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>
31 #include "print-tree.h"
32 #include "transaction.h"
36 #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;
118 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
120 return (cache->flags & bits) == bits;
123 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
125 atomic_inc(&cache->count);
128 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
130 if (atomic_dec_and_test(&cache->count)) {
131 WARN_ON(cache->pinned > 0);
132 WARN_ON(cache->reserved > 0);
133 kfree(cache->free_space_ctl);
139 * this adds the block group to the fs_info rb tree for the block group
142 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
143 struct btrfs_block_group_cache *block_group)
146 struct rb_node *parent = NULL;
147 struct btrfs_block_group_cache *cache;
149 spin_lock(&info->block_group_cache_lock);
150 p = &info->block_group_cache_tree.rb_node;
154 cache = rb_entry(parent, struct btrfs_block_group_cache,
156 if (block_group->key.objectid < cache->key.objectid) {
158 } else if (block_group->key.objectid > cache->key.objectid) {
161 spin_unlock(&info->block_group_cache_lock);
166 rb_link_node(&block_group->cache_node, parent, p);
167 rb_insert_color(&block_group->cache_node,
168 &info->block_group_cache_tree);
170 if (info->first_logical_byte > block_group->key.objectid)
171 info->first_logical_byte = block_group->key.objectid;
173 spin_unlock(&info->block_group_cache_lock);
179 * This will return the block group at or after bytenr if contains is 0, else
180 * it will return the block group that contains the bytenr
182 static struct btrfs_block_group_cache *
183 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
186 struct btrfs_block_group_cache *cache, *ret = NULL;
190 spin_lock(&info->block_group_cache_lock);
191 n = info->block_group_cache_tree.rb_node;
194 cache = rb_entry(n, struct btrfs_block_group_cache,
196 end = cache->key.objectid + cache->key.offset - 1;
197 start = cache->key.objectid;
199 if (bytenr < start) {
200 if (!contains && (!ret || start < ret->key.objectid))
203 } else if (bytenr > start) {
204 if (contains && bytenr <= end) {
215 btrfs_get_block_group(ret);
216 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
217 info->first_logical_byte = ret->key.objectid;
219 spin_unlock(&info->block_group_cache_lock);
224 static int add_excluded_extent(struct btrfs_root *root,
225 u64 start, u64 num_bytes)
227 u64 end = start + num_bytes - 1;
228 set_extent_bits(&root->fs_info->freed_extents[0],
229 start, end, EXTENT_UPTODATE, GFP_NOFS);
230 set_extent_bits(&root->fs_info->freed_extents[1],
231 start, end, EXTENT_UPTODATE, GFP_NOFS);
235 static void free_excluded_extents(struct btrfs_root *root,
236 struct btrfs_block_group_cache *cache)
240 start = cache->key.objectid;
241 end = start + cache->key.offset - 1;
243 clear_extent_bits(&root->fs_info->freed_extents[0],
244 start, end, EXTENT_UPTODATE, GFP_NOFS);
245 clear_extent_bits(&root->fs_info->freed_extents[1],
246 start, end, EXTENT_UPTODATE, GFP_NOFS);
249 static int exclude_super_stripes(struct btrfs_root *root,
250 struct btrfs_block_group_cache *cache)
257 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
258 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
259 cache->bytes_super += stripe_len;
260 ret = add_excluded_extent(root, cache->key.objectid,
266 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
267 bytenr = btrfs_sb_offset(i);
268 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
269 cache->key.objectid, bytenr,
270 0, &logical, &nr, &stripe_len);
277 if (logical[nr] > cache->key.objectid +
281 if (logical[nr] + stripe_len <= cache->key.objectid)
285 if (start < cache->key.objectid) {
286 start = cache->key.objectid;
287 len = (logical[nr] + stripe_len) - start;
289 len = min_t(u64, stripe_len,
290 cache->key.objectid +
291 cache->key.offset - start);
294 cache->bytes_super += len;
295 ret = add_excluded_extent(root, start, len);
307 static struct btrfs_caching_control *
308 get_caching_control(struct btrfs_block_group_cache *cache)
310 struct btrfs_caching_control *ctl;
312 spin_lock(&cache->lock);
313 if (cache->cached != BTRFS_CACHE_STARTED) {
314 spin_unlock(&cache->lock);
318 /* We're loading it the fast way, so we don't have a caching_ctl. */
319 if (!cache->caching_ctl) {
320 spin_unlock(&cache->lock);
324 ctl = cache->caching_ctl;
325 atomic_inc(&ctl->count);
326 spin_unlock(&cache->lock);
330 static void put_caching_control(struct btrfs_caching_control *ctl)
332 if (atomic_dec_and_test(&ctl->count))
337 * this is only called by cache_block_group, since we could have freed extents
338 * we need to check the pinned_extents for any extents that can't be used yet
339 * since their free space will be released as soon as the transaction commits.
341 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
342 struct btrfs_fs_info *info, u64 start, u64 end)
344 u64 extent_start, extent_end, size, total_added = 0;
347 while (start < end) {
348 ret = find_first_extent_bit(info->pinned_extents, start,
349 &extent_start, &extent_end,
350 EXTENT_DIRTY | EXTENT_UPTODATE,
355 if (extent_start <= start) {
356 start = extent_end + 1;
357 } else if (extent_start > start && extent_start < end) {
358 size = extent_start - start;
360 ret = btrfs_add_free_space(block_group, start,
362 BUG_ON(ret); /* -ENOMEM or logic error */
363 start = extent_end + 1;
372 ret = btrfs_add_free_space(block_group, start, size);
373 BUG_ON(ret); /* -ENOMEM or logic error */
379 static noinline void caching_thread(struct btrfs_work *work)
381 struct btrfs_block_group_cache *block_group;
382 struct btrfs_fs_info *fs_info;
383 struct btrfs_caching_control *caching_ctl;
384 struct btrfs_root *extent_root;
385 struct btrfs_path *path;
386 struct extent_buffer *leaf;
387 struct btrfs_key key;
393 caching_ctl = container_of(work, struct btrfs_caching_control, work);
394 block_group = caching_ctl->block_group;
395 fs_info = block_group->fs_info;
396 extent_root = fs_info->extent_root;
398 path = btrfs_alloc_path();
402 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
405 * We don't want to deadlock with somebody trying to allocate a new
406 * extent for the extent root while also trying to search the extent
407 * root to add free space. So we skip locking and search the commit
408 * root, since its read-only
410 path->skip_locking = 1;
411 path->search_commit_root = 1;
416 key.type = BTRFS_EXTENT_ITEM_KEY;
418 mutex_lock(&caching_ctl->mutex);
419 /* need to make sure the commit_root doesn't disappear */
420 down_read(&fs_info->extent_commit_sem);
422 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
426 leaf = path->nodes[0];
427 nritems = btrfs_header_nritems(leaf);
430 if (btrfs_fs_closing(fs_info) > 1) {
435 if (path->slots[0] < nritems) {
436 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
438 ret = find_next_key(path, 0, &key);
442 if (need_resched()) {
443 caching_ctl->progress = last;
444 btrfs_release_path(path);
445 up_read(&fs_info->extent_commit_sem);
446 mutex_unlock(&caching_ctl->mutex);
451 ret = btrfs_next_leaf(extent_root, path);
456 leaf = path->nodes[0];
457 nritems = btrfs_header_nritems(leaf);
461 if (key.objectid < block_group->key.objectid) {
466 if (key.objectid >= block_group->key.objectid +
467 block_group->key.offset)
470 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
471 key.type == BTRFS_METADATA_ITEM_KEY) {
472 total_found += add_new_free_space(block_group,
475 if (key.type == BTRFS_METADATA_ITEM_KEY)
476 last = key.objectid +
477 fs_info->tree_root->leafsize;
479 last = key.objectid + key.offset;
481 if (total_found > (1024 * 1024 * 2)) {
483 wake_up(&caching_ctl->wait);
490 total_found += add_new_free_space(block_group, fs_info, last,
491 block_group->key.objectid +
492 block_group->key.offset);
493 caching_ctl->progress = (u64)-1;
495 spin_lock(&block_group->lock);
496 block_group->caching_ctl = NULL;
497 block_group->cached = BTRFS_CACHE_FINISHED;
498 spin_unlock(&block_group->lock);
501 btrfs_free_path(path);
502 up_read(&fs_info->extent_commit_sem);
504 free_excluded_extents(extent_root, block_group);
506 mutex_unlock(&caching_ctl->mutex);
508 wake_up(&caching_ctl->wait);
510 put_caching_control(caching_ctl);
511 btrfs_put_block_group(block_group);
514 static int cache_block_group(struct btrfs_block_group_cache *cache,
518 struct btrfs_fs_info *fs_info = cache->fs_info;
519 struct btrfs_caching_control *caching_ctl;
522 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
526 INIT_LIST_HEAD(&caching_ctl->list);
527 mutex_init(&caching_ctl->mutex);
528 init_waitqueue_head(&caching_ctl->wait);
529 caching_ctl->block_group = cache;
530 caching_ctl->progress = cache->key.objectid;
531 atomic_set(&caching_ctl->count, 1);
532 caching_ctl->work.func = caching_thread;
534 spin_lock(&cache->lock);
536 * This should be a rare occasion, but this could happen I think in the
537 * case where one thread starts to load the space cache info, and then
538 * some other thread starts a transaction commit which tries to do an
539 * allocation while the other thread is still loading the space cache
540 * info. The previous loop should have kept us from choosing this block
541 * group, but if we've moved to the state where we will wait on caching
542 * block groups we need to first check if we're doing a fast load here,
543 * so we can wait for it to finish, otherwise we could end up allocating
544 * from a block group who's cache gets evicted for one reason or
547 while (cache->cached == BTRFS_CACHE_FAST) {
548 struct btrfs_caching_control *ctl;
550 ctl = cache->caching_ctl;
551 atomic_inc(&ctl->count);
552 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
553 spin_unlock(&cache->lock);
557 finish_wait(&ctl->wait, &wait);
558 put_caching_control(ctl);
559 spin_lock(&cache->lock);
562 if (cache->cached != BTRFS_CACHE_NO) {
563 spin_unlock(&cache->lock);
567 WARN_ON(cache->caching_ctl);
568 cache->caching_ctl = caching_ctl;
569 cache->cached = BTRFS_CACHE_FAST;
570 spin_unlock(&cache->lock);
572 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
573 ret = load_free_space_cache(fs_info, cache);
575 spin_lock(&cache->lock);
577 cache->caching_ctl = NULL;
578 cache->cached = BTRFS_CACHE_FINISHED;
579 cache->last_byte_to_unpin = (u64)-1;
581 if (load_cache_only) {
582 cache->caching_ctl = NULL;
583 cache->cached = BTRFS_CACHE_NO;
585 cache->cached = BTRFS_CACHE_STARTED;
588 spin_unlock(&cache->lock);
589 wake_up(&caching_ctl->wait);
591 put_caching_control(caching_ctl);
592 free_excluded_extents(fs_info->extent_root, cache);
597 * We are not going to do the fast caching, set cached to the
598 * appropriate value and wakeup any waiters.
