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"
39 #undef SCRAMBLE_DELAYED_REFS
42 * control flags for do_chunk_alloc's force field
43 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
44 * if we really need one.
46 * CHUNK_ALLOC_LIMITED means to only try and allocate one
47 * if we have very few chunks already allocated. This is
48 * used as part of the clustering code to help make sure
49 * we have a good pool of storage to cluster in, without
50 * filling the FS with empty chunks
52 * CHUNK_ALLOC_FORCE means it must try to allocate one
56 CHUNK_ALLOC_NO_FORCE = 0,
57 CHUNK_ALLOC_LIMITED = 1,
58 CHUNK_ALLOC_FORCE = 2,
62 * Control how reservations are dealt with.
64 * RESERVE_FREE - freeing a reservation.
65 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
67 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
68 * bytes_may_use as the ENOSPC accounting is done elsewhere
73 RESERVE_ALLOC_NO_ACCOUNT = 2,
76 static int update_block_group(struct btrfs_root *root,
77 u64 bytenr, u64 num_bytes, int alloc);
78 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
79 struct btrfs_root *root,
80 u64 bytenr, u64 num_bytes, u64 parent,
81 u64 root_objectid, u64 owner_objectid,
82 u64 owner_offset, int refs_to_drop,
83 struct btrfs_delayed_extent_op *extra_op);
84 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
85 struct extent_buffer *leaf,
86 struct btrfs_extent_item *ei);
87 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
88 struct btrfs_root *root,
89 u64 parent, u64 root_objectid,
90 u64 flags, u64 owner, u64 offset,
91 struct btrfs_key *ins, int ref_mod);
92 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
93 struct btrfs_root *root,
94 u64 parent, u64 root_objectid,
95 u64 flags, struct btrfs_disk_key *key,
96 int level, struct btrfs_key *ins);
97 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
98 struct btrfs_root *extent_root, u64 flags,
100 static int find_next_key(struct btrfs_path *path, int level,
101 struct btrfs_key *key);
102 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
103 int dump_block_groups);
104 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
105 u64 num_bytes, int reserve);
106 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
108 int btrfs_pin_extent(struct btrfs_root *root,
109 u64 bytenr, u64 num_bytes, int reserved);
112 block_group_cache_done(struct btrfs_block_group_cache *cache)
115 return cache->cached == BTRFS_CACHE_FINISHED ||
116 cache->cached == BTRFS_CACHE_ERROR;
119 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
121 return (cache->flags & bits) == bits;
124 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
126 atomic_inc(&cache->count);
129 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
131 if (atomic_dec_and_test(&cache->count)) {
132 WARN_ON(cache->pinned > 0);
133 WARN_ON(cache->reserved > 0);
134 kfree(cache->free_space_ctl);
140 * this adds the block group to the fs_info rb tree for the block group
143 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
144 struct btrfs_block_group_cache *block_group)
147 struct rb_node *parent = NULL;
148 struct btrfs_block_group_cache *cache;
150 spin_lock(&info->block_group_cache_lock);
151 p = &info->block_group_cache_tree.rb_node;
155 cache = rb_entry(parent, struct btrfs_block_group_cache,
157 if (block_group->key.objectid < cache->key.objectid) {
159 } else if (block_group->key.objectid > cache->key.objectid) {
162 spin_unlock(&info->block_group_cache_lock);
167 rb_link_node(&block_group->cache_node, parent, p);
168 rb_insert_color(&block_group->cache_node,
169 &info->block_group_cache_tree);
171 if (info->first_logical_byte > block_group->key.objectid)
172 info->first_logical_byte = block_group->key.objectid;
174 spin_unlock(&info->block_group_cache_lock);
180 * This will return the block group at or after bytenr if contains is 0, else
181 * it will return the block group that contains the bytenr
183 static struct btrfs_block_group_cache *
184 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
187 struct btrfs_block_group_cache *cache, *ret = NULL;
191 spin_lock(&info->block_group_cache_lock);
192 n = info->block_group_cache_tree.rb_node;
195 cache = rb_entry(n, struct btrfs_block_group_cache,
197 end = cache->key.objectid + cache->key.offset - 1;
198 start = cache->key.objectid;
200 if (bytenr < start) {
201 if (!contains && (!ret || start < ret->key.objectid))
204 } else if (bytenr > start) {
205 if (contains && bytenr <= end) {
216 btrfs_get_block_group(ret);
217 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
218 info->first_logical_byte = ret->key.objectid;
220 spin_unlock(&info->block_group_cache_lock);
225 static int add_excluded_extent(struct btrfs_root *root,
226 u64 start, u64 num_bytes)
228 u64 end = start + num_bytes - 1;
229 set_extent_bits(&root->fs_info->freed_extents[0],
230 start, end, EXTENT_UPTODATE, GFP_NOFS);
231 set_extent_bits(&root->fs_info->freed_extents[1],
232 start, end, EXTENT_UPTODATE, GFP_NOFS);
236 static void free_excluded_extents(struct btrfs_root *root,
237 struct btrfs_block_group_cache *cache)
241 start = cache->key.objectid;
242 end = start + cache->key.offset - 1;
244 clear_extent_bits(&root->fs_info->freed_extents[0],
245 start, end, EXTENT_UPTODATE, GFP_NOFS);
246 clear_extent_bits(&root->fs_info->freed_extents[1],
247 start, end, EXTENT_UPTODATE, GFP_NOFS);
250 static int exclude_super_stripes(struct btrfs_root *root,
251 struct btrfs_block_group_cache *cache)
258 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
259 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
260 cache->bytes_super += stripe_len;
261 ret = add_excluded_extent(root, cache->key.objectid,
267 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
268 bytenr = btrfs_sb_offset(i);
269 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
270 cache->key.objectid, bytenr,
271 0, &logical, &nr, &stripe_len);
278 if (logical[nr] > cache->key.objectid +
282 if (logical[nr] + stripe_len <= cache->key.objectid)
286 if (start < cache->key.objectid) {
287 start = cache->key.objectid;
288 len = (logical[nr] + stripe_len) - start;
290 len = min_t(u64, stripe_len,
291 cache->key.objectid +
292 cache->key.offset - start);
295 cache->bytes_super += len;
296 ret = add_excluded_extent(root, start, len);
308 static struct btrfs_caching_control *
309 get_caching_control(struct btrfs_block_group_cache *cache)
311 struct btrfs_caching_control *ctl;
313 spin_lock(&cache->lock);
314 if (cache->cached != BTRFS_CACHE_STARTED) {
315 spin_unlock(&cache->lock);
319 /* We're loading it the fast way, so we don't have a caching_ctl. */
320 if (!cache->caching_ctl) {
321 spin_unlock(&cache->lock);
325 ctl = cache->caching_ctl;
326 atomic_inc(&ctl->count);
327 spin_unlock(&cache->lock);
331 static void put_caching_control(struct btrfs_caching_control *ctl)
333 if (atomic_dec_and_test(&ctl->count))
338 * this is only called by cache_block_group, since we could have freed extents
339 * we need to check the pinned_extents for any extents that can't be used yet
340 * since their free space will be released as soon as the transaction commits.
342 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
343 struct btrfs_fs_info *info, u64 start, u64 end)
345 u64 extent_start, extent_end, size, total_added = 0;
348 while (start < end) {
349 ret = find_first_extent_bit(info->pinned_extents, start,
350 &extent_start, &extent_end,
351 EXTENT_DIRTY | EXTENT_UPTODATE,
356 if (extent_start <= start) {
357 start = extent_end + 1;
358 } else if (extent_start > start && extent_start < end) {
359 size = extent_start - start;
361 ret = btrfs_add_free_space(block_group, start,
363 BUG_ON(ret); /* -ENOMEM or logic error */
364 start = extent_end + 1;
373 ret = btrfs_add_free_space(block_group, start, size);
374 BUG_ON(ret); /* -ENOMEM or logic error */
380 static noinline void caching_thread(struct btrfs_work *work)
382 struct btrfs_block_group_cache *block_group;
383 struct btrfs_fs_info *fs_info;
384 struct btrfs_caching_control *caching_ctl;
385 struct btrfs_root *extent_root;
386 struct btrfs_path *path;
387 struct extent_buffer *leaf;
388 struct btrfs_key key;
394 caching_ctl = container_of(work, struct btrfs_caching_control, work);
395 block_group = caching_ctl->block_group;
396 fs_info = block_group->fs_info;
397 extent_root = fs_info->extent_root;
399 path = btrfs_alloc_path();
403 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
406 * We don't want to deadlock with somebody trying to allocate a new
407 * extent for the extent root while also trying to search the extent
408 * root to add free space. So we skip locking and search the commit
409 * root, since its read-only
411 path->skip_locking = 1;
412 path->search_commit_root = 1;
417 key.type = BTRFS_EXTENT_ITEM_KEY;
419 mutex_lock(&caching_ctl->mutex);
420 /* need to make sure the commit_root doesn't disappear */
421 down_read(&fs_info->commit_root_sem);
424 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
428 leaf = path->nodes[0];
429 nritems = btrfs_header_nritems(leaf);
432 if (btrfs_fs_closing(fs_info) > 1) {
437 if (path->slots[0] < nritems) {
438 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
440 ret = find_next_key(path, 0, &key);
444 if (need_resched() ||
445 rwsem_is_contended(&fs_info->commit_root_sem)) {
446 caching_ctl->progress = last;
447 btrfs_release_path(path);
448 up_read(&fs_info->commit_root_sem);
449 mutex_unlock(&caching_ctl->mutex);
454 ret = btrfs_next_leaf(extent_root, path);
459 leaf = path->nodes[0];
460 nritems = btrfs_header_nritems(leaf);
464 if (key.objectid < last) {
467 key.type = BTRFS_EXTENT_ITEM_KEY;
469 caching_ctl->progress = last;
470 btrfs_release_path(path);
474 if (key.objectid < block_group->key.objectid) {
479 if (key.objectid >= block_group->key.objectid +
480 block_group->key.offset)
483 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
484 key.type == BTRFS_METADATA_ITEM_KEY) {
485 total_found += add_new_free_space(block_group,
488 if (key.type == BTRFS_METADATA_ITEM_KEY)
489 last = key.objectid +
490 fs_info->tree_root->leafsize;
492 last = key.objectid + key.offset;
494 if (total_found > (1024 * 1024 * 2)) {
496 wake_up(&caching_ctl->wait);
503 total_found += add_new_free_space(block_group, fs_info, last,
504 block_group->key.objectid +
505 block_group->key.offset);
506 caching_ctl->progress = (u64)-1;
508 spin_lock(&block_group->lock);
509 block_group->caching_ctl = NULL;
510 block_group->cached = BTRFS_CACHE_FINISHED;
511 spin_unlock(&block_group->lock);
514 btrfs_free_path(path);
515 up_read(&fs_info->commit_root_sem);
517 free_excluded_extents(extent_root, block_group);
519 mutex_unlock(&caching_ctl->mutex);
522 spin_lock(&block_group->lock);
523 block_group->caching_ctl = NULL;
524 block_group->cached = BTRFS_CACHE_ERROR;
525 spin_unlock(&block_group->lock);
527 wake_up(&caching_ctl->wait);
529 put_caching_control(caching_ctl);
530 btrfs_put_block_group(block_group);
533 static int cache_block_group(struct btrfs_block_group_cache *cache,
537 struct btrfs_fs_info *fs_info = cache->fs_info;
538 struct btrfs_caching_control *caching_ctl;
541 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
545 INIT_LIST_HEAD(&caching_ctl->list);
546 mutex_init(&caching_ctl->mutex);
547 init_waitqueue_head(&caching_ctl->wait);
548 caching_ctl->block_group = cache;
549 caching_ctl->progress = cache->key.objectid;
550 atomic_set(&caching_ctl->count, 1);
551 btrfs_init_work(&caching_ctl->work, caching_thread, NULL, NULL);
553 spin_lock(&cache->lock);
555 * This should be a rare occasion, but this could happen I think in the
556 * case where one thread starts to load the space cache info, and then
557 * some other thread starts a transaction commit which tries to do an
558 * allocation while the other thread is still loading the space cache
559 * info. The previous loop should have kept us from choosing this block
560 * group, but if we've moved to the state where we will wait on caching
561 * block groups we need to first check if we're doing a fast load here,
562 * so we can wait for it to finish, otherwise we could end up allocating
563 * from a block group who's cache gets evicted for one reason or
566 while (cache->cached == BTRFS_CACHE_FAST) {
567 struct btrfs_caching_control *ctl;
569 ctl = cache->caching_ctl;
570 atomic_inc(&ctl->count);
571 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
572 spin_unlock(&cache->lock);
576 finish_wait(&ctl->wait, &wait);
577 put_caching_control(ctl);
578 spin_lock(&cache->lock);
581 if (cache->cached != BTRFS_CACHE_NO) {
582 spin_unlock(&cache->lock);
586 WARN_ON(cache->caching_ctl);
587 cache->caching_ctl = caching_ctl;
588 cache->cached = BTRFS_CACHE_FAST;
589 spin_unlock(&cache->lock);
591 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
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;
600 if (load_cache_only) {
601 cache->caching_ctl = NULL;
602 cache->cached = BTRFS_CACHE_NO;
604 cache->cached = BTRFS_CACHE_STARTED;
607 spin_unlock(&cache->lock);
608 wake_up(&caching_ctl->wait);
610 put_caching_control(caching_ctl);
611 free_excluded_extents(fs_info->extent_root, cache);
616 * We are not going to do the fast caching, set cached to the
617 * appropriate value and wakeup any waiters.
619 spin_lock(&cache->lock);
620 if (load_cache_only) {
621 cache->caching_ctl = NULL;
622 cache->cached = BTRFS_CACHE_NO;
624 cache->cached = BTRFS_CACHE_STARTED;
626 spin_unlock(&cache->lock);
627 wake_up(&caching_ctl->wait);
630 if (load_cache_only) {
631 put_caching_control(caching_ctl);
635 down_write(&fs_info->commit_root_sem);
636 atomic_inc(&caching_ctl->count);
637 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
638 up_write(&fs_info->commit_root_sem);
640 btrfs_get_block_group(cache);
642 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
648 * return the block group that starts at or after bytenr
650 static struct btrfs_block_group_cache *
651 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
653 struct btrfs_block_group_cache *cache;
655 cache = block_group_cache_tree_search(info, bytenr, 0);
661 * return the block group that contains the given bytenr
663 struct btrfs_block_group_cache *btrfs_lookup_block_group(
664 struct btrfs_fs_info *info,
667 struct btrfs_block_group_cache *cache;
669 cache = block_group_cache_tree_search(info, bytenr, 1);
674 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
677 struct list_head *head = &info->space_info;
678 struct btrfs_space_info *found;
680 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
683 list_for_each_entry_rcu(found, head, list) {
684 if (found->flags & flags) {
694 * after adding space to the filesystem, we need to clear the full flags
695 * on all the space infos.
697 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
699 struct list_head *head = &info->space_info;
700 struct btrfs_space_info *found;
703 list_for_each_entry_rcu(found, head, list)
708 /* simple helper to search for an existing extent at a given offset */
709 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
712 struct btrfs_key key;
713 struct btrfs_path *path;
715 path = btrfs_alloc_path();
719 key.objectid = start;
721 key.type = BTRFS_EXTENT_ITEM_KEY;
722 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
725 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
726 if (key.objectid == start &&
727 key.type == BTRFS_METADATA_ITEM_KEY)
730 btrfs_free_path(path);
735 * helper function to lookup reference count and flags of a tree block.
737 * the head node for delayed ref is used to store the sum of all the
738 * reference count modifications queued up in the rbtree. the head
739 * node may also store the extent flags to set. This way you can check
740 * to see what the reference count and extent flags would be if all of
741 * the delayed refs are not processed.
743 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
744 struct btrfs_root *root, u64 bytenr,
745 u64 offset, int metadata, u64 *refs, u64 *flags)
747 struct btrfs_delayed_ref_head *head;
748 struct btrfs_delayed_ref_root *delayed_refs;
749 struct btrfs_path *path;
750 struct btrfs_extent_item *ei;
751 struct extent_buffer *leaf;
752 struct btrfs_key key;
759 * If we don't have skinny metadata, don't bother doing anything
762 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
763 offset = root->leafsize;
767 path = btrfs_alloc_path();
772 path->skip_locking = 1;
773 path->search_commit_root = 1;
777 key.objectid = bytenr;
780 key.type = BTRFS_METADATA_ITEM_KEY;
782 key.type = BTRFS_EXTENT_ITEM_KEY;
785 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
790 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
791 if (path->slots[0]) {
793 btrfs_item_key_to_cpu(path->nodes[0], &key,
795 if (key.objectid == bytenr &&
796 key.type == BTRFS_EXTENT_ITEM_KEY &&
797 key.offset == root->leafsize)
801 key.objectid = bytenr;
802 key.type = BTRFS_EXTENT_ITEM_KEY;
803 key.offset = root->leafsize;
804 btrfs_release_path(path);
810 leaf = path->nodes[0];
811 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
812 if (item_size >= sizeof(*ei)) {
813 ei = btrfs_item_ptr(leaf, path->slots[0],
814 struct btrfs_extent_item);
815 num_refs = btrfs_extent_refs(leaf, ei);
816 extent_flags = btrfs_extent_flags(leaf, ei);
818 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
819 struct btrfs_extent_item_v0 *ei0;
820 BUG_ON(item_size != sizeof(*ei0));
821 ei0 = btrfs_item_ptr(leaf, path->slots[0],
822 struct btrfs_extent_item_v0);
823 num_refs = btrfs_extent_refs_v0(leaf, ei0);
824 /* FIXME: this isn't correct for data */
825 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
830 BUG_ON(num_refs == 0);
840 delayed_refs = &trans->transaction->delayed_refs;
841 spin_lock(&delayed_refs->lock);
842 head = btrfs_find_delayed_ref_head(trans, bytenr);
844 if (!mutex_trylock(&head->mutex)) {
845 atomic_inc(&head->node.refs);
846 spin_unlock(&delayed_refs->lock);
848 btrfs_release_path(path);
851 * Mutex was contended, block until it's released and try
854 mutex_lock(&head->mutex);
855 mutex_unlock(&head->mutex);
856 btrfs_put_delayed_ref(&head->node);
859 spin_lock(&head->lock);
860 if (head->extent_op && head->extent_op->update_flags)
861 extent_flags |= head->extent_op->flags_to_set;
863 BUG_ON(num_refs == 0);
865 num_refs += head->node.ref_mod;
866 spin_unlock(&head->lock);
867 mutex_unlock(&head->mutex);
869 spin_unlock(&delayed_refs->lock);
871 WARN_ON(num_refs == 0);
875 *flags = extent_flags;
877 btrfs_free_path(path);
882 * Back reference rules. Back refs have three main goals:
884 * 1) differentiate between all holders of references to an extent so that
885 * when a reference is dropped we can make sure it was a valid reference
886 * before freeing the extent.
888 * 2) Provide enough information to quickly find the holders of an extent
889 * if we notice a given block is corrupted or bad.
891 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
892 * maintenance. This is actually the same as #2, but with a slightly
893 * different use case.
895 * There are two kinds of back refs. The implicit back refs is optimized
896 * for pointers in non-shared tree blocks. For a given pointer in a block,
897 * back refs of this kind provide information about the block's owner tree
898 * and the pointer's key. These information allow us to find the block by
899 * b-tree searching. The full back refs is for pointers in tree blocks not
900 * referenced by their owner trees. The location of tree block is recorded
901 * in the back refs. Actually the full back refs is generic, and can be
902 * used in all cases the implicit back refs is used. The major shortcoming
903 * of the full back refs is its overhead. Every time a tree block gets
904 * COWed, we have to update back refs entry for all pointers in it.
906 * For a newly allocated tree block, we use implicit back refs for
907 * pointers in it. This means most tree related operations only involve
908 * implicit back refs. For a tree block created in old transaction, the
909 * only way to drop a reference to it is COW it. So we can detect the
910 * event that tree block loses its owner tree's reference and do the
911 * back refs conversion.
913 * When a tree block is COW'd through a tree, there are four cases:
915 * The reference count of the block is one and the tree is the block's
916 * owner tree. Nothing to do in this case.
918 * The reference count of the block is one and the tree is not the
919 * block's owner tree. In this case, full back refs is used for pointers
920 * in the block. Remove these full back refs, add implicit back refs for
921 * every pointers in the new block.
923 * The reference count of the block is greater than one and the tree is
924 * the block's owner tree. In this case, implicit back refs is used for
925 * pointers in the block. Add full back refs for every pointers in the
926 * block, increase lower level extents' reference counts. The original
927 * implicit back refs are entailed to the new block.
929 * The reference count of the block is greater than one and the tree is
930 * not the block's owner tree. Add implicit back refs for every pointer in
931 * the new block, increase lower level extents' reference count.
933 * Back Reference Key composing:
935 * The key objectid corresponds to the first byte in the extent,
936 * The key type is used to differentiate between types of back refs.
937 * There are different meanings of the key offset for different types
940 * File extents can be referenced by:
942 * - multiple snapshots, subvolumes, or different generations in one subvol
943 * - different files inside a single subvolume
944 * - different offsets inside a file (bookend extents in file.c)
946 * The extent ref structure for the implicit back refs has fields for:
948 * - Objectid of the subvolume root
949 * - objectid of the file holding the reference
950 * - original offset in the file
951 * - how many bookend extents
953 * The key offset for the implicit back refs is hash of the first
956 * The extent ref structure for the full back refs has field for:
958 * - number of pointers in the tree leaf
960 * The key offset for the implicit back refs is the first byte of
963 * When a file extent is allocated, The implicit back refs is used.
964 * the fields are filled in:
966 * (root_key.objectid, inode objectid, offset in file, 1)
968 * When a file extent is removed file truncation, we find the
969 * corresponding implicit back refs and check the following fields:
971 * (btrfs_header_owner(leaf), inode objectid, offset in file)
973 * Btree extents can be referenced by:
975 * - Different subvolumes
977 * Both the implicit back refs and the full back refs for tree blocks
978 * only consist of key. The key offset for the implicit back refs is
979 * objectid of block's owner tree. The key offset for the full back refs
980 * is the first byte of parent block.
982 * When implicit back refs is used, information about the lowest key and
983 * level of the tree block are required. These information are stored in
984 * tree block info structure.
