1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/bitops.h>
3 #include <linux/slab.h>
6 #include <linux/pagemap.h>
7 #include <linux/page-flags.h>
8 #include <linux/spinlock.h>
9 #include <linux/blkdev.h>
10 #include <linux/swap.h>
11 #include <linux/writeback.h>
12 #include <linux/pagevec.h>
13 #include <linux/prefetch.h>
14 #include <linux/cleancache.h>
15 #include "extent_io.h"
16 #include "extent_map.h"
18 #include "btrfs_inode.h"
20 #include "check-integrity.h"
22 #include "rcu-string.h"
25 static struct kmem_cache *extent_state_cache;
26 static struct kmem_cache *extent_buffer_cache;
27 static struct bio_set *btrfs_bioset;
29 static inline bool extent_state_in_tree(const struct extent_state *state)
31 return !RB_EMPTY_NODE(&state->rb_node);
34 #ifdef CONFIG_BTRFS_DEBUG
35 static LIST_HEAD(buffers);
36 static LIST_HEAD(states);
38 static DEFINE_SPINLOCK(leak_lock);
41 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
45 spin_lock_irqsave(&leak_lock, flags);
47 spin_unlock_irqrestore(&leak_lock, flags);
51 void btrfs_leak_debug_del(struct list_head *entry)
55 spin_lock_irqsave(&leak_lock, flags);
57 spin_unlock_irqrestore(&leak_lock, flags);
61 void btrfs_leak_debug_check(void)
63 struct extent_state *state;
64 struct extent_buffer *eb;
66 while (!list_empty(&states)) {
67 state = list_entry(states.next, struct extent_state, leak_list);
68 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
69 state->start, state->end, state->state,
70 extent_state_in_tree(state),
71 refcount_read(&state->refs));
72 list_del(&state->leak_list);
73 kmem_cache_free(extent_state_cache, state);
76 while (!list_empty(&buffers)) {
77 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
78 pr_err("BTRFS: buffer leak start %llu len %lu refs %d\n",
79 eb->start, eb->len, atomic_read(&eb->refs));
80 list_del(&eb->leak_list);
81 kmem_cache_free(extent_buffer_cache, eb);
85 #define btrfs_debug_check_extent_io_range(tree, start, end) \
86 __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
87 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
88 struct extent_io_tree *tree, u64 start, u64 end)
90 if (tree->ops && tree->ops->check_extent_io_range)
91 tree->ops->check_extent_io_range(tree->private_data, caller,
95 #define btrfs_leak_debug_add(new, head) do {} while (0)
96 #define btrfs_leak_debug_del(entry) do {} while (0)
97 #define btrfs_leak_debug_check() do {} while (0)
98 #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0)
101 #define BUFFER_LRU_MAX 64
106 struct rb_node rb_node;
109 struct extent_page_data {
111 struct extent_io_tree *tree;
112 get_extent_t *get_extent;
113 unsigned long bio_flags;
115 /* tells writepage not to lock the state bits for this range
116 * it still does the unlocking
118 unsigned int extent_locked:1;
120 /* tells the submit_bio code to use REQ_SYNC */
121 unsigned int sync_io:1;
124 static void add_extent_changeset(struct extent_state *state, unsigned bits,
125 struct extent_changeset *changeset,
132 if (set && (state->state & bits) == bits)
134 if (!set && (state->state & bits) == 0)
136 changeset->bytes_changed += state->end - state->start + 1;
137 ret = ulist_add(&changeset->range_changed, state->start, state->end,
143 static noinline void flush_write_bio(void *data);
144 static inline struct btrfs_fs_info *
145 tree_fs_info(struct extent_io_tree *tree)
148 return tree->ops->tree_fs_info(tree->private_data);
152 int __init extent_io_init(void)
154 extent_state_cache = kmem_cache_create("btrfs_extent_state",
155 sizeof(struct extent_state), 0,
156 SLAB_MEM_SPREAD, NULL);
157 if (!extent_state_cache)
160 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
161 sizeof(struct extent_buffer), 0,
162 SLAB_MEM_SPREAD, NULL);
163 if (!extent_buffer_cache)
164 goto free_state_cache;
166 btrfs_bioset = bioset_create(BIO_POOL_SIZE,
167 offsetof(struct btrfs_io_bio, bio),
170 goto free_buffer_cache;
172 if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE))
178 bioset_free(btrfs_bioset);
182 kmem_cache_destroy(extent_buffer_cache);
183 extent_buffer_cache = NULL;
186 kmem_cache_destroy(extent_state_cache);
187 extent_state_cache = NULL;
191 void extent_io_exit(void)
193 btrfs_leak_debug_check();
196 * Make sure all delayed rcu free are flushed before we
200 kmem_cache_destroy(extent_state_cache);
201 kmem_cache_destroy(extent_buffer_cache);
203 bioset_free(btrfs_bioset);
206 void extent_io_tree_init(struct extent_io_tree *tree,
209 tree->state = RB_ROOT;
211 tree->dirty_bytes = 0;
212 spin_lock_init(&tree->lock);
213 tree->private_data = private_data;
216 static struct extent_state *alloc_extent_state(gfp_t mask)
218 struct extent_state *state;
221 * The given mask might be not appropriate for the slab allocator,
222 * drop the unsupported bits
224 mask &= ~(__GFP_DMA32|__GFP_HIGHMEM);
225 state = kmem_cache_alloc(extent_state_cache, mask);
229 state->failrec = NULL;
230 RB_CLEAR_NODE(&state->rb_node);
231 btrfs_leak_debug_add(&state->leak_list, &states);
232 refcount_set(&state->refs, 1);
233 init_waitqueue_head(&state->wq);
234 trace_alloc_extent_state(state, mask, _RET_IP_);
238 void free_extent_state(struct extent_state *state)
242 if (refcount_dec_and_test(&state->refs)) {
243 WARN_ON(extent_state_in_tree(state));
244 btrfs_leak_debug_del(&state->leak_list);
245 trace_free_extent_state(state, _RET_IP_);
246 kmem_cache_free(extent_state_cache, state);
250 static struct rb_node *tree_insert(struct rb_root *root,
251 struct rb_node *search_start,
253 struct rb_node *node,
254 struct rb_node ***p_in,
255 struct rb_node **parent_in)
258 struct rb_node *parent = NULL;
259 struct tree_entry *entry;
261 if (p_in && parent_in) {
267 p = search_start ? &search_start : &root->rb_node;
270 entry = rb_entry(parent, struct tree_entry, rb_node);
272 if (offset < entry->start)
274 else if (offset > entry->end)
281 rb_link_node(node, parent, p);
282 rb_insert_color(node, root);
286 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
287 struct rb_node **prev_ret,
288 struct rb_node **next_ret,
289 struct rb_node ***p_ret,
290 struct rb_node **parent_ret)
292 struct rb_root *root = &tree->state;
293 struct rb_node **n = &root->rb_node;
294 struct rb_node *prev = NULL;
295 struct rb_node *orig_prev = NULL;
296 struct tree_entry *entry;
297 struct tree_entry *prev_entry = NULL;
301 entry = rb_entry(prev, struct tree_entry, rb_node);
304 if (offset < entry->start)
306 else if (offset > entry->end)
319 while (prev && offset > prev_entry->end) {
320 prev = rb_next(prev);
321 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
328 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
329 while (prev && offset < prev_entry->start) {
330 prev = rb_prev(prev);
331 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
338 static inline struct rb_node *
339 tree_search_for_insert(struct extent_io_tree *tree,
341 struct rb_node ***p_ret,
342 struct rb_node **parent_ret)
344 struct rb_node *prev = NULL;
347 ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret);
353 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
356 return tree_search_for_insert(tree, offset, NULL, NULL);
359 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
360 struct extent_state *other)
362 if (tree->ops && tree->ops->merge_extent_hook)
363 tree->ops->merge_extent_hook(tree->private_data, new, other);
367 * utility function to look for merge candidates inside a given range.
368 * Any extents with matching state are merged together into a single
369 * extent in the tree. Extents with EXTENT_IO in their state field
370 * are not merged because the end_io handlers need to be able to do
371 * operations on them without sleeping (or doing allocations/splits).
373 * This should be called with the tree lock held.
375 static void merge_state(struct extent_io_tree *tree,
376 struct extent_state *state)
378 struct extent_state *other;
379 struct rb_node *other_node;
381 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
384 other_node = rb_prev(&state->rb_node);
386 other = rb_entry(other_node, struct extent_state, rb_node);
387 if (other->end == state->start - 1 &&
388 other->state == state->state) {
389 merge_cb(tree, state, other);
390 state->start = other->start;
391 rb_erase(&other->rb_node, &tree->state);
392 RB_CLEAR_NODE(&other->rb_node);
393 free_extent_state(other);
396 other_node = rb_next(&state->rb_node);
398 other = rb_entry(other_node, struct extent_state, rb_node);
399 if (other->start == state->end + 1 &&
400 other->state == state->state) {
401 merge_cb(tree, state, other);
402 state->end = other->end;
403 rb_erase(&other->rb_node, &tree->state);
404 RB_CLEAR_NODE(&other->rb_node);
405 free_extent_state(other);
410 static void set_state_cb(struct extent_io_tree *tree,
411 struct extent_state *state, unsigned *bits)
413 if (tree->ops && tree->ops->set_bit_hook)
414 tree->ops->set_bit_hook(tree->private_data, state, bits);
417 static void clear_state_cb(struct extent_io_tree *tree,
418 struct extent_state *state, unsigned *bits)
420 if (tree->ops && tree->ops->clear_bit_hook)
421 tree->ops->clear_bit_hook(tree->private_data, state, bits);
424 static void set_state_bits(struct extent_io_tree *tree,
425 struct extent_state *state, unsigned *bits,
426 struct extent_changeset *changeset);
429 * insert an extent_state struct into the tree. 'bits' are set on the
430 * struct before it is inserted.
432 * This may return -EEXIST if the extent is already there, in which case the
433 * state struct is freed.
435 * The tree lock is not taken internally. This is a utility function and
436 * probably isn't what you want to call (see set/clear_extent_bit).
438 static int insert_state(struct extent_io_tree *tree,
439 struct extent_state *state, u64 start, u64 end,
441 struct rb_node **parent,
442 unsigned *bits, struct extent_changeset *changeset)
444 struct rb_node *node;
447 WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n",
449 state->start = start;
452 set_state_bits(tree, state, bits, changeset);
454 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
456 struct extent_state *found;
457 found = rb_entry(node, struct extent_state, rb_node);
458 pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n",
459 found->start, found->end, start, end);
462 merge_state(tree, state);
466 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
469 if (tree->ops && tree->ops->split_extent_hook)
470 tree->ops->split_extent_hook(tree->private_data, orig, split);
474 * split a given extent state struct in two, inserting the preallocated
475 * struct 'prealloc' as the newly created second half. 'split' indicates an
476 * offset inside 'orig' where it should be split.
479 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
480 * are two extent state structs in the tree:
481 * prealloc: [orig->start, split - 1]
482 * orig: [ split, orig->end ]
484 * The tree locks are not taken by this function. They need to be held
487 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
488 struct extent_state *prealloc, u64 split)
490 struct rb_node *node;
492 split_cb(tree, orig, split);
494 prealloc->start = orig->start;
495 prealloc->end = split - 1;
496 prealloc->state = orig->state;
499 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
500 &prealloc->rb_node, NULL, NULL);
502 free_extent_state(prealloc);
508 static struct extent_state *next_state(struct extent_state *state)
510 struct rb_node *next = rb_next(&state->rb_node);
512 return rb_entry(next, struct extent_state, rb_node);
518 * utility function to clear some bits in an extent state struct.
519 * it will optionally wake up any one waiting on this state (wake == 1).
521 * If no bits are set on the state struct after clearing things, the
522 * struct is freed and removed from the tree
524 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
525 struct extent_state *state,
526 unsigned *bits, int wake,
527 struct extent_changeset *changeset)
529 struct extent_state *next;
530 unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS;
532 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
533 u64 range = state->end - state->start + 1;
534 WARN_ON(range > tree->dirty_bytes);
535 tree->dirty_bytes -= range;
537 clear_state_cb(tree, state, bits);
538 add_extent_changeset(state, bits_to_clear, changeset, 0);
539 state->state &= ~bits_to_clear;
542 if (state->state == 0) {
543 next = next_state(state);
544 if (extent_state_in_tree(state)) {
545 rb_erase(&state->rb_node, &tree->state);
546 RB_CLEAR_NODE(&state->rb_node);
547 free_extent_state(state);
552 merge_state(tree, state);
553 next = next_state(state);
558 static struct extent_state *
559 alloc_extent_state_atomic(struct extent_state *prealloc)
562 prealloc = alloc_extent_state(GFP_ATOMIC);
567 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
569 btrfs_panic(tree_fs_info(tree), err,
570 "Locking error: Extent tree was modified by another thread while locked.");
574 * clear some bits on a range in the tree. This may require splitting
575 * or inserting elements in the tree, so the gfp mask is used to
576 * indicate which allocations or sleeping are allowed.
578 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
579 * the given range from the tree regardless of state (ie for truncate).
581 * the range [start, end] is inclusive.
583 * This takes the tree lock, and returns 0 on success and < 0 on error.
585 static int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
586 unsigned bits, int wake, int delete,
587 struct extent_state **cached_state,
588 gfp_t mask, struct extent_changeset *changeset)
590 struct extent_state *state;
591 struct extent_state *cached;
592 struct extent_state *prealloc = NULL;
593 struct rb_node *node;
598 btrfs_debug_check_extent_io_range(tree, start, end);
600 if (bits & EXTENT_DELALLOC)
601 bits |= EXTENT_NORESERVE;
604 bits |= ~EXTENT_CTLBITS;
605 bits |= EXTENT_FIRST_DELALLOC;
607 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
610 if (!prealloc && gfpflags_allow_blocking(mask)) {
612 * Don't care for allocation failure here because we might end
613 * up not needing the pre-allocated extent state at all, which
614 * is the case if we only have in the tree extent states that
615 * cover our input range and don't cover too any other range.
616 * If we end up needing a new extent state we allocate it later.
618 prealloc = alloc_extent_state(mask);
621 spin_lock(&tree->lock);
623 cached = *cached_state;
626 *cached_state = NULL;
630 if (cached && extent_state_in_tree(cached) &&
631 cached->start <= start && cached->end > start) {
633 refcount_dec(&cached->refs);
638 free_extent_state(cached);
641 * this search will find the extents that end after
644 node = tree_search(tree, start);
647 state = rb_entry(node, struct extent_state, rb_node);
649 if (state->start > end)
651 WARN_ON(state->end < start);
652 last_end = state->end;
654 /* the state doesn't have the wanted bits, go ahead */
655 if (!(state->state & bits)) {
656 state = next_state(state);
661 * | ---- desired range ---- |
663 * | ------------- state -------------- |
665 * We need to split the extent we found, and may flip
666 * bits on second half.
668 * If the extent we found extends past our range, we
669 * just split and search again. It'll get split again
670 * the next time though.
672 * If the extent we found is inside our range, we clear
673 * the desired bit on it.
676 if (state->start < start) {
677 prealloc = alloc_extent_state_atomic(prealloc);
679 err = split_state(tree, state, prealloc, start);
681 extent_io_tree_panic(tree, err);
686 if (state->end <= end) {
687 state = clear_state_bit(tree, state, &bits, wake,
694 * | ---- desired range ---- |
696 * We need to split the extent, and clear the bit
699 if (state->start <= end && state->end > end) {
700 prealloc = alloc_extent_state_atomic(prealloc);
702 err = split_state(tree, state, prealloc, end + 1);
704 extent_io_tree_panic(tree, err);
709 clear_state_bit(tree, prealloc, &bits, wake, changeset);
715 state = clear_state_bit(tree, state, &bits, wake, changeset);
717 if (last_end == (u64)-1)
719 start = last_end + 1;
720 if (start <= end && state && !need_resched())
726 spin_unlock(&tree->lock);
727 if (gfpflags_allow_blocking(mask))
732 spin_unlock(&tree->lock);
734 free_extent_state(prealloc);
740 static void wait_on_state(struct extent_io_tree *tree,
741 struct extent_state *state)
742 __releases(tree->lock)
743 __acquires(tree->lock)
746 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
747 spin_unlock(&tree->lock);
749 spin_lock(&tree->lock);
750 finish_wait(&state->wq, &wait);
754 * waits for one or more bits to clear on a range in the state tree.
