloc) | |
503 | return -ENOMEM; | |
504 | } | |
505 | ||
cad321ad | 506 | spin_lock(&tree->lock); |
2c64c53d CM |
507 | if (cached_state) { |
508 | cached = *cached_state; | |
2ac55d41 JB |
509 | |
510 | if (clear) { | |
511 | *cached_state = NULL; | |
512 | cached_state = NULL; | |
513 | } | |
514 | ||
df98b6e2 JB |
515 | if (cached && cached->tree && cached->start <= start && |
516 | cached->end > start) { | |
2ac55d41 JB |
517 | if (clear) |
518 | atomic_dec(&cached->refs); | |
2c64c53d | 519 | state = cached; |
42daec29 | 520 | goto hit_next; |
2c64c53d | 521 | } |
2ac55d41 JB |
522 | if (clear) |
523 | free_extent_state(cached); | |
2c64c53d | 524 | } |
d1310b2e CM |
525 | /* |
526 | * this search will find the extents that end after | |
527 | * our range starts | |
528 | */ | |
80ea96b1 | 529 | node = tree_search(tree, start); |
d1310b2e CM |
530 | if (!node) |
531 | goto out; | |
532 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 533 | hit_next: |
d1310b2e CM |
534 | if (state->start > end) |
535 | goto out; | |
536 | WARN_ON(state->end < start); | |
5c939df5 | 537 | last_end = state->end; |
d1310b2e | 538 | |
0449314a | 539 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
540 | if (!(state->state & bits)) { |
541 | state = next_state(state); | |
0449314a | 542 | goto next; |
cdc6a395 | 543 | } |
0449314a | 544 | |
d1310b2e CM |
545 | /* |
546 | * | ---- desired range ---- | | |
547 | * | state | or | |
548 | * | ------------- state -------------- | | |
549 | * | |
550 | * We need to split the extent we found, and may flip | |
551 | * bits on second half. | |
552 | * | |
553 | * If the extent we found extends past our range, we | |
554 | * just split and search again. It'll get split again | |
555 | * the next time though. | |
556 | * | |
557 | * If the extent we found is inside our range, we clear | |
558 | * the desired bit on it. | |
559 | */ | |
560 | ||
561 | if (state->start < start) { | |
8233767a XG |
562 | prealloc = alloc_extent_state_atomic(prealloc); |
563 | BUG_ON(!prealloc); | |
d1310b2e | 564 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
565 | if (err) |
566 | extent_io_tree_panic(tree, err); | |
567 | ||
d1310b2e CM |
568 | prealloc = NULL; |
569 | if (err) | |
570 | goto out; | |
571 | if (state->end <= end) { | |
d1ac6e41 LB |
572 | state = clear_state_bit(tree, state, &bits, wake); |
573 | goto next; | |
d1310b2e CM |
574 | } |
575 | goto search_again; | |
576 | } | |
577 | /* | |
578 | * | ---- desired range ---- | | |
579 | * | state | | |
580 | * We need to split the extent, and clear the bit | |
581 | * on the first half | |
582 | */ | |
583 | if (state->start <= end && state->end > end) { | |
8233767a XG |
584 | prealloc = alloc_extent_state_atomic(prealloc); |
585 | BUG_ON(!prealloc); | |
d1310b2e | 586 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
587 | if (err) |
588 | extent_io_tree_panic(tree, err); | |
589 | ||
d1310b2e CM |
590 | if (wake) |
591 | wake_up(&state->wq); | |
42daec29 | 592 | |
6763af84 | 593 | clear_state_bit(tree, prealloc, &bits, wake); |
9ed74f2d | 594 | |
d1310b2e CM |
595 | prealloc = NULL; |
596 | goto out; | |
597 | } | |
42daec29 | 598 | |
cdc6a395 | 599 | state = clear_state_bit(tree, state, &bits, wake); |
0449314a | 600 | next: |
5c939df5 YZ |
601 | if (last_end == (u64)-1) |
602 | goto out; | |
603 | start = last_end + 1; | |
cdc6a395 | 604 | if (start <= end && state && !need_resched()) |
692e5759 | 605 | goto hit_next; |
d1310b2e CM |
606 | goto search_again; |
607 | ||
608 | out: | |
cad321ad | 609 | spin_unlock(&tree->lock); |
d1310b2e CM |
610 | if (prealloc) |
611 | free_extent_state(prealloc); | |
612 | ||
6763af84 | 613 | return 0; |
d1310b2e CM |
614 | |
615 | search_again: | |
616 | if (start > end) | |
617 | goto out; | |
cad321ad | 618 | spin_unlock(&tree->lock); |
d1310b2e CM |
619 | if (mask & __GFP_WAIT) |
620 | cond_resched(); | |
621 | goto again; | |
622 | } | |
d1310b2e | 623 | |
143bede5 JM |
624 | static void wait_on_state(struct extent_io_tree *tree, |
625 | struct extent_state *state) | |
641f5219 CH |
626 | __releases(tree->lock) |
627 | __acquires(tree->lock) | |
d1310b2e CM |
628 | { |
629 | DEFINE_WAIT(wait); | |
630 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 631 | spin_unlock(&tree->lock); |
d1310b2e | 632 | schedule(); |
cad321ad | 633 | spin_lock(&tree->lock); |
d1310b2e | 634 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
635 | } |
636 | ||
637 | /* | |
638 | * waits for one or more bits to clear on a range in the state tree. | |
639 | * The range [start, end] is inclusive. | |
640 | * The tree lock is taken by this function | |
641 | */ | |
143bede5 | 642 | void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits) |
d1310b2e CM |
643 | { |
644 | struct extent_state *state; | |
645 | struct rb_node *node; | |
646 | ||
cad321ad | 647 | spin_lock(&tree->lock); |
d1310b2e CM |
648 | again: |
649 | while (1) { | |
650 | /* | |
651 | * this search will find all the extents that end after | |
652 | * our range starts | |
653 | */ | |
80ea96b1 | 654 | node = tree_search(tree, start); |
d1310b2e CM |
655 | if (!node) |
656 | break; | |
657 | ||
658 | state = rb_entry(node, struct extent_state, rb_node); | |
659 | ||
660 | if (state->start > end) | |
661 | goto out; | |
662 | ||
663 | if (state->state & bits) { | |
664 | start = state->start; | |
665 | atomic_inc(&state->refs); | |
666 | wait_on_state(tree, state); | |
667 | free_extent_state(state); | |
668 | goto again; | |
669 | } | |
670 | start = state->end + 1; | |
671 | ||
672 | if (start > end) | |
673 | break; | |
674 | ||
ded91f08 | 675 | cond_resched_lock(&tree->lock); |
d1310b2e CM |
676 | } |
677 | out: | |
cad321ad | 678 | spin_unlock(&tree->lock); |
d1310b2e | 679 | } |
d1310b2e | 680 | |
1bf85046 | 681 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 682 | struct extent_state *state, |
0ca1f7ce | 683 | int *bits) |
d1310b2e | 684 | { |
0ca1f7ce | 685 | int bits_to_set = *bits & ~EXTENT_CTLBITS; |
9ed74f2d | 686 | |
1bf85046 | 687 | set_state_cb(tree, state, bits); |
0ca1f7ce | 688 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
689 | u64 range = state->end - state->start + 1; |
690 | tree->dirty_bytes += range; | |
691 | } | |
0ca1f7ce | 692 | state->state |= bits_to_set; |
d1310b2e CM |
693 | } |
694 | ||
2c64c53d CM |
695 | static void cache_state(struct extent_state *state, |
696 | struct extent_state **cached_ptr) | |
697 | { | |
698 | if (cached_ptr && !(*cached_ptr)) { | |
699 | if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) { | |
700 | *cached_ptr = state; | |
701 | atomic_inc(&state->refs); | |
702 | } | |
703 | } | |
704 | } | |
705 | ||
507903b8 AJ |
706 | static void uncache_state(struct extent_state **cached_ptr) |
707 | { | |
708 | if (cached_ptr && (*cached_ptr)) { | |
709 | struct extent_state *state = *cached_ptr; | |
109b36a2 CM |
710 | *cached_ptr = NULL; |
711 | free_extent_state(state); | |
507903b8 AJ |
712 | } |
713 | } | |
714 | ||
d1310b2e | 715 | /* |
1edbb734 CM |
716 | * set some bits on a range in the tree. This may require allocations or |
717 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 718 | * |
1edbb734 CM |
719 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
720 | * part of the range already has the desired bits set. The start of the | |
721 | * existing range is returned in failed_start in this case. | |
d1310b2e | 722 | * |
1edbb734 | 723 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 724 | */ |
1edbb734 | 725 | |
3fbe5c02 JM |
726 | static int __must_check |
727 | __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
728 | int bits, int exclusive_bits, u64 *failed_start, | |
729 | struct extent_state **cached_state, gfp_t mask) | |
d1310b2e CM |
730 | { |
731 | struct extent_state *state; | |
732 | struct extent_state *prealloc = NULL; | |
733 | struct rb_node *node; | |
d1310b2e | 734 | int err = 0; |
d1310b2e CM |
735 | u64 last_start; |
736 | u64 last_end; | |
42daec29 | 737 | |
0ca1f7ce | 738 | bits |= EXTENT_FIRST_DELALLOC; |
d1310b2e CM |
739 | again: |
740 | if (!prealloc && (mask & __GFP_WAIT)) { | |
741 | prealloc = alloc_extent_state(mask); | |
8233767a | 742 | BUG_ON(!prealloc); |
d1310b2e CM |
743 | } |
744 | ||
cad321ad | 745 | spin_lock(&tree->lock); |
9655d298 CM |
746 | if (cached_state && *cached_state) { |
747 | state = *cached_state; | |
df98b6e2 JB |
748 | if (state->start <= start && state->end > start && |
749 | state->tree) { | |
9655d298 CM |
750 | node = &state->rb_node; |
751 | goto hit_next; | |
752 | } | |
753 | } | |
d1310b2e CM |
754 | /* |
755 | * this search will find all the extents that end after | |
756 | * our range starts. | |
757 | */ | |
80ea96b1 | 758 | node = tree_search(tree, start); |
d1310b2e | 759 | if (!node) { |
8233767a XG |
760 | prealloc = alloc_extent_state_atomic(prealloc); |
761 | BUG_ON(!prealloc); | |
0ca1f7ce | 762 | err = insert_state(tree, prealloc, start, end, &bits); |
c2d904e0 JM |
763 | if (err) |
764 | extent_io_tree_panic(tree, err); | |
765 | ||
d1310b2e | 766 | prealloc = NULL; |
d1310b2e CM |
767 | goto out; |
768 | } | |
d1310b2e | 769 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 770 | hit_next: |
d1310b2e CM |
771 | last_start = state->start; |
772 | last_end = state->end; | |
773 | ||
774 | /* | |
775 | * | ---- desired range ---- | | |
776 | * | state | | |
777 | * | |
778 | * Just lock what we found and keep going | |
779 | */ | |
780 | if (state->start == start && state->end <= end) { | |
1edbb734 | 781 | if (state->state & exclusive_bits) { |
d1310b2e CM |
782 | *failed_start = state->start; |
783 | err = -EEXIST; | |
784 | goto out; | |
785 | } | |
42daec29 | 786 | |
1bf85046 | 787 | set_state_bits(tree, state, &bits); |
2c64c53d | 788 | cache_state(state, cached_state); |
d1310b2e | 789 | merge_state(tree, state); |
5c939df5 YZ |
790 | if (last_end == (u64)-1) |
791 | goto out; | |
792 | start = last_end + 1; | |
d1ac6e41 LB |
793 | state = next_state(state); |
794 | if (start < end && state && state->start == start && | |
795 | !need_resched()) | |
796 | goto hit_next; | |
d1310b2e CM |
797 | goto search_again; |
798 | } | |
799 | ||
800 | /* | |
801 | * | ---- desired range ---- | | |
802 | * | state | | |
803 | * or | |
804 | * | ------------- state -------------- | | |
805 | * | |
806 | * We need to split the extent we found, and may flip bits on | |
807 | * second half. | |
808 | * | |
809 | * If the extent we found extends past our | |
810 | * range, we just split and search again. It'll get split | |
811 | * again the next time though. | |
812 | * | |
813 | * If the extent we found is inside our range, we set the | |
814 | * desired bit on it. | |
815 | */ | |
816 | if (state->start < start) { | |
1edbb734 | 817 | if (state->state & exclusive_bits) { |
d1310b2e CM |
818 | *failed_start = start; |
819 | err = -EEXIST; | |
820 | goto out; | |
821 | } | |
8233767a XG |
822 | |
823 | prealloc = alloc_extent_state_atomic(prealloc); | |
824 | BUG_ON(!prealloc); | |
d1310b2e | 825 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
826 | if (err) |
827 | extent_io_tree_panic(tree, err); | |
828 | ||
d1310b2e CM |
829 | prealloc = NULL; |
830 | if (err) | |
831 | goto out; | |
832 | if (state->end <= end) { | |
1bf85046 | 833 | set_state_bits(tree, state, &bits); |
2c64c53d | 834 | cache_state(state, cached_state); |
d1310b2e | 835 | merge_state(tree, state); |
5c939df5 YZ |
836 | if (last_end == (u64)-1) |
837 | goto out; | |
838 | start = last_end + 1; | |
d1ac6e41 LB |
839 | state = next_state(state); |
840 | if (start < end && state && state->start == start && | |
841 | !need_resched()) | |
842 | goto hit_next; | |
d1310b2e CM |
843 | } |
844 | goto search_again; | |
845 | } | |
846 | /* | |
847 | * | ---- desired range ---- | | |
848 | * | state | or | state | | |
849 | * | |
850 | * There's a hole, we need to insert something in it and | |
851 | * ignore the extent we found. | |
852 | */ | |
853 | if (state->start > start) { | |
854 | u64 this_end; | |
855 | if (end < last_start) | |
856 | this_end = end; | |
857 | else | |
d397712b | 858 | this_end = last_start - 1; |
8233767a XG |
859 | |
860 | prealloc = alloc_extent_state_atomic(prealloc); | |
861 | BUG_ON(!prealloc); | |
c7f895a2 XG |
862 | |
863 | /* | |
864 | * Avoid to free 'prealloc' if it can be merged with | |
865 | * the later extent. | |
866 | */ | |
d1310b2e | 867 | err = insert_state(tree, prealloc, start, this_end, |
0ca1f7ce | 868 | &bits); |
c2d904e0 JM |
869 | if (err) |
870 | extent_io_tree_panic(tree, err); | |
871 | ||
9ed74f2d JB |
872 | cache_state(prealloc, cached_state); |
873 | prealloc = NULL; | |
d1310b2e CM |
874 | start = this_end + 1; |
875 | goto search_again; | |
876 | } | |
877 | /* | |
878 | * | ---- desired range ---- | | |
879 | * | state | | |
880 | * We need to split the extent, and set the bit | |
881 | * on the first half | |
882 | */ | |
883 | if (state->start <= end && state->end > end) { | |
1edbb734 | 884 | if (state->state & exclusive_bits) { |
d1310b2e CM |
885 | *failed_start = start; |
886 | err = -EEXIST; | |
887 | goto out; | |
888 | } | |
8233767a XG |
889 | |
890 | prealloc = alloc_extent_state_atomic(prealloc); | |
891 | BUG_ON(!prealloc); | |
d1310b2e | 892 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
893 | if (err) |
894 | extent_io_tree_panic(tree, err); | |
d1310b2e | 895 | |
1bf85046 | 896 | set_state_bits(tree, prealloc, &bits); |
2c64c53d | 897 | cache_state(prealloc, cached_state); |
d1310b2e CM |
898 | merge_state(tree, prealloc); |
899 | prealloc = NULL; | |
900 | goto out; | |
901 | } | |
902 | ||
903 | goto search_again; | |
904 | ||
905 | out: | |
cad321ad | 906 | spin_unlock(&tree->lock); |
d1310b2e CM |
907 | if (prealloc) |
908 | free_extent_state(prealloc); | |
909 | ||
910 | return err; | |
911 | ||
912 | search_again: | |
913 | if (start > end) | |
914 | goto out; | |
cad321ad | 915 | spin_unlock(&tree->lock); |
d1310b2e CM |
916 | if (mask & __GFP_WAIT) |
917 | cond_resched(); | |
918 | goto again; | |
919 | } | |
d1310b2e | 920 | |
3fbe5c02 JM |
921 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, int bits, |
922 | u64 *failed_start, struct extent_state **cached_state, | |
923 | gfp_t mask) | |
924 | { | |
925 | return __set_extent_bit(tree, start, end, bits, 0, failed_start, | |
926 | cached_state, mask); | |
927 | } | |
928 | ||
929 | ||
462d6fac JB |
930 | /** |
931 | * convert_extent - convert all bits in a given range from one bit to another | |
932 | * @tree: the io tree to search | |
933 | * @start: the start offset in bytes | |
934 | * @end: the end offset in bytes (inclusive) | |
935 | * @bits: the bits to set in this range | |
936 | * @clear_bits: the bits to clear in this range | |
937 | * @mask: the allocation mask | |
938 | * | |
939 | * This will go through and set bits for the given range. If any states exist | |
940 | * already in this range they are set with the given bit and cleared of the | |
941 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
942 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
943 | * boundary bits like LOCK. | |
944 | */ | |
945 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
946 | int bits, int clear_bits, gfp_t mask) | |
947 | { | |
948 | struct extent_state *state; | |
949 | struct extent_state *prealloc = NULL; | |
950 | struct rb_node *node; | |
951 | int err = 0; | |
952 | u64 last_start; | |
953 | u64 last_end; | |
954 | ||
955 | again: | |
956 | if (!prealloc && (mask & __GFP_WAIT)) { | |
957 | prealloc = alloc_extent_state(mask); | |
958 | if (!prealloc) | |
959 | return -ENOMEM; | |
960 | } | |
961 | ||
962 | spin_lock(&tree->lock); | |
963 | /* | |
964 | * this search will find all the extents that end after | |
965 | * our range starts. | |
966 | */ | |
967 | node = tree_search(tree, start); | |
968 | if (!node) { | |
969 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
970 | if (!prealloc) { |
971 | err = -ENOMEM; | |
972 | goto out; | |
973 | } | |
462d6fac JB |
974 | err = insert_state(tree, prealloc, start, end, &bits); |
975 | prealloc = NULL; | |
c2d904e0 JM |
976 | if (err) |
977 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
978 | goto out; |
979 | } | |
980 | state = rb_entry(node, struct extent_state, rb_node); | |
981 | hit_next: | |
982 | last_start = state->start; | |
983 | last_end = state->end; | |
984 | ||
985 | /* | |
986 | * | ---- desired range ---- | | |
987 | * | state | | |
988 | * | |
989 | * Just lock what we found and keep going | |
990 | */ | |
991 | if (state->start == start && state->end <= end) { | |
462d6fac | 992 | set_state_bits(tree, state, &bits); |
d1ac6e41 | 993 | state = clear_state_bit(tree, state, &clear_bits, 0); |
462d6fac JB |
994 | if (last_end == (u64)-1) |
995 | goto out; | |
462d6fac | 996 | start = last_end + 1; |
d1ac6e41 LB |
997 | if (start < end && state && state->start == start && |
998 | !need_resched()) | |
999 | goto hit_next; | |
462d6fac JB |
1000 | goto search_again; |
1001 | } | |
1002 | ||
1003 | /* | |
1004 | * | ---- desired range ---- | | |
1005 | * | state | | |
1006 | * or | |
1007 | * | ------------- state -------------- | | |
1008 | * | |
1009 | * We need to split the extent we found, and may flip bits on | |
1010 | * second half. | |
1011 | * | |
1012 | * If the extent we found extends past our | |
1013 | * range, we just split and search again. It'll get split | |
1014 | * again the next time though. | |
1015 | * | |
1016 | * If the extent we found is inside our range, we set the | |
1017 | * desired bit on it. | |
1018 | */ | |
1019 | if (state->start < start) { | |
1020 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1021 | if (!prealloc) { |
1022 | err = -ENOMEM; | |
1023 | goto out; | |
1024 | } | |
462d6fac | 1025 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1026 | if (err) |
1027 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1028 | prealloc = NULL; |
1029 | if (err) | |
1030 | goto out; | |
1031 | if (state->end <= end) { | |
1032 | set_state_bits(tree, state, &bits); | |
d1ac6e41 | 1033 | state = clear_state_bit(tree, state, &clear_bits, 0); |
462d6fac JB |
1034 | if (last_end == (u64)-1) |
1035 | goto out; | |
1036 | start = last_end + 1; | |
d1ac6e41 LB |
1037 | if (start < end && state && state->start == start && |
1038 | !need_resched()) | |
1039 | goto hit_next; | |
462d6fac JB |
1040 | } |
1041 | goto search_again; | |
1042 | } | |
1043 | /* | |
1044 | * | ---- desired range ---- | | |
1045 | * | state | or | state | | |
1046 | * | |
1047 | * There's a hole, we need to insert something in it and | |
1048 | * ignore the extent we found. | |
1049 | */ | |
1050 | if (state->start > start) { | |
1051 | u64 this_end; | |
1052 | if (end < last_start) | |
1053 | this_end = end; | |
1054 | else | |
1055 | this_end = last_start - 1; | |
1056 | ||
1057 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1058 | if (!prealloc) { |
1059 | err = -ENOMEM; | |
1060 | goto out; | |
1061 | } | |
462d6fac JB |
1062 | |
1063 | /* | |
1064 | * Avoid to free 'prealloc' if it can be merged with | |
1065 | * the later extent. | |
1066 | */ | |
1067 | err = insert_state(tree, prealloc, start, this_end, | |
1068 | &bits); | |
c2d904e0 JM |
1069 | if (err) |
1070 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1071 | prealloc = NULL; |
1072 | start = this_end + 1; | |
1073 | goto search_again; | |
1074 | } | |
1075 | /* | |
1076 | * | ---- desired range ---- | | |
1077 | * | state | | |
1078 | * We need to split the extent, and set the bit | |
1079 | * on the first half | |
1080 | */ | |
1081 | if (state->start <= end && state->end > end) { | |
1082 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1083 | if (!prealloc) { |
1084 | err = -ENOMEM; | |
1085 | goto out; | |
1086 | } | |
462d6fac JB |
1087 | |
1088 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1089 | if (err) |
1090 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1091 | |
1092 | set_state_bits(tree, prealloc, &bits); | |
1093 | clear_state_bit(tree, prealloc, &clear_bits, 0); | |
462d6fac JB |
1094 | prealloc = NULL; |
1095 | goto out; | |
1096 | } | |
1097 | ||
1098 | goto search_again; | |
1099 | ||
1100 | out: | |
1101 | spin_unlock(&tree->lock); | |
1102 | if (prealloc) | |
