btrfs: drop null testing before destroy functions
[linux-2.6-block.git] / fs / btrfs / extent_io.c
CommitLineData
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>
d1310b2e
CM
7#include <linux/spinlock.h>
8#include <linux/blkdev.h>
9#include <linux/swap.h>
d1310b2e
CM
10#include <linux/writeback.h>
11#include <linux/pagevec.h>
268bb0ce 12#include <linux/prefetch.h>
90a887c9 13#include <linux/cleancache.h>
d1310b2e
CM
14#include "extent_io.h"
15#include "extent_map.h"
902b22f3
DW
16#include "ctree.h"
17#include "btrfs_inode.h"
4a54c8c1 18#include "volumes.h"
21adbd5c 19#include "check-integrity.h"
0b32f4bb 20#include "locking.h"
606686ee 21#include "rcu-string.h"
fe09e16c 22#include "backref.h"
d1310b2e 23
d1310b2e
CM
24static struct kmem_cache *extent_state_cache;
25static struct kmem_cache *extent_buffer_cache;
9be3395b 26static struct bio_set *btrfs_bioset;
d1310b2e 27
27a3507d
FM
28static inline bool extent_state_in_tree(const struct extent_state *state)
29{
30 return !RB_EMPTY_NODE(&state->rb_node);
31}
32
6d49ba1b 33#ifdef CONFIG_BTRFS_DEBUG
d1310b2e
CM
34static LIST_HEAD(buffers);
35static LIST_HEAD(states);
4bef0848 36
d397712b 37static DEFINE_SPINLOCK(leak_lock);
6d49ba1b
ES
38
39static inline
40void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
41{
42 unsigned long flags;
43
44 spin_lock_irqsave(&leak_lock, flags);
45 list_add(new, head);
46 spin_unlock_irqrestore(&leak_lock, flags);
47}
48
49static inline
50void btrfs_leak_debug_del(struct list_head *entry)
51{
52 unsigned long flags;
53
54 spin_lock_irqsave(&leak_lock, flags);
55 list_del(entry);
56 spin_unlock_irqrestore(&leak_lock, flags);
57}
58
59static inline
60void btrfs_leak_debug_check(void)
61{
62 struct extent_state *state;
63 struct extent_buffer *eb;
64
65 while (!list_empty(&states)) {
66 state = list_entry(states.next, struct extent_state, leak_list);
9ee49a04 67 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
27a3507d
FM
68 state->start, state->end, state->state,
69 extent_state_in_tree(state),
c1c9ff7c 70 atomic_read(&state->refs));
6d49ba1b
ES
71 list_del(&state->leak_list);
72 kmem_cache_free(extent_state_cache, state);
73 }
74
75 while (!list_empty(&buffers)) {
76 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
efe120a0 77 printk(KERN_ERR "BTRFS: buffer leak start %llu len %lu "
c1c9ff7c
GU
78 "refs %d\n",
79 eb->start, eb->len, atomic_read(&eb->refs));
6d49ba1b
ES
80 list_del(&eb->leak_list);
81 kmem_cache_free(extent_buffer_cache, eb);
82 }
83}
8d599ae1 84
a5dee37d
JB
85#define btrfs_debug_check_extent_io_range(tree, start, end) \
86 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
8d599ae1 87static inline void __btrfs_debug_check_extent_io_range(const char *caller,
a5dee37d 88 struct extent_io_tree *tree, u64 start, u64 end)
8d599ae1 89{
a5dee37d
JB
90 struct inode *inode;
91 u64 isize;
8d599ae1 92
a5dee37d
JB
93 if (!tree->mapping)
94 return;
8d599ae1 95
a5dee37d
JB
96 inode = tree->mapping->host;
97 isize = i_size_read(inode);
8d599ae1 98 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
94647322
DS
99 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
100 "%s: ino %llu isize %llu odd range [%llu,%llu]",
c1c9ff7c 101 caller, btrfs_ino(inode), isize, start, end);
8d599ae1
DS
102 }
103}
6d49ba1b
ES
104#else
105#define btrfs_leak_debug_add(new, head) do {} while (0)
106#define btrfs_leak_debug_del(entry) do {} while (0)
107#define btrfs_leak_debug_check() do {} while (0)
8d599ae1 108#define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
4bef0848 109#endif
d1310b2e 110
d1310b2e
CM
111#define BUFFER_LRU_MAX 64
112
113struct tree_entry {
114 u64 start;
115 u64 end;
d1310b2e
CM
116 struct rb_node rb_node;
117};
118
119struct extent_page_data {
120 struct bio *bio;
121 struct extent_io_tree *tree;
122 get_extent_t *get_extent;
de0022b9 123 unsigned long bio_flags;
771ed689
CM
124
125 /* tells writepage not to lock the state bits for this range
126 * it still does the unlocking
127 */
ffbd517d
CM
128 unsigned int extent_locked:1;
129
130 /* tells the submit_bio code to use a WRITE_SYNC */
131 unsigned int sync_io:1;
d1310b2e
CM
132};
133
d38ed27f
QW
134static void add_extent_changeset(struct extent_state *state, unsigned bits,
135 struct extent_changeset *changeset,
136 int set)
137{
138 int ret;
139
140 if (!changeset)
141 return;
142 if (set && (state->state & bits) == bits)
143 return;
fefdc557
QW
144 if (!set && (state->state & bits) == 0)
145 return;
d38ed27f
QW
146 changeset->bytes_changed += state->end - state->start + 1;
147 ret = ulist_add(changeset->range_changed, state->start, state->end,
148 GFP_ATOMIC);
149 /* ENOMEM */
150 BUG_ON(ret < 0);
151}
152
0b32f4bb 153static noinline void flush_write_bio(void *data);
c2d904e0
JM
154static inline struct btrfs_fs_info *
155tree_fs_info(struct extent_io_tree *tree)
156{
a5dee37d
JB
157 if (!tree->mapping)
158 return NULL;
c2d904e0
JM
159 return btrfs_sb(tree->mapping->host->i_sb);
160}
0b32f4bb 161
d1310b2e
CM
162int __init extent_io_init(void)
163{
837e1972 164 extent_state_cache = kmem_cache_create("btrfs_extent_state",
9601e3f6
CH
165 sizeof(struct extent_state), 0,
166 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
d1310b2e
CM
167 if (!extent_state_cache)
168 return -ENOMEM;
169
837e1972 170 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
9601e3f6
CH
171 sizeof(struct extent_buffer), 0,
172 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
d1310b2e
CM
173 if (!extent_buffer_cache)
174 goto free_state_cache;
9be3395b
CM
175
176 btrfs_bioset = bioset_create(BIO_POOL_SIZE,
177 offsetof(struct btrfs_io_bio, bio));
178 if (!btrfs_bioset)
179 goto free_buffer_cache;
b208c2f7
DW
180
181 if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE))
182 goto free_bioset;
183
d1310b2e
CM
184 return 0;
185
b208c2f7
DW
186free_bioset:
187 bioset_free(btrfs_bioset);
188 btrfs_bioset = NULL;
189
9be3395b
CM
190free_buffer_cache:
191 kmem_cache_destroy(extent_buffer_cache);
192 extent_buffer_cache = NULL;
193
d1310b2e
CM
194free_state_cache:
195 kmem_cache_destroy(extent_state_cache);
9be3395b 196 extent_state_cache = NULL;
d1310b2e
CM
197 return -ENOMEM;
198}
199
200void extent_io_exit(void)
201{
6d49ba1b 202 btrfs_leak_debug_check();
8c0a8537
KS
203
204 /*
205 * Make sure all delayed rcu free are flushed before we
206 * destroy caches.
207 */
208 rcu_barrier();
5598e900
KM
209 kmem_cache_destroy(extent_state_cache);
210 kmem_cache_destroy(extent_buffer_cache);
9be3395b
CM
211 if (btrfs_bioset)
212 bioset_free(btrfs_bioset);
d1310b2e
CM
213}
214
215void extent_io_tree_init(struct extent_io_tree *tree,
f993c883 216 struct address_space *mapping)
d1310b2e 217{
6bef4d31 218 tree->state = RB_ROOT;
d1310b2e
CM
219 tree->ops = NULL;
220 tree->dirty_bytes = 0;
70dec807 221 spin_lock_init(&tree->lock);
d1310b2e 222 tree->mapping = mapping;
d1310b2e 223}
d1310b2e 224
b2950863 225static struct extent_state *alloc_extent_state(gfp_t mask)
d1310b2e
CM
226{
227 struct extent_state *state;
d1310b2e
CM
228
229 state = kmem_cache_alloc(extent_state_cache, mask);
2b114d1d 230 if (!state)
d1310b2e
CM
231 return state;
232 state->state = 0;
47dc196a 233 state->failrec = NULL;
27a3507d 234 RB_CLEAR_NODE(&state->rb_node);
6d49ba1b 235 btrfs_leak_debug_add(&state->leak_list, &states);
d1310b2e
CM
236 atomic_set(&state->refs, 1);
237 init_waitqueue_head(&state->wq);
143bede5 238 trace_alloc_extent_state(state, mask, _RET_IP_);
d1310b2e
CM
239 return state;
240}
d1310b2e 241
4845e44f 242void free_extent_state(struct extent_state *state)
d1310b2e 243{
d1310b2e
CM
244 if (!state)
245 return;
246 if (atomic_dec_and_test(&state->refs)) {
27a3507d 247 WARN_ON(extent_state_in_tree(state));
6d49ba1b 248 btrfs_leak_debug_del(&state->leak_list);
143bede5 249 trace_free_extent_state(state, _RET_IP_);
d1310b2e
CM
250 kmem_cache_free(extent_state_cache, state);
251 }
252}
d1310b2e 253
f2071b21
FM
254static struct rb_node *tree_insert(struct rb_root *root,
255 struct rb_node *search_start,
256 u64 offset,
12cfbad9
FDBM
257 struct rb_node *node,
258 struct rb_node ***p_in,
259 struct rb_node **parent_in)
d1310b2e 260{
f2071b21 261 struct rb_node **p;
d397712b 262 struct rb_node *parent = NULL;
d1310b2e
CM
263 struct tree_entry *entry;
264
12cfbad9
FDBM
265 if (p_in && parent_in) {
266 p = *p_in;
267 parent = *parent_in;
268 goto do_insert;
269 }
270
f2071b21 271 p = search_start ? &search_start : &root->rb_node;
d397712b 272 while (*p) {
d1310b2e
CM
273 parent = *p;
274 entry = rb_entry(parent, struct tree_entry, rb_node);
275
276 if (offset < entry->start)
277 p = &(*p)->rb_left;
278 else if (offset > entry->end)
279 p = &(*p)->rb_right;
280 else
281 return parent;
282 }
283
12cfbad9 284do_insert:
d1310b2e
CM
285 rb_link_node(node, parent, p);
286 rb_insert_color(node, root);
287 return NULL;
288}
289
80ea96b1 290static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
12cfbad9
FDBM
291 struct rb_node **prev_ret,
292 struct rb_node **next_ret,
293 struct rb_node ***p_ret,
294 struct rb_node **parent_ret)
d1310b2e 295{
80ea96b1 296 struct rb_root *root = &tree->state;
12cfbad9 297 struct rb_node **n = &root->rb_node;
d1310b2e
CM
298 struct rb_node *prev = NULL;
299 struct rb_node *orig_prev = NULL;
300 struct tree_entry *entry;
301 struct tree_entry *prev_entry = NULL;
302
12cfbad9
FDBM
303 while (*n) {
304 prev = *n;
305 entry = rb_entry(prev, struct tree_entry, rb_node);
d1310b2e
CM
306 prev_entry = entry;
307
308 if (offset < entry->start)
12cfbad9 309 n = &(*n)->rb_left;
d1310b2e 310 else if (offset > entry->end)
12cfbad9 311 n = &(*n)->rb_right;
d397712b 312 else
12cfbad9 313 return *n;
d1310b2e
CM
314 }
315
12cfbad9
FDBM
316 if (p_ret)
317 *p_ret = n;
318 if (parent_ret)
319 *parent_ret = prev;
320
d1310b2e
CM
321 if (prev_ret) {
322 orig_prev = prev;
d397712b 323 while (prev && offset > prev_entry->end) {
d1310b2e
CM
324 prev = rb_next(prev);
325 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
326 }
327 *prev_ret = prev;
328 prev = orig_prev;
329 }
330
331 if (next_ret) {
332 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
d397712b 333 while (prev && offset < prev_entry->start) {
d1310b2e
CM
334 prev = rb_prev(prev);
335 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
336 }
337 *next_ret = prev;
338 }
339 return NULL;
340}
341
12cfbad9
FDBM
342static inline struct rb_node *
343tree_search_for_insert(struct extent_io_tree *tree,
344 u64 offset,
345 struct rb_node ***p_ret,
346 struct rb_node **parent_ret)
d1310b2e 347{
70dec807 348 struct rb_node *prev = NULL;
d1310b2e 349 struct rb_node *ret;
70dec807 350
12cfbad9 351 ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret);
d397712b 352 if (!ret)
d1310b2e
CM
353 return prev;
354 return ret;
355}
356
12cfbad9
FDBM
357static inline struct rb_node *tree_search(struct extent_io_tree *tree,
358 u64 offset)
359{
360 return tree_search_for_insert(tree, offset, NULL, NULL);
361}
362
9ed74f2d
JB
363static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
364 struct extent_state *other)
365{
366 if (tree->ops && tree->ops->merge_extent_hook)
367 tree->ops->merge_extent_hook(tree->mapping->host, new,
368 other);
369}
370
d1310b2e
CM
371/*
372 * utility function to look for merge candidates inside a given range.
373 * Any extents with matching state are merged together into a single
374 * extent in the tree. Extents with EXTENT_IO in their state field
375 * are not merged because the end_io handlers need to be able to do
376 * operations on them without sleeping (or doing allocations/splits).
377 *
378 * This should be called with the tree lock held.
379 */
1bf85046
JM
380static void merge_state(struct extent_io_tree *tree,
381 struct extent_state *state)
d1310b2e
CM
382{
383 struct extent_state *other;
384 struct rb_node *other_node;
385
5b21f2ed 386 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
1bf85046 387 return;
d1310b2e
CM
388
389 other_node = rb_prev(&state->rb_node);
390 if (other_node) {
391 other = rb_entry(other_node, struct extent_state, rb_node);
392 if (other->end == state->start - 1 &&
393 other->state == state->state) {
9ed74f2d 394 merge_cb(tree, state, other);
d1310b2e 395 state->start = other->start;
d1310b2e 396 rb_erase(&other->rb_node, &tree->state);
27a3507d 397 RB_CLEAR_NODE(&other->rb_node);
d1310b2e
CM
398 free_extent_state(other);
399 }
400 }
401 other_node = rb_next(&state->rb_node);
402 if (other_node) {
403 other = rb_entry(other_node, struct extent_state, rb_node);
404 if (other->start == state->end + 1 &&
405 other->state == state->state) {
9ed74f2d 406 merge_cb(tree, state, other);
df98b6e2 407 state->end = other->end;
df98b6e2 408 rb_erase(&other->rb_node, &tree->state);
27a3507d 409 RB_CLEAR_NODE(&other->rb_node);
df98b6e2 410 free_extent_state(other);
d1310b2e
CM
411 }
412 }
d1310b2e
CM
413}
414
1bf85046 415static void set_state_cb(struct extent_io_tree *tree,
9ee49a04 416 struct extent_state *state, unsigned *bits)
291d673e 417{
1bf85046
JM
418 if (tree->ops && tree->ops->set_bit_hook)
419 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
291d673e
CM
420}
421
422static void clear_state_cb(struct extent_io_tree *tree,
9ee49a04 423 struct extent_state *state, unsigned *bits)
291d673e 424{
9ed74f2d
JB
425 if (tree->ops && tree->ops->clear_bit_hook)
426 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
291d673e
CM
427}
428
3150b699 429static void set_state_bits(struct extent_io_tree *tree,
d38ed27f
QW
430 struct extent_state *state, unsigned *bits,
431 struct extent_changeset *changeset);
3150b699 432
d1310b2e
CM
433/*
434 * insert an extent_state struct into the tree. 'bits' are set on the
435 * struct before it is inserted.
436 *
437 * This may return -EEXIST if the extent is already there, in which case the
438 * state struct is freed.
439 *
440 * The tree lock is not taken internally. This is a utility function and
441 * probably isn't what you want to call (see set/clear_extent_bit).
442 */
443static int insert_state(struct extent_io_tree *tree,
444 struct extent_state *state, u64 start, u64 end,
12cfbad9
FDBM
445 struct rb_node ***p,
446 struct rb_node **parent,
d38ed27f 447 unsigned *bits, struct extent_changeset *changeset)
d1310b2e
CM
448{
449 struct rb_node *node;
450
31b1a2bd 451 if (end < start)
efe120a0 452 WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n",
c1c9ff7c 453 end, start);
d1310b2e
CM
454 state->start = start;
455 state->end = end;
9ed74f2d 456
d38ed27f 457 set_state_bits(tree, state, bits, changeset);
3150b699 458
f2071b21 459 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
d1310b2e
CM
460 if (node) {
461 struct extent_state *found;
462 found = rb_entry(node, struct extent_state, rb_node);
efe120a0 463 printk(KERN_ERR "BTRFS: found node %llu %llu on insert of "
c1c9ff7c
GU
464 "%llu %llu\n",
465 found->start, found->end, start, end);
d1310b2e
CM
466 return -EEXIST;
467 }
468 merge_state(tree, state);
469 return 0;
470}
471
1bf85046 472static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
9ed74f2d
JB
473 u64 split)
474{
475 if (tree->ops && tree->ops->split_extent_hook)
1bf85046 476 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
9ed74f2d
JB
477}
478
d1310b2e
CM
479/*
480 * split a given extent state struct in two, inserting the preallocated
481 * struct 'prealloc' as the newly created second half. 'split' indicates an
482 * offset inside 'orig' where it should be split.
483 *
484 * Before calling,
485 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
486 * are two extent state structs in the tree:
487 * prealloc: [orig->start, split - 1]
488 * orig: [ split, orig->end ]
489 *
490 * The tree locks are not taken by this function. They need to be held
491 * by the caller.
492 */
493static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
494 struct extent_state *prealloc, u64 split)
495{
496 struct rb_node *node;
9ed74f2d
JB
497
498 split_cb(tree, orig, split);
499
d1310b2e
CM
500 prealloc->start = orig->start;
501 prealloc->end = split - 1;
502 prealloc->state = orig->state;
503 orig->start = split;
504
f2071b21
FM
505 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
506 &prealloc->rb_node, NULL, NULL);
d1310b2e 507 if (node) {
d1310b2e
CM
508 free_extent_state(prealloc);
509 return -EEXIST;
510 }
511 return 0;
512}
513
cdc6a395
LZ
514static struct extent_state *next_state(struct extent_state *state)
515{
516 struct rb_node *next = rb_next(&state->rb_node);
517 if (next)
518 return rb_entry(next, struct extent_state, rb_node);
519 else
520 return NULL;
521}
522
d1310b2e
CM
523/*
524 * utility function to clear some bits in an extent state struct.
1b303fc0 525 * it will optionally wake up any one waiting on this state (wake == 1).
d1310b2e
CM
526 *
527 * If no bits are set on the state struct after clearing things, the
528 * struct is freed and removed from the tree
529 */
cdc6a395
LZ
530static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
531 struct extent_state *state,
fefdc557
QW
532 unsigned *bits, int wake,
533 struct extent_changeset *changeset)
d1310b2e 534{
cdc6a395 535 struct extent_state *next;
9ee49a04 536 unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS;
d1310b2e 537
0ca1f7ce 538 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
d1310b2e
CM
539 u64 range = state->end - state->start + 1;
540 WARN_ON(range > tree->dirty_bytes);
541 tree->dirty_bytes -= range;
542 }
291d673e 543 clear_state_cb(tree, state, bits);
fefdc557 544 add_extent_changeset(state, bits_to_clear, changeset, 0);
32c00aff 545 state->state &= ~bits_to_clear;
d1310b2e
CM
546 if (wake)
547 wake_up(&state->wq);
0ca1f7ce 548 if (state->state == 0) {
cdc6a395 549 next = next_state(state);
27a3507d 550 if (extent_state_in_tree(state)) {
d1310b2e 551 rb_erase(&state->rb_node, &tree->state);
27a3507d 552 RB_CLEAR_NODE(&state->rb_node);
d1310b2e
CM
553 free_extent_state(state);
554 } else {
555 WARN_ON(1);
556 }
557 } else {
558 merge_state(tree, state);
cdc6a395 559 next = next_state(state);
d1310b2e 560 }
cdc6a395 561 return next;
d1310b2e
CM
562}
563
8233767a
XG
564static struct extent_state *
565alloc_extent_state_atomic(struct extent_state *prealloc)
566{
567 if (!prealloc)
568 prealloc = alloc_extent_state(GFP_ATOMIC);
569
570 return prealloc;
571}
572
48a3b636 573static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
c2d904e0
JM
574{
575 btrfs_panic(tree_fs_info(tree), err, "Locking error: "
576 "Extent tree was modified by another "
577 "thread while locked.");
578}
579
d1310b2e
CM
580/*
581 * clear some bits on a range in the tree. This may require splitting
582 * or inserting elements in the tree, so the gfp mask is used to
583 * indicate which allocations or sleeping are allowed.
584 *
585 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
586 * the given range from the tree regardless of state (ie for truncate).
587 *
588 * the range [start, end] is inclusive.
589 *
6763af84 590 * This takes the tree lock, and returns 0 on success and < 0 on error.
d1310b2e 591 */
fefdc557
QW
592static int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
593 unsigned bits, int wake, int delete,
594 struct extent_state **cached_state,
595 gfp_t mask, struct extent_changeset *changeset)
d1310b2e
CM
596{
597 struct extent_state *state;
2c64c53d 598 struct extent_state *cached;
d1310b2e
CM
599 struct extent_state *prealloc = NULL;
600 struct rb_node *node;
5c939df5 601 u64 last_end;
d1310b2e 602 int err;
2ac55d41 603 int clear = 0;
d1310b2e 604
a5dee37d 605 btrfs_debug_check_extent_io_range(tree, start, end);
8d599ae1 606
7ee9e440
JB
607 if (bits & EXTENT_DELALLOC)
608 bits |= EXTENT_NORESERVE;
609
0ca1f7ce
YZ
610 if (delete)
611 bits |= ~EXTENT_CTLBITS;
612 bits |= EXTENT_FIRST_DELALLOC;
613
2ac55d41
JB
614 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
615 clear = 1;
d1310b2e 616again:
d0164adc 617 if (!prealloc && gfpflags_allow_blocking(mask)) {
c7bc6319
FM
618 /*
619 * Don't care for allocation failure here because we might end
620 * up not needing the pre-allocated extent state at all, which
621 * is the case if we only have in the tree extent states that
622 * cover our input range and don't cover too any other range.
623 * If we end up needing a new extent state we allocate it later.
624 */
d1310b2e 625 prealloc = alloc_extent_state(mask);
d1310b2e
CM
626 }
627
cad321ad 628 spin_lock(&tree->lock);
2c64c53d
CM
629 if (cached_state) {
630 cached = *cached_state;
2ac55d41
JB
631
632 if (clear) {
633 *cached_state = NULL;
634 cached_state = NULL;
635 }
636
27a3507d
FM
637 if (cached && extent_state_in_tree(cached) &&
638 cached->start <= start && cached->end > start) {
2ac55d41
JB
639 if (clear)
640 atomic_dec(&cached->refs);
2c64c53d 641 state = cached;
42daec29 642 goto hit_next;
2c64c53d 643 }
2ac55d41
JB
644 if (clear)
645 free_extent_state(cached);
2c64c53d 646 }
d1310b2e
CM
647 /*
648 * this search will find the extents that end after
649 * our range starts
650 */
80ea96b1 651 node = tree_search(tree, start);
d1310b2e
CM
652 if (!node)
653 goto out;
654 state = rb_entry(node, struct extent_state, rb_node);
2c64c53d 655hit_next:
d1310b2e
CM
656 if (state->start > end)
657 goto out;
658 WARN_ON(state->end < start);
5c939df5 659 last_end = state->end;
d1310b2e 660
0449314a 661 /* the state doesn't have the wanted bits, go ahead */
cdc6a395
LZ
662 if (!(state->state & bits)) {
663 state = next_state(state);
0449314a 664 goto next;
cdc6a395 665 }
0449314a 666
d1310b2e
CM
667 /*
668 * | ---- desired range ---- |
669 * | state | or
670 * | ------------- state -------------- |
671 *
672 * We need to split the extent we found, and may flip
673 * bits on second half.
674 *
675 * If the extent we found extends past our range, we
676 * just split and search again. It'll get split again
677 * the next time though.
678 *
679 * If the extent we found is inside our range, we clear
680 * the desired bit on it.
681 */
682
683 if (state->start < start) {
8233767a
XG
684 prealloc = alloc_extent_state_atomic(prealloc);
685 BUG_ON(!prealloc);
d1310b2e 686 err = split_state(tree, state, prealloc, start);
c2d904e0
JM
687 if (err)
688 extent_io_tree_panic(tree, err);
689
d1310b2e
CM
690 prealloc = NULL;
691 if (err)
692 goto out;
693 if (state->end <= end) {
fefdc557
QW
694 state = clear_state_bit(tree, state, &bits, wake,
695 changeset);
d1ac6e41 696 goto next;
d1310b2e
CM
697 }
698 goto search_again;
699 }
700 /*
701 * | ---- desired range ---- |
702 * | state |
703 * We need to split the extent, and clear the bit
704 * on the first half
705 */
706 if (state->start <= end && state->end > end) {
8233767a
XG
707 prealloc = alloc_extent_state_atomic(prealloc);
708 BUG_ON(!prealloc);
d1310b2e 709 err = split_state(tree, state, prealloc, end + 1);
c2d904e0
JM
710 if (err)
711 extent_io_tree_panic(tree, err);
712
d1310b2e
CM
713 if (wake)
714 wake_up(&state->wq);
42daec29 715
fefdc557 716 clear_state_bit(tree, prealloc, &bits, wake, changeset);
9ed74f2d 717
d1310b2e
CM
718 prealloc = NULL;
719 goto out;
720 }
42daec29 721
fefdc557 722 state = clear_state_bit(tree, state, &bits, wake, changeset);
0449314a 723next:
5c939df5
YZ
724 if (last_end == (u64)-1)
725 goto out;
726 start = last_end + 1;
cdc6a395 727 if (start <= end && state && !need_resched())
692e5759 728 goto hit_next;
d1310b2e
CM
729 goto search_again;
730
731out:
cad321ad 732 spin_unlock(&tree->lock);
d1310b2e
CM
733 if (prealloc)
734 free_extent_state(prealloc);
735
6763af84 736 return 0;
d1310b2e
CM
737
738search_again:
739 if (start > end)
740 goto out;
cad321ad 741 spin_unlock(&tree->lock);
d0164adc 742 if (gfpflags_allow_blocking(mask))
d1310b2e
CM
743 cond_resched();
744 goto again;
745}
d1310b2e 746
143bede5
JM
747static void wait_on_state(struct extent_io_tree *tree,
748 struct extent_state *state)
641f5219
CH
749 __releases(tree->lock)
750 __acquires(tree->lock)
d1310b2e
CM
751{
752 DEFINE_WAIT(wait);
753 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
cad321ad 754 spin_unlock(&tree->lock);
d1310b2e 755 schedule();
cad321ad 756 spin_lock(&tree->lock);
d1310b2e 757 finish_wait(&state->wq, &wait);
d1310b2e
CM
758}
759
760/*
761 * waits for one or more bits to clear on a range in the state tree.
