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]);
1724 for (; i < ret; i++)
1738 if (err && index_ret)
1739 *index_ret = start_index + pages_locked - 1;
1743 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1744 u64 delalloc_end, struct page *locked_page,
1745 unsigned clear_bits,
1746 unsigned long page_ops)
1748 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0,
1751 __process_pages_contig(inode->i_mapping, locked_page,
1752 start >> PAGE_SHIFT, end >> PAGE_SHIFT,
1757 * count the number of bytes in the tree that have a given bit(s)
1758 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1759 * cached. The total number found is returned.
1761 u64 count_range_bits(struct extent_io_tree *tree,
1762 u64 *start, u64 search_end, u64 max_bytes,
1763 unsigned bits, int contig)
1765 struct rb_node *node;
1766 struct extent_state *state;
1767 u64 cur_start = *start;
1768 u64 total_bytes = 0;
1772 if (WARN_ON(search_end <= cur_start))
1775 spin_lock(&tree->lock);
1776 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1777 total_bytes = tree->dirty_bytes;
1781 * this search will find all the extents that end after
1784 node = tree_search(tree, cur_start);
1789 state = rb_entry(node, struct extent_state, rb_node);
1790 if (state->start > search_end)
1792 if (contig && found && state->start > last + 1)
1794 if (state->end >= cur_start && (state->state & bits) == bits) {
1795 total_bytes += min(search_end, state->end) + 1 -
1796 max(cur_start, state->start);
1797 if (total_bytes >= max_bytes)
1800 *start = max(cur_start, state->start);
1804 } else if (contig && found) {
1807 node = rb_next(node);
1812 spin_unlock(&tree->lock);
1817 * set the private field for a given byte offset in the tree. If there isn't
1818 * an extent_state there already, this does nothing.
1820 static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start,
1821 struct io_failure_record *failrec)
1823 struct rb_node *node;
1824 struct extent_state *state;
1827 spin_lock(&tree->lock);
1829 * this search will find all the extents that end after
1832 node = tree_search(tree, start);
1837 state = rb_entry(node, struct extent_state, rb_node);
1838 if (state->start != start) {
1842 state->failrec = failrec;
1844 spin_unlock(&tree->lock);
1848 static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start,
1849 struct io_failure_record **failrec)
1851 struct rb_node *node;
1852 struct extent_state *state;
1855 spin_lock(&tree->lock);
1857 * this search will find all the extents that end after
1860 node = tree_search(tree, start);
1865 state = rb_entry(node, struct extent_state, rb_node);
1866 if (state->start != start) {
1870 *failrec = state->failrec;
1872 spin_unlock(&tree->lock);
1877 * searches a range in the state tree for a given mask.
1878 * If 'filled' == 1, this returns 1 only if every extent in the tree
1879 * has the bits set. Otherwise, 1 is returned if any bit in the
1880 * range is found set.
1882 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1883 unsigned bits, int filled, struct extent_state *cached)
1885 struct extent_state *state = NULL;
1886 struct rb_node *node;
1889 spin_lock(&tree->lock);
1890 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
1891 cached->end > start)
1892 node = &cached->rb_node;
1894 node = tree_search(tree, start);
1895 while (node && start <= end) {
1896 state = rb_entry(node, struct extent_state, rb_node);
1898 if (filled && state->start > start) {
1903 if (state->start > end)
1906 if (state->state & bits) {
1910 } else if (filled) {
1915 if (state->end == (u64)-1)
1918 start = state->end + 1;
1921 node = rb_next(node);
1928 spin_unlock(&tree->lock);
1933 * helper function to set a given page up to date if all the
1934 * extents in the tree for that page are up to date
1936 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1938 u64 start = page_offset(page);
1939 u64 end = start + PAGE_SIZE - 1;
1940 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1941 SetPageUptodate(page);
1944 int free_io_failure(struct extent_io_tree *failure_tree,
1945 struct extent_io_tree *io_tree,
1946 struct io_failure_record *rec)
1951 set_state_failrec(failure_tree, rec->start, NULL);
1952 ret = clear_extent_bits(failure_tree, rec->start,
1953 rec->start + rec->len - 1,
1954 EXTENT_LOCKED | EXTENT_DIRTY);
1958 ret = clear_extent_bits(io_tree, rec->start,
1959 rec->start + rec->len - 1,
1969 * this bypasses the standard btrfs submit functions deliberately, as
1970 * the standard behavior is to write all copies in a raid setup. here we only
1971 * want to write the one bad copy. so we do the mapping for ourselves and issue
1972 * submit_bio directly.
1973 * to avoid any synchronization issues, wait for the data after writing, which
1974 * actually prevents the read that triggered the error from finishing.
1975 * currently, there can be no more than two copies of every data bit. thus,
1976 * exactly one rewrite is required.
1978 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
1979 u64 length, u64 logical, struct page *page,
1980 unsigned int pg_offset, int mirror_num)
1983 struct btrfs_device *dev;
1986 struct btrfs_bio *bbio = NULL;
1989 ASSERT(!(fs_info->sb->s_flags & MS_RDONLY));
1990 BUG_ON(!mirror_num);
1992 bio = btrfs_io_bio_alloc(1);
1993 bio->bi_iter.bi_size = 0;
1994 map_length = length;
1997 * Avoid races with device replace and make sure our bbio has devices
1998 * associated to its stripes that don't go away while we are doing the
1999 * read repair operation.
2001 btrfs_bio_counter_inc_blocked(fs_info);
2002 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2004 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2005 * to update all raid stripes, but here we just want to correct
2006 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2007 * stripe's dev and sector.
2009 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2010 &map_length, &bbio, 0);
2012 btrfs_bio_counter_dec(fs_info);
2016 ASSERT(bbio->mirror_num == 1);
2018 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2019 &map_length, &bbio, mirror_num);
2021 btrfs_bio_counter_dec(fs_info);
2025 BUG_ON(mirror_num != bbio->mirror_num);
2028 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2029 bio->bi_iter.bi_sector = sector;
2030 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2031 btrfs_put_bbio(bbio);
2032 if (!dev || !dev->bdev || !dev->writeable) {
2033 btrfs_bio_counter_dec(fs_info);
2037 bio_set_dev(bio, dev->bdev);
2038 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2039 bio_add_page(bio, page, length, pg_offset);
2041 if (btrfsic_submit_bio_wait(bio)) {
2042 /* try to remap that extent elsewhere? */
2043 btrfs_bio_counter_dec(fs_info);
2045 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2049 btrfs_info_rl_in_rcu(fs_info,
2050 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2052 rcu_str_deref(dev->name), sector);
2053 btrfs_bio_counter_dec(fs_info);
2058 int repair_eb_io_failure(struct btrfs_fs_info *fs_info,
2059 struct extent_buffer *eb, int mirror_num)
2061 u64 start = eb->start;
2062 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2065 if (sb_rdonly(fs_info->sb))
2068 for (i = 0; i < num_pages; i++) {
2069 struct page *p = eb->pages[i];
2071 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2072 start - page_offset(p), mirror_num);
2082 * each time an IO finishes, we do a fast check in the IO failure tree
2083 * to see if we need to process or clean up an io_failure_record
2085 int clean_io_failure(struct btrfs_fs_info *fs_info,
2086 struct extent_io_tree *failure_tree,
2087 struct extent_io_tree *io_tree, u64 start,
2088 struct page *page, u64 ino, unsigned int pg_offset)
2091 struct io_failure_record *failrec;
2092 struct extent_state *state;
2097 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2102 ret = get_state_failrec(failure_tree, start, &failrec);
2106 BUG_ON(!failrec->this_mirror);
2108 if (failrec->in_validation) {
2109 /* there was no real error, just free the record */
2110 btrfs_debug(fs_info,
2111 "clean_io_failure: freeing dummy error at %llu",
2115 if (sb_rdonly(fs_info->sb))
2118 spin_lock(&io_tree->lock);
2119 state = find_first_extent_bit_state(io_tree,
2122 spin_unlock(&io_tree->lock);
2124 if (state && state->start <= failrec->start &&
2125 state->end >= failrec->start + failrec->len - 1) {
2126 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2128 if (num_copies > 1) {
2129 repair_io_failure(fs_info, ino, start, failrec->len,
2130 failrec->logical, page, pg_offset,
2131 failrec->failed_mirror);
2136 free_io_failure(failure_tree, io_tree, failrec);
2142 * Can be called when
2143 * - hold extent lock
2144 * - under ordered extent
2145 * - the inode is freeing
2147 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2149 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2150 struct io_failure_record *failrec;
2151 struct extent_state *state, *next;
2153 if (RB_EMPTY_ROOT(&failure_tree->state))
2156 spin_lock(&failure_tree->lock);
2157 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2159 if (state->start > end)
2162 ASSERT(state->end <= end);
2164 next = next_state(state);
2166 failrec = state->failrec;
2167 free_extent_state(state);
2172 spin_unlock(&failure_tree->lock);
2175 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2176 struct io_failure_record **failrec_ret)
2178 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2179 struct io_failure_record *failrec;
2180 struct extent_map *em;
2181 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2182 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2183 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2187 ret = get_state_failrec(failure_tree, start, &failrec);
2189 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2193 failrec->start = start;
2194 failrec->len = end - start + 1;
2195 failrec->this_mirror = 0;
2196 failrec->bio_flags = 0;
2197 failrec->in_validation = 0;
2199 read_lock(&em_tree->lock);
2200 em = lookup_extent_mapping(em_tree, start, failrec->len);
2202 read_unlock(&em_tree->lock);
2207 if (em->start > start || em->start + em->len <= start) {
2208 free_extent_map(em);
2211 read_unlock(&em_tree->lock);
2217 logical = start - em->start;
2218 logical = em->block_start + logical;
2219 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2220 logical = em->block_start;
2221 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2222 extent_set_compress_type(&failrec->bio_flags,
2226 btrfs_debug(fs_info,
2227 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2228 logical, start, failrec->len);
2230 failrec->logical = logical;
2231 free_extent_map(em);
2233 /* set the bits in the private failure tree */
2234 ret = set_extent_bits(failure_tree, start, end,
2235 EXTENT_LOCKED | EXTENT_DIRTY);
2237 ret = set_state_failrec(failure_tree, start, failrec);
2238 /* set the bits in the inode's tree */
2240 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2246 btrfs_debug(fs_info,
2247 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2248 failrec->logical, failrec->start, failrec->len,
2249 failrec->in_validation);
2251 * when data can be on disk more than twice, add to failrec here
2252 * (e.g. with a list for failed_mirror) to make
2253 * clean_io_failure() clean all those errors at once.
2257 *failrec_ret = failrec;
2262 bool btrfs_check_repairable(struct inode *inode, struct bio *failed_bio,
2263 struct io_failure_record *failrec, int failed_mirror)
2265 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2268 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2269 if (num_copies == 1) {
2271 * we only have a single copy of the data, so don't bother with
2272 * all the retry and error correction code that follows. no
2273 * matter what the error is, it is very likely to persist.
2275 btrfs_debug(fs_info,
2276 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2277 num_copies, failrec->this_mirror, failed_mirror);
2282 * there are two premises:
2283 * a) deliver good data to the caller
2284 * b) correct the bad sectors on disk
2286 if (failed_bio->bi_vcnt > 1) {
2288 * to fulfill b), we need to know the exact failing sectors, as
2289 * we don't want to rewrite any more than the failed ones. thus,
2290 * we need separate read requests for the failed bio
2292 * if the following BUG_ON triggers, our validation request got
2293 * merged. we need separate requests for our algorithm to work.
2295 BUG_ON(failrec->in_validation);
2296 failrec->in_validation = 1;
2297 failrec->this_mirror = failed_mirror;
2300 * we're ready to fulfill a) and b) alongside. get a good copy
2301 * of the failed sector and if we succeed, we have setup
2302 * everything for repair_io_failure to do the rest for us.