1103 | free_extent_state(prealloc); | |
1104 | ||
1105 | return err; | |
1106 | ||
1107 | search_again: | |
1108 | if (start > end) | |
1109 | goto out; | |
1110 | spin_unlock(&tree->lock); | |
1111 | if (mask & __GFP_WAIT) | |
1112 | cond_resched(); | |
1113 | goto again; | |
1114 | } | |
1115 | ||
d1310b2e CM |
1116 | /* wrappers around set/clear extent bit */ |
1117 | int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
1118 | gfp_t mask) | |
1119 | { | |
3fbe5c02 | 1120 | return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL, |
2c64c53d | 1121 | NULL, mask); |
d1310b2e | 1122 | } |
d1310b2e CM |
1123 | |
1124 | int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
1125 | int bits, gfp_t mask) | |
1126 | { | |
3fbe5c02 | 1127 | return set_extent_bit(tree, start, end, bits, NULL, |
2c64c53d | 1128 | NULL, mask); |
d1310b2e | 1129 | } |
d1310b2e CM |
1130 | |
1131 | int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, | |
1132 | int bits, gfp_t mask) | |
1133 | { | |
2c64c53d | 1134 | return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask); |
d1310b2e | 1135 | } |
d1310b2e CM |
1136 | |
1137 | int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, | |
2ac55d41 | 1138 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e CM |
1139 | { |
1140 | return set_extent_bit(tree, start, end, | |
fee187d9 | 1141 | EXTENT_DELALLOC | EXTENT_UPTODATE, |
3fbe5c02 | 1142 | NULL, cached_state, mask); |
d1310b2e | 1143 | } |
d1310b2e CM |
1144 | |
1145 | int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, | |
1146 | gfp_t mask) | |
1147 | { | |
1148 | return clear_extent_bit(tree, start, end, | |
32c00aff | 1149 | EXTENT_DIRTY | EXTENT_DELALLOC | |
0ca1f7ce | 1150 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask); |
d1310b2e | 1151 | } |
d1310b2e CM |
1152 | |
1153 | int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end, | |
1154 | gfp_t mask) | |
1155 | { | |
3fbe5c02 | 1156 | return set_extent_bit(tree, start, end, EXTENT_NEW, NULL, |
2c64c53d | 1157 | NULL, mask); |
d1310b2e | 1158 | } |
d1310b2e | 1159 | |
d1310b2e | 1160 | int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
507903b8 | 1161 | struct extent_state **cached_state, gfp_t mask) |
d1310b2e | 1162 | { |
507903b8 | 1163 | return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, |
3fbe5c02 | 1164 | cached_state, mask); |
d1310b2e | 1165 | } |
d1310b2e | 1166 | |
5fd02043 JB |
1167 | int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end, |
1168 | struct extent_state **cached_state, gfp_t mask) | |
d1310b2e | 1169 | { |
2c64c53d | 1170 | return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, |
2ac55d41 | 1171 | cached_state, mask); |
d1310b2e | 1172 | } |
d1310b2e | 1173 | |
d352ac68 CM |
1174 | /* |
1175 | * either insert or lock state struct between start and end use mask to tell | |
1176 | * us if waiting is desired. | |
1177 | */ | |
1edbb734 | 1178 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
d0082371 | 1179 | int bits, struct extent_state **cached_state) |
d1310b2e CM |
1180 | { |
1181 | int err; | |
1182 | u64 failed_start; | |
1183 | while (1) { | |
3fbe5c02 JM |
1184 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits, |
1185 | EXTENT_LOCKED, &failed_start, | |
1186 | cached_state, GFP_NOFS); | |
d0082371 | 1187 | if (err == -EEXIST) { |
d1310b2e CM |
1188 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1189 | start = failed_start; | |
d0082371 | 1190 | } else |
d1310b2e | 1191 | break; |
d1310b2e CM |
1192 | WARN_ON(start > end); |
1193 | } | |
1194 | return err; | |
1195 | } | |
d1310b2e | 1196 | |
d0082371 | 1197 | int lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
1edbb734 | 1198 | { |
d0082371 | 1199 | return lock_extent_bits(tree, start, end, 0, NULL); |
1edbb734 CM |
1200 | } |
1201 | ||
d0082371 | 1202 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1203 | { |
1204 | int err; | |
1205 | u64 failed_start; | |
1206 | ||
3fbe5c02 JM |
1207 | err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
1208 | &failed_start, NULL, GFP_NOFS); | |
6643558d YZ |
1209 | if (err == -EEXIST) { |
1210 | if (failed_start > start) | |
1211 | clear_extent_bit(tree, start, failed_start - 1, | |
d0082371 | 1212 | EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS); |
25179201 | 1213 | return 0; |
6643558d | 1214 | } |
25179201 JB |
1215 | return 1; |
1216 | } | |
25179201 | 1217 | |
2c64c53d CM |
1218 | int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end, |
1219 | struct extent_state **cached, gfp_t mask) | |
1220 | { | |
1221 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, | |
1222 | mask); | |
1223 | } | |
1224 | ||
d0082371 | 1225 | int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e | 1226 | { |
2c64c53d | 1227 | return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL, |
d0082371 | 1228 | GFP_NOFS); |
d1310b2e | 1229 | } |
d1310b2e | 1230 | |
d1310b2e CM |
1231 | /* |
1232 | * helper function to set both pages and extents in the tree writeback | |
1233 | */ | |
b2950863 | 1234 | static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
d1310b2e CM |
1235 | { |
1236 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1237 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1238 | struct page *page; | |
1239 | ||
1240 | while (index <= end_index) { | |
1241 | page = find_get_page(tree->mapping, index); | |
79787eaa | 1242 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ |
d1310b2e CM |
1243 | set_page_writeback(page); |
1244 | page_cache_release(page); | |
1245 | index++; | |
1246 | } | |
d1310b2e CM |
1247 | return 0; |
1248 | } | |
d1310b2e | 1249 | |
d352ac68 CM |
1250 | /* find the first state struct with 'bits' set after 'start', and |
1251 | * return it. tree->lock must be held. NULL will returned if | |
1252 | * nothing was found after 'start' | |
1253 | */ | |
d7fc640e CM |
1254 | struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree, |
1255 | u64 start, int bits) | |
1256 | { | |
1257 | struct rb_node *node; | |
1258 | struct extent_state *state; | |
1259 | ||
1260 | /* | |
1261 | * this search will find all the extents that end after | |
1262 | * our range starts. | |
1263 | */ | |
1264 | node = tree_search(tree, start); | |
d397712b | 1265 | if (!node) |
d7fc640e | 1266 | goto out; |
d7fc640e | 1267 | |
d397712b | 1268 | while (1) { |
d7fc640e | 1269 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1270 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1271 | return state; |
d397712b | 1272 | |
d7fc640e CM |
1273 | node = rb_next(node); |
1274 | if (!node) | |
1275 | break; | |
1276 | } | |
1277 | out: | |
1278 | return NULL; | |
1279 | } | |
d7fc640e | 1280 | |
69261c4b XG |
1281 | /* |
1282 | * find the first offset in the io tree with 'bits' set. zero is | |
1283 | * returned if we find something, and *start_ret and *end_ret are | |
1284 | * set to reflect the state struct that was found. | |
1285 | * | |
477d7eaf | 1286 | * If nothing was found, 1 is returned. If found something, return 0. |
69261c4b XG |
1287 | */ |
1288 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
1289 | u64 *start_ret, u64 *end_ret, int bits) | |
1290 | { | |
1291 | struct extent_state *state; | |
1292 | int ret = 1; | |
1293 | ||
1294 | spin_lock(&tree->lock); | |
1295 | state = find_first_extent_bit_state(tree, start, bits); | |
1296 | if (state) { | |
1297 | *start_ret = state->start; | |
1298 | *end_ret = state->end; | |
1299 | ret = 0; | |
1300 | } | |
1301 | spin_unlock(&tree->lock); | |
1302 | return ret; | |
1303 | } | |
1304 | ||
d352ac68 CM |
1305 | /* |
1306 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1307 | * more than 'max_bytes'. start and end are used to return the range, | |
1308 | * | |
1309 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1310 | */ | |
c8b97818 | 1311 | static noinline u64 find_delalloc_range(struct extent_io_tree *tree, |
c2a128d2 JB |
1312 | u64 *start, u64 *end, u64 max_bytes, |
1313 | struct extent_state **cached_state) | |
d1310b2e CM |
1314 | { |
1315 | struct rb_node *node; | |
1316 | struct extent_state *state; | |
1317 | u64 cur_start = *start; | |
1318 | u64 found = 0; | |
1319 | u64 total_bytes = 0; | |
1320 | ||
cad321ad | 1321 | spin_lock(&tree->lock); |
c8b97818 | 1322 | |
d1310b2e CM |
1323 | /* |
1324 | * this search will find all the extents that end after | |
1325 | * our range starts. | |
1326 | */ | |
80ea96b1 | 1327 | node = tree_search(tree, cur_start); |
2b114d1d | 1328 | if (!node) { |
3b951516 CM |
1329 | if (!found) |
1330 | *end = (u64)-1; | |
d1310b2e CM |
1331 | goto out; |
1332 | } | |
1333 | ||
d397712b | 1334 | while (1) { |
d1310b2e | 1335 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1336 | if (found && (state->start != cur_start || |
1337 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1338 | goto out; |
1339 | } | |
1340 | if (!(state->state & EXTENT_DELALLOC)) { | |
1341 | if (!found) | |
1342 | *end = state->end; | |
1343 | goto out; | |
1344 | } | |
c2a128d2 | 1345 | if (!found) { |
d1310b2e | 1346 | *start = state->start; |
c2a128d2 JB |
1347 | *cached_state = state; |
1348 | atomic_inc(&state->refs); | |
1349 | } | |
d1310b2e CM |
1350 | found++; |
1351 | *end = state->end; | |
1352 | cur_start = state->end + 1; | |
1353 | node = rb_next(node); | |
1354 | if (!node) | |
1355 | break; | |
1356 | total_bytes += state->end - state->start + 1; | |
1357 | if (total_bytes >= max_bytes) | |
1358 | break; | |
1359 | } | |
1360 | out: | |
cad321ad | 1361 | spin_unlock(&tree->lock); |
d1310b2e CM |
1362 | return found; |
1363 | } | |
1364 | ||
143bede5 JM |
1365 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1366 | struct page *locked_page, | |
1367 | u64 start, u64 end) | |
c8b97818 CM |
1368 | { |
1369 | int ret; | |
1370 | struct page *pages[16]; | |
1371 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1372 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1373 | unsigned long nr_pages = end_index - index + 1; | |
1374 | int i; | |
1375 | ||
1376 | if (index == locked_page->index && end_index == index) | |
143bede5 | 1377 | return; |
c8b97818 | 1378 | |
d397712b | 1379 | while (nr_pages > 0) { |
c8b97818 | 1380 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1381 | min_t(unsigned long, nr_pages, |
1382 | ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1383 | for (i = 0; i < ret; i++) { |
1384 | if (pages[i] != locked_page) | |
1385 | unlock_page(pages[i]); | |
1386 | page_cache_release(pages[i]); | |
1387 | } | |
1388 | nr_pages -= ret; | |
1389 | index += ret; | |
1390 | cond_resched(); | |
1391 | } | |
c8b97818 CM |
1392 | } |
1393 | ||
1394 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1395 | struct page *locked_page, | |
1396 | u64 delalloc_start, | |
1397 | u64 delalloc_end) | |
1398 | { | |
1399 | unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT; | |
1400 | unsigned long start_index = index; | |
1401 | unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT; | |
1402 | unsigned long pages_locked = 0; | |
1403 | struct page *pages[16]; | |
1404 | unsigned long nrpages; | |
1405 | int ret; | |
1406 | int i; | |
1407 | ||
1408 | /* the caller is responsible for locking the start index */ | |
1409 | if (index == locked_page->index && index == end_index) | |
1410 | return 0; | |
1411 | ||
1412 | /* skip the page at the start index */ | |
1413 | nrpages = end_index - index + 1; | |
d397712b | 1414 | while (nrpages > 0) { |
c8b97818 | 1415 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1416 | min_t(unsigned long, |
1417 | nrpages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
1418 | if (ret == 0) { |
1419 | ret = -EAGAIN; | |
1420 | goto done; | |
1421 | } | |
1422 | /* now we have an array of pages, lock them all */ | |
1423 | for (i = 0; i < ret; i++) { | |
1424 | /* | |
1425 | * the caller is taking responsibility for | |
1426 | * locked_page | |
1427 | */ | |
771ed689 | 1428 | if (pages[i] != locked_page) { |
c8b97818 | 1429 | lock_page(pages[i]); |
f2b1c41c CM |
1430 | if (!PageDirty(pages[i]) || |
1431 | pages[i]->mapping != inode->i_mapping) { | |
771ed689 CM |
1432 | ret = -EAGAIN; |
1433 | unlock_page(pages[i]); | |
1434 | page_cache_release(pages[i]); | |
1435 | goto done; | |
1436 | } | |
1437 | } | |
c8b97818 | 1438 | page_cache_release(pages[i]); |
771ed689 | 1439 | pages_locked++; |
c8b97818 | 1440 | } |
c8b97818 CM |
1441 | nrpages -= ret; |
1442 | index += ret; | |
1443 | cond_resched(); | |
1444 | } | |
1445 | ret = 0; | |
1446 | done: | |
1447 | if (ret && pages_locked) { | |
1448 | __unlock_for_delalloc(inode, locked_page, | |
1449 | delalloc_start, | |
1450 | ((u64)(start_index + pages_locked - 1)) << | |
1451 | PAGE_CACHE_SHIFT); | |
1452 | } | |
1453 | return ret; | |
1454 | } | |
1455 | ||
1456 | /* | |
1457 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1458 | * more than 'max_bytes'. start and end are used to return the range, | |
1459 | * | |
1460 | * 1 is returned if we find something, 0 if nothing was in the tree | |
1461 | */ | |
1462 | static noinline u64 find_lock_delalloc_range(struct inode *inode, | |
1463 | struct extent_io_tree *tree, | |
1464 | struct page *locked_page, | |
1465 | u64 *start, u64 *end, | |
1466 | u64 max_bytes) | |
1467 | { | |
1468 | u64 delalloc_start; | |
1469 | u64 delalloc_end; | |
1470 | u64 found; | |
9655d298 | 1471 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1472 | int ret; |
1473 | int loops = 0; | |
1474 | ||
1475 | again: | |
1476 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1477 | delalloc_start = *start; | |
1478 | delalloc_end = 0; | |
1479 | found = find_delalloc_range(tree, &delalloc_start, &delalloc_end, | |
c2a128d2 | 1480 | max_bytes, &cached_state); |
70b99e69 | 1481 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
1482 | *start = delalloc_start; |
1483 | *end = delalloc_end; | |
c2a128d2 | 1484 | free_extent_state(cached_state); |
c8b97818 CM |
1485 | return found; |
1486 | } | |
1487 | ||
70b99e69 CM |
1488 | /* |
1489 | * start comes from the offset of locked_page. We have to lock | |
1490 | * pages in order, so we can't process delalloc bytes before | |
1491 | * locked_page | |
1492 | */ | |
d397712b | 1493 | if (delalloc_start < *start) |
70b99e69 | 1494 | delalloc_start = *start; |
70b99e69 | 1495 | |
c8b97818 CM |
1496 | /* |
1497 | * make sure to limit the number of pages we try to lock down | |
1498 | * if we're looping. | |
1499 | */ | |
d397712b | 1500 | if (delalloc_end + 1 - delalloc_start > max_bytes && loops) |
771ed689 | 1501 | delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1; |
d397712b | 1502 | |
c8b97818 CM |
1503 | /* step two, lock all the pages after the page that has start */ |
1504 | ret = lock_delalloc_pages(inode, locked_page, | |
1505 | delalloc_start, delalloc_end); | |
1506 | if (ret == -EAGAIN) { | |
1507 | /* some of the pages are gone, lets avoid looping by | |
1508 | * shortening the size of the delalloc range we're searching | |
1509 | */ | |
9655d298 | 1510 | free_extent_state(cached_state); |
c8b97818 CM |
1511 | if (!loops) { |
1512 | unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1); | |
1513 | max_bytes = PAGE_CACHE_SIZE - offset; | |
1514 | loops = 1; | |
1515 | goto again; | |
1516 | } else { | |
1517 | found = 0; | |
1518 | goto out_failed; | |
1519 | } | |
1520 | } | |
79787eaa | 1521 | BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */ |
c8b97818 CM |
1522 | |
1523 | /* step three, lock the state bits for the whole range */ | |
d0082371 | 1524 | lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state); |
c8b97818 CM |
1525 | |
1526 | /* then test to make sure it is all still delalloc */ | |
1527 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 1528 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 1529 | if (!ret) { |
9655d298 CM |
1530 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
1531 | &cached_state, GFP_NOFS); | |
c8b97818 CM |
1532 | __unlock_for_delalloc(inode, locked_page, |
1533 | delalloc_start, delalloc_end); | |
1534 | cond_resched(); | |
1535 | goto again; | |
1536 | } | |
9655d298 | 1537 | free_extent_state(cached_state); |
c8b97818 CM |
1538 | *start = delalloc_start; |
1539 | *end = delalloc_end; | |
1540 | out_failed: | |
1541 | return found; | |
1542 | } | |
1543 | ||
1544 | int extent_clear_unlock_delalloc(struct inode *inode, | |
1545 | struct extent_io_tree *tree, | |
1546 | u64 start, u64 end, struct page *locked_page, | |
a791e35e | 1547 | unsigned long op) |
c8b97818 CM |
1548 | { |
1549 | int ret; | |
1550 | struct page *pages[16]; | |
1551 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
1552 | unsigned long end_index = end >> PAGE_CACHE_SHIFT; | |
1553 | unsigned long nr_pages = end_index - index + 1; | |
1554 | int i; | |
771ed689 | 1555 | int clear_bits = 0; |
c8b97818 | 1556 | |
a791e35e | 1557 | if (op & EXTENT_CLEAR_UNLOCK) |
771ed689 | 1558 | clear_bits |= EXTENT_LOCKED; |
a791e35e | 1559 | if (op & EXTENT_CLEAR_DIRTY) |
c8b97818 CM |
1560 | clear_bits |= EXTENT_DIRTY; |
1561 | ||
a791e35e | 1562 | if (op & EXTENT_CLEAR_DELALLOC) |
771ed689 CM |
1563 | clear_bits |= EXTENT_DELALLOC; |
1564 | ||
2c64c53d | 1565 | clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS); |
32c00aff JB |
1566 | if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | |
1567 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK | | |
1568 | EXTENT_SET_PRIVATE2))) | |
771ed689 | 1569 | return 0; |
c8b97818 | 1570 | |
d397712b | 1571 | while (nr_pages > 0) { |
c8b97818 | 1572 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
1573 | min_t(unsigned long, |
1574 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 | 1575 | for (i = 0; i < ret; i++) { |
8b62b72b | 1576 | |
a791e35e | 1577 | if (op & EXTENT_SET_PRIVATE2) |
8b62b72b CM |
1578 | SetPagePrivate2(pages[i]); |
1579 | ||
c8b97818 CM |
1580 | if (pages[i] == locked_page) { |
1581 | page_cache_release(pages[i]); | |
1582 | continue; | |
1583 | } | |
a791e35e | 1584 | if (op & EXTENT_CLEAR_DIRTY) |
c8b97818 | 1585 | clear_page_dirty_for_io(pages[i]); |
a791e35e | 1586 | if (op & EXTENT_SET_WRITEBACK) |
c8b97818 | 1587 | set_page_writeback(pages[i]); |
a791e35e | 1588 | if (op & EXTENT_END_WRITEBACK) |
c8b97818 | 1589 | end_page_writeback(pages[i]); |
a791e35e | 1590 | if (op & EXTENT_CLEAR_UNLOCK_PAGE) |
771ed689 | 1591 | unlock_page(pages[i]); |
c8b97818 CM |
1592 | page_cache_release(pages[i]); |
1593 | } | |
1594 | nr_pages -= ret; | |
1595 | index += ret; | |
1596 | cond_resched(); | |
1597 | } | |
1598 | return 0; | |
1599 | } | |
c8b97818 | 1600 | |
d352ac68 CM |
1601 | /* |
1602 | * count the number of bytes in the tree that have a given bit(s) | |
1603 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
1604 | * cached. The total number found is returned. | |
1605 | */ | |
d1310b2e CM |
1606 | u64 count_range_bits(struct extent_io_tree *tree, |
1607 | u64 *start, u64 search_end, u64 max_bytes, | |
ec29ed5b | 1608 | unsigned long bits, int contig) |
d1310b2e CM |
1609 | { |
1610 | struct rb_node *node; | |
1611 | struct extent_state *state; | |
1612 | u64 cur_start = *start; | |
1613 | u64 total_bytes = 0; | |
ec29ed5b | 1614 | u64 last = 0; |
d1310b2e CM |
1615 | int found = 0; |
1616 | ||
1617 | if (search_end <= cur_start) { | |
d1310b2e CM |
1618 | WARN_ON(1); |
1619 | return 0; | |
1620 | } | |
1621 | ||
cad321ad | 1622 | spin_lock(&tree->lock); |
d1310b2e CM |
1623 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
1624 | total_bytes = tree->dirty_bytes; | |
1625 | goto out; | |
1626 | } | |
1627 | /* | |
1628 | * this search will find all the extents that end after | |
1629 | * our range starts. | |
1630 | */ | |
80ea96b1 | 1631 | node = tree_search(tree, cur_start); |
d397712b | 1632 | if (!node) |
d1310b2e | 1633 | goto out; |
d1310b2e | 1634 | |
d397712b | 1635 | while (1) { |
d1310b2e CM |
1636 | state = rb_entry(node, struct extent_state, rb_node); |
1637 | if (state->start > search_end) | |
1638 | break; | |
ec29ed5b CM |
1639 | if (contig && found && state->start > last + 1) |
1640 | break; | |
1641 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
1642 | total_bytes += min(search_end, state->end) + 1 - |
1643 | max(cur_start, state->start); | |
1644 | if (total_bytes >= max_bytes) | |
