2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
35 #include "free-space-cache.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE = 0,
58 CHUNK_ALLOC_LIMITED = 1,
59 CHUNK_ALLOC_FORCE = 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT = 2,
77 static int update_block_group(struct btrfs_trans_handle *trans,
78 struct btrfs_root *root, u64 bytenr,
79 u64 num_bytes, int alloc);
80 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
81 struct btrfs_root *root,
82 struct btrfs_delayed_ref_node *node, u64 parent,
83 u64 root_objectid, u64 owner_objectid,
84 u64 owner_offset, int refs_to_drop,
85 struct btrfs_delayed_extent_op *extra_op);
86 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
87 struct extent_buffer *leaf,
88 struct btrfs_extent_item *ei);
89 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
90 struct btrfs_root *root,
91 u64 parent, u64 root_objectid,
92 u64 flags, u64 owner, u64 offset,
93 struct btrfs_key *ins, int ref_mod);
94 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
95 struct btrfs_root *root,
96 u64 parent, u64 root_objectid,
97 u64 flags, struct btrfs_disk_key *key,
98 int level, struct btrfs_key *ins,
100 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
101 struct btrfs_root *extent_root, u64 flags,
103 static int find_next_key(struct btrfs_path *path, int level,
104 struct btrfs_key *key);
105 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
106 int dump_block_groups);
107 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
108 u64 num_bytes, int reserve,
110 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
112 int btrfs_pin_extent(struct btrfs_root *root,
113 u64 bytenr, u64 num_bytes, int reserved);
116 block_group_cache_done(struct btrfs_block_group_cache *cache)
119 return cache->cached == BTRFS_CACHE_FINISHED ||
120 cache->cached == BTRFS_CACHE_ERROR;
123 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
125 return (cache->flags & bits) == bits;
128 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
130 atomic_inc(&cache->count);
133 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
135 if (atomic_dec_and_test(&cache->count)) {
136 WARN_ON(cache->pinned > 0);
137 WARN_ON(cache->reserved > 0);
138 kfree(cache->free_space_ctl);
144 * this adds the block group to the fs_info rb tree for the block group
147 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
148 struct btrfs_block_group_cache *block_group)
151 struct rb_node *parent = NULL;
152 struct btrfs_block_group_cache *cache;
154 spin_lock(&info->block_group_cache_lock);
155 p = &info->block_group_cache_tree.rb_node;
159 cache = rb_entry(parent, struct btrfs_block_group_cache,
161 if (block_group->key.objectid < cache->key.objectid) {
163 } else if (block_group->key.objectid > cache->key.objectid) {
166 spin_unlock(&info->block_group_cache_lock);
171 rb_link_node(&block_group->cache_node, parent, p);
172 rb_insert_color(&block_group->cache_node,
173 &info->block_group_cache_tree);
175 if (info->first_logical_byte > block_group->key.objectid)
176 info->first_logical_byte = block_group->key.objectid;
178 spin_unlock(&info->block_group_cache_lock);
184 * This will return the block group at or after bytenr if contains is 0, else
185 * it will return the block group that contains the bytenr
187 static struct btrfs_block_group_cache *
188 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
191 struct btrfs_block_group_cache *cache, *ret = NULL;
195 spin_lock(&info->block_group_cache_lock);
196 n = info->block_group_cache_tree.rb_node;
199 cache = rb_entry(n, struct btrfs_block_group_cache,
201 end = cache->key.objectid + cache->key.offset - 1;
202 start = cache->key.objectid;
204 if (bytenr < start) {
205 if (!contains && (!ret || start < ret->key.objectid))
208 } else if (bytenr > start) {
209 if (contains && bytenr <= end) {
220 btrfs_get_block_group(ret);
221 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
222 info->first_logical_byte = ret->key.objectid;
224 spin_unlock(&info->block_group_cache_lock);
229 static int add_excluded_extent(struct btrfs_root *root,
230 u64 start, u64 num_bytes)
232 u64 end = start + num_bytes - 1;
233 set_extent_bits(&root->fs_info->freed_extents[0],
234 start, end, EXTENT_UPTODATE, GFP_NOFS);
235 set_extent_bits(&root->fs_info->freed_extents[1],
236 start, end, EXTENT_UPTODATE, GFP_NOFS);
240 static void free_excluded_extents(struct btrfs_root *root,
241 struct btrfs_block_group_cache *cache)
245 start = cache->key.objectid;
246 end = start + cache->key.offset - 1;
248 clear_extent_bits(&root->fs_info->freed_extents[0],
249 start, end, EXTENT_UPTODATE, GFP_NOFS);
250 clear_extent_bits(&root->fs_info->freed_extents[1],
251 start, end, EXTENT_UPTODATE, GFP_NOFS);
254 static int exclude_super_stripes(struct btrfs_root *root,
255 struct btrfs_block_group_cache *cache)
262 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
263 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
264 cache->bytes_super += stripe_len;
265 ret = add_excluded_extent(root, cache->key.objectid,
271 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
272 bytenr = btrfs_sb_offset(i);
273 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
274 cache->key.objectid, bytenr,
275 0, &logical, &nr, &stripe_len);
282 if (logical[nr] > cache->key.objectid +
286 if (logical[nr] + stripe_len <= cache->key.objectid)
290 if (start < cache->key.objectid) {
291 start = cache->key.objectid;
292 len = (logical[nr] + stripe_len) - start;
294 len = min_t(u64, stripe_len,
295 cache->key.objectid +
296 cache->key.offset - start);
299 cache->bytes_super += len;
300 ret = add_excluded_extent(root, start, len);
312 static struct btrfs_caching_control *
313 get_caching_control(struct btrfs_block_group_cache *cache)
315 struct btrfs_caching_control *ctl;
317 spin_lock(&cache->lock);
318 if (!cache->caching_ctl) {
319 spin_unlock(&cache->lock);
323 ctl = cache->caching_ctl;
324 atomic_inc(&ctl->count);
325 spin_unlock(&cache->lock);
329 static void put_caching_control(struct btrfs_caching_control *ctl)
331 if (atomic_dec_and_test(&ctl->count))
336 * this is only called by cache_block_group, since we could have freed extents
337 * we need to check the pinned_extents for any extents that can't be used yet
338 * since their free space will be released as soon as the transaction commits.
340 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
341 struct btrfs_fs_info *info, u64 start, u64 end)
343 u64 extent_start, extent_end, size, total_added = 0;
346 while (start < end) {
347 ret = find_first_extent_bit(info->pinned_extents, start,
348 &extent_start, &extent_end,
349 EXTENT_DIRTY | EXTENT_UPTODATE,
354 if (extent_start <= start) {
355 start = extent_end + 1;
356 } else if (extent_start > start && extent_start < end) {
357 size = extent_start - start;
359 ret = btrfs_add_free_space(block_group, start,
361 BUG_ON(ret); /* -ENOMEM or logic error */
362 start = extent_end + 1;
371 ret = btrfs_add_free_space(block_group, start, size);
372 BUG_ON(ret); /* -ENOMEM or logic error */
378 static noinline void caching_thread(struct btrfs_work *work)
380 struct btrfs_block_group_cache *block_group;
381 struct btrfs_fs_info *fs_info;
382 struct btrfs_caching_control *caching_ctl;
383 struct btrfs_root *extent_root;
384 struct btrfs_path *path;
385 struct extent_buffer *leaf;
386 struct btrfs_key key;
392 caching_ctl = container_of(work, struct btrfs_caching_control, work);
393 block_group = caching_ctl->block_group;
394 fs_info = block_group->fs_info;
395 extent_root = fs_info->extent_root;
397 path = btrfs_alloc_path();
401 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
404 * We don't want to deadlock with somebody trying to allocate a new
405 * extent for the extent root while also trying to search the extent
406 * root to add free space. So we skip locking and search the commit
407 * root, since its read-only
409 path->skip_locking = 1;
410 path->search_commit_root = 1;
415 key.type = BTRFS_EXTENT_ITEM_KEY;
417 mutex_lock(&caching_ctl->mutex);
418 /* need to make sure the commit_root doesn't disappear */
419 down_read(&fs_info->commit_root_sem);
422 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
426 leaf = path->nodes[0];
427 nritems = btrfs_header_nritems(leaf);
430 if (btrfs_fs_closing(fs_info) > 1) {
435 if (path->slots[0] < nritems) {
436 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
438 ret = find_next_key(path, 0, &key);
442 if (need_resched() ||
443 rwsem_is_contended(&fs_info->commit_root_sem)) {
444 caching_ctl->progress = last;
445 btrfs_release_path(path);
446 up_read(&fs_info->commit_root_sem);
447 mutex_unlock(&caching_ctl->mutex);
452 ret = btrfs_next_leaf(extent_root, path);
457 leaf = path->nodes[0];
458 nritems = btrfs_header_nritems(leaf);
462 if (key.objectid < last) {
465 key.type = BTRFS_EXTENT_ITEM_KEY;
467 caching_ctl->progress = last;
468 btrfs_release_path(path);
472 if (key.objectid < block_group->key.objectid) {
477 if (key.objectid >= block_group->key.objectid +
478 block_group->key.offset)
481 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
482 key.type == BTRFS_METADATA_ITEM_KEY) {
483 total_found += add_new_free_space(block_group,
486 if (key.type == BTRFS_METADATA_ITEM_KEY)
487 last = key.objectid +
488 fs_info->tree_root->nodesize;
490 last = key.objectid + key.offset;
492 if (total_found > (1024 * 1024 * 2)) {
494 wake_up(&caching_ctl->wait);
501 total_found += add_new_free_space(block_group, fs_info, last,
502 block_group->key.objectid +
503 block_group->key.offset);
504 caching_ctl->progress = (u64)-1;
506 spin_lock(&block_group->lock);
507 block_group->caching_ctl = NULL;
508 block_group->cached = BTRFS_CACHE_FINISHED;
509 spin_unlock(&block_group->lock);
512 btrfs_free_path(path);
513 up_read(&fs_info->commit_root_sem);
515 free_excluded_extents(extent_root, block_group);
517 mutex_unlock(&caching_ctl->mutex);
520 spin_lock(&block_group->lock);
521 block_group->caching_ctl = NULL;
522 block_group->cached = BTRFS_CACHE_ERROR;
523 spin_unlock(&block_group->lock);
525 wake_up(&caching_ctl->wait);
527 put_caching_control(caching_ctl);
528 btrfs_put_block_group(block_group);
531 static int cache_block_group(struct btrfs_block_group_cache *cache,
535 struct btrfs_fs_info *fs_info = cache->fs_info;
536 struct btrfs_caching_control *caching_ctl;
539 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
543 INIT_LIST_HEAD(&caching_ctl->list);
544 mutex_init(&caching_ctl->mutex);
545 init_waitqueue_head(&caching_ctl->wait);
546 caching_ctl->block_group = cache;
547 caching_ctl->progress = cache->key.objectid;
548 atomic_set(&caching_ctl->count, 1);
549 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
550 caching_thread, NULL, NULL);
552 spin_lock(&cache->lock);
554 * This should be a rare occasion, but this could happen I think in the
555 * case where one thread starts to load the space cache info, and then
556 * some other thread starts a transaction commit which tries to do an
557 * allocation while the other thread is still loading the space cache
558 * info. The previous loop should have kept us from choosing this block
559 * group, but if we've moved to the state where we will wait on caching
560 * block groups we need to first check if we're doing a fast load here,
561 * so we can wait for it to finish, otherwise we could end up allocating
562 * from a block group who's cache gets evicted for one reason or
565 while (cache->cached == BTRFS_CACHE_FAST) {
566 struct btrfs_caching_control *ctl;
568 ctl = cache->caching_ctl;
569 atomic_inc(&ctl->count);
570 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
571 spin_unlock(&cache->lock);
575 finish_wait(&ctl->wait, &wait);
576 put_caching_control(ctl);
577 spin_lock(&cache->lock);
580 if (cache->cached != BTRFS_CACHE_NO) {
581 spin_unlock(&cache->lock);
585 WARN_ON(cache->caching_ctl);
586 cache->caching_ctl = caching_ctl;
587 cache->cached = BTRFS_CACHE_FAST;
588 spin_unlock(&cache->lock);
590 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
591 mutex_lock(&caching_ctl->mutex);
592 ret = load_free_space_cache(fs_info, cache);
594 spin_lock(&cache->lock);
596 cache->caching_ctl = NULL;
597 cache->cached = BTRFS_CACHE_FINISHED;
598 cache->last_byte_to_unpin = (u64)-1;
599 caching_ctl->progress = (u64)-1;
601 if (load_cache_only) {
602 cache->caching_ctl = NULL;
603 cache->cached = BTRFS_CACHE_NO;
605 cache->cached = BTRFS_CACHE_STARTED;
606 cache->has_caching_ctl = 1;
609 spin_unlock(&cache->lock);
610 mutex_unlock(&caching_ctl->mutex);
612 wake_up(&caching_ctl->wait);
614 put_caching_control(caching_ctl);
615 free_excluded_extents(fs_info->extent_root, cache);
620 * We are not going to do the fast caching, set cached to the
621 * appropriate value and wakeup any waiters.
623 spin_lock(&cache->lock);
624 if (load_cache_only) {
625 cache->caching_ctl = NULL;
626 cache->cached = BTRFS_CACHE_NO;
628 cache->cached = BTRFS_CACHE_STARTED;
629 cache->has_caching_ctl = 1;
631 spin_unlock(&cache->lock);
632 wake_up(&caching_ctl->wait);
635 if (load_cache_only) {
636 put_caching_control(caching_ctl);
640 down_write(&fs_info->commit_root_sem);
641 atomic_inc(&caching_ctl->count);
642 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
643 up_write(&fs_info->commit_root_sem);
645 btrfs_get_block_group(cache);
647 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
653 * return the block group that starts at or after bytenr
655 static struct btrfs_block_group_cache *
656 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
658 struct btrfs_block_group_cache *cache;
660 cache = block_group_cache_tree_search(info, bytenr, 0);
666 * return the block group that contains the given bytenr
668 struct btrfs_block_group_cache *btrfs_lookup_block_group(
669 struct btrfs_fs_info *info,
672 struct btrfs_block_group_cache *cache;
674 cache = block_group_cache_tree_search(info, bytenr, 1);
679 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
682 struct list_head *head = &info->space_info;
683 struct btrfs_space_info *found;
685 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
688 list_for_each_entry_rcu(found, head, list) {
689 if (found->flags & flags) {
699 * after adding space to the filesystem, we need to clear the full flags
700 * on all the space infos.
702 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
704 struct list_head *head = &info->space_info;
705 struct btrfs_space_info *found;
708 list_for_each_entry_rcu(found, head, list)
713 /* simple helper to search for an existing data extent at a given offset */
714 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
717 struct btrfs_key key;
718 struct btrfs_path *path;
720 path = btrfs_alloc_path();
724 key.objectid = start;
726 key.type = BTRFS_EXTENT_ITEM_KEY;
727 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
729 btrfs_free_path(path);
734 * helper function to lookup reference count and flags of a tree block.
736 * the head node for delayed ref is used to store the sum of all the
737 * reference count modifications queued up in the rbtree. the head
738 * node may also store the extent flags to set. This way you can check
739 * to see what the reference count and extent flags would be if all of
740 * the delayed refs are not processed.
742 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
743 struct btrfs_root *root, u64 bytenr,
744 u64 offset, int metadata, u64 *refs, u64 *flags)
746 struct btrfs_delayed_ref_head *head;
747 struct btrfs_delayed_ref_root *delayed_refs;
748 struct btrfs_path *path;
749 struct btrfs_extent_item *ei;
750 struct extent_buffer *leaf;
751 struct btrfs_key key;
758 * If we don't have skinny metadata, don't bother doing anything
761 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
762 offset = root->nodesize;
766 path = btrfs_alloc_path();
771 path->skip_locking = 1;
772 path->search_commit_root = 1;
776 key.objectid = bytenr;
779 key.type = BTRFS_METADATA_ITEM_KEY;
781 key.type = BTRFS_EXTENT_ITEM_KEY;
783 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
788 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
789 if (path->slots[0]) {
791 btrfs_item_key_to_cpu(path->nodes[0], &key,
793 if (key.objectid == bytenr &&
794 key.type == BTRFS_EXTENT_ITEM_KEY &&
795 key.offset == root->nodesize)
801 leaf = path->nodes[0];
802 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
803 if (item_size >= sizeof(*ei)) {
804 ei = btrfs_item_ptr(leaf, path->slots[0],
805 struct btrfs_extent_item);
806 num_refs = btrfs_extent_refs(leaf, ei);
807 extent_flags = btrfs_extent_flags(leaf, ei);
809 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
810 struct btrfs_extent_item_v0 *ei0;
811 BUG_ON(item_size != sizeof(*ei0));
812 ei0 = btrfs_item_ptr(leaf, path->slots[0],
813 struct btrfs_extent_item_v0);
814 num_refs = btrfs_extent_refs_v0(leaf, ei0);
815 /* FIXME: this isn't correct for data */
816 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
821 BUG_ON(num_refs == 0);
831 delayed_refs = &trans->transaction->delayed_refs;
832 spin_lock(&delayed_refs->lock);
833 head = btrfs_find_delayed_ref_head(trans, bytenr);
835 if (!mutex_trylock(&head->mutex)) {
836 atomic_inc(&head->node.refs);
837 spin_unlock(&delayed_refs->lock);
839 btrfs_release_path(path);
842 * Mutex was contended, block until it's released and try
845 mutex_lock(&head->mutex);
846 mutex_unlock(&head->mutex);
847 btrfs_put_delayed_ref(&head->node);
850 spin_lock(&head->lock);
851 if (head->extent_op && head->extent_op->update_flags)
852 extent_flags |= head->extent_op->flags_to_set;
854 BUG_ON(num_refs == 0);
856 num_refs += head->node.ref_mod;
857 spin_unlock(&head->lock);
858 mutex_unlock(&head->mutex);
860 spin_unlock(&delayed_refs->lock);
862 WARN_ON(num_refs == 0);
866 *flags = extent_flags;
868 btrfs_free_path(path);
873 * Back reference rules. Back refs have three main goals:
875 * 1) differentiate between all holders of references to an extent so that
876 * when a reference is dropped we can make sure it was a valid reference
877 * before freeing the extent.
879 * 2) Provide enough information to quickly find the holders of an extent
880 * if we notice a given block is corrupted or bad.
882 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
883 * maintenance. This is actually the same as #2, but with a slightly
884 * different use case.
886 * There are two kinds of back refs. The implicit back refs is optimized
887 * for pointers in non-shared tree blocks. For a given pointer in a block,
888 * back refs of this kind provide information about the block's owner tree
889 * and the pointer's key. These information allow us to find the block by
890 * b-tree searching. The full back refs is for pointers in tree blocks not
891 * referenced by their owner trees. The location of tree block is recorded
892 * in the back refs. Actually the full back refs is generic, and can be
893 * used in all cases the implicit back refs is used. The major shortcoming
894 * of the full back refs is its overhead. Every time a tree block gets
895 * COWed, we have to update back refs entry for all pointers in it.
897 * For a newly allocated tree block, we use implicit back refs for
898 * pointers in it. This means most tree related operations only involve
899 * implicit back refs. For a tree block created in old transaction, the
900 * only way to drop a reference to it is COW it. So we can detect the
901 * event that tree block loses its owner tree's reference and do the
902 * back refs conversion.
904 * When a tree block is COW'd through a tree, there are four cases:
906 * The reference count of the block is one and the tree is the block's
907 * owner tree. Nothing to do in this case.
909 * The reference count of the block is one and the tree is not the
910 * block's owner tree. In this case, full back refs is used for pointers
911 * in the block. Remove these full back refs, add implicit back refs for
912 * every pointers in the new block.
914 * The reference count of the block is greater than one and the tree is
915 * the block's owner tree. In this case, implicit back refs is used for
916 * pointers in the block. Add full back refs for every pointers in the
917 * block, increase lower level extents' reference counts. The original
918 * implicit back refs are entailed to the new block.
920 * The reference count of the block is greater than one and the tree is
921 * not the block's owner tree. Add implicit back refs for every pointer in
922 * the new block, increase lower level extents' reference count.
924 * Back Reference Key composing:
926 * The key objectid corresponds to the first byte in the extent,
927 * The key type is used to differentiate between types of back refs.
928 * There are different meanings of the key offset for different types
931 * File extents can be referenced by:
933 * - multiple snapshots, subvolumes, or different generations in one subvol
934 * - different files inside a single subvolume
935 * - different offsets inside a file (bookend extents in file.c)
937 * The extent ref structure for the implicit back refs has fields for:
939 * - Objectid of the subvolume root
940 * - objectid of the file holding the reference
941 * - original offset in the file
942 * - how many bookend extents
944 * The key offset for the implicit back refs is hash of the first
947 * The extent ref structure for the full back refs has field for:
949 * - number of pointers in the tree leaf
951 * The key offset for the implicit back refs is the first byte of
954 * When a file extent is allocated, The implicit back refs is used.
955 * the fields are filled in:
957 * (root_key.objectid, inode objectid, offset in file, 1)
959 * When a file extent is removed file truncation, we find the
960 * corresponding implicit back refs and check the following fields:
962 * (btrfs_header_owner(leaf), inode objectid, offset in file)
964 * Btree extents can be referenced by:
966 * - Different subvolumes
968 * Both the implicit back refs and the full back refs for tree blocks
969 * only consist of key. The key offset for the implicit back refs is
970 * objectid of block's owner tree. The key offset for the full back refs
971 * is the first byte of parent block.
973 * When implicit back refs is used, information about the lowest key and
974 * level of the tree block are required. These information are stored in
975 * tree block info structure.
978 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
979 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
980 struct btrfs_root *root,
981 struct btrfs_path *path,
982 u64 owner, u32 extra_size)
984 struct btrfs_extent_item *item;
985 struct btrfs_extent_item_v0 *ei0;
986 struct btrfs_extent_ref_v0 *ref0;
987 struct btrfs_tree_block_info *bi;
988 struct extent_buffer *leaf;
989 struct btrfs_key key;
990 struct btrfs_key found_key;
991 u32 new_size = sizeof(*item);
995 leaf = path->nodes[0];
996 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
998 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
999 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1000 struct btrfs_extent_item_v0);
1001 refs = btrfs_extent_refs_v0(leaf, ei0);
1003 if (owner == (u64)-1) {
1005 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1006 ret = btrfs_next_leaf(root, path);
1009 BUG_ON(ret > 0); /* Corruption */
1010 leaf = path->nodes[0];
1012 btrfs_item_key_to_cpu(leaf, &found_key,
1014 BUG_ON(key.objectid != found_key.objectid);
1015 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1019 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1020 struct btrfs_extent_ref_v0);
1021 owner = btrfs_ref_objectid_v0(leaf, ref0);
1025 btrfs_release_path(path);
1027 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1028 new_size += sizeof(*bi);
1030 new_size -= sizeof(*ei0);
1031 ret = btrfs_search_slot(trans, root, &key, path,
1032 new_size + extra_size, 1);
1035 BUG_ON(ret); /* Corruption */
1037 btrfs_extend_item(root, path, new_size);
1039 leaf = path->nodes[0];
1040 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1041 btrfs_set_extent_refs(leaf, item, refs);
1042 /* FIXME: get real generation */
1043 btrfs_set_extent_generation(leaf, item, 0);
1044 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1045 btrfs_set_extent_flags(leaf, item,
1046 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1047 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1048 bi = (struct btrfs_tree_block_info *)(item + 1);
1049 /* FIXME: get first key of the block */
1050 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1051 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1053 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1055 btrfs_mark_buffer_dirty(leaf);
1060 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1062 u32 high_crc = ~(u32)0;
1063 u32 low_crc = ~(u32)0;
1066 lenum = cpu_to_le64(root_objectid);
1067 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1068 lenum = cpu_to_le64(owner);
1069 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1070 lenum = cpu_to_le64(offset);
1071 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1073 return ((u64)high_crc << 31) ^ (u64)low_crc;
1076 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1077 struct btrfs_extent_data_ref *ref)
1079 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1080 btrfs_extent_data_ref_objectid(leaf, ref),
1081 btrfs_extent_data_ref_offset(leaf, ref));
1084 static int match_extent_data_ref(struct extent_buffer *leaf,
1085 struct btrfs_extent_data_ref *ref,
1086 u64 root_objectid, u64 owner, u64 offset)
1088 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1089 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1090 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1095 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1096 struct btrfs_root *root,
1097 struct btrfs_path *path,
1098 u64 bytenr, u64 parent,
1100 u64 owner, u64 offset)
1102 struct btrfs_key key;
1103 struct btrfs_extent_data_ref *ref;
1104 struct extent_buffer *leaf;
1110 key.objectid = bytenr;
1112 key.type = BTRFS_SHARED_DATA_REF_KEY;
1113 key.offset = parent;
1115 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1116 key.offset = hash_extent_data_ref(root_objectid,
1121 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1130 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1131 key.type = BTRFS_EXTENT_REF_V0_KEY;
1132 btrfs_release_path(path);
1133 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1144 leaf = path->nodes[0];
1145 nritems = btrfs_header_nritems(leaf);
1147 if (path->slots[0] >= nritems) {
1148 ret = btrfs_next_leaf(root, path);
1154 leaf = path->nodes[0];
1155 nritems = btrfs_header_nritems(leaf);
1159 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1160 if (key.objectid != bytenr ||
1161 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1164 ref = btrfs_item_ptr(leaf, path->slots[0],
1165 struct btrfs_extent_data_ref);
1167 if (match_extent_data_ref(leaf, ref, root_objectid,
1170 btrfs_release_path(path);
1182 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1183 struct btrfs_root *root,
1184 struct btrfs_path *path,
1185 u64 bytenr, u64 parent,
1186 u64 root_objectid, u64 owner,
1187 u64 offset, int refs_to_add)
1189 struct btrfs_key key;
1190 struct extent_buffer *leaf;
1195 key.objectid = bytenr;
1197 key.type = BTRFS_SHARED_DATA_REF_KEY;
1198 key.offset = parent;
1199 size = sizeof(struct btrfs_shared_data_ref);
1201 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1202 key.offset = hash_extent_data_ref(root_objectid,
1204 size = sizeof(struct btrfs_extent_data_ref);
1207 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1208 if (ret && ret != -EEXIST)
1211 leaf = path->nodes[0];
1213 struct btrfs_shared_data_ref *ref;
1214 ref = btrfs_item_ptr(leaf, path->slots[0],
1215 struct btrfs_shared_data_ref);
1217 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1219 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1220 num_refs += refs_to_add;
1221 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1224 struct btrfs_extent_data_ref *ref;
1225 while (ret == -EEXIST) {
1226 ref = btrfs_item_ptr(leaf, path->slots[0],
1227 struct btrfs_extent_data_ref);
1228 if (match_extent_data_ref(leaf, ref, root_objectid,
1231 btrfs_release_path(path);
1233 ret = btrfs_insert_empty_item(trans, root, path, &key,
1235 if (ret && ret != -EEXIST)
1238 leaf = path->nodes[0];
1240 ref = btrfs_item_ptr(leaf, path->slots[0],
1241 struct btrfs_extent_data_ref);
1243 btrfs_set_extent_data_ref_root(leaf, ref,
1245 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1246 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1247 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1249 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1250 num_refs += refs_to_add;
1251 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1254 btrfs_mark_buffer_dirty(leaf);
1257 btrfs_release_path(path);
1261 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1262 struct btrfs_root *root,
1263 struct btrfs_path *path,
1264 int refs_to_drop, int *last_ref)
1266 struct btrfs_key key;
1267 struct btrfs_extent_data_ref *ref1 = NULL;
1268 struct btrfs_shared_data_ref *ref2 = NULL;
1269 struct extent_buffer *leaf;
1273 leaf = path->nodes[0];
1274 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1276 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1277 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1278 struct btrfs_extent_data_ref);
1279 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1280 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1281 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1282 struct btrfs_shared_data_ref);
1283 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1284 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1285 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1286 struct btrfs_extent_ref_v0 *ref0;
1287 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1288 struct btrfs_extent_ref_v0);
1289 num_refs = btrfs_ref_count_v0(leaf, ref0);
1295 BUG_ON(num_refs < refs_to_drop);
1296 num_refs -= refs_to_drop;
1298 if (num_refs == 0) {
1299 ret = btrfs_del_item(trans, root, path);
1302 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1303 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1304 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1305 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1306 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1308 struct btrfs_extent_ref_v0 *ref0;
1309 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1310 struct btrfs_extent_ref_v0);
1311 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1314 btrfs_mark_buffer_dirty(leaf);
1319 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1320 struct btrfs_extent_inline_ref *iref)
1322 struct btrfs_key key;
1323 struct extent_buffer *leaf;
1324 struct btrfs_extent_data_ref *ref1;
1325 struct btrfs_shared_data_ref *ref2;
1328 leaf = path->nodes[0];
1329 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1331 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1332 BTRFS_EXTENT_DATA_REF_KEY) {
1333 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1334 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1336 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1337 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1339 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1340 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1341 struct btrfs_extent_data_ref);
1342 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1343 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1344 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1345 struct btrfs_shared_data_ref);
1346 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1347 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1348 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1349 struct btrfs_extent_ref_v0 *ref0;
1350 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1351 struct btrfs_extent_ref_v0);
1352 num_refs = btrfs_ref_count_v0(leaf, ref0);
1360 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1361 struct btrfs_root *root,
1362 struct btrfs_path *path,
1363 u64 bytenr, u64 parent,
1366 struct btrfs_key key;
1369 key.objectid = bytenr;
1371 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1372 key.offset = parent;
1374 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1375 key.offset = root_objectid;
1378 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1381 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1382 if (ret == -ENOENT && parent) {
1383 btrfs_release_path(path);
1384 key.type = BTRFS_EXTENT_REF_V0_KEY;
1385 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1393 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1394 struct btrfs_root *root,
1395 struct btrfs_path *path,
1396 u64 bytenr, u64 parent,
1399 struct btrfs_key key;
1402 key.objectid = bytenr;
1404 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1405 key.offset = parent;
1407 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1408 key.offset = root_objectid;
1411 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1412 btrfs_release_path(path);
1416 static inline int extent_ref_type(u64 parent, u64 owner)
1419 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1421 type = BTRFS_SHARED_BLOCK_REF_KEY;
1423 type = BTRFS_TREE_BLOCK_REF_KEY;
1426 type = BTRFS_SHARED_DATA_REF_KEY;
1428 type = BTRFS_EXTENT_DATA_REF_KEY;
1433 static int find_next_key(struct btrfs_path *path, int level,
1434 struct btrfs_key *key)
1437 for (; level < BTRFS_MAX_LEVEL; level++) {
1438 if (!path->nodes[level])
1440 if (path->slots[level] + 1 >=
1441 btrfs_header_nritems(path->nodes[level]))
1444 btrfs_item_key_to_cpu(path->nodes[level], key,
1445 path->slots[level] + 1);
1447 btrfs_node_key_to_cpu(path->nodes[level], key,
1448 path->slots[level] + 1);
1455 * look for inline back ref. if back ref is found, *ref_ret is set
1456 * to the address of inline back ref, and 0 is returned.