2304 if (failrec->in_validation) {
2305 BUG_ON(failrec->this_mirror != failed_mirror);
2306 failrec->in_validation = 0;
2307 failrec->this_mirror = 0;
2309 failrec->failed_mirror = failed_mirror;
2310 failrec->this_mirror++;
2311 if (failrec->this_mirror == failed_mirror)
2312 failrec->this_mirror++;
2315 if (failrec->this_mirror > num_copies) {
2316 btrfs_debug(fs_info,
2317 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2318 num_copies, failrec->this_mirror, failed_mirror);
2326 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2327 struct io_failure_record *failrec,
2328 struct page *page, int pg_offset, int icsum,
2329 bio_end_io_t *endio_func, void *data)
2331 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2333 struct btrfs_io_bio *btrfs_failed_bio;
2334 struct btrfs_io_bio *btrfs_bio;
2336 bio = btrfs_io_bio_alloc(1);
2337 bio->bi_end_io = endio_func;
2338 bio->bi_iter.bi_sector = failrec->logical >> 9;
2339 bio_set_dev(bio, fs_info->fs_devices->latest_bdev);
2340 bio->bi_iter.bi_size = 0;
2341 bio->bi_private = data;
2343 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2344 if (btrfs_failed_bio->csum) {
2345 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2347 btrfs_bio = btrfs_io_bio(bio);
2348 btrfs_bio->csum = btrfs_bio->csum_inline;
2350 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2354 bio_add_page(bio, page, failrec->len, pg_offset);
2360 * this is a generic handler for readpage errors (default
2361 * readpage_io_failed_hook). if other copies exist, read those and write back
2362 * good data to the failed position. does not investigate in remapping the
2363 * failed extent elsewhere, hoping the device will be smart enough to do this as
2367 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2368 struct page *page, u64 start, u64 end,
2371 struct io_failure_record *failrec;
2372 struct inode *inode = page->mapping->host;
2373 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2374 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2377 blk_status_t status;
2380 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2382 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2386 if (!btrfs_check_repairable(inode, failed_bio, failrec,
2388 free_io_failure(failure_tree, tree, failrec);
2392 if (failed_bio->bi_vcnt > 1)
2393 read_mode |= REQ_FAILFAST_DEV;
2395 phy_offset >>= inode->i_sb->s_blocksize_bits;
2396 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2397 start - page_offset(page),
2398 (int)phy_offset, failed_bio->bi_end_io,
2400 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
2402 btrfs_debug(btrfs_sb(inode->i_sb),
2403 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2404 read_mode, failrec->this_mirror, failrec->in_validation);
2406 status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror,
2407 failrec->bio_flags, 0);
2409 free_io_failure(failure_tree, tree, failrec);
2411 ret = blk_status_to_errno(status);
2417 /* lots and lots of room for performance fixes in the end_bio funcs */
2419 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2421 int uptodate = (err == 0);
2422 struct extent_io_tree *tree;
2425 tree = &BTRFS_I(page->mapping->host)->io_tree;
2427 if (tree->ops && tree->ops->writepage_end_io_hook)
2428 tree->ops->writepage_end_io_hook(page, start, end, NULL,
2432 ClearPageUptodate(page);
2434 ret = err < 0 ? err : -EIO;
2435 mapping_set_error(page->mapping, ret);
2440 * after a writepage IO is done, we need to:
2441 * clear the uptodate bits on error
2442 * clear the writeback bits in the extent tree for this IO
2443 * end_page_writeback if the page has no more pending IO
2445 * Scheduling is not allowed, so the extent state tree is expected
2446 * to have one and only one object corresponding to this IO.
2448 static void end_bio_extent_writepage(struct bio *bio)
2450 int error = blk_status_to_errno(bio->bi_status);
2451 struct bio_vec *bvec;
2456 ASSERT(!bio_flagged(bio, BIO_CLONED));
2457 bio_for_each_segment_all(bvec, bio, i) {
2458 struct page *page = bvec->bv_page;
2459 struct inode *inode = page->mapping->host;
2460 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2462 /* We always issue full-page reads, but if some block
2463 * in a page fails to read, blk_update_request() will
2464 * advance bv_offset and adjust bv_len to compensate.
2465 * Print a warning for nonzero offsets, and an error
2466 * if they don't add up to a full page. */
2467 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2468 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2470 "partial page write in btrfs with offset %u and length %u",
2471 bvec->bv_offset, bvec->bv_len);
2474 "incomplete page write in btrfs with offset %u and length %u",
2475 bvec->bv_offset, bvec->bv_len);
2478 start = page_offset(page);
2479 end = start + bvec->bv_offset + bvec->bv_len - 1;
2481 end_extent_writepage(page, error, start, end);
2482 end_page_writeback(page);
2489 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2492 struct extent_state *cached = NULL;
2493 u64 end = start + len - 1;
2495 if (uptodate && tree->track_uptodate)
2496 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2497 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2501 * after a readpage IO is done, we need to:
2502 * clear the uptodate bits on error
2503 * set the uptodate bits if things worked
2504 * set the page up to date if all extents in the tree are uptodate
2505 * clear the lock bit in the extent tree
2506 * unlock the page if there are no other extents locked for it
2508 * Scheduling is not allowed, so the extent state tree is expected
2509 * to have one and only one object corresponding to this IO.
2511 static void end_bio_extent_readpage(struct bio *bio)
2513 struct bio_vec *bvec;
2514 int uptodate = !bio->bi_status;
2515 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2516 struct extent_io_tree *tree, *failure_tree;
2521 u64 extent_start = 0;
2527 ASSERT(!bio_flagged(bio, BIO_CLONED));
2528 bio_for_each_segment_all(bvec, bio, i) {
2529 struct page *page = bvec->bv_page;
2530 struct inode *inode = page->mapping->host;
2531 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2533 btrfs_debug(fs_info,
2534 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2535 (u64)bio->bi_iter.bi_sector, bio->bi_status,
2536 io_bio->mirror_num);
2537 tree = &BTRFS_I(inode)->io_tree;
2538 failure_tree = &BTRFS_I(inode)->io_failure_tree;
2540 /* We always issue full-page reads, but if some block
2541 * in a page fails to read, blk_update_request() will
2542 * advance bv_offset and adjust bv_len to compensate.
2543 * Print a warning for nonzero offsets, and an error
2544 * if they don't add up to a full page. */
2545 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2546 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2548 "partial page read in btrfs with offset %u and length %u",
2549 bvec->bv_offset, bvec->bv_len);
2552 "incomplete page read in btrfs with offset %u and length %u",
2553 bvec->bv_offset, bvec->bv_len);
2556 start = page_offset(page);
2557 end = start + bvec->bv_offset + bvec->bv_len - 1;
2560 mirror = io_bio->mirror_num;
2561 if (likely(uptodate && tree->ops)) {
2562 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2568 clean_io_failure(BTRFS_I(inode)->root->fs_info,
2569 failure_tree, tree, start,
2571 btrfs_ino(BTRFS_I(inode)), 0);
2574 if (likely(uptodate))
2578 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2579 if (ret == -EAGAIN) {
2581 * Data inode's readpage_io_failed_hook() always
2584 * The generic bio_readpage_error handles errors
2585 * the following way: If possible, new read
2586 * requests are created and submitted and will
2587 * end up in end_bio_extent_readpage as well (if
2588 * we're lucky, not in the !uptodate case). In
2589 * that case it returns 0 and we just go on with
2590 * the next page in our bio. If it can't handle
2591 * the error it will return -EIO and we remain
2592 * responsible for that page.
2594 ret = bio_readpage_error(bio, offset, page,
2595 start, end, mirror);
2597 uptodate = !bio->bi_status;
2604 * metadata's readpage_io_failed_hook() always returns
2605 * -EIO and fixes nothing. -EIO is also returned if
2606 * data inode error could not be fixed.
2608 ASSERT(ret == -EIO);
2611 if (likely(uptodate)) {
2612 loff_t i_size = i_size_read(inode);
2613 pgoff_t end_index = i_size >> PAGE_SHIFT;
2616 /* Zero out the end if this page straddles i_size */
2617 off = i_size & (PAGE_SIZE-1);
2618 if (page->index == end_index && off)
2619 zero_user_segment(page, off, PAGE_SIZE);
2620 SetPageUptodate(page);
2622 ClearPageUptodate(page);
2628 if (unlikely(!uptodate)) {
2630 endio_readpage_release_extent(tree,
2636 endio_readpage_release_extent(tree, start,
2637 end - start + 1, 0);
2638 } else if (!extent_len) {
2639 extent_start = start;
2640 extent_len = end + 1 - start;
2641 } else if (extent_start + extent_len == start) {
2642 extent_len += end + 1 - start;
2644 endio_readpage_release_extent(tree, extent_start,
2645 extent_len, uptodate);
2646 extent_start = start;
2647 extent_len = end + 1 - start;
2652 endio_readpage_release_extent(tree, extent_start, extent_len,
2655 io_bio->end_io(io_bio, blk_status_to_errno(bio->bi_status));
2660 * Initialize the members up to but not including 'bio'. Use after allocating a
2661 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2662 * 'bio' because use of __GFP_ZERO is not supported.