1645 | break; | |
1646 | if (!found) { | |
af60bed2 | 1647 | *start = max(cur_start, state->start); |
d1310b2e CM |
1648 | found = 1; |
1649 | } | |
ec29ed5b CM |
1650 | last = state->end; |
1651 | } else if (contig && found) { | |
1652 | break; | |
d1310b2e CM |
1653 | } |
1654 | node = rb_next(node); | |
1655 | if (!node) | |
1656 | break; | |
1657 | } | |
1658 | out: | |
cad321ad | 1659 | spin_unlock(&tree->lock); |
d1310b2e CM |
1660 | return total_bytes; |
1661 | } | |
b2950863 | 1662 | |
d352ac68 CM |
1663 | /* |
1664 | * set the private field for a given byte offset in the tree. If there isn't | |
1665 | * an extent_state there already, this does nothing. | |
1666 | */ | |
d1310b2e CM |
1667 | int set_state_private(struct extent_io_tree *tree, u64 start, u64 private) |
1668 | { | |
1669 | struct rb_node *node; | |
1670 | struct extent_state *state; | |
1671 | int ret = 0; | |
1672 | ||
cad321ad | 1673 | spin_lock(&tree->lock); |
d1310b2e CM |
1674 | /* |
1675 | * this search will find all the extents that end after | |
1676 | * our range starts. | |
1677 | */ | |
80ea96b1 | 1678 | node = tree_search(tree, start); |
2b114d1d | 1679 | if (!node) { |
d1310b2e CM |
1680 | ret = -ENOENT; |
1681 | goto out; | |
1682 | } | |
1683 | state = rb_entry(node, struct extent_state, rb_node); | |
1684 | if (state->start != start) { | |
1685 | ret = -ENOENT; | |
1686 | goto out; | |
1687 | } | |
1688 | state->private = private; | |
1689 | out: | |
cad321ad | 1690 | spin_unlock(&tree->lock); |
d1310b2e CM |
1691 | return ret; |
1692 | } | |
1693 | ||
1694 | int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private) | |
1695 | { | |
1696 | struct rb_node *node; | |
1697 | struct extent_state *state; | |
1698 | int ret = 0; | |
1699 | ||
cad321ad | 1700 | spin_lock(&tree->lock); |
d1310b2e CM |
1701 | /* |
1702 | * this search will find all the extents that end after | |
1703 | * our range starts. | |
1704 | */ | |
80ea96b1 | 1705 | node = tree_search(tree, start); |
2b114d1d | 1706 | if (!node) { |
d1310b2e CM |
1707 | ret = -ENOENT; |
1708 | goto out; | |
1709 | } | |
1710 | state = rb_entry(node, struct extent_state, rb_node); | |
1711 | if (state->start != start) { | |
1712 | ret = -ENOENT; | |
1713 | goto out; | |
1714 | } | |
1715 | *private = state->private; | |
1716 | out: | |
cad321ad | 1717 | spin_unlock(&tree->lock); |
d1310b2e CM |
1718 | return ret; |
1719 | } | |
1720 | ||
1721 | /* | |
1722 | * searches a range in the state tree for a given mask. | |
70dec807 | 1723 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
1724 | * has the bits set. Otherwise, 1 is returned if any bit in the |
1725 | * range is found set. | |
1726 | */ | |
1727 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
9655d298 | 1728 | int bits, int filled, struct extent_state *cached) |
d1310b2e CM |
1729 | { |
1730 | struct extent_state *state = NULL; | |
1731 | struct rb_node *node; | |
1732 | int bitset = 0; | |
d1310b2e | 1733 | |
cad321ad | 1734 | spin_lock(&tree->lock); |
df98b6e2 JB |
1735 | if (cached && cached->tree && cached->start <= start && |
1736 | cached->end > start) | |
9655d298 CM |
1737 | node = &cached->rb_node; |
1738 | else | |
1739 | node = tree_search(tree, start); | |
d1310b2e CM |
1740 | while (node && start <= end) { |
1741 | state = rb_entry(node, struct extent_state, rb_node); | |
1742 | ||
1743 | if (filled && state->start > start) { | |
1744 | bitset = 0; | |
1745 | break; | |
1746 | } | |
1747 | ||
1748 | if (state->start > end) | |
1749 | break; | |
1750 | ||
1751 | if (state->state & bits) { | |
1752 | bitset = 1; | |
1753 | if (!filled) | |
1754 | break; | |
1755 | } else if (filled) { | |
1756 | bitset = 0; | |
1757 | break; | |
1758 | } | |
46562cec CM |
1759 | |
1760 | if (state->end == (u64)-1) | |
1761 | break; | |
1762 | ||
d1310b2e CM |
1763 | start = state->end + 1; |
1764 | if (start > end) | |
1765 | break; | |
1766 | node = rb_next(node); | |
1767 | if (!node) { | |
1768 | if (filled) | |
1769 | bitset = 0; | |
1770 | break; | |
1771 | } | |
1772 | } | |
cad321ad | 1773 | spin_unlock(&tree->lock); |
d1310b2e CM |
1774 | return bitset; |
1775 | } | |
d1310b2e CM |
1776 | |
1777 | /* | |
1778 | * helper function to set a given page up to date if all the | |
1779 | * extents in the tree for that page are up to date | |
1780 | */ | |
143bede5 | 1781 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e CM |
1782 | { |
1783 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1784 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
9655d298 | 1785 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 1786 | SetPageUptodate(page); |
d1310b2e CM |
1787 | } |
1788 | ||
1789 | /* | |
1790 | * helper function to unlock a page if all the extents in the tree | |
1791 | * for that page are unlocked | |
1792 | */ | |
143bede5 | 1793 | static void check_page_locked(struct extent_io_tree *tree, struct page *page) |
d1310b2e CM |
1794 | { |
1795 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
1796 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
9655d298 | 1797 | if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) |
d1310b2e | 1798 | unlock_page(page); |
d1310b2e CM |
1799 | } |
1800 | ||
1801 | /* | |
1802 | * helper function to end page writeback if all the extents | |
1803 | * in the tree for that page are done with writeback | |
1804 | */ | |
143bede5 JM |
1805 | static void check_page_writeback(struct extent_io_tree *tree, |
1806 | struct page *page) | |
d1310b2e | 1807 | { |
1edbb734 | 1808 | end_page_writeback(page); |
d1310b2e CM |
1809 | } |
1810 | ||
4a54c8c1 JS |
1811 | /* |
1812 | * When IO fails, either with EIO or csum verification fails, we | |
1813 | * try other mirrors that might have a good copy of the data. This | |
1814 | * io_failure_record is used to record state as we go through all the | |
1815 | * mirrors. If another mirror has good data, the page is set up to date | |
1816 | * and things continue. If a good mirror can't be found, the original | |
1817 | * bio end_io callback is called to indicate things have failed. | |
1818 | */ | |
1819 | struct io_failure_record { | |
1820 | struct page *page; | |
1821 | u64 start; | |
1822 | u64 len; | |
1823 | u64 logical; | |
1824 | unsigned long bio_flags; | |
1825 | int this_mirror; | |
1826 | int failed_mirror; | |
1827 | int in_validation; | |
1828 | }; | |
1829 | ||
1830 | static int free_io_failure(struct inode *inode, struct io_failure_record *rec, | |
1831 | int did_repair) | |
1832 | { | |
1833 | int ret; | |
1834 | int err = 0; | |
1835 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
1836 | ||
1837 | set_state_private(failure_tree, rec->start, 0); | |
1838 | ret = clear_extent_bits(failure_tree, rec->start, | |
1839 | rec->start + rec->len - 1, | |
1840 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1841 | if (ret) | |
1842 | err = ret; | |
1843 | ||
1844 | if (did_repair) { | |
1845 | ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start, | |
1846 | rec->start + rec->len - 1, | |
1847 | EXTENT_DAMAGED, GFP_NOFS); | |
1848 | if (ret && !err) | |
1849 | err = ret; | |
1850 | } | |
1851 | ||
1852 | kfree(rec); | |
1853 | return err; | |
1854 | } | |
1855 | ||
1856 | static void repair_io_failure_callback(struct bio *bio, int err) | |
1857 | { | |
1858 | complete(bio->bi_private); | |
1859 | } | |
1860 | ||
1861 | /* | |
1862 | * this bypasses the standard btrfs submit functions deliberately, as | |
1863 | * the standard behavior is to write all copies in a raid setup. here we only | |
1864 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
1865 | * submit_bio directly. | |
1866 | * to avoid any synchonization issues, wait for the data after writing, which | |
1867 | * actually prevents the read that triggered the error from finishing. | |
1868 | * currently, there can be no more than two copies of every data bit. thus, | |
1869 | * exactly one rewrite is required. | |
1870 | */ | |
1871 | int repair_io_failure(struct btrfs_mapping_tree *map_tree, u64 start, | |
1872 | u64 length, u64 logical, struct page *page, | |
1873 | int mirror_num) | |
1874 | { | |
1875 | struct bio *bio; | |
1876 | struct btrfs_device *dev; | |
1877 | DECLARE_COMPLETION_ONSTACK(compl); | |
1878 | u64 map_length = 0; | |
1879 | u64 sector; | |
1880 | struct btrfs_bio *bbio = NULL; | |
1881 | int ret; | |
1882 | ||
1883 | BUG_ON(!mirror_num); | |
1884 | ||
1885 | bio = bio_alloc(GFP_NOFS, 1); | |
1886 | if (!bio) | |
1887 | return -EIO; | |
1888 | bio->bi_private = &compl; | |
1889 | bio->bi_end_io = repair_io_failure_callback; | |
1890 | bio->bi_size = 0; | |
1891 | map_length = length; | |
1892 | ||
1893 | ret = btrfs_map_block(map_tree, WRITE, logical, | |
1894 | &map_length, &bbio, mirror_num); | |
1895 | if (ret) { | |
1896 | bio_put(bio); | |
1897 | return -EIO; | |
1898 | } | |
1899 | BUG_ON(mirror_num != bbio->mirror_num); | |
1900 | sector = bbio->stripes[mirror_num-1].physical >> 9; | |
1901 | bio->bi_sector = sector; | |
1902 | dev = bbio->stripes[mirror_num-1].dev; | |
1903 | kfree(bbio); | |
1904 | if (!dev || !dev->bdev || !dev->writeable) { | |
1905 | bio_put(bio); | |
1906 | return -EIO; | |
1907 | } | |
1908 | bio->bi_bdev = dev->bdev; | |
1909 | bio_add_page(bio, page, length, start-page_offset(page)); | |
21adbd5c | 1910 | btrfsic_submit_bio(WRITE_SYNC, bio); |
4a54c8c1 JS |
1911 | wait_for_completion(&compl); |
1912 | ||
1913 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) { | |
1914 | /* try to remap that extent elsewhere? */ | |
1915 | bio_put(bio); | |
442a4f63 | 1916 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
1917 | return -EIO; |
1918 | } | |
1919 | ||
1920 | printk(KERN_INFO "btrfs read error corrected: ino %lu off %llu (dev %s " | |
1921 | "sector %llu)\n", page->mapping->host->i_ino, start, | |
1922 | dev->name, sector); | |
1923 | ||
1924 | bio_put(bio); | |
1925 | return 0; | |
1926 | } | |
1927 | ||
ea466794 JB |
1928 | int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb, |
1929 | int mirror_num) | |
1930 | { | |
1931 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
1932 | u64 start = eb->start; | |
1933 | unsigned long i, num_pages = num_extent_pages(eb->start, eb->len); | |
d95603b2 | 1934 | int ret = 0; |
ea466794 JB |
1935 | |
1936 | for (i = 0; i < num_pages; i++) { | |
1937 | struct page *p = extent_buffer_page(eb, i); | |
1938 | ret = repair_io_failure(map_tree, start, PAGE_CACHE_SIZE, | |
1939 | start, p, mirror_num); | |
1940 | if (ret) | |
1941 | break; | |
1942 | start += PAGE_CACHE_SIZE; | |
1943 | } | |
1944 | ||
1945 | return ret; | |
1946 | } | |
1947 | ||
4a54c8c1 JS |
1948 | /* |
1949 | * each time an IO finishes, we do a fast check in the IO failure tree | |
1950 | * to see if we need to process or clean up an io_failure_record | |
1951 | */ | |
1952 | static int clean_io_failure(u64 start, struct page *page) | |
1953 | { | |
1954 | u64 private; | |
1955 | u64 private_failure; | |
1956 | struct io_failure_record *failrec; | |
1957 | struct btrfs_mapping_tree *map_tree; | |
1958 | struct extent_state *state; | |
1959 | int num_copies; | |
1960 | int did_repair = 0; | |
1961 | int ret; | |
1962 | struct inode *inode = page->mapping->host; | |
1963 | ||
1964 | private = 0; | |
1965 | ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
1966 | (u64)-1, 1, EXTENT_DIRTY, 0); | |
1967 | if (!ret) | |
1968 | return 0; | |
1969 | ||
1970 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start, | |
1971 | &private_failure); | |
1972 | if (ret) | |
1973 | return 0; | |
1974 | ||
1975 | failrec = (struct io_failure_record *)(unsigned long) private_failure; | |
1976 | BUG_ON(!failrec->this_mirror); | |
1977 | ||
1978 | if (failrec->in_validation) { | |
1979 | /* there was no real error, just free the record */ | |
1980 | pr_debug("clean_io_failure: freeing dummy error at %llu\n", | |
1981 | failrec->start); | |
1982 | did_repair = 1; | |
1983 | goto out; | |
1984 | } | |
1985 | ||
1986 | spin_lock(&BTRFS_I(inode)->io_tree.lock); | |
1987 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, | |
1988 | failrec->start, | |
1989 | EXTENT_LOCKED); | |
1990 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); | |
1991 | ||
1992 | if (state && state->start == failrec->start) { | |
1993 | map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree; | |
1994 | num_copies = btrfs_num_copies(map_tree, failrec->logical, | |
1995 | failrec->len); | |
1996 | if (num_copies > 1) { | |
1997 | ret = repair_io_failure(map_tree, start, failrec->len, | |
1998 | failrec->logical, page, | |
1999 | failrec->failed_mirror); | |
2000 | did_repair = !ret; | |
2001 | } | |
2002 | } | |
2003 | ||
2004 | out: | |
2005 | if (!ret) | |
2006 | ret = free_io_failure(inode, failrec, did_repair); | |
2007 | ||
2008 | return ret; | |
2009 | } | |
2010 | ||
2011 | /* | |
2012 | * this is a generic handler for readpage errors (default | |
2013 | * readpage_io_failed_hook). if other copies exist, read those and write back | |
2014 | * good data to the failed position. does not investigate in remapping the | |
2015 | * failed extent elsewhere, hoping the device will be smart enough to do this as | |
2016 | * needed | |
2017 | */ | |
2018 | ||
2019 | static int bio_readpage_error(struct bio *failed_bio, struct page *page, | |
2020 | u64 start, u64 end, int failed_mirror, | |
2021 | struct extent_state *state) | |
2022 | { | |
2023 | struct io_failure_record *failrec = NULL; | |
2024 | u64 private; | |
2025 | struct extent_map *em; | |
2026 | struct inode *inode = page->mapping->host; | |
2027 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
2028 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2029 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
2030 | struct bio *bio; | |
2031 | int num_copies; | |
2032 | int ret; | |
2033 | int read_mode; | |
2034 | u64 logical; | |
2035 | ||
2036 | BUG_ON(failed_bio->bi_rw & REQ_WRITE); | |
2037 | ||
2038 | ret = get_state_private(failure_tree, start, &private); | |
2039 | if (ret) { | |
2040 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); | |
2041 | if (!failrec) | |
2042 | return -ENOMEM; | |
2043 | failrec->start = start; | |
2044 | failrec->len = end - start + 1; | |
2045 | failrec->this_mirror = 0; | |
2046 | failrec->bio_flags = 0; | |
2047 | failrec->in_validation = 0; | |
2048 | ||
2049 | read_lock(&em_tree->lock); | |
2050 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2051 | if (!em) { | |
2052 | read_unlock(&em_tree->lock); | |
2053 | kfree(failrec); | |
2054 | return -EIO; | |
2055 | } | |
2056 | ||
2057 | if (em->start > start || em->start + em->len < start) { | |
2058 | free_extent_map(em); | |
2059 | em = NULL; | |
2060 | } | |
2061 | read_unlock(&em_tree->lock); | |
2062 | ||
2063 | if (!em || IS_ERR(em)) { | |
2064 | kfree(failrec); | |
2065 | return -EIO; | |
2066 | } | |
2067 | logical = start - em->start; | |
2068 | logical = em->block_start + logical; | |
2069 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2070 | logical = em->block_start; | |
2071 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2072 | extent_set_compress_type(&failrec->bio_flags, | |
2073 | em->compress_type); | |
2074 | } | |
2075 | pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, " | |
2076 | "len=%llu\n", logical, start, failrec->len); | |
2077 | failrec->logical = logical; | |
2078 | free_extent_map(em); | |
2079 | ||
2080 | /* set the bits in the private failure tree */ | |
2081 | ret = set_extent_bits(failure_tree, start, end, | |
2082 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
2083 | if (ret >= 0) | |
2084 | ret = set_state_private(failure_tree, start, | |
2085 | (u64)(unsigned long)failrec); | |
2086 | /* set the bits in the inode's tree */ | |
2087 | if (ret >= 0) | |
2088 | ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED, | |
2089 | GFP_NOFS); | |
2090 | if (ret < 0) { | |
2091 | kfree(failrec); | |
2092 | return ret; | |
2093 | } | |
2094 | } else { | |
2095 | failrec = (struct io_failure_record *)(unsigned long)private; | |
2096 | pr_debug("bio_readpage_error: (found) logical=%llu, " | |
2097 | "start=%llu, len=%llu, validation=%d\n", | |
2098 | failrec->logical, failrec->start, failrec->len, | |
2099 | failrec->in_validation); | |
2100 | /* | |
2101 | * when data can be on disk more than twice, add to failrec here | |
2102 | * (e.g. with a list for failed_mirror) to make | |
2103 | * clean_io_failure() clean all those errors at once. | |
2104 | */ | |
2105 | } | |
2106 | num_copies = btrfs_num_copies( | |
2107 | &BTRFS_I(inode)->root->fs_info->mapping_tree, | |
2108 | failrec->logical, failrec->len); | |
2109 | if (num_copies == 1) { | |
2110 | /* | |
2111 | * we only have a single copy of the data, so don't bother with | |
2112 | * all the retry and error correction code that follows. no | |
2113 | * matter what the error is, it is very likely to persist. | |
2114 | */ | |
2115 | pr_debug("bio_readpage_error: cannot repair, num_copies == 1. " | |
2116 | "state=%p, num_copies=%d, next_mirror %d, " | |
2117 | "failed_mirror %d\n", state, num_copies, | |
2118 | failrec->this_mirror, failed_mirror); | |
2119 | free_io_failure(inode, failrec, 0); | |
2120 | return -EIO; | |
2121 | } | |
2122 | ||
2123 | if (!state) { | |
2124 | spin_lock(&tree->lock); | |
2125 | state = find_first_extent_bit_state(tree, failrec->start, | |
2126 | EXTENT_LOCKED); | |
2127 | if (state && state->start != failrec->start) | |
2128 | state = NULL; | |
2129 | spin_unlock(&tree->lock); | |
2130 | } | |
2131 | ||
2132 | /* | |
2133 | * there are two premises: | |
2134 | * a) deliver good data to the caller | |
2135 | * b) correct the bad sectors on disk | |
2136 | */ | |
2137 | if (failed_bio->bi_vcnt > 1) { | |
2138 | /* | |
2139 | * to fulfill b), we need to know the exact failing sectors, as | |
2140 | * we don't want to rewrite any more than the failed ones. thus, | |
2141 | * we need separate read requests for the failed bio | |
2142 | * | |
2143 | * if the following BUG_ON triggers, our validation request got | |
2144 | * merged. we need separate requests for our algorithm to work. | |
2145 | */ | |
2146 | BUG_ON(failrec->in_validation); | |
2147 | failrec->in_validation = 1; | |
2148 | failrec->this_mirror = failed_mirror; | |
2149 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; | |
2150 | } else { | |
2151 | /* | |
2152 | * we're ready to fulfill a) and b) alongside. get a good copy | |
2153 | * of the failed sector and if we succeed, we have setup | |
2154 | * everything for repair_io_failure to do the rest for us. | |
2155 | */ | |
2156 | if (failrec->in_validation) { | |
2157 | BUG_ON(failrec->this_mirror != failed_mirror); | |
2158 | failrec->in_validation = 0; | |
2159 | failrec->this_mirror = 0; | |
2160 | } | |
2161 | failrec->failed_mirror = failed_mirror; | |
2162 | failrec->this_mirror++; | |
2163 | if (failrec->this_mirror == failed_mirror) | |
2164 | failrec->this_mirror++; | |
2165 | read_mode = READ_SYNC; | |
2166 | } | |
2167 | ||
2168 | if (!state || failrec->this_mirror > num_copies) { | |
2169 | pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, " | |
2170 | "next_mirror %d, failed_mirror %d\n", state, | |
2171 | num_copies, failrec->this_mirror, failed_mirror); | |
2172 | free_io_failure(inode, failrec, 0); | |
2173 | return -EIO; | |
2174 | } | |
2175 | ||
2176 | bio = bio_alloc(GFP_NOFS, 1); | |
e627ee7b TI |
2177 | if (!bio) { |
2178 | free_io_failure(inode, failrec, 0); | |
2179 | return -EIO; | |
2180 | } | |
4a54c8c1 JS |
2181 | bio->bi_private = state; |
2182 | bio->bi_end_io = failed_bio->bi_end_io; | |
2183 | bio->bi_sector = failrec->logical >> 9; | |
2184 | bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; | |
2185 | bio->bi_size = 0; | |
2186 | ||
2187 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); | |
2188 | ||
2189 | pr_debug("bio_readpage_error: submitting new read[%#x] to " | |
2190 | "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode, | |
2191 | failrec->this_mirror, num_copies, failrec->in_validation); | |
2192 | ||
013bd4c3 TI |
2193 | ret = tree->ops->submit_bio_hook(inode, read_mode, bio, |
2194 | failrec->this_mirror, | |
2195 | failrec->bio_flags, 0); | |
2196 | return ret; | |
4a54c8c1 JS |
2197 | } |
2198 | ||
d1310b2e CM |