1458 * if back ref isn't found, *ref_ret is set to the address where it
1459 * should be inserted, and -ENOENT is returned.
1461 * if insert is true and there are too many inline back refs, the path
1462 * points to the extent item, and -EAGAIN is returned.
1464 * NOTE: inline back refs are ordered in the same way that back ref
1465 * items in the tree are ordered.
1467 static noinline_for_stack
1468 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1469 struct btrfs_root *root,
1470 struct btrfs_path *path,
1471 struct btrfs_extent_inline_ref **ref_ret,
1472 u64 bytenr, u64 num_bytes,
1473 u64 parent, u64 root_objectid,
1474 u64 owner, u64 offset, int insert)
1476 struct btrfs_key key;
1477 struct extent_buffer *leaf;
1478 struct btrfs_extent_item *ei;
1479 struct btrfs_extent_inline_ref *iref;
1489 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1492 key.objectid = bytenr;
1493 key.type = BTRFS_EXTENT_ITEM_KEY;
1494 key.offset = num_bytes;
1496 want = extent_ref_type(parent, owner);
1498 extra_size = btrfs_extent_inline_ref_size(want);
1499 path->keep_locks = 1;
1504 * Owner is our parent level, so we can just add one to get the level
1505 * for the block we are interested in.
1507 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1508 key.type = BTRFS_METADATA_ITEM_KEY;
1513 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1520 * We may be a newly converted file system which still has the old fat
1521 * extent entries for metadata, so try and see if we have one of those.
1523 if (ret > 0 && skinny_metadata) {
1524 skinny_metadata = false;
1525 if (path->slots[0]) {
1527 btrfs_item_key_to_cpu(path->nodes[0], &key,
1529 if (key.objectid == bytenr &&
1530 key.type == BTRFS_EXTENT_ITEM_KEY &&
1531 key.offset == num_bytes)
1535 key.objectid = bytenr;
1536 key.type = BTRFS_EXTENT_ITEM_KEY;
1537 key.offset = num_bytes;
1538 btrfs_release_path(path);
1543 if (ret && !insert) {
1546 } else if (WARN_ON(ret)) {
1551 leaf = path->nodes[0];
1552 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1553 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1554 if (item_size < sizeof(*ei)) {
1559 ret = convert_extent_item_v0(trans, root, path, owner,
1565 leaf = path->nodes[0];
1566 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1569 BUG_ON(item_size < sizeof(*ei));
1571 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1572 flags = btrfs_extent_flags(leaf, ei);
1574 ptr = (unsigned long)(ei + 1);
1575 end = (unsigned long)ei + item_size;
1577 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1578 ptr += sizeof(struct btrfs_tree_block_info);
1588 iref = (struct btrfs_extent_inline_ref *)ptr;
1589 type = btrfs_extent_inline_ref_type(leaf, iref);
1593 ptr += btrfs_extent_inline_ref_size(type);
1597 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1598 struct btrfs_extent_data_ref *dref;
1599 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1600 if (match_extent_data_ref(leaf, dref, root_objectid,
1605 if (hash_extent_data_ref_item(leaf, dref) <
1606 hash_extent_data_ref(root_objectid, owner, offset))
1610 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1612 if (parent == ref_offset) {
1616 if (ref_offset < parent)
1619 if (root_objectid == ref_offset) {
1623 if (ref_offset < root_objectid)
1627 ptr += btrfs_extent_inline_ref_size(type);
1629 if (err == -ENOENT && insert) {
1630 if (item_size + extra_size >=
1631 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1636 * To add new inline back ref, we have to make sure
1637 * there is no corresponding back ref item.
1638 * For simplicity, we just do not add new inline back
1639 * ref if there is any kind of item for this block
1641 if (find_next_key(path, 0, &key) == 0 &&
1642 key.objectid == bytenr &&
1643 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1648 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1651 path->keep_locks = 0;
1652 btrfs_unlock_up_safe(path, 1);
1658 * helper to add new inline back ref
1660 static noinline_for_stack
1661 void setup_inline_extent_backref(struct btrfs_root *root,
1662 struct btrfs_path *path,
1663 struct btrfs_extent_inline_ref *iref,
1664 u64 parent, u64 root_objectid,
1665 u64 owner, u64 offset, int refs_to_add,
1666 struct btrfs_delayed_extent_op *extent_op)
1668 struct extent_buffer *leaf;
1669 struct btrfs_extent_item *ei;
1672 unsigned long item_offset;
1677 leaf = path->nodes[0];
1678 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1679 item_offset = (unsigned long)iref - (unsigned long)ei;
1681 type = extent_ref_type(parent, owner);
1682 size = btrfs_extent_inline_ref_size(type);
1684 btrfs_extend_item(root, path, size);
1686 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1687 refs = btrfs_extent_refs(leaf, ei);
1688 refs += refs_to_add;
1689 btrfs_set_extent_refs(leaf, ei, refs);
1691 __run_delayed_extent_op(extent_op, leaf, ei);
1693 ptr = (unsigned long)ei + item_offset;
1694 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1695 if (ptr < end - size)
1696 memmove_extent_buffer(leaf, ptr + size, ptr,
1699 iref = (struct btrfs_extent_inline_ref *)ptr;
1700 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1701 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1702 struct btrfs_extent_data_ref *dref;
1703 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1704 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1705 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1706 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1707 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1708 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1709 struct btrfs_shared_data_ref *sref;
1710 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1711 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1712 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1713 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1714 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1716 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1718 btrfs_mark_buffer_dirty(leaf);
1721 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1722 struct btrfs_root *root,
1723 struct btrfs_path *path,
1724 struct btrfs_extent_inline_ref **ref_ret,
1725 u64 bytenr, u64 num_bytes, u64 parent,
1726 u64 root_objectid, u64 owner, u64 offset)
1730 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1731 bytenr, num_bytes, parent,
1732 root_objectid, owner, offset, 0);
1736 btrfs_release_path(path);
1739 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1740 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1743 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1744 root_objectid, owner, offset);
1750 * helper to update/remove inline back ref
1752 static noinline_for_stack
1753 void update_inline_extent_backref(struct btrfs_root *root,
1754 struct btrfs_path *path,
1755 struct btrfs_extent_inline_ref *iref,
1757 struct btrfs_delayed_extent_op *extent_op,
1760 struct extent_buffer *leaf;
1761 struct btrfs_extent_item *ei;
1762 struct btrfs_extent_data_ref *dref = NULL;
1763 struct btrfs_shared_data_ref *sref = NULL;
1771 leaf = path->nodes[0];
1772 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1773 refs = btrfs_extent_refs(leaf, ei);
1774 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1775 refs += refs_to_mod;
1776 btrfs_set_extent_refs(leaf, ei, refs);
1778 __run_delayed_extent_op(extent_op, leaf, ei);
1780 type = btrfs_extent_inline_ref_type(leaf, iref);
1782 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1783 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1784 refs = btrfs_extent_data_ref_count(leaf, dref);
1785 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1786 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1787 refs = btrfs_shared_data_ref_count(leaf, sref);
1790 BUG_ON(refs_to_mod != -1);
1793 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1794 refs += refs_to_mod;
1797 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1798 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1800 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1803 size = btrfs_extent_inline_ref_size(type);
1804 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1805 ptr = (unsigned long)iref;
1806 end = (unsigned long)ei + item_size;
1807 if (ptr + size < end)
1808 memmove_extent_buffer(leaf, ptr, ptr + size,
1811 btrfs_truncate_item(root, path, item_size, 1);
1813 btrfs_mark_buffer_dirty(leaf);
1816 static noinline_for_stack
1817 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1818 struct btrfs_root *root,
1819 struct btrfs_path *path,
1820 u64 bytenr, u64 num_bytes, u64 parent,
1821 u64 root_objectid, u64 owner,
1822 u64 offset, int refs_to_add,
1823 struct btrfs_delayed_extent_op *extent_op)
1825 struct btrfs_extent_inline_ref *iref;
1828 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1829 bytenr, num_bytes, parent,
1830 root_objectid, owner, offset, 1);
1832 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1833 update_inline_extent_backref(root, path, iref,
1834 refs_to_add, extent_op, NULL);
1835 } else if (ret == -ENOENT) {
1836 setup_inline_extent_backref(root, path, iref, parent,
1837 root_objectid, owner, offset,
1838 refs_to_add, extent_op);
1844 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1845 struct btrfs_root *root,
1846 struct btrfs_path *path,
1847 u64 bytenr, u64 parent, u64 root_objectid,
1848 u64 owner, u64 offset, int refs_to_add)
1851 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1852 BUG_ON(refs_to_add != 1);
1853 ret = insert_tree_block_ref(trans, root, path, bytenr,
1854 parent, root_objectid);
1856 ret = insert_extent_data_ref(trans, root, path, bytenr,
1857 parent, root_objectid,
1858 owner, offset, refs_to_add);
1863 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1864 struct btrfs_root *root,
1865 struct btrfs_path *path,
1866 struct btrfs_extent_inline_ref *iref,
1867 int refs_to_drop, int is_data, int *last_ref)
1871 BUG_ON(!is_data && refs_to_drop != 1);
1873 update_inline_extent_backref(root, path, iref,
1874 -refs_to_drop, NULL, last_ref);
1875 } else if (is_data) {
1876 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1880 ret = btrfs_del_item(trans, root, path);
1885 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1886 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1887 u64 *discarded_bytes)
1890 u64 bytes_left, end;
1891 u64 aligned_start = ALIGN(start, 1 << 9);
1893 if (WARN_ON(start != aligned_start)) {
1894 len -= aligned_start - start;
1895 len = round_down(len, 1 << 9);
1896 start = aligned_start;
1899 *discarded_bytes = 0;
1907 /* Skip any superblocks on this device. */
1908 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1909 u64 sb_start = btrfs_sb_offset(j);
1910 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1911 u64 size = sb_start - start;
1913 if (!in_range(sb_start, start, bytes_left) &&
1914 !in_range(sb_end, start, bytes_left) &&
1915 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1919 * Superblock spans beginning of range. Adjust start and
1922 if (sb_start <= start) {
1923 start += sb_end - start;
1928 bytes_left = end - start;
1933 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
1936 *discarded_bytes += size;
1937 else if (ret != -EOPNOTSUPP)
1946 bytes_left = end - start;
1950 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
1953 *discarded_bytes += bytes_left;
1958 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1959 u64 num_bytes, u64 *actual_bytes)
1962 u64 discarded_bytes = 0;
1963 struct btrfs_bio *bbio = NULL;
1966 /* Tell the block device(s) that the sectors can be discarded */
1967 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1968 bytenr, &num_bytes, &bbio, 0);
1969 /* Error condition is -ENOMEM */
1971 struct btrfs_bio_stripe *stripe = bbio->stripes;
1975 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1977 if (!stripe->dev->can_discard)
1980 ret = btrfs_issue_discard(stripe->dev->bdev,
1985 discarded_bytes += bytes;
1986 else if (ret != -EOPNOTSUPP)
1987 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1990 * Just in case we get back EOPNOTSUPP for some reason,
1991 * just ignore the return value so we don't screw up
1992 * people calling discard_extent.
1996 btrfs_put_bbio(bbio);
2000 *actual_bytes = discarded_bytes;
2003 if (ret == -EOPNOTSUPP)
2008 /* Can return -ENOMEM */
2009 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2010 struct btrfs_root *root,
2011 u64 bytenr, u64 num_bytes, u64 parent,
2012 u64 root_objectid, u64 owner, u64 offset,
2016 struct btrfs_fs_info *fs_info = root->fs_info;
2018 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2019 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2021 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2022 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2024 parent, root_objectid, (int)owner,
2025 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
2027 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2029 parent, root_objectid, owner, offset,
2030 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
2035 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2036 struct btrfs_root *root,
2037 struct btrfs_delayed_ref_node *node,
2038 u64 parent, u64 root_objectid,
2039 u64 owner, u64 offset, int refs_to_add,
2040 struct btrfs_delayed_extent_op *extent_op)
2042 struct btrfs_fs_info *fs_info = root->fs_info;
2043 struct btrfs_path *path;
2044 struct extent_buffer *leaf;
2045 struct btrfs_extent_item *item;
2046 struct btrfs_key key;
2047 u64 bytenr = node->bytenr;
2048 u64 num_bytes = node->num_bytes;
2051 int no_quota = node->no_quota;
2053 path = btrfs_alloc_path();
2057 if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
2061 path->leave_spinning = 1;
2062 /* this will setup the path even if it fails to insert the back ref */
2063 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2064 bytenr, num_bytes, parent,
2065 root_objectid, owner, offset,
2066 refs_to_add, extent_op);
2067 if ((ret < 0 && ret != -EAGAIN) || !ret)
2071 * Ok we had -EAGAIN which means we didn't have space to insert and
2072 * inline extent ref, so just update the reference count and add a
2075 leaf = path->nodes[0];
2076 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2077 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2078 refs = btrfs_extent_refs(leaf, item);
2079 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2081 __run_delayed_extent_op(extent_op, leaf, item);
2083 btrfs_mark_buffer_dirty(leaf);
2084 btrfs_release_path(path);
2087 path->leave_spinning = 1;
2088 /* now insert the actual backref */
2089 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2090 path, bytenr, parent, root_objectid,
2091 owner, offset, refs_to_add);
2093 btrfs_abort_transaction(trans, root, ret);
2095 btrfs_free_path(path);
2099 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2100 struct btrfs_root *root,
2101 struct btrfs_delayed_ref_node *node,
2102 struct btrfs_delayed_extent_op *extent_op,
2103 int insert_reserved)
2106 struct btrfs_delayed_data_ref *ref;
2107 struct btrfs_key ins;
2112 ins.objectid = node->bytenr;
2113 ins.offset = node->num_bytes;
2114 ins.type = BTRFS_EXTENT_ITEM_KEY;
2116 ref = btrfs_delayed_node_to_data_ref(node);
2117 trace_run_delayed_data_ref(node, ref, node->action);
2119 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2120 parent = ref->parent;
2121 ref_root = ref->root;
2123 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2125 flags |= extent_op->flags_to_set;
2126 ret = alloc_reserved_file_extent(trans, root,
2127 parent, ref_root, flags,
2128 ref->objectid, ref->offset,
2129 &ins, node->ref_mod);
2130 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2131 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2132 ref_root, ref->objectid,
2133 ref->offset, node->ref_mod,
2135 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2136 ret = __btrfs_free_extent(trans, root, node, parent,
2137 ref_root, ref->objectid,
2138 ref->offset, node->ref_mod,
2146 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2147 struct extent_buffer *leaf,
2148 struct btrfs_extent_item *ei)
2150 u64 flags = btrfs_extent_flags(leaf, ei);
2151 if (extent_op->update_flags) {
2152 flags |= extent_op->flags_to_set;
2153 btrfs_set_extent_flags(leaf, ei, flags);
2156 if (extent_op->update_key) {
2157 struct btrfs_tree_block_info *bi;
2158 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2159 bi = (struct btrfs_tree_block_info *)(ei + 1);
2160 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2164 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2165 struct btrfs_root *root,
2166 struct btrfs_delayed_ref_node *node,
2167 struct btrfs_delayed_extent_op *extent_op)
2169 struct btrfs_key key;
2170 struct btrfs_path *path;
2171 struct btrfs_extent_item *ei;
2172 struct extent_buffer *leaf;
2176 int metadata = !extent_op->is_data;
2181 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2184 path = btrfs_alloc_path();
2188 key.objectid = node->bytenr;
2191 key.type = BTRFS_METADATA_ITEM_KEY;
2192 key.offset = extent_op->level;
2194 key.type = BTRFS_EXTENT_ITEM_KEY;
2195 key.offset = node->num_bytes;
2200 path->leave_spinning = 1;
2201 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2209 if (path->slots[0] > 0) {
2211 btrfs_item_key_to_cpu(path->nodes[0], &key,
2213 if (key.objectid == node->bytenr &&
2214 key.type == BTRFS_EXTENT_ITEM_KEY &&
2215 key.offset == node->num_bytes)
2219 btrfs_release_path(path);
2222 key.objectid = node->bytenr;
2223 key.offset = node->num_bytes;
2224 key.type = BTRFS_EXTENT_ITEM_KEY;
2233 leaf = path->nodes[0];
2234 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2235 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2236 if (item_size < sizeof(*ei)) {
2237 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2243 leaf = path->nodes[0];
2244 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2247 BUG_ON(item_size < sizeof(*ei));
2248 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2249 __run_delayed_extent_op(extent_op, leaf, ei);
2251 btrfs_mark_buffer_dirty(leaf);
2253 btrfs_free_path(path);
2257 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2258 struct btrfs_root *root,
2259 struct btrfs_delayed_ref_node *node,
2260 struct btrfs_delayed_extent_op *extent_op,
2261 int insert_reserved)
2264 struct btrfs_delayed_tree_ref *ref;
2265 struct btrfs_key ins;
2268 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2271 ref = btrfs_delayed_node_to_tree_ref(node);
2272 trace_run_delayed_tree_ref(node, ref, node->action);
2274 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2275 parent = ref->parent;
2276 ref_root = ref->root;
2278 ins.objectid = node->bytenr;
2279 if (skinny_metadata) {
2280 ins.offset = ref->level;
2281 ins.type = BTRFS_METADATA_ITEM_KEY;
2283 ins.offset = node->num_bytes;
2284 ins.type = BTRFS_EXTENT_ITEM_KEY;
2287 BUG_ON(node->ref_mod != 1);
2288 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2289 BUG_ON(!extent_op || !extent_op->update_flags);
2290 ret = alloc_reserved_tree_block(trans, root,
2292 extent_op->flags_to_set,
2296 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2297 ret = __btrfs_inc_extent_ref(trans, root, node,
2301 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2302 ret = __btrfs_free_extent(trans, root, node,
2304 ref->level, 0, 1, extent_op);
2311 /* helper function to actually process a single delayed ref entry */
2312 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2313 struct btrfs_root *root,
2314 struct btrfs_delayed_ref_node *node,
2315 struct btrfs_delayed_extent_op *extent_op,
2316 int insert_reserved)
2320 if (trans->aborted) {
2321 if (insert_reserved)
2322 btrfs_pin_extent(root, node->bytenr,
2323 node->num_bytes, 1);
2327 if (btrfs_delayed_ref_is_head(node)) {
2328 struct btrfs_delayed_ref_head *head;
2330 * we've hit the end of the chain and we were supposed
2331 * to insert this extent into the tree. But, it got
2332 * deleted before we ever needed to insert it, so all
2333 * we have to do is clean up the accounting
2336 head = btrfs_delayed_node_to_head(node);
2337 trace_run_delayed_ref_head(node, head, node->action);
2339 if (insert_reserved) {
2340 btrfs_pin_extent(root, node->bytenr,
2341 node->num_bytes, 1);
2342 if (head->is_data) {
2343 ret = btrfs_del_csums(trans, root,
2351 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2352 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2353 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2355 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2356 node->type == BTRFS_SHARED_DATA_REF_KEY)
2357 ret = run_delayed_data_ref(trans, root, node, extent_op,
2364 static inline struct btrfs_delayed_ref_node *
2365 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2367 struct btrfs_delayed_ref_node *ref;
2369 if (list_empty(&head->ref_list))
2373 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2374 * This is to prevent a ref count from going down to zero, which deletes
2375 * the extent item from the extent tree, when there still are references
2376 * to add, which would fail because they would not find the extent item.
2378 list_for_each_entry(ref, &head->ref_list, list) {
2379 if (ref->action == BTRFS_ADD_DELAYED_REF)
2383 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2388 * Returns 0 on success or if called with an already aborted transaction.
2389 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2391 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2392 struct btrfs_root *root,
2395 struct btrfs_delayed_ref_root *delayed_refs;
2396 struct btrfs_delayed_ref_node *ref;
2397 struct btrfs_delayed_ref_head *locked_ref = NULL;
2398 struct btrfs_delayed_extent_op *extent_op;
2399 struct btrfs_fs_info *fs_info = root->fs_info;
2400 ktime_t start = ktime_get();
2402 unsigned long count = 0;
2403 unsigned long actual_count = 0;
2404 int must_insert_reserved = 0;
2406 delayed_refs = &trans->transaction->delayed_refs;
2412 spin_lock(&delayed_refs->lock);
2413 locked_ref = btrfs_select_ref_head(trans);
2415 spin_unlock(&delayed_refs->lock);
2419 /* grab the lock that says we are going to process
2420 * all the refs for this head */
2421 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2422 spin_unlock(&delayed_refs->lock);
2424 * we may have dropped the spin lock to get the head
2425 * mutex lock, and that might have given someone else
2426 * time to free the head. If that's true, it has been
2427 * removed from our list and we can move on.
2429 if (ret == -EAGAIN) {
2436 spin_lock(&locked_ref->lock);
2439 * locked_ref is the head node, so we have to go one
2440 * node back for any delayed ref updates
2442 ref = select_delayed_ref(locked_ref);
2444 if (ref && ref->seq &&
2445 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2446 spin_unlock(&locked_ref->lock);
2447 btrfs_delayed_ref_unlock(locked_ref);
2448 spin_lock(&delayed_refs->lock);
2449 locked_ref->processing = 0;
2450 delayed_refs->num_heads_ready++;
2451 spin_unlock(&delayed_refs->lock);
2459 * record the must insert reserved flag before we
2460 * drop the spin lock.
2462 must_insert_reserved = locked_ref->must_insert_reserved;
2463 locked_ref->must_insert_reserved = 0;
2465 extent_op = locked_ref->extent_op;
2466 locked_ref->extent_op = NULL;
2471 /* All delayed refs have been processed, Go ahead
2472 * and send the head node to run_one_delayed_ref,
2473 * so that any accounting fixes can happen
2475 ref = &locked_ref->node;
2477 if (extent_op && must_insert_reserved) {
2478 btrfs_free_delayed_extent_op(extent_op);
2483 spin_unlock(&locked_ref->lock);
2484 ret = run_delayed_extent_op(trans, root,
2486 btrfs_free_delayed_extent_op(extent_op);
2490 * Need to reset must_insert_reserved if
2491 * there was an error so the abort stuff
2492 * can cleanup the reserved space
2495 if (must_insert_reserved)
2496 locked_ref->must_insert_reserved = 1;
2497 locked_ref->processing = 0;
2498 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2499 btrfs_delayed_ref_unlock(locked_ref);
2506 * Need to drop our head ref lock and re-aqcuire the
2507 * delayed ref lock and then re-check to make sure
2510 spin_unlock(&locked_ref->lock);
2511 spin_lock(&delayed_refs->lock);
2512 spin_lock(&locked_ref->lock);
2513 if (!list_empty(&locked_ref->ref_list) ||
2514 locked_ref->extent_op) {
2515 spin_unlock(&locked_ref->lock);
2516 spin_unlock(&delayed_refs->lock);
2520 delayed_refs->num_heads--;
2521 rb_erase(&locked_ref->href_node,
2522 &delayed_refs->href_root);
2523 spin_unlock(&delayed_refs->lock);
2527 list_del(&ref->list);
2529 atomic_dec(&delayed_refs->num_entries);
2531 if (!btrfs_delayed_ref_is_head(ref)) {
2533 * when we play the delayed ref, also correct the
2536 switch (ref->action) {
2537 case BTRFS_ADD_DELAYED_REF:
2538 case BTRFS_ADD_DELAYED_EXTENT:
2539 locked_ref->node.ref_mod -= ref->ref_mod;
2541 case BTRFS_DROP_DELAYED_REF:
2542 locked_ref->node.ref_mod += ref->ref_mod;
2548 spin_unlock(&locked_ref->lock);
2550 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2551 must_insert_reserved);
2553 btrfs_free_delayed_extent_op(extent_op);
2555 locked_ref->processing = 0;
2556 btrfs_delayed_ref_unlock(locked_ref);
2557 btrfs_put_delayed_ref(ref);
2558 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2563 * If this node is a head, that means all the refs in this head
2564 * have been dealt with, and we will pick the next head to deal
2565 * with, so we must unlock the head and drop it from the cluster
2566 * list before we release it.
2568 if (btrfs_delayed_ref_is_head(ref)) {
2569 if (locked_ref->is_data &&
2570 locked_ref->total_ref_mod < 0) {
2571 spin_lock(&delayed_refs->lock);
2572 delayed_refs->pending_csums -= ref->num_bytes;
2573 spin_unlock(&delayed_refs->lock);
2575 btrfs_delayed_ref_unlock(locked_ref);
2578 btrfs_put_delayed_ref(ref);
2584 * We don't want to include ref heads since we can have empty ref heads
2585 * and those will drastically skew our runtime down since we just do
2586 * accounting, no actual extent tree updates.
2588 if (actual_count > 0) {
2589 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2593 * We weigh the current average higher than our current runtime
2594 * to avoid large swings in the average.
2596 spin_lock(&delayed_refs->lock);
2597 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2598 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2599 spin_unlock(&delayed_refs->lock);
2604 #ifdef SCRAMBLE_DELAYED_REFS
2606 * Normally delayed refs get processed in ascending bytenr order. This
2607 * correlates in most cases to the order added. To expose dependencies on this
2608 * order, we start to process the tree in the middle instead of the beginning
2610 static u64 find_middle(struct rb_root *root)
2612 struct rb_node *n = root->rb_node;
2613 struct btrfs_delayed_ref_node *entry;
2616 u64 first = 0, last = 0;
2620 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2621 first = entry->bytenr;
2625 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2626 last = entry->bytenr;
2631 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2632 WARN_ON(!entry->in_tree);
2634 middle = entry->bytenr;
2647 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2651 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2652 sizeof(struct btrfs_extent_inline_ref));
2653 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2654 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2657 * We don't ever fill up leaves all the way so multiply by 2 just to be
2658 * closer to what we're really going to want to ouse.
2660 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2664 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2665 * would require to store the csums for that many bytes.
2667 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2670 u64 num_csums_per_leaf;
2673 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2674 num_csums_per_leaf = div64_u64(csum_size,
2675 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2676 num_csums = div64_u64(csum_bytes, root->sectorsize);
2677 num_csums += num_csums_per_leaf - 1;
2678 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2682 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2683 struct btrfs_root *root)
2685 struct btrfs_block_rsv *global_rsv;
2686 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2687 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2688 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2689 u64 num_bytes, num_dirty_bgs_bytes;
2692 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2693 num_heads = heads_to_leaves(root, num_heads);
2695 num_bytes += (num_heads - 1) * root->nodesize;
2697 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2698 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2700 global_rsv = &root->fs_info->global_block_rsv;
2703 * If we can't allocate any more chunks lets make sure we have _lots_ of
2704 * wiggle room since running delayed refs can create more delayed refs.
2706 if (global_rsv->space_info->full) {
2707 num_dirty_bgs_bytes <<= 1;
2711 spin_lock(&global_rsv->lock);
2712 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2714 spin_unlock(&global_rsv->lock);
2718 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2719 struct btrfs_root *root)
2721 struct btrfs_fs_info *fs_info = root->fs_info;
2723 atomic_read(&trans->transaction->delayed_refs.num_entries);
2728 avg_runtime = fs_info->avg_delayed_ref_runtime;
2729 val = num_entries * avg_runtime;
2730 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2732 if (val >= NSEC_PER_SEC / 2)
2735 return btrfs_check_space_for_delayed_refs(trans, root);
2738 struct async_delayed_refs {
2739 struct btrfs_root *root;
2743 struct completion wait;
2744 struct btrfs_work work;
2747 static void delayed_ref_async_start(struct btrfs_work *work)
2749 struct async_delayed_refs *async;
2750 struct btrfs_trans_handle *trans;
2753 async = container_of(work, struct async_delayed_refs, work);
2755 trans = btrfs_join_transaction(async->root);
2756 if (IS_ERR(trans)) {
2757 async->error = PTR_ERR(trans);
2762 * trans->sync means that when we call end_transaciton, we won't
2763 * wait on delayed refs
2766 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2770 ret = btrfs_end_transaction(trans, async->root);
2771 if (ret && !async->error)
2775 complete(&async->wait);
2780 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2781 unsigned long count, int wait)
2783 struct async_delayed_refs *async;
2786 async = kmalloc(sizeof(*async), GFP_NOFS);
2790 async->root = root->fs_info->tree_root;
2791 async->count = count;
2797 init_completion(&async->wait);
2799 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2800 delayed_ref_async_start, NULL, NULL);
2802 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2805 wait_for_completion(&async->wait);
2814 * this starts processing the delayed reference count updates and
2815 * extent insertions we have queued up so far. count can be
2816 * 0, which means to process everything in the tree at the start
2817 * of the run (but not newly added entries), or it can be some target
2818 * number you'd like to process.