2664 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
2666 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
2670 * The following helpers allocate a bio. As it's backed by a bioset, it'll
2671 * never fail. We're returning a bio right now but you can call btrfs_io_bio
2672 * for the appropriate container_of magic
2674 struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte)
2678 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, btrfs_bioset);
2679 bio_set_dev(bio, bdev);
2680 bio->bi_iter.bi_sector = first_byte >> 9;
2681 btrfs_io_bio_init(btrfs_io_bio(bio));
2685 struct bio *btrfs_bio_clone(struct bio *bio)
2687 struct btrfs_io_bio *btrfs_bio;
2690 /* Bio allocation backed by a bioset does not fail */
2691 new = bio_clone_fast(bio, GFP_NOFS, btrfs_bioset);
2692 btrfs_bio = btrfs_io_bio(new);
2693 btrfs_io_bio_init(btrfs_bio);
2694 btrfs_bio->iter = bio->bi_iter;
2698 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
2702 /* Bio allocation backed by a bioset does not fail */
2703 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, btrfs_bioset);
2704 btrfs_io_bio_init(btrfs_io_bio(bio));
2708 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
2711 struct btrfs_io_bio *btrfs_bio;
2713 /* this will never fail when it's backed by a bioset */
2714 bio = bio_clone_fast(orig, GFP_NOFS, btrfs_bioset);
2717 btrfs_bio = btrfs_io_bio(bio);
2718 btrfs_io_bio_init(btrfs_bio);
2720 bio_trim(bio, offset >> 9, size >> 9);
2721 btrfs_bio->iter = bio->bi_iter;
2725 static int __must_check submit_one_bio(struct bio *bio, int mirror_num,
2726 unsigned long bio_flags)
2728 blk_status_t ret = 0;
2729 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2730 struct page *page = bvec->bv_page;
2731 struct extent_io_tree *tree = bio->bi_private;
2734 start = page_offset(page) + bvec->bv_offset;
2736 bio->bi_private = NULL;
2740 ret = tree->ops->submit_bio_hook(tree->private_data, bio,
2741 mirror_num, bio_flags, start);
2743 btrfsic_submit_bio(bio);
2746 return blk_status_to_errno(ret);
2749 static int merge_bio(struct extent_io_tree *tree, struct page *page,
2750 unsigned long offset, size_t size, struct bio *bio,
2751 unsigned long bio_flags)
2755 ret = tree->ops->merge_bio_hook(page, offset, size, bio,
2762 * @opf: bio REQ_OP_* and REQ_* flags as one value
2764 static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree,
2765 struct writeback_control *wbc,
2766 struct page *page, sector_t sector,
2767 size_t size, unsigned long offset,
2768 struct block_device *bdev,
2769 struct bio **bio_ret,
2770 bio_end_io_t end_io_func,
2772 unsigned long prev_bio_flags,
2773 unsigned long bio_flags,
2774 bool force_bio_submit)
2779 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2780 size_t page_size = min_t(size_t, size, PAGE_SIZE);
2782 if (bio_ret && *bio_ret) {
2785 contig = bio->bi_iter.bi_sector == sector;
2787 contig = bio_end_sector(bio) == sector;
2789 if (prev_bio_flags != bio_flags || !contig ||
2791 merge_bio(tree, page, offset, page_size, bio, bio_flags) ||
2792 bio_add_page(bio, page, page_size, offset) < page_size) {
2793 ret = submit_one_bio(bio, mirror_num, prev_bio_flags);
2801 wbc_account_io(wbc, page, page_size);
2806 bio = btrfs_bio_alloc(bdev, (u64)sector << 9);
2807 bio_add_page(bio, page, page_size, offset);
2808 bio->bi_end_io = end_io_func;
2809 bio->bi_private = tree;
2810 bio->bi_write_hint = page->mapping->host->i_write_hint;
2813 wbc_init_bio(wbc, bio);
2814 wbc_account_io(wbc, page, page_size);
2820 ret = submit_one_bio(bio, mirror_num, bio_flags);
2825 static void attach_extent_buffer_page(struct extent_buffer *eb,
2828 if (!PagePrivate(page)) {
2829 SetPagePrivate(page);
2831 set_page_private(page, (unsigned long)eb);
2833 WARN_ON(page->private != (unsigned long)eb);
2837 void set_page_extent_mapped(struct page *page)
2839 if (!PagePrivate(page)) {
2840 SetPagePrivate(page);
2842 set_page_private(page, EXTENT_PAGE_PRIVATE);
2846 static struct extent_map *
2847 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2848 u64 start, u64 len, get_extent_t *get_extent,
2849 struct extent_map **em_cached)
2851 struct extent_map *em;
2853 if (em_cached && *em_cached) {
2855 if (extent_map_in_tree(em) && start >= em->start &&
2856 start < extent_map_end(em)) {
2857 refcount_inc(&em->refs);
2861 free_extent_map(em);
2865 em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0);
2866 if (em_cached && !IS_ERR_OR_NULL(em)) {
2868 refcount_inc(&em->refs);
2874 * basic readpage implementation. Locked extent state structs are inserted
2875 * into the tree that are removed when the IO is done (by the end_io
2877 * XXX JDM: This needs looking at to ensure proper page locking
2878 * return 0 on success, otherwise return error
2880 static int __do_readpage(struct extent_io_tree *tree,
2882 get_extent_t *get_extent,
2883 struct extent_map **em_cached,
2884 struct bio **bio, int mirror_num,
2885 unsigned long *bio_flags, unsigned int read_flags,
2888 struct inode *inode = page->mapping->host;
2889 u64 start = page_offset(page);
2890 u64 page_end = start + PAGE_SIZE - 1;
2894 u64 last_byte = i_size_read(inode);
2898 struct extent_map *em;
2899 struct block_device *bdev;
2902 size_t pg_offset = 0;
2904 size_t disk_io_size;
2905 size_t blocksize = inode->i_sb->s_blocksize;
2906 unsigned long this_bio_flag = 0;
2908 set_page_extent_mapped(page);
2911 if (!PageUptodate(page)) {
2912 if (cleancache_get_page(page) == 0) {
2913 BUG_ON(blocksize != PAGE_SIZE);
2914 unlock_extent(tree, start, end);
2919 if (page->index == last_byte >> PAGE_SHIFT) {
2921 size_t zero_offset = last_byte & (PAGE_SIZE - 1);
2924 iosize = PAGE_SIZE - zero_offset;
2925 userpage = kmap_atomic(page);
2926 memset(userpage + zero_offset, 0, iosize);
2927 flush_dcache_page(page);
2928 kunmap_atomic(userpage);
2931 while (cur <= end) {
2932 bool force_bio_submit = false;
2934 if (cur >= last_byte) {
2936 struct extent_state *cached = NULL;
2938 iosize = PAGE_SIZE - pg_offset;
2939 userpage = kmap_atomic(page);
2940 memset(userpage + pg_offset, 0, iosize);
2941 flush_dcache_page(page);
2942 kunmap_atomic(userpage);
2943 set_extent_uptodate(tree, cur, cur + iosize - 1,
2945 unlock_extent_cached(tree, cur,
2950 em = __get_extent_map(inode, page, pg_offset, cur,
2951 end - cur + 1, get_extent, em_cached);
2952 if (IS_ERR_OR_NULL(em)) {
2954 unlock_extent(tree, cur, end);
2957 extent_offset = cur - em->start;
2958 BUG_ON(extent_map_end(em) <= cur);
2961 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2962 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2963 extent_set_compress_type(&this_bio_flag,
2967 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2968 cur_end = min(extent_map_end(em) - 1, end);
2969 iosize = ALIGN(iosize, blocksize);
2970 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2971 disk_io_size = em->block_len;
2972 sector = em->block_start >> 9;
2974 sector = (em->block_start + extent_offset) >> 9;
2975 disk_io_size = iosize;
2978 block_start = em->block_start;
2979 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2980 block_start = EXTENT_MAP_HOLE;
2983 * If we have a file range that points to a compressed extent
2984 * and it's followed by a consecutive file range that points to
2985 * to the same compressed extent (possibly with a different
2986 * offset and/or length, so it either points to the whole extent
2987 * or only part of it), we must make sure we do not submit a
2988 * single bio to populate the pages for the 2 ranges because
2989 * this makes the compressed extent read zero out the pages
2990 * belonging to the 2nd range. Imagine the following scenario:
2993 * [0 - 8K] [8K - 24K]
2996 * points to extent X, points to extent X,
2997 * offset 4K, length of 8K offset 0, length 16K
2999 * [extent X, compressed length = 4K uncompressed length = 16K]
3001 * If the bio to read the compressed extent covers both ranges,
3002 * it will decompress extent X into the pages belonging to the
3003 * first range and then it will stop, zeroing out the remaining
3004 * pages that belong to the other range that points to extent X.
3005 * So here we make sure we submit 2 bios, one for the first
3006 * range and another one for the third range. Both will target
3007 * the same physical extent from disk, but we can't currently
3008 * make the compressed bio endio callback populate the pages
3009 * for both ranges because each compressed bio is tightly
3010 * coupled with a single extent map, and each range can have
3011 * an extent map with a different offset value relative to the
3012 * uncompressed data of our extent and different lengths. This
3013 * is a corner case so we prioritize correctness over
3014 * non-optimal behavior (submitting 2 bios for the same extent).
3016 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3017 prev_em_start && *prev_em_start != (u64)-1 &&
3018 *prev_em_start != em->start)
3019 force_bio_submit = true;
3022 *prev_em_start = em->start;
3024 free_extent_map(em);
3027 /* we've found a hole, just zero and go on */
3028 if (block_start == EXTENT_MAP_HOLE) {
3030 struct extent_state *cached = NULL;
3032 userpage = kmap_atomic(page);
3033 memset(userpage + pg_offset, 0, iosize);
3034 flush_dcache_page(page);
3035 kunmap_atomic(userpage);
3037 set_extent_uptodate(tree, cur, cur + iosize - 1,
3039 unlock_extent_cached(tree, cur,
3043 pg_offset += iosize;
3046 /* the get_extent function already copied into the page */
3047 if (test_range_bit(tree, cur, cur_end,
3048 EXTENT_UPTODATE, 1, NULL)) {
3049 check_page_uptodate(tree, page);
3050 unlock_extent(tree, cur, cur + iosize - 1);
3052 pg_offset += iosize;
3055 /* we have an inline extent but it didn't get marked up
3056 * to date. Error out
3058 if (block_start == EXTENT_MAP_INLINE) {
3060 unlock_extent(tree, cur, cur + iosize - 1);
3062 pg_offset += iosize;
3066 ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL,
3067 page, sector, disk_io_size, pg_offset,
3069 end_bio_extent_readpage, mirror_num,
3075 *bio_flags = this_bio_flag;
3078 unlock_extent(tree, cur, cur + iosize - 1);
3082 pg_offset += iosize;
3086 if (!PageError(page))
3087 SetPageUptodate(page);
3093 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
3094 struct page *pages[], int nr_pages,
3096 get_extent_t *get_extent,
3097 struct extent_map **em_cached,
3098 struct bio **bio, int mirror_num,
3099 unsigned long *bio_flags,
3102 struct inode *inode;
3103 struct btrfs_ordered_extent *ordered;
3106 inode = pages[0]->mapping->host;
3108 lock_extent(tree, start, end);
3109 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3113 unlock_extent(tree, start, end);
3114 btrfs_start_ordered_extent(inode, ordered, 1);
3115 btrfs_put_ordered_extent(ordered);
3118 for (index = 0; index < nr_pages; index++) {
3119 __do_readpage(tree, pages[index], get_extent, em_cached, bio,
3120 mirror_num, bio_flags, 0, prev_em_start);
3121 put_page(pages[index]);
3125 static void __extent_readpages(struct extent_io_tree *tree,
3126 struct page *pages[],
3127 int nr_pages, get_extent_t *get_extent,
3128 struct extent_map **em_cached,
3129 struct bio **bio, int mirror_num,
3130 unsigned long *bio_flags,
3137 int first_index = 0;
3139 for (index = 0; index < nr_pages; index++) {
3140 page_start = page_offset(pages[index]);
3143 end = start + PAGE_SIZE - 1;
3144 first_index = index;
3145 } else if (end + 1 == page_start) {
3148 __do_contiguous_readpages(tree, &pages[first_index],
3149 index - first_index, start,
3150 end, get_extent, em_cached,
3151 bio, mirror_num, bio_flags,
3154 end = start + PAGE_SIZE - 1;
3155 first_index = index;
3160 __do_contiguous_readpages(tree, &pages[first_index],
3161 index - first_index, start,
3162 end, get_extent, em_cached, bio,
3163 mirror_num, bio_flags,
3167 static int __extent_read_full_page(struct extent_io_tree *tree,
3169 get_extent_t *get_extent,
3170 struct bio **bio, int mirror_num,
3171 unsigned long *bio_flags,
3172 unsigned int read_flags)
3174 struct inode *inode = page->mapping->host;
3175 struct btrfs_ordered_extent *ordered;
3176 u64 start = page_offset(page);
3177 u64 end = start + PAGE_SIZE - 1;
3181 lock_extent(tree, start, end);
3182 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3186 unlock_extent(tree, start, end);
3187 btrfs_start_ordered_extent(inode, ordered, 1);
3188 btrfs_put_ordered_extent(ordered);
3191 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3192 bio_flags, read_flags, NULL);
3196 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3197 get_extent_t *get_extent, int mirror_num)
3199 struct bio *bio = NULL;
3200 unsigned long bio_flags = 0;
3203 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3206 ret = submit_one_bio(bio, mirror_num, bio_flags);
3210 static void update_nr_written(struct writeback_control *wbc,
3211 unsigned long nr_written)
3213 wbc->nr_to_write -= nr_written;
3217 * helper for __extent_writepage, doing all of the delayed allocation setup.
3219 * This returns 1 if our fill_delalloc function did all the work required
3220 * to write the page (copy into inline extent). In this case the IO has
3221 * been started and the page is already unlocked.
3223 * This returns 0 if all went well (page still locked)
3224 * This returns < 0 if there were errors (page still locked)
3226 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3227 struct page *page, struct writeback_control *wbc,
3228 struct extent_page_data *epd,
3230 unsigned long *nr_written)
3232 struct extent_io_tree *tree = epd->tree;
3233 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3235 u64 delalloc_to_write = 0;
3236 u64 delalloc_end = 0;
3238 int page_started = 0;
3240 if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc)
3243 while (delalloc_end < page_end) {
3244 nr_delalloc = find_lock_delalloc_range(inode, tree,
3248 BTRFS_MAX_EXTENT_SIZE);
3249 if (nr_delalloc == 0) {
3250 delalloc_start = delalloc_end + 1;
3253 ret = tree->ops->fill_delalloc(inode, page,
3258 /* File system has been set read-only */
3261 /* fill_delalloc should be return < 0 for error
3262 * but just in case, we use > 0 here meaning the
3263 * IO is started, so we don't want to return > 0
3264 * unless things are going well.