762 * The range [start, end] is inclusive.
763 * The tree lock is taken by this function
764 */
41074888
DS
765static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
766 unsigned long bits)
d1310b2e
CM
767{
768 struct extent_state *state;
769 struct rb_node *node;
770
a5dee37d 771 btrfs_debug_check_extent_io_range(tree, start, end);
8d599ae1 772
cad321ad 773 spin_lock(&tree->lock);
d1310b2e
CM
774again:
775 while (1) {
776 /*
777 * this search will find all the extents that end after
778 * our range starts
779 */
80ea96b1 780 node = tree_search(tree, start);
c50d3e71 781process_node:
d1310b2e
CM
782 if (!node)
783 break;
784
785 state = rb_entry(node, struct extent_state, rb_node);
786
787 if (state->start > end)
788 goto out;
789
790 if (state->state & bits) {
791 start = state->start;
792 atomic_inc(&state->refs);
793 wait_on_state(tree, state);
794 free_extent_state(state);
795 goto again;
796 }
797 start = state->end + 1;
798
799 if (start > end)
800 break;
801
c50d3e71
FM
802 if (!cond_resched_lock(&tree->lock)) {
803 node = rb_next(node);
804 goto process_node;
805 }
d1310b2e
CM
806 }
807out:
cad321ad 808 spin_unlock(&tree->lock);
d1310b2e 809}
d1310b2e 810
1bf85046 811static void set_state_bits(struct extent_io_tree *tree,
d1310b2e 812 struct extent_state *state,
d38ed27f 813 unsigned *bits, struct extent_changeset *changeset)
d1310b2e 814{
9ee49a04 815 unsigned bits_to_set = *bits & ~EXTENT_CTLBITS;
9ed74f2d 816
1bf85046 817 set_state_cb(tree, state, bits);
0ca1f7ce 818 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
d1310b2e
CM
819 u64 range = state->end - state->start + 1;
820 tree->dirty_bytes += range;
821 }
d38ed27f 822 add_extent_changeset(state, bits_to_set, changeset, 1);
0ca1f7ce 823 state->state |= bits_to_set;
d1310b2e
CM
824}
825
e38e2ed7
FM
826static void cache_state_if_flags(struct extent_state *state,
827 struct extent_state **cached_ptr,
9ee49a04 828 unsigned flags)
2c64c53d
CM
829{
830 if (cached_ptr && !(*cached_ptr)) {
e38e2ed7 831 if (!flags || (state->state & flags)) {
2c64c53d
CM
832 *cached_ptr = state;
833 atomic_inc(&state->refs);
834 }
835 }
836}
837
e38e2ed7
FM
838static void cache_state(struct extent_state *state,
839 struct extent_state **cached_ptr)
840{
841 return cache_state_if_flags(state, cached_ptr,
842 EXTENT_IOBITS | EXTENT_BOUNDARY);
843}
844
d1310b2e 845/*
1edbb734
CM
846 * set some bits on a range in the tree. This may require allocations or
847 * sleeping, so the gfp mask is used to indicate what is allowed.
d1310b2e 848 *
1edbb734
CM
849 * If any of the exclusive bits are set, this will fail with -EEXIST if some
850 * part of the range already has the desired bits set. The start of the
851 * existing range is returned in failed_start in this case.
d1310b2e 852 *
1edbb734 853 * [start, end] is inclusive This takes the tree lock.
d1310b2e 854 */
1edbb734 855
3fbe5c02
JM
856static int __must_check
857__set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
9ee49a04 858 unsigned bits, unsigned exclusive_bits,
41074888 859 u64 *failed_start, struct extent_state **cached_state,
d38ed27f 860 gfp_t mask, struct extent_changeset *changeset)
d1310b2e
CM
861{
862 struct extent_state *state;
863 struct extent_state *prealloc = NULL;
864 struct rb_node *node;
12cfbad9
FDBM
865 struct rb_node **p;
866 struct rb_node *parent;
d1310b2e 867 int err = 0;
d1310b2e
CM
868 u64 last_start;
869 u64 last_end;
42daec29 870
a5dee37d 871 btrfs_debug_check_extent_io_range(tree, start, end);
8d599ae1 872
0ca1f7ce 873 bits |= EXTENT_FIRST_DELALLOC;
d1310b2e 874again:
d0164adc 875 if (!prealloc && gfpflags_allow_blocking(mask)) {
d1310b2e 876 prealloc = alloc_extent_state(mask);
8233767a 877 BUG_ON(!prealloc);
d1310b2e
CM
878 }
879
cad321ad 880 spin_lock(&tree->lock);
9655d298
CM
881 if (cached_state && *cached_state) {
882 state = *cached_state;
df98b6e2 883 if (state->start <= start && state->end > start &&
27a3507d 884 extent_state_in_tree(state)) {
9655d298
CM
885 node = &state->rb_node;
886 goto hit_next;
887 }
888 }
d1310b2e
CM
889 /*
890 * this search will find all the extents that end after
891 * our range starts.
892 */
12cfbad9 893 node = tree_search_for_insert(tree, start, &p, &parent);
d1310b2e 894 if (!node) {
8233767a
XG
895 prealloc = alloc_extent_state_atomic(prealloc);
896 BUG_ON(!prealloc);
12cfbad9 897 err = insert_state(tree, prealloc, start, end,
d38ed27f 898 &p, &parent, &bits, changeset);
c2d904e0
JM
899 if (err)
900 extent_io_tree_panic(tree, err);
901
c42ac0bc 902 cache_state(prealloc, cached_state);
d1310b2e 903 prealloc = NULL;
d1310b2e
CM
904 goto out;
905 }
d1310b2e 906 state = rb_entry(node, struct extent_state, rb_node);
40431d6c 907hit_next:
d1310b2e
CM
908 last_start = state->start;
909 last_end = state->end;
910
911 /*
912 * | ---- desired range ---- |
913 * | state |
914 *
915 * Just lock what we found and keep going
916 */
917 if (state->start == start && state->end <= end) {
1edbb734 918 if (state->state & exclusive_bits) {
d1310b2e
CM
919 *failed_start = state->start;
920 err = -EEXIST;
921 goto out;
922 }
42daec29 923
d38ed27f 924 set_state_bits(tree, state, &bits, changeset);
2c64c53d 925 cache_state(state, cached_state);
d1310b2e 926 merge_state(tree, state);
5c939df5
YZ
927 if (last_end == (u64)-1)
928 goto out;
929 start = last_end + 1;
d1ac6e41
LB
930 state = next_state(state);
931 if (start < end && state && state->start == start &&
932 !need_resched())
933 goto hit_next;
d1310b2e
CM
934 goto search_again;
935 }
936
937 /*
938 * | ---- desired range ---- |
939 * | state |
940 * or
941 * | ------------- state -------------- |
942 *
943 * We need to split the extent we found, and may flip bits on
944 * second half.
945 *
946 * If the extent we found extends past our
947 * range, we just split and search again. It'll get split
948 * again the next time though.
949 *
950 * If the extent we found is inside our range, we set the
951 * desired bit on it.
952 */
953 if (state->start < start) {
1edbb734 954 if (state->state & exclusive_bits) {
d1310b2e
CM
955 *failed_start = start;
956 err = -EEXIST;
957 goto out;
958 }
8233767a
XG
959
960 prealloc = alloc_extent_state_atomic(prealloc);
961 BUG_ON(!prealloc);
d1310b2e 962 err = split_state(tree, state, prealloc, start);
c2d904e0
JM
963 if (err)
964 extent_io_tree_panic(tree, err);
965
d1310b2e
CM
966 prealloc = NULL;
967 if (err)
968 goto out;
969 if (state->end <= end) {
d38ed27f 970 set_state_bits(tree, state, &bits, changeset);
2c64c53d 971 cache_state(state, cached_state);
d1310b2e 972 merge_state(tree, state);
5c939df5
YZ
973 if (last_end == (u64)-1)
974 goto out;
975 start = last_end + 1;
d1ac6e41
LB
976 state = next_state(state);
977 if (start < end && state && state->start == start &&
978 !need_resched())
979 goto hit_next;
d1310b2e
CM
980 }
981 goto search_again;
982 }
983 /*
984 * | ---- desired range ---- |
985 * | state | or | state |
986 *
987 * There's a hole, we need to insert something in it and
988 * ignore the extent we found.
989 */
990 if (state->start > start) {
991 u64 this_end;
992 if (end < last_start)
993 this_end = end;
994 else
d397712b 995 this_end = last_start - 1;
8233767a
XG
996
997 prealloc = alloc_extent_state_atomic(prealloc);
998 BUG_ON(!prealloc);
c7f895a2
XG
999
1000 /*
1001 * Avoid to free 'prealloc' if it can be merged with
1002 * the later extent.
1003 */
d1310b2e 1004 err = insert_state(tree, prealloc, start, this_end,
d38ed27f 1005 NULL, NULL, &bits, changeset);
c2d904e0
JM
1006 if (err)
1007 extent_io_tree_panic(tree, err);
1008
9ed74f2d
JB
1009 cache_state(prealloc, cached_state);
1010 prealloc = NULL;
d1310b2e
CM
1011 start = this_end + 1;
1012 goto search_again;
1013 }
1014 /*
1015 * | ---- desired range ---- |
1016 * | state |
1017 * We need to split the extent, and set the bit
1018 * on the first half
1019 */
1020 if (state->start <= end && state->end > end) {
1edbb734 1021 if (state->state & exclusive_bits) {
d1310b2e
CM
1022 *failed_start = start;
1023 err = -EEXIST;
1024 goto out;
1025 }
8233767a
XG
1026
1027 prealloc = alloc_extent_state_atomic(prealloc);
1028 BUG_ON(!prealloc);
d1310b2e 1029 err = split_state(tree, state, prealloc, end + 1);
c2d904e0
JM
1030 if (err)
1031 extent_io_tree_panic(tree, err);
d1310b2e 1032
d38ed27f 1033 set_state_bits(tree, prealloc, &bits, changeset);
2c64c53d 1034 cache_state(prealloc, cached_state);
d1310b2e
CM
1035 merge_state(tree, prealloc);
1036 prealloc = NULL;
1037 goto out;
1038 }
1039
1040 goto search_again;
1041
1042out:
cad321ad 1043 spin_unlock(&tree->lock);
d1310b2e
CM
1044 if (prealloc)
1045 free_extent_state(prealloc);
1046
1047 return err;
1048
1049search_again:
1050 if (start > end)
1051 goto out;
cad321ad 1052 spin_unlock(&tree->lock);
d0164adc 1053 if (gfpflags_allow_blocking(mask))
d1310b2e
CM
1054 cond_resched();
1055 goto again;
1056}
d1310b2e 1057
41074888 1058int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
9ee49a04 1059 unsigned bits, u64 * failed_start,
41074888 1060 struct extent_state **cached_state, gfp_t mask)
3fbe5c02
JM
1061{
1062 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
d38ed27f 1063 cached_state, mask, NULL);
3fbe5c02
JM
1064}
1065
1066
462d6fac 1067/**
10983f2e
LB
1068 * convert_extent_bit - convert all bits in a given range from one bit to
1069 * another
462d6fac
JB
1070 * @tree: the io tree to search
1071 * @start: the start offset in bytes
1072 * @end: the end offset in bytes (inclusive)
1073 * @bits: the bits to set in this range
1074 * @clear_bits: the bits to clear in this range
e6138876 1075 * @cached_state: state that we're going to cache
462d6fac
JB
1076 * @mask: the allocation mask
1077 *
1078 * This will go through and set bits for the given range. If any states exist
1079 * already in this range they are set with the given bit and cleared of the
1080 * clear_bits. This is only meant to be used by things that are mergeable, ie
1081 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1082 * boundary bits like LOCK.
1083 */
1084int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
9ee49a04 1085 unsigned bits, unsigned clear_bits,
e6138876 1086 struct extent_state **cached_state, gfp_t mask)
462d6fac
JB
1087{
1088 struct extent_state *state;
1089 struct extent_state *prealloc = NULL;
1090 struct rb_node *node;
12cfbad9
FDBM
1091 struct rb_node **p;
1092 struct rb_node *parent;
462d6fac
JB
1093 int err = 0;
1094 u64 last_start;
1095 u64 last_end;
c8fd3de7 1096 bool first_iteration = true;
462d6fac 1097
a5dee37d 1098 btrfs_debug_check_extent_io_range(tree, start, end);
8d599ae1 1099
462d6fac 1100again:
d0164adc 1101 if (!prealloc && gfpflags_allow_blocking(mask)) {
c8fd3de7
FM
1102 /*
1103 * Best effort, don't worry if extent state allocation fails
1104 * here for the first iteration. We might have a cached state
1105 * that matches exactly the target range, in which case no
1106 * extent state allocations are needed. We'll only know this
1107 * after locking the tree.
1108 */
462d6fac 1109 prealloc = alloc_extent_state(mask);
c8fd3de7 1110 if (!prealloc && !first_iteration)
462d6fac
JB
1111 return -ENOMEM;
1112 }
1113
1114 spin_lock(&tree->lock);
e6138876
JB
1115 if (cached_state && *cached_state) {
1116 state = *cached_state;
1117 if (state->start <= start && state->end > start &&
27a3507d 1118 extent_state_in_tree(state)) {
e6138876
JB
1119 node = &state->rb_node;
1120 goto hit_next;
1121 }
1122 }
1123
462d6fac
JB
1124 /*
1125 * this search will find all the extents that end after
1126 * our range starts.
1127 */
12cfbad9 1128 node = tree_search_for_insert(tree, start, &p, &parent);
462d6fac
JB
1129 if (!node) {
1130 prealloc = alloc_extent_state_atomic(prealloc);
1cf4ffdb
LB
1131 if (!prealloc) {
1132 err = -ENOMEM;
1133 goto out;
1134 }
12cfbad9 1135 err = insert_state(tree, prealloc, start, end,
d38ed27f 1136 &p, &parent, &bits, NULL);
c2d904e0
JM
1137 if (err)
1138 extent_io_tree_panic(tree, err);
c42ac0bc
FDBM
1139 cache_state(prealloc, cached_state);
1140 prealloc = NULL;
462d6fac
JB
1141 goto out;
1142 }
1143 state = rb_entry(node, struct extent_state, rb_node);
1144hit_next:
1145 last_start = state->start;
1146 last_end = state->end;
1147
1148 /*
1149 * | ---- desired range ---- |
1150 * | state |
1151 *
1152 * Just lock what we found and keep going
1153 */
1154 if (state->start == start && state->end <= end) {
d38ed27f 1155 set_state_bits(tree, state, &bits, NULL);
e6138876 1156 cache_state(state, cached_state);
fefdc557 1157 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
462d6fac
JB
1158 if (last_end == (u64)-1)
1159 goto out;
462d6fac 1160 start = last_end + 1;
d1ac6e41
LB
1161 if (start < end && state && state->start == start &&
1162 !need_resched())
1163 goto hit_next;
462d6fac
JB
1164 goto search_again;
1165 }
1166
1167 /*
1168 * | ---- desired range ---- |
1169 * | state |
1170 * or
1171 * | ------------- state -------------- |
1172 *
1173 * We need to split the extent we found, and may flip bits on
1174 * second half.
1175 *
1176 * If the extent we found extends past our
1177 * range, we just split and search again. It'll get split
1178 * again the next time though.
1179 *
1180 * If the extent we found is inside our range, we set the
1181 * desired bit on it.
1182 */
1183 if (state->start < start) {
1184 prealloc = alloc_extent_state_atomic(prealloc);
1cf4ffdb
LB
1185 if (!prealloc) {
1186 err = -ENOMEM;
1187 goto out;
1188 }
462d6fac 1189 err = split_state(tree, state, prealloc, start);
c2d904e0
JM
1190 if (err)
1191 extent_io_tree_panic(tree, err);
462d6fac
JB
1192 prealloc = NULL;
1193 if (err)
1194 goto out;
1195 if (state->end <= end) {
d38ed27f 1196 set_state_bits(tree, state, &bits, NULL);
e6138876 1197 cache_state(state, cached_state);
fefdc557
QW
1198 state = clear_state_bit(tree, state, &clear_bits, 0,
1199 NULL);
462d6fac
JB
1200 if (last_end == (u64)-1)
1201 goto out;
1202 start = last_end + 1;
d1ac6e41
LB
1203 if (start < end && state && state->start == start &&
1204 !need_resched())
1205 goto hit_next;
462d6fac
JB
1206 }
1207 goto search_again;
1208 }
1209 /*
1210 * | ---- desired range ---- |
1211 * | state | or | state |
1212 *
1213 * There's a hole, we need to insert something in it and
1214 * ignore the extent we found.
1215 */
1216 if (state->start > start) {
1217 u64 this_end;
1218 if (end < last_start)
1219 this_end = end;
1220 else
1221 this_end = last_start - 1;
1222
1223 prealloc = alloc_extent_state_atomic(prealloc);
1cf4ffdb
LB
1224 if (!prealloc) {
1225 err = -ENOMEM;
1226 goto out;
1227 }
462d6fac
JB
1228
1229 /*
1230 * Avoid to free 'prealloc' if it can be merged with
1231 * the later extent.
1232 */
1233 err = insert_state(tree, prealloc, start, this_end,
d38ed27f 1234 NULL, NULL, &bits, NULL);
c2d904e0
JM
1235 if (err)
1236 extent_io_tree_panic(tree, err);
e6138876 1237 cache_state(prealloc, cached_state);
462d6fac
JB
1238 prealloc = NULL;
1239 start = this_end + 1;
1240 goto search_again;
1241 }
1242 /*
1243 * | ---- desired range ---- |
1244 * | state |
1245 * We need to split the extent, and set the bit
1246 * on the first half
1247 */
1248 if (state->start <= end && state->end > end) {
1249 prealloc = alloc_extent_state_atomic(prealloc);
1cf4ffdb
LB
1250 if (!prealloc) {
1251 err = -ENOMEM;
1252 goto out;
1253 }
462d6fac
JB
1254
1255 err = split_state(tree, state, prealloc, end + 1);
c2d904e0
JM
1256 if (err)
1257 extent_io_tree_panic(tree, err);
462d6fac 1258
d38ed27f 1259 set_state_bits(tree, prealloc, &bits, NULL);
e6138876 1260 cache_state(prealloc, cached_state);
fefdc557 1261 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
462d6fac
JB
1262 prealloc = NULL;
1263 goto out;
1264 }
1265
1266 goto search_again;
1267
1268out:
1269 spin_unlock(&tree->lock);
1270 if (prealloc)
1271 free_extent_state(prealloc);
1272
1273 return err;
1274
1275search_again:
1276 if (start > end)
1277 goto out;
1278 spin_unlock(&tree->lock);
d0164adc 1279 if (gfpflags_allow_blocking(mask))
462d6fac 1280 cond_resched();
c8fd3de7 1281 first_iteration = false;
462d6fac
JB
1282 goto again;
1283}
1284
d1310b2e 1285/* wrappers around set/clear extent bit */
d38ed27f
QW
1286int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1287 unsigned bits, gfp_t mask,
1288 struct extent_changeset *changeset)
1289{
1290 /*
1291 * We don't support EXTENT_LOCKED yet, as current changeset will
1292 * record any bits changed, so for EXTENT_LOCKED case, it will
1293 * either fail with -EEXIST or changeset will record the whole
1294 * range.
1295 */
1296 BUG_ON(bits & EXTENT_LOCKED);
1297
1298 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, mask,
1299 changeset);
1300}
1301
fefdc557
QW
1302int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1303 unsigned bits, int wake, int delete,
1304 struct extent_state **cached, gfp_t mask)
1305{
1306 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1307 cached, mask, NULL);
1308}
1309
fefdc557
QW
1310int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1311 unsigned bits, gfp_t mask,
1312 struct extent_changeset *changeset)
1313{
1314 /*
1315 * Don't support EXTENT_LOCKED case, same reason as
1316 * set_record_extent_bits().
1317 */
1318 BUG_ON(bits & EXTENT_LOCKED);
1319
1320 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask,
1321 changeset);
1322}
1323
d352ac68
CM
1324/*
1325 * either insert or lock state struct between start and end use mask to tell
1326 * us if waiting is desired.
1327 */
1edbb734 1328int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
ff13db41 1329 struct extent_state **cached_state)
d1310b2e
CM
1330{
1331 int err;
1332 u64 failed_start;
9ee49a04 1333
d1310b2e 1334 while (1) {
ff13db41 1335 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
3fbe5c02 1336 EXTENT_LOCKED, &failed_start,
d38ed27f 1337 cached_state, GFP_NOFS, NULL);
d0082371 1338 if (err == -EEXIST) {
d1310b2e
CM
1339 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1340 start = failed_start;
d0082371 1341 } else
d1310b2e 1342 break;
d1310b2e
CM
1343 WARN_ON(start > end);
1344 }
1345 return err;
1346}
d1310b2e 1347
d0082371 1348int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
25179201
JB
1349{
1350 int err;
1351 u64 failed_start;
1352
3fbe5c02 1353 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
d38ed27f 1354 &failed_start, NULL, GFP_NOFS, NULL);
6643558d
YZ
1355 if (err == -EEXIST) {
1356 if (failed_start > start)
1357 clear_extent_bit(tree, start, failed_start - 1,
d0082371 1358 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
25179201 1359 return 0;
6643558d 1360 }
25179201
JB
1361 return 1;
1362}
25179201 1363
bd1fa4f0 1364void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
4adaa611
CM
1365{
1366 unsigned long index = start >> PAGE_CACHE_SHIFT;
1367 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1368 struct page *page;
1369
1370 while (index <= end_index) {
1371 page = find_get_page(inode->i_mapping, index);
1372 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1373 clear_page_dirty_for_io(page);
1374 page_cache_release(page);
1375 index++;
1376 }
4adaa611
CM
1377}
1378
f6311572 1379void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
4adaa611
CM
1380{
1381 unsigned long index = start >> PAGE_CACHE_SHIFT;
1382 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1383 struct page *page;
1384
1385 while (index <= end_index) {
1386 page = find_get_page(inode->i_mapping, index);
1387 BUG_ON(!page); /* Pages should be in the extent_io_tree */
4adaa611 1388 __set_page_dirty_nobuffers(page);
8d38633c 1389 account_page_redirty(page);
4adaa611
CM
1390 page_cache_release(page);
1391 index++;
1392 }
4adaa611
CM
1393}
1394
d1310b2e
CM
1395/*
1396 * helper function to set both pages and extents in the tree writeback
1397 */
35de6db2 1398static void set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
d1310b2e
CM
1399{
1400 unsigned long index = start >> PAGE_CACHE_SHIFT;
1401 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1402 struct page *page;
1403
1404 while (index <= end_index) {
1405 page = find_get_page(tree->mapping, index);
79787eaa 1406 BUG_ON(!page); /* Pages should be in the extent_io_tree */
d1310b2e
CM
1407 set_page_writeback(page);
1408 page_cache_release(page);
1409 index++;
1410 }
d1310b2e 1411}
d1310b2e 1412
d352ac68
CM
1413/* find the first state struct with 'bits' set after 'start', and
1414 * return it. tree->lock must be held. NULL will returned if
1415 * nothing was found after 'start'
1416 */
48a3b636
ES
1417static struct extent_state *
1418find_first_extent_bit_state(struct extent_io_tree *tree,
9ee49a04 1419 u64 start, unsigned bits)
d7fc640e
CM
1420{
1421 struct rb_node *node;
1422 struct extent_state *state;
1423
1424 /*
1425 * this search will find all the extents that end after
1426 * our range starts.
1427 */
1428 node = tree_search(tree, start);
d397712b 1429 if (!node)
d7fc640e 1430 goto out;
d7fc640e 1431
d397712b 1432 while (1) {
d7fc640e 1433 state = rb_entry(node, struct extent_state, rb_node);
d397712b 1434 if (state->end >= start && (state->state & bits))
d7fc640e 1435 return state;
d397712b 1436
d7fc640e
CM
1437 node = rb_next(node);
1438 if (!node)
1439 break;
1440 }
1441out:
1442 return NULL;
1443}
d7fc640e 1444
69261c4b
XG
1445/*
1446 * find the first offset in the io tree with 'bits' set. zero is
1447 * returned if we find something, and *start_ret and *end_ret are
1448 * set to reflect the state struct that was found.
1449 *
477d7eaf 1450 * If nothing was found, 1 is returned. If found something, return 0.
69261c4b
XG
1451 */
1452int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
9ee49a04 1453 u64 *start_ret, u64 *end_ret, unsigned bits,
e6138876 1454 struct extent_state **cached_state)
69261c4b
XG
1455{
1456 struct extent_state *state;
e6138876 1457 struct rb_node *n;
69261c4b
XG
1458 int ret = 1;
1459
1460 spin_lock(&tree->lock);
e6138876
JB
1461 if (cached_state && *cached_state) {
1462 state = *cached_state;
27a3507d 1463 if (state->end == start - 1 && extent_state_in_tree(state)) {
e6138876
JB
1464 n = rb_next(&state->rb_node);
1465 while (n) {
1466 state = rb_entry(n, struct extent_state,
1467 rb_node);
1468 if (state->state & bits)
1469 goto got_it;
1470 n = rb_next(n);
1471 }
1472 free_extent_state(*cached_state);
1473 *cached_state = NULL;
1474 goto out;
1475 }
1476 free_extent_state(*cached_state);
1477 *cached_state = NULL;
1478 }
1479
69261c4b 1480 state = find_first_extent_bit_state(tree, start, bits);
e6138876 1481got_it:
69261c4b 1482 if (state) {
e38e2ed7 1483 cache_state_if_flags(state, cached_state, 0);
69261c4b
XG
1484 *start_ret = state->start;
1485 *end_ret = state->end;
1486 ret = 0;
1487 }
e6138876 1488out:
69261c4b
XG
1489 spin_unlock(&tree->lock);
1490 return ret;
1491}
1492
d352ac68
CM
1493/*
1494 * find a contiguous range of bytes in the file marked as delalloc, not
1495 * more than 'max_bytes'. start and end are used to return the range,
1496 *
1497 * 1 is returned if we find something, 0 if nothing was in the tree
1498 */
c8b97818 1499static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
c2a128d2
JB
1500 u64 *start, u64 *end, u64 max_bytes,
1501 struct extent_state **cached_state)
d1310b2e
CM
1502{
1503 struct rb_node *node;
1504 struct extent_state *state;
1505 u64 cur_start = *start;
1506 u64 found = 0;
1507 u64 total_bytes = 0;
1508
cad321ad 1509 spin_lock(&tree->lock);
c8b97818 1510
d1310b2e
CM
1511 /*
1512 * this search will find all the extents that end after
1513 * our range starts.