600 spin_lock(&cache->lock);
601 if (load_cache_only) {
602 cache->caching_ctl = NULL;
603 cache->cached = BTRFS_CACHE_NO;
605 cache->cached = BTRFS_CACHE_STARTED;
607 spin_unlock(&cache->lock);
608 wake_up(&caching_ctl->wait);
611 if (load_cache_only) {
612 put_caching_control(caching_ctl);
616 down_write(&fs_info->extent_commit_sem);
617 atomic_inc(&caching_ctl->count);
618 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
619 up_write(&fs_info->extent_commit_sem);
621 btrfs_get_block_group(cache);
623 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
629 * return the block group that starts at or after bytenr
631 static struct btrfs_block_group_cache *
632 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
634 struct btrfs_block_group_cache *cache;
636 cache = block_group_cache_tree_search(info, bytenr, 0);
642 * return the block group that contains the given bytenr
644 struct btrfs_block_group_cache *btrfs_lookup_block_group(
645 struct btrfs_fs_info *info,
648 struct btrfs_block_group_cache *cache;
650 cache = block_group_cache_tree_search(info, bytenr, 1);
655 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
658 struct list_head *head = &info->space_info;
659 struct btrfs_space_info *found;
661 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
664 list_for_each_entry_rcu(found, head, list) {
665 if (found->flags & flags) {
675 * after adding space to the filesystem, we need to clear the full flags
676 * on all the space infos.
678 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
680 struct list_head *head = &info->space_info;
681 struct btrfs_space_info *found;
684 list_for_each_entry_rcu(found, head, list)
689 /* simple helper to search for an existing extent at a given offset */
690 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
693 struct btrfs_key key;
694 struct btrfs_path *path;
696 path = btrfs_alloc_path();
700 key.objectid = start;
702 key.type = BTRFS_EXTENT_ITEM_KEY;
703 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
706 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
707 if (key.objectid == start &&
708 key.type == BTRFS_METADATA_ITEM_KEY)
711 btrfs_free_path(path);
716 * helper function to lookup reference count and flags of a tree block.
718 * the head node for delayed ref is used to store the sum of all the
719 * reference count modifications queued up in the rbtree. the head
720 * node may also store the extent flags to set. This way you can check
721 * to see what the reference count and extent flags would be if all of
722 * the delayed refs are not processed.
724 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
725 struct btrfs_root *root, u64 bytenr,
726 u64 offset, int metadata, u64 *refs, u64 *flags)
728 struct btrfs_delayed_ref_head *head;
729 struct btrfs_delayed_ref_root *delayed_refs;
730 struct btrfs_path *path;
731 struct btrfs_extent_item *ei;
732 struct extent_buffer *leaf;
733 struct btrfs_key key;
740 * If we don't have skinny metadata, don't bother doing anything
743 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
744 offset = root->leafsize;
748 path = btrfs_alloc_path();
753 key.objectid = bytenr;
754 key.type = BTRFS_METADATA_ITEM_KEY;
757 key.objectid = bytenr;
758 key.type = BTRFS_EXTENT_ITEM_KEY;
763 path->skip_locking = 1;
764 path->search_commit_root = 1;
767 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
772 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
773 key.type = BTRFS_EXTENT_ITEM_KEY;
774 key.offset = root->leafsize;
775 btrfs_release_path(path);
780 leaf = path->nodes[0];
781 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
782 if (item_size >= sizeof(*ei)) {
783 ei = btrfs_item_ptr(leaf, path->slots[0],
784 struct btrfs_extent_item);
785 num_refs = btrfs_extent_refs(leaf, ei);
786 extent_flags = btrfs_extent_flags(leaf, ei);
788 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
789 struct btrfs_extent_item_v0 *ei0;
790 BUG_ON(item_size != sizeof(*ei0));
791 ei0 = btrfs_item_ptr(leaf, path->slots[0],
792 struct btrfs_extent_item_v0);
793 num_refs = btrfs_extent_refs_v0(leaf, ei0);
794 /* FIXME: this isn't correct for data */
795 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
800 BUG_ON(num_refs == 0);
810 delayed_refs = &trans->transaction->delayed_refs;
811 spin_lock(&delayed_refs->lock);
812 head = btrfs_find_delayed_ref_head(trans, bytenr);
814 if (!mutex_trylock(&head->mutex)) {
815 atomic_inc(&head->node.refs);
816 spin_unlock(&delayed_refs->lock);
818 btrfs_release_path(path);
821 * Mutex was contended, block until it's released and try
824 mutex_lock(&head->mutex);
825 mutex_unlock(&head->mutex);
826 btrfs_put_delayed_ref(&head->node);
829 if (head->extent_op && head->extent_op->update_flags)
830 extent_flags |= head->extent_op->flags_to_set;
832 BUG_ON(num_refs == 0);
834 num_refs += head->node.ref_mod;
835 mutex_unlock(&head->mutex);
837 spin_unlock(&delayed_refs->lock);
839 WARN_ON(num_refs == 0);
843 *flags = extent_flags;
845 btrfs_free_path(path);
850 * Back reference rules. Back refs have three main goals:
852 * 1) differentiate between all holders of references to an extent so that
853 * when a reference is dropped we can make sure it was a valid reference
854 * before freeing the extent.
856 * 2) Provide enough information to quickly find the holders of an extent
857 * if we notice a given block is corrupted or bad.
859 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
860 * maintenance. This is actually the same as #2, but with a slightly
861 * different use case.
863 * There are two kinds of back refs. The implicit back refs is optimized
864 * for pointers in non-shared tree blocks. For a given pointer in a block,
865 * back refs of this kind provide information about the block's owner tree
866 * and the pointer's key. These information allow us to find the block by
867 * b-tree searching. The full back refs is for pointers in tree blocks not
868 * referenced by their owner trees. The location of tree block is recorded
869 * in the back refs. Actually the full back refs is generic, and can be
870 * used in all cases the implicit back refs is used. The major shortcoming
871 * of the full back refs is its overhead. Every time a tree block gets
872 * COWed, we have to update back refs entry for all pointers in it.
874 * For a newly allocated tree block, we use implicit back refs for
875 * pointers in it. This means most tree related operations only involve
876 * implicit back refs. For a tree block created in old transaction, the
877 * only way to drop a reference to it is COW it. So we can detect the
878 * event that tree block loses its owner tree's reference and do the
879 * back refs conversion.
881 * When a tree block is COW'd through a tree, there are four cases:
883 * The reference count of the block is one and the tree is the block's
884 * owner tree. Nothing to do in this case.
886 * The reference count of the block is one and the tree is not the
887 * block's owner tree. In this case, full back refs is used for pointers
888 * in the block. Remove these full back refs, add implicit back refs for
889 * every pointers in the new block.
891 * The reference count of the block is greater than one and the tree is
892 * the block's owner tree. In this case, implicit back refs is used for
893 * pointers in the block. Add full back refs for every pointers in the
894 * block, increase lower level extents' reference counts. The original
895 * implicit back refs are entailed to the new block.
897 * The reference count of the block is greater than one and the tree is
898 * not the block's owner tree. Add implicit back refs for every pointer in
899 * the new block, increase lower level extents' reference count.
901 * Back Reference Key composing:
903 * The key objectid corresponds to the first byte in the extent,
904 * The key type is used to differentiate between types of back refs.
905 * There are different meanings of the key offset for different types
908 * File extents can be referenced by:
910 * - multiple snapshots, subvolumes, or different generations in one subvol
911 * - different files inside a single subvolume
912 * - different offsets inside a file (bookend extents in file.c)
914 * The extent ref structure for the implicit back refs has fields for:
916 * - Objectid of the subvolume root
917 * - objectid of the file holding the reference
918 * - original offset in the file
919 * - how many bookend extents
921 * The key offset for the implicit back refs is hash of the first
924 * The extent ref structure for the full back refs has field for:
926 * - number of pointers in the tree leaf
928 * The key offset for the implicit back refs is the first byte of
931 * When a file extent is allocated, The implicit back refs is used.
932 * the fields are filled in:
934 * (root_key.objectid, inode objectid, offset in file, 1)
936 * When a file extent is removed file truncation, we find the
937 * corresponding implicit back refs and check the following fields:
939 * (btrfs_header_owner(leaf), inode objectid, offset in file)
941 * Btree extents can be referenced by:
943 * - Different subvolumes
945 * Both the implicit back refs and the full back refs for tree blocks
946 * only consist of key. The key offset for the implicit back refs is
947 * objectid of block's owner tree. The key offset for the full back refs
948 * is the first byte of parent block.
950 * When implicit back refs is used, information about the lowest key and
951 * level of the tree block are required. These information are stored in
952 * tree block info structure.