3266 ret = ret < 0 ? ret : -EIO;
3270 * delalloc_end is already one less than the total length, so
3271 * we don't subtract one from PAGE_SIZE
3273 delalloc_to_write += (delalloc_end - delalloc_start +
3274 PAGE_SIZE) >> PAGE_SHIFT;
3275 delalloc_start = delalloc_end + 1;
3277 if (wbc->nr_to_write < delalloc_to_write) {
3280 if (delalloc_to_write < thresh * 2)
3281 thresh = delalloc_to_write;
3282 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3286 /* did the fill delalloc function already unlock and start
3291 * we've unlocked the page, so we can't update
3292 * the mapping's writeback index, just update
3295 wbc->nr_to_write -= *nr_written;
3306 * helper for __extent_writepage. This calls the writepage start hooks,
3307 * and does the loop to map the page into extents and bios.
3309 * We return 1 if the IO is started and the page is unlocked,
3310 * 0 if all went well (page still locked)
3311 * < 0 if there were errors (page still locked)
3313 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3315 struct writeback_control *wbc,
3316 struct extent_page_data *epd,
3318 unsigned long nr_written,
3319 unsigned int write_flags, int *nr_ret)
3321 struct extent_io_tree *tree = epd->tree;
3322 u64 start = page_offset(page);
3323 u64 page_end = start + PAGE_SIZE - 1;
3330 struct extent_map *em;
3331 struct block_device *bdev;
3332 size_t pg_offset = 0;
3338 if (tree->ops && tree->ops->writepage_start_hook) {
3339 ret = tree->ops->writepage_start_hook(page, start,
3342 /* Fixup worker will requeue */
3344 wbc->pages_skipped++;
3346 redirty_page_for_writepage(wbc, page);
3348 update_nr_written(wbc, nr_written);
3355 * we don't want to touch the inode after unlocking the page,
3356 * so we update the mapping writeback index now
3358 update_nr_written(wbc, nr_written + 1);
3361 if (i_size <= start) {
3362 if (tree->ops && tree->ops->writepage_end_io_hook)
3363 tree->ops->writepage_end_io_hook(page, start,
3368 blocksize = inode->i_sb->s_blocksize;
3370 while (cur <= end) {
3373 if (cur >= i_size) {
3374 if (tree->ops && tree->ops->writepage_end_io_hook)
3375 tree->ops->writepage_end_io_hook(page, cur,
3379 em = epd->get_extent(BTRFS_I(inode), page, pg_offset, cur,
3381 if (IS_ERR_OR_NULL(em)) {
3383 ret = PTR_ERR_OR_ZERO(em);
3387 extent_offset = cur - em->start;
3388 em_end = extent_map_end(em);
3389 BUG_ON(em_end <= cur);
3391 iosize = min(em_end - cur, end - cur + 1);
3392 iosize = ALIGN(iosize, blocksize);
3393 sector = (em->block_start + extent_offset) >> 9;
3395 block_start = em->block_start;
3396 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3397 free_extent_map(em);
3401 * compressed and inline extents are written through other
3404 if (compressed || block_start == EXTENT_MAP_HOLE ||
3405 block_start == EXTENT_MAP_INLINE) {
3407 * end_io notification does not happen here for
3408 * compressed extents
3410 if (!compressed && tree->ops &&
3411 tree->ops->writepage_end_io_hook)
3412 tree->ops->writepage_end_io_hook(page, cur,
3415 else if (compressed) {
3416 /* we don't want to end_page_writeback on
3417 * a compressed extent. this happens
3424 pg_offset += iosize;
3428 set_range_writeback(tree, cur, cur + iosize - 1);
3429 if (!PageWriteback(page)) {
3430 btrfs_err(BTRFS_I(inode)->root->fs_info,
3431 "page %lu not writeback, cur %llu end %llu",
3432 page->index, cur, end);
3435 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3436 page, sector, iosize, pg_offset,
3438 end_bio_extent_writepage,
3442 if (PageWriteback(page))
3443 end_page_writeback(page);
3447 pg_offset += iosize;
3456 * the writepage semantics are similar to regular writepage. extent
3457 * records are inserted to lock ranges in the tree, and as dirty areas
3458 * are found, they are marked writeback. Then the lock bits are removed
3459 * and the end_io handler clears the writeback ranges
3461 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3464 struct inode *inode = page->mapping->host;
3465 struct extent_page_data *epd = data;
3466 u64 start = page_offset(page);
3467 u64 page_end = start + PAGE_SIZE - 1;
3470 size_t pg_offset = 0;
3471 loff_t i_size = i_size_read(inode);
3472 unsigned long end_index = i_size >> PAGE_SHIFT;
3473 unsigned int write_flags = 0;
3474 unsigned long nr_written = 0;
3476 write_flags = wbc_to_write_flags(wbc);
3478 trace___extent_writepage(page, inode, wbc);
3480 WARN_ON(!PageLocked(page));
3482 ClearPageError(page);
3484 pg_offset = i_size & (PAGE_SIZE - 1);
3485 if (page->index > end_index ||
3486 (page->index == end_index && !pg_offset)) {
3487 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3492 if (page->index == end_index) {
3495 userpage = kmap_atomic(page);
3496 memset(userpage + pg_offset, 0,
3497 PAGE_SIZE - pg_offset);
3498 kunmap_atomic(userpage);
3499 flush_dcache_page(page);
3504 set_page_extent_mapped(page);
3506 ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written);
3512 ret = __extent_writepage_io(inode, page, wbc, epd,
3513 i_size, nr_written, write_flags, &nr);
3519 /* make sure the mapping tag for page dirty gets cleared */
3520 set_page_writeback(page);
3521 end_page_writeback(page);
3523 if (PageError(page)) {
3524 ret = ret < 0 ? ret : -EIO;
3525 end_extent_writepage(page, ret, start, page_end);
3534 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3536 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3537 TASK_UNINTERRUPTIBLE);
3540 static noinline_for_stack int
3541 lock_extent_buffer_for_io(struct extent_buffer *eb,
3542 struct btrfs_fs_info *fs_info,
3543 struct extent_page_data *epd)
3545 unsigned long i, num_pages;
3549 if (!btrfs_try_tree_write_lock(eb)) {
3551 flush_write_bio(epd);
3552 btrfs_tree_lock(eb);
3555 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3556 btrfs_tree_unlock(eb);
3560 flush_write_bio(epd);
3564 wait_on_extent_buffer_writeback(eb);
3565 btrfs_tree_lock(eb);
3566 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3568 btrfs_tree_unlock(eb);
3573 * We need to do this to prevent races in people who check if the eb is
3574 * under IO since we can end up having no IO bits set for a short period
3577 spin_lock(&eb->refs_lock);
3578 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3579 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3580 spin_unlock(&eb->refs_lock);
3581 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3582 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3584 fs_info->dirty_metadata_batch);
3587 spin_unlock(&eb->refs_lock);
3590 btrfs_tree_unlock(eb);
3595 num_pages = num_extent_pages(eb->start, eb->len);
3596 for (i = 0; i < num_pages; i++) {
3597 struct page *p = eb->pages[i];
3599 if (!trylock_page(p)) {
3601 flush_write_bio(epd);
3611 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3613 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3614 smp_mb__after_atomic();
3615 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3618 static void set_btree_ioerr(struct page *page)
3620 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3623 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3627 * If writeback for a btree extent that doesn't belong to a log tree
3628 * failed, increment the counter transaction->eb_write_errors.
3629 * We do this because while the transaction is running and before it's
3630 * committing (when we call filemap_fdata[write|wait]_range against
3631 * the btree inode), we might have
3632 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3633 * returns an error or an error happens during writeback, when we're
3634 * committing the transaction we wouldn't know about it, since the pages
3635 * can be no longer dirty nor marked anymore for writeback (if a
3636 * subsequent modification to the extent buffer didn't happen before the
3637 * transaction commit), which makes filemap_fdata[write|wait]_range not
3638 * able to find the pages tagged with SetPageError at transaction
3639 * commit time. So if this happens we must abort the transaction,
3640 * otherwise we commit a super block with btree roots that point to
3641 * btree nodes/leafs whose content on disk is invalid - either garbage
3642 * or the content of some node/leaf from a past generation that got
3643 * cowed or deleted and is no longer valid.
3645 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3646 * not be enough - we need to distinguish between log tree extents vs
3647 * non-log tree extents, and the next filemap_fdatawait_range() call
3648 * will catch and clear such errors in the mapping - and that call might
3649 * be from a log sync and not from a transaction commit. Also, checking
3650 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3651 * not done and would not be reliable - the eb might have been released
3652 * from memory and reading it back again means that flag would not be
3653 * set (since it's a runtime flag, not persisted on disk).
3655 * Using the flags below in the btree inode also makes us achieve the
3656 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3657 * writeback for all dirty pages and before filemap_fdatawait_range()
3658 * is called, the writeback for all dirty pages had already finished
3659 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3660 * filemap_fdatawait_range() would return success, as it could not know
3661 * that writeback errors happened (the pages were no longer tagged for
3664 switch (eb->log_index) {
3666 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3669 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3672 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3675 BUG(); /* unexpected, logic error */
3679 static void end_bio_extent_buffer_writepage(struct bio *bio)
3681 struct bio_vec *bvec;
3682 struct extent_buffer *eb;
3685 ASSERT(!bio_flagged(bio, BIO_CLONED));
3686 bio_for_each_segment_all(bvec, bio, i) {
3687 struct page *page = bvec->bv_page;
3689 eb = (struct extent_buffer *)page->private;
3691 done = atomic_dec_and_test(&eb->io_pages);
3693 if (bio->bi_status ||
3694 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3695 ClearPageUptodate(page);
3696 set_btree_ioerr(page);
3699 end_page_writeback(page);
3704 end_extent_buffer_writeback(eb);
3710 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3711 struct btrfs_fs_info *fs_info,
3712 struct writeback_control *wbc,
3713 struct extent_page_data *epd)
3715 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3716 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3717 u64 offset = eb->start;
3719 unsigned long i, num_pages;
3720 unsigned long bio_flags = 0;
3721 unsigned long start, end;
3722 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
3725 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3726 num_pages = num_extent_pages(eb->start, eb->len);
3727 atomic_set(&eb->io_pages, num_pages);
3728 if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3729 bio_flags = EXTENT_BIO_TREE_LOG;
3731 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3732 nritems = btrfs_header_nritems(eb);
3733 if (btrfs_header_level(eb) > 0) {
3734 end = btrfs_node_key_ptr_offset(nritems);
3736 memzero_extent_buffer(eb, end, eb->len - end);
3740 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3742 start = btrfs_item_nr_offset(nritems);
3743 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, eb);
3744 memzero_extent_buffer(eb, start, end - start);
3747 for (i = 0; i < num_pages; i++) {
3748 struct page *p = eb->pages[i];
3750 clear_page_dirty_for_io(p);
3751 set_page_writeback(p);
3752 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3753 p, offset >> 9, PAGE_SIZE, 0, bdev,
3755 end_bio_extent_buffer_writepage,
3756 0, epd->bio_flags, bio_flags, false);
3757 epd->bio_flags = bio_flags;
3760 if (PageWriteback(p))
3761 end_page_writeback(p);
3762 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3763 end_extent_buffer_writeback(eb);
3767 offset += PAGE_SIZE;
3768 update_nr_written(wbc, 1);
3772 if (unlikely(ret)) {
3773 for (; i < num_pages; i++) {
3774 struct page *p = eb->pages[i];
3775 clear_page_dirty_for_io(p);
3783 int btree_write_cache_pages(struct address_space *mapping,
3784 struct writeback_control *wbc)
3786 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3787 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3788 struct extent_buffer *eb, *prev_eb = NULL;
3789 struct extent_page_data epd = {
3793 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3798 int nr_to_write_done = 0;
3799 struct pagevec pvec;
3802 pgoff_t end; /* Inclusive */
3806 pagevec_init(&pvec, 0);
3807 if (wbc->range_cyclic) {
3808 index = mapping->writeback_index; /* Start from prev offset */
3811 index = wbc->range_start >> PAGE_SHIFT;
3812 end = wbc->range_end >> PAGE_SHIFT;
3815 if (wbc->sync_mode == WB_SYNC_ALL)
3816 tag = PAGECACHE_TAG_TOWRITE;
3818 tag = PAGECACHE_TAG_DIRTY;
3820 if (wbc->sync_mode == WB_SYNC_ALL)
3821 tag_pages_for_writeback(mapping, index, end);
3822 while (!done && !nr_to_write_done && (index <= end) &&
3823 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3824 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3828 for (i = 0; i < nr_pages; i++) {
3829 struct page *page = pvec.pages[i];
3831 if (!PagePrivate(page))
3834 if (!wbc->range_cyclic && page->index > end) {
3839 spin_lock(&mapping->private_lock);
3840 if (!PagePrivate(page)) {
3841 spin_unlock(&mapping->private_lock);
3845 eb = (struct extent_buffer *)page->private;
3848 * Shouldn't happen and normally this would be a BUG_ON
3849 * but no sense in crashing the users box for something
3850 * we can survive anyway.