1514 */
80ea96b1 1515 node = tree_search(tree, cur_start);
2b114d1d 1516 if (!node) {
3b951516
CM
1517 if (!found)
1518 *end = (u64)-1;
d1310b2e
CM
1519 goto out;
1520 }
1521
d397712b 1522 while (1) {
d1310b2e 1523 state = rb_entry(node, struct extent_state, rb_node);
5b21f2ed
ZY
1524 if (found && (state->start != cur_start ||
1525 (state->state & EXTENT_BOUNDARY))) {
d1310b2e
CM
1526 goto out;
1527 }
1528 if (!(state->state & EXTENT_DELALLOC)) {
1529 if (!found)
1530 *end = state->end;
1531 goto out;
1532 }
c2a128d2 1533 if (!found) {
d1310b2e 1534 *start = state->start;
c2a128d2
JB
1535 *cached_state = state;
1536 atomic_inc(&state->refs);
1537 }
d1310b2e
CM
1538 found++;
1539 *end = state->end;
1540 cur_start = state->end + 1;
1541 node = rb_next(node);
d1310b2e 1542 total_bytes += state->end - state->start + 1;
7bf811a5 1543 if (total_bytes >= max_bytes)
573aecaf 1544 break;
573aecaf 1545 if (!node)
d1310b2e
CM
1546 break;
1547 }
1548out:
cad321ad 1549 spin_unlock(&tree->lock);
d1310b2e
CM
1550 return found;
1551}
1552
143bede5
JM
1553static noinline void __unlock_for_delalloc(struct inode *inode,
1554 struct page *locked_page,
1555 u64 start, u64 end)
c8b97818
CM
1556{
1557 int ret;
1558 struct page *pages[16];
1559 unsigned long index = start >> PAGE_CACHE_SHIFT;
1560 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1561 unsigned long nr_pages = end_index - index + 1;
1562 int i;
1563
1564 if (index == locked_page->index && end_index == index)
143bede5 1565 return;
c8b97818 1566
d397712b 1567 while (nr_pages > 0) {
c8b97818 1568 ret = find_get_pages_contig(inode->i_mapping, index,
5b050f04
CM
1569 min_t(unsigned long, nr_pages,
1570 ARRAY_SIZE(pages)), pages);
c8b97818
CM
1571 for (i = 0; i < ret; i++) {
1572 if (pages[i] != locked_page)
1573 unlock_page(pages[i]);
1574 page_cache_release(pages[i]);
1575 }
1576 nr_pages -= ret;
1577 index += ret;
1578 cond_resched();
1579 }
c8b97818
CM
1580}
1581
1582static noinline int lock_delalloc_pages(struct inode *inode,
1583 struct page *locked_page,
1584 u64 delalloc_start,
1585 u64 delalloc_end)
1586{
1587 unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1588 unsigned long start_index = index;
1589 unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1590 unsigned long pages_locked = 0;
1591 struct page *pages[16];
1592 unsigned long nrpages;
1593 int ret;
1594 int i;
1595
1596 /* the caller is responsible for locking the start index */
1597 if (index == locked_page->index && index == end_index)
1598 return 0;
1599
1600 /* skip the page at the start index */
1601 nrpages = end_index - index + 1;
d397712b 1602 while (nrpages > 0) {
c8b97818 1603 ret = find_get_pages_contig(inode->i_mapping, index,
5b050f04
CM
1604 min_t(unsigned long,
1605 nrpages, ARRAY_SIZE(pages)), pages);
c8b97818
CM
1606 if (ret == 0) {
1607 ret = -EAGAIN;
1608 goto done;
1609 }
1610 /* now we have an array of pages, lock them all */
1611 for (i = 0; i < ret; i++) {
1612 /*
1613 * the caller is taking responsibility for
1614 * locked_page
1615 */
771ed689 1616 if (pages[i] != locked_page) {
c8b97818 1617 lock_page(pages[i]);
f2b1c41c
CM
1618 if (!PageDirty(pages[i]) ||
1619 pages[i]->mapping != inode->i_mapping) {
771ed689
CM
1620 ret = -EAGAIN;
1621 unlock_page(pages[i]);
1622 page_cache_release(pages[i]);
1623 goto done;
1624 }
1625 }
c8b97818 1626 page_cache_release(pages[i]);
771ed689 1627 pages_locked++;
c8b97818 1628 }
c8b97818
CM
1629 nrpages -= ret;
1630 index += ret;
1631 cond_resched();
1632 }
1633 ret = 0;
1634done:
1635 if (ret && pages_locked) {
1636 __unlock_for_delalloc(inode, locked_page,
1637 delalloc_start,
1638 ((u64)(start_index + pages_locked - 1)) <<
1639 PAGE_CACHE_SHIFT);
1640 }
1641 return ret;
1642}
1643
1644/*
1645 * find a contiguous range of bytes in the file marked as delalloc, not
1646 * more than 'max_bytes'. start and end are used to return the range,
1647 *
1648 * 1 is returned if we find something, 0 if nothing was in the tree
1649 */
294e30fe
JB
1650STATIC u64 find_lock_delalloc_range(struct inode *inode,
1651 struct extent_io_tree *tree,
1652 struct page *locked_page, u64 *start,
1653 u64 *end, u64 max_bytes)
c8b97818
CM
1654{
1655 u64 delalloc_start;
1656 u64 delalloc_end;
1657 u64 found;
9655d298 1658 struct extent_state *cached_state = NULL;
c8b97818
CM
1659 int ret;
1660 int loops = 0;
1661
1662again:
1663 /* step one, find a bunch of delalloc bytes starting at start */
1664 delalloc_start = *start;
1665 delalloc_end = 0;
1666 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
c2a128d2 1667 max_bytes, &cached_state);
70b99e69 1668 if (!found || delalloc_end <= *start) {
c8b97818
CM
1669 *start = delalloc_start;
1670 *end = delalloc_end;
c2a128d2 1671 free_extent_state(cached_state);
385fe0be 1672 return 0;
c8b97818
CM
1673 }
1674
70b99e69
CM
1675 /*
1676 * start comes from the offset of locked_page. We have to lock
1677 * pages in order, so we can't process delalloc bytes before
1678 * locked_page
1679 */
d397712b 1680 if (delalloc_start < *start)
70b99e69 1681 delalloc_start = *start;
70b99e69 1682
c8b97818
CM
1683 /*
1684 * make sure to limit the number of pages we try to lock down
c8b97818 1685 */
7bf811a5
JB
1686 if (delalloc_end + 1 - delalloc_start > max_bytes)
1687 delalloc_end = delalloc_start + max_bytes - 1;
d397712b 1688
c8b97818
CM
1689 /* step two, lock all the pages after the page that has start */
1690 ret = lock_delalloc_pages(inode, locked_page,
1691 delalloc_start, delalloc_end);
1692 if (ret == -EAGAIN) {
1693 /* some of the pages are gone, lets avoid looping by
1694 * shortening the size of the delalloc range we're searching
1695 */
9655d298 1696 free_extent_state(cached_state);
7d788742 1697 cached_state = NULL;
c8b97818 1698 if (!loops) {
7bf811a5 1699 max_bytes = PAGE_CACHE_SIZE;
c8b97818
CM
1700 loops = 1;
1701 goto again;
1702 } else {
1703 found = 0;
1704 goto out_failed;
1705 }
1706 }
79787eaa 1707 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
c8b97818
CM
1708
1709 /* step three, lock the state bits for the whole range */
ff13db41 1710 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
c8b97818
CM
1711
1712 /* then test to make sure it is all still delalloc */
1713 ret = test_range_bit(tree, delalloc_start, delalloc_end,
9655d298 1714 EXTENT_DELALLOC, 1, cached_state);
c8b97818 1715 if (!ret) {
9655d298
CM
1716 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1717 &cached_state, GFP_NOFS);
c8b97818
CM
1718 __unlock_for_delalloc(inode, locked_page,
1719 delalloc_start, delalloc_end);
1720 cond_resched();
1721 goto again;
1722 }
9655d298 1723 free_extent_state(cached_state);
c8b97818
CM
1724 *start = delalloc_start;
1725 *end = delalloc_end;
1726out_failed:
1727 return found;
1728}
1729
a9d93e17 1730void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
c2790a2e 1731 struct page *locked_page,
9ee49a04 1732 unsigned clear_bits,
c2790a2e 1733 unsigned long page_ops)
c8b97818 1734{
c2790a2e 1735 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
c8b97818
CM
1736 int ret;
1737 struct page *pages[16];
1738 unsigned long index = start >> PAGE_CACHE_SHIFT;
1739 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1740 unsigned long nr_pages = end_index - index + 1;
1741 int i;
771ed689 1742
2c64c53d 1743 clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
c2790a2e 1744 if (page_ops == 0)
a9d93e17 1745 return;
c8b97818 1746
704de49d
FM
1747 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1748 mapping_set_error(inode->i_mapping, -EIO);
1749
d397712b 1750 while (nr_pages > 0) {
c8b97818 1751 ret = find_get_pages_contig(inode->i_mapping, index,
5b050f04
CM
1752 min_t(unsigned long,
1753 nr_pages, ARRAY_SIZE(pages)), pages);
c8b97818 1754 for (i = 0; i < ret; i++) {
8b62b72b 1755
c2790a2e 1756 if (page_ops & PAGE_SET_PRIVATE2)
8b62b72b
CM
1757 SetPagePrivate2(pages[i]);
1758
c8b97818
CM
1759 if (pages[i] == locked_page) {
1760 page_cache_release(pages[i]);
1761 continue;
1762 }
c2790a2e 1763 if (page_ops & PAGE_CLEAR_DIRTY)
c8b97818 1764 clear_page_dirty_for_io(pages[i]);
c2790a2e 1765 if (page_ops & PAGE_SET_WRITEBACK)
c8b97818 1766 set_page_writeback(pages[i]);
704de49d
FM
1767 if (page_ops & PAGE_SET_ERROR)
1768 SetPageError(pages[i]);
c2790a2e 1769 if (page_ops & PAGE_END_WRITEBACK)
c8b97818 1770 end_page_writeback(pages[i]);
c2790a2e 1771 if (page_ops & PAGE_UNLOCK)
771ed689 1772 unlock_page(pages[i]);
c8b97818
CM
1773 page_cache_release(pages[i]);
1774 }
1775 nr_pages -= ret;
1776 index += ret;
1777 cond_resched();
1778 }
c8b97818 1779}
c8b97818 1780
d352ac68
CM
1781/*
1782 * count the number of bytes in the tree that have a given bit(s)
1783 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1784 * cached. The total number found is returned.
1785 */
d1310b2e
CM
1786u64 count_range_bits(struct extent_io_tree *tree,
1787 u64 *start, u64 search_end, u64 max_bytes,
9ee49a04 1788 unsigned bits, int contig)
d1310b2e
CM
1789{
1790 struct rb_node *node;
1791 struct extent_state *state;
1792 u64 cur_start = *start;
1793 u64 total_bytes = 0;
ec29ed5b 1794 u64 last = 0;
d1310b2e
CM
1795 int found = 0;
1796
fae7f21c 1797 if (WARN_ON(search_end <= cur_start))
d1310b2e 1798 return 0;
d1310b2e 1799
cad321ad 1800 spin_lock(&tree->lock);
d1310b2e
CM
1801 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1802 total_bytes = tree->dirty_bytes;
1803 goto out;
1804 }
1805 /*
1806 * this search will find all the extents that end after
1807 * our range starts.
1808 */
80ea96b1 1809 node = tree_search(tree, cur_start);
d397712b 1810 if (!node)
d1310b2e 1811 goto out;
d1310b2e 1812
d397712b 1813 while (1) {
d1310b2e
CM
1814 state = rb_entry(node, struct extent_state, rb_node);
1815 if (state->start > search_end)
1816 break;
ec29ed5b
CM
1817 if (contig && found && state->start > last + 1)
1818 break;
1819 if (state->end >= cur_start && (state->state & bits) == bits) {
d1310b2e
CM
1820 total_bytes += min(search_end, state->end) + 1 -
1821 max(cur_start, state->start);
1822 if (total_bytes >= max_bytes)
1823 break;
1824 if (!found) {
af60bed2 1825 *start = max(cur_start, state->start);
d1310b2e
CM
1826 found = 1;
1827 }
ec29ed5b
CM
1828 last = state->end;
1829 } else if (contig && found) {
1830 break;
d1310b2e
CM
1831 }
1832 node = rb_next(node);
1833 if (!node)
1834 break;
1835 }
1836out:
cad321ad 1837 spin_unlock(&tree->lock);
d1310b2e
CM
1838 return total_bytes;
1839}
b2950863 1840
d352ac68
CM
1841/*
1842 * set the private field for a given byte offset in the tree. If there isn't
1843 * an extent_state there already, this does nothing.
1844 */
47dc196a
DS
1845static int set_state_failrec(struct extent_io_tree *tree, u64 start,
1846 struct io_failure_record *failrec)
d1310b2e
CM
1847{
1848 struct rb_node *node;
1849 struct extent_state *state;
1850 int ret = 0;
1851
cad321ad 1852 spin_lock(&tree->lock);
d1310b2e
CM
1853 /*
1854 * this search will find all the extents that end after
1855 * our range starts.
1856 */
80ea96b1 1857 node = tree_search(tree, start);
2b114d1d 1858 if (!node) {
d1310b2e
CM
1859 ret = -ENOENT;
1860 goto out;
1861 }
1862 state = rb_entry(node, struct extent_state, rb_node);
1863 if (state->start != start) {
1864 ret = -ENOENT;
1865 goto out;
1866 }
47dc196a 1867 state->failrec = failrec;
d1310b2e 1868out:
cad321ad 1869 spin_unlock(&tree->lock);
d1310b2e
CM
1870 return ret;
1871}
1872
47dc196a
DS
1873static int get_state_failrec(struct extent_io_tree *tree, u64 start,
1874 struct io_failure_record **failrec)
d1310b2e
CM
1875{
1876 struct rb_node *node;
1877 struct extent_state *state;
1878 int ret = 0;
1879
cad321ad 1880 spin_lock(&tree->lock);
d1310b2e
CM
1881 /*
1882 * this search will find all the extents that end after
1883 * our range starts.
1884 */
80ea96b1 1885 node = tree_search(tree, start);
2b114d1d 1886 if (!node) {
d1310b2e
CM
1887 ret = -ENOENT;
1888 goto out;
1889 }
1890 state = rb_entry(node, struct extent_state, rb_node);
1891 if (state->start != start) {
1892 ret = -ENOENT;
1893 goto out;
1894 }
47dc196a 1895 *failrec = state->failrec;
d1310b2e 1896out:
cad321ad 1897 spin_unlock(&tree->lock);
d1310b2e
CM
1898 return ret;
1899}
1900
1901/*
1902 * searches a range in the state tree for a given mask.
70dec807 1903 * If 'filled' == 1, this returns 1 only if every extent in the tree
d1310b2e
CM
1904 * has the bits set. Otherwise, 1 is returned if any bit in the
1905 * range is found set.
1906 */
1907int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
9ee49a04 1908 unsigned bits, int filled, struct extent_state *cached)
d1310b2e
CM
1909{
1910 struct extent_state *state = NULL;
1911 struct rb_node *node;
1912 int bitset = 0;
d1310b2e 1913
cad321ad 1914 spin_lock(&tree->lock);
27a3507d 1915 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
df98b6e2 1916 cached->end > start)
9655d298
CM
1917 node = &cached->rb_node;
1918 else
1919 node = tree_search(tree, start);
d1310b2e
CM
1920 while (node && start <= end) {
1921 state = rb_entry(node, struct extent_state, rb_node);
1922
1923 if (filled && state->start > start) {
1924 bitset = 0;
1925 break;
1926 }
1927
1928 if (state->start > end)
1929 break;
1930
1931 if (state->state & bits) {
1932 bitset = 1;
1933 if (!filled)
1934 break;
1935 } else if (filled) {
1936 bitset = 0;
1937 break;
1938 }
46562cec
CM
1939
1940 if (state->end == (u64)-1)
1941 break;
1942
d1310b2e
CM
1943 start = state->end + 1;
1944 if (start > end)
1945 break;
1946 node = rb_next(node);
1947 if (!node) {
1948 if (filled)
1949 bitset = 0;
1950 break;
1951 }
1952 }
cad321ad 1953 spin_unlock(&tree->lock);
d1310b2e
CM
1954 return bitset;
1955}
d1310b2e
CM
1956
1957/*
1958 * helper function to set a given page up to date if all the
1959 * extents in the tree for that page are up to date
1960 */
143bede5 1961static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
d1310b2e 1962{
4eee4fa4 1963 u64 start = page_offset(page);
d1310b2e 1964 u64 end = start + PAGE_CACHE_SIZE - 1;
9655d298 1965 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
d1310b2e 1966 SetPageUptodate(page);
d1310b2e
CM
1967}
1968
8b110e39 1969int free_io_failure(struct inode *inode, struct io_failure_record *rec)
4a54c8c1
JS
1970{
1971 int ret;
1972 int err = 0;
1973 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1974
47dc196a 1975 set_state_failrec(failure_tree, rec->start, NULL);
4a54c8c1
JS
1976 ret = clear_extent_bits(failure_tree, rec->start,
1977 rec->start + rec->len - 1,
1978 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1979 if (ret)
1980 err = ret;
1981
53b381b3
DW
1982 ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1983 rec->start + rec->len - 1,
1984 EXTENT_DAMAGED, GFP_NOFS);
1985 if (ret && !err)
1986 err = ret;
4a54c8c1
JS
1987
1988 kfree(rec);
1989 return err;
1990}
1991
4a54c8c1
JS
1992/*
1993 * this bypasses the standard btrfs submit functions deliberately, as
1994 * the standard behavior is to write all copies in a raid setup. here we only
1995 * want to write the one bad copy. so we do the mapping for ourselves and issue
1996 * submit_bio directly.
3ec706c8 1997 * to avoid any synchronization issues, wait for the data after writing, which
4a54c8c1
JS
1998 * actually prevents the read that triggered the error from finishing.
1999 * currently, there can be no more than two copies of every data bit. thus,
2000 * exactly one rewrite is required.
2001 */
1203b681
MX
2002int repair_io_failure(struct inode *inode, u64 start, u64 length, u64 logical,
2003 struct page *page, unsigned int pg_offset, int mirror_num)
4a54c8c1 2004{
1203b681 2005 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4a54c8c1
JS
2006 struct bio *bio;
2007 struct btrfs_device *dev;
4a54c8c1
JS
2008 u64 map_length = 0;
2009 u64 sector;
2010 struct btrfs_bio *bbio = NULL;
53b381b3 2011 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4a54c8c1
JS
2012 int ret;
2013
908960c6 2014 ASSERT(!(fs_info->sb->s_flags & MS_RDONLY));
4a54c8c1
JS
2015 BUG_ON(!mirror_num);
2016
53b381b3
DW
2017 /* we can't repair anything in raid56 yet */
2018 if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num))
2019 return 0;
2020
9be3395b 2021 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
4a54c8c1
JS
2022 if (!bio)
2023 return -EIO;
4f024f37 2024 bio->bi_iter.bi_size = 0;
4a54c8c1
JS
2025 map_length = length;
2026
3ec706c8 2027 ret = btrfs_map_block(fs_info, WRITE, logical,
4a54c8c1
JS
2028 &map_length, &bbio, mirror_num);
2029 if (ret) {
2030 bio_put(bio);
2031 return -EIO;
2032 }
2033 BUG_ON(mirror_num != bbio->mirror_num);
2034 sector = bbio->stripes[mirror_num-1].physical >> 9;
4f024f37 2035 bio->bi_iter.bi_sector = sector;
4a54c8c1 2036 dev = bbio->stripes[mirror_num-1].dev;
6e9606d2 2037 btrfs_put_bbio(bbio);
4a54c8c1
JS
2038 if (!dev || !dev->bdev || !dev->writeable) {
2039 bio_put(bio);
2040 return -EIO;
2041 }
2042 bio->bi_bdev = dev->bdev;
ffdd2018 2043 bio_add_page(bio, page, length, pg_offset);
4a54c8c1 2044
33879d45 2045 if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) {
4a54c8c1
JS
2046 /* try to remap that extent elsewhere? */
2047 bio_put(bio);
442a4f63 2048 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
4a54c8c1
JS
2049 return -EIO;
2050 }
2051
b14af3b4
DS
2052 btrfs_info_rl_in_rcu(fs_info,
2053 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
1203b681
MX
2054 btrfs_ino(inode), start,
2055 rcu_str_deref(dev->name), sector);
4a54c8c1
JS
2056 bio_put(bio);
2057 return 0;
2058}
2059
ea466794
JB
2060int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
2061 int mirror_num)
2062{
ea466794
JB
2063 u64 start = eb->start;
2064 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
d95603b2 2065 int ret = 0;
ea466794 2066
908960c6
ID
2067 if (root->fs_info->sb->s_flags & MS_RDONLY)
2068 return -EROFS;
2069
ea466794 2070 for (i = 0; i < num_pages; i++) {
fb85fc9a 2071 struct page *p = eb->pages[i];
1203b681
MX
2072
2073 ret = repair_io_failure(root->fs_info->btree_inode, start,
2074 PAGE_CACHE_SIZE, start, p,
2075 start - page_offset(p), mirror_num);
ea466794
JB
2076 if (ret)
2077 break;
2078 start += PAGE_CACHE_SIZE;
2079 }
2080
2081 return ret;
2082}
2083
4a54c8c1
JS
2084/*
2085 * each time an IO finishes, we do a fast check in the IO failure tree
2086 * to see if we need to process or clean up an io_failure_record
2087 */
8b110e39
MX
2088int clean_io_failure(struct inode *inode, u64 start, struct page *page,
2089 unsigned int pg_offset)
4a54c8c1
JS
2090{
2091 u64 private;
4a54c8c1 2092 struct io_failure_record *failrec;
908960c6 2093 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4a54c8c1
JS
2094 struct extent_state *state;
2095 int num_copies;
4a54c8c1 2096 int ret;
4a54c8c1
JS
2097
2098 private = 0;
2099 ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
2100 (u64)-1, 1, EXTENT_DIRTY, 0);
2101 if (!ret)
2102 return 0;
2103
47dc196a
DS
2104 ret = get_state_failrec(&BTRFS_I(inode)->io_failure_tree, start,
2105 &failrec);
4a54c8c1
JS
2106 if (ret)
2107 return 0;
2108
4a54c8c1
JS
2109 BUG_ON(!failrec->this_mirror);
2110
2111 if (failrec->in_validation) {
2112 /* there was no real error, just free the record */
2113 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2114 failrec->start);
4a54c8c1
JS
2115 goto out;
2116 }
908960c6
ID
2117 if (fs_info->sb->s_flags & MS_RDONLY)
2118 goto out;
4a54c8c1
JS
2119
2120 spin_lock(&BTRFS_I(inode)->io_tree.lock);
2121 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
2122 failrec->start,
2123 EXTENT_LOCKED);
2124 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
2125
883d0de4
MX
2126 if (state && state->start <= failrec->start &&
2127 state->end >= failrec->start + failrec->len - 1) {
3ec706c8
SB
2128 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2129 failrec->len);
4a54c8c1 2130 if (num_copies > 1) {
1203b681 2131 repair_io_failure(inode, start, failrec->len,
454ff3de 2132 failrec->logical, page,
1203b681 2133 pg_offset, failrec->failed_mirror);
4a54c8c1
JS
2134 }
2135 }
2136
2137out:
454ff3de 2138 free_io_failure(inode, failrec);
4a54c8c1 2139
454ff3de 2140 return 0;
4a54c8c1
JS
2141}
2142
f612496b
MX
2143/*
2144 * Can be called when
2145 * - hold extent lock
2146 * - under ordered extent
2147 * - the inode is freeing
2148 */
2149void btrfs_free_io_failure_record(struct inode *inode, u64 start, u64 end)
2150{
2151 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2152 struct io_failure_record *failrec;
2153 struct extent_state *state, *next;
2154
2155 if (RB_EMPTY_ROOT(&failure_tree->state))
2156 return;
2157
2158 spin_lock(&failure_tree->lock);
2159 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2160 while (state) {
2161 if (state->start > end)
2162 break;
2163
2164 ASSERT(state->end <= end);
2165
2166 next = next_state(state);
2167
47dc196a 2168 failrec = state->failrec;
f612496b
MX
2169 free_extent_state(state);
2170 kfree(failrec);
2171
2172 state = next;
2173 }
2174 spin_unlock(&failure_tree->lock);
2175}
2176
2fe6303e 2177int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
47dc196a 2178 struct io_failure_record **failrec_ret)
4a54c8c1 2179{
2fe6303e 2180 struct io_failure_record *failrec;
4a54c8c1 2181 struct extent_map *em;
4a54c8c1
JS
2182 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2183 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2184 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4a54c8c1 2185 int ret;
4a54c8c1
JS
2186 u64 logical;
2187
47dc196a 2188 ret = get_state_failrec(failure_tree, start, &failrec);
4a54c8c1
JS
2189 if (ret) {
2190 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2191 if (!failrec)
2192 return -ENOMEM;
2fe6303e 2193
4a54c8c1
JS
2194 failrec->start = start;
2195 failrec->len = end - start + 1;
2196 failrec->this_mirror = 0;
2197 failrec->bio_flags = 0;
2198 failrec->in_validation = 0;
2199
2200 read_lock(&em_tree->lock);
2201 em = lookup_extent_mapping(em_tree, start, failrec->len);
2202 if (!em) {
2203 read_unlock(&em_tree->lock);
2204 kfree(failrec);
2205 return -EIO;
2206 }
2207
68ba990f 2208 if (em->start > start || em->start + em->len <= start) {
4a54c8c1
JS
2209 free_extent_map(em);
2210 em = NULL;
2211 }
2212 read_unlock(&em_tree->lock);
7a2d6a64 2213 if (!em) {
4a54c8c1
JS
2214 kfree(failrec);
2215 return -EIO;
2216 }
2fe6303e 2217
4a54c8c1
JS
2218 logical = start - em->start;
2219 logical = em->block_start + logical;
2220 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2221 logical = em->block_start;
2222 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2223 extent_set_compress_type(&failrec->bio_flags,
2224 em->compress_type);
2225 }
2fe6303e
MX
2226
2227 pr_debug("Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu\n",
2228 logical, start, failrec->len);
2229
4a54c8c1
JS
2230 failrec->logical = logical;
2231 free_extent_map(em);
2232
2233 /* set the bits in the private failure tree */
2234 ret = set_extent_bits(failure_tree, start, end,
2235 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2236 if (ret >= 0)
47dc196a 2237 ret = set_state_failrec(failure_tree, start, failrec);
4a54c8c1
JS
2238 /* set the bits in the inode's tree */
2239 if (ret >= 0)
2240 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2241 GFP_NOFS);
2242 if (ret < 0) {
2243 kfree(failrec);
2244 return ret;
2245 }
2246 } else {
2fe6303e 2247 pr_debug("Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d\n",
4a54c8c1
JS
2248 failrec->logical, failrec->start, failrec->len,
2249 failrec->in_validation);
2250 /*
2251 * when data can be on disk more than twice, add to failrec here
2252 * (e.g. with a list for failed_mirror) to make
2253 * clean_io_failure() clean all those errors at once.