955 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
956 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
957 struct btrfs_root *root,
958 struct btrfs_path *path,
959 u64 owner, u32 extra_size)
961 struct btrfs_extent_item *item;
962 struct btrfs_extent_item_v0 *ei0;
963 struct btrfs_extent_ref_v0 *ref0;
964 struct btrfs_tree_block_info *bi;
965 struct extent_buffer *leaf;
966 struct btrfs_key key;
967 struct btrfs_key found_key;
968 u32 new_size = sizeof(*item);
972 leaf = path->nodes[0];
973 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
975 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
976 ei0 = btrfs_item_ptr(leaf, path->slots[0],
977 struct btrfs_extent_item_v0);
978 refs = btrfs_extent_refs_v0(leaf, ei0);
980 if (owner == (u64)-1) {
982 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
983 ret = btrfs_next_leaf(root, path);
986 BUG_ON(ret > 0); /* Corruption */
987 leaf = path->nodes[0];
989 btrfs_item_key_to_cpu(leaf, &found_key,
991 BUG_ON(key.objectid != found_key.objectid);
992 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
996 ref0 = btrfs_item_ptr(leaf, path->slots[0],
997 struct btrfs_extent_ref_v0);
998 owner = btrfs_ref_objectid_v0(leaf, ref0);
1002 btrfs_release_path(path);
1004 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1005 new_size += sizeof(*bi);
1007 new_size -= sizeof(*ei0);
1008 ret = btrfs_search_slot(trans, root, &key, path,
1009 new_size + extra_size, 1);
1012 BUG_ON(ret); /* Corruption */
1014 btrfs_extend_item(root, path, new_size);
1016 leaf = path->nodes[0];
1017 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1018 btrfs_set_extent_refs(leaf, item, refs);
1019 /* FIXME: get real generation */
1020 btrfs_set_extent_generation(leaf, item, 0);
1021 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1022 btrfs_set_extent_flags(leaf, item,
1023 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1024 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1025 bi = (struct btrfs_tree_block_info *)(item + 1);
1026 /* FIXME: get first key of the block */
1027 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1028 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1030 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1032 btrfs_mark_buffer_dirty(leaf);
1037 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1039 u32 high_crc = ~(u32)0;
1040 u32 low_crc = ~(u32)0;
1043 lenum = cpu_to_le64(root_objectid);
1044 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1045 lenum = cpu_to_le64(owner);
1046 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1047 lenum = cpu_to_le64(offset);
1048 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1050 return ((u64)high_crc << 31) ^ (u64)low_crc;
1053 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1054 struct btrfs_extent_data_ref *ref)
1056 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1057 btrfs_extent_data_ref_objectid(leaf, ref),
1058 btrfs_extent_data_ref_offset(leaf, ref));
1061 static int match_extent_data_ref(struct extent_buffer *leaf,
1062 struct btrfs_extent_data_ref *ref,
1063 u64 root_objectid, u64 owner, u64 offset)
1065 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1066 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1067 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1072 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1073 struct btrfs_root *root,
1074 struct btrfs_path *path,
1075 u64 bytenr, u64 parent,
1077 u64 owner, u64 offset)
1079 struct btrfs_key key;
1080 struct btrfs_extent_data_ref *ref;
1081 struct extent_buffer *leaf;
1087 key.objectid = bytenr;
1089 key.type = BTRFS_SHARED_DATA_REF_KEY;
1090 key.offset = parent;
1092 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1093 key.offset = hash_extent_data_ref(root_objectid,
1098 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1107 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1108 key.type = BTRFS_EXTENT_REF_V0_KEY;
1109 btrfs_release_path(path);
1110 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1121 leaf = path->nodes[0];
1122 nritems = btrfs_header_nritems(leaf);
1124 if (path->slots[0] >= nritems) {
1125 ret = btrfs_next_leaf(root, path);
1131 leaf = path->nodes[0];
1132 nritems = btrfs_header_nritems(leaf);
1136 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1137 if (key.objectid != bytenr ||
1138 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1141 ref = btrfs_item_ptr(leaf, path->slots[0],
1142 struct btrfs_extent_data_ref);
1144 if (match_extent_data_ref(leaf, ref, root_objectid,
1147 btrfs_release_path(path);
1159 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1160 struct btrfs_root *root,
1161 struct btrfs_path *path,
1162 u64 bytenr, u64 parent,
1163 u64 root_objectid, u64 owner,
1164 u64 offset, int refs_to_add)
1166 struct btrfs_key key;
1167 struct extent_buffer *leaf;
1172 key.objectid = bytenr;
1174 key.type = BTRFS_SHARED_DATA_REF_KEY;
1175 key.offset = parent;
1176 size = sizeof(struct btrfs_shared_data_ref);
1178 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1179 key.offset = hash_extent_data_ref(root_objectid,
1181 size = sizeof(struct btrfs_extent_data_ref);
1184 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1185 if (ret && ret != -EEXIST)
1188 leaf = path->nodes[0];
1190 struct btrfs_shared_data_ref *ref;
1191 ref = btrfs_item_ptr(leaf, path->slots[0],
1192 struct btrfs_shared_data_ref);
1194 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1196 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1197 num_refs += refs_to_add;
1198 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1201 struct btrfs_extent_data_ref *ref;
1202 while (ret == -EEXIST) {
1203 ref = btrfs_item_ptr(leaf, path->slots[0],
1204 struct btrfs_extent_data_ref);
1205 if (match_extent_data_ref(leaf, ref, root_objectid,
1208 btrfs_release_path(path);
1210 ret = btrfs_insert_empty_item(trans, root, path, &key,
1212 if (ret && ret != -EEXIST)
1215 leaf = path->nodes[0];
1217 ref = btrfs_item_ptr(leaf, path->slots[0],
1218 struct btrfs_extent_data_ref);
1220 btrfs_set_extent_data_ref_root(leaf, ref,
1222 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1223 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1224 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1226 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1227 num_refs += refs_to_add;
1228 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1231 btrfs_mark_buffer_dirty(leaf);
1234 btrfs_release_path(path);
1238 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1239 struct btrfs_root *root,
1240 struct btrfs_path *path,
1243 struct btrfs_key key;
1244 struct btrfs_extent_data_ref *ref1 = NULL;
1245 struct btrfs_shared_data_ref *ref2 = NULL;
1246 struct extent_buffer *leaf;
1250 leaf = path->nodes[0];
1251 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1253 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1254 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1255 struct btrfs_extent_data_ref);
1256 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1257 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1258 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1259 struct btrfs_shared_data_ref);
1260 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1261 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1262 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1263 struct btrfs_extent_ref_v0 *ref0;
1264 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1265 struct btrfs_extent_ref_v0);
1266 num_refs = btrfs_ref_count_v0(leaf, ref0);
1272 BUG_ON(num_refs < refs_to_drop);
1273 num_refs -= refs_to_drop;
1275 if (num_refs == 0) {
1276 ret = btrfs_del_item(trans, root, path);
1278 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1279 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1280 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1281 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1282 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1284 struct btrfs_extent_ref_v0 *ref0;
1285 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1286 struct btrfs_extent_ref_v0);
1287 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1290 btrfs_mark_buffer_dirty(leaf);
1295 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1296 struct btrfs_path *path,
1297 struct btrfs_extent_inline_ref *iref)
1299 struct btrfs_key key;
1300 struct extent_buffer *leaf;
1301 struct btrfs_extent_data_ref *ref1;
1302 struct btrfs_shared_data_ref *ref2;
1305 leaf = path->nodes[0];
1306 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1308 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1309 BTRFS_EXTENT_DATA_REF_KEY) {
1310 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1311 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1313 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1314 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1316 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1317 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1318 struct btrfs_extent_data_ref);
1319 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1320 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1321 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1322 struct btrfs_shared_data_ref);
1323 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1324 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1325 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1326 struct btrfs_extent_ref_v0 *ref0;
1327 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1328 struct btrfs_extent_ref_v0);
1329 num_refs = btrfs_ref_count_v0(leaf, ref0);
1337 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1338 struct btrfs_root *root,
1339 struct btrfs_path *path,
1340 u64 bytenr, u64 parent,
1343 struct btrfs_key key;
1346 key.objectid = bytenr;
1348 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1349 key.offset = parent;
1351 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1352 key.offset = root_objectid;
1355 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1358 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1359 if (ret == -ENOENT && parent) {
1360 btrfs_release_path(path);
1361 key.type = BTRFS_EXTENT_REF_V0_KEY;
1362 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1370 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1371 struct btrfs_root *root,
1372 struct btrfs_path *path,
1373 u64 bytenr, u64 parent,
1376 struct btrfs_key key;
1379 key.objectid = bytenr;
1381 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1382 key.offset = parent;
1384 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1385 key.offset = root_objectid;
1388 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1389 btrfs_release_path(path);
1393 static inline int extent_ref_type(u64 parent, u64 owner)
1396 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1398 type = BTRFS_SHARED_BLOCK_REF_KEY;
1400 type = BTRFS_TREE_BLOCK_REF_KEY;
1403 type = BTRFS_SHARED_DATA_REF_KEY;
1405 type = BTRFS_EXTENT_DATA_REF_KEY;
1410 static int find_next_key(struct btrfs_path *path, int level,
1411 struct btrfs_key *key)
1414 for (; level < BTRFS_MAX_LEVEL; level++) {
1415 if (!path->nodes[level])
1417 if (path->slots[level] + 1 >=
1418 btrfs_header_nritems(path->nodes[level]))
1421 btrfs_item_key_to_cpu(path->nodes[level], key,
1422 path->slots[level] + 1);
1424 btrfs_node_key_to_cpu(path->nodes[level], key,
1425 path->slots[level] + 1);
1432 * look for inline back ref. if back ref is found, *ref_ret is set
1433 * to the address of inline back ref, and 0 is returned.
1435 * if back ref isn't found, *ref_ret is set to the address where it
1436 * should be inserted, and -ENOENT is returned.
1438 * if insert is true and there are too many inline back refs, the path
1439 * points to the extent item, and -EAGAIN is returned.
1441 * NOTE: inline back refs are ordered in the same way that back ref
1442 * items in the tree are ordered.
1444 static noinline_for_stack
1445 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1446 struct btrfs_root *root,
1447 struct btrfs_path *path,
1448 struct btrfs_extent_inline_ref **ref_ret,
1449 u64 bytenr, u64 num_bytes,
1450 u64 parent, u64 root_objectid,
1451 u64 owner, u64 offset, int insert)
1453 struct btrfs_key key;
1454 struct extent_buffer *leaf;
1455 struct btrfs_extent_item *ei;
1456 struct btrfs_extent_inline_ref *iref;
1466 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1469 key.objectid = bytenr;
1470 key.type = BTRFS_EXTENT_ITEM_KEY;
1471 key.offset = num_bytes;
1473 want = extent_ref_type(parent, owner);
1475 extra_size = btrfs_extent_inline_ref_size(want);
1476 path->keep_locks = 1;
1481 * Owner is our parent level, so we can just add one to get the level
1482 * for the block we are interested in.