3853 spin_unlock(&mapping->private_lock);
3857 if (eb == prev_eb) {
3858 spin_unlock(&mapping->private_lock);
3862 ret = atomic_inc_not_zero(&eb->refs);
3863 spin_unlock(&mapping->private_lock);
3868 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3870 free_extent_buffer(eb);
3872 } else if (ret < 0) {
3874 free_extent_buffer(eb);
3878 ret = write_one_eb(eb, fs_info, wbc, &epd);
3881 free_extent_buffer(eb);
3884 free_extent_buffer(eb);
3887 * The filesystem may choose to bump up nr_to_write.
3888 * We have to make sure to honor the new nr_to_write
3891 nr_to_write_done = (wbc->sync_mode == WB_SYNC_NONE &&
3892 wbc->nr_to_write <= 0);
3894 pagevec_release(&pvec);
3897 if (!scanned && !done) {
3899 * We hit the last page and there is more work to be done: wrap
3900 * back to the start of the file
3906 flush_write_bio(&epd);
3911 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3912 * @mapping: address space structure to write
3913 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3914 * @writepage: function called for each page
3915 * @data: data passed to writepage function
3917 * If a page is already under I/O, write_cache_pages() skips it, even
3918 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3919 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3920 * and msync() need to guarantee that all the data which was dirty at the time
3921 * the call was made get new I/O started against them. If wbc->sync_mode is
3922 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3923 * existing IO to complete.
3925 static int extent_write_cache_pages(struct address_space *mapping,
3926 struct writeback_control *wbc,
3927 writepage_t writepage, void *data,
3928 void (*flush_fn)(void *))
3930 struct inode *inode = mapping->host;
3933 int nr_to_write_done = 0;
3934 struct pagevec pvec;
3937 pgoff_t end; /* Inclusive */
3939 int range_whole = 0;
3944 * We have to hold onto the inode so that ordered extents can do their
3945 * work when the IO finishes. The alternative to this is failing to add
3946 * an ordered extent if the igrab() fails there and that is a huge pain
3947 * to deal with, so instead just hold onto the inode throughout the
3948 * writepages operation. If it fails here we are freeing up the inode
3949 * anyway and we'd rather not waste our time writing out stuff that is
3950 * going to be truncated anyway.
3955 pagevec_init(&pvec, 0);
3956 if (wbc->range_cyclic) {
3957 index = mapping->writeback_index; /* Start from prev offset */
3960 index = wbc->range_start >> PAGE_SHIFT;
3961 end = wbc->range_end >> PAGE_SHIFT;
3962 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3966 if (wbc->sync_mode == WB_SYNC_ALL)
3967 tag = PAGECACHE_TAG_TOWRITE;
3969 tag = PAGECACHE_TAG_DIRTY;
3971 if (wbc->sync_mode == WB_SYNC_ALL)
3972 tag_pages_for_writeback(mapping, index, end);
3974 while (!done && !nr_to_write_done && (index <= end) &&
3975 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3976 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3980 for (i = 0; i < nr_pages; i++) {
3981 struct page *page = pvec.pages[i];
3983 done_index = page->index;
3985 * At this point we hold neither mapping->tree_lock nor
3986 * lock on the page itself: the page may be truncated or
3987 * invalidated (changing page->mapping to NULL), or even
3988 * swizzled back from swapper_space to tmpfs file
3991 if (!trylock_page(page)) {
3996 if (unlikely(page->mapping != mapping)) {
4001 if (!wbc->range_cyclic && page->index > end) {
4007 if (wbc->sync_mode != WB_SYNC_NONE) {
4008 if (PageWriteback(page))
4010 wait_on_page_writeback(page);
4013 if (PageWriteback(page) ||
4014 !clear_page_dirty_for_io(page)) {
4019 ret = (*writepage)(page, wbc, data);
4021 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
4027 * done_index is set past this page,
4028 * so media errors will not choke
4029 * background writeout for the entire
4030 * file. This has consequences for
4031 * range_cyclic semantics (ie. it may
4032 * not be suitable for data integrity
4035 done_index = page->index + 1;
4041 * the filesystem may choose to bump up nr_to_write.
4042 * We have to make sure to honor the new nr_to_write
4045 nr_to_write_done = wbc->nr_to_write <= 0;
4047 pagevec_release(&pvec);
4050 if (!scanned && !done) {
4052 * We hit the last page and there is more work to be done: wrap
4053 * back to the start of the file
4059 * If we're looping we could run into a page that is locked by a
4060 * writer and that writer could be waiting on writeback for a
4061 * page in our current bio, and thus deadlock, so flush the
4064 flush_write_bio(data);
4068 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4069 mapping->writeback_index = done_index;
4071 btrfs_add_delayed_iput(inode);
4075 static void flush_epd_write_bio(struct extent_page_data *epd)
4080 ret = submit_one_bio(epd->bio, 0, epd->bio_flags);
4081 BUG_ON(ret < 0); /* -ENOMEM */
4086 static noinline void flush_write_bio(void *data)
4088 struct extent_page_data *epd = data;
4089 flush_epd_write_bio(epd);
4092 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
4093 get_extent_t *get_extent,
4094 struct writeback_control *wbc)
4097 struct extent_page_data epd = {
4100 .get_extent = get_extent,
4102 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4106 ret = __extent_writepage(page, wbc, &epd);
4108 flush_epd_write_bio(&epd);
4112 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
4113 u64 start, u64 end, get_extent_t *get_extent,
4117 struct address_space *mapping = inode->i_mapping;
4119 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4122 struct extent_page_data epd = {
4125 .get_extent = get_extent,
4127 .sync_io = mode == WB_SYNC_ALL,
4130 struct writeback_control wbc_writepages = {
4132 .nr_to_write = nr_pages * 2,
4133 .range_start = start,
4134 .range_end = end + 1,
4137 while (start <= end) {
4138 page = find_get_page(mapping, start >> PAGE_SHIFT);
4139 if (clear_page_dirty_for_io(page))
4140 ret = __extent_writepage(page, &wbc_writepages, &epd);
4142 if (tree->ops && tree->ops->writepage_end_io_hook)
4143 tree->ops->writepage_end_io_hook(page, start,
4144 start + PAGE_SIZE - 1,
4152 flush_epd_write_bio(&epd);
4156 int extent_writepages(struct extent_io_tree *tree,
4157 struct address_space *mapping,
4158 get_extent_t *get_extent,
4159 struct writeback_control *wbc)
4162 struct extent_page_data epd = {
4165 .get_extent = get_extent,
4167 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4171 ret = extent_write_cache_pages(mapping, wbc, __extent_writepage, &epd,
4173 flush_epd_write_bio(&epd);
4177 int extent_readpages(struct extent_io_tree *tree,
4178 struct address_space *mapping,
4179 struct list_head *pages, unsigned nr_pages,
4180 get_extent_t get_extent)
4182 struct bio *bio = NULL;
4184 unsigned long bio_flags = 0;
4185 struct page *pagepool[16];
4187 struct extent_map *em_cached = NULL;
4189 u64 prev_em_start = (u64)-1;
4191 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
4192 page = list_entry(pages->prev, struct page, lru);
4194 prefetchw(&page->flags);
4195 list_del(&page->lru);
4196 if (add_to_page_cache_lru(page, mapping,
4198 readahead_gfp_mask(mapping))) {
4203 pagepool[nr++] = page;
4204 if (nr < ARRAY_SIZE(pagepool))
4206 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4207 &bio, 0, &bio_flags, &prev_em_start);
4211 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4212 &bio, 0, &bio_flags, &prev_em_start);
4215 free_extent_map(em_cached);
4217 BUG_ON(!list_empty(pages));
4219 return submit_one_bio(bio, 0, bio_flags);
4224 * basic invalidatepage code, this waits on any locked or writeback
4225 * ranges corresponding to the page, and then deletes any extent state
4226 * records from the tree
4228 int extent_invalidatepage(struct extent_io_tree *tree,
4229 struct page *page, unsigned long offset)
4231 struct extent_state *cached_state = NULL;
4232 u64 start = page_offset(page);
4233 u64 end = start + PAGE_SIZE - 1;
4234 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4236 start += ALIGN(offset, blocksize);
4240 lock_extent_bits(tree, start, end, &cached_state);
4241 wait_on_page_writeback(page);
4242 clear_extent_bit(tree, start, end,
4243 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4244 EXTENT_DO_ACCOUNTING,
4245 1, 1, &cached_state, GFP_NOFS);
4250 * a helper for releasepage, this tests for areas of the page that
4251 * are locked or under IO and drops the related state bits if it is safe
4254 static int try_release_extent_state(struct extent_map_tree *map,
4255 struct extent_io_tree *tree,
4256 struct page *page, gfp_t mask)
4258 u64 start = page_offset(page);
4259 u64 end = start + PAGE_SIZE - 1;
4262 if (test_range_bit(tree, start, end,
4263 EXTENT_IOBITS, 0, NULL))
4267 * at this point we can safely clear everything except the
4268 * locked bit and the nodatasum bit
4270 ret = clear_extent_bit(tree, start, end,
4271 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4274 /* if clear_extent_bit failed for enomem reasons,
4275 * we can't allow the release to continue.
4286 * a helper for releasepage. As long as there are no locked extents
4287 * in the range corresponding to the page, both state records and extent
4288 * map records are removed
4290 int try_release_extent_mapping(struct extent_map_tree *map,
4291 struct extent_io_tree *tree, struct page *page,
4294 struct extent_map *em;
4295 u64 start = page_offset(page);
4296 u64 end = start + PAGE_SIZE - 1;
4297 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
4299 if (gfpflags_allow_blocking(mask) &&
4300 page->mapping->host->i_size > SZ_16M) {
4302 while (start <= end) {
4303 len = end - start + 1;
4304 write_lock(&map->lock);
4305 em = lookup_extent_mapping(map, start, len);
4307 write_unlock(&map->lock);
4310 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4311 em->start != start) {
4312 write_unlock(&map->lock);
4313 free_extent_map(em);
4316 if (!test_range_bit(tree, em->start,
4317 extent_map_end(em) - 1,
4318 EXTENT_LOCKED | EXTENT_WRITEBACK,
4320 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4321 &btrfs_inode->runtime_flags);
4322 remove_extent_mapping(map, em);
4323 /* once for the rb tree */
4324 free_extent_map(em);
4326 start = extent_map_end(em);
4327 write_unlock(&map->lock);
4330 free_extent_map(em);
4332 cond_resched(); /* Allow large-extent preemption. */
4335 return try_release_extent_state(map, tree, page, mask);
4339 * helper function for fiemap, which doesn't want to see any holes.
