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 /* tells writepage not to lock the state bits for this range
113 * it still does the unlocking
115 unsigned int extent_locked:1;
117 /* tells the submit_bio code to use REQ_SYNC */
118 unsigned int sync_io:1;
121 static void add_extent_changeset(struct extent_state *state, unsigned bits,
122 struct extent_changeset *changeset,
129 if (set && (state->state & bits) == bits)
131 if (!set && (state->state & bits) == 0)
133 changeset->bytes_changed += state->end - state->start + 1;
134 ret = ulist_add(&changeset->range_changed, state->start, state->end,
140 static noinline void flush_write_bio(void *data);
141 static inline struct btrfs_fs_info *
142 tree_fs_info(struct extent_io_tree *tree)
145 return tree->ops->tree_fs_info(tree->private_data);
149 int __init extent_io_init(void)
151 extent_state_cache = kmem_cache_create("btrfs_extent_state",
152 sizeof(struct extent_state), 0,
153 SLAB_MEM_SPREAD, NULL);
154 if (!extent_state_cache)
157 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
158 sizeof(struct extent_buffer), 0,
159 SLAB_MEM_SPREAD, NULL);
160 if (!extent_buffer_cache)
161 goto free_state_cache;
163 btrfs_bioset = bioset_create(BIO_POOL_SIZE,
164 offsetof(struct btrfs_io_bio, bio),
167 goto free_buffer_cache;
169 if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE))
175 bioset_free(btrfs_bioset);
179 kmem_cache_destroy(extent_buffer_cache);
180 extent_buffer_cache = NULL;
183 kmem_cache_destroy(extent_state_cache);
184 extent_state_cache = NULL;
188 void extent_io_exit(void)
190 btrfs_leak_debug_check();
193 * Make sure all delayed rcu free are flushed before we
197 kmem_cache_destroy(extent_state_cache);
198 kmem_cache_destroy(extent_buffer_cache);
200 bioset_free(btrfs_bioset);
203 void extent_io_tree_init(struct extent_io_tree *tree,
206 tree->state = RB_ROOT;
208 tree->dirty_bytes = 0;
209 spin_lock_init(&tree->lock);
210 tree->private_data = private_data;
213 static struct extent_state *alloc_extent_state(gfp_t mask)
215 struct extent_state *state;
218 * The given mask might be not appropriate for the slab allocator,
219 * drop the unsupported bits
221 mask &= ~(__GFP_DMA32|__GFP_HIGHMEM);
222 state = kmem_cache_alloc(extent_state_cache, mask);
226 state->failrec = NULL;
227 RB_CLEAR_NODE(&state->rb_node);
228 btrfs_leak_debug_add(&state->leak_list, &states);
229 refcount_set(&state->refs, 1);
230 init_waitqueue_head(&state->wq);
231 trace_alloc_extent_state(state, mask, _RET_IP_);
235 void free_extent_state(struct extent_state *state)
239 if (refcount_dec_and_test(&state->refs)) {
240 WARN_ON(extent_state_in_tree(state));
241 btrfs_leak_debug_del(&state->leak_list);
242 trace_free_extent_state(state, _RET_IP_);
243 kmem_cache_free(extent_state_cache, state);
247 static struct rb_node *tree_insert(struct rb_root *root,
248 struct rb_node *search_start,
250 struct rb_node *node,
251 struct rb_node ***p_in,
252 struct rb_node **parent_in)
255 struct rb_node *parent = NULL;
256 struct tree_entry *entry;
258 if (p_in && parent_in) {
264 p = search_start ? &search_start : &root->rb_node;
267 entry = rb_entry(parent, struct tree_entry, rb_node);
269 if (offset < entry->start)
271 else if (offset > entry->end)
278 rb_link_node(node, parent, p);
279 rb_insert_color(node, root);
283 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
284 struct rb_node **prev_ret,
285 struct rb_node **next_ret,
286 struct rb_node ***p_ret,
287 struct rb_node **parent_ret)
289 struct rb_root *root = &tree->state;
290 struct rb_node **n = &root->rb_node;
291 struct rb_node *prev = NULL;
292 struct rb_node *orig_prev = NULL;
293 struct tree_entry *entry;
294 struct tree_entry *prev_entry = NULL;
298 entry = rb_entry(prev, struct tree_entry, rb_node);
301 if (offset < entry->start)
303 else if (offset > entry->end)
316 while (prev && offset > prev_entry->end) {
317 prev = rb_next(prev);
318 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
325 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
326 while (prev && offset < prev_entry->start) {
327 prev = rb_prev(prev);
328 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
335 static inline struct rb_node *
336 tree_search_for_insert(struct extent_io_tree *tree,
338 struct rb_node ***p_ret,
339 struct rb_node **parent_ret)
341 struct rb_node *prev = NULL;
344 ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret);
350 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
353 return tree_search_for_insert(tree, offset, NULL, NULL);
356 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
357 struct extent_state *other)
359 if (tree->ops && tree->ops->merge_extent_hook)
360 tree->ops->merge_extent_hook(tree->private_data, new, other);
364 * utility function to look for merge candidates inside a given range.
365 * Any extents with matching state are merged together into a single
366 * extent in the tree. Extents with EXTENT_IO in their state field
367 * are not merged because the end_io handlers need to be able to do
368 * operations on them without sleeping (or doing allocations/splits).
370 * This should be called with the tree lock held.
372 static void merge_state(struct extent_io_tree *tree,
373 struct extent_state *state)
375 struct extent_state *other;
376 struct rb_node *other_node;
378 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
381 other_node = rb_prev(&state->rb_node);
383 other = rb_entry(other_node, struct extent_state, rb_node);
384 if (other->end == state->start - 1 &&
385 other->state == state->state) {
386 merge_cb(tree, state, other);
387 state->start = other->start;
388 rb_erase(&other->rb_node, &tree->state);
389 RB_CLEAR_NODE(&other->rb_node);
390 free_extent_state(other);
393 other_node = rb_next(&state->rb_node);
395 other = rb_entry(other_node, struct extent_state, rb_node);
396 if (other->start == state->end + 1 &&
397 other->state == state->state) {
398 merge_cb(tree, state, other);
399 state->end = other->end;
400 rb_erase(&other->rb_node, &tree->state);
401 RB_CLEAR_NODE(&other->rb_node);
402 free_extent_state(other);
407 static void set_state_cb(struct extent_io_tree *tree,
408 struct extent_state *state, unsigned *bits)
410 if (tree->ops && tree->ops->set_bit_hook)
411 tree->ops->set_bit_hook(tree->private_data, state, bits);
414 static void clear_state_cb(struct extent_io_tree *tree,
415 struct extent_state *state, unsigned *bits)
417 if (tree->ops && tree->ops->clear_bit_hook)
418 tree->ops->clear_bit_hook(tree->private_data, state, bits);
421 static void set_state_bits(struct extent_io_tree *tree,
422 struct extent_state *state, unsigned *bits,
423 struct extent_changeset *changeset);
426 * insert an extent_state struct into the tree. 'bits' are set on the
427 * struct before it is inserted.
429 * This may return -EEXIST if the extent is already there, in which case the
430 * state struct is freed.
432 * The tree lock is not taken internally. This is a utility function and
433 * probably isn't what you want to call (see set/clear_extent_bit).
435 static int insert_state(struct extent_io_tree *tree,
436 struct extent_state *state, u64 start, u64 end,
438 struct rb_node **parent,
439 unsigned *bits, struct extent_changeset *changeset)
441 struct rb_node *node;
444 WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n",
446 state->start = start;
449 set_state_bits(tree, state, bits, changeset);
451 node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
453 struct extent_state *found;
454 found = rb_entry(node, struct extent_state, rb_node);
455 pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n",
456 found->start, found->end, start, end);
459 merge_state(tree, state);
463 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
466 if (tree->ops && tree->ops->split_extent_hook)
467 tree->ops->split_extent_hook(tree->private_data, orig, split);
471 * split a given extent state struct in two, inserting the preallocated
472 * struct 'prealloc' as the newly created second half. 'split' indicates an
473 * offset inside 'orig' where it should be split.
476 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
477 * are two extent state structs in the tree:
478 * prealloc: [orig->start, split - 1]
479 * orig: [ split, orig->end ]
481 * The tree locks are not taken by this function. They need to be held
484 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
485 struct extent_state *prealloc, u64 split)
487 struct rb_node *node;
489 split_cb(tree, orig, split);
491 prealloc->start = orig->start;
492 prealloc->end = split - 1;
493 prealloc->state = orig->state;
496 node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
497 &prealloc->rb_node, NULL, NULL);
499 free_extent_state(prealloc);
505 static struct extent_state *next_state(struct extent_state *state)
507 struct rb_node *next = rb_next(&state->rb_node);
509 return rb_entry(next, struct extent_state, rb_node);
515 * utility function to clear some bits in an extent state struct.
516 * it will optionally wake up any one waiting on this state (wake == 1).
518 * If no bits are set on the state struct after clearing things, the
519 * struct is freed and removed from the tree
521 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
522 struct extent_state *state,
523 unsigned *bits, int wake,
524 struct extent_changeset *changeset)
526 struct extent_state *next;
527 unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS;
529 if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
530 u64 range = state->end - state->start + 1;
531 WARN_ON(range > tree->dirty_bytes);
532 tree->dirty_bytes -= range;
534 clear_state_cb(tree, state, bits);
535 add_extent_changeset(state, bits_to_clear, changeset, 0);
536 state->state &= ~bits_to_clear;
539 if (state->state == 0) {
540 next = next_state(state);
541 if (extent_state_in_tree(state)) {
542 rb_erase(&state->rb_node, &tree->state);
543 RB_CLEAR_NODE(&state->rb_node);
544 free_extent_state(state);
549 merge_state(tree, state);
550 next = next_state(state);
555 static struct extent_state *
556 alloc_extent_state_atomic(struct extent_state *prealloc)
559 prealloc = alloc_extent_state(GFP_ATOMIC);
564 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
566 btrfs_panic(tree_fs_info(tree), err,
567 "Locking error: Extent tree was modified by another thread while locked.");
571 * clear some bits on a range in the tree. This may require splitting
572 * or inserting elements in the tree, so the gfp mask is used to
573 * indicate which allocations or sleeping are allowed.
575 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
576 * the given range from the tree regardless of state (ie for truncate).
578 * the range [start, end] is inclusive.
580 * This takes the tree lock, and returns 0 on success and < 0 on error.
582 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
583 unsigned bits, int wake, int delete,
584 struct extent_state **cached_state,
585 gfp_t mask, struct extent_changeset *changeset)
587 struct extent_state *state;
588 struct extent_state *cached;
589 struct extent_state *prealloc = NULL;
590 struct rb_node *node;
595 btrfs_debug_check_extent_io_range(tree, start, end);
597 if (bits & EXTENT_DELALLOC)
598 bits |= EXTENT_NORESERVE;
601 bits |= ~EXTENT_CTLBITS;
602 bits |= EXTENT_FIRST_DELALLOC;
604 if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
607 if (!prealloc && gfpflags_allow_blocking(mask)) {
609 * Don't care for allocation failure here because we might end
610 * up not needing the pre-allocated extent state at all, which
611 * is the case if we only have in the tree extent states that
612 * cover our input range and don't cover too any other range.
613 * If we end up needing a new extent state we allocate it later.
615 prealloc = alloc_extent_state(mask);
618 spin_lock(&tree->lock);
620 cached = *cached_state;
623 *cached_state = NULL;
627 if (cached && extent_state_in_tree(cached) &&
628 cached->start <= start && cached->end > start) {
630 refcount_dec(&cached->refs);
635 free_extent_state(cached);
638 * this search will find the extents that end after
641 node = tree_search(tree, start);
644 state = rb_entry(node, struct extent_state, rb_node);
646 if (state->start > end)
648 WARN_ON(state->end < start);
649 last_end = state->end;
651 /* the state doesn't have the wanted bits, go ahead */
652 if (!(state->state & bits)) {
653 state = next_state(state);
658 * | ---- desired range ---- |
660 * | ------------- state -------------- |
662 * We need to split the extent we found, and may flip
663 * bits on second half.
665 * If the extent we found extends past our range, we
666 * just split and search again. It'll get split again
667 * the next time though.
669 * If the extent we found is inside our range, we clear
670 * the desired bit on it.
673 if (state->start < start) {
674 prealloc = alloc_extent_state_atomic(prealloc);
676 err = split_state(tree, state, prealloc, start);
678 extent_io_tree_panic(tree, err);
683 if (state->end <= end) {
684 state = clear_state_bit(tree, state, &bits, wake,
691 * | ---- desired range ---- |
693 * We need to split the extent, and clear the bit
696 if (state->start <= end && state->end > end) {
697 prealloc = alloc_extent_state_atomic(prealloc);
699 err = split_state(tree, state, prealloc, end + 1);
701 extent_io_tree_panic(tree, err);
706 clear_state_bit(tree, prealloc, &bits, wake, changeset);
712 state = clear_state_bit(tree, state, &bits, wake, changeset);
714 if (last_end == (u64)-1)
716 start = last_end + 1;
717 if (start <= end && state && !need_resched())
723 spin_unlock(&tree->lock);
724 if (gfpflags_allow_blocking(mask))
729 spin_unlock(&tree->lock);
731 free_extent_state(prealloc);
737 static void wait_on_state(struct extent_io_tree *tree,
738 struct extent_state *state)
739 __releases(tree->lock)
740 __acquires(tree->lock)
743 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
744 spin_unlock(&tree->lock);
746 spin_lock(&tree->lock);
747 finish_wait(&state->wq, &wait);
751 * waits for one or more bits to clear on a range in the state tree.
752 * The range [start, end] is inclusive.
753 * The tree lock is taken by this function
755 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
758 struct extent_state *state;
759 struct rb_node *node;
761 btrfs_debug_check_extent_io_range(tree, start, end);
763 spin_lock(&tree->lock);
767 * this search will find all the extents that end after
770 node = tree_search(tree, start);
775 state = rb_entry(node, struct extent_state, rb_node);
777 if (state->start > end)
780 if (state->state & bits) {
781 start = state->start;
782 refcount_inc(&state->refs);
783 wait_on_state(tree, state);
784 free_extent_state(state);
787 start = state->end + 1;
792 if (!cond_resched_lock(&tree->lock)) {
793 node = rb_next(node);
798 spin_unlock(&tree->lock);
801 static void set_state_bits(struct extent_io_tree *tree,
802 struct extent_state *state,
803 unsigned *bits, struct extent_changeset *changeset)
805 unsigned bits_to_set = *bits & ~EXTENT_CTLBITS;
807 set_state_cb(tree, state, bits);
808 if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
809 u64 range = state->end - state->start + 1;
810 tree->dirty_bytes += range;
812 add_extent_changeset(state, bits_to_set, changeset, 1);
813 state->state |= bits_to_set;
816 static void cache_state_if_flags(struct extent_state *state,
817 struct extent_state **cached_ptr,
820 if (cached_ptr && !(*cached_ptr)) {
821 if (!flags || (state->state & flags)) {
823 refcount_inc(&state->refs);
828 static void cache_state(struct extent_state *state,
829 struct extent_state **cached_ptr)
831 return cache_state_if_flags(state, cached_ptr,
832 EXTENT_IOBITS | EXTENT_BOUNDARY);
836 * set some bits on a range in the tree. This may require allocations or
837 * sleeping, so the gfp mask is used to indicate what is allowed.
839 * If any of the exclusive bits are set, this will fail with -EEXIST if some
840 * part of the range already has the desired bits set. The start of the
841 * existing range is returned in failed_start in this case.
843 * [start, end] is inclusive This takes the tree lock.
846 static int __must_check
847 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
848 unsigned bits, unsigned exclusive_bits,
849 u64 *failed_start, struct extent_state **cached_state,
850 gfp_t mask, struct extent_changeset *changeset)
852 struct extent_state *state;
853 struct extent_state *prealloc = NULL;
854 struct rb_node *node;
856 struct rb_node *parent;
861 btrfs_debug_check_extent_io_range(tree, start, end);
863 bits |= EXTENT_FIRST_DELALLOC;
865 if (!prealloc && gfpflags_allow_blocking(mask)) {
867 * Don't care for allocation failure here because we might end
868 * up not needing the pre-allocated extent state at all, which
869 * is the case if we only have in the tree extent states that
870 * cover our input range and don't cover too any other range.