755 * The range [start, end] is inclusive.
756 * The tree lock is taken by this function
758 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
761 struct extent_state *state;
762 struct rb_node *node;
764 btrfs_debug_check_extent_io_range(tree, start, end);
766 spin_lock(&tree->lock);
770 * this search will find all the extents that end after
773 node = tree_search(tree, start);
778 state = rb_entry(node, struct extent_state, rb_node);
780 if (state->start > end)
783 if (state->state & bits) {
784 start = state->start;
785 refcount_inc(&state->refs);
786 wait_on_state(tree, state);
787 free_extent_state(state);
790 start = state->end + 1;
795 if (!cond_resched_lock(&tree->lock)) {
796 node = rb_next(node);
801 spin_unlock(&tree->lock);
804 static void set_state_bits(struct extent_io_tree *tree,
805 struct extent_state *state,
806 unsigned *bits, struct extent_changeset *changeset)
808 unsigned bits_to_set = *bits & ~EXTENT_CTLBITS;
810 set_state_cb(tree, state, bits);
811 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
812 u64 range = state->end - state->start + 1;
813 tree->dirty_bytes += range;
815 add_extent_changeset(state, bits_to_set, changeset, 1);
816 state->state |= bits_to_set;
819 static void cache_state_if_flags(struct extent_state *state,
820 struct extent_state **cached_ptr,
823 if (cached_ptr && !(*cached_ptr)) {
824 if (!flags || (state->state & flags)) {
826 refcount_inc(&state->refs);
831 static void cache_state(struct extent_state *state,
832 struct extent_state **cached_ptr)
834 return cache_state_if_flags(state, cached_ptr,
835 EXTENT_IOBITS | EXTENT_BOUNDARY);
839 * set some bits on a range in the tree. This may require allocations or
840 * sleeping, so the gfp mask is used to indicate what is allowed.
842 * If any of the exclusive bits are set, this will fail with -EEXIST if some
843 * part of the range already has the desired bits set. The start of the
844 * existing range is returned in failed_start in this case.
846 * [start, end] is inclusive This takes the tree lock.
849 static int __must_check
850 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
851 unsigned bits, unsigned exclusive_bits,
852 u64 *failed_start, struct extent_state **cached_state,
853 gfp_t mask, struct extent_changeset *changeset)
855 struct extent_state *state;
856 struct extent_state *prealloc = NULL;
857 struct rb_node *node;
859 struct rb_node *parent;
864 btrfs_debug_check_extent_io_range(tree, start, end);
866 bits |= EXTENT_FIRST_DELALLOC;
868 if (!prealloc && gfpflags_allow_blocking(mask)) {
870 * Don't care for allocation failure here because we might end
871 * up not needing the pre-allocated extent state at all, which
872 * is the case if we only have in the tree extent states that
873 * cover our input range and don't cover too any other range.
874 * If we end up needing a new extent state we allocate it later.
876 prealloc = alloc_extent_state(mask);
879 spin_lock(&tree->lock);
880 if (cached_state && *cached_state) {
881 state = *cached_state;
882 if (state->start <= start && state->end > start &&
883 extent_state_in_tree(state)) {
884 node = &state->rb_node;
889 * this search will find all the extents that end after
892 node = tree_search_for_insert(tree, start, &p, &parent);
894 prealloc = alloc_extent_state_atomic(prealloc);
896 err = insert_state(tree, prealloc, start, end,
897 &p, &parent, &bits, changeset);
899 extent_io_tree_panic(tree, err);
901 cache_state(prealloc, cached_state);
905 state = rb_entry(node, struct extent_state, rb_node);
907 last_start = state->start;
908 last_end = state->end;
911 * | ---- desired range ---- |
914 * Just lock what we found and keep going
916 if (state->start == start && state->end <= end) {
917 if (state->state & exclusive_bits) {
918 *failed_start = state->start;
923 set_state_bits(tree, state, &bits, changeset);
924 cache_state(state, cached_state);
925 merge_state(tree, state);
926 if (last_end == (u64)-1)
928 start = last_end + 1;
929 state = next_state(state);
930 if (start < end && state && state->start == start &&
937 * | ---- desired range ---- |
940 * | ------------- state -------------- |
942 * We need to split the extent we found, and may flip bits on
945 * If the extent we found extends past our
946 * range, we just split and search again. It'll get split
947 * again the next time though.
949 * If the extent we found is inside our range, we set the
952 if (state->start < start) {
953 if (state->state & exclusive_bits) {
954 *failed_start = start;
959 prealloc = alloc_extent_state_atomic(prealloc);
961 err = split_state(tree, state, prealloc, start);
963 extent_io_tree_panic(tree, err);
968 if (state->end <= end) {
969 set_state_bits(tree, state, &bits, changeset);
970 cache_state(state, cached_state);
971 merge_state(tree, state);
972 if (last_end == (u64)-1)
974 start = last_end + 1;
975 state = next_state(state);
976 if (start < end && state && state->start == start &&
983 * | ---- desired range ---- |
984 * | state | or | state |
986 * There's a hole, we need to insert something in it and
987 * ignore the extent we found.
989 if (state->start > start) {
991 if (end < last_start)
994 this_end = last_start - 1;
996 prealloc = alloc_extent_state_atomic(prealloc);
1000 * Avoid to free 'prealloc' if it can be merged with
1003 err = insert_state(tree, prealloc, start, this_end,
1004 NULL, NULL, &bits, changeset);
1006 extent_io_tree_panic(tree, err);
1008 cache_state(prealloc, cached_state);
1010 start = this_end + 1;
1014 * | ---- desired range ---- |
1016 * We need to split the extent, and set the bit
1019 if (state->start <= end && state->end > end) {
1020 if (state->state & exclusive_bits) {
1021 *failed_start = start;
1026 prealloc = alloc_extent_state_atomic(prealloc);
1028 err = split_state(tree, state, prealloc, end + 1);
1030 extent_io_tree_panic(tree, err);
1032 set_state_bits(tree, prealloc, &bits, changeset);
1033 cache_state(prealloc, cached_state);
1034 merge_state(tree, prealloc);
1042 spin_unlock(&tree->lock);
1043 if (gfpflags_allow_blocking(mask))
1048 spin_unlock(&tree->lock);
1050 free_extent_state(prealloc);
1056 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1057 unsigned bits, u64 * failed_start,
1058 struct extent_state **cached_state, gfp_t mask)
1060 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1061 cached_state, mask, NULL);
1066 * convert_extent_bit - convert all bits in a given range from one bit to
1068 * @tree: the io tree to search
1069 * @start: the start offset in bytes
1070 * @end: the end offset in bytes (inclusive)
1071 * @bits: the bits to set in this range
1072 * @clear_bits: the bits to clear in this range
1073 * @cached_state: state that we're going to cache
1075 * This will go through and set bits for the given range. If any states exist
1076 * already in this range they are set with the given bit and cleared of the
1077 * clear_bits. This is only meant to be used by things that are mergeable, ie
1078 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1079 * boundary bits like LOCK.
1081 * All allocations are done with GFP_NOFS.
1083 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1084 unsigned bits, unsigned clear_bits,
1085 struct extent_state **cached_state)
1087 struct extent_state *state;
1088 struct extent_state *prealloc = NULL;
1089 struct rb_node *node;
1091 struct rb_node *parent;
1095 bool first_iteration = true;
1097 btrfs_debug_check_extent_io_range(tree, start, end);
1102 * Best effort, don't worry if extent state allocation fails
1103 * here for the first iteration. We might have a cached state
1104 * that matches exactly the target range, in which case no
1105 * extent state allocations are needed. We'll only know this
1106 * after locking the tree.
1108 prealloc = alloc_extent_state(GFP_NOFS);
1109 if (!prealloc && !first_iteration)
1113 spin_lock(&tree->lock);
1114 if (cached_state && *cached_state) {
1115 state = *cached_state;
1116 if (state->start <= start && state->end > start &&
1117 extent_state_in_tree(state)) {
1118 node = &state->rb_node;
1124 * this search will find all the extents that end after
1127 node = tree_search_for_insert(tree, start, &p, &parent);
1129 prealloc = alloc_extent_state_atomic(prealloc);
1134 err = insert_state(tree, prealloc, start, end,
1135 &p, &parent, &bits, NULL);
1137 extent_io_tree_panic(tree, err);
1138 cache_state(prealloc, cached_state);
1142 state = rb_entry(node, struct extent_state, rb_node);
1144 last_start = state->start;
1145 last_end = state->end;
1148 * | ---- desired range ---- |
1151 * Just lock what we found and keep going
1153 if (state->start == start && state->end <= end) {
1154 set_state_bits(tree, state, &bits, NULL);
1155 cache_state(state, cached_state);
1156 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1157 if (last_end == (u64)-1)
1159 start = last_end + 1;
1160 if (start < end && state && state->start == start &&
1167 * | ---- desired range ---- |
1170 * | ------------- state -------------- |
1172 * We need to split the extent we found, and may flip bits on
1175 * If the extent we found extends past our
1176 * range, we just split and search again. It'll get split
1177 * again the next time though.
1179 * If the extent we found is inside our range, we set the
1180 * desired bit on it.
1182 if (state->start < start) {
1183 prealloc = alloc_extent_state_atomic(prealloc);
1188 err = split_state(tree, state, prealloc, start);
1190 extent_io_tree_panic(tree, err);
1194 if (state->end <= end) {
1195 set_state_bits(tree, state, &bits, NULL);
1196 cache_state(state, cached_state);
1197 state = clear_state_bit(tree, state, &clear_bits, 0,
1199 if (last_end == (u64)-1)
1201 start = last_end + 1;
1202 if (start < end && state && state->start == start &&
1209 * | ---- desired range ---- |
1210 * | state | or | state |
1212 * There's a hole, we need to insert something in it and
1213 * ignore the extent we found.
1215 if (state->start > start) {
1217 if (end < last_start)
1220 this_end = last_start - 1;
1222 prealloc = alloc_extent_state_atomic(prealloc);
1229 * Avoid to free 'prealloc' if it can be merged with
1232 err = insert_state(tree, prealloc, start, this_end,
1233 NULL, NULL, &bits, NULL);
1235 extent_io_tree_panic(tree, err);
1236 cache_state(prealloc, cached_state);
1238 start = this_end + 1;
1242 * | ---- desired range ---- |
1244 * We need to split the extent, and set the bit
1247 if (state->start <= end && state->end > end) {
1248 prealloc = alloc_extent_state_atomic(prealloc);
1254 err = split_state(tree, state, prealloc, end + 1);
1256 extent_io_tree_panic(tree, err);
1258 set_state_bits(tree, prealloc, &bits, NULL);
1259 cache_state(prealloc, cached_state);
1260 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1268 spin_unlock(&tree->lock);
1270 first_iteration = false;
1274 spin_unlock(&tree->lock);
1276 free_extent_state(prealloc);
1281 /* wrappers around set/clear extent bit */
1282 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1283 unsigned bits, struct extent_changeset *changeset)
1286 * We don't support EXTENT_LOCKED yet, as current changeset will
1287 * record any bits changed, so for EXTENT_LOCKED case, it will
1288 * either fail with -EEXIST or changeset will record the whole
1291 BUG_ON(bits & EXTENT_LOCKED);
1293 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1297 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1298 unsigned bits, int wake, int delete,
1299 struct extent_state **cached, gfp_t mask)
1301 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1302 cached, mask, NULL);
1305 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1306 unsigned bits, struct extent_changeset *changeset)
1309 * Don't support EXTENT_LOCKED case, same reason as
1310 * set_record_extent_bits().
1312 BUG_ON(bits & EXTENT_LOCKED);
1314 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1319 * either insert or lock state struct between start and end use mask to tell
1320 * us if waiting is desired.
1322 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1323 struct extent_state **cached_state)
1329 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
1330 EXTENT_LOCKED, &failed_start,
1331 cached_state, GFP_NOFS, NULL);
1332 if (err == -EEXIST) {
1333 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1334 start = failed_start;
1337 WARN_ON(start > end);
1342 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1347 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1348 &failed_start, NULL, GFP_NOFS, NULL);
1349 if (err == -EEXIST) {
1350 if (failed_start > start)
1351 clear_extent_bit(tree, start, failed_start - 1,
1352 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1358 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1360 unsigned long index = start >> PAGE_SHIFT;
1361 unsigned long end_index = end >> PAGE_SHIFT;
1364 while (index <= end_index) {
1365 page = find_get_page(inode->i_mapping, index);
1366 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1367 clear_page_dirty_for_io(page);
1373 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1375 unsigned long index = start >> PAGE_SHIFT;
1376 unsigned long end_index = end >> PAGE_SHIFT;
1379 while (index <= end_index) {
1380 page = find_get_page(inode->i_mapping, index);
1381 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1382 __set_page_dirty_nobuffers(page);
1383 account_page_redirty(page);
1390 * helper function to set both pages and extents in the tree writeback
1392 static void set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1394 tree->ops->set_range_writeback(tree->private_data, start, end);
1397 /* find the first state struct with 'bits' set after 'start', and
1398 * return it. tree->lock must be held. NULL will returned if
1399 * nothing was found after 'start'
1401 static struct extent_state *
1402 find_first_extent_bit_state(struct extent_io_tree *tree,
1403 u64 start, unsigned bits)
1405 struct rb_node *node;
1406 struct extent_state *state;
1409 * this search will find all the extents that end after
1412 node = tree_search(tree, start);
1417 state = rb_entry(node, struct extent_state, rb_node);
1418 if (state->end >= start && (state->state & bits))
1421 node = rb_next(node);
1430 * find the first offset in the io tree with 'bits' set. zero is
1431 * returned if we find something, and *start_ret and *end_ret are
1432 * set to reflect the state struct that was found.
1434 * If nothing was found, 1 is returned. If found something, return 0.