2820 * Returns 0 on success or if called with an aborted transaction
2821 * Returns <0 on error and aborts the transaction
2823 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2824 struct btrfs_root *root, unsigned long count)
2826 struct rb_node *node;
2827 struct btrfs_delayed_ref_root *delayed_refs;
2828 struct btrfs_delayed_ref_head *head;
2830 int run_all = count == (unsigned long)-1;
2832 /* We'll clean this up in btrfs_cleanup_transaction */
2836 if (root == root->fs_info->extent_root)
2837 root = root->fs_info->tree_root;
2839 delayed_refs = &trans->transaction->delayed_refs;
2841 count = atomic_read(&delayed_refs->num_entries) * 2;
2844 #ifdef SCRAMBLE_DELAYED_REFS
2845 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2847 ret = __btrfs_run_delayed_refs(trans, root, count);
2849 btrfs_abort_transaction(trans, root, ret);
2854 if (!list_empty(&trans->new_bgs))
2855 btrfs_create_pending_block_groups(trans, root);
2857 spin_lock(&delayed_refs->lock);
2858 node = rb_first(&delayed_refs->href_root);
2860 spin_unlock(&delayed_refs->lock);
2863 count = (unsigned long)-1;
2866 head = rb_entry(node, struct btrfs_delayed_ref_head,
2868 if (btrfs_delayed_ref_is_head(&head->node)) {
2869 struct btrfs_delayed_ref_node *ref;
2872 atomic_inc(&ref->refs);
2874 spin_unlock(&delayed_refs->lock);
2876 * Mutex was contended, block until it's
2877 * released and try again
2879 mutex_lock(&head->mutex);
2880 mutex_unlock(&head->mutex);
2882 btrfs_put_delayed_ref(ref);
2888 node = rb_next(node);
2890 spin_unlock(&delayed_refs->lock);
2895 assert_qgroups_uptodate(trans);
2899 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2900 struct btrfs_root *root,
2901 u64 bytenr, u64 num_bytes, u64 flags,
2902 int level, int is_data)
2904 struct btrfs_delayed_extent_op *extent_op;
2907 extent_op = btrfs_alloc_delayed_extent_op();
2911 extent_op->flags_to_set = flags;
2912 extent_op->update_flags = 1;
2913 extent_op->update_key = 0;
2914 extent_op->is_data = is_data ? 1 : 0;
2915 extent_op->level = level;
2917 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2918 num_bytes, extent_op);
2920 btrfs_free_delayed_extent_op(extent_op);
2924 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2925 struct btrfs_root *root,
2926 struct btrfs_path *path,
2927 u64 objectid, u64 offset, u64 bytenr)
2929 struct btrfs_delayed_ref_head *head;
2930 struct btrfs_delayed_ref_node *ref;
2931 struct btrfs_delayed_data_ref *data_ref;
2932 struct btrfs_delayed_ref_root *delayed_refs;
2935 delayed_refs = &trans->transaction->delayed_refs;
2936 spin_lock(&delayed_refs->lock);
2937 head = btrfs_find_delayed_ref_head(trans, bytenr);
2939 spin_unlock(&delayed_refs->lock);
2943 if (!mutex_trylock(&head->mutex)) {
2944 atomic_inc(&head->node.refs);
2945 spin_unlock(&delayed_refs->lock);
2947 btrfs_release_path(path);
2950 * Mutex was contended, block until it's released and let
2953 mutex_lock(&head->mutex);
2954 mutex_unlock(&head->mutex);
2955 btrfs_put_delayed_ref(&head->node);
2958 spin_unlock(&delayed_refs->lock);
2960 spin_lock(&head->lock);
2961 list_for_each_entry(ref, &head->ref_list, list) {
2962 /* If it's a shared ref we know a cross reference exists */
2963 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2968 data_ref = btrfs_delayed_node_to_data_ref(ref);
2971 * If our ref doesn't match the one we're currently looking at
2972 * then we have a cross reference.
2974 if (data_ref->root != root->root_key.objectid ||
2975 data_ref->objectid != objectid ||
2976 data_ref->offset != offset) {
2981 spin_unlock(&head->lock);
2982 mutex_unlock(&head->mutex);
2986 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2987 struct btrfs_root *root,
2988 struct btrfs_path *path,
2989 u64 objectid, u64 offset, u64 bytenr)
2991 struct btrfs_root *extent_root = root->fs_info->extent_root;
2992 struct extent_buffer *leaf;
2993 struct btrfs_extent_data_ref *ref;
2994 struct btrfs_extent_inline_ref *iref;
2995 struct btrfs_extent_item *ei;
2996 struct btrfs_key key;
3000 key.objectid = bytenr;
3001 key.offset = (u64)-1;
3002 key.type = BTRFS_EXTENT_ITEM_KEY;
3004 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3007 BUG_ON(ret == 0); /* Corruption */
3010 if (path->slots[0] == 0)
3014 leaf = path->nodes[0];
3015 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3017 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3021 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3022 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3023 if (item_size < sizeof(*ei)) {
3024 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3028 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3030 if (item_size != sizeof(*ei) +
3031 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3034 if (btrfs_extent_generation(leaf, ei) <=
3035 btrfs_root_last_snapshot(&root->root_item))
3038 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3039 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3040 BTRFS_EXTENT_DATA_REF_KEY)
3043 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3044 if (btrfs_extent_refs(leaf, ei) !=
3045 btrfs_extent_data_ref_count(leaf, ref) ||
3046 btrfs_extent_data_ref_root(leaf, ref) !=
3047 root->root_key.objectid ||
3048 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3049 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3057 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3058 struct btrfs_root *root,
3059 u64 objectid, u64 offset, u64 bytenr)
3061 struct btrfs_path *path;
3065 path = btrfs_alloc_path();
3070 ret = check_committed_ref(trans, root, path, objectid,
3072 if (ret && ret != -ENOENT)
3075 ret2 = check_delayed_ref(trans, root, path, objectid,
3077 } while (ret2 == -EAGAIN);
3079 if (ret2 && ret2 != -ENOENT) {
3084 if (ret != -ENOENT || ret2 != -ENOENT)
3087 btrfs_free_path(path);
3088 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3093 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3094 struct btrfs_root *root,
3095 struct extent_buffer *buf,
3096 int full_backref, int inc)
3103 struct btrfs_key key;
3104 struct btrfs_file_extent_item *fi;
3108 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3109 u64, u64, u64, u64, u64, u64, int);
3112 if (btrfs_test_is_dummy_root(root))
3115 ref_root = btrfs_header_owner(buf);
3116 nritems = btrfs_header_nritems(buf);
3117 level = btrfs_header_level(buf);
3119 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3123 process_func = btrfs_inc_extent_ref;
3125 process_func = btrfs_free_extent;
3128 parent = buf->start;
3132 for (i = 0; i < nritems; i++) {
3134 btrfs_item_key_to_cpu(buf, &key, i);
3135 if (key.type != BTRFS_EXTENT_DATA_KEY)
3137 fi = btrfs_item_ptr(buf, i,
3138 struct btrfs_file_extent_item);
3139 if (btrfs_file_extent_type(buf, fi) ==
3140 BTRFS_FILE_EXTENT_INLINE)
3142 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3146 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3147 key.offset -= btrfs_file_extent_offset(buf, fi);
3148 ret = process_func(trans, root, bytenr, num_bytes,
3149 parent, ref_root, key.objectid,
3154 bytenr = btrfs_node_blockptr(buf, i);
3155 num_bytes = root->nodesize;
3156 ret = process_func(trans, root, bytenr, num_bytes,
3157 parent, ref_root, level - 1, 0,
3168 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3169 struct extent_buffer *buf, int full_backref)
3171 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3174 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3175 struct extent_buffer *buf, int full_backref)
3177 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3180 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3181 struct btrfs_root *root,
3182 struct btrfs_path *path,
3183 struct btrfs_block_group_cache *cache)
3186 struct btrfs_root *extent_root = root->fs_info->extent_root;
3188 struct extent_buffer *leaf;
3190 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3197 leaf = path->nodes[0];
3198 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3199 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3200 btrfs_mark_buffer_dirty(leaf);
3202 btrfs_release_path(path);
3207 static struct btrfs_block_group_cache *
3208 next_block_group(struct btrfs_root *root,
3209 struct btrfs_block_group_cache *cache)
3211 struct rb_node *node;
3213 spin_lock(&root->fs_info->block_group_cache_lock);
3215 /* If our block group was removed, we need a full search. */
3216 if (RB_EMPTY_NODE(&cache->cache_node)) {
3217 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3219 spin_unlock(&root->fs_info->block_group_cache_lock);
3220 btrfs_put_block_group(cache);
3221 cache = btrfs_lookup_first_block_group(root->fs_info,
3225 node = rb_next(&cache->cache_node);
3226 btrfs_put_block_group(cache);
3228 cache = rb_entry(node, struct btrfs_block_group_cache,
3230 btrfs_get_block_group(cache);
3233 spin_unlock(&root->fs_info->block_group_cache_lock);
3237 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3238 struct btrfs_trans_handle *trans,
3239 struct btrfs_path *path)
3241 struct btrfs_root *root = block_group->fs_info->tree_root;
3242 struct inode *inode = NULL;
3244 int dcs = BTRFS_DC_ERROR;
3250 * If this block group is smaller than 100 megs don't bother caching the
3253 if (block_group->key.offset < (100 * 1024 * 1024)) {
3254 spin_lock(&block_group->lock);
3255 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3256 spin_unlock(&block_group->lock);
3263 inode = lookup_free_space_inode(root, block_group, path);
3264 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3265 ret = PTR_ERR(inode);
3266 btrfs_release_path(path);
3270 if (IS_ERR(inode)) {
3274 if (block_group->ro)
3277 ret = create_free_space_inode(root, trans, block_group, path);
3283 /* We've already setup this transaction, go ahead and exit */
3284 if (block_group->cache_generation == trans->transid &&
3285 i_size_read(inode)) {
3286 dcs = BTRFS_DC_SETUP;
3291 * We want to set the generation to 0, that way if anything goes wrong
3292 * from here on out we know not to trust this cache when we load up next
3295 BTRFS_I(inode)->generation = 0;
3296 ret = btrfs_update_inode(trans, root, inode);
3299 * So theoretically we could recover from this, simply set the
3300 * super cache generation to 0 so we know to invalidate the
3301 * cache, but then we'd have to keep track of the block groups
3302 * that fail this way so we know we _have_ to reset this cache
3303 * before the next commit or risk reading stale cache. So to
3304 * limit our exposure to horrible edge cases lets just abort the
3305 * transaction, this only happens in really bad situations
3308 btrfs_abort_transaction(trans, root, ret);
3313 if (i_size_read(inode) > 0) {
3314 ret = btrfs_check_trunc_cache_free_space(root,
3315 &root->fs_info->global_block_rsv);
3319 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3324 spin_lock(&block_group->lock);
3325 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3326 !btrfs_test_opt(root, SPACE_CACHE)) {
3328 * don't bother trying to write stuff out _if_
3329 * a) we're not cached,
3330 * b) we're with nospace_cache mount option.
3332 dcs = BTRFS_DC_WRITTEN;
3333 spin_unlock(&block_group->lock);
3336 spin_unlock(&block_group->lock);
3339 * Try to preallocate enough space based on how big the block group is.
3340 * Keep in mind this has to include any pinned space which could end up
3341 * taking up quite a bit since it's not folded into the other space
3344 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3349 num_pages *= PAGE_CACHE_SIZE;
3351 ret = btrfs_check_data_free_space(inode, num_pages, num_pages);
3355 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3356 num_pages, num_pages,
3359 dcs = BTRFS_DC_SETUP;
3360 btrfs_free_reserved_data_space(inode, num_pages);
3365 btrfs_release_path(path);
3367 spin_lock(&block_group->lock);
3368 if (!ret && dcs == BTRFS_DC_SETUP)
3369 block_group->cache_generation = trans->transid;
3370 block_group->disk_cache_state = dcs;
3371 spin_unlock(&block_group->lock);
3376 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3377 struct btrfs_root *root)
3379 struct btrfs_block_group_cache *cache, *tmp;
3380 struct btrfs_transaction *cur_trans = trans->transaction;
3381 struct btrfs_path *path;
3383 if (list_empty(&cur_trans->dirty_bgs) ||
3384 !btrfs_test_opt(root, SPACE_CACHE))
3387 path = btrfs_alloc_path();
3391 /* Could add new block groups, use _safe just in case */
3392 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3394 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3395 cache_save_setup(cache, trans, path);
3398 btrfs_free_path(path);
3403 * transaction commit does final block group cache writeback during a
3404 * critical section where nothing is allowed to change the FS. This is
3405 * required in order for the cache to actually match the block group,
3406 * but can introduce a lot of latency into the commit.
3408 * So, btrfs_start_dirty_block_groups is here to kick off block group
3409 * cache IO. There's a chance we'll have to redo some of it if the
3410 * block group changes again during the commit, but it greatly reduces
3411 * the commit latency by getting rid of the easy block groups while
3412 * we're still allowing others to join the commit.
3414 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3415 struct btrfs_root *root)
3417 struct btrfs_block_group_cache *cache;
3418 struct btrfs_transaction *cur_trans = trans->transaction;
3421 struct btrfs_path *path = NULL;
3423 struct list_head *io = &cur_trans->io_bgs;
3424 int num_started = 0;
3427 spin_lock(&cur_trans->dirty_bgs_lock);
3428 if (list_empty(&cur_trans->dirty_bgs)) {
3429 spin_unlock(&cur_trans->dirty_bgs_lock);
3432 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3433 spin_unlock(&cur_trans->dirty_bgs_lock);
3437 * make sure all the block groups on our dirty list actually
3440 btrfs_create_pending_block_groups(trans, root);
3443 path = btrfs_alloc_path();
3449 * cache_write_mutex is here only to save us from balance or automatic
3450 * removal of empty block groups deleting this block group while we are
3451 * writing out the cache
3453 mutex_lock(&trans->transaction->cache_write_mutex);
3454 while (!list_empty(&dirty)) {
3455 cache = list_first_entry(&dirty,
3456 struct btrfs_block_group_cache,
3459 * this can happen if something re-dirties a block
3460 * group that is already under IO. Just wait for it to
3461 * finish and then do it all again
3463 if (!list_empty(&cache->io_list)) {
3464 list_del_init(&cache->io_list);
3465 btrfs_wait_cache_io(root, trans, cache,
3466 &cache->io_ctl, path,
3467 cache->key.objectid);
3468 btrfs_put_block_group(cache);
3473 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3474 * if it should update the cache_state. Don't delete
3475 * until after we wait.
3477 * Since we're not running in the commit critical section
3478 * we need the dirty_bgs_lock to protect from update_block_group
3480 spin_lock(&cur_trans->dirty_bgs_lock);
3481 list_del_init(&cache->dirty_list);
3482 spin_unlock(&cur_trans->dirty_bgs_lock);
3486 cache_save_setup(cache, trans, path);
3488 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3489 cache->io_ctl.inode = NULL;
3490 ret = btrfs_write_out_cache(root, trans, cache, path);
3491 if (ret == 0 && cache->io_ctl.inode) {
3496 * the cache_write_mutex is protecting
3499 list_add_tail(&cache->io_list, io);
3502 * if we failed to write the cache, the
3503 * generation will be bad and life goes on
3509 ret = write_one_cache_group(trans, root, path, cache);
3511 * Our block group might still be attached to the list
3512 * of new block groups in the transaction handle of some
3513 * other task (struct btrfs_trans_handle->new_bgs). This
3514 * means its block group item isn't yet in the extent
3515 * tree. If this happens ignore the error, as we will
3516 * try again later in the critical section of the
3517 * transaction commit.
3519 if (ret == -ENOENT) {
3521 spin_lock(&cur_trans->dirty_bgs_lock);
3522 if (list_empty(&cache->dirty_list)) {
3523 list_add_tail(&cache->dirty_list,
3524 &cur_trans->dirty_bgs);
3525 btrfs_get_block_group(cache);
3527 spin_unlock(&cur_trans->dirty_bgs_lock);
3529 btrfs_abort_transaction(trans, root, ret);
3533 /* if its not on the io list, we need to put the block group */
3535 btrfs_put_block_group(cache);
3541 * Avoid blocking other tasks for too long. It might even save
3542 * us from writing caches for block groups that are going to be
3545 mutex_unlock(&trans->transaction->cache_write_mutex);
3546 mutex_lock(&trans->transaction->cache_write_mutex);
3548 mutex_unlock(&trans->transaction->cache_write_mutex);
3551 * go through delayed refs for all the stuff we've just kicked off
3552 * and then loop back (just once)
3554 ret = btrfs_run_delayed_refs(trans, root, 0);
3555 if (!ret && loops == 0) {
3557 spin_lock(&cur_trans->dirty_bgs_lock);
3558 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3560 * dirty_bgs_lock protects us from concurrent block group
3561 * deletes too (not just cache_write_mutex).
3563 if (!list_empty(&dirty)) {
3564 spin_unlock(&cur_trans->dirty_bgs_lock);
3567 spin_unlock(&cur_trans->dirty_bgs_lock);
3570 btrfs_free_path(path);
3574 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3575 struct btrfs_root *root)
3577 struct btrfs_block_group_cache *cache;
3578 struct btrfs_transaction *cur_trans = trans->transaction;
3581 struct btrfs_path *path;
3582 struct list_head *io = &cur_trans->io_bgs;
3583 int num_started = 0;
3585 path = btrfs_alloc_path();
3590 * We don't need the lock here since we are protected by the transaction
3591 * commit. We want to do the cache_save_setup first and then run the
3592 * delayed refs to make sure we have the best chance at doing this all
3595 while (!list_empty(&cur_trans->dirty_bgs)) {
3596 cache = list_first_entry(&cur_trans->dirty_bgs,
3597 struct btrfs_block_group_cache,
3601 * this can happen if cache_save_setup re-dirties a block
3602 * group that is already under IO. Just wait for it to
3603 * finish and then do it all again
3605 if (!list_empty(&cache->io_list)) {
3606 list_del_init(&cache->io_list);
3607 btrfs_wait_cache_io(root, trans, cache,
3608 &cache->io_ctl, path,
3609 cache->key.objectid);
3610 btrfs_put_block_group(cache);
3614 * don't remove from the dirty list until after we've waited
3617 list_del_init(&cache->dirty_list);
3620 cache_save_setup(cache, trans, path);
3623 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3625 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3626 cache->io_ctl.inode = NULL;
3627 ret = btrfs_write_out_cache(root, trans, cache, path);
3628 if (ret == 0 && cache->io_ctl.inode) {
3631 list_add_tail(&cache->io_list, io);
3634 * if we failed to write the cache, the
3635 * generation will be bad and life goes on
3641 ret = write_one_cache_group(trans, root, path, cache);
3643 btrfs_abort_transaction(trans, root, ret);
3646 /* if its not on the io list, we need to put the block group */
3648 btrfs_put_block_group(cache);
3651 while (!list_empty(io)) {
3652 cache = list_first_entry(io, struct btrfs_block_group_cache,
3654 list_del_init(&cache->io_list);
3655 btrfs_wait_cache_io(root, trans, cache,
3656 &cache->io_ctl, path, cache->key.objectid);
3657 btrfs_put_block_group(cache);
3660 btrfs_free_path(path);
3664 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3666 struct btrfs_block_group_cache *block_group;
3669 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3670 if (!block_group || block_group->ro)
3673 btrfs_put_block_group(block_group);
3677 static const char *alloc_name(u64 flags)
3680 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3682 case BTRFS_BLOCK_GROUP_METADATA:
3684 case BTRFS_BLOCK_GROUP_DATA:
3686 case BTRFS_BLOCK_GROUP_SYSTEM:
3690 return "invalid-combination";
3694 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3695 u64 total_bytes, u64 bytes_used,
3696 struct btrfs_space_info **space_info)
3698 struct btrfs_space_info *found;
3703 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3704 BTRFS_BLOCK_GROUP_RAID10))
3709 found = __find_space_info(info, flags);
3711 spin_lock(&found->lock);
3712 found->total_bytes += total_bytes;
3713 found->disk_total += total_bytes * factor;
3714 found->bytes_used += bytes_used;
3715 found->disk_used += bytes_used * factor;
3716 if (total_bytes > 0)
3718 spin_unlock(&found->lock);
3719 *space_info = found;
3722 found = kzalloc(sizeof(*found), GFP_NOFS);
3726 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3732 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3733 INIT_LIST_HEAD(&found->block_groups[i]);
3734 init_rwsem(&found->groups_sem);
3735 spin_lock_init(&found->lock);
3736 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3737 found->total_bytes = total_bytes;
3738 found->disk_total = total_bytes * factor;
3739 found->bytes_used = bytes_used;
3740 found->disk_used = bytes_used * factor;
3741 found->bytes_pinned = 0;
3742 found->bytes_reserved = 0;
3743 found->bytes_readonly = 0;
3744 found->bytes_may_use = 0;
3745 if (total_bytes > 0)
3749 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3750 found->chunk_alloc = 0;
3752 init_waitqueue_head(&found->wait);
3753 INIT_LIST_HEAD(&found->ro_bgs);
3755 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3756 info->space_info_kobj, "%s",
3757 alloc_name(found->flags));
3763 *space_info = found;
3764 list_add_rcu(&found->list, &info->space_info);
3765 if (flags & BTRFS_BLOCK_GROUP_DATA)
3766 info->data_sinfo = found;
3771 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3773 u64 extra_flags = chunk_to_extended(flags) &
3774 BTRFS_EXTENDED_PROFILE_MASK;
3776 write_seqlock(&fs_info->profiles_lock);
3777 if (flags & BTRFS_BLOCK_GROUP_DATA)
3778 fs_info->avail_data_alloc_bits |= extra_flags;
3779 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3780 fs_info->avail_metadata_alloc_bits |= extra_flags;
3781 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3782 fs_info->avail_system_alloc_bits |= extra_flags;
3783 write_sequnlock(&fs_info->profiles_lock);
3787 * returns target flags in extended format or 0 if restripe for this
3788 * chunk_type is not in progress
3790 * should be called with either volume_mutex or balance_lock held
3792 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3794 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3800 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3801 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3802 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3803 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3804 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3805 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3806 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3807 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3808 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3815 * @flags: available profiles in extended format (see ctree.h)
3817 * Returns reduced profile in chunk format. If profile changing is in
3818 * progress (either running or paused) picks the target profile (if it's
3819 * already available), otherwise falls back to plain reducing.
3821 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3823 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3828 * see if restripe for this chunk_type is in progress, if so
3829 * try to reduce to the target profile
3831 spin_lock(&root->fs_info->balance_lock);
3832 target = get_restripe_target(root->fs_info, flags);
3834 /* pick target profile only if it's already available */
3835 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3836 spin_unlock(&root->fs_info->balance_lock);
3837 return extended_to_chunk(target);
3840 spin_unlock(&root->fs_info->balance_lock);
3842 /* First, mask out the RAID levels which aren't possible */
3843 if (num_devices == 1)
3844 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3845 BTRFS_BLOCK_GROUP_RAID5);
3846 if (num_devices < 3)
3847 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3848 if (num_devices < 4)
3849 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3851 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3852 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3853 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3856 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3857 tmp = BTRFS_BLOCK_GROUP_RAID6;
3858 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3859 tmp = BTRFS_BLOCK_GROUP_RAID5;
3860 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3861 tmp = BTRFS_BLOCK_GROUP_RAID10;
3862 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3863 tmp = BTRFS_BLOCK_GROUP_RAID1;
3864 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3865 tmp = BTRFS_BLOCK_GROUP_RAID0;
3867 return extended_to_chunk(flags | tmp);
3870 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3877 seq = read_seqbegin(&root->fs_info->profiles_lock);
3879 if (flags & BTRFS_BLOCK_GROUP_DATA)
3880 flags |= root->fs_info->avail_data_alloc_bits;
3881 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3882 flags |= root->fs_info->avail_system_alloc_bits;
3883 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3884 flags |= root->fs_info->avail_metadata_alloc_bits;
3885 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3887 return btrfs_reduce_alloc_profile(root, flags);
3890 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3896 flags = BTRFS_BLOCK_GROUP_DATA;
3897 else if (root == root->fs_info->chunk_root)
3898 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3900 flags = BTRFS_BLOCK_GROUP_METADATA;
3902 ret = get_alloc_profile(root, flags);
3907 * This will check the space that the inode allocates from to make sure we have
3908 * enough space for bytes.
3910 int btrfs_check_data_free_space(struct inode *inode, u64 bytes, u64 write_bytes)
3912 struct btrfs_space_info *data_sinfo;
3913 struct btrfs_root *root = BTRFS_I(inode)->root;
3914 struct btrfs_fs_info *fs_info = root->fs_info;
3917 int need_commit = 2;
3918 int have_pinned_space;
3920 /* make sure bytes are sectorsize aligned */
3921 bytes = ALIGN(bytes, root->sectorsize);
3923 if (btrfs_is_free_space_inode(inode)) {
3925 ASSERT(current->journal_info);
3928 data_sinfo = fs_info->data_sinfo;
3933 /* make sure we have enough space to handle the data first */
3934 spin_lock(&data_sinfo->lock);
3935 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3936 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3937 data_sinfo->bytes_may_use;
3939 if (used + bytes > data_sinfo->total_bytes) {
3940 struct btrfs_trans_handle *trans;
3943 * if we don't have enough free bytes in this space then we need
3944 * to alloc a new chunk.
3946 if (!data_sinfo->full) {
3949 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3950 spin_unlock(&data_sinfo->lock);
3952 alloc_target = btrfs_get_alloc_profile(root, 1);
3954 * It is ugly that we don't call nolock join
3955 * transaction for the free space inode case here.
3956 * But it is safe because we only do the data space
3957 * reservation for the free space cache in the
3958 * transaction context, the common join transaction
3959 * just increase the counter of the current transaction
3960 * handler, doesn't try to acquire the trans_lock of
3963 trans = btrfs_join_transaction(root);
3965 return PTR_ERR(trans);
3967 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3969 CHUNK_ALLOC_NO_FORCE);
3970 btrfs_end_transaction(trans, root);
3975 have_pinned_space = 1;
3981 data_sinfo = fs_info->data_sinfo;
3987 * If we don't have enough pinned space to deal with this
3988 * allocation, and no removed chunk in current transaction,
3989 * don't bother committing the transaction.
3991 have_pinned_space = percpu_counter_compare(
3992 &data_sinfo->total_bytes_pinned,
3993 used + bytes - data_sinfo->total_bytes);
3994 spin_unlock(&data_sinfo->lock);
3996 /* commit the current transaction and try again */
3999 !atomic_read(&root->fs_info->open_ioctl_trans)) {
4002 if (need_commit > 0)
4003 btrfs_wait_ordered_roots(fs_info, -1);
4005 trans = btrfs_join_transaction(root);
4007 return PTR_ERR(trans);
4008 if (have_pinned_space >= 0 ||
4009 trans->transaction->have_free_bgs ||
4011 ret = btrfs_commit_transaction(trans, root);
4015 * make sure that all running delayed iput are
4018 down_write(&root->fs_info->delayed_iput_sem);
4019 up_write(&root->fs_info->delayed_iput_sem);
4022 btrfs_end_transaction(trans, root);
4026 trace_btrfs_space_reservation(root->fs_info,
4027 "space_info:enospc",
4028 data_sinfo->flags, bytes, 1);
4031 ret = btrfs_qgroup_reserve(root, write_bytes);
4034 data_sinfo->bytes_may_use += bytes;
4035 trace_btrfs_space_reservation(root->fs_info, "space_info",
4036 data_sinfo->flags, bytes, 1);
4038 spin_unlock(&data_sinfo->lock);
4044 * Called if we need to clear a data reservation for this inode.
4046 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
4048 struct btrfs_root *root = BTRFS_I(inode)->root;
4049 struct btrfs_space_info *data_sinfo;
4051 /* make sure bytes are sectorsize aligned */
4052 bytes = ALIGN(bytes, root->sectorsize);
4054 data_sinfo = root->fs_info->data_sinfo;
4055 spin_lock(&data_sinfo->lock);
4056 WARN_ON(data_sinfo->bytes_may_use < bytes);
4057 data_sinfo->bytes_may_use -= bytes;
4058 trace_btrfs_space_reservation(root->fs_info, "space_info",
4059 data_sinfo->flags, bytes, 0);
4060 spin_unlock(&data_sinfo->lock);
4063 static void force_metadata_allocation(struct btrfs_fs_info *info)
4065 struct list_head *head = &info->space_info;
4066 struct btrfs_space_info *found;
4069 list_for_each_entry_rcu(found, head, list) {
4070 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4071 found->force_alloc = CHUNK_ALLOC_FORCE;
4076 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4078 return (global->size << 1);
4081 static int should_alloc_chunk(struct btrfs_root *root,
4082 struct btrfs_space_info *sinfo, int force)
4084 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4085 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4086 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4089 if (force == CHUNK_ALLOC_FORCE)
4093 * We need to take into account the global rsv because for all intents
4094 * and purposes it's used space. Don't worry about locking the
4095 * global_rsv, it doesn't change except when the transaction commits.