871 * If we end up needing a new extent state we allocate it later.
873 prealloc = alloc_extent_state(mask);
876 spin_lock(&tree->lock);
877 if (cached_state && *cached_state) {
878 state = *cached_state;
879 if (state->start <= start && state->end > start &&
880 extent_state_in_tree(state)) {
881 node = &state->rb_node;
886 * this search will find all the extents that end after
889 node = tree_search_for_insert(tree, start, &p, &parent);
891 prealloc = alloc_extent_state_atomic(prealloc);
893 err = insert_state(tree, prealloc, start, end,
894 &p, &parent, &bits, changeset);
896 extent_io_tree_panic(tree, err);
898 cache_state(prealloc, cached_state);
902 state = rb_entry(node, struct extent_state, rb_node);
904 last_start = state->start;
905 last_end = state->end;
908 * | ---- desired range ---- |
911 * Just lock what we found and keep going
913 if (state->start == start && state->end <= end) {
914 if (state->state & exclusive_bits) {
915 *failed_start = state->start;
920 set_state_bits(tree, state, &bits, changeset);
921 cache_state(state, cached_state);
922 merge_state(tree, state);
923 if (last_end == (u64)-1)
925 start = last_end + 1;
926 state = next_state(state);
927 if (start < end && state && state->start == start &&
934 * | ---- desired range ---- |
937 * | ------------- state -------------- |
939 * We need to split the extent we found, and may flip bits on
942 * If the extent we found extends past our
943 * range, we just split and search again. It'll get split
944 * again the next time though.
946 * If the extent we found is inside our range, we set the
949 if (state->start < start) {
950 if (state->state & exclusive_bits) {
951 *failed_start = start;
956 prealloc = alloc_extent_state_atomic(prealloc);
958 err = split_state(tree, state, prealloc, start);
960 extent_io_tree_panic(tree, err);
965 if (state->end <= end) {
966 set_state_bits(tree, state, &bits, changeset);
967 cache_state(state, cached_state);
968 merge_state(tree, state);
969 if (last_end == (u64)-1)
971 start = last_end + 1;
972 state = next_state(state);
973 if (start < end && state && state->start == start &&
980 * | ---- desired range ---- |
981 * | state | or | state |
983 * There's a hole, we need to insert something in it and
984 * ignore the extent we found.
986 if (state->start > start) {
988 if (end < last_start)
991 this_end = last_start - 1;
993 prealloc = alloc_extent_state_atomic(prealloc);
997 * Avoid to free 'prealloc' if it can be merged with
1000 err = insert_state(tree, prealloc, start, this_end,
1001 NULL, NULL, &bits, changeset);
1003 extent_io_tree_panic(tree, err);
1005 cache_state(prealloc, cached_state);
1007 start = this_end + 1;
1011 * | ---- desired range ---- |
1013 * We need to split the extent, and set the bit
1016 if (state->start <= end && state->end > end) {
1017 if (state->state & exclusive_bits) {
1018 *failed_start = start;
1023 prealloc = alloc_extent_state_atomic(prealloc);
1025 err = split_state(tree, state, prealloc, end + 1);
1027 extent_io_tree_panic(tree, err);
1029 set_state_bits(tree, prealloc, &bits, changeset);
1030 cache_state(prealloc, cached_state);
1031 merge_state(tree, prealloc);
1039 spin_unlock(&tree->lock);
1040 if (gfpflags_allow_blocking(mask))
1045 spin_unlock(&tree->lock);
1047 free_extent_state(prealloc);
1053 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1054 unsigned bits, u64 * failed_start,
1055 struct extent_state **cached_state, gfp_t mask)
1057 return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1058 cached_state, mask, NULL);
1063 * convert_extent_bit - convert all bits in a given range from one bit to
1065 * @tree: the io tree to search
1066 * @start: the start offset in bytes
1067 * @end: the end offset in bytes (inclusive)
1068 * @bits: the bits to set in this range
1069 * @clear_bits: the bits to clear in this range
1070 * @cached_state: state that we're going to cache
1072 * This will go through and set bits for the given range. If any states exist
1073 * already in this range they are set with the given bit and cleared of the
1074 * clear_bits. This is only meant to be used by things that are mergeable, ie
1075 * converting from say DELALLOC to DIRTY. This is not meant to be used with
1076 * boundary bits like LOCK.
1078 * All allocations are done with GFP_NOFS.
1080 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1081 unsigned bits, unsigned clear_bits,
1082 struct extent_state **cached_state)
1084 struct extent_state *state;
1085 struct extent_state *prealloc = NULL;
1086 struct rb_node *node;
1088 struct rb_node *parent;
1092 bool first_iteration = true;
1094 btrfs_debug_check_extent_io_range(tree, start, end);
1099 * Best effort, don't worry if extent state allocation fails
1100 * here for the first iteration. We might have a cached state
1101 * that matches exactly the target range, in which case no
1102 * extent state allocations are needed. We'll only know this
1103 * after locking the tree.
1105 prealloc = alloc_extent_state(GFP_NOFS);
1106 if (!prealloc && !first_iteration)
1110 spin_lock(&tree->lock);
1111 if (cached_state && *cached_state) {
1112 state = *cached_state;
1113 if (state->start <= start && state->end > start &&
1114 extent_state_in_tree(state)) {
1115 node = &state->rb_node;
1121 * this search will find all the extents that end after
1124 node = tree_search_for_insert(tree, start, &p, &parent);
1126 prealloc = alloc_extent_state_atomic(prealloc);
1131 err = insert_state(tree, prealloc, start, end,
1132 &p, &parent, &bits, NULL);
1134 extent_io_tree_panic(tree, err);
1135 cache_state(prealloc, cached_state);
1139 state = rb_entry(node, struct extent_state, rb_node);
1141 last_start = state->start;
1142 last_end = state->end;
1145 * | ---- desired range ---- |
1148 * Just lock what we found and keep going
1150 if (state->start == start && state->end <= end) {
1151 set_state_bits(tree, state, &bits, NULL);
1152 cache_state(state, cached_state);
1153 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1154 if (last_end == (u64)-1)
1156 start = last_end + 1;
1157 if (start < end && state && state->start == start &&
1164 * | ---- desired range ---- |
1167 * | ------------- state -------------- |
1169 * We need to split the extent we found, and may flip bits on
1172 * If the extent we found extends past our
1173 * range, we just split and search again. It'll get split
1174 * again the next time though.
1176 * If the extent we found is inside our range, we set the
1177 * desired bit on it.
1179 if (state->start < start) {
1180 prealloc = alloc_extent_state_atomic(prealloc);
1185 err = split_state(tree, state, prealloc, start);
1187 extent_io_tree_panic(tree, err);
1191 if (state->end <= end) {
1192 set_state_bits(tree, state, &bits, NULL);
1193 cache_state(state, cached_state);
1194 state = clear_state_bit(tree, state, &clear_bits, 0,
1196 if (last_end == (u64)-1)
1198 start = last_end + 1;
1199 if (start < end && state && state->start == start &&
1206 * | ---- desired range ---- |
1207 * | state | or | state |
1209 * There's a hole, we need to insert something in it and
1210 * ignore the extent we found.
1212 if (state->start > start) {
1214 if (end < last_start)
1217 this_end = last_start - 1;
1219 prealloc = alloc_extent_state_atomic(prealloc);
1226 * Avoid to free 'prealloc' if it can be merged with
1229 err = insert_state(tree, prealloc, start, this_end,
1230 NULL, NULL, &bits, NULL);
1232 extent_io_tree_panic(tree, err);
1233 cache_state(prealloc, cached_state);
1235 start = this_end + 1;
1239 * | ---- desired range ---- |
1241 * We need to split the extent, and set the bit
1244 if (state->start <= end && state->end > end) {
1245 prealloc = alloc_extent_state_atomic(prealloc);
1251 err = split_state(tree, state, prealloc, end + 1);
1253 extent_io_tree_panic(tree, err);
1255 set_state_bits(tree, prealloc, &bits, NULL);
1256 cache_state(prealloc, cached_state);
1257 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1265 spin_unlock(&tree->lock);
1267 first_iteration = false;
1271 spin_unlock(&tree->lock);
1273 free_extent_state(prealloc);
1278 /* wrappers around set/clear extent bit */
1279 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1280 unsigned bits, struct extent_changeset *changeset)
1283 * We don't support EXTENT_LOCKED yet, as current changeset will
1284 * record any bits changed, so for EXTENT_LOCKED case, it will
1285 * either fail with -EEXIST or changeset will record the whole
1288 BUG_ON(bits & EXTENT_LOCKED);
1290 return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1294 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1295 unsigned bits, int wake, int delete,
1296 struct extent_state **cached)
1298 return __clear_extent_bit(tree, start, end, bits, wake, delete,
1299 cached, GFP_NOFS, NULL);
1302 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1303 unsigned bits, struct extent_changeset *changeset)
1306 * Don't support EXTENT_LOCKED case, same reason as
1307 * set_record_extent_bits().
1309 BUG_ON(bits & EXTENT_LOCKED);
1311 return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1316 * either insert or lock state struct between start and end use mask to tell
1317 * us if waiting is desired.
1319 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1320 struct extent_state **cached_state)
1326 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
1327 EXTENT_LOCKED, &failed_start,
1328 cached_state, GFP_NOFS, NULL);
1329 if (err == -EEXIST) {
1330 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1331 start = failed_start;
1334 WARN_ON(start > end);
1339 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1344 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1345 &failed_start, NULL, GFP_NOFS, NULL);
1346 if (err == -EEXIST) {
1347 if (failed_start > start)
1348 clear_extent_bit(tree, start, failed_start - 1,
1349 EXTENT_LOCKED, 1, 0, NULL);
1355 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1357 unsigned long index = start >> PAGE_SHIFT;
1358 unsigned long end_index = end >> PAGE_SHIFT;
1361 while (index <= end_index) {
1362 page = find_get_page(inode->i_mapping, index);
1363 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1364 clear_page_dirty_for_io(page);
1370 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1372 unsigned long index = start >> PAGE_SHIFT;
1373 unsigned long end_index = end >> PAGE_SHIFT;
1376 while (index <= end_index) {
1377 page = find_get_page(inode->i_mapping, index);
1378 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1379 __set_page_dirty_nobuffers(page);
1380 account_page_redirty(page);
1387 * helper function to set both pages and extents in the tree writeback
1389 static void set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1391 tree->ops->set_range_writeback(tree->private_data, start, end);
1394 /* find the first state struct with 'bits' set after 'start', and
1395 * return it. tree->lock must be held. NULL will returned if
1396 * nothing was found after 'start'
1398 static struct extent_state *
1399 find_first_extent_bit_state(struct extent_io_tree *tree,
1400 u64 start, unsigned bits)
1402 struct rb_node *node;
1403 struct extent_state *state;
1406 * this search will find all the extents that end after
1409 node = tree_search(tree, start);
1414 state = rb_entry(node, struct extent_state, rb_node);
1415 if (state->end >= start && (state->state & bits))
1418 node = rb_next(node);
1427 * find the first offset in the io tree with 'bits' set. zero is
1428 * returned if we find something, and *start_ret and *end_ret are
1429 * set to reflect the state struct that was found.
1431 * If nothing was found, 1 is returned. If found something, return 0.
1433 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1434 u64 *start_ret, u64 *end_ret, unsigned bits,
1435 struct extent_state **cached_state)
1437 struct extent_state *state;
1441 spin_lock(&tree->lock);
1442 if (cached_state && *cached_state) {
1443 state = *cached_state;
1444 if (state->end == start - 1 && extent_state_in_tree(state)) {
1445 n = rb_next(&state->rb_node);
1447 state = rb_entry(n, struct extent_state,
1449 if (state->state & bits)
1453 free_extent_state(*cached_state);
1454 *cached_state = NULL;
1457 free_extent_state(*cached_state);
1458 *cached_state = NULL;
1461 state = find_first_extent_bit_state(tree, start, bits);
1464 cache_state_if_flags(state, cached_state, 0);
1465 *start_ret = state->start;
1466 *end_ret = state->end;
1470 spin_unlock(&tree->lock);
1475 * find a contiguous range of bytes in the file marked as delalloc, not
1476 * more than 'max_bytes'. start and end are used to return the range,
1478 * 1 is returned if we find something, 0 if nothing was in the tree
1480 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1481 u64 *start, u64 *end, u64 max_bytes,
1482 struct extent_state **cached_state)
1484 struct rb_node *node;
1485 struct extent_state *state;
1486 u64 cur_start = *start;
1488 u64 total_bytes = 0;
1490 spin_lock(&tree->lock);
1493 * this search will find all the extents that end after
1496 node = tree_search(tree, cur_start);
1504 state = rb_entry(node, struct extent_state, rb_node);
1505 if (found && (state->start != cur_start ||
1506 (state->state & EXTENT_BOUNDARY))) {
1509 if (!(state->state & EXTENT_DELALLOC)) {
1515 *start = state->start;
1516 *cached_state = state;
1517 refcount_inc(&state->refs);
1521 cur_start = state->end + 1;
1522 node = rb_next(node);
1523 total_bytes += state->end - state->start + 1;
1524 if (total_bytes >= max_bytes)
1530 spin_unlock(&tree->lock);
1534 static int __process_pages_contig(struct address_space *mapping,
1535 struct page *locked_page,
1536 pgoff_t start_index, pgoff_t end_index,
1537 unsigned long page_ops, pgoff_t *index_ret);
1539 static noinline void __unlock_for_delalloc(struct inode *inode,
1540 struct page *locked_page,
1543 unsigned long index = start >> PAGE_SHIFT;
1544 unsigned long end_index = end >> PAGE_SHIFT;
1546 ASSERT(locked_page);
1547 if (index == locked_page->index && end_index == index)
1550 __process_pages_contig(inode->i_mapping, locked_page, index, end_index,
1554 static noinline int lock_delalloc_pages(struct inode *inode,
1555 struct page *locked_page,
1559 unsigned long index = delalloc_start >> PAGE_SHIFT;
1560 unsigned long index_ret = index;
1561 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1564 ASSERT(locked_page);
1565 if (index == locked_page->index && index == end_index)
1568 ret = __process_pages_contig(inode->i_mapping, locked_page, index,
1569 end_index, PAGE_LOCK, &index_ret);
1571 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1572 (u64)index_ret << PAGE_SHIFT);
1577 * find a contiguous range of bytes in the file marked as delalloc, not
1578 * more than 'max_bytes'. start and end are used to return the range,
1580 * 1 is returned if we find something, 0 if nothing was in the tree
1582 STATIC u64 find_lock_delalloc_range(struct inode *inode,
1583 struct extent_io_tree *tree,
1584 struct page *locked_page, u64 *start,
1585 u64 *end, u64 max_bytes)
1590 struct extent_state *cached_state = NULL;
1595 /* step one, find a bunch of delalloc bytes starting at start */
1596 delalloc_start = *start;
1598 found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1599 max_bytes, &cached_state);
1600 if (!found || delalloc_end <= *start) {
1601 *start = delalloc_start;
1602 *end = delalloc_end;
1603 free_extent_state(cached_state);
1608 * start comes from the offset of locked_page. We have to lock
1609 * pages in order, so we can't process delalloc bytes before
1612 if (delalloc_start < *start)
1613 delalloc_start = *start;
1616 * make sure to limit the number of pages we try to lock down
1618 if (delalloc_end + 1 - delalloc_start > max_bytes)
1619 delalloc_end = delalloc_start + max_bytes - 1;
1621 /* step two, lock all the pages after the page that has start */
1622 ret = lock_delalloc_pages(inode, locked_page,
1623 delalloc_start, delalloc_end);
1624 if (ret == -EAGAIN) {
1625 /* some of the pages are gone, lets avoid looping by
1626 * shortening the size of the delalloc range we're searching
1628 free_extent_state(cached_state);
1629 cached_state = NULL;
1631 max_bytes = PAGE_SIZE;
1639 BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1641 /* step three, lock the state bits for the whole range */
1642 lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1644 /* then test to make sure it is all still delalloc */
1645 ret = test_range_bit(tree, delalloc_start, delalloc_end,
1646 EXTENT_DELALLOC, 1, cached_state);
1648 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1649 &cached_state, GFP_NOFS);
1650 __unlock_for_delalloc(inode, locked_page,
1651 delalloc_start, delalloc_end);
1655 free_extent_state(cached_state);
1656 *start = delalloc_start;
1657 *end = delalloc_end;
1662 static int __process_pages_contig(struct address_space *mapping,
1663 struct page *locked_page,
1664 pgoff_t start_index, pgoff_t end_index,
1665 unsigned long page_ops, pgoff_t *index_ret)
1667 unsigned long nr_pages = end_index - start_index + 1;
1668 unsigned long pages_locked = 0;
1669 pgoff_t index = start_index;
1670 struct page *pages[16];
1675 if (page_ops & PAGE_LOCK) {
1676 ASSERT(page_ops == PAGE_LOCK);
1677 ASSERT(index_ret && *index_ret == start_index);
1680 if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1681 mapping_set_error(mapping, -EIO);
1683 while (nr_pages > 0) {
1684 ret = find_get_pages_contig(mapping, index,
1685 min_t(unsigned long,
1686 nr_pages, ARRAY_SIZE(pages)), pages);
1689 * Only if we're going to lock these pages,
1690 * can we find nothing at @index.