2199 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2200 | ||
87826df0 JM |
2201 | int end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
2202 | { | |
2203 | int uptodate = (err == 0); | |
2204 | struct extent_io_tree *tree; | |
2205 | int ret; | |
2206 | ||
2207 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
2208 | ||
2209 | if (tree->ops && tree->ops->writepage_end_io_hook) { | |
2210 | ret = tree->ops->writepage_end_io_hook(page, start, | |
2211 | end, NULL, uptodate); | |
2212 | if (ret) | |
2213 | uptodate = 0; | |
2214 | } | |
2215 | ||
87826df0 | 2216 | if (!uptodate) { |
87826df0 JM |
2217 | ClearPageUptodate(page); |
2218 | SetPageError(page); | |
2219 | } | |
2220 | return 0; | |
2221 | } | |
2222 | ||
d1310b2e CM |
2223 | /* |
2224 | * after a writepage IO is done, we need to: | |
2225 | * clear the uptodate bits on error | |
2226 | * clear the writeback bits in the extent tree for this IO | |
2227 | * end_page_writeback if the page has no more pending IO | |
2228 | * | |
2229 | * Scheduling is not allowed, so the extent state tree is expected | |
2230 | * to have one and only one object corresponding to this IO. | |
2231 | */ | |
d1310b2e | 2232 | static void end_bio_extent_writepage(struct bio *bio, int err) |
d1310b2e | 2233 | { |
d1310b2e | 2234 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
902b22f3 | 2235 | struct extent_io_tree *tree; |
d1310b2e CM |
2236 | u64 start; |
2237 | u64 end; | |
2238 | int whole_page; | |
2239 | ||
d1310b2e CM |
2240 | do { |
2241 | struct page *page = bvec->bv_page; | |
902b22f3 DW |
2242 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
2243 | ||
d1310b2e CM |
2244 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
2245 | bvec->bv_offset; | |
2246 | end = start + bvec->bv_len - 1; | |
2247 | ||
2248 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
2249 | whole_page = 1; | |
2250 | else | |
2251 | whole_page = 0; | |
2252 | ||
2253 | if (--bvec >= bio->bi_io_vec) | |
2254 | prefetchw(&bvec->bv_page->flags); | |
1259ab75 | 2255 | |
87826df0 JM |
2256 | if (end_extent_writepage(page, err, start, end)) |
2257 | continue; | |
70dec807 | 2258 | |
d1310b2e CM |
2259 | if (whole_page) |
2260 | end_page_writeback(page); | |
2261 | else | |
2262 | check_page_writeback(tree, page); | |
d1310b2e | 2263 | } while (bvec >= bio->bi_io_vec); |
2b1f55b0 | 2264 | |
d1310b2e | 2265 | bio_put(bio); |
d1310b2e CM |
2266 | } |
2267 | ||
2268 | /* | |
2269 | * after a readpage IO is done, we need to: | |
2270 | * clear the uptodate bits on error | |
2271 | * set the uptodate bits if things worked | |
2272 | * set the page up to date if all extents in the tree are uptodate | |
2273 | * clear the lock bit in the extent tree | |
2274 | * unlock the page if there are no other extents locked for it | |
2275 | * | |
2276 | * Scheduling is not allowed, so the extent state tree is expected | |
2277 | * to have one and only one object corresponding to this IO. | |
2278 | */ | |
d1310b2e | 2279 | static void end_bio_extent_readpage(struct bio *bio, int err) |
d1310b2e CM |
2280 | { |
2281 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
4125bf76 CM |
2282 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
2283 | struct bio_vec *bvec = bio->bi_io_vec; | |
902b22f3 | 2284 | struct extent_io_tree *tree; |
d1310b2e CM |
2285 | u64 start; |
2286 | u64 end; | |
2287 | int whole_page; | |
5cf1ab56 | 2288 | int mirror; |
d1310b2e CM |
2289 | int ret; |
2290 | ||
d20f7043 CM |
2291 | if (err) |
2292 | uptodate = 0; | |
2293 | ||
d1310b2e CM |
2294 | do { |
2295 | struct page *page = bvec->bv_page; | |
507903b8 AJ |
2296 | struct extent_state *cached = NULL; |
2297 | struct extent_state *state; | |
2298 | ||
4a54c8c1 JS |
2299 | pr_debug("end_bio_extent_readpage: bi_vcnt=%d, idx=%d, err=%d, " |
2300 | "mirror=%ld\n", bio->bi_vcnt, bio->bi_idx, err, | |
2301 | (long int)bio->bi_bdev); | |
902b22f3 DW |
2302 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
2303 | ||
d1310b2e CM |
2304 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + |
2305 | bvec->bv_offset; | |
2306 | end = start + bvec->bv_len - 1; | |
2307 | ||
2308 | if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) | |
2309 | whole_page = 1; | |
2310 | else | |
2311 | whole_page = 0; | |
2312 | ||
4125bf76 | 2313 | if (++bvec <= bvec_end) |
d1310b2e CM |
2314 | prefetchw(&bvec->bv_page->flags); |
2315 | ||
507903b8 | 2316 | spin_lock(&tree->lock); |
0d399205 | 2317 | state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED); |
109b36a2 | 2318 | if (state && state->start == start) { |
507903b8 AJ |
2319 | /* |
2320 | * take a reference on the state, unlock will drop | |
2321 | * the ref | |
2322 | */ | |
2323 | cache_state(state, &cached); | |
2324 | } | |
2325 | spin_unlock(&tree->lock); | |
2326 | ||
5cf1ab56 | 2327 | mirror = (int)(unsigned long)bio->bi_bdev; |
d1310b2e | 2328 | if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) { |
70dec807 | 2329 | ret = tree->ops->readpage_end_io_hook(page, start, end, |
5cf1ab56 | 2330 | state, mirror); |
442a4f63 SB |
2331 | if (ret) { |
2332 | /* no IO indicated but software detected errors | |
2333 | * in the block, either checksum errors or | |
2334 | * issues with the contents */ | |
2335 | struct btrfs_root *root = | |
2336 | BTRFS_I(page->mapping->host)->root; | |
2337 | struct btrfs_device *device; | |
2338 | ||
d1310b2e | 2339 | uptodate = 0; |
442a4f63 SB |
2340 | device = btrfs_find_device_for_logical( |
2341 | root, start, mirror); | |
2342 | if (device) | |
2343 | btrfs_dev_stat_inc_and_print(device, | |
2344 | BTRFS_DEV_STAT_CORRUPTION_ERRS); | |
2345 | } else { | |
4a54c8c1 | 2346 | clean_io_failure(start, page); |
442a4f63 | 2347 | } |
d1310b2e | 2348 | } |
ea466794 | 2349 | |
ea466794 | 2350 | if (!uptodate && tree->ops && tree->ops->readpage_io_failed_hook) { |
5cf1ab56 | 2351 | ret = tree->ops->readpage_io_failed_hook(page, mirror); |
ea466794 JB |
2352 | if (!ret && !err && |
2353 | test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
2354 | uptodate = 1; | |
2355 | } else if (!uptodate) { | |
f4a8e656 JS |
2356 | /* |
2357 | * The generic bio_readpage_error handles errors the | |
2358 | * following way: If possible, new read requests are | |
2359 | * created and submitted and will end up in | |
2360 | * end_bio_extent_readpage as well (if we're lucky, not | |
2361 | * in the !uptodate case). In that case it returns 0 and | |
2362 | * we just go on with the next page in our bio. If it | |
2363 | * can't handle the error it will return -EIO and we | |
2364 | * remain responsible for that page. | |
2365 | */ | |
5cf1ab56 | 2366 | ret = bio_readpage_error(bio, page, start, end, mirror, NULL); |
7e38326f | 2367 | if (ret == 0) { |
3b951516 CM |
2368 | uptodate = |
2369 | test_bit(BIO_UPTODATE, &bio->bi_flags); | |
d20f7043 CM |
2370 | if (err) |
2371 | uptodate = 0; | |
507903b8 | 2372 | uncache_state(&cached); |
7e38326f CM |
2373 | continue; |
2374 | } | |
2375 | } | |
d1310b2e | 2376 | |
0b32f4bb | 2377 | if (uptodate && tree->track_uptodate) { |
507903b8 | 2378 | set_extent_uptodate(tree, start, end, &cached, |
902b22f3 | 2379 | GFP_ATOMIC); |
771ed689 | 2380 | } |
507903b8 | 2381 | unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC); |
d1310b2e | 2382 | |
70dec807 CM |
2383 | if (whole_page) { |
2384 | if (uptodate) { | |
2385 | SetPageUptodate(page); | |
2386 | } else { | |
2387 | ClearPageUptodate(page); | |
2388 | SetPageError(page); | |
2389 | } | |
d1310b2e | 2390 | unlock_page(page); |
70dec807 CM |
2391 | } else { |
2392 | if (uptodate) { | |
2393 | check_page_uptodate(tree, page); | |
2394 | } else { | |
2395 | ClearPageUptodate(page); | |
2396 | SetPageError(page); | |
2397 | } | |
d1310b2e | 2398 | check_page_locked(tree, page); |
70dec807 | 2399 | } |
4125bf76 | 2400 | } while (bvec <= bvec_end); |
d1310b2e CM |
2401 | |
2402 | bio_put(bio); | |
d1310b2e CM |
2403 | } |
2404 | ||
88f794ed MX |
2405 | struct bio * |
2406 | btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, | |
2407 | gfp_t gfp_flags) | |
d1310b2e CM |
2408 | { |
2409 | struct bio *bio; | |
2410 | ||
2411 | bio = bio_alloc(gfp_flags, nr_vecs); | |
2412 | ||
2413 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
2414 | while (!bio && (nr_vecs /= 2)) | |
2415 | bio = bio_alloc(gfp_flags, nr_vecs); | |
2416 | } | |
2417 | ||
2418 | if (bio) { | |
e1c4b745 | 2419 | bio->bi_size = 0; |
d1310b2e CM |
2420 | bio->bi_bdev = bdev; |
2421 | bio->bi_sector = first_sector; | |
2422 | } | |
2423 | return bio; | |
2424 | } | |
2425 | ||
79787eaa JM |
2426 | /* |
2427 | * Since writes are async, they will only return -ENOMEM. | |
2428 | * Reads can return the full range of I/O error conditions. | |
2429 | */ | |
355808c2 JM |
2430 | static int __must_check submit_one_bio(int rw, struct bio *bio, |
2431 | int mirror_num, unsigned long bio_flags) | |
d1310b2e | 2432 | { |
d1310b2e | 2433 | int ret = 0; |
70dec807 CM |
2434 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; |
2435 | struct page *page = bvec->bv_page; | |
2436 | struct extent_io_tree *tree = bio->bi_private; | |
70dec807 | 2437 | u64 start; |
70dec807 CM |
2438 | |
2439 | start = ((u64)page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset; | |
70dec807 | 2440 | |
902b22f3 | 2441 | bio->bi_private = NULL; |
d1310b2e CM |
2442 | |
2443 | bio_get(bio); | |
2444 | ||
065631f6 | 2445 | if (tree->ops && tree->ops->submit_bio_hook) |
6b82ce8d | 2446 | ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio, |
eaf25d93 | 2447 | mirror_num, bio_flags, start); |
0b86a832 | 2448 | else |
21adbd5c | 2449 | btrfsic_submit_bio(rw, bio); |
4a54c8c1 | 2450 | |
d1310b2e CM |
2451 | if (bio_flagged(bio, BIO_EOPNOTSUPP)) |
2452 | ret = -EOPNOTSUPP; | |
2453 | bio_put(bio); | |
2454 | return ret; | |
2455 | } | |
2456 | ||
3444a972 JM |
2457 | static int merge_bio(struct extent_io_tree *tree, struct page *page, |
2458 | unsigned long offset, size_t size, struct bio *bio, | |
2459 | unsigned long bio_flags) | |
2460 | { | |
2461 | int ret = 0; | |
2462 | if (tree->ops && tree->ops->merge_bio_hook) | |
2463 | ret = tree->ops->merge_bio_hook(page, offset, size, bio, | |
2464 | bio_flags); | |
2465 | BUG_ON(ret < 0); | |
2466 | return ret; | |
2467 | ||
2468 | } | |
2469 | ||
d1310b2e CM |
2470 | static int submit_extent_page(int rw, struct extent_io_tree *tree, |
2471 | struct page *page, sector_t sector, | |
2472 | size_t size, unsigned long offset, | |
2473 | struct block_device *bdev, | |
2474 | struct bio **bio_ret, | |
2475 | unsigned long max_pages, | |
f188591e | 2476 | bio_end_io_t end_io_func, |
c8b97818 CM |
2477 | int mirror_num, |
2478 | unsigned long prev_bio_flags, | |
2479 | unsigned long bio_flags) | |
d1310b2e CM |
2480 | { |
2481 | int ret = 0; | |
2482 | struct bio *bio; | |
2483 | int nr; | |
c8b97818 CM |
2484 | int contig = 0; |
2485 | int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED; | |
2486 | int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED; | |
5b050f04 | 2487 | size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE); |
d1310b2e CM |
2488 | |
2489 | if (bio_ret && *bio_ret) { | |
2490 | bio = *bio_ret; | |
c8b97818 CM |
2491 | if (old_compressed) |
2492 | contig = bio->bi_sector == sector; | |
2493 | else | |
2494 | contig = bio->bi_sector + (bio->bi_size >> 9) == | |
2495 | sector; | |
2496 | ||
2497 | if (prev_bio_flags != bio_flags || !contig || | |
3444a972 | 2498 | merge_bio(tree, page, offset, page_size, bio, bio_flags) || |
c8b97818 CM |
2499 | bio_add_page(bio, page, page_size, offset) < page_size) { |
2500 | ret = submit_one_bio(rw, bio, mirror_num, | |
2501 | prev_bio_flags); | |
79787eaa JM |
2502 | if (ret < 0) |
2503 | return ret; | |
d1310b2e CM |
2504 | bio = NULL; |
2505 | } else { | |
2506 | return 0; | |
2507 | } | |
2508 | } | |
c8b97818 CM |
2509 | if (this_compressed) |
2510 | nr = BIO_MAX_PAGES; | |
2511 | else | |
2512 | nr = bio_get_nr_vecs(bdev); | |
2513 | ||
88f794ed | 2514 | bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH); |
5df67083 TI |
2515 | if (!bio) |
2516 | return -ENOMEM; | |
70dec807 | 2517 | |
c8b97818 | 2518 | bio_add_page(bio, page, page_size, offset); |
d1310b2e CM |
2519 | bio->bi_end_io = end_io_func; |
2520 | bio->bi_private = tree; | |
70dec807 | 2521 | |
d397712b | 2522 | if (bio_ret) |
d1310b2e | 2523 | *bio_ret = bio; |
d397712b | 2524 | else |
c8b97818 | 2525 | ret = submit_one_bio(rw, bio, mirror_num, bio_flags); |
d1310b2e CM |
2526 | |
2527 | return ret; | |
2528 | } | |
2529 | ||
4f2de97a | 2530 | void attach_extent_buffer_page(struct extent_buffer *eb, struct page *page) |
d1310b2e CM |
2531 | { |
2532 | if (!PagePrivate(page)) { | |
2533 | SetPagePrivate(page); | |
d1310b2e | 2534 | page_cache_get(page); |
4f2de97a JB |
2535 | set_page_private(page, (unsigned long)eb); |
2536 | } else { | |
2537 | WARN_ON(page->private != (unsigned long)eb); | |
d1310b2e CM |
2538 | } |
2539 | } | |
2540 | ||
4f2de97a | 2541 | void set_page_extent_mapped(struct page *page) |
d1310b2e | 2542 | { |
4f2de97a JB |
2543 | if (!PagePrivate(page)) { |
2544 | SetPagePrivate(page); | |
2545 | page_cache_get(page); | |
2546 | set_page_private(page, EXTENT_PAGE_PRIVATE); | |
2547 | } | |
d1310b2e CM |
2548 | } |
2549 | ||
2550 | /* | |
2551 | * basic readpage implementation. Locked extent state structs are inserted | |
2552 | * into the tree that are removed when the IO is done (by the end_io | |
2553 | * handlers) | |
79787eaa | 2554 | * XXX JDM: This needs looking at to ensure proper page locking |
d1310b2e CM |
2555 | */ |
2556 | static int __extent_read_full_page(struct extent_io_tree *tree, | |
2557 | struct page *page, | |
2558 | get_extent_t *get_extent, | |
c8b97818 CM |
2559 | struct bio **bio, int mirror_num, |
2560 | unsigned long *bio_flags) | |
d1310b2e CM |
2561 | { |
2562 | struct inode *inode = page->mapping->host; | |
2563 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2564 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2565 | u64 end; | |
2566 | u64 cur = start; | |
2567 | u64 extent_offset; | |
2568 | u64 last_byte = i_size_read(inode); | |
2569 | u64 block_start; | |
2570 | u64 cur_end; | |
2571 | sector_t sector; | |
2572 | struct extent_map *em; | |
2573 | struct block_device *bdev; | |
11c65dcc | 2574 | struct btrfs_ordered_extent *ordered; |
d1310b2e CM |
2575 | int ret; |
2576 | int nr = 0; | |
306e16ce | 2577 | size_t pg_offset = 0; |
d1310b2e | 2578 | size_t iosize; |
c8b97818 | 2579 | size_t disk_io_size; |
d1310b2e | 2580 | size_t blocksize = inode->i_sb->s_blocksize; |
c8b97818 | 2581 | unsigned long this_bio_flag = 0; |
d1310b2e CM |
2582 | |
2583 | set_page_extent_mapped(page); | |
2584 | ||
90a887c9 DM |
2585 | if (!PageUptodate(page)) { |
2586 | if (cleancache_get_page(page) == 0) { | |
2587 | BUG_ON(blocksize != PAGE_SIZE); | |
2588 | goto out; | |
2589 | } | |
2590 | } | |
2591 | ||
d1310b2e | 2592 | end = page_end; |
11c65dcc | 2593 | while (1) { |
d0082371 | 2594 | lock_extent(tree, start, end); |
11c65dcc JB |
2595 | ordered = btrfs_lookup_ordered_extent(inode, start); |
2596 | if (!ordered) | |
2597 | break; | |
d0082371 | 2598 | unlock_extent(tree, start, end); |
11c65dcc JB |
2599 | btrfs_start_ordered_extent(inode, ordered, 1); |
2600 | btrfs_put_ordered_extent(ordered); | |
2601 | } | |
d1310b2e | 2602 | |
c8b97818 CM |
2603 | if (page->index == last_byte >> PAGE_CACHE_SHIFT) { |
2604 | char *userpage; | |
2605 | size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1); | |
2606 | ||
2607 | if (zero_offset) { | |
2608 | iosize = PAGE_CACHE_SIZE - zero_offset; | |
7ac687d9 | 2609 | userpage = kmap_atomic(page); |
c8b97818 CM |
2610 | memset(userpage + zero_offset, 0, iosize); |
2611 | flush_dcache_page(page); | |
7ac687d9 | 2612 | kunmap_atomic(userpage); |
c8b97818 CM |
2613 | } |
2614 | } | |
d1310b2e CM |
2615 | while (cur <= end) { |
2616 | if (cur >= last_byte) { | |
2617 | char *userpage; | |
507903b8 AJ |
2618 | struct extent_state *cached = NULL; |
2619 | ||
306e16ce | 2620 | iosize = PAGE_CACHE_SIZE - pg_offset; |
7ac687d9 | 2621 | userpage = kmap_atomic(page); |
306e16ce | 2622 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2623 | flush_dcache_page(page); |
7ac687d9 | 2624 | kunmap_atomic(userpage); |
d1310b2e | 2625 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 AJ |
2626 | &cached, GFP_NOFS); |
2627 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
2628 | &cached, GFP_NOFS); | |
d1310b2e CM |
2629 | break; |
2630 | } | |
306e16ce | 2631 | em = get_extent(inode, page, pg_offset, cur, |
d1310b2e | 2632 | end - cur + 1, 0); |
c704005d | 2633 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 2634 | SetPageError(page); |
d0082371 | 2635 | unlock_extent(tree, cur, end); |
d1310b2e CM |
2636 | break; |
2637 | } | |
d1310b2e CM |
2638 | extent_offset = cur - em->start; |
2639 | BUG_ON(extent_map_end(em) <= cur); | |
2640 | BUG_ON(end < cur); | |
2641 | ||
261507a0 | 2642 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
c8b97818 | 2643 | this_bio_flag = EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
2644 | extent_set_compress_type(&this_bio_flag, |
2645 | em->compress_type); | |
2646 | } | |
c8b97818 | 2647 | |
d1310b2e CM |
2648 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
2649 | cur_end = min(extent_map_end(em) - 1, end); | |
2650 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
c8b97818 CM |
2651 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) { |
2652 | disk_io_size = em->block_len; | |
2653 | sector = em->block_start >> 9; | |
2654 | } else { | |
2655 | sector = (em->block_start + extent_offset) >> 9; | |
2656 | disk_io_size = iosize; | |
2657 | } | |
d1310b2e CM |
2658 | bdev = em->bdev; |
2659 | block_start = em->block_start; | |
d899e052 YZ |
2660 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
2661 | block_start = EXTENT_MAP_HOLE; | |
d1310b2e CM |
2662 | free_extent_map(em); |
2663 | em = NULL; | |
2664 | ||
2665 | /* we've found a hole, just zero and go on */ | |
2666 | if (block_start == EXTENT_MAP_HOLE) { | |
2667 | char *userpage; | |
507903b8 AJ |
2668 | struct extent_state *cached = NULL; |
2669 | ||
7ac687d9 | 2670 | userpage = kmap_atomic(page); |
306e16ce | 2671 | memset(userpage + pg_offset, 0, iosize); |
d1310b2e | 2672 | flush_dcache_page(page); |
7ac687d9 | 2673 | kunmap_atomic(userpage); |
d1310b2e CM |
2674 | |
2675 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 AJ |
2676 | &cached, GFP_NOFS); |
2677 | unlock_extent_cached(tree, cur, cur + iosize - 1, | |
2678 | &cached, GFP_NOFS); | |
d1310b2e | 2679 | cur = cur + iosize; |
306e16ce | 2680 | pg_offset += iosize; |
d1310b2e CM |
2681 | continue; |
2682 | } | |
2683 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
2684 | if (test_range_bit(tree, cur, cur_end, |
2685 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 2686 | check_page_uptodate(tree, page); |
d0082371 | 2687 | unlock_extent(tree, cur, cur + iosize - 1); |
d1310b2e | 2688 | cur = cur + iosize; |
306e16ce | 2689 | pg_offset += iosize; |
d1310b2e CM |
2690 | continue; |
2691 | } | |
70dec807 CM |
2692 | /* we have an inline extent but it didn't get marked up |
2693 | * to date. Error out | |
2694 | */ | |
2695 | if (block_start == EXTENT_MAP_INLINE) { | |
2696 | SetPageError(page); | |
d0082371 | 2697 | unlock_extent(tree, cur, cur + iosize - 1); |
70dec807 | 2698 | cur = cur + iosize; |
306e16ce | 2699 | pg_offset += iosize; |
70dec807 CM |
2700 | continue; |
2701 | } | |
d1310b2e CM |
2702 | |
2703 | ret = 0; | |
2704 | if (tree->ops && tree->ops->readpage_io_hook) { | |
2705 | ret = tree->ops->readpage_io_hook(page, cur, | |
2706 | cur + iosize - 1); | |
2707 | } | |
2708 | if (!ret) { | |
89642229 CM |
2709 | unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1; |
2710 | pnr -= page->index; | |
d1310b2e | 2711 | ret = submit_extent_page(READ, tree, page, |
306e16ce | 2712 | sector, disk_io_size, pg_offset, |
89642229 | 2713 | bdev, bio, pnr, |
c8b97818 CM |
2714 | end_bio_extent_readpage, mirror_num, |
2715 | *bio_flags, | |
2716 | this_bio_flag); | |
79787eaa | 2717 | BUG_ON(ret == -ENOMEM); |
89642229 | 2718 | nr++; |
c8b97818 | 2719 | *bio_flags = this_bio_flag; |
d1310b2e CM |
2720 | } |
2721 | if (ret) | |
2722 | SetPageError(page); | |