1436 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1437 u64 *start_ret, u64 *end_ret, unsigned bits,
1438 struct extent_state **cached_state)
1440 struct extent_state *state;
1444 spin_lock(&tree->lock);
1445 if (cached_state && *cached_state) {
1446 state = *cached_state;
1447 if (state->end == start - 1 && extent_state_in_tree(state)) {
1448 n = rb_next(&state->rb_node);
1450 state = rb_entry(n, struct extent_state,
1452 if (state->state & bits)
1456 free_extent_state(*cached_state);
1457 *cached_state = NULL;
1460 free_extent_state(*cached_state);
1461 *cached_state = NULL;
1464 state = find_first_extent_bit_state(tree, start, bits);
1467 cache_state_if_flags(state, cached_state, 0);
1468 *start_ret = state->start;
1469 *end_ret = state->end;
1473 spin_unlock(&tree->lock);
1478 * find a contiguous range of bytes in the file marked as delalloc, not
1479 * more than 'max_bytes'. start and end are used to return the range,
1481 * 1 is returned if we find something, 0 if nothing was in the tree
1483 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1484 u64 *start, u64 *end, u64 max_bytes,
1485 struct extent_state **cached_state)
1487 struct rb_node *node;
1488 struct extent_state *state;
1489 u64 cur_start = *start;
1491 u64 total_bytes = 0;
1493 spin_lock(&tree->lock);
1496 * this search will find all the extents that end after
1499 node = tree_search(tree, cur_start);
1507 state = rb_entry(node, struct extent_state, rb_node);
1508 if (found && (state->start != cur_start ||
1509 (state->state & EXTENT_BOUNDARY))) {
1512 if (!(state->state & EXTENT_DELALLOC)) {
1518 *start = state->start;
1519 *cached_state = state;
1520 refcount_inc(&state->refs);
1524 cur_start = state->end + 1;
1525 node = rb_next(node);
1526 total_bytes += state->end - state->start + 1;
1527 if (total_bytes >= max_bytes)
1533 spin_unlock(&tree->lock);
1537 static int __process_pages_contig(struct address_space *mapping,
1538 struct page *locked_page,
1539 pgoff_t start_index, pgoff_t end_index,
1540 unsigned long page_ops, pgoff_t *index_ret);
1542 static noinline void __unlock_for_delalloc(struct inode *inode,
1543 struct page *locked_page,
1546 unsigned long index = start >> PAGE_SHIFT;
1547 unsigned long end_index = end >> PAGE_SHIFT;
1549 ASSERT(locked_page);
1550 if (index == locked_page->index && end_index == index)
1553 __process_pages_contig(inode->i_mapping, locked_page, index, end_index,
1557 static noinline int lock_delalloc_pages(struct inode *inode,
1558 struct page *locked_page,
1562 unsigned long index = delalloc_start >> PAGE_SHIFT;
1563 unsigned long index_ret = index;
1564 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1567 ASSERT(locked_page);
1568 if (index == locked_page->index && index == end_index)
1571 ret = __process_pages_contig(inode->i_mapping, locked_page, index,
1572 end_index, PAGE_LOCK, &index_ret);
1574 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1575 (u64)index_ret << PAGE_SHIFT);
1580 * find a contiguous range of bytes in the file marked as delalloc, not
1581 * more than 'max_bytes'. start and end are used to return the range,
1583 * 1 is returned if we find something, 0 if nothing was in the tree
1585 STATIC u64 find_lock_delalloc_range(struct inode *inode,
1586 struct extent_io_tree *tree,
1587 struct page *locked_page, u64 *start,
1588 u64 *end, u64 max_bytes)
1593 struct extent_state *cached_state = NULL;
1598 /* step one, find a bunch of delalloc bytes starting at start */
1599 delalloc_start = *start;
1601 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1602 max_bytes, &cached_state);
1603 if (!found || delalloc_end <= *start) {
1604 *start = delalloc_start;
1605 *end = delalloc_end;
1606 free_extent_state(cached_state);
1611 * start comes from the offset of locked_page. We have to lock
1612 * pages in order, so we can't process delalloc bytes before
1615 if (delalloc_start < *start)
1616 delalloc_start = *start;
1619 * make sure to limit the number of pages we try to lock down
1621 if (delalloc_end + 1 - delalloc_start > max_bytes)
1622 delalloc_end = delalloc_start + max_bytes - 1;
1624 /* step two, lock all the pages after the page that has start */
1625 ret = lock_delalloc_pages(inode, locked_page,
1626 delalloc_start, delalloc_end);
1627 if (ret == -EAGAIN) {
1628 /* some of the pages are gone, lets avoid looping by
1629 * shortening the size of the delalloc range we're searching
1631 free_extent_state(cached_state);
1632 cached_state = NULL;
1634 max_bytes = PAGE_SIZE;
1642 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1644 /* step three, lock the state bits for the whole range */
1645 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1647 /* then test to make sure it is all still delalloc */
1648 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1649 EXTENT_DELALLOC, 1, cached_state);
1651 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1652 &cached_state, GFP_NOFS);
1653 __unlock_for_delalloc(inode, locked_page,
1654 delalloc_start, delalloc_end);
1658 free_extent_state(cached_state);
1659 *start = delalloc_start;
1660 *end = delalloc_end;
1665 static int __process_pages_contig(struct address_space *mapping,
1666 struct page *locked_page,
1667 pgoff_t start_index, pgoff_t end_index,
1668 unsigned long page_ops, pgoff_t *index_ret)
1670 unsigned long nr_pages = end_index - start_index + 1;
1671 unsigned long pages_locked = 0;
1672 pgoff_t index = start_index;
1673 struct page *pages[16];
1678 if (page_ops & PAGE_LOCK) {
1679 ASSERT(page_ops == PAGE_LOCK);
1680 ASSERT(index_ret && *index_ret == start_index);
1683 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1684 mapping_set_error(mapping, -EIO);
1686 while (nr_pages > 0) {
1687 ret = find_get_pages_contig(mapping, index,
1688 min_t(unsigned long,
1689 nr_pages, ARRAY_SIZE(pages)), pages);
1692 * Only if we're going to lock these pages,
1693 * can we find nothing at @index.
1695 ASSERT(page_ops & PAGE_LOCK);
1700 for (i = 0; i < ret; i++) {
1701 if (page_ops & PAGE_SET_PRIVATE2)
1702 SetPagePrivate2(pages[i]);
1704 if (pages[i] == locked_page) {
1709 if (page_ops & PAGE_CLEAR_DIRTY)
1710 clear_page_dirty_for_io(pages[i]);
1711 if (page_ops & PAGE_SET_WRITEBACK)
1712 set_page_writeback(pages[i]);
1713 if (page_ops & PAGE_SET_ERROR)
1714 SetPageError(pages[i]);
1715 if (page_ops & PAGE_END_WRITEBACK)
1716 end_page_writeback(pages[i]);
1717 if (page_ops & PAGE_UNLOCK)
1718 unlock_page(pages[i]);
1719 if (page_ops & PAGE_LOCK) {
1720 lock_page(pages[i]);
1721 if (!PageDirty(pages[i]) ||
1722 pages[i]->mapping != mapping) {
1723 unlock_page(pages[i]);
1737 if (err && index_ret)
1738 *index_ret = start_index + pages_locked - 1;
1742 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1743 u64 delalloc_end, struct page *locked_page,
1744 unsigned clear_bits,
1745 unsigned long page_ops)
1747 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0,
1750 __process_pages_contig(inode->i_mapping, locked_page,
1751 start >> PAGE_SHIFT, end >> PAGE_SHIFT,
1756 * count the number of bytes in the tree that have a given bit(s)
1757 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1758 * cached. The total number found is returned.
1760 u64 count_range_bits(struct extent_io_tree *tree,
1761 u64 *start, u64 search_end, u64 max_bytes,
1762 unsigned bits, int contig)
1764 struct rb_node *node;
1765 struct extent_state *state;
1766 u64 cur_start = *start;
1767 u64 total_bytes = 0;
1771 if (WARN_ON(search_end <= cur_start))
1774 spin_lock(&tree->lock);
1775 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1776 total_bytes = tree->dirty_bytes;
1780 * this search will find all the extents that end after
1783 node = tree_search(tree, cur_start);
1788 state = rb_entry(node, struct extent_state, rb_node);
1789 if (state->start > search_end)
1791 if (contig && found && state->start > last + 1)
1793 if (state->end >= cur_start && (state->state & bits) == bits) {
1794 total_bytes += min(search_end, state->end) + 1 -
1795 max(cur_start, state->start);
1796 if (total_bytes >= max_bytes)
1799 *start = max(cur_start, state->start);
1803 } else if (contig && found) {
1806 node = rb_next(node);
1811 spin_unlock(&tree->lock);
1816 * set the private field for a given byte offset in the tree. If there isn't
1817 * an extent_state there already, this does nothing.
1819 static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start,
1820 struct io_failure_record *failrec)
1822 struct rb_node *node;
1823 struct extent_state *state;
1826 spin_lock(&tree->lock);
1828 * this search will find all the extents that end after
1831 node = tree_search(tree, start);
1836 state = rb_entry(node, struct extent_state, rb_node);
1837 if (state->start != start) {
1841 state->failrec = failrec;
1843 spin_unlock(&tree->lock);
1847 static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start,
1848 struct io_failure_record **failrec)
1850 struct rb_node *node;
1851 struct extent_state *state;
1854 spin_lock(&tree->lock);
1856 * this search will find all the extents that end after
1859 node = tree_search(tree, start);
1864 state = rb_entry(node, struct extent_state, rb_node);
1865 if (state->start != start) {
1869 *failrec = state->failrec;
1871 spin_unlock(&tree->lock);
1876 * searches a range in the state tree for a given mask.
1877 * If 'filled' == 1, this returns 1 only if every extent in the tree
1878 * has the bits set. Otherwise, 1 is returned if any bit in the
1879 * range is found set.
1881 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1882 unsigned bits, int filled, struct extent_state *cached)
1884 struct extent_state *state = NULL;
1885 struct rb_node *node;
1888 spin_lock(&tree->lock);
1889 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
1890 cached->end > start)
1891 node = &cached->rb_node;
1893 node = tree_search(tree, start);
1894 while (node && start <= end) {
1895 state = rb_entry(node, struct extent_state, rb_node);
1897 if (filled && state->start > start) {
1902 if (state->start > end)
1905 if (state->state & bits) {
1909 } else if (filled) {
1914 if (state->end == (u64)-1)
1917 start = state->end + 1;
1920 node = rb_next(node);
1927 spin_unlock(&tree->lock);
1932 * helper function to set a given page up to date if all the
1933 * extents in the tree for that page are up to date
1935 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1937 u64 start = page_offset(page);
1938 u64 end = start + PAGE_SIZE - 1;
1939 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1940 SetPageUptodate(page);
1943 int free_io_failure(struct extent_io_tree *failure_tree,
1944 struct extent_io_tree *io_tree,
1945 struct io_failure_record *rec)
1950 set_state_failrec(failure_tree, rec->start, NULL);
1951 ret = clear_extent_bits(failure_tree, rec->start,
1952 rec->start + rec->len - 1,
1953 EXTENT_LOCKED | EXTENT_DIRTY);
1957 ret = clear_extent_bits(io_tree, rec->start,
1958 rec->start + rec->len - 1,
1968 * this bypasses the standard btrfs submit functions deliberately, as
1969 * the standard behavior is to write all copies in a raid setup. here we only
1970 * want to write the one bad copy. so we do the mapping for ourselves and issue
1971 * submit_bio directly.
1972 * to avoid any synchronization issues, wait for the data after writing, which
1973 * actually prevents the read that triggered the error from finishing.
1974 * currently, there can be no more than two copies of every data bit. thus,
1975 * exactly one rewrite is required.
1977 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
1978 u64 length, u64 logical, struct page *page,
1979 unsigned int pg_offset, int mirror_num)
1982 struct btrfs_device *dev;
1985 struct btrfs_bio *bbio = NULL;
1988 ASSERT(!(fs_info->sb->s_flags & MS_RDONLY));
1989 BUG_ON(!mirror_num);
1991 bio = btrfs_io_bio_alloc(1);
1992 bio->bi_iter.bi_size = 0;
1993 map_length = length;
1996 * Avoid races with device replace and make sure our bbio has devices
1997 * associated to its stripes that don't go away while we are doing the
1998 * read repair operation.
2000 btrfs_bio_counter_inc_blocked(fs_info);
2001 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2003 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2004 * to update all raid stripes, but here we just want to correct
2005 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2006 * stripe's dev and sector.
2008 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2009 &map_length, &bbio, 0);
2011 btrfs_bio_counter_dec(fs_info);
2015 ASSERT(bbio->mirror_num == 1);
2017 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2018 &map_length, &bbio, mirror_num);
2020 btrfs_bio_counter_dec(fs_info);
2024 BUG_ON(mirror_num != bbio->mirror_num);
2027 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2028 bio->bi_iter.bi_sector = sector;
2029 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2030 btrfs_put_bbio(bbio);
2031 if (!dev || !dev->bdev || !dev->writeable) {
2032 btrfs_bio_counter_dec(fs_info);
2036 bio_set_dev(bio, dev->bdev);
2037 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2038 bio_add_page(bio, page, length, pg_offset);
2040 if (btrfsic_submit_bio_wait(bio)) {
2041 /* try to remap that extent elsewhere? */
2042 btrfs_bio_counter_dec(fs_info);
2044 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2048 btrfs_info_rl_in_rcu(fs_info,
2049 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2051 rcu_str_deref(dev->name), sector);
2052 btrfs_bio_counter_dec(fs_info);
2057 int repair_eb_io_failure(struct btrfs_fs_info *fs_info,
2058 struct extent_buffer *eb, int mirror_num)
2060 u64 start = eb->start;
2061 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2064 if (sb_rdonly(fs_info->sb))
2067 for (i = 0; i < num_pages; i++) {
2068 struct page *p = eb->pages[i];
2070 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2071 start - page_offset(p), mirror_num);
2081 * each time an IO finishes, we do a fast check in the IO failure tree
2082 * to see if we need to process or clean up an io_failure_record
2084 int clean_io_failure(struct btrfs_fs_info *fs_info,
2085 struct extent_io_tree *failure_tree,
2086 struct extent_io_tree *io_tree, u64 start,
2087 struct page *page, u64 ino, unsigned int pg_offset)
2090 struct io_failure_record *failrec;
2091 struct extent_state *state;
2096 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2101 ret = get_state_failrec(failure_tree, start, &failrec);
2105 BUG_ON(!failrec->this_mirror);
2107 if (failrec->in_validation) {
2108 /* there was no real error, just free the record */
2109 btrfs_debug(fs_info,
2110 "clean_io_failure: freeing dummy error at %llu",
2114 if (sb_rdonly(fs_info->sb))
2117 spin_lock(&io_tree->lock);
2118 state = find_first_extent_bit_state(io_tree,
2121 spin_unlock(&io_tree->lock);
2123 if (state && state->start <= failrec->start &&
2124 state->end >= failrec->start + failrec->len - 1) {
2125 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2127 if (num_copies > 1) {
2128 repair_io_failure(fs_info, ino, start, failrec->len,
2129 failrec->logical, page, pg_offset,
2130 failrec->failed_mirror);
2135 free_io_failure(failure_tree, io_tree, failrec);
2141 * Can be called when
2142 * - hold extent lock
2143 * - under ordered extent
2144 * - the inode is freeing
2146 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2148 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2149 struct io_failure_record *failrec;
2150 struct extent_state *state, *next;
2152 if (RB_EMPTY_ROOT(&failure_tree->state))
2155 spin_lock(&failure_tree->lock);
2156 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2158 if (state->start > end)
2161 ASSERT(state->end <= end);
2163 next = next_state(state);
2165 failrec = state->failrec;
2166 free_extent_state(state);
2171 spin_unlock(&failure_tree->lock);
2174 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2175 struct io_failure_record **failrec_ret)
2177 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2178 struct io_failure_record *failrec;
2179 struct extent_map *em;
2180 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2181 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2182 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2186 ret = get_state_failrec(failure_tree, start, &failrec);
2188 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2192 failrec->start = start;
2193 failrec->len = end - start + 1;
2194 failrec->this_mirror = 0;
2195 failrec->bio_flags = 0;
2196 failrec->in_validation = 0;
2198 read_lock(&em_tree->lock);
2199 em = lookup_extent_mapping(em_tree, start, failrec->len);
2201 read_unlock(&em_tree->lock);
2206 if (em->start > start || em->start + em->len <= start) {
2207 free_extent_map(em);
2210 read_unlock(&em_tree->lock);
2216 logical = start - em->start;
2217 logical = em->block_start + logical;
2218 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2219 logical = em->block_start;
2220 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2221 extent_set_compress_type(&failrec->bio_flags,
2225 btrfs_debug(fs_info,
2226 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2227 logical, start, failrec->len);
2229 failrec->logical = logical;
2230 free_extent_map(em);
2232 /* set the bits in the private failure tree */
2233 ret = set_extent_bits(failure_tree, start, end,
2234 EXTENT_LOCKED | EXTENT_DIRTY);
2236 ret = set_state_failrec(failure_tree, start, failrec);
2237 /* set the bits in the inode's tree */
2239 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2245 btrfs_debug(fs_info,
2246 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2247 failrec->logical, failrec->start, failrec->len,
2248 failrec->in_validation);
2250 * when data can be on disk more than twice, add to failrec here
2251 * (e.g. with a list for failed_mirror) to make
2252 * clean_io_failure() clean all those errors at once.
2256 *failrec_ret = failrec;
2261 bool btrfs_check_repairable(struct inode *inode, struct bio *failed_bio,
2262 struct io_failure_record *failrec, int failed_mirror)
2264 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2267 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2268 if (num_copies == 1) {
2270 * we only have a single copy of the data, so don't bother with
2271 * all the retry and error correction code that follows. no
2272 * matter what the error is, it is very likely to persist.
2274 btrfs_debug(fs_info,
2275 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2276 num_copies, failrec->this_mirror, failed_mirror);
2281 * there are two premises:
2282 * a) deliver good data to the caller
2283 * b) correct the bad sectors on disk
2285 if (failed_bio->bi_vcnt > 1) {
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
2291 * if the following BUG_ON triggers, our validation request got
2292 * merged. we need separate requests for our algorithm to work.
2294 BUG_ON(failrec->in_validation);
2295 failrec->in_validation = 1;
2296 failrec->this_mirror = failed_mirror;
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.