1692 ASSERT(page_ops & PAGE_LOCK);
1697 for (i = 0; i < ret; i++) {
1698 if (page_ops & PAGE_SET_PRIVATE2)
1699 SetPagePrivate2(pages[i]);
1701 if (pages[i] == locked_page) {
1706 if (page_ops & PAGE_CLEAR_DIRTY)
1707 clear_page_dirty_for_io(pages[i]);
1708 if (page_ops & PAGE_SET_WRITEBACK)
1709 set_page_writeback(pages[i]);
1710 if (page_ops & PAGE_SET_ERROR)
1711 SetPageError(pages[i]);
1712 if (page_ops & PAGE_END_WRITEBACK)
1713 end_page_writeback(pages[i]);
1714 if (page_ops & PAGE_UNLOCK)
1715 unlock_page(pages[i]);
1716 if (page_ops & PAGE_LOCK) {
1717 lock_page(pages[i]);
1718 if (!PageDirty(pages[i]) ||
1719 pages[i]->mapping != mapping) {
1720 unlock_page(pages[i]);
1734 if (err && index_ret)
1735 *index_ret = start_index + pages_locked - 1;
1739 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1740 u64 delalloc_end, struct page *locked_page,
1741 unsigned clear_bits,
1742 unsigned long page_ops)
1744 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0,
1747 __process_pages_contig(inode->i_mapping, locked_page,
1748 start >> PAGE_SHIFT, end >> PAGE_SHIFT,
1753 * count the number of bytes in the tree that have a given bit(s)
1754 * set. This can be fairly slow, except for EXTENT_DIRTY which is
1755 * cached. The total number found is returned.
1757 u64 count_range_bits(struct extent_io_tree *tree,
1758 u64 *start, u64 search_end, u64 max_bytes,
1759 unsigned bits, int contig)
1761 struct rb_node *node;
1762 struct extent_state *state;
1763 u64 cur_start = *start;
1764 u64 total_bytes = 0;
1768 if (WARN_ON(search_end <= cur_start))
1771 spin_lock(&tree->lock);
1772 if (cur_start == 0 && bits == EXTENT_DIRTY) {
1773 total_bytes = tree->dirty_bytes;
1777 * this search will find all the extents that end after
1780 node = tree_search(tree, cur_start);
1785 state = rb_entry(node, struct extent_state, rb_node);
1786 if (state->start > search_end)
1788 if (contig && found && state->start > last + 1)
1790 if (state->end >= cur_start && (state->state & bits) == bits) {
1791 total_bytes += min(search_end, state->end) + 1 -
1792 max(cur_start, state->start);
1793 if (total_bytes >= max_bytes)
1796 *start = max(cur_start, state->start);
1800 } else if (contig && found) {
1803 node = rb_next(node);
1808 spin_unlock(&tree->lock);
1813 * set the private field for a given byte offset in the tree. If there isn't
1814 * an extent_state there already, this does nothing.
1816 static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start,
1817 struct io_failure_record *failrec)
1819 struct rb_node *node;
1820 struct extent_state *state;
1823 spin_lock(&tree->lock);
1825 * this search will find all the extents that end after
1828 node = tree_search(tree, start);
1833 state = rb_entry(node, struct extent_state, rb_node);
1834 if (state->start != start) {
1838 state->failrec = failrec;
1840 spin_unlock(&tree->lock);
1844 static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start,
1845 struct io_failure_record **failrec)
1847 struct rb_node *node;
1848 struct extent_state *state;
1851 spin_lock(&tree->lock);
1853 * this search will find all the extents that end after
1856 node = tree_search(tree, start);
1861 state = rb_entry(node, struct extent_state, rb_node);
1862 if (state->start != start) {
1866 *failrec = state->failrec;
1868 spin_unlock(&tree->lock);
1873 * searches a range in the state tree for a given mask.
1874 * If 'filled' == 1, this returns 1 only if every extent in the tree
1875 * has the bits set. Otherwise, 1 is returned if any bit in the
1876 * range is found set.
1878 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1879 unsigned bits, int filled, struct extent_state *cached)
1881 struct extent_state *state = NULL;
1882 struct rb_node *node;
1885 spin_lock(&tree->lock);
1886 if (cached && extent_state_in_tree(cached) && cached->start <= start &&
1887 cached->end > start)
1888 node = &cached->rb_node;
1890 node = tree_search(tree, start);
1891 while (node && start <= end) {
1892 state = rb_entry(node, struct extent_state, rb_node);
1894 if (filled && state->start > start) {
1899 if (state->start > end)
1902 if (state->state & bits) {
1906 } else if (filled) {
1911 if (state->end == (u64)-1)
1914 start = state->end + 1;
1917 node = rb_next(node);
1924 spin_unlock(&tree->lock);
1929 * helper function to set a given page up to date if all the
1930 * extents in the tree for that page are up to date
1932 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1934 u64 start = page_offset(page);
1935 u64 end = start + PAGE_SIZE - 1;
1936 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1937 SetPageUptodate(page);
1940 int free_io_failure(struct extent_io_tree *failure_tree,
1941 struct extent_io_tree *io_tree,
1942 struct io_failure_record *rec)
1947 set_state_failrec(failure_tree, rec->start, NULL);
1948 ret = clear_extent_bits(failure_tree, rec->start,
1949 rec->start + rec->len - 1,
1950 EXTENT_LOCKED | EXTENT_DIRTY);
1954 ret = clear_extent_bits(io_tree, rec->start,
1955 rec->start + rec->len - 1,
1965 * this bypasses the standard btrfs submit functions deliberately, as
1966 * the standard behavior is to write all copies in a raid setup. here we only
1967 * want to write the one bad copy. so we do the mapping for ourselves and issue
1968 * submit_bio directly.
1969 * to avoid any synchronization issues, wait for the data after writing, which
1970 * actually prevents the read that triggered the error from finishing.
1971 * currently, there can be no more than two copies of every data bit. thus,
1972 * exactly one rewrite is required.
1974 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
1975 u64 length, u64 logical, struct page *page,
1976 unsigned int pg_offset, int mirror_num)
1979 struct btrfs_device *dev;
1982 struct btrfs_bio *bbio = NULL;
1985 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
1986 BUG_ON(!mirror_num);
1988 bio = btrfs_io_bio_alloc(1);
1989 bio->bi_iter.bi_size = 0;
1990 map_length = length;
1993 * Avoid races with device replace and make sure our bbio has devices
1994 * associated to its stripes that don't go away while we are doing the
1995 * read repair operation.
1997 btrfs_bio_counter_inc_blocked(fs_info);
1998 if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2000 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2001 * to update all raid stripes, but here we just want to correct
2002 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2003 * stripe's dev and sector.
2005 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2006 &map_length, &bbio, 0);
2008 btrfs_bio_counter_dec(fs_info);
2012 ASSERT(bbio->mirror_num == 1);
2014 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2015 &map_length, &bbio, mirror_num);
2017 btrfs_bio_counter_dec(fs_info);
2021 BUG_ON(mirror_num != bbio->mirror_num);
2024 sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2025 bio->bi_iter.bi_sector = sector;
2026 dev = bbio->stripes[bbio->mirror_num - 1].dev;
2027 btrfs_put_bbio(bbio);
2028 if (!dev || !dev->bdev || !dev->writeable) {
2029 btrfs_bio_counter_dec(fs_info);
2033 bio_set_dev(bio, dev->bdev);
2034 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2035 bio_add_page(bio, page, length, pg_offset);
2037 if (btrfsic_submit_bio_wait(bio)) {
2038 /* try to remap that extent elsewhere? */
2039 btrfs_bio_counter_dec(fs_info);
2041 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2045 btrfs_info_rl_in_rcu(fs_info,
2046 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2048 rcu_str_deref(dev->name), sector);
2049 btrfs_bio_counter_dec(fs_info);
2054 int repair_eb_io_failure(struct btrfs_fs_info *fs_info,
2055 struct extent_buffer *eb, int mirror_num)
2057 u64 start = eb->start;
2058 unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2061 if (sb_rdonly(fs_info->sb))
2064 for (i = 0; i < num_pages; i++) {
2065 struct page *p = eb->pages[i];
2067 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2068 start - page_offset(p), mirror_num);
2078 * each time an IO finishes, we do a fast check in the IO failure tree
2079 * to see if we need to process or clean up an io_failure_record
2081 int clean_io_failure(struct btrfs_fs_info *fs_info,
2082 struct extent_io_tree *failure_tree,
2083 struct extent_io_tree *io_tree, u64 start,
2084 struct page *page, u64 ino, unsigned int pg_offset)
2087 struct io_failure_record *failrec;
2088 struct extent_state *state;
2093 ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2098 ret = get_state_failrec(failure_tree, start, &failrec);
2102 BUG_ON(!failrec->this_mirror);
2104 if (failrec->in_validation) {
2105 /* there was no real error, just free the record */
2106 btrfs_debug(fs_info,
2107 "clean_io_failure: freeing dummy error at %llu",
2111 if (sb_rdonly(fs_info->sb))
2114 spin_lock(&io_tree->lock);
2115 state = find_first_extent_bit_state(io_tree,
2118 spin_unlock(&io_tree->lock);
2120 if (state && state->start <= failrec->start &&
2121 state->end >= failrec->start + failrec->len - 1) {
2122 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2124 if (num_copies > 1) {
2125 repair_io_failure(fs_info, ino, start, failrec->len,
2126 failrec->logical, page, pg_offset,
2127 failrec->failed_mirror);
2132 free_io_failure(failure_tree, io_tree, failrec);
2138 * Can be called when
2139 * - hold extent lock
2140 * - under ordered extent
2141 * - the inode is freeing
2143 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2145 struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2146 struct io_failure_record *failrec;
2147 struct extent_state *state, *next;
2149 if (RB_EMPTY_ROOT(&failure_tree->state))
2152 spin_lock(&failure_tree->lock);
2153 state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2155 if (state->start > end)
2158 ASSERT(state->end <= end);
2160 next = next_state(state);
2162 failrec = state->failrec;
2163 free_extent_state(state);
2168 spin_unlock(&failure_tree->lock);
2171 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2172 struct io_failure_record **failrec_ret)
2174 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2175 struct io_failure_record *failrec;
2176 struct extent_map *em;
2177 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2178 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2179 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2183 ret = get_state_failrec(failure_tree, start, &failrec);
2185 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2189 failrec->start = start;
2190 failrec->len = end - start + 1;
2191 failrec->this_mirror = 0;
2192 failrec->bio_flags = 0;
2193 failrec->in_validation = 0;
2195 read_lock(&em_tree->lock);
2196 em = lookup_extent_mapping(em_tree, start, failrec->len);
2198 read_unlock(&em_tree->lock);
2203 if (em->start > start || em->start + em->len <= start) {
2204 free_extent_map(em);
2207 read_unlock(&em_tree->lock);
2213 logical = start - em->start;
2214 logical = em->block_start + logical;
2215 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2216 logical = em->block_start;
2217 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2218 extent_set_compress_type(&failrec->bio_flags,
2222 btrfs_debug(fs_info,
2223 "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2224 logical, start, failrec->len);
2226 failrec->logical = logical;
2227 free_extent_map(em);
2229 /* set the bits in the private failure tree */
2230 ret = set_extent_bits(failure_tree, start, end,
2231 EXTENT_LOCKED | EXTENT_DIRTY);
2233 ret = set_state_failrec(failure_tree, start, failrec);
2234 /* set the bits in the inode's tree */
2236 ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2242 btrfs_debug(fs_info,
2243 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2244 failrec->logical, failrec->start, failrec->len,
2245 failrec->in_validation);
2247 * when data can be on disk more than twice, add to failrec here
2248 * (e.g. with a list for failed_mirror) to make
2249 * clean_io_failure() clean all those errors at once.
2253 *failrec_ret = failrec;
2258 bool btrfs_check_repairable(struct inode *inode, struct bio *failed_bio,
2259 struct io_failure_record *failrec, int failed_mirror)
2261 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2264 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2265 if (num_copies == 1) {
2267 * we only have a single copy of the data, so don't bother with
2268 * all the retry and error correction code that follows. no
2269 * matter what the error is, it is very likely to persist.
2271 btrfs_debug(fs_info,
2272 "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2273 num_copies, failrec->this_mirror, failed_mirror);
2278 * there are two premises:
2279 * a) deliver good data to the caller
2280 * b) correct the bad sectors on disk
2282 if (failed_bio->bi_vcnt > 1) {
2284 * to fulfill b), we need to know the exact failing sectors, as
2285 * we don't want to rewrite any more than the failed ones. thus,
2286 * we need separate read requests for the failed bio
2288 * if the following BUG_ON triggers, our validation request got
2289 * merged. we need separate requests for our algorithm to work.
2291 BUG_ON(failrec->in_validation);
2292 failrec->in_validation = 1;
2293 failrec->this_mirror = failed_mirror;
2296 * we're ready to fulfill a) and b) alongside. get a good copy
2297 * of the failed sector and if we succeed, we have setup
2298 * everything for repair_io_failure to do the rest for us.