4097 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4098 num_allocated += calc_global_rsv_need_space(global_rsv);
4101 * in limited mode, we want to have some free space up to
4102 * about 1% of the FS size.
4104 if (force == CHUNK_ALLOC_LIMITED) {
4105 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4106 thresh = max_t(u64, 64 * 1024 * 1024,
4107 div_factor_fine(thresh, 1));
4109 if (num_bytes - num_allocated < thresh)
4113 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4118 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4122 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4123 BTRFS_BLOCK_GROUP_RAID0 |
4124 BTRFS_BLOCK_GROUP_RAID5 |
4125 BTRFS_BLOCK_GROUP_RAID6))
4126 num_dev = root->fs_info->fs_devices->rw_devices;
4127 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4130 num_dev = 1; /* DUP or single */
4136 * If @is_allocation is true, reserve space in the system space info necessary
4137 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4140 void check_system_chunk(struct btrfs_trans_handle *trans,
4141 struct btrfs_root *root,
4144 struct btrfs_space_info *info;
4151 * Needed because we can end up allocating a system chunk and for an
4152 * atomic and race free space reservation in the chunk block reserve.
4154 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4156 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4157 spin_lock(&info->lock);
4158 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4159 info->bytes_reserved - info->bytes_readonly -
4160 info->bytes_may_use;
4161 spin_unlock(&info->lock);
4163 num_devs = get_profile_num_devs(root, type);
4165 /* num_devs device items to update and 1 chunk item to add or remove */
4166 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4167 btrfs_calc_trans_metadata_size(root, 1);
4169 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4170 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4171 left, thresh, type);
4172 dump_space_info(info, 0, 0);
4175 if (left < thresh) {
4178 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4180 * Ignore failure to create system chunk. We might end up not
4181 * needing it, as we might not need to COW all nodes/leafs from
4182 * the paths we visit in the chunk tree (they were already COWed
4183 * or created in the current transaction for example).
4185 ret = btrfs_alloc_chunk(trans, root, flags);
4189 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4190 &root->fs_info->chunk_block_rsv,
4191 thresh, BTRFS_RESERVE_NO_FLUSH);
4193 trans->chunk_bytes_reserved += thresh;
4197 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4198 struct btrfs_root *extent_root, u64 flags, int force)
4200 struct btrfs_space_info *space_info;
4201 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4202 int wait_for_alloc = 0;
4205 /* Don't re-enter if we're already allocating a chunk */
4206 if (trans->allocating_chunk)
4209 space_info = __find_space_info(extent_root->fs_info, flags);
4211 ret = update_space_info(extent_root->fs_info, flags,
4213 BUG_ON(ret); /* -ENOMEM */
4215 BUG_ON(!space_info); /* Logic error */
4218 spin_lock(&space_info->lock);
4219 if (force < space_info->force_alloc)
4220 force = space_info->force_alloc;
4221 if (space_info->full) {
4222 if (should_alloc_chunk(extent_root, space_info, force))
4226 spin_unlock(&space_info->lock);
4230 if (!should_alloc_chunk(extent_root, space_info, force)) {
4231 spin_unlock(&space_info->lock);
4233 } else if (space_info->chunk_alloc) {
4236 space_info->chunk_alloc = 1;
4239 spin_unlock(&space_info->lock);
4241 mutex_lock(&fs_info->chunk_mutex);
4244 * The chunk_mutex is held throughout the entirety of a chunk
4245 * allocation, so once we've acquired the chunk_mutex we know that the
4246 * other guy is done and we need to recheck and see if we should
4249 if (wait_for_alloc) {
4250 mutex_unlock(&fs_info->chunk_mutex);
4255 trans->allocating_chunk = true;
4258 * If we have mixed data/metadata chunks we want to make sure we keep
4259 * allocating mixed chunks instead of individual chunks.
4261 if (btrfs_mixed_space_info(space_info))
4262 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4265 * if we're doing a data chunk, go ahead and make sure that
4266 * we keep a reasonable number of metadata chunks allocated in the
4269 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4270 fs_info->data_chunk_allocations++;
4271 if (!(fs_info->data_chunk_allocations %
4272 fs_info->metadata_ratio))
4273 force_metadata_allocation(fs_info);
4277 * Check if we have enough space in SYSTEM chunk because we may need
4278 * to update devices.
4280 check_system_chunk(trans, extent_root, flags);
4282 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4283 trans->allocating_chunk = false;
4285 spin_lock(&space_info->lock);
4286 if (ret < 0 && ret != -ENOSPC)
4289 space_info->full = 1;
4293 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4295 space_info->chunk_alloc = 0;
4296 spin_unlock(&space_info->lock);
4297 mutex_unlock(&fs_info->chunk_mutex);
4299 * When we allocate a new chunk we reserve space in the chunk block
4300 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4301 * add new nodes/leafs to it if we end up needing to do it when
4302 * inserting the chunk item and updating device items as part of the
4303 * second phase of chunk allocation, performed by
4304 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4305 * large number of new block groups to create in our transaction
4306 * handle's new_bgs list to avoid exhausting the chunk block reserve
4307 * in extreme cases - like having a single transaction create many new
4308 * block groups when starting to write out the free space caches of all
4309 * the block groups that were made dirty during the lifetime of the
4312 if (trans->chunk_bytes_reserved >= (2 * 1024 * 1024ull)) {
4313 btrfs_create_pending_block_groups(trans, trans->root);
4314 btrfs_trans_release_chunk_metadata(trans);
4319 static int can_overcommit(struct btrfs_root *root,
4320 struct btrfs_space_info *space_info, u64 bytes,
4321 enum btrfs_reserve_flush_enum flush)
4323 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4324 u64 profile = btrfs_get_alloc_profile(root, 0);
4329 used = space_info->bytes_used + space_info->bytes_reserved +
4330 space_info->bytes_pinned + space_info->bytes_readonly;
4333 * We only want to allow over committing if we have lots of actual space
4334 * free, but if we don't have enough space to handle the global reserve
4335 * space then we could end up having a real enospc problem when trying
4336 * to allocate a chunk or some other such important allocation.
4338 spin_lock(&global_rsv->lock);
4339 space_size = calc_global_rsv_need_space(global_rsv);
4340 spin_unlock(&global_rsv->lock);
4341 if (used + space_size >= space_info->total_bytes)
4344 used += space_info->bytes_may_use;
4346 spin_lock(&root->fs_info->free_chunk_lock);
4347 avail = root->fs_info->free_chunk_space;
4348 spin_unlock(&root->fs_info->free_chunk_lock);
4351 * If we have dup, raid1 or raid10 then only half of the free
4352 * space is actually useable. For raid56, the space info used
4353 * doesn't include the parity drive, so we don't have to
4356 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4357 BTRFS_BLOCK_GROUP_RAID1 |
4358 BTRFS_BLOCK_GROUP_RAID10))
4362 * If we aren't flushing all things, let us overcommit up to
4363 * 1/2th of the space. If we can flush, don't let us overcommit
4364 * too much, let it overcommit up to 1/8 of the space.
4366 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4371 if (used + bytes < space_info->total_bytes + avail)
4376 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4377 unsigned long nr_pages, int nr_items)
4379 struct super_block *sb = root->fs_info->sb;
4381 if (down_read_trylock(&sb->s_umount)) {
4382 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4383 up_read(&sb->s_umount);
4386 * We needn't worry the filesystem going from r/w to r/o though
4387 * we don't acquire ->s_umount mutex, because the filesystem
4388 * should guarantee the delalloc inodes list be empty after
4389 * the filesystem is readonly(all dirty pages are written to
4392 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4393 if (!current->journal_info)
4394 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4398 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4403 bytes = btrfs_calc_trans_metadata_size(root, 1);
4404 nr = (int)div64_u64(to_reclaim, bytes);
4410 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4413 * shrink metadata reservation for delalloc
4415 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4418 struct btrfs_block_rsv *block_rsv;
4419 struct btrfs_space_info *space_info;
4420 struct btrfs_trans_handle *trans;
4424 unsigned long nr_pages;
4427 enum btrfs_reserve_flush_enum flush;
4429 /* Calc the number of the pages we need flush for space reservation */
4430 items = calc_reclaim_items_nr(root, to_reclaim);
4431 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4433 trans = (struct btrfs_trans_handle *)current->journal_info;
4434 block_rsv = &root->fs_info->delalloc_block_rsv;
4435 space_info = block_rsv->space_info;
4437 delalloc_bytes = percpu_counter_sum_positive(
4438 &root->fs_info->delalloc_bytes);
4439 if (delalloc_bytes == 0) {
4443 btrfs_wait_ordered_roots(root->fs_info, items);
4448 while (delalloc_bytes && loops < 3) {
4449 max_reclaim = min(delalloc_bytes, to_reclaim);
4450 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4451 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4453 * We need to wait for the async pages to actually start before
4456 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4460 if (max_reclaim <= nr_pages)
4463 max_reclaim -= nr_pages;
4465 wait_event(root->fs_info->async_submit_wait,
4466 atomic_read(&root->fs_info->async_delalloc_pages) <=
4470 flush = BTRFS_RESERVE_FLUSH_ALL;
4472 flush = BTRFS_RESERVE_NO_FLUSH;
4473 spin_lock(&space_info->lock);
4474 if (can_overcommit(root, space_info, orig, flush)) {
4475 spin_unlock(&space_info->lock);
4478 spin_unlock(&space_info->lock);
4481 if (wait_ordered && !trans) {
4482 btrfs_wait_ordered_roots(root->fs_info, items);
4484 time_left = schedule_timeout_killable(1);
4488 delalloc_bytes = percpu_counter_sum_positive(
4489 &root->fs_info->delalloc_bytes);
4494 * maybe_commit_transaction - possibly commit the transaction if its ok to
4495 * @root - the root we're allocating for
4496 * @bytes - the number of bytes we want to reserve
4497 * @force - force the commit
4499 * This will check to make sure that committing the transaction will actually
4500 * get us somewhere and then commit the transaction if it does. Otherwise it
4501 * will return -ENOSPC.
4503 static int may_commit_transaction(struct btrfs_root *root,
4504 struct btrfs_space_info *space_info,
4505 u64 bytes, int force)
4507 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4508 struct btrfs_trans_handle *trans;
4510 trans = (struct btrfs_trans_handle *)current->journal_info;
4517 /* See if there is enough pinned space to make this reservation */
4518 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4523 * See if there is some space in the delayed insertion reservation for
4526 if (space_info != delayed_rsv->space_info)
4529 spin_lock(&delayed_rsv->lock);
4530 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4531 bytes - delayed_rsv->size) >= 0) {
4532 spin_unlock(&delayed_rsv->lock);
4535 spin_unlock(&delayed_rsv->lock);
4538 trans = btrfs_join_transaction(root);
4542 return btrfs_commit_transaction(trans, root);
4546 FLUSH_DELAYED_ITEMS_NR = 1,
4547 FLUSH_DELAYED_ITEMS = 2,
4549 FLUSH_DELALLOC_WAIT = 4,
4554 static int flush_space(struct btrfs_root *root,
4555 struct btrfs_space_info *space_info, u64 num_bytes,
4556 u64 orig_bytes, int state)
4558 struct btrfs_trans_handle *trans;
4563 case FLUSH_DELAYED_ITEMS_NR:
4564 case FLUSH_DELAYED_ITEMS:
4565 if (state == FLUSH_DELAYED_ITEMS_NR)
4566 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4570 trans = btrfs_join_transaction(root);
4571 if (IS_ERR(trans)) {
4572 ret = PTR_ERR(trans);
4575 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4576 btrfs_end_transaction(trans, root);
4578 case FLUSH_DELALLOC:
4579 case FLUSH_DELALLOC_WAIT:
4580 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4581 state == FLUSH_DELALLOC_WAIT);
4584 trans = btrfs_join_transaction(root);
4585 if (IS_ERR(trans)) {
4586 ret = PTR_ERR(trans);
4589 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4590 btrfs_get_alloc_profile(root, 0),
4591 CHUNK_ALLOC_NO_FORCE);
4592 btrfs_end_transaction(trans, root);
4597 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4608 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4609 struct btrfs_space_info *space_info)
4615 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4617 spin_lock(&space_info->lock);
4618 if (can_overcommit(root, space_info, to_reclaim,
4619 BTRFS_RESERVE_FLUSH_ALL)) {
4624 used = space_info->bytes_used + space_info->bytes_reserved +
4625 space_info->bytes_pinned + space_info->bytes_readonly +
4626 space_info->bytes_may_use;
4627 if (can_overcommit(root, space_info, 1024 * 1024,
4628 BTRFS_RESERVE_FLUSH_ALL))
4629 expected = div_factor_fine(space_info->total_bytes, 95);
4631 expected = div_factor_fine(space_info->total_bytes, 90);
4633 if (used > expected)
4634 to_reclaim = used - expected;
4637 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4638 space_info->bytes_reserved);
4640 spin_unlock(&space_info->lock);
4645 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4646 struct btrfs_fs_info *fs_info, u64 used)
4648 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4650 /* If we're just plain full then async reclaim just slows us down. */
4651 if (space_info->bytes_used >= thresh)
4654 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4655 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4658 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4659 struct btrfs_fs_info *fs_info,
4664 spin_lock(&space_info->lock);
4666 * We run out of space and have not got any free space via flush_space,
4667 * so don't bother doing async reclaim.
4669 if (flush_state > COMMIT_TRANS && space_info->full) {
4670 spin_unlock(&space_info->lock);
4674 used = space_info->bytes_used + space_info->bytes_reserved +
4675 space_info->bytes_pinned + space_info->bytes_readonly +
4676 space_info->bytes_may_use;
4677 if (need_do_async_reclaim(space_info, fs_info, used)) {
4678 spin_unlock(&space_info->lock);
4681 spin_unlock(&space_info->lock);
4686 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4688 struct btrfs_fs_info *fs_info;
4689 struct btrfs_space_info *space_info;
4693 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4694 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4696 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4701 flush_state = FLUSH_DELAYED_ITEMS_NR;
4703 flush_space(fs_info->fs_root, space_info, to_reclaim,
4704 to_reclaim, flush_state);
4706 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4709 } while (flush_state < COMMIT_TRANS);
4712 void btrfs_init_async_reclaim_work(struct work_struct *work)
4714 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4718 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4719 * @root - the root we're allocating for
4720 * @block_rsv - the block_rsv we're allocating for
4721 * @orig_bytes - the number of bytes we want
4722 * @flush - whether or not we can flush to make our reservation
4724 * This will reserve orgi_bytes number of bytes from the space info associated
4725 * with the block_rsv. If there is not enough space it will make an attempt to
4726 * flush out space to make room. It will do this by flushing delalloc if
4727 * possible or committing the transaction. If flush is 0 then no attempts to
4728 * regain reservations will be made and this will fail if there is not enough
4731 static int reserve_metadata_bytes(struct btrfs_root *root,
4732 struct btrfs_block_rsv *block_rsv,
4734 enum btrfs_reserve_flush_enum flush)
4736 struct btrfs_space_info *space_info = block_rsv->space_info;
4738 u64 num_bytes = orig_bytes;
4739 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4741 bool flushing = false;
4745 spin_lock(&space_info->lock);
4747 * We only want to wait if somebody other than us is flushing and we
4748 * are actually allowed to flush all things.
4750 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4751 space_info->flush) {
4752 spin_unlock(&space_info->lock);
4754 * If we have a trans handle we can't wait because the flusher
4755 * may have to commit the transaction, which would mean we would
4756 * deadlock since we are waiting for the flusher to finish, but
4757 * hold the current transaction open.
4759 if (current->journal_info)
4761 ret = wait_event_killable(space_info->wait, !space_info->flush);
4762 /* Must have been killed, return */
4766 spin_lock(&space_info->lock);
4770 used = space_info->bytes_used + space_info->bytes_reserved +
4771 space_info->bytes_pinned + space_info->bytes_readonly +
4772 space_info->bytes_may_use;
4775 * The idea here is that we've not already over-reserved the block group
4776 * then we can go ahead and save our reservation first and then start
4777 * flushing if we need to. Otherwise if we've already overcommitted
4778 * lets start flushing stuff first and then come back and try to make
4781 if (used <= space_info->total_bytes) {
4782 if (used + orig_bytes <= space_info->total_bytes) {
4783 space_info->bytes_may_use += orig_bytes;
4784 trace_btrfs_space_reservation(root->fs_info,
4785 "space_info", space_info->flags, orig_bytes, 1);
4789 * Ok set num_bytes to orig_bytes since we aren't
4790 * overocmmitted, this way we only try and reclaim what
4793 num_bytes = orig_bytes;
4797 * Ok we're over committed, set num_bytes to the overcommitted
4798 * amount plus the amount of bytes that we need for this
4801 num_bytes = used - space_info->total_bytes +
4805 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4806 space_info->bytes_may_use += orig_bytes;
4807 trace_btrfs_space_reservation(root->fs_info, "space_info",
4808 space_info->flags, orig_bytes,
4814 * Couldn't make our reservation, save our place so while we're trying
4815 * to reclaim space we can actually use it instead of somebody else
4816 * stealing it from us.
4818 * We make the other tasks wait for the flush only when we can flush
4821 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4823 space_info->flush = 1;
4824 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4827 * We will do the space reservation dance during log replay,
4828 * which means we won't have fs_info->fs_root set, so don't do
4829 * the async reclaim as we will panic.
4831 if (!root->fs_info->log_root_recovering &&
4832 need_do_async_reclaim(space_info, root->fs_info, used) &&
4833 !work_busy(&root->fs_info->async_reclaim_work))
4834 queue_work(system_unbound_wq,
4835 &root->fs_info->async_reclaim_work);
4837 spin_unlock(&space_info->lock);
4839 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4842 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4847 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4848 * would happen. So skip delalloc flush.
4850 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4851 (flush_state == FLUSH_DELALLOC ||
4852 flush_state == FLUSH_DELALLOC_WAIT))
4853 flush_state = ALLOC_CHUNK;
4857 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4858 flush_state < COMMIT_TRANS)
4860 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4861 flush_state <= COMMIT_TRANS)
4865 if (ret == -ENOSPC &&
4866 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4867 struct btrfs_block_rsv *global_rsv =
4868 &root->fs_info->global_block_rsv;
4870 if (block_rsv != global_rsv &&
4871 !block_rsv_use_bytes(global_rsv, orig_bytes))
4875 trace_btrfs_space_reservation(root->fs_info,
4876 "space_info:enospc",
4877 space_info->flags, orig_bytes, 1);
4879 spin_lock(&space_info->lock);
4880 space_info->flush = 0;
4881 wake_up_all(&space_info->wait);
4882 spin_unlock(&space_info->lock);
4887 static struct btrfs_block_rsv *get_block_rsv(
4888 const struct btrfs_trans_handle *trans,
4889 const struct btrfs_root *root)
4891 struct btrfs_block_rsv *block_rsv = NULL;
4893 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4894 block_rsv = trans->block_rsv;
4896 if (root == root->fs_info->csum_root && trans->adding_csums)
4897 block_rsv = trans->block_rsv;
4899 if (root == root->fs_info->uuid_root)
4900 block_rsv = trans->block_rsv;
4903 block_rsv = root->block_rsv;
4906 block_rsv = &root->fs_info->empty_block_rsv;
4911 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4915 spin_lock(&block_rsv->lock);
4916 if (block_rsv->reserved >= num_bytes) {
4917 block_rsv->reserved -= num_bytes;
4918 if (block_rsv->reserved < block_rsv->size)
4919 block_rsv->full = 0;
4922 spin_unlock(&block_rsv->lock);
4926 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4927 u64 num_bytes, int update_size)
4929 spin_lock(&block_rsv->lock);
4930 block_rsv->reserved += num_bytes;
4932 block_rsv->size += num_bytes;
4933 else if (block_rsv->reserved >= block_rsv->size)
4934 block_rsv->full = 1;
4935 spin_unlock(&block_rsv->lock);
4938 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4939 struct btrfs_block_rsv *dest, u64 num_bytes,
4942 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4945 if (global_rsv->space_info != dest->space_info)
4948 spin_lock(&global_rsv->lock);
4949 min_bytes = div_factor(global_rsv->size, min_factor);
4950 if (global_rsv->reserved < min_bytes + num_bytes) {
4951 spin_unlock(&global_rsv->lock);
4954 global_rsv->reserved -= num_bytes;
4955 if (global_rsv->reserved < global_rsv->size)
4956 global_rsv->full = 0;
4957 spin_unlock(&global_rsv->lock);
4959 block_rsv_add_bytes(dest, num_bytes, 1);
4963 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4964 struct btrfs_block_rsv *block_rsv,
4965 struct btrfs_block_rsv *dest, u64 num_bytes)
4967 struct btrfs_space_info *space_info = block_rsv->space_info;
4969 spin_lock(&block_rsv->lock);
4970 if (num_bytes == (u64)-1)
4971 num_bytes = block_rsv->size;
4972 block_rsv->size -= num_bytes;
4973 if (block_rsv->reserved >= block_rsv->size) {
4974 num_bytes = block_rsv->reserved - block_rsv->size;
4975 block_rsv->reserved = block_rsv->size;
4976 block_rsv->full = 1;
4980 spin_unlock(&block_rsv->lock);
4982 if (num_bytes > 0) {
4984 spin_lock(&dest->lock);
4988 bytes_to_add = dest->size - dest->reserved;
4989 bytes_to_add = min(num_bytes, bytes_to_add);
4990 dest->reserved += bytes_to_add;
4991 if (dest->reserved >= dest->size)
4993 num_bytes -= bytes_to_add;
4995 spin_unlock(&dest->lock);
4998 spin_lock(&space_info->lock);
4999 space_info->bytes_may_use -= num_bytes;
5000 trace_btrfs_space_reservation(fs_info, "space_info",
5001 space_info->flags, num_bytes, 0);
5002 spin_unlock(&space_info->lock);
5007 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
5008 struct btrfs_block_rsv *dst, u64 num_bytes)
5012 ret = block_rsv_use_bytes(src, num_bytes);
5016 block_rsv_add_bytes(dst, num_bytes, 1);
5020 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5022 memset(rsv, 0, sizeof(*rsv));
5023 spin_lock_init(&rsv->lock);
5027 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5028 unsigned short type)
5030 struct btrfs_block_rsv *block_rsv;
5031 struct btrfs_fs_info *fs_info = root->fs_info;
5033 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5037 btrfs_init_block_rsv(block_rsv, type);
5038 block_rsv->space_info = __find_space_info(fs_info,
5039 BTRFS_BLOCK_GROUP_METADATA);
5043 void btrfs_free_block_rsv(struct btrfs_root *root,
5044 struct btrfs_block_rsv *rsv)
5048 btrfs_block_rsv_release(root, rsv, (u64)-1);
5052 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5057 int btrfs_block_rsv_add(struct btrfs_root *root,
5058 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5059 enum btrfs_reserve_flush_enum flush)
5066 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5068 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5075 int btrfs_block_rsv_check(struct btrfs_root *root,
5076 struct btrfs_block_rsv *block_rsv, int min_factor)
5084 spin_lock(&block_rsv->lock);
5085 num_bytes = div_factor(block_rsv->size, min_factor);
5086 if (block_rsv->reserved >= num_bytes)
5088 spin_unlock(&block_rsv->lock);
5093 int btrfs_block_rsv_refill(struct btrfs_root *root,
5094 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5095 enum btrfs_reserve_flush_enum flush)
5103 spin_lock(&block_rsv->lock);
5104 num_bytes = min_reserved;
5105 if (block_rsv->reserved >= num_bytes)
5108 num_bytes -= block_rsv->reserved;
5109 spin_unlock(&block_rsv->lock);
5114 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5116 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5123 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5124 struct btrfs_block_rsv *dst_rsv,
5127 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5130 void btrfs_block_rsv_release(struct btrfs_root *root,
5131 struct btrfs_block_rsv *block_rsv,
5134 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5135 if (global_rsv == block_rsv ||
5136 block_rsv->space_info != global_rsv->space_info)
5138 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5143 * helper to calculate size of global block reservation.
5144 * the desired value is sum of space used by extent tree,
5145 * checksum tree and root tree
5147 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5149 struct btrfs_space_info *sinfo;
5153 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5155 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5156 spin_lock(&sinfo->lock);
5157 data_used = sinfo->bytes_used;
5158 spin_unlock(&sinfo->lock);
5160 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5161 spin_lock(&sinfo->lock);
5162 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5164 meta_used = sinfo->bytes_used;
5165 spin_unlock(&sinfo->lock);
5167 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5169 num_bytes += div_u64(data_used + meta_used, 50);
5171 if (num_bytes * 3 > meta_used)
5172 num_bytes = div_u64(meta_used, 3);
5174 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5177 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5179 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5180 struct btrfs_space_info *sinfo = block_rsv->space_info;
5183 num_bytes = calc_global_metadata_size(fs_info);
5185 spin_lock(&sinfo->lock);
5186 spin_lock(&block_rsv->lock);
5188 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5190 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5191 sinfo->bytes_reserved + sinfo->bytes_readonly +
5192 sinfo->bytes_may_use;
5194 if (sinfo->total_bytes > num_bytes) {
5195 num_bytes = sinfo->total_bytes - num_bytes;
5196 block_rsv->reserved += num_bytes;
5197 sinfo->bytes_may_use += num_bytes;
5198 trace_btrfs_space_reservation(fs_info, "space_info",
5199 sinfo->flags, num_bytes, 1);
5202 if (block_rsv->reserved >= block_rsv->size) {
5203 num_bytes = block_rsv->reserved - block_rsv->size;
5204 sinfo->bytes_may_use -= num_bytes;
5205 trace_btrfs_space_reservation(fs_info, "space_info",
5206 sinfo->flags, num_bytes, 0);
5207 block_rsv->reserved = block_rsv->size;
5208 block_rsv->full = 1;
5211 spin_unlock(&block_rsv->lock);
5212 spin_unlock(&sinfo->lock);
5215 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5217 struct btrfs_space_info *space_info;
5219 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5220 fs_info->chunk_block_rsv.space_info = space_info;
5222 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5223 fs_info->global_block_rsv.space_info = space_info;
5224 fs_info->delalloc_block_rsv.space_info = space_info;
5225 fs_info->trans_block_rsv.space_info = space_info;
5226 fs_info->empty_block_rsv.space_info = space_info;
5227 fs_info->delayed_block_rsv.space_info = space_info;
5229 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5230 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5231 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5232 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5233 if (fs_info->quota_root)
5234 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5235 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5237 update_global_block_rsv(fs_info);
5240 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5242 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5244 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5245 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5246 WARN_ON(fs_info->trans_block_rsv.size > 0);
5247 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5248 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5249 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5250 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5251 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5254 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5255 struct btrfs_root *root)
5257 if (!trans->block_rsv)
5260 if (!trans->bytes_reserved)
5263 trace_btrfs_space_reservation(root->fs_info, "transaction",
5264 trans->transid, trans->bytes_reserved, 0);
5265 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5266 trans->bytes_reserved = 0;
5270 * To be called after all the new block groups attached to the transaction
5271 * handle have been created (btrfs_create_pending_block_groups()).
5273 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5275 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5277 if (!trans->chunk_bytes_reserved)
5280 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5282 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5283 trans->chunk_bytes_reserved);
5284 trans->chunk_bytes_reserved = 0;
5287 /* Can only return 0 or -ENOSPC */
5288 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5289 struct inode *inode)
5291 struct btrfs_root *root = BTRFS_I(inode)->root;
5292 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5293 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5296 * We need to hold space in order to delete our orphan item once we've
5297 * added it, so this takes the reservation so we can release it later
5298 * when we are truly done with the orphan item.
5300 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5301 trace_btrfs_space_reservation(root->fs_info, "orphan",
5302 btrfs_ino(inode), num_bytes, 1);
5303 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5306 void btrfs_orphan_release_metadata(struct inode *inode)
5308 struct btrfs_root *root = BTRFS_I(inode)->root;
5309 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5310 trace_btrfs_space_reservation(root->fs_info, "orphan",
5311 btrfs_ino(inode), num_bytes, 0);
5312 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5316 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5317 * root: the root of the parent directory
5318 * rsv: block reservation
5319 * items: the number of items that we need do reservation
5320 * qgroup_reserved: used to return the reserved size in qgroup
5322 * This function is used to reserve the space for snapshot/subvolume
5323 * creation and deletion. Those operations are different with the
5324 * common file/directory operations, they change two fs/file trees
5325 * and root tree, the number of items that the qgroup reserves is
5326 * different with the free space reservation. So we can not use
5327 * the space reseravtion mechanism in start_transaction().
5329 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5330 struct btrfs_block_rsv *rsv,
5332 u64 *qgroup_reserved,
5333 bool use_global_rsv)
5337 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5339 if (root->fs_info->quota_enabled) {
5340 /* One for parent inode, two for dir entries */
5341 num_bytes = 3 * root->nodesize;
5342 ret = btrfs_qgroup_reserve(root, num_bytes);
5349 *qgroup_reserved = num_bytes;
5351 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5352 rsv->space_info = __find_space_info(root->fs_info,
5353 BTRFS_BLOCK_GROUP_METADATA);
5354 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5355 BTRFS_RESERVE_FLUSH_ALL);
5357 if (ret == -ENOSPC && use_global_rsv)
5358 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5361 if (*qgroup_reserved)
5362 btrfs_qgroup_free(root, *qgroup_reserved);
5368 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5369 struct btrfs_block_rsv *rsv,
5370 u64 qgroup_reserved)
5372 btrfs_block_rsv_release(root, rsv, (u64)-1);
5376 * drop_outstanding_extent - drop an outstanding extent
5377 * @inode: the inode we're dropping the extent for
5378 * @num_bytes: the number of bytes we're relaseing.