1484 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1485 key.type = BTRFS_METADATA_ITEM_KEY;
1490 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1497 * We may be a newly converted file system which still has the old fat
1498 * extent entries for metadata, so try and see if we have one of those.
1500 if (ret > 0 && skinny_metadata) {
1501 skinny_metadata = false;
1502 if (path->slots[0]) {
1504 btrfs_item_key_to_cpu(path->nodes[0], &key,
1506 if (key.objectid == bytenr &&
1507 key.type == BTRFS_EXTENT_ITEM_KEY &&
1508 key.offset == num_bytes)
1512 key.type = BTRFS_EXTENT_ITEM_KEY;
1513 key.offset = num_bytes;
1514 btrfs_release_path(path);
1519 if (ret && !insert) {
1528 leaf = path->nodes[0];
1529 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1530 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1531 if (item_size < sizeof(*ei)) {
1536 ret = convert_extent_item_v0(trans, root, path, owner,
1542 leaf = path->nodes[0];
1543 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1546 BUG_ON(item_size < sizeof(*ei));
1548 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1549 flags = btrfs_extent_flags(leaf, ei);
1551 ptr = (unsigned long)(ei + 1);
1552 end = (unsigned long)ei + item_size;
1554 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1555 ptr += sizeof(struct btrfs_tree_block_info);
1565 iref = (struct btrfs_extent_inline_ref *)ptr;
1566 type = btrfs_extent_inline_ref_type(leaf, iref);
1570 ptr += btrfs_extent_inline_ref_size(type);
1574 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1575 struct btrfs_extent_data_ref *dref;
1576 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1577 if (match_extent_data_ref(leaf, dref, root_objectid,
1582 if (hash_extent_data_ref_item(leaf, dref) <
1583 hash_extent_data_ref(root_objectid, owner, offset))
1587 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1589 if (parent == ref_offset) {
1593 if (ref_offset < parent)
1596 if (root_objectid == ref_offset) {
1600 if (ref_offset < root_objectid)
1604 ptr += btrfs_extent_inline_ref_size(type);
1606 if (err == -ENOENT && insert) {
1607 if (item_size + extra_size >=
1608 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1613 * To add new inline back ref, we have to make sure
1614 * there is no corresponding back ref item.
1615 * For simplicity, we just do not add new inline back
1616 * ref if there is any kind of item for this block
1618 if (find_next_key(path, 0, &key) == 0 &&
1619 key.objectid == bytenr &&
1620 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1625 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1628 path->keep_locks = 0;
1629 btrfs_unlock_up_safe(path, 1);
1635 * helper to add new inline back ref
1637 static noinline_for_stack
1638 void setup_inline_extent_backref(struct btrfs_root *root,
1639 struct btrfs_path *path,
1640 struct btrfs_extent_inline_ref *iref,
1641 u64 parent, u64 root_objectid,
1642 u64 owner, u64 offset, int refs_to_add,
1643 struct btrfs_delayed_extent_op *extent_op)
1645 struct extent_buffer *leaf;
1646 struct btrfs_extent_item *ei;
1649 unsigned long item_offset;
1654 leaf = path->nodes[0];
1655 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1656 item_offset = (unsigned long)iref - (unsigned long)ei;
1658 type = extent_ref_type(parent, owner);
1659 size = btrfs_extent_inline_ref_size(type);
1661 btrfs_extend_item(root, path, size);
1663 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1664 refs = btrfs_extent_refs(leaf, ei);
1665 refs += refs_to_add;
1666 btrfs_set_extent_refs(leaf, ei, refs);
1668 __run_delayed_extent_op(extent_op, leaf, ei);
1670 ptr = (unsigned long)ei + item_offset;
1671 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1672 if (ptr < end - size)
1673 memmove_extent_buffer(leaf, ptr + size, ptr,
1676 iref = (struct btrfs_extent_inline_ref *)ptr;
1677 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1678 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1679 struct btrfs_extent_data_ref *dref;
1680 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1681 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1682 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1683 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1684 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1685 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1686 struct btrfs_shared_data_ref *sref;
1687 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1688 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1689 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1690 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1691 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1693 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1695 btrfs_mark_buffer_dirty(leaf);
1698 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1699 struct btrfs_root *root,
1700 struct btrfs_path *path,
1701 struct btrfs_extent_inline_ref **ref_ret,
1702 u64 bytenr, u64 num_bytes, u64 parent,
1703 u64 root_objectid, u64 owner, u64 offset)
1707 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1708 bytenr, num_bytes, parent,
1709 root_objectid, owner, offset, 0);
1713 btrfs_release_path(path);
1716 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1717 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1720 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1721 root_objectid, owner, offset);
1727 * helper to update/remove inline back ref
1729 static noinline_for_stack
1730 void update_inline_extent_backref(struct btrfs_root *root,
1731 struct btrfs_path *path,
1732 struct btrfs_extent_inline_ref *iref,
1734 struct btrfs_delayed_extent_op *extent_op)
1736 struct extent_buffer *leaf;
1737 struct btrfs_extent_item *ei;
1738 struct btrfs_extent_data_ref *dref = NULL;
1739 struct btrfs_shared_data_ref *sref = NULL;
1747 leaf = path->nodes[0];
1748 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1749 refs = btrfs_extent_refs(leaf, ei);
1750 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1751 refs += refs_to_mod;
1752 btrfs_set_extent_refs(leaf, ei, refs);
1754 __run_delayed_extent_op(extent_op, leaf, ei);
1756 type = btrfs_extent_inline_ref_type(leaf, iref);
1758 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1759 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1760 refs = btrfs_extent_data_ref_count(leaf, dref);
1761 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1762 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1763 refs = btrfs_shared_data_ref_count(leaf, sref);
1766 BUG_ON(refs_to_mod != -1);
1769 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1770 refs += refs_to_mod;
1773 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1774 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1776 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1778 size = btrfs_extent_inline_ref_size(type);
1779 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1780 ptr = (unsigned long)iref;
1781 end = (unsigned long)ei + item_size;
1782 if (ptr + size < end)
1783 memmove_extent_buffer(leaf, ptr, ptr + size,
1786 btrfs_truncate_item(root, path, item_size, 1);
1788 btrfs_mark_buffer_dirty(leaf);
1791 static noinline_for_stack
1792 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1793 struct btrfs_root *root,
1794 struct btrfs_path *path,
1795 u64 bytenr, u64 num_bytes, u64 parent,
1796 u64 root_objectid, u64 owner,
1797 u64 offset, int refs_to_add,
1798 struct btrfs_delayed_extent_op *extent_op)
1800 struct btrfs_extent_inline_ref *iref;
1803 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1804 bytenr, num_bytes, parent,
1805 root_objectid, owner, offset, 1);
1807 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1808 update_inline_extent_backref(root, path, iref,
1809 refs_to_add, extent_op);
1810 } else if (ret == -ENOENT) {
1811 setup_inline_extent_backref(root, path, iref, parent,
1812 root_objectid, owner, offset,
1813 refs_to_add, extent_op);
1819 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1820 struct btrfs_root *root,
1821 struct btrfs_path *path,
1822 u64 bytenr, u64 parent, u64 root_objectid,
1823 u64 owner, u64 offset, int refs_to_add)
1826 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1827 BUG_ON(refs_to_add != 1);
1828 ret = insert_tree_block_ref(trans, root, path, bytenr,
1829 parent, root_objectid);
1831 ret = insert_extent_data_ref(trans, root, path, bytenr,
1832 parent, root_objectid,
1833 owner, offset, refs_to_add);
1838 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1839 struct btrfs_root *root,
1840 struct btrfs_path *path,
1841 struct btrfs_extent_inline_ref *iref,
1842 int refs_to_drop, int is_data)
1846 BUG_ON(!is_data && refs_to_drop != 1);
1848 update_inline_extent_backref(root, path, iref,
1849 -refs_to_drop, NULL);
1850 } else if (is_data) {
1851 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1853 ret = btrfs_del_item(trans, root, path);
1858 static int btrfs_issue_discard(struct block_device *bdev,
1861 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1864 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1865 u64 num_bytes, u64 *actual_bytes)
1868 u64 discarded_bytes = 0;
1869 struct btrfs_bio *bbio = NULL;
1872 /* Tell the block device(s) that the sectors can be discarded */
1873 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1874 bytenr, &num_bytes, &bbio, 0);
1875 /* Error condition is -ENOMEM */
1877 struct btrfs_bio_stripe *stripe = bbio->stripes;
1881 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1882 if (!stripe->dev->can_discard)
1885 ret = btrfs_issue_discard(stripe->dev->bdev,
1889 discarded_bytes += stripe->length;
1890 else if (ret != -EOPNOTSUPP)
1891 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1894 * Just in case we get back EOPNOTSUPP for some reason,
1895 * just ignore the return value so we don't screw up
1896 * people calling discard_extent.