2300 if (failrec->in_validation) {
2301 BUG_ON(failrec->this_mirror != failed_mirror);
2302 failrec->in_validation = 0;
2303 failrec->this_mirror = 0;
2305 failrec->failed_mirror = failed_mirror;
2306 failrec->this_mirror++;
2307 if (failrec->this_mirror == failed_mirror)
2308 failrec->this_mirror++;
2311 if (failrec->this_mirror > num_copies) {
2312 btrfs_debug(fs_info,
2313 "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2314 num_copies, failrec->this_mirror, failed_mirror);
2322 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2323 struct io_failure_record *failrec,
2324 struct page *page, int pg_offset, int icsum,
2325 bio_end_io_t *endio_func, void *data)
2327 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2329 struct btrfs_io_bio *btrfs_failed_bio;
2330 struct btrfs_io_bio *btrfs_bio;
2332 bio = btrfs_io_bio_alloc(1);
2333 bio->bi_end_io = endio_func;
2334 bio->bi_iter.bi_sector = failrec->logical >> 9;
2335 bio_set_dev(bio, fs_info->fs_devices->latest_bdev);
2336 bio->bi_iter.bi_size = 0;
2337 bio->bi_private = data;
2339 btrfs_failed_bio = btrfs_io_bio(failed_bio);
2340 if (btrfs_failed_bio->csum) {
2341 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2343 btrfs_bio = btrfs_io_bio(bio);
2344 btrfs_bio->csum = btrfs_bio->csum_inline;
2346 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2350 bio_add_page(bio, page, failrec->len, pg_offset);
2356 * this is a generic handler for readpage errors (default
2357 * readpage_io_failed_hook). if other copies exist, read those and write back
2358 * good data to the failed position. does not investigate in remapping the
2359 * failed extent elsewhere, hoping the device will be smart enough to do this as
2363 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2364 struct page *page, u64 start, u64 end,
2367 struct io_failure_record *failrec;
2368 struct inode *inode = page->mapping->host;
2369 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2370 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2373 blk_status_t status;
2376 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2378 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2382 if (!btrfs_check_repairable(inode, failed_bio, failrec,
2384 free_io_failure(failure_tree, tree, failrec);
2388 if (failed_bio->bi_vcnt > 1)
2389 read_mode |= REQ_FAILFAST_DEV;
2391 phy_offset >>= inode->i_sb->s_blocksize_bits;
2392 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2393 start - page_offset(page),
2394 (int)phy_offset, failed_bio->bi_end_io,
2396 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
2398 btrfs_debug(btrfs_sb(inode->i_sb),
2399 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2400 read_mode, failrec->this_mirror, failrec->in_validation);
2402 status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror,
2403 failrec->bio_flags, 0);
2405 free_io_failure(failure_tree, tree, failrec);
2407 ret = blk_status_to_errno(status);
2413 /* lots and lots of room for performance fixes in the end_bio funcs */
2415 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2417 int uptodate = (err == 0);
2418 struct extent_io_tree *tree;
2421 tree = &BTRFS_I(page->mapping->host)->io_tree;
2423 if (tree->ops && tree->ops->writepage_end_io_hook)
2424 tree->ops->writepage_end_io_hook(page, start, end, NULL,
2428 ClearPageUptodate(page);
2430 ret = err < 0 ? err : -EIO;
2431 mapping_set_error(page->mapping, ret);
2436 * after a writepage IO is done, we need to:
2437 * clear the uptodate bits on error
2438 * clear the writeback bits in the extent tree for this IO
2439 * end_page_writeback if the page has no more pending IO
2441 * Scheduling is not allowed, so the extent state tree is expected
2442 * to have one and only one object corresponding to this IO.
2444 static void end_bio_extent_writepage(struct bio *bio)
2446 int error = blk_status_to_errno(bio->bi_status);
2447 struct bio_vec *bvec;
2452 ASSERT(!bio_flagged(bio, BIO_CLONED));
2453 bio_for_each_segment_all(bvec, bio, i) {
2454 struct page *page = bvec->bv_page;
2455 struct inode *inode = page->mapping->host;
2456 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2458 /* We always issue full-page reads, but if some block
2459 * in a page fails to read, blk_update_request() will
2460 * advance bv_offset and adjust bv_len to compensate.
2461 * Print a warning for nonzero offsets, and an error
2462 * if they don't add up to a full page. */
2463 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2464 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2466 "partial page write in btrfs with offset %u and length %u",
2467 bvec->bv_offset, bvec->bv_len);
2470 "incomplete page write in btrfs with offset %u and length %u",
2471 bvec->bv_offset, bvec->bv_len);
2474 start = page_offset(page);
2475 end = start + bvec->bv_offset + bvec->bv_len - 1;
2477 end_extent_writepage(page, error, start, end);
2478 end_page_writeback(page);
2485 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2488 struct extent_state *cached = NULL;
2489 u64 end = start + len - 1;
2491 if (uptodate && tree->track_uptodate)
2492 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2493 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2497 * after a readpage IO is done, we need to:
2498 * clear the uptodate bits on error
2499 * set the uptodate bits if things worked
2500 * set the page up to date if all extents in the tree are uptodate
2501 * clear the lock bit in the extent tree
2502 * unlock the page if there are no other extents locked for it
2504 * Scheduling is not allowed, so the extent state tree is expected
2505 * to have one and only one object corresponding to this IO.
2507 static void end_bio_extent_readpage(struct bio *bio)
2509 struct bio_vec *bvec;
2510 int uptodate = !bio->bi_status;
2511 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2512 struct extent_io_tree *tree, *failure_tree;
2517 u64 extent_start = 0;
2523 ASSERT(!bio_flagged(bio, BIO_CLONED));
2524 bio_for_each_segment_all(bvec, bio, i) {
2525 struct page *page = bvec->bv_page;
2526 struct inode *inode = page->mapping->host;
2527 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2529 btrfs_debug(fs_info,
2530 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2531 (u64)bio->bi_iter.bi_sector, bio->bi_status,
2532 io_bio->mirror_num);
2533 tree = &BTRFS_I(inode)->io_tree;
2534 failure_tree = &BTRFS_I(inode)->io_failure_tree;
2536 /* We always issue full-page reads, but if some block
2537 * in a page fails to read, blk_update_request() will
2538 * advance bv_offset and adjust bv_len to compensate.
2539 * Print a warning for nonzero offsets, and an error
2540 * if they don't add up to a full page. */
2541 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2542 if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2544 "partial page read in btrfs with offset %u and length %u",
2545 bvec->bv_offset, bvec->bv_len);
2548 "incomplete page read in btrfs with offset %u and length %u",
2549 bvec->bv_offset, bvec->bv_len);
2552 start = page_offset(page);
2553 end = start + bvec->bv_offset + bvec->bv_len - 1;
2556 mirror = io_bio->mirror_num;
2557 if (likely(uptodate && tree->ops)) {
2558 ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2564 clean_io_failure(BTRFS_I(inode)->root->fs_info,
2565 failure_tree, tree, start,
2567 btrfs_ino(BTRFS_I(inode)), 0);
2570 if (likely(uptodate))
2574 ret = tree->ops->readpage_io_failed_hook(page, mirror);
2575 if (ret == -EAGAIN) {
2577 * Data inode's readpage_io_failed_hook() always
2580 * The generic bio_readpage_error handles errors
2581 * the following way: If possible, new read
2582 * requests are created and submitted and will
2583 * end up in end_bio_extent_readpage as well (if
2584 * we're lucky, not in the !uptodate case). In
2585 * that case it returns 0 and we just go on with
2586 * the next page in our bio. If it can't handle
2587 * the error it will return -EIO and we remain
2588 * responsible for that page.
2590 ret = bio_readpage_error(bio, offset, page,
2591 start, end, mirror);
2593 uptodate = !bio->bi_status;
2600 * metadata's readpage_io_failed_hook() always returns
2601 * -EIO and fixes nothing. -EIO is also returned if
2602 * data inode error could not be fixed.
2604 ASSERT(ret == -EIO);
2607 if (likely(uptodate)) {
2608 loff_t i_size = i_size_read(inode);
2609 pgoff_t end_index = i_size >> PAGE_SHIFT;
2612 /* Zero out the end if this page straddles i_size */
2613 off = i_size & (PAGE_SIZE-1);
2614 if (page->index == end_index && off)
2615 zero_user_segment(page, off, PAGE_SIZE);
2616 SetPageUptodate(page);
2618 ClearPageUptodate(page);
2624 if (unlikely(!uptodate)) {
2626 endio_readpage_release_extent(tree,
2632 endio_readpage_release_extent(tree, start,
2633 end - start + 1, 0);
2634 } else if (!extent_len) {
2635 extent_start = start;
2636 extent_len = end + 1 - start;
2637 } else if (extent_start + extent_len == start) {
2638 extent_len += end + 1 - start;
2640 endio_readpage_release_extent(tree, extent_start,
2641 extent_len, uptodate);
2642 extent_start = start;
2643 extent_len = end + 1 - start;
2648 endio_readpage_release_extent(tree, extent_start, extent_len,
2651 io_bio->end_io(io_bio, blk_status_to_errno(bio->bi_status));
2656 * Initialize the members up to but not including 'bio'. Use after allocating a
2657 * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2658 * 'bio' because use of __GFP_ZERO is not supported.
2660 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
2662 memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
2666 * The following helpers allocate a bio. As it's backed by a bioset, it'll
2667 * never fail. We're returning a bio right now but you can call btrfs_io_bio
2668 * for the appropriate container_of magic
2670 struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte)
2674 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, btrfs_bioset);
2675 bio_set_dev(bio, bdev);
2676 bio->bi_iter.bi_sector = first_byte >> 9;
2677 btrfs_io_bio_init(btrfs_io_bio(bio));
2681 struct bio *btrfs_bio_clone(struct bio *bio)
2683 struct btrfs_io_bio *btrfs_bio;
2686 /* Bio allocation backed by a bioset does not fail */
2687 new = bio_clone_fast(bio, GFP_NOFS, btrfs_bioset);
2688 btrfs_bio = btrfs_io_bio(new);
2689 btrfs_io_bio_init(btrfs_bio);
2690 btrfs_bio->iter = bio->bi_iter;
2694 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
2698 /* Bio allocation backed by a bioset does not fail */
2699 bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, btrfs_bioset);
2700 btrfs_io_bio_init(btrfs_io_bio(bio));
2704 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
2707 struct btrfs_io_bio *btrfs_bio;
2709 /* this will never fail when it's backed by a bioset */
2710 bio = bio_clone_fast(orig, GFP_NOFS, btrfs_bioset);
2713 btrfs_bio = btrfs_io_bio(bio);
2714 btrfs_io_bio_init(btrfs_bio);
2716 bio_trim(bio, offset >> 9, size >> 9);
2717 btrfs_bio->iter = bio->bi_iter;
2721 static int __must_check submit_one_bio(struct bio *bio, int mirror_num,
2722 unsigned long bio_flags)
2724 blk_status_t ret = 0;
2725 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2726 struct page *page = bvec->bv_page;
2727 struct extent_io_tree *tree = bio->bi_private;
2730 start = page_offset(page) + bvec->bv_offset;
2732 bio->bi_private = NULL;
2736 ret = tree->ops->submit_bio_hook(tree->private_data, bio,
2737 mirror_num, bio_flags, start);
2739 btrfsic_submit_bio(bio);
2742 return blk_status_to_errno(ret);
2745 static int merge_bio(struct extent_io_tree *tree, struct page *page,
2746 unsigned long offset, size_t size, struct bio *bio,
2747 unsigned long bio_flags)
2751 ret = tree->ops->merge_bio_hook(page, offset, size, bio,
2758 * @opf: bio REQ_OP_* and REQ_* flags as one value
2760 static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree,
2761 struct writeback_control *wbc,
2762 struct page *page, u64 offset,
2763 size_t size, unsigned long pg_offset,
2764 struct block_device *bdev,
2765 struct bio **bio_ret,
2766 bio_end_io_t end_io_func,
2768 unsigned long prev_bio_flags,
2769 unsigned long bio_flags,
2770 bool force_bio_submit)
2775 int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2776 size_t page_size = min_t(size_t, size, PAGE_SIZE);
2777 sector_t sector = offset >> 9;
2779 if (bio_ret && *bio_ret) {
2782 contig = bio->bi_iter.bi_sector == sector;
2784 contig = bio_end_sector(bio) == sector;
2786 if (prev_bio_flags != bio_flags || !contig ||
2788 merge_bio(tree, page, pg_offset, page_size, bio, bio_flags) ||
2789 bio_add_page(bio, page, page_size, pg_offset) < page_size) {
2790 ret = submit_one_bio(bio, mirror_num, prev_bio_flags);
2798 wbc_account_io(wbc, page, page_size);
2803 bio = btrfs_bio_alloc(bdev, offset);
2804 bio_add_page(bio, page, page_size, pg_offset);
2805 bio->bi_end_io = end_io_func;
2806 bio->bi_private = tree;
2807 bio->bi_write_hint = page->mapping->host->i_write_hint;
2810 wbc_init_bio(wbc, bio);
2811 wbc_account_io(wbc, page, page_size);
2817 ret = submit_one_bio(bio, mirror_num, bio_flags);
2822 static void attach_extent_buffer_page(struct extent_buffer *eb,
2825 if (!PagePrivate(page)) {
2826 SetPagePrivate(page);
2828 set_page_private(page, (unsigned long)eb);
2830 WARN_ON(page->private != (unsigned long)eb);
2834 void set_page_extent_mapped(struct page *page)
2836 if (!PagePrivate(page)) {
2837 SetPagePrivate(page);
2839 set_page_private(page, EXTENT_PAGE_PRIVATE);
2843 static struct extent_map *
2844 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2845 u64 start, u64 len, get_extent_t *get_extent,
2846 struct extent_map **em_cached)
2848 struct extent_map *em;
2850 if (em_cached && *em_cached) {
2852 if (extent_map_in_tree(em) && start >= em->start &&
2853 start < extent_map_end(em)) {
2854 refcount_inc(&em->refs);
2858 free_extent_map(em);
2862 em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0);
2863 if (em_cached && !IS_ERR_OR_NULL(em)) {
2865 refcount_inc(&em->refs);
2871 * basic readpage implementation. Locked extent state structs are inserted
2872 * into the tree that are removed when the IO is done (by the end_io
2874 * XXX JDM: This needs looking at to ensure proper page locking
2875 * return 0 on success, otherwise return error
2877 static int __do_readpage(struct extent_io_tree *tree,
2879 get_extent_t *get_extent,
2880 struct extent_map **em_cached,
2881 struct bio **bio, int mirror_num,
2882 unsigned long *bio_flags, unsigned int read_flags,
2885 struct inode *inode = page->mapping->host;
2886 u64 start = page_offset(page);
2887 u64 page_end = start + PAGE_SIZE - 1;
2891 u64 last_byte = i_size_read(inode);
2894 struct extent_map *em;
2895 struct block_device *bdev;
2898 size_t pg_offset = 0;
2900 size_t disk_io_size;
2901 size_t blocksize = inode->i_sb->s_blocksize;
2902 unsigned long this_bio_flag = 0;
2904 set_page_extent_mapped(page);
2907 if (!PageUptodate(page)) {
2908 if (cleancache_get_page(page) == 0) {
2909 BUG_ON(blocksize != PAGE_SIZE);
2910 unlock_extent(tree, start, end);
2915 if (page->index == last_byte >> PAGE_SHIFT) {
2917 size_t zero_offset = last_byte & (PAGE_SIZE - 1);
2920 iosize = PAGE_SIZE - zero_offset;
2921 userpage = kmap_atomic(page);
2922 memset(userpage + zero_offset, 0, iosize);
2923 flush_dcache_page(page);
2924 kunmap_atomic(userpage);
2927 while (cur <= end) {
2928 bool force_bio_submit = false;
2931 if (cur >= last_byte) {
2933 struct extent_state *cached = NULL;
2935 iosize = PAGE_SIZE - pg_offset;
2936 userpage = kmap_atomic(page);
2937 memset(userpage + pg_offset, 0, iosize);
2938 flush_dcache_page(page);
2939 kunmap_atomic(userpage);
2940 set_extent_uptodate(tree, cur, cur + iosize - 1,
2942 unlock_extent_cached(tree, cur,
2947 em = __get_extent_map(inode, page, pg_offset, cur,
2948 end - cur + 1, get_extent, em_cached);
2949 if (IS_ERR_OR_NULL(em)) {
2951 unlock_extent(tree, cur, end);
2954 extent_offset = cur - em->start;
2955 BUG_ON(extent_map_end(em) <= cur);
2958 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2959 this_bio_flag |= EXTENT_BIO_COMPRESSED;
2960 extent_set_compress_type(&this_bio_flag,
2964 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2965 cur_end = min(extent_map_end(em) - 1, end);
2966 iosize = ALIGN(iosize, blocksize);
2967 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2968 disk_io_size = em->block_len;
2969 offset = em->block_start;
2971 offset = em->block_start + extent_offset;
2972 disk_io_size = iosize;
2975 block_start = em->block_start;
2976 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2977 block_start = EXTENT_MAP_HOLE;
2980 * If we have a file range that points to a compressed extent
2981 * and it's followed by a consecutive file range that points to
2982 * to the same compressed extent (possibly with a different
2983 * offset and/or length, so it either points to the whole extent
2984 * or only part of it), we must make sure we do not submit a
2985 * single bio to populate the pages for the 2 ranges because
2986 * this makes the compressed extent read zero out the pages
2987 * belonging to the 2nd range. Imagine the following scenario:
2990 * [0 - 8K] [8K - 24K]
2993 * points to extent X, points to extent X,
2994 * offset 4K, length of 8K offset 0, length 16K
2996 * [extent X, compressed length = 4K uncompressed length = 16K]
2998 * If the bio to read the compressed extent covers both ranges,
2999 * it will decompress extent X into the pages belonging to the
3000 * first range and then it will stop, zeroing out the remaining
3001 * pages that belong to the other range that points to extent X.