2723 | cur = cur + iosize; | |
306e16ce | 2724 | pg_offset += iosize; |
d1310b2e | 2725 | } |
90a887c9 | 2726 | out: |
d1310b2e CM |
2727 | if (!nr) { |
2728 | if (!PageError(page)) | |
2729 | SetPageUptodate(page); | |
2730 | unlock_page(page); | |
2731 | } | |
2732 | return 0; | |
2733 | } | |
2734 | ||
2735 | int extent_read_full_page(struct extent_io_tree *tree, struct page *page, | |
8ddc7d9c | 2736 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
2737 | { |
2738 | struct bio *bio = NULL; | |
c8b97818 | 2739 | unsigned long bio_flags = 0; |
d1310b2e CM |
2740 | int ret; |
2741 | ||
8ddc7d9c | 2742 | ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num, |
c8b97818 | 2743 | &bio_flags); |
d1310b2e | 2744 | if (bio) |
8ddc7d9c | 2745 | ret = submit_one_bio(READ, bio, mirror_num, bio_flags); |
d1310b2e CM |
2746 | return ret; |
2747 | } | |
d1310b2e | 2748 | |
11c8349b CM |
2749 | static noinline void update_nr_written(struct page *page, |
2750 | struct writeback_control *wbc, | |
2751 | unsigned long nr_written) | |
2752 | { | |
2753 | wbc->nr_to_write -= nr_written; | |
2754 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && | |
2755 | wbc->range_start == 0 && wbc->range_end == LLONG_MAX)) | |
2756 | page->mapping->writeback_index = page->index + nr_written; | |
2757 | } | |
2758 | ||
d1310b2e CM |
2759 | /* |
2760 | * the writepage semantics are similar to regular writepage. extent | |
2761 | * records are inserted to lock ranges in the tree, and as dirty areas | |
2762 | * are found, they are marked writeback. Then the lock bits are removed | |
2763 | * and the end_io handler clears the writeback ranges | |
2764 | */ | |
2765 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
2766 | void *data) | |
2767 | { | |
2768 | struct inode *inode = page->mapping->host; | |
2769 | struct extent_page_data *epd = data; | |
2770 | struct extent_io_tree *tree = epd->tree; | |
2771 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2772 | u64 delalloc_start; | |
2773 | u64 page_end = start + PAGE_CACHE_SIZE - 1; | |
2774 | u64 end; | |
2775 | u64 cur = start; | |
2776 | u64 extent_offset; | |
2777 | u64 last_byte = i_size_read(inode); | |
2778 | u64 block_start; | |
2779 | u64 iosize; | |
2780 | sector_t sector; | |
2c64c53d | 2781 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
2782 | struct extent_map *em; |
2783 | struct block_device *bdev; | |
2784 | int ret; | |
2785 | int nr = 0; | |
7f3c74fb | 2786 | size_t pg_offset = 0; |
d1310b2e CM |
2787 | size_t blocksize; |
2788 | loff_t i_size = i_size_read(inode); | |
2789 | unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; | |
2790 | u64 nr_delalloc; | |
2791 | u64 delalloc_end; | |
c8b97818 CM |
2792 | int page_started; |
2793 | int compressed; | |
ffbd517d | 2794 | int write_flags; |
771ed689 | 2795 | unsigned long nr_written = 0; |
9e487107 | 2796 | bool fill_delalloc = true; |
d1310b2e | 2797 | |
ffbd517d | 2798 | if (wbc->sync_mode == WB_SYNC_ALL) |
721a9602 | 2799 | write_flags = WRITE_SYNC; |
ffbd517d CM |
2800 | else |
2801 | write_flags = WRITE; | |
2802 | ||
1abe9b8a | 2803 | trace___extent_writepage(page, inode, wbc); |
2804 | ||
d1310b2e | 2805 | WARN_ON(!PageLocked(page)); |
bf0da8c1 CM |
2806 | |
2807 | ClearPageError(page); | |
2808 | ||
7f3c74fb | 2809 | pg_offset = i_size & (PAGE_CACHE_SIZE - 1); |
211c17f5 | 2810 | if (page->index > end_index || |
7f3c74fb | 2811 | (page->index == end_index && !pg_offset)) { |
39be25cd | 2812 | page->mapping->a_ops->invalidatepage(page, 0); |
d1310b2e CM |
2813 | unlock_page(page); |
2814 | return 0; | |
2815 | } | |
2816 | ||
2817 | if (page->index == end_index) { | |
2818 | char *userpage; | |
2819 | ||
7ac687d9 | 2820 | userpage = kmap_atomic(page); |
7f3c74fb CM |
2821 | memset(userpage + pg_offset, 0, |
2822 | PAGE_CACHE_SIZE - pg_offset); | |
7ac687d9 | 2823 | kunmap_atomic(userpage); |
211c17f5 | 2824 | flush_dcache_page(page); |
d1310b2e | 2825 | } |
7f3c74fb | 2826 | pg_offset = 0; |
d1310b2e CM |
2827 | |
2828 | set_page_extent_mapped(page); | |
2829 | ||
9e487107 JB |
2830 | if (!tree->ops || !tree->ops->fill_delalloc) |
2831 | fill_delalloc = false; | |
2832 | ||
d1310b2e CM |
2833 | delalloc_start = start; |
2834 | delalloc_end = 0; | |
c8b97818 | 2835 | page_started = 0; |
9e487107 | 2836 | if (!epd->extent_locked && fill_delalloc) { |
f85d7d6c | 2837 | u64 delalloc_to_write = 0; |
11c8349b CM |
2838 | /* |
2839 | * make sure the wbc mapping index is at least updated | |
2840 | * to this page. | |
2841 | */ | |
2842 | update_nr_written(page, wbc, 0); | |
2843 | ||
d397712b | 2844 | while (delalloc_end < page_end) { |
771ed689 | 2845 | nr_delalloc = find_lock_delalloc_range(inode, tree, |
c8b97818 CM |
2846 | page, |
2847 | &delalloc_start, | |
d1310b2e CM |
2848 | &delalloc_end, |
2849 | 128 * 1024 * 1024); | |
771ed689 CM |
2850 | if (nr_delalloc == 0) { |
2851 | delalloc_start = delalloc_end + 1; | |
2852 | continue; | |
2853 | } | |
013bd4c3 TI |
2854 | ret = tree->ops->fill_delalloc(inode, page, |
2855 | delalloc_start, | |
2856 | delalloc_end, | |
2857 | &page_started, | |
2858 | &nr_written); | |
79787eaa JM |
2859 | /* File system has been set read-only */ |
2860 | if (ret) { | |
2861 | SetPageError(page); | |
2862 | goto done; | |
2863 | } | |
f85d7d6c CM |
2864 | /* |
2865 | * delalloc_end is already one less than the total | |
2866 | * length, so we don't subtract one from | |
2867 | * PAGE_CACHE_SIZE | |
2868 | */ | |
2869 | delalloc_to_write += (delalloc_end - delalloc_start + | |
2870 | PAGE_CACHE_SIZE) >> | |
2871 | PAGE_CACHE_SHIFT; | |
d1310b2e | 2872 | delalloc_start = delalloc_end + 1; |
d1310b2e | 2873 | } |
f85d7d6c CM |
2874 | if (wbc->nr_to_write < delalloc_to_write) { |
2875 | int thresh = 8192; | |
2876 | ||
2877 | if (delalloc_to_write < thresh * 2) | |
2878 | thresh = delalloc_to_write; | |
2879 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
2880 | thresh); | |
2881 | } | |
c8b97818 | 2882 | |
771ed689 CM |
2883 | /* did the fill delalloc function already unlock and start |
2884 | * the IO? | |
2885 | */ | |
2886 | if (page_started) { | |
2887 | ret = 0; | |
11c8349b CM |
2888 | /* |
2889 | * we've unlocked the page, so we can't update | |
2890 | * the mapping's writeback index, just update | |
2891 | * nr_to_write. | |
2892 | */ | |
2893 | wbc->nr_to_write -= nr_written; | |
2894 | goto done_unlocked; | |
771ed689 | 2895 | } |
c8b97818 | 2896 | } |
247e743c | 2897 | if (tree->ops && tree->ops->writepage_start_hook) { |
c8b97818 CM |
2898 | ret = tree->ops->writepage_start_hook(page, start, |
2899 | page_end); | |
87826df0 JM |
2900 | if (ret) { |
2901 | /* Fixup worker will requeue */ | |
2902 | if (ret == -EBUSY) | |
2903 | wbc->pages_skipped++; | |
2904 | else | |
2905 | redirty_page_for_writepage(wbc, page); | |
11c8349b | 2906 | update_nr_written(page, wbc, nr_written); |
247e743c | 2907 | unlock_page(page); |
771ed689 | 2908 | ret = 0; |
11c8349b | 2909 | goto done_unlocked; |
247e743c CM |
2910 | } |
2911 | } | |
2912 | ||
11c8349b CM |
2913 | /* |
2914 | * we don't want to touch the inode after unlocking the page, | |
2915 | * so we update the mapping writeback index now | |
2916 | */ | |
2917 | update_nr_written(page, wbc, nr_written + 1); | |
771ed689 | 2918 | |
d1310b2e | 2919 | end = page_end; |
d1310b2e | 2920 | if (last_byte <= start) { |
e6dcd2dc CM |
2921 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2922 | tree->ops->writepage_end_io_hook(page, start, | |
2923 | page_end, NULL, 1); | |
d1310b2e CM |
2924 | goto done; |
2925 | } | |
2926 | ||
d1310b2e CM |
2927 | blocksize = inode->i_sb->s_blocksize; |
2928 | ||
2929 | while (cur <= end) { | |
2930 | if (cur >= last_byte) { | |
e6dcd2dc CM |
2931 | if (tree->ops && tree->ops->writepage_end_io_hook) |
2932 | tree->ops->writepage_end_io_hook(page, cur, | |
2933 | page_end, NULL, 1); | |
d1310b2e CM |
2934 | break; |
2935 | } | |
7f3c74fb | 2936 | em = epd->get_extent(inode, page, pg_offset, cur, |
d1310b2e | 2937 | end - cur + 1, 1); |
c704005d | 2938 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e CM |
2939 | SetPageError(page); |
2940 | break; | |
2941 | } | |
2942 | ||
2943 | extent_offset = cur - em->start; | |
2944 | BUG_ON(extent_map_end(em) <= cur); | |
2945 | BUG_ON(end < cur); | |
2946 | iosize = min(extent_map_end(em) - cur, end - cur + 1); | |
2947 | iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); | |
2948 | sector = (em->block_start + extent_offset) >> 9; | |
2949 | bdev = em->bdev; | |
2950 | block_start = em->block_start; | |
c8b97818 | 2951 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
d1310b2e CM |
2952 | free_extent_map(em); |
2953 | em = NULL; | |
2954 | ||
c8b97818 CM |
2955 | /* |
2956 | * compressed and inline extents are written through other | |
2957 | * paths in the FS | |
2958 | */ | |
2959 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 2960 | block_start == EXTENT_MAP_INLINE) { |
c8b97818 CM |
2961 | /* |
2962 | * end_io notification does not happen here for | |
2963 | * compressed extents | |
2964 | */ | |
2965 | if (!compressed && tree->ops && | |
2966 | tree->ops->writepage_end_io_hook) | |
e6dcd2dc CM |
2967 | tree->ops->writepage_end_io_hook(page, cur, |
2968 | cur + iosize - 1, | |
2969 | NULL, 1); | |
c8b97818 CM |
2970 | else if (compressed) { |
2971 | /* we don't want to end_page_writeback on | |
2972 | * a compressed extent. this happens | |
2973 | * elsewhere | |
2974 | */ | |
2975 | nr++; | |
2976 | } | |
2977 | ||
2978 | cur += iosize; | |
7f3c74fb | 2979 | pg_offset += iosize; |
d1310b2e CM |
2980 | continue; |
2981 | } | |
d1310b2e CM |
2982 | /* leave this out until we have a page_mkwrite call */ |
2983 | if (0 && !test_range_bit(tree, cur, cur + iosize - 1, | |
9655d298 | 2984 | EXTENT_DIRTY, 0, NULL)) { |
d1310b2e | 2985 | cur = cur + iosize; |
7f3c74fb | 2986 | pg_offset += iosize; |
d1310b2e CM |
2987 | continue; |
2988 | } | |
c8b97818 | 2989 | |
d1310b2e CM |
2990 | if (tree->ops && tree->ops->writepage_io_hook) { |
2991 | ret = tree->ops->writepage_io_hook(page, cur, | |
2992 | cur + iosize - 1); | |
2993 | } else { | |
2994 | ret = 0; | |
2995 | } | |
1259ab75 | 2996 | if (ret) { |
d1310b2e | 2997 | SetPageError(page); |
1259ab75 | 2998 | } else { |
d1310b2e | 2999 | unsigned long max_nr = end_index + 1; |
7f3c74fb | 3000 | |
d1310b2e CM |
3001 | set_range_writeback(tree, cur, cur + iosize - 1); |
3002 | if (!PageWriteback(page)) { | |
d397712b CM |
3003 | printk(KERN_ERR "btrfs warning page %lu not " |
3004 | "writeback, cur %llu end %llu\n", | |
3005 | page->index, (unsigned long long)cur, | |
d1310b2e CM |
3006 | (unsigned long long)end); |
3007 | } | |
3008 | ||
ffbd517d CM |
3009 | ret = submit_extent_page(write_flags, tree, page, |
3010 | sector, iosize, pg_offset, | |
3011 | bdev, &epd->bio, max_nr, | |
c8b97818 CM |
3012 | end_bio_extent_writepage, |
3013 | 0, 0, 0); | |
d1310b2e CM |
3014 | if (ret) |
3015 | SetPageError(page); | |
3016 | } | |
3017 | cur = cur + iosize; | |
7f3c74fb | 3018 | pg_offset += iosize; |
d1310b2e CM |
3019 | nr++; |
3020 | } | |
3021 | done: | |
3022 | if (nr == 0) { | |
3023 | /* make sure the mapping tag for page dirty gets cleared */ | |
3024 | set_page_writeback(page); | |
3025 | end_page_writeback(page); | |
3026 | } | |
d1310b2e | 3027 | unlock_page(page); |
771ed689 | 3028 | |
11c8349b CM |
3029 | done_unlocked: |
3030 | ||
2c64c53d CM |
3031 | /* drop our reference on any cached states */ |
3032 | free_extent_state(cached_state); | |
d1310b2e CM |
3033 | return 0; |
3034 | } | |
3035 | ||
0b32f4bb JB |
3036 | static int eb_wait(void *word) |
3037 | { | |
3038 | io_schedule(); | |
3039 | return 0; | |
3040 | } | |
3041 | ||
3042 | static void wait_on_extent_buffer_writeback(struct extent_buffer *eb) | |
3043 | { | |
3044 | wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait, | |
3045 | TASK_UNINTERRUPTIBLE); | |
3046 | } | |
3047 | ||
3048 | static int lock_extent_buffer_for_io(struct extent_buffer *eb, | |
3049 | struct btrfs_fs_info *fs_info, | |
3050 | struct extent_page_data *epd) | |
3051 | { | |
3052 | unsigned long i, num_pages; | |
3053 | int flush = 0; | |
3054 | int ret = 0; | |
3055 | ||
3056 | if (!btrfs_try_tree_write_lock(eb)) { | |
3057 | flush = 1; | |
3058 | flush_write_bio(epd); | |
3059 | btrfs_tree_lock(eb); | |
3060 | } | |
3061 | ||
3062 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
3063 | btrfs_tree_unlock(eb); | |
3064 | if (!epd->sync_io) | |
3065 | return 0; | |
3066 | if (!flush) { | |
3067 | flush_write_bio(epd); | |
3068 | flush = 1; | |
3069 | } | |
a098d8e8 CM |
3070 | while (1) { |
3071 | wait_on_extent_buffer_writeback(eb); | |
3072 | btrfs_tree_lock(eb); | |
3073 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
3074 | break; | |
0b32f4bb | 3075 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
3076 | } |
3077 | } | |
3078 | ||
3079 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { | |
3080 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
3081 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); | |
3082 | spin_lock(&fs_info->delalloc_lock); | |
3083 | if (fs_info->dirty_metadata_bytes >= eb->len) | |
3084 | fs_info->dirty_metadata_bytes -= eb->len; | |
3085 | else | |
3086 | WARN_ON(1); | |
3087 | spin_unlock(&fs_info->delalloc_lock); | |
3088 | ret = 1; | |
3089 | } | |
3090 | ||
3091 | btrfs_tree_unlock(eb); | |
3092 | ||
3093 | if (!ret) | |
3094 | return ret; | |
3095 | ||
3096 | num_pages = num_extent_pages(eb->start, eb->len); | |
3097 | for (i = 0; i < num_pages; i++) { | |
3098 | struct page *p = extent_buffer_page(eb, i); | |
3099 | ||
3100 | if (!trylock_page(p)) { | |
3101 | if (!flush) { | |
3102 | flush_write_bio(epd); | |
3103 | flush = 1; | |
3104 | } | |
3105 | lock_page(p); | |
3106 | } | |
3107 | } | |
3108 | ||
3109 | return ret; | |
3110 | } | |
3111 | ||
3112 | static void end_extent_buffer_writeback(struct extent_buffer *eb) | |
3113 | { | |
3114 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
3115 | smp_mb__after_clear_bit(); | |
3116 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); | |
3117 | } | |
3118 | ||
3119 | static void end_bio_extent_buffer_writepage(struct bio *bio, int err) | |
3120 | { | |
3121 | int uptodate = err == 0; | |
3122 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
3123 | struct extent_buffer *eb; | |
3124 | int done; | |
3125 | ||
3126 | do { | |
3127 | struct page *page = bvec->bv_page; | |
3128 | ||
3129 | bvec--; | |
3130 | eb = (struct extent_buffer *)page->private; | |
3131 | BUG_ON(!eb); | |
3132 | done = atomic_dec_and_test(&eb->io_pages); | |
3133 | ||
3134 | if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) { | |
3135 | set_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
3136 | ClearPageUptodate(page); | |
3137 | SetPageError(page); | |
3138 | } | |
3139 | ||
3140 | end_page_writeback(page); | |
3141 | ||
3142 | if (!done) | |
3143 | continue; | |
3144 | ||
3145 | end_extent_buffer_writeback(eb); | |
3146 | } while (bvec >= bio->bi_io_vec); | |
3147 | ||
3148 | bio_put(bio); | |
3149 | ||
3150 | } | |
3151 | ||
3152 | static int write_one_eb(struct extent_buffer *eb, | |
3153 | struct btrfs_fs_info *fs_info, | |
3154 | struct writeback_control *wbc, | |
3155 | struct extent_page_data *epd) | |
3156 | { | |
3157 | struct block_device *bdev = fs_info->fs_devices->latest_bdev; | |
3158 | u64 offset = eb->start; | |
3159 | unsigned long i, num_pages; | |
3160 | int rw = (epd->sync_io ? WRITE_SYNC : WRITE); | |
d7dbe9e7 | 3161 | int ret = 0; |
0b32f4bb JB |
3162 | |
3163 | clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
3164 | num_pages = num_extent_pages(eb->start, eb->len); | |
3165 | atomic_set(&eb->io_pages, num_pages); | |
3166 | for (i = 0; i < num_pages; i++) { | |
3167 | struct page *p = extent_buffer_page(eb, i); | |
3168 | ||
3169 | clear_page_dirty_for_io(p); | |
3170 | set_page_writeback(p); | |
3171 | ret = submit_extent_page(rw, eb->tree, p, offset >> 9, | |
3172 | PAGE_CACHE_SIZE, 0, bdev, &epd->bio, | |
3173 | -1, end_bio_extent_buffer_writepage, | |
3174 | 0, 0, 0); | |
3175 | if (ret) { | |
3176 | set_bit(EXTENT_BUFFER_IOERR, &eb->bflags); | |
3177 | SetPageError(p); | |
3178 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) | |
3179 | end_extent_buffer_writeback(eb); | |
3180 | ret = -EIO; | |
3181 | break; | |
3182 | } | |
3183 | offset += PAGE_CACHE_SIZE; | |
3184 | update_nr_written(p, wbc, 1); | |
3185 | unlock_page(p); | |
3186 | } | |
3187 | ||
3188 | if (unlikely(ret)) { | |
3189 | for (; i < num_pages; i++) { | |
3190 | struct page *p = extent_buffer_page(eb, i); | |
3191 | unlock_page(p); | |
3192 | } | |
3193 | } | |
3194 | ||
3195 | return ret; | |
3196 | } | |
3197 | ||
3198 | int btree_write_cache_pages(struct address_space *mapping, | |
3199 | struct writeback_control *wbc) | |
3200 | { | |
3201 | struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree; | |
3202 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; | |
3203 | struct extent_buffer *eb, *prev_eb = NULL; | |
3204 | struct extent_page_data epd = { | |
3205 | .bio = NULL, | |
3206 | .tree = tree, | |
3207 | .extent_locked = 0, | |
3208 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
3209 | }; | |
3210 | int ret = 0; | |
3211 | int done = 0; | |
3212 | int nr_to_write_done = 0; | |
3213 | struct pagevec pvec; | |
3214 | int nr_pages; | |
3215 | pgoff_t index; | |
3216 | pgoff_t end; /* Inclusive */ | |
3217 | int scanned = 0; | |
3218 | int tag; | |
3219 | ||
3220 | pagevec_init(&pvec, 0); | |
3221 | if (wbc->range_cyclic) { | |
3222 | index = mapping->writeback_index; /* Start from prev offset */ | |
3223 | end = -1; | |
3224 | } else { | |
3225 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
3226 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
3227 | scanned = 1; | |
3228 | } | |
3229 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3230 | tag = PAGECACHE_TAG_TOWRITE; | |
3231 | else | |
3232 | tag = PAGECACHE_TAG_DIRTY; | |
3233 | retry: | |
3234 | if (wbc->sync_mode == WB_SYNC_ALL) | |
3235 | tag_pages_for_writeback(mapping, index, end); | |
3236 | while (!done && !nr_to_write_done && (index <= end) && | |
3237 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, | |
3238 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
3239 | unsigned i; | |
3240 | ||
3241 | scanned = 1; | |
3242 | for (i = 0; i < nr_pages; i++) { | |
3243 | struct page *page = pvec.pages[i]; | |
3244 | ||
3245 | if (!PagePrivate(page)) | |
3246 | continue; | |
3247 | ||
3248 | if (!wbc->range_cyclic && page->index > end) { | |
3249 | done = 1; | |
3250 | break; | |
3251 | } | |
3252 | ||
3253 | eb = (struct extent_buffer *)page->private; | |
3254 | if (!eb) { | |
3255 | WARN_ON(1); | |
3256 | continue; | |
3257 | } | |
3258 | ||
3259 | if (eb == prev_eb) | |
3260 | continue; | |
3261 | ||
3262 | if (!atomic_inc_not_zero(&eb->refs)) { | |
3263 | WARN_ON(1); | |
3264 | continue; | |
3265 | } | |
3266 | ||
3267 | prev_eb = eb; | |
3268 | ret = lock_extent_buffer_for_io(eb, fs_info, &epd); | |
3269 | if (!ret) { | |
3270 | free_extent_buffer(eb); | |
3271 | continue; | |
3272 | } | |
3273 | ||
3274 | ret = write_one_eb(eb, fs_info, wbc, &epd); | |
3275 | if (ret) { | |
3276 | done = 1; | |
3277 | free_extent_buffer(eb); | |
3278 | break; | |
3279 | } | |
3280 | free_extent_buffer(eb); | |
3281 | ||
3282 | /* | |
3283 | * the filesystem may choose to bump up nr_to_write. | |
3284 | * We have to make sure to honor the new nr_to_write | |
3285 | * at any time | |
3286 | */ | |
3287 | nr_to_write_done = wbc->nr_to_write <= 0; | |
3288 | } | |
3289 | pagevec_release(&pvec); | |
3290 | cond_resched(); | |
3291 | } | |
3292 | if (!scanned && !done) { | |
3293 | /* | |
3294 | * We hit the last page and there is more work to be done: wrap | |
3295 | * back to the start of the file | |
3296 | */ | |
3297 | scanned = 1; | |
3298 | index = 0; | |
3299 | goto retry; | |
3300 | } | |
3301 | flush_write_bio(&epd); | |
3302 | return ret; | |
3303 | } | |
3304 | ||
d1310b2e | 3305 | /** |
4bef0848 | 3306 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
d1310b2e CM |
3307 | * @mapping: address space structure to write |
3308 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
3309 | * @writepage: function called for each page | |