1904 *actual_bytes = discarded_bytes;
1907 if (ret == -EOPNOTSUPP)
1912 /* Can return -ENOMEM */
1913 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1914 struct btrfs_root *root,
1915 u64 bytenr, u64 num_bytes, u64 parent,
1916 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1919 struct btrfs_fs_info *fs_info = root->fs_info;
1921 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1922 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1924 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1925 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1927 parent, root_objectid, (int)owner,
1928 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1930 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1932 parent, root_objectid, owner, offset,
1933 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1938 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1939 struct btrfs_root *root,
1940 u64 bytenr, u64 num_bytes,
1941 u64 parent, u64 root_objectid,
1942 u64 owner, u64 offset, int refs_to_add,
1943 struct btrfs_delayed_extent_op *extent_op)
1945 struct btrfs_path *path;
1946 struct extent_buffer *leaf;
1947 struct btrfs_extent_item *item;
1952 path = btrfs_alloc_path();
1957 path->leave_spinning = 1;
1958 /* this will setup the path even if it fails to insert the back ref */
1959 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1960 path, bytenr, num_bytes, parent,
1961 root_objectid, owner, offset,
1962 refs_to_add, extent_op);
1966 if (ret != -EAGAIN) {
1971 leaf = path->nodes[0];
1972 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1973 refs = btrfs_extent_refs(leaf, item);
1974 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1976 __run_delayed_extent_op(extent_op, leaf, item);
1978 btrfs_mark_buffer_dirty(leaf);
1979 btrfs_release_path(path);
1982 path->leave_spinning = 1;
1984 /* now insert the actual backref */
1985 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1986 path, bytenr, parent, root_objectid,
1987 owner, offset, refs_to_add);
1989 btrfs_abort_transaction(trans, root, ret);
1991 btrfs_free_path(path);
1995 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1996 struct btrfs_root *root,
1997 struct btrfs_delayed_ref_node *node,
1998 struct btrfs_delayed_extent_op *extent_op,
1999 int insert_reserved)
2002 struct btrfs_delayed_data_ref *ref;
2003 struct btrfs_key ins;
2008 ins.objectid = node->bytenr;
2009 ins.offset = node->num_bytes;
2010 ins.type = BTRFS_EXTENT_ITEM_KEY;
2012 ref = btrfs_delayed_node_to_data_ref(node);
2013 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2014 parent = ref->parent;
2016 ref_root = ref->root;
2018 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2020 flags |= extent_op->flags_to_set;
2021 ret = alloc_reserved_file_extent(trans, root,
2022 parent, ref_root, flags,
2023 ref->objectid, ref->offset,
2024 &ins, node->ref_mod);
2025 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2026 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2027 node->num_bytes, parent,
2028 ref_root, ref->objectid,
2029 ref->offset, node->ref_mod,
2031 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2032 ret = __btrfs_free_extent(trans, root, node->bytenr,
2033 node->num_bytes, parent,
2034 ref_root, ref->objectid,
2035 ref->offset, node->ref_mod,
2043 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2044 struct extent_buffer *leaf,
2045 struct btrfs_extent_item *ei)
2047 u64 flags = btrfs_extent_flags(leaf, ei);
2048 if (extent_op->update_flags) {
2049 flags |= extent_op->flags_to_set;
2050 btrfs_set_extent_flags(leaf, ei, flags);
2053 if (extent_op->update_key) {
2054 struct btrfs_tree_block_info *bi;
2055 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2056 bi = (struct btrfs_tree_block_info *)(ei + 1);
2057 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2061 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2062 struct btrfs_root *root,
2063 struct btrfs_delayed_ref_node *node,
2064 struct btrfs_delayed_extent_op *extent_op)
2066 struct btrfs_key key;
2067 struct btrfs_path *path;
2068 struct btrfs_extent_item *ei;
2069 struct extent_buffer *leaf;
2073 int metadata = !extent_op->is_data;
2078 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2081 path = btrfs_alloc_path();
2085 key.objectid = node->bytenr;
2088 key.type = BTRFS_METADATA_ITEM_KEY;
2089 key.offset = extent_op->level;
2091 key.type = BTRFS_EXTENT_ITEM_KEY;
2092 key.offset = node->num_bytes;
2097 path->leave_spinning = 1;
2098 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2106 btrfs_release_path(path);
2109 key.offset = node->num_bytes;
2110 key.type = BTRFS_EXTENT_ITEM_KEY;
2117 leaf = path->nodes[0];
2118 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2119 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2120 if (item_size < sizeof(*ei)) {
2121 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2127 leaf = path->nodes[0];
2128 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2131 BUG_ON(item_size < sizeof(*ei));
2132 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2133 __run_delayed_extent_op(extent_op, leaf, ei);
2135 btrfs_mark_buffer_dirty(leaf);
2137 btrfs_free_path(path);
2141 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2142 struct btrfs_root *root,
2143 struct btrfs_delayed_ref_node *node,
2144 struct btrfs_delayed_extent_op *extent_op,
2145 int insert_reserved)
2148 struct btrfs_delayed_tree_ref *ref;
2149 struct btrfs_key ins;
2152 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2155 ref = btrfs_delayed_node_to_tree_ref(node);
2156 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2157 parent = ref->parent;
2159 ref_root = ref->root;
2161 ins.objectid = node->bytenr;
2162 if (skinny_metadata) {
2163 ins.offset = ref->level;
2164 ins.type = BTRFS_METADATA_ITEM_KEY;
2166 ins.offset = node->num_bytes;
2167 ins.type = BTRFS_EXTENT_ITEM_KEY;
2170 BUG_ON(node->ref_mod != 1);
2171 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2172 BUG_ON(!extent_op || !extent_op->update_flags);
2173 ret = alloc_reserved_tree_block(trans, root,
2175 extent_op->flags_to_set,
2178 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2179 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2180 node->num_bytes, parent, ref_root,
2181 ref->level, 0, 1, extent_op);
2182 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2183 ret = __btrfs_free_extent(trans, root, node->bytenr,
2184 node->num_bytes, parent, ref_root,
2185 ref->level, 0, 1, extent_op);
2192 /* helper function to actually process a single delayed ref entry */
2193 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2194 struct btrfs_root *root,
2195 struct btrfs_delayed_ref_node *node,
2196 struct btrfs_delayed_extent_op *extent_op,
2197 int insert_reserved)
2204 if (btrfs_delayed_ref_is_head(node)) {
2205 struct btrfs_delayed_ref_head *head;
2207 * we've hit the end of the chain and we were supposed
2208 * to insert this extent into the tree. But, it got
2209 * deleted before we ever needed to insert it, so all
2210 * we have to do is clean up the accounting
2213 head = btrfs_delayed_node_to_head(node);
2214 if (insert_reserved) {
2215 btrfs_pin_extent(root, node->bytenr,
2216 node->num_bytes, 1);
2217 if (head->is_data) {
2218 ret = btrfs_del_csums(trans, root,
2226 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2227 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2228 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2230 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2231 node->type == BTRFS_SHARED_DATA_REF_KEY)
2232 ret = run_delayed_data_ref(trans, root, node, extent_op,
2239 static noinline struct btrfs_delayed_ref_node *
2240 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2242 struct rb_node *node;
2243 struct btrfs_delayed_ref_node *ref;
2244 int action = BTRFS_ADD_DELAYED_REF;
2247 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2248 * this prevents ref count from going down to zero when
2249 * there still are pending delayed ref.
2251 node = rb_prev(&head->node.rb_node);
2255 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2257 if (ref->bytenr != head->node.bytenr)
2259 if (ref->action == action)
2261 node = rb_prev(node);
2263 if (action == BTRFS_ADD_DELAYED_REF) {
2264 action = BTRFS_DROP_DELAYED_REF;
2271 * Returns 0 on success or if called with an already aborted transaction.
2272 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2274 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2275 struct btrfs_root *root,
2276 struct list_head *cluster)
2278 struct btrfs_delayed_ref_root *delayed_refs;
2279 struct btrfs_delayed_ref_node *ref;
2280 struct btrfs_delayed_ref_head *locked_ref = NULL;
2281 struct btrfs_delayed_extent_op *extent_op;
2282 struct btrfs_fs_info *fs_info = root->fs_info;
2285 int must_insert_reserved = 0;
2287 delayed_refs = &trans->transaction->delayed_refs;
2290 /* pick a new head ref from the cluster list */
2291 if (list_empty(cluster))
2294 locked_ref = list_entry(cluster->next,
2295 struct btrfs_delayed_ref_head, cluster);
2297 /* grab the lock that says we are going to process
2298 * all the refs for this head */
2299 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2302 * we may have dropped the spin lock to get the head
2303 * mutex lock, and that might have given someone else
2304 * time to free the head. If that's true, it has been
2305 * removed from our list and we can move on.
2307 if (ret == -EAGAIN) {
2315 * We need to try and merge add/drops of the same ref since we
2316 * can run into issues with relocate dropping the implicit ref
2317 * and then it being added back again before the drop can
2318 * finish. If we merged anything we need to re-loop so we can
2321 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2325 * locked_ref is the head node, so we have to go one
2326 * node back for any delayed ref updates
2328 ref = select_delayed_ref(locked_ref);
2330 if (ref && ref->seq &&
2331 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2333 * there are still refs with lower seq numbers in the
2334 * process of being added. Don't run this ref yet.
2336 list_del_init(&locked_ref->cluster);
2337 btrfs_delayed_ref_unlock(locked_ref);
2339 delayed_refs->num_heads_ready++;
2340 spin_unlock(&delayed_refs->lock);
2342 spin_lock(&delayed_refs->lock);
2347 * record the must insert reserved flag before we
2348 * drop the spin lock.
2350 must_insert_reserved = locked_ref->must_insert_reserved;
2351 locked_ref->must_insert_reserved = 0;
2353 extent_op = locked_ref->extent_op;
2354 locked_ref->extent_op = NULL;
2357 /* All delayed refs have been processed, Go ahead
2358 * and send the head node to run_one_delayed_ref,
2359 * so that any accounting fixes can happen
2361 ref = &locked_ref->node;
2363 if (extent_op && must_insert_reserved) {
2364 btrfs_free_delayed_extent_op(extent_op);
2369 spin_unlock(&delayed_refs->lock);
2371 ret = run_delayed_extent_op(trans, root,
2373 btrfs_free_delayed_extent_op(extent_op);
2376 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2377 spin_lock(&delayed_refs->lock);
2378 btrfs_delayed_ref_unlock(locked_ref);
2387 rb_erase(&ref->rb_node, &delayed_refs->root);
2388 delayed_refs->num_entries--;
2389 if (!btrfs_delayed_ref_is_head(ref)) {
2391 * when we play the delayed ref, also correct the
2394 switch (ref->action) {
2395 case BTRFS_ADD_DELAYED_REF:
2396 case BTRFS_ADD_DELAYED_EXTENT:
2397 locked_ref->node.ref_mod -= ref->ref_mod;
2399 case BTRFS_DROP_DELAYED_REF:
2400 locked_ref->node.ref_mod += ref->ref_mod;
2406 spin_unlock(&delayed_refs->lock);
2408 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2409 must_insert_reserved);
2411 btrfs_free_delayed_extent_op(extent_op);
2413 btrfs_delayed_ref_unlock(locked_ref);
2414 btrfs_put_delayed_ref(ref);
2415 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2416 spin_lock(&delayed_refs->lock);
2421 * If this node is a head, that means all the refs in this head
2422 * have been dealt with, and we will pick the next head to deal
2423 * with, so we must unlock the head and drop it from the cluster
2424 * list before we release it.