3002 * So here we make sure we submit 2 bios, one for the first
3003 * range and another one for the third range. Both will target
3004 * the same physical extent from disk, but we can't currently
3005 * make the compressed bio endio callback populate the pages
3006 * for both ranges because each compressed bio is tightly
3007 * coupled with a single extent map, and each range can have
3008 * an extent map with a different offset value relative to the
3009 * uncompressed data of our extent and different lengths. This
3010 * is a corner case so we prioritize correctness over
3011 * non-optimal behavior (submitting 2 bios for the same extent).
3013 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3014 prev_em_start && *prev_em_start != (u64)-1 &&
3015 *prev_em_start != em->orig_start)
3016 force_bio_submit = true;
3019 *prev_em_start = em->orig_start;
3021 free_extent_map(em);
3024 /* we've found a hole, just zero and go on */
3025 if (block_start == EXTENT_MAP_HOLE) {
3027 struct extent_state *cached = NULL;
3029 userpage = kmap_atomic(page);
3030 memset(userpage + pg_offset, 0, iosize);
3031 flush_dcache_page(page);
3032 kunmap_atomic(userpage);
3034 set_extent_uptodate(tree, cur, cur + iosize - 1,
3036 unlock_extent_cached(tree, cur,
3040 pg_offset += iosize;
3043 /* the get_extent function already copied into the page */
3044 if (test_range_bit(tree, cur, cur_end,
3045 EXTENT_UPTODATE, 1, NULL)) {
3046 check_page_uptodate(tree, page);
3047 unlock_extent(tree, cur, cur + iosize - 1);
3049 pg_offset += iosize;
3052 /* we have an inline extent but it didn't get marked up
3053 * to date. Error out
3055 if (block_start == EXTENT_MAP_INLINE) {
3057 unlock_extent(tree, cur, cur + iosize - 1);
3059 pg_offset += iosize;
3063 ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL,
3064 page, offset, disk_io_size,
3065 pg_offset, bdev, bio,
3066 end_bio_extent_readpage, mirror_num,
3072 *bio_flags = this_bio_flag;
3075 unlock_extent(tree, cur, cur + iosize - 1);
3079 pg_offset += iosize;
3083 if (!PageError(page))
3084 SetPageUptodate(page);
3090 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
3091 struct page *pages[], int nr_pages,
3093 get_extent_t *get_extent,
3094 struct extent_map **em_cached,
3096 unsigned long *bio_flags,
3099 struct inode *inode;
3100 struct btrfs_ordered_extent *ordered;
3103 inode = pages[0]->mapping->host;
3105 lock_extent(tree, start, end);
3106 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3110 unlock_extent(tree, start, end);
3111 btrfs_start_ordered_extent(inode, ordered, 1);
3112 btrfs_put_ordered_extent(ordered);
3115 for (index = 0; index < nr_pages; index++) {
3116 __do_readpage(tree, pages[index], get_extent, em_cached, bio,
3117 0, bio_flags, 0, prev_em_start);
3118 put_page(pages[index]);
3122 static void __extent_readpages(struct extent_io_tree *tree,
3123 struct page *pages[],
3124 int nr_pages, get_extent_t *get_extent,
3125 struct extent_map **em_cached,
3126 struct bio **bio, unsigned long *bio_flags,
3133 int first_index = 0;
3135 for (index = 0; index < nr_pages; index++) {
3136 page_start = page_offset(pages[index]);
3139 end = start + PAGE_SIZE - 1;
3140 first_index = index;
3141 } else if (end + 1 == page_start) {
3144 __do_contiguous_readpages(tree, &pages[first_index],
3145 index - first_index, start,
3146 end, get_extent, em_cached,
3150 end = start + PAGE_SIZE - 1;
3151 first_index = index;
3156 __do_contiguous_readpages(tree, &pages[first_index],
3157 index - first_index, start,
3158 end, get_extent, em_cached, bio,
3159 bio_flags, prev_em_start);
3162 static int __extent_read_full_page(struct extent_io_tree *tree,
3164 get_extent_t *get_extent,
3165 struct bio **bio, int mirror_num,
3166 unsigned long *bio_flags,
3167 unsigned int read_flags)
3169 struct inode *inode = page->mapping->host;
3170 struct btrfs_ordered_extent *ordered;
3171 u64 start = page_offset(page);
3172 u64 end = start + PAGE_SIZE - 1;
3176 lock_extent(tree, start, end);
3177 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3181 unlock_extent(tree, start, end);
3182 btrfs_start_ordered_extent(inode, ordered, 1);
3183 btrfs_put_ordered_extent(ordered);
3186 ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3187 bio_flags, read_flags, NULL);
3191 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3192 get_extent_t *get_extent, int mirror_num)
3194 struct bio *bio = NULL;
3195 unsigned long bio_flags = 0;
3198 ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3201 ret = submit_one_bio(bio, mirror_num, bio_flags);
3205 static void update_nr_written(struct writeback_control *wbc,
3206 unsigned long nr_written)
3208 wbc->nr_to_write -= nr_written;
3212 * helper for __extent_writepage, doing all of the delayed allocation setup.
3214 * This returns 1 if our fill_delalloc function did all the work required
3215 * to write the page (copy into inline extent). In this case the IO has
3216 * been started and the page is already unlocked.
3218 * This returns 0 if all went well (page still locked)
3219 * This returns < 0 if there were errors (page still locked)
3221 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3222 struct page *page, struct writeback_control *wbc,
3223 struct extent_page_data *epd,
3225 unsigned long *nr_written)
3227 struct extent_io_tree *tree = epd->tree;
3228 u64 page_end = delalloc_start + PAGE_SIZE - 1;
3230 u64 delalloc_to_write = 0;
3231 u64 delalloc_end = 0;
3233 int page_started = 0;
3235 if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc)
3238 while (delalloc_end < page_end) {
3239 nr_delalloc = find_lock_delalloc_range(inode, tree,
3243 BTRFS_MAX_EXTENT_SIZE);
3244 if (nr_delalloc == 0) {
3245 delalloc_start = delalloc_end + 1;
3248 ret = tree->ops->fill_delalloc(inode, page,
3253 /* File system has been set read-only */
3256 /* fill_delalloc should be return < 0 for error
3257 * but just in case, we use > 0 here meaning the
3258 * IO is started, so we don't want to return > 0
3259 * unless things are going well.
3261 ret = ret < 0 ? ret : -EIO;
3265 * delalloc_end is already one less than the total length, so
3266 * we don't subtract one from PAGE_SIZE
3268 delalloc_to_write += (delalloc_end - delalloc_start +
3269 PAGE_SIZE) >> PAGE_SHIFT;
3270 delalloc_start = delalloc_end + 1;
3272 if (wbc->nr_to_write < delalloc_to_write) {
3275 if (delalloc_to_write < thresh * 2)
3276 thresh = delalloc_to_write;
3277 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3281 /* did the fill delalloc function already unlock and start
3286 * we've unlocked the page, so we can't update
3287 * the mapping's writeback index, just update
3290 wbc->nr_to_write -= *nr_written;
3301 * helper for __extent_writepage. This calls the writepage start hooks,
3302 * and does the loop to map the page into extents and bios.
3304 * We return 1 if the IO is started and the page is unlocked,
3305 * 0 if all went well (page still locked)
3306 * < 0 if there were errors (page still locked)
3308 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3310 struct writeback_control *wbc,
3311 struct extent_page_data *epd,
3313 unsigned long nr_written,
3314 unsigned int write_flags, int *nr_ret)
3316 struct extent_io_tree *tree = epd->tree;
3317 u64 start = page_offset(page);
3318 u64 page_end = start + PAGE_SIZE - 1;
3324 struct extent_map *em;
3325 struct block_device *bdev;
3326 size_t pg_offset = 0;
3332 if (tree->ops && tree->ops->writepage_start_hook) {
3333 ret = tree->ops->writepage_start_hook(page, start,
3336 /* Fixup worker will requeue */
3338 wbc->pages_skipped++;
3340 redirty_page_for_writepage(wbc, page);
3342 update_nr_written(wbc, nr_written);
3349 * we don't want to touch the inode after unlocking the page,
3350 * so we update the mapping writeback index now
3352 update_nr_written(wbc, nr_written + 1);
3355 if (i_size <= start) {
3356 if (tree->ops && tree->ops->writepage_end_io_hook)
3357 tree->ops->writepage_end_io_hook(page, start,
3362 blocksize = inode->i_sb->s_blocksize;
3364 while (cur <= end) {
3368 if (cur >= i_size) {
3369 if (tree->ops && tree->ops->writepage_end_io_hook)
3370 tree->ops->writepage_end_io_hook(page, cur,
3374 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, cur,
3376 if (IS_ERR_OR_NULL(em)) {
3378 ret = PTR_ERR_OR_ZERO(em);
3382 extent_offset = cur - em->start;
3383 em_end = extent_map_end(em);
3384 BUG_ON(em_end <= cur);
3386 iosize = min(em_end - cur, end - cur + 1);
3387 iosize = ALIGN(iosize, blocksize);
3388 offset = em->block_start + extent_offset;
3390 block_start = em->block_start;
3391 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3392 free_extent_map(em);
3396 * compressed and inline extents are written through other
3399 if (compressed || block_start == EXTENT_MAP_HOLE ||
3400 block_start == EXTENT_MAP_INLINE) {
3402 * end_io notification does not happen here for
3403 * compressed extents
3405 if (!compressed && tree->ops &&
3406 tree->ops->writepage_end_io_hook)
3407 tree->ops->writepage_end_io_hook(page, cur,
3410 else if (compressed) {
3411 /* we don't want to end_page_writeback on
3412 * a compressed extent. this happens
3419 pg_offset += iosize;
3423 set_range_writeback(tree, cur, cur + iosize - 1);
3424 if (!PageWriteback(page)) {
3425 btrfs_err(BTRFS_I(inode)->root->fs_info,
3426 "page %lu not writeback, cur %llu end %llu",
3427 page->index, cur, end);
3430 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3431 page, offset, iosize, pg_offset,
3433 end_bio_extent_writepage,
3437 if (PageWriteback(page))
3438 end_page_writeback(page);
3442 pg_offset += iosize;
3451 * the writepage semantics are similar to regular writepage. extent
3452 * records are inserted to lock ranges in the tree, and as dirty areas
3453 * are found, they are marked writeback. Then the lock bits are removed
3454 * and the end_io handler clears the writeback ranges
3456 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3459 struct inode *inode = page->mapping->host;
3460 struct extent_page_data *epd = data;
3461 u64 start = page_offset(page);
3462 u64 page_end = start + PAGE_SIZE - 1;
3465 size_t pg_offset = 0;
3466 loff_t i_size = i_size_read(inode);
3467 unsigned long end_index = i_size >> PAGE_SHIFT;
3468 unsigned int write_flags = 0;
3469 unsigned long nr_written = 0;
3471 write_flags = wbc_to_write_flags(wbc);
3473 trace___extent_writepage(page, inode, wbc);
3475 WARN_ON(!PageLocked(page));
3477 ClearPageError(page);
3479 pg_offset = i_size & (PAGE_SIZE - 1);
3480 if (page->index > end_index ||
3481 (page->index == end_index && !pg_offset)) {
3482 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3487 if (page->index == end_index) {
3490 userpage = kmap_atomic(page);
3491 memset(userpage + pg_offset, 0,
3492 PAGE_SIZE - pg_offset);
3493 kunmap_atomic(userpage);
3494 flush_dcache_page(page);
3499 set_page_extent_mapped(page);
3501 ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written);
3507 ret = __extent_writepage_io(inode, page, wbc, epd,
3508 i_size, nr_written, write_flags, &nr);
3514 /* make sure the mapping tag for page dirty gets cleared */
3515 set_page_writeback(page);
3516 end_page_writeback(page);
3518 if (PageError(page)) {
3519 ret = ret < 0 ? ret : -EIO;
3520 end_extent_writepage(page, ret, start, page_end);
3529 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3531 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3532 TASK_UNINTERRUPTIBLE);
3535 static noinline_for_stack int
3536 lock_extent_buffer_for_io(struct extent_buffer *eb,
3537 struct btrfs_fs_info *fs_info,
3538 struct extent_page_data *epd)
3540 unsigned long i, num_pages;
3544 if (!btrfs_try_tree_write_lock(eb)) {
3546 flush_write_bio(epd);
3547 btrfs_tree_lock(eb);
3550 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3551 btrfs_tree_unlock(eb);
3555 flush_write_bio(epd);
3559 wait_on_extent_buffer_writeback(eb);
3560 btrfs_tree_lock(eb);
3561 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3563 btrfs_tree_unlock(eb);
3568 * We need to do this to prevent races in people who check if the eb is
3569 * under IO since we can end up having no IO bits set for a short period
3572 spin_lock(&eb->refs_lock);
3573 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3574 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3575 spin_unlock(&eb->refs_lock);
3576 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3577 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3579 fs_info->dirty_metadata_batch);
3582 spin_unlock(&eb->refs_lock);
3585 btrfs_tree_unlock(eb);
3590 num_pages = num_extent_pages(eb->start, eb->len);
3591 for (i = 0; i < num_pages; i++) {
3592 struct page *p = eb->pages[i];
3594 if (!trylock_page(p)) {
3596 flush_write_bio(epd);
3606 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3608 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3609 smp_mb__after_atomic();
3610 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3613 static void set_btree_ioerr(struct page *page)
3615 struct extent_buffer *eb = (struct extent_buffer *)page->private;
3618 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3622 * If writeback for a btree extent that doesn't belong to a log tree
3623 * failed, increment the counter transaction->eb_write_errors.
3624 * We do this because while the transaction is running and before it's
3625 * committing (when we call filemap_fdata[write|wait]_range against
3626 * the btree inode), we might have
3627 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3628 * returns an error or an error happens during writeback, when we're
3629 * committing the transaction we wouldn't know about it, since the pages
3630 * can be no longer dirty nor marked anymore for writeback (if a
3631 * subsequent modification to the extent buffer didn't happen before the
3632 * transaction commit), which makes filemap_fdata[write|wait]_range not
3633 * able to find the pages tagged with SetPageError at transaction
3634 * commit time. So if this happens we must abort the transaction,
3635 * otherwise we commit a super block with btree roots that point to
3636 * btree nodes/leafs whose content on disk is invalid - either garbage
3637 * or the content of some node/leaf from a past generation that got
3638 * cowed or deleted and is no longer valid.
3640 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3641 * not be enough - we need to distinguish between log tree extents vs
3642 * non-log tree extents, and the next filemap_fdatawait_range() call
3643 * will catch and clear such errors in the mapping - and that call might
3644 * be from a log sync and not from a transaction commit. Also, checking
3645 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3646 * not done and would not be reliable - the eb might have been released
3647 * from memory and reading it back again means that flag would not be
3648 * set (since it's a runtime flag, not persisted on disk).