3310 | * @data: data passed to writepage function | |
3311 | * | |
3312 | * If a page is already under I/O, write_cache_pages() skips it, even | |
3313 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
3314 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
3315 | * and msync() need to guarantee that all the data which was dirty at the time | |
3316 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
3317 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
3318 | * existing IO to complete. | |
3319 | */ | |
b2950863 | 3320 | static int extent_write_cache_pages(struct extent_io_tree *tree, |
4bef0848 CM |
3321 | struct address_space *mapping, |
3322 | struct writeback_control *wbc, | |
d2c3f4f6 CM |
3323 | writepage_t writepage, void *data, |
3324 | void (*flush_fn)(void *)) | |
d1310b2e | 3325 | { |
d1310b2e CM |
3326 | int ret = 0; |
3327 | int done = 0; | |
f85d7d6c | 3328 | int nr_to_write_done = 0; |
d1310b2e CM |
3329 | struct pagevec pvec; |
3330 | int nr_pages; | |
3331 | pgoff_t index; | |
3332 | pgoff_t end; /* Inclusive */ | |
3333 | int scanned = 0; | |
f7aaa06b | 3334 | int tag; |
d1310b2e | 3335 | |
d1310b2e CM |
3336 | pagevec_init(&pvec, 0); |
3337 | if (wbc->range_cyclic) { | |
3338 | index = mapping->writeback_index; /* Start from prev offset */ | |
3339 | end = -1; | |
3340 | } else { | |
3341 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
3342 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
3343 | scanned = 1; |
3344 | } | |
f7aaa06b JB |
3345 | if (wbc->sync_mode == WB_SYNC_ALL) |
3346 | tag = PAGECACHE_TAG_TOWRITE; | |
3347 | else | |
3348 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 3349 | retry: |
f7aaa06b JB |
3350 | if (wbc->sync_mode == WB_SYNC_ALL) |
3351 | tag_pages_for_writeback(mapping, index, end); | |
f85d7d6c | 3352 | while (!done && !nr_to_write_done && (index <= end) && |
f7aaa06b JB |
3353 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
3354 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
d1310b2e CM |
3355 | unsigned i; |
3356 | ||
3357 | scanned = 1; | |
3358 | for (i = 0; i < nr_pages; i++) { | |
3359 | struct page *page = pvec.pages[i]; | |
3360 | ||
3361 | /* | |
3362 | * At this point we hold neither mapping->tree_lock nor | |
3363 | * lock on the page itself: the page may be truncated or | |
3364 | * invalidated (changing page->mapping to NULL), or even | |
3365 | * swizzled back from swapper_space to tmpfs file | |
3366 | * mapping | |
3367 | */ | |
01d658f2 CM |
3368 | if (tree->ops && |
3369 | tree->ops->write_cache_pages_lock_hook) { | |
3370 | tree->ops->write_cache_pages_lock_hook(page, | |
3371 | data, flush_fn); | |
3372 | } else { | |
3373 | if (!trylock_page(page)) { | |
3374 | flush_fn(data); | |
3375 | lock_page(page); | |
3376 | } | |
3377 | } | |
d1310b2e CM |
3378 | |
3379 | if (unlikely(page->mapping != mapping)) { | |
3380 | unlock_page(page); | |
3381 | continue; | |
3382 | } | |
3383 | ||
3384 | if (!wbc->range_cyclic && page->index > end) { | |
3385 | done = 1; | |
3386 | unlock_page(page); | |
3387 | continue; | |
3388 | } | |
3389 | ||
d2c3f4f6 | 3390 | if (wbc->sync_mode != WB_SYNC_NONE) { |
0e6bd956 CM |
3391 | if (PageWriteback(page)) |
3392 | flush_fn(data); | |
d1310b2e | 3393 | wait_on_page_writeback(page); |
d2c3f4f6 | 3394 | } |
d1310b2e CM |
3395 | |
3396 | if (PageWriteback(page) || | |
3397 | !clear_page_dirty_for_io(page)) { | |
3398 | unlock_page(page); | |
3399 | continue; | |
3400 | } | |
3401 | ||
3402 | ret = (*writepage)(page, wbc, data); | |
3403 | ||
3404 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { | |
3405 | unlock_page(page); | |
3406 | ret = 0; | |
3407 | } | |
f85d7d6c | 3408 | if (ret) |
d1310b2e | 3409 | done = 1; |
f85d7d6c CM |
3410 | |
3411 | /* | |
3412 | * the filesystem may choose to bump up nr_to_write. | |
3413 | * We have to make sure to honor the new nr_to_write | |
3414 | * at any time | |
3415 | */ | |
3416 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
3417 | } |
3418 | pagevec_release(&pvec); | |
3419 | cond_resched(); | |
3420 | } | |
3421 | if (!scanned && !done) { | |
3422 | /* | |
3423 | * We hit the last page and there is more work to be done: wrap | |
3424 | * back to the start of the file | |
3425 | */ | |
3426 | scanned = 1; | |
3427 | index = 0; | |
3428 | goto retry; | |
3429 | } | |
d1310b2e CM |
3430 | return ret; |
3431 | } | |
d1310b2e | 3432 | |
ffbd517d | 3433 | static void flush_epd_write_bio(struct extent_page_data *epd) |
d2c3f4f6 | 3434 | { |
d2c3f4f6 | 3435 | if (epd->bio) { |
355808c2 JM |
3436 | int rw = WRITE; |
3437 | int ret; | |
3438 | ||
ffbd517d | 3439 | if (epd->sync_io) |
355808c2 JM |
3440 | rw = WRITE_SYNC; |
3441 | ||
3442 | ret = submit_one_bio(rw, epd->bio, 0, 0); | |
79787eaa | 3443 | BUG_ON(ret < 0); /* -ENOMEM */ |
d2c3f4f6 CM |
3444 | epd->bio = NULL; |
3445 | } | |
3446 | } | |
3447 | ||
ffbd517d CM |
3448 | static noinline void flush_write_bio(void *data) |
3449 | { | |
3450 | struct extent_page_data *epd = data; | |
3451 | flush_epd_write_bio(epd); | |
3452 | } | |
3453 | ||
d1310b2e CM |
3454 | int extent_write_full_page(struct extent_io_tree *tree, struct page *page, |
3455 | get_extent_t *get_extent, | |
3456 | struct writeback_control *wbc) | |
3457 | { | |
3458 | int ret; | |
d1310b2e CM |
3459 | struct extent_page_data epd = { |
3460 | .bio = NULL, | |
3461 | .tree = tree, | |
3462 | .get_extent = get_extent, | |
771ed689 | 3463 | .extent_locked = 0, |
ffbd517d | 3464 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e | 3465 | }; |
d1310b2e | 3466 | |
d1310b2e CM |
3467 | ret = __extent_writepage(page, wbc, &epd); |
3468 | ||
ffbd517d | 3469 | flush_epd_write_bio(&epd); |
d1310b2e CM |
3470 | return ret; |
3471 | } | |
d1310b2e | 3472 | |
771ed689 CM |
3473 | int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, |
3474 | u64 start, u64 end, get_extent_t *get_extent, | |
3475 | int mode) | |
3476 | { | |
3477 | int ret = 0; | |
3478 | struct address_space *mapping = inode->i_mapping; | |
3479 | struct page *page; | |
3480 | unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >> | |
3481 | PAGE_CACHE_SHIFT; | |
3482 | ||
3483 | struct extent_page_data epd = { | |
3484 | .bio = NULL, | |
3485 | .tree = tree, | |
3486 | .get_extent = get_extent, | |
3487 | .extent_locked = 1, | |
ffbd517d | 3488 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
3489 | }; |
3490 | struct writeback_control wbc_writepages = { | |
771ed689 | 3491 | .sync_mode = mode, |
771ed689 CM |
3492 | .nr_to_write = nr_pages * 2, |
3493 | .range_start = start, | |
3494 | .range_end = end + 1, | |
3495 | }; | |
3496 | ||
d397712b | 3497 | while (start <= end) { |
771ed689 CM |
3498 | page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); |
3499 | if (clear_page_dirty_for_io(page)) | |
3500 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
3501 | else { | |
3502 | if (tree->ops && tree->ops->writepage_end_io_hook) | |
3503 | tree->ops->writepage_end_io_hook(page, start, | |
3504 | start + PAGE_CACHE_SIZE - 1, | |
3505 | NULL, 1); | |
3506 | unlock_page(page); | |
3507 | } | |
3508 | page_cache_release(page); | |
3509 | start += PAGE_CACHE_SIZE; | |
3510 | } | |
3511 | ||
ffbd517d | 3512 | flush_epd_write_bio(&epd); |
771ed689 CM |
3513 | return ret; |
3514 | } | |
d1310b2e CM |
3515 | |
3516 | int extent_writepages(struct extent_io_tree *tree, | |
3517 | struct address_space *mapping, | |
3518 | get_extent_t *get_extent, | |
3519 | struct writeback_control *wbc) | |
3520 | { | |
3521 | int ret = 0; | |
3522 | struct extent_page_data epd = { | |
3523 | .bio = NULL, | |
3524 | .tree = tree, | |
3525 | .get_extent = get_extent, | |
771ed689 | 3526 | .extent_locked = 0, |
ffbd517d | 3527 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
3528 | }; |
3529 | ||
4bef0848 | 3530 | ret = extent_write_cache_pages(tree, mapping, wbc, |
d2c3f4f6 CM |
3531 | __extent_writepage, &epd, |
3532 | flush_write_bio); | |
ffbd517d | 3533 | flush_epd_write_bio(&epd); |
d1310b2e CM |
3534 | return ret; |
3535 | } | |
d1310b2e CM |
3536 | |
3537 | int extent_readpages(struct extent_io_tree *tree, | |
3538 | struct address_space *mapping, | |
3539 | struct list_head *pages, unsigned nr_pages, | |
3540 | get_extent_t get_extent) | |
3541 | { | |
3542 | struct bio *bio = NULL; | |
3543 | unsigned page_idx; | |
c8b97818 | 3544 | unsigned long bio_flags = 0; |
d1310b2e | 3545 | |
d1310b2e CM |
3546 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
3547 | struct page *page = list_entry(pages->prev, struct page, lru); | |
3548 | ||
3549 | prefetchw(&page->flags); | |
3550 | list_del(&page->lru); | |
28ecb609 | 3551 | if (!add_to_page_cache_lru(page, mapping, |
43e817a1 | 3552 | page->index, GFP_NOFS)) { |
f188591e | 3553 | __extent_read_full_page(tree, page, get_extent, |
c8b97818 | 3554 | &bio, 0, &bio_flags); |
d1310b2e CM |
3555 | } |
3556 | page_cache_release(page); | |
3557 | } | |
d1310b2e CM |
3558 | BUG_ON(!list_empty(pages)); |
3559 | if (bio) | |
79787eaa | 3560 | return submit_one_bio(READ, bio, 0, bio_flags); |
d1310b2e CM |
3561 | return 0; |
3562 | } | |
d1310b2e CM |
3563 | |
3564 | /* | |
3565 | * basic invalidatepage code, this waits on any locked or writeback | |
3566 | * ranges corresponding to the page, and then deletes any extent state | |
3567 | * records from the tree | |
3568 | */ | |
3569 | int extent_invalidatepage(struct extent_io_tree *tree, | |
3570 | struct page *page, unsigned long offset) | |
3571 | { | |
2ac55d41 | 3572 | struct extent_state *cached_state = NULL; |
d1310b2e CM |
3573 | u64 start = ((u64)page->index << PAGE_CACHE_SHIFT); |
3574 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
3575 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; | |
3576 | ||
d397712b | 3577 | start += (offset + blocksize - 1) & ~(blocksize - 1); |
d1310b2e CM |
3578 | if (start > end) |
3579 | return 0; | |
3580 | ||
d0082371 | 3581 | lock_extent_bits(tree, start, end, 0, &cached_state); |
1edbb734 | 3582 | wait_on_page_writeback(page); |
d1310b2e | 3583 | clear_extent_bit(tree, start, end, |
32c00aff JB |
3584 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
3585 | EXTENT_DO_ACCOUNTING, | |
2ac55d41 | 3586 | 1, 1, &cached_state, GFP_NOFS); |
d1310b2e CM |
3587 | return 0; |
3588 | } | |
d1310b2e | 3589 | |
7b13b7b1 CM |
3590 | /* |
3591 | * a helper for releasepage, this tests for areas of the page that | |
3592 | * are locked or under IO and drops the related state bits if it is safe | |
3593 | * to drop the page. | |
3594 | */ | |
3595 | int try_release_extent_state(struct extent_map_tree *map, | |
3596 | struct extent_io_tree *tree, struct page *page, | |
3597 | gfp_t mask) | |
3598 | { | |
3599 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
3600 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
3601 | int ret = 1; | |
3602 | ||
211f90e6 | 3603 | if (test_range_bit(tree, start, end, |
8b62b72b | 3604 | EXTENT_IOBITS, 0, NULL)) |
7b13b7b1 CM |
3605 | ret = 0; |
3606 | else { | |
3607 | if ((mask & GFP_NOFS) == GFP_NOFS) | |
3608 | mask = GFP_NOFS; | |
11ef160f CM |
3609 | /* |
3610 | * at this point we can safely clear everything except the | |
3611 | * locked bit and the nodatasum bit | |
3612 | */ | |
e3f24cc5 | 3613 | ret = clear_extent_bit(tree, start, end, |
11ef160f CM |
3614 | ~(EXTENT_LOCKED | EXTENT_NODATASUM), |
3615 | 0, 0, NULL, mask); | |
e3f24cc5 CM |
3616 | |
3617 | /* if clear_extent_bit failed for enomem reasons, | |
3618 | * we can't allow the release to continue. | |
3619 | */ | |
3620 | if (ret < 0) | |
3621 | ret = 0; | |
3622 | else | |
3623 | ret = 1; | |
7b13b7b1 CM |
3624 | } |
3625 | return ret; | |
3626 | } | |
7b13b7b1 | 3627 | |
d1310b2e CM |
3628 | /* |
3629 | * a helper for releasepage. As long as there are no locked extents | |
3630 | * in the range corresponding to the page, both state records and extent | |
3631 | * map records are removed | |
3632 | */ | |
3633 | int try_release_extent_mapping(struct extent_map_tree *map, | |
70dec807 CM |
3634 | struct extent_io_tree *tree, struct page *page, |
3635 | gfp_t mask) | |
d1310b2e CM |
3636 | { |
3637 | struct extent_map *em; | |
3638 | u64 start = (u64)page->index << PAGE_CACHE_SHIFT; | |
3639 | u64 end = start + PAGE_CACHE_SIZE - 1; | |
7b13b7b1 | 3640 | |
70dec807 CM |
3641 | if ((mask & __GFP_WAIT) && |
3642 | page->mapping->host->i_size > 16 * 1024 * 1024) { | |
39b5637f | 3643 | u64 len; |
70dec807 | 3644 | while (start <= end) { |
39b5637f | 3645 | len = end - start + 1; |
890871be | 3646 | write_lock(&map->lock); |
39b5637f | 3647 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 3648 | if (!em) { |
890871be | 3649 | write_unlock(&map->lock); |
70dec807 CM |
3650 | break; |
3651 | } | |
7f3c74fb CM |
3652 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
3653 | em->start != start) { | |
890871be | 3654 | write_unlock(&map->lock); |
70dec807 CM |
3655 | free_extent_map(em); |
3656 | break; | |
3657 | } | |
3658 | if (!test_range_bit(tree, em->start, | |
3659 | extent_map_end(em) - 1, | |
8b62b72b | 3660 | EXTENT_LOCKED | EXTENT_WRITEBACK, |
9655d298 | 3661 | 0, NULL)) { |
70dec807 CM |
3662 | remove_extent_mapping(map, em); |
3663 | /* once for the rb tree */ | |
3664 | free_extent_map(em); | |
3665 | } | |
3666 | start = extent_map_end(em); | |
890871be | 3667 | write_unlock(&map->lock); |
70dec807 CM |
3668 | |
3669 | /* once for us */ | |
d1310b2e CM |
3670 | free_extent_map(em); |
3671 | } | |
d1310b2e | 3672 | } |
7b13b7b1 | 3673 | return try_release_extent_state(map, tree, page, mask); |
d1310b2e | 3674 | } |
d1310b2e | 3675 | |
ec29ed5b CM |
3676 | /* |
3677 | * helper function for fiemap, which doesn't want to see any holes. | |
3678 | * This maps until we find something past 'last' | |
3679 | */ | |
3680 | static struct extent_map *get_extent_skip_holes(struct inode *inode, | |
3681 | u64 offset, | |
3682 | u64 last, | |
3683 | get_extent_t *get_extent) | |
3684 | { | |
3685 | u64 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
3686 | struct extent_map *em; | |
3687 | u64 len; | |
3688 | ||
3689 | if (offset >= last) | |
3690 | return NULL; | |
3691 | ||
3692 | while(1) { | |
3693 | len = last - offset; | |
3694 | if (len == 0) | |
3695 | break; | |
3696 | len = (len + sectorsize - 1) & ~(sectorsize - 1); | |
3697 | em = get_extent(inode, NULL, 0, offset, len, 0); | |
c704005d | 3698 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
3699 | return em; |
3700 | ||
3701 | /* if this isn't a hole return it */ | |
3702 | if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) && | |
3703 | em->block_start != EXTENT_MAP_HOLE) { | |
3704 | return em; | |
3705 | } | |
3706 | ||
3707 | /* this is a hole, advance to the next extent */ | |
3708 | offset = extent_map_end(em); | |
3709 | free_extent_map(em); | |
3710 | if (offset >= last) | |
3711 | break; | |
3712 | } | |
3713 | return NULL; | |
3714 | } | |
3715 | ||
1506fcc8 YS |
3716 | int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
3717 | __u64 start, __u64 len, get_extent_t *get_extent) | |
3718 | { | |
975f84fe | 3719 | int ret = 0; |
1506fcc8 YS |
3720 | u64 off = start; |
3721 | u64 max = start + len; | |
3722 | u32 flags = 0; | |
975f84fe JB |
3723 | u32 found_type; |
3724 | u64 last; | |
ec29ed5b | 3725 | u64 last_for_get_extent = 0; |
1506fcc8 | 3726 | u64 disko = 0; |
ec29ed5b | 3727 | u64 isize = i_size_read(inode); |
975f84fe | 3728 | struct btrfs_key found_key; |
1506fcc8 | 3729 | struct extent_map *em = NULL; |
2ac55d41 | 3730 | struct extent_state *cached_state = NULL; |
975f84fe JB |
3731 | struct btrfs_path *path; |
3732 | struct btrfs_file_extent_item *item; | |
1506fcc8 | 3733 | int end = 0; |
ec29ed5b CM |
3734 | u64 em_start = 0; |
3735 | u64 em_len = 0; | |
3736 | u64 em_end = 0; | |
1506fcc8 | 3737 | unsigned long emflags; |
1506fcc8 YS |
3738 | |
3739 | if (len == 0) | |
3740 | return -EINVAL; | |
3741 | ||
975f84fe JB |
3742 | path = btrfs_alloc_path(); |
3743 | if (!path) | |
3744 | return -ENOMEM; | |
3745 | path->leave_spinning = 1; | |
3746 | ||
4d479cf0 JB |
3747 | start = ALIGN(start, BTRFS_I(inode)->root->sectorsize); |
3748 | len = ALIGN(len, BTRFS_I(inode)->root->sectorsize); | |
3749 | ||
ec29ed5b CM |
3750 | /* |
3751 | * lookup the last file extent. We're not using i_size here | |
3752 | * because there might be preallocation past i_size | |
3753 | */ | |
975f84fe | 3754 | ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root, |
33345d01 | 3755 | path, btrfs_ino(inode), -1, 0); |
975f84fe JB |
3756 | if (ret < 0) { |
3757 | btrfs_free_path(path); | |
3758 | return ret; | |
3759 | } | |
3760 | WARN_ON(!ret); | |
3761 | path->slots[0]--; | |
3762 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3763 | struct btrfs_file_extent_item); | |
3764 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
3765 | found_type = btrfs_key_type(&found_key); | |
3766 | ||
ec29ed5b | 3767 | /* No extents, but there might be delalloc bits */ |
33345d01 | 3768 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 3769 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
3770 | /* have to trust i_size as the end */ |
3771 | last = (u64)-1; | |
3772 | last_for_get_extent = isize; | |
3773 | } else { | |
3774 | /* | |
3775 | * remember the start of the last extent. There are a | |
3776 | * bunch of different factors that go into the length of the | |
3777 | * extent, so its much less complex to remember where it started | |
3778 | */ | |
3779 | last = found_key.offset; | |
3780 | last_for_get_extent = last + 1; | |
975f84fe | 3781 | } |
975f84fe JB |
3782 | btrfs_free_path(path); |
3783 | ||
ec29ed5b CM |
3784 | /* |
3785 | * we might have some extents allocated but more delalloc past those | |
3786 | * extents. so, we trust isize unless the start of the last extent is | |
3787 | * beyond isize | |
3788 | */ | |
3789 | if (last < isize) { | |
3790 | last = (u64)-1; | |
3791 | last_for_get_extent = isize; | |
3792 | } | |
3793 | ||
2ac55d41 | 3794 | lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len, 0, |
d0082371 | 3795 | &cached_state); |
ec29ed5b | 3796 | |
4d479cf0 | 3797 | em = get_extent_skip_holes(inode, start, last_for_get_extent, |
ec29ed5b | 3798 | get_extent); |
1506fcc8 YS |
3799 | if (!em) |
3800 | goto out; | |
3801 | if (IS_ERR(em)) { | |
3802 | ret = PTR_ERR(em); | |
3803 | goto out; | |
3804 | } | |
975f84fe | 3805 | |
1506fcc8 | 3806 | while (!end) { |
ea8efc74 CM |
3807 | u64 offset_in_extent; |
3808 | ||
3809 | /* break if the extent we found is outside the range */ | |
3810 | if (em->start >= max || extent_map_end(em) < off) | |
3811 | break; | |
3812 | ||
3813 | /* | |
3814 | * get_extent may return an extent that starts before our | |
3815 | * requested range. We have to make sure the ranges | |
3816 | * we return to fiemap always move forward and don't | |
3817 | * overlap, so adjust the offsets here | |
3818 | */ | |
3819 | em_start = max(em->start, off); | |
1506fcc8 | 3820 | |
ea8efc74 CM |
3821 | /* |
3822 | * record the offset from the start of the extent | |
3823 | * for adjusting the disk offset below | |
3824 | */ | |
3825 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 3826 | em_end = extent_map_end(em); |
ea8efc74 | 3827 | em_len = em_end - em_start; |
ec29ed5b | 3828 | emflags = em->flags; |
1506fcc8 YS |
3829 | disko = 0; |
3830 | flags = 0; | |
3831 | ||
ea8efc74 CM |
3832 | /* |
3833 | * bump off for our next call to get_extent | |
3834 | */ | |
3835 | off = extent_map_end(em); | |
3836 | if (off >= max) | |
3837 | end = 1; | |
3838 | ||
93dbfad7 | 3839 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
3840 | end = 1; |
3841 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 3842 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
3843 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
3844 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 3845 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
3846 | flags |= (FIEMAP_EXTENT_DELALLOC | |
3847 | FIEMAP_EXTENT_UNKNOWN); | |
93dbfad7 | 3848 | } else { |
ea8efc74 | 3849 | disko = em->block_start + offset_in_extent; |
1506fcc8 YS |
3850 | } |
3851 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
3852 | flags |= FIEMAP_EXTENT_ENCODED; | |
3853 | ||
1506fcc8 YS |
3854 | free_extent_map(em); |