987 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
988 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
989 struct btrfs_root *root,
990 struct btrfs_path *path,
991 u64 owner, u32 extra_size)
993 struct btrfs_extent_item *item;
994 struct btrfs_extent_item_v0 *ei0;
995 struct btrfs_extent_ref_v0 *ref0;
996 struct btrfs_tree_block_info *bi;
997 struct extent_buffer *leaf;
998 struct btrfs_key key;
999 struct btrfs_key found_key;
1000 u32 new_size = sizeof(*item);
1004 leaf = path->nodes[0];
1005 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1007 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1008 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1009 struct btrfs_extent_item_v0);
1010 refs = btrfs_extent_refs_v0(leaf, ei0);
1012 if (owner == (u64)-1) {
1014 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1015 ret = btrfs_next_leaf(root, path);
1018 BUG_ON(ret > 0); /* Corruption */
1019 leaf = path->nodes[0];
1021 btrfs_item_key_to_cpu(leaf, &found_key,
1023 BUG_ON(key.objectid != found_key.objectid);
1024 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1028 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1029 struct btrfs_extent_ref_v0);
1030 owner = btrfs_ref_objectid_v0(leaf, ref0);
1034 btrfs_release_path(path);
1036 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1037 new_size += sizeof(*bi);
1039 new_size -= sizeof(*ei0);
1040 ret = btrfs_search_slot(trans, root, &key, path,
1041 new_size + extra_size, 1);
1044 BUG_ON(ret); /* Corruption */
1046 btrfs_extend_item(root, path, new_size);
1048 leaf = path->nodes[0];
1049 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1050 btrfs_set_extent_refs(leaf, item, refs);
1051 /* FIXME: get real generation */
1052 btrfs_set_extent_generation(leaf, item, 0);
1053 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1054 btrfs_set_extent_flags(leaf, item,
1055 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1056 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1057 bi = (struct btrfs_tree_block_info *)(item + 1);
1058 /* FIXME: get first key of the block */
1059 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1060 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1062 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1064 btrfs_mark_buffer_dirty(leaf);
1069 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1071 u32 high_crc = ~(u32)0;
1072 u32 low_crc = ~(u32)0;
1075 lenum = cpu_to_le64(root_objectid);
1076 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1077 lenum = cpu_to_le64(owner);
1078 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1079 lenum = cpu_to_le64(offset);
1080 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1082 return ((u64)high_crc << 31) ^ (u64)low_crc;
1085 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1086 struct btrfs_extent_data_ref *ref)
1088 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1089 btrfs_extent_data_ref_objectid(leaf, ref),
1090 btrfs_extent_data_ref_offset(leaf, ref));
1093 static int match_extent_data_ref(struct extent_buffer *leaf,
1094 struct btrfs_extent_data_ref *ref,
1095 u64 root_objectid, u64 owner, u64 offset)
1097 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1098 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1099 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1104 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1105 struct btrfs_root *root,
1106 struct btrfs_path *path,
1107 u64 bytenr, u64 parent,
1109 u64 owner, u64 offset)
1111 struct btrfs_key key;
1112 struct btrfs_extent_data_ref *ref;
1113 struct extent_buffer *leaf;
1119 key.objectid = bytenr;
1121 key.type = BTRFS_SHARED_DATA_REF_KEY;
1122 key.offset = parent;
1124 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1125 key.offset = hash_extent_data_ref(root_objectid,
1130 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1139 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1140 key.type = BTRFS_EXTENT_REF_V0_KEY;
1141 btrfs_release_path(path);
1142 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1153 leaf = path->nodes[0];
1154 nritems = btrfs_header_nritems(leaf);
1156 if (path->slots[0] >= nritems) {
1157 ret = btrfs_next_leaf(root, path);
1163 leaf = path->nodes[0];
1164 nritems = btrfs_header_nritems(leaf);
1168 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1169 if (key.objectid != bytenr ||
1170 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1173 ref = btrfs_item_ptr(leaf, path->slots[0],
1174 struct btrfs_extent_data_ref);
1176 if (match_extent_data_ref(leaf, ref, root_objectid,
1179 btrfs_release_path(path);
1191 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1192 struct btrfs_root *root,
1193 struct btrfs_path *path,
1194 u64 bytenr, u64 parent,
1195 u64 root_objectid, u64 owner,
1196 u64 offset, int refs_to_add)
1198 struct btrfs_key key;
1199 struct extent_buffer *leaf;
1204 key.objectid = bytenr;
1206 key.type = BTRFS_SHARED_DATA_REF_KEY;
1207 key.offset = parent;
1208 size = sizeof(struct btrfs_shared_data_ref);
1210 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1211 key.offset = hash_extent_data_ref(root_objectid,
1213 size = sizeof(struct btrfs_extent_data_ref);
1216 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1217 if (ret && ret != -EEXIST)
1220 leaf = path->nodes[0];
1222 struct btrfs_shared_data_ref *ref;
1223 ref = btrfs_item_ptr(leaf, path->slots[0],
1224 struct btrfs_shared_data_ref);
1226 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1228 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1229 num_refs += refs_to_add;
1230 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1233 struct btrfs_extent_data_ref *ref;
1234 while (ret == -EEXIST) {
1235 ref = btrfs_item_ptr(leaf, path->slots[0],
1236 struct btrfs_extent_data_ref);
1237 if (match_extent_data_ref(leaf, ref, root_objectid,
1240 btrfs_release_path(path);
1242 ret = btrfs_insert_empty_item(trans, root, path, &key,
1244 if (ret && ret != -EEXIST)
1247 leaf = path->nodes[0];
1249 ref = btrfs_item_ptr(leaf, path->slots[0],
1250 struct btrfs_extent_data_ref);
1252 btrfs_set_extent_data_ref_root(leaf, ref,
1254 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1255 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1256 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1258 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1259 num_refs += refs_to_add;
1260 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1263 btrfs_mark_buffer_dirty(leaf);
1266 btrfs_release_path(path);
1270 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1271 struct btrfs_root *root,
1272 struct btrfs_path *path,
1275 struct btrfs_key key;
1276 struct btrfs_extent_data_ref *ref1 = NULL;
1277 struct btrfs_shared_data_ref *ref2 = NULL;
1278 struct extent_buffer *leaf;
1282 leaf = path->nodes[0];
1283 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1285 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1286 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1287 struct btrfs_extent_data_ref);
1288 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1289 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1290 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1291 struct btrfs_shared_data_ref);
1292 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1293 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1294 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1295 struct btrfs_extent_ref_v0 *ref0;
1296 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1297 struct btrfs_extent_ref_v0);
1298 num_refs = btrfs_ref_count_v0(leaf, ref0);
1304 BUG_ON(num_refs < refs_to_drop);
1305 num_refs -= refs_to_drop;
1307 if (num_refs == 0) {
1308 ret = btrfs_del_item(trans, root, path);
1310 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1311 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1312 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1313 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1314 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1316 struct btrfs_extent_ref_v0 *ref0;
1317 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1318 struct btrfs_extent_ref_v0);
1319 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1322 btrfs_mark_buffer_dirty(leaf);
1327 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1328 struct btrfs_path *path,
1329 struct btrfs_extent_inline_ref *iref)
1331 struct btrfs_key key;
1332 struct extent_buffer *leaf;
1333 struct btrfs_extent_data_ref *ref1;
1334 struct btrfs_shared_data_ref *ref2;
1337 leaf = path->nodes[0];
1338 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1340 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1341 BTRFS_EXTENT_DATA_REF_KEY) {
1342 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1343 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1345 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1346 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1348 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1349 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1350 struct btrfs_extent_data_ref);
1351 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1352 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1353 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1354 struct btrfs_shared_data_ref);
1355 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1356 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1357 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1358 struct btrfs_extent_ref_v0 *ref0;
1359 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1360 struct btrfs_extent_ref_v0);
1361 num_refs = btrfs_ref_count_v0(leaf, ref0);
1369 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1370 struct btrfs_root *root,
1371 struct btrfs_path *path,
1372 u64 bytenr, u64 parent,
1375 struct btrfs_key key;
1378 key.objectid = bytenr;
1380 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1381 key.offset = parent;
1383 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1384 key.offset = root_objectid;
1387 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1390 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1391 if (ret == -ENOENT && parent) {
1392 btrfs_release_path(path);
1393 key.type = BTRFS_EXTENT_REF_V0_KEY;
1394 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1402 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1403 struct btrfs_root *root,
1404 struct btrfs_path *path,
1405 u64 bytenr, u64 parent,
1408 struct btrfs_key key;
1411 key.objectid = bytenr;
1413 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1414 key.offset = parent;
1416 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1417 key.offset = root_objectid;
1420 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1421 btrfs_release_path(path);
1425 static inline int extent_ref_type(u64 parent, u64 owner)
1428 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1430 type = BTRFS_SHARED_BLOCK_REF_KEY;
1432 type = BTRFS_TREE_BLOCK_REF_KEY;
1435 type = BTRFS_SHARED_DATA_REF_KEY;
1437 type = BTRFS_EXTENT_DATA_REF_KEY;
1442 static int find_next_key(struct btrfs_path *path, int level,
1443 struct btrfs_key *key)
1446 for (; level < BTRFS_MAX_LEVEL; level++) {
1447 if (!path->nodes[level])
1449 if (path->slots[level] + 1 >=
1450 btrfs_header_nritems(path->nodes[level]))
1453 btrfs_item_key_to_cpu(path->nodes[level], key,
1454 path->slots[level] + 1);
1456 btrfs_node_key_to_cpu(path->nodes[level], key,
1457 path->slots[level] + 1);
1464 * look for inline back ref. if back ref is found, *ref_ret is set
1465 * to the address of inline back ref, and 0 is returned.
1467 * if back ref isn't found, *ref_ret is set to the address where it
1468 * should be inserted, and -ENOENT is returned.
1470 * if insert is true and there are too many inline back refs, the path
1471 * points to the extent item, and -EAGAIN is returned.
1473 * NOTE: inline back refs are ordered in the same way that back ref
1474 * items in the tree are ordered.
1476 static noinline_for_stack
1477 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1478 struct btrfs_root *root,
1479 struct btrfs_path *path,
1480 struct btrfs_extent_inline_ref **ref_ret,
1481 u64 bytenr, u64 num_bytes,
1482 u64 parent, u64 root_objectid,
1483 u64 owner, u64 offset, int insert)
1485 struct btrfs_key key;
1486 struct extent_buffer *leaf;
1487 struct btrfs_extent_item *ei;
1488 struct btrfs_extent_inline_ref *iref;
1498 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1501 key.objectid = bytenr;
1502 key.type = BTRFS_EXTENT_ITEM_KEY;
1503 key.offset = num_bytes;
1505 want = extent_ref_type(parent, owner);
1507 extra_size = btrfs_extent_inline_ref_size(want);
1508 path->keep_locks = 1;
1513 * Owner is our parent level, so we can just add one to get the level
1514 * for the block we are interested in.
1516 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1517 key.type = BTRFS_METADATA_ITEM_KEY;
1522 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1529 * We may be a newly converted file system which still has the old fat
1530 * extent entries for metadata, so try and see if we have one of those.
1532 if (ret > 0 && skinny_metadata) {
1533 skinny_metadata = false;
1534 if (path->slots[0]) {
1536 btrfs_item_key_to_cpu(path->nodes[0], &key,
1538 if (key.objectid == bytenr &&
1539 key.type == BTRFS_EXTENT_ITEM_KEY &&
1540 key.offset == num_bytes)
1544 key.objectid = bytenr;
1545 key.type = BTRFS_EXTENT_ITEM_KEY;
1546 key.offset = num_bytes;
1547 btrfs_release_path(path);
1552 if (ret && !insert) {
1555 } else if (WARN_ON(ret)) {
1560 leaf = path->nodes[0];
1561 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1562 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1563 if (item_size < sizeof(*ei)) {
1568 ret = convert_extent_item_v0(trans, root, path, owner,
1574 leaf = path->nodes[0];
1575 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1578 BUG_ON(item_size < sizeof(*ei));
1580 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1581 flags = btrfs_extent_flags(leaf, ei);
1583 ptr = (unsigned long)(ei + 1);
1584 end = (unsigned long)ei + item_size;
1586 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1587 ptr += sizeof(struct btrfs_tree_block_info);
1597 iref = (struct btrfs_extent_inline_ref *)ptr;
1598 type = btrfs_extent_inline_ref_type(leaf, iref);
1602 ptr += btrfs_extent_inline_ref_size(type);
1606 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1607 struct btrfs_extent_data_ref *dref;
1608 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1609 if (match_extent_data_ref(leaf, dref, root_objectid,
1614 if (hash_extent_data_ref_item(leaf, dref) <
1615 hash_extent_data_ref(root_objectid, owner, offset))
1619 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1621 if (parent == ref_offset) {
1625 if (ref_offset < parent)
1628 if (root_objectid == ref_offset) {
1632 if (ref_offset < root_objectid)
1636 ptr += btrfs_extent_inline_ref_size(type);
1638 if (err == -ENOENT && insert) {
1639 if (item_size + extra_size >=
1640 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1645 * To add new inline back ref, we have to make sure
1646 * there is no corresponding back ref item.
1647 * For simplicity, we just do not add new inline back
1648 * ref if there is any kind of item for this block
1650 if (find_next_key(path, 0, &key) == 0 &&
1651 key.objectid == bytenr &&
1652 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1657 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1660 path->keep_locks = 0;
1661 btrfs_unlock_up_safe(path, 1);
1667 * helper to add new inline back ref
1669 static noinline_for_stack
1670 void setup_inline_extent_backref(struct btrfs_root *root,
1671 struct btrfs_path *path,
1672 struct btrfs_extent_inline_ref *iref,
1673 u64 parent, u64 root_objectid,
1674 u64 owner, u64 offset, int refs_to_add,
1675 struct btrfs_delayed_extent_op *extent_op)
1677 struct extent_buffer *leaf;
1678 struct btrfs_extent_item *ei;
1681 unsigned long item_offset;
1686 leaf = path->nodes[0];
1687 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1688 item_offset = (unsigned long)iref - (unsigned long)ei;
1690 type = extent_ref_type(parent, owner);
1691 size = btrfs_extent_inline_ref_size(type);
1693 btrfs_extend_item(root, path, size);
1695 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1696 refs = btrfs_extent_refs(leaf, ei);
1697 refs += refs_to_add;
1698 btrfs_set_extent_refs(leaf, ei, refs);
1700 __run_delayed_extent_op(extent_op, leaf, ei);
1702 ptr = (unsigned long)ei + item_offset;
1703 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1704 if (ptr < end - size)
1705 memmove_extent_buffer(leaf, ptr + size, ptr,
1708 iref = (struct btrfs_extent_inline_ref *)ptr;
1709 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1710 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1711 struct btrfs_extent_data_ref *dref;
1712 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1713 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1714 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1715 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1716 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1717 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1718 struct btrfs_shared_data_ref *sref;
1719 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1720 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1721 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1722 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1723 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1725 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1727 btrfs_mark_buffer_dirty(leaf);
1730 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1731 struct btrfs_root *root,
1732 struct btrfs_path *path,
1733 struct btrfs_extent_inline_ref **ref_ret,
1734 u64 bytenr, u64 num_bytes, u64 parent,
1735 u64 root_objectid, u64 owner, u64 offset)
1739 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1740 bytenr, num_bytes, parent,
1741 root_objectid, owner, offset, 0);
1745 btrfs_release_path(path);
1748 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1749 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1752 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1753 root_objectid, owner, offset);
1759 * helper to update/remove inline back ref
1761 static noinline_for_stack
1762 void update_inline_extent_backref(struct btrfs_root *root,
1763 struct btrfs_path *path,
1764 struct btrfs_extent_inline_ref *iref,
1766 struct btrfs_delayed_extent_op *extent_op)
1768 struct extent_buffer *leaf;
1769 struct btrfs_extent_item *ei;
1770 struct btrfs_extent_data_ref *dref = NULL;
1771 struct btrfs_shared_data_ref *sref = NULL;
1779 leaf = path->nodes[0];
1780 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1781 refs = btrfs_extent_refs(leaf, ei);
1782 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1783 refs += refs_to_mod;
1784 btrfs_set_extent_refs(leaf, ei, refs);
1786 __run_delayed_extent_op(extent_op, leaf, ei);
1788 type = btrfs_extent_inline_ref_type(leaf, iref);
1790 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1791 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1792 refs = btrfs_extent_data_ref_count(leaf, dref);
1793 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1794 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1795 refs = btrfs_shared_data_ref_count(leaf, sref);
1798 BUG_ON(refs_to_mod != -1);
1801 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1802 refs += refs_to_mod;
1805 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1806 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1808 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1810 size = btrfs_extent_inline_ref_size(type);
1811 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1812 ptr = (unsigned long)iref;
1813 end = (unsigned long)ei + item_size;
1814 if (ptr + size < end)
1815 memmove_extent_buffer(leaf, ptr, ptr + size,
1818 btrfs_truncate_item(root, path, item_size, 1);
1820 btrfs_mark_buffer_dirty(leaf);
1823 static noinline_for_stack
1824 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1825 struct btrfs_root *root,
1826 struct btrfs_path *path,
1827 u64 bytenr, u64 num_bytes, u64 parent,
1828 u64 root_objectid, u64 owner,
1829 u64 offset, int refs_to_add,
1830 struct btrfs_delayed_extent_op *extent_op)
1832 struct btrfs_extent_inline_ref *iref;
1835 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1836 bytenr, num_bytes, parent,
1837 root_objectid, owner, offset, 1);
1839 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1840 update_inline_extent_backref(root, path, iref,
1841 refs_to_add, extent_op);
1842 } else if (ret == -ENOENT) {
1843 setup_inline_extent_backref(root, path, iref, parent,
1844 root_objectid, owner, offset,
1845 refs_to_add, extent_op);
1851 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1852 struct btrfs_root *root,
1853 struct btrfs_path *path,
1854 u64 bytenr, u64 parent, u64 root_objectid,
1855 u64 owner, u64 offset, int refs_to_add)
1858 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1859 BUG_ON(refs_to_add != 1);
1860 ret = insert_tree_block_ref(trans, root, path, bytenr,
1861 parent, root_objectid);
1863 ret = insert_extent_data_ref(trans, root, path, bytenr,
1864 parent, root_objectid,
1865 owner, offset, refs_to_add);
1870 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1871 struct btrfs_root *root,
1872 struct btrfs_path *path,
1873 struct btrfs_extent_inline_ref *iref,
1874 int refs_to_drop, int is_data)
1878 BUG_ON(!is_data && refs_to_drop != 1);
1880 update_inline_extent_backref(root, path, iref,
1881 -refs_to_drop, NULL);
1882 } else if (is_data) {
1883 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1885 ret = btrfs_del_item(trans, root, path);
1890 static int btrfs_issue_discard(struct block_device *bdev,
1893 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1896 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1897 u64 num_bytes, u64 *actual_bytes)
1900 u64 discarded_bytes = 0;
1901 struct btrfs_bio *bbio = NULL;
1904 /* Tell the block device(s) that the sectors can be discarded */
1905 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1906 bytenr, &num_bytes, &bbio, 0);
1907 /* Error condition is -ENOMEM */
1909 struct btrfs_bio_stripe *stripe = bbio->stripes;
1913 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1914 if (!stripe->dev->can_discard)
1917 ret = btrfs_issue_discard(stripe->dev->bdev,
1921 discarded_bytes += stripe->length;
1922 else if (ret != -EOPNOTSUPP)
1923 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1926 * Just in case we get back EOPNOTSUPP for some reason,
1927 * just ignore the return value so we don't screw up
1928 * people calling discard_extent.
1936 *actual_bytes = discarded_bytes;
1939 if (ret == -EOPNOTSUPP)
1944 /* Can return -ENOMEM */
1945 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1946 struct btrfs_root *root,
1947 u64 bytenr, u64 num_bytes, u64 parent,
1948 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1951 struct btrfs_fs_info *fs_info = root->fs_info;
1953 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1954 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1956 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1957 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1959 parent, root_objectid, (int)owner,
1960 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1962 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1964 parent, root_objectid, owner, offset,
1965 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1970 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1971 struct btrfs_root *root,
1972 u64 bytenr, u64 num_bytes,
1973 u64 parent, u64 root_objectid,
1974 u64 owner, u64 offset, int refs_to_add,
1975 struct btrfs_delayed_extent_op *extent_op)
1977 struct btrfs_path *path;
1978 struct extent_buffer *leaf;
1979 struct btrfs_extent_item *item;
1983 path = btrfs_alloc_path();
1988 path->leave_spinning = 1;
1989 /* this will setup the path even if it fails to insert the back ref */
1990 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1991 path, bytenr, num_bytes, parent,
1992 root_objectid, owner, offset,
1993 refs_to_add, extent_op);
1997 leaf = path->nodes[0];
1998 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1999 refs = btrfs_extent_refs(leaf, item);
2000 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2002 __run_delayed_extent_op(extent_op, leaf, item);
2004 btrfs_mark_buffer_dirty(leaf);
2005 btrfs_release_path(path);
2008 path->leave_spinning = 1;
2010 /* now insert the actual backref */
2011 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2012 path, bytenr, parent, root_objectid,
2013 owner, offset, refs_to_add);
2015 btrfs_abort_transaction(trans, root, ret);
2017 btrfs_free_path(path);
2021 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2022 struct btrfs_root *root,
2023 struct btrfs_delayed_ref_node *node,
2024 struct btrfs_delayed_extent_op *extent_op,
2025 int insert_reserved)
2028 struct btrfs_delayed_data_ref *ref;
2029 struct btrfs_key ins;
2034 ins.objectid = node->bytenr;
2035 ins.offset = node->num_bytes;
2036 ins.type = BTRFS_EXTENT_ITEM_KEY;
2038 ref = btrfs_delayed_node_to_data_ref(node);
2039 trace_run_delayed_data_ref(node, ref, node->action);
2041 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2042 parent = ref->parent;
2044 ref_root = ref->root;
2046 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2048 flags |= extent_op->flags_to_set;
2049 ret = alloc_reserved_file_extent(trans, root,
2050 parent, ref_root, flags,
2051 ref->objectid, ref->offset,
2052 &ins, node->ref_mod);
2053 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2054 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2055 node->num_bytes, parent,
2056 ref_root, ref->objectid,
2057 ref->offset, node->ref_mod,
2059 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2060 ret = __btrfs_free_extent(trans, root, node->bytenr,
2061 node->num_bytes, parent,
2062 ref_root, ref->objectid,
2063 ref->offset, node->ref_mod,
2071 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2072 struct extent_buffer *leaf,
2073 struct btrfs_extent_item *ei)
2075 u64 flags = btrfs_extent_flags(leaf, ei);
2076 if (extent_op->update_flags) {
2077 flags |= extent_op->flags_to_set;
2078 btrfs_set_extent_flags(leaf, ei, flags);
2081 if (extent_op->update_key) {
2082 struct btrfs_tree_block_info *bi;
2083 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2084 bi = (struct btrfs_tree_block_info *)(ei + 1);
2085 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2089 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2090 struct btrfs_root *root,
2091 struct btrfs_delayed_ref_node *node,
2092 struct btrfs_delayed_extent_op *extent_op)
2094 struct btrfs_key key;
2095 struct btrfs_path *path;
2096 struct btrfs_extent_item *ei;
2097 struct extent_buffer *leaf;
2101 int metadata = !extent_op->is_data;
2106 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2109 path = btrfs_alloc_path();
2113 key.objectid = node->bytenr;
2116 key.type = BTRFS_METADATA_ITEM_KEY;
2117 key.offset = extent_op->level;
2119 key.type = BTRFS_EXTENT_ITEM_KEY;
2120 key.offset = node->num_bytes;
2125 path->leave_spinning = 1;
2126 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2134 if (path->slots[0] > 0) {
2136 btrfs_item_key_to_cpu(path->nodes[0], &key,
2138 if (key.objectid == node->bytenr &&
2139 key.type == BTRFS_EXTENT_ITEM_KEY &&
2140 key.offset == node->num_bytes)
2144 btrfs_release_path(path);
2147 key.objectid = node->bytenr;
2148 key.offset = node->num_bytes;
2149 key.type = BTRFS_EXTENT_ITEM_KEY;
2158 leaf = path->nodes[0];
2159 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2160 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2161 if (item_size < sizeof(*ei)) {
2162 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2168 leaf = path->nodes[0];
2169 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2172 BUG_ON(item_size < sizeof(*ei));
2173 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2174 __run_delayed_extent_op(extent_op, leaf, ei);
2176 btrfs_mark_buffer_dirty(leaf);
2178 btrfs_free_path(path);
2182 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2183 struct btrfs_root *root,
2184 struct btrfs_delayed_ref_node *node,
2185 struct btrfs_delayed_extent_op *extent_op,
2186 int insert_reserved)
2189 struct btrfs_delayed_tree_ref *ref;
2190 struct btrfs_key ins;
2193 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2196 ref = btrfs_delayed_node_to_tree_ref(node);
2197 trace_run_delayed_tree_ref(node, ref, node->action);
2199 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2200 parent = ref->parent;
2202 ref_root = ref->root;
2204 ins.objectid = node->bytenr;
2205 if (skinny_metadata) {
2206 ins.offset = ref->level;
2207 ins.type = BTRFS_METADATA_ITEM_KEY;
2209 ins.offset = node->num_bytes;
2210 ins.type = BTRFS_EXTENT_ITEM_KEY;
2213 BUG_ON(node->ref_mod != 1);
2214 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2215 BUG_ON(!extent_op || !extent_op->update_flags);
2216 ret = alloc_reserved_tree_block(trans, root,
2218 extent_op->flags_to_set,
2221 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2222 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2223 node->num_bytes, parent, ref_root,
2224 ref->level, 0, 1, extent_op);
2225 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2226 ret = __btrfs_free_extent(trans, root, node->bytenr,
2227 node->num_bytes, parent, ref_root,
2228 ref->level, 0, 1, extent_op);
2235 /* helper function to actually process a single delayed ref entry */
2236 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2237 struct btrfs_root *root,
2238 struct btrfs_delayed_ref_node *node,
2239 struct btrfs_delayed_extent_op *extent_op,
2240 int insert_reserved)
2244 if (trans->aborted) {
2245 if (insert_reserved)
2246 btrfs_pin_extent(root, node->bytenr,
2247 node->num_bytes, 1);
2251 if (btrfs_delayed_ref_is_head(node)) {
2252 struct btrfs_delayed_ref_head *head;
2254 * we've hit the end of the chain and we were supposed
2255 * to insert this extent into the tree. But, it got
2256 * deleted before we ever needed to insert it, so all
2257 * we have to do is clean up the accounting
2260 head = btrfs_delayed_node_to_head(node);
2261 trace_run_delayed_ref_head(node, head, node->action);
2263 if (insert_reserved) {
2264 btrfs_pin_extent(root, node->bytenr,
2265 node->num_bytes, 1);
2266 if (head->is_data) {
2267 ret = btrfs_del_csums(trans, root,
2275 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2276 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2277 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2279 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2280 node->type == BTRFS_SHARED_DATA_REF_KEY)
2281 ret = run_delayed_data_ref(trans, root, node, extent_op,
2288 static noinline struct btrfs_delayed_ref_node *
2289 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2291 struct rb_node *node;
2292 struct btrfs_delayed_ref_node *ref, *last = NULL;;
2295 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2296 * this prevents ref count from going down to zero when
2297 * there still are pending delayed ref.
2299 node = rb_first(&head->ref_root);
2301 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2303 if (ref->action == BTRFS_ADD_DELAYED_REF)
2305 else if (last == NULL)
2307 node = rb_next(node);
2313 * Returns 0 on success or if called with an already aborted transaction.
2314 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2316 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2317 struct btrfs_root *root,
2320 struct btrfs_delayed_ref_root *delayed_refs;
2321 struct btrfs_delayed_ref_node *ref;
2322 struct btrfs_delayed_ref_head *locked_ref = NULL;
2323 struct btrfs_delayed_extent_op *extent_op;
2324 struct btrfs_fs_info *fs_info = root->fs_info;
2325 ktime_t start = ktime_get();
2327 unsigned long count = 0;
2328 unsigned long actual_count = 0;
2329 int must_insert_reserved = 0;
2331 delayed_refs = &trans->transaction->delayed_refs;
2337 spin_lock(&delayed_refs->lock);
2338 locked_ref = btrfs_select_ref_head(trans);
2340 spin_unlock(&delayed_refs->lock);
2344 /* grab the lock that says we are going to process
2345 * all the refs for this head */
2346 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2347 spin_unlock(&delayed_refs->lock);
2349 * we may have dropped the spin lock to get the head
2350 * mutex lock, and that might have given someone else
2351 * time to free the head. If that's true, it has been
2352 * removed from our list and we can move on.