2254 */
2255 }
2fe6303e
MX
2256
2257 *failrec_ret = failrec;
2258
2259 return 0;
2260}
2261
2262int btrfs_check_repairable(struct inode *inode, struct bio *failed_bio,
2263 struct io_failure_record *failrec, int failed_mirror)
2264{
2265 int num_copies;
2266
5d964051
SB
2267 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
2268 failrec->logical, failrec->len);
4a54c8c1
JS
2269 if (num_copies == 1) {
2270 /*
2271 * we only have a single copy of the data, so don't bother with
2272 * all the retry and error correction code that follows. no
2273 * matter what the error is, it is very likely to persist.
2274 */
2fe6303e 2275 pr_debug("Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
09a7f7a2 2276 num_copies, failrec->this_mirror, failed_mirror);
2fe6303e 2277 return 0;
4a54c8c1
JS
2278 }
2279
4a54c8c1
JS
2280 /*
2281 * there are two premises:
2282 * a) deliver good data to the caller
2283 * b) correct the bad sectors on disk
2284 */
2285 if (failed_bio->bi_vcnt > 1) {
2286 /*
2287 * to fulfill b), we need to know the exact failing sectors, as
2288 * we don't want to rewrite any more than the failed ones. thus,
2289 * we need separate read requests for the failed bio
2290 *
2291 * if the following BUG_ON triggers, our validation request got
2292 * merged. we need separate requests for our algorithm to work.
2293 */
2294 BUG_ON(failrec->in_validation);
2295 failrec->in_validation = 1;
2296 failrec->this_mirror = failed_mirror;
4a54c8c1
JS
2297 } else {
2298 /*
2299 * we're ready to fulfill a) and b) alongside. get a good copy
2300 * of the failed sector and if we succeed, we have setup
2301 * everything for repair_io_failure to do the rest for us.
2302 */
2303 if (failrec->in_validation) {
2304 BUG_ON(failrec->this_mirror != failed_mirror);
2305 failrec->in_validation = 0;
2306 failrec->this_mirror = 0;
2307 }
2308 failrec->failed_mirror = failed_mirror;
2309 failrec->this_mirror++;
2310 if (failrec->this_mirror == failed_mirror)
2311 failrec->this_mirror++;
4a54c8c1
JS
2312 }
2313
facc8a22 2314 if (failrec->this_mirror > num_copies) {
2fe6303e 2315 pr_debug("Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
4a54c8c1 2316 num_copies, failrec->this_mirror, failed_mirror);
2fe6303e 2317 return 0;
4a54c8c1
JS
2318 }
2319
2fe6303e
MX
2320 return 1;
2321}
2322
2323
2324struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2325 struct io_failure_record *failrec,
2326 struct page *page, int pg_offset, int icsum,
8b110e39 2327 bio_end_io_t *endio_func, void *data)
2fe6303e
MX
2328{
2329 struct bio *bio;
2330 struct btrfs_io_bio *btrfs_failed_bio;
2331 struct btrfs_io_bio *btrfs_bio;
2332
9be3395b 2333 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2fe6303e
MX
2334 if (!bio)
2335 return NULL;
2336
2337 bio->bi_end_io = endio_func;
4f024f37 2338 bio->bi_iter.bi_sector = failrec->logical >> 9;
4a54c8c1 2339 bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
4f024f37 2340 bio->bi_iter.bi_size = 0;
8b110e39 2341 bio->bi_private = data;
4a54c8c1 2342
facc8a22
MX
2343 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2344 if (btrfs_failed_bio->csum) {
2345 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2346 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2347
2348 btrfs_bio = btrfs_io_bio(bio);
2349 btrfs_bio->csum = btrfs_bio->csum_inline;
2fe6303e
MX
2350 icsum *= csum_size;
2351 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
facc8a22
MX
2352 csum_size);
2353 }
2354
2fe6303e
MX
2355 bio_add_page(bio, page, failrec->len, pg_offset);
2356
2357 return bio;
2358}
2359
2360/*
2361 * this is a generic handler for readpage errors (default
2362 * readpage_io_failed_hook). if other copies exist, read those and write back
2363 * good data to the failed position. does not investigate in remapping the
2364 * failed extent elsewhere, hoping the device will be smart enough to do this as
2365 * needed
2366 */
2367
2368static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2369 struct page *page, u64 start, u64 end,
2370 int failed_mirror)
2371{
2372 struct io_failure_record *failrec;
2373 struct inode *inode = page->mapping->host;
2374 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2375 struct bio *bio;
2376 int read_mode;
2377 int ret;
2378
2379 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2380
2381 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2382 if (ret)
2383 return ret;
2384
2385 ret = btrfs_check_repairable(inode, failed_bio, failrec, failed_mirror);
2386 if (!ret) {
2387 free_io_failure(inode, failrec);
2388 return -EIO;
2389 }
2390
2391 if (failed_bio->bi_vcnt > 1)
2392 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2393 else
2394 read_mode = READ_SYNC;
2395
2396 phy_offset >>= inode->i_sb->s_blocksize_bits;
2397 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2398 start - page_offset(page),
8b110e39
MX
2399 (int)phy_offset, failed_bio->bi_end_io,
2400 NULL);
2fe6303e
MX
2401 if (!bio) {
2402 free_io_failure(inode, failrec);
2403 return -EIO;
2404 }
4a54c8c1 2405
2fe6303e
MX
2406 pr_debug("Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d\n",
2407 read_mode, failrec->this_mirror, failrec->in_validation);
4a54c8c1 2408
013bd4c3
TI
2409 ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2410 failrec->this_mirror,
2411 failrec->bio_flags, 0);
6c387ab2 2412 if (ret) {
454ff3de 2413 free_io_failure(inode, failrec);
6c387ab2
MX
2414 bio_put(bio);
2415 }
2416
013bd4c3 2417 return ret;
4a54c8c1
JS
2418}
2419
d1310b2e
CM
2420/* lots and lots of room for performance fixes in the end_bio funcs */
2421
b5227c07 2422void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
87826df0
JM
2423{
2424 int uptodate = (err == 0);
2425 struct extent_io_tree *tree;
3e2426bd 2426 int ret = 0;
87826df0
JM
2427
2428 tree = &BTRFS_I(page->mapping->host)->io_tree;
2429
2430 if (tree->ops && tree->ops->writepage_end_io_hook) {
2431 ret = tree->ops->writepage_end_io_hook(page, start,
2432 end, NULL, uptodate);
2433 if (ret)
2434 uptodate = 0;
2435 }
2436
87826df0 2437 if (!uptodate) {
87826df0
JM
2438 ClearPageUptodate(page);
2439 SetPageError(page);
5dca6eea
LB
2440 ret = ret < 0 ? ret : -EIO;
2441 mapping_set_error(page->mapping, ret);
87826df0 2442 }
87826df0
JM
2443}
2444
d1310b2e
CM
2445/*
2446 * after a writepage IO is done, we need to:
2447 * clear the uptodate bits on error
2448 * clear the writeback bits in the extent tree for this IO
2449 * end_page_writeback if the page has no more pending IO
2450 *
2451 * Scheduling is not allowed, so the extent state tree is expected
2452 * to have one and only one object corresponding to this IO.
2453 */
4246a0b6 2454static void end_bio_extent_writepage(struct bio *bio)
d1310b2e 2455{
2c30c71b 2456 struct bio_vec *bvec;
d1310b2e
CM
2457 u64 start;
2458 u64 end;
2c30c71b 2459 int i;
d1310b2e 2460
2c30c71b 2461 bio_for_each_segment_all(bvec, bio, i) {
d1310b2e 2462 struct page *page = bvec->bv_page;
902b22f3 2463
17a5adcc
AO
2464 /* We always issue full-page reads, but if some block
2465 * in a page fails to read, blk_update_request() will
2466 * advance bv_offset and adjust bv_len to compensate.
2467 * Print a warning for nonzero offsets, and an error
2468 * if they don't add up to a full page. */
efe120a0
FH
2469 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) {
2470 if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE)
2471 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2472 "partial page write in btrfs with offset %u and length %u",
2473 bvec->bv_offset, bvec->bv_len);
2474 else
2475 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2476 "incomplete page write in btrfs with offset %u and "
2477 "length %u",
2478 bvec->bv_offset, bvec->bv_len);
2479 }
d1310b2e 2480
17a5adcc
AO
2481 start = page_offset(page);
2482 end = start + bvec->bv_offset + bvec->bv_len - 1;
d1310b2e 2483
b5227c07 2484 end_extent_writepage(page, bio->bi_error, start, end);
17a5adcc 2485 end_page_writeback(page);
2c30c71b 2486 }
2b1f55b0 2487
d1310b2e 2488 bio_put(bio);
d1310b2e
CM
2489}
2490
883d0de4
MX
2491static void
2492endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2493 int uptodate)
2494{
2495 struct extent_state *cached = NULL;
2496 u64 end = start + len - 1;
2497
2498 if (uptodate && tree->track_uptodate)
2499 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2500 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2501}
2502
d1310b2e
CM
2503/*
2504 * after a readpage IO is done, we need to:
2505 * clear the uptodate bits on error
2506 * set the uptodate bits if things worked
2507 * set the page up to date if all extents in the tree are uptodate
2508 * clear the lock bit in the extent tree
2509 * unlock the page if there are no other extents locked for it
2510 *
2511 * Scheduling is not allowed, so the extent state tree is expected
2512 * to have one and only one object corresponding to this IO.
2513 */
4246a0b6 2514static void end_bio_extent_readpage(struct bio *bio)
d1310b2e 2515{
2c30c71b 2516 struct bio_vec *bvec;
4246a0b6 2517 int uptodate = !bio->bi_error;
facc8a22 2518 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
902b22f3 2519 struct extent_io_tree *tree;
facc8a22 2520 u64 offset = 0;
d1310b2e
CM
2521 u64 start;
2522 u64 end;
facc8a22 2523 u64 len;
883d0de4
MX
2524 u64 extent_start = 0;
2525 u64 extent_len = 0;
5cf1ab56 2526 int mirror;
d1310b2e 2527 int ret;
2c30c71b 2528 int i;
d1310b2e 2529
2c30c71b 2530 bio_for_each_segment_all(bvec, bio, i) {
d1310b2e 2531 struct page *page = bvec->bv_page;
a71754fc 2532 struct inode *inode = page->mapping->host;
507903b8 2533
be3940c0 2534 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
4246a0b6
CH
2535 "mirror=%u\n", (u64)bio->bi_iter.bi_sector,
2536 bio->bi_error, io_bio->mirror_num);
a71754fc 2537 tree = &BTRFS_I(inode)->io_tree;
902b22f3 2538
17a5adcc
AO
2539 /* We always issue full-page reads, but if some block
2540 * in a page fails to read, blk_update_request() will
2541 * advance bv_offset and adjust bv_len to compensate.
2542 * Print a warning for nonzero offsets, and an error
2543 * if they don't add up to a full page. */
efe120a0
FH
2544 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) {
2545 if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE)
2546 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2547 "partial page read in btrfs with offset %u and length %u",
2548 bvec->bv_offset, bvec->bv_len);
2549 else
2550 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2551 "incomplete page read in btrfs with offset %u and "
2552 "length %u",
2553 bvec->bv_offset, bvec->bv_len);
2554 }
d1310b2e 2555
17a5adcc
AO
2556 start = page_offset(page);
2557 end = start + bvec->bv_offset + bvec->bv_len - 1;
facc8a22 2558 len = bvec->bv_len;
d1310b2e 2559
9be3395b 2560 mirror = io_bio->mirror_num;
f2a09da9
MX
2561 if (likely(uptodate && tree->ops &&
2562 tree->ops->readpage_end_io_hook)) {
facc8a22
MX
2563 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2564 page, start, end,
2565 mirror);
5ee0844d 2566 if (ret)
d1310b2e 2567 uptodate = 0;
5ee0844d 2568 else
1203b681 2569 clean_io_failure(inode, start, page, 0);
d1310b2e 2570 }
ea466794 2571
f2a09da9
MX
2572 if (likely(uptodate))
2573 goto readpage_ok;
2574
2575 if (tree->ops && tree->ops->readpage_io_failed_hook) {
5cf1ab56 2576 ret = tree->ops->readpage_io_failed_hook(page, mirror);
4246a0b6 2577 if (!ret && !bio->bi_error)
ea466794 2578 uptodate = 1;
f2a09da9 2579 } else {
f4a8e656
JS
2580 /*
2581 * The generic bio_readpage_error handles errors the
2582 * following way: If possible, new read requests are
2583 * created and submitted and will end up in
2584 * end_bio_extent_readpage as well (if we're lucky, not
2585 * in the !uptodate case). In that case it returns 0 and
2586 * we just go on with the next page in our bio. If it
2587 * can't handle the error it will return -EIO and we
2588 * remain responsible for that page.
2589 */
facc8a22
MX
2590 ret = bio_readpage_error(bio, offset, page, start, end,
2591 mirror);
7e38326f 2592 if (ret == 0) {
4246a0b6 2593 uptodate = !bio->bi_error;
38c1c2e4 2594 offset += len;
7e38326f
CM
2595 continue;
2596 }
2597 }
f2a09da9 2598readpage_ok:
883d0de4 2599 if (likely(uptodate)) {
a71754fc
JB
2600 loff_t i_size = i_size_read(inode);
2601 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
a583c026 2602 unsigned off;
a71754fc
JB
2603
2604 /* Zero out the end if this page straddles i_size */
a583c026
LB
2605 off = i_size & (PAGE_CACHE_SIZE-1);
2606 if (page->index == end_index && off)
2607 zero_user_segment(page, off, PAGE_CACHE_SIZE);
17a5adcc 2608 SetPageUptodate(page);
70dec807 2609 } else {
17a5adcc
AO
2610 ClearPageUptodate(page);
2611 SetPageError(page);
70dec807 2612 }
17a5adcc 2613 unlock_page(page);
facc8a22 2614 offset += len;
883d0de4
MX
2615
2616 if (unlikely(!uptodate)) {
2617 if (extent_len) {
2618 endio_readpage_release_extent(tree,
2619 extent_start,
2620 extent_len, 1);
2621 extent_start = 0;
2622 extent_len = 0;
2623 }
2624 endio_readpage_release_extent(tree, start,
2625 end - start + 1, 0);
2626 } else if (!extent_len) {
2627 extent_start = start;
2628 extent_len = end + 1 - start;
2629 } else if (extent_start + extent_len == start) {
2630 extent_len += end + 1 - start;
2631 } else {
2632 endio_readpage_release_extent(tree, extent_start,
2633 extent_len, uptodate);
2634 extent_start = start;
2635 extent_len = end + 1 - start;
2636 }
2c30c71b 2637 }
d1310b2e 2638
883d0de4
MX
2639 if (extent_len)
2640 endio_readpage_release_extent(tree, extent_start, extent_len,
2641 uptodate);
facc8a22 2642 if (io_bio->end_io)
4246a0b6 2643 io_bio->end_io(io_bio, bio->bi_error);
d1310b2e 2644 bio_put(bio);
d1310b2e
CM
2645}
2646
9be3395b
CM
2647/*
2648 * this allocates from the btrfs_bioset. We're returning a bio right now
2649 * but you can call btrfs_io_bio for the appropriate container_of magic
2650 */
88f794ed
MX
2651struct bio *
2652btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2653 gfp_t gfp_flags)
d1310b2e 2654{
facc8a22 2655 struct btrfs_io_bio *btrfs_bio;
d1310b2e
CM
2656 struct bio *bio;
2657
9be3395b 2658 bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset);
d1310b2e
CM
2659
2660 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
9be3395b
CM
2661 while (!bio && (nr_vecs /= 2)) {
2662 bio = bio_alloc_bioset(gfp_flags,
2663 nr_vecs, btrfs_bioset);
2664 }
d1310b2e
CM
2665 }
2666
2667 if (bio) {
2668 bio->bi_bdev = bdev;
4f024f37 2669 bio->bi_iter.bi_sector = first_sector;
facc8a22
MX
2670 btrfs_bio = btrfs_io_bio(bio);
2671 btrfs_bio->csum = NULL;
2672 btrfs_bio->csum_allocated = NULL;
2673 btrfs_bio->end_io = NULL;
d1310b2e
CM
2674 }
2675 return bio;
2676}
2677
9be3395b
CM
2678struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask)
2679{
23ea8e5a
MX
2680 struct btrfs_io_bio *btrfs_bio;
2681 struct bio *new;
9be3395b 2682
23ea8e5a
MX
2683 new = bio_clone_bioset(bio, gfp_mask, btrfs_bioset);
2684 if (new) {
2685 btrfs_bio = btrfs_io_bio(new);
2686 btrfs_bio->csum = NULL;
2687 btrfs_bio->csum_allocated = NULL;
2688 btrfs_bio->end_io = NULL;
3a9508b0
CM
2689
2690#ifdef CONFIG_BLK_CGROUP
da2f0f74
CM
2691 /* FIXME, put this into bio_clone_bioset */
2692 if (bio->bi_css)
2693 bio_associate_blkcg(new, bio->bi_css);
3a9508b0 2694#endif
23ea8e5a
MX
2695 }
2696 return new;
2697}
9be3395b
CM
2698
2699/* this also allocates from the btrfs_bioset */
2700struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
2701{
facc8a22
MX
2702 struct btrfs_io_bio *btrfs_bio;
2703 struct bio *bio;
2704
2705 bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset);
2706 if (bio) {
2707 btrfs_bio = btrfs_io_bio(bio);
2708 btrfs_bio->csum = NULL;
2709 btrfs_bio->csum_allocated = NULL;
2710 btrfs_bio->end_io = NULL;
2711 }
2712 return bio;
9be3395b
CM
2713}
2714
2715
355808c2
JM
2716static int __must_check submit_one_bio(int rw, struct bio *bio,
2717 int mirror_num, unsigned long bio_flags)
d1310b2e 2718{
d1310b2e 2719 int ret = 0;
70dec807
CM
2720 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2721 struct page *page = bvec->bv_page;
2722 struct extent_io_tree *tree = bio->bi_private;
70dec807 2723 u64 start;
70dec807 2724
4eee4fa4 2725 start = page_offset(page) + bvec->bv_offset;
70dec807 2726
902b22f3 2727 bio->bi_private = NULL;
d1310b2e
CM
2728
2729 bio_get(bio);
2730
065631f6 2731 if (tree->ops && tree->ops->submit_bio_hook)
6b82ce8d 2732 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
eaf25d93 2733 mirror_num, bio_flags, start);
0b86a832 2734 else
21adbd5c 2735 btrfsic_submit_bio(rw, bio);
4a54c8c1 2736
d1310b2e
CM
2737 bio_put(bio);
2738 return ret;
2739}
2740
64a16701 2741static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page,
3444a972
JM
2742 unsigned long offset, size_t size, struct bio *bio,
2743 unsigned long bio_flags)
2744{
2745 int ret = 0;
2746 if (tree->ops && tree->ops->merge_bio_hook)
64a16701 2747 ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio,
3444a972
JM
2748 bio_flags);
2749 BUG_ON(ret < 0);
2750 return ret;
2751
2752}
2753
d1310b2e 2754static int submit_extent_page(int rw, struct extent_io_tree *tree,
da2f0f74 2755 struct writeback_control *wbc,
d1310b2e
CM
2756 struct page *page, sector_t sector,
2757 size_t size, unsigned long offset,
2758 struct block_device *bdev,
2759 struct bio **bio_ret,
2760 unsigned long max_pages,
f188591e 2761 bio_end_io_t end_io_func,
c8b97818
CM
2762 int mirror_num,
2763 unsigned long prev_bio_flags,
005efedf
FM
2764 unsigned long bio_flags,
2765 bool force_bio_submit)
d1310b2e
CM
2766{
2767 int ret = 0;
2768 struct bio *bio;
c8b97818 2769 int contig = 0;
c8b97818 2770 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
5b050f04 2771 size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
d1310b2e
CM
2772
2773 if (bio_ret && *bio_ret) {
2774 bio = *bio_ret;
c8b97818 2775 if (old_compressed)
4f024f37 2776 contig = bio->bi_iter.bi_sector == sector;
c8b97818 2777 else
f73a1c7d 2778 contig = bio_end_sector(bio) == sector;
c8b97818
CM
2779
2780 if (prev_bio_flags != bio_flags || !contig ||
005efedf 2781 force_bio_submit ||
64a16701 2782 merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
c8b97818
CM
2783 bio_add_page(bio, page, page_size, offset) < page_size) {
2784 ret = submit_one_bio(rw, bio, mirror_num,
2785 prev_bio_flags);
289454ad
NA
2786 if (ret < 0) {
2787 *bio_ret = NULL;
79787eaa 2788 return ret;
289454ad 2789 }
d1310b2e
CM
2790 bio = NULL;
2791 } else {
da2f0f74
CM
2792 if (wbc)
2793 wbc_account_io(wbc, page, page_size);
d1310b2e
CM
2794 return 0;
2795 }
2796 }
c8b97818 2797
b54ffb73
KO
2798 bio = btrfs_bio_alloc(bdev, sector, BIO_MAX_PAGES,
2799 GFP_NOFS | __GFP_HIGH);
5df67083
TI
2800 if (!bio)
2801 return -ENOMEM;
70dec807 2802
c8b97818 2803 bio_add_page(bio, page, page_size, offset);
d1310b2e
CM
2804 bio->bi_end_io = end_io_func;
2805 bio->bi_private = tree;
da2f0f74
CM
2806 if (wbc) {
2807 wbc_init_bio(wbc, bio);
2808 wbc_account_io(wbc, page, page_size);
2809 }
70dec807 2810
d397712b 2811 if (bio_ret)
d1310b2e 2812 *bio_ret = bio;
d397712b 2813 else
c8b97818 2814 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
d1310b2e
CM
2815
2816 return ret;
2817}
2818
48a3b636
ES
2819static void attach_extent_buffer_page(struct extent_buffer *eb,
2820 struct page *page)
d1310b2e
CM
2821{
2822 if (!PagePrivate(page)) {
2823 SetPagePrivate(page);
d1310b2e 2824 page_cache_get(page);
4f2de97a
JB
2825 set_page_private(page, (unsigned long)eb);
2826 } else {
2827 WARN_ON(page->private != (unsigned long)eb);
d1310b2e
CM
2828 }
2829}
2830
4f2de97a 2831void set_page_extent_mapped(struct page *page)
d1310b2e 2832{
4f2de97a
JB
2833 if (!PagePrivate(page)) {
2834 SetPagePrivate(page);
2835 page_cache_get(page);
2836 set_page_private(page, EXTENT_PAGE_PRIVATE);
2837 }
d1310b2e
CM
2838}
2839
125bac01
MX
2840static struct extent_map *
2841__get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2842 u64 start, u64 len, get_extent_t *get_extent,
2843 struct extent_map **em_cached)
2844{
2845 struct extent_map *em;
2846
2847 if (em_cached && *em_cached) {
2848 em = *em_cached;
cbc0e928 2849 if (extent_map_in_tree(em) && start >= em->start &&
125bac01
MX
2850 start < extent_map_end(em)) {
2851 atomic_inc(&em->refs);
2852 return em;
2853 }
2854
2855 free_extent_map(em);
2856 *em_cached = NULL;
2857 }
2858
2859 em = get_extent(inode, page, pg_offset, start, len, 0);
2860 if (em_cached && !IS_ERR_OR_NULL(em)) {
2861 BUG_ON(*em_cached);
2862 atomic_inc(&em->refs);
2863 *em_cached = em;
2864 }
2865 return em;
2866}
d1310b2e
CM
2867/*
2868 * basic readpage implementation. Locked extent state structs are inserted
2869 * into the tree that are removed when the IO is done (by the end_io
2870 * handlers)
79787eaa 2871 * XXX JDM: This needs looking at to ensure proper page locking
d1310b2e 2872 */
9974090b
MX
2873static int __do_readpage(struct extent_io_tree *tree,
2874 struct page *page,
2875 get_extent_t *get_extent,
125bac01 2876 struct extent_map **em_cached,
9974090b 2877 struct bio **bio, int mirror_num,
005efedf
FM
2878 unsigned long *bio_flags, int rw,
2879 u64 *prev_em_start)
d1310b2e
CM
2880{
2881 struct inode *inode = page->mapping->host;
4eee4fa4 2882 u64 start = page_offset(page);
d1310b2e
CM
2883 u64 page_end = start + PAGE_CACHE_SIZE - 1;
2884 u64 end;
2885 u64 cur = start;
2886 u64 extent_offset;
2887 u64 last_byte = i_size_read(inode);
2888 u64 block_start;
2889 u64 cur_end;
2890 sector_t sector;
2891 struct extent_map *em;
2892 struct block_device *bdev;
2893 int ret;
2894 int nr = 0;
4b384318 2895 int parent_locked = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
306e16ce 2896 size_t pg_offset = 0;
d1310b2e 2897 size_t iosize;
c8b97818 2898 size_t disk_io_size;
d1310b2e 2899 size_t blocksize = inode->i_sb->s_blocksize;
4b384318 2900 unsigned long this_bio_flag = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
d1310b2e
CM
2901
2902 set_page_extent_mapped(page);
2903
9974090b 2904 end = page_end;
90a887c9
DM
2905 if (!PageUptodate(page)) {
2906 if (cleancache_get_page(page) == 0) {
2907 BUG_ON(blocksize != PAGE_SIZE);
9974090b 2908 unlock_extent(tree, start, end);
90a887c9
DM
2909 goto out;
2910 }
2911 }
2912
c8b97818
CM
2913 if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2914 char *userpage;
2915 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2916
2917 if (zero_offset) {
2918 iosize = PAGE_CACHE_SIZE - zero_offset;
7ac687d9 2919 userpage = kmap_atomic(page);
c8b97818
CM
2920 memset(userpage + zero_offset, 0, iosize);
2921 flush_dcache_page(page);
7ac687d9 2922 kunmap_atomic(userpage);
c8b97818
CM
2923 }
2924 }
d1310b2e 2925 while (cur <= end) {
c8f2f24b 2926 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
005efedf 2927 bool force_bio_submit = false;
c8f2f24b 2928
d1310b2e
CM
2929 if (cur >= last_byte) {
2930 char *userpage;
507903b8
AJ
2931 struct extent_state *cached = NULL;
2932
306e16ce 2933 iosize = PAGE_CACHE_SIZE - pg_offset;
7ac687d9 2934 userpage = kmap_atomic(page);
306e16ce 2935 memset(userpage + pg_offset, 0, iosize);
d1310b2e 2936 flush_dcache_page(page);
7ac687d9 2937 kunmap_atomic(userpage);
d1310b2e 2938 set_extent_uptodate(tree, cur, cur + iosize - 1,
507903b8 2939 &cached, GFP_NOFS);
4b384318
MF
2940 if (!parent_locked)
2941 unlock_extent_cached(tree, cur,
2942 cur + iosize - 1,
2943 &cached, GFP_NOFS);
d1310b2e
CM
2944 break;
2945 }
125bac01
MX
2946 em = __get_extent_map(inode, page, pg_offset, cur,
2947 end - cur + 1, get_extent, em_cached);
c704005d 2948 if (IS_ERR_OR_NULL(em)) {
d1310b2e 2949 SetPageError(page);
4b384318
MF
2950 if (!parent_locked)
2951 unlock_extent(tree, cur, end);
d1310b2e
CM
2952 break;
2953 }
d1310b2e
CM
2954 extent_offset = cur - em->start;
2955 BUG_ON(extent_map_end(em) <= cur);
2956 BUG_ON(end < cur);
2957
261507a0 2958 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4b384318 2959 this_bio_flag |= EXTENT_BIO_COMPRESSED;
261507a0
LZ
2960 extent_set_compress_type(&this_bio_flag,
2961 em->compress_type);
2962 }
c8b97818 2963
d1310b2e
CM
2964 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2965 cur_end = min(extent_map_end(em) - 1, end);
fda2832f 2966 iosize = ALIGN(iosize, blocksize);
c8b97818
CM
2967 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2968 disk_io_size = em->block_len;
2969 sector = em->block_start >> 9;
2970 } else {
2971 sector = (em->block_start + extent_offset) >> 9;
2972 disk_io_size = iosize;
2973 }
d1310b2e
CM
2974 bdev = em->bdev;
2975 block_start = em->block_start;
d899e052
YZ
2976 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2977 block_start = EXTENT_MAP_HOLE;
005efedf
FM
2978
2979 /*
2980 * If we have a file range that points to a compressed extent
2981 * and it's followed by a consecutive file range that points to
2982 * to the same compressed extent (possibly with a different
2983 * offset and/or length, so it either points to the whole extent
2984 * or only part of it), we must make sure we do not submit a
2985 * single bio to populate the pages for the 2 ranges because
2986 * this makes the compressed extent read zero out the pages
2987 * belonging to the 2nd range. Imagine the following scenario:
2988 *
2989 * File layout
2990 * [0 - 8K] [8K - 24K]
2991 * | |
2992 * | |
2993 * points to extent X, points to extent X,
2994 * offset 4K, length of 8K offset 0, length 16K
2995 *
2996 * [extent X, compressed length = 4K uncompressed length = 16K]
2997 *
2998 * If the bio to read the compressed extent covers both ranges,
2999 * it will decompress extent X into the pages belonging to the
3000 * first range and then it will stop, zeroing out the remaining
3001 * pages that belong to the other range that points to extent X.