5380 * This is called when we are freeing up an outstanding extent, either called
5381 * after an error or after an extent is written. This will return the number of
5382 * reserved extents that need to be freed. This must be called with
5383 * BTRFS_I(inode)->lock held.
5385 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5387 unsigned drop_inode_space = 0;
5388 unsigned dropped_extents = 0;
5389 unsigned num_extents = 0;
5391 num_extents = (unsigned)div64_u64(num_bytes +
5392 BTRFS_MAX_EXTENT_SIZE - 1,
5393 BTRFS_MAX_EXTENT_SIZE);
5394 ASSERT(num_extents);
5395 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5396 BTRFS_I(inode)->outstanding_extents -= num_extents;
5398 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5399 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5400 &BTRFS_I(inode)->runtime_flags))
5401 drop_inode_space = 1;
5404 * If we have more or the same amount of outsanding extents than we have
5405 * reserved then we need to leave the reserved extents count alone.
5407 if (BTRFS_I(inode)->outstanding_extents >=
5408 BTRFS_I(inode)->reserved_extents)
5409 return drop_inode_space;
5411 dropped_extents = BTRFS_I(inode)->reserved_extents -
5412 BTRFS_I(inode)->outstanding_extents;
5413 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5414 return dropped_extents + drop_inode_space;
5418 * calc_csum_metadata_size - return the amount of metada space that must be
5419 * reserved/free'd for the given bytes.
5420 * @inode: the inode we're manipulating
5421 * @num_bytes: the number of bytes in question
5422 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5424 * This adjusts the number of csum_bytes in the inode and then returns the
5425 * correct amount of metadata that must either be reserved or freed. We
5426 * calculate how many checksums we can fit into one leaf and then divide the
5427 * number of bytes that will need to be checksumed by this value to figure out
5428 * how many checksums will be required. If we are adding bytes then the number
5429 * may go up and we will return the number of additional bytes that must be
5430 * reserved. If it is going down we will return the number of bytes that must
5433 * This must be called with BTRFS_I(inode)->lock held.
5435 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5438 struct btrfs_root *root = BTRFS_I(inode)->root;
5439 u64 old_csums, num_csums;
5441 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5442 BTRFS_I(inode)->csum_bytes == 0)
5445 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5447 BTRFS_I(inode)->csum_bytes += num_bytes;
5449 BTRFS_I(inode)->csum_bytes -= num_bytes;
5450 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5452 /* No change, no need to reserve more */
5453 if (old_csums == num_csums)
5457 return btrfs_calc_trans_metadata_size(root,
5458 num_csums - old_csums);
5460 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5463 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5465 struct btrfs_root *root = BTRFS_I(inode)->root;
5466 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5469 unsigned nr_extents = 0;
5470 int extra_reserve = 0;
5471 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5473 bool delalloc_lock = true;
5477 /* If we are a free space inode we need to not flush since we will be in
5478 * the middle of a transaction commit. We also don't need the delalloc
5479 * mutex since we won't race with anybody. We need this mostly to make
5480 * lockdep shut its filthy mouth.
5482 if (btrfs_is_free_space_inode(inode)) {
5483 flush = BTRFS_RESERVE_NO_FLUSH;
5484 delalloc_lock = false;
5487 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5488 btrfs_transaction_in_commit(root->fs_info))
5489 schedule_timeout(1);
5492 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5494 num_bytes = ALIGN(num_bytes, root->sectorsize);
5496 spin_lock(&BTRFS_I(inode)->lock);
5497 nr_extents = (unsigned)div64_u64(num_bytes +
5498 BTRFS_MAX_EXTENT_SIZE - 1,
5499 BTRFS_MAX_EXTENT_SIZE);
5500 BTRFS_I(inode)->outstanding_extents += nr_extents;
5503 if (BTRFS_I(inode)->outstanding_extents >
5504 BTRFS_I(inode)->reserved_extents)
5505 nr_extents = BTRFS_I(inode)->outstanding_extents -
5506 BTRFS_I(inode)->reserved_extents;
5509 * Add an item to reserve for updating the inode when we complete the
5512 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5513 &BTRFS_I(inode)->runtime_flags)) {
5518 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5519 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5520 csum_bytes = BTRFS_I(inode)->csum_bytes;
5521 spin_unlock(&BTRFS_I(inode)->lock);
5523 if (root->fs_info->quota_enabled) {
5524 ret = btrfs_qgroup_reserve(root, nr_extents * root->nodesize);
5529 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5530 if (unlikely(ret)) {
5531 if (root->fs_info->quota_enabled)
5532 btrfs_qgroup_free(root, nr_extents * root->nodesize);
5536 spin_lock(&BTRFS_I(inode)->lock);
5537 if (extra_reserve) {
5538 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5539 &BTRFS_I(inode)->runtime_flags);
5542 BTRFS_I(inode)->reserved_extents += nr_extents;
5543 spin_unlock(&BTRFS_I(inode)->lock);
5546 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5549 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5550 btrfs_ino(inode), to_reserve, 1);
5551 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5556 spin_lock(&BTRFS_I(inode)->lock);
5557 dropped = drop_outstanding_extent(inode, num_bytes);
5559 * If the inodes csum_bytes is the same as the original
5560 * csum_bytes then we know we haven't raced with any free()ers
5561 * so we can just reduce our inodes csum bytes and carry on.
5563 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5564 calc_csum_metadata_size(inode, num_bytes, 0);
5566 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5570 * This is tricky, but first we need to figure out how much we
5571 * free'd from any free-ers that occured during this
5572 * reservation, so we reset ->csum_bytes to the csum_bytes
5573 * before we dropped our lock, and then call the free for the
5574 * number of bytes that were freed while we were trying our
5577 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5578 BTRFS_I(inode)->csum_bytes = csum_bytes;
5579 to_free = calc_csum_metadata_size(inode, bytes, 0);
5583 * Now we need to see how much we would have freed had we not
5584 * been making this reservation and our ->csum_bytes were not
5585 * artificially inflated.
5587 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5588 bytes = csum_bytes - orig_csum_bytes;
5589 bytes = calc_csum_metadata_size(inode, bytes, 0);
5592 * Now reset ->csum_bytes to what it should be. If bytes is
5593 * more than to_free then we would have free'd more space had we
5594 * not had an artificially high ->csum_bytes, so we need to free
5595 * the remainder. If bytes is the same or less then we don't
5596 * need to do anything, the other free-ers did the correct
5599 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5600 if (bytes > to_free)
5601 to_free = bytes - to_free;
5605 spin_unlock(&BTRFS_I(inode)->lock);
5607 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5610 btrfs_block_rsv_release(root, block_rsv, to_free);
5611 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5612 btrfs_ino(inode), to_free, 0);
5615 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5620 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5621 * @inode: the inode to release the reservation for
5622 * @num_bytes: the number of bytes we're releasing
5624 * This will release the metadata reservation for an inode. This can be called
5625 * once we complete IO for a given set of bytes to release their metadata
5628 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5630 struct btrfs_root *root = BTRFS_I(inode)->root;
5634 num_bytes = ALIGN(num_bytes, root->sectorsize);
5635 spin_lock(&BTRFS_I(inode)->lock);
5636 dropped = drop_outstanding_extent(inode, num_bytes);
5639 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5640 spin_unlock(&BTRFS_I(inode)->lock);
5642 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5644 if (btrfs_test_is_dummy_root(root))
5647 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5648 btrfs_ino(inode), to_free, 0);
5650 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5655 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5656 * @inode: inode we're writing to
5657 * @num_bytes: the number of bytes we want to allocate
5659 * This will do the following things
5661 * o reserve space in the data space info for num_bytes
5662 * o reserve space in the metadata space info based on number of outstanding
5663 * extents and how much csums will be needed
5664 * o add to the inodes ->delalloc_bytes
5665 * o add it to the fs_info's delalloc inodes list.
5667 * This will return 0 for success and -ENOSPC if there is no space left.
5669 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5673 ret = btrfs_check_data_free_space(inode, num_bytes, num_bytes);
5677 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5679 btrfs_free_reserved_data_space(inode, num_bytes);
5687 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5688 * @inode: inode we're releasing space for
5689 * @num_bytes: the number of bytes we want to free up
5691 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5692 * called in the case that we don't need the metadata AND data reservations
5693 * anymore. So if there is an error or we insert an inline extent.
5695 * This function will release the metadata space that was not used and will
5696 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5697 * list if there are no delalloc bytes left.
5699 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5701 btrfs_delalloc_release_metadata(inode, num_bytes);
5702 btrfs_free_reserved_data_space(inode, num_bytes);
5705 static int update_block_group(struct btrfs_trans_handle *trans,
5706 struct btrfs_root *root, u64 bytenr,
5707 u64 num_bytes, int alloc)
5709 struct btrfs_block_group_cache *cache = NULL;
5710 struct btrfs_fs_info *info = root->fs_info;
5711 u64 total = num_bytes;
5716 /* block accounting for super block */
5717 spin_lock(&info->delalloc_root_lock);
5718 old_val = btrfs_super_bytes_used(info->super_copy);
5720 old_val += num_bytes;
5722 old_val -= num_bytes;
5723 btrfs_set_super_bytes_used(info->super_copy, old_val);
5724 spin_unlock(&info->delalloc_root_lock);
5727 cache = btrfs_lookup_block_group(info, bytenr);
5730 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5731 BTRFS_BLOCK_GROUP_RAID1 |
5732 BTRFS_BLOCK_GROUP_RAID10))
5737 * If this block group has free space cache written out, we
5738 * need to make sure to load it if we are removing space. This
5739 * is because we need the unpinning stage to actually add the
5740 * space back to the block group, otherwise we will leak space.
5742 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5743 cache_block_group(cache, 1);
5745 byte_in_group = bytenr - cache->key.objectid;
5746 WARN_ON(byte_in_group > cache->key.offset);
5748 spin_lock(&cache->space_info->lock);
5749 spin_lock(&cache->lock);
5751 if (btrfs_test_opt(root, SPACE_CACHE) &&
5752 cache->disk_cache_state < BTRFS_DC_CLEAR)
5753 cache->disk_cache_state = BTRFS_DC_CLEAR;
5755 old_val = btrfs_block_group_used(&cache->item);
5756 num_bytes = min(total, cache->key.offset - byte_in_group);
5758 old_val += num_bytes;
5759 btrfs_set_block_group_used(&cache->item, old_val);
5760 cache->reserved -= num_bytes;
5761 cache->space_info->bytes_reserved -= num_bytes;
5762 cache->space_info->bytes_used += num_bytes;
5763 cache->space_info->disk_used += num_bytes * factor;
5764 spin_unlock(&cache->lock);
5765 spin_unlock(&cache->space_info->lock);
5767 old_val -= num_bytes;
5768 btrfs_set_block_group_used(&cache->item, old_val);
5769 cache->pinned += num_bytes;
5770 cache->space_info->bytes_pinned += num_bytes;
5771 cache->space_info->bytes_used -= num_bytes;
5772 cache->space_info->disk_used -= num_bytes * factor;
5773 spin_unlock(&cache->lock);
5774 spin_unlock(&cache->space_info->lock);
5776 set_extent_dirty(info->pinned_extents,
5777 bytenr, bytenr + num_bytes - 1,
5778 GFP_NOFS | __GFP_NOFAIL);
5780 * No longer have used bytes in this block group, queue
5784 spin_lock(&info->unused_bgs_lock);
5785 if (list_empty(&cache->bg_list)) {
5786 btrfs_get_block_group(cache);
5787 list_add_tail(&cache->bg_list,
5790 spin_unlock(&info->unused_bgs_lock);
5794 spin_lock(&trans->transaction->dirty_bgs_lock);
5795 if (list_empty(&cache->dirty_list)) {
5796 list_add_tail(&cache->dirty_list,
5797 &trans->transaction->dirty_bgs);
5798 trans->transaction->num_dirty_bgs++;
5799 btrfs_get_block_group(cache);
5801 spin_unlock(&trans->transaction->dirty_bgs_lock);
5803 btrfs_put_block_group(cache);
5805 bytenr += num_bytes;
5810 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5812 struct btrfs_block_group_cache *cache;
5815 spin_lock(&root->fs_info->block_group_cache_lock);
5816 bytenr = root->fs_info->first_logical_byte;
5817 spin_unlock(&root->fs_info->block_group_cache_lock);
5819 if (bytenr < (u64)-1)
5822 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5826 bytenr = cache->key.objectid;
5827 btrfs_put_block_group(cache);
5832 static int pin_down_extent(struct btrfs_root *root,
5833 struct btrfs_block_group_cache *cache,
5834 u64 bytenr, u64 num_bytes, int reserved)
5836 spin_lock(&cache->space_info->lock);
5837 spin_lock(&cache->lock);
5838 cache->pinned += num_bytes;
5839 cache->space_info->bytes_pinned += num_bytes;
5841 cache->reserved -= num_bytes;
5842 cache->space_info->bytes_reserved -= num_bytes;
5844 spin_unlock(&cache->lock);
5845 spin_unlock(&cache->space_info->lock);
5847 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5848 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5850 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5855 * this function must be called within transaction
5857 int btrfs_pin_extent(struct btrfs_root *root,
5858 u64 bytenr, u64 num_bytes, int reserved)
5860 struct btrfs_block_group_cache *cache;
5862 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5863 BUG_ON(!cache); /* Logic error */
5865 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5867 btrfs_put_block_group(cache);
5872 * this function must be called within transaction
5874 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5875 u64 bytenr, u64 num_bytes)
5877 struct btrfs_block_group_cache *cache;
5880 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5885 * pull in the free space cache (if any) so that our pin
5886 * removes the free space from the cache. We have load_only set
5887 * to one because the slow code to read in the free extents does check
5888 * the pinned extents.
5890 cache_block_group(cache, 1);
5892 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5894 /* remove us from the free space cache (if we're there at all) */
5895 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5896 btrfs_put_block_group(cache);
5900 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5903 struct btrfs_block_group_cache *block_group;
5904 struct btrfs_caching_control *caching_ctl;
5906 block_group = btrfs_lookup_block_group(root->fs_info, start);
5910 cache_block_group(block_group, 0);
5911 caching_ctl = get_caching_control(block_group);
5915 BUG_ON(!block_group_cache_done(block_group));
5916 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5918 mutex_lock(&caching_ctl->mutex);
5920 if (start >= caching_ctl->progress) {
5921 ret = add_excluded_extent(root, start, num_bytes);
5922 } else if (start + num_bytes <= caching_ctl->progress) {
5923 ret = btrfs_remove_free_space(block_group,
5926 num_bytes = caching_ctl->progress - start;
5927 ret = btrfs_remove_free_space(block_group,
5932 num_bytes = (start + num_bytes) -
5933 caching_ctl->progress;
5934 start = caching_ctl->progress;
5935 ret = add_excluded_extent(root, start, num_bytes);
5938 mutex_unlock(&caching_ctl->mutex);
5939 put_caching_control(caching_ctl);
5941 btrfs_put_block_group(block_group);
5945 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5946 struct extent_buffer *eb)
5948 struct btrfs_file_extent_item *item;
5949 struct btrfs_key key;
5953 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5956 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5957 btrfs_item_key_to_cpu(eb, &key, i);
5958 if (key.type != BTRFS_EXTENT_DATA_KEY)
5960 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5961 found_type = btrfs_file_extent_type(eb, item);
5962 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5964 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5966 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5967 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5968 __exclude_logged_extent(log, key.objectid, key.offset);
5975 * btrfs_update_reserved_bytes - update the block_group and space info counters
5976 * @cache: The cache we are manipulating
5977 * @num_bytes: The number of bytes in question
5978 * @reserve: One of the reservation enums
5979 * @delalloc: The blocks are allocated for the delalloc write
5981 * This is called by the allocator when it reserves space, or by somebody who is
5982 * freeing space that was never actually used on disk. For example if you
5983 * reserve some space for a new leaf in transaction A and before transaction A
5984 * commits you free that leaf, you call this with reserve set to 0 in order to
5985 * clear the reservation.
5987 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5988 * ENOSPC accounting. For data we handle the reservation through clearing the
5989 * delalloc bits in the io_tree. We have to do this since we could end up
5990 * allocating less disk space for the amount of data we have reserved in the
5991 * case of compression.
5993 * If this is a reservation and the block group has become read only we cannot
5994 * make the reservation and return -EAGAIN, otherwise this function always
5997 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5998 u64 num_bytes, int reserve, int delalloc)
6000 struct btrfs_space_info *space_info = cache->space_info;
6003 spin_lock(&space_info->lock);
6004 spin_lock(&cache->lock);
6005 if (reserve != RESERVE_FREE) {
6009 cache->reserved += num_bytes;
6010 space_info->bytes_reserved += num_bytes;
6011 if (reserve == RESERVE_ALLOC) {
6012 trace_btrfs_space_reservation(cache->fs_info,
6013 "space_info", space_info->flags,
6015 space_info->bytes_may_use -= num_bytes;
6019 cache->delalloc_bytes += num_bytes;
6023 space_info->bytes_readonly += num_bytes;
6024 cache->reserved -= num_bytes;
6025 space_info->bytes_reserved -= num_bytes;
6028 cache->delalloc_bytes -= num_bytes;
6030 spin_unlock(&cache->lock);
6031 spin_unlock(&space_info->lock);
6035 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6036 struct btrfs_root *root)
6038 struct btrfs_fs_info *fs_info = root->fs_info;
6039 struct btrfs_caching_control *next;
6040 struct btrfs_caching_control *caching_ctl;
6041 struct btrfs_block_group_cache *cache;
6043 down_write(&fs_info->commit_root_sem);
6045 list_for_each_entry_safe(caching_ctl, next,
6046 &fs_info->caching_block_groups, list) {
6047 cache = caching_ctl->block_group;
6048 if (block_group_cache_done(cache)) {
6049 cache->last_byte_to_unpin = (u64)-1;
6050 list_del_init(&caching_ctl->list);
6051 put_caching_control(caching_ctl);
6053 cache->last_byte_to_unpin = caching_ctl->progress;
6057 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6058 fs_info->pinned_extents = &fs_info->freed_extents[1];
6060 fs_info->pinned_extents = &fs_info->freed_extents[0];
6062 up_write(&fs_info->commit_root_sem);
6064 update_global_block_rsv(fs_info);
6067 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6068 const bool return_free_space)
6070 struct btrfs_fs_info *fs_info = root->fs_info;
6071 struct btrfs_block_group_cache *cache = NULL;
6072 struct btrfs_space_info *space_info;
6073 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6077 while (start <= end) {
6080 start >= cache->key.objectid + cache->key.offset) {
6082 btrfs_put_block_group(cache);
6083 cache = btrfs_lookup_block_group(fs_info, start);
6084 BUG_ON(!cache); /* Logic error */
6087 len = cache->key.objectid + cache->key.offset - start;
6088 len = min(len, end + 1 - start);
6090 if (start < cache->last_byte_to_unpin) {
6091 len = min(len, cache->last_byte_to_unpin - start);
6092 if (return_free_space)
6093 btrfs_add_free_space(cache, start, len);
6097 space_info = cache->space_info;
6099 spin_lock(&space_info->lock);
6100 spin_lock(&cache->lock);
6101 cache->pinned -= len;
6102 space_info->bytes_pinned -= len;
6103 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6105 space_info->bytes_readonly += len;
6108 spin_unlock(&cache->lock);
6109 if (!readonly && global_rsv->space_info == space_info) {
6110 spin_lock(&global_rsv->lock);
6111 if (!global_rsv->full) {
6112 len = min(len, global_rsv->size -
6113 global_rsv->reserved);
6114 global_rsv->reserved += len;
6115 space_info->bytes_may_use += len;
6116 if (global_rsv->reserved >= global_rsv->size)
6117 global_rsv->full = 1;
6119 spin_unlock(&global_rsv->lock);
6121 spin_unlock(&space_info->lock);
6125 btrfs_put_block_group(cache);
6129 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6130 struct btrfs_root *root)
6132 struct btrfs_fs_info *fs_info = root->fs_info;
6133 struct btrfs_block_group_cache *block_group, *tmp;
6134 struct list_head *deleted_bgs;
6135 struct extent_io_tree *unpin;
6140 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6141 unpin = &fs_info->freed_extents[1];
6143 unpin = &fs_info->freed_extents[0];
6145 while (!trans->aborted) {
6146 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6147 ret = find_first_extent_bit(unpin, 0, &start, &end,
6148 EXTENT_DIRTY, NULL);
6150 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6154 if (btrfs_test_opt(root, DISCARD))
6155 ret = btrfs_discard_extent(root, start,
6156 end + 1 - start, NULL);
6158 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6159 unpin_extent_range(root, start, end, true);
6160 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6165 * Transaction is finished. We don't need the lock anymore. We
6166 * do need to clean up the block groups in case of a transaction
6169 deleted_bgs = &trans->transaction->deleted_bgs;
6170 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6174 if (!trans->aborted)
6175 ret = btrfs_discard_extent(root,
6176 block_group->key.objectid,
6177 block_group->key.offset,
6180 list_del_init(&block_group->bg_list);
6181 btrfs_put_block_group_trimming(block_group);
6182 btrfs_put_block_group(block_group);
6185 const char *errstr = btrfs_decode_error(ret);
6187 "Discard failed while removing blockgroup: errno=%d %s\n",
6195 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6196 u64 owner, u64 root_objectid)
6198 struct btrfs_space_info *space_info;
6201 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6202 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6203 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6205 flags = BTRFS_BLOCK_GROUP_METADATA;
6207 flags = BTRFS_BLOCK_GROUP_DATA;
6210 space_info = __find_space_info(fs_info, flags);
6211 BUG_ON(!space_info); /* Logic bug */
6212 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6216 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6217 struct btrfs_root *root,
6218 struct btrfs_delayed_ref_node *node, u64 parent,
6219 u64 root_objectid, u64 owner_objectid,
6220 u64 owner_offset, int refs_to_drop,
6221 struct btrfs_delayed_extent_op *extent_op)
6223 struct btrfs_key key;
6224 struct btrfs_path *path;
6225 struct btrfs_fs_info *info = root->fs_info;
6226 struct btrfs_root *extent_root = info->extent_root;
6227 struct extent_buffer *leaf;
6228 struct btrfs_extent_item *ei;
6229 struct btrfs_extent_inline_ref *iref;
6232 int extent_slot = 0;
6233 int found_extent = 0;
6235 int no_quota = node->no_quota;
6238 u64 bytenr = node->bytenr;
6239 u64 num_bytes = node->num_bytes;
6241 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6244 if (!info->quota_enabled || !is_fstree(root_objectid))
6247 path = btrfs_alloc_path();
6252 path->leave_spinning = 1;
6254 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6255 BUG_ON(!is_data && refs_to_drop != 1);
6258 skinny_metadata = 0;
6260 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6261 bytenr, num_bytes, parent,
6262 root_objectid, owner_objectid,
6265 extent_slot = path->slots[0];
6266 while (extent_slot >= 0) {
6267 btrfs_item_key_to_cpu(path->nodes[0], &key,
6269 if (key.objectid != bytenr)
6271 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6272 key.offset == num_bytes) {
6276 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6277 key.offset == owner_objectid) {
6281 if (path->slots[0] - extent_slot > 5)
6285 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6286 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6287 if (found_extent && item_size < sizeof(*ei))
6290 if (!found_extent) {
6292 ret = remove_extent_backref(trans, extent_root, path,
6294 is_data, &last_ref);
6296 btrfs_abort_transaction(trans, extent_root, ret);
6299 btrfs_release_path(path);
6300 path->leave_spinning = 1;
6302 key.objectid = bytenr;
6303 key.type = BTRFS_EXTENT_ITEM_KEY;
6304 key.offset = num_bytes;
6306 if (!is_data && skinny_metadata) {
6307 key.type = BTRFS_METADATA_ITEM_KEY;
6308 key.offset = owner_objectid;
6311 ret = btrfs_search_slot(trans, extent_root,
6313 if (ret > 0 && skinny_metadata && path->slots[0]) {
6315 * Couldn't find our skinny metadata item,
6316 * see if we have ye olde extent item.
6319 btrfs_item_key_to_cpu(path->nodes[0], &key,
6321 if (key.objectid == bytenr &&
6322 key.type == BTRFS_EXTENT_ITEM_KEY &&
6323 key.offset == num_bytes)
6327 if (ret > 0 && skinny_metadata) {
6328 skinny_metadata = false;
6329 key.objectid = bytenr;
6330 key.type = BTRFS_EXTENT_ITEM_KEY;
6331 key.offset = num_bytes;
6332 btrfs_release_path(path);
6333 ret = btrfs_search_slot(trans, extent_root,
6338 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6341 btrfs_print_leaf(extent_root,
6345 btrfs_abort_transaction(trans, extent_root, ret);
6348 extent_slot = path->slots[0];
6350 } else if (WARN_ON(ret == -ENOENT)) {
6351 btrfs_print_leaf(extent_root, path->nodes[0]);
6353 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6354 bytenr, parent, root_objectid, owner_objectid,
6356 btrfs_abort_transaction(trans, extent_root, ret);
6359 btrfs_abort_transaction(trans, extent_root, ret);
6363 leaf = path->nodes[0];
6364 item_size = btrfs_item_size_nr(leaf, extent_slot);
6365 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6366 if (item_size < sizeof(*ei)) {
6367 BUG_ON(found_extent || extent_slot != path->slots[0]);
6368 ret = convert_extent_item_v0(trans, extent_root, path,
6371 btrfs_abort_transaction(trans, extent_root, ret);
6375 btrfs_release_path(path);
6376 path->leave_spinning = 1;
6378 key.objectid = bytenr;
6379 key.type = BTRFS_EXTENT_ITEM_KEY;
6380 key.offset = num_bytes;
6382 ret = btrfs_search_slot(trans, extent_root, &key, path,
6385 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6387 btrfs_print_leaf(extent_root, path->nodes[0]);
6390 btrfs_abort_transaction(trans, extent_root, ret);
6394 extent_slot = path->slots[0];
6395 leaf = path->nodes[0];
6396 item_size = btrfs_item_size_nr(leaf, extent_slot);
6399 BUG_ON(item_size < sizeof(*ei));
6400 ei = btrfs_item_ptr(leaf, extent_slot,
6401 struct btrfs_extent_item);
6402 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6403 key.type == BTRFS_EXTENT_ITEM_KEY) {
6404 struct btrfs_tree_block_info *bi;
6405 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6406 bi = (struct btrfs_tree_block_info *)(ei + 1);
6407 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6410 refs = btrfs_extent_refs(leaf, ei);
6411 if (refs < refs_to_drop) {
6412 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6413 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6415 btrfs_abort_transaction(trans, extent_root, ret);
6418 refs -= refs_to_drop;
6422 __run_delayed_extent_op(extent_op, leaf, ei);
6424 * In the case of inline back ref, reference count will
6425 * be updated by remove_extent_backref
6428 BUG_ON(!found_extent);
6430 btrfs_set_extent_refs(leaf, ei, refs);
6431 btrfs_mark_buffer_dirty(leaf);
6434 ret = remove_extent_backref(trans, extent_root, path,
6436 is_data, &last_ref);
6438 btrfs_abort_transaction(trans, extent_root, ret);
6442 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6446 BUG_ON(is_data && refs_to_drop !=
6447 extent_data_ref_count(path, iref));
6449 BUG_ON(path->slots[0] != extent_slot);
6451 BUG_ON(path->slots[0] != extent_slot + 1);
6452 path->slots[0] = extent_slot;
6458 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6461 btrfs_abort_transaction(trans, extent_root, ret);
6464 btrfs_release_path(path);
6467 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6469 btrfs_abort_transaction(trans, extent_root, ret);
6474 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6476 btrfs_abort_transaction(trans, extent_root, ret);
6480 btrfs_release_path(path);
6483 btrfs_free_path(path);
6488 * when we free an block, it is possible (and likely) that we free the last
6489 * delayed ref for that extent as well. This searches the delayed ref tree for
6490 * a given extent, and if there are no other delayed refs to be processed, it
6491 * removes it from the tree.
6493 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6494 struct btrfs_root *root, u64 bytenr)
6496 struct btrfs_delayed_ref_head *head;
6497 struct btrfs_delayed_ref_root *delayed_refs;
6500 delayed_refs = &trans->transaction->delayed_refs;
6501 spin_lock(&delayed_refs->lock);
6502 head = btrfs_find_delayed_ref_head(trans, bytenr);
6504 goto out_delayed_unlock;
6506 spin_lock(&head->lock);
6507 if (!list_empty(&head->ref_list))
6510 if (head->extent_op) {
6511 if (!head->must_insert_reserved)
6513 btrfs_free_delayed_extent_op(head->extent_op);
6514 head->extent_op = NULL;
6518 * waiting for the lock here would deadlock. If someone else has it
6519 * locked they are already in the process of dropping it anyway
6521 if (!mutex_trylock(&head->mutex))
6525 * at this point we have a head with no other entries. Go
6526 * ahead and process it.
6528 head->node.in_tree = 0;
6529 rb_erase(&head->href_node, &delayed_refs->href_root);
6531 atomic_dec(&delayed_refs->num_entries);
6534 * we don't take a ref on the node because we're removing it from the
6535 * tree, so we just steal the ref the tree was holding.