2354 if (ret == -EAGAIN) {
2362 * We need to try and merge add/drops of the same ref since we
2363 * can run into issues with relocate dropping the implicit ref
2364 * and then it being added back again before the drop can
2365 * finish. If we merged anything we need to re-loop so we can
2368 spin_lock(&locked_ref->lock);
2369 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2373 * locked_ref is the head node, so we have to go one
2374 * node back for any delayed ref updates
2376 ref = select_delayed_ref(locked_ref);
2378 if (ref && ref->seq &&
2379 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2380 spin_unlock(&locked_ref->lock);
2381 btrfs_delayed_ref_unlock(locked_ref);
2382 spin_lock(&delayed_refs->lock);
2383 locked_ref->processing = 0;
2384 delayed_refs->num_heads_ready++;
2385 spin_unlock(&delayed_refs->lock);
2393 * record the must insert reserved flag before we
2394 * drop the spin lock.
2396 must_insert_reserved = locked_ref->must_insert_reserved;
2397 locked_ref->must_insert_reserved = 0;
2399 extent_op = locked_ref->extent_op;
2400 locked_ref->extent_op = NULL;
2405 /* All delayed refs have been processed, Go ahead
2406 * and send the head node to run_one_delayed_ref,
2407 * so that any accounting fixes can happen
2409 ref = &locked_ref->node;
2411 if (extent_op && must_insert_reserved) {
2412 btrfs_free_delayed_extent_op(extent_op);
2417 spin_unlock(&locked_ref->lock);
2418 ret = run_delayed_extent_op(trans, root,
2420 btrfs_free_delayed_extent_op(extent_op);
2424 * Need to reset must_insert_reserved if
2425 * there was an error so the abort stuff
2426 * can cleanup the reserved space
2429 if (must_insert_reserved)
2430 locked_ref->must_insert_reserved = 1;
2431 locked_ref->processing = 0;
2432 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2433 btrfs_delayed_ref_unlock(locked_ref);
2440 * Need to drop our head ref lock and re-aqcuire the
2441 * delayed ref lock and then re-check to make sure
2444 spin_unlock(&locked_ref->lock);
2445 spin_lock(&delayed_refs->lock);
2446 spin_lock(&locked_ref->lock);
2447 if (rb_first(&locked_ref->ref_root) ||
2448 locked_ref->extent_op) {
2449 spin_unlock(&locked_ref->lock);
2450 spin_unlock(&delayed_refs->lock);
2454 delayed_refs->num_heads--;
2455 rb_erase(&locked_ref->href_node,
2456 &delayed_refs->href_root);
2457 spin_unlock(&delayed_refs->lock);
2461 rb_erase(&ref->rb_node, &locked_ref->ref_root);
2463 atomic_dec(&delayed_refs->num_entries);
2465 if (!btrfs_delayed_ref_is_head(ref)) {
2467 * when we play the delayed ref, also correct the
2470 switch (ref->action) {
2471 case BTRFS_ADD_DELAYED_REF:
2472 case BTRFS_ADD_DELAYED_EXTENT:
2473 locked_ref->node.ref_mod -= ref->ref_mod;
2475 case BTRFS_DROP_DELAYED_REF:
2476 locked_ref->node.ref_mod += ref->ref_mod;
2482 spin_unlock(&locked_ref->lock);
2484 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2485 must_insert_reserved);
2487 btrfs_free_delayed_extent_op(extent_op);
2489 locked_ref->processing = 0;
2490 btrfs_delayed_ref_unlock(locked_ref);
2491 btrfs_put_delayed_ref(ref);
2492 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2497 * If this node is a head, that means all the refs in this head
2498 * have been dealt with, and we will pick the next head to deal
2499 * with, so we must unlock the head and drop it from the cluster
2500 * list before we release it.
2502 if (btrfs_delayed_ref_is_head(ref)) {
2503 btrfs_delayed_ref_unlock(locked_ref);
2506 btrfs_put_delayed_ref(ref);
2512 * We don't want to include ref heads since we can have empty ref heads
2513 * and those will drastically skew our runtime down since we just do
2514 * accounting, no actual extent tree updates.
2516 if (actual_count > 0) {
2517 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2521 * We weigh the current average higher than our current runtime
2522 * to avoid large swings in the average.
2524 spin_lock(&delayed_refs->lock);
2525 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2526 avg = div64_u64(avg, 4);
2527 fs_info->avg_delayed_ref_runtime = avg;
2528 spin_unlock(&delayed_refs->lock);
2533 #ifdef SCRAMBLE_DELAYED_REFS
2535 * Normally delayed refs get processed in ascending bytenr order. This
2536 * correlates in most cases to the order added. To expose dependencies on this
2537 * order, we start to process the tree in the middle instead of the beginning
2539 static u64 find_middle(struct rb_root *root)
2541 struct rb_node *n = root->rb_node;
2542 struct btrfs_delayed_ref_node *entry;
2545 u64 first = 0, last = 0;
2549 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2550 first = entry->bytenr;
2554 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2555 last = entry->bytenr;
2560 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2561 WARN_ON(!entry->in_tree);
2563 middle = entry->bytenr;
2576 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2577 struct btrfs_fs_info *fs_info)
2579 struct qgroup_update *qgroup_update;
2582 if (list_empty(&trans->qgroup_ref_list) !=
2583 !trans->delayed_ref_elem.seq) {
2584 /* list without seq or seq without list */
2586 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2587 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2588 (u32)(trans->delayed_ref_elem.seq >> 32),
2589 (u32)trans->delayed_ref_elem.seq);
2593 if (!trans->delayed_ref_elem.seq)
2596 while (!list_empty(&trans->qgroup_ref_list)) {
2597 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2598 struct qgroup_update, list);
2599 list_del(&qgroup_update->list);
2601 ret = btrfs_qgroup_account_ref(
2602 trans, fs_info, qgroup_update->node,
2603 qgroup_update->extent_op);
2604 kfree(qgroup_update);
2607 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2612 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2616 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2617 sizeof(struct btrfs_extent_inline_ref));
2618 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2619 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2622 * We don't ever fill up leaves all the way so multiply by 2 just to be
2623 * closer to what we're really going to want to ouse.
2625 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2628 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2629 struct btrfs_root *root)
2631 struct btrfs_block_rsv *global_rsv;
2632 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2636 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2637 num_heads = heads_to_leaves(root, num_heads);
2639 num_bytes += (num_heads - 1) * root->leafsize;
2641 global_rsv = &root->fs_info->global_block_rsv;
2644 * If we can't allocate any more chunks lets make sure we have _lots_ of
2645 * wiggle room since running delayed refs can create more delayed refs.
2647 if (global_rsv->space_info->full)
2650 spin_lock(&global_rsv->lock);
2651 if (global_rsv->reserved <= num_bytes)
2653 spin_unlock(&global_rsv->lock);
2657 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2658 struct btrfs_root *root)
2660 struct btrfs_fs_info *fs_info = root->fs_info;
2662 atomic_read(&trans->transaction->delayed_refs.num_entries);
2666 avg_runtime = fs_info->avg_delayed_ref_runtime;
2667 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2670 return btrfs_check_space_for_delayed_refs(trans, root);
2674 * this starts processing the delayed reference count updates and
2675 * extent insertions we have queued up so far. count can be
2676 * 0, which means to process everything in the tree at the start
2677 * of the run (but not newly added entries), or it can be some target
2678 * number you'd like to process.
2680 * Returns 0 on success or if called with an aborted transaction
2681 * Returns <0 on error and aborts the transaction
2683 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2684 struct btrfs_root *root, unsigned long count)
2686 struct rb_node *node;
2687 struct btrfs_delayed_ref_root *delayed_refs;
2688 struct btrfs_delayed_ref_head *head;
2690 int run_all = count == (unsigned long)-1;
2693 /* We'll clean this up in btrfs_cleanup_transaction */
2697 if (root == root->fs_info->extent_root)
2698 root = root->fs_info->tree_root;
2700 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2702 delayed_refs = &trans->transaction->delayed_refs;
2704 count = atomic_read(&delayed_refs->num_entries) * 2;
2709 #ifdef SCRAMBLE_DELAYED_REFS
2710 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2712 ret = __btrfs_run_delayed_refs(trans, root, count);
2714 btrfs_abort_transaction(trans, root, ret);
2719 if (!list_empty(&trans->new_bgs))
2720 btrfs_create_pending_block_groups(trans, root);
2722 spin_lock(&delayed_refs->lock);
2723 node = rb_first(&delayed_refs->href_root);
2725 spin_unlock(&delayed_refs->lock);
2728 count = (unsigned long)-1;
2731 head = rb_entry(node, struct btrfs_delayed_ref_head,
2733 if (btrfs_delayed_ref_is_head(&head->node)) {
2734 struct btrfs_delayed_ref_node *ref;
2737 atomic_inc(&ref->refs);
2739 spin_unlock(&delayed_refs->lock);
2741 * Mutex was contended, block until it's
2742 * released and try again
2744 mutex_lock(&head->mutex);
2745 mutex_unlock(&head->mutex);
2747 btrfs_put_delayed_ref(ref);
2753 node = rb_next(node);
2755 spin_unlock(&delayed_refs->lock);
2760 assert_qgroups_uptodate(trans);
2764 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2765 struct btrfs_root *root,
2766 u64 bytenr, u64 num_bytes, u64 flags,
2767 int level, int is_data)
2769 struct btrfs_delayed_extent_op *extent_op;
2772 extent_op = btrfs_alloc_delayed_extent_op();
2776 extent_op->flags_to_set = flags;
2777 extent_op->update_flags = 1;
2778 extent_op->update_key = 0;
2779 extent_op->is_data = is_data ? 1 : 0;
2780 extent_op->level = level;
2782 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2783 num_bytes, extent_op);
2785 btrfs_free_delayed_extent_op(extent_op);
2789 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2790 struct btrfs_root *root,
2791 struct btrfs_path *path,
2792 u64 objectid, u64 offset, u64 bytenr)
2794 struct btrfs_delayed_ref_head *head;
2795 struct btrfs_delayed_ref_node *ref;
2796 struct btrfs_delayed_data_ref *data_ref;
2797 struct btrfs_delayed_ref_root *delayed_refs;
2798 struct rb_node *node;
2801 delayed_refs = &trans->transaction->delayed_refs;
2802 spin_lock(&delayed_refs->lock);
2803 head = btrfs_find_delayed_ref_head(trans, bytenr);
2805 spin_unlock(&delayed_refs->lock);
2809 if (!mutex_trylock(&head->mutex)) {
2810 atomic_inc(&head->node.refs);
2811 spin_unlock(&delayed_refs->lock);
2813 btrfs_release_path(path);
2816 * Mutex was contended, block until it's released and let
2819 mutex_lock(&head->mutex);
2820 mutex_unlock(&head->mutex);
2821 btrfs_put_delayed_ref(&head->node);
2824 spin_unlock(&delayed_refs->lock);
2826 spin_lock(&head->lock);
2827 node = rb_first(&head->ref_root);
2829 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2830 node = rb_next(node);
2832 /* If it's a shared ref we know a cross reference exists */
2833 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2838 data_ref = btrfs_delayed_node_to_data_ref(ref);
2841 * If our ref doesn't match the one we're currently looking at
2842 * then we have a cross reference.
2844 if (data_ref->root != root->root_key.objectid ||
2845 data_ref->objectid != objectid ||
2846 data_ref->offset != offset) {
2851 spin_unlock(&head->lock);
2852 mutex_unlock(&head->mutex);
2856 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2857 struct btrfs_root *root,
2858 struct btrfs_path *path,
2859 u64 objectid, u64 offset, u64 bytenr)
2861 struct btrfs_root *extent_root = root->fs_info->extent_root;
2862 struct extent_buffer *leaf;
2863 struct btrfs_extent_data_ref *ref;
2864 struct btrfs_extent_inline_ref *iref;
2865 struct btrfs_extent_item *ei;
2866 struct btrfs_key key;
2870 key.objectid = bytenr;
2871 key.offset = (u64)-1;
2872 key.type = BTRFS_EXTENT_ITEM_KEY;
2874 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2877 BUG_ON(ret == 0); /* Corruption */
2880 if (path->slots[0] == 0)
2884 leaf = path->nodes[0];
2885 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2887 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2891 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2892 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2893 if (item_size < sizeof(*ei)) {
2894 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2898 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2900 if (item_size != sizeof(*ei) +
2901 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2904 if (btrfs_extent_generation(leaf, ei) <=
2905 btrfs_root_last_snapshot(&root->root_item))
2908 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2909 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2910 BTRFS_EXTENT_DATA_REF_KEY)
2913 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2914 if (btrfs_extent_refs(leaf, ei) !=
2915 btrfs_extent_data_ref_count(leaf, ref) ||
2916 btrfs_extent_data_ref_root(leaf, ref) !=
2917 root->root_key.objectid ||
2918 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2919 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2927 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2928 struct btrfs_root *root,
2929 u64 objectid, u64 offset, u64 bytenr)
2931 struct btrfs_path *path;
2935 path = btrfs_alloc_path();
2940 ret = check_committed_ref(trans, root, path, objectid,
2942 if (ret && ret != -ENOENT)
2945 ret2 = check_delayed_ref(trans, root, path, objectid,
2947 } while (ret2 == -EAGAIN);
2949 if (ret2 && ret2 != -ENOENT) {
2954 if (ret != -ENOENT || ret2 != -ENOENT)
2957 btrfs_free_path(path);
2958 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2963 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2964 struct btrfs_root *root,
2965 struct extent_buffer *buf,
2966 int full_backref, int inc, int for_cow)
2973 struct btrfs_key key;
2974 struct btrfs_file_extent_item *fi;
2978 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2979 u64, u64, u64, u64, u64, u64, int);
2981 ref_root = btrfs_header_owner(buf);
2982 nritems = btrfs_header_nritems(buf);
2983 level = btrfs_header_level(buf);
2985 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
2989 process_func = btrfs_inc_extent_ref;
2991 process_func = btrfs_free_extent;
2994 parent = buf->start;
2998 for (i = 0; i < nritems; i++) {
3000 btrfs_item_key_to_cpu(buf, &key, i);
3001 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3003 fi = btrfs_item_ptr(buf, i,
3004 struct btrfs_file_extent_item);
3005 if (btrfs_file_extent_type(buf, fi) ==
3006 BTRFS_FILE_EXTENT_INLINE)
3008 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3012 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3013 key.offset -= btrfs_file_extent_offset(buf, fi);
3014 ret = process_func(trans, root, bytenr, num_bytes,
3015 parent, ref_root, key.objectid,
3016 key.offset, for_cow);
3020 bytenr = btrfs_node_blockptr(buf, i);
3021 num_bytes = btrfs_level_size(root, level - 1);
3022 ret = process_func(trans, root, bytenr, num_bytes,
3023 parent, ref_root, level - 1, 0,
3034 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3035 struct extent_buffer *buf, int full_backref, int for_cow)
3037 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3040 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3041 struct extent_buffer *buf, int full_backref, int for_cow)
3043 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3046 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3047 struct btrfs_root *root,
3048 struct btrfs_path *path,
3049 struct btrfs_block_group_cache *cache)
3052 struct btrfs_root *extent_root = root->fs_info->extent_root;
3054 struct extent_buffer *leaf;
3056 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3059 BUG_ON(ret); /* Corruption */
3061 leaf = path->nodes[0];
3062 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3063 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3064 btrfs_mark_buffer_dirty(leaf);
3065 btrfs_release_path(path);
3068 btrfs_abort_transaction(trans, root, ret);
3075 static struct btrfs_block_group_cache *
3076 next_block_group(struct btrfs_root *root,
3077 struct btrfs_block_group_cache *cache)
3079 struct rb_node *node;
3080 spin_lock(&root->fs_info->block_group_cache_lock);
3081 node = rb_next(&cache->cache_node);
3082 btrfs_put_block_group(cache);
3084 cache = rb_entry(node, struct btrfs_block_group_cache,
3086 btrfs_get_block_group(cache);
3089 spin_unlock(&root->fs_info->block_group_cache_lock);
3093 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3094 struct btrfs_trans_handle *trans,
3095 struct btrfs_path *path)
3097 struct btrfs_root *root = block_group->fs_info->tree_root;
3098 struct inode *inode = NULL;
3100 int dcs = BTRFS_DC_ERROR;
3106 * If this block group is smaller than 100 megs don't bother caching the
3109 if (block_group->key.offset < (100 * 1024 * 1024)) {
3110 spin_lock(&block_group->lock);
3111 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3112 spin_unlock(&block_group->lock);
3117 inode = lookup_free_space_inode(root, block_group, path);
3118 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3119 ret = PTR_ERR(inode);
3120 btrfs_release_path(path);
3124 if (IS_ERR(inode)) {
3128 if (block_group->ro)
3131 ret = create_free_space_inode(root, trans, block_group, path);
3137 /* We've already setup this transaction, go ahead and exit */
3138 if (block_group->cache_generation == trans->transid &&
3139 i_size_read(inode)) {
3140 dcs = BTRFS_DC_SETUP;
3145 * We want to set the generation to 0, that way if anything goes wrong
3146 * from here on out we know not to trust this cache when we load up next
3149 BTRFS_I(inode)->generation = 0;
3150 ret = btrfs_update_inode(trans, root, inode);
3153 if (i_size_read(inode) > 0) {
3154 ret = btrfs_check_trunc_cache_free_space(root,
3155 &root->fs_info->global_block_rsv);
3159 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3164 spin_lock(&block_group->lock);
3165 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3166 !btrfs_test_opt(root, SPACE_CACHE)) {
3168 * don't bother trying to write stuff out _if_
3169 * a) we're not cached,
3170 * b) we're with nospace_cache mount option.
3172 dcs = BTRFS_DC_WRITTEN;
3173 spin_unlock(&block_group->lock);
3176 spin_unlock(&block_group->lock);
3179 * Try to preallocate enough space based on how big the block group is.
3180 * Keep in mind this has to include any pinned space which could end up
3181 * taking up quite a bit since it's not folded into the other space
3184 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3189 num_pages *= PAGE_CACHE_SIZE;
3191 ret = btrfs_check_data_free_space(inode, num_pages);
3195 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3196 num_pages, num_pages,
3199 dcs = BTRFS_DC_SETUP;
3200 btrfs_free_reserved_data_space(inode, num_pages);
3205 btrfs_release_path(path);
3207 spin_lock(&block_group->lock);
3208 if (!ret && dcs == BTRFS_DC_SETUP)
3209 block_group->cache_generation = trans->transid;
3210 block_group->disk_cache_state = dcs;
3211 spin_unlock(&block_group->lock);
3216 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3217 struct btrfs_root *root)
3219 struct btrfs_block_group_cache *cache;
3221 struct btrfs_path *path;
3224 path = btrfs_alloc_path();
3230 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3232 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3234 cache = next_block_group(root, cache);
3242 err = cache_save_setup(cache, trans, path);
3243 last = cache->key.objectid + cache->key.offset;
3244 btrfs_put_block_group(cache);
3249 err = btrfs_run_delayed_refs(trans, root,
3251 if (err) /* File system offline */
3255 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3257 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3258 btrfs_put_block_group(cache);
3264 cache = next_block_group(root, cache);
3273 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3274 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3276 last = cache->key.objectid + cache->key.offset;
3278 err = write_one_cache_group(trans, root, path, cache);
3279 btrfs_put_block_group(cache);
3280 if (err) /* File system offline */
3286 * I don't think this is needed since we're just marking our
3287 * preallocated extent as written, but just in case it can't
3291 err = btrfs_run_delayed_refs(trans, root,
3293 if (err) /* File system offline */
3297 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3300 * Really this shouldn't happen, but it could if we
3301 * couldn't write the entire preallocated extent and
3302 * splitting the extent resulted in a new block.
3305 btrfs_put_block_group(cache);
3308 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3310 cache = next_block_group(root, cache);
3319 err = btrfs_write_out_cache(root, trans, cache, path);
3322 * If we didn't have an error then the cache state is still
3323 * NEED_WRITE, so we can set it to WRITTEN.
3325 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3326 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3327 last = cache->key.objectid + cache->key.offset;
3328 btrfs_put_block_group(cache);
3332 btrfs_free_path(path);
3336 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3338 struct btrfs_block_group_cache *block_group;
3341 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3342 if (!block_group || block_group->ro)
3345 btrfs_put_block_group(block_group);
3349 static const char *alloc_name(u64 flags)
3352 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3354 case BTRFS_BLOCK_GROUP_METADATA:
3356 case BTRFS_BLOCK_GROUP_DATA:
3358 case BTRFS_BLOCK_GROUP_SYSTEM:
3362 return "invalid-combination";
3366 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3367 u64 total_bytes, u64 bytes_used,
3368 struct btrfs_space_info **space_info)
3370 struct btrfs_space_info *found;
3375 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3376 BTRFS_BLOCK_GROUP_RAID10))
3381 found = __find_space_info(info, flags);
3383 spin_lock(&found->lock);
3384 found->total_bytes += total_bytes;
3385 found->disk_total += total_bytes * factor;
3386 found->bytes_used += bytes_used;
3387 found->disk_used += bytes_used * factor;
3389 spin_unlock(&found->lock);
3390 *space_info = found;
3393 found = kzalloc(sizeof(*found), GFP_NOFS);
3397 ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3403 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
3404 INIT_LIST_HEAD(&found->block_groups[i]);
3405 kobject_init(&found->block_group_kobjs[i], &btrfs_raid_ktype);
3407 init_rwsem(&found->groups_sem);
3408 spin_lock_init(&found->lock);
3409 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3410 found->total_bytes = total_bytes;
3411 found->disk_total = total_bytes * factor;
3412 found->bytes_used = bytes_used;
3413 found->disk_used = bytes_used * factor;
3414 found->bytes_pinned = 0;
3415 found->bytes_reserved = 0;
3416 found->bytes_readonly = 0;
3417 found->bytes_may_use = 0;
3419 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3420 found->chunk_alloc = 0;
3422 init_waitqueue_head(&found->wait);
3424 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3425 info->space_info_kobj, "%s",
3426 alloc_name(found->flags));
3432 *space_info = found;
3433 list_add_rcu(&found->list, &info->space_info);
3434 if (flags & BTRFS_BLOCK_GROUP_DATA)
3435 info->data_sinfo = found;
3440 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3442 u64 extra_flags = chunk_to_extended(flags) &
3443 BTRFS_EXTENDED_PROFILE_MASK;
3445 write_seqlock(&fs_info->profiles_lock);
3446 if (flags & BTRFS_BLOCK_GROUP_DATA)
3447 fs_info->avail_data_alloc_bits |= extra_flags;
3448 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3449 fs_info->avail_metadata_alloc_bits |= extra_flags;
3450 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3451 fs_info->avail_system_alloc_bits |= extra_flags;
3452 write_sequnlock(&fs_info->profiles_lock);
3456 * returns target flags in extended format or 0 if restripe for this
3457 * chunk_type is not in progress
3459 * should be called with either volume_mutex or balance_lock held
3461 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3463 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3469 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3470 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3471 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3472 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3473 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3474 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3475 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3476 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3477 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3484 * @flags: available profiles in extended format (see ctree.h)
3486 * Returns reduced profile in chunk format. If profile changing is in
3487 * progress (either running or paused) picks the target profile (if it's
3488 * already available), otherwise falls back to plain reducing.
3490 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3493 * we add in the count of missing devices because we want
3494 * to make sure that any RAID levels on a degraded FS
3495 * continue to be honored.
3497 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3498 root->fs_info->fs_devices->missing_devices;
3503 * see if restripe for this chunk_type is in progress, if so
3504 * try to reduce to the target profile
3506 spin_lock(&root->fs_info->balance_lock);
3507 target = get_restripe_target(root->fs_info, flags);
3509 /* pick target profile only if it's already available */
3510 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3511 spin_unlock(&root->fs_info->balance_lock);
3512 return extended_to_chunk(target);
3515 spin_unlock(&root->fs_info->balance_lock);
3517 /* First, mask out the RAID levels which aren't possible */
3518 if (num_devices == 1)
3519 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3520 BTRFS_BLOCK_GROUP_RAID5);
3521 if (num_devices < 3)
3522 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3523 if (num_devices < 4)
3524 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3526 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3527 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3528 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3531 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3532 tmp = BTRFS_BLOCK_GROUP_RAID6;
3533 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3534 tmp = BTRFS_BLOCK_GROUP_RAID5;
3535 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3536 tmp = BTRFS_BLOCK_GROUP_RAID10;
3537 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3538 tmp = BTRFS_BLOCK_GROUP_RAID1;
3539 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3540 tmp = BTRFS_BLOCK_GROUP_RAID0;
3542 return extended_to_chunk(flags | tmp);
3545 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3552 seq = read_seqbegin(&root->fs_info->profiles_lock);
3554 if (flags & BTRFS_BLOCK_GROUP_DATA)
3555 flags |= root->fs_info->avail_data_alloc_bits;
3556 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3557 flags |= root->fs_info->avail_system_alloc_bits;
3558 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3559 flags |= root->fs_info->avail_metadata_alloc_bits;
3560 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3562 return btrfs_reduce_alloc_profile(root, flags);
3565 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3571 flags = BTRFS_BLOCK_GROUP_DATA;
3572 else if (root == root->fs_info->chunk_root)
3573 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3575 flags = BTRFS_BLOCK_GROUP_METADATA;
3577 ret = get_alloc_profile(root, flags);
3582 * This will check the space that the inode allocates from to make sure we have
3583 * enough space for bytes.