3002 * So here we make sure we submit 2 bios, one for the first
3003 * range and another one for the third range. Both will target
3004 * the same physical extent from disk, but we can't currently
3005 * make the compressed bio endio callback populate the pages
3006 * for both ranges because each compressed bio is tightly
3007 * coupled with a single extent map, and each range can have
3008 * an extent map with a different offset value relative to the
3009 * uncompressed data of our extent and different lengths. This
3010 * is a corner case so we prioritize correctness over
3011 * non-optimal behavior (submitting 2 bios for the same extent).
3012 */
3013 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3014 prev_em_start && *prev_em_start != (u64)-1 &&
3015 *prev_em_start != em->orig_start)
3016 force_bio_submit = true;
3017
3018 if (prev_em_start)
3019 *prev_em_start = em->orig_start;
3020
d1310b2e
CM
3021 free_extent_map(em);
3022 em = NULL;
3023
3024 /* we've found a hole, just zero and go on */
3025 if (block_start == EXTENT_MAP_HOLE) {
3026 char *userpage;
507903b8
AJ
3027 struct extent_state *cached = NULL;
3028
7ac687d9 3029 userpage = kmap_atomic(page);
306e16ce 3030 memset(userpage + pg_offset, 0, iosize);
d1310b2e 3031 flush_dcache_page(page);
7ac687d9 3032 kunmap_atomic(userpage);
d1310b2e
CM
3033
3034 set_extent_uptodate(tree, cur, cur + iosize - 1,
507903b8 3035 &cached, GFP_NOFS);
5e6ecb36
FM
3036 if (parent_locked)
3037 free_extent_state(cached);
3038 else
3039 unlock_extent_cached(tree, cur,
3040 cur + iosize - 1,
3041 &cached, GFP_NOFS);
d1310b2e 3042 cur = cur + iosize;
306e16ce 3043 pg_offset += iosize;
d1310b2e
CM
3044 continue;
3045 }
3046 /* the get_extent function already copied into the page */
9655d298
CM
3047 if (test_range_bit(tree, cur, cur_end,
3048 EXTENT_UPTODATE, 1, NULL)) {
a1b32a59 3049 check_page_uptodate(tree, page);
4b384318
MF
3050 if (!parent_locked)
3051 unlock_extent(tree, cur, cur + iosize - 1);
d1310b2e 3052 cur = cur + iosize;
306e16ce 3053 pg_offset += iosize;
d1310b2e
CM
3054 continue;
3055 }
70dec807
CM
3056 /* we have an inline extent but it didn't get marked up
3057 * to date. Error out
3058 */
3059 if (block_start == EXTENT_MAP_INLINE) {
3060 SetPageError(page);
4b384318
MF
3061 if (!parent_locked)
3062 unlock_extent(tree, cur, cur + iosize - 1);
70dec807 3063 cur = cur + iosize;
306e16ce 3064 pg_offset += iosize;
70dec807
CM
3065 continue;
3066 }
d1310b2e 3067
c8f2f24b 3068 pnr -= page->index;
da2f0f74 3069 ret = submit_extent_page(rw, tree, NULL, page,
306e16ce 3070 sector, disk_io_size, pg_offset,
89642229 3071 bdev, bio, pnr,
c8b97818
CM
3072 end_bio_extent_readpage, mirror_num,
3073 *bio_flags,
005efedf
FM
3074 this_bio_flag,
3075 force_bio_submit);
c8f2f24b
JB
3076 if (!ret) {
3077 nr++;
3078 *bio_flags = this_bio_flag;
3079 } else {
d1310b2e 3080 SetPageError(page);
4b384318
MF
3081 if (!parent_locked)
3082 unlock_extent(tree, cur, cur + iosize - 1);
edd33c99 3083 }
d1310b2e 3084 cur = cur + iosize;
306e16ce 3085 pg_offset += iosize;
d1310b2e 3086 }
90a887c9 3087out:
d1310b2e
CM
3088 if (!nr) {
3089 if (!PageError(page))
3090 SetPageUptodate(page);
3091 unlock_page(page);
3092 }
3093 return 0;
3094}
3095
9974090b
MX
3096static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
3097 struct page *pages[], int nr_pages,
3098 u64 start, u64 end,
3099 get_extent_t *get_extent,
125bac01 3100 struct extent_map **em_cached,
9974090b 3101 struct bio **bio, int mirror_num,
808f80b4
FM
3102 unsigned long *bio_flags, int rw,
3103 u64 *prev_em_start)
9974090b
MX
3104{
3105 struct inode *inode;
3106 struct btrfs_ordered_extent *ordered;
3107 int index;
3108
3109 inode = pages[0]->mapping->host;
3110 while (1) {
3111 lock_extent(tree, start, end);
3112 ordered = btrfs_lookup_ordered_range(inode, start,
3113 end - start + 1);
3114 if (!ordered)
3115 break;
3116 unlock_extent(tree, start, end);
3117 btrfs_start_ordered_extent(inode, ordered, 1);
3118 btrfs_put_ordered_extent(ordered);
3119 }
3120
3121 for (index = 0; index < nr_pages; index++) {
125bac01 3122 __do_readpage(tree, pages[index], get_extent, em_cached, bio,
808f80b4 3123 mirror_num, bio_flags, rw, prev_em_start);
9974090b
MX
3124 page_cache_release(pages[index]);
3125 }
3126}
3127
3128static void __extent_readpages(struct extent_io_tree *tree,
3129 struct page *pages[],
3130 int nr_pages, get_extent_t *get_extent,
125bac01 3131 struct extent_map **em_cached,
9974090b 3132 struct bio **bio, int mirror_num,
808f80b4
FM
3133 unsigned long *bio_flags, int rw,
3134 u64 *prev_em_start)
9974090b 3135{
35a3621b 3136 u64 start = 0;
9974090b
MX
3137 u64 end = 0;
3138 u64 page_start;
3139 int index;
35a3621b 3140 int first_index = 0;
9974090b
MX
3141
3142 for (index = 0; index < nr_pages; index++) {
3143 page_start = page_offset(pages[index]);
3144 if (!end) {
3145 start = page_start;
3146 end = start + PAGE_CACHE_SIZE - 1;
3147 first_index = index;
3148 } else if (end + 1 == page_start) {
3149 end += PAGE_CACHE_SIZE;
3150 } else {
3151 __do_contiguous_readpages(tree, &pages[first_index],
3152 index - first_index, start,
125bac01
MX
3153 end, get_extent, em_cached,
3154 bio, mirror_num, bio_flags,
808f80b4 3155 rw, prev_em_start);
9974090b
MX
3156 start = page_start;
3157 end = start + PAGE_CACHE_SIZE - 1;
3158 first_index = index;
3159 }
3160 }
3161
3162 if (end)
3163 __do_contiguous_readpages(tree, &pages[first_index],
3164 index - first_index, start,
125bac01 3165 end, get_extent, em_cached, bio,
808f80b4
FM
3166 mirror_num, bio_flags, rw,
3167 prev_em_start);
9974090b
MX
3168}
3169
3170static int __extent_read_full_page(struct extent_io_tree *tree,
3171 struct page *page,
3172 get_extent_t *get_extent,
3173 struct bio **bio, int mirror_num,
3174 unsigned long *bio_flags, int rw)
3175{
3176 struct inode *inode = page->mapping->host;
3177 struct btrfs_ordered_extent *ordered;
3178 u64 start = page_offset(page);
3179 u64 end = start + PAGE_CACHE_SIZE - 1;
3180 int ret;
3181
3182 while (1) {
3183 lock_extent(tree, start, end);
3184 ordered = btrfs_lookup_ordered_extent(inode, start);
3185 if (!ordered)
3186 break;
3187 unlock_extent(tree, start, end);
3188 btrfs_start_ordered_extent(inode, ordered, 1);
3189 btrfs_put_ordered_extent(ordered);
3190 }
3191
125bac01 3192 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
005efedf 3193 bio_flags, rw, NULL);
9974090b
MX
3194 return ret;
3195}
3196
d1310b2e 3197int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
8ddc7d9c 3198 get_extent_t *get_extent, int mirror_num)
d1310b2e
CM
3199{
3200 struct bio *bio = NULL;
c8b97818 3201 unsigned long bio_flags = 0;
d1310b2e
CM
3202 int ret;
3203
8ddc7d9c 3204 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
d4c7ca86 3205 &bio_flags, READ);
d1310b2e 3206 if (bio)
8ddc7d9c 3207 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
d1310b2e
CM
3208 return ret;
3209}
d1310b2e 3210
4b384318
MF
3211int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page,
3212 get_extent_t *get_extent, int mirror_num)
3213{
3214 struct bio *bio = NULL;
3215 unsigned long bio_flags = EXTENT_BIO_PARENT_LOCKED;
3216 int ret;
3217
3218 ret = __do_readpage(tree, page, get_extent, NULL, &bio, mirror_num,
005efedf 3219 &bio_flags, READ, NULL);
4b384318
MF
3220 if (bio)
3221 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3222 return ret;
3223}
3224
11c8349b
CM
3225static noinline void update_nr_written(struct page *page,
3226 struct writeback_control *wbc,
3227 unsigned long nr_written)
3228{
3229 wbc->nr_to_write -= nr_written;
3230 if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
3231 wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
3232 page->mapping->writeback_index = page->index + nr_written;
3233}
3234
d1310b2e 3235/*
40f76580
CM
3236 * helper for __extent_writepage, doing all of the delayed allocation setup.
3237 *
3238 * This returns 1 if our fill_delalloc function did all the work required
3239 * to write the page (copy into inline extent). In this case the IO has
3240 * been started and the page is already unlocked.
3241 *
3242 * This returns 0 if all went well (page still locked)
3243 * This returns < 0 if there were errors (page still locked)
d1310b2e 3244 */
40f76580
CM
3245static noinline_for_stack int writepage_delalloc(struct inode *inode,
3246 struct page *page, struct writeback_control *wbc,
3247 struct extent_page_data *epd,
3248 u64 delalloc_start,
3249 unsigned long *nr_written)
3250{
3251 struct extent_io_tree *tree = epd->tree;
3252 u64 page_end = delalloc_start + PAGE_CACHE_SIZE - 1;
3253 u64 nr_delalloc;
3254 u64 delalloc_to_write = 0;
3255 u64 delalloc_end = 0;
3256 int ret;
3257 int page_started = 0;
3258
3259 if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc)
3260 return 0;
3261
3262 while (delalloc_end < page_end) {
3263 nr_delalloc = find_lock_delalloc_range(inode, tree,
3264 page,
3265 &delalloc_start,
3266 &delalloc_end,
dcab6a3b 3267 BTRFS_MAX_EXTENT_SIZE);
40f76580
CM
3268 if (nr_delalloc == 0) {
3269 delalloc_start = delalloc_end + 1;
3270 continue;
3271 }
3272 ret = tree->ops->fill_delalloc(inode, page,
3273 delalloc_start,
3274 delalloc_end,
3275 &page_started,
3276 nr_written);
3277 /* File system has been set read-only */
3278 if (ret) {
3279 SetPageError(page);
3280 /* fill_delalloc should be return < 0 for error
3281 * but just in case, we use > 0 here meaning the
3282 * IO is started, so we don't want to return > 0
3283 * unless things are going well.
3284 */
3285 ret = ret < 0 ? ret : -EIO;
3286 goto done;
3287 }
3288 /*
3289 * delalloc_end is already one less than the total
3290 * length, so we don't subtract one from
3291 * PAGE_CACHE_SIZE
3292 */
3293 delalloc_to_write += (delalloc_end - delalloc_start +
3294 PAGE_CACHE_SIZE) >>
3295 PAGE_CACHE_SHIFT;
3296 delalloc_start = delalloc_end + 1;
3297 }
3298 if (wbc->nr_to_write < delalloc_to_write) {
3299 int thresh = 8192;
3300
3301 if (delalloc_to_write < thresh * 2)
3302 thresh = delalloc_to_write;
3303 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3304 thresh);
3305 }
3306
3307 /* did the fill delalloc function already unlock and start
3308 * the IO?
3309 */
3310 if (page_started) {
3311 /*
3312 * we've unlocked the page, so we can't update
3313 * the mapping's writeback index, just update
3314 * nr_to_write.
3315 */
3316 wbc->nr_to_write -= *nr_written;
3317 return 1;
3318 }
3319
3320 ret = 0;
3321
3322done:
3323 return ret;
3324}
3325
3326/*
3327 * helper for __extent_writepage. This calls the writepage start hooks,
3328 * and does the loop to map the page into extents and bios.
3329 *
3330 * We return 1 if the IO is started and the page is unlocked,
3331 * 0 if all went well (page still locked)
3332 * < 0 if there were errors (page still locked)
3333 */
3334static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3335 struct page *page,
3336 struct writeback_control *wbc,
3337 struct extent_page_data *epd,
3338 loff_t i_size,
3339 unsigned long nr_written,
3340 int write_flags, int *nr_ret)
d1310b2e 3341{
d1310b2e 3342 struct extent_io_tree *tree = epd->tree;
4eee4fa4 3343 u64 start = page_offset(page);
d1310b2e
CM
3344 u64 page_end = start + PAGE_CACHE_SIZE - 1;
3345 u64 end;
3346 u64 cur = start;
3347 u64 extent_offset;
d1310b2e
CM
3348 u64 block_start;
3349 u64 iosize;
3350 sector_t sector;
2c64c53d 3351 struct extent_state *cached_state = NULL;
d1310b2e
CM
3352 struct extent_map *em;
3353 struct block_device *bdev;
7f3c74fb 3354 size_t pg_offset = 0;
d1310b2e 3355 size_t blocksize;
40f76580
CM
3356 int ret = 0;
3357 int nr = 0;
3358 bool compressed;
c8b97818 3359
247e743c 3360 if (tree->ops && tree->ops->writepage_start_hook) {
c8b97818
CM
3361 ret = tree->ops->writepage_start_hook(page, start,
3362 page_end);
87826df0
JM
3363 if (ret) {
3364 /* Fixup worker will requeue */
3365 if (ret == -EBUSY)
3366 wbc->pages_skipped++;
3367 else
3368 redirty_page_for_writepage(wbc, page);
40f76580 3369
11c8349b 3370 update_nr_written(page, wbc, nr_written);
247e743c 3371 unlock_page(page);
40f76580 3372 ret = 1;
11c8349b 3373 goto done_unlocked;
247e743c
CM
3374 }
3375 }
3376
11c8349b
CM
3377 /*
3378 * we don't want to touch the inode after unlocking the page,
3379 * so we update the mapping writeback index now
3380 */
3381 update_nr_written(page, wbc, nr_written + 1);
771ed689 3382
d1310b2e 3383 end = page_end;
40f76580 3384 if (i_size <= start) {
e6dcd2dc
CM
3385 if (tree->ops && tree->ops->writepage_end_io_hook)
3386 tree->ops->writepage_end_io_hook(page, start,
3387 page_end, NULL, 1);
d1310b2e
CM
3388 goto done;
3389 }
3390
d1310b2e
CM
3391 blocksize = inode->i_sb->s_blocksize;
3392
3393 while (cur <= end) {
40f76580
CM
3394 u64 em_end;
3395 if (cur >= i_size) {
e6dcd2dc
CM
3396 if (tree->ops && tree->ops->writepage_end_io_hook)
3397 tree->ops->writepage_end_io_hook(page, cur,
3398 page_end, NULL, 1);
d1310b2e
CM
3399 break;
3400 }
7f3c74fb 3401 em = epd->get_extent(inode, page, pg_offset, cur,
d1310b2e 3402 end - cur + 1, 1);
c704005d 3403 if (IS_ERR_OR_NULL(em)) {
d1310b2e 3404 SetPageError(page);
61391d56 3405 ret = PTR_ERR_OR_ZERO(em);
d1310b2e
CM
3406 break;
3407 }
3408
3409 extent_offset = cur - em->start;
40f76580
CM
3410 em_end = extent_map_end(em);
3411 BUG_ON(em_end <= cur);
d1310b2e 3412 BUG_ON(end < cur);
40f76580 3413 iosize = min(em_end - cur, end - cur + 1);
fda2832f 3414 iosize = ALIGN(iosize, blocksize);
d1310b2e
CM
3415 sector = (em->block_start + extent_offset) >> 9;
3416 bdev = em->bdev;
3417 block_start = em->block_start;
c8b97818 3418 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
d1310b2e
CM
3419 free_extent_map(em);
3420 em = NULL;
3421
c8b97818
CM
3422 /*
3423 * compressed and inline extents are written through other
3424 * paths in the FS
3425 */
3426 if (compressed || block_start == EXTENT_MAP_HOLE ||
d1310b2e 3427 block_start == EXTENT_MAP_INLINE) {
c8b97818
CM
3428 /*
3429 * end_io notification does not happen here for
3430 * compressed extents
3431 */
3432 if (!compressed && tree->ops &&
3433 tree->ops->writepage_end_io_hook)
e6dcd2dc
CM
3434 tree->ops->writepage_end_io_hook(page, cur,
3435 cur + iosize - 1,
3436 NULL, 1);
c8b97818
CM
3437 else if (compressed) {
3438 /* we don't want to end_page_writeback on
3439 * a compressed extent. this happens
3440 * elsewhere
3441 */
3442 nr++;
3443 }
3444
3445 cur += iosize;
7f3c74fb 3446 pg_offset += iosize;
d1310b2e
CM
3447 continue;
3448 }
c8b97818 3449
d1310b2e
CM
3450 if (tree->ops && tree->ops->writepage_io_hook) {
3451 ret = tree->ops->writepage_io_hook(page, cur,
3452 cur + iosize - 1);
3453 } else {
3454 ret = 0;
3455 }
1259ab75 3456 if (ret) {
d1310b2e 3457 SetPageError(page);
1259ab75 3458 } else {
40f76580 3459 unsigned long max_nr = (i_size >> PAGE_CACHE_SHIFT) + 1;
7f3c74fb 3460
d1310b2e
CM
3461 set_range_writeback(tree, cur, cur + iosize - 1);
3462 if (!PageWriteback(page)) {
efe120a0
FH
3463 btrfs_err(BTRFS_I(inode)->root->fs_info,
3464 "page %lu not writeback, cur %llu end %llu",
c1c9ff7c 3465 page->index, cur, end);
d1310b2e
CM
3466 }
3467
da2f0f74 3468 ret = submit_extent_page(write_flags, tree, wbc, page,
ffbd517d
CM
3469 sector, iosize, pg_offset,
3470 bdev, &epd->bio, max_nr,
c8b97818 3471 end_bio_extent_writepage,
005efedf 3472 0, 0, 0, false);
d1310b2e
CM
3473 if (ret)
3474 SetPageError(page);
3475 }
3476 cur = cur + iosize;
7f3c74fb 3477 pg_offset += iosize;
d1310b2e
CM
3478 nr++;
3479 }
40f76580
CM
3480done:
3481 *nr_ret = nr;
3482
3483done_unlocked:
3484
3485 /* drop our reference on any cached states */
3486 free_extent_state(cached_state);
3487 return ret;
3488}
3489
3490/*
3491 * the writepage semantics are similar to regular writepage. extent
3492 * records are inserted to lock ranges in the tree, and as dirty areas
3493 * are found, they are marked writeback. Then the lock bits are removed
3494 * and the end_io handler clears the writeback ranges
3495 */
3496static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3497 void *data)
3498{
3499 struct inode *inode = page->mapping->host;
3500 struct extent_page_data *epd = data;
3501 u64 start = page_offset(page);
3502 u64 page_end = start + PAGE_CACHE_SIZE - 1;
3503 int ret;
3504 int nr = 0;
3505 size_t pg_offset = 0;
3506 loff_t i_size = i_size_read(inode);
3507 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
3508 int write_flags;
3509 unsigned long nr_written = 0;
3510
3511 if (wbc->sync_mode == WB_SYNC_ALL)
3512 write_flags = WRITE_SYNC;
3513 else
3514 write_flags = WRITE;
3515
3516 trace___extent_writepage(page, inode, wbc);
3517
3518 WARN_ON(!PageLocked(page));
3519
3520 ClearPageError(page);
3521
3522 pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
3523 if (page->index > end_index ||
3524 (page->index == end_index && !pg_offset)) {
3525 page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE);
3526 unlock_page(page);
3527 return 0;
3528 }
3529
3530 if (page->index == end_index) {
3531 char *userpage;
3532
3533 userpage = kmap_atomic(page);
3534 memset(userpage + pg_offset, 0,
3535 PAGE_CACHE_SIZE - pg_offset);
3536 kunmap_atomic(userpage);
3537 flush_dcache_page(page);
3538 }
3539
3540 pg_offset = 0;
3541
3542 set_page_extent_mapped(page);
3543
3544 ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written);
3545 if (ret == 1)
3546 goto done_unlocked;
3547 if (ret)
3548 goto done;
3549
3550 ret = __extent_writepage_io(inode, page, wbc, epd,
3551 i_size, nr_written, write_flags, &nr);
3552 if (ret == 1)
3553 goto done_unlocked;
3554
d1310b2e
CM
3555done:
3556 if (nr == 0) {
3557 /* make sure the mapping tag for page dirty gets cleared */
3558 set_page_writeback(page);
3559 end_page_writeback(page);
3560 }
61391d56
FM
3561 if (PageError(page)) {
3562 ret = ret < 0 ? ret : -EIO;
3563 end_extent_writepage(page, ret, start, page_end);
3564 }
d1310b2e 3565 unlock_page(page);
40f76580 3566 return ret;
771ed689 3567
11c8349b 3568done_unlocked:
d1310b2e
CM
3569 return 0;
3570}
3571
fd8b2b61 3572void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
0b32f4bb 3573{
74316201
N
3574 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3575 TASK_UNINTERRUPTIBLE);
0b32f4bb
JB
3576}
3577
0e378df1
CM
3578static noinline_for_stack int
3579lock_extent_buffer_for_io(struct extent_buffer *eb,
3580 struct btrfs_fs_info *fs_info,
3581 struct extent_page_data *epd)
0b32f4bb
JB
3582{
3583 unsigned long i, num_pages;
3584 int flush = 0;
3585 int ret = 0;
3586
3587 if (!btrfs_try_tree_write_lock(eb)) {
3588 flush = 1;
3589 flush_write_bio(epd);
3590 btrfs_tree_lock(eb);
3591 }
3592
3593 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3594 btrfs_tree_unlock(eb);
3595 if (!epd->sync_io)
3596 return 0;
3597 if (!flush) {
3598 flush_write_bio(epd);
3599 flush = 1;
3600 }
a098d8e8
CM
3601 while (1) {
3602 wait_on_extent_buffer_writeback(eb);
3603 btrfs_tree_lock(eb);
3604 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3605 break;
0b32f4bb 3606 btrfs_tree_unlock(eb);
0b32f4bb
JB
3607 }
3608 }
3609
51561ffe
JB
3610 /*
3611 * We need to do this to prevent races in people who check if the eb is
3612 * under IO since we can end up having no IO bits set for a short period
3613 * of time.