2303 if (failrec->in_validation) {
2304 BUG_ON(failrec->this_mirror != failed_mirror);
2305 failrec->in_validation = 0;
2306 failrec->this_mirror = 0;
2308 failrec->failed_mirror = failed_mirror;
2309 failrec->this_mirror++;
2310 if (failrec->this_mirror == failed_mirror)
2311 failrec->this_mirror++;
2314 if (failrec->this_mirror > num_copies) {
2315 btrfs_debug(fs_info,
2316 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2317 num_copies, failrec->this_mirror, failed_mirror);
2325 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2326 struct io_failure_record *failrec,
2327 struct page *page, int pg_offset, int icsum,
2328 bio_end_io_t *endio_func, void *data)
2330 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2332 struct btrfs_io_bio *btrfs_failed_bio;
2333 struct btrfs_io_bio *btrfs_bio;
2335 bio = btrfs_io_bio_alloc(1);
2336 bio->bi_end_io = endio_func;
2337 bio->bi_iter.bi_sector = failrec->logical >> 9;
2338 bio_set_dev(bio, fs_info->fs_devices->latest_bdev);
2339 bio->bi_iter.bi_size = 0;
2340 bio->bi_private = data;
2342 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2343 if (btrfs_failed_bio->csum) {
2344 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2346 btrfs_bio = btrfs_io_bio(bio);
2347 btrfs_bio->csum = btrfs_bio->csum_inline;
2349 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2353 bio_add_page(bio, page, failrec->len, pg_offset);
2359 * this is a generic handler for readpage errors (default
2360 * readpage_io_failed_hook). if other copies exist, read those and write back
2361 * good data to the failed position. does not investigate in remapping the
2362 * failed extent elsewhere, hoping the device will be smart enough to do this as
2366 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2367 struct page *page, u64 start, u64 end,
2370 struct io_failure_record *failrec;
2371 struct inode *inode = page->mapping->host;
2372 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2373 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2376 blk_status_t status;
2379 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2381 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2385 if (!btrfs_check_repairable(inode, failed_bio, failrec,
2387 free_io_failure(failure_tree, tree, failrec);
2391 if (failed_bio->bi_vcnt > 1)
2392 read_mode |= REQ_FAILFAST_DEV;
2394 phy_offset >>= inode->i_sb->s_blocksize_bits;
2395 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2396 start - page_offset(page),
2397 (int)phy_offset, failed_bio->bi_end_io,
2399 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
2401 btrfs_debug(btrfs_sb(inode->i_sb),
2402 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2403 read_mode, failrec->this_mirror, failrec->in_validation);
2405 status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror,
2406 failrec->bio_flags, 0);
2408 free_io_failure(failure_tree, tree, failrec);
2410 ret = blk_status_to_errno(status);
2416 /* lots and lots of room for performance fixes in the end_bio funcs */
2418 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2420 int uptodate = (err == 0);
2421 struct extent_io_tree *tree;
2424 tree = &BTRFS_I(page->mapping->host)->io_tree;
2426 if (tree->ops && tree->ops->writepage_end_io_hook)
2427 tree->ops->writepage_end_io_hook(page, start, end, NULL,
2431 ClearPageUptodate(page);
2433 ret = err < 0 ? err : -EIO;
2434 mapping_set_error(page->mapping, ret);
2439 * after a writepage IO is done, we need to:
2440 * clear the uptodate bits on error
2441 * clear the writeback bits in the extent tree for this IO
2442 * end_page_writeback if the page has no more pending IO
2444 * Scheduling is not allowed, so the extent state tree is expected
2445 * to have one and only one object corresponding to this IO.
2447 static void end_bio_extent_writepage(struct bio *bio)
2449 int error = blk_status_to_errno(bio->bi_status);
2450 struct bio_vec *bvec;
2455 ASSERT(!bio_flagged(bio, BIO_CLONED));
2456 bio_for_each_segment_all(bvec, bio, i) {
2457 struct page *page = bvec->bv_page;
2458 struct inode *inode = page->mapping->host;
2459 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2461 /* We always issue full-page reads, but if some block
2462 * in a page fails to read, blk_update_request() will
2463 * advance bv_offset and adjust bv_len to compensate.
2464 * Print a warning for nonzero offsets, and an error
2465 * if they don't add up to a full page. */
2466 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2467 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2469 "partial page write in btrfs with offset %u and length %u",
2470 bvec->bv_offset, bvec->bv_len);
2473 "incomplete page write in btrfs with offset %u and length %u",
2474 bvec->bv_offset, bvec->bv_len);
2477 start = page_offset(page);
2478 end = start + bvec->bv_offset + bvec->bv_len - 1;
2480 end_extent_writepage(page, error, start, end);
2481 end_page_writeback(page);
2488 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2491 struct extent_state *cached = NULL;
2492 u64 end = start + len - 1;
2494 if (uptodate && tree->track_uptodate)
2495 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2496 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2500 * after a readpage IO is done, we need to:
2501 * clear the uptodate bits on error
2502 * set the uptodate bits if things worked
2503 * set the page up to date if all extents in the tree are uptodate
2504 * clear the lock bit in the extent tree
2505 * unlock the page if there are no other extents locked for it
2507 * Scheduling is not allowed, so the extent state tree is expected
2508 * to have one and only one object corresponding to this IO.
2510 static void end_bio_extent_readpage(struct bio *bio)
2512 struct bio_vec *bvec;
2513 int uptodate = !bio->bi_status;
2514 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2515 struct extent_io_tree *tree, *failure_tree;
2520 u64 extent_start = 0;
2526 ASSERT(!bio_flagged(bio, BIO_CLONED));
2527 bio_for_each_segment_all(bvec, bio, i) {
2528 struct page *page = bvec->bv_page;
2529 struct inode *inode = page->mapping->host;
2530 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2532 btrfs_debug(fs_info,
2533 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2534 (u64)bio->bi_iter.bi_sector, bio->bi_status,
2535 io_bio->mirror_num);
2536 tree = &BTRFS_I(inode)->io_tree;
2537 failure_tree = &BTRFS_I(inode)->io_failure_tree;
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. */
2544 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2545 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2547 "partial page read in btrfs with offset %u and length %u",
2548 bvec->bv_offset, bvec->bv_len);
2551 "incomplete page read in btrfs with offset %u and length %u",
2552 bvec->bv_offset, bvec->bv_len);
2555 start = page_offset(page);
2556 end = start + bvec->bv_offset + bvec->bv_len - 1;
2559 mirror = io_bio->mirror_num;
2560 if (likely(uptodate && tree->ops)) {
2561 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2567 clean_io_failure(BTRFS_I(inode)->root->fs_info,
2568 failure_tree, tree, start,
2570 btrfs_ino(BTRFS_I(inode)), 0);
2573 if (likely(uptodate))
2577 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2578 if (ret == -EAGAIN) {
2580 * Data inode's readpage_io_failed_hook() always
2583 * The generic bio_readpage_error handles errors
2584 * the following way: If possible, new read
2585 * requests are created and submitted and will
2586 * end up in end_bio_extent_readpage as well (if
2587 * we're lucky, not in the !uptodate case). In
2588 * that case it returns 0 and we just go on with
2589 * the next page in our bio. If it can't handle
2590 * the error it will return -EIO and we remain
2591 * responsible for that page.
2593 ret = bio_readpage_error(bio, offset, page,
2594 start, end, mirror);
2596 uptodate = !bio->bi_status;
2603 * metadata's readpage_io_failed_hook() always returns
2604 * -EIO and fixes nothing. -EIO is also returned if
2605 * data inode error could not be fixed.
2607 ASSERT(ret == -EIO);
2610 if (likely(uptodate)) {
2611 loff_t i_size = i_size_read(inode);
2612 pgoff_t end_index = i_size >> PAGE_SHIFT;
2615 /* Zero out the end if this page straddles i_size */
2616 off = i_size & (PAGE_SIZE-1);
2617 if (page->index == end_index && off)
2618 zero_user_segment(page, off, PAGE_SIZE);
2619 SetPageUptodate(page);
2621 ClearPageUptodate(page);
2627 if (unlikely(!uptodate)) {
2629 endio_readpage_release_extent(tree,
2635 endio_readpage_release_extent(tree, start,
2636 end - start + 1, 0);
2637 } else if (!extent_len) {
2638 extent_start = start;
2639 extent_len = end + 1 - start;
2640 } else if (extent_start + extent_len == start) {
2641 extent_len += end + 1 - start;
2643 endio_readpage_release_extent(tree, extent_start,
2644 extent_len, uptodate);
2645 extent_start = start;
2646 extent_len = end + 1 - start;
2651 endio_readpage_release_extent(tree, extent_start, extent_len,
2654 io_bio->end_io(io_bio, blk_status_to_errno(bio->bi_status));
2659 * Initialize the members up to but not including 'bio'. Use after allocating a
2660 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2661 * 'bio' because use of __GFP_ZERO is not supported.
2663 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
2665 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
2669 * The following helpers allocate a bio. As it's backed by a bioset, it'll
2670 * never fail. We're returning a bio right now but you can call btrfs_io_bio
2671 * for the appropriate container_of magic
2673 struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte)
2677 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, btrfs_bioset);
2678 bio_set_dev(bio, bdev);
2679 bio->bi_iter.bi_sector = first_byte >> 9;
2680 btrfs_io_bio_init(btrfs_io_bio(bio));
2684 struct bio *btrfs_bio_clone(struct bio *bio)
2686 struct btrfs_io_bio *btrfs_bio;
2689 /* Bio allocation backed by a bioset does not fail */
2690 new = bio_clone_fast(bio, GFP_NOFS, btrfs_bioset);
2691 btrfs_bio = btrfs_io_bio(new);
2692 btrfs_io_bio_init(btrfs_bio);
2693 btrfs_bio->iter = bio->bi_iter;
2697 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
2701 /* Bio allocation backed by a bioset does not fail */
2702 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, btrfs_bioset);
2703 btrfs_io_bio_init(btrfs_io_bio(bio));
2707 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
2710 struct btrfs_io_bio *btrfs_bio;
2712 /* this will never fail when it's backed by a bioset */
2713 bio = bio_clone_fast(orig, GFP_NOFS, btrfs_bioset);
2716 btrfs_bio = btrfs_io_bio(bio);
2717 btrfs_io_bio_init(btrfs_bio);
2719 bio_trim(bio, offset >> 9, size >> 9);
2720 btrfs_bio->iter = bio->bi_iter;
2724 static int __must_check submit_one_bio(struct bio *bio, int mirror_num,
2725 unsigned long bio_flags)
2727 blk_status_t ret = 0;
2728 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2729 struct page *page = bvec->bv_page;
2730 struct extent_io_tree *tree = bio->bi_private;
2733 start = page_offset(page) + bvec->bv_offset;
2735 bio->bi_private = NULL;
2739 ret = tree->ops->submit_bio_hook(tree->private_data, bio,
2740 mirror_num, bio_flags, start);
2742 btrfsic_submit_bio(bio);
2745 return blk_status_to_errno(ret);
2748 static int merge_bio(struct extent_io_tree *tree, struct page *page,
2749 unsigned long offset, size_t size, struct bio *bio,
2750 unsigned long bio_flags)
2754 ret = tree->ops->merge_bio_hook(page, offset, size, bio,
2761 * @opf: bio REQ_OP_* and REQ_* flags as one value
2763 static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree,
2764 struct writeback_control *wbc,
2765 struct page *page, sector_t sector,
2766 size_t size, unsigned long offset,
2767 struct block_device *bdev,
2768 struct bio **bio_ret,
2769 bio_end_io_t end_io_func,
2771 unsigned long prev_bio_flags,
2772 unsigned long bio_flags,
2773 bool force_bio_submit)
2778 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2779 size_t page_size = min_t(size_t, size, PAGE_SIZE);
2781 if (bio_ret && *bio_ret) {
2784 contig = bio->bi_iter.bi_sector == sector;
2786 contig = bio_end_sector(bio) == sector;
2788 if (prev_bio_flags != bio_flags || !contig ||
2790 merge_bio(tree, page, offset, page_size, bio, bio_flags) ||
2791 bio_add_page(bio, page, page_size, offset) < page_size) {
2792 ret = submit_one_bio(bio, mirror_num, prev_bio_flags);
2800 wbc_account_io(wbc, page, page_size);
2805 bio = btrfs_bio_alloc(bdev, (u64)sector << 9);
2806 bio_add_page(bio, page, page_size, offset);
2807 bio->bi_end_io = end_io_func;
2808 bio->bi_private = tree;
2809 bio->bi_write_hint = page->mapping->host->i_write_hint;
2812 wbc_init_bio(wbc, bio);
2813 wbc_account_io(wbc, page, page_size);
2819 ret = submit_one_bio(bio, mirror_num, bio_flags);
2824 static void attach_extent_buffer_page(struct extent_buffer *eb,
2827 if (!PagePrivate(page)) {
2828 SetPagePrivate(page);
2830 set_page_private(page, (unsigned long)eb);
2832 WARN_ON(page->private != (unsigned long)eb);
2836 void set_page_extent_mapped(struct page *page)
2838 if (!PagePrivate(page)) {
2839 SetPagePrivate(page);
2841 set_page_private(page, EXTENT_PAGE_PRIVATE);
2845 static struct extent_map *
2846 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2847 u64 start, u64 len, get_extent_t *get_extent,
2848 struct extent_map **em_cached)
2850 struct extent_map *em;
2852 if (em_cached && *em_cached) {
2854 if (extent_map_in_tree(em) && start >= em->start &&
2855 start < extent_map_end(em)) {
2856 refcount_inc(&em->refs);
2860 free_extent_map(em);
2864 em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0);
2865 if (em_cached && !IS_ERR_OR_NULL(em)) {
2867 refcount_inc(&em->refs);
2873 * basic readpage implementation. Locked extent state structs are inserted
2874 * into the tree that are removed when the IO is done (by the end_io
2876 * XXX JDM: This needs looking at to ensure proper page locking
2877 * return 0 on success, otherwise return error
2879 static int __do_readpage(struct extent_io_tree *tree,
2881 get_extent_t *get_extent,
2882 struct extent_map **em_cached,
2883 struct bio **bio, int mirror_num,
2884 unsigned long *bio_flags, unsigned int read_flags,
2887 struct inode *inode = page->mapping->host;
2888 u64 start = page_offset(page);
2889 u64 page_end = start + PAGE_SIZE - 1;
2893 u64 last_byte = i_size_read(inode);
2897 struct extent_map *em;
2898 struct block_device *bdev;
2901 size_t pg_offset = 0;
2903 size_t disk_io_size;
2904 size_t blocksize = inode->i_sb->s_blocksize;
2905 unsigned long this_bio_flag = 0;
2907 set_page_extent_mapped(page);
2910 if (!PageUptodate(page)) {
2911 if (cleancache_get_page(page) == 0) {
2912 BUG_ON(blocksize != PAGE_SIZE);
2913 unlock_extent(tree, start, end);
2918 if (page->index == last_byte >> PAGE_SHIFT) {
2920 size_t zero_offset = last_byte & (PAGE_SIZE - 1);
2923 iosize = PAGE_SIZE - zero_offset;
2924 userpage = kmap_atomic(page);
2925 memset(userpage + zero_offset, 0, iosize);
2926 flush_dcache_page(page);
2927 kunmap_atomic(userpage);
2930 while (cur <= end) {
2931 bool force_bio_submit = false;
2933 if (cur >= last_byte) {
2935 struct extent_state *cached = NULL;
2937 iosize = PAGE_SIZE - pg_offset;
2938 userpage = kmap_atomic(page);
2939 memset(userpage + pg_offset, 0, iosize);
2940 flush_dcache_page(page);
2941 kunmap_atomic(userpage);
2942 set_extent_uptodate(tree, cur, cur + iosize - 1,
2944 unlock_extent_cached(tree, cur,
2949 em = __get_extent_map(inode, page, pg_offset, cur,
2950 end - cur + 1, get_extent, em_cached);
2951 if (IS_ERR_OR_NULL(em)) {
2953 unlock_extent(tree, cur, end);
2956 extent_offset = cur - em->start;
2957 BUG_ON(extent_map_end(em) <= cur);
2960 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2961 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2962 extent_set_compress_type(&this_bio_flag,
2966 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2967 cur_end = min(extent_map_end(em) - 1, end);
2968 iosize = ALIGN(iosize, blocksize);
2969 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2970 disk_io_size = em->block_len;
2971 sector = em->block_start >> 9;
2973 sector = (em->block_start + extent_offset) >> 9;
2974 disk_io_size = iosize;
2977 block_start = em->block_start;
2978 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2979 block_start = EXTENT_MAP_HOLE;
2982 * If we have a file range that points to a compressed extent
2983 * and it's followed by a consecutive file range that points to
2984 * to the same compressed extent (possibly with a different
2985 * offset and/or length, so it either points to the whole extent
2986 * or only part of it), we must make sure we do not submit a
2987 * single bio to populate the pages for the 2 ranges because
2988 * this makes the compressed extent read zero out the pages
2989 * belonging to the 2nd range. Imagine the following scenario:
2992 * [0 - 8K] [8K - 24K]
2995 * points to extent X, points to extent X,
2996 * offset 4K, length of 8K offset 0, length 16K
2998 * [extent X, compressed length = 4K uncompressed length = 16K]
3000 * If the bio to read the compressed extent covers both ranges,
3001 * it will decompress extent X into the pages belonging to the
3002 * first range and then it will stop, zeroing out the remaining
3003 * pages that belong to the other range that points to extent X.