3585 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3587 struct btrfs_space_info *data_sinfo;
3588 struct btrfs_root *root = BTRFS_I(inode)->root;
3589 struct btrfs_fs_info *fs_info = root->fs_info;
3591 int ret = 0, committed = 0, alloc_chunk = 1;
3593 /* make sure bytes are sectorsize aligned */
3594 bytes = ALIGN(bytes, root->sectorsize);
3596 if (btrfs_is_free_space_inode(inode)) {
3598 ASSERT(current->journal_info);
3601 data_sinfo = fs_info->data_sinfo;
3606 /* make sure we have enough space to handle the data first */
3607 spin_lock(&data_sinfo->lock);
3608 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3609 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3610 data_sinfo->bytes_may_use;
3612 if (used + bytes > data_sinfo->total_bytes) {
3613 struct btrfs_trans_handle *trans;
3616 * if we don't have enough free bytes in this space then we need
3617 * to alloc a new chunk.
3619 if (!data_sinfo->full && alloc_chunk) {
3622 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3623 spin_unlock(&data_sinfo->lock);
3625 alloc_target = btrfs_get_alloc_profile(root, 1);
3627 * It is ugly that we don't call nolock join
3628 * transaction for the free space inode case here.
3629 * But it is safe because we only do the data space
3630 * reservation for the free space cache in the
3631 * transaction context, the common join transaction
3632 * just increase the counter of the current transaction
3633 * handler, doesn't try to acquire the trans_lock of
3636 trans = btrfs_join_transaction(root);
3638 return PTR_ERR(trans);
3640 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3642 CHUNK_ALLOC_NO_FORCE);
3643 btrfs_end_transaction(trans, root);
3652 data_sinfo = fs_info->data_sinfo;
3658 * If we don't have enough pinned space to deal with this
3659 * allocation don't bother committing the transaction.
3661 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3664 spin_unlock(&data_sinfo->lock);
3666 /* commit the current transaction and try again */
3669 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3672 trans = btrfs_join_transaction(root);
3674 return PTR_ERR(trans);
3675 ret = btrfs_commit_transaction(trans, root);
3681 trace_btrfs_space_reservation(root->fs_info,
3682 "space_info:enospc",
3683 data_sinfo->flags, bytes, 1);
3686 data_sinfo->bytes_may_use += bytes;
3687 trace_btrfs_space_reservation(root->fs_info, "space_info",
3688 data_sinfo->flags, bytes, 1);
3689 spin_unlock(&data_sinfo->lock);
3695 * Called if we need to clear a data reservation for this inode.
3697 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3699 struct btrfs_root *root = BTRFS_I(inode)->root;
3700 struct btrfs_space_info *data_sinfo;
3702 /* make sure bytes are sectorsize aligned */
3703 bytes = ALIGN(bytes, root->sectorsize);
3705 data_sinfo = root->fs_info->data_sinfo;
3706 spin_lock(&data_sinfo->lock);
3707 WARN_ON(data_sinfo->bytes_may_use < bytes);
3708 data_sinfo->bytes_may_use -= bytes;
3709 trace_btrfs_space_reservation(root->fs_info, "space_info",
3710 data_sinfo->flags, bytes, 0);
3711 spin_unlock(&data_sinfo->lock);
3714 static void force_metadata_allocation(struct btrfs_fs_info *info)
3716 struct list_head *head = &info->space_info;
3717 struct btrfs_space_info *found;
3720 list_for_each_entry_rcu(found, head, list) {
3721 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3722 found->force_alloc = CHUNK_ALLOC_FORCE;
3727 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3729 return (global->size << 1);
3732 static int should_alloc_chunk(struct btrfs_root *root,
3733 struct btrfs_space_info *sinfo, int force)
3735 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3736 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3737 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3740 if (force == CHUNK_ALLOC_FORCE)
3744 * We need to take into account the global rsv because for all intents
3745 * and purposes it's used space. Don't worry about locking the
3746 * global_rsv, it doesn't change except when the transaction commits.
3748 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3749 num_allocated += calc_global_rsv_need_space(global_rsv);
3752 * in limited mode, we want to have some free space up to
3753 * about 1% of the FS size.
3755 if (force == CHUNK_ALLOC_LIMITED) {
3756 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3757 thresh = max_t(u64, 64 * 1024 * 1024,
3758 div_factor_fine(thresh, 1));
3760 if (num_bytes - num_allocated < thresh)
3764 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3769 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3773 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3774 BTRFS_BLOCK_GROUP_RAID0 |
3775 BTRFS_BLOCK_GROUP_RAID5 |
3776 BTRFS_BLOCK_GROUP_RAID6))
3777 num_dev = root->fs_info->fs_devices->rw_devices;
3778 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3781 num_dev = 1; /* DUP or single */
3783 /* metadata for updaing devices and chunk tree */
3784 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3787 static void check_system_chunk(struct btrfs_trans_handle *trans,
3788 struct btrfs_root *root, u64 type)
3790 struct btrfs_space_info *info;
3794 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3795 spin_lock(&info->lock);
3796 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3797 info->bytes_reserved - info->bytes_readonly;
3798 spin_unlock(&info->lock);
3800 thresh = get_system_chunk_thresh(root, type);
3801 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3802 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3803 left, thresh, type);
3804 dump_space_info(info, 0, 0);
3807 if (left < thresh) {
3810 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3811 btrfs_alloc_chunk(trans, root, flags);
3815 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3816 struct btrfs_root *extent_root, u64 flags, int force)
3818 struct btrfs_space_info *space_info;
3819 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3820 int wait_for_alloc = 0;
3823 /* Don't re-enter if we're already allocating a chunk */
3824 if (trans->allocating_chunk)
3827 space_info = __find_space_info(extent_root->fs_info, flags);
3829 ret = update_space_info(extent_root->fs_info, flags,
3831 BUG_ON(ret); /* -ENOMEM */
3833 BUG_ON(!space_info); /* Logic error */
3836 spin_lock(&space_info->lock);
3837 if (force < space_info->force_alloc)
3838 force = space_info->force_alloc;
3839 if (space_info->full) {
3840 if (should_alloc_chunk(extent_root, space_info, force))
3844 spin_unlock(&space_info->lock);
3848 if (!should_alloc_chunk(extent_root, space_info, force)) {
3849 spin_unlock(&space_info->lock);
3851 } else if (space_info->chunk_alloc) {
3854 space_info->chunk_alloc = 1;
3857 spin_unlock(&space_info->lock);
3859 mutex_lock(&fs_info->chunk_mutex);
3862 * The chunk_mutex is held throughout the entirety of a chunk
3863 * allocation, so once we've acquired the chunk_mutex we know that the
3864 * other guy is done and we need to recheck and see if we should
3867 if (wait_for_alloc) {
3868 mutex_unlock(&fs_info->chunk_mutex);
3873 trans->allocating_chunk = true;
3876 * If we have mixed data/metadata chunks we want to make sure we keep
3877 * allocating mixed chunks instead of individual chunks.
3879 if (btrfs_mixed_space_info(space_info))
3880 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3883 * if we're doing a data chunk, go ahead and make sure that
3884 * we keep a reasonable number of metadata chunks allocated in the
3887 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3888 fs_info->data_chunk_allocations++;
3889 if (!(fs_info->data_chunk_allocations %
3890 fs_info->metadata_ratio))
3891 force_metadata_allocation(fs_info);
3895 * Check if we have enough space in SYSTEM chunk because we may need
3896 * to update devices.
3898 check_system_chunk(trans, extent_root, flags);
3900 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3901 trans->allocating_chunk = false;
3903 spin_lock(&space_info->lock);
3904 if (ret < 0 && ret != -ENOSPC)
3907 space_info->full = 1;
3911 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3913 space_info->chunk_alloc = 0;
3914 spin_unlock(&space_info->lock);
3915 mutex_unlock(&fs_info->chunk_mutex);
3919 static int can_overcommit(struct btrfs_root *root,
3920 struct btrfs_space_info *space_info, u64 bytes,
3921 enum btrfs_reserve_flush_enum flush)
3923 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3924 u64 profile = btrfs_get_alloc_profile(root, 0);
3929 used = space_info->bytes_used + space_info->bytes_reserved +
3930 space_info->bytes_pinned + space_info->bytes_readonly;
3933 * We only want to allow over committing if we have lots of actual space
3934 * free, but if we don't have enough space to handle the global reserve
3935 * space then we could end up having a real enospc problem when trying
3936 * to allocate a chunk or some other such important allocation.
3938 spin_lock(&global_rsv->lock);
3939 space_size = calc_global_rsv_need_space(global_rsv);
3940 spin_unlock(&global_rsv->lock);
3941 if (used + space_size >= space_info->total_bytes)
3944 used += space_info->bytes_may_use;
3946 spin_lock(&root->fs_info->free_chunk_lock);
3947 avail = root->fs_info->free_chunk_space;
3948 spin_unlock(&root->fs_info->free_chunk_lock);
3951 * If we have dup, raid1 or raid10 then only half of the free
3952 * space is actually useable. For raid56, the space info used
3953 * doesn't include the parity drive, so we don't have to
3956 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3957 BTRFS_BLOCK_GROUP_RAID1 |
3958 BTRFS_BLOCK_GROUP_RAID10))
3962 * If we aren't flushing all things, let us overcommit up to
3963 * 1/2th of the space. If we can flush, don't let us overcommit
3964 * too much, let it overcommit up to 1/8 of the space.
3966 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3971 if (used + bytes < space_info->total_bytes + avail)
3976 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3977 unsigned long nr_pages, int nr_items)
3979 struct super_block *sb = root->fs_info->sb;
3981 if (down_read_trylock(&sb->s_umount)) {
3982 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
3983 up_read(&sb->s_umount);
3986 * We needn't worry the filesystem going from r/w to r/o though
3987 * we don't acquire ->s_umount mutex, because the filesystem
3988 * should guarantee the delalloc inodes list be empty after
3989 * the filesystem is readonly(all dirty pages are written to
3992 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
3993 if (!current->journal_info)
3994 btrfs_wait_ordered_roots(root->fs_info, nr_items);
3998 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4003 bytes = btrfs_calc_trans_metadata_size(root, 1);
4004 nr = (int)div64_u64(to_reclaim, bytes);
4010 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4013 * shrink metadata reservation for delalloc
4015 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4018 struct btrfs_block_rsv *block_rsv;
4019 struct btrfs_space_info *space_info;
4020 struct btrfs_trans_handle *trans;
4024 unsigned long nr_pages;
4027 enum btrfs_reserve_flush_enum flush;
4029 /* Calc the number of the pages we need flush for space reservation */
4030 items = calc_reclaim_items_nr(root, to_reclaim);
4031 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4033 trans = (struct btrfs_trans_handle *)current->journal_info;
4034 block_rsv = &root->fs_info->delalloc_block_rsv;
4035 space_info = block_rsv->space_info;
4037 delalloc_bytes = percpu_counter_sum_positive(
4038 &root->fs_info->delalloc_bytes);
4039 if (delalloc_bytes == 0) {
4043 btrfs_wait_ordered_roots(root->fs_info, items);
4048 while (delalloc_bytes && loops < 3) {
4049 max_reclaim = min(delalloc_bytes, to_reclaim);
4050 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4051 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4053 * We need to wait for the async pages to actually start before
4056 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4060 if (max_reclaim <= nr_pages)
4063 max_reclaim -= nr_pages;
4065 wait_event(root->fs_info->async_submit_wait,
4066 atomic_read(&root->fs_info->async_delalloc_pages) <=
4070 flush = BTRFS_RESERVE_FLUSH_ALL;
4072 flush = BTRFS_RESERVE_NO_FLUSH;
4073 spin_lock(&space_info->lock);
4074 if (can_overcommit(root, space_info, orig, flush)) {
4075 spin_unlock(&space_info->lock);
4078 spin_unlock(&space_info->lock);
4081 if (wait_ordered && !trans) {
4082 btrfs_wait_ordered_roots(root->fs_info, items);
4084 time_left = schedule_timeout_killable(1);
4088 delalloc_bytes = percpu_counter_sum_positive(
4089 &root->fs_info->delalloc_bytes);
4094 * maybe_commit_transaction - possibly commit the transaction if its ok to
4095 * @root - the root we're allocating for
4096 * @bytes - the number of bytes we want to reserve
4097 * @force - force the commit
4099 * This will check to make sure that committing the transaction will actually
4100 * get us somewhere and then commit the transaction if it does. Otherwise it
4101 * will return -ENOSPC.
4103 static int may_commit_transaction(struct btrfs_root *root,
4104 struct btrfs_space_info *space_info,
4105 u64 bytes, int force)
4107 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4108 struct btrfs_trans_handle *trans;
4110 trans = (struct btrfs_trans_handle *)current->journal_info;
4117 /* See if there is enough pinned space to make this reservation */
4118 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4123 * See if there is some space in the delayed insertion reservation for
4126 if (space_info != delayed_rsv->space_info)
4129 spin_lock(&delayed_rsv->lock);
4130 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4131 bytes - delayed_rsv->size) >= 0) {
4132 spin_unlock(&delayed_rsv->lock);
4135 spin_unlock(&delayed_rsv->lock);
4138 trans = btrfs_join_transaction(root);
4142 return btrfs_commit_transaction(trans, root);
4146 FLUSH_DELAYED_ITEMS_NR = 1,
4147 FLUSH_DELAYED_ITEMS = 2,
4149 FLUSH_DELALLOC_WAIT = 4,
4154 static int flush_space(struct btrfs_root *root,
4155 struct btrfs_space_info *space_info, u64 num_bytes,
4156 u64 orig_bytes, int state)
4158 struct btrfs_trans_handle *trans;
4163 case FLUSH_DELAYED_ITEMS_NR:
4164 case FLUSH_DELAYED_ITEMS:
4165 if (state == FLUSH_DELAYED_ITEMS_NR)
4166 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4170 trans = btrfs_join_transaction(root);
4171 if (IS_ERR(trans)) {
4172 ret = PTR_ERR(trans);
4175 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4176 btrfs_end_transaction(trans, root);
4178 case FLUSH_DELALLOC:
4179 case FLUSH_DELALLOC_WAIT:
4180 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4181 state == FLUSH_DELALLOC_WAIT);
4184 trans = btrfs_join_transaction(root);
4185 if (IS_ERR(trans)) {
4186 ret = PTR_ERR(trans);
4189 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4190 btrfs_get_alloc_profile(root, 0),
4191 CHUNK_ALLOC_NO_FORCE);
4192 btrfs_end_transaction(trans, root);
4197 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4208 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4209 struct btrfs_space_info *space_info)
4215 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4217 spin_lock(&space_info->lock);
4218 if (can_overcommit(root, space_info, to_reclaim,
4219 BTRFS_RESERVE_FLUSH_ALL)) {
4224 used = space_info->bytes_used + space_info->bytes_reserved +
4225 space_info->bytes_pinned + space_info->bytes_readonly +
4226 space_info->bytes_may_use;
4227 if (can_overcommit(root, space_info, 1024 * 1024,
4228 BTRFS_RESERVE_FLUSH_ALL))
4229 expected = div_factor_fine(space_info->total_bytes, 95);
4231 expected = div_factor_fine(space_info->total_bytes, 90);
4233 if (used > expected)
4234 to_reclaim = used - expected;
4237 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4238 space_info->bytes_reserved);
4240 spin_unlock(&space_info->lock);
4245 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4246 struct btrfs_fs_info *fs_info, u64 used)
4248 return (used >= div_factor_fine(space_info->total_bytes, 98) &&
4249 !btrfs_fs_closing(fs_info) &&
4250 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4253 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4254 struct btrfs_fs_info *fs_info)
4258 spin_lock(&space_info->lock);
4259 used = space_info->bytes_used + space_info->bytes_reserved +
4260 space_info->bytes_pinned + space_info->bytes_readonly +
4261 space_info->bytes_may_use;
4262 if (need_do_async_reclaim(space_info, fs_info, used)) {
4263 spin_unlock(&space_info->lock);
4266 spin_unlock(&space_info->lock);
4271 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4273 struct btrfs_fs_info *fs_info;
4274 struct btrfs_space_info *space_info;
4278 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4279 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4281 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4286 flush_state = FLUSH_DELAYED_ITEMS_NR;
4288 flush_space(fs_info->fs_root, space_info, to_reclaim,
4289 to_reclaim, flush_state);
4291 if (!btrfs_need_do_async_reclaim(space_info, fs_info))
4293 } while (flush_state <= COMMIT_TRANS);
4295 if (btrfs_need_do_async_reclaim(space_info, fs_info))
4296 queue_work(system_unbound_wq, work);
4299 void btrfs_init_async_reclaim_work(struct work_struct *work)
4301 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4305 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4306 * @root - the root we're allocating for
4307 * @block_rsv - the block_rsv we're allocating for
4308 * @orig_bytes - the number of bytes we want
4309 * @flush - whether or not we can flush to make our reservation
4311 * This will reserve orgi_bytes number of bytes from the space info associated
4312 * with the block_rsv. If there is not enough space it will make an attempt to
4313 * flush out space to make room. It will do this by flushing delalloc if
4314 * possible or committing the transaction. If flush is 0 then no attempts to
4315 * regain reservations will be made and this will fail if there is not enough
4318 static int reserve_metadata_bytes(struct btrfs_root *root,
4319 struct btrfs_block_rsv *block_rsv,
4321 enum btrfs_reserve_flush_enum flush)
4323 struct btrfs_space_info *space_info = block_rsv->space_info;
4325 u64 num_bytes = orig_bytes;
4326 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4328 bool flushing = false;
4332 spin_lock(&space_info->lock);
4334 * We only want to wait if somebody other than us is flushing and we
4335 * are actually allowed to flush all things.
4337 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4338 space_info->flush) {
4339 spin_unlock(&space_info->lock);
4341 * If we have a trans handle we can't wait because the flusher
4342 * may have to commit the transaction, which would mean we would
4343 * deadlock since we are waiting for the flusher to finish, but
4344 * hold the current transaction open.
4346 if (current->journal_info)
4348 ret = wait_event_killable(space_info->wait, !space_info->flush);
4349 /* Must have been killed, return */
4353 spin_lock(&space_info->lock);
4357 used = space_info->bytes_used + space_info->bytes_reserved +
4358 space_info->bytes_pinned + space_info->bytes_readonly +
4359 space_info->bytes_may_use;
4362 * The idea here is that we've not already over-reserved the block group
4363 * then we can go ahead and save our reservation first and then start
4364 * flushing if we need to. Otherwise if we've already overcommitted
4365 * lets start flushing stuff first and then come back and try to make
4368 if (used <= space_info->total_bytes) {
4369 if (used + orig_bytes <= space_info->total_bytes) {
4370 space_info->bytes_may_use += orig_bytes;
4371 trace_btrfs_space_reservation(root->fs_info,
4372 "space_info", space_info->flags, orig_bytes, 1);
4376 * Ok set num_bytes to orig_bytes since we aren't
4377 * overocmmitted, this way we only try and reclaim what
4380 num_bytes = orig_bytes;
4384 * Ok we're over committed, set num_bytes to the overcommitted
4385 * amount plus the amount of bytes that we need for this
4388 num_bytes = used - space_info->total_bytes +
4392 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4393 space_info->bytes_may_use += orig_bytes;
4394 trace_btrfs_space_reservation(root->fs_info, "space_info",
4395 space_info->flags, orig_bytes,
4401 * Couldn't make our reservation, save our place so while we're trying
4402 * to reclaim space we can actually use it instead of somebody else
4403 * stealing it from us.
4405 * We make the other tasks wait for the flush only when we can flush
4408 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4410 space_info->flush = 1;
4411 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4413 if (need_do_async_reclaim(space_info, root->fs_info, used) &&
4414 !work_busy(&root->fs_info->async_reclaim_work))
4415 queue_work(system_unbound_wq,
4416 &root->fs_info->async_reclaim_work);
4418 spin_unlock(&space_info->lock);
4420 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4423 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4428 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4429 * would happen. So skip delalloc flush.
4431 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4432 (flush_state == FLUSH_DELALLOC ||
4433 flush_state == FLUSH_DELALLOC_WAIT))
4434 flush_state = ALLOC_CHUNK;
4438 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4439 flush_state < COMMIT_TRANS)
4441 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4442 flush_state <= COMMIT_TRANS)
4446 if (ret == -ENOSPC &&
4447 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4448 struct btrfs_block_rsv *global_rsv =
4449 &root->fs_info->global_block_rsv;
4451 if (block_rsv != global_rsv &&
4452 !block_rsv_use_bytes(global_rsv, orig_bytes))
4456 trace_btrfs_space_reservation(root->fs_info,
4457 "space_info:enospc",
4458 space_info->flags, orig_bytes, 1);
4460 spin_lock(&space_info->lock);
4461 space_info->flush = 0;
4462 wake_up_all(&space_info->wait);
4463 spin_unlock(&space_info->lock);
4468 static struct btrfs_block_rsv *get_block_rsv(
4469 const struct btrfs_trans_handle *trans,
4470 const struct btrfs_root *root)
4472 struct btrfs_block_rsv *block_rsv = NULL;
4474 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4475 block_rsv = trans->block_rsv;
4477 if (root == root->fs_info->csum_root && trans->adding_csums)
4478 block_rsv = trans->block_rsv;
4480 if (root == root->fs_info->uuid_root)
4481 block_rsv = trans->block_rsv;
4484 block_rsv = root->block_rsv;
4487 block_rsv = &root->fs_info->empty_block_rsv;
4492 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4496 spin_lock(&block_rsv->lock);
4497 if (block_rsv->reserved >= num_bytes) {
4498 block_rsv->reserved -= num_bytes;
4499 if (block_rsv->reserved < block_rsv->size)
4500 block_rsv->full = 0;
4503 spin_unlock(&block_rsv->lock);
4507 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4508 u64 num_bytes, int update_size)
4510 spin_lock(&block_rsv->lock);
4511 block_rsv->reserved += num_bytes;
4513 block_rsv->size += num_bytes;
4514 else if (block_rsv->reserved >= block_rsv->size)
4515 block_rsv->full = 1;
4516 spin_unlock(&block_rsv->lock);
4519 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4520 struct btrfs_block_rsv *dest, u64 num_bytes,
4523 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4526 if (global_rsv->space_info != dest->space_info)
4529 spin_lock(&global_rsv->lock);
4530 min_bytes = div_factor(global_rsv->size, min_factor);
4531 if (global_rsv->reserved < min_bytes + num_bytes) {
4532 spin_unlock(&global_rsv->lock);
4535 global_rsv->reserved -= num_bytes;
4536 if (global_rsv->reserved < global_rsv->size)
4537 global_rsv->full = 0;
4538 spin_unlock(&global_rsv->lock);
4540 block_rsv_add_bytes(dest, num_bytes, 1);
4544 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4545 struct btrfs_block_rsv *block_rsv,
4546 struct btrfs_block_rsv *dest, u64 num_bytes)
4548 struct btrfs_space_info *space_info = block_rsv->space_info;
4550 spin_lock(&block_rsv->lock);
4551 if (num_bytes == (u64)-1)
4552 num_bytes = block_rsv->size;
4553 block_rsv->size -= num_bytes;
4554 if (block_rsv->reserved >= block_rsv->size) {
4555 num_bytes = block_rsv->reserved - block_rsv->size;
4556 block_rsv->reserved = block_rsv->size;
4557 block_rsv->full = 1;
4561 spin_unlock(&block_rsv->lock);
4563 if (num_bytes > 0) {
4565 spin_lock(&dest->lock);
4569 bytes_to_add = dest->size - dest->reserved;
4570 bytes_to_add = min(num_bytes, bytes_to_add);
4571 dest->reserved += bytes_to_add;
4572 if (dest->reserved >= dest->size)
4574 num_bytes -= bytes_to_add;
4576 spin_unlock(&dest->lock);
4579 spin_lock(&space_info->lock);
4580 space_info->bytes_may_use -= num_bytes;
4581 trace_btrfs_space_reservation(fs_info, "space_info",
4582 space_info->flags, num_bytes, 0);
4583 spin_unlock(&space_info->lock);
4588 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4589 struct btrfs_block_rsv *dst, u64 num_bytes)
4593 ret = block_rsv_use_bytes(src, num_bytes);
4597 block_rsv_add_bytes(dst, num_bytes, 1);
4601 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4603 memset(rsv, 0, sizeof(*rsv));
4604 spin_lock_init(&rsv->lock);
4608 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4609 unsigned short type)
4611 struct btrfs_block_rsv *block_rsv;
4612 struct btrfs_fs_info *fs_info = root->fs_info;
4614 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4618 btrfs_init_block_rsv(block_rsv, type);
4619 block_rsv->space_info = __find_space_info(fs_info,
4620 BTRFS_BLOCK_GROUP_METADATA);
4624 void btrfs_free_block_rsv(struct btrfs_root *root,
4625 struct btrfs_block_rsv *rsv)
4629 btrfs_block_rsv_release(root, rsv, (u64)-1);
4633 int btrfs_block_rsv_add(struct btrfs_root *root,
4634 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4635 enum btrfs_reserve_flush_enum flush)
4642 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4644 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4651 int btrfs_block_rsv_check(struct btrfs_root *root,
4652 struct btrfs_block_rsv *block_rsv, int min_factor)
4660 spin_lock(&block_rsv->lock);
4661 num_bytes = div_factor(block_rsv->size, min_factor);
4662 if (block_rsv->reserved >= num_bytes)
4664 spin_unlock(&block_rsv->lock);
4669 int btrfs_block_rsv_refill(struct btrfs_root *root,
4670 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4671 enum btrfs_reserve_flush_enum flush)
4679 spin_lock(&block_rsv->lock);
4680 num_bytes = min_reserved;
4681 if (block_rsv->reserved >= num_bytes)
4684 num_bytes -= block_rsv->reserved;
4685 spin_unlock(&block_rsv->lock);
4690 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4692 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4699 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4700 struct btrfs_block_rsv *dst_rsv,
4703 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4706 void btrfs_block_rsv_release(struct btrfs_root *root,
4707 struct btrfs_block_rsv *block_rsv,
4710 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4711 if (global_rsv == block_rsv ||
4712 block_rsv->space_info != global_rsv->space_info)
4714 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4719 * helper to calculate size of global block reservation.