3614 */
3615 spin_lock(&eb->refs_lock);
0b32f4bb
JB
3616 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3617 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
51561ffe 3618 spin_unlock(&eb->refs_lock);
0b32f4bb 3619 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
e2d84521
MX
3620 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
3621 -eb->len,
3622 fs_info->dirty_metadata_batch);
0b32f4bb 3623 ret = 1;
51561ffe
JB
3624 } else {
3625 spin_unlock(&eb->refs_lock);
0b32f4bb
JB
3626 }
3627
3628 btrfs_tree_unlock(eb);
3629
3630 if (!ret)
3631 return ret;
3632
3633 num_pages = num_extent_pages(eb->start, eb->len);
3634 for (i = 0; i < num_pages; i++) {
fb85fc9a 3635 struct page *p = eb->pages[i];
0b32f4bb
JB
3636
3637 if (!trylock_page(p)) {
3638 if (!flush) {
3639 flush_write_bio(epd);
3640 flush = 1;
3641 }
3642 lock_page(p);
3643 }
3644 }
3645
3646 return ret;
3647}
3648
3649static void end_extent_buffer_writeback(struct extent_buffer *eb)
3650{
3651 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
4e857c58 3652 smp_mb__after_atomic();
0b32f4bb
JB
3653 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3654}
3655
656f30db
FM
3656static void set_btree_ioerr(struct page *page)
3657{
3658 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3659 struct btrfs_inode *btree_ino = BTRFS_I(eb->fs_info->btree_inode);
3660
3661 SetPageError(page);
3662 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3663 return;
3664
3665 /*
3666 * If writeback for a btree extent that doesn't belong to a log tree
3667 * failed, increment the counter transaction->eb_write_errors.
3668 * We do this because while the transaction is running and before it's
3669 * committing (when we call filemap_fdata[write|wait]_range against
3670 * the btree inode), we might have
3671 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3672 * returns an error or an error happens during writeback, when we're
3673 * committing the transaction we wouldn't know about it, since the pages
3674 * can be no longer dirty nor marked anymore for writeback (if a
3675 * subsequent modification to the extent buffer didn't happen before the
3676 * transaction commit), which makes filemap_fdata[write|wait]_range not
3677 * able to find the pages tagged with SetPageError at transaction
3678 * commit time. So if this happens we must abort the transaction,
3679 * otherwise we commit a super block with btree roots that point to
3680 * btree nodes/leafs whose content on disk is invalid - either garbage
3681 * or the content of some node/leaf from a past generation that got
3682 * cowed or deleted and is no longer valid.
3683 *
3684 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3685 * not be enough - we need to distinguish between log tree extents vs
3686 * non-log tree extents, and the next filemap_fdatawait_range() call
3687 * will catch and clear such errors in the mapping - and that call might
3688 * be from a log sync and not from a transaction commit. Also, checking
3689 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3690 * not done and would not be reliable - the eb might have been released
3691 * from memory and reading it back again means that flag would not be
3692 * set (since it's a runtime flag, not persisted on disk).
3693 *
3694 * Using the flags below in the btree inode also makes us achieve the
3695 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3696 * writeback for all dirty pages and before filemap_fdatawait_range()
3697 * is called, the writeback for all dirty pages had already finished
3698 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3699 * filemap_fdatawait_range() would return success, as it could not know
3700 * that writeback errors happened (the pages were no longer tagged for
3701 * writeback).
3702 */
3703 switch (eb->log_index) {
3704 case -1:
3705 set_bit(BTRFS_INODE_BTREE_ERR, &btree_ino->runtime_flags);
3706 break;
3707 case 0:
3708 set_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags);
3709 break;
3710 case 1:
3711 set_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags);
3712 break;
3713 default:
3714 BUG(); /* unexpected, logic error */
3715 }
3716}
3717
4246a0b6 3718static void end_bio_extent_buffer_writepage(struct bio *bio)
0b32f4bb 3719{
2c30c71b 3720 struct bio_vec *bvec;
0b32f4bb 3721 struct extent_buffer *eb;
2c30c71b 3722 int i, done;
0b32f4bb 3723
2c30c71b 3724 bio_for_each_segment_all(bvec, bio, i) {
0b32f4bb
JB
3725 struct page *page = bvec->bv_page;
3726
0b32f4bb
JB
3727 eb = (struct extent_buffer *)page->private;
3728 BUG_ON(!eb);
3729 done = atomic_dec_and_test(&eb->io_pages);
3730
4246a0b6
CH
3731 if (bio->bi_error ||
3732 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
0b32f4bb 3733 ClearPageUptodate(page);
656f30db 3734 set_btree_ioerr(page);
0b32f4bb
JB
3735 }
3736
3737 end_page_writeback(page);
3738
3739 if (!done)
3740 continue;
3741
3742 end_extent_buffer_writeback(eb);
2c30c71b 3743 }
0b32f4bb
JB
3744
3745 bio_put(bio);
0b32f4bb
JB
3746}
3747
0e378df1 3748static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
0b32f4bb
JB
3749 struct btrfs_fs_info *fs_info,
3750 struct writeback_control *wbc,
3751 struct extent_page_data *epd)
3752{
3753 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
f28491e0 3754 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
0b32f4bb
JB
3755 u64 offset = eb->start;
3756 unsigned long i, num_pages;
de0022b9 3757 unsigned long bio_flags = 0;
d4c7ca86 3758 int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META;
d7dbe9e7 3759 int ret = 0;
0b32f4bb 3760
656f30db 3761 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
0b32f4bb
JB
3762 num_pages = num_extent_pages(eb->start, eb->len);
3763 atomic_set(&eb->io_pages, num_pages);
de0022b9
JB
3764 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3765 bio_flags = EXTENT_BIO_TREE_LOG;
3766
0b32f4bb 3767 for (i = 0; i < num_pages; i++) {
fb85fc9a 3768 struct page *p = eb->pages[i];
0b32f4bb
JB
3769
3770 clear_page_dirty_for_io(p);
3771 set_page_writeback(p);
da2f0f74 3772 ret = submit_extent_page(rw, tree, wbc, p, offset >> 9,
0b32f4bb
JB
3773 PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3774 -1, end_bio_extent_buffer_writepage,
005efedf 3775 0, epd->bio_flags, bio_flags, false);
de0022b9 3776 epd->bio_flags = bio_flags;
0b32f4bb 3777 if (ret) {
656f30db 3778 set_btree_ioerr(p);
55e3bd2e 3779 end_page_writeback(p);
0b32f4bb
JB
3780 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3781 end_extent_buffer_writeback(eb);
3782 ret = -EIO;
3783 break;
3784 }
3785 offset += PAGE_CACHE_SIZE;
3786 update_nr_written(p, wbc, 1);
3787 unlock_page(p);
3788 }
3789
3790 if (unlikely(ret)) {
3791 for (; i < num_pages; i++) {
bbf65cf0 3792 struct page *p = eb->pages[i];
81465028 3793 clear_page_dirty_for_io(p);
0b32f4bb
JB
3794 unlock_page(p);
3795 }
3796 }
3797
3798 return ret;
3799}
3800
3801int btree_write_cache_pages(struct address_space *mapping,
3802 struct writeback_control *wbc)
3803{
3804 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3805 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3806 struct extent_buffer *eb, *prev_eb = NULL;
3807 struct extent_page_data epd = {
3808 .bio = NULL,
3809 .tree = tree,
3810 .extent_locked = 0,
3811 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
de0022b9 3812 .bio_flags = 0,
0b32f4bb
JB
3813 };
3814 int ret = 0;
3815 int done = 0;
3816 int nr_to_write_done = 0;
3817 struct pagevec pvec;
3818 int nr_pages;
3819 pgoff_t index;
3820 pgoff_t end; /* Inclusive */
3821 int scanned = 0;
3822 int tag;
3823
3824 pagevec_init(&pvec, 0);
3825 if (wbc->range_cyclic) {
3826 index = mapping->writeback_index; /* Start from prev offset */
3827 end = -1;
3828 } else {
3829 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3830 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3831 scanned = 1;
3832 }
3833 if (wbc->sync_mode == WB_SYNC_ALL)
3834 tag = PAGECACHE_TAG_TOWRITE;
3835 else
3836 tag = PAGECACHE_TAG_DIRTY;
3837retry:
3838 if (wbc->sync_mode == WB_SYNC_ALL)
3839 tag_pages_for_writeback(mapping, index, end);
3840 while (!done && !nr_to_write_done && (index <= end) &&
3841 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3842 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3843 unsigned i;
3844
3845 scanned = 1;
3846 for (i = 0; i < nr_pages; i++) {
3847 struct page *page = pvec.pages[i];
3848
3849 if (!PagePrivate(page))
3850 continue;
3851
3852 if (!wbc->range_cyclic && page->index > end) {
3853 done = 1;
3854 break;
3855 }
3856
b5bae261
JB
3857 spin_lock(&mapping->private_lock);
3858 if (!PagePrivate(page)) {
3859 spin_unlock(&mapping->private_lock);
3860 continue;
3861 }
3862
0b32f4bb 3863 eb = (struct extent_buffer *)page->private;
b5bae261
JB
3864
3865 /*
3866 * Shouldn't happen and normally this would be a BUG_ON
3867 * but no sense in crashing the users box for something
3868 * we can survive anyway.
3869 */
fae7f21c 3870 if (WARN_ON(!eb)) {
b5bae261 3871 spin_unlock(&mapping->private_lock);
0b32f4bb
JB
3872 continue;
3873 }
3874
b5bae261
JB
3875 if (eb == prev_eb) {
3876 spin_unlock(&mapping->private_lock);
0b32f4bb 3877 continue;
b5bae261 3878 }
0b32f4bb 3879
b5bae261
JB
3880 ret = atomic_inc_not_zero(&eb->refs);
3881 spin_unlock(&mapping->private_lock);
3882 if (!ret)
0b32f4bb 3883 continue;
0b32f4bb
JB
3884
3885 prev_eb = eb;
3886 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3887 if (!ret) {
3888 free_extent_buffer(eb);
3889 continue;
3890 }
3891
3892 ret = write_one_eb(eb, fs_info, wbc, &epd);
3893 if (ret) {
3894 done = 1;
3895 free_extent_buffer(eb);
3896 break;
3897 }
3898 free_extent_buffer(eb);
3899
3900 /*
3901 * the filesystem may choose to bump up nr_to_write.
3902 * We have to make sure to honor the new nr_to_write
3903 * at any time
3904 */
3905 nr_to_write_done = wbc->nr_to_write <= 0;
3906 }
3907 pagevec_release(&pvec);
3908 cond_resched();
3909 }
3910 if (!scanned && !done) {
3911 /*
3912 * We hit the last page and there is more work to be done: wrap
3913 * back to the start of the file
3914 */
3915 scanned = 1;
3916 index = 0;
3917 goto retry;
3918 }
3919 flush_write_bio(&epd);
3920 return ret;
3921}
3922
d1310b2e 3923/**
4bef0848 3924 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
d1310b2e
CM
3925 * @mapping: address space structure to write
3926 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3927 * @writepage: function called for each page
3928 * @data: data passed to writepage function
3929 *
3930 * If a page is already under I/O, write_cache_pages() skips it, even
3931 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3932 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3933 * and msync() need to guarantee that all the data which was dirty at the time
3934 * the call was made get new I/O started against them. If wbc->sync_mode is
3935 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3936 * existing IO to complete.
3937 */
b2950863 3938static int extent_write_cache_pages(struct extent_io_tree *tree,
4bef0848
CM
3939 struct address_space *mapping,
3940 struct writeback_control *wbc,
d2c3f4f6
CM
3941 writepage_t writepage, void *data,
3942 void (*flush_fn)(void *))
d1310b2e 3943{
7fd1a3f7 3944 struct inode *inode = mapping->host;
d1310b2e
CM
3945 int ret = 0;
3946 int done = 0;
61391d56 3947 int err = 0;
f85d7d6c 3948 int nr_to_write_done = 0;
d1310b2e
CM
3949 struct pagevec pvec;
3950 int nr_pages;
3951 pgoff_t index;
3952 pgoff_t end; /* Inclusive */
3953 int scanned = 0;
f7aaa06b 3954 int tag;
d1310b2e 3955
7fd1a3f7
JB
3956 /*
3957 * We have to hold onto the inode so that ordered extents can do their
3958 * work when the IO finishes. The alternative to this is failing to add
3959 * an ordered extent if the igrab() fails there and that is a huge pain
3960 * to deal with, so instead just hold onto the inode throughout the
3961 * writepages operation. If it fails here we are freeing up the inode
3962 * anyway and we'd rather not waste our time writing out stuff that is
3963 * going to be truncated anyway.
3964 */
3965 if (!igrab(inode))
3966 return 0;
3967
d1310b2e
CM
3968 pagevec_init(&pvec, 0);
3969 if (wbc->range_cyclic) {
3970 index = mapping->writeback_index; /* Start from prev offset */
3971 end = -1;
3972 } else {
3973 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3974 end = wbc->range_end >> PAGE_CACHE_SHIFT;
d1310b2e
CM
3975 scanned = 1;
3976 }
f7aaa06b
JB
3977 if (wbc->sync_mode == WB_SYNC_ALL)
3978 tag = PAGECACHE_TAG_TOWRITE;
3979 else
3980 tag = PAGECACHE_TAG_DIRTY;
d1310b2e 3981retry:
f7aaa06b
JB
3982 if (wbc->sync_mode == WB_SYNC_ALL)
3983 tag_pages_for_writeback(mapping, index, end);
f85d7d6c 3984 while (!done && !nr_to_write_done && (index <= end) &&
f7aaa06b
JB
3985 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3986 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
d1310b2e
CM
3987 unsigned i;
3988
3989 scanned = 1;
3990 for (i = 0; i < nr_pages; i++) {
3991 struct page *page = pvec.pages[i];
3992
3993 /*
3994 * At this point we hold neither mapping->tree_lock nor
3995 * lock on the page itself: the page may be truncated or
3996 * invalidated (changing page->mapping to NULL), or even
3997 * swizzled back from swapper_space to tmpfs file
3998 * mapping
3999 */
c8f2f24b
JB
4000 if (!trylock_page(page)) {
4001 flush_fn(data);
4002 lock_page(page);
01d658f2 4003 }
d1310b2e
CM
4004
4005 if (unlikely(page->mapping != mapping)) {
4006 unlock_page(page);
4007 continue;
4008 }
4009
4010 if (!wbc->range_cyclic && page->index > end) {
4011 done = 1;
4012 unlock_page(page);
4013 continue;
4014 }
4015
d2c3f4f6 4016 if (wbc->sync_mode != WB_SYNC_NONE) {
0e6bd956
CM
4017 if (PageWriteback(page))
4018 flush_fn(data);
d1310b2e 4019 wait_on_page_writeback(page);
d2c3f4f6 4020 }
d1310b2e
CM
4021
4022 if (PageWriteback(page) ||
4023 !clear_page_dirty_for_io(page)) {
4024 unlock_page(page);
4025 continue;
4026 }
4027
4028 ret = (*writepage)(page, wbc, data);
4029
4030 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
4031 unlock_page(page);
4032 ret = 0;
4033 }
61391d56
FM
4034 if (!err && ret < 0)
4035 err = ret;
f85d7d6c
CM
4036
4037 /*
4038 * the filesystem may choose to bump up nr_to_write.
4039 * We have to make sure to honor the new nr_to_write
4040 * at any time
4041 */
4042 nr_to_write_done = wbc->nr_to_write <= 0;
d1310b2e
CM
4043 }
4044 pagevec_release(&pvec);
4045 cond_resched();
4046 }
61391d56 4047 if (!scanned && !done && !err) {
d1310b2e
CM
4048 /*
4049 * We hit the last page and there is more work to be done: wrap
4050 * back to the start of the file
4051 */
4052 scanned = 1;
4053 index = 0;
4054 goto retry;
4055 }
7fd1a3f7 4056 btrfs_add_delayed_iput(inode);
61391d56 4057 return err;
d1310b2e 4058}
d1310b2e 4059
ffbd517d 4060static void flush_epd_write_bio(struct extent_page_data *epd)
d2c3f4f6 4061{
d2c3f4f6 4062 if (epd->bio) {
355808c2
JM
4063 int rw = WRITE;
4064 int ret;
4065
ffbd517d 4066 if (epd->sync_io)
355808c2
JM
4067 rw = WRITE_SYNC;
4068
de0022b9 4069 ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags);
79787eaa 4070 BUG_ON(ret < 0); /* -ENOMEM */
d2c3f4f6
CM
4071 epd->bio = NULL;
4072 }
4073}
4074
ffbd517d
CM
4075static noinline void flush_write_bio(void *data)
4076{
4077 struct extent_page_data *epd = data;
4078 flush_epd_write_bio(epd);
4079}
4080
d1310b2e
CM
4081int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
4082 get_extent_t *get_extent,
4083 struct writeback_control *wbc)
4084{
4085 int ret;
d1310b2e
CM
4086 struct extent_page_data epd = {
4087 .bio = NULL,
4088 .tree = tree,
4089 .get_extent = get_extent,
771ed689 4090 .extent_locked = 0,
ffbd517d 4091 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
de0022b9 4092 .bio_flags = 0,
d1310b2e 4093 };
d1310b2e 4094
d1310b2e
CM
4095 ret = __extent_writepage(page, wbc, &epd);
4096
ffbd517d 4097 flush_epd_write_bio(&epd);
d1310b2e
CM
4098 return ret;
4099}
d1310b2e 4100
771ed689
CM
4101int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
4102 u64 start, u64 end, get_extent_t *get_extent,
4103 int mode)
4104{
4105 int ret = 0;
4106 struct address_space *mapping = inode->i_mapping;
4107 struct page *page;
4108 unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
4109 PAGE_CACHE_SHIFT;
4110
4111 struct extent_page_data epd = {
4112 .bio = NULL,
4113 .tree = tree,
4114 .get_extent = get_extent,
4115 .extent_locked = 1,
ffbd517d 4116 .sync_io = mode == WB_SYNC_ALL,
de0022b9 4117 .bio_flags = 0,
771ed689
CM
4118 };
4119 struct writeback_control wbc_writepages = {
771ed689 4120 .sync_mode = mode,
771ed689
CM
4121 .nr_to_write = nr_pages * 2,
4122 .range_start = start,
4123 .range_end = end + 1,
4124 };
4125
d397712b 4126 while (start <= end) {
771ed689
CM
4127 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
4128 if (clear_page_dirty_for_io(page))
4129 ret = __extent_writepage(page, &wbc_writepages, &epd);
4130 else {
4131 if (tree->ops && tree->ops->writepage_end_io_hook)
4132 tree->ops->writepage_end_io_hook(page, start,
4133 start + PAGE_CACHE_SIZE - 1,
4134 NULL, 1);
4135 unlock_page(page);
4136 }
4137 page_cache_release(page);
4138 start += PAGE_CACHE_SIZE;
4139 }
4140
ffbd517d 4141 flush_epd_write_bio(&epd);
771ed689
CM
4142 return ret;
4143}
d1310b2e
CM
4144
4145int extent_writepages(struct extent_io_tree *tree,
4146 struct address_space *mapping,
4147 get_extent_t *get_extent,
4148 struct writeback_control *wbc)
4149{
4150 int ret = 0;
4151 struct extent_page_data epd = {
4152 .bio = NULL,
4153 .tree = tree,
4154 .get_extent = get_extent,
771ed689 4155 .extent_locked = 0,
ffbd517d 4156 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
de0022b9 4157 .bio_flags = 0,
d1310b2e
CM
4158 };
4159
4bef0848 4160 ret = extent_write_cache_pages(tree, mapping, wbc,
d2c3f4f6
CM
4161 __extent_writepage, &epd,
4162 flush_write_bio);
ffbd517d 4163 flush_epd_write_bio(&epd);
d1310b2e
CM
4164 return ret;
4165}
d1310b2e
CM
4166
4167int extent_readpages(struct extent_io_tree *tree,
4168 struct address_space *mapping,
4169 struct list_head *pages, unsigned nr_pages,
4170 get_extent_t get_extent)
4171{
4172 struct bio *bio = NULL;
4173 unsigned page_idx;
c8b97818 4174 unsigned long bio_flags = 0;
67c9684f
LB
4175 struct page *pagepool[16];
4176 struct page *page;
125bac01 4177 struct extent_map *em_cached = NULL;
67c9684f 4178 int nr = 0;
808f80b4 4179 u64 prev_em_start = (u64)-1;
d1310b2e 4180
d1310b2e 4181 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
67c9684f 4182 page = list_entry(pages->prev, struct page, lru);
d1310b2e
CM
4183
4184 prefetchw(&page->flags);
4185 list_del(&page->lru);
67c9684f 4186 if (add_to_page_cache_lru(page, mapping,
43e817a1 4187 page->index, GFP_NOFS)) {
67c9684f
LB
4188 page_cache_release(page);
4189 continue;
d1310b2e 4190 }
67c9684f
LB
4191
4192 pagepool[nr++] = page;
4193 if (nr < ARRAY_SIZE(pagepool))
4194 continue;
125bac01 4195 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
808f80b4 4196 &bio, 0, &bio_flags, READ, &prev_em_start);
67c9684f 4197 nr = 0;
d1310b2e 4198 }
9974090b 4199 if (nr)
125bac01 4200 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
808f80b4 4201 &bio, 0, &bio_flags, READ, &prev_em_start);
67c9684f 4202
125bac01
MX
4203 if (em_cached)
4204 free_extent_map(em_cached);
4205
d1310b2e
CM
4206 BUG_ON(!list_empty(pages));
4207 if (bio)
79787eaa 4208 return submit_one_bio(READ, bio, 0, bio_flags);
d1310b2e
CM
4209 return 0;
4210}
d1310b2e
CM
4211
4212/*
4213 * basic invalidatepage code, this waits on any locked or writeback
4214 * ranges corresponding to the page, and then deletes any extent state
4215 * records from the tree
4216 */
4217int extent_invalidatepage(struct extent_io_tree *tree,
4218 struct page *page, unsigned long offset)
4219{
2ac55d41 4220 struct extent_state *cached_state = NULL;
4eee4fa4 4221 u64 start = page_offset(page);
d1310b2e
CM
4222 u64 end = start + PAGE_CACHE_SIZE - 1;
4223 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4224
fda2832f 4225 start += ALIGN(offset, blocksize);
d1310b2e
CM
4226 if (start > end)
4227 return 0;
4228
ff13db41 4229 lock_extent_bits(tree, start, end, &cached_state);
1edbb734 4230 wait_on_page_writeback(page);
d1310b2e 4231 clear_extent_bit(tree, start, end,
32c00aff
JB
4232 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4233 EXTENT_DO_ACCOUNTING,
2ac55d41 4234 1, 1, &cached_state, GFP_NOFS);
d1310b2e
CM
4235 return 0;
4236}
d1310b2e 4237
7b13b7b1
CM
4238/*
4239 * a helper for releasepage, this tests for areas of the page that
4240 * are locked or under IO and drops the related state bits if it is safe
4241 * to drop the page.
4242 */
48a3b636
ES
4243static int try_release_extent_state(struct extent_map_tree *map,
4244 struct extent_io_tree *tree,
4245 struct page *page, gfp_t mask)
7b13b7b1 4246{
4eee4fa4 4247 u64 start = page_offset(page);
7b13b7b1
CM
4248 u64 end = start + PAGE_CACHE_SIZE - 1;
4249 int ret = 1;
4250
211f90e6 4251 if (test_range_bit(tree, start, end,
8b62b72b 4252 EXTENT_IOBITS, 0, NULL))
7b13b7b1
CM
4253 ret = 0;
4254 else {
4255 if ((mask & GFP_NOFS) == GFP_NOFS)
4256 mask = GFP_NOFS;
11ef160f
CM
4257 /*
4258 * at this point we can safely clear everything except the
4259 * locked bit and the nodatasum bit
4260 */
e3f24cc5 4261 ret = clear_extent_bit(tree, start, end,
11ef160f
CM
4262 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4263 0, 0, NULL, mask);
e3f24cc5
CM
4264
4265 /* if clear_extent_bit failed for enomem reasons,
4266 * we can't allow the release to continue.
4267 */
4268 if (ret < 0)
4269 ret = 0;
4270 else
4271 ret = 1;
7b13b7b1
CM
4272 }
4273 return ret;
4274}
7b13b7b1 4275
d1310b2e
CM
4276/*
4277 * a helper for releasepage. As long as there are no locked extents
4278 * in the range corresponding to the page, both state records and extent
4279 * map records are removed
4280 */
4281int try_release_extent_mapping(struct extent_map_tree *map,
70dec807
CM
4282 struct extent_io_tree *tree, struct page *page,
4283 gfp_t mask)
d1310b2e
CM
4284{
4285 struct extent_map *em;
4eee4fa4 4286 u64 start = page_offset(page);
d1310b2e 4287 u64 end = start + PAGE_CACHE_SIZE - 1;
7b13b7b1 4288
d0164adc 4289 if (gfpflags_allow_blocking(mask) &&
ee22184b 4290 page->mapping->host->i_size > SZ_16M) {
39b5637f 4291 u64 len;
70dec807 4292 while (start <= end) {
39b5637f 4293 len = end - start + 1;
890871be 4294 write_lock(&map->lock);
39b5637f 4295 em = lookup_extent_mapping(map, start, len);
285190d9 4296 if (!em) {
890871be 4297 write_unlock(&map->lock);
70dec807
CM
4298 break;
4299 }
7f3c74fb
CM
4300 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4301 em->start != start) {
890871be 4302 write_unlock(&map->lock);
70dec807
CM
4303 free_extent_map(em);
4304 break;
4305 }
4306 if (!test_range_bit(tree, em->start,
4307 extent_map_end(em) - 1,
8b62b72b 4308 EXTENT_LOCKED | EXTENT_WRITEBACK,
9655d298 4309 0, NULL)) {
70dec807
CM
4310 remove_extent_mapping(map, em);
4311 /* once for the rb tree */
4312 free_extent_map(em);
4313 }
4314 start = extent_map_end(em);
890871be 4315 write_unlock(&map->lock);
70dec807
CM
4316
4317 /* once for us */
d1310b2e
CM
4318 free_extent_map(em);
4319 }
d1310b2e 4320 }
7b13b7b1 4321 return try_release_extent_state(map, tree, page, mask);
d1310b2e 4322}
d1310b2e 4323
ec29ed5b
CM
4324/*
4325 * helper function for fiemap, which doesn't want to see any holes.