3650 * Using the flags below in the btree inode also makes us achieve the
3651 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3652 * writeback for all dirty pages and before filemap_fdatawait_range()
3653 * is called, the writeback for all dirty pages had already finished
3654 * with errors - because we were not using AS_EIO/AS_ENOSPC,
3655 * filemap_fdatawait_range() would return success, as it could not know
3656 * that writeback errors happened (the pages were no longer tagged for
3659 switch (eb->log_index) {
3661 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3664 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3667 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3670 BUG(); /* unexpected, logic error */
3674 static void end_bio_extent_buffer_writepage(struct bio *bio)
3676 struct bio_vec *bvec;
3677 struct extent_buffer *eb;
3680 ASSERT(!bio_flagged(bio, BIO_CLONED));
3681 bio_for_each_segment_all(bvec, bio, i) {
3682 struct page *page = bvec->bv_page;
3684 eb = (struct extent_buffer *)page->private;
3686 done = atomic_dec_and_test(&eb->io_pages);
3688 if (bio->bi_status ||
3689 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3690 ClearPageUptodate(page);
3691 set_btree_ioerr(page);
3694 end_page_writeback(page);
3699 end_extent_buffer_writeback(eb);
3705 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3706 struct btrfs_fs_info *fs_info,
3707 struct writeback_control *wbc,
3708 struct extent_page_data *epd)
3710 struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3711 struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3712 u64 offset = eb->start;
3714 unsigned long i, num_pages;
3715 unsigned long start, end;
3716 unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
3719 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3720 num_pages = num_extent_pages(eb->start, eb->len);
3721 atomic_set(&eb->io_pages, num_pages);
3723 /* set btree blocks beyond nritems with 0 to avoid stale content. */
3724 nritems = btrfs_header_nritems(eb);
3725 if (btrfs_header_level(eb) > 0) {
3726 end = btrfs_node_key_ptr_offset(nritems);
3728 memzero_extent_buffer(eb, end, eb->len - end);
3732 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3734 start = btrfs_item_nr_offset(nritems);
3735 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, eb);
3736 memzero_extent_buffer(eb, start, end - start);
3739 for (i = 0; i < num_pages; i++) {
3740 struct page *p = eb->pages[i];
3742 clear_page_dirty_for_io(p);
3743 set_page_writeback(p);
3744 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3745 p, offset, PAGE_SIZE, 0, bdev,
3747 end_bio_extent_buffer_writepage,
3751 if (PageWriteback(p))
3752 end_page_writeback(p);
3753 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3754 end_extent_buffer_writeback(eb);
3758 offset += PAGE_SIZE;
3759 update_nr_written(wbc, 1);
3763 if (unlikely(ret)) {
3764 for (; i < num_pages; i++) {
3765 struct page *p = eb->pages[i];
3766 clear_page_dirty_for_io(p);
3774 int btree_write_cache_pages(struct address_space *mapping,
3775 struct writeback_control *wbc)
3777 struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3778 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3779 struct extent_buffer *eb, *prev_eb = NULL;
3780 struct extent_page_data epd = {
3784 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3788 int nr_to_write_done = 0;
3789 struct pagevec pvec;
3792 pgoff_t end; /* Inclusive */
3796 pagevec_init(&pvec);
3797 if (wbc->range_cyclic) {
3798 index = mapping->writeback_index; /* Start from prev offset */
3801 index = wbc->range_start >> PAGE_SHIFT;
3802 end = wbc->range_end >> PAGE_SHIFT;
3805 if (wbc->sync_mode == WB_SYNC_ALL)
3806 tag = PAGECACHE_TAG_TOWRITE;
3808 tag = PAGECACHE_TAG_DIRTY;
3810 if (wbc->sync_mode == WB_SYNC_ALL)
3811 tag_pages_for_writeback(mapping, index, end);
3812 while (!done && !nr_to_write_done && (index <= end) &&
3813 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
3818 for (i = 0; i < nr_pages; i++) {
3819 struct page *page = pvec.pages[i];
3821 if (!PagePrivate(page))
3824 spin_lock(&mapping->private_lock);
3825 if (!PagePrivate(page)) {
3826 spin_unlock(&mapping->private_lock);
3830 eb = (struct extent_buffer *)page->private;
3833 * Shouldn't happen and normally this would be a BUG_ON
3834 * but no sense in crashing the users box for something
3835 * we can survive anyway.
3838 spin_unlock(&mapping->private_lock);
3842 if (eb == prev_eb) {
3843 spin_unlock(&mapping->private_lock);
3847 ret = atomic_inc_not_zero(&eb->refs);
3848 spin_unlock(&mapping->private_lock);
3853 ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3855 free_extent_buffer(eb);
3859 ret = write_one_eb(eb, fs_info, wbc, &epd);
3862 free_extent_buffer(eb);
3865 free_extent_buffer(eb);
3868 * the filesystem may choose to bump up nr_to_write.
3869 * We have to make sure to honor the new nr_to_write
3872 nr_to_write_done = wbc->nr_to_write <= 0;
3874 pagevec_release(&pvec);
3877 if (!scanned && !done) {
3879 * We hit the last page and there is more work to be done: wrap
3880 * back to the start of the file
3886 flush_write_bio(&epd);
3891 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3892 * @mapping: address space structure to write
3893 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3894 * @writepage: function called for each page
3895 * @data: data passed to writepage function
3897 * If a page is already under I/O, write_cache_pages() skips it, even
3898 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
3899 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
3900 * and msync() need to guarantee that all the data which was dirty at the time
3901 * the call was made get new I/O started against them. If wbc->sync_mode is
3902 * WB_SYNC_ALL then we were called for data integrity and we must wait for
3903 * existing IO to complete.
3905 static int extent_write_cache_pages(struct address_space *mapping,
3906 struct writeback_control *wbc,
3907 writepage_t writepage, void *data,
3908 void (*flush_fn)(void *))
3910 struct inode *inode = mapping->host;
3913 int nr_to_write_done = 0;
3914 struct pagevec pvec;
3917 pgoff_t end; /* Inclusive */
3919 int range_whole = 0;
3924 * We have to hold onto the inode so that ordered extents can do their
3925 * work when the IO finishes. The alternative to this is failing to add
3926 * an ordered extent if the igrab() fails there and that is a huge pain
3927 * to deal with, so instead just hold onto the inode throughout the
3928 * writepages operation. If it fails here we are freeing up the inode
3929 * anyway and we'd rather not waste our time writing out stuff that is
3930 * going to be truncated anyway.
3935 pagevec_init(&pvec);
3936 if (wbc->range_cyclic) {
3937 index = mapping->writeback_index; /* Start from prev offset */
3940 index = wbc->range_start >> PAGE_SHIFT;
3941 end = wbc->range_end >> PAGE_SHIFT;
3942 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3946 if (wbc->sync_mode == WB_SYNC_ALL)
3947 tag = PAGECACHE_TAG_TOWRITE;
3949 tag = PAGECACHE_TAG_DIRTY;
3951 if (wbc->sync_mode == WB_SYNC_ALL)
3952 tag_pages_for_writeback(mapping, index, end);
3954 while (!done && !nr_to_write_done && (index <= end) &&
3955 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
3956 &index, end, tag))) {
3960 for (i = 0; i < nr_pages; i++) {
3961 struct page *page = pvec.pages[i];
3963 done_index = page->index;
3965 * At this point we hold neither mapping->tree_lock nor
3966 * lock on the page itself: the page may be truncated or
3967 * invalidated (changing page->mapping to NULL), or even
3968 * swizzled back from swapper_space to tmpfs file
3971 if (!trylock_page(page)) {
3976 if (unlikely(page->mapping != mapping)) {
3981 if (wbc->sync_mode != WB_SYNC_NONE) {
3982 if (PageWriteback(page))
3984 wait_on_page_writeback(page);
3987 if (PageWriteback(page) ||
3988 !clear_page_dirty_for_io(page)) {
3993 ret = (*writepage)(page, wbc, data);
3995 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
4001 * done_index is set past this page,
4002 * so media errors will not choke
4003 * background writeout for the entire
4004 * file. This has consequences for
4005 * range_cyclic semantics (ie. it may
4006 * not be suitable for data integrity
4009 done_index = page->index + 1;
4015 * the filesystem may choose to bump up nr_to_write.
4016 * We have to make sure to honor the new nr_to_write
4019 nr_to_write_done = wbc->nr_to_write <= 0;
4021 pagevec_release(&pvec);
4024 if (!scanned && !done) {
4026 * We hit the last page and there is more work to be done: wrap
4027 * back to the start of the file
4034 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4035 mapping->writeback_index = done_index;
4037 btrfs_add_delayed_iput(inode);
4041 static void flush_epd_write_bio(struct extent_page_data *epd)
4046 ret = submit_one_bio(epd->bio, 0, 0);
4047 BUG_ON(ret < 0); /* -ENOMEM */
4052 static noinline void flush_write_bio(void *data)
4054 struct extent_page_data *epd = data;
4055 flush_epd_write_bio(epd);
4058 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
4059 struct writeback_control *wbc)
4062 struct extent_page_data epd = {
4066 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4069 ret = __extent_writepage(page, wbc, &epd);
4071 flush_epd_write_bio(&epd);
4075 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
4076 u64 start, u64 end, int mode)
4079 struct address_space *mapping = inode->i_mapping;
4081 unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4084 struct extent_page_data epd = {
4088 .sync_io = mode == WB_SYNC_ALL,
4090 struct writeback_control wbc_writepages = {
4092 .nr_to_write = nr_pages * 2,
4093 .range_start = start,
4094 .range_end = end + 1,
4097 while (start <= end) {
4098 page = find_get_page(mapping, start >> PAGE_SHIFT);
4099 if (clear_page_dirty_for_io(page))
4100 ret = __extent_writepage(page, &wbc_writepages, &epd);
4102 if (tree->ops && tree->ops->writepage_end_io_hook)
4103 tree->ops->writepage_end_io_hook(page, start,
4104 start + PAGE_SIZE - 1,
4112 flush_epd_write_bio(&epd);
4116 int extent_writepages(struct extent_io_tree *tree,
4117 struct address_space *mapping,
4118 struct writeback_control *wbc)
4121 struct extent_page_data epd = {
4125 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4128 ret = extent_write_cache_pages(mapping, wbc, __extent_writepage, &epd,
4130 flush_epd_write_bio(&epd);
4134 int extent_readpages(struct extent_io_tree *tree,
4135 struct address_space *mapping,
4136 struct list_head *pages, unsigned nr_pages,
4137 get_extent_t get_extent)
4139 struct bio *bio = NULL;
4141 unsigned long bio_flags = 0;
4142 struct page *pagepool[16];
4144 struct extent_map *em_cached = NULL;
4146 u64 prev_em_start = (u64)-1;
4148 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
4149 page = list_entry(pages->prev, struct page, lru);
4151 prefetchw(&page->flags);
4152 list_del(&page->lru);
4153 if (add_to_page_cache_lru(page, mapping,
4155 readahead_gfp_mask(mapping))) {
4160 pagepool[nr++] = page;
4161 if (nr < ARRAY_SIZE(pagepool))
4163 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4164 &bio, &bio_flags, &prev_em_start);
4168 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4169 &bio, &bio_flags, &prev_em_start);
4172 free_extent_map(em_cached);
4174 BUG_ON(!list_empty(pages));
4176 return submit_one_bio(bio, 0, bio_flags);
4181 * basic invalidatepage code, this waits on any locked or writeback
4182 * ranges corresponding to the page, and then deletes any extent state
4183 * records from the tree
4185 int extent_invalidatepage(struct extent_io_tree *tree,
4186 struct page *page, unsigned long offset)
4188 struct extent_state *cached_state = NULL;
4189 u64 start = page_offset(page);
4190 u64 end = start + PAGE_SIZE - 1;
4191 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4193 start += ALIGN(offset, blocksize);
4197 lock_extent_bits(tree, start, end, &cached_state);
4198 wait_on_page_writeback(page);
4199 clear_extent_bit(tree, start, end,
4200 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4201 EXTENT_DO_ACCOUNTING,
4202 1, 1, &cached_state);
4207 * a helper for releasepage, this tests for areas of the page that
4208 * are locked or under IO and drops the related state bits if it is safe
4211 static int try_release_extent_state(struct extent_map_tree *map,
4212 struct extent_io_tree *tree,
4213 struct page *page, gfp_t mask)
4215 u64 start = page_offset(page);
4216 u64 end = start + PAGE_SIZE - 1;
4219 if (test_range_bit(tree, start, end,
4220 EXTENT_IOBITS, 0, NULL))
4224 * at this point we can safely clear everything except the
4225 * locked bit and the nodatasum bit
4227 ret = __clear_extent_bit(tree, start, end,
4228 ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4229 0, 0, NULL, mask, NULL);
4231 /* if clear_extent_bit failed for enomem reasons,
4232 * we can't allow the release to continue.
4243 * a helper for releasepage. As long as there are no locked extents
4244 * in the range corresponding to the page, both state records and extent
4245 * map records are removed
4247 int try_release_extent_mapping(struct extent_map_tree *map,
4248 struct extent_io_tree *tree, struct page *page,
4251 struct extent_map *em;
4252 u64 start = page_offset(page);
4253 u64 end = start + PAGE_SIZE - 1;
4255 if (gfpflags_allow_blocking(mask) &&
4256 page->mapping->host->i_size > SZ_16M) {
4258 while (start <= end) {
4259 len = end - start + 1;
4260 write_lock(&map->lock);
4261 em = lookup_extent_mapping(map, start, len);
4263 write_unlock(&map->lock);
4266 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4267 em->start != start) {
4268 write_unlock(&map->lock);
4269 free_extent_map(em);
4272 if (!test_range_bit(tree, em->start,
4273 extent_map_end(em) - 1,
4274 EXTENT_LOCKED | EXTENT_WRITEBACK,
4276 remove_extent_mapping(map, em);
4277 /* once for the rb tree */
4278 free_extent_map(em);
4280 start = extent_map_end(em);
4281 write_unlock(&map->lock);
4284 free_extent_map(em);
4287 return try_release_extent_state(map, tree, page, mask);
4291 * helper function for fiemap, which doesn't want to see any holes.
4292 * This maps until we find something past 'last'
4294 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4297 get_extent_t *get_extent)
4299 u64 sectorsize = btrfs_inode_sectorsize(inode);
4300 struct extent_map *em;
4307 len = last - offset;
4310 len = ALIGN(len, sectorsize);
4311 em = get_extent(BTRFS_I(inode), NULL, 0, offset, len, 0);
4312 if (IS_ERR_OR_NULL(em))
4315 /* if this isn't a hole return it */
4316 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4317 em->block_start != EXTENT_MAP_HOLE) {
4321 /* this is a hole, advance to the next extent */
4322 offset = extent_map_end(em);
4323 free_extent_map(em);
4331 * To cache previous fiemap extent
4333 * Will be used for merging fiemap extent
4335 struct fiemap_cache {
4344 * Helper to submit fiemap extent.
4346 * Will try to merge current fiemap extent specified by @offset, @phys,
4347 * @len and @flags with cached one.
4348 * And only when we fails to merge, cached one will be submitted as
4351 * Return value is the same as fiemap_fill_next_extent().
4353 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4354 struct fiemap_cache *cache,
4355 u64 offset, u64 phys, u64 len, u32 flags)
4363 * Sanity check, extent_fiemap() should have ensured that new
4364 * fiemap extent won't overlap with cahced one.
4367 * NOTE: Physical address can overlap, due to compression
4369 if (cache->offset + cache->len > offset) {
4375 * Only merges fiemap extents if
4376 * 1) Their logical addresses are continuous
4378 * 2) Their physical addresses are continuous
4379 * So truly compressed (physical size smaller than logical size)
4380 * extents won't get merged with each other
4382 * 3) Share same flags except FIEMAP_EXTENT_LAST
4383 * So regular extent won't get merged with prealloc extent
4385 if (cache->offset + cache->len == offset &&
4386 cache->phys + cache->len == phys &&
4387 (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4388 (flags & ~FIEMAP_EXTENT_LAST)) {
4390 cache->flags |= flags;
4391 goto try_submit_last;
4394 /* Not mergeable, need to submit cached one */
4395 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4396 cache->len, cache->flags);
4397 cache->cached = false;
4401 cache->cached = true;
4402 cache->offset = offset;
4405 cache->flags = flags;
4407 if (cache->flags & FIEMAP_EXTENT_LAST) {
4408 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4409 cache->phys, cache->len, cache->flags);
4410 cache->cached = false;
4416 * Emit last fiemap cache
4418 * The last fiemap cache may still be cached in the following case:
4420 * |<- Fiemap range ->|
4421 * |<------------ First extent ----------->|
4423 * In this case, the first extent range will be cached but not emitted.
4424 * So we must emit it before ending extent_fiemap().