4720 * the desired value is sum of space used by extent tree,
4721 * checksum tree and root tree
4723 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4725 struct btrfs_space_info *sinfo;
4729 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4731 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4732 spin_lock(&sinfo->lock);
4733 data_used = sinfo->bytes_used;
4734 spin_unlock(&sinfo->lock);
4736 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4737 spin_lock(&sinfo->lock);
4738 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4740 meta_used = sinfo->bytes_used;
4741 spin_unlock(&sinfo->lock);
4743 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4745 num_bytes += div64_u64(data_used + meta_used, 50);
4747 if (num_bytes * 3 > meta_used)
4748 num_bytes = div64_u64(meta_used, 3);
4750 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4753 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4755 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4756 struct btrfs_space_info *sinfo = block_rsv->space_info;
4759 num_bytes = calc_global_metadata_size(fs_info);
4761 spin_lock(&sinfo->lock);
4762 spin_lock(&block_rsv->lock);
4764 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4766 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4767 sinfo->bytes_reserved + sinfo->bytes_readonly +
4768 sinfo->bytes_may_use;
4770 if (sinfo->total_bytes > num_bytes) {
4771 num_bytes = sinfo->total_bytes - num_bytes;
4772 block_rsv->reserved += num_bytes;
4773 sinfo->bytes_may_use += num_bytes;
4774 trace_btrfs_space_reservation(fs_info, "space_info",
4775 sinfo->flags, num_bytes, 1);
4778 if (block_rsv->reserved >= block_rsv->size) {
4779 num_bytes = block_rsv->reserved - block_rsv->size;
4780 sinfo->bytes_may_use -= num_bytes;
4781 trace_btrfs_space_reservation(fs_info, "space_info",
4782 sinfo->flags, num_bytes, 0);
4783 block_rsv->reserved = block_rsv->size;
4784 block_rsv->full = 1;
4787 spin_unlock(&block_rsv->lock);
4788 spin_unlock(&sinfo->lock);
4791 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4793 struct btrfs_space_info *space_info;
4795 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4796 fs_info->chunk_block_rsv.space_info = space_info;
4798 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4799 fs_info->global_block_rsv.space_info = space_info;
4800 fs_info->delalloc_block_rsv.space_info = space_info;
4801 fs_info->trans_block_rsv.space_info = space_info;
4802 fs_info->empty_block_rsv.space_info = space_info;
4803 fs_info->delayed_block_rsv.space_info = space_info;
4805 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4806 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4807 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4808 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4809 if (fs_info->quota_root)
4810 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4811 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4813 update_global_block_rsv(fs_info);
4816 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4818 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4820 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4821 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4822 WARN_ON(fs_info->trans_block_rsv.size > 0);
4823 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4824 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4825 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4826 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4827 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4830 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4831 struct btrfs_root *root)
4833 if (!trans->block_rsv)
4836 if (!trans->bytes_reserved)
4839 trace_btrfs_space_reservation(root->fs_info, "transaction",
4840 trans->transid, trans->bytes_reserved, 0);
4841 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4842 trans->bytes_reserved = 0;
4845 /* Can only return 0 or -ENOSPC */
4846 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4847 struct inode *inode)
4849 struct btrfs_root *root = BTRFS_I(inode)->root;
4850 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4851 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4854 * We need to hold space in order to delete our orphan item once we've
4855 * added it, so this takes the reservation so we can release it later
4856 * when we are truly done with the orphan item.
4858 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4859 trace_btrfs_space_reservation(root->fs_info, "orphan",
4860 btrfs_ino(inode), num_bytes, 1);
4861 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4864 void btrfs_orphan_release_metadata(struct inode *inode)
4866 struct btrfs_root *root = BTRFS_I(inode)->root;
4867 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4868 trace_btrfs_space_reservation(root->fs_info, "orphan",
4869 btrfs_ino(inode), num_bytes, 0);
4870 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4874 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4875 * root: the root of the parent directory
4876 * rsv: block reservation
4877 * items: the number of items that we need do reservation
4878 * qgroup_reserved: used to return the reserved size in qgroup
4880 * This function is used to reserve the space for snapshot/subvolume
4881 * creation and deletion. Those operations are different with the
4882 * common file/directory operations, they change two fs/file trees
4883 * and root tree, the number of items that the qgroup reserves is
4884 * different with the free space reservation. So we can not use
4885 * the space reseravtion mechanism in start_transaction().
4887 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4888 struct btrfs_block_rsv *rsv,
4890 u64 *qgroup_reserved,
4891 bool use_global_rsv)
4895 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4897 if (root->fs_info->quota_enabled) {
4898 /* One for parent inode, two for dir entries */
4899 num_bytes = 3 * root->leafsize;
4900 ret = btrfs_qgroup_reserve(root, num_bytes);
4907 *qgroup_reserved = num_bytes;
4909 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4910 rsv->space_info = __find_space_info(root->fs_info,
4911 BTRFS_BLOCK_GROUP_METADATA);
4912 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4913 BTRFS_RESERVE_FLUSH_ALL);
4915 if (ret == -ENOSPC && use_global_rsv)
4916 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4919 if (*qgroup_reserved)
4920 btrfs_qgroup_free(root, *qgroup_reserved);
4926 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4927 struct btrfs_block_rsv *rsv,
4928 u64 qgroup_reserved)
4930 btrfs_block_rsv_release(root, rsv, (u64)-1);
4931 if (qgroup_reserved)
4932 btrfs_qgroup_free(root, qgroup_reserved);
4936 * drop_outstanding_extent - drop an outstanding extent
4937 * @inode: the inode we're dropping the extent for
4939 * This is called when we are freeing up an outstanding extent, either called
4940 * after an error or after an extent is written. This will return the number of
4941 * reserved extents that need to be freed. This must be called with
4942 * BTRFS_I(inode)->lock held.
4944 static unsigned drop_outstanding_extent(struct inode *inode)
4946 unsigned drop_inode_space = 0;
4947 unsigned dropped_extents = 0;
4949 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4950 BTRFS_I(inode)->outstanding_extents--;
4952 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4953 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4954 &BTRFS_I(inode)->runtime_flags))
4955 drop_inode_space = 1;
4958 * If we have more or the same amount of outsanding extents than we have
4959 * reserved then we need to leave the reserved extents count alone.
4961 if (BTRFS_I(inode)->outstanding_extents >=
4962 BTRFS_I(inode)->reserved_extents)
4963 return drop_inode_space;
4965 dropped_extents = BTRFS_I(inode)->reserved_extents -
4966 BTRFS_I(inode)->outstanding_extents;
4967 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4968 return dropped_extents + drop_inode_space;
4972 * calc_csum_metadata_size - return the amount of metada space that must be
4973 * reserved/free'd for the given bytes.
4974 * @inode: the inode we're manipulating
4975 * @num_bytes: the number of bytes in question
4976 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4978 * This adjusts the number of csum_bytes in the inode and then returns the
4979 * correct amount of metadata that must either be reserved or freed. We
4980 * calculate how many checksums we can fit into one leaf and then divide the
4981 * number of bytes that will need to be checksumed by this value to figure out
4982 * how many checksums will be required. If we are adding bytes then the number
4983 * may go up and we will return the number of additional bytes that must be
4984 * reserved. If it is going down we will return the number of bytes that must
4987 * This must be called with BTRFS_I(inode)->lock held.
4989 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4992 struct btrfs_root *root = BTRFS_I(inode)->root;
4994 int num_csums_per_leaf;
4998 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4999 BTRFS_I(inode)->csum_bytes == 0)
5002 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
5004 BTRFS_I(inode)->csum_bytes += num_bytes;
5006 BTRFS_I(inode)->csum_bytes -= num_bytes;
5007 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
5008 num_csums_per_leaf = (int)div64_u64(csum_size,
5009 sizeof(struct btrfs_csum_item) +
5010 sizeof(struct btrfs_disk_key));
5011 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
5012 num_csums = num_csums + num_csums_per_leaf - 1;
5013 num_csums = num_csums / num_csums_per_leaf;
5015 old_csums = old_csums + num_csums_per_leaf - 1;
5016 old_csums = old_csums / num_csums_per_leaf;
5018 /* No change, no need to reserve more */
5019 if (old_csums == num_csums)
5023 return btrfs_calc_trans_metadata_size(root,
5024 num_csums - old_csums);
5026 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5029 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5031 struct btrfs_root *root = BTRFS_I(inode)->root;
5032 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5035 unsigned nr_extents = 0;
5036 int extra_reserve = 0;
5037 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5039 bool delalloc_lock = true;
5043 /* If we are a free space inode we need to not flush since we will be in
5044 * the middle of a transaction commit. We also don't need the delalloc
5045 * mutex since we won't race with anybody. We need this mostly to make
5046 * lockdep shut its filthy mouth.
5048 if (btrfs_is_free_space_inode(inode)) {
5049 flush = BTRFS_RESERVE_NO_FLUSH;
5050 delalloc_lock = false;
5053 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5054 btrfs_transaction_in_commit(root->fs_info))
5055 schedule_timeout(1);
5058 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5060 num_bytes = ALIGN(num_bytes, root->sectorsize);
5062 spin_lock(&BTRFS_I(inode)->lock);
5063 BTRFS_I(inode)->outstanding_extents++;
5065 if (BTRFS_I(inode)->outstanding_extents >
5066 BTRFS_I(inode)->reserved_extents)
5067 nr_extents = BTRFS_I(inode)->outstanding_extents -
5068 BTRFS_I(inode)->reserved_extents;
5071 * Add an item to reserve for updating the inode when we complete the
5074 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5075 &BTRFS_I(inode)->runtime_flags)) {
5080 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5081 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5082 csum_bytes = BTRFS_I(inode)->csum_bytes;
5083 spin_unlock(&BTRFS_I(inode)->lock);
5085 if (root->fs_info->quota_enabled) {
5086 ret = btrfs_qgroup_reserve(root, num_bytes +
5087 nr_extents * root->leafsize);
5092 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5093 if (unlikely(ret)) {
5094 if (root->fs_info->quota_enabled)
5095 btrfs_qgroup_free(root, num_bytes +
5096 nr_extents * root->leafsize);
5100 spin_lock(&BTRFS_I(inode)->lock);
5101 if (extra_reserve) {
5102 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5103 &BTRFS_I(inode)->runtime_flags);
5106 BTRFS_I(inode)->reserved_extents += nr_extents;
5107 spin_unlock(&BTRFS_I(inode)->lock);
5110 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5113 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5114 btrfs_ino(inode), to_reserve, 1);
5115 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5120 spin_lock(&BTRFS_I(inode)->lock);
5121 dropped = drop_outstanding_extent(inode);
5123 * If the inodes csum_bytes is the same as the original
5124 * csum_bytes then we know we haven't raced with any free()ers
5125 * so we can just reduce our inodes csum bytes and carry on.
5127 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5128 calc_csum_metadata_size(inode, num_bytes, 0);
5130 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5134 * This is tricky, but first we need to figure out how much we
5135 * free'd from any free-ers that occured during this
5136 * reservation, so we reset ->csum_bytes to the csum_bytes
5137 * before we dropped our lock, and then call the free for the
5138 * number of bytes that were freed while we were trying our
5141 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5142 BTRFS_I(inode)->csum_bytes = csum_bytes;
5143 to_free = calc_csum_metadata_size(inode, bytes, 0);
5147 * Now we need to see how much we would have freed had we not
5148 * been making this reservation and our ->csum_bytes were not
5149 * artificially inflated.
5151 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5152 bytes = csum_bytes - orig_csum_bytes;
5153 bytes = calc_csum_metadata_size(inode, bytes, 0);
5156 * Now reset ->csum_bytes to what it should be. If bytes is
5157 * more than to_free then we would have free'd more space had we
5158 * not had an artificially high ->csum_bytes, so we need to free
5159 * the remainder. If bytes is the same or less then we don't
5160 * need to do anything, the other free-ers did the correct
5163 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5164 if (bytes > to_free)
5165 to_free = bytes - to_free;
5169 spin_unlock(&BTRFS_I(inode)->lock);
5171 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5174 btrfs_block_rsv_release(root, block_rsv, to_free);
5175 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5176 btrfs_ino(inode), to_free, 0);
5179 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5184 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5185 * @inode: the inode to release the reservation for
5186 * @num_bytes: the number of bytes we're releasing
5188 * This will release the metadata reservation for an inode. This can be called
5189 * once we complete IO for a given set of bytes to release their metadata
5192 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5194 struct btrfs_root *root = BTRFS_I(inode)->root;
5198 num_bytes = ALIGN(num_bytes, root->sectorsize);
5199 spin_lock(&BTRFS_I(inode)->lock);
5200 dropped = drop_outstanding_extent(inode);
5203 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5204 spin_unlock(&BTRFS_I(inode)->lock);
5206 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5208 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5209 btrfs_ino(inode), to_free, 0);
5210 if (root->fs_info->quota_enabled) {
5211 btrfs_qgroup_free(root, num_bytes +
5212 dropped * root->leafsize);
5215 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5220 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5221 * @inode: inode we're writing to
5222 * @num_bytes: the number of bytes we want to allocate
5224 * This will do the following things
5226 * o reserve space in the data space info for num_bytes
5227 * o reserve space in the metadata space info based on number of outstanding
5228 * extents and how much csums will be needed
5229 * o add to the inodes ->delalloc_bytes
5230 * o add it to the fs_info's delalloc inodes list.
5232 * This will return 0 for success and -ENOSPC if there is no space left.
5234 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5238 ret = btrfs_check_data_free_space(inode, num_bytes);
5242 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5244 btrfs_free_reserved_data_space(inode, num_bytes);
5252 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5253 * @inode: inode we're releasing space for
5254 * @num_bytes: the number of bytes we want to free up
5256 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5257 * called in the case that we don't need the metadata AND data reservations
5258 * anymore. So if there is an error or we insert an inline extent.
5260 * This function will release the metadata space that was not used and will
5261 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5262 * list if there are no delalloc bytes left.
5264 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5266 btrfs_delalloc_release_metadata(inode, num_bytes);
5267 btrfs_free_reserved_data_space(inode, num_bytes);
5270 static int update_block_group(struct btrfs_root *root,
5271 u64 bytenr, u64 num_bytes, int alloc)
5273 struct btrfs_block_group_cache *cache = NULL;
5274 struct btrfs_fs_info *info = root->fs_info;
5275 u64 total = num_bytes;
5280 /* block accounting for super block */
5281 spin_lock(&info->delalloc_root_lock);
5282 old_val = btrfs_super_bytes_used(info->super_copy);
5284 old_val += num_bytes;
5286 old_val -= num_bytes;
5287 btrfs_set_super_bytes_used(info->super_copy, old_val);
5288 spin_unlock(&info->delalloc_root_lock);
5291 cache = btrfs_lookup_block_group(info, bytenr);
5294 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5295 BTRFS_BLOCK_GROUP_RAID1 |
5296 BTRFS_BLOCK_GROUP_RAID10))
5301 * If this block group has free space cache written out, we
5302 * need to make sure to load it if we are removing space. This
5303 * is because we need the unpinning stage to actually add the
5304 * space back to the block group, otherwise we will leak space.
5306 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5307 cache_block_group(cache, 1);
5309 byte_in_group = bytenr - cache->key.objectid;
5310 WARN_ON(byte_in_group > cache->key.offset);
5312 spin_lock(&cache->space_info->lock);
5313 spin_lock(&cache->lock);
5315 if (btrfs_test_opt(root, SPACE_CACHE) &&
5316 cache->disk_cache_state < BTRFS_DC_CLEAR)
5317 cache->disk_cache_state = BTRFS_DC_CLEAR;
5320 old_val = btrfs_block_group_used(&cache->item);
5321 num_bytes = min(total, cache->key.offset - byte_in_group);
5323 old_val += num_bytes;
5324 btrfs_set_block_group_used(&cache->item, old_val);
5325 cache->reserved -= num_bytes;
5326 cache->space_info->bytes_reserved -= num_bytes;
5327 cache->space_info->bytes_used += num_bytes;
5328 cache->space_info->disk_used += num_bytes * factor;
5329 spin_unlock(&cache->lock);
5330 spin_unlock(&cache->space_info->lock);
5332 old_val -= num_bytes;
5333 btrfs_set_block_group_used(&cache->item, old_val);
5334 cache->pinned += num_bytes;
5335 cache->space_info->bytes_pinned += num_bytes;
5336 cache->space_info->bytes_used -= num_bytes;
5337 cache->space_info->disk_used -= num_bytes * factor;
5338 spin_unlock(&cache->lock);
5339 spin_unlock(&cache->space_info->lock);
5341 set_extent_dirty(info->pinned_extents,
5342 bytenr, bytenr + num_bytes - 1,
5343 GFP_NOFS | __GFP_NOFAIL);
5345 btrfs_put_block_group(cache);
5347 bytenr += num_bytes;
5352 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5354 struct btrfs_block_group_cache *cache;
5357 spin_lock(&root->fs_info->block_group_cache_lock);
5358 bytenr = root->fs_info->first_logical_byte;
5359 spin_unlock(&root->fs_info->block_group_cache_lock);
5361 if (bytenr < (u64)-1)
5364 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5368 bytenr = cache->key.objectid;
5369 btrfs_put_block_group(cache);
5374 static int pin_down_extent(struct btrfs_root *root,
5375 struct btrfs_block_group_cache *cache,
5376 u64 bytenr, u64 num_bytes, int reserved)
5378 spin_lock(&cache->space_info->lock);
5379 spin_lock(&cache->lock);
5380 cache->pinned += num_bytes;
5381 cache->space_info->bytes_pinned += num_bytes;
5383 cache->reserved -= num_bytes;
5384 cache->space_info->bytes_reserved -= num_bytes;
5386 spin_unlock(&cache->lock);
5387 spin_unlock(&cache->space_info->lock);
5389 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5390 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5392 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5397 * this function must be called within transaction
5399 int btrfs_pin_extent(struct btrfs_root *root,
5400 u64 bytenr, u64 num_bytes, int reserved)
5402 struct btrfs_block_group_cache *cache;
5404 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5405 BUG_ON(!cache); /* Logic error */
5407 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5409 btrfs_put_block_group(cache);
5414 * this function must be called within transaction
5416 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5417 u64 bytenr, u64 num_bytes)
5419 struct btrfs_block_group_cache *cache;
5422 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5427 * pull in the free space cache (if any) so that our pin
5428 * removes the free space from the cache. We have load_only set
5429 * to one because the slow code to read in the free extents does check
5430 * the pinned extents.
5432 cache_block_group(cache, 1);
5434 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5436 /* remove us from the free space cache (if we're there at all) */
5437 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5438 btrfs_put_block_group(cache);
5442 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5445 struct btrfs_block_group_cache *block_group;
5446 struct btrfs_caching_control *caching_ctl;
5448 block_group = btrfs_lookup_block_group(root->fs_info, start);
5452 cache_block_group(block_group, 0);
5453 caching_ctl = get_caching_control(block_group);
5457 BUG_ON(!block_group_cache_done(block_group));
5458 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5460 mutex_lock(&caching_ctl->mutex);
5462 if (start >= caching_ctl->progress) {
5463 ret = add_excluded_extent(root, start, num_bytes);
5464 } else if (start + num_bytes <= caching_ctl->progress) {
5465 ret = btrfs_remove_free_space(block_group,
5468 num_bytes = caching_ctl->progress - start;
5469 ret = btrfs_remove_free_space(block_group,
5474 num_bytes = (start + num_bytes) -
5475 caching_ctl->progress;
5476 start = caching_ctl->progress;
5477 ret = add_excluded_extent(root, start, num_bytes);
5480 mutex_unlock(&caching_ctl->mutex);
5481 put_caching_control(caching_ctl);
5483 btrfs_put_block_group(block_group);
5487 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5488 struct extent_buffer *eb)
5490 struct btrfs_file_extent_item *item;
5491 struct btrfs_key key;
5495 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5498 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5499 btrfs_item_key_to_cpu(eb, &key, i);
5500 if (key.type != BTRFS_EXTENT_DATA_KEY)
5502 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5503 found_type = btrfs_file_extent_type(eb, item);
5504 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5506 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5508 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5509 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5510 __exclude_logged_extent(log, key.objectid, key.offset);
5517 * btrfs_update_reserved_bytes - update the block_group and space info counters
5518 * @cache: The cache we are manipulating
5519 * @num_bytes: The number of bytes in question
5520 * @reserve: One of the reservation enums
5522 * This is called by the allocator when it reserves space, or by somebody who is
5523 * freeing space that was never actually used on disk. For example if you
5524 * reserve some space for a new leaf in transaction A and before transaction A
5525 * commits you free that leaf, you call this with reserve set to 0 in order to
5526 * clear the reservation.
5528 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5529 * ENOSPC accounting. For data we handle the reservation through clearing the
5530 * delalloc bits in the io_tree. We have to do this since we could end up
5531 * allocating less disk space for the amount of data we have reserved in the
5532 * case of compression.
5534 * If this is a reservation and the block group has become read only we cannot
5535 * make the reservation and return -EAGAIN, otherwise this function always
5538 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5539 u64 num_bytes, int reserve)
5541 struct btrfs_space_info *space_info = cache->space_info;
5544 spin_lock(&space_info->lock);
5545 spin_lock(&cache->lock);
5546 if (reserve != RESERVE_FREE) {
5550 cache->reserved += num_bytes;
5551 space_info->bytes_reserved += num_bytes;
5552 if (reserve == RESERVE_ALLOC) {
5553 trace_btrfs_space_reservation(cache->fs_info,
5554 "space_info", space_info->flags,
5556 space_info->bytes_may_use -= num_bytes;
5561 space_info->bytes_readonly += num_bytes;
5562 cache->reserved -= num_bytes;
5563 space_info->bytes_reserved -= num_bytes;
5565 spin_unlock(&cache->lock);
5566 spin_unlock(&space_info->lock);
5570 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5571 struct btrfs_root *root)
5573 struct btrfs_fs_info *fs_info = root->fs_info;
5574 struct btrfs_caching_control *next;
5575 struct btrfs_caching_control *caching_ctl;
5576 struct btrfs_block_group_cache *cache;
5577 struct btrfs_space_info *space_info;
5579 down_write(&fs_info->commit_root_sem);
5581 list_for_each_entry_safe(caching_ctl, next,
5582 &fs_info->caching_block_groups, list) {
5583 cache = caching_ctl->block_group;
5584 if (block_group_cache_done(cache)) {
5585 cache->last_byte_to_unpin = (u64)-1;
5586 list_del_init(&caching_ctl->list);
5587 put_caching_control(caching_ctl);
5589 cache->last_byte_to_unpin = caching_ctl->progress;
5593 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5594 fs_info->pinned_extents = &fs_info->freed_extents[1];
5596 fs_info->pinned_extents = &fs_info->freed_extents[0];
5598 up_write(&fs_info->commit_root_sem);
5600 list_for_each_entry_rcu(space_info, &fs_info->space_info, list)
5601 percpu_counter_set(&space_info->total_bytes_pinned, 0);
5603 update_global_block_rsv(fs_info);
5606 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5608 struct btrfs_fs_info *fs_info = root->fs_info;
5609 struct btrfs_block_group_cache *cache = NULL;
5610 struct btrfs_space_info *space_info;
5611 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5615 while (start <= end) {
5618 start >= cache->key.objectid + cache->key.offset) {
5620 btrfs_put_block_group(cache);
5621 cache = btrfs_lookup_block_group(fs_info, start);
5622 BUG_ON(!cache); /* Logic error */
5625 len = cache->key.objectid + cache->key.offset - start;
5626 len = min(len, end + 1 - start);
5628 if (start < cache->last_byte_to_unpin) {
5629 len = min(len, cache->last_byte_to_unpin - start);
5630 btrfs_add_free_space(cache, start, len);
5634 space_info = cache->space_info;
5636 spin_lock(&space_info->lock);
5637 spin_lock(&cache->lock);
5638 cache->pinned -= len;
5639 space_info->bytes_pinned -= len;
5641 space_info->bytes_readonly += len;
5644 spin_unlock(&cache->lock);
5645 if (!readonly && global_rsv->space_info == space_info) {
5646 spin_lock(&global_rsv->lock);
5647 if (!global_rsv->full) {
5648 len = min(len, global_rsv->size -
5649 global_rsv->reserved);
5650 global_rsv->reserved += len;
5651 space_info->bytes_may_use += len;
5652 if (global_rsv->reserved >= global_rsv->size)
5653 global_rsv->full = 1;
5655 spin_unlock(&global_rsv->lock);
5657 spin_unlock(&space_info->lock);
5661 btrfs_put_block_group(cache);
5665 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5666 struct btrfs_root *root)
5668 struct btrfs_fs_info *fs_info = root->fs_info;
5669 struct extent_io_tree *unpin;
5677 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5678 unpin = &fs_info->freed_extents[1];
5680 unpin = &fs_info->freed_extents[0];
5683 ret = find_first_extent_bit(unpin, 0, &start, &end,
5684 EXTENT_DIRTY, NULL);
5688 if (btrfs_test_opt(root, DISCARD))
5689 ret = btrfs_discard_extent(root, start,
5690 end + 1 - start, NULL);
5692 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5693 unpin_extent_range(root, start, end);
5700 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5701 u64 owner, u64 root_objectid)
5703 struct btrfs_space_info *space_info;
5706 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5707 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5708 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5710 flags = BTRFS_BLOCK_GROUP_METADATA;
5712 flags = BTRFS_BLOCK_GROUP_DATA;
5715 space_info = __find_space_info(fs_info, flags);
5716 BUG_ON(!space_info); /* Logic bug */
5717 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5721 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5722 struct btrfs_root *root,
5723 u64 bytenr, u64 num_bytes, u64 parent,
5724 u64 root_objectid, u64 owner_objectid,
5725 u64 owner_offset, int refs_to_drop,
5726 struct btrfs_delayed_extent_op *extent_op)
5728 struct btrfs_key key;
5729 struct btrfs_path *path;
5730 struct btrfs_fs_info *info = root->fs_info;
5731 struct btrfs_root *extent_root = info->extent_root;
5732 struct extent_buffer *leaf;
5733 struct btrfs_extent_item *ei;
5734 struct btrfs_extent_inline_ref *iref;
5737 int extent_slot = 0;
5738 int found_extent = 0;
5742 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5745 path = btrfs_alloc_path();
5750 path->leave_spinning = 1;
5752 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5753 BUG_ON(!is_data && refs_to_drop != 1);
5756 skinny_metadata = 0;
5758 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5759 bytenr, num_bytes, parent,
5760 root_objectid, owner_objectid,
5763 extent_slot = path->slots[0];
5764 while (extent_slot >= 0) {
5765 btrfs_item_key_to_cpu(path->nodes[0], &key,
5767 if (key.objectid != bytenr)
5769 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5770 key.offset == num_bytes) {
5774 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5775 key.offset == owner_objectid) {
5779 if (path->slots[0] - extent_slot > 5)
5783 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5784 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5785 if (found_extent && item_size < sizeof(*ei))
5788 if (!found_extent) {
5790 ret = remove_extent_backref(trans, extent_root, path,
5794 btrfs_abort_transaction(trans, extent_root, ret);
5797 btrfs_release_path(path);
5798 path->leave_spinning = 1;
5800 key.objectid = bytenr;
5801 key.type = BTRFS_EXTENT_ITEM_KEY;
5802 key.offset = num_bytes;
5804 if (!is_data && skinny_metadata) {
5805 key.type = BTRFS_METADATA_ITEM_KEY;
5806 key.offset = owner_objectid;
5809 ret = btrfs_search_slot(trans, extent_root,
5811 if (ret > 0 && skinny_metadata && path->slots[0]) {
5813 * Couldn't find our skinny metadata item,
5814 * see if we have ye olde extent item.