3004 * So here we make sure we submit 2 bios, one for the first
3005 * range and another one for the third range. Both will target
3006 * the same physical extent from disk, but we can't currently
3007 * make the compressed bio endio callback populate the pages
3008 * for both ranges because each compressed bio is tightly
3009 * coupled with a single extent map, and each range can have
3010 * an extent map with a different offset value relative to the
3011 * uncompressed data of our extent and different lengths. This
3012 * is a corner case so we prioritize correctness over
3013 * non-optimal behavior (submitting 2 bios for the same extent).
3015 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3016 prev_em_start && *prev_em_start != (u64)-1 &&
3017 *prev_em_start != em->orig_start)
3018 force_bio_submit = true;
3021 *prev_em_start = em->orig_start;
3023 free_extent_map(em);
3026 /* we've found a hole, just zero and go on */
3027 if (block_start == EXTENT_MAP_HOLE) {
3029 struct extent_state *cached = NULL;
3031 userpage = kmap_atomic(page);
3032 memset(userpage + pg_offset, 0, iosize);
3033 flush_dcache_page(page);
3034 kunmap_atomic(userpage);
3036 set_extent_uptodate(tree, cur, cur + iosize - 1,
3038 unlock_extent_cached(tree, cur,
3042 pg_offset += iosize;
3045 /* the get_extent function already copied into the page */
3046 if (test_range_bit(tree, cur, cur_end,
3047 EXTENT_UPTODATE, 1, NULL)) {
3048 check_page_uptodate(tree, page);
3049 unlock_extent(tree, cur, cur + iosize - 1);
3051 pg_offset += iosize;
3054 /* we have an inline extent but it didn't get marked up
3055 * to date. Error out
3057 if (block_start == EXTENT_MAP_INLINE) {
3059 unlock_extent(tree, cur, cur + iosize - 1);
3061 pg_offset += iosize;
3065 ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL,
3066 page, sector, disk_io_size, pg_offset,
3068 end_bio_extent_readpage, mirror_num,
3074 *bio_flags = this_bio_flag;
3077 unlock_extent(tree, cur, cur + iosize - 1);
3081 pg_offset += iosize;
3085 if (!PageError(page))
3086 SetPageUptodate(page);
3092 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
3093 struct page *pages[], int nr_pages,
3095 get_extent_t *get_extent,
3096 struct extent_map **em_cached,
3097 struct bio **bio, int mirror_num,
3098 unsigned long *bio_flags,
3101 struct inode *inode;
3102 struct btrfs_ordered_extent *ordered;
3105 inode = pages[0]->mapping->host;
3107 lock_extent(tree, start, end);
3108 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3112 unlock_extent(tree, start, end);
3113 btrfs_start_ordered_extent(inode, ordered, 1);
3114 btrfs_put_ordered_extent(ordered);
3117 for (index = 0; index < nr_pages; index++) {
3118 __do_readpage(tree, pages[index], get_extent, em_cached, bio,
3119 mirror_num, bio_flags, 0, prev_em_start);
3120 put_page(pages[index]);
3124 static void __extent_readpages(struct extent_io_tree *tree,
3125 struct page *pages[],
3126 int nr_pages, get_extent_t *get_extent,
3127 struct extent_map **em_cached,
3128 struct bio **bio, int mirror_num,
3129 unsigned long *bio_flags,
3136 int first_index = 0;
3138 for (index = 0; index < nr_pages; index++) {
3139 page_start = page_offset(pages[index]);
3142 end = start + PAGE_SIZE - 1;
3143 first_index = index;
3144 } else if (end + 1 == page_start) {
3147 __do_contiguous_readpages(tree, &pages[first_index],
3148 index - first_index, start,
3149 end, get_extent, em_cached,
3150 bio, mirror_num, bio_flags,
3153 end = start + PAGE_SIZE - 1;
3154 first_index = index;
3159 __do_contiguous_readpages(tree, &pages[first_index],
3160 index - first_index, start,
3161 end, get_extent, em_cached, bio,
3162 mirror_num, bio_flags,
3166 static int __extent_read_full_page(struct extent_io_tree *tree,
3168 get_extent_t *get_extent,
3169 struct bio **bio, int mirror_num,
3170 unsigned long *bio_flags,
3171 unsigned int read_flags)
3173 struct inode *inode = page->mapping->host;
3174 struct btrfs_ordered_extent *ordered;
3175 u64 start = page_offset(page);
3176 u64 end = start + PAGE_SIZE - 1;
3180 lock_extent(tree, start, end);
3181 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3185 unlock_extent(tree, start, end);
3186 btrfs_start_ordered_extent(inode, ordered, 1);
3187 btrfs_put_ordered_extent(ordered);
3190 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3191 bio_flags, read_flags, NULL);
3195 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3196 get_extent_t *get_extent, int mirror_num)
3198 struct bio *bio = NULL;
3199 unsigned long bio_flags = 0;
3202 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3205 ret = submit_one_bio(bio, mirror_num, bio_flags);
3209 static void update_nr_written(struct writeback_control *wbc,
3210 unsigned long nr_written)
3212 wbc->nr_to_write -= nr_written;
3216 * helper for __extent_writepage, doing all of the delayed allocation setup.
3218 * This returns 1 if our fill_delalloc function did all the work required
3219 * to write the page (copy into inline extent). In this case the IO has
3220 * been started and the page is already unlocked.
3222 * This returns 0 if all went well (page still locked)
3223 * This returns < 0 if there were errors (page still locked)
3225 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3226 struct page *page, struct writeback_control *wbc,
3227 struct extent_page_data *epd,
3229 unsigned long *nr_written)
3231 struct extent_io_tree *tree = epd->tree;
3232 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3234 u64 delalloc_to_write = 0;
3235 u64 delalloc_end = 0;
3237 int page_started = 0;
3239 if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc)
3242 while (delalloc_end < page_end) {
3243 nr_delalloc = find_lock_delalloc_range(inode, tree,
3247 BTRFS_MAX_EXTENT_SIZE);
3248 if (nr_delalloc == 0) {
3249 delalloc_start = delalloc_end + 1;
3252 ret = tree->ops->fill_delalloc(inode, page,
3257 /* File system has been set read-only */
3260 /* fill_delalloc should be return < 0 for error
3261 * but just in case, we use > 0 here meaning the
3262 * IO is started, so we don't want to return > 0
3263 * unless things are going well.
3265 ret = ret < 0 ? ret : -EIO;
3269 * delalloc_end is already one less than the total length, so
3270 * we don't subtract one from PAGE_SIZE
3272 delalloc_to_write += (delalloc_end - delalloc_start +
3273 PAGE_SIZE) >> PAGE_SHIFT;
3274 delalloc_start = delalloc_end + 1;
3276 if (wbc->nr_to_write < delalloc_to_write) {
3279 if (delalloc_to_write < thresh * 2)
3280 thresh = delalloc_to_write;
3281 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3285 /* did the fill delalloc function already unlock and start
3290 * we've unlocked the page, so we can't update
3291 * the mapping's writeback index, just update
3294 wbc->nr_to_write -= *nr_written;
3305 * helper for __extent_writepage. This calls the writepage start hooks,
3306 * and does the loop to map the page into extents and bios.
3308 * We return 1 if the IO is started and the page is unlocked,
3309 * 0 if all went well (page still locked)
3310 * < 0 if there were errors (page still locked)
3312 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3314 struct writeback_control *wbc,
3315 struct extent_page_data *epd,
3317 unsigned long nr_written,
3318 unsigned int write_flags, int *nr_ret)
3320 struct extent_io_tree *tree = epd->tree;
3321 u64 start = page_offset(page);
3322 u64 page_end = start + PAGE_SIZE - 1;
3329 struct extent_map *em;
3330 struct block_device *bdev;
3331 size_t pg_offset = 0;
3337 if (tree->ops && tree->ops->writepage_start_hook) {
3338 ret = tree->ops->writepage_start_hook(page, start,
3341 /* Fixup worker will requeue */
3343 wbc->pages_skipped++;
3345 redirty_page_for_writepage(wbc, page);
3347 update_nr_written(wbc, nr_written);
3354 * we don't want to touch the inode after unlocking the page,
3355 * so we update the mapping writeback index now
3357 update_nr_written(wbc, nr_written + 1);
3360 if (i_size <= start) {
3361 if (tree->ops && tree->ops->writepage_end_io_hook)
3362 tree->ops->writepage_end_io_hook(page, start,
3367 blocksize = inode->i_sb->s_blocksize;
3369 while (cur <= end) {
3372 if (cur >= i_size) {
3373 if (tree->ops && tree->ops->writepage_end_io_hook)
3374 tree->ops->writepage_end_io_hook(page, cur,
3378 em = epd->get_extent(BTRFS_I(inode), page, pg_offset, cur,
3380 if (IS_ERR_OR_NULL(em)) {
3382 ret = PTR_ERR_OR_ZERO(em);
3386 extent_offset = cur - em->start;
3387 em_end = extent_map_end(em);
3388 BUG_ON(em_end <= cur);
3390 iosize = min(em_end - cur, end - cur + 1);
3391 iosize = ALIGN(iosize, blocksize);
3392 sector = (em->block_start + extent_offset) >> 9;
3394 block_start = em->block_start;
3395 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3396 free_extent_map(em);
3400 * compressed and inline extents are written through other
3403 if (compressed || block_start == EXTENT_MAP_HOLE ||
3404 block_start == EXTENT_MAP_INLINE) {
3406 * end_io notification does not happen here for
3407 * compressed extents
3409 if (!compressed && tree->ops &&
3410 tree->ops->writepage_end_io_hook)
3411 tree->ops->writepage_end_io_hook(page, cur,
3414 else if (compressed) {
3415 /* we don't want to end_page_writeback on
3416 * a compressed extent. this happens
3423 pg_offset += iosize;
3427 set_range_writeback(tree, cur, cur + iosize - 1);
3428 if (!PageWriteback(page)) {
3429 btrfs_err(BTRFS_I(inode)->root->fs_info,
3430 "page %lu not writeback, cur %llu end %llu",
3431 page->index, cur, end);
3434 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3435 page, sector, iosize, pg_offset,
3437 end_bio_extent_writepage,
3441 if (PageWriteback(page))
3442 end_page_writeback(page);
3446 pg_offset += iosize;
3455 * the writepage semantics are similar to regular writepage. extent
3456 * records are inserted to lock ranges in the tree, and as dirty areas
3457 * are found, they are marked writeback. Then the lock bits are removed
3458 * and the end_io handler clears the writeback ranges
3460 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3463 struct inode *inode = page->mapping->host;
3464 struct extent_page_data *epd = data;
3465 u64 start = page_offset(page);
3466 u64 page_end = start + PAGE_SIZE - 1;
3469 size_t pg_offset = 0;
3470 loff_t i_size = i_size_read(inode);
3471 unsigned long end_index = i_size >> PAGE_SHIFT;
3472 unsigned int write_flags = 0;
3473 unsigned long nr_written = 0;
3475 write_flags = wbc_to_write_flags(wbc);
3477 trace___extent_writepage(page, inode, wbc);
3479 WARN_ON(!PageLocked(page));
3481 ClearPageError(page);
3483 pg_offset = i_size & (PAGE_SIZE - 1);
3484 if (page->index > end_index ||
3485 (page->index == end_index && !pg_offset)) {
3486 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3491 if (page->index == end_index) {
3494 userpage = kmap_atomic(page);
3495 memset(userpage + pg_offset, 0,
3496 PAGE_SIZE - pg_offset);
3497 kunmap_atomic(userpage);
3498 flush_dcache_page(page);
3503 set_page_extent_mapped(page);
3505 ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written);
3511 ret = __extent_writepage_io(inode, page, wbc, epd,
3512 i_size, nr_written, write_flags, &nr);
3518 /* make sure the mapping tag for page dirty gets cleared */
3519 set_page_writeback(page);
3520 end_page_writeback(page);
3522 if (PageError(page)) {
3523 ret = ret < 0 ? ret : -EIO;
3524 end_extent_writepage(page, ret, start, page_end);
3533 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3535 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3536 TASK_UNINTERRUPTIBLE);
3539 static noinline_for_stack int
3540 lock_extent_buffer_for_io(struct extent_buffer *eb,
3541 struct btrfs_fs_info *fs_info,
3542 struct extent_page_data *epd)
3544 unsigned long i, num_pages;
3548 if (!btrfs_try_tree_write_lock(eb)) {
3550 flush_write_bio(epd);
3551 btrfs_tree_lock(eb);
3554 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3555 btrfs_tree_unlock(eb);
3559 flush_write_bio(epd);
3563 wait_on_extent_buffer_writeback(eb);
3564 btrfs_tree_lock(eb);
3565 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3567 btrfs_tree_unlock(eb);
3572 * We need to do this to prevent races in people who check if the eb is
3573 * under IO since we can end up having no IO bits set for a short period
3576 spin_lock(&eb->refs_lock);
3577 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3578 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3579 spin_unlock(&eb->refs_lock);
3580 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3581 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3583 fs_info->dirty_metadata_batch);
3586 spin_unlock(&eb->refs_lock);
3589 btrfs_tree_unlock(eb);
3594 num_pages = num_extent_pages(eb->start, eb->len);
3595 for (i = 0; i < num_pages; i++) {
3596 struct page *p = eb->pages[i];
3598 if (!trylock_page(p)) {
3600 flush_write_bio(epd);
3610 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3612 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3613 smp_mb__after_atomic();
3614 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3617 static void set_btree_ioerr(struct page *page)
3619 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3622 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3626 * If writeback for a btree extent that doesn't belong to a log tree
3627 * failed, increment the counter transaction->eb_write_errors.
3628 * We do this because while the transaction is running and before it's
3629 * committing (when we call filemap_fdata[write|wait]_range against
3630 * the btree inode), we might have
3631 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3632 * returns an error or an error happens during writeback, when we're
3633 * committing the transaction we wouldn't know about it, since the pages
3634 * can be no longer dirty nor marked anymore for writeback (if a
3635 * subsequent modification to the extent buffer didn't happen before the
3636 * transaction commit), which makes filemap_fdata[write|wait]_range not
3637 * able to find the pages tagged with SetPageError at transaction
3638 * commit time. So if this happens we must abort the transaction,
3639 * otherwise we commit a super block with btree roots that point to
3640 * btree nodes/leafs whose content on disk is invalid - either garbage
3641 * or the content of some node/leaf from a past generation that got
3642 * cowed or deleted and is no longer valid.
3644 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3645 * not be enough - we need to distinguish between log tree extents vs
3646 * non-log tree extents, and the next filemap_fdatawait_range() call
3647 * will catch and clear such errors in the mapping - and that call might
3648 * be from a log sync and not from a transaction commit. Also, checking
3649 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3650 * not done and would not be reliable - the eb might have been released
3651 * from memory and reading it back again means that flag would not be
3652 * set (since it's a runtime flag, not persisted on disk).