6537 delayed_refs->num_heads--;
6538 if (head->processing == 0)
6539 delayed_refs->num_heads_ready--;
6540 head->processing = 0;
6541 spin_unlock(&head->lock);
6542 spin_unlock(&delayed_refs->lock);
6544 BUG_ON(head->extent_op);
6545 if (head->must_insert_reserved)
6548 mutex_unlock(&head->mutex);
6549 btrfs_put_delayed_ref(&head->node);
6552 spin_unlock(&head->lock);
6555 spin_unlock(&delayed_refs->lock);
6559 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6560 struct btrfs_root *root,
6561 struct extent_buffer *buf,
6562 u64 parent, int last_ref)
6567 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6568 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6569 buf->start, buf->len,
6570 parent, root->root_key.objectid,
6571 btrfs_header_level(buf),
6572 BTRFS_DROP_DELAYED_REF, NULL, 0);
6573 BUG_ON(ret); /* -ENOMEM */
6579 if (btrfs_header_generation(buf) == trans->transid) {
6580 struct btrfs_block_group_cache *cache;
6582 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6583 ret = check_ref_cleanup(trans, root, buf->start);
6588 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6590 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6591 pin_down_extent(root, cache, buf->start, buf->len, 1);
6592 btrfs_put_block_group(cache);
6596 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6598 btrfs_add_free_space(cache, buf->start, buf->len);
6599 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6600 btrfs_put_block_group(cache);
6601 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6606 add_pinned_bytes(root->fs_info, buf->len,
6607 btrfs_header_level(buf),
6608 root->root_key.objectid);
6611 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6614 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6617 /* Can return -ENOMEM */
6618 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6619 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6620 u64 owner, u64 offset, int no_quota)
6623 struct btrfs_fs_info *fs_info = root->fs_info;
6625 if (btrfs_test_is_dummy_root(root))
6628 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6631 * tree log blocks never actually go into the extent allocation
6632 * tree, just update pinning info and exit early.
6634 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6635 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6636 /* unlocks the pinned mutex */
6637 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6639 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6640 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6642 parent, root_objectid, (int)owner,
6643 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6645 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6647 parent, root_objectid, owner,
6648 offset, BTRFS_DROP_DELAYED_REF,
6655 * when we wait for progress in the block group caching, its because
6656 * our allocation attempt failed at least once. So, we must sleep
6657 * and let some progress happen before we try again.
6659 * This function will sleep at least once waiting for new free space to
6660 * show up, and then it will check the block group free space numbers
6661 * for our min num_bytes. Another option is to have it go ahead
6662 * and look in the rbtree for a free extent of a given size, but this
6665 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6666 * any of the information in this block group.
6668 static noinline void
6669 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6672 struct btrfs_caching_control *caching_ctl;
6674 caching_ctl = get_caching_control(cache);
6678 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6679 (cache->free_space_ctl->free_space >= num_bytes));
6681 put_caching_control(caching_ctl);
6685 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6687 struct btrfs_caching_control *caching_ctl;
6690 caching_ctl = get_caching_control(cache);
6692 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6694 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6695 if (cache->cached == BTRFS_CACHE_ERROR)
6697 put_caching_control(caching_ctl);
6701 int __get_raid_index(u64 flags)
6703 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6704 return BTRFS_RAID_RAID10;
6705 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6706 return BTRFS_RAID_RAID1;
6707 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6708 return BTRFS_RAID_DUP;
6709 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6710 return BTRFS_RAID_RAID0;
6711 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6712 return BTRFS_RAID_RAID5;
6713 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6714 return BTRFS_RAID_RAID6;
6716 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6719 int get_block_group_index(struct btrfs_block_group_cache *cache)
6721 return __get_raid_index(cache->flags);
6724 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6725 [BTRFS_RAID_RAID10] = "raid10",
6726 [BTRFS_RAID_RAID1] = "raid1",
6727 [BTRFS_RAID_DUP] = "dup",
6728 [BTRFS_RAID_RAID0] = "raid0",
6729 [BTRFS_RAID_SINGLE] = "single",
6730 [BTRFS_RAID_RAID5] = "raid5",
6731 [BTRFS_RAID_RAID6] = "raid6",
6734 static const char *get_raid_name(enum btrfs_raid_types type)
6736 if (type >= BTRFS_NR_RAID_TYPES)
6739 return btrfs_raid_type_names[type];
6742 enum btrfs_loop_type {
6743 LOOP_CACHING_NOWAIT = 0,
6744 LOOP_CACHING_WAIT = 1,
6745 LOOP_ALLOC_CHUNK = 2,
6746 LOOP_NO_EMPTY_SIZE = 3,
6750 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6754 down_read(&cache->data_rwsem);
6758 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6761 btrfs_get_block_group(cache);
6763 down_read(&cache->data_rwsem);
6766 static struct btrfs_block_group_cache *
6767 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6768 struct btrfs_free_cluster *cluster,
6771 struct btrfs_block_group_cache *used_bg;
6772 bool locked = false;
6774 spin_lock(&cluster->refill_lock);
6776 if (used_bg == cluster->block_group)
6779 up_read(&used_bg->data_rwsem);
6780 btrfs_put_block_group(used_bg);
6783 used_bg = cluster->block_group;
6787 if (used_bg == block_group)
6790 btrfs_get_block_group(used_bg);
6795 if (down_read_trylock(&used_bg->data_rwsem))
6798 spin_unlock(&cluster->refill_lock);
6799 down_read(&used_bg->data_rwsem);
6805 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6809 up_read(&cache->data_rwsem);
6810 btrfs_put_block_group(cache);
6814 * walks the btree of allocated extents and find a hole of a given size.
6815 * The key ins is changed to record the hole:
6816 * ins->objectid == start position
6817 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6818 * ins->offset == the size of the hole.
6819 * Any available blocks before search_start are skipped.
6821 * If there is no suitable free space, we will record the max size of
6822 * the free space extent currently.
6824 static noinline int find_free_extent(struct btrfs_root *orig_root,
6825 u64 num_bytes, u64 empty_size,
6826 u64 hint_byte, struct btrfs_key *ins,
6827 u64 flags, int delalloc)
6830 struct btrfs_root *root = orig_root->fs_info->extent_root;
6831 struct btrfs_free_cluster *last_ptr = NULL;
6832 struct btrfs_block_group_cache *block_group = NULL;
6833 u64 search_start = 0;
6834 u64 max_extent_size = 0;
6835 int empty_cluster = 2 * 1024 * 1024;
6836 struct btrfs_space_info *space_info;
6838 int index = __get_raid_index(flags);
6839 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6840 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6841 bool failed_cluster_refill = false;
6842 bool failed_alloc = false;
6843 bool use_cluster = true;
6844 bool have_caching_bg = false;
6846 WARN_ON(num_bytes < root->sectorsize);
6847 ins->type = BTRFS_EXTENT_ITEM_KEY;
6851 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6853 space_info = __find_space_info(root->fs_info, flags);
6855 btrfs_err(root->fs_info, "No space info for %llu", flags);
6860 * If the space info is for both data and metadata it means we have a
6861 * small filesystem and we can't use the clustering stuff.
6863 if (btrfs_mixed_space_info(space_info))
6864 use_cluster = false;
6866 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6867 last_ptr = &root->fs_info->meta_alloc_cluster;
6868 if (!btrfs_test_opt(root, SSD))
6869 empty_cluster = 64 * 1024;
6872 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6873 btrfs_test_opt(root, SSD)) {
6874 last_ptr = &root->fs_info->data_alloc_cluster;
6878 spin_lock(&last_ptr->lock);
6879 if (last_ptr->block_group)
6880 hint_byte = last_ptr->window_start;
6881 spin_unlock(&last_ptr->lock);
6884 search_start = max(search_start, first_logical_byte(root, 0));
6885 search_start = max(search_start, hint_byte);
6890 if (search_start == hint_byte) {
6891 block_group = btrfs_lookup_block_group(root->fs_info,
6894 * we don't want to use the block group if it doesn't match our
6895 * allocation bits, or if its not cached.
6897 * However if we are re-searching with an ideal block group
6898 * picked out then we don't care that the block group is cached.
6900 if (block_group && block_group_bits(block_group, flags) &&
6901 block_group->cached != BTRFS_CACHE_NO) {
6902 down_read(&space_info->groups_sem);
6903 if (list_empty(&block_group->list) ||
6906 * someone is removing this block group,
6907 * we can't jump into the have_block_group
6908 * target because our list pointers are not
6911 btrfs_put_block_group(block_group);
6912 up_read(&space_info->groups_sem);
6914 index = get_block_group_index(block_group);
6915 btrfs_lock_block_group(block_group, delalloc);
6916 goto have_block_group;
6918 } else if (block_group) {
6919 btrfs_put_block_group(block_group);
6923 have_caching_bg = false;
6924 down_read(&space_info->groups_sem);
6925 list_for_each_entry(block_group, &space_info->block_groups[index],
6930 btrfs_grab_block_group(block_group, delalloc);
6931 search_start = block_group->key.objectid;
6934 * this can happen if we end up cycling through all the
6935 * raid types, but we want to make sure we only allocate
6936 * for the proper type.
6938 if (!block_group_bits(block_group, flags)) {
6939 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6940 BTRFS_BLOCK_GROUP_RAID1 |
6941 BTRFS_BLOCK_GROUP_RAID5 |
6942 BTRFS_BLOCK_GROUP_RAID6 |
6943 BTRFS_BLOCK_GROUP_RAID10;
6946 * if they asked for extra copies and this block group
6947 * doesn't provide them, bail. This does allow us to
6948 * fill raid0 from raid1.
6950 if ((flags & extra) && !(block_group->flags & extra))
6955 cached = block_group_cache_done(block_group);
6956 if (unlikely(!cached)) {
6957 ret = cache_block_group(block_group, 0);
6962 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6964 if (unlikely(block_group->ro))
6968 * Ok we want to try and use the cluster allocator, so
6972 struct btrfs_block_group_cache *used_block_group;
6973 unsigned long aligned_cluster;
6975 * the refill lock keeps out other
6976 * people trying to start a new cluster
6978 used_block_group = btrfs_lock_cluster(block_group,
6981 if (!used_block_group)
6982 goto refill_cluster;
6984 if (used_block_group != block_group &&
6985 (used_block_group->ro ||
6986 !block_group_bits(used_block_group, flags)))
6987 goto release_cluster;
6989 offset = btrfs_alloc_from_cluster(used_block_group,
6992 used_block_group->key.objectid,
6995 /* we have a block, we're done */
6996 spin_unlock(&last_ptr->refill_lock);
6997 trace_btrfs_reserve_extent_cluster(root,
6999 search_start, num_bytes);
7000 if (used_block_group != block_group) {
7001 btrfs_release_block_group(block_group,
7003 block_group = used_block_group;
7008 WARN_ON(last_ptr->block_group != used_block_group);
7010 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7011 * set up a new clusters, so lets just skip it
7012 * and let the allocator find whatever block
7013 * it can find. If we reach this point, we
7014 * will have tried the cluster allocator
7015 * plenty of times and not have found
7016 * anything, so we are likely way too
7017 * fragmented for the clustering stuff to find
7020 * However, if the cluster is taken from the
7021 * current block group, release the cluster
7022 * first, so that we stand a better chance of
7023 * succeeding in the unclustered
7025 if (loop >= LOOP_NO_EMPTY_SIZE &&
7026 used_block_group != block_group) {
7027 spin_unlock(&last_ptr->refill_lock);
7028 btrfs_release_block_group(used_block_group,
7030 goto unclustered_alloc;
7034 * this cluster didn't work out, free it and
7037 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7039 if (used_block_group != block_group)
7040 btrfs_release_block_group(used_block_group,
7043 if (loop >= LOOP_NO_EMPTY_SIZE) {
7044 spin_unlock(&last_ptr->refill_lock);
7045 goto unclustered_alloc;
7048 aligned_cluster = max_t(unsigned long,
7049 empty_cluster + empty_size,
7050 block_group->full_stripe_len);
7052 /* allocate a cluster in this block group */
7053 ret = btrfs_find_space_cluster(root, block_group,
7054 last_ptr, search_start,
7059 * now pull our allocation out of this
7062 offset = btrfs_alloc_from_cluster(block_group,
7068 /* we found one, proceed */
7069 spin_unlock(&last_ptr->refill_lock);
7070 trace_btrfs_reserve_extent_cluster(root,
7071 block_group, search_start,
7075 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7076 && !failed_cluster_refill) {
7077 spin_unlock(&last_ptr->refill_lock);
7079 failed_cluster_refill = true;
7080 wait_block_group_cache_progress(block_group,
7081 num_bytes + empty_cluster + empty_size);
7082 goto have_block_group;
7086 * at this point we either didn't find a cluster
7087 * or we weren't able to allocate a block from our
7088 * cluster. Free the cluster we've been trying
7089 * to use, and go to the next block group
7091 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7092 spin_unlock(&last_ptr->refill_lock);
7097 spin_lock(&block_group->free_space_ctl->tree_lock);
7099 block_group->free_space_ctl->free_space <
7100 num_bytes + empty_cluster + empty_size) {
7101 if (block_group->free_space_ctl->free_space >
7104 block_group->free_space_ctl->free_space;
7105 spin_unlock(&block_group->free_space_ctl->tree_lock);
7108 spin_unlock(&block_group->free_space_ctl->tree_lock);
7110 offset = btrfs_find_space_for_alloc(block_group, search_start,
7111 num_bytes, empty_size,
7114 * If we didn't find a chunk, and we haven't failed on this
7115 * block group before, and this block group is in the middle of
7116 * caching and we are ok with waiting, then go ahead and wait
7117 * for progress to be made, and set failed_alloc to true.
7119 * If failed_alloc is true then we've already waited on this
7120 * block group once and should move on to the next block group.
7122 if (!offset && !failed_alloc && !cached &&
7123 loop > LOOP_CACHING_NOWAIT) {
7124 wait_block_group_cache_progress(block_group,
7125 num_bytes + empty_size);
7126 failed_alloc = true;
7127 goto have_block_group;
7128 } else if (!offset) {
7130 have_caching_bg = true;
7134 search_start = ALIGN(offset, root->stripesize);
7136 /* move on to the next group */
7137 if (search_start + num_bytes >
7138 block_group->key.objectid + block_group->key.offset) {
7139 btrfs_add_free_space(block_group, offset, num_bytes);
7143 if (offset < search_start)
7144 btrfs_add_free_space(block_group, offset,
7145 search_start - offset);
7146 BUG_ON(offset > search_start);
7148 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7149 alloc_type, delalloc);
7150 if (ret == -EAGAIN) {
7151 btrfs_add_free_space(block_group, offset, num_bytes);
7155 /* we are all good, lets return */
7156 ins->objectid = search_start;
7157 ins->offset = num_bytes;
7159 trace_btrfs_reserve_extent(orig_root, block_group,
7160 search_start, num_bytes);
7161 btrfs_release_block_group(block_group, delalloc);
7164 failed_cluster_refill = false;
7165 failed_alloc = false;
7166 BUG_ON(index != get_block_group_index(block_group));
7167 btrfs_release_block_group(block_group, delalloc);
7169 up_read(&space_info->groups_sem);
7171 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7174 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7178 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7179 * caching kthreads as we move along
7180 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7181 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7182 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7185 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7188 if (loop == LOOP_ALLOC_CHUNK) {
7189 struct btrfs_trans_handle *trans;
7192 trans = current->journal_info;
7196 trans = btrfs_join_transaction(root);
7198 if (IS_ERR(trans)) {
7199 ret = PTR_ERR(trans);
7203 ret = do_chunk_alloc(trans, root, flags,
7206 * Do not bail out on ENOSPC since we
7207 * can do more things.
7209 if (ret < 0 && ret != -ENOSPC)
7210 btrfs_abort_transaction(trans,
7215 btrfs_end_transaction(trans, root);
7220 if (loop == LOOP_NO_EMPTY_SIZE) {
7226 } else if (!ins->objectid) {
7228 } else if (ins->objectid) {
7233 ins->offset = max_extent_size;
7237 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7238 int dump_block_groups)
7240 struct btrfs_block_group_cache *cache;
7243 spin_lock(&info->lock);
7244 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7246 info->total_bytes - info->bytes_used - info->bytes_pinned -
7247 info->bytes_reserved - info->bytes_readonly,
7248 (info->full) ? "" : "not ");
7249 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7250 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7251 info->total_bytes, info->bytes_used, info->bytes_pinned,
7252 info->bytes_reserved, info->bytes_may_use,
7253 info->bytes_readonly);
7254 spin_unlock(&info->lock);
7256 if (!dump_block_groups)
7259 down_read(&info->groups_sem);
7261 list_for_each_entry(cache, &info->block_groups[index], list) {
7262 spin_lock(&cache->lock);
7263 printk(KERN_INFO "BTRFS: "
7264 "block group %llu has %llu bytes, "
7265 "%llu used %llu pinned %llu reserved %s\n",
7266 cache->key.objectid, cache->key.offset,
7267 btrfs_block_group_used(&cache->item), cache->pinned,
7268 cache->reserved, cache->ro ? "[readonly]" : "");
7269 btrfs_dump_free_space(cache, bytes);
7270 spin_unlock(&cache->lock);
7272 if (++index < BTRFS_NR_RAID_TYPES)
7274 up_read(&info->groups_sem);
7277 int btrfs_reserve_extent(struct btrfs_root *root,
7278 u64 num_bytes, u64 min_alloc_size,
7279 u64 empty_size, u64 hint_byte,
7280 struct btrfs_key *ins, int is_data, int delalloc)
7282 bool final_tried = false;
7286 flags = btrfs_get_alloc_profile(root, is_data);
7288 WARN_ON(num_bytes < root->sectorsize);
7289 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7292 if (ret == -ENOSPC) {
7293 if (!final_tried && ins->offset) {
7294 num_bytes = min(num_bytes >> 1, ins->offset);
7295 num_bytes = round_down(num_bytes, root->sectorsize);
7296 num_bytes = max(num_bytes, min_alloc_size);
7297 if (num_bytes == min_alloc_size)
7300 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7301 struct btrfs_space_info *sinfo;
7303 sinfo = __find_space_info(root->fs_info, flags);
7304 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7307 dump_space_info(sinfo, num_bytes, 1);
7314 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7316 int pin, int delalloc)
7318 struct btrfs_block_group_cache *cache;
7321 cache = btrfs_lookup_block_group(root->fs_info, start);
7323 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7329 pin_down_extent(root, cache, start, len, 1);
7331 if (btrfs_test_opt(root, DISCARD))
7332 ret = btrfs_discard_extent(root, start, len, NULL);
7333 btrfs_add_free_space(cache, start, len);
7334 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7337 btrfs_put_block_group(cache);
7339 trace_btrfs_reserved_extent_free(root, start, len);
7344 int btrfs_free_reserved_extent(struct btrfs_root *root,
7345 u64 start, u64 len, int delalloc)
7347 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7350 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7353 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7356 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7357 struct btrfs_root *root,
7358 u64 parent, u64 root_objectid,
7359 u64 flags, u64 owner, u64 offset,
7360 struct btrfs_key *ins, int ref_mod)
7363 struct btrfs_fs_info *fs_info = root->fs_info;
7364 struct btrfs_extent_item *extent_item;
7365 struct btrfs_extent_inline_ref *iref;
7366 struct btrfs_path *path;
7367 struct extent_buffer *leaf;
7372 type = BTRFS_SHARED_DATA_REF_KEY;
7374 type = BTRFS_EXTENT_DATA_REF_KEY;
7376 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7378 path = btrfs_alloc_path();
7382 path->leave_spinning = 1;
7383 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7386 btrfs_free_path(path);
7390 leaf = path->nodes[0];
7391 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7392 struct btrfs_extent_item);
7393 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7394 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7395 btrfs_set_extent_flags(leaf, extent_item,
7396 flags | BTRFS_EXTENT_FLAG_DATA);
7398 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7399 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7401 struct btrfs_shared_data_ref *ref;
7402 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7403 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7404 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7406 struct btrfs_extent_data_ref *ref;
7407 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7408 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7409 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7410 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7411 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7414 btrfs_mark_buffer_dirty(path->nodes[0]);
7415 btrfs_free_path(path);
7417 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7418 if (ret) { /* -ENOENT, logic error */
7419 btrfs_err(fs_info, "update block group failed for %llu %llu",
7420 ins->objectid, ins->offset);
7423 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7427 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7428 struct btrfs_root *root,
7429 u64 parent, u64 root_objectid,
7430 u64 flags, struct btrfs_disk_key *key,
7431 int level, struct btrfs_key *ins,
7435 struct btrfs_fs_info *fs_info = root->fs_info;
7436 struct btrfs_extent_item *extent_item;
7437 struct btrfs_tree_block_info *block_info;
7438 struct btrfs_extent_inline_ref *iref;
7439 struct btrfs_path *path;
7440 struct extent_buffer *leaf;
7441 u32 size = sizeof(*extent_item) + sizeof(*iref);
7442 u64 num_bytes = ins->offset;
7443 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7446 if (!skinny_metadata)
7447 size += sizeof(*block_info);
7449 path = btrfs_alloc_path();
7451 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7456 path->leave_spinning = 1;
7457 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7460 btrfs_free_path(path);
7461 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7466 leaf = path->nodes[0];
7467 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7468 struct btrfs_extent_item);
7469 btrfs_set_extent_refs(leaf, extent_item, 1);
7470 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7471 btrfs_set_extent_flags(leaf, extent_item,
7472 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7474 if (skinny_metadata) {
7475 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7476 num_bytes = root->nodesize;
7478 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7479 btrfs_set_tree_block_key(leaf, block_info, key);
7480 btrfs_set_tree_block_level(leaf, block_info, level);
7481 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7485 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7486 btrfs_set_extent_inline_ref_type(leaf, iref,
7487 BTRFS_SHARED_BLOCK_REF_KEY);
7488 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7490 btrfs_set_extent_inline_ref_type(leaf, iref,
7491 BTRFS_TREE_BLOCK_REF_KEY);
7492 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7495 btrfs_mark_buffer_dirty(leaf);
7496 btrfs_free_path(path);
7498 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7500 if (ret) { /* -ENOENT, logic error */
7501 btrfs_err(fs_info, "update block group failed for %llu %llu",
7502 ins->objectid, ins->offset);
7506 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7510 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7511 struct btrfs_root *root,
7512 u64 root_objectid, u64 owner,
7513 u64 offset, struct btrfs_key *ins)
7517 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7519 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7521 root_objectid, owner, offset,
7522 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7527 * this is used by the tree logging recovery code. It records that
7528 * an extent has been allocated and makes sure to clear the free
7529 * space cache bits as well
7531 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7532 struct btrfs_root *root,
7533 u64 root_objectid, u64 owner, u64 offset,
7534 struct btrfs_key *ins)
7537 struct btrfs_block_group_cache *block_group;
7540 * Mixed block groups will exclude before processing the log so we only
7541 * need to do the exlude dance if this fs isn't mixed.
7543 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7544 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7549 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7553 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7554 RESERVE_ALLOC_NO_ACCOUNT, 0);
7555 BUG_ON(ret); /* logic error */
7556 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7557 0, owner, offset, ins, 1);
7558 btrfs_put_block_group(block_group);
7562 static struct extent_buffer *
7563 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7564 u64 bytenr, int level)
7566 struct extent_buffer *buf;
7568 buf = btrfs_find_create_tree_block(root, bytenr);
7570 return ERR_PTR(-ENOMEM);
7571 btrfs_set_header_generation(buf, trans->transid);
7572 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7573 btrfs_tree_lock(buf);
7574 clean_tree_block(trans, root->fs_info, buf);
7575 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7577 btrfs_set_lock_blocking(buf);
7578 btrfs_set_buffer_uptodate(buf);
7580 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7581 buf->log_index = root->log_transid % 2;
7583 * we allow two log transactions at a time, use different
7584 * EXENT bit to differentiate dirty pages.
7586 if (buf->log_index == 0)
7587 set_extent_dirty(&root->dirty_log_pages, buf->start,
7588 buf->start + buf->len - 1, GFP_NOFS);
7590 set_extent_new(&root->dirty_log_pages, buf->start,
7591 buf->start + buf->len - 1, GFP_NOFS);
7593 buf->log_index = -1;
7594 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7595 buf->start + buf->len - 1, GFP_NOFS);
7597 trans->blocks_used++;
7598 /* this returns a buffer locked for blocking */
7602 static struct btrfs_block_rsv *
7603 use_block_rsv(struct btrfs_trans_handle *trans,
7604 struct btrfs_root *root, u32 blocksize)
7606 struct btrfs_block_rsv *block_rsv;
7607 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7609 bool global_updated = false;
7611 block_rsv = get_block_rsv(trans, root);
7613 if (unlikely(block_rsv->size == 0))
7616 ret = block_rsv_use_bytes(block_rsv, blocksize);
7620 if (block_rsv->failfast)
7621 return ERR_PTR(ret);
7623 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7624 global_updated = true;
7625 update_global_block_rsv(root->fs_info);
7629 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7630 static DEFINE_RATELIMIT_STATE(_rs,
7631 DEFAULT_RATELIMIT_INTERVAL * 10,
7632 /*DEFAULT_RATELIMIT_BURST*/ 1);
7633 if (__ratelimit(&_rs))
7635 "BTRFS: block rsv returned %d\n", ret);
7638 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7639 BTRFS_RESERVE_NO_FLUSH);
7643 * If we couldn't reserve metadata bytes try and use some from
7644 * the global reserve if its space type is the same as the global
7647 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7648 block_rsv->space_info == global_rsv->space_info) {
7649 ret = block_rsv_use_bytes(global_rsv, blocksize);
7653 return ERR_PTR(ret);
7656 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7657 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7659 block_rsv_add_bytes(block_rsv, blocksize, 0);
7660 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7664 * finds a free extent and does all the dirty work required for allocation
7665 * returns the tree buffer or an ERR_PTR on error.
7667 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7668 struct btrfs_root *root,
7669 u64 parent, u64 root_objectid,
7670 struct btrfs_disk_key *key, int level,
7671 u64 hint, u64 empty_size)
7673 struct btrfs_key ins;
7674 struct btrfs_block_rsv *block_rsv;
7675 struct extent_buffer *buf;
7676 struct btrfs_delayed_extent_op *extent_op;
7679 u32 blocksize = root->nodesize;
7680 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7683 if (btrfs_test_is_dummy_root(root)) {
7684 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7687 root->alloc_bytenr += blocksize;
7691 block_rsv = use_block_rsv(trans, root, blocksize);
7692 if (IS_ERR(block_rsv))
7693 return ERR_CAST(block_rsv);
7695 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7696 empty_size, hint, &ins, 0, 0);
7700 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7703 goto out_free_reserved;
7706 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7708 parent = ins.objectid;
7709 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7713 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7714 extent_op = btrfs_alloc_delayed_extent_op();
7720 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7722 memset(&extent_op->key, 0, sizeof(extent_op->key));
7723 extent_op->flags_to_set = flags;
7724 if (skinny_metadata)
7725 extent_op->update_key = 0;
7727 extent_op->update_key = 1;
7728 extent_op->update_flags = 1;
7729 extent_op->is_data = 0;
7730 extent_op->level = level;
7732 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7733 ins.objectid, ins.offset,
7734 parent, root_objectid, level,
7735 BTRFS_ADD_DELAYED_EXTENT,
7738 goto out_free_delayed;
7743 btrfs_free_delayed_extent_op(extent_op);
7745 free_extent_buffer(buf);
7747 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
7749 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7750 return ERR_PTR(ret);
7753 struct walk_control {
7754 u64 refs[BTRFS_MAX_LEVEL];
7755 u64 flags[BTRFS_MAX_LEVEL];
7756 struct btrfs_key update_progress;
7767 #define DROP_REFERENCE 1
7768 #define UPDATE_BACKREF 2
7770 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7771 struct btrfs_root *root,
7772 struct walk_control *wc,
7773 struct btrfs_path *path)
7781 struct btrfs_key key;
7782 struct extent_buffer *eb;
7787 if (path->slots[wc->level] < wc->reada_slot) {
7788 wc->reada_count = wc->reada_count * 2 / 3;
7789 wc->reada_count = max(wc->reada_count, 2);
7791 wc->reada_count = wc->reada_count * 3 / 2;
7792 wc->reada_count = min_t(int, wc->reada_count,
7793 BTRFS_NODEPTRS_PER_BLOCK(root));
7796 eb = path->nodes[wc->level];
7797 nritems = btrfs_header_nritems(eb);
7798 blocksize = root->nodesize;
7800 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7801 if (nread >= wc->reada_count)
7805 bytenr = btrfs_node_blockptr(eb, slot);
7806 generation = btrfs_node_ptr_generation(eb, slot);
7808 if (slot == path->slots[wc->level])
7811 if (wc->stage == UPDATE_BACKREF &&
7812 generation <= root->root_key.offset)
7815 /* We don't lock the tree block, it's OK to be racy here */
7816 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7817 wc->level - 1, 1, &refs,
7819 /* We don't care about errors in readahead. */
7824 if (wc->stage == DROP_REFERENCE) {
7828 if (wc->level == 1 &&
7829 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7831 if (!wc->update_ref ||
7832 generation <= root->root_key.offset)
7834 btrfs_node_key_to_cpu(eb, &key, slot);
7835 ret = btrfs_comp_cpu_keys(&key,
7836 &wc->update_progress);
7840 if (wc->level == 1 &&
7841 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7845 readahead_tree_block(root, bytenr);
7848 wc->reada_slot = slot;
7852 * TODO: Modify related function to add related node/leaf to dirty_extent_root,
7853 * for later qgroup accounting.
7855 * Current, this function does nothing.