4340 * This maps until we find something past 'last'
4342 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4345 get_extent_t *get_extent)
4347 u64 sectorsize = btrfs_inode_sectorsize(inode);
4348 struct extent_map *em;
4355 len = last - offset;
4358 len = ALIGN(len, sectorsize);
4359 em = get_extent(BTRFS_I(inode), NULL, 0, offset, len, 0);
4360 if (IS_ERR_OR_NULL(em))
4363 /* if this isn't a hole return it */
4364 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4365 em->block_start != EXTENT_MAP_HOLE) {
4369 /* this is a hole, advance to the next extent */
4370 offset = extent_map_end(em);
4371 free_extent_map(em);
4379 * To cache previous fiemap extent
4381 * Will be used for merging fiemap extent
4383 struct fiemap_cache {
4392 * Helper to submit fiemap extent.
4394 * Will try to merge current fiemap extent specified by @offset, @phys,
4395 * @len and @flags with cached one.
4396 * And only when we fails to merge, cached one will be submitted as
4399 * Return value is the same as fiemap_fill_next_extent().
4401 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4402 struct fiemap_cache *cache,
4403 u64 offset, u64 phys, u64 len, u32 flags)
4411 * Sanity check, extent_fiemap() should have ensured that new
4412 * fiemap extent won't overlap with cahced one.
4415 * NOTE: Physical address can overlap, due to compression
4417 if (cache->offset + cache->len > offset) {
4423 * Only merges fiemap extents if
4424 * 1) Their logical addresses are continuous
4426 * 2) Their physical addresses are continuous
4427 * So truly compressed (physical size smaller than logical size)
4428 * extents won't get merged with each other
4430 * 3) Share same flags except FIEMAP_EXTENT_LAST
4431 * So regular extent won't get merged with prealloc extent
4433 if (cache->offset + cache->len == offset &&
4434 cache->phys + cache->len == phys &&
4435 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4436 (flags & ~FIEMAP_EXTENT_LAST)) {
4438 cache->flags |= flags;
4439 goto try_submit_last;
4442 /* Not mergeable, need to submit cached one */
4443 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4444 cache->len, cache->flags);
4445 cache->cached = false;
4449 cache->cached = true;
4450 cache->offset = offset;
4453 cache->flags = flags;
4455 if (cache->flags & FIEMAP_EXTENT_LAST) {
4456 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4457 cache->phys, cache->len, cache->flags);
4458 cache->cached = false;
4464 * Emit last fiemap cache
4466 * The last fiemap cache may still be cached in the following case:
4468 * |<- Fiemap range ->|
4469 * |<------------ First extent ----------->|
4471 * In this case, the first extent range will be cached but not emitted.
4472 * So we must emit it before ending extent_fiemap().
4474 static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info,
4475 struct fiemap_extent_info *fieinfo,
4476 struct fiemap_cache *cache)
4483 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4484 cache->len, cache->flags);
4485 cache->cached = false;
4491 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4492 __u64 start, __u64 len, get_extent_t *get_extent)
4496 u64 max = start + len;
4500 u64 last_for_get_extent = 0;
4502 u64 isize = i_size_read(inode);
4503 struct btrfs_key found_key;
4504 struct extent_map *em = NULL;
4505 struct extent_state *cached_state = NULL;
4506 struct btrfs_path *path;
4507 struct btrfs_root *root = BTRFS_I(inode)->root;
4508 struct fiemap_cache cache = { 0 };
4517 path = btrfs_alloc_path();
4520 path->leave_spinning = 1;
4522 start = round_down(start, btrfs_inode_sectorsize(inode));
4523 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
4526 * lookup the last file extent. We're not using i_size here
4527 * because there might be preallocation past i_size
4529 ret = btrfs_lookup_file_extent(NULL, root, path,
4530 btrfs_ino(BTRFS_I(inode)), -1, 0);
4532 btrfs_free_path(path);
4541 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4542 found_type = found_key.type;
4544 /* No extents, but there might be delalloc bits */
4545 if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) ||
4546 found_type != BTRFS_EXTENT_DATA_KEY) {
4547 /* have to trust i_size as the end */
4549 last_for_get_extent = isize;
4552 * remember the start of the last extent. There are a
4553 * bunch of different factors that go into the length of the
4554 * extent, so its much less complex to remember where it started
4556 last = found_key.offset;
4557 last_for_get_extent = last + 1;
4559 btrfs_release_path(path);
4562 * we might have some extents allocated but more delalloc past those
4563 * extents. so, we trust isize unless the start of the last extent is
4568 last_for_get_extent = isize;
4571 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4574 em = get_extent_skip_holes(inode, start, last_for_get_extent,
4584 u64 offset_in_extent = 0;
4586 /* break if the extent we found is outside the range */
4587 if (em->start >= max || extent_map_end(em) < off)
4591 * get_extent may return an extent that starts before our
4592 * requested range. We have to make sure the ranges
4593 * we return to fiemap always move forward and don't
4594 * overlap, so adjust the offsets here
4596 em_start = max(em->start, off);
4599 * record the offset from the start of the extent
4600 * for adjusting the disk offset below. Only do this if the
4601 * extent isn't compressed since our in ram offset may be past
4602 * what we have actually allocated on disk.
4604 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4605 offset_in_extent = em_start - em->start;
4606 em_end = extent_map_end(em);
4607 em_len = em_end - em_start;
4612 * bump off for our next call to get_extent
4614 off = extent_map_end(em);
4618 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4620 flags |= FIEMAP_EXTENT_LAST;
4621 } else if (em->block_start == EXTENT_MAP_INLINE) {
4622 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4623 FIEMAP_EXTENT_NOT_ALIGNED);
4624 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4625 flags |= (FIEMAP_EXTENT_DELALLOC |
4626 FIEMAP_EXTENT_UNKNOWN);
4627 } else if (fieinfo->fi_extents_max) {
4628 u64 bytenr = em->block_start -
4629 (em->start - em->orig_start);
4631 disko = em->block_start + offset_in_extent;
4634 * As btrfs supports shared space, this information
4635 * can be exported to userspace tools via
4636 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4637 * then we're just getting a count and we can skip the
4640 ret = btrfs_check_shared(root,
4641 btrfs_ino(BTRFS_I(inode)),
4646 flags |= FIEMAP_EXTENT_SHARED;
4649 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4650 flags |= FIEMAP_EXTENT_ENCODED;
4651 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4652 flags |= FIEMAP_EXTENT_UNWRITTEN;
4654 free_extent_map(em);
4656 if ((em_start >= last) || em_len == (u64)-1 ||
4657 (last == (u64)-1 && isize <= em_end)) {
4658 flags |= FIEMAP_EXTENT_LAST;
4662 /* now scan forward to see if this is really the last extent. */
4663 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4670 flags |= FIEMAP_EXTENT_LAST;
4673 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4683 ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache);
4684 free_extent_map(em);
4686 btrfs_free_path(path);
4687 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4688 &cached_state, GFP_NOFS);
4692 static void __free_extent_buffer(struct extent_buffer *eb)
4694 btrfs_leak_debug_del(&eb->leak_list);
4695 kmem_cache_free(extent_buffer_cache, eb);
4698 int extent_buffer_under_io(struct extent_buffer *eb)
4700 return (atomic_read(&eb->io_pages) ||
4701 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4702 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4706 * Helper for releasing extent buffer page.
4708 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb)
4710 unsigned long index;
4712 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4714 BUG_ON(extent_buffer_under_io(eb));
4716 index = num_extent_pages(eb->start, eb->len);
4722 page = eb->pages[index];
4726 spin_lock(&page->mapping->private_lock);
4728 * We do this since we'll remove the pages after we've
4729 * removed the eb from the radix tree, so we could race
4730 * and have this page now attached to the new eb. So
4731 * only clear page_private if it's still connected to
4734 if (PagePrivate(page) &&
4735 page->private == (unsigned long)eb) {
4736 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4737 BUG_ON(PageDirty(page));
4738 BUG_ON(PageWriteback(page));
4740 * We need to make sure we haven't be attached
4743 ClearPagePrivate(page);
4744 set_page_private(page, 0);
4745 /* One for the page private */
4750 spin_unlock(&page->mapping->private_lock);
4752 /* One for when we allocated the page */
4754 } while (index != 0);
4758 * Helper for releasing the extent buffer.
4760 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4762 btrfs_release_extent_buffer_page(eb);
4763 __free_extent_buffer(eb);
4766 static struct extent_buffer *
4767 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4770 struct extent_buffer *eb = NULL;
4772 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4775 eb->fs_info = fs_info;
4777 rwlock_init(&eb->lock);
4778 atomic_set(&eb->write_locks, 0);
4779 atomic_set(&eb->read_locks, 0);
4780 atomic_set(&eb->blocking_readers, 0);
4781 atomic_set(&eb->blocking_writers, 0);
4782 atomic_set(&eb->spinning_readers, 0);
4783 atomic_set(&eb->spinning_writers, 0);
4784 eb->lock_nested = 0;
4785 init_waitqueue_head(&eb->write_lock_wq);
4786 init_waitqueue_head(&eb->read_lock_wq);
4788 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4790 spin_lock_init(&eb->refs_lock);
4791 atomic_set(&eb->refs, 1);
4792 atomic_set(&eb->io_pages, 0);
4795 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4797 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4798 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4799 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4804 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4808 struct extent_buffer *new;
4809 unsigned long num_pages = num_extent_pages(src->start, src->len);
4811 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4815 for (i = 0; i < num_pages; i++) {
4816 p = alloc_page(GFP_NOFS);
4818 btrfs_release_extent_buffer(new);
4821 attach_extent_buffer_page(new, p);
4822 WARN_ON(PageDirty(p));
4825 copy_page(page_address(p), page_address(src->pages[i]));
4828 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4829 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4834 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4835 u64 start, unsigned long len)
4837 struct extent_buffer *eb;
4838 unsigned long num_pages;
4841 num_pages = num_extent_pages(start, len);
4843 eb = __alloc_extent_buffer(fs_info, start, len);
4847 for (i = 0; i < num_pages; i++) {
4848 eb->pages[i] = alloc_page(GFP_NOFS);
4852 set_extent_buffer_uptodate(eb);
4853 btrfs_set_header_nritems(eb, 0);
4854 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4859 __free_page(eb->pages[i - 1]);
4860 __free_extent_buffer(eb);
4864 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4867 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
4870 static void check_buffer_tree_ref(struct extent_buffer *eb)
4874 * The TREE_REF bit is first set when the extent_buffer is added
4875 * to the radix tree. It is also reset, if unset, when a new reference
4876 * is created by find_extent_buffer.
4878 * It is only cleared in two cases: freeing the last non-tree
4879 * reference to the extent_buffer when its STALE bit is set or
4880 * calling releasepage when the tree reference is the only reference.
4882 * In both cases, care is taken to ensure that the extent_buffer's
4883 * pages are not under io. However, releasepage can be concurrently
4884 * called with creating new references, which is prone to race
4885 * conditions between the calls to check_buffer_tree_ref in those
4886 * codepaths and clearing TREE_REF in try_release_extent_buffer.
4888 * The actual lifetime of the extent_buffer in the radix tree is
4889 * adequately protected by the refcount, but the TREE_REF bit and
4890 * its corresponding reference are not. To protect against this
4891 * class of races, we call check_buffer_tree_ref from the codepaths
4892 * which trigger io after they set eb->io_pages. Note that once io is
4893 * initiated, TREE_REF can no longer be cleared, so that is the
4894 * moment at which any such race is best fixed.