3654 * Using the flags below in the btree inode also makes us achieve the
3655 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3656 * writeback for all dirty pages and before filemap_fdatawait_range()
3657 * is called, the writeback for all dirty pages had already finished
3658 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3659 * filemap_fdatawait_range() would return success, as it could not know
3660 * that writeback errors happened (the pages were no longer tagged for
3663 switch (eb->log_index) {
3665 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3668 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3671 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3674 BUG(); /* unexpected, logic error */
3678 static void end_bio_extent_buffer_writepage(struct bio *bio)
3680 struct bio_vec *bvec;
3681 struct extent_buffer *eb;
3684 ASSERT(!bio_flagged(bio, BIO_CLONED));
3685 bio_for_each_segment_all(bvec, bio, i) {
3686 struct page *page = bvec->bv_page;
3688 eb = (struct extent_buffer *)page->private;
3690 done = atomic_dec_and_test(&eb->io_pages);
3692 if (bio->bi_status ||
3693 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3694 ClearPageUptodate(page);
3695 set_btree_ioerr(page);
3698 end_page_writeback(page);
3703 end_extent_buffer_writeback(eb);
3709 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3710 struct btrfs_fs_info *fs_info,
3711 struct writeback_control *wbc,
3712 struct extent_page_data *epd)
3714 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3715 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3716 u64 offset = eb->start;
3718 unsigned long i, num_pages;
3719 unsigned long bio_flags = 0;
3720 unsigned long start, end;
3721 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
3724 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3725 num_pages = num_extent_pages(eb->start, eb->len);
3726 atomic_set(&eb->io_pages, num_pages);
3727 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3728 bio_flags = EXTENT_BIO_TREE_LOG;
3730 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3731 nritems = btrfs_header_nritems(eb);
3732 if (btrfs_header_level(eb) > 0) {
3733 end = btrfs_node_key_ptr_offset(nritems);
3735 memzero_extent_buffer(eb, end, eb->len - end);
3739 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3741 start = btrfs_item_nr_offset(nritems);
3742 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, eb);
3743 memzero_extent_buffer(eb, start, end - start);
3746 for (i = 0; i < num_pages; i++) {
3747 struct page *p = eb->pages[i];
3749 clear_page_dirty_for_io(p);
3750 set_page_writeback(p);
3751 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3752 p, offset >> 9, PAGE_SIZE, 0, bdev,
3754 end_bio_extent_buffer_writepage,
3755 0, epd->bio_flags, bio_flags, false);
3756 epd->bio_flags = bio_flags;
3759 if (PageWriteback(p))
3760 end_page_writeback(p);
3761 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3762 end_extent_buffer_writeback(eb);
3766 offset += PAGE_SIZE;
3767 update_nr_written(wbc, 1);
3771 if (unlikely(ret)) {
3772 for (; i < num_pages; i++) {
3773 struct page *p = eb->pages[i];
3774 clear_page_dirty_for_io(p);
3782 int btree_write_cache_pages(struct address_space *mapping,
3783 struct writeback_control *wbc)
3785 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3786 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3787 struct extent_buffer *eb, *prev_eb = NULL;
3788 struct extent_page_data epd = {
3792 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3797 int nr_to_write_done = 0;
3798 struct pagevec pvec;
3801 pgoff_t end; /* Inclusive */
3805 pagevec_init(&pvec, 0);
3806 if (wbc->range_cyclic) {
3807 index = mapping->writeback_index; /* Start from prev offset */
3810 index = wbc->range_start >> PAGE_SHIFT;
3811 end = wbc->range_end >> PAGE_SHIFT;
3814 if (wbc->sync_mode == WB_SYNC_ALL)
3815 tag = PAGECACHE_TAG_TOWRITE;
3817 tag = PAGECACHE_TAG_DIRTY;
3819 if (wbc->sync_mode == WB_SYNC_ALL)
3820 tag_pages_for_writeback(mapping, index, end);
3821 while (!done && !nr_to_write_done && (index <= end) &&
3822 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3823 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3827 for (i = 0; i < nr_pages; i++) {
3828 struct page *page = pvec.pages[i];
3830 if (!PagePrivate(page))
3833 if (!wbc->range_cyclic && page->index > end) {
3838 spin_lock(&mapping->private_lock);
3839 if (!PagePrivate(page)) {
3840 spin_unlock(&mapping->private_lock);
3844 eb = (struct extent_buffer *)page->private;
3847 * Shouldn't happen and normally this would be a BUG_ON
3848 * but no sense in crashing the users box for something
3849 * we can survive anyway.
3852 spin_unlock(&mapping->private_lock);
3856 if (eb == prev_eb) {
3857 spin_unlock(&mapping->private_lock);
3861 ret = atomic_inc_not_zero(&eb->refs);
3862 spin_unlock(&mapping->private_lock);
3867 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3869 free_extent_buffer(eb);
3873 ret = write_one_eb(eb, fs_info, wbc, &epd);
3876 free_extent_buffer(eb);
3879 free_extent_buffer(eb);
3882 * the filesystem may choose to bump up nr_to_write.
3883 * We have to make sure to honor the new nr_to_write
3886 nr_to_write_done = wbc->nr_to_write <= 0;
3888 pagevec_release(&pvec);
3891 if (!scanned && !done) {
3893 * We hit the last page and there is more work to be done: wrap
3894 * back to the start of the file
3900 flush_write_bio(&epd);
3905 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3906 * @mapping: address space structure to write
3907 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3908 * @writepage: function called for each page
3909 * @data: data passed to writepage function
3911 * If a page is already under I/O, write_cache_pages() skips it, even
3912 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3913 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3914 * and msync() need to guarantee that all the data which was dirty at the time
3915 * the call was made get new I/O started against them. If wbc->sync_mode is
3916 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3917 * existing IO to complete.
3919 static int extent_write_cache_pages(struct address_space *mapping,
3920 struct writeback_control *wbc,
3921 writepage_t writepage, void *data,
3922 void (*flush_fn)(void *))
3924 struct inode *inode = mapping->host;
3927 int nr_to_write_done = 0;
3928 struct pagevec pvec;
3931 pgoff_t end; /* Inclusive */
3933 int range_whole = 0;
3938 * We have to hold onto the inode so that ordered extents can do their
3939 * work when the IO finishes. The alternative to this is failing to add
3940 * an ordered extent if the igrab() fails there and that is a huge pain
3941 * to deal with, so instead just hold onto the inode throughout the
3942 * writepages operation. If it fails here we are freeing up the inode
3943 * anyway and we'd rather not waste our time writing out stuff that is
3944 * going to be truncated anyway.
3949 pagevec_init(&pvec, 0);
3950 if (wbc->range_cyclic) {
3951 index = mapping->writeback_index; /* Start from prev offset */
3954 index = wbc->range_start >> PAGE_SHIFT;
3955 end = wbc->range_end >> PAGE_SHIFT;
3956 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3960 if (wbc->sync_mode == WB_SYNC_ALL)
3961 tag = PAGECACHE_TAG_TOWRITE;
3963 tag = PAGECACHE_TAG_DIRTY;
3965 if (wbc->sync_mode == WB_SYNC_ALL)
3966 tag_pages_for_writeback(mapping, index, end);
3968 while (!done && !nr_to_write_done && (index <= end) &&
3969 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3970 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3974 for (i = 0; i < nr_pages; i++) {
3975 struct page *page = pvec.pages[i];
3977 done_index = page->index;
3979 * At this point we hold neither mapping->tree_lock nor
3980 * lock on the page itself: the page may be truncated or
3981 * invalidated (changing page->mapping to NULL), or even
3982 * swizzled back from swapper_space to tmpfs file
3985 if (!trylock_page(page)) {
3990 if (unlikely(page->mapping != mapping)) {
3995 if (!wbc->range_cyclic && page->index > end) {
4001 if (wbc->sync_mode != WB_SYNC_NONE) {
4002 if (PageWriteback(page))
4004 wait_on_page_writeback(page);
4007 if (PageWriteback(page) ||
4008 !clear_page_dirty_for_io(page)) {
4013 ret = (*writepage)(page, wbc, data);
4015 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
4021 * done_index is set past this page,
4022 * so media errors will not choke
4023 * background writeout for the entire
4024 * file. This has consequences for
4025 * range_cyclic semantics (ie. it may
4026 * not be suitable for data integrity
4029 done_index = page->index + 1;
4035 * the filesystem may choose to bump up nr_to_write.
4036 * We have to make sure to honor the new nr_to_write
4039 nr_to_write_done = wbc->nr_to_write <= 0;
4041 pagevec_release(&pvec);
4044 if (!scanned && !done) {
4046 * We hit the last page and there is more work to be done: wrap
4047 * back to the start of the file
4054 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4055 mapping->writeback_index = done_index;
4057 btrfs_add_delayed_iput(inode);
4061 static void flush_epd_write_bio(struct extent_page_data *epd)
4066 ret = submit_one_bio(epd->bio, 0, epd->bio_flags);
4067 BUG_ON(ret < 0); /* -ENOMEM */
4072 static noinline void flush_write_bio(void *data)
4074 struct extent_page_data *epd = data;
4075 flush_epd_write_bio(epd);
4078 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
4079 get_extent_t *get_extent,
4080 struct writeback_control *wbc)
4083 struct extent_page_data epd = {
4086 .get_extent = get_extent,
4088 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4092 ret = __extent_writepage(page, wbc, &epd);
4094 flush_epd_write_bio(&epd);
4098 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
4099 u64 start, u64 end, get_extent_t *get_extent,
4103 struct address_space *mapping = inode->i_mapping;
4105 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4108 struct extent_page_data epd = {
4111 .get_extent = get_extent,
4113 .sync_io = mode == WB_SYNC_ALL,
4116 struct writeback_control wbc_writepages = {
4118 .nr_to_write = nr_pages * 2,
4119 .range_start = start,
4120 .range_end = end + 1,
4123 while (start <= end) {
4124 page = find_get_page(mapping, start >> PAGE_SHIFT);
4125 if (clear_page_dirty_for_io(page))
4126 ret = __extent_writepage(page, &wbc_writepages, &epd);
4128 if (tree->ops && tree->ops->writepage_end_io_hook)
4129 tree->ops->writepage_end_io_hook(page, start,
4130 start + PAGE_SIZE - 1,
4138 flush_epd_write_bio(&epd);
4142 int extent_writepages(struct extent_io_tree *tree,
4143 struct address_space *mapping,
4144 get_extent_t *get_extent,
4145 struct writeback_control *wbc)
4148 struct extent_page_data epd = {
4151 .get_extent = get_extent,
4153 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4157 ret = extent_write_cache_pages(mapping, wbc, __extent_writepage, &epd,
4159 flush_epd_write_bio(&epd);
4163 int extent_readpages(struct extent_io_tree *tree,
4164 struct address_space *mapping,
4165 struct list_head *pages, unsigned nr_pages,
4166 get_extent_t get_extent)
4168 struct bio *bio = NULL;
4170 unsigned long bio_flags = 0;
4171 struct page *pagepool[16];
4173 struct extent_map *em_cached = NULL;
4175 u64 prev_em_start = (u64)-1;
4177 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
4178 page = list_entry(pages->prev, struct page, lru);
4180 prefetchw(&page->flags);
4181 list_del(&page->lru);
4182 if (add_to_page_cache_lru(page, mapping,
4184 readahead_gfp_mask(mapping))) {
4189 pagepool[nr++] = page;
4190 if (nr < ARRAY_SIZE(pagepool))
4192 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4193 &bio, 0, &bio_flags, &prev_em_start);
4197 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4198 &bio, 0, &bio_flags, &prev_em_start);
4201 free_extent_map(em_cached);
4203 BUG_ON(!list_empty(pages));
4205 return submit_one_bio(bio, 0, bio_flags);
4210 * basic invalidatepage code, this waits on any locked or writeback
4211 * ranges corresponding to the page, and then deletes any extent state
4212 * records from the tree
4214 int extent_invalidatepage(struct extent_io_tree *tree,
4215 struct page *page, unsigned long offset)
4217 struct extent_state *cached_state = NULL;
4218 u64 start = page_offset(page);
4219 u64 end = start + PAGE_SIZE - 1;
4220 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4222 start += ALIGN(offset, blocksize);
4226 lock_extent_bits(tree, start, end, &cached_state);
4227 wait_on_page_writeback(page);
4228 clear_extent_bit(tree, start, end,
4229 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4230 EXTENT_DO_ACCOUNTING,
4231 1, 1, &cached_state, GFP_NOFS);
4236 * a helper for releasepage, this tests for areas of the page that
4237 * are locked or under IO and drops the related state bits if it is safe
4240 static int try_release_extent_state(struct extent_map_tree *map,
4241 struct extent_io_tree *tree,
4242 struct page *page, gfp_t mask)
4244 u64 start = page_offset(page);
4245 u64 end = start + PAGE_SIZE - 1;
4248 if (test_range_bit(tree, start, end,
4249 EXTENT_IOBITS, 0, NULL))
4253 * at this point we can safely clear everything except the
4254 * locked bit and the nodatasum bit
4256 ret = clear_extent_bit(tree, start, end,
4257 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4260 /* if clear_extent_bit failed for enomem reasons,
4261 * we can't allow the release to continue.
4272 * a helper for releasepage. As long as there are no locked extents
4273 * in the range corresponding to the page, both state records and extent
4274 * map records are removed
4276 int try_release_extent_mapping(struct extent_map_tree *map,
4277 struct extent_io_tree *tree, struct page *page,
4280 struct extent_map *em;
4281 u64 start = page_offset(page);
4282 u64 end = start + PAGE_SIZE - 1;
4284 if (gfpflags_allow_blocking(mask) &&
4285 page->mapping->host->i_size > SZ_16M) {
4287 while (start <= end) {
4288 len = end - start + 1;
4289 write_lock(&map->lock);
4290 em = lookup_extent_mapping(map, start, len);
4292 write_unlock(&map->lock);
4295 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4296 em->start != start) {
4297 write_unlock(&map->lock);
4298 free_extent_map(em);
4301 if (!test_range_bit(tree, em->start,
4302 extent_map_end(em) - 1,
4303 EXTENT_LOCKED | EXTENT_WRITEBACK,
4305 remove_extent_mapping(map, em);
4306 /* once for the rb tree */
4307 free_extent_map(em);
4309 start = extent_map_end(em);
4310 write_unlock(&map->lock);
4313 free_extent_map(em);
4316 return try_release_extent_state(map, tree, page, mask);
4320 * helper function for fiemap, which doesn't want to see any holes.
4321 * This maps until we find something past 'last'
4323 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4326 get_extent_t *get_extent)
4328 u64 sectorsize = btrfs_inode_sectorsize(inode);
4329 struct extent_map *em;
4336 len = last - offset;
4339 len = ALIGN(len, sectorsize);
4340 em = get_extent(BTRFS_I(inode), NULL, 0, offset, len, 0);
4341 if (IS_ERR_OR_NULL(em))
4344 /* if this isn't a hole return it */
4345 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4346 em->block_start != EXTENT_MAP_HOLE) {
4350 /* this is a hole, advance to the next extent */
4351 offset = extent_map_end(em);
4352 free_extent_map(em);
4360 * To cache previous fiemap extent
4362 * Will be used for merging fiemap extent
4364 struct fiemap_cache {
4373 * Helper to submit fiemap extent.
4375 * Will try to merge current fiemap extent specified by @offset, @phys,
4376 * @len and @flags with cached one.
4377 * And only when we fails to merge, cached one will be submitted as
4380 * Return value is the same as fiemap_fill_next_extent().
4382 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4383 struct fiemap_cache *cache,
4384 u64 offset, u64 phys, u64 len, u32 flags)
4392 * Sanity check, extent_fiemap() should have ensured that new
4393 * fiemap extent won't overlap with cahced one.
4396 * NOTE: Physical address can overlap, due to compression
4398 if (cache->offset + cache->len > offset) {
4404 * Only merges fiemap extents if
4405 * 1) Their logical addresses are continuous
4407 * 2) Their physical addresses are continuous
4408 * So truly compressed (physical size smaller than logical size)
4409 * extents won't get merged with each other
4411 * 3) Share same flags except FIEMAP_EXTENT_LAST
4412 * So regular extent won't get merged with prealloc extent
4414 if (cache->offset + cache->len == offset &&
4415 cache->phys + cache->len == phys &&
4416 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4417 (flags & ~FIEMAP_EXTENT_LAST)) {
4419 cache->flags |= flags;
4420 goto try_submit_last;
4423 /* Not mergeable, need to submit cached one */
4424 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4425 cache->len, cache->flags);
4426 cache->cached = false;
4430 cache->cached = true;
4431 cache->offset = offset;
4434 cache->flags = flags;
4436 if (cache->flags & FIEMAP_EXTENT_LAST) {
4437 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4438 cache->phys, cache->len, cache->flags);
4439 cache->cached = false;
4445 * Emit last fiemap cache
4447 * The last fiemap cache may still be cached in the following case:
4449 * |<- Fiemap range ->|
4450 * |<------------ First extent ----------->|
4452 * In this case, the first extent range will be cached but not emitted.