4426 static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info,
4427 struct fiemap_extent_info *fieinfo,
4428 struct fiemap_cache *cache)
4435 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4436 cache->len, cache->flags);
4437 cache->cached = false;
4443 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4444 __u64 start, __u64 len)
4448 u64 max = start + len;
4452 u64 last_for_get_extent = 0;
4454 u64 isize = i_size_read(inode);
4455 struct btrfs_key found_key;
4456 struct extent_map *em = NULL;
4457 struct extent_state *cached_state = NULL;
4458 struct btrfs_path *path;
4459 struct btrfs_root *root = BTRFS_I(inode)->root;
4460 struct fiemap_cache cache = { 0 };
4469 path = btrfs_alloc_path();
4472 path->leave_spinning = 1;
4474 start = round_down(start, btrfs_inode_sectorsize(inode));
4475 len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
4478 * lookup the last file extent. We're not using i_size here
4479 * because there might be preallocation past i_size
4481 ret = btrfs_lookup_file_extent(NULL, root, path,
4482 btrfs_ino(BTRFS_I(inode)), -1, 0);
4484 btrfs_free_path(path);
4493 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4494 found_type = found_key.type;
4496 /* No extents, but there might be delalloc bits */
4497 if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) ||
4498 found_type != BTRFS_EXTENT_DATA_KEY) {
4499 /* have to trust i_size as the end */
4501 last_for_get_extent = isize;
4504 * remember the start of the last extent. There are a
4505 * bunch of different factors that go into the length of the
4506 * extent, so its much less complex to remember where it started
4508 last = found_key.offset;
4509 last_for_get_extent = last + 1;
4511 btrfs_release_path(path);
4514 * we might have some extents allocated but more delalloc past those
4515 * extents. so, we trust isize unless the start of the last extent is
4520 last_for_get_extent = isize;
4523 lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4526 em = get_extent_skip_holes(inode, start, last_for_get_extent,
4527 btrfs_get_extent_fiemap);
4536 u64 offset_in_extent = 0;
4538 /* break if the extent we found is outside the range */
4539 if (em->start >= max || extent_map_end(em) < off)
4543 * get_extent may return an extent that starts before our
4544 * requested range. We have to make sure the ranges
4545 * we return to fiemap always move forward and don't
4546 * overlap, so adjust the offsets here
4548 em_start = max(em->start, off);
4551 * record the offset from the start of the extent
4552 * for adjusting the disk offset below. Only do this if the
4553 * extent isn't compressed since our in ram offset may be past
4554 * what we have actually allocated on disk.
4556 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4557 offset_in_extent = em_start - em->start;
4558 em_end = extent_map_end(em);
4559 em_len = em_end - em_start;
4564 * bump off for our next call to get_extent
4566 off = extent_map_end(em);
4570 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4572 flags |= FIEMAP_EXTENT_LAST;
4573 } else if (em->block_start == EXTENT_MAP_INLINE) {
4574 flags |= (FIEMAP_EXTENT_DATA_INLINE |
4575 FIEMAP_EXTENT_NOT_ALIGNED);
4576 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4577 flags |= (FIEMAP_EXTENT_DELALLOC |
4578 FIEMAP_EXTENT_UNKNOWN);
4579 } else if (fieinfo->fi_extents_max) {
4580 u64 bytenr = em->block_start -
4581 (em->start - em->orig_start);
4583 disko = em->block_start + offset_in_extent;
4586 * As btrfs supports shared space, this information
4587 * can be exported to userspace tools via
4588 * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0
4589 * then we're just getting a count and we can skip the
4592 ret = btrfs_check_shared(root,
4593 btrfs_ino(BTRFS_I(inode)),
4598 flags |= FIEMAP_EXTENT_SHARED;
4601 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4602 flags |= FIEMAP_EXTENT_ENCODED;
4603 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4604 flags |= FIEMAP_EXTENT_UNWRITTEN;
4606 free_extent_map(em);
4608 if ((em_start >= last) || em_len == (u64)-1 ||
4609 (last == (u64)-1 && isize <= em_end)) {
4610 flags |= FIEMAP_EXTENT_LAST;
4614 /* now scan forward to see if this is really the last extent. */
4615 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4616 btrfs_get_extent_fiemap);
4622 flags |= FIEMAP_EXTENT_LAST;
4625 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4635 ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache);
4636 free_extent_map(em);
4638 btrfs_free_path(path);
4639 unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4640 &cached_state, GFP_NOFS);
4644 static void __free_extent_buffer(struct extent_buffer *eb)
4646 btrfs_leak_debug_del(&eb->leak_list);
4647 kmem_cache_free(extent_buffer_cache, eb);
4650 int extent_buffer_under_io(struct extent_buffer *eb)
4652 return (atomic_read(&eb->io_pages) ||
4653 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4654 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4658 * Helper for releasing extent buffer page.
4660 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb)
4662 unsigned long index;
4664 int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4666 BUG_ON(extent_buffer_under_io(eb));
4668 index = num_extent_pages(eb->start, eb->len);
4674 page = eb->pages[index];
4678 spin_lock(&page->mapping->private_lock);
4680 * We do this since we'll remove the pages after we've
4681 * removed the eb from the radix tree, so we could race
4682 * and have this page now attached to the new eb. So
4683 * only clear page_private if it's still connected to
4686 if (PagePrivate(page) &&
4687 page->private == (unsigned long)eb) {
4688 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4689 BUG_ON(PageDirty(page));
4690 BUG_ON(PageWriteback(page));
4692 * We need to make sure we haven't be attached
4695 ClearPagePrivate(page);
4696 set_page_private(page, 0);
4697 /* One for the page private */
4702 spin_unlock(&page->mapping->private_lock);
4704 /* One for when we allocated the page */
4706 } while (index != 0);
4710 * Helper for releasing the extent buffer.
4712 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4714 btrfs_release_extent_buffer_page(eb);
4715 __free_extent_buffer(eb);
4718 static struct extent_buffer *
4719 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4722 struct extent_buffer *eb = NULL;
4724 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4727 eb->fs_info = fs_info;
4729 rwlock_init(&eb->lock);
4730 atomic_set(&eb->write_locks, 0);
4731 atomic_set(&eb->read_locks, 0);
4732 atomic_set(&eb->blocking_readers, 0);
4733 atomic_set(&eb->blocking_writers, 0);
4734 atomic_set(&eb->spinning_readers, 0);
4735 atomic_set(&eb->spinning_writers, 0);
4736 eb->lock_nested = 0;
4737 init_waitqueue_head(&eb->write_lock_wq);
4738 init_waitqueue_head(&eb->read_lock_wq);
4740 btrfs_leak_debug_add(&eb->leak_list, &buffers);
4742 spin_lock_init(&eb->refs_lock);
4743 atomic_set(&eb->refs, 1);
4744 atomic_set(&eb->io_pages, 0);
4747 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4749 BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4750 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4751 BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4756 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4760 struct extent_buffer *new;
4761 unsigned long num_pages = num_extent_pages(src->start, src->len);
4763 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4767 for (i = 0; i < num_pages; i++) {
4768 p = alloc_page(GFP_NOFS);
4770 btrfs_release_extent_buffer(new);
4773 attach_extent_buffer_page(new, p);
4774 WARN_ON(PageDirty(p));
4777 copy_page(page_address(p), page_address(src->pages[i]));
4780 set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4781 set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4786 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4787 u64 start, unsigned long len)
4789 struct extent_buffer *eb;
4790 unsigned long num_pages;
4793 num_pages = num_extent_pages(start, len);
4795 eb = __alloc_extent_buffer(fs_info, start, len);
4799 for (i = 0; i < num_pages; i++) {
4800 eb->pages[i] = alloc_page(GFP_NOFS);
4804 set_extent_buffer_uptodate(eb);
4805 btrfs_set_header_nritems(eb, 0);
4806 set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4811 __free_page(eb->pages[i - 1]);
4812 __free_extent_buffer(eb);
4816 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4819 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
4822 static void check_buffer_tree_ref(struct extent_buffer *eb)
4825 /* the ref bit is tricky. We have to make sure it is set
4826 * if we have the buffer dirty. Otherwise the
4827 * code to free a buffer can end up dropping a dirty
4830 * Once the ref bit is set, it won't go away while the
4831 * buffer is dirty or in writeback, and it also won't
4832 * go away while we have the reference count on the
4835 * We can't just set the ref bit without bumping the
4836 * ref on the eb because free_extent_buffer might
4837 * see the ref bit and try to clear it. If this happens
4838 * free_extent_buffer might end up dropping our original
4839 * ref by mistake and freeing the page before we are able
4840 * to add one more ref.
4842 * So bump the ref count first, then set the bit. If someone
4843 * beat us to it, drop the ref we added.
4845 refs = atomic_read(&eb->refs);
4846 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4849 spin_lock(&eb->refs_lock);
4850 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4851 atomic_inc(&eb->refs);
4852 spin_unlock(&eb->refs_lock);
4855 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4856 struct page *accessed)
4858 unsigned long num_pages, i;
4860 check_buffer_tree_ref(eb);
4862 num_pages = num_extent_pages(eb->start, eb->len);
4863 for (i = 0; i < num_pages; i++) {
4864 struct page *p = eb->pages[i];
4867 mark_page_accessed(p);
4871 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4874 struct extent_buffer *eb;
4877 eb = radix_tree_lookup(&fs_info->buffer_radix,
4878 start >> PAGE_SHIFT);
4879 if (eb && atomic_inc_not_zero(&eb->refs)) {
4882 * Lock our eb's refs_lock to avoid races with
4883 * free_extent_buffer. When we get our eb it might be flagged
4884 * with EXTENT_BUFFER_STALE and another task running
4885 * free_extent_buffer might have seen that flag set,
4886 * eb->refs == 2, that the buffer isn't under IO (dirty and
4887 * writeback flags not set) and it's still in the tree (flag
4888 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4889 * of decrementing the extent buffer's reference count twice.
4890 * So here we could race and increment the eb's reference count,
4891 * clear its stale flag, mark it as dirty and drop our reference
4892 * before the other task finishes executing free_extent_buffer,
4893 * which would later result in an attempt to free an extent
4894 * buffer that is dirty.
4896 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
4897 spin_lock(&eb->refs_lock);
4898 spin_unlock(&eb->refs_lock);
4900 mark_extent_buffer_accessed(eb, NULL);
4908 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4909 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4912 struct extent_buffer *eb, *exists = NULL;
4915 eb = find_extent_buffer(fs_info, start);
4918 eb = alloc_dummy_extent_buffer(fs_info, start);
4921 eb->fs_info = fs_info;
4923 ret = radix_tree_preload(GFP_NOFS);
4926 spin_lock(&fs_info->buffer_lock);
4927 ret = radix_tree_insert(&fs_info->buffer_radix,
4928 start >> PAGE_SHIFT, eb);
4929 spin_unlock(&fs_info->buffer_lock);
4930 radix_tree_preload_end();
4931 if (ret == -EEXIST) {
4932 exists = find_extent_buffer(fs_info, start);
4938 check_buffer_tree_ref(eb);
4939 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4942 * We will free dummy extent buffer's if they come into
4943 * free_extent_buffer with a ref count of 2, but if we are using this we
4944 * want the buffers to stay in memory until we're done with them, so
4945 * bump the ref count again.
4947 atomic_inc(&eb->refs);
4950 btrfs_release_extent_buffer(eb);
4955 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4958 unsigned long len = fs_info->nodesize;
4959 unsigned long num_pages = num_extent_pages(start, len);
4961 unsigned long index = start >> PAGE_SHIFT;
4962 struct extent_buffer *eb;
4963 struct extent_buffer *exists = NULL;
4965 struct address_space *mapping = fs_info->btree_inode->i_mapping;
4969 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
4970 btrfs_err(fs_info, "bad tree block start %llu", start);
4971 return ERR_PTR(-EINVAL);
4974 eb = find_extent_buffer(fs_info, start);
4978 eb = __alloc_extent_buffer(fs_info, start, len);
4980 return ERR_PTR(-ENOMEM);
4982 for (i = 0; i < num_pages; i++, index++) {
4983 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
4985 exists = ERR_PTR(-ENOMEM);
4989 spin_lock(&mapping->private_lock);
4990 if (PagePrivate(p)) {
4992 * We could have already allocated an eb for this page
4993 * and attached one so lets see if we can get a ref on
4994 * the existing eb, and if we can we know it's good and
4995 * we can just return that one, else we know we can just
4996 * overwrite page->private.
4998 exists = (struct extent_buffer *)p->private;
4999 if (atomic_inc_not_zero(&exists->refs)) {
5000 spin_unlock(&mapping->private_lock);
5003 mark_extent_buffer_accessed(exists, p);
5009 * Do this so attach doesn't complain and we need to
5010 * drop the ref the old guy had.
5012 ClearPagePrivate(p);
5013 WARN_ON(PageDirty(p));
5016 attach_extent_buffer_page(eb, p);
5017 spin_unlock(&mapping->private_lock);
5018 WARN_ON(PageDirty(p));
5020 if (!PageUptodate(p))
5024 * see below about how we avoid a nasty race with release page
5025 * and why we unlock later
5029 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5031 ret = radix_tree_preload(GFP_NOFS);
5033 exists = ERR_PTR(ret);
5037 spin_lock(&fs_info->buffer_lock);
5038 ret = radix_tree_insert(&fs_info->buffer_radix,
5039 start >> PAGE_SHIFT, eb);
5040 spin_unlock(&fs_info->buffer_lock);
5041 radix_tree_preload_end();
5042 if (ret == -EEXIST) {
5043 exists = find_extent_buffer(fs_info, start);
5049 /* add one reference for the tree */
5050 check_buffer_tree_ref(eb);
5051 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5054 * there is a race where release page may have
5055 * tried to find this extent buffer in the radix
5056 * but failed. It will tell the VM it is safe to
5057 * reclaim the, and it will clear the page private bit.
5058 * We must make sure to set the page private bit properly
5059 * after the extent buffer is in the radix tree so
5060 * it doesn't get lost
5062 SetPageChecked(eb->pages[0]);
5063 for (i = 1; i < num_pages; i++) {
5065 ClearPageChecked(p);
5068 unlock_page(eb->pages[0]);
5072 WARN_ON(!atomic_dec_and_test(&eb->refs));
5073 for (i = 0; i < num_pages; i++) {
5075 unlock_page(eb->pages[i]);
5078 btrfs_release_extent_buffer(eb);
5082 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5084 struct extent_buffer *eb =
5085 container_of(head, struct extent_buffer, rcu_head);
5087 __free_extent_buffer(eb);
5090 /* Expects to have eb->eb_lock already held */
5091 static int release_extent_buffer(struct extent_buffer *eb)
5093 WARN_ON(atomic_read(&eb->refs) == 0);
5094 if (atomic_dec_and_test(&eb->refs)) {
5095 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5096 struct btrfs_fs_info *fs_info = eb->fs_info;
5098 spin_unlock(&eb->refs_lock);
5100 spin_lock(&fs_info->buffer_lock);
5101 radix_tree_delete(&fs_info->buffer_radix,
5102 eb->start >> PAGE_SHIFT);
5103 spin_unlock(&fs_info->buffer_lock);
5105 spin_unlock(&eb->refs_lock);
5108 /* Should be safe to release our pages at this point */
5109 btrfs_release_extent_buffer_page(eb);
5110 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5111 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) {
5112 __free_extent_buffer(eb);
5116 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5119 spin_unlock(&eb->refs_lock);
5124 void free_extent_buffer(struct extent_buffer *eb)
5132 refs = atomic_read(&eb->refs);
5135 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5140 spin_lock(&eb->refs_lock);
5141 if (atomic_read(&eb->refs) == 2 &&
5142 test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
5143 atomic_dec(&eb->refs);
5145 if (atomic_read(&eb->refs) == 2 &&
5146 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5147 !extent_buffer_under_io(eb) &&
5148 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5149 atomic_dec(&eb->refs);
5152 * I know this is terrible, but it's temporary until we stop tracking
5153 * the uptodate bits and such for the extent buffers.