4896 refs = atomic_read(&eb->refs);
4897 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4900 spin_lock(&eb->refs_lock);
4901 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4902 atomic_inc(&eb->refs);
4903 spin_unlock(&eb->refs_lock);
4906 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4907 struct page *accessed)
4909 unsigned long num_pages, i;
4911 check_buffer_tree_ref(eb);
4913 num_pages = num_extent_pages(eb->start, eb->len);
4914 for (i = 0; i < num_pages; i++) {
4915 struct page *p = eb->pages[i];
4918 mark_page_accessed(p);
4922 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4925 struct extent_buffer *eb;
4928 eb = radix_tree_lookup(&fs_info->buffer_radix,
4929 start >> PAGE_SHIFT);
4930 if (eb && atomic_inc_not_zero(&eb->refs)) {
4933 * Lock our eb's refs_lock to avoid races with
4934 * free_extent_buffer. When we get our eb it might be flagged
4935 * with EXTENT_BUFFER_STALE and another task running
4936 * free_extent_buffer might have seen that flag set,
4937 * eb->refs == 2, that the buffer isn't under IO (dirty and
4938 * writeback flags not set) and it's still in the tree (flag
4939 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4940 * of decrementing the extent buffer's reference count twice.
4941 * So here we could race and increment the eb's reference count,
4942 * clear its stale flag, mark it as dirty and drop our reference
4943 * before the other task finishes executing free_extent_buffer,
4944 * which would later result in an attempt to free an extent
4945 * buffer that is dirty.
4947 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
4948 spin_lock(&eb->refs_lock);
4949 spin_unlock(&eb->refs_lock);
4951 mark_extent_buffer_accessed(eb, NULL);
4959 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4960 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4963 struct extent_buffer *eb, *exists = NULL;
4966 eb = find_extent_buffer(fs_info, start);
4969 eb = alloc_dummy_extent_buffer(fs_info, start);
4971 return ERR_PTR(-ENOMEM);
4972 eb->fs_info = fs_info;
4974 ret = radix_tree_preload(GFP_NOFS);
4976 exists = ERR_PTR(ret);
4979 spin_lock(&fs_info->buffer_lock);
4980 ret = radix_tree_insert(&fs_info->buffer_radix,
4981 start >> PAGE_SHIFT, eb);
4982 spin_unlock(&fs_info->buffer_lock);
4983 radix_tree_preload_end();
4984 if (ret == -EEXIST) {
4985 exists = find_extent_buffer(fs_info, start);
4991 check_buffer_tree_ref(eb);
4992 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4995 * We will free dummy extent buffer's if they come into
4996 * free_extent_buffer with a ref count of 2, but if we are using this we
4997 * want the buffers to stay in memory until we're done with them, so
4998 * bump the ref count again.
5000 atomic_inc(&eb->refs);
5003 btrfs_release_extent_buffer(eb);
5008 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
5011 unsigned long len = fs_info->nodesize;
5012 unsigned long num_pages = num_extent_pages(start, len);
5014 unsigned long index = start >> PAGE_SHIFT;
5015 struct extent_buffer *eb;
5016 struct extent_buffer *exists = NULL;
5018 struct address_space *mapping = fs_info->btree_inode->i_mapping;
5022 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
5023 btrfs_err(fs_info, "bad tree block start %llu", start);
5024 return ERR_PTR(-EINVAL);
5027 eb = find_extent_buffer(fs_info, start);
5031 eb = __alloc_extent_buffer(fs_info, start, len);
5033 return ERR_PTR(-ENOMEM);
5035 for (i = 0; i < num_pages; i++, index++) {
5036 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
5038 exists = ERR_PTR(-ENOMEM);
5042 spin_lock(&mapping->private_lock);
5043 if (PagePrivate(p)) {
5045 * We could have already allocated an eb for this page
5046 * and attached one so lets see if we can get a ref on
5047 * the existing eb, and if we can we know it's good and
5048 * we can just return that one, else we know we can just
5049 * overwrite page->private.
5051 exists = (struct extent_buffer *)p->private;
5052 if (atomic_inc_not_zero(&exists->refs)) {
5053 spin_unlock(&mapping->private_lock);
5056 mark_extent_buffer_accessed(exists, p);
5062 * Do this so attach doesn't complain and we need to
5063 * drop the ref the old guy had.
5065 ClearPagePrivate(p);
5066 WARN_ON(PageDirty(p));
5069 attach_extent_buffer_page(eb, p);
5070 spin_unlock(&mapping->private_lock);
5071 WARN_ON(PageDirty(p));
5073 if (!PageUptodate(p))
5077 * see below about how we avoid a nasty race with release page
5078 * and why we unlock later
5082 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5084 ret = radix_tree_preload(GFP_NOFS);
5086 exists = ERR_PTR(ret);
5090 spin_lock(&fs_info->buffer_lock);
5091 ret = radix_tree_insert(&fs_info->buffer_radix,
5092 start >> PAGE_SHIFT, eb);
5093 spin_unlock(&fs_info->buffer_lock);
5094 radix_tree_preload_end();
5095 if (ret == -EEXIST) {
5096 exists = find_extent_buffer(fs_info, start);
5102 /* add one reference for the tree */
5103 check_buffer_tree_ref(eb);
5104 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5107 * there is a race where release page may have
5108 * tried to find this extent buffer in the radix
5109 * but failed. It will tell the VM it is safe to
5110 * reclaim the, and it will clear the page private bit.
5111 * We must make sure to set the page private bit properly
5112 * after the extent buffer is in the radix tree so
5113 * it doesn't get lost
5115 SetPageChecked(eb->pages[0]);
5116 for (i = 1; i < num_pages; i++) {
5118 ClearPageChecked(p);
5121 unlock_page(eb->pages[0]);
5125 WARN_ON(!atomic_dec_and_test(&eb->refs));
5126 for (i = 0; i < num_pages; i++) {
5128 unlock_page(eb->pages[i]);
5131 btrfs_release_extent_buffer(eb);
5135 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5137 struct extent_buffer *eb =
5138 container_of(head, struct extent_buffer, rcu_head);
5140 __free_extent_buffer(eb);
5143 /* Expects to have eb->eb_lock already held */
5144 static int release_extent_buffer(struct extent_buffer *eb)
5146 WARN_ON(atomic_read(&eb->refs) == 0);
5147 if (atomic_dec_and_test(&eb->refs)) {
5148 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5149 struct btrfs_fs_info *fs_info = eb->fs_info;
5151 spin_unlock(&eb->refs_lock);
5153 spin_lock(&fs_info->buffer_lock);
5154 radix_tree_delete(&fs_info->buffer_radix,
5155 eb->start >> PAGE_SHIFT);
5156 spin_unlock(&fs_info->buffer_lock);
5158 spin_unlock(&eb->refs_lock);
5161 /* Should be safe to release our pages at this point */
5162 btrfs_release_extent_buffer_page(eb);
5163 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5164 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) {
5165 __free_extent_buffer(eb);
5169 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5172 spin_unlock(&eb->refs_lock);
5177 void free_extent_buffer(struct extent_buffer *eb)
5185 refs = atomic_read(&eb->refs);
5188 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5193 spin_lock(&eb->refs_lock);
5194 if (atomic_read(&eb->refs) == 2 &&
5195 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
5196 atomic_dec(&eb->refs);
5198 if (atomic_read(&eb->refs) == 2 &&
5199 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5200 !extent_buffer_under_io(eb) &&
5201 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5202 atomic_dec(&eb->refs);
5205 * I know this is terrible, but it's temporary until we stop tracking
5206 * the uptodate bits and such for the extent buffers.
5208 release_extent_buffer(eb);
5211 void free_extent_buffer_stale(struct extent_buffer *eb)
5216 spin_lock(&eb->refs_lock);
5217 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5219 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5220 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5221 atomic_dec(&eb->refs);
5222 release_extent_buffer(eb);
5225 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5228 unsigned long num_pages;
5231 num_pages = num_extent_pages(eb->start, eb->len);
5233 for (i = 0; i < num_pages; i++) {
5234 page = eb->pages[i];
5235 if (!PageDirty(page))
5239 WARN_ON(!PagePrivate(page));
5241 clear_page_dirty_for_io(page);
5242 spin_lock_irq(&page->mapping->tree_lock);
5243 if (!PageDirty(page)) {
5244 radix_tree_tag_clear(&page->mapping->page_tree,
5246 PAGECACHE_TAG_DIRTY);
5248 spin_unlock_irq(&page->mapping->tree_lock);
5249 ClearPageError(page);
5252 WARN_ON(atomic_read(&eb->refs) == 0);
5255 int set_extent_buffer_dirty(struct extent_buffer *eb)
5258 unsigned long num_pages;
5261 check_buffer_tree_ref(eb);
5263 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5265 num_pages = num_extent_pages(eb->start, eb->len);
5266 WARN_ON(atomic_read(&eb->refs) == 0);
5267 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5269 for (i = 0; i < num_pages; i++)
5270 set_page_dirty(eb->pages[i]);
5274 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5278 unsigned long num_pages;
5280 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5281 num_pages = num_extent_pages(eb->start, eb->len);
5282 for (i = 0; i < num_pages; i++) {
5283 page = eb->pages[i];
5285 ClearPageUptodate(page);
5289 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5293 unsigned long num_pages;
5295 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5296 num_pages = num_extent_pages(eb->start, eb->len);
5297 for (i = 0; i < num_pages; i++) {
5298 page = eb->pages[i];
5299 SetPageUptodate(page);
5303 int extent_buffer_uptodate(struct extent_buffer *eb)
5305 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5308 int read_extent_buffer_pages(struct extent_io_tree *tree,
5309 struct extent_buffer *eb, int wait,
5310 get_extent_t *get_extent, int mirror_num)
5316 int locked_pages = 0;
5317 int all_uptodate = 1;
5318 unsigned long num_pages;
5319 unsigned long num_reads = 0;
5320 struct bio *bio = NULL;
5321 unsigned long bio_flags = 0;
5323 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5326 num_pages = num_extent_pages(eb->start, eb->len);
5327 for (i = 0; i < num_pages; i++) {
5328 page = eb->pages[i];
5329 if (wait == WAIT_NONE) {
5330 if (!trylock_page(page))
5338 * We need to firstly lock all pages to make sure that
5339 * the uptodate bit of our pages won't be affected by
5340 * clear_extent_buffer_uptodate().
5342 for (i = 0; i < num_pages; i++) {
5343 page = eb->pages[i];
5344 if (!PageUptodate(page)) {
5351 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5355 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5356 eb->read_mirror = 0;
5357 atomic_set(&eb->io_pages, num_reads);
5359 * It is possible for releasepage to clear the TREE_REF bit before we
5360 * set io_pages. See check_buffer_tree_ref for a more detailed comment.
5362 check_buffer_tree_ref(eb);
5363 for (i = 0; i < num_pages; i++) {
5364 page = eb->pages[i];
5366 if (!PageUptodate(page)) {
5368 atomic_dec(&eb->io_pages);
5373 ClearPageError(page);
5374 err = __extent_read_full_page(tree, page,
5376 mirror_num, &bio_flags,
5381 * We use &bio in above __extent_read_full_page,
5382 * so we ensure that if it returns error, the
5383 * current page fails to add itself to bio and
5384 * it's been unlocked.
5386 * We must dec io_pages by ourselves.
5388 atomic_dec(&eb->io_pages);
5396 err = submit_one_bio(bio, mirror_num, bio_flags);
5401 if (ret || wait != WAIT_COMPLETE)
5404 for (i = 0; i < num_pages; i++) {
5405 page = eb->pages[i];
5406 wait_on_page_locked(page);
5407 if (!PageUptodate(page))
5414 while (locked_pages > 0) {
5416 page = eb->pages[locked_pages];
5422 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5423 unsigned long start, unsigned long len)
5429 char *dst = (char *)dstv;
5430 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5431 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5433 if (start + len > eb->len) {
5434 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5435 eb->start, eb->len, start, len);
5436 memset(dst, 0, len);
5440 offset = (start_offset + start) & (PAGE_SIZE - 1);
5443 page = eb->pages[i];
5445 cur = min(len, (PAGE_SIZE - offset));
5446 kaddr = page_address(page);
5447 memcpy(dst, kaddr + offset, cur);
5456 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
5458 unsigned long start, unsigned long len)
5464 char __user *dst = (char __user *)dstv;
5465 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5466 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5469 WARN_ON(start > eb->len);
5470 WARN_ON(start + len > eb->start + eb->len);
5472 offset = (start_offset + start) & (PAGE_SIZE - 1);
5475 page = eb->pages[i];
5477 cur = min(len, (PAGE_SIZE - offset));
5478 kaddr = page_address(page);
5479 if (probe_user_write(dst, kaddr + offset, cur)) {
5494 * return 0 if the item is found within a page.