4453 * So we must emit it before ending extent_fiemap().
4455 static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info,
4456 struct fiemap_extent_info *fieinfo,
4457 struct fiemap_cache *cache)
4464 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4465 cache->len, cache->flags);
4466 cache->cached = false;
4472 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4473 __u64 start, __u64 len, get_extent_t *get_extent)
4477 u64 max = start + len;
4481 u64 last_for_get_extent = 0;
4483 u64 isize = i_size_read(inode);
4484 struct btrfs_key found_key;
4485 struct extent_map *em = NULL;
4486 struct extent_state *cached_state = NULL;
4487 struct btrfs_path *path;
4488 struct btrfs_root *root = BTRFS_I(inode)->root;
4489 struct fiemap_cache cache = { 0 };
4498 path = btrfs_alloc_path();
4501 path->leave_spinning = 1;
4503 start = round_down(start, btrfs_inode_sectorsize(inode));
4504 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
4507 * lookup the last file extent. We're not using i_size here
4508 * because there might be preallocation past i_size
4510 ret = btrfs_lookup_file_extent(NULL, root, path,
4511 btrfs_ino(BTRFS_I(inode)), -1, 0);
4513 btrfs_free_path(path);
4522 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4523 found_type = found_key.type;
4525 /* No extents, but there might be delalloc bits */
4526 if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) ||
4527 found_type != BTRFS_EXTENT_DATA_KEY) {
4528 /* have to trust i_size as the end */
4530 last_for_get_extent = isize;
4533 * remember the start of the last extent. There are a
4534 * bunch of different factors that go into the length of the
4535 * extent, so its much less complex to remember where it started
4537 last = found_key.offset;
4538 last_for_get_extent = last + 1;
4540 btrfs_release_path(path);
4543 * we might have some extents allocated but more delalloc past those
4544 * extents. so, we trust isize unless the start of the last extent is
4549 last_for_get_extent = isize;
4552 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4555 em = get_extent_skip_holes(inode, start, last_for_get_extent,
4565 u64 offset_in_extent = 0;
4567 /* break if the extent we found is outside the range */
4568 if (em->start >= max || extent_map_end(em) < off)
4572 * get_extent may return an extent that starts before our
4573 * requested range. We have to make sure the ranges
4574 * we return to fiemap always move forward and don't
4575 * overlap, so adjust the offsets here
4577 em_start = max(em->start, off);
4580 * record the offset from the start of the extent
4581 * for adjusting the disk offset below. Only do this if the
4582 * extent isn't compressed since our in ram offset may be past
4583 * what we have actually allocated on disk.
4585 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4586 offset_in_extent = em_start - em->start;
4587 em_end = extent_map_end(em);
4588 em_len = em_end - em_start;
4593 * bump off for our next call to get_extent
4595 off = extent_map_end(em);
4599 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4601 flags |= FIEMAP_EXTENT_LAST;
4602 } else if (em->block_start == EXTENT_MAP_INLINE) {
4603 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4604 FIEMAP_EXTENT_NOT_ALIGNED);
4605 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4606 flags |= (FIEMAP_EXTENT_DELALLOC |
4607 FIEMAP_EXTENT_UNKNOWN);
4608 } else if (fieinfo->fi_extents_max) {
4609 u64 bytenr = em->block_start -
4610 (em->start - em->orig_start);
4612 disko = em->block_start + offset_in_extent;
4615 * As btrfs supports shared space, this information
4616 * can be exported to userspace tools via
4617 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4618 * then we're just getting a count and we can skip the
4621 ret = btrfs_check_shared(root,
4622 btrfs_ino(BTRFS_I(inode)),
4627 flags |= FIEMAP_EXTENT_SHARED;
4630 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4631 flags |= FIEMAP_EXTENT_ENCODED;
4632 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4633 flags |= FIEMAP_EXTENT_UNWRITTEN;
4635 free_extent_map(em);
4637 if ((em_start >= last) || em_len == (u64)-1 ||
4638 (last == (u64)-1 && isize <= em_end)) {
4639 flags |= FIEMAP_EXTENT_LAST;
4643 /* now scan forward to see if this is really the last extent. */
4644 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4651 flags |= FIEMAP_EXTENT_LAST;
4654 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4664 ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache);
4665 free_extent_map(em);
4667 btrfs_free_path(path);
4668 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4669 &cached_state, GFP_NOFS);
4673 static void __free_extent_buffer(struct extent_buffer *eb)
4675 btrfs_leak_debug_del(&eb->leak_list);
4676 kmem_cache_free(extent_buffer_cache, eb);
4679 int extent_buffer_under_io(struct extent_buffer *eb)
4681 return (atomic_read(&eb->io_pages) ||
4682 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4683 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4687 * Helper for releasing extent buffer page.
4689 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb)
4691 unsigned long index;
4693 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4695 BUG_ON(extent_buffer_under_io(eb));
4697 index = num_extent_pages(eb->start, eb->len);
4703 page = eb->pages[index];
4707 spin_lock(&page->mapping->private_lock);
4709 * We do this since we'll remove the pages after we've
4710 * removed the eb from the radix tree, so we could race
4711 * and have this page now attached to the new eb. So
4712 * only clear page_private if it's still connected to
4715 if (PagePrivate(page) &&
4716 page->private == (unsigned long)eb) {
4717 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4718 BUG_ON(PageDirty(page));
4719 BUG_ON(PageWriteback(page));
4721 * We need to make sure we haven't be attached
4724 ClearPagePrivate(page);
4725 set_page_private(page, 0);
4726 /* One for the page private */
4731 spin_unlock(&page->mapping->private_lock);
4733 /* One for when we allocated the page */
4735 } while (index != 0);
4739 * Helper for releasing the extent buffer.
4741 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4743 btrfs_release_extent_buffer_page(eb);
4744 __free_extent_buffer(eb);
4747 static struct extent_buffer *
4748 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4751 struct extent_buffer *eb = NULL;
4753 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4756 eb->fs_info = fs_info;
4758 rwlock_init(&eb->lock);
4759 atomic_set(&eb->write_locks, 0);
4760 atomic_set(&eb->read_locks, 0);
4761 atomic_set(&eb->blocking_readers, 0);
4762 atomic_set(&eb->blocking_writers, 0);
4763 atomic_set(&eb->spinning_readers, 0);
4764 atomic_set(&eb->spinning_writers, 0);
4765 eb->lock_nested = 0;
4766 init_waitqueue_head(&eb->write_lock_wq);
4767 init_waitqueue_head(&eb->read_lock_wq);
4769 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4771 spin_lock_init(&eb->refs_lock);
4772 atomic_set(&eb->refs, 1);
4773 atomic_set(&eb->io_pages, 0);
4776 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4778 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4779 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4780 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4785 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4789 struct extent_buffer *new;
4790 unsigned long num_pages = num_extent_pages(src->start, src->len);
4792 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4796 for (i = 0; i < num_pages; i++) {
4797 p = alloc_page(GFP_NOFS);
4799 btrfs_release_extent_buffer(new);
4802 attach_extent_buffer_page(new, p);
4803 WARN_ON(PageDirty(p));
4806 copy_page(page_address(p), page_address(src->pages[i]));
4809 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4810 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4815 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4816 u64 start, unsigned long len)
4818 struct extent_buffer *eb;
4819 unsigned long num_pages;
4822 num_pages = num_extent_pages(start, len);
4824 eb = __alloc_extent_buffer(fs_info, start, len);
4828 for (i = 0; i < num_pages; i++) {
4829 eb->pages[i] = alloc_page(GFP_NOFS);
4833 set_extent_buffer_uptodate(eb);
4834 btrfs_set_header_nritems(eb, 0);
4835 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4840 __free_page(eb->pages[i - 1]);
4841 __free_extent_buffer(eb);
4845 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4848 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
4851 static void check_buffer_tree_ref(struct extent_buffer *eb)
4854 /* the ref bit is tricky. We have to make sure it is set
4855 * if we have the buffer dirty. Otherwise the
4856 * code to free a buffer can end up dropping a dirty
4859 * Once the ref bit is set, it won't go away while the
4860 * buffer is dirty or in writeback, and it also won't
4861 * go away while we have the reference count on the
4864 * We can't just set the ref bit without bumping the
4865 * ref on the eb because free_extent_buffer might
4866 * see the ref bit and try to clear it. If this happens
4867 * free_extent_buffer might end up dropping our original
4868 * ref by mistake and freeing the page before we are able
4869 * to add one more ref.
4871 * So bump the ref count first, then set the bit. If someone
4872 * beat us to it, drop the ref we added.
4874 refs = atomic_read(&eb->refs);
4875 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4878 spin_lock(&eb->refs_lock);
4879 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4880 atomic_inc(&eb->refs);
4881 spin_unlock(&eb->refs_lock);
4884 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4885 struct page *accessed)
4887 unsigned long num_pages, i;
4889 check_buffer_tree_ref(eb);
4891 num_pages = num_extent_pages(eb->start, eb->len);
4892 for (i = 0; i < num_pages; i++) {
4893 struct page *p = eb->pages[i];
4896 mark_page_accessed(p);
4900 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4903 struct extent_buffer *eb;
4906 eb = radix_tree_lookup(&fs_info->buffer_radix,
4907 start >> PAGE_SHIFT);
4908 if (eb && atomic_inc_not_zero(&eb->refs)) {
4911 * Lock our eb's refs_lock to avoid races with
4912 * free_extent_buffer. When we get our eb it might be flagged
4913 * with EXTENT_BUFFER_STALE and another task running
4914 * free_extent_buffer might have seen that flag set,
4915 * eb->refs == 2, that the buffer isn't under IO (dirty and
4916 * writeback flags not set) and it's still in the tree (flag
4917 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4918 * of decrementing the extent buffer's reference count twice.
4919 * So here we could race and increment the eb's reference count,
4920 * clear its stale flag, mark it as dirty and drop our reference
4921 * before the other task finishes executing free_extent_buffer,
4922 * which would later result in an attempt to free an extent
4923 * buffer that is dirty.
4925 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
4926 spin_lock(&eb->refs_lock);
4927 spin_unlock(&eb->refs_lock);
4929 mark_extent_buffer_accessed(eb, NULL);
4937 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4938 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4941 struct extent_buffer *eb, *exists = NULL;
4944 eb = find_extent_buffer(fs_info, start);
4947 eb = alloc_dummy_extent_buffer(fs_info, start);
4950 eb->fs_info = fs_info;
4952 ret = radix_tree_preload(GFP_NOFS);
4955 spin_lock(&fs_info->buffer_lock);
4956 ret = radix_tree_insert(&fs_info->buffer_radix,
4957 start >> PAGE_SHIFT, eb);
4958 spin_unlock(&fs_info->buffer_lock);
4959 radix_tree_preload_end();
4960 if (ret == -EEXIST) {
4961 exists = find_extent_buffer(fs_info, start);
4967 check_buffer_tree_ref(eb);
4968 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4971 * We will free dummy extent buffer's if they come into
4972 * free_extent_buffer with a ref count of 2, but if we are using this we
4973 * want the buffers to stay in memory until we're done with them, so
4974 * bump the ref count again.
4976 atomic_inc(&eb->refs);
4979 btrfs_release_extent_buffer(eb);
4984 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4987 unsigned long len = fs_info->nodesize;
4988 unsigned long num_pages = num_extent_pages(start, len);
4990 unsigned long index = start >> PAGE_SHIFT;
4991 struct extent_buffer *eb;
4992 struct extent_buffer *exists = NULL;
4994 struct address_space *mapping = fs_info->btree_inode->i_mapping;
4998 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
4999 btrfs_err(fs_info, "bad tree block start %llu", start);
5000 return ERR_PTR(-EINVAL);
5003 eb = find_extent_buffer(fs_info, start);
5007 eb = __alloc_extent_buffer(fs_info, start, len);
5009 return ERR_PTR(-ENOMEM);
5011 for (i = 0; i < num_pages; i++, index++) {
5012 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
5014 exists = ERR_PTR(-ENOMEM);
5018 spin_lock(&mapping->private_lock);
5019 if (PagePrivate(p)) {
5021 * We could have already allocated an eb for this page
5022 * and attached one so lets see if we can get a ref on
5023 * the existing eb, and if we can we know it's good and
5024 * we can just return that one, else we know we can just
5025 * overwrite page->private.
5027 exists = (struct extent_buffer *)p->private;
5028 if (atomic_inc_not_zero(&exists->refs)) {
5029 spin_unlock(&mapping->private_lock);
5032 mark_extent_buffer_accessed(exists, p);
5038 * Do this so attach doesn't complain and we need to
5039 * drop the ref the old guy had.
5041 ClearPagePrivate(p);
5042 WARN_ON(PageDirty(p));
5045 attach_extent_buffer_page(eb, p);
5046 spin_unlock(&mapping->private_lock);
5047 WARN_ON(PageDirty(p));
5049 if (!PageUptodate(p))
5053 * see below about how we avoid a nasty race with release page
5054 * and why we unlock later
5058 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5060 ret = radix_tree_preload(GFP_NOFS);
5062 exists = ERR_PTR(ret);
5066 spin_lock(&fs_info->buffer_lock);
5067 ret = radix_tree_insert(&fs_info->buffer_radix,
5068 start >> PAGE_SHIFT, eb);
5069 spin_unlock(&fs_info->buffer_lock);
5070 radix_tree_preload_end();
5071 if (ret == -EEXIST) {
5072 exists = find_extent_buffer(fs_info, start);
5078 /* add one reference for the tree */
5079 check_buffer_tree_ref(eb);
5080 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5083 * there is a race where release page may have
5084 * tried to find this extent buffer in the radix
5085 * but failed. It will tell the VM it is safe to
5086 * reclaim the, and it will clear the page private bit.
5087 * We must make sure to set the page private bit properly
5088 * after the extent buffer is in the radix tree so
5089 * it doesn't get lost
5091 SetPageChecked(eb->pages[0]);
5092 for (i = 1; i < num_pages; i++) {
5094 ClearPageChecked(p);
5097 unlock_page(eb->pages[0]);
5101 WARN_ON(!atomic_dec_and_test(&eb->refs));
5102 for (i = 0; i < num_pages; i++) {
5104 unlock_page(eb->pages[i]);
5107 btrfs_release_extent_buffer(eb);
5111 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5113 struct extent_buffer *eb =
5114 container_of(head, struct extent_buffer, rcu_head);
5116 __free_extent_buffer(eb);
5119 /* Expects to have eb->eb_lock already held */
5120 static int release_extent_buffer(struct extent_buffer *eb)
5122 WARN_ON(atomic_read(&eb->refs) == 0);
5123 if (atomic_dec_and_test(&eb->refs)) {
5124 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5125 struct btrfs_fs_info *fs_info = eb->fs_info;
5127 spin_unlock(&eb->refs_lock);
5129 spin_lock(&fs_info->buffer_lock);
5130 radix_tree_delete(&fs_info->buffer_radix,
5131 eb->start >> PAGE_SHIFT);
5132 spin_unlock(&fs_info->buffer_lock);
5134 spin_unlock(&eb->refs_lock);
5137 /* Should be safe to release our pages at this point */
5138 btrfs_release_extent_buffer_page(eb);
5139 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5140 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) {
5141 __free_extent_buffer(eb);
5145 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5148 spin_unlock(&eb->refs_lock);
5153 void free_extent_buffer(struct extent_buffer *eb)
5161 refs = atomic_read(&eb->refs);
5164 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5169 spin_lock(&eb->refs_lock);
5170 if (atomic_read(&eb->refs) == 2 &&
5171 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
5172 atomic_dec(&eb->refs);
5174 if (atomic_read(&eb->refs) == 2 &&
5175 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5176 !extent_buffer_under_io(eb) &&
5177 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5178 atomic_dec(&eb->refs);
5181 * I know this is terrible, but it's temporary until we stop tracking
5182 * the uptodate bits and such for the extent buffers.