2426 if (btrfs_delayed_ref_is_head(ref)) {
2427 list_del_init(&locked_ref->cluster);
2428 btrfs_delayed_ref_unlock(locked_ref);
2431 btrfs_put_delayed_ref(ref);
2435 spin_lock(&delayed_refs->lock);
2440 #ifdef SCRAMBLE_DELAYED_REFS
2442 * Normally delayed refs get processed in ascending bytenr order. This
2443 * correlates in most cases to the order added. To expose dependencies on this
2444 * order, we start to process the tree in the middle instead of the beginning
2446 static u64 find_middle(struct rb_root *root)
2448 struct rb_node *n = root->rb_node;
2449 struct btrfs_delayed_ref_node *entry;
2452 u64 first = 0, last = 0;
2456 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2457 first = entry->bytenr;
2461 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2462 last = entry->bytenr;
2467 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2468 WARN_ON(!entry->in_tree);
2470 middle = entry->bytenr;
2483 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2484 struct btrfs_fs_info *fs_info)
2486 struct qgroup_update *qgroup_update;
2489 if (list_empty(&trans->qgroup_ref_list) !=
2490 !trans->delayed_ref_elem.seq) {
2491 /* list without seq or seq without list */
2493 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2494 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2495 (u32)(trans->delayed_ref_elem.seq >> 32),
2496 (u32)trans->delayed_ref_elem.seq);
2500 if (!trans->delayed_ref_elem.seq)
2503 while (!list_empty(&trans->qgroup_ref_list)) {
2504 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2505 struct qgroup_update, list);
2506 list_del(&qgroup_update->list);
2508 ret = btrfs_qgroup_account_ref(
2509 trans, fs_info, qgroup_update->node,
2510 qgroup_update->extent_op);
2511 kfree(qgroup_update);
2514 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2519 static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2522 int val = atomic_read(&delayed_refs->ref_seq);
2524 if (val < seq || val >= seq + count)
2530 * this starts processing the delayed reference count updates and
2531 * extent insertions we have queued up so far. count can be
2532 * 0, which means to process everything in the tree at the start
2533 * of the run (but not newly added entries), or it can be some target
2534 * number you'd like to process.
2536 * Returns 0 on success or if called with an aborted transaction
2537 * Returns <0 on error and aborts the transaction
2539 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2540 struct btrfs_root *root, unsigned long count)
2542 struct rb_node *node;
2543 struct btrfs_delayed_ref_root *delayed_refs;
2544 struct btrfs_delayed_ref_node *ref;
2545 struct list_head cluster;
2548 int run_all = count == (unsigned long)-1;
2552 /* We'll clean this up in btrfs_cleanup_transaction */
2556 if (root == root->fs_info->extent_root)
2557 root = root->fs_info->tree_root;
2559 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2561 delayed_refs = &trans->transaction->delayed_refs;
2562 INIT_LIST_HEAD(&cluster);
2564 count = delayed_refs->num_entries * 2;
2568 if (!run_all && !run_most) {
2570 int seq = atomic_read(&delayed_refs->ref_seq);
2573 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2575 DEFINE_WAIT(__wait);
2576 if (delayed_refs->num_entries < 16348)
2579 prepare_to_wait(&delayed_refs->wait, &__wait,
2580 TASK_UNINTERRUPTIBLE);
2582 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2585 finish_wait(&delayed_refs->wait, &__wait);
2587 if (!refs_newer(delayed_refs, seq, 256))
2592 finish_wait(&delayed_refs->wait, &__wait);
2598 atomic_inc(&delayed_refs->procs_running_refs);
2603 spin_lock(&delayed_refs->lock);
2605 #ifdef SCRAMBLE_DELAYED_REFS
2606 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2610 if (!(run_all || run_most) &&
2611 delayed_refs->num_heads_ready < 64)
2615 * go find something we can process in the rbtree. We start at
2616 * the beginning of the tree, and then build a cluster
2617 * of refs to process starting at the first one we are able to
2620 delayed_start = delayed_refs->run_delayed_start;
2621 ret = btrfs_find_ref_cluster(trans, &cluster,
2622 delayed_refs->run_delayed_start);
2626 ret = run_clustered_refs(trans, root, &cluster);
2628 btrfs_release_ref_cluster(&cluster);
2629 spin_unlock(&delayed_refs->lock);
2630 btrfs_abort_transaction(trans, root, ret);
2631 atomic_dec(&delayed_refs->procs_running_refs);
2635 atomic_add(ret, &delayed_refs->ref_seq);
2637 count -= min_t(unsigned long, ret, count);
2642 if (delayed_start >= delayed_refs->run_delayed_start) {
2645 * btrfs_find_ref_cluster looped. let's do one
2646 * more cycle. if we don't run any delayed ref
2647 * during that cycle (because we can't because
2648 * all of them are blocked), bail out.
2653 * no runnable refs left, stop trying
2660 /* refs were run, let's reset staleness detection */
2666 if (!list_empty(&trans->new_bgs)) {
2667 spin_unlock(&delayed_refs->lock);
2668 btrfs_create_pending_block_groups(trans, root);
2669 spin_lock(&delayed_refs->lock);
2672 node = rb_first(&delayed_refs->root);
2675 count = (unsigned long)-1;
2678 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2680 if (btrfs_delayed_ref_is_head(ref)) {
2681 struct btrfs_delayed_ref_head *head;
2683 head = btrfs_delayed_node_to_head(ref);
2684 atomic_inc(&ref->refs);
2686 spin_unlock(&delayed_refs->lock);
2688 * Mutex was contended, block until it's
2689 * released and try again
2691 mutex_lock(&head->mutex);
2692 mutex_unlock(&head->mutex);
2694 btrfs_put_delayed_ref(ref);
2698 node = rb_next(node);
2700 spin_unlock(&delayed_refs->lock);
2701 schedule_timeout(1);
2705 atomic_dec(&delayed_refs->procs_running_refs);
2707 if (waitqueue_active(&delayed_refs->wait))
2708 wake_up(&delayed_refs->wait);
2710 spin_unlock(&delayed_refs->lock);
2711 assert_qgroups_uptodate(trans);
2715 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2716 struct btrfs_root *root,
2717 u64 bytenr, u64 num_bytes, u64 flags,
2718 int level, int is_data)
2720 struct btrfs_delayed_extent_op *extent_op;
2723 extent_op = btrfs_alloc_delayed_extent_op();
2727 extent_op->flags_to_set = flags;
2728 extent_op->update_flags = 1;
2729 extent_op->update_key = 0;
2730 extent_op->is_data = is_data ? 1 : 0;
2731 extent_op->level = level;
2733 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2734 num_bytes, extent_op);
2736 btrfs_free_delayed_extent_op(extent_op);
2740 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2741 struct btrfs_root *root,
2742 struct btrfs_path *path,
2743 u64 objectid, u64 offset, u64 bytenr)
2745 struct btrfs_delayed_ref_head *head;
2746 struct btrfs_delayed_ref_node *ref;
2747 struct btrfs_delayed_data_ref *data_ref;
2748 struct btrfs_delayed_ref_root *delayed_refs;
2749 struct rb_node *node;
2753 delayed_refs = &trans->transaction->delayed_refs;
2754 spin_lock(&delayed_refs->lock);
2755 head = btrfs_find_delayed_ref_head(trans, bytenr);
2759 if (!mutex_trylock(&head->mutex)) {
2760 atomic_inc(&head->node.refs);
2761 spin_unlock(&delayed_refs->lock);
2763 btrfs_release_path(path);
2766 * Mutex was contended, block until it's released and let
2769 mutex_lock(&head->mutex);
2770 mutex_unlock(&head->mutex);
2771 btrfs_put_delayed_ref(&head->node);
2775 node = rb_prev(&head->node.rb_node);
2779 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2781 if (ref->bytenr != bytenr)
2785 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2788 data_ref = btrfs_delayed_node_to_data_ref(ref);
2790 node = rb_prev(node);
2794 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2795 if (ref->bytenr == bytenr && ref->seq == seq)
2799 if (data_ref->root != root->root_key.objectid ||
2800 data_ref->objectid != objectid || data_ref->offset != offset)
2805 mutex_unlock(&head->mutex);
2807 spin_unlock(&delayed_refs->lock);
2811 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2812 struct btrfs_root *root,
2813 struct btrfs_path *path,
2814 u64 objectid, u64 offset, u64 bytenr)
2816 struct btrfs_root *extent_root = root->fs_info->extent_root;
2817 struct extent_buffer *leaf;
2818 struct btrfs_extent_data_ref *ref;
2819 struct btrfs_extent_inline_ref *iref;
2820 struct btrfs_extent_item *ei;
2821 struct btrfs_key key;
2825 key.objectid = bytenr;
2826 key.offset = (u64)-1;
2827 key.type = BTRFS_EXTENT_ITEM_KEY;
2829 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2832 BUG_ON(ret == 0); /* Corruption */
2835 if (path->slots[0] == 0)
2839 leaf = path->nodes[0];
2840 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2842 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2846 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2847 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2848 if (item_size < sizeof(*ei)) {
2849 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2853 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2855 if (item_size != sizeof(*ei) +
2856 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2859 if (btrfs_extent_generation(leaf, ei) <=
2860 btrfs_root_last_snapshot(&root->root_item))
2863 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2864 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2865 BTRFS_EXTENT_DATA_REF_KEY)
2868 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2869 if (btrfs_extent_refs(leaf, ei) !=
2870 btrfs_extent_data_ref_count(leaf, ref) ||
2871 btrfs_extent_data_ref_root(leaf, ref) !=
2872 root->root_key.objectid ||
2873 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2874 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2882 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2883 struct btrfs_root *root,
2884 u64 objectid, u64 offset, u64 bytenr)
2886 struct btrfs_path *path;
2890 path = btrfs_alloc_path();
2895 ret = check_committed_ref(trans, root, path, objectid,
2897 if (ret && ret != -ENOENT)
2900 ret2 = check_delayed_ref(trans, root, path, objectid,
2902 } while (ret2 == -EAGAIN);
2904 if (ret2 && ret2 != -ENOENT) {
2909 if (ret != -ENOENT || ret2 != -ENOENT)
2912 btrfs_free_path(path);
2913 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2918 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2919 struct btrfs_root *root,
2920 struct extent_buffer *buf,
2921 int full_backref, int inc, int for_cow)
2928 struct btrfs_key key;
2929 struct btrfs_file_extent_item *fi;
2933 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2934 u64, u64, u64, u64, u64, u64, int);
2936 ref_root = btrfs_header_owner(buf);
2937 nritems = btrfs_header_nritems(buf);
2938 level = btrfs_header_level(buf);