4326 * This maps until we find something past 'last'
4327 */
4328static struct extent_map *get_extent_skip_holes(struct inode *inode,
4329 u64 offset,
4330 u64 last,
4331 get_extent_t *get_extent)
4332{
4333 u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
4334 struct extent_map *em;
4335 u64 len;
4336
4337 if (offset >= last)
4338 return NULL;
4339
67871254 4340 while (1) {
ec29ed5b
CM
4341 len = last - offset;
4342 if (len == 0)
4343 break;
fda2832f 4344 len = ALIGN(len, sectorsize);
ec29ed5b 4345 em = get_extent(inode, NULL, 0, offset, len, 0);
c704005d 4346 if (IS_ERR_OR_NULL(em))
ec29ed5b
CM
4347 return em;
4348
4349 /* if this isn't a hole return it */
4350 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4351 em->block_start != EXTENT_MAP_HOLE) {
4352 return em;
4353 }
4354
4355 /* this is a hole, advance to the next extent */
4356 offset = extent_map_end(em);
4357 free_extent_map(em);
4358 if (offset >= last)
4359 break;
4360 }
4361 return NULL;
4362}
4363
1506fcc8
YS
4364int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4365 __u64 start, __u64 len, get_extent_t *get_extent)
4366{
975f84fe 4367 int ret = 0;
1506fcc8
YS
4368 u64 off = start;
4369 u64 max = start + len;
4370 u32 flags = 0;
975f84fe
JB
4371 u32 found_type;
4372 u64 last;
ec29ed5b 4373 u64 last_for_get_extent = 0;
1506fcc8 4374 u64 disko = 0;
ec29ed5b 4375 u64 isize = i_size_read(inode);
975f84fe 4376 struct btrfs_key found_key;
1506fcc8 4377 struct extent_map *em = NULL;
2ac55d41 4378 struct extent_state *cached_state = NULL;
975f84fe 4379 struct btrfs_path *path;
dc046b10 4380 struct btrfs_root *root = BTRFS_I(inode)->root;
1506fcc8 4381 int end = 0;
ec29ed5b
CM
4382 u64 em_start = 0;
4383 u64 em_len = 0;
4384 u64 em_end = 0;
1506fcc8
YS
4385
4386 if (len == 0)
4387 return -EINVAL;
4388
975f84fe
JB
4389 path = btrfs_alloc_path();
4390 if (!path)
4391 return -ENOMEM;
4392 path->leave_spinning = 1;
4393
2c91943b
QW
4394 start = round_down(start, BTRFS_I(inode)->root->sectorsize);
4395 len = round_up(max, BTRFS_I(inode)->root->sectorsize) - start;
4d479cf0 4396
ec29ed5b
CM
4397 /*
4398 * lookup the last file extent. We're not using i_size here
4399 * because there might be preallocation past i_size
4400 */
dc046b10
JB
4401 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1,
4402 0);
975f84fe
JB
4403 if (ret < 0) {
4404 btrfs_free_path(path);
4405 return ret;
4406 }
4407 WARN_ON(!ret);
4408 path->slots[0]--;
975f84fe 4409 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
962a298f 4410 found_type = found_key.type;
975f84fe 4411
ec29ed5b 4412 /* No extents, but there might be delalloc bits */
33345d01 4413 if (found_key.objectid != btrfs_ino(inode) ||
975f84fe 4414 found_type != BTRFS_EXTENT_DATA_KEY) {
ec29ed5b
CM
4415 /* have to trust i_size as the end */
4416 last = (u64)-1;
4417 last_for_get_extent = isize;
4418 } else {
4419 /*
4420 * remember the start of the last extent. There are a
4421 * bunch of different factors that go into the length of the
4422 * extent, so its much less complex to remember where it started
4423 */
4424 last = found_key.offset;
4425 last_for_get_extent = last + 1;
975f84fe 4426 }
fe09e16c 4427 btrfs_release_path(path);
975f84fe 4428
ec29ed5b
CM
4429 /*
4430 * we might have some extents allocated but more delalloc past those
4431 * extents. so, we trust isize unless the start of the last extent is
4432 * beyond isize
4433 */
4434 if (last < isize) {
4435 last = (u64)-1;
4436 last_for_get_extent = isize;
4437 }
4438
ff13db41 4439 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
d0082371 4440 &cached_state);
ec29ed5b 4441
4d479cf0 4442 em = get_extent_skip_holes(inode, start, last_for_get_extent,
ec29ed5b 4443 get_extent);
1506fcc8
YS
4444 if (!em)
4445 goto out;
4446 if (IS_ERR(em)) {
4447 ret = PTR_ERR(em);
4448 goto out;
4449 }
975f84fe 4450
1506fcc8 4451 while (!end) {
b76bb701 4452 u64 offset_in_extent = 0;
ea8efc74
CM
4453
4454 /* break if the extent we found is outside the range */
4455 if (em->start >= max || extent_map_end(em) < off)
4456 break;
4457
4458 /*
4459 * get_extent may return an extent that starts before our
4460 * requested range. We have to make sure the ranges
4461 * we return to fiemap always move forward and don't
4462 * overlap, so adjust the offsets here
4463 */
4464 em_start = max(em->start, off);
1506fcc8 4465
ea8efc74
CM
4466 /*
4467 * record the offset from the start of the extent
b76bb701
JB
4468 * for adjusting the disk offset below. Only do this if the
4469 * extent isn't compressed since our in ram offset may be past
4470 * what we have actually allocated on disk.
ea8efc74 4471 */
b76bb701
JB
4472 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4473 offset_in_extent = em_start - em->start;
ec29ed5b 4474 em_end = extent_map_end(em);
ea8efc74 4475 em_len = em_end - em_start;
1506fcc8
YS
4476 disko = 0;
4477 flags = 0;
4478
ea8efc74
CM
4479 /*
4480 * bump off for our next call to get_extent
4481 */
4482 off = extent_map_end(em);
4483 if (off >= max)
4484 end = 1;
4485
93dbfad7 4486 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
1506fcc8
YS
4487 end = 1;
4488 flags |= FIEMAP_EXTENT_LAST;
93dbfad7 4489 } else if (em->block_start == EXTENT_MAP_INLINE) {
1506fcc8
YS
4490 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4491 FIEMAP_EXTENT_NOT_ALIGNED);
93dbfad7 4492 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
1506fcc8
YS
4493 flags |= (FIEMAP_EXTENT_DELALLOC |
4494 FIEMAP_EXTENT_UNKNOWN);
dc046b10
JB
4495 } else if (fieinfo->fi_extents_max) {
4496 u64 bytenr = em->block_start -
4497 (em->start - em->orig_start);
fe09e16c 4498
ea8efc74 4499 disko = em->block_start + offset_in_extent;
fe09e16c
LB
4500
4501 /*
4502 * As btrfs supports shared space, this information
4503 * can be exported to userspace tools via
dc046b10
JB
4504 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4505 * then we're just getting a count and we can skip the
4506 * lookup stuff.
fe09e16c 4507 */
dc046b10
JB
4508 ret = btrfs_check_shared(NULL, root->fs_info,
4509 root->objectid,
4510 btrfs_ino(inode), bytenr);
4511 if (ret < 0)
fe09e16c 4512 goto out_free;
dc046b10 4513 if (ret)
fe09e16c 4514 flags |= FIEMAP_EXTENT_SHARED;
dc046b10 4515 ret = 0;
1506fcc8
YS
4516 }
4517 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4518 flags |= FIEMAP_EXTENT_ENCODED;
0d2b2372
JB
4519 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4520 flags |= FIEMAP_EXTENT_UNWRITTEN;
1506fcc8 4521
1506fcc8
YS
4522 free_extent_map(em);
4523 em = NULL;
ec29ed5b
CM
4524 if ((em_start >= last) || em_len == (u64)-1 ||
4525 (last == (u64)-1 && isize <= em_end)) {
1506fcc8
YS
4526 flags |= FIEMAP_EXTENT_LAST;
4527 end = 1;
4528 }
4529
ec29ed5b
CM
4530 /* now scan forward to see if this is really the last extent. */
4531 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4532 get_extent);
4533 if (IS_ERR(em)) {
4534 ret = PTR_ERR(em);
4535 goto out;
4536 }
4537 if (!em) {
975f84fe
JB
4538 flags |= FIEMAP_EXTENT_LAST;
4539 end = 1;
4540 }
ec29ed5b
CM
4541 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
4542 em_len, flags);
26e726af
CS
4543 if (ret) {
4544 if (ret == 1)
4545 ret = 0;
ec29ed5b 4546 goto out_free;
26e726af 4547 }
1506fcc8
YS
4548 }
4549out_free:
4550 free_extent_map(em);
4551out:
fe09e16c 4552 btrfs_free_path(path);
a52f4cd2 4553 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
2ac55d41 4554 &cached_state, GFP_NOFS);
1506fcc8
YS
4555 return ret;
4556}
4557
727011e0
CM
4558static void __free_extent_buffer(struct extent_buffer *eb)
4559{
6d49ba1b 4560 btrfs_leak_debug_del(&eb->leak_list);
727011e0
CM
4561 kmem_cache_free(extent_buffer_cache, eb);
4562}
4563
a26e8c9f 4564int extent_buffer_under_io(struct extent_buffer *eb)
db7f3436
JB
4565{
4566 return (atomic_read(&eb->io_pages) ||
4567 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4568 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4569}
4570
4571/*
4572 * Helper for releasing extent buffer page.
4573 */
a50924e3 4574static void btrfs_release_extent_buffer_page(struct extent_buffer *eb)
db7f3436
JB
4575{
4576 unsigned long index;
db7f3436
JB
4577 struct page *page;
4578 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4579
4580 BUG_ON(extent_buffer_under_io(eb));
4581
a50924e3
DS
4582 index = num_extent_pages(eb->start, eb->len);
4583 if (index == 0)
db7f3436
JB
4584 return;
4585
4586 do {
4587 index--;
fb85fc9a 4588 page = eb->pages[index];
5d2361db
FL
4589 if (!page)
4590 continue;
4591 if (mapped)
db7f3436 4592 spin_lock(&page->mapping->private_lock);
5d2361db
FL
4593 /*
4594 * We do this since we'll remove the pages after we've
4595 * removed the eb from the radix tree, so we could race
4596 * and have this page now attached to the new eb. So
4597 * only clear page_private if it's still connected to
4598 * this eb.
4599 */
4600 if (PagePrivate(page) &&
4601 page->private == (unsigned long)eb) {
4602 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4603 BUG_ON(PageDirty(page));
4604 BUG_ON(PageWriteback(page));
db7f3436 4605 /*
5d2361db
FL
4606 * We need to make sure we haven't be attached
4607 * to a new eb.
db7f3436 4608 */
5d2361db
FL
4609 ClearPagePrivate(page);
4610 set_page_private(page, 0);
4611 /* One for the page private */
db7f3436
JB
4612 page_cache_release(page);
4613 }
5d2361db
FL
4614
4615 if (mapped)
4616 spin_unlock(&page->mapping->private_lock);
4617
4618 /* One for when we alloced the page */
4619 page_cache_release(page);
a50924e3 4620 } while (index != 0);
db7f3436
JB
4621}
4622
4623/*
4624 * Helper for releasing the extent buffer.
4625 */
4626static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4627{
a50924e3 4628 btrfs_release_extent_buffer_page(eb);
db7f3436
JB
4629 __free_extent_buffer(eb);
4630}
4631
f28491e0
JB
4632static struct extent_buffer *
4633__alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
23d79d81 4634 unsigned long len)
d1310b2e
CM
4635{
4636 struct extent_buffer *eb = NULL;
4637
d1b5c567 4638 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
d1310b2e
CM
4639 eb->start = start;
4640 eb->len = len;
f28491e0 4641 eb->fs_info = fs_info;
815a51c7 4642 eb->bflags = 0;
bd681513
CM
4643 rwlock_init(&eb->lock);
4644 atomic_set(&eb->write_locks, 0);
4645 atomic_set(&eb->read_locks, 0);
4646 atomic_set(&eb->blocking_readers, 0);
4647 atomic_set(&eb->blocking_writers, 0);
4648 atomic_set(&eb->spinning_readers, 0);
4649 atomic_set(&eb->spinning_writers, 0);
5b25f70f 4650 eb->lock_nested = 0;
bd681513
CM
4651 init_waitqueue_head(&eb->write_lock_wq);
4652 init_waitqueue_head(&eb->read_lock_wq);
b4ce94de 4653
6d49ba1b
ES
4654 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4655
3083ee2e 4656 spin_lock_init(&eb->refs_lock);
d1310b2e 4657 atomic_set(&eb->refs, 1);
0b32f4bb 4658 atomic_set(&eb->io_pages, 0);
727011e0 4659
b8dae313
DS
4660 /*
4661 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4662 */
4663 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4664 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4665 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
d1310b2e
CM
4666
4667 return eb;
4668}
4669
815a51c7
JS
4670struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4671{
4672 unsigned long i;
4673 struct page *p;
4674 struct extent_buffer *new;
4675 unsigned long num_pages = num_extent_pages(src->start, src->len);
4676
3f556f78 4677 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
815a51c7
JS
4678 if (new == NULL)
4679 return NULL;
4680
4681 for (i = 0; i < num_pages; i++) {
9ec72677 4682 p = alloc_page(GFP_NOFS);
db7f3436
JB
4683 if (!p) {
4684 btrfs_release_extent_buffer(new);
4685 return NULL;
4686 }
815a51c7
JS
4687 attach_extent_buffer_page(new, p);
4688 WARN_ON(PageDirty(p));
4689 SetPageUptodate(p);
4690 new->pages[i] = p;
4691 }
4692
4693 copy_extent_buffer(new, src, 0, 0, src->len);
4694 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4695 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4696
4697 return new;
4698}
4699
0f331229
OS
4700struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4701 u64 start, unsigned long len)
815a51c7
JS
4702{
4703 struct extent_buffer *eb;
3f556f78 4704 unsigned long num_pages;
815a51c7
JS
4705 unsigned long i;
4706
0f331229 4707 num_pages = num_extent_pages(start, len);
3f556f78
DS
4708
4709 eb = __alloc_extent_buffer(fs_info, start, len);
815a51c7
JS
4710 if (!eb)
4711 return NULL;
4712
4713 for (i = 0; i < num_pages; i++) {
9ec72677 4714 eb->pages[i] = alloc_page(GFP_NOFS);
815a51c7
JS
4715 if (!eb->pages[i])
4716 goto err;
4717 }
4718 set_extent_buffer_uptodate(eb);
4719 btrfs_set_header_nritems(eb, 0);
4720 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4721
4722 return eb;
4723err:
84167d19
SB
4724 for (; i > 0; i--)
4725 __free_page(eb->pages[i - 1]);
815a51c7
JS
4726 __free_extent_buffer(eb);
4727 return NULL;
4728}
4729
0f331229
OS
4730struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4731 u64 start)
4732{
4733 unsigned long len;
4734
4735 if (!fs_info) {
4736 /*
4737 * Called only from tests that don't always have a fs_info
4738 * available, but we know that nodesize is 4096
4739 */
4740 len = 4096;
4741 } else {
4742 len = fs_info->tree_root->nodesize;
4743 }
4744
4745 return __alloc_dummy_extent_buffer(fs_info, start, len);
4746}
4747
0b32f4bb
JB
4748static void check_buffer_tree_ref(struct extent_buffer *eb)
4749{
242e18c7 4750 int refs;
0b32f4bb
JB
4751 /* the ref bit is tricky. We have to make sure it is set
4752 * if we have the buffer dirty. Otherwise the
4753 * code to free a buffer can end up dropping a dirty
4754 * page
4755 *
4756 * Once the ref bit is set, it won't go away while the
4757 * buffer is dirty or in writeback, and it also won't
4758 * go away while we have the reference count on the
4759 * eb bumped.
4760 *
4761 * We can't just set the ref bit without bumping the
4762 * ref on the eb because free_extent_buffer might
4763 * see the ref bit and try to clear it. If this happens
4764 * free_extent_buffer might end up dropping our original
4765 * ref by mistake and freeing the page before we are able
4766 * to add one more ref.
4767 *
4768 * So bump the ref count first, then set the bit. If someone
4769 * beat us to it, drop the ref we added.
4770 */
242e18c7
CM
4771 refs = atomic_read(&eb->refs);
4772 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4773 return;
4774
594831c4
JB
4775 spin_lock(&eb->refs_lock);
4776 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
0b32f4bb 4777 atomic_inc(&eb->refs);
594831c4 4778 spin_unlock(&eb->refs_lock);
0b32f4bb
JB
4779}
4780
2457aec6
MG
4781static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4782 struct page *accessed)
5df4235e
JB
4783{
4784 unsigned long num_pages, i;
4785
0b32f4bb
JB
4786 check_buffer_tree_ref(eb);
4787
5df4235e
JB
4788 num_pages = num_extent_pages(eb->start, eb->len);
4789 for (i = 0; i < num_pages; i++) {
fb85fc9a
DS
4790 struct page *p = eb->pages[i];
4791
2457aec6
MG
4792 if (p != accessed)
4793 mark_page_accessed(p);
5df4235e
JB
4794 }
4795}
4796
f28491e0
JB
4797struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4798 u64 start)
452c75c3
CS
4799{
4800 struct extent_buffer *eb;
4801
4802 rcu_read_lock();
f28491e0
JB
4803 eb = radix_tree_lookup(&fs_info->buffer_radix,
4804 start >> PAGE_CACHE_SHIFT);
452c75c3
CS
4805 if (eb && atomic_inc_not_zero(&eb->refs)) {
4806 rcu_read_unlock();
062c19e9
FM
4807 /*
4808 * Lock our eb's refs_lock to avoid races with
4809 * free_extent_buffer. When we get our eb it might be flagged
4810 * with EXTENT_BUFFER_STALE and another task running
4811 * free_extent_buffer might have seen that flag set,
4812 * eb->refs == 2, that the buffer isn't under IO (dirty and
4813 * writeback flags not set) and it's still in the tree (flag
4814 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4815 * of decrementing the extent buffer's reference count twice.
4816 * So here we could race and increment the eb's reference count,
4817 * clear its stale flag, mark it as dirty and drop our reference
4818 * before the other task finishes executing free_extent_buffer,
4819 * which would later result in an attempt to free an extent
4820 * buffer that is dirty.
4821 */
4822 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
4823 spin_lock(&eb->refs_lock);
4824 spin_unlock(&eb->refs_lock);
4825 }
2457aec6 4826 mark_extent_buffer_accessed(eb, NULL);
452c75c3
CS
4827 return eb;
4828 }
4829 rcu_read_unlock();
4830
4831 return NULL;
4832}
4833
faa2dbf0
JB
4834#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4835struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
ce3e6984 4836 u64 start)
faa2dbf0
JB
4837{
4838 struct extent_buffer *eb, *exists = NULL;
4839 int ret;
4840
4841 eb = find_extent_buffer(fs_info, start);
4842 if (eb)
4843 return eb;
3f556f78 4844 eb = alloc_dummy_extent_buffer(fs_info, start);
faa2dbf0
JB
4845 if (!eb)
4846 return NULL;
4847 eb->fs_info = fs_info;
4848again:
4849 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4850 if (ret)
4851 goto free_eb;
4852 spin_lock(&fs_info->buffer_lock);
4853 ret = radix_tree_insert(&fs_info->buffer_radix,
4854 start >> PAGE_CACHE_SHIFT, eb);
4855 spin_unlock(&fs_info->buffer_lock);
4856 radix_tree_preload_end();
4857 if (ret == -EEXIST) {
4858 exists = find_extent_buffer(fs_info, start);
4859 if (exists)
4860 goto free_eb;
4861 else
4862 goto again;
4863 }
4864 check_buffer_tree_ref(eb);
4865 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4866
4867 /*
4868 * We will free dummy extent buffer's if they come into
4869 * free_extent_buffer with a ref count of 2, but if we are using this we
4870 * want the buffers to stay in memory until we're done with them, so
4871 * bump the ref count again.
4872 */
4873 atomic_inc(&eb->refs);
4874 return eb;
4875free_eb:
4876 btrfs_release_extent_buffer(eb);
4877 return exists;
4878}
4879#endif
4880
f28491e0 4881struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
ce3e6984 4882 u64 start)
d1310b2e 4883{
ce3e6984 4884 unsigned long len = fs_info->tree_root->nodesize;
d1310b2e
CM
4885 unsigned long num_pages = num_extent_pages(start, len);
4886 unsigned long i;
4887 unsigned long index = start >> PAGE_CACHE_SHIFT;
4888 struct extent_buffer *eb;
6af118ce 4889 struct extent_buffer *exists = NULL;
d1310b2e 4890 struct page *p;
f28491e0 4891 struct address_space *mapping = fs_info->btree_inode->i_mapping;
d1310b2e 4892 int uptodate = 1;
19fe0a8b 4893 int ret;
d1310b2e 4894
f28491e0 4895 eb = find_extent_buffer(fs_info, start);
452c75c3 4896 if (eb)
6af118ce 4897 return eb;
6af118ce 4898
23d79d81 4899 eb = __alloc_extent_buffer(fs_info, start, len);
2b114d1d 4900 if (!eb)
d1310b2e
CM
4901 return NULL;
4902
727011e0 4903 for (i = 0; i < num_pages; i++, index++) {
d1b5c567 4904 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
4804b382 4905 if (!p)
6af118ce 4906 goto free_eb;
4f2de97a
JB
4907
4908 spin_lock(&mapping->private_lock);
4909 if (PagePrivate(p)) {
4910 /*
4911 * We could have already allocated an eb for this page
4912 * and attached one so lets see if we can get a ref on
4913 * the existing eb, and if we can we know it's good and
4914 * we can just return that one, else we know we can just
4915 * overwrite page->private.
4916 */
4917 exists = (struct extent_buffer *)p->private;
4918 if (atomic_inc_not_zero(&exists->refs)) {
4919 spin_unlock(&mapping->private_lock);
4920 unlock_page(p);
17de39ac 4921 page_cache_release(p);
2457aec6 4922 mark_extent_buffer_accessed(exists, p);
4f2de97a
JB
4923 goto free_eb;
4924 }
5ca64f45 4925 exists = NULL;
4f2de97a 4926
0b32f4bb 4927 /*
4f2de97a
JB
4928 * Do this so attach doesn't complain and we need to
4929 * drop the ref the old guy had.
4930 */
4931 ClearPagePrivate(p);
0b32f4bb 4932 WARN_ON(PageDirty(p));
4f2de97a 4933 page_cache_release(p);
d1310b2e 4934 }
4f2de97a
JB
4935 attach_extent_buffer_page(eb, p);
4936 spin_unlock(&mapping->private_lock);
0b32f4bb 4937 WARN_ON(PageDirty(p));
727011e0 4938 eb->pages[i] = p;
d1310b2e
CM
4939 if (!PageUptodate(p))
4940 uptodate = 0;
eb14ab8e
CM
4941
4942 /*
4943 * see below about how we avoid a nasty race with release page
4944 * and why we unlock later
4945 */
d1310b2e
CM
4946 }
4947 if (uptodate)
b4ce94de 4948 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
115391d2 4949again:
19fe0a8b
MX
4950 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4951 if (ret)
4952 goto free_eb;
4953
f28491e0
JB
4954 spin_lock(&fs_info->buffer_lock);
4955 ret = radix_tree_insert(&fs_info->buffer_radix,
4956 start >> PAGE_CACHE_SHIFT, eb);
4957 spin_unlock(&fs_info->buffer_lock);
452c75c3 4958 radix_tree_preload_end();
19fe0a8b 4959 if (ret == -EEXIST) {
f28491e0 4960 exists = find_extent_buffer(fs_info, start);
452c75c3
CS
4961 if (exists)
4962 goto free_eb;
4963 else
115391d2 4964 goto again;
6af118ce 4965 }
6af118ce 4966 /* add one reference for the tree */
0b32f4bb 4967 check_buffer_tree_ref(eb);
34b41ace 4968 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
eb14ab8e
CM
4969
4970 /*
4971 * there is a race where release page may have
4972 * tried to find this extent buffer in the radix
4973 * but failed. It will tell the VM it is safe to
4974 * reclaim the, and it will clear the page private bit.
4975 * We must make sure to set the page private bit properly
4976 * after the extent buffer is in the radix tree so
4977 * it doesn't get lost
4978 */
727011e0
CM
4979 SetPageChecked(eb->pages[0]);
4980 for (i = 1; i < num_pages; i++) {
fb85fc9a 4981 p = eb->pages[i];
727011e0
CM
4982 ClearPageChecked(p);
4983 unlock_page(p);
4984 }
4985 unlock_page(eb->pages[0]);
d1310b2e
CM
4986 return eb;
4987
6af118ce 4988free_eb:
5ca64f45 4989 WARN_ON(!atomic_dec_and_test(&eb->refs));
727011e0
CM
4990 for (i = 0; i < num_pages; i++) {
4991 if (eb->pages[i])
4992 unlock_page(eb->pages[i]);
4993 }
eb14ab8e 4994
897ca6e9 4995 btrfs_release_extent_buffer(eb);
6af118ce 4996 return exists;
d1310b2e 4997}
d1310b2e 4998
3083ee2e
JB
4999static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5000{
5001 struct extent_buffer *eb =
5002 container_of(head, struct extent_buffer, rcu_head);
5003
5004 __free_extent_buffer(eb);
5005}
5006
3083ee2e 5007/* Expects to have eb->eb_lock already held */
f7a52a40 5008static int release_extent_buffer(struct extent_buffer *eb)
3083ee2e
JB
5009{
5010 WARN_ON(atomic_read(&eb->refs) == 0);
5011 if (atomic_dec_and_test(&eb->refs)) {
34b41ace 5012 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
f28491e0 5013 struct btrfs_fs_info *fs_info = eb->fs_info;
3083ee2e 5014
815a51c7 5015 spin_unlock(&eb->refs_lock);
3083ee2e 5016
f28491e0
JB
5017 spin_lock(&fs_info->buffer_lock);
5018 radix_tree_delete(&fs_info->buffer_radix,
815a51c7 5019 eb->start >> PAGE_CACHE_SHIFT);
f28491e0 5020 spin_unlock(&fs_info->buffer_lock);
34b41ace
JB
5021 } else {
5022 spin_unlock(&eb->refs_lock);
815a51c7 5023 }
3083ee2e
JB
5024
5025 /* Should be safe to release our pages at this point */
a50924e3 5026 btrfs_release_extent_buffer_page(eb);
bcb7e449
JB
5027#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5028 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) {
5029 __free_extent_buffer(eb);
5030 return 1;
5031 }
5032#endif
3083ee2e 5033 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
e64860aa 5034 return 1;
3083ee2e
JB
5035 }
5036 spin_unlock(&eb->refs_lock);
e64860aa
JB
5037
5038 return 0;
3083ee2e
JB
5039}
5040
d1310b2e
CM
5041void free_extent_buffer(struct extent_buffer *eb)
5042{
242e18c7
CM
5043 int refs;
5044 int old;
d1310b2e
CM
5045 if (!eb)
5046 return;
5047
242e18c7
CM
5048 while (1) {
5049 refs = atomic_read(&eb->refs);
5050 if (refs <= 3)
5051 break;
5052 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5053 if (old == refs)
5054 return;
5055 }
5056
3083ee2e 5057 spin_lock(&eb->refs_lock);
815a51c7
JS
5058 if (atomic_read(&eb->refs) == 2 &&
5059 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
5060 atomic_dec(&eb->refs);
5061
3083ee2e
JB
5062 if (atomic_read(&eb->refs) == 2 &&
5063 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
0b32f4bb 5064 !extent_buffer_under_io(eb) &&
3083ee2e
JB
5065 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5066 atomic_dec(&eb->refs);
5067
5068 /*
5069 * I know this is terrible, but it's temporary until we stop tracking
5070 * the uptodate bits and such for the extent buffers.