5155 release_extent_buffer(eb);
5158 void free_extent_buffer_stale(struct extent_buffer *eb)
5163 spin_lock(&eb->refs_lock);
5164 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5166 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5167 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5168 atomic_dec(&eb->refs);
5169 release_extent_buffer(eb);
5172 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5175 unsigned long num_pages;
5178 num_pages = num_extent_pages(eb->start, eb->len);
5180 for (i = 0; i < num_pages; i++) {
5181 page = eb->pages[i];
5182 if (!PageDirty(page))
5186 WARN_ON(!PagePrivate(page));
5188 clear_page_dirty_for_io(page);
5189 spin_lock_irq(&page->mapping->tree_lock);
5190 if (!PageDirty(page)) {
5191 radix_tree_tag_clear(&page->mapping->page_tree,
5193 PAGECACHE_TAG_DIRTY);
5195 spin_unlock_irq(&page->mapping->tree_lock);
5196 ClearPageError(page);
5199 WARN_ON(atomic_read(&eb->refs) == 0);
5202 int set_extent_buffer_dirty(struct extent_buffer *eb)
5205 unsigned long num_pages;
5208 check_buffer_tree_ref(eb);
5210 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5212 num_pages = num_extent_pages(eb->start, eb->len);
5213 WARN_ON(atomic_read(&eb->refs) == 0);
5214 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5216 for (i = 0; i < num_pages; i++)
5217 set_page_dirty(eb->pages[i]);
5221 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5225 unsigned long num_pages;
5227 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5228 num_pages = num_extent_pages(eb->start, eb->len);
5229 for (i = 0; i < num_pages; i++) {
5230 page = eb->pages[i];
5232 ClearPageUptodate(page);
5236 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5240 unsigned long num_pages;
5242 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5243 num_pages = num_extent_pages(eb->start, eb->len);
5244 for (i = 0; i < num_pages; i++) {
5245 page = eb->pages[i];
5246 SetPageUptodate(page);
5250 int extent_buffer_uptodate(struct extent_buffer *eb)
5252 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5255 int read_extent_buffer_pages(struct extent_io_tree *tree,
5256 struct extent_buffer *eb, int wait,
5257 get_extent_t *get_extent, int mirror_num)
5263 int locked_pages = 0;
5264 int all_uptodate = 1;
5265 unsigned long num_pages;
5266 unsigned long num_reads = 0;
5267 struct bio *bio = NULL;
5268 unsigned long bio_flags = 0;
5270 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5273 num_pages = num_extent_pages(eb->start, eb->len);
5274 for (i = 0; i < num_pages; i++) {
5275 page = eb->pages[i];
5276 if (wait == WAIT_NONE) {
5277 if (!trylock_page(page))
5285 * We need to firstly lock all pages to make sure that
5286 * the uptodate bit of our pages won't be affected by
5287 * clear_extent_buffer_uptodate().
5289 for (i = 0; i < num_pages; i++) {
5290 page = eb->pages[i];
5291 if (!PageUptodate(page)) {
5298 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5302 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5303 eb->read_mirror = 0;
5304 atomic_set(&eb->io_pages, num_reads);
5305 for (i = 0; i < num_pages; i++) {
5306 page = eb->pages[i];
5308 if (!PageUptodate(page)) {
5310 atomic_dec(&eb->io_pages);
5315 ClearPageError(page);
5316 err = __extent_read_full_page(tree, page,
5318 mirror_num, &bio_flags,
5323 * We use &bio in above __extent_read_full_page,
5324 * so we ensure that if it returns error, the
5325 * current page fails to add itself to bio and
5326 * it's been unlocked.
5328 * We must dec io_pages by ourselves.
5330 atomic_dec(&eb->io_pages);
5338 err = submit_one_bio(bio, mirror_num, bio_flags);
5343 if (ret || wait != WAIT_COMPLETE)
5346 for (i = 0; i < num_pages; i++) {
5347 page = eb->pages[i];
5348 wait_on_page_locked(page);
5349 if (!PageUptodate(page))
5356 while (locked_pages > 0) {
5358 page = eb->pages[locked_pages];
5364 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5365 unsigned long start, unsigned long len)
5371 char *dst = (char *)dstv;
5372 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5373 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5375 if (start + len > eb->len) {
5376 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5377 eb->start, eb->len, start, len);
5378 memset(dst, 0, len);
5382 offset = (start_offset + start) & (PAGE_SIZE - 1);
5385 page = eb->pages[i];
5387 cur = min(len, (PAGE_SIZE - offset));
5388 kaddr = page_address(page);
5389 memcpy(dst, kaddr + offset, cur);
5398 int read_extent_buffer_to_user(const struct extent_buffer *eb,
5400 unsigned long start, unsigned long len)
5406 char __user *dst = (char __user *)dstv;
5407 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5408 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5411 WARN_ON(start > eb->len);
5412 WARN_ON(start + len > eb->start + eb->len);
5414 offset = (start_offset + start) & (PAGE_SIZE - 1);
5417 page = eb->pages[i];
5419 cur = min(len, (PAGE_SIZE - offset));
5420 kaddr = page_address(page);
5421 if (copy_to_user(dst, kaddr + offset, cur)) {
5436 * return 0 if the item is found within a page.
5437 * return 1 if the item spans two pages.
5438 * return -EINVAL otherwise.
5440 int map_private_extent_buffer(const struct extent_buffer *eb,
5441 unsigned long start, unsigned long min_len,
5442 char **map, unsigned long *map_start,
5443 unsigned long *map_len)
5445 size_t offset = start & (PAGE_SIZE - 1);
5448 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5449 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5450 unsigned long end_i = (start_offset + start + min_len - 1) >>
5453 if (start + min_len > eb->len) {
5454 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5455 eb->start, eb->len, start, min_len);
5463 offset = start_offset;
5467 *map_start = ((u64)i << PAGE_SHIFT) - start_offset;
5471 kaddr = page_address(p);
5472 *map = kaddr + offset;
5473 *map_len = PAGE_SIZE - offset;
5477 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
5478 unsigned long start, unsigned long len)
5484 char *ptr = (char *)ptrv;
5485 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5486 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5489 WARN_ON(start > eb->len);
5490 WARN_ON(start + len > eb->start + eb->len);
5492 offset = (start_offset + start) & (PAGE_SIZE - 1);
5495 page = eb->pages[i];
5497 cur = min(len, (PAGE_SIZE - offset));
5499 kaddr = page_address(page);
5500 ret = memcmp(ptr, kaddr + offset, cur);
5512 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb,
5517 WARN_ON(!PageUptodate(eb->pages[0]));
5518 kaddr = page_address(eb->pages[0]);
5519 memcpy(kaddr + offsetof(struct btrfs_header, chunk_tree_uuid), srcv,
5523 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *srcv)
5527 WARN_ON(!PageUptodate(eb->pages[0]));
5528 kaddr = page_address(eb->pages[0]);
5529 memcpy(kaddr + offsetof(struct btrfs_header, fsid), srcv,
5533 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5534 unsigned long start, unsigned long len)
5540 char *src = (char *)srcv;
5541 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5542 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5544 WARN_ON(start > eb->len);
5545 WARN_ON(start + len > eb->start + eb->len);
5547 offset = (start_offset + start) & (PAGE_SIZE - 1);
5550 page = eb->pages[i];
5551 WARN_ON(!PageUptodate(page));
5553 cur = min(len, PAGE_SIZE - offset);
5554 kaddr = page_address(page);
5555 memcpy(kaddr + offset, src, cur);
5564 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start,
5571 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5572 unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5574 WARN_ON(start > eb->len);
5575 WARN_ON(start + len > eb->start + eb->len);
5577 offset = (start_offset + start) & (PAGE_SIZE - 1);
5580 page = eb->pages[i];
5581 WARN_ON(!PageUptodate(page));
5583 cur = min(len, PAGE_SIZE - offset);
5584 kaddr = page_address(page);
5585 memset(kaddr + offset, 0, cur);
5593 void copy_extent_buffer_full(struct extent_buffer *dst,
5594 struct extent_buffer *src)
5599 ASSERT(dst->len == src->len);
5601 num_pages = num_extent_pages(dst->start, dst->len);
5602 for (i = 0; i < num_pages; i++)
5603 copy_page(page_address(dst->pages[i]),
5604 page_address(src->pages[i]));
5607 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5608 unsigned long dst_offset, unsigned long src_offset,
5611 u64 dst_len = dst->len;
5616 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5617 unsigned long i = (start_offset + dst_offset) >> PAGE_SHIFT;
5619 WARN_ON(src->len != dst_len);
5621 offset = (start_offset + dst_offset) &
5625 page = dst->pages[i];
5626 WARN_ON(!PageUptodate(page));
5628 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5630 kaddr = page_address(page);
5631 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5640 void le_bitmap_set(u8 *map, unsigned int start, int len)
5642 u8 *p = map + BIT_BYTE(start);
5643 const unsigned int size = start + len;
5644 int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE);
5645 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start);
5647 while (len - bits_to_set >= 0) {
5650 bits_to_set = BITS_PER_BYTE;
5655 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5660 void le_bitmap_clear(u8 *map, unsigned int start, int len)
5662 u8 *p = map + BIT_BYTE(start);
5663 const unsigned int size = start + len;
5664 int bits_to_clear = BITS_PER_BYTE - (start % BITS_PER_BYTE);
5665 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(start);
5667 while (len - bits_to_clear >= 0) {
5668 *p &= ~mask_to_clear;
5669 len -= bits_to_clear;
5670 bits_to_clear = BITS_PER_BYTE;
5675 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5676 *p &= ~mask_to_clear;
5681 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5683 * @eb: the extent buffer
5684 * @start: offset of the bitmap item in the extent buffer
5686 * @page_index: return index of the page in the extent buffer that contains the
5688 * @page_offset: return offset into the page given by page_index
5690 * This helper hides the ugliness of finding the byte in an extent buffer which
5691 * contains a given bit.
5693 static inline void eb_bitmap_offset(struct extent_buffer *eb,
5694 unsigned long start, unsigned long nr,
5695 unsigned long *page_index,
5696 size_t *page_offset)
5698 size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5699 size_t byte_offset = BIT_BYTE(nr);
5703 * The byte we want is the offset of the extent buffer + the offset of
5704 * the bitmap item in the extent buffer + the offset of the byte in the
5707 offset = start_offset + start + byte_offset;
5709 *page_index = offset >> PAGE_SHIFT;
5710 *page_offset = offset & (PAGE_SIZE - 1);
5714 * extent_buffer_test_bit - determine whether a bit in a bitmap item is set
5715 * @eb: the extent buffer
5716 * @start: offset of the bitmap item in the extent buffer
5717 * @nr: bit number to test
5719 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
5727 eb_bitmap_offset(eb, start, nr, &i, &offset);
5728 page = eb->pages[i];
5729 WARN_ON(!PageUptodate(page));
5730 kaddr = page_address(page);
5731 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5735 * extent_buffer_bitmap_set - set an area of a bitmap
5736 * @eb: the extent buffer
5737 * @start: offset of the bitmap item in the extent buffer
5738 * @pos: bit number of the first bit
5739 * @len: number of bits to set
5741 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
5742 unsigned long pos, unsigned long len)
5748 const unsigned int size = pos + len;
5749 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5750 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5752 eb_bitmap_offset(eb, start, pos, &i, &offset);
5753 page = eb->pages[i];
5754 WARN_ON(!PageUptodate(page));
5755 kaddr = page_address(page);
5757 while (len >= bits_to_set) {
5758 kaddr[offset] |= mask_to_set;
5760 bits_to_set = BITS_PER_BYTE;
5762 if (++offset >= PAGE_SIZE && len > 0) {
5764 page = eb->pages[++i];
5765 WARN_ON(!PageUptodate(page));
5766 kaddr = page_address(page);
5770 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5771 kaddr[offset] |= mask_to_set;
5777 * extent_buffer_bitmap_clear - clear an area of a bitmap
5778 * @eb: the extent buffer
5779 * @start: offset of the bitmap item in the extent buffer
5780 * @pos: bit number of the first bit
5781 * @len: number of bits to clear
5783 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
5784 unsigned long pos, unsigned long len)
5790 const unsigned int size = pos + len;
5791 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5792 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5794 eb_bitmap_offset(eb, start, pos, &i, &offset);
5795 page = eb->pages[i];
5796 WARN_ON(!PageUptodate(page));
5797 kaddr = page_address(page);
5799 while (len >= bits_to_clear) {
5800 kaddr[offset] &= ~mask_to_clear;
5801 len -= bits_to_clear;
5802 bits_to_clear = BITS_PER_BYTE;
5804 if (++offset >= PAGE_SIZE && len > 0) {
5806 page = eb->pages[++i];
5807 WARN_ON(!PageUptodate(page));
5808 kaddr = page_address(page);
5812 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5813 kaddr[offset] &= ~mask_to_clear;
5817 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5819 unsigned long distance = (src > dst) ? src - dst : dst - src;
5820 return distance < len;
5823 static void copy_pages(struct page *dst_page, struct page *src_page,
5824 unsigned long dst_off, unsigned long src_off,
5827 char *dst_kaddr = page_address(dst_page);
5829 int must_memmove = 0;
5831 if (dst_page != src_page) {
5832 src_kaddr = page_address(src_page);
5834 src_kaddr = dst_kaddr;
5835 if (areas_overlap(src_off, dst_off, len))
5840 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5842 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5845 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5846 unsigned long src_offset, unsigned long len)
5848 struct btrfs_fs_info *fs_info = dst->fs_info;
5850 size_t dst_off_in_page;
5851 size_t src_off_in_page;
5852 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5853 unsigned long dst_i;
5854 unsigned long src_i;
5856 if (src_offset + len > dst->len) {
5858 "memmove bogus src_offset %lu move len %lu dst len %lu",
5859 src_offset, len, dst->len);
5862 if (dst_offset + len > dst->len) {
5864 "memmove bogus dst_offset %lu move len %lu dst len %lu",
5865 dst_offset, len, dst->len);
5870 dst_off_in_page = (start_offset + dst_offset) &
5872 src_off_in_page = (start_offset + src_offset) &
5875 dst_i = (start_offset + dst_offset) >> PAGE_SHIFT;
5876 src_i = (start_offset + src_offset) >> PAGE_SHIFT;
5878 cur = min(len, (unsigned long)(PAGE_SIZE -
5880 cur = min_t(unsigned long, cur,
5881 (unsigned long)(PAGE_SIZE - dst_off_in_page));
5883 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5884 dst_off_in_page, src_off_in_page, cur);
5892 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5893 unsigned long src_offset, unsigned long len)
5895 struct btrfs_fs_info *fs_info = dst->fs_info;
5897 size_t dst_off_in_page;
5898 size_t src_off_in_page;
5899 unsigned long dst_end = dst_offset + len - 1;
5900 unsigned long src_end = src_offset + len - 1;
5901 size_t start_offset = dst->start & ((u64)PAGE_SIZE - 1);
5902 unsigned long dst_i;
5903 unsigned long src_i;
5905 if (src_offset + len > dst->len) {
5907 "memmove bogus src_offset %lu move len %lu len %lu",
5908 src_offset, len, dst->len);
5911 if (dst_offset + len > dst->len) {
5913 "memmove bogus dst_offset %lu move len %lu len %lu",
5914 dst_offset, len, dst->len);
5917 if (dst_offset < src_offset) {
5918 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5922 dst_i = (start_offset + dst_end) >> PAGE_SHIFT;
5923 src_i = (start_offset + src_end) >> PAGE_SHIFT;
5925 dst_off_in_page = (start_offset + dst_end) &
5927 src_off_in_page = (start_offset + src_end) &
5930 cur = min_t(unsigned long, len, src_off_in_page + 1);
5931 cur = min(cur, dst_off_in_page + 1);
5932 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5933 dst_off_in_page - cur + 1,
5934 src_off_in_page - cur + 1, cur);
5942 int try_release_extent_buffer(struct page *page)
5944 struct extent_buffer *eb;
5947 * We need to make sure nobody is attaching this page to an eb right
5950 spin_lock(&page->mapping->private_lock);
5951 if (!PagePrivate(page)) {
5952 spin_unlock(&page->mapping->private_lock);
5956 eb = (struct extent_buffer *)page->private;
5960 * This is a little awful but should be ok, we need to make sure that
5961 * the eb doesn't disappear out from under us while we're looking at
5964 spin_lock(&eb->refs_lock);
5965 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5966 spin_unlock(&eb->refs_lock);
5967 spin_unlock(&page->mapping->private_lock);
5970 spin_unlock(&page->mapping->private_lock);
5973 * If tree ref isn't set then we know the ref on this eb is a real ref,
5974 * so just return, this page will likely be freed soon anyway.
5976 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5977 spin_unlock(&eb->refs_lock);
5981 return release_extent_buffer(eb);