5817 btrfs_item_key_to_cpu(path->nodes[0], &key,
5819 if (key.objectid == bytenr &&
5820 key.type == BTRFS_EXTENT_ITEM_KEY &&
5821 key.offset == num_bytes)
5825 if (ret > 0 && skinny_metadata) {
5826 skinny_metadata = false;
5827 key.objectid = bytenr;
5828 key.type = BTRFS_EXTENT_ITEM_KEY;
5829 key.offset = num_bytes;
5830 btrfs_release_path(path);
5831 ret = btrfs_search_slot(trans, extent_root,
5836 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5839 btrfs_print_leaf(extent_root,
5843 btrfs_abort_transaction(trans, extent_root, ret);
5846 extent_slot = path->slots[0];
5848 } else if (WARN_ON(ret == -ENOENT)) {
5849 btrfs_print_leaf(extent_root, path->nodes[0]);
5851 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5852 bytenr, parent, root_objectid, owner_objectid,
5854 btrfs_abort_transaction(trans, extent_root, ret);
5857 btrfs_abort_transaction(trans, extent_root, ret);
5861 leaf = path->nodes[0];
5862 item_size = btrfs_item_size_nr(leaf, extent_slot);
5863 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5864 if (item_size < sizeof(*ei)) {
5865 BUG_ON(found_extent || extent_slot != path->slots[0]);
5866 ret = convert_extent_item_v0(trans, extent_root, path,
5869 btrfs_abort_transaction(trans, extent_root, ret);
5873 btrfs_release_path(path);
5874 path->leave_spinning = 1;
5876 key.objectid = bytenr;
5877 key.type = BTRFS_EXTENT_ITEM_KEY;
5878 key.offset = num_bytes;
5880 ret = btrfs_search_slot(trans, extent_root, &key, path,
5883 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5885 btrfs_print_leaf(extent_root, path->nodes[0]);
5888 btrfs_abort_transaction(trans, extent_root, ret);
5892 extent_slot = path->slots[0];
5893 leaf = path->nodes[0];
5894 item_size = btrfs_item_size_nr(leaf, extent_slot);
5897 BUG_ON(item_size < sizeof(*ei));
5898 ei = btrfs_item_ptr(leaf, extent_slot,
5899 struct btrfs_extent_item);
5900 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5901 key.type == BTRFS_EXTENT_ITEM_KEY) {
5902 struct btrfs_tree_block_info *bi;
5903 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5904 bi = (struct btrfs_tree_block_info *)(ei + 1);
5905 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5908 refs = btrfs_extent_refs(leaf, ei);
5909 if (refs < refs_to_drop) {
5910 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5911 "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5913 btrfs_abort_transaction(trans, extent_root, ret);
5916 refs -= refs_to_drop;
5920 __run_delayed_extent_op(extent_op, leaf, ei);
5922 * In the case of inline back ref, reference count will
5923 * be updated by remove_extent_backref
5926 BUG_ON(!found_extent);
5928 btrfs_set_extent_refs(leaf, ei, refs);
5929 btrfs_mark_buffer_dirty(leaf);
5932 ret = remove_extent_backref(trans, extent_root, path,
5936 btrfs_abort_transaction(trans, extent_root, ret);
5940 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
5944 BUG_ON(is_data && refs_to_drop !=
5945 extent_data_ref_count(root, path, iref));
5947 BUG_ON(path->slots[0] != extent_slot);
5949 BUG_ON(path->slots[0] != extent_slot + 1);
5950 path->slots[0] = extent_slot;
5955 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5958 btrfs_abort_transaction(trans, extent_root, ret);
5961 btrfs_release_path(path);
5964 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5966 btrfs_abort_transaction(trans, extent_root, ret);
5971 ret = update_block_group(root, bytenr, num_bytes, 0);
5973 btrfs_abort_transaction(trans, extent_root, ret);
5978 btrfs_free_path(path);
5983 * when we free an block, it is possible (and likely) that we free the last
5984 * delayed ref for that extent as well. This searches the delayed ref tree for
5985 * a given extent, and if there are no other delayed refs to be processed, it
5986 * removes it from the tree.
5988 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5989 struct btrfs_root *root, u64 bytenr)
5991 struct btrfs_delayed_ref_head *head;
5992 struct btrfs_delayed_ref_root *delayed_refs;
5995 delayed_refs = &trans->transaction->delayed_refs;
5996 spin_lock(&delayed_refs->lock);
5997 head = btrfs_find_delayed_ref_head(trans, bytenr);
5999 goto out_delayed_unlock;
6001 spin_lock(&head->lock);
6002 if (rb_first(&head->ref_root))
6005 if (head->extent_op) {
6006 if (!head->must_insert_reserved)
6008 btrfs_free_delayed_extent_op(head->extent_op);
6009 head->extent_op = NULL;
6013 * waiting for the lock here would deadlock. If someone else has it
6014 * locked they are already in the process of dropping it anyway
6016 if (!mutex_trylock(&head->mutex))
6020 * at this point we have a head with no other entries. Go
6021 * ahead and process it.
6023 head->node.in_tree = 0;
6024 rb_erase(&head->href_node, &delayed_refs->href_root);
6026 atomic_dec(&delayed_refs->num_entries);
6029 * we don't take a ref on the node because we're removing it from the
6030 * tree, so we just steal the ref the tree was holding.
6032 delayed_refs->num_heads--;
6033 if (head->processing == 0)
6034 delayed_refs->num_heads_ready--;
6035 head->processing = 0;
6036 spin_unlock(&head->lock);
6037 spin_unlock(&delayed_refs->lock);
6039 BUG_ON(head->extent_op);
6040 if (head->must_insert_reserved)
6043 mutex_unlock(&head->mutex);
6044 btrfs_put_delayed_ref(&head->node);
6047 spin_unlock(&head->lock);
6050 spin_unlock(&delayed_refs->lock);
6054 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6055 struct btrfs_root *root,
6056 struct extent_buffer *buf,
6057 u64 parent, int last_ref)
6059 struct btrfs_block_group_cache *cache = NULL;
6063 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6064 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6065 buf->start, buf->len,
6066 parent, root->root_key.objectid,
6067 btrfs_header_level(buf),
6068 BTRFS_DROP_DELAYED_REF, NULL, 0);
6069 BUG_ON(ret); /* -ENOMEM */
6075 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6077 if (btrfs_header_generation(buf) == trans->transid) {
6078 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6079 ret = check_ref_cleanup(trans, root, buf->start);
6084 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6085 pin_down_extent(root, cache, buf->start, buf->len, 1);
6089 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6091 btrfs_add_free_space(cache, buf->start, buf->len);
6092 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
6093 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6098 add_pinned_bytes(root->fs_info, buf->len,
6099 btrfs_header_level(buf),
6100 root->root_key.objectid);
6103 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6106 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6107 btrfs_put_block_group(cache);
6110 /* Can return -ENOMEM */
6111 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6112 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6113 u64 owner, u64 offset, int for_cow)
6116 struct btrfs_fs_info *fs_info = root->fs_info;
6118 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6121 * tree log blocks never actually go into the extent allocation
6122 * tree, just update pinning info and exit early.
6124 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6125 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6126 /* unlocks the pinned mutex */
6127 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6129 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6130 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6132 parent, root_objectid, (int)owner,
6133 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
6135 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6137 parent, root_objectid, owner,
6138 offset, BTRFS_DROP_DELAYED_REF,
6144 static u64 stripe_align(struct btrfs_root *root,
6145 struct btrfs_block_group_cache *cache,
6146 u64 val, u64 num_bytes)
6148 u64 ret = ALIGN(val, root->stripesize);
6153 * when we wait for progress in the block group caching, its because
6154 * our allocation attempt failed at least once. So, we must sleep
6155 * and let some progress happen before we try again.
6157 * This function will sleep at least once waiting for new free space to
6158 * show up, and then it will check the block group free space numbers
6159 * for our min num_bytes. Another option is to have it go ahead
6160 * and look in the rbtree for a free extent of a given size, but this
6163 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6164 * any of the information in this block group.
6166 static noinline void
6167 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6170 struct btrfs_caching_control *caching_ctl;
6172 caching_ctl = get_caching_control(cache);
6176 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6177 (cache->free_space_ctl->free_space >= num_bytes));
6179 put_caching_control(caching_ctl);
6183 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6185 struct btrfs_caching_control *caching_ctl;
6188 caching_ctl = get_caching_control(cache);
6190 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6192 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6193 if (cache->cached == BTRFS_CACHE_ERROR)
6195 put_caching_control(caching_ctl);
6199 int __get_raid_index(u64 flags)
6201 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6202 return BTRFS_RAID_RAID10;
6203 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6204 return BTRFS_RAID_RAID1;
6205 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6206 return BTRFS_RAID_DUP;
6207 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6208 return BTRFS_RAID_RAID0;
6209 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6210 return BTRFS_RAID_RAID5;
6211 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6212 return BTRFS_RAID_RAID6;
6214 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6217 int get_block_group_index(struct btrfs_block_group_cache *cache)
6219 return __get_raid_index(cache->flags);
6222 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6223 [BTRFS_RAID_RAID10] = "raid10",
6224 [BTRFS_RAID_RAID1] = "raid1",
6225 [BTRFS_RAID_DUP] = "dup",
6226 [BTRFS_RAID_RAID0] = "raid0",
6227 [BTRFS_RAID_SINGLE] = "single",
6228 [BTRFS_RAID_RAID5] = "raid5",
6229 [BTRFS_RAID_RAID6] = "raid6",
6232 static const char *get_raid_name(enum btrfs_raid_types type)
6234 if (type >= BTRFS_NR_RAID_TYPES)
6237 return btrfs_raid_type_names[type];
6240 enum btrfs_loop_type {
6241 LOOP_CACHING_NOWAIT = 0,
6242 LOOP_CACHING_WAIT = 1,
6243 LOOP_ALLOC_CHUNK = 2,
6244 LOOP_NO_EMPTY_SIZE = 3,
6248 * walks the btree of allocated extents and find a hole of a given size.
6249 * The key ins is changed to record the hole:
6250 * ins->objectid == start position
6251 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6252 * ins->offset == the size of the hole.
6253 * Any available blocks before search_start are skipped.
6255 * If there is no suitable free space, we will record the max size of
6256 * the free space extent currently.
6258 static noinline int find_free_extent(struct btrfs_root *orig_root,
6259 u64 num_bytes, u64 empty_size,
6260 u64 hint_byte, struct btrfs_key *ins,
6264 struct btrfs_root *root = orig_root->fs_info->extent_root;
6265 struct btrfs_free_cluster *last_ptr = NULL;
6266 struct btrfs_block_group_cache *block_group = NULL;
6267 u64 search_start = 0;
6268 u64 max_extent_size = 0;
6269 int empty_cluster = 2 * 1024 * 1024;
6270 struct btrfs_space_info *space_info;
6272 int index = __get_raid_index(flags);
6273 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6274 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6275 bool failed_cluster_refill = false;
6276 bool failed_alloc = false;
6277 bool use_cluster = true;
6278 bool have_caching_bg = false;
6280 WARN_ON(num_bytes < root->sectorsize);
6281 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
6285 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6287 space_info = __find_space_info(root->fs_info, flags);
6289 btrfs_err(root->fs_info, "No space info for %llu", flags);
6294 * If the space info is for both data and metadata it means we have a
6295 * small filesystem and we can't use the clustering stuff.
6297 if (btrfs_mixed_space_info(space_info))
6298 use_cluster = false;
6300 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6301 last_ptr = &root->fs_info->meta_alloc_cluster;
6302 if (!btrfs_test_opt(root, SSD))
6303 empty_cluster = 64 * 1024;
6306 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6307 btrfs_test_opt(root, SSD)) {
6308 last_ptr = &root->fs_info->data_alloc_cluster;
6312 spin_lock(&last_ptr->lock);
6313 if (last_ptr->block_group)
6314 hint_byte = last_ptr->window_start;
6315 spin_unlock(&last_ptr->lock);
6318 search_start = max(search_start, first_logical_byte(root, 0));
6319 search_start = max(search_start, hint_byte);
6324 if (search_start == hint_byte) {
6325 block_group = btrfs_lookup_block_group(root->fs_info,
6328 * we don't want to use the block group if it doesn't match our
6329 * allocation bits, or if its not cached.
6331 * However if we are re-searching with an ideal block group
6332 * picked out then we don't care that the block group is cached.
6334 if (block_group && block_group_bits(block_group, flags) &&
6335 block_group->cached != BTRFS_CACHE_NO) {
6336 down_read(&space_info->groups_sem);
6337 if (list_empty(&block_group->list) ||
6340 * someone is removing this block group,
6341 * we can't jump into the have_block_group
6342 * target because our list pointers are not
6345 btrfs_put_block_group(block_group);
6346 up_read(&space_info->groups_sem);
6348 index = get_block_group_index(block_group);
6349 goto have_block_group;
6351 } else if (block_group) {
6352 btrfs_put_block_group(block_group);
6356 have_caching_bg = false;
6357 down_read(&space_info->groups_sem);
6358 list_for_each_entry(block_group, &space_info->block_groups[index],
6363 btrfs_get_block_group(block_group);
6364 search_start = block_group->key.objectid;
6367 * this can happen if we end up cycling through all the
6368 * raid types, but we want to make sure we only allocate
6369 * for the proper type.
6371 if (!block_group_bits(block_group, flags)) {
6372 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6373 BTRFS_BLOCK_GROUP_RAID1 |
6374 BTRFS_BLOCK_GROUP_RAID5 |
6375 BTRFS_BLOCK_GROUP_RAID6 |
6376 BTRFS_BLOCK_GROUP_RAID10;
6379 * if they asked for extra copies and this block group
6380 * doesn't provide them, bail. This does allow us to
6381 * fill raid0 from raid1.
6383 if ((flags & extra) && !(block_group->flags & extra))
6388 cached = block_group_cache_done(block_group);
6389 if (unlikely(!cached)) {
6390 ret = cache_block_group(block_group, 0);
6395 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6397 if (unlikely(block_group->ro))
6401 * Ok we want to try and use the cluster allocator, so
6405 struct btrfs_block_group_cache *used_block_group;
6406 unsigned long aligned_cluster;
6408 * the refill lock keeps out other
6409 * people trying to start a new cluster
6411 spin_lock(&last_ptr->refill_lock);
6412 used_block_group = last_ptr->block_group;
6413 if (used_block_group != block_group &&
6414 (!used_block_group ||
6415 used_block_group->ro ||
6416 !block_group_bits(used_block_group, flags)))
6417 goto refill_cluster;
6419 if (used_block_group != block_group)
6420 btrfs_get_block_group(used_block_group);
6422 offset = btrfs_alloc_from_cluster(used_block_group,
6425 used_block_group->key.objectid,
6428 /* we have a block, we're done */
6429 spin_unlock(&last_ptr->refill_lock);
6430 trace_btrfs_reserve_extent_cluster(root,
6432 search_start, num_bytes);
6433 if (used_block_group != block_group) {
6434 btrfs_put_block_group(block_group);
6435 block_group = used_block_group;
6440 WARN_ON(last_ptr->block_group != used_block_group);
6441 if (used_block_group != block_group)
6442 btrfs_put_block_group(used_block_group);
6444 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6445 * set up a new clusters, so lets just skip it
6446 * and let the allocator find whatever block
6447 * it can find. If we reach this point, we
6448 * will have tried the cluster allocator
6449 * plenty of times and not have found
6450 * anything, so we are likely way too
6451 * fragmented for the clustering stuff to find
6454 * However, if the cluster is taken from the
6455 * current block group, release the cluster
6456 * first, so that we stand a better chance of
6457 * succeeding in the unclustered
6459 if (loop >= LOOP_NO_EMPTY_SIZE &&
6460 last_ptr->block_group != block_group) {
6461 spin_unlock(&last_ptr->refill_lock);
6462 goto unclustered_alloc;
6466 * this cluster didn't work out, free it and
6469 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6471 if (loop >= LOOP_NO_EMPTY_SIZE) {
6472 spin_unlock(&last_ptr->refill_lock);
6473 goto unclustered_alloc;
6476 aligned_cluster = max_t(unsigned long,
6477 empty_cluster + empty_size,
6478 block_group->full_stripe_len);
6480 /* allocate a cluster in this block group */
6481 ret = btrfs_find_space_cluster(root, block_group,
6482 last_ptr, search_start,
6487 * now pull our allocation out of this
6490 offset = btrfs_alloc_from_cluster(block_group,
6496 /* we found one, proceed */
6497 spin_unlock(&last_ptr->refill_lock);
6498 trace_btrfs_reserve_extent_cluster(root,
6499 block_group, search_start,
6503 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6504 && !failed_cluster_refill) {
6505 spin_unlock(&last_ptr->refill_lock);
6507 failed_cluster_refill = true;
6508 wait_block_group_cache_progress(block_group,
6509 num_bytes + empty_cluster + empty_size);
6510 goto have_block_group;
6514 * at this point we either didn't find a cluster
6515 * or we weren't able to allocate a block from our
6516 * cluster. Free the cluster we've been trying
6517 * to use, and go to the next block group
6519 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6520 spin_unlock(&last_ptr->refill_lock);
6525 spin_lock(&block_group->free_space_ctl->tree_lock);
6527 block_group->free_space_ctl->free_space <
6528 num_bytes + empty_cluster + empty_size) {
6529 if (block_group->free_space_ctl->free_space >
6532 block_group->free_space_ctl->free_space;
6533 spin_unlock(&block_group->free_space_ctl->tree_lock);
6536 spin_unlock(&block_group->free_space_ctl->tree_lock);
6538 offset = btrfs_find_space_for_alloc(block_group, search_start,
6539 num_bytes, empty_size,
6542 * If we didn't find a chunk, and we haven't failed on this
6543 * block group before, and this block group is in the middle of
6544 * caching and we are ok with waiting, then go ahead and wait
6545 * for progress to be made, and set failed_alloc to true.
6547 * If failed_alloc is true then we've already waited on this
6548 * block group once and should move on to the next block group.
6550 if (!offset && !failed_alloc && !cached &&
6551 loop > LOOP_CACHING_NOWAIT) {
6552 wait_block_group_cache_progress(block_group,
6553 num_bytes + empty_size);
6554 failed_alloc = true;
6555 goto have_block_group;
6556 } else if (!offset) {
6558 have_caching_bg = true;
6562 search_start = stripe_align(root, block_group,
6565 /* move on to the next group */
6566 if (search_start + num_bytes >
6567 block_group->key.objectid + block_group->key.offset) {
6568 btrfs_add_free_space(block_group, offset, num_bytes);
6572 if (offset < search_start)
6573 btrfs_add_free_space(block_group, offset,
6574 search_start - offset);
6575 BUG_ON(offset > search_start);
6577 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
6579 if (ret == -EAGAIN) {
6580 btrfs_add_free_space(block_group, offset, num_bytes);
6584 /* we are all good, lets return */
6585 ins->objectid = search_start;
6586 ins->offset = num_bytes;
6588 trace_btrfs_reserve_extent(orig_root, block_group,
6589 search_start, num_bytes);
6590 btrfs_put_block_group(block_group);
6593 failed_cluster_refill = false;
6594 failed_alloc = false;
6595 BUG_ON(index != get_block_group_index(block_group));
6596 btrfs_put_block_group(block_group);
6598 up_read(&space_info->groups_sem);
6600 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6603 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6607 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6608 * caching kthreads as we move along
6609 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6610 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6611 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6614 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6617 if (loop == LOOP_ALLOC_CHUNK) {
6618 struct btrfs_trans_handle *trans;
6620 trans = btrfs_join_transaction(root);
6621 if (IS_ERR(trans)) {
6622 ret = PTR_ERR(trans);
6626 ret = do_chunk_alloc(trans, root, flags,
6629 * Do not bail out on ENOSPC since we
6630 * can do more things.