5071 */
f7a52a40 5072 release_extent_buffer(eb);
3083ee2e
JB
5073}
5074
5075void free_extent_buffer_stale(struct extent_buffer *eb)
5076{
5077 if (!eb)
d1310b2e
CM
5078 return;
5079
3083ee2e
JB
5080 spin_lock(&eb->refs_lock);
5081 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5082
0b32f4bb 5083 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
3083ee2e
JB
5084 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5085 atomic_dec(&eb->refs);
f7a52a40 5086 release_extent_buffer(eb);
d1310b2e 5087}
d1310b2e 5088
1d4284bd 5089void clear_extent_buffer_dirty(struct extent_buffer *eb)
d1310b2e 5090{
d1310b2e
CM
5091 unsigned long i;
5092 unsigned long num_pages;
5093 struct page *page;
5094
d1310b2e
CM
5095 num_pages = num_extent_pages(eb->start, eb->len);
5096
5097 for (i = 0; i < num_pages; i++) {
fb85fc9a 5098 page = eb->pages[i];
b9473439 5099 if (!PageDirty(page))
d2c3f4f6
CM
5100 continue;
5101
a61e6f29 5102 lock_page(page);
eb14ab8e
CM
5103 WARN_ON(!PagePrivate(page));
5104
d1310b2e 5105 clear_page_dirty_for_io(page);
0ee0fda0 5106 spin_lock_irq(&page->mapping->tree_lock);
d1310b2e
CM
5107 if (!PageDirty(page)) {
5108 radix_tree_tag_clear(&page->mapping->page_tree,
5109 page_index(page),
5110 PAGECACHE_TAG_DIRTY);
5111 }
0ee0fda0 5112 spin_unlock_irq(&page->mapping->tree_lock);
bf0da8c1 5113 ClearPageError(page);
a61e6f29 5114 unlock_page(page);
d1310b2e 5115 }
0b32f4bb 5116 WARN_ON(atomic_read(&eb->refs) == 0);
d1310b2e 5117}
d1310b2e 5118
0b32f4bb 5119int set_extent_buffer_dirty(struct extent_buffer *eb)
d1310b2e
CM
5120{
5121 unsigned long i;
5122 unsigned long num_pages;
b9473439 5123 int was_dirty = 0;
d1310b2e 5124
0b32f4bb
JB
5125 check_buffer_tree_ref(eb);
5126
b9473439 5127 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
0b32f4bb 5128
d1310b2e 5129 num_pages = num_extent_pages(eb->start, eb->len);
3083ee2e 5130 WARN_ON(atomic_read(&eb->refs) == 0);
0b32f4bb
JB
5131 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5132
b9473439 5133 for (i = 0; i < num_pages; i++)
fb85fc9a 5134 set_page_dirty(eb->pages[i]);
b9473439 5135 return was_dirty;
d1310b2e 5136}
d1310b2e 5137
69ba3927 5138void clear_extent_buffer_uptodate(struct extent_buffer *eb)
1259ab75
CM
5139{
5140 unsigned long i;
5141 struct page *page;
5142 unsigned long num_pages;
5143
b4ce94de 5144 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
0b32f4bb 5145 num_pages = num_extent_pages(eb->start, eb->len);
1259ab75 5146 for (i = 0; i < num_pages; i++) {
fb85fc9a 5147 page = eb->pages[i];
33958dc6
CM
5148 if (page)
5149 ClearPageUptodate(page);
1259ab75 5150 }
1259ab75
CM
5151}
5152
09c25a8c 5153void set_extent_buffer_uptodate(struct extent_buffer *eb)
d1310b2e
CM
5154{
5155 unsigned long i;
5156 struct page *page;
5157 unsigned long num_pages;
5158
0b32f4bb 5159 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
d1310b2e 5160 num_pages = num_extent_pages(eb->start, eb->len);
d1310b2e 5161 for (i = 0; i < num_pages; i++) {
fb85fc9a 5162 page = eb->pages[i];
d1310b2e
CM
5163 SetPageUptodate(page);
5164 }
d1310b2e 5165}
d1310b2e 5166
0b32f4bb 5167int extent_buffer_uptodate(struct extent_buffer *eb)
d1310b2e 5168{
0b32f4bb 5169 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
d1310b2e 5170}
d1310b2e
CM
5171
5172int read_extent_buffer_pages(struct extent_io_tree *tree,
bb82ab88 5173 struct extent_buffer *eb, u64 start, int wait,
f188591e 5174 get_extent_t *get_extent, int mirror_num)
d1310b2e
CM
5175{
5176 unsigned long i;
5177 unsigned long start_i;
5178 struct page *page;
5179 int err;
5180 int ret = 0;
ce9adaa5
CM
5181 int locked_pages = 0;
5182 int all_uptodate = 1;
d1310b2e 5183 unsigned long num_pages;
727011e0 5184 unsigned long num_reads = 0;
a86c12c7 5185 struct bio *bio = NULL;
c8b97818 5186 unsigned long bio_flags = 0;
a86c12c7 5187
b4ce94de 5188 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
d1310b2e
CM
5189 return 0;
5190
d1310b2e
CM
5191 if (start) {
5192 WARN_ON(start < eb->start);
5193 start_i = (start >> PAGE_CACHE_SHIFT) -
5194 (eb->start >> PAGE_CACHE_SHIFT);
5195 } else {
5196 start_i = 0;
5197 }
5198
5199 num_pages = num_extent_pages(eb->start, eb->len);
5200 for (i = start_i; i < num_pages; i++) {
fb85fc9a 5201 page = eb->pages[i];
bb82ab88 5202 if (wait == WAIT_NONE) {
2db04966 5203 if (!trylock_page(page))
ce9adaa5 5204 goto unlock_exit;
d1310b2e
CM
5205 } else {
5206 lock_page(page);
5207 }
ce9adaa5 5208 locked_pages++;
727011e0
CM
5209 if (!PageUptodate(page)) {
5210 num_reads++;
ce9adaa5 5211 all_uptodate = 0;
727011e0 5212 }
ce9adaa5
CM
5213 }
5214 if (all_uptodate) {
5215 if (start_i == 0)
b4ce94de 5216 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
ce9adaa5
CM
5217 goto unlock_exit;
5218 }
5219
656f30db 5220 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5cf1ab56 5221 eb->read_mirror = 0;
0b32f4bb 5222 atomic_set(&eb->io_pages, num_reads);
ce9adaa5 5223 for (i = start_i; i < num_pages; i++) {
fb85fc9a 5224 page = eb->pages[i];
ce9adaa5 5225 if (!PageUptodate(page)) {
f188591e 5226 ClearPageError(page);
a86c12c7 5227 err = __extent_read_full_page(tree, page,
f188591e 5228 get_extent, &bio,
d4c7ca86
JB
5229 mirror_num, &bio_flags,
5230 READ | REQ_META);
d397712b 5231 if (err)
d1310b2e 5232 ret = err;
d1310b2e
CM
5233 } else {
5234 unlock_page(page);
5235 }
5236 }
5237
355808c2 5238 if (bio) {
d4c7ca86
JB
5239 err = submit_one_bio(READ | REQ_META, bio, mirror_num,
5240 bio_flags);
79787eaa
JM
5241 if (err)
5242 return err;
355808c2 5243 }
a86c12c7 5244
bb82ab88 5245 if (ret || wait != WAIT_COMPLETE)
d1310b2e 5246 return ret;
d397712b 5247
d1310b2e 5248 for (i = start_i; i < num_pages; i++) {
fb85fc9a 5249 page = eb->pages[i];
d1310b2e 5250 wait_on_page_locked(page);
d397712b 5251 if (!PageUptodate(page))
d1310b2e 5252 ret = -EIO;
d1310b2e 5253 }
d397712b 5254
d1310b2e 5255 return ret;
ce9adaa5
CM
5256
5257unlock_exit:
5258 i = start_i;
d397712b 5259 while (locked_pages > 0) {
fb85fc9a 5260 page = eb->pages[i];
ce9adaa5
CM
5261 i++;
5262 unlock_page(page);
5263 locked_pages--;
5264 }
5265 return ret;
d1310b2e 5266}
d1310b2e
CM
5267
5268void read_extent_buffer(struct extent_buffer *eb, void *dstv,
5269 unsigned long start,
5270 unsigned long len)
5271{
5272 size_t cur;
5273 size_t offset;
5274 struct page *page;
5275 char *kaddr;
5276 char *dst = (char *)dstv;
5277 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5278 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
d1310b2e
CM
5279
5280 WARN_ON(start > eb->len);
5281 WARN_ON(start + len > eb->start + eb->len);
5282
778746b5 5283 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
d1310b2e 5284
d397712b 5285 while (len > 0) {
fb85fc9a 5286 page = eb->pages[i];
d1310b2e
CM
5287
5288 cur = min(len, (PAGE_CACHE_SIZE - offset));
a6591715 5289 kaddr = page_address(page);
d1310b2e 5290 memcpy(dst, kaddr + offset, cur);
d1310b2e
CM
5291
5292 dst += cur;
5293 len -= cur;
5294 offset = 0;
5295 i++;
5296 }
5297}
d1310b2e 5298
550ac1d8
GH
5299int read_extent_buffer_to_user(struct extent_buffer *eb, void __user *dstv,
5300 unsigned long start,
5301 unsigned long len)
5302{
5303 size_t cur;
5304 size_t offset;
5305 struct page *page;
5306 char *kaddr;
5307 char __user *dst = (char __user *)dstv;
5308 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5309 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5310 int ret = 0;
5311
5312 WARN_ON(start > eb->len);
5313 WARN_ON(start + len > eb->start + eb->len);
5314
5315 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5316
5317 while (len > 0) {
fb85fc9a 5318 page = eb->pages[i];
550ac1d8
GH
5319
5320 cur = min(len, (PAGE_CACHE_SIZE - offset));
5321 kaddr = page_address(page);
5322 if (copy_to_user(dst, kaddr + offset, cur)) {
5323 ret = -EFAULT;
5324 break;
5325 }
5326
5327 dst += cur;
5328 len -= cur;
5329 offset = 0;
5330 i++;
5331 }
5332
5333 return ret;
5334}
5335
d1310b2e 5336int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
a6591715 5337 unsigned long min_len, char **map,
d1310b2e 5338 unsigned long *map_start,
a6591715 5339 unsigned long *map_len)
d1310b2e
CM
5340{
5341 size_t offset = start & (PAGE_CACHE_SIZE - 1);
5342 char *kaddr;
5343 struct page *p;
5344 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5345 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5346 unsigned long end_i = (start_offset + start + min_len - 1) >>
5347 PAGE_CACHE_SHIFT;
5348
5349 if (i != end_i)
5350 return -EINVAL;
5351
5352 if (i == 0) {
5353 offset = start_offset;
5354 *map_start = 0;
5355 } else {
5356 offset = 0;
5357 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
5358 }
d397712b 5359
d1310b2e 5360 if (start + min_len > eb->len) {
31b1a2bd 5361 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
c1c9ff7c
GU
5362 "wanted %lu %lu\n",
5363 eb->start, eb->len, start, min_len);
85026533 5364 return -EINVAL;
d1310b2e
CM
5365 }
5366
fb85fc9a 5367 p = eb->pages[i];
a6591715 5368 kaddr = page_address(p);
d1310b2e
CM
5369 *map = kaddr + offset;
5370 *map_len = PAGE_CACHE_SIZE - offset;
5371 return 0;
5372}
d1310b2e 5373
d1310b2e
CM
5374int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
5375 unsigned long start,
5376 unsigned long len)
5377{
5378 size_t cur;
5379 size_t offset;
5380 struct page *page;
5381 char *kaddr;
5382 char *ptr = (char *)ptrv;
5383 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5384 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5385 int ret = 0;
5386
5387 WARN_ON(start > eb->len);
5388 WARN_ON(start + len > eb->start + eb->len);
5389
778746b5 5390 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
d1310b2e 5391
d397712b 5392 while (len > 0) {
fb85fc9a 5393 page = eb->pages[i];
d1310b2e
CM
5394
5395 cur = min(len, (PAGE_CACHE_SIZE - offset));
5396
a6591715 5397 kaddr = page_address(page);
d1310b2e 5398 ret = memcmp(ptr, kaddr + offset, cur);
d1310b2e
CM
5399 if (ret)
5400 break;
5401
5402 ptr += cur;
5403 len -= cur;
5404 offset = 0;
5405 i++;
5406 }
5407 return ret;
5408}
d1310b2e
CM
5409
5410void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5411 unsigned long start, unsigned long len)
5412{
5413 size_t cur;
5414 size_t offset;
5415 struct page *page;
5416 char *kaddr;
5417 char *src = (char *)srcv;
5418 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5419 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5420
5421 WARN_ON(start > eb->len);
5422 WARN_ON(start + len > eb->start + eb->len);
5423
778746b5 5424 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
d1310b2e 5425
d397712b 5426 while (len > 0) {
fb85fc9a 5427 page = eb->pages[i];
d1310b2e
CM
5428 WARN_ON(!PageUptodate(page));
5429
5430 cur = min(len, PAGE_CACHE_SIZE - offset);
a6591715 5431 kaddr = page_address(page);
d1310b2e 5432 memcpy(kaddr + offset, src, cur);
d1310b2e
CM
5433
5434 src += cur;
5435 len -= cur;
5436 offset = 0;
5437 i++;
5438 }
5439}
d1310b2e
CM
5440
5441void memset_extent_buffer(struct extent_buffer *eb, char c,
5442 unsigned long start, unsigned long len)
5443{
5444 size_t cur;
5445 size_t offset;
5446 struct page *page;
5447 char *kaddr;
5448 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5449 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5450
5451 WARN_ON(start > eb->len);
5452 WARN_ON(start + len > eb->start + eb->len);
5453
778746b5 5454 offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
d1310b2e 5455
d397712b 5456 while (len > 0) {
fb85fc9a 5457 page = eb->pages[i];
d1310b2e
CM
5458 WARN_ON(!PageUptodate(page));
5459
5460 cur = min(len, PAGE_CACHE_SIZE - offset);
a6591715 5461 kaddr = page_address(page);
d1310b2e 5462 memset(kaddr + offset, c, cur);
d1310b2e
CM
5463
5464 len -= cur;
5465 offset = 0;
5466 i++;
5467 }
5468}
d1310b2e
CM
5469
5470void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5471 unsigned long dst_offset, unsigned long src_offset,
5472 unsigned long len)
5473{
5474 u64 dst_len = dst->len;
5475 size_t cur;
5476 size_t offset;
5477 struct page *page;
5478 char *kaddr;
5479 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5480 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5481
5482 WARN_ON(src->len != dst_len);
5483
5484 offset = (start_offset + dst_offset) &
778746b5 5485 (PAGE_CACHE_SIZE - 1);
d1310b2e 5486
d397712b 5487 while (len > 0) {
fb85fc9a 5488 page = dst->pages[i];
d1310b2e
CM
5489 WARN_ON(!PageUptodate(page));
5490
5491 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
5492
a6591715 5493 kaddr = page_address(page);
d1310b2e 5494 read_extent_buffer(src, kaddr + offset, src_offset, cur);
d1310b2e
CM
5495
5496 src_offset += cur;
5497 len -= cur;
5498 offset = 0;
5499 i++;
5500 }
5501}
d1310b2e 5502
3e1e8bb7
OS
5503/*
5504 * The extent buffer bitmap operations are done with byte granularity because
5505 * bitmap items are not guaranteed to be aligned to a word and therefore a
5506 * single word in a bitmap may straddle two pages in the extent buffer.
5507 */
5508#define BIT_BYTE(nr) ((nr) / BITS_PER_BYTE)
5509#define BYTE_MASK ((1 << BITS_PER_BYTE) - 1)
5510#define BITMAP_FIRST_BYTE_MASK(start) \
5511 ((BYTE_MASK << ((start) & (BITS_PER_BYTE - 1))) & BYTE_MASK)
5512#define BITMAP_LAST_BYTE_MASK(nbits) \
5513 (BYTE_MASK >> (-(nbits) & (BITS_PER_BYTE - 1)))
5514
5515/*
5516 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5517 * given bit number
5518 * @eb: the extent buffer
5519 * @start: offset of the bitmap item in the extent buffer
5520 * @nr: bit number
5521 * @page_index: return index of the page in the extent buffer that contains the
5522 * given bit number
5523 * @page_offset: return offset into the page given by page_index
5524 *
5525 * This helper hides the ugliness of finding the byte in an extent buffer which
5526 * contains a given bit.
5527 */
5528static inline void eb_bitmap_offset(struct extent_buffer *eb,
5529 unsigned long start, unsigned long nr,
5530 unsigned long *page_index,
5531 size_t *page_offset)
5532{
5533 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5534 size_t byte_offset = BIT_BYTE(nr);
5535 size_t offset;
5536
5537 /*
5538 * The byte we want is the offset of the extent buffer + the offset of
5539 * the bitmap item in the extent buffer + the offset of the byte in the
5540 * bitmap item.
5541 */
5542 offset = start_offset + start + byte_offset;
5543
5544 *page_index = offset >> PAGE_CACHE_SHIFT;
5545 *page_offset = offset & (PAGE_CACHE_SIZE - 1);
5546}
5547
5548/**
5549 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5550 * @eb: the extent buffer
5551 * @start: offset of the bitmap item in the extent buffer
5552 * @nr: bit number to test
5553 */
5554int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
5555 unsigned long nr)
5556{
5557 char *kaddr;
5558 struct page *page;
5559 unsigned long i;
5560 size_t offset;
5561
5562 eb_bitmap_offset(eb, start, nr, &i, &offset);
5563 page = eb->pages[i];
5564 WARN_ON(!PageUptodate(page));
5565 kaddr = page_address(page);
5566 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5567}
5568
5569/**
5570 * extent_buffer_bitmap_set - set an area of a bitmap
5571 * @eb: the extent buffer
5572 * @start: offset of the bitmap item in the extent buffer
5573 * @pos: bit number of the first bit
5574 * @len: number of bits to set
5575 */
5576void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
5577 unsigned long pos, unsigned long len)
5578{
5579 char *kaddr;
5580 struct page *page;
5581 unsigned long i;
5582 size_t offset;
5583 const unsigned int size = pos + len;
5584 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5585 unsigned int mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5586
5587 eb_bitmap_offset(eb, start, pos, &i, &offset);
5588 page = eb->pages[i];
5589 WARN_ON(!PageUptodate(page));
5590 kaddr = page_address(page);
5591
5592 while (len >= bits_to_set) {
5593 kaddr[offset] |= mask_to_set;
5594 len -= bits_to_set;
5595 bits_to_set = BITS_PER_BYTE;
5596 mask_to_set = ~0U;
5597 if (++offset >= PAGE_CACHE_SIZE && len > 0) {
5598 offset = 0;
5599 page = eb->pages[++i];
5600 WARN_ON(!PageUptodate(page));
5601 kaddr = page_address(page);
5602 }
5603 }
5604 if (len) {
5605 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5606 kaddr[offset] |= mask_to_set;
5607 }
5608}
5609
5610
5611/**
5612 * extent_buffer_bitmap_clear - clear an area of a bitmap
5613 * @eb: the extent buffer
5614 * @start: offset of the bitmap item in the extent buffer
5615 * @pos: bit number of the first bit
5616 * @len: number of bits to clear
5617 */
5618void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
5619 unsigned long pos, unsigned long len)
5620{
5621 char *kaddr;
5622 struct page *page;
5623 unsigned long i;
5624 size_t offset;
5625 const unsigned int size = pos + len;
5626 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5627 unsigned int mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5628
5629 eb_bitmap_offset(eb, start, pos, &i, &offset);
5630 page = eb->pages[i];
5631 WARN_ON(!PageUptodate(page));
5632 kaddr = page_address(page);
5633
5634 while (len >= bits_to_clear) {
5635 kaddr[offset] &= ~mask_to_clear;
5636 len -= bits_to_clear;
5637 bits_to_clear = BITS_PER_BYTE;
5638 mask_to_clear = ~0U;
5639 if (++offset >= PAGE_CACHE_SIZE && len > 0) {
5640 offset = 0;
5641 page = eb->pages[++i];
5642 WARN_ON(!PageUptodate(page));
5643 kaddr = page_address(page);
5644 }
5645 }
5646 if (len) {
5647 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5648 kaddr[offset] &= ~mask_to_clear;
5649 }
5650}
5651
3387206f
ST
5652static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5653{
5654 unsigned long distance = (src > dst) ? src - dst : dst - src;
5655 return distance < len;
5656}
5657
d1310b2e
CM
5658static void copy_pages(struct page *dst_page, struct page *src_page,
5659 unsigned long dst_off, unsigned long src_off,
5660 unsigned long len)
5661{
a6591715 5662 char *dst_kaddr = page_address(dst_page);
d1310b2e 5663 char *src_kaddr;
727011e0 5664 int must_memmove = 0;
d1310b2e 5665
3387206f 5666 if (dst_page != src_page) {
a6591715 5667 src_kaddr = page_address(src_page);
3387206f 5668 } else {
d1310b2e 5669 src_kaddr = dst_kaddr;
727011e0
CM
5670 if (areas_overlap(src_off, dst_off, len))
5671 must_memmove = 1;
3387206f 5672 }
d1310b2e 5673
727011e0
CM
5674 if (must_memmove)
5675 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5676 else
5677 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
d1310b2e
CM
5678}
5679
5680void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5681 unsigned long src_offset, unsigned long len)
5682{
5683 size_t cur;
5684 size_t dst_off_in_page;
5685 size_t src_off_in_page;
5686 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5687 unsigned long dst_i;
5688 unsigned long src_i;
5689
5690 if (src_offset + len > dst->len) {
f14d104d
DS
5691 btrfs_err(dst->fs_info,
5692 "memmove bogus src_offset %lu move "
5693 "len %lu dst len %lu", src_offset, len, dst->len);
d1310b2e
CM
5694 BUG_ON(1);
5695 }
5696 if (dst_offset + len > dst->len) {
f14d104d
DS
5697 btrfs_err(dst->fs_info,
5698 "memmove bogus dst_offset %lu move "
5699 "len %lu dst len %lu", dst_offset, len, dst->len);
d1310b2e
CM
5700 BUG_ON(1);
5701 }
5702
d397712b 5703 while (len > 0) {
d1310b2e 5704 dst_off_in_page = (start_offset + dst_offset) &
778746b5 5705 (PAGE_CACHE_SIZE - 1);
d1310b2e 5706 src_off_in_page = (start_offset + src_offset) &
778746b5 5707 (PAGE_CACHE_SIZE - 1);
d1310b2e
CM
5708
5709 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5710 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
5711
5712 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
5713 src_off_in_page));
5714 cur = min_t(unsigned long, cur,
5715 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
5716
fb85fc9a 5717 copy_pages(dst->pages[dst_i], dst->pages[src_i],
d1310b2e
CM
5718 dst_off_in_page, src_off_in_page, cur);
5719
5720 src_offset += cur;
5721 dst_offset += cur;
5722 len -= cur;
5723 }
5724}
d1310b2e
CM
5725
5726void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5727 unsigned long src_offset, unsigned long len)
5728{
5729 size_t cur;
5730 size_t dst_off_in_page;
5731 size_t src_off_in_page;
5732 unsigned long dst_end = dst_offset + len - 1;
5733 unsigned long src_end = src_offset + len - 1;
5734 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5735 unsigned long dst_i;
5736 unsigned long src_i;
5737
5738 if (src_offset + len > dst->len) {
f14d104d
DS
5739 btrfs_err(dst->fs_info, "memmove bogus src_offset %lu move "
5740 "len %lu len %lu", src_offset, len, dst->len);
d1310b2e
CM
5741 BUG_ON(1);
5742 }
5743 if (dst_offset + len > dst->len) {
f14d104d
DS
5744 btrfs_err(dst->fs_info, "memmove bogus dst_offset %lu move "
5745 "len %lu len %lu", dst_offset, len, dst->len);
d1310b2e
CM
5746 BUG_ON(1);
5747 }
727011e0 5748 if (dst_offset < src_offset) {
d1310b2e
CM
5749 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5750 return;
5751 }
d397712b 5752 while (len > 0) {
d1310b2e
CM
5753 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
5754 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
5755
5756 dst_off_in_page = (start_offset + dst_end) &
778746b5 5757 (PAGE_CACHE_SIZE - 1);
d1310b2e 5758 src_off_in_page = (start_offset + src_end) &
778746b5 5759 (PAGE_CACHE_SIZE - 1);
d1310b2e
CM
5760
5761 cur = min_t(unsigned long, len, src_off_in_page + 1);
5762 cur = min(cur, dst_off_in_page + 1);
fb85fc9a 5763 copy_pages(dst->pages[dst_i], dst->pages[src_i],
d1310b2e
CM
5764 dst_off_in_page - cur + 1,
5765 src_off_in_page - cur + 1, cur);
5766
5767 dst_end -= cur;
5768 src_end -= cur;
5769 len -= cur;
5770 }
5771}
6af118ce 5772
f7a52a40 5773int try_release_extent_buffer(struct page *page)
19fe0a8b 5774{
6af118ce 5775 struct extent_buffer *eb;
6af118ce 5776
3083ee2e
JB
5777 /*
5778 * We need to make sure noboody is attaching this page to an eb right
5779 * now.
5780 */
5781 spin_lock(&page->mapping->private_lock);
5782 if (!PagePrivate(page)) {
5783 spin_unlock(&page->mapping->private_lock);
4f2de97a 5784 return 1;
45f49bce 5785 }
6af118ce 5786
3083ee2e
JB
5787 eb = (struct extent_buffer *)page->private;
5788 BUG_ON(!eb);
19fe0a8b
MX
5789
5790 /*
3083ee2e
JB
5791 * This is a little awful but should be ok, we need to make sure that
5792 * the eb doesn't disappear out from under us while we're looking at
5793 * this page.
19fe0a8b 5794 */
3083ee2e 5795 spin_lock(&eb->refs_lock);
0b32f4bb 5796 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
3083ee2e
JB
5797 spin_unlock(&eb->refs_lock);
5798 spin_unlock(&page->mapping->private_lock);
5799 return 0;
b9473439 5800 }
3083ee2e 5801 spin_unlock(&page->mapping->private_lock);
897ca6e9 5802
19fe0a8b 5803 /*
3083ee2e
JB
5804 * If tree ref isn't set then we know the ref on this eb is a real ref,
5805 * so just return, this page will likely be freed soon anyway.
19fe0a8b 5806 */
3083ee2e
JB
5807 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5808 spin_unlock(&eb->refs_lock);
5809 return 0;
b9473439 5810 }
19fe0a8b 5811
f7a52a40 5812 return release_extent_buffer(eb);
6af118ce 5813}