5495 * return 1 if the item spans two pages.
5496 * return -EINVAL otherwise.
5498 int map_private_extent_buffer(const struct extent_buffer *eb,
5499 unsigned long start, unsigned long min_len,
5500 char **map, unsigned long *map_start,
5501 unsigned long *map_len)
5503 size_t offset = start & (PAGE_SIZE - 1);
5506 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5507 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5508 unsigned long end_i = (start_offset + start + min_len - 1) >>
5511 if (start + min_len > eb->len) {
5512 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5513 eb->start, eb->len, start, min_len);
5521 offset = start_offset;
5525 *map_start = ((u64)i << PAGE_SHIFT) - start_offset;
5529 kaddr = page_address(p);
5530 *map = kaddr + offset;
5531 *map_len = PAGE_SIZE - offset;
5535 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
5536 unsigned long start, unsigned long len)
5542 char *ptr = (char *)ptrv;
5543 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5544 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5547 WARN_ON(start > eb->len);
5548 WARN_ON(start + len > eb->start + eb->len);
5550 offset = (start_offset + start) & (PAGE_SIZE - 1);
5553 page = eb->pages[i];
5555 cur = min(len, (PAGE_SIZE - offset));
5557 kaddr = page_address(page);
5558 ret = memcmp(ptr, kaddr + offset, cur);
5570 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb,
5575 WARN_ON(!PageUptodate(eb->pages[0]));
5576 kaddr = page_address(eb->pages[0]);
5577 memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv,
5581 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv)
5585 WARN_ON(!PageUptodate(eb->pages[0]));
5586 kaddr = page_address(eb->pages[0]);
5587 memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv,
5591 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5592 unsigned long start, unsigned long len)
5598 char *src = (char *)srcv;
5599 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5600 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5602 WARN_ON(start > eb->len);
5603 WARN_ON(start + len > eb->start + eb->len);
5605 offset = (start_offset + start) & (PAGE_SIZE - 1);
5608 page = eb->pages[i];
5609 WARN_ON(!PageUptodate(page));
5611 cur = min(len, PAGE_SIZE - offset);
5612 kaddr = page_address(page);
5613 memcpy(kaddr + offset, src, cur);
5622 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start,
5629 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5630 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5632 WARN_ON(start > eb->len);
5633 WARN_ON(start + len > eb->start + eb->len);
5635 offset = (start_offset + start) & (PAGE_SIZE - 1);
5638 page = eb->pages[i];
5639 WARN_ON(!PageUptodate(page));
5641 cur = min(len, PAGE_SIZE - offset);
5642 kaddr = page_address(page);
5643 memset(kaddr + offset, 0, cur);
5651 void copy_extent_buffer_full(struct extent_buffer *dst,
5652 struct extent_buffer *src)
5657 ASSERT(dst->len == src->len);
5659 num_pages = num_extent_pages(dst->start, dst->len);
5660 for (i = 0; i < num_pages; i++)
5661 copy_page(page_address(dst->pages[i]),
5662 page_address(src->pages[i]));
5665 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5666 unsigned long dst_offset, unsigned long src_offset,
5669 u64 dst_len = dst->len;
5674 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5675 unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT;
5677 WARN_ON(src->len != dst_len);
5679 offset = (start_offset + dst_offset) &
5683 page = dst->pages[i];
5684 WARN_ON(!PageUptodate(page));
5686 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5688 kaddr = page_address(page);
5689 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5698 void le_bitmap_set(u8 *map, unsigned int start, int len)
5700 u8 *p = map + BIT_BYTE(start);
5701 const unsigned int size = start + len;
5702 int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE);
5703 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start);
5705 while (len - bits_to_set >= 0) {
5708 bits_to_set = BITS_PER_BYTE;
5713 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5718 void le_bitmap_clear(u8 *map, unsigned int start, int len)
5720 u8 *p = map + BIT_BYTE(start);
5721 const unsigned int size = start + len;
5722 int bits_to_clear = BITS_PER_BYTE - (start % BITS_PER_BYTE);
5723 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(start);
5725 while (len - bits_to_clear >= 0) {
5726 *p &= ~mask_to_clear;
5727 len -= bits_to_clear;
5728 bits_to_clear = BITS_PER_BYTE;
5733 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5734 *p &= ~mask_to_clear;
5739 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5741 * @eb: the extent buffer
5742 * @start: offset of the bitmap item in the extent buffer
5744 * @page_index: return index of the page in the extent buffer that contains the
5746 * @page_offset: return offset into the page given by page_index
5748 * This helper hides the ugliness of finding the byte in an extent buffer which
5749 * contains a given bit.
5751 static inline void eb_bitmap_offset(struct extent_buffer *eb,
5752 unsigned long start, unsigned long nr,
5753 unsigned long *page_index,
5754 size_t *page_offset)
5756 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5757 size_t byte_offset = BIT_BYTE(nr);
5761 * The byte we want is the offset of the extent buffer + the offset of
5762 * the bitmap item in the extent buffer + the offset of the byte in the
5765 offset = start_offset + start + byte_offset;
5767 *page_index = offset >> PAGE_SHIFT;
5768 *page_offset = offset & (PAGE_SIZE - 1);
5772 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5773 * @eb: the extent buffer
5774 * @start: offset of the bitmap item in the extent buffer
5775 * @nr: bit number to test
5777 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
5785 eb_bitmap_offset(eb, start, nr, &i, &offset);
5786 page = eb->pages[i];
5787 WARN_ON(!PageUptodate(page));
5788 kaddr = page_address(page);
5789 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5793 * extent_buffer_bitmap_set - set an area of a bitmap
5794 * @eb: the extent buffer
5795 * @start: offset of the bitmap item in the extent buffer
5796 * @pos: bit number of the first bit
5797 * @len: number of bits to set
5799 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
5800 unsigned long pos, unsigned long len)
5806 const unsigned int size = pos + len;
5807 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5808 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5810 eb_bitmap_offset(eb, start, pos, &i, &offset);
5811 page = eb->pages[i];
5812 WARN_ON(!PageUptodate(page));
5813 kaddr = page_address(page);
5815 while (len >= bits_to_set) {
5816 kaddr[offset] |= mask_to_set;
5818 bits_to_set = BITS_PER_BYTE;
5820 if (++offset >= PAGE_SIZE && len > 0) {
5822 page = eb->pages[++i];
5823 WARN_ON(!PageUptodate(page));
5824 kaddr = page_address(page);
5828 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5829 kaddr[offset] |= mask_to_set;
5835 * extent_buffer_bitmap_clear - clear an area of a bitmap
5836 * @eb: the extent buffer
5837 * @start: offset of the bitmap item in the extent buffer
5838 * @pos: bit number of the first bit
5839 * @len: number of bits to clear
5841 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
5842 unsigned long pos, unsigned long len)
5848 const unsigned int size = pos + len;
5849 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5850 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5852 eb_bitmap_offset(eb, start, pos, &i, &offset);
5853 page = eb->pages[i];
5854 WARN_ON(!PageUptodate(page));
5855 kaddr = page_address(page);
5857 while (len >= bits_to_clear) {
5858 kaddr[offset] &= ~mask_to_clear;
5859 len -= bits_to_clear;
5860 bits_to_clear = BITS_PER_BYTE;
5862 if (++offset >= PAGE_SIZE && len > 0) {
5864 page = eb->pages[++i];
5865 WARN_ON(!PageUptodate(page));
5866 kaddr = page_address(page);
5870 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5871 kaddr[offset] &= ~mask_to_clear;
5875 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5877 unsigned long distance = (src > dst) ? src - dst : dst - src;
5878 return distance < len;
5881 static void copy_pages(struct page *dst_page, struct page *src_page,
5882 unsigned long dst_off, unsigned long src_off,
5885 char *dst_kaddr = page_address(dst_page);
5887 int must_memmove = 0;
5889 if (dst_page != src_page) {
5890 src_kaddr = page_address(src_page);
5892 src_kaddr = dst_kaddr;
5893 if (areas_overlap(src_off, dst_off, len))
5898 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5900 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5903 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5904 unsigned long src_offset, unsigned long len)
5906 struct btrfs_fs_info *fs_info = dst->fs_info;
5908 size_t dst_off_in_page;
5909 size_t src_off_in_page;
5910 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5911 unsigned long dst_i;
5912 unsigned long src_i;
5914 if (src_offset + len > dst->len) {
5916 "memmove bogus src_offset %lu move len %lu dst len %lu",
5917 src_offset, len, dst->len);
5920 if (dst_offset + len > dst->len) {
5922 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5923 dst_offset, len, dst->len);
5928 dst_off_in_page = (start_offset + dst_offset) &
5930 src_off_in_page = (start_offset + src_offset) &
5933 dst_i = (start_offset + dst_offset) >> PAGE_SHIFT;
5934 src_i = (start_offset + src_offset) >> PAGE_SHIFT;
5936 cur = min(len, (unsigned long)(PAGE_SIZE -
5938 cur = min_t(unsigned long, cur,
5939 (unsigned long)(PAGE_SIZE - dst_off_in_page));
5941 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5942 dst_off_in_page, src_off_in_page, cur);
5950 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5951 unsigned long src_offset, unsigned long len)
5953 struct btrfs_fs_info *fs_info = dst->fs_info;
5955 size_t dst_off_in_page;
5956 size_t src_off_in_page;
5957 unsigned long dst_end = dst_offset + len - 1;
5958 unsigned long src_end = src_offset + len - 1;
5959 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5960 unsigned long dst_i;
5961 unsigned long src_i;
5963 if (src_offset + len > dst->len) {
5965 "memmove bogus src_offset %lu move len %lu len %lu",
5966 src_offset, len, dst->len);
5969 if (dst_offset + len > dst->len) {
5971 "memmove bogus dst_offset %lu move len %lu len %lu",
5972 dst_offset, len, dst->len);
5975 if (dst_offset < src_offset) {
5976 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5980 dst_i = (start_offset + dst_end) >> PAGE_SHIFT;
5981 src_i = (start_offset + src_end) >> PAGE_SHIFT;
5983 dst_off_in_page = (start_offset + dst_end) &
5985 src_off_in_page = (start_offset + src_end) &
5988 cur = min_t(unsigned long, len, src_off_in_page + 1);
5989 cur = min(cur, dst_off_in_page + 1);
5990 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5991 dst_off_in_page - cur + 1,
5992 src_off_in_page - cur + 1, cur);
6000 int try_release_extent_buffer(struct page *page)
6002 struct extent_buffer *eb;
6005 * We need to make sure nobody is attaching this page to an eb right
6008 spin_lock(&page->mapping->private_lock);
6009 if (!PagePrivate(page)) {
6010 spin_unlock(&page->mapping->private_lock);
6014 eb = (struct extent_buffer *)page->private;
6018 * This is a little awful but should be ok, we need to make sure that
6019 * the eb doesn't disappear out from under us while we're looking at
6022 spin_lock(&eb->refs_lock);
6023 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
6024 spin_unlock(&eb->refs_lock);
6025 spin_unlock(&page->mapping->private_lock);
6028 spin_unlock(&page->mapping->private_lock);
6031 * If tree ref isn't set then we know the ref on this eb is a real ref,
6032 * so just return, this page will likely be freed soon anyway.
6034 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
6035 spin_unlock(&eb->refs_lock);
6039 return release_extent_buffer(eb);