6632 if (ret < 0 && ret != -ENOSPC)
6633 btrfs_abort_transaction(trans,
6637 btrfs_end_transaction(trans, root);
6642 if (loop == LOOP_NO_EMPTY_SIZE) {
6648 } else if (!ins->objectid) {
6650 } else if (ins->objectid) {
6655 ins->offset = max_extent_size;
6659 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6660 int dump_block_groups)
6662 struct btrfs_block_group_cache *cache;
6665 spin_lock(&info->lock);
6666 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
6668 info->total_bytes - info->bytes_used - info->bytes_pinned -
6669 info->bytes_reserved - info->bytes_readonly,
6670 (info->full) ? "" : "not ");
6671 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
6672 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6673 info->total_bytes, info->bytes_used, info->bytes_pinned,
6674 info->bytes_reserved, info->bytes_may_use,
6675 info->bytes_readonly);
6676 spin_unlock(&info->lock);
6678 if (!dump_block_groups)
6681 down_read(&info->groups_sem);
6683 list_for_each_entry(cache, &info->block_groups[index], list) {
6684 spin_lock(&cache->lock);
6685 printk(KERN_INFO "BTRFS: "
6686 "block group %llu has %llu bytes, "
6687 "%llu used %llu pinned %llu reserved %s\n",
6688 cache->key.objectid, cache->key.offset,
6689 btrfs_block_group_used(&cache->item), cache->pinned,
6690 cache->reserved, cache->ro ? "[readonly]" : "");
6691 btrfs_dump_free_space(cache, bytes);
6692 spin_unlock(&cache->lock);
6694 if (++index < BTRFS_NR_RAID_TYPES)
6696 up_read(&info->groups_sem);
6699 int btrfs_reserve_extent(struct btrfs_root *root,
6700 u64 num_bytes, u64 min_alloc_size,
6701 u64 empty_size, u64 hint_byte,
6702 struct btrfs_key *ins, int is_data)
6704 bool final_tried = false;
6708 flags = btrfs_get_alloc_profile(root, is_data);
6710 WARN_ON(num_bytes < root->sectorsize);
6711 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6714 if (ret == -ENOSPC) {
6715 if (!final_tried && ins->offset) {
6716 num_bytes = min(num_bytes >> 1, ins->offset);
6717 num_bytes = round_down(num_bytes, root->sectorsize);
6718 num_bytes = max(num_bytes, min_alloc_size);
6719 if (num_bytes == min_alloc_size)
6722 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6723 struct btrfs_space_info *sinfo;
6725 sinfo = __find_space_info(root->fs_info, flags);
6726 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6729 dump_space_info(sinfo, num_bytes, 1);
6736 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6737 u64 start, u64 len, int pin)
6739 struct btrfs_block_group_cache *cache;
6742 cache = btrfs_lookup_block_group(root->fs_info, start);
6744 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6749 if (btrfs_test_opt(root, DISCARD))
6750 ret = btrfs_discard_extent(root, start, len, NULL);
6753 pin_down_extent(root, cache, start, len, 1);
6755 btrfs_add_free_space(cache, start, len);
6756 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6758 btrfs_put_block_group(cache);
6760 trace_btrfs_reserved_extent_free(root, start, len);
6765 int btrfs_free_reserved_extent(struct btrfs_root *root,
6768 return __btrfs_free_reserved_extent(root, start, len, 0);
6771 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6774 return __btrfs_free_reserved_extent(root, start, len, 1);
6777 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6778 struct btrfs_root *root,
6779 u64 parent, u64 root_objectid,
6780 u64 flags, u64 owner, u64 offset,
6781 struct btrfs_key *ins, int ref_mod)
6784 struct btrfs_fs_info *fs_info = root->fs_info;
6785 struct btrfs_extent_item *extent_item;
6786 struct btrfs_extent_inline_ref *iref;
6787 struct btrfs_path *path;
6788 struct extent_buffer *leaf;
6793 type = BTRFS_SHARED_DATA_REF_KEY;
6795 type = BTRFS_EXTENT_DATA_REF_KEY;
6797 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6799 path = btrfs_alloc_path();
6803 path->leave_spinning = 1;
6804 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6807 btrfs_free_path(path);
6811 leaf = path->nodes[0];
6812 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6813 struct btrfs_extent_item);
6814 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6815 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6816 btrfs_set_extent_flags(leaf, extent_item,
6817 flags | BTRFS_EXTENT_FLAG_DATA);
6819 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6820 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6822 struct btrfs_shared_data_ref *ref;
6823 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6824 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6825 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6827 struct btrfs_extent_data_ref *ref;
6828 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6829 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6830 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6831 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6832 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6835 btrfs_mark_buffer_dirty(path->nodes[0]);
6836 btrfs_free_path(path);
6838 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6839 if (ret) { /* -ENOENT, logic error */
6840 btrfs_err(fs_info, "update block group failed for %llu %llu",
6841 ins->objectid, ins->offset);
6844 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6848 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6849 struct btrfs_root *root,
6850 u64 parent, u64 root_objectid,
6851 u64 flags, struct btrfs_disk_key *key,
6852 int level, struct btrfs_key *ins)
6855 struct btrfs_fs_info *fs_info = root->fs_info;
6856 struct btrfs_extent_item *extent_item;
6857 struct btrfs_tree_block_info *block_info;
6858 struct btrfs_extent_inline_ref *iref;
6859 struct btrfs_path *path;
6860 struct extent_buffer *leaf;
6861 u32 size = sizeof(*extent_item) + sizeof(*iref);
6862 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6865 if (!skinny_metadata)
6866 size += sizeof(*block_info);
6868 path = btrfs_alloc_path();
6870 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6875 path->leave_spinning = 1;
6876 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6879 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6881 btrfs_free_path(path);
6885 leaf = path->nodes[0];
6886 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6887 struct btrfs_extent_item);
6888 btrfs_set_extent_refs(leaf, extent_item, 1);
6889 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6890 btrfs_set_extent_flags(leaf, extent_item,
6891 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6893 if (skinny_metadata) {
6894 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6896 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6897 btrfs_set_tree_block_key(leaf, block_info, key);
6898 btrfs_set_tree_block_level(leaf, block_info, level);
6899 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6903 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6904 btrfs_set_extent_inline_ref_type(leaf, iref,
6905 BTRFS_SHARED_BLOCK_REF_KEY);
6906 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6908 btrfs_set_extent_inline_ref_type(leaf, iref,
6909 BTRFS_TREE_BLOCK_REF_KEY);
6910 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6913 btrfs_mark_buffer_dirty(leaf);
6914 btrfs_free_path(path);
6916 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6917 if (ret) { /* -ENOENT, logic error */
6918 btrfs_err(fs_info, "update block group failed for %llu %llu",
6919 ins->objectid, ins->offset);
6923 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->leafsize);
6927 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6928 struct btrfs_root *root,
6929 u64 root_objectid, u64 owner,
6930 u64 offset, struct btrfs_key *ins)
6934 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6936 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6938 root_objectid, owner, offset,
6939 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6944 * this is used by the tree logging recovery code. It records that
6945 * an extent has been allocated and makes sure to clear the free
6946 * space cache bits as well
6948 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6949 struct btrfs_root *root,
6950 u64 root_objectid, u64 owner, u64 offset,
6951 struct btrfs_key *ins)
6954 struct btrfs_block_group_cache *block_group;
6957 * Mixed block groups will exclude before processing the log so we only
6958 * need to do the exlude dance if this fs isn't mixed.
6960 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6961 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6966 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6970 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6971 RESERVE_ALLOC_NO_ACCOUNT);
6972 BUG_ON(ret); /* logic error */
6973 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6974 0, owner, offset, ins, 1);
6975 btrfs_put_block_group(block_group);
6979 static struct extent_buffer *
6980 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6981 u64 bytenr, u32 blocksize, int level)
6983 struct extent_buffer *buf;
6985 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6987 return ERR_PTR(-ENOMEM);
6988 btrfs_set_header_generation(buf, trans->transid);
6989 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6990 btrfs_tree_lock(buf);
6991 clean_tree_block(trans, root, buf);
6992 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6994 btrfs_set_lock_blocking(buf);
6995 btrfs_set_buffer_uptodate(buf);
6997 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6999 * we allow two log transactions at a time, use different
7000 * EXENT bit to differentiate dirty pages.
7002 if (root->log_transid % 2 == 0)
7003 set_extent_dirty(&root->dirty_log_pages, buf->start,
7004 buf->start + buf->len - 1, GFP_NOFS);
7006 set_extent_new(&root->dirty_log_pages, buf->start,
7007 buf->start + buf->len - 1, GFP_NOFS);
7009 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7010 buf->start + buf->len - 1, GFP_NOFS);
7012 trans->blocks_used++;
7013 /* this returns a buffer locked for blocking */
7017 static struct btrfs_block_rsv *
7018 use_block_rsv(struct btrfs_trans_handle *trans,
7019 struct btrfs_root *root, u32 blocksize)
7021 struct btrfs_block_rsv *block_rsv;
7022 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7024 bool global_updated = false;
7026 block_rsv = get_block_rsv(trans, root);
7028 if (unlikely(block_rsv->size == 0))
7031 ret = block_rsv_use_bytes(block_rsv, blocksize);
7035 if (block_rsv->failfast)
7036 return ERR_PTR(ret);
7038 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7039 global_updated = true;
7040 update_global_block_rsv(root->fs_info);
7044 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7045 static DEFINE_RATELIMIT_STATE(_rs,
7046 DEFAULT_RATELIMIT_INTERVAL * 10,
7047 /*DEFAULT_RATELIMIT_BURST*/ 1);
7048 if (__ratelimit(&_rs))
7050 "BTRFS: block rsv returned %d\n", ret);
7053 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7054 BTRFS_RESERVE_NO_FLUSH);
7058 * If we couldn't reserve metadata bytes try and use some from
7059 * the global reserve if its space type is the same as the global
7062 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7063 block_rsv->space_info == global_rsv->space_info) {
7064 ret = block_rsv_use_bytes(global_rsv, blocksize);
7068 return ERR_PTR(ret);
7071 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7072 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7074 block_rsv_add_bytes(block_rsv, blocksize, 0);
7075 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7079 * finds a free extent and does all the dirty work required for allocation
7080 * returns the key for the extent through ins, and a tree buffer for
7081 * the first block of the extent through buf.
7083 * returns the tree buffer or NULL.
7085 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
7086 struct btrfs_root *root, u32 blocksize,
7087 u64 parent, u64 root_objectid,
7088 struct btrfs_disk_key *key, int level,
7089 u64 hint, u64 empty_size)
7091 struct btrfs_key ins;
7092 struct btrfs_block_rsv *block_rsv;
7093 struct extent_buffer *buf;
7096 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7099 block_rsv = use_block_rsv(trans, root, blocksize);
7100 if (IS_ERR(block_rsv))
7101 return ERR_CAST(block_rsv);
7103 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7104 empty_size, hint, &ins, 0);
7106 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7107 return ERR_PTR(ret);
7110 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
7112 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
7114 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7116 parent = ins.objectid;
7117 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7121 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7122 struct btrfs_delayed_extent_op *extent_op;
7123 extent_op = btrfs_alloc_delayed_extent_op();
7124 BUG_ON(!extent_op); /* -ENOMEM */
7126 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7128 memset(&extent_op->key, 0, sizeof(extent_op->key));
7129 extent_op->flags_to_set = flags;
7130 if (skinny_metadata)
7131 extent_op->update_key = 0;
7133 extent_op->update_key = 1;
7134 extent_op->update_flags = 1;
7135 extent_op->is_data = 0;
7136 extent_op->level = level;
7138 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7140 ins.offset, parent, root_objectid,
7141 level, BTRFS_ADD_DELAYED_EXTENT,
7143 BUG_ON(ret); /* -ENOMEM */
7148 struct walk_control {
7149 u64 refs[BTRFS_MAX_LEVEL];
7150 u64 flags[BTRFS_MAX_LEVEL];
7151 struct btrfs_key update_progress;
7162 #define DROP_REFERENCE 1
7163 #define UPDATE_BACKREF 2
7165 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7166 struct btrfs_root *root,
7167 struct walk_control *wc,
7168 struct btrfs_path *path)
7176 struct btrfs_key key;
7177 struct extent_buffer *eb;
7182 if (path->slots[wc->level] < wc->reada_slot) {
7183 wc->reada_count = wc->reada_count * 2 / 3;
7184 wc->reada_count = max(wc->reada_count, 2);
7186 wc->reada_count = wc->reada_count * 3 / 2;
7187 wc->reada_count = min_t(int, wc->reada_count,
7188 BTRFS_NODEPTRS_PER_BLOCK(root));
7191 eb = path->nodes[wc->level];
7192 nritems = btrfs_header_nritems(eb);
7193 blocksize = btrfs_level_size(root, wc->level - 1);
7195 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7196 if (nread >= wc->reada_count)
7200 bytenr = btrfs_node_blockptr(eb, slot);
7201 generation = btrfs_node_ptr_generation(eb, slot);
7203 if (slot == path->slots[wc->level])
7206 if (wc->stage == UPDATE_BACKREF &&
7207 generation <= root->root_key.offset)
7210 /* We don't lock the tree block, it's OK to be racy here */
7211 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7212 wc->level - 1, 1, &refs,
7214 /* We don't care about errors in readahead. */
7219 if (wc->stage == DROP_REFERENCE) {
7223 if (wc->level == 1 &&
7224 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7226 if (!wc->update_ref ||
7227 generation <= root->root_key.offset)
7229 btrfs_node_key_to_cpu(eb, &key, slot);
7230 ret = btrfs_comp_cpu_keys(&key,
7231 &wc->update_progress);
7235 if (wc->level == 1 &&
7236 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7240 ret = readahead_tree_block(root, bytenr, blocksize,
7246 wc->reada_slot = slot;
7250 * helper to process tree block while walking down the tree.
7252 * when wc->stage == UPDATE_BACKREF, this function updates
7253 * back refs for pointers in the block.
7255 * NOTE: return value 1 means we should stop walking down.
7257 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7258 struct btrfs_root *root,
7259 struct btrfs_path *path,
7260 struct walk_control *wc, int lookup_info)
7262 int level = wc->level;
7263 struct extent_buffer *eb = path->nodes[level];
7264 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7267 if (wc->stage == UPDATE_BACKREF &&
7268 btrfs_header_owner(eb) != root->root_key.objectid)
7272 * when reference count of tree block is 1, it won't increase
7273 * again. once full backref flag is set, we never clear it.
7276 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7277 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7278 BUG_ON(!path->locks[level]);
7279 ret = btrfs_lookup_extent_info(trans, root,
7280 eb->start, level, 1,
7283 BUG_ON(ret == -ENOMEM);
7286 BUG_ON(wc->refs[level] == 0);
7289 if (wc->stage == DROP_REFERENCE) {
7290 if (wc->refs[level] > 1)
7293 if (path->locks[level] && !wc->keep_locks) {
7294 btrfs_tree_unlock_rw(eb, path->locks[level]);
7295 path->locks[level] = 0;
7300 /* wc->stage == UPDATE_BACKREF */
7301 if (!(wc->flags[level] & flag)) {
7302 BUG_ON(!path->locks[level]);
7303 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
7304 BUG_ON(ret); /* -ENOMEM */
7305 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
7306 BUG_ON(ret); /* -ENOMEM */
7307 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7309 btrfs_header_level(eb), 0);
7310 BUG_ON(ret); /* -ENOMEM */
7311 wc->flags[level] |= flag;
7315 * the block is shared by multiple trees, so it's not good to
7316 * keep the tree lock
7318 if (path->locks[level] && level > 0) {
7319 btrfs_tree_unlock_rw(eb, path->locks[level]);
7320 path->locks[level] = 0;
7326 * helper to process tree block pointer.
7328 * when wc->stage == DROP_REFERENCE, this function checks
7329 * reference count of the block pointed to. if the block
7330 * is shared and we need update back refs for the subtree
7331 * rooted at the block, this function changes wc->stage to
7332 * UPDATE_BACKREF. if the block is shared and there is no
7333 * need to update back, this function drops the reference
7336 * NOTE: return value 1 means we should stop walking down.
7338 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7339 struct btrfs_root *root,
7340 struct btrfs_path *path,
7341 struct walk_control *wc, int *lookup_info)
7347 struct btrfs_key key;
7348 struct extent_buffer *next;
7349 int level = wc->level;
7353 generation = btrfs_node_ptr_generation(path->nodes[level],
7354 path->slots[level]);
7356 * if the lower level block was created before the snapshot
7357 * was created, we know there is no need to update back refs
7360 if (wc->stage == UPDATE_BACKREF &&
7361 generation <= root->root_key.offset) {
7366 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7367 blocksize = btrfs_level_size(root, level - 1);
7369 next = btrfs_find_tree_block(root, bytenr, blocksize);
7371 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7374 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7378 btrfs_tree_lock(next);
7379 btrfs_set_lock_blocking(next);
7381 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7382 &wc->refs[level - 1],
7383 &wc->flags[level - 1]);
7385 btrfs_tree_unlock(next);
7389 if (unlikely(wc->refs[level - 1] == 0)) {
7390 btrfs_err(root->fs_info, "Missing references.");
7395 if (wc->stage == DROP_REFERENCE) {
7396 if (wc->refs[level - 1] > 1) {
7398 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7401 if (!wc->update_ref ||
7402 generation <= root->root_key.offset)
7405 btrfs_node_key_to_cpu(path->nodes[level], &key,
7406 path->slots[level]);
7407 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7411 wc->stage = UPDATE_BACKREF;
7412 wc->shared_level = level - 1;
7416 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7420 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7421 btrfs_tree_unlock(next);
7422 free_extent_buffer(next);
7428 if (reada && level == 1)
7429 reada_walk_down(trans, root, wc, path);
7430 next = read_tree_block(root, bytenr, blocksize, generation);
7431 if (!next || !extent_buffer_uptodate(next)) {
7432 free_extent_buffer(next);
7435 btrfs_tree_lock(next);
7436 btrfs_set_lock_blocking(next);
7440 BUG_ON(level != btrfs_header_level(next));
7441 path->nodes[level] = next;
7442 path->slots[level] = 0;
7443 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7449 wc->refs[level - 1] = 0;
7450 wc->flags[level - 1] = 0;
7451 if (wc->stage == DROP_REFERENCE) {
7452 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7453 parent = path->nodes[level]->start;
7455 BUG_ON(root->root_key.objectid !=
7456 btrfs_header_owner(path->nodes[level]));
7460 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7461 root->root_key.objectid, level - 1, 0, 0);
7462 BUG_ON(ret); /* -ENOMEM */
7464 btrfs_tree_unlock(next);
7465 free_extent_buffer(next);
7471 * helper to process tree block while walking up the tree.
7473 * when wc->stage == DROP_REFERENCE, this function drops
7474 * reference count on the block.
7476 * when wc->stage == UPDATE_BACKREF, this function changes
7477 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7478 * to UPDATE_BACKREF previously while processing the block.
7480 * NOTE: return value 1 means we should stop walking up.
7482 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7483 struct btrfs_root *root,
7484 struct btrfs_path *path,
7485 struct walk_control *wc)
7488 int level = wc->level;
7489 struct extent_buffer *eb = path->nodes[level];
7492 if (wc->stage == UPDATE_BACKREF) {
7493 BUG_ON(wc->shared_level < level);
7494 if (level < wc->shared_level)
7497 ret = find_next_key(path, level + 1, &wc->update_progress);
7501 wc->stage = DROP_REFERENCE;
7502 wc->shared_level = -1;
7503 path->slots[level] = 0;
7506 * check reference count again if the block isn't locked.
7507 * we should start walking down the tree again if reference
7510 if (!path->locks[level]) {
7512 btrfs_tree_lock(eb);
7513 btrfs_set_lock_blocking(eb);
7514 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7516 ret = btrfs_lookup_extent_info(trans, root,
7517 eb->start, level, 1,
7521 btrfs_tree_unlock_rw(eb, path->locks[level]);
7522 path->locks[level] = 0;
7525 BUG_ON(wc->refs[level] == 0);
7526 if (wc->refs[level] == 1) {
7527 btrfs_tree_unlock_rw(eb, path->locks[level]);
7528 path->locks[level] = 0;
7534 /* wc->stage == DROP_REFERENCE */
7535 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7537 if (wc->refs[level] == 1) {
7539 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7540 ret = btrfs_dec_ref(trans, root, eb, 1,
7543 ret = btrfs_dec_ref(trans, root, eb, 0,
7545 BUG_ON(ret); /* -ENOMEM */
7547 /* make block locked assertion in clean_tree_block happy */
7548 if (!path->locks[level] &&
7549 btrfs_header_generation(eb) == trans->transid) {
7550 btrfs_tree_lock(eb);
7551 btrfs_set_lock_blocking(eb);
7552 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7554 clean_tree_block(trans, root, eb);
7557 if (eb == root->node) {
7558 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7561 BUG_ON(root->root_key.objectid !=
7562 btrfs_header_owner(eb));
7564 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7565 parent = path->nodes[level + 1]->start;
7567 BUG_ON(root->root_key.objectid !=
7568 btrfs_header_owner(path->nodes[level + 1]));
7571 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7573 wc->refs[level] = 0;
7574 wc->flags[level] = 0;
7578 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7579 struct btrfs_root *root,
7580 struct btrfs_path *path,
7581 struct walk_control *wc)
7583 int level = wc->level;
7584 int lookup_info = 1;
7587 while (level >= 0) {
7588 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7595 if (path->slots[level] >=
7596 btrfs_header_nritems(path->nodes[level]))
7599 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7601 path->slots[level]++;
7610 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7611 struct btrfs_root *root,
7612 struct btrfs_path *path,
7613 struct walk_control *wc, int max_level)
7615 int level = wc->level;
7618 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7619 while (level < max_level && path->nodes[level]) {
7621 if (path->slots[level] + 1 <
7622 btrfs_header_nritems(path->nodes[level])) {
7623 path->slots[level]++;
7626 ret = walk_up_proc(trans, root, path, wc);
7630 if (path->locks[level]) {
7631 btrfs_tree_unlock_rw(path->nodes[level],
7632 path->locks[level]);
7633 path->locks[level] = 0;
7635 free_extent_buffer(path->nodes[level]);
7636 path->nodes[level] = NULL;
7644 * drop a subvolume tree.
7646 * this function traverses the tree freeing any blocks that only
7647 * referenced by the tree.
7649 * when a shared tree block is found. this function decreases its
7650 * reference count by one. if update_ref is true, this function
7651 * also make sure backrefs for the shared block and all lower level
7652 * blocks are properly updated.
7654 * If called with for_reloc == 0, may exit early with -EAGAIN
7656 int btrfs_drop_snapshot(struct btrfs_root *root,
7657 struct btrfs_block_rsv *block_rsv, int update_ref,
7660 struct btrfs_path *path;
7661 struct btrfs_trans_handle *trans;
7662 struct btrfs_root *tree_root = root->fs_info->tree_root;
7663 struct btrfs_root_item *root_item = &root->root_item;
7664 struct walk_control *wc;
7665 struct btrfs_key key;
7669 bool root_dropped = false;
7671 path = btrfs_alloc_path();
7677 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7679 btrfs_free_path(path);
7684 trans = btrfs_start_transaction(tree_root, 0);
7685 if (IS_ERR(trans)) {
7686 err = PTR_ERR(trans);
7691 trans->block_rsv = block_rsv;
7693 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7694 level = btrfs_header_level(root->node);
7695 path->nodes[level] = btrfs_lock_root_node(root);
7696 btrfs_set_lock_blocking(path->nodes[level]);
7697 path->slots[level] = 0;
7698 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7699 memset(&wc->update_progress, 0,
7700 sizeof(wc->update_progress));
7702 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7703 memcpy(&wc->update_progress, &key,
7704 sizeof(wc->update_progress));
7706 level = root_item->drop_level;
7708 path->lowest_level = level;
7709 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7710 path->lowest_level = 0;
7718 * unlock our path, this is safe because only this
7719 * function is allowed to delete this snapshot
7721 btrfs_unlock_up_safe(path, 0);
7723 level = btrfs_header_level(root->node);
7725 btrfs_tree_lock(path->nodes[level]);
7726 btrfs_set_lock_blocking(path->nodes[level]);
7727 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7729 ret = btrfs_lookup_extent_info(trans, root,
7730 path->nodes[level]->start,
7731 level, 1, &wc->refs[level],
7737 BUG_ON(wc->refs[level] == 0);
7739 if (level == root_item->drop_level)
7742 btrfs_tree_unlock(path->nodes[level]);
7743 path->locks[level] = 0;
7744 WARN_ON(wc->refs[level] != 1);
7750 wc->shared_level = -1;
7751 wc->stage = DROP_REFERENCE;
7752 wc->update_ref = update_ref;
7754 wc->for_reloc = for_reloc;
7755 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7759 ret = walk_down_tree(trans, root, path, wc);
7765 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7772 BUG_ON(wc->stage != DROP_REFERENCE);
7776 if (wc->stage == DROP_REFERENCE) {
7778 btrfs_node_key(path->nodes[level],
7779 &root_item->drop_progress,
7780 path->slots[level]);
7781 root_item->drop_level = level;
7784 BUG_ON(wc->level == 0);
7785 if (btrfs_should_end_transaction(trans, tree_root) ||
7786 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
7787 ret = btrfs_update_root(trans, tree_root,
7791 btrfs_abort_transaction(trans, tree_root, ret);
7796 btrfs_end_transaction_throttle(trans, tree_root);
7797 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7798 pr_debug("BTRFS: drop snapshot early exit\n");
7803 trans = btrfs_start_transaction(tree_root, 0);
7804 if (IS_ERR(trans)) {
7805 err = PTR_ERR(trans);
7809 trans->block_rsv = block_rsv;
7812 btrfs_release_path(path);
7816 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7818 btrfs_abort_transaction(trans, tree_root, ret);
7822 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7823 ret = btrfs_find_root(tree_root, &root->root_key, path,
7826 btrfs_abort_transaction(trans, tree_root, ret);
7829 } else if (ret > 0) {
7830 /* if we fail to delete the orphan item this time
7831 * around, it'll get picked up the next time.
7833 * The most common failure here is just -ENOENT.
7835 btrfs_del_orphan_item(trans, tree_root,
7836 root->root_key.objectid);
7840 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
7841 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
7843 free_extent_buffer(root->node);
7844 free_extent_buffer(root->commit_root);
7845 btrfs_put_fs_root(root);
7847 root_dropped = true;
7849 btrfs_end_transaction_throttle(trans, tree_root);
7852 btrfs_free_path(path);
7855 * So if we need to stop dropping the snapshot for whatever reason we
7856 * need to make sure to add it back to the dead root list so that we
7857 * keep trying to do the work later. This also cleans up roots if we
7858 * don't have it in the radix (like when we recover after a power fail
7859 * or unmount) so we don't leak memory.
7861 if (!for_reloc && root_dropped == false)
7862 btrfs_add_dead_root(root);
7863 if (err && err != -EAGAIN)
7864 btrfs_std_error(root->fs_info, err);
7869 * drop subtree rooted at tree block 'node'.
7871 * NOTE: this function will unlock and release tree block 'node'
7872 * only used by relocation code
7874 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7875 struct btrfs_root *root,
7876 struct extent_buffer *node,
7877 struct extent_buffer *parent)
7879 struct btrfs_path *path;
7880 struct walk_control *wc;
7886 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7888 path = btrfs_alloc_path();
7892 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7894 btrfs_free_path(path);
7898 btrfs_assert_tree_locked(parent);
7899 parent_level = btrfs_header_level(parent);
7900 extent_buffer_get(parent);
7901 path->nodes[parent_level] = parent;
7902 path->slots[parent_level] = btrfs_header_nritems(parent);
7904 btrfs_assert_tree_locked(node);
7905 level = btrfs_header_level(node);
7906 path->nodes[level] = node;
7907 path->slots[level] = 0;
7908 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7910 wc->refs[parent_level] = 1;
7911 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7913 wc->shared_level = -1;
7914 wc->stage = DROP_REFERENCE;
7918 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7921 wret = walk_down_tree(trans, root, path, wc);
7927 wret = walk_up_tree(trans, root, path, wc, parent_level);
7935 btrfs_free_path(path);
7939 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7945 * if restripe for this chunk_type is on pick target profile and
7946 * return, otherwise do the usual balance
7948 stripped = get_restripe_target(root->fs_info, flags);
7950 return extended_to_chunk(stripped);
7953 * we add in the count of missing devices because we want
7954 * to make sure that any RAID levels on a degraded FS
7955 * continue to be honored.