2940 if (!root->ref_cows && level == 0)
2944 process_func = btrfs_inc_extent_ref;
2946 process_func = btrfs_free_extent;
2949 parent = buf->start;
2953 for (i = 0; i < nritems; i++) {
2955 btrfs_item_key_to_cpu(buf, &key, i);
2956 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2958 fi = btrfs_item_ptr(buf, i,
2959 struct btrfs_file_extent_item);
2960 if (btrfs_file_extent_type(buf, fi) ==
2961 BTRFS_FILE_EXTENT_INLINE)
2963 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2967 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2968 key.offset -= btrfs_file_extent_offset(buf, fi);
2969 ret = process_func(trans, root, bytenr, num_bytes,
2970 parent, ref_root, key.objectid,
2971 key.offset, for_cow);
2975 bytenr = btrfs_node_blockptr(buf, i);
2976 num_bytes = btrfs_level_size(root, level - 1);
2977 ret = process_func(trans, root, bytenr, num_bytes,
2978 parent, ref_root, level - 1, 0,
2989 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2990 struct extent_buffer *buf, int full_backref, int for_cow)
2992 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2995 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2996 struct extent_buffer *buf, int full_backref, int for_cow)
2998 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3001 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3002 struct btrfs_root *root,
3003 struct btrfs_path *path,
3004 struct btrfs_block_group_cache *cache)
3007 struct btrfs_root *extent_root = root->fs_info->extent_root;
3009 struct extent_buffer *leaf;
3011 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3014 BUG_ON(ret); /* Corruption */
3016 leaf = path->nodes[0];
3017 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3018 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3019 btrfs_mark_buffer_dirty(leaf);
3020 btrfs_release_path(path);
3023 btrfs_abort_transaction(trans, root, ret);
3030 static struct btrfs_block_group_cache *
3031 next_block_group(struct btrfs_root *root,
3032 struct btrfs_block_group_cache *cache)
3034 struct rb_node *node;
3035 spin_lock(&root->fs_info->block_group_cache_lock);
3036 node = rb_next(&cache->cache_node);
3037 btrfs_put_block_group(cache);
3039 cache = rb_entry(node, struct btrfs_block_group_cache,
3041 btrfs_get_block_group(cache);
3044 spin_unlock(&root->fs_info->block_group_cache_lock);
3048 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3049 struct btrfs_trans_handle *trans,
3050 struct btrfs_path *path)
3052 struct btrfs_root *root = block_group->fs_info->tree_root;
3053 struct inode *inode = NULL;
3055 int dcs = BTRFS_DC_ERROR;
3061 * If this block group is smaller than 100 megs don't bother caching the
3064 if (block_group->key.offset < (100 * 1024 * 1024)) {
3065 spin_lock(&block_group->lock);
3066 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3067 spin_unlock(&block_group->lock);
3072 inode = lookup_free_space_inode(root, block_group, path);
3073 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3074 ret = PTR_ERR(inode);
3075 btrfs_release_path(path);
3079 if (IS_ERR(inode)) {
3083 if (block_group->ro)
3086 ret = create_free_space_inode(root, trans, block_group, path);
3092 /* We've already setup this transaction, go ahead and exit */
3093 if (block_group->cache_generation == trans->transid &&
3094 i_size_read(inode)) {
3095 dcs = BTRFS_DC_SETUP;
3100 * We want to set the generation to 0, that way if anything goes wrong
3101 * from here on out we know not to trust this cache when we load up next
3104 BTRFS_I(inode)->generation = 0;
3105 ret = btrfs_update_inode(trans, root, inode);
3108 if (i_size_read(inode) > 0) {
3109 ret = btrfs_check_trunc_cache_free_space(root,
3110 &root->fs_info->global_block_rsv);
3114 ret = btrfs_truncate_free_space_cache(root, trans, path,
3120 spin_lock(&block_group->lock);
3121 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3122 !btrfs_test_opt(root, SPACE_CACHE)) {
3124 * don't bother trying to write stuff out _if_
3125 * a) we're not cached,
3126 * b) we're with nospace_cache mount option.
3128 dcs = BTRFS_DC_WRITTEN;
3129 spin_unlock(&block_group->lock);
3132 spin_unlock(&block_group->lock);
3135 * Try to preallocate enough space based on how big the block group is.
3136 * Keep in mind this has to include any pinned space which could end up
3137 * taking up quite a bit since it's not folded into the other space
3140 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3145 num_pages *= PAGE_CACHE_SIZE;
3147 ret = btrfs_check_data_free_space(inode, num_pages);
3151 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3152 num_pages, num_pages,
3155 dcs = BTRFS_DC_SETUP;
3156 btrfs_free_reserved_data_space(inode, num_pages);
3161 btrfs_release_path(path);
3163 spin_lock(&block_group->lock);
3164 if (!ret && dcs == BTRFS_DC_SETUP)
3165 block_group->cache_generation = trans->transid;
3166 block_group->disk_cache_state = dcs;
3167 spin_unlock(&block_group->lock);
3172 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3173 struct btrfs_root *root)
3175 struct btrfs_block_group_cache *cache;
3177 struct btrfs_path *path;
3180 path = btrfs_alloc_path();
3186 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3188 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3190 cache = next_block_group(root, cache);
3198 err = cache_save_setup(cache, trans, path);
3199 last = cache->key.objectid + cache->key.offset;
3200 btrfs_put_block_group(cache);
3205 err = btrfs_run_delayed_refs(trans, root,
3207 if (err) /* File system offline */
3211 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3213 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3214 btrfs_put_block_group(cache);
3220 cache = next_block_group(root, cache);
3229 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3230 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3232 last = cache->key.objectid + cache->key.offset;
3234 err = write_one_cache_group(trans, root, path, cache);
3235 if (err) /* File system offline */
3238 btrfs_put_block_group(cache);
3243 * I don't think this is needed since we're just marking our
3244 * preallocated extent as written, but just in case it can't
3248 err = btrfs_run_delayed_refs(trans, root,
3250 if (err) /* File system offline */
3254 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3257 * Really this shouldn't happen, but it could if we
3258 * couldn't write the entire preallocated extent and
3259 * splitting the extent resulted in a new block.
3262 btrfs_put_block_group(cache);
3265 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3267 cache = next_block_group(root, cache);
3276 err = btrfs_write_out_cache(root, trans, cache, path);
3279 * If we didn't have an error then the cache state is still
3280 * NEED_WRITE, so we can set it to WRITTEN.
3282 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3283 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3284 last = cache->key.objectid + cache->key.offset;
3285 btrfs_put_block_group(cache);
3289 btrfs_free_path(path);
3293 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3295 struct btrfs_block_group_cache *block_group;
3298 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3299 if (!block_group || block_group->ro)
3302 btrfs_put_block_group(block_group);
3306 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3307 u64 total_bytes, u64 bytes_used,
3308 struct btrfs_space_info **space_info)
3310 struct btrfs_space_info *found;
3314 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3315 BTRFS_BLOCK_GROUP_RAID10))
3320 found = __find_space_info(info, flags);
3322 spin_lock(&found->lock);
3323 found->total_bytes += total_bytes;
3324 found->disk_total += total_bytes * factor;
3325 found->bytes_used += bytes_used;
3326 found->disk_used += bytes_used * factor;
3328 spin_unlock(&found->lock);
3329 *space_info = found;
3332 found = kzalloc(sizeof(*found), GFP_NOFS);
3336 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3337 INIT_LIST_HEAD(&found->block_groups[i]);
3338 init_rwsem(&found->groups_sem);
3339 spin_lock_init(&found->lock);
3340 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3341 found->total_bytes = total_bytes;
3342 found->disk_total = total_bytes * factor;
3343 found->bytes_used = bytes_used;
3344 found->disk_used = bytes_used * factor;
3345 found->bytes_pinned = 0;
3346 found->bytes_reserved = 0;
3347 found->bytes_readonly = 0;
3348 found->bytes_may_use = 0;
3350 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3351 found->chunk_alloc = 0;
3353 init_waitqueue_head(&found->wait);
3354 *space_info = found;
3355 list_add_rcu(&found->list, &info->space_info);
3356 if (flags & BTRFS_BLOCK_GROUP_DATA)
3357 info->data_sinfo = found;
3361 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3363 u64 extra_flags = chunk_to_extended(flags) &
3364 BTRFS_EXTENDED_PROFILE_MASK;
3366 write_seqlock(&fs_info->profiles_lock);
3367 if (flags & BTRFS_BLOCK_GROUP_DATA)
3368 fs_info->avail_data_alloc_bits |= extra_flags;
3369 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3370 fs_info->avail_metadata_alloc_bits |= extra_flags;
3371 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3372 fs_info->avail_system_alloc_bits |= extra_flags;
3373 write_sequnlock(&fs_info->profiles_lock);
3377 * returns target flags in extended format or 0 if restripe for this
3378 * chunk_type is not in progress
3380 * should be called with either volume_mutex or balance_lock held
3382 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3384 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3390 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3391 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3392 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3393 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3394 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3395 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3396 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3397 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3398 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3405 * @flags: available profiles in extended format (see ctree.h)
3407 * Returns reduced profile in chunk format. If profile changing is in
3408 * progress (either running or paused) picks the target profile (if it's
3409 * already available), otherwise falls back to plain reducing.
3411 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3414 * we add in the count of missing devices because we want
3415 * to make sure that any RAID levels on a degraded FS
3416 * continue to be honored.