5184 release_extent_buffer(eb);
5187 void free_extent_buffer_stale(struct extent_buffer *eb)
5192 spin_lock(&eb->refs_lock);
5193 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5195 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5196 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5197 atomic_dec(&eb->refs);
5198 release_extent_buffer(eb);
5201 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5204 unsigned long num_pages;
5207 num_pages = num_extent_pages(eb->start, eb->len);
5209 for (i = 0; i < num_pages; i++) {
5210 page = eb->pages[i];
5211 if (!PageDirty(page))
5215 WARN_ON(!PagePrivate(page));
5217 clear_page_dirty_for_io(page);
5218 spin_lock_irq(&page->mapping->tree_lock);
5219 if (!PageDirty(page)) {
5220 radix_tree_tag_clear(&page->mapping->page_tree,
5222 PAGECACHE_TAG_DIRTY);
5224 spin_unlock_irq(&page->mapping->tree_lock);
5225 ClearPageError(page);
5228 WARN_ON(atomic_read(&eb->refs) == 0);
5231 int set_extent_buffer_dirty(struct extent_buffer *eb)
5234 unsigned long num_pages;
5237 check_buffer_tree_ref(eb);
5239 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5241 num_pages = num_extent_pages(eb->start, eb->len);
5242 WARN_ON(atomic_read(&eb->refs) == 0);
5243 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5245 for (i = 0; i < num_pages; i++)
5246 set_page_dirty(eb->pages[i]);
5250 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5254 unsigned long num_pages;
5256 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5257 num_pages = num_extent_pages(eb->start, eb->len);
5258 for (i = 0; i < num_pages; i++) {
5259 page = eb->pages[i];
5261 ClearPageUptodate(page);
5265 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5269 unsigned long num_pages;
5271 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5272 num_pages = num_extent_pages(eb->start, eb->len);
5273 for (i = 0; i < num_pages; i++) {
5274 page = eb->pages[i];
5275 SetPageUptodate(page);
5279 int extent_buffer_uptodate(struct extent_buffer *eb)
5281 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5284 int read_extent_buffer_pages(struct extent_io_tree *tree,
5285 struct extent_buffer *eb, int wait,
5286 get_extent_t *get_extent, int mirror_num)
5292 int locked_pages = 0;
5293 int all_uptodate = 1;
5294 unsigned long num_pages;
5295 unsigned long num_reads = 0;
5296 struct bio *bio = NULL;
5297 unsigned long bio_flags = 0;
5299 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5302 num_pages = num_extent_pages(eb->start, eb->len);
5303 for (i = 0; i < num_pages; i++) {
5304 page = eb->pages[i];
5305 if (wait == WAIT_NONE) {
5306 if (!trylock_page(page))
5314 * We need to firstly lock all pages to make sure that
5315 * the uptodate bit of our pages won't be affected by
5316 * clear_extent_buffer_uptodate().
5318 for (i = 0; i < num_pages; i++) {
5319 page = eb->pages[i];
5320 if (!PageUptodate(page)) {
5327 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5331 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5332 eb->read_mirror = 0;
5333 atomic_set(&eb->io_pages, num_reads);
5334 for (i = 0; i < num_pages; i++) {
5335 page = eb->pages[i];
5337 if (!PageUptodate(page)) {
5339 atomic_dec(&eb->io_pages);
5344 ClearPageError(page);
5345 err = __extent_read_full_page(tree, page,
5347 mirror_num, &bio_flags,
5352 * We use &bio in above __extent_read_full_page,
5353 * so we ensure that if it returns error, the
5354 * current page fails to add itself to bio and
5355 * it's been unlocked.
5357 * We must dec io_pages by ourselves.
5359 atomic_dec(&eb->io_pages);
5367 err = submit_one_bio(bio, mirror_num, bio_flags);
5372 if (ret || wait != WAIT_COMPLETE)
5375 for (i = 0; i < num_pages; i++) {
5376 page = eb->pages[i];
5377 wait_on_page_locked(page);
5378 if (!PageUptodate(page))
5385 while (locked_pages > 0) {
5387 page = eb->pages[locked_pages];
5393 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5394 unsigned long start, unsigned long len)
5400 char *dst = (char *)dstv;
5401 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5402 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5404 if (start + len > eb->len) {
5405 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5406 eb->start, eb->len, start, len);
5407 memset(dst, 0, len);
5411 offset = (start_offset + start) & (PAGE_SIZE - 1);
5414 page = eb->pages[i];
5416 cur = min(len, (PAGE_SIZE - offset));
5417 kaddr = page_address(page);
5418 memcpy(dst, kaddr + offset, cur);
5427 int read_extent_buffer_to_user(const struct extent_buffer *eb,
5429 unsigned long start, unsigned long len)
5435 char __user *dst = (char __user *)dstv;
5436 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5437 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5440 WARN_ON(start > eb->len);
5441 WARN_ON(start + len > eb->start + eb->len);
5443 offset = (start_offset + start) & (PAGE_SIZE - 1);
5446 page = eb->pages[i];
5448 cur = min(len, (PAGE_SIZE - offset));
5449 kaddr = page_address(page);
5450 if (copy_to_user(dst, kaddr + offset, cur)) {
5465 * return 0 if the item is found within a page.
5466 * return 1 if the item spans two pages.
5467 * return -EINVAL otherwise.
5469 int map_private_extent_buffer(const struct extent_buffer *eb,
5470 unsigned long start, unsigned long min_len,
5471 char **map, unsigned long *map_start,
5472 unsigned long *map_len)
5474 size_t offset = start & (PAGE_SIZE - 1);
5477 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5478 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5479 unsigned long end_i = (start_offset + start + min_len - 1) >>
5482 if (start + min_len > eb->len) {
5483 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5484 eb->start, eb->len, start, min_len);
5492 offset = start_offset;
5496 *map_start = ((u64)i << PAGE_SHIFT) - start_offset;
5500 kaddr = page_address(p);
5501 *map = kaddr + offset;
5502 *map_len = PAGE_SIZE - offset;
5506 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
5507 unsigned long start, unsigned long len)
5513 char *ptr = (char *)ptrv;
5514 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5515 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5518 WARN_ON(start > eb->len);
5519 WARN_ON(start + len > eb->start + eb->len);
5521 offset = (start_offset + start) & (PAGE_SIZE - 1);
5524 page = eb->pages[i];
5526 cur = min(len, (PAGE_SIZE - offset));
5528 kaddr = page_address(page);
5529 ret = memcmp(ptr, kaddr + offset, cur);
5541 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb,
5546 WARN_ON(!PageUptodate(eb->pages[0]));
5547 kaddr = page_address(eb->pages[0]);
5548 memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv,
5552 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv)
5556 WARN_ON(!PageUptodate(eb->pages[0]));
5557 kaddr = page_address(eb->pages[0]);
5558 memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv,
5562 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5563 unsigned long start, unsigned long len)
5569 char *src = (char *)srcv;
5570 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5571 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5573 WARN_ON(start > eb->len);
5574 WARN_ON(start + len > eb->start + eb->len);
5576 offset = (start_offset + start) & (PAGE_SIZE - 1);
5579 page = eb->pages[i];
5580 WARN_ON(!PageUptodate(page));
5582 cur = min(len, PAGE_SIZE - offset);
5583 kaddr = page_address(page);
5584 memcpy(kaddr + offset, src, cur);
5593 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start,
5600 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5601 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5603 WARN_ON(start > eb->len);
5604 WARN_ON(start + len > eb->start + eb->len);
5606 offset = (start_offset + start) & (PAGE_SIZE - 1);
5609 page = eb->pages[i];
5610 WARN_ON(!PageUptodate(page));
5612 cur = min(len, PAGE_SIZE - offset);
5613 kaddr = page_address(page);
5614 memset(kaddr + offset, 0, cur);
5622 void copy_extent_buffer_full(struct extent_buffer *dst,
5623 struct extent_buffer *src)
5628 ASSERT(dst->len == src->len);
5630 num_pages = num_extent_pages(dst->start, dst->len);
5631 for (i = 0; i < num_pages; i++)
5632 copy_page(page_address(dst->pages[i]),
5633 page_address(src->pages[i]));
5636 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5637 unsigned long dst_offset, unsigned long src_offset,
5640 u64 dst_len = dst->len;
5645 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5646 unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT;
5648 WARN_ON(src->len != dst_len);
5650 offset = (start_offset + dst_offset) &
5654 page = dst->pages[i];
5655 WARN_ON(!PageUptodate(page));
5657 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5659 kaddr = page_address(page);
5660 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5669 void le_bitmap_set(u8 *map, unsigned int start, int len)
5671 u8 *p = map + BIT_BYTE(start);
5672 const unsigned int size = start + len;
5673 int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE);
5674 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start);
5676 while (len - bits_to_set >= 0) {
5679 bits_to_set = BITS_PER_BYTE;
5684 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5689 void le_bitmap_clear(u8 *map, unsigned int start, int len)
5691 u8 *p = map + BIT_BYTE(start);
5692 const unsigned int size = start + len;
5693 int bits_to_clear = BITS_PER_BYTE - (start % BITS_PER_BYTE);
5694 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(start);
5696 while (len - bits_to_clear >= 0) {
5697 *p &= ~mask_to_clear;
5698 len -= bits_to_clear;
5699 bits_to_clear = BITS_PER_BYTE;
5704 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5705 *p &= ~mask_to_clear;
5710 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5712 * @eb: the extent buffer
5713 * @start: offset of the bitmap item in the extent buffer
5715 * @page_index: return index of the page in the extent buffer that contains the
5717 * @page_offset: return offset into the page given by page_index
5719 * This helper hides the ugliness of finding the byte in an extent buffer which
5720 * contains a given bit.
5722 static inline void eb_bitmap_offset(struct extent_buffer *eb,
5723 unsigned long start, unsigned long nr,
5724 unsigned long *page_index,
5725 size_t *page_offset)
5727 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5728 size_t byte_offset = BIT_BYTE(nr);
5732 * The byte we want is the offset of the extent buffer + the offset of
5733 * the bitmap item in the extent buffer + the offset of the byte in the
5736 offset = start_offset + start + byte_offset;
5738 *page_index = offset >> PAGE_SHIFT;
5739 *page_offset = offset & (PAGE_SIZE - 1);
5743 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5744 * @eb: the extent buffer
5745 * @start: offset of the bitmap item in the extent buffer
5746 * @nr: bit number to test
5748 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
5756 eb_bitmap_offset(eb, start, nr, &i, &offset);
5757 page = eb->pages[i];
5758 WARN_ON(!PageUptodate(page));
5759 kaddr = page_address(page);
5760 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5764 * extent_buffer_bitmap_set - set an area of a bitmap
5765 * @eb: the extent buffer
5766 * @start: offset of the bitmap item in the extent buffer
5767 * @pos: bit number of the first bit
5768 * @len: number of bits to set
5770 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
5771 unsigned long pos, unsigned long len)
5777 const unsigned int size = pos + len;
5778 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5779 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5781 eb_bitmap_offset(eb, start, pos, &i, &offset);
5782 page = eb->pages[i];
5783 WARN_ON(!PageUptodate(page));
5784 kaddr = page_address(page);
5786 while (len >= bits_to_set) {
5787 kaddr[offset] |= mask_to_set;
5789 bits_to_set = BITS_PER_BYTE;
5791 if (++offset >= PAGE_SIZE && len > 0) {
5793 page = eb->pages[++i];
5794 WARN_ON(!PageUptodate(page));
5795 kaddr = page_address(page);
5799 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5800 kaddr[offset] |= mask_to_set;
5806 * extent_buffer_bitmap_clear - clear an area of a bitmap
5807 * @eb: the extent buffer
5808 * @start: offset of the bitmap item in the extent buffer
5809 * @pos: bit number of the first bit
5810 * @len: number of bits to clear
5812 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
5813 unsigned long pos, unsigned long len)
5819 const unsigned int size = pos + len;
5820 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5821 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5823 eb_bitmap_offset(eb, start, pos, &i, &offset);
5824 page = eb->pages[i];
5825 WARN_ON(!PageUptodate(page));
5826 kaddr = page_address(page);
5828 while (len >= bits_to_clear) {
5829 kaddr[offset] &= ~mask_to_clear;
5830 len -= bits_to_clear;
5831 bits_to_clear = BITS_PER_BYTE;
5833 if (++offset >= PAGE_SIZE && len > 0) {
5835 page = eb->pages[++i];
5836 WARN_ON(!PageUptodate(page));
5837 kaddr = page_address(page);
5841 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5842 kaddr[offset] &= ~mask_to_clear;
5846 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5848 unsigned long distance = (src > dst) ? src - dst : dst - src;
5849 return distance < len;
5852 static void copy_pages(struct page *dst_page, struct page *src_page,
5853 unsigned long dst_off, unsigned long src_off,
5856 char *dst_kaddr = page_address(dst_page);
5858 int must_memmove = 0;
5860 if (dst_page != src_page) {
5861 src_kaddr = page_address(src_page);
5863 src_kaddr = dst_kaddr;
5864 if (areas_overlap(src_off, dst_off, len))
5869 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5871 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5874 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5875 unsigned long src_offset, unsigned long len)
5877 struct btrfs_fs_info *fs_info = dst->fs_info;
5879 size_t dst_off_in_page;
5880 size_t src_off_in_page;
5881 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5882 unsigned long dst_i;
5883 unsigned long src_i;
5885 if (src_offset + len > dst->len) {
5887 "memmove bogus src_offset %lu move len %lu dst len %lu",
5888 src_offset, len, dst->len);
5891 if (dst_offset + len > dst->len) {
5893 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5894 dst_offset, len, dst->len);
5899 dst_off_in_page = (start_offset + dst_offset) &
5901 src_off_in_page = (start_offset + src_offset) &
5904 dst_i = (start_offset + dst_offset) >> PAGE_SHIFT;
5905 src_i = (start_offset + src_offset) >> PAGE_SHIFT;
5907 cur = min(len, (unsigned long)(PAGE_SIZE -
5909 cur = min_t(unsigned long, cur,
5910 (unsigned long)(PAGE_SIZE - dst_off_in_page));
5912 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5913 dst_off_in_page, src_off_in_page, cur);
5921 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5922 unsigned long src_offset, unsigned long len)
5924 struct btrfs_fs_info *fs_info = dst->fs_info;
5926 size_t dst_off_in_page;
5927 size_t src_off_in_page;
5928 unsigned long dst_end = dst_offset + len - 1;
5929 unsigned long src_end = src_offset + len - 1;
5930 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5931 unsigned long dst_i;
5932 unsigned long src_i;
5934 if (src_offset + len > dst->len) {
5936 "memmove bogus src_offset %lu move len %lu len %lu",
5937 src_offset, len, dst->len);
5940 if (dst_offset + len > dst->len) {
5942 "memmove bogus dst_offset %lu move len %lu len %lu",
5943 dst_offset, len, dst->len);
5946 if (dst_offset < src_offset) {
5947 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5951 dst_i = (start_offset + dst_end) >> PAGE_SHIFT;
5952 src_i = (start_offset + src_end) >> PAGE_SHIFT;
5954 dst_off_in_page = (start_offset + dst_end) &
5956 src_off_in_page = (start_offset + src_end) &
5959 cur = min_t(unsigned long, len, src_off_in_page + 1);
5960 cur = min(cur, dst_off_in_page + 1);
5961 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5962 dst_off_in_page - cur + 1,
5963 src_off_in_page - cur + 1, cur);
5971 int try_release_extent_buffer(struct page *page)
5973 struct extent_buffer *eb;
5976 * We need to make sure nobody is attaching this page to an eb right
5979 spin_lock(&page->mapping->private_lock);
5980 if (!PagePrivate(page)) {
5981 spin_unlock(&page->mapping->private_lock);
5985 eb = (struct extent_buffer *)page->private;
5989 * This is a little awful but should be ok, we need to make sure that
5990 * the eb doesn't disappear out from under us while we're looking at
5993 spin_lock(&eb->refs_lock);
5994 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5995 spin_unlock(&eb->refs_lock);
5996 spin_unlock(&page->mapping->private_lock);
5999 spin_unlock(&page->mapping->private_lock);
6002 * If tree ref isn't set then we know the ref on this eb is a real ref,
6003 * so just return, this page will likely be freed soon anyway.
6005 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
6006 spin_unlock(&eb->refs_lock);
6010 return release_extent_buffer(eb);