3855 | em = NULL; | |
ec29ed5b CM |
3856 | if ((em_start >= last) || em_len == (u64)-1 || |
3857 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
3858 | flags |= FIEMAP_EXTENT_LAST; |
3859 | end = 1; | |
3860 | } | |
3861 | ||
ec29ed5b CM |
3862 | /* now scan forward to see if this is really the last extent. */ |
3863 | em = get_extent_skip_holes(inode, off, last_for_get_extent, | |
3864 | get_extent); | |
3865 | if (IS_ERR(em)) { | |
3866 | ret = PTR_ERR(em); | |
3867 | goto out; | |
3868 | } | |
3869 | if (!em) { | |
975f84fe JB |
3870 | flags |= FIEMAP_EXTENT_LAST; |
3871 | end = 1; | |
3872 | } | |
ec29ed5b CM |
3873 | ret = fiemap_fill_next_extent(fieinfo, em_start, disko, |
3874 | em_len, flags); | |
3875 | if (ret) | |
3876 | goto out_free; | |
1506fcc8 YS |
3877 | } |
3878 | out_free: | |
3879 | free_extent_map(em); | |
3880 | out: | |
2ac55d41 JB |
3881 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len, |
3882 | &cached_state, GFP_NOFS); | |
1506fcc8 YS |
3883 | return ret; |
3884 | } | |
3885 | ||
4a54c8c1 | 3886 | inline struct page *extent_buffer_page(struct extent_buffer *eb, |
d1310b2e CM |
3887 | unsigned long i) |
3888 | { | |
727011e0 | 3889 | return eb->pages[i]; |
d1310b2e CM |
3890 | } |
3891 | ||
4a54c8c1 | 3892 | inline unsigned long num_extent_pages(u64 start, u64 len) |
728131d8 | 3893 | { |
6af118ce CM |
3894 | return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) - |
3895 | (start >> PAGE_CACHE_SHIFT); | |
728131d8 CM |
3896 | } |
3897 | ||
727011e0 CM |
3898 | static void __free_extent_buffer(struct extent_buffer *eb) |
3899 | { | |
3900 | #if LEAK_DEBUG | |
3901 | unsigned long flags; | |
3902 | spin_lock_irqsave(&leak_lock, flags); | |
3903 | list_del(&eb->leak_list); | |
3904 | spin_unlock_irqrestore(&leak_lock, flags); | |
3905 | #endif | |
3906 | if (eb->pages && eb->pages != eb->inline_pages) | |
3907 | kfree(eb->pages); | |
3908 | kmem_cache_free(extent_buffer_cache, eb); | |
3909 | } | |
3910 | ||
d1310b2e CM |
3911 | static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree, |
3912 | u64 start, | |
3913 | unsigned long len, | |
3914 | gfp_t mask) | |
3915 | { | |
3916 | struct extent_buffer *eb = NULL; | |
3935127c | 3917 | #if LEAK_DEBUG |
2d2ae547 | 3918 | unsigned long flags; |
4bef0848 | 3919 | #endif |
d1310b2e | 3920 | |
d1310b2e | 3921 | eb = kmem_cache_zalloc(extent_buffer_cache, mask); |
91ca338d TI |
3922 | if (eb == NULL) |
3923 | return NULL; | |
d1310b2e CM |
3924 | eb->start = start; |
3925 | eb->len = len; | |
4f2de97a | 3926 | eb->tree = tree; |
815a51c7 | 3927 | eb->bflags = 0; |
bd681513 CM |
3928 | rwlock_init(&eb->lock); |
3929 | atomic_set(&eb->write_locks, 0); | |
3930 | atomic_set(&eb->read_locks, 0); | |
3931 | atomic_set(&eb->blocking_readers, 0); | |
3932 | atomic_set(&eb->blocking_writers, 0); | |
3933 | atomic_set(&eb->spinning_readers, 0); | |
3934 | atomic_set(&eb->spinning_writers, 0); | |
5b25f70f | 3935 | eb->lock_nested = 0; |
bd681513 CM |
3936 | init_waitqueue_head(&eb->write_lock_wq); |
3937 | init_waitqueue_head(&eb->read_lock_wq); | |
b4ce94de | 3938 | |
3935127c | 3939 | #if LEAK_DEBUG |
2d2ae547 CM |
3940 | spin_lock_irqsave(&leak_lock, flags); |
3941 | list_add(&eb->leak_list, &buffers); | |
3942 | spin_unlock_irqrestore(&leak_lock, flags); | |
4bef0848 | 3943 | #endif |
3083ee2e | 3944 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 3945 | atomic_set(&eb->refs, 1); |
0b32f4bb | 3946 | atomic_set(&eb->io_pages, 0); |
727011e0 CM |
3947 | |
3948 | if (len > MAX_INLINE_EXTENT_BUFFER_SIZE) { | |
3949 | struct page **pages; | |
3950 | int num_pages = (len + PAGE_CACHE_SIZE - 1) >> | |
3951 | PAGE_CACHE_SHIFT; | |
3952 | pages = kzalloc(num_pages, mask); | |
3953 | if (!pages) { | |
3954 | __free_extent_buffer(eb); | |
3955 | return NULL; | |
3956 | } | |
3957 | eb->pages = pages; | |
3958 | } else { | |
3959 | eb->pages = eb->inline_pages; | |
3960 | } | |
d1310b2e CM |
3961 | |
3962 | return eb; | |
3963 | } | |
3964 | ||
815a51c7 JS |
3965 | struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src) |
3966 | { | |
3967 | unsigned long i; | |
3968 | struct page *p; | |
3969 | struct extent_buffer *new; | |
3970 | unsigned long num_pages = num_extent_pages(src->start, src->len); | |
3971 | ||
3972 | new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_ATOMIC); | |
3973 | if (new == NULL) | |
3974 | return NULL; | |
3975 | ||
3976 | for (i = 0; i < num_pages; i++) { | |
3977 | p = alloc_page(GFP_ATOMIC); | |
3978 | BUG_ON(!p); | |
3979 | attach_extent_buffer_page(new, p); | |
3980 | WARN_ON(PageDirty(p)); | |
3981 | SetPageUptodate(p); | |
3982 | new->pages[i] = p; | |
3983 | } | |
3984 | ||
3985 | copy_extent_buffer(new, src, 0, 0, src->len); | |
3986 | set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags); | |
3987 | set_bit(EXTENT_BUFFER_DUMMY, &new->bflags); | |
3988 | ||
3989 | return new; | |
3990 | } | |
3991 | ||
3992 | struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len) | |
3993 | { | |
3994 | struct extent_buffer *eb; | |
3995 | unsigned long num_pages = num_extent_pages(0, len); | |
3996 | unsigned long i; | |
3997 | ||
3998 | eb = __alloc_extent_buffer(NULL, start, len, GFP_ATOMIC); | |
3999 | if (!eb) | |
4000 | return NULL; | |
4001 | ||
4002 | for (i = 0; i < num_pages; i++) { | |
4003 | eb->pages[i] = alloc_page(GFP_ATOMIC); | |
4004 | if (!eb->pages[i]) | |
4005 | goto err; | |
4006 | } | |
4007 | set_extent_buffer_uptodate(eb); | |
4008 | btrfs_set_header_nritems(eb, 0); | |
4009 | set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); | |
4010 | ||
4011 | return eb; | |
4012 | err: | |
4013 | for (i--; i > 0; i--) | |
4014 | __free_page(eb->pages[i]); | |
4015 | __free_extent_buffer(eb); | |
4016 | return NULL; | |
4017 | } | |
4018 | ||
0b32f4bb | 4019 | static int extent_buffer_under_io(struct extent_buffer *eb) |
d1310b2e | 4020 | { |
0b32f4bb JB |
4021 | return (atomic_read(&eb->io_pages) || |
4022 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
4023 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
d1310b2e CM |
4024 | } |
4025 | ||
897ca6e9 MX |
4026 | /* |
4027 | * Helper for releasing extent buffer page. | |
4028 | */ | |
4029 | static void btrfs_release_extent_buffer_page(struct extent_buffer *eb, | |
4030 | unsigned long start_idx) | |
4031 | { | |
4032 | unsigned long index; | |
39bab87b | 4033 | unsigned long num_pages; |
897ca6e9 | 4034 | struct page *page; |
815a51c7 | 4035 | int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags); |
897ca6e9 | 4036 | |
0b32f4bb | 4037 | BUG_ON(extent_buffer_under_io(eb)); |
897ca6e9 | 4038 | |
39bab87b WSH |
4039 | num_pages = num_extent_pages(eb->start, eb->len); |
4040 | index = start_idx + num_pages; | |
897ca6e9 MX |
4041 | if (start_idx >= index) |
4042 | return; | |
4043 | ||
4044 | do { | |
4045 | index--; | |
4046 | page = extent_buffer_page(eb, index); | |
815a51c7 | 4047 | if (page && mapped) { |
4f2de97a JB |
4048 | spin_lock(&page->mapping->private_lock); |
4049 | /* | |
4050 | * We do this since we'll remove the pages after we've | |
4051 | * removed the eb from the radix tree, so we could race | |
4052 | * and have this page now attached to the new eb. So | |
4053 | * only clear page_private if it's still connected to | |
4054 | * this eb. | |
4055 | */ | |
4056 | if (PagePrivate(page) && | |
4057 | page->private == (unsigned long)eb) { | |
0b32f4bb | 4058 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); |
3083ee2e JB |
4059 | BUG_ON(PageDirty(page)); |
4060 | BUG_ON(PageWriteback(page)); | |
4f2de97a JB |
4061 | /* |
4062 | * We need to make sure we haven't be attached | |
4063 | * to a new eb. | |
4064 | */ | |
4065 | ClearPagePrivate(page); | |
4066 | set_page_private(page, 0); | |
4067 | /* One for the page private */ | |
4068 | page_cache_release(page); | |
4069 | } | |
4070 | spin_unlock(&page->mapping->private_lock); | |
4071 | ||
815a51c7 JS |
4072 | } |
4073 | if (page) { | |
4f2de97a | 4074 | /* One for when we alloced the page */ |
897ca6e9 | 4075 | page_cache_release(page); |
4f2de97a | 4076 | } |
897ca6e9 MX |
4077 | } while (index != start_idx); |
4078 | } | |
4079 | ||
4080 | /* | |
4081 | * Helper for releasing the extent buffer. | |
4082 | */ | |
4083 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
4084 | { | |
4085 | btrfs_release_extent_buffer_page(eb, 0); | |
4086 | __free_extent_buffer(eb); | |
4087 | } | |
4088 | ||
0b32f4bb JB |
4089 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
4090 | { | |
4091 | /* the ref bit is tricky. We have to make sure it is set | |
4092 | * if we have the buffer dirty. Otherwise the | |
4093 | * code to free a buffer can end up dropping a dirty | |
4094 | * page | |
4095 | * | |
4096 | * Once the ref bit is set, it won't go away while the | |
4097 | * buffer is dirty or in writeback, and it also won't | |
4098 | * go away while we have the reference count on the | |
4099 | * eb bumped. | |
4100 | * | |
4101 | * We can't just set the ref bit without bumping the | |
4102 | * ref on the eb because free_extent_buffer might | |
4103 | * see the ref bit and try to clear it. If this happens | |
4104 | * free_extent_buffer might end up dropping our original | |
4105 | * ref by mistake and freeing the page before we are able | |
4106 | * to add one more ref. | |
4107 | * | |
4108 | * So bump the ref count first, then set the bit. If someone | |
4109 | * beat us to it, drop the ref we added. | |
4110 | */ | |
4111 | if (!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { | |
4112 | atomic_inc(&eb->refs); | |
4113 | if (test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
4114 | atomic_dec(&eb->refs); | |
4115 | } | |
4116 | } | |
4117 | ||
5df4235e JB |
4118 | static void mark_extent_buffer_accessed(struct extent_buffer *eb) |
4119 | { | |
4120 | unsigned long num_pages, i; | |
4121 | ||
0b32f4bb JB |
4122 | check_buffer_tree_ref(eb); |
4123 | ||
5df4235e JB |
4124 | num_pages = num_extent_pages(eb->start, eb->len); |
4125 | for (i = 0; i < num_pages; i++) { | |
4126 | struct page *p = extent_buffer_page(eb, i); | |
4127 | mark_page_accessed(p); | |
4128 | } | |
4129 | } | |
4130 | ||
d1310b2e | 4131 | struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree, |
727011e0 | 4132 | u64 start, unsigned long len) |
d1310b2e CM |
4133 | { |
4134 | unsigned long num_pages = num_extent_pages(start, len); | |
4135 | unsigned long i; | |
4136 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
4137 | struct extent_buffer *eb; | |
6af118ce | 4138 | struct extent_buffer *exists = NULL; |
d1310b2e CM |
4139 | struct page *p; |
4140 | struct address_space *mapping = tree->mapping; | |
4141 | int uptodate = 1; | |
19fe0a8b | 4142 | int ret; |
d1310b2e | 4143 | |
19fe0a8b MX |
4144 | rcu_read_lock(); |
4145 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); | |
4146 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
4147 | rcu_read_unlock(); | |
5df4235e | 4148 | mark_extent_buffer_accessed(eb); |
6af118ce CM |
4149 | return eb; |
4150 | } | |
19fe0a8b | 4151 | rcu_read_unlock(); |
6af118ce | 4152 | |
ba144192 | 4153 | eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS); |
2b114d1d | 4154 | if (!eb) |
d1310b2e CM |
4155 | return NULL; |
4156 | ||
727011e0 | 4157 | for (i = 0; i < num_pages; i++, index++) { |
a6591715 | 4158 | p = find_or_create_page(mapping, index, GFP_NOFS); |
d1310b2e CM |
4159 | if (!p) { |
4160 | WARN_ON(1); | |
6af118ce | 4161 | goto free_eb; |
d1310b2e | 4162 | } |
4f2de97a JB |
4163 | |
4164 | spin_lock(&mapping->private_lock); | |
4165 | if (PagePrivate(p)) { | |
4166 | /* | |
4167 | * We could have already allocated an eb for this page | |
4168 | * and attached one so lets see if we can get a ref on | |
4169 | * the existing eb, and if we can we know it's good and | |
4170 | * we can just return that one, else we know we can just | |
4171 | * overwrite page->private. | |
4172 | */ | |
4173 | exists = (struct extent_buffer *)p->private; | |
4174 | if (atomic_inc_not_zero(&exists->refs)) { | |
4175 | spin_unlock(&mapping->private_lock); | |
4176 | unlock_page(p); | |
17de39ac | 4177 | page_cache_release(p); |
5df4235e | 4178 | mark_extent_buffer_accessed(exists); |
4f2de97a JB |
4179 | goto free_eb; |
4180 | } | |
4181 | ||
0b32f4bb | 4182 | /* |
4f2de97a JB |
4183 | * Do this so attach doesn't complain and we need to |
4184 | * drop the ref the old guy had. | |
4185 | */ | |
4186 | ClearPagePrivate(p); | |
0b32f4bb | 4187 | WARN_ON(PageDirty(p)); |
4f2de97a | 4188 | page_cache_release(p); |
d1310b2e | 4189 | } |
4f2de97a JB |
4190 | attach_extent_buffer_page(eb, p); |
4191 | spin_unlock(&mapping->private_lock); | |
0b32f4bb | 4192 | WARN_ON(PageDirty(p)); |
d1310b2e | 4193 | mark_page_accessed(p); |
727011e0 | 4194 | eb->pages[i] = p; |
d1310b2e CM |
4195 | if (!PageUptodate(p)) |
4196 | uptodate = 0; | |
eb14ab8e CM |
4197 | |
4198 | /* | |
4199 | * see below about how we avoid a nasty race with release page | |
4200 | * and why we unlock later | |
4201 | */ | |
d1310b2e CM |
4202 | } |
4203 | if (uptodate) | |
b4ce94de | 4204 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 4205 | again: |
19fe0a8b MX |
4206 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); |
4207 | if (ret) | |
4208 | goto free_eb; | |
4209 | ||
6af118ce | 4210 | spin_lock(&tree->buffer_lock); |
19fe0a8b MX |
4211 | ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb); |
4212 | if (ret == -EEXIST) { | |
4213 | exists = radix_tree_lookup(&tree->buffer, | |
4214 | start >> PAGE_CACHE_SHIFT); | |
115391d2 JB |
4215 | if (!atomic_inc_not_zero(&exists->refs)) { |
4216 | spin_unlock(&tree->buffer_lock); | |
4217 | radix_tree_preload_end(); | |
115391d2 JB |
4218 | exists = NULL; |
4219 | goto again; | |
4220 | } | |
6af118ce | 4221 | spin_unlock(&tree->buffer_lock); |
19fe0a8b | 4222 | radix_tree_preload_end(); |
5df4235e | 4223 | mark_extent_buffer_accessed(exists); |
6af118ce CM |
4224 | goto free_eb; |
4225 | } | |
6af118ce | 4226 | /* add one reference for the tree */ |
3083ee2e | 4227 | spin_lock(&eb->refs_lock); |
0b32f4bb | 4228 | check_buffer_tree_ref(eb); |
3083ee2e | 4229 | spin_unlock(&eb->refs_lock); |
f044ba78 | 4230 | spin_unlock(&tree->buffer_lock); |
19fe0a8b | 4231 | radix_tree_preload_end(); |
eb14ab8e CM |
4232 | |
4233 | /* | |
4234 | * there is a race where release page may have | |
4235 | * tried to find this extent buffer in the radix | |
4236 | * but failed. It will tell the VM it is safe to | |
4237 | * reclaim the, and it will clear the page private bit. | |
4238 | * We must make sure to set the page private bit properly | |
4239 | * after the extent buffer is in the radix tree so | |
4240 | * it doesn't get lost | |
4241 | */ | |
727011e0 CM |
4242 | SetPageChecked(eb->pages[0]); |
4243 | for (i = 1; i < num_pages; i++) { | |
4244 | p = extent_buffer_page(eb, i); | |
727011e0 CM |
4245 | ClearPageChecked(p); |
4246 | unlock_page(p); | |
4247 | } | |
4248 | unlock_page(eb->pages[0]); | |
d1310b2e CM |
4249 | return eb; |
4250 | ||
6af118ce | 4251 | free_eb: |
727011e0 CM |
4252 | for (i = 0; i < num_pages; i++) { |
4253 | if (eb->pages[i]) | |
4254 | unlock_page(eb->pages[i]); | |
4255 | } | |
eb14ab8e | 4256 | |
17de39ac | 4257 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
897ca6e9 | 4258 | btrfs_release_extent_buffer(eb); |
6af118ce | 4259 | return exists; |
d1310b2e | 4260 | } |
d1310b2e CM |
4261 | |
4262 | struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree, | |
f09d1f60 | 4263 | u64 start, unsigned long len) |
d1310b2e | 4264 | { |
d1310b2e | 4265 | struct extent_buffer *eb; |
d1310b2e | 4266 | |
19fe0a8b MX |
4267 | rcu_read_lock(); |
4268 | eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT); | |
4269 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
4270 | rcu_read_unlock(); | |
5df4235e | 4271 | mark_extent_buffer_accessed(eb); |
19fe0a8b MX |
4272 | return eb; |
4273 | } | |
4274 | rcu_read_unlock(); | |
0f9dd46c | 4275 | |
19fe0a8b | 4276 | return NULL; |
d1310b2e | 4277 | } |
d1310b2e | 4278 | |
3083ee2e JB |
4279 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
4280 | { | |
4281 | struct extent_buffer *eb = | |
4282 | container_of(head, struct extent_buffer, rcu_head); | |
4283 | ||
4284 | __free_extent_buffer(eb); | |
4285 | } | |
4286 | ||
3083ee2e JB |
4287 | /* Expects to have eb->eb_lock already held */ |
4288 | static void release_extent_buffer(struct extent_buffer *eb, gfp_t mask) | |
4289 | { | |
4290 | WARN_ON(atomic_read(&eb->refs) == 0); | |
4291 | if (atomic_dec_and_test(&eb->refs)) { | |
815a51c7 JS |
4292 | if (test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) { |
4293 | spin_unlock(&eb->refs_lock); | |
4294 | } else { | |
4295 | struct extent_io_tree *tree = eb->tree; | |
3083ee2e | 4296 | |
815a51c7 | 4297 | spin_unlock(&eb->refs_lock); |
3083ee2e | 4298 | |
815a51c7 JS |
4299 | spin_lock(&tree->buffer_lock); |
4300 | radix_tree_delete(&tree->buffer, | |
4301 | eb->start >> PAGE_CACHE_SHIFT); | |
4302 | spin_unlock(&tree->buffer_lock); | |
4303 | } | |
3083ee2e JB |
4304 | |
4305 | /* Should be safe to release our pages at this point */ | |
4306 | btrfs_release_extent_buffer_page(eb, 0); | |
4307 | ||
4308 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); | |
4309 | return; | |
4310 | } | |
4311 | spin_unlock(&eb->refs_lock); | |
4312 | } | |
4313 | ||
d1310b2e CM |
4314 | void free_extent_buffer(struct extent_buffer *eb) |
4315 | { | |
d1310b2e CM |
4316 | if (!eb) |
4317 | return; | |
4318 | ||
3083ee2e | 4319 | spin_lock(&eb->refs_lock); |
815a51c7 JS |
4320 | if (atomic_read(&eb->refs) == 2 && |
4321 | test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) | |
4322 | atomic_dec(&eb->refs); | |
4323 | ||
3083ee2e JB |
4324 | if (atomic_read(&eb->refs) == 2 && |
4325 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 4326 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
4327 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
4328 | atomic_dec(&eb->refs); | |
4329 | ||
4330 | /* | |
4331 | * I know this is terrible, but it's temporary until we stop tracking | |
4332 | * the uptodate bits and such for the extent buffers. | |
4333 | */ | |
4334 | release_extent_buffer(eb, GFP_ATOMIC); | |
4335 | } | |
4336 | ||
4337 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
4338 | { | |
4339 | if (!eb) | |
d1310b2e CM |
4340 | return; |
4341 | ||
3083ee2e JB |
4342 | spin_lock(&eb->refs_lock); |
4343 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
4344 | ||
0b32f4bb | 4345 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
4346 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
4347 | atomic_dec(&eb->refs); | |
4348 | release_extent_buffer(eb, GFP_NOFS); | |
d1310b2e | 4349 | } |
d1310b2e | 4350 | |
1d4284bd | 4351 | void clear_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 4352 | { |
d1310b2e CM |
4353 | unsigned long i; |
4354 | unsigned long num_pages; | |
4355 | struct page *page; | |
4356 | ||
d1310b2e CM |
4357 | num_pages = num_extent_pages(eb->start, eb->len); |
4358 | ||
4359 | for (i = 0; i < num_pages; i++) { | |
4360 | page = extent_buffer_page(eb, i); | |
b9473439 | 4361 | if (!PageDirty(page)) |
d2c3f4f6 CM |
4362 | continue; |
4363 | ||
a61e6f29 | 4364 | lock_page(page); |
eb14ab8e CM |
4365 | WARN_ON(!PagePrivate(page)); |
4366 | ||
d1310b2e | 4367 | clear_page_dirty_for_io(page); |
0ee0fda0 | 4368 | spin_lock_irq(&page->mapping->tree_lock); |
d1310b2e CM |
4369 | if (!PageDirty(page)) { |
4370 | radix_tree_tag_clear(&page->mapping->page_tree, | |
4371 | page_index(page), | |
4372 | PAGECACHE_TAG_DIRTY); | |
4373 | } | |
0ee0fda0 | 4374 | spin_unlock_irq(&page->mapping->tree_lock); |
bf0da8c1 | 4375 | ClearPageError(page); |
a61e6f29 | 4376 | unlock_page(page); |
d1310b2e | 4377 | } |
0b32f4bb | 4378 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 4379 | } |
d1310b2e | 4380 | |
0b32f4bb | 4381 | int set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e CM |
4382 | { |
4383 | unsigned long i; | |
4384 | unsigned long num_pages; | |
b9473439 | 4385 | int was_dirty = 0; |
d1310b2e | 4386 | |
0b32f4bb JB |
4387 | check_buffer_tree_ref(eb); |
4388 | ||
b9473439 | 4389 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 4390 | |