7857 static int account_leaf_items(struct btrfs_trans_handle *trans,
7858 struct btrfs_root *root,
7859 struct extent_buffer *eb)
7861 int nr = btrfs_header_nritems(eb);
7863 struct btrfs_key key;
7864 struct btrfs_file_extent_item *fi;
7865 u64 bytenr, num_bytes;
7867 for (i = 0; i < nr; i++) {
7868 btrfs_item_key_to_cpu(eb, &key, i);
7870 if (key.type != BTRFS_EXTENT_DATA_KEY)
7873 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7874 /* filter out non qgroup-accountable extents */
7875 extent_type = btrfs_file_extent_type(eb, fi);
7877 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7880 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7884 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7890 * Walk up the tree from the bottom, freeing leaves and any interior
7891 * nodes which have had all slots visited. If a node (leaf or
7892 * interior) is freed, the node above it will have it's slot
7893 * incremented. The root node will never be freed.
7895 * At the end of this function, we should have a path which has all
7896 * slots incremented to the next position for a search. If we need to
7897 * read a new node it will be NULL and the node above it will have the
7898 * correct slot selected for a later read.
7900 * If we increment the root nodes slot counter past the number of
7901 * elements, 1 is returned to signal completion of the search.
7903 static int adjust_slots_upwards(struct btrfs_root *root,
7904 struct btrfs_path *path, int root_level)
7908 struct extent_buffer *eb;
7910 if (root_level == 0)
7913 while (level <= root_level) {
7914 eb = path->nodes[level];
7915 nr = btrfs_header_nritems(eb);
7916 path->slots[level]++;
7917 slot = path->slots[level];
7918 if (slot >= nr || level == 0) {
7920 * Don't free the root - we will detect this
7921 * condition after our loop and return a
7922 * positive value for caller to stop walking the tree.
7924 if (level != root_level) {
7925 btrfs_tree_unlock_rw(eb, path->locks[level]);
7926 path->locks[level] = 0;
7928 free_extent_buffer(eb);
7929 path->nodes[level] = NULL;
7930 path->slots[level] = 0;
7934 * We have a valid slot to walk back down
7935 * from. Stop here so caller can process these
7944 eb = path->nodes[root_level];
7945 if (path->slots[root_level] >= btrfs_header_nritems(eb))
7952 * root_eb is the subtree root and is locked before this function is called.
7953 * TODO: Modify this function to mark all (including complete shared node)
7954 * to dirty_extent_root to allow it get accounted in qgroup.
7956 static int account_shared_subtree(struct btrfs_trans_handle *trans,
7957 struct btrfs_root *root,
7958 struct extent_buffer *root_eb,
7964 struct extent_buffer *eb = root_eb;
7965 struct btrfs_path *path = NULL;
7967 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
7968 BUG_ON(root_eb == NULL);
7970 if (!root->fs_info->quota_enabled)
7973 if (!extent_buffer_uptodate(root_eb)) {
7974 ret = btrfs_read_buffer(root_eb, root_gen);
7979 if (root_level == 0) {
7980 ret = account_leaf_items(trans, root, root_eb);
7984 path = btrfs_alloc_path();
7989 * Walk down the tree. Missing extent blocks are filled in as
7990 * we go. Metadata is accounted every time we read a new
7993 * When we reach a leaf, we account for file extent items in it,
7994 * walk back up the tree (adjusting slot pointers as we go)
7995 * and restart the search process.
7997 extent_buffer_get(root_eb); /* For path */
7998 path->nodes[root_level] = root_eb;
7999 path->slots[root_level] = 0;
8000 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8003 while (level >= 0) {
8004 if (path->nodes[level] == NULL) {
8009 /* We need to get child blockptr/gen from
8010 * parent before we can read it. */
8011 eb = path->nodes[level + 1];
8012 parent_slot = path->slots[level + 1];
8013 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8014 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8016 eb = read_tree_block(root, child_bytenr, child_gen);
8020 } else if (!extent_buffer_uptodate(eb)) {
8021 free_extent_buffer(eb);
8026 path->nodes[level] = eb;
8027 path->slots[level] = 0;
8029 btrfs_tree_read_lock(eb);
8030 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8031 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8035 ret = account_leaf_items(trans, root, path->nodes[level]);
8039 /* Nonzero return here means we completed our search */
8040 ret = adjust_slots_upwards(root, path, root_level);
8044 /* Restart search with new slots */
8053 btrfs_free_path(path);
8059 * helper to process tree block while walking down the tree.
8061 * when wc->stage == UPDATE_BACKREF, this function updates
8062 * back refs for pointers in the block.
8064 * NOTE: return value 1 means we should stop walking down.
8066 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8067 struct btrfs_root *root,
8068 struct btrfs_path *path,
8069 struct walk_control *wc, int lookup_info)
8071 int level = wc->level;
8072 struct extent_buffer *eb = path->nodes[level];
8073 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8076 if (wc->stage == UPDATE_BACKREF &&
8077 btrfs_header_owner(eb) != root->root_key.objectid)
8081 * when reference count of tree block is 1, it won't increase
8082 * again. once full backref flag is set, we never clear it.
8085 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8086 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8087 BUG_ON(!path->locks[level]);
8088 ret = btrfs_lookup_extent_info(trans, root,
8089 eb->start, level, 1,
8092 BUG_ON(ret == -ENOMEM);
8095 BUG_ON(wc->refs[level] == 0);
8098 if (wc->stage == DROP_REFERENCE) {
8099 if (wc->refs[level] > 1)
8102 if (path->locks[level] && !wc->keep_locks) {
8103 btrfs_tree_unlock_rw(eb, path->locks[level]);
8104 path->locks[level] = 0;
8109 /* wc->stage == UPDATE_BACKREF */
8110 if (!(wc->flags[level] & flag)) {
8111 BUG_ON(!path->locks[level]);
8112 ret = btrfs_inc_ref(trans, root, eb, 1);
8113 BUG_ON(ret); /* -ENOMEM */
8114 ret = btrfs_dec_ref(trans, root, eb, 0);
8115 BUG_ON(ret); /* -ENOMEM */
8116 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8118 btrfs_header_level(eb), 0);
8119 BUG_ON(ret); /* -ENOMEM */
8120 wc->flags[level] |= flag;
8124 * the block is shared by multiple trees, so it's not good to
8125 * keep the tree lock
8127 if (path->locks[level] && level > 0) {
8128 btrfs_tree_unlock_rw(eb, path->locks[level]);
8129 path->locks[level] = 0;
8135 * helper to process tree block pointer.
8137 * when wc->stage == DROP_REFERENCE, this function checks
8138 * reference count of the block pointed to. if the block
8139 * is shared and we need update back refs for the subtree
8140 * rooted at the block, this function changes wc->stage to
8141 * UPDATE_BACKREF. if the block is shared and there is no
8142 * need to update back, this function drops the reference
8145 * NOTE: return value 1 means we should stop walking down.
8147 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8148 struct btrfs_root *root,
8149 struct btrfs_path *path,
8150 struct walk_control *wc, int *lookup_info)
8156 struct btrfs_key key;
8157 struct extent_buffer *next;
8158 int level = wc->level;
8161 bool need_account = false;
8163 generation = btrfs_node_ptr_generation(path->nodes[level],
8164 path->slots[level]);
8166 * if the lower level block was created before the snapshot
8167 * was created, we know there is no need to update back refs
8170 if (wc->stage == UPDATE_BACKREF &&
8171 generation <= root->root_key.offset) {
8176 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8177 blocksize = root->nodesize;
8179 next = btrfs_find_tree_block(root->fs_info, bytenr);
8181 next = btrfs_find_create_tree_block(root, bytenr);
8184 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8188 btrfs_tree_lock(next);
8189 btrfs_set_lock_blocking(next);
8191 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8192 &wc->refs[level - 1],
8193 &wc->flags[level - 1]);
8195 btrfs_tree_unlock(next);
8199 if (unlikely(wc->refs[level - 1] == 0)) {
8200 btrfs_err(root->fs_info, "Missing references.");
8205 if (wc->stage == DROP_REFERENCE) {
8206 if (wc->refs[level - 1] > 1) {
8207 need_account = true;
8209 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8212 if (!wc->update_ref ||
8213 generation <= root->root_key.offset)
8216 btrfs_node_key_to_cpu(path->nodes[level], &key,
8217 path->slots[level]);
8218 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8222 wc->stage = UPDATE_BACKREF;
8223 wc->shared_level = level - 1;
8227 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8231 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8232 btrfs_tree_unlock(next);
8233 free_extent_buffer(next);
8239 if (reada && level == 1)
8240 reada_walk_down(trans, root, wc, path);
8241 next = read_tree_block(root, bytenr, generation);
8243 return PTR_ERR(next);
8244 } else if (!extent_buffer_uptodate(next)) {
8245 free_extent_buffer(next);
8248 btrfs_tree_lock(next);
8249 btrfs_set_lock_blocking(next);
8253 BUG_ON(level != btrfs_header_level(next));
8254 path->nodes[level] = next;
8255 path->slots[level] = 0;
8256 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8262 wc->refs[level - 1] = 0;
8263 wc->flags[level - 1] = 0;
8264 if (wc->stage == DROP_REFERENCE) {
8265 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8266 parent = path->nodes[level]->start;
8268 BUG_ON(root->root_key.objectid !=
8269 btrfs_header_owner(path->nodes[level]));
8274 ret = account_shared_subtree(trans, root, next,
8275 generation, level - 1);
8277 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8278 "%d accounting shared subtree. Quota "
8279 "is out of sync, rescan required.\n",
8280 root->fs_info->sb->s_id, ret);
8283 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8284 root->root_key.objectid, level - 1, 0, 0);
8285 BUG_ON(ret); /* -ENOMEM */
8287 btrfs_tree_unlock(next);
8288 free_extent_buffer(next);
8294 * helper to process tree block while walking up the tree.
8296 * when wc->stage == DROP_REFERENCE, this function drops
8297 * reference count on the block.
8299 * when wc->stage == UPDATE_BACKREF, this function changes
8300 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8301 * to UPDATE_BACKREF previously while processing the block.
8303 * NOTE: return value 1 means we should stop walking up.
8305 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8306 struct btrfs_root *root,
8307 struct btrfs_path *path,
8308 struct walk_control *wc)
8311 int level = wc->level;
8312 struct extent_buffer *eb = path->nodes[level];
8315 if (wc->stage == UPDATE_BACKREF) {
8316 BUG_ON(wc->shared_level < level);
8317 if (level < wc->shared_level)
8320 ret = find_next_key(path, level + 1, &wc->update_progress);
8324 wc->stage = DROP_REFERENCE;
8325 wc->shared_level = -1;
8326 path->slots[level] = 0;
8329 * check reference count again if the block isn't locked.
8330 * we should start walking down the tree again if reference
8333 if (!path->locks[level]) {
8335 btrfs_tree_lock(eb);
8336 btrfs_set_lock_blocking(eb);
8337 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8339 ret = btrfs_lookup_extent_info(trans, root,
8340 eb->start, level, 1,
8344 btrfs_tree_unlock_rw(eb, path->locks[level]);
8345 path->locks[level] = 0;
8348 BUG_ON(wc->refs[level] == 0);
8349 if (wc->refs[level] == 1) {
8350 btrfs_tree_unlock_rw(eb, path->locks[level]);
8351 path->locks[level] = 0;
8357 /* wc->stage == DROP_REFERENCE */
8358 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8360 if (wc->refs[level] == 1) {
8362 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8363 ret = btrfs_dec_ref(trans, root, eb, 1);
8365 ret = btrfs_dec_ref(trans, root, eb, 0);
8366 BUG_ON(ret); /* -ENOMEM */
8367 ret = account_leaf_items(trans, root, eb);
8369 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8370 "%d accounting leaf items. Quota "
8371 "is out of sync, rescan required.\n",
8372 root->fs_info->sb->s_id, ret);
8375 /* make block locked assertion in clean_tree_block happy */
8376 if (!path->locks[level] &&
8377 btrfs_header_generation(eb) == trans->transid) {
8378 btrfs_tree_lock(eb);
8379 btrfs_set_lock_blocking(eb);
8380 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8382 clean_tree_block(trans, root->fs_info, eb);
8385 if (eb == root->node) {
8386 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8389 BUG_ON(root->root_key.objectid !=
8390 btrfs_header_owner(eb));
8392 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8393 parent = path->nodes[level + 1]->start;
8395 BUG_ON(root->root_key.objectid !=
8396 btrfs_header_owner(path->nodes[level + 1]));
8399 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8401 wc->refs[level] = 0;
8402 wc->flags[level] = 0;
8406 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8407 struct btrfs_root *root,
8408 struct btrfs_path *path,
8409 struct walk_control *wc)
8411 int level = wc->level;
8412 int lookup_info = 1;
8415 while (level >= 0) {
8416 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8423 if (path->slots[level] >=
8424 btrfs_header_nritems(path->nodes[level]))
8427 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8429 path->slots[level]++;
8438 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8439 struct btrfs_root *root,
8440 struct btrfs_path *path,
8441 struct walk_control *wc, int max_level)
8443 int level = wc->level;
8446 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8447 while (level < max_level && path->nodes[level]) {
8449 if (path->slots[level] + 1 <
8450 btrfs_header_nritems(path->nodes[level])) {
8451 path->slots[level]++;
8454 ret = walk_up_proc(trans, root, path, wc);
8458 if (path->locks[level]) {
8459 btrfs_tree_unlock_rw(path->nodes[level],
8460 path->locks[level]);
8461 path->locks[level] = 0;
8463 free_extent_buffer(path->nodes[level]);
8464 path->nodes[level] = NULL;
8472 * drop a subvolume tree.
8474 * this function traverses the tree freeing any blocks that only
8475 * referenced by the tree.
8477 * when a shared tree block is found. this function decreases its
8478 * reference count by one. if update_ref is true, this function
8479 * also make sure backrefs for the shared block and all lower level
8480 * blocks are properly updated.
8482 * If called with for_reloc == 0, may exit early with -EAGAIN
8484 int btrfs_drop_snapshot(struct btrfs_root *root,
8485 struct btrfs_block_rsv *block_rsv, int update_ref,
8488 struct btrfs_path *path;
8489 struct btrfs_trans_handle *trans;
8490 struct btrfs_root *tree_root = root->fs_info->tree_root;
8491 struct btrfs_root_item *root_item = &root->root_item;
8492 struct walk_control *wc;
8493 struct btrfs_key key;
8497 bool root_dropped = false;
8499 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8501 path = btrfs_alloc_path();
8507 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8509 btrfs_free_path(path);
8514 trans = btrfs_start_transaction(tree_root, 0);
8515 if (IS_ERR(trans)) {
8516 err = PTR_ERR(trans);
8521 trans->block_rsv = block_rsv;
8523 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8524 level = btrfs_header_level(root->node);
8525 path->nodes[level] = btrfs_lock_root_node(root);
8526 btrfs_set_lock_blocking(path->nodes[level]);
8527 path->slots[level] = 0;
8528 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8529 memset(&wc->update_progress, 0,
8530 sizeof(wc->update_progress));
8532 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8533 memcpy(&wc->update_progress, &key,
8534 sizeof(wc->update_progress));
8536 level = root_item->drop_level;
8538 path->lowest_level = level;
8539 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8540 path->lowest_level = 0;
8548 * unlock our path, this is safe because only this
8549 * function is allowed to delete this snapshot
8551 btrfs_unlock_up_safe(path, 0);
8553 level = btrfs_header_level(root->node);
8555 btrfs_tree_lock(path->nodes[level]);
8556 btrfs_set_lock_blocking(path->nodes[level]);
8557 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8559 ret = btrfs_lookup_extent_info(trans, root,
8560 path->nodes[level]->start,
8561 level, 1, &wc->refs[level],
8567 BUG_ON(wc->refs[level] == 0);
8569 if (level == root_item->drop_level)
8572 btrfs_tree_unlock(path->nodes[level]);
8573 path->locks[level] = 0;
8574 WARN_ON(wc->refs[level] != 1);
8580 wc->shared_level = -1;
8581 wc->stage = DROP_REFERENCE;
8582 wc->update_ref = update_ref;
8584 wc->for_reloc = for_reloc;
8585 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8589 ret = walk_down_tree(trans, root, path, wc);
8595 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8602 BUG_ON(wc->stage != DROP_REFERENCE);
8606 if (wc->stage == DROP_REFERENCE) {
8608 btrfs_node_key(path->nodes[level],
8609 &root_item->drop_progress,
8610 path->slots[level]);
8611 root_item->drop_level = level;
8614 BUG_ON(wc->level == 0);
8615 if (btrfs_should_end_transaction(trans, tree_root) ||
8616 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8617 ret = btrfs_update_root(trans, tree_root,
8621 btrfs_abort_transaction(trans, tree_root, ret);
8626 btrfs_end_transaction_throttle(trans, tree_root);
8627 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8628 pr_debug("BTRFS: drop snapshot early exit\n");
8633 trans = btrfs_start_transaction(tree_root, 0);
8634 if (IS_ERR(trans)) {
8635 err = PTR_ERR(trans);
8639 trans->block_rsv = block_rsv;
8642 btrfs_release_path(path);
8646 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8648 btrfs_abort_transaction(trans, tree_root, ret);
8652 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8653 ret = btrfs_find_root(tree_root, &root->root_key, path,
8656 btrfs_abort_transaction(trans, tree_root, ret);
8659 } else if (ret > 0) {
8660 /* if we fail to delete the orphan item this time
8661 * around, it'll get picked up the next time.
8663 * The most common failure here is just -ENOENT.
8665 btrfs_del_orphan_item(trans, tree_root,
8666 root->root_key.objectid);
8670 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8671 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
8673 free_extent_buffer(root->node);
8674 free_extent_buffer(root->commit_root);
8675 btrfs_put_fs_root(root);
8677 root_dropped = true;
8679 btrfs_end_transaction_throttle(trans, tree_root);
8682 btrfs_free_path(path);
8685 * So if we need to stop dropping the snapshot for whatever reason we
8686 * need to make sure to add it back to the dead root list so that we
8687 * keep trying to do the work later. This also cleans up roots if we
8688 * don't have it in the radix (like when we recover after a power fail
8689 * or unmount) so we don't leak memory.
8691 if (!for_reloc && root_dropped == false)
8692 btrfs_add_dead_root(root);
8693 if (err && err != -EAGAIN)
8694 btrfs_std_error(root->fs_info, err);
8699 * drop subtree rooted at tree block 'node'.
8701 * NOTE: this function will unlock and release tree block 'node'
8702 * only used by relocation code
8704 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8705 struct btrfs_root *root,
8706 struct extent_buffer *node,
8707 struct extent_buffer *parent)
8709 struct btrfs_path *path;
8710 struct walk_control *wc;
8716 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8718 path = btrfs_alloc_path();
8722 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8724 btrfs_free_path(path);
8728 btrfs_assert_tree_locked(parent);
8729 parent_level = btrfs_header_level(parent);
8730 extent_buffer_get(parent);
8731 path->nodes[parent_level] = parent;
8732 path->slots[parent_level] = btrfs_header_nritems(parent);
8734 btrfs_assert_tree_locked(node);
8735 level = btrfs_header_level(node);
8736 path->nodes[level] = node;
8737 path->slots[level] = 0;
8738 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8740 wc->refs[parent_level] = 1;
8741 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8743 wc->shared_level = -1;
8744 wc->stage = DROP_REFERENCE;
8748 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8751 wret = walk_down_tree(trans, root, path, wc);
8757 wret = walk_up_tree(trans, root, path, wc, parent_level);
8765 btrfs_free_path(path);
8769 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8775 * if restripe for this chunk_type is on pick target profile and
8776 * return, otherwise do the usual balance
8778 stripped = get_restripe_target(root->fs_info, flags);
8780 return extended_to_chunk(stripped);
8782 num_devices = root->fs_info->fs_devices->rw_devices;
8784 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8785 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8786 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8788 if (num_devices == 1) {
8789 stripped |= BTRFS_BLOCK_GROUP_DUP;
8790 stripped = flags & ~stripped;
8792 /* turn raid0 into single device chunks */
8793 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8796 /* turn mirroring into duplication */
8797 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8798 BTRFS_BLOCK_GROUP_RAID10))
8799 return stripped | BTRFS_BLOCK_GROUP_DUP;
8801 /* they already had raid on here, just return */
8802 if (flags & stripped)
8805 stripped |= BTRFS_BLOCK_GROUP_DUP;
8806 stripped = flags & ~stripped;
8808 /* switch duplicated blocks with raid1 */
8809 if (flags & BTRFS_BLOCK_GROUP_DUP)
8810 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8812 /* this is drive concat, leave it alone */
8818 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8820 struct btrfs_space_info *sinfo = cache->space_info;
8822 u64 min_allocable_bytes;
8826 * We need some metadata space and system metadata space for
8827 * allocating chunks in some corner cases until we force to set
8828 * it to be readonly.
8831 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8833 min_allocable_bytes = 1 * 1024 * 1024;
8835 min_allocable_bytes = 0;
8837 spin_lock(&sinfo->lock);
8838 spin_lock(&cache->lock);
8846 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8847 cache->bytes_super - btrfs_block_group_used(&cache->item);
8849 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8850 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8851 min_allocable_bytes <= sinfo->total_bytes) {
8852 sinfo->bytes_readonly += num_bytes;
8854 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8858 spin_unlock(&cache->lock);
8859 spin_unlock(&sinfo->lock);
8863 int btrfs_inc_block_group_ro(struct btrfs_root *root,
8864 struct btrfs_block_group_cache *cache)
8867 struct btrfs_trans_handle *trans;
8872 trans = btrfs_join_transaction(root);
8874 return PTR_ERR(trans);
8877 * we're not allowed to set block groups readonly after the dirty
8878 * block groups cache has started writing. If it already started,
8879 * back off and let this transaction commit
8881 mutex_lock(&root->fs_info->ro_block_group_mutex);
8882 if (trans->transaction->dirty_bg_run) {
8883 u64 transid = trans->transid;
8885 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8886 btrfs_end_transaction(trans, root);
8888 ret = btrfs_wait_for_commit(root, transid);
8895 * if we are changing raid levels, try to allocate a corresponding
8896 * block group with the new raid level.
8898 alloc_flags = update_block_group_flags(root, cache->flags);
8899 if (alloc_flags != cache->flags) {
8900 ret = do_chunk_alloc(trans, root, alloc_flags,
8903 * ENOSPC is allowed here, we may have enough space
8904 * already allocated at the new raid level to
8913 ret = inc_block_group_ro(cache, 0);
8916 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8917 ret = do_chunk_alloc(trans, root, alloc_flags,
8921 ret = inc_block_group_ro(cache, 0);
8923 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
8924 alloc_flags = update_block_group_flags(root, cache->flags);
8925 lock_chunks(root->fs_info->chunk_root);
8926 check_system_chunk(trans, root, alloc_flags);
8927 unlock_chunks(root->fs_info->chunk_root);
8929 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8931 btrfs_end_transaction(trans, root);
8935 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8936 struct btrfs_root *root, u64 type)
8938 u64 alloc_flags = get_alloc_profile(root, type);
8939 return do_chunk_alloc(trans, root, alloc_flags,
8944 * helper to account the unused space of all the readonly block group in the
8945 * space_info. takes mirrors into account.
8947 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8949 struct btrfs_block_group_cache *block_group;
8953 /* It's df, we don't care if it's racey */
8954 if (list_empty(&sinfo->ro_bgs))
8957 spin_lock(&sinfo->lock);
8958 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
8959 spin_lock(&block_group->lock);
8961 if (!block_group->ro) {
8962 spin_unlock(&block_group->lock);
8966 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8967 BTRFS_BLOCK_GROUP_RAID10 |
8968 BTRFS_BLOCK_GROUP_DUP))
8973 free_bytes += (block_group->key.offset -
8974 btrfs_block_group_used(&block_group->item)) *
8977 spin_unlock(&block_group->lock);
8979 spin_unlock(&sinfo->lock);
8984 void btrfs_dec_block_group_ro(struct btrfs_root *root,
8985 struct btrfs_block_group_cache *cache)
8987 struct btrfs_space_info *sinfo = cache->space_info;
8992 spin_lock(&sinfo->lock);
8993 spin_lock(&cache->lock);
8995 num_bytes = cache->key.offset - cache->reserved -
8996 cache->pinned - cache->bytes_super -
8997 btrfs_block_group_used(&cache->item);
8998 sinfo->bytes_readonly -= num_bytes;
8999 list_del_init(&cache->ro_list);
9001 spin_unlock(&cache->lock);
9002 spin_unlock(&sinfo->lock);
9006 * checks to see if its even possible to relocate this block group.
9008 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9009 * ok to go ahead and try.
9011 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9013 struct btrfs_block_group_cache *block_group;
9014 struct btrfs_space_info *space_info;
9015 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9016 struct btrfs_device *device;
9017 struct btrfs_trans_handle *trans;
9026 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9028 /* odd, couldn't find the block group, leave it alone */
9032 min_free = btrfs_block_group_used(&block_group->item);
9034 /* no bytes used, we're good */
9038 space_info = block_group->space_info;
9039 spin_lock(&space_info->lock);
9041 full = space_info->full;
9044 * if this is the last block group we have in this space, we can't
9045 * relocate it unless we're able to allocate a new chunk below.
9047 * Otherwise, we need to make sure we have room in the space to handle
9048 * all of the extents from this block group. If we can, we're good
9050 if ((space_info->total_bytes != block_group->key.offset) &&
9051 (space_info->bytes_used + space_info->bytes_reserved +
9052 space_info->bytes_pinned + space_info->bytes_readonly +
9053 min_free < space_info->total_bytes)) {
9054 spin_unlock(&space_info->lock);
9057 spin_unlock(&space_info->lock);
9060 * ok we don't have enough space, but maybe we have free space on our
9061 * devices to allocate new chunks for relocation, so loop through our
9062 * alloc devices and guess if we have enough space. if this block
9063 * group is going to be restriped, run checks against the target
9064 * profile instead of the current one.
9076 target = get_restripe_target(root->fs_info, block_group->flags);
9078 index = __get_raid_index(extended_to_chunk(target));
9081 * this is just a balance, so if we were marked as full
9082 * we know there is no space for a new chunk
9087 index = get_block_group_index(block_group);
9090 if (index == BTRFS_RAID_RAID10) {
9094 } else if (index == BTRFS_RAID_RAID1) {
9096 } else if (index == BTRFS_RAID_DUP) {
9099 } else if (index == BTRFS_RAID_RAID0) {
9100 dev_min = fs_devices->rw_devices;
9101 min_free = div64_u64(min_free, dev_min);
9104 /* We need to do this so that we can look at pending chunks */
9105 trans = btrfs_join_transaction(root);
9106 if (IS_ERR(trans)) {
9107 ret = PTR_ERR(trans);
9111 mutex_lock(&root->fs_info->chunk_mutex);
9112 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9116 * check to make sure we can actually find a chunk with enough
9117 * space to fit our block group in.
9119 if (device->total_bytes > device->bytes_used + min_free &&
9120 !device->is_tgtdev_for_dev_replace) {
9121 ret = find_free_dev_extent(trans, device, min_free,
9126 if (dev_nr >= dev_min)
9132 mutex_unlock(&root->fs_info->chunk_mutex);
9133 btrfs_end_transaction(trans, root);
9135 btrfs_put_block_group(block_group);
9139 static int find_first_block_group(struct btrfs_root *root,
9140 struct btrfs_path *path, struct btrfs_key *key)
9143 struct btrfs_key found_key;
9144 struct extent_buffer *leaf;
9147 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9152 slot = path->slots[0];
9153 leaf = path->nodes[0];
9154 if (slot >= btrfs_header_nritems(leaf)) {
9155 ret = btrfs_next_leaf(root, path);
9162 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9164 if (found_key.objectid >= key->objectid &&
9165 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9175 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9177 struct btrfs_block_group_cache *block_group;
9181 struct inode *inode;
9183 block_group = btrfs_lookup_first_block_group(info, last);
9184 while (block_group) {
9185 spin_lock(&block_group->lock);
9186 if (block_group->iref)
9188 spin_unlock(&block_group->lock);
9189 block_group = next_block_group(info->tree_root,
9199 inode = block_group->inode;
9200 block_group->iref = 0;
9201 block_group->inode = NULL;
9202 spin_unlock(&block_group->lock);
9204 last = block_group->key.objectid + block_group->key.offset;
9205 btrfs_put_block_group(block_group);
9209 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9211 struct btrfs_block_group_cache *block_group;
9212 struct btrfs_space_info *space_info;
9213 struct btrfs_caching_control *caching_ctl;
9216 down_write(&info->commit_root_sem);
9217 while (!list_empty(&info->caching_block_groups)) {
9218 caching_ctl = list_entry(info->caching_block_groups.next,
9219 struct btrfs_caching_control, list);
9220 list_del(&caching_ctl->list);
9221 put_caching_control(caching_ctl);
9223 up_write(&info->commit_root_sem);
9225 spin_lock(&info->unused_bgs_lock);
9226 while (!list_empty(&info->unused_bgs)) {
9227 block_group = list_first_entry(&info->unused_bgs,
9228 struct btrfs_block_group_cache,
9230 list_del_init(&block_group->bg_list);
9231 btrfs_put_block_group(block_group);
9233 spin_unlock(&info->unused_bgs_lock);
9235 spin_lock(&info->block_group_cache_lock);
9236 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9237 block_group = rb_entry(n, struct btrfs_block_group_cache,
9239 rb_erase(&block_group->cache_node,
9240 &info->block_group_cache_tree);
9241 RB_CLEAR_NODE(&block_group->cache_node);
9242 spin_unlock(&info->block_group_cache_lock);
9244 down_write(&block_group->space_info->groups_sem);
9245 list_del(&block_group->list);
9246 up_write(&block_group->space_info->groups_sem);
9248 if (block_group->cached == BTRFS_CACHE_STARTED)
9249 wait_block_group_cache_done(block_group);
9252 * We haven't cached this block group, which means we could
9253 * possibly have excluded extents on this block group.