7957 num_devices = root->fs_info->fs_devices->rw_devices +
7958 root->fs_info->fs_devices->missing_devices;
7960 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7961 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7962 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7964 if (num_devices == 1) {
7965 stripped |= BTRFS_BLOCK_GROUP_DUP;
7966 stripped = flags & ~stripped;
7968 /* turn raid0 into single device chunks */
7969 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7972 /* turn mirroring into duplication */
7973 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7974 BTRFS_BLOCK_GROUP_RAID10))
7975 return stripped | BTRFS_BLOCK_GROUP_DUP;
7977 /* they already had raid on here, just return */
7978 if (flags & stripped)
7981 stripped |= BTRFS_BLOCK_GROUP_DUP;
7982 stripped = flags & ~stripped;
7984 /* switch duplicated blocks with raid1 */
7985 if (flags & BTRFS_BLOCK_GROUP_DUP)
7986 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7988 /* this is drive concat, leave it alone */
7994 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7996 struct btrfs_space_info *sinfo = cache->space_info;
7998 u64 min_allocable_bytes;
8003 * We need some metadata space and system metadata space for
8004 * allocating chunks in some corner cases until we force to set
8005 * it to be readonly.
8008 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8010 min_allocable_bytes = 1 * 1024 * 1024;
8012 min_allocable_bytes = 0;
8014 spin_lock(&sinfo->lock);
8015 spin_lock(&cache->lock);
8022 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8023 cache->bytes_super - btrfs_block_group_used(&cache->item);
8025 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8026 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8027 min_allocable_bytes <= sinfo->total_bytes) {
8028 sinfo->bytes_readonly += num_bytes;
8033 spin_unlock(&cache->lock);
8034 spin_unlock(&sinfo->lock);
8038 int btrfs_set_block_group_ro(struct btrfs_root *root,
8039 struct btrfs_block_group_cache *cache)
8042 struct btrfs_trans_handle *trans;
8048 trans = btrfs_join_transaction(root);
8050 return PTR_ERR(trans);
8052 alloc_flags = update_block_group_flags(root, cache->flags);
8053 if (alloc_flags != cache->flags) {
8054 ret = do_chunk_alloc(trans, root, alloc_flags,
8060 ret = set_block_group_ro(cache, 0);
8063 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8064 ret = do_chunk_alloc(trans, root, alloc_flags,
8068 ret = set_block_group_ro(cache, 0);
8070 btrfs_end_transaction(trans, root);
8074 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8075 struct btrfs_root *root, u64 type)
8077 u64 alloc_flags = get_alloc_profile(root, type);
8078 return do_chunk_alloc(trans, root, alloc_flags,
8083 * helper to account the unused space of all the readonly block group in the
8084 * list. takes mirrors into account.
8086 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8088 struct btrfs_block_group_cache *block_group;
8092 list_for_each_entry(block_group, groups_list, list) {
8093 spin_lock(&block_group->lock);
8095 if (!block_group->ro) {
8096 spin_unlock(&block_group->lock);
8100 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8101 BTRFS_BLOCK_GROUP_RAID10 |
8102 BTRFS_BLOCK_GROUP_DUP))
8107 free_bytes += (block_group->key.offset -
8108 btrfs_block_group_used(&block_group->item)) *
8111 spin_unlock(&block_group->lock);
8118 * helper to account the unused space of all the readonly block group in the
8119 * space_info. takes mirrors into account.
8121 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8126 spin_lock(&sinfo->lock);
8128 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8129 if (!list_empty(&sinfo->block_groups[i]))
8130 free_bytes += __btrfs_get_ro_block_group_free_space(
8131 &sinfo->block_groups[i]);
8133 spin_unlock(&sinfo->lock);
8138 void btrfs_set_block_group_rw(struct btrfs_root *root,
8139 struct btrfs_block_group_cache *cache)
8141 struct btrfs_space_info *sinfo = cache->space_info;
8146 spin_lock(&sinfo->lock);
8147 spin_lock(&cache->lock);
8148 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8149 cache->bytes_super - btrfs_block_group_used(&cache->item);
8150 sinfo->bytes_readonly -= num_bytes;
8152 spin_unlock(&cache->lock);
8153 spin_unlock(&sinfo->lock);
8157 * checks to see if its even possible to relocate this block group.
8159 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8160 * ok to go ahead and try.
8162 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8164 struct btrfs_block_group_cache *block_group;
8165 struct btrfs_space_info *space_info;
8166 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8167 struct btrfs_device *device;
8168 struct btrfs_trans_handle *trans;
8177 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8179 /* odd, couldn't find the block group, leave it alone */
8183 min_free = btrfs_block_group_used(&block_group->item);
8185 /* no bytes used, we're good */
8189 space_info = block_group->space_info;
8190 spin_lock(&space_info->lock);
8192 full = space_info->full;
8195 * if this is the last block group we have in this space, we can't
8196 * relocate it unless we're able to allocate a new chunk below.
8198 * Otherwise, we need to make sure we have room in the space to handle
8199 * all of the extents from this block group. If we can, we're good
8201 if ((space_info->total_bytes != block_group->key.offset) &&
8202 (space_info->bytes_used + space_info->bytes_reserved +
8203 space_info->bytes_pinned + space_info->bytes_readonly +
8204 min_free < space_info->total_bytes)) {
8205 spin_unlock(&space_info->lock);
8208 spin_unlock(&space_info->lock);
8211 * ok we don't have enough space, but maybe we have free space on our
8212 * devices to allocate new chunks for relocation, so loop through our
8213 * alloc devices and guess if we have enough space. if this block
8214 * group is going to be restriped, run checks against the target
8215 * profile instead of the current one.
8227 target = get_restripe_target(root->fs_info, block_group->flags);
8229 index = __get_raid_index(extended_to_chunk(target));
8232 * this is just a balance, so if we were marked as full
8233 * we know there is no space for a new chunk
8238 index = get_block_group_index(block_group);
8241 if (index == BTRFS_RAID_RAID10) {
8245 } else if (index == BTRFS_RAID_RAID1) {
8247 } else if (index == BTRFS_RAID_DUP) {
8250 } else if (index == BTRFS_RAID_RAID0) {
8251 dev_min = fs_devices->rw_devices;
8252 do_div(min_free, dev_min);
8255 /* We need to do this so that we can look at pending chunks */
8256 trans = btrfs_join_transaction(root);
8257 if (IS_ERR(trans)) {
8258 ret = PTR_ERR(trans);
8262 mutex_lock(&root->fs_info->chunk_mutex);
8263 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8267 * check to make sure we can actually find a chunk with enough
8268 * space to fit our block group in.
8270 if (device->total_bytes > device->bytes_used + min_free &&
8271 !device->is_tgtdev_for_dev_replace) {
8272 ret = find_free_dev_extent(trans, device, min_free,
8277 if (dev_nr >= dev_min)
8283 mutex_unlock(&root->fs_info->chunk_mutex);
8284 btrfs_end_transaction(trans, root);
8286 btrfs_put_block_group(block_group);
8290 static int find_first_block_group(struct btrfs_root *root,
8291 struct btrfs_path *path, struct btrfs_key *key)
8294 struct btrfs_key found_key;
8295 struct extent_buffer *leaf;
8298 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8303 slot = path->slots[0];
8304 leaf = path->nodes[0];
8305 if (slot >= btrfs_header_nritems(leaf)) {
8306 ret = btrfs_next_leaf(root, path);
8313 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8315 if (found_key.objectid >= key->objectid &&
8316 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8326 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8328 struct btrfs_block_group_cache *block_group;
8332 struct inode *inode;
8334 block_group = btrfs_lookup_first_block_group(info, last);
8335 while (block_group) {
8336 spin_lock(&block_group->lock);
8337 if (block_group->iref)
8339 spin_unlock(&block_group->lock);
8340 block_group = next_block_group(info->tree_root,
8350 inode = block_group->inode;
8351 block_group->iref = 0;
8352 block_group->inode = NULL;
8353 spin_unlock(&block_group->lock);
8355 last = block_group->key.objectid + block_group->key.offset;
8356 btrfs_put_block_group(block_group);
8360 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8362 struct btrfs_block_group_cache *block_group;
8363 struct btrfs_space_info *space_info;
8364 struct btrfs_caching_control *caching_ctl;
8367 down_write(&info->commit_root_sem);
8368 while (!list_empty(&info->caching_block_groups)) {
8369 caching_ctl = list_entry(info->caching_block_groups.next,
8370 struct btrfs_caching_control, list);
8371 list_del(&caching_ctl->list);
8372 put_caching_control(caching_ctl);
8374 up_write(&info->commit_root_sem);
8376 spin_lock(&info->block_group_cache_lock);
8377 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8378 block_group = rb_entry(n, struct btrfs_block_group_cache,
8380 rb_erase(&block_group->cache_node,
8381 &info->block_group_cache_tree);
8382 spin_unlock(&info->block_group_cache_lock);
8384 down_write(&block_group->space_info->groups_sem);
8385 list_del(&block_group->list);
8386 up_write(&block_group->space_info->groups_sem);
8388 if (block_group->cached == BTRFS_CACHE_STARTED)
8389 wait_block_group_cache_done(block_group);
8392 * We haven't cached this block group, which means we could
8393 * possibly have excluded extents on this block group.
8395 if (block_group->cached == BTRFS_CACHE_NO ||
8396 block_group->cached == BTRFS_CACHE_ERROR)
8397 free_excluded_extents(info->extent_root, block_group);
8399 btrfs_remove_free_space_cache(block_group);
8400 btrfs_put_block_group(block_group);
8402 spin_lock(&info->block_group_cache_lock);
8404 spin_unlock(&info->block_group_cache_lock);
8406 /* now that all the block groups are freed, go through and
8407 * free all the space_info structs. This is only called during
8408 * the final stages of unmount, and so we know nobody is
8409 * using them. We call synchronize_rcu() once before we start,
8410 * just to be on the safe side.
8414 release_global_block_rsv(info);
8416 while (!list_empty(&info->space_info)) {
8419 space_info = list_entry(info->space_info.next,
8420 struct btrfs_space_info,
8422 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8423 if (WARN_ON(space_info->bytes_pinned > 0 ||
8424 space_info->bytes_reserved > 0 ||
8425 space_info->bytes_may_use > 0)) {
8426 dump_space_info(space_info, 0, 0);
8429 list_del(&space_info->list);
8430 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
8431 struct kobject *kobj;
8432 kobj = &space_info->block_group_kobjs[i];
8438 kobject_del(&space_info->kobj);
8439 kobject_put(&space_info->kobj);
8444 static void __link_block_group(struct btrfs_space_info *space_info,
8445 struct btrfs_block_group_cache *cache)
8447 int index = get_block_group_index(cache);
8450 down_write(&space_info->groups_sem);
8451 if (list_empty(&space_info->block_groups[index]))
8453 list_add_tail(&cache->list, &space_info->block_groups[index]);
8454 up_write(&space_info->groups_sem);
8457 struct kobject *kobj = &space_info->block_group_kobjs[index];
8460 kobject_get(&space_info->kobj); /* put in release */
8461 ret = kobject_add(kobj, &space_info->kobj, "%s",
8462 get_raid_name(index));
8464 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
8465 kobject_put(&space_info->kobj);
8470 static struct btrfs_block_group_cache *
8471 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
8473 struct btrfs_block_group_cache *cache;
8475 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8479 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8481 if (!cache->free_space_ctl) {
8486 cache->key.objectid = start;
8487 cache->key.offset = size;
8488 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8490 cache->sectorsize = root->sectorsize;
8491 cache->fs_info = root->fs_info;
8492 cache->full_stripe_len = btrfs_full_stripe_len(root,
8493 &root->fs_info->mapping_tree,
8495 atomic_set(&cache->count, 1);
8496 spin_lock_init(&cache->lock);
8497 INIT_LIST_HEAD(&cache->list);
8498 INIT_LIST_HEAD(&cache->cluster_list);
8499 INIT_LIST_HEAD(&cache->new_bg_list);
8500 btrfs_init_free_space_ctl(cache);
8505 int btrfs_read_block_groups(struct btrfs_root *root)
8507 struct btrfs_path *path;
8509 struct btrfs_block_group_cache *cache;
8510 struct btrfs_fs_info *info = root->fs_info;
8511 struct btrfs_space_info *space_info;
8512 struct btrfs_key key;
8513 struct btrfs_key found_key;
8514 struct extent_buffer *leaf;
8518 root = info->extent_root;
8521 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8522 path = btrfs_alloc_path();
8527 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8528 if (btrfs_test_opt(root, SPACE_CACHE) &&
8529 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8531 if (btrfs_test_opt(root, CLEAR_CACHE))
8535 ret = find_first_block_group(root, path, &key);
8541 leaf = path->nodes[0];
8542 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8544 cache = btrfs_create_block_group_cache(root, found_key.objectid,
8553 * When we mount with old space cache, we need to
8554 * set BTRFS_DC_CLEAR and set dirty flag.
8556 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8557 * truncate the old free space cache inode and
8559 * b) Setting 'dirty flag' makes sure that we flush
8560 * the new space cache info onto disk.
8562 cache->disk_cache_state = BTRFS_DC_CLEAR;
8563 if (btrfs_test_opt(root, SPACE_CACHE))
8567 read_extent_buffer(leaf, &cache->item,
8568 btrfs_item_ptr_offset(leaf, path->slots[0]),
8569 sizeof(cache->item));
8570 cache->flags = btrfs_block_group_flags(&cache->item);
8572 key.objectid = found_key.objectid + found_key.offset;
8573 btrfs_release_path(path);
8576 * We need to exclude the super stripes now so that the space
8577 * info has super bytes accounted for, otherwise we'll think
8578 * we have more space than we actually do.
8580 ret = exclude_super_stripes(root, cache);
8583 * We may have excluded something, so call this just in
8586 free_excluded_extents(root, cache);
8587 btrfs_put_block_group(cache);
8592 * check for two cases, either we are full, and therefore
8593 * don't need to bother with the caching work since we won't
8594 * find any space, or we are empty, and we can just add all
8595 * the space in and be done with it. This saves us _alot_ of
8596 * time, particularly in the full case.
8598 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8599 cache->last_byte_to_unpin = (u64)-1;
8600 cache->cached = BTRFS_CACHE_FINISHED;
8601 free_excluded_extents(root, cache);
8602 } else if (btrfs_block_group_used(&cache->item) == 0) {
8603 cache->last_byte_to_unpin = (u64)-1;
8604 cache->cached = BTRFS_CACHE_FINISHED;
8605 add_new_free_space(cache, root->fs_info,
8607 found_key.objectid +
8609 free_excluded_extents(root, cache);
8612 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8614 btrfs_remove_free_space_cache(cache);
8615 btrfs_put_block_group(cache);
8619 ret = update_space_info(info, cache->flags, found_key.offset,
8620 btrfs_block_group_used(&cache->item),
8623 btrfs_remove_free_space_cache(cache);
8624 spin_lock(&info->block_group_cache_lock);
8625 rb_erase(&cache->cache_node,
8626 &info->block_group_cache_tree);
8627 spin_unlock(&info->block_group_cache_lock);
8628 btrfs_put_block_group(cache);
8632 cache->space_info = space_info;
8633 spin_lock(&cache->space_info->lock);
8634 cache->space_info->bytes_readonly += cache->bytes_super;
8635 spin_unlock(&cache->space_info->lock);
8637 __link_block_group(space_info, cache);
8639 set_avail_alloc_bits(root->fs_info, cache->flags);
8640 if (btrfs_chunk_readonly(root, cache->key.objectid))
8641 set_block_group_ro(cache, 1);
8644 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8645 if (!(get_alloc_profile(root, space_info->flags) &
8646 (BTRFS_BLOCK_GROUP_RAID10 |
8647 BTRFS_BLOCK_GROUP_RAID1 |
8648 BTRFS_BLOCK_GROUP_RAID5 |
8649 BTRFS_BLOCK_GROUP_RAID6 |
8650 BTRFS_BLOCK_GROUP_DUP)))
8653 * avoid allocating from un-mirrored block group if there are
8654 * mirrored block groups.
8656 list_for_each_entry(cache,
8657 &space_info->block_groups[BTRFS_RAID_RAID0],
8659 set_block_group_ro(cache, 1);
8660 list_for_each_entry(cache,
8661 &space_info->block_groups[BTRFS_RAID_SINGLE],
8663 set_block_group_ro(cache, 1);
8666 init_global_block_rsv(info);
8669 btrfs_free_path(path);
8673 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8674 struct btrfs_root *root)
8676 struct btrfs_block_group_cache *block_group, *tmp;
8677 struct btrfs_root *extent_root = root->fs_info->extent_root;
8678 struct btrfs_block_group_item item;
8679 struct btrfs_key key;
8682 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8684 list_del_init(&block_group->new_bg_list);
8689 spin_lock(&block_group->lock);
8690 memcpy(&item, &block_group->item, sizeof(item));
8691 memcpy(&key, &block_group->key, sizeof(key));
8692 spin_unlock(&block_group->lock);
8694 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8697 btrfs_abort_transaction(trans, extent_root, ret);
8698 ret = btrfs_finish_chunk_alloc(trans, extent_root,
8699 key.objectid, key.offset);
8701 btrfs_abort_transaction(trans, extent_root, ret);
8705 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8706 struct btrfs_root *root, u64 bytes_used,
8707 u64 type, u64 chunk_objectid, u64 chunk_offset,
8711 struct btrfs_root *extent_root;
8712 struct btrfs_block_group_cache *cache;
8714 extent_root = root->fs_info->extent_root;
8716 btrfs_set_log_full_commit(root->fs_info, trans);
8718 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
8722 btrfs_set_block_group_used(&cache->item, bytes_used);
8723 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8724 btrfs_set_block_group_flags(&cache->item, type);
8726 cache->flags = type;
8727 cache->last_byte_to_unpin = (u64)-1;
8728 cache->cached = BTRFS_CACHE_FINISHED;
8729 ret = exclude_super_stripes(root, cache);
8732 * We may have excluded something, so call this just in
8735 free_excluded_extents(root, cache);
8736 btrfs_put_block_group(cache);
8740 add_new_free_space(cache, root->fs_info, chunk_offset,
8741 chunk_offset + size);
8743 free_excluded_extents(root, cache);
8745 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8747 btrfs_remove_free_space_cache(cache);
8748 btrfs_put_block_group(cache);
8752 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8753 &cache->space_info);
8755 btrfs_remove_free_space_cache(cache);
8756 spin_lock(&root->fs_info->block_group_cache_lock);
8757 rb_erase(&cache->cache_node,
8758 &root->fs_info->block_group_cache_tree);
8759 spin_unlock(&root->fs_info->block_group_cache_lock);
8760 btrfs_put_block_group(cache);
8763 update_global_block_rsv(root->fs_info);
8765 spin_lock(&cache->space_info->lock);
8766 cache->space_info->bytes_readonly += cache->bytes_super;
8767 spin_unlock(&cache->space_info->lock);
8769 __link_block_group(cache->space_info, cache);
8771 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8773 set_avail_alloc_bits(extent_root->fs_info, type);
8778 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8780 u64 extra_flags = chunk_to_extended(flags) &
8781 BTRFS_EXTENDED_PROFILE_MASK;
8783 write_seqlock(&fs_info->profiles_lock);
8784 if (flags & BTRFS_BLOCK_GROUP_DATA)
8785 fs_info->avail_data_alloc_bits &= ~extra_flags;
8786 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8787 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8788 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8789 fs_info->avail_system_alloc_bits &= ~extra_flags;
8790 write_sequnlock(&fs_info->profiles_lock);
8793 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8794 struct btrfs_root *root, u64 group_start)
8796 struct btrfs_path *path;
8797 struct btrfs_block_group_cache *block_group;
8798 struct btrfs_free_cluster *cluster;
8799 struct btrfs_root *tree_root = root->fs_info->tree_root;
8800 struct btrfs_key key;
8801 struct inode *inode;
8806 root = root->fs_info->extent_root;
8808 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8809 BUG_ON(!block_group);
8810 BUG_ON(!block_group->ro);
8813 * Free the reserved super bytes from this block group before
8816 free_excluded_extents(root, block_group);
8818 memcpy(&key, &block_group->key, sizeof(key));
8819 index = get_block_group_index(block_group);
8820 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8821 BTRFS_BLOCK_GROUP_RAID1 |
8822 BTRFS_BLOCK_GROUP_RAID10))
8827 /* make sure this block group isn't part of an allocation cluster */
8828 cluster = &root->fs_info->data_alloc_cluster;
8829 spin_lock(&cluster->refill_lock);
8830 btrfs_return_cluster_to_free_space(block_group, cluster);
8831 spin_unlock(&cluster->refill_lock);
8834 * make sure this block group isn't part of a metadata
8835 * allocation cluster
8837 cluster = &root->fs_info->meta_alloc_cluster;
8838 spin_lock(&cluster->refill_lock);
8839 btrfs_return_cluster_to_free_space(block_group, cluster);
8840 spin_unlock(&cluster->refill_lock);
8842 path = btrfs_alloc_path();
8848 inode = lookup_free_space_inode(tree_root, block_group, path);
8849 if (!IS_ERR(inode)) {
8850 ret = btrfs_orphan_add(trans, inode);
8852 btrfs_add_delayed_iput(inode);
8856 /* One for the block groups ref */
8857 spin_lock(&block_group->lock);
8858 if (block_group->iref) {
8859 block_group->iref = 0;
8860 block_group->inode = NULL;
8861 spin_unlock(&block_group->lock);
8864 spin_unlock(&block_group->lock);
8866 /* One for our lookup ref */
8867 btrfs_add_delayed_iput(inode);
8870 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8871 key.offset = block_group->key.objectid;
8874 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8878 btrfs_release_path(path);
8880 ret = btrfs_del_item(trans, tree_root, path);
8883 btrfs_release_path(path);
8886 spin_lock(&root->fs_info->block_group_cache_lock);
8887 rb_erase(&block_group->cache_node,
8888 &root->fs_info->block_group_cache_tree);
8890 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8891 root->fs_info->first_logical_byte = (u64)-1;
8892 spin_unlock(&root->fs_info->block_group_cache_lock);
8894 down_write(&block_group->space_info->groups_sem);
8896 * we must use list_del_init so people can check to see if they
8897 * are still on the list after taking the semaphore
8899 list_del_init(&block_group->list);
8900 if (list_empty(&block_group->space_info->block_groups[index])) {
8901 kobject_del(&block_group->space_info->block_group_kobjs[index]);
8902 kobject_put(&block_group->space_info->block_group_kobjs[index]);
8903 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8905 up_write(&block_group->space_info->groups_sem);
8907 if (block_group->cached == BTRFS_CACHE_STARTED)
8908 wait_block_group_cache_done(block_group);
8910 btrfs_remove_free_space_cache(block_group);
8912 spin_lock(&block_group->space_info->lock);
8913 block_group->space_info->total_bytes -= block_group->key.offset;
8914 block_group->space_info->bytes_readonly -= block_group->key.offset;
8915 block_group->space_info->disk_total -= block_group->key.offset * factor;
8916 spin_unlock(&block_group->space_info->lock);
8918 memcpy(&key, &block_group->key, sizeof(key));
8920 btrfs_clear_space_info_full(root->fs_info);
8922 btrfs_put_block_group(block_group);
8923 btrfs_put_block_group(block_group);
8925 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8931 ret = btrfs_del_item(trans, root, path);
8933 btrfs_free_path(path);
8937 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8939 struct btrfs_space_info *space_info;
8940 struct btrfs_super_block *disk_super;
8946 disk_super = fs_info->super_copy;
8947 if (!btrfs_super_root(disk_super))
8950 features = btrfs_super_incompat_flags(disk_super);
8951 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8954 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8955 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8960 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8961 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8963 flags = BTRFS_BLOCK_GROUP_METADATA;
8964 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8968 flags = BTRFS_BLOCK_GROUP_DATA;
8969 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8975 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8977 return unpin_extent_range(root, start, end);
8980 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8981 u64 num_bytes, u64 *actual_bytes)
8983 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8986 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8988 struct btrfs_fs_info *fs_info = root->fs_info;
8989 struct btrfs_block_group_cache *cache = NULL;
8994 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8998 * try to trim all FS space, our block group may start from non-zero.
9000 if (range->len == total_bytes)
9001 cache = btrfs_lookup_first_block_group(fs_info, range->start);
9003 cache = btrfs_lookup_block_group(fs_info, range->start);
9006 if (cache->key.objectid >= (range->start + range->len)) {
9007 btrfs_put_block_group(cache);
9011 start = max(range->start, cache->key.objectid);
9012 end = min(range->start + range->len,
9013 cache->key.objectid + cache->key.offset);
9015 if (end - start >= range->minlen) {
9016 if (!block_group_cache_done(cache)) {
9017 ret = cache_block_group(cache, 0);
9019 btrfs_put_block_group(cache);
9022 ret = wait_block_group_cache_done(cache);
9024 btrfs_put_block_group(cache);
9028 ret = btrfs_trim_block_group(cache,
9034 trimmed += group_trimmed;
9036 btrfs_put_block_group(cache);
9041 cache = next_block_group(fs_info->tree_root, cache);
9044 range->len = trimmed;
9049 * btrfs_{start,end}_write() is similar to mnt_{want, drop}_write(),
9050 * they are used to prevent the some tasks writing data into the page cache
9051 * by nocow before the subvolume is snapshoted, but flush the data into
9052 * the disk after the snapshot creation.
9054 void btrfs_end_nocow_write(struct btrfs_root *root)
9056 percpu_counter_dec(&root->subv_writers->counter);
9058 * Make sure counter is updated before we wake up
9062 if (waitqueue_active(&root->subv_writers->wait))
9063 wake_up(&root->subv_writers->wait);
9066 int btrfs_start_nocow_write(struct btrfs_root *root)
9068 if (unlikely(atomic_read(&root->will_be_snapshoted)))
9071 percpu_counter_inc(&root->subv_writers->counter);
9073 * Make sure counter is updated before we check for snapshot creation.
9076 if (unlikely(atomic_read(&root->will_be_snapshoted))) {
9077 btrfs_end_nocow_write(root);