d1310b2e | 4391 | num_pages = num_extent_pages(eb->start, eb->len); |
3083ee2e | 4392 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
4393 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
4394 | ||
b9473439 | 4395 | for (i = 0; i < num_pages; i++) |
0b32f4bb | 4396 | set_page_dirty(extent_buffer_page(eb, i)); |
b9473439 | 4397 | return was_dirty; |
d1310b2e | 4398 | } |
d1310b2e | 4399 | |
0b32f4bb | 4400 | static int range_straddles_pages(u64 start, u64 len) |
19b6caf4 CM |
4401 | { |
4402 | if (len < PAGE_CACHE_SIZE) | |
4403 | return 1; | |
4404 | if (start & (PAGE_CACHE_SIZE - 1)) | |
4405 | return 1; | |
4406 | if ((start + len) & (PAGE_CACHE_SIZE - 1)) | |
4407 | return 1; | |
4408 | return 0; | |
4409 | } | |
4410 | ||
0b32f4bb | 4411 | int clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 CM |
4412 | { |
4413 | unsigned long i; | |
4414 | struct page *page; | |
4415 | unsigned long num_pages; | |
4416 | ||
b4ce94de | 4417 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
0b32f4bb | 4418 | num_pages = num_extent_pages(eb->start, eb->len); |
1259ab75 CM |
4419 | for (i = 0; i < num_pages; i++) { |
4420 | page = extent_buffer_page(eb, i); | |
33958dc6 CM |
4421 | if (page) |
4422 | ClearPageUptodate(page); | |
1259ab75 CM |
4423 | } |
4424 | return 0; | |
4425 | } | |
4426 | ||
0b32f4bb | 4427 | int set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e CM |
4428 | { |
4429 | unsigned long i; | |
4430 | struct page *page; | |
4431 | unsigned long num_pages; | |
4432 | ||
0b32f4bb | 4433 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 4434 | num_pages = num_extent_pages(eb->start, eb->len); |
d1310b2e CM |
4435 | for (i = 0; i < num_pages; i++) { |
4436 | page = extent_buffer_page(eb, i); | |
d1310b2e CM |
4437 | SetPageUptodate(page); |
4438 | } | |
4439 | return 0; | |
4440 | } | |
d1310b2e | 4441 | |
ce9adaa5 CM |
4442 | int extent_range_uptodate(struct extent_io_tree *tree, |
4443 | u64 start, u64 end) | |
4444 | { | |
4445 | struct page *page; | |
4446 | int ret; | |
4447 | int pg_uptodate = 1; | |
4448 | int uptodate; | |
4449 | unsigned long index; | |
4450 | ||
0b32f4bb | 4451 | if (range_straddles_pages(start, end - start + 1)) { |
19b6caf4 CM |
4452 | ret = test_range_bit(tree, start, end, |
4453 | EXTENT_UPTODATE, 1, NULL); | |
4454 | if (ret) | |
4455 | return 1; | |
4456 | } | |
d397712b | 4457 | while (start <= end) { |
ce9adaa5 CM |
4458 | index = start >> PAGE_CACHE_SHIFT; |
4459 | page = find_get_page(tree->mapping, index); | |
8bedd51b MH |
4460 | if (!page) |
4461 | return 1; | |
ce9adaa5 CM |
4462 | uptodate = PageUptodate(page); |
4463 | page_cache_release(page); | |
4464 | if (!uptodate) { | |
4465 | pg_uptodate = 0; | |
4466 | break; | |
4467 | } | |
4468 | start += PAGE_CACHE_SIZE; | |
4469 | } | |
4470 | return pg_uptodate; | |
4471 | } | |
4472 | ||
0b32f4bb | 4473 | int extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 4474 | { |
0b32f4bb | 4475 | return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
d1310b2e | 4476 | } |
d1310b2e CM |
4477 | |
4478 | int read_extent_buffer_pages(struct extent_io_tree *tree, | |
bb82ab88 | 4479 | struct extent_buffer *eb, u64 start, int wait, |
f188591e | 4480 | get_extent_t *get_extent, int mirror_num) |
d1310b2e CM |
4481 | { |
4482 | unsigned long i; | |
4483 | unsigned long start_i; | |
4484 | struct page *page; | |
4485 | int err; | |
4486 | int ret = 0; | |
ce9adaa5 CM |
4487 | int locked_pages = 0; |
4488 | int all_uptodate = 1; | |
d1310b2e | 4489 | unsigned long num_pages; |
727011e0 | 4490 | unsigned long num_reads = 0; |
a86c12c7 | 4491 | struct bio *bio = NULL; |
c8b97818 | 4492 | unsigned long bio_flags = 0; |
a86c12c7 | 4493 | |
b4ce94de | 4494 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
4495 | return 0; |
4496 | ||
d1310b2e CM |
4497 | if (start) { |
4498 | WARN_ON(start < eb->start); | |
4499 | start_i = (start >> PAGE_CACHE_SHIFT) - | |
4500 | (eb->start >> PAGE_CACHE_SHIFT); | |
4501 | } else { | |
4502 | start_i = 0; | |
4503 | } | |
4504 | ||
4505 | num_pages = num_extent_pages(eb->start, eb->len); | |
4506 | for (i = start_i; i < num_pages; i++) { | |
4507 | page = extent_buffer_page(eb, i); | |
bb82ab88 | 4508 | if (wait == WAIT_NONE) { |
2db04966 | 4509 | if (!trylock_page(page)) |
ce9adaa5 | 4510 | goto unlock_exit; |
d1310b2e CM |
4511 | } else { |
4512 | lock_page(page); | |
4513 | } | |
ce9adaa5 | 4514 | locked_pages++; |
727011e0 CM |
4515 | if (!PageUptodate(page)) { |
4516 | num_reads++; | |
ce9adaa5 | 4517 | all_uptodate = 0; |
727011e0 | 4518 | } |
ce9adaa5 CM |
4519 | } |
4520 | if (all_uptodate) { | |
4521 | if (start_i == 0) | |
b4ce94de | 4522 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
4523 | goto unlock_exit; |
4524 | } | |
4525 | ||
ea466794 | 4526 | clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags); |
5cf1ab56 | 4527 | eb->read_mirror = 0; |
0b32f4bb | 4528 | atomic_set(&eb->io_pages, num_reads); |
ce9adaa5 CM |
4529 | for (i = start_i; i < num_pages; i++) { |
4530 | page = extent_buffer_page(eb, i); | |
ce9adaa5 | 4531 | if (!PageUptodate(page)) { |
f188591e | 4532 | ClearPageError(page); |
a86c12c7 | 4533 | err = __extent_read_full_page(tree, page, |
f188591e | 4534 | get_extent, &bio, |
c8b97818 | 4535 | mirror_num, &bio_flags); |
d397712b | 4536 | if (err) |
d1310b2e | 4537 | ret = err; |
d1310b2e CM |
4538 | } else { |
4539 | unlock_page(page); | |
4540 | } | |
4541 | } | |
4542 | ||
355808c2 JM |
4543 | if (bio) { |
4544 | err = submit_one_bio(READ, bio, mirror_num, bio_flags); | |
79787eaa JM |
4545 | if (err) |
4546 | return err; | |
355808c2 | 4547 | } |
a86c12c7 | 4548 | |
bb82ab88 | 4549 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 4550 | return ret; |
d397712b | 4551 | |
d1310b2e CM |
4552 | for (i = start_i; i < num_pages; i++) { |
4553 | page = extent_buffer_page(eb, i); | |
4554 | wait_on_page_locked(page); | |
d397712b | 4555 | if (!PageUptodate(page)) |
d1310b2e | 4556 | ret = -EIO; |
d1310b2e | 4557 | } |
d397712b | 4558 | |
d1310b2e | 4559 | return ret; |
ce9adaa5 CM |
4560 | |
4561 | unlock_exit: | |
4562 | i = start_i; | |
d397712b | 4563 | while (locked_pages > 0) { |
ce9adaa5 CM |
4564 | page = extent_buffer_page(eb, i); |
4565 | i++; | |
4566 | unlock_page(page); | |
4567 | locked_pages--; | |
4568 | } | |
4569 | return ret; | |
d1310b2e | 4570 | } |
d1310b2e CM |
4571 | |
4572 | void read_extent_buffer(struct extent_buffer *eb, void *dstv, | |
4573 | unsigned long start, | |
4574 | unsigned long len) | |
4575 | { | |
4576 | size_t cur; | |
4577 | size_t offset; | |
4578 | struct page *page; | |
4579 | char *kaddr; | |
4580 | char *dst = (char *)dstv; | |
4581 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4582 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
d1310b2e CM |
4583 | |
4584 | WARN_ON(start > eb->len); | |
4585 | WARN_ON(start + len > eb->start + eb->len); | |
4586 | ||
4587 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4588 | ||
d397712b | 4589 | while (len > 0) { |
d1310b2e | 4590 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
4591 | |
4592 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
a6591715 | 4593 | kaddr = page_address(page); |
d1310b2e | 4594 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
4595 | |
4596 | dst += cur; | |
4597 | len -= cur; | |
4598 | offset = 0; | |
4599 | i++; | |
4600 | } | |
4601 | } | |
d1310b2e CM |
4602 | |
4603 | int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, | |
a6591715 | 4604 | unsigned long min_len, char **map, |
d1310b2e | 4605 | unsigned long *map_start, |
a6591715 | 4606 | unsigned long *map_len) |
d1310b2e CM |
4607 | { |
4608 | size_t offset = start & (PAGE_CACHE_SIZE - 1); | |
4609 | char *kaddr; | |
4610 | struct page *p; | |
4611 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4612 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4613 | unsigned long end_i = (start_offset + start + min_len - 1) >> | |
4614 | PAGE_CACHE_SHIFT; | |
4615 | ||
4616 | if (i != end_i) | |
4617 | return -EINVAL; | |
4618 | ||
4619 | if (i == 0) { | |
4620 | offset = start_offset; | |
4621 | *map_start = 0; | |
4622 | } else { | |
4623 | offset = 0; | |
4624 | *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; | |
4625 | } | |
d397712b | 4626 | |
d1310b2e | 4627 | if (start + min_len > eb->len) { |
d397712b CM |
4628 | printk(KERN_ERR "btrfs bad mapping eb start %llu len %lu, " |
4629 | "wanted %lu %lu\n", (unsigned long long)eb->start, | |
4630 | eb->len, start, min_len); | |
d1310b2e | 4631 | WARN_ON(1); |
85026533 | 4632 | return -EINVAL; |
d1310b2e CM |
4633 | } |
4634 | ||
4635 | p = extent_buffer_page(eb, i); | |
a6591715 | 4636 | kaddr = page_address(p); |
d1310b2e CM |
4637 | *map = kaddr + offset; |
4638 | *map_len = PAGE_CACHE_SIZE - offset; | |
4639 | return 0; | |
4640 | } | |
d1310b2e | 4641 | |
d1310b2e CM |
4642 | int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, |
4643 | unsigned long start, | |
4644 | unsigned long len) | |
4645 | { | |
4646 | size_t cur; | |
4647 | size_t offset; | |
4648 | struct page *page; | |
4649 | char *kaddr; | |
4650 | char *ptr = (char *)ptrv; | |
4651 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4652 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4653 | int ret = 0; | |
4654 | ||
4655 | WARN_ON(start > eb->len); | |
4656 | WARN_ON(start + len > eb->start + eb->len); | |
4657 | ||
4658 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4659 | ||
d397712b | 4660 | while (len > 0) { |
d1310b2e | 4661 | page = extent_buffer_page(eb, i); |
d1310b2e CM |
4662 | |
4663 | cur = min(len, (PAGE_CACHE_SIZE - offset)); | |
4664 | ||
a6591715 | 4665 | kaddr = page_address(page); |
d1310b2e | 4666 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
4667 | if (ret) |
4668 | break; | |
4669 | ||
4670 | ptr += cur; | |
4671 | len -= cur; | |
4672 | offset = 0; | |
4673 | i++; | |
4674 | } | |
4675 | return ret; | |
4676 | } | |
d1310b2e CM |
4677 | |
4678 | void write_extent_buffer(struct extent_buffer *eb, const void *srcv, | |
4679 | unsigned long start, unsigned long len) | |
4680 | { | |
4681 | size_t cur; | |
4682 | size_t offset; | |
4683 | struct page *page; | |
4684 | char *kaddr; | |
4685 | char *src = (char *)srcv; | |
4686 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4687 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4688 | ||
4689 | WARN_ON(start > eb->len); | |
4690 | WARN_ON(start + len > eb->start + eb->len); | |
4691 | ||
4692 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4693 | ||
d397712b | 4694 | while (len > 0) { |
d1310b2e CM |
4695 | page = extent_buffer_page(eb, i); |
4696 | WARN_ON(!PageUptodate(page)); | |
4697 | ||
4698 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 4699 | kaddr = page_address(page); |
d1310b2e | 4700 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
4701 | |
4702 | src += cur; | |
4703 | len -= cur; | |
4704 | offset = 0; | |
4705 | i++; | |
4706 | } | |
4707 | } | |
d1310b2e CM |
4708 | |
4709 | void memset_extent_buffer(struct extent_buffer *eb, char c, | |
4710 | unsigned long start, unsigned long len) | |
4711 | { | |
4712 | size_t cur; | |
4713 | size_t offset; | |
4714 | struct page *page; | |
4715 | char *kaddr; | |
4716 | size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4717 | unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; | |
4718 | ||
4719 | WARN_ON(start > eb->len); | |
4720 | WARN_ON(start + len > eb->start + eb->len); | |
4721 | ||
4722 | offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4723 | ||
d397712b | 4724 | while (len > 0) { |
d1310b2e CM |
4725 | page = extent_buffer_page(eb, i); |
4726 | WARN_ON(!PageUptodate(page)); | |
4727 | ||
4728 | cur = min(len, PAGE_CACHE_SIZE - offset); | |
a6591715 | 4729 | kaddr = page_address(page); |
d1310b2e | 4730 | memset(kaddr + offset, c, cur); |
d1310b2e CM |
4731 | |
4732 | len -= cur; | |
4733 | offset = 0; | |
4734 | i++; | |
4735 | } | |
4736 | } | |
d1310b2e CM |
4737 | |
4738 | void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, | |
4739 | unsigned long dst_offset, unsigned long src_offset, | |
4740 | unsigned long len) | |
4741 | { | |
4742 | u64 dst_len = dst->len; | |
4743 | size_t cur; | |
4744 | size_t offset; | |
4745 | struct page *page; | |
4746 | char *kaddr; | |
4747 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4748 | unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
4749 | ||
4750 | WARN_ON(src->len != dst_len); | |
4751 | ||
4752 | offset = (start_offset + dst_offset) & | |
4753 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4754 | ||
d397712b | 4755 | while (len > 0) { |
d1310b2e CM |
4756 | page = extent_buffer_page(dst, i); |
4757 | WARN_ON(!PageUptodate(page)); | |
4758 | ||
4759 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); | |
4760 | ||
a6591715 | 4761 | kaddr = page_address(page); |
d1310b2e | 4762 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
4763 | |
4764 | src_offset += cur; | |
4765 | len -= cur; | |
4766 | offset = 0; | |
4767 | i++; | |
4768 | } | |
4769 | } | |
d1310b2e CM |
4770 | |
4771 | static void move_pages(struct page *dst_page, struct page *src_page, | |
4772 | unsigned long dst_off, unsigned long src_off, | |
4773 | unsigned long len) | |
4774 | { | |
a6591715 | 4775 | char *dst_kaddr = page_address(dst_page); |
d1310b2e CM |
4776 | if (dst_page == src_page) { |
4777 | memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len); | |
4778 | } else { | |
a6591715 | 4779 | char *src_kaddr = page_address(src_page); |
d1310b2e CM |
4780 | char *p = dst_kaddr + dst_off + len; |
4781 | char *s = src_kaddr + src_off + len; | |
4782 | ||
4783 | while (len--) | |
4784 | *--p = *--s; | |
d1310b2e | 4785 | } |
d1310b2e CM |
4786 | } |
4787 | ||
3387206f ST |
4788 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
4789 | { | |
4790 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
4791 | return distance < len; | |
4792 | } | |
4793 | ||
d1310b2e CM |
4794 | static void copy_pages(struct page *dst_page, struct page *src_page, |
4795 | unsigned long dst_off, unsigned long src_off, | |
4796 | unsigned long len) | |
4797 | { | |
a6591715 | 4798 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 4799 | char *src_kaddr; |
727011e0 | 4800 | int must_memmove = 0; |
d1310b2e | 4801 | |
3387206f | 4802 | if (dst_page != src_page) { |
a6591715 | 4803 | src_kaddr = page_address(src_page); |
3387206f | 4804 | } else { |
d1310b2e | 4805 | src_kaddr = dst_kaddr; |
727011e0 CM |
4806 | if (areas_overlap(src_off, dst_off, len)) |
4807 | must_memmove = 1; | |
3387206f | 4808 | } |
d1310b2e | 4809 | |
727011e0 CM |
4810 | if (must_memmove) |
4811 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
4812 | else | |
4813 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
4814 | } |
4815 | ||
4816 | void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
4817 | unsigned long src_offset, unsigned long len) | |
4818 | { | |
4819 | size_t cur; | |
4820 | size_t dst_off_in_page; | |
4821 | size_t src_off_in_page; | |
4822 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4823 | unsigned long dst_i; | |
4824 | unsigned long src_i; | |
4825 | ||
4826 | if (src_offset + len > dst->len) { | |
d397712b CM |
4827 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
4828 | "len %lu dst len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
4829 | BUG_ON(1); |
4830 | } | |
4831 | if (dst_offset + len > dst->len) { | |
d397712b CM |
4832 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
4833 | "len %lu dst len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
4834 | BUG_ON(1); |
4835 | } | |
4836 | ||
d397712b | 4837 | while (len > 0) { |
d1310b2e CM |
4838 | dst_off_in_page = (start_offset + dst_offset) & |
4839 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4840 | src_off_in_page = (start_offset + src_offset) & | |
4841 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4842 | ||
4843 | dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; | |
4844 | src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; | |
4845 | ||
4846 | cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - | |
4847 | src_off_in_page)); | |
4848 | cur = min_t(unsigned long, cur, | |
4849 | (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); | |
4850 | ||
4851 | copy_pages(extent_buffer_page(dst, dst_i), | |
4852 | extent_buffer_page(dst, src_i), | |
4853 | dst_off_in_page, src_off_in_page, cur); | |
4854 | ||
4855 | src_offset += cur; | |
4856 | dst_offset += cur; | |
4857 | len -= cur; | |
4858 | } | |
4859 | } | |
d1310b2e CM |
4860 | |
4861 | void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, | |
4862 | unsigned long src_offset, unsigned long len) | |
4863 | { | |
4864 | size_t cur; | |
4865 | size_t dst_off_in_page; | |
4866 | size_t src_off_in_page; | |
4867 | unsigned long dst_end = dst_offset + len - 1; | |
4868 | unsigned long src_end = src_offset + len - 1; | |
4869 | size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); | |
4870 | unsigned long dst_i; | |
4871 | unsigned long src_i; | |
4872 | ||
4873 | if (src_offset + len > dst->len) { | |
d397712b CM |
4874 | printk(KERN_ERR "btrfs memmove bogus src_offset %lu move " |
4875 | "len %lu len %lu\n", src_offset, len, dst->len); | |
d1310b2e CM |
4876 | BUG_ON(1); |
4877 | } | |
4878 | if (dst_offset + len > dst->len) { | |
d397712b CM |
4879 | printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move " |
4880 | "len %lu len %lu\n", dst_offset, len, dst->len); | |
d1310b2e CM |
4881 | BUG_ON(1); |
4882 | } | |
727011e0 | 4883 | if (dst_offset < src_offset) { |
d1310b2e CM |
4884 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
4885 | return; | |
4886 | } | |
d397712b | 4887 | while (len > 0) { |
d1310b2e CM |
4888 | dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; |
4889 | src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; | |
4890 | ||
4891 | dst_off_in_page = (start_offset + dst_end) & | |
4892 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4893 | src_off_in_page = (start_offset + src_end) & | |
4894 | ((unsigned long)PAGE_CACHE_SIZE - 1); | |
4895 | ||
4896 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
4897 | cur = min(cur, dst_off_in_page + 1); | |
4898 | move_pages(extent_buffer_page(dst, dst_i), | |
4899 | extent_buffer_page(dst, src_i), | |
4900 | dst_off_in_page - cur + 1, | |
4901 | src_off_in_page - cur + 1, cur); | |
4902 | ||
4903 | dst_end -= cur; | |
4904 | src_end -= cur; | |
4905 | len -= cur; | |
4906 | } | |
4907 | } | |
6af118ce | 4908 | |
3083ee2e | 4909 | int try_release_extent_buffer(struct page *page, gfp_t mask) |
19fe0a8b | 4910 | { |
6af118ce | 4911 | struct extent_buffer *eb; |
6af118ce | 4912 | |
3083ee2e JB |
4913 | /* |
4914 | * We need to make sure noboody is attaching this page to an eb right | |
4915 | * now. | |
4916 | */ | |
4917 | spin_lock(&page->mapping->private_lock); | |
4918 | if (!PagePrivate(page)) { | |
4919 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 4920 | return 1; |
45f49bce | 4921 | } |
6af118ce | 4922 | |
3083ee2e JB |
4923 | eb = (struct extent_buffer *)page->private; |
4924 | BUG_ON(!eb); | |
19fe0a8b MX |
4925 | |
4926 | /* | |
3083ee2e JB |
4927 | * This is a little awful but should be ok, we need to make sure that |
4928 | * the eb doesn't disappear out from under us while we're looking at | |
4929 | * this page. | |
19fe0a8b | 4930 | */ |
3083ee2e | 4931 | spin_lock(&eb->refs_lock); |
0b32f4bb | 4932 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
4933 | spin_unlock(&eb->refs_lock); |
4934 | spin_unlock(&page->mapping->private_lock); | |
4935 | return 0; | |
b9473439 | 4936 | } |
3083ee2e | 4937 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 4938 | |
3083ee2e JB |
4939 | if ((mask & GFP_NOFS) == GFP_NOFS) |
4940 | mask = GFP_NOFS; | |
19fe0a8b | 4941 | |
19fe0a8b | 4942 | /* |
3083ee2e JB |
4943 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
4944 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 4945 | */ |
3083ee2e JB |
4946 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
4947 | spin_unlock(&eb->refs_lock); | |
4948 | return 0; | |
b9473439 | 4949 | } |
3083ee2e | 4950 | release_extent_buffer(eb, mask); |
19fe0a8b | 4951 | |
3083ee2e | 4952 | return 1; |
6af118ce | 4953 | } |