9255 if (block_group->cached == BTRFS_CACHE_NO ||
9256 block_group->cached == BTRFS_CACHE_ERROR)
9257 free_excluded_extents(info->extent_root, block_group);
9259 btrfs_remove_free_space_cache(block_group);
9260 btrfs_put_block_group(block_group);
9262 spin_lock(&info->block_group_cache_lock);
9264 spin_unlock(&info->block_group_cache_lock);
9266 /* now that all the block groups are freed, go through and
9267 * free all the space_info structs. This is only called during
9268 * the final stages of unmount, and so we know nobody is
9269 * using them. We call synchronize_rcu() once before we start,
9270 * just to be on the safe side.
9274 release_global_block_rsv(info);
9276 while (!list_empty(&info->space_info)) {
9279 space_info = list_entry(info->space_info.next,
9280 struct btrfs_space_info,
9282 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9283 if (WARN_ON(space_info->bytes_pinned > 0 ||
9284 space_info->bytes_reserved > 0 ||
9285 space_info->bytes_may_use > 0)) {
9286 dump_space_info(space_info, 0, 0);
9289 list_del(&space_info->list);
9290 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9291 struct kobject *kobj;
9292 kobj = space_info->block_group_kobjs[i];
9293 space_info->block_group_kobjs[i] = NULL;
9299 kobject_del(&space_info->kobj);
9300 kobject_put(&space_info->kobj);
9305 static void __link_block_group(struct btrfs_space_info *space_info,
9306 struct btrfs_block_group_cache *cache)
9308 int index = get_block_group_index(cache);
9311 down_write(&space_info->groups_sem);
9312 if (list_empty(&space_info->block_groups[index]))
9314 list_add_tail(&cache->list, &space_info->block_groups[index]);
9315 up_write(&space_info->groups_sem);
9318 struct raid_kobject *rkobj;
9321 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9324 rkobj->raid_type = index;
9325 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9326 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9327 "%s", get_raid_name(index));
9329 kobject_put(&rkobj->kobj);
9332 space_info->block_group_kobjs[index] = &rkobj->kobj;
9337 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9340 static struct btrfs_block_group_cache *
9341 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9343 struct btrfs_block_group_cache *cache;
9345 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9349 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9351 if (!cache->free_space_ctl) {
9356 cache->key.objectid = start;
9357 cache->key.offset = size;
9358 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9360 cache->sectorsize = root->sectorsize;
9361 cache->fs_info = root->fs_info;
9362 cache->full_stripe_len = btrfs_full_stripe_len(root,
9363 &root->fs_info->mapping_tree,
9365 atomic_set(&cache->count, 1);
9366 spin_lock_init(&cache->lock);
9367 init_rwsem(&cache->data_rwsem);
9368 INIT_LIST_HEAD(&cache->list);
9369 INIT_LIST_HEAD(&cache->cluster_list);
9370 INIT_LIST_HEAD(&cache->bg_list);
9371 INIT_LIST_HEAD(&cache->ro_list);
9372 INIT_LIST_HEAD(&cache->dirty_list);
9373 INIT_LIST_HEAD(&cache->io_list);
9374 btrfs_init_free_space_ctl(cache);
9375 atomic_set(&cache->trimming, 0);
9380 int btrfs_read_block_groups(struct btrfs_root *root)
9382 struct btrfs_path *path;
9384 struct btrfs_block_group_cache *cache;
9385 struct btrfs_fs_info *info = root->fs_info;
9386 struct btrfs_space_info *space_info;
9387 struct btrfs_key key;
9388 struct btrfs_key found_key;
9389 struct extent_buffer *leaf;
9393 root = info->extent_root;
9396 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9397 path = btrfs_alloc_path();
9402 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9403 if (btrfs_test_opt(root, SPACE_CACHE) &&
9404 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9406 if (btrfs_test_opt(root, CLEAR_CACHE))
9410 ret = find_first_block_group(root, path, &key);
9416 leaf = path->nodes[0];
9417 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9419 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9428 * When we mount with old space cache, we need to
9429 * set BTRFS_DC_CLEAR and set dirty flag.
9431 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9432 * truncate the old free space cache inode and
9434 * b) Setting 'dirty flag' makes sure that we flush
9435 * the new space cache info onto disk.
9437 if (btrfs_test_opt(root, SPACE_CACHE))
9438 cache->disk_cache_state = BTRFS_DC_CLEAR;
9441 read_extent_buffer(leaf, &cache->item,
9442 btrfs_item_ptr_offset(leaf, path->slots[0]),
9443 sizeof(cache->item));
9444 cache->flags = btrfs_block_group_flags(&cache->item);
9446 key.objectid = found_key.objectid + found_key.offset;
9447 btrfs_release_path(path);
9450 * We need to exclude the super stripes now so that the space
9451 * info has super bytes accounted for, otherwise we'll think
9452 * we have more space than we actually do.
9454 ret = exclude_super_stripes(root, cache);
9457 * We may have excluded something, so call this just in
9460 free_excluded_extents(root, cache);
9461 btrfs_put_block_group(cache);
9466 * check for two cases, either we are full, and therefore
9467 * don't need to bother with the caching work since we won't
9468 * find any space, or we are empty, and we can just add all
9469 * the space in and be done with it. This saves us _alot_ of
9470 * time, particularly in the full case.
9472 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9473 cache->last_byte_to_unpin = (u64)-1;
9474 cache->cached = BTRFS_CACHE_FINISHED;
9475 free_excluded_extents(root, cache);
9476 } else if (btrfs_block_group_used(&cache->item) == 0) {
9477 cache->last_byte_to_unpin = (u64)-1;
9478 cache->cached = BTRFS_CACHE_FINISHED;
9479 add_new_free_space(cache, root->fs_info,
9481 found_key.objectid +
9483 free_excluded_extents(root, cache);
9486 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9488 btrfs_remove_free_space_cache(cache);
9489 btrfs_put_block_group(cache);
9493 ret = update_space_info(info, cache->flags, found_key.offset,
9494 btrfs_block_group_used(&cache->item),
9497 btrfs_remove_free_space_cache(cache);
9498 spin_lock(&info->block_group_cache_lock);
9499 rb_erase(&cache->cache_node,
9500 &info->block_group_cache_tree);
9501 RB_CLEAR_NODE(&cache->cache_node);
9502 spin_unlock(&info->block_group_cache_lock);
9503 btrfs_put_block_group(cache);
9507 cache->space_info = space_info;
9508 spin_lock(&cache->space_info->lock);
9509 cache->space_info->bytes_readonly += cache->bytes_super;
9510 spin_unlock(&cache->space_info->lock);
9512 __link_block_group(space_info, cache);
9514 set_avail_alloc_bits(root->fs_info, cache->flags);
9515 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9516 inc_block_group_ro(cache, 1);
9517 } else if (btrfs_block_group_used(&cache->item) == 0) {
9518 spin_lock(&info->unused_bgs_lock);
9519 /* Should always be true but just in case. */
9520 if (list_empty(&cache->bg_list)) {
9521 btrfs_get_block_group(cache);
9522 list_add_tail(&cache->bg_list,
9525 spin_unlock(&info->unused_bgs_lock);
9529 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9530 if (!(get_alloc_profile(root, space_info->flags) &
9531 (BTRFS_BLOCK_GROUP_RAID10 |
9532 BTRFS_BLOCK_GROUP_RAID1 |
9533 BTRFS_BLOCK_GROUP_RAID5 |
9534 BTRFS_BLOCK_GROUP_RAID6 |
9535 BTRFS_BLOCK_GROUP_DUP)))
9538 * avoid allocating from un-mirrored block group if there are
9539 * mirrored block groups.
9541 list_for_each_entry(cache,
9542 &space_info->block_groups[BTRFS_RAID_RAID0],
9544 inc_block_group_ro(cache, 1);
9545 list_for_each_entry(cache,
9546 &space_info->block_groups[BTRFS_RAID_SINGLE],
9548 inc_block_group_ro(cache, 1);
9551 init_global_block_rsv(info);
9554 btrfs_free_path(path);
9558 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9559 struct btrfs_root *root)
9561 struct btrfs_block_group_cache *block_group, *tmp;
9562 struct btrfs_root *extent_root = root->fs_info->extent_root;
9563 struct btrfs_block_group_item item;
9564 struct btrfs_key key;
9567 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9571 spin_lock(&block_group->lock);
9572 memcpy(&item, &block_group->item, sizeof(item));
9573 memcpy(&key, &block_group->key, sizeof(key));
9574 spin_unlock(&block_group->lock);
9576 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9579 btrfs_abort_transaction(trans, extent_root, ret);
9580 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9581 key.objectid, key.offset);
9583 btrfs_abort_transaction(trans, extent_root, ret);
9585 list_del_init(&block_group->bg_list);
9589 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9590 struct btrfs_root *root, u64 bytes_used,
9591 u64 type, u64 chunk_objectid, u64 chunk_offset,
9595 struct btrfs_root *extent_root;
9596 struct btrfs_block_group_cache *cache;
9598 extent_root = root->fs_info->extent_root;
9600 btrfs_set_log_full_commit(root->fs_info, trans);
9602 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9606 btrfs_set_block_group_used(&cache->item, bytes_used);
9607 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9608 btrfs_set_block_group_flags(&cache->item, type);
9610 cache->flags = type;
9611 cache->last_byte_to_unpin = (u64)-1;
9612 cache->cached = BTRFS_CACHE_FINISHED;
9613 ret = exclude_super_stripes(root, cache);
9616 * We may have excluded something, so call this just in
9619 free_excluded_extents(root, cache);
9620 btrfs_put_block_group(cache);
9624 add_new_free_space(cache, root->fs_info, chunk_offset,
9625 chunk_offset + size);
9627 free_excluded_extents(root, cache);
9630 * Call to ensure the corresponding space_info object is created and
9631 * assigned to our block group, but don't update its counters just yet.
9632 * We want our bg to be added to the rbtree with its ->space_info set.
9634 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
9635 &cache->space_info);
9637 btrfs_remove_free_space_cache(cache);
9638 btrfs_put_block_group(cache);
9642 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9644 btrfs_remove_free_space_cache(cache);
9645 btrfs_put_block_group(cache);
9650 * Now that our block group has its ->space_info set and is inserted in
9651 * the rbtree, update the space info's counters.
9653 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9654 &cache->space_info);
9656 btrfs_remove_free_space_cache(cache);
9657 spin_lock(&root->fs_info->block_group_cache_lock);
9658 rb_erase(&cache->cache_node,
9659 &root->fs_info->block_group_cache_tree);
9660 RB_CLEAR_NODE(&cache->cache_node);
9661 spin_unlock(&root->fs_info->block_group_cache_lock);
9662 btrfs_put_block_group(cache);
9665 update_global_block_rsv(root->fs_info);
9667 spin_lock(&cache->space_info->lock);
9668 cache->space_info->bytes_readonly += cache->bytes_super;
9669 spin_unlock(&cache->space_info->lock);
9671 __link_block_group(cache->space_info, cache);
9673 list_add_tail(&cache->bg_list, &trans->new_bgs);
9675 set_avail_alloc_bits(extent_root->fs_info, type);
9680 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9682 u64 extra_flags = chunk_to_extended(flags) &
9683 BTRFS_EXTENDED_PROFILE_MASK;
9685 write_seqlock(&fs_info->profiles_lock);
9686 if (flags & BTRFS_BLOCK_GROUP_DATA)
9687 fs_info->avail_data_alloc_bits &= ~extra_flags;
9688 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9689 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9690 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9691 fs_info->avail_system_alloc_bits &= ~extra_flags;
9692 write_sequnlock(&fs_info->profiles_lock);
9695 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9696 struct btrfs_root *root, u64 group_start,
9697 struct extent_map *em)
9699 struct btrfs_path *path;
9700 struct btrfs_block_group_cache *block_group;
9701 struct btrfs_free_cluster *cluster;
9702 struct btrfs_root *tree_root = root->fs_info->tree_root;
9703 struct btrfs_key key;
9704 struct inode *inode;
9705 struct kobject *kobj = NULL;
9709 struct btrfs_caching_control *caching_ctl = NULL;
9712 root = root->fs_info->extent_root;
9714 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9715 BUG_ON(!block_group);
9716 BUG_ON(!block_group->ro);
9719 * Free the reserved super bytes from this block group before
9722 free_excluded_extents(root, block_group);
9724 memcpy(&key, &block_group->key, sizeof(key));
9725 index = get_block_group_index(block_group);
9726 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9727 BTRFS_BLOCK_GROUP_RAID1 |
9728 BTRFS_BLOCK_GROUP_RAID10))
9733 /* make sure this block group isn't part of an allocation cluster */
9734 cluster = &root->fs_info->data_alloc_cluster;
9735 spin_lock(&cluster->refill_lock);
9736 btrfs_return_cluster_to_free_space(block_group, cluster);
9737 spin_unlock(&cluster->refill_lock);
9740 * make sure this block group isn't part of a metadata
9741 * allocation cluster
9743 cluster = &root->fs_info->meta_alloc_cluster;
9744 spin_lock(&cluster->refill_lock);
9745 btrfs_return_cluster_to_free_space(block_group, cluster);
9746 spin_unlock(&cluster->refill_lock);
9748 path = btrfs_alloc_path();
9755 * get the inode first so any iput calls done for the io_list
9756 * aren't the final iput (no unlinks allowed now)
9758 inode = lookup_free_space_inode(tree_root, block_group, path);
9760 mutex_lock(&trans->transaction->cache_write_mutex);
9762 * make sure our free spache cache IO is done before remove the
9765 spin_lock(&trans->transaction->dirty_bgs_lock);
9766 if (!list_empty(&block_group->io_list)) {
9767 list_del_init(&block_group->io_list);
9769 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
9771 spin_unlock(&trans->transaction->dirty_bgs_lock);
9772 btrfs_wait_cache_io(root, trans, block_group,
9773 &block_group->io_ctl, path,
9774 block_group->key.objectid);
9775 btrfs_put_block_group(block_group);
9776 spin_lock(&trans->transaction->dirty_bgs_lock);
9779 if (!list_empty(&block_group->dirty_list)) {
9780 list_del_init(&block_group->dirty_list);
9781 btrfs_put_block_group(block_group);
9783 spin_unlock(&trans->transaction->dirty_bgs_lock);
9784 mutex_unlock(&trans->transaction->cache_write_mutex);
9786 if (!IS_ERR(inode)) {
9787 ret = btrfs_orphan_add(trans, inode);
9789 btrfs_add_delayed_iput(inode);
9793 /* One for the block groups ref */
9794 spin_lock(&block_group->lock);
9795 if (block_group->iref) {
9796 block_group->iref = 0;
9797 block_group->inode = NULL;
9798 spin_unlock(&block_group->lock);
9801 spin_unlock(&block_group->lock);
9803 /* One for our lookup ref */
9804 btrfs_add_delayed_iput(inode);
9807 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9808 key.offset = block_group->key.objectid;
9811 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9815 btrfs_release_path(path);
9817 ret = btrfs_del_item(trans, tree_root, path);
9820 btrfs_release_path(path);
9823 spin_lock(&root->fs_info->block_group_cache_lock);
9824 rb_erase(&block_group->cache_node,
9825 &root->fs_info->block_group_cache_tree);
9826 RB_CLEAR_NODE(&block_group->cache_node);
9828 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9829 root->fs_info->first_logical_byte = (u64)-1;
9830 spin_unlock(&root->fs_info->block_group_cache_lock);
9832 down_write(&block_group->space_info->groups_sem);
9834 * we must use list_del_init so people can check to see if they
9835 * are still on the list after taking the semaphore
9837 list_del_init(&block_group->list);
9838 if (list_empty(&block_group->space_info->block_groups[index])) {
9839 kobj = block_group->space_info->block_group_kobjs[index];
9840 block_group->space_info->block_group_kobjs[index] = NULL;
9841 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9843 up_write(&block_group->space_info->groups_sem);
9849 if (block_group->has_caching_ctl)
9850 caching_ctl = get_caching_control(block_group);
9851 if (block_group->cached == BTRFS_CACHE_STARTED)
9852 wait_block_group_cache_done(block_group);
9853 if (block_group->has_caching_ctl) {
9854 down_write(&root->fs_info->commit_root_sem);
9856 struct btrfs_caching_control *ctl;
9858 list_for_each_entry(ctl,
9859 &root->fs_info->caching_block_groups, list)
9860 if (ctl->block_group == block_group) {
9862 atomic_inc(&caching_ctl->count);
9867 list_del_init(&caching_ctl->list);
9868 up_write(&root->fs_info->commit_root_sem);
9870 /* Once for the caching bgs list and once for us. */
9871 put_caching_control(caching_ctl);
9872 put_caching_control(caching_ctl);
9876 spin_lock(&trans->transaction->dirty_bgs_lock);
9877 if (!list_empty(&block_group->dirty_list)) {
9880 if (!list_empty(&block_group->io_list)) {
9883 spin_unlock(&trans->transaction->dirty_bgs_lock);
9884 btrfs_remove_free_space_cache(block_group);
9886 spin_lock(&block_group->space_info->lock);
9887 list_del_init(&block_group->ro_list);
9889 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
9890 WARN_ON(block_group->space_info->total_bytes
9891 < block_group->key.offset);
9892 WARN_ON(block_group->space_info->bytes_readonly
9893 < block_group->key.offset);
9894 WARN_ON(block_group->space_info->disk_total
9895 < block_group->key.offset * factor);
9897 block_group->space_info->total_bytes -= block_group->key.offset;
9898 block_group->space_info->bytes_readonly -= block_group->key.offset;
9899 block_group->space_info->disk_total -= block_group->key.offset * factor;
9901 spin_unlock(&block_group->space_info->lock);
9903 memcpy(&key, &block_group->key, sizeof(key));
9906 if (!list_empty(&em->list)) {
9907 /* We're in the transaction->pending_chunks list. */
9908 free_extent_map(em);
9910 spin_lock(&block_group->lock);
9911 block_group->removed = 1;
9913 * At this point trimming can't start on this block group, because we
9914 * removed the block group from the tree fs_info->block_group_cache_tree
9915 * so no one can't find it anymore and even if someone already got this
9916 * block group before we removed it from the rbtree, they have already
9917 * incremented block_group->trimming - if they didn't, they won't find
9918 * any free space entries because we already removed them all when we
9919 * called btrfs_remove_free_space_cache().
9921 * And we must not remove the extent map from the fs_info->mapping_tree
9922 * to prevent the same logical address range and physical device space
9923 * ranges from being reused for a new block group. This is because our
9924 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9925 * completely transactionless, so while it is trimming a range the
9926 * currently running transaction might finish and a new one start,
9927 * allowing for new block groups to be created that can reuse the same
9928 * physical device locations unless we take this special care.
9930 * There may also be an implicit trim operation if the file system
9931 * is mounted with -odiscard. The same protections must remain
9932 * in place until the extents have been discarded completely when
9933 * the transaction commit has completed.
9935 remove_em = (atomic_read(&block_group->trimming) == 0);
9937 * Make sure a trimmer task always sees the em in the pinned_chunks list
9938 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9939 * before checking block_group->removed).
9943 * Our em might be in trans->transaction->pending_chunks which
9944 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9945 * and so is the fs_info->pinned_chunks list.
9947 * So at this point we must be holding the chunk_mutex to avoid
9948 * any races with chunk allocation (more specifically at
9949 * volumes.c:contains_pending_extent()), to ensure it always
9950 * sees the em, either in the pending_chunks list or in the
9951 * pinned_chunks list.
9953 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
9955 spin_unlock(&block_group->lock);
9958 struct extent_map_tree *em_tree;
9960 em_tree = &root->fs_info->mapping_tree.map_tree;
9961 write_lock(&em_tree->lock);
9963 * The em might be in the pending_chunks list, so make sure the
9964 * chunk mutex is locked, since remove_extent_mapping() will
9965 * delete us from that list.
9967 remove_extent_mapping(em_tree, em);
9968 write_unlock(&em_tree->lock);
9969 /* once for the tree */
9970 free_extent_map(em);
9973 unlock_chunks(root);
9975 btrfs_put_block_group(block_group);
9976 btrfs_put_block_group(block_group);
9978 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
9984 ret = btrfs_del_item(trans, root, path);
9986 btrfs_free_path(path);
9991 * Process the unused_bgs list and remove any that don't have any allocated
9992 * space inside of them.
9994 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
9996 struct btrfs_block_group_cache *block_group;
9997 struct btrfs_space_info *space_info;
9998 struct btrfs_root *root = fs_info->extent_root;
9999 struct btrfs_trans_handle *trans;
10002 if (!fs_info->open)
10005 spin_lock(&fs_info->unused_bgs_lock);
10006 while (!list_empty(&fs_info->unused_bgs)) {
10010 block_group = list_first_entry(&fs_info->unused_bgs,
10011 struct btrfs_block_group_cache,
10013 space_info = block_group->space_info;
10014 list_del_init(&block_group->bg_list);
10015 if (ret || btrfs_mixed_space_info(space_info)) {
10016 btrfs_put_block_group(block_group);
10019 spin_unlock(&fs_info->unused_bgs_lock);
10021 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
10023 /* Don't want to race with allocators so take the groups_sem */
10024 down_write(&space_info->groups_sem);
10025 spin_lock(&block_group->lock);
10026 if (block_group->reserved ||
10027 btrfs_block_group_used(&block_group->item) ||
10030 * We want to bail if we made new allocations or have
10031 * outstanding allocations in this block group. We do
10032 * the ro check in case balance is currently acting on
10033 * this block group.
10035 spin_unlock(&block_group->lock);
10036 up_write(&space_info->groups_sem);
10039 spin_unlock(&block_group->lock);
10041 /* We don't want to force the issue, only flip if it's ok. */
10042 ret = inc_block_group_ro(block_group, 0);
10043 up_write(&space_info->groups_sem);
10050 * Want to do this before we do anything else so we can recover
10051 * properly if we fail to join the transaction.
10053 /* 1 for btrfs_orphan_reserve_metadata() */
10054 trans = btrfs_start_transaction(root, 1);
10055 if (IS_ERR(trans)) {
10056 btrfs_dec_block_group_ro(root, block_group);
10057 ret = PTR_ERR(trans);
10062 * We could have pending pinned extents for this block group,
10063 * just delete them, we don't care about them anymore.
10065 start = block_group->key.objectid;
10066 end = start + block_group->key.offset - 1;
10068 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10069 * btrfs_finish_extent_commit(). If we are at transaction N,
10070 * another task might be running finish_extent_commit() for the
10071 * previous transaction N - 1, and have seen a range belonging
10072 * to the block group in freed_extents[] before we were able to
10073 * clear the whole block group range from freed_extents[]. This
10074 * means that task can lookup for the block group after we
10075 * unpinned it from freed_extents[] and removed it, leading to
10076 * a BUG_ON() at btrfs_unpin_extent_range().
10078 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10079 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10080 EXTENT_DIRTY, GFP_NOFS);
10082 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10083 btrfs_dec_block_group_ro(root, block_group);
10086 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10087 EXTENT_DIRTY, GFP_NOFS);
10089 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10090 btrfs_dec_block_group_ro(root, block_group);
10093 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10095 /* Reset pinned so btrfs_put_block_group doesn't complain */
10096 spin_lock(&space_info->lock);
10097 spin_lock(&block_group->lock);
10099 space_info->bytes_pinned -= block_group->pinned;
10100 space_info->bytes_readonly += block_group->pinned;
10101 percpu_counter_add(&space_info->total_bytes_pinned,
10102 -block_group->pinned);
10103 block_group->pinned = 0;
10105 spin_unlock(&block_group->lock);
10106 spin_unlock(&space_info->lock);
10108 /* DISCARD can flip during remount */
10109 trimming = btrfs_test_opt(root, DISCARD);
10111 /* Implicit trim during transaction commit. */
10113 btrfs_get_block_group_trimming(block_group);
10116 * Btrfs_remove_chunk will abort the transaction if things go
10119 ret = btrfs_remove_chunk(trans, root,
10120 block_group->key.objectid);
10124 btrfs_put_block_group_trimming(block_group);
10129 * If we're not mounted with -odiscard, we can just forget
10130 * about this block group. Otherwise we'll need to wait
10131 * until transaction commit to do the actual discard.
10134 WARN_ON(!list_empty(&block_group->bg_list));
10135 spin_lock(&trans->transaction->deleted_bgs_lock);
10136 list_move(&block_group->bg_list,
10137 &trans->transaction->deleted_bgs);
10138 spin_unlock(&trans->transaction->deleted_bgs_lock);
10139 btrfs_get_block_group(block_group);
10142 btrfs_end_transaction(trans, root);
10144 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
10145 btrfs_put_block_group(block_group);
10146 spin_lock(&fs_info->unused_bgs_lock);
10148 spin_unlock(&fs_info->unused_bgs_lock);
10151 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10153 struct btrfs_space_info *space_info;
10154 struct btrfs_super_block *disk_super;
10160 disk_super = fs_info->super_copy;
10161 if (!btrfs_super_root(disk_super))
10164 features = btrfs_super_incompat_flags(disk_super);
10165 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10168 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10169 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10174 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10175 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10177 flags = BTRFS_BLOCK_GROUP_METADATA;
10178 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10182 flags = BTRFS_BLOCK_GROUP_DATA;
10183 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10189 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10191 return unpin_extent_range(root, start, end, false);
10195 * It used to be that old block groups would be left around forever.
10196 * Iterating over them would be enough to trim unused space. Since we
10197 * now automatically remove them, we also need to iterate over unallocated
10200 * We don't want a transaction for this since the discard may take a
10201 * substantial amount of time. We don't require that a transaction be
10202 * running, but we do need to take a running transaction into account
10203 * to ensure that we're not discarding chunks that were released in
10204 * the current transaction.
10206 * Holding the chunks lock will prevent other threads from allocating
10207 * or releasing chunks, but it won't prevent a running transaction
10208 * from committing and releasing the memory that the pending chunks
10209 * list head uses. For that, we need to take a reference to the
10212 static int btrfs_trim_free_extents(struct btrfs_device *device,
10213 u64 minlen, u64 *trimmed)
10215 u64 start = 0, len = 0;
10220 /* Not writeable = nothing to do. */
10221 if (!device->writeable)
10224 /* No free space = nothing to do. */
10225 if (device->total_bytes <= device->bytes_used)
10231 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
10232 struct btrfs_transaction *trans;
10235 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10239 down_read(&fs_info->commit_root_sem);
10241 spin_lock(&fs_info->trans_lock);
10242 trans = fs_info->running_transaction;
10244 atomic_inc(&trans->use_count);
10245 spin_unlock(&fs_info->trans_lock);
10247 ret = find_free_dev_extent_start(trans, device, minlen, start,
10250 btrfs_put_transaction(trans);
10253 up_read(&fs_info->commit_root_sem);
10254 mutex_unlock(&fs_info->chunk_mutex);
10255 if (ret == -ENOSPC)
10260 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10261 up_read(&fs_info->commit_root_sem);
10262 mutex_unlock(&fs_info->chunk_mutex);
10270 if (fatal_signal_pending(current)) {
10271 ret = -ERESTARTSYS;
10281 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10283 struct btrfs_fs_info *fs_info = root->fs_info;
10284 struct btrfs_block_group_cache *cache = NULL;
10285 struct btrfs_device *device;
10286 struct list_head *devices;
10291 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10295 * try to trim all FS space, our block group may start from non-zero.
10297 if (range->len == total_bytes)
10298 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10300 cache = btrfs_lookup_block_group(fs_info, range->start);
10303 if (cache->key.objectid >= (range->start + range->len)) {
10304 btrfs_put_block_group(cache);
10308 start = max(range->start, cache->key.objectid);
10309 end = min(range->start + range->len,
10310 cache->key.objectid + cache->key.offset);
10312 if (end - start >= range->minlen) {
10313 if (!block_group_cache_done(cache)) {
10314 ret = cache_block_group(cache, 0);
10316 btrfs_put_block_group(cache);
10319 ret = wait_block_group_cache_done(cache);
10321 btrfs_put_block_group(cache);
10325 ret = btrfs_trim_block_group(cache,
10331 trimmed += group_trimmed;
10333 btrfs_put_block_group(cache);
10338 cache = next_block_group(fs_info->tree_root, cache);
10341 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
10342 devices = &root->fs_info->fs_devices->alloc_list;
10343 list_for_each_entry(device, devices, dev_alloc_list) {
10344 ret = btrfs_trim_free_extents(device, range->minlen,
10349 trimmed += group_trimmed;
10351 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
10353 range->len = trimmed;
10358 * btrfs_{start,end}_write_no_snapshoting() are similar to
10359 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10360 * data into the page cache through nocow before the subvolume is snapshoted,
10361 * but flush the data into disk after the snapshot creation, or to prevent
10362 * operations while snapshoting is ongoing and that cause the snapshot to be
10363 * inconsistent (writes followed by expanding truncates for example).
10365 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10367 percpu_counter_dec(&root->subv_writers->counter);
10369 * Make sure counter is updated before we wake up
10373 if (waitqueue_active(&root->subv_writers->wait))
10374 wake_up(&root->subv_writers->wait);
10377 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10379 if (atomic_read(&root->will_be_snapshoted))
10382 percpu_counter_inc(&root->subv_writers->counter);
10384 * Make sure counter is updated before we check for snapshot creation.
10387 if (atomic_read(&root->will_be_snapshoted)) {
10388 btrfs_end_write_no_snapshoting(root);