3418 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3419 root->fs_info->fs_devices->missing_devices;
3424 * see if restripe for this chunk_type is in progress, if so
3425 * try to reduce to the target profile
3427 spin_lock(&root->fs_info->balance_lock);
3428 target = get_restripe_target(root->fs_info, flags);
3430 /* pick target profile only if it's already available */
3431 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3432 spin_unlock(&root->fs_info->balance_lock);
3433 return extended_to_chunk(target);
3436 spin_unlock(&root->fs_info->balance_lock);
3438 /* First, mask out the RAID levels which aren't possible */
3439 if (num_devices == 1)
3440 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3441 BTRFS_BLOCK_GROUP_RAID5);
3442 if (num_devices < 3)
3443 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3444 if (num_devices < 4)
3445 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3447 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3448 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3449 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3452 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3453 tmp = BTRFS_BLOCK_GROUP_RAID6;
3454 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3455 tmp = BTRFS_BLOCK_GROUP_RAID5;
3456 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3457 tmp = BTRFS_BLOCK_GROUP_RAID10;
3458 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3459 tmp = BTRFS_BLOCK_GROUP_RAID1;
3460 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3461 tmp = BTRFS_BLOCK_GROUP_RAID0;
3463 return extended_to_chunk(flags | tmp);
3466 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3471 seq = read_seqbegin(&root->fs_info->profiles_lock);
3473 if (flags & BTRFS_BLOCK_GROUP_DATA)
3474 flags |= root->fs_info->avail_data_alloc_bits;
3475 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3476 flags |= root->fs_info->avail_system_alloc_bits;
3477 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3478 flags |= root->fs_info->avail_metadata_alloc_bits;
3479 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3481 return btrfs_reduce_alloc_profile(root, flags);
3484 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3490 flags = BTRFS_BLOCK_GROUP_DATA;
3491 else if (root == root->fs_info->chunk_root)
3492 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3494 flags = BTRFS_BLOCK_GROUP_METADATA;
3496 ret = get_alloc_profile(root, flags);
3501 * This will check the space that the inode allocates from to make sure we have
3502 * enough space for bytes.
3504 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3506 struct btrfs_space_info *data_sinfo;
3507 struct btrfs_root *root = BTRFS_I(inode)->root;
3508 struct btrfs_fs_info *fs_info = root->fs_info;
3510 int ret = 0, committed = 0, alloc_chunk = 1;
3512 /* make sure bytes are sectorsize aligned */
3513 bytes = ALIGN(bytes, root->sectorsize);
3515 if (root == root->fs_info->tree_root ||
3516 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3521 data_sinfo = fs_info->data_sinfo;
3526 /* make sure we have enough space to handle the data first */
3527 spin_lock(&data_sinfo->lock);
3528 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3529 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3530 data_sinfo->bytes_may_use;
3532 if (used + bytes > data_sinfo->total_bytes) {
3533 struct btrfs_trans_handle *trans;
3536 * if we don't have enough free bytes in this space then we need
3537 * to alloc a new chunk.
3539 if (!data_sinfo->full && alloc_chunk) {
3542 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3543 spin_unlock(&data_sinfo->lock);
3545 alloc_target = btrfs_get_alloc_profile(root, 1);
3546 trans = btrfs_join_transaction(root);
3548 return PTR_ERR(trans);
3550 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3552 CHUNK_ALLOC_NO_FORCE);
3553 btrfs_end_transaction(trans, root);
3562 data_sinfo = fs_info->data_sinfo;
3568 * If we have less pinned bytes than we want to allocate then
3569 * don't bother committing the transaction, it won't help us.
3571 if (data_sinfo->bytes_pinned < bytes)
3573 spin_unlock(&data_sinfo->lock);
3575 /* commit the current transaction and try again */
3578 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3580 trans = btrfs_join_transaction(root);
3582 return PTR_ERR(trans);
3583 ret = btrfs_commit_transaction(trans, root);
3591 data_sinfo->bytes_may_use += bytes;
3592 trace_btrfs_space_reservation(root->fs_info, "space_info",
3593 data_sinfo->flags, bytes, 1);
3594 spin_unlock(&data_sinfo->lock);
3600 * Called if we need to clear a data reservation for this inode.
3602 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3604 struct btrfs_root *root = BTRFS_I(inode)->root;
3605 struct btrfs_space_info *data_sinfo;
3607 /* make sure bytes are sectorsize aligned */
3608 bytes = ALIGN(bytes, root->sectorsize);
3610 data_sinfo = root->fs_info->data_sinfo;
3611 spin_lock(&data_sinfo->lock);
3612 data_sinfo->bytes_may_use -= bytes;
3613 trace_btrfs_space_reservation(root->fs_info, "space_info",
3614 data_sinfo->flags, bytes, 0);
3615 spin_unlock(&data_sinfo->lock);
3618 static void force_metadata_allocation(struct btrfs_fs_info *info)
3620 struct list_head *head = &info->space_info;
3621 struct btrfs_space_info *found;
3624 list_for_each_entry_rcu(found, head, list) {
3625 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3626 found->force_alloc = CHUNK_ALLOC_FORCE;
3631 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3633 return (global->size << 1);
3636 static int should_alloc_chunk(struct btrfs_root *root,
3637 struct btrfs_space_info *sinfo, int force)
3639 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3640 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3641 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3644 if (force == CHUNK_ALLOC_FORCE)
3648 * We need to take into account the global rsv because for all intents
3649 * and purposes it's used space. Don't worry about locking the
3650 * global_rsv, it doesn't change except when the transaction commits.
3652 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3653 num_allocated += calc_global_rsv_need_space(global_rsv);
3656 * in limited mode, we want to have some free space up to
3657 * about 1% of the FS size.
3659 if (force == CHUNK_ALLOC_LIMITED) {
3660 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3661 thresh = max_t(u64, 64 * 1024 * 1024,
3662 div_factor_fine(thresh, 1));
3664 if (num_bytes - num_allocated < thresh)
3668 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3673 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3677 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3678 BTRFS_BLOCK_GROUP_RAID0 |
3679 BTRFS_BLOCK_GROUP_RAID5 |
3680 BTRFS_BLOCK_GROUP_RAID6))
3681 num_dev = root->fs_info->fs_devices->rw_devices;
3682 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3685 num_dev = 1; /* DUP or single */
3687 /* metadata for updaing devices and chunk tree */
3688 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3691 static void check_system_chunk(struct btrfs_trans_handle *trans,
3692 struct btrfs_root *root, u64 type)
3694 struct btrfs_space_info *info;
3698 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3699 spin_lock(&info->lock);
3700 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3701 info->bytes_reserved - info->bytes_readonly;
3702 spin_unlock(&info->lock);
3704 thresh = get_system_chunk_thresh(root, type);
3705 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3706 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3707 left, thresh, type);
3708 dump_space_info(info, 0, 0);
3711 if (left < thresh) {
3714 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3715 btrfs_alloc_chunk(trans, root, flags);
3719 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3720 struct btrfs_root *extent_root, u64 flags, int force)
3722 struct btrfs_space_info *space_info;
3723 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3724 int wait_for_alloc = 0;
3727 /* Don't re-enter if we're already allocating a chunk */
3728 if (trans->allocating_chunk)
3731 space_info = __find_space_info(extent_root->fs_info, flags);
3733 ret = update_space_info(extent_root->fs_info, flags,
3735 BUG_ON(ret); /* -ENOMEM */
3737 BUG_ON(!space_info); /* Logic error */
3740 spin_lock(&space_info->lock);
3741 if (force < space_info->force_alloc)
3742 force = space_info->force_alloc;
3743 if (space_info->full) {
3744 spin_unlock(&space_info->lock);
3748 if (!should_alloc_chunk(extent_root, space_info, force)) {
3749 spin_unlock(&space_info->lock);
3751 } else if (space_info->chunk_alloc) {
3754 space_info->chunk_alloc = 1;
3757 spin_unlock(&space_info->lock);
3759 mutex_lock(&fs_info->chunk_mutex);
3762 * The chunk_mutex is held throughout the entirety of a chunk
3763 * allocation, so once we've acquired the chunk_mutex we know that the
3764 * other guy is done and we need to recheck and see if we should
3767 if (wait_for_alloc) {
3768 mutex_unlock(&fs_info->chunk_mutex);
3773 trans->allocating_chunk = true;
3776 * If we have mixed data/metadata chunks we want to make sure we keep
3777 * allocating mixed chunks instead of individual chunks.
3779 if (btrfs_mixed_space_info(space_info))
3780 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3783 * if we're doing a data chunk, go ahead and make sure that
3784 * we keep a reasonable number of metadata chunks allocated in the
3787 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3788 fs_info->data_chunk_allocations++;
3789 if (!(fs_info->data_chunk_allocations %
3790 fs_info->metadata_ratio))
3791 force_metadata_allocation(fs_info);
3795 * Check if we have enough space in SYSTEM chunk because we may need
3796 * to update devices.
3798 check_system_chunk(trans, extent_root, flags);
3800 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3801 trans->allocating_chunk = false;
3803 spin_lock(&space_info->lock);
3804 if (ret < 0 && ret != -ENOSPC)
3807 space_info->full = 1;
3811 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3813 space_info->chunk_alloc = 0;
3814 spin_unlock(&space_info->lock);
3815 mutex_unlock(&fs_info->chunk_mutex);
3819 static int can_overcommit(struct btrfs_root *root,
3820 struct btrfs_space_info *space_info, u64 bytes,
3821 enum btrfs_reserve_flush_enum flush)
3823 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3824 u64 profile = btrfs_get_alloc_profile(root, 0);
3830 used = space_info->bytes_used + space_info->bytes_reserved +
3831 space_info->bytes_pinned + space_info->bytes_readonly;
3834 * We only want to allow over committing if we have lots of actual space
3835 * free, but if we don't have enough space to handle the global reserve
3836 * space then we could end up having a real enospc problem when trying
3837 * to allocate a chunk or some other such important allocation.
3839 spin_lock(&global_rsv->lock);
3840 space_size = calc_global_rsv_need_space(global_rsv);
3841 spin_unlock(&global_rsv->lock);
3842 if (used + space_size >= space_info->total_bytes)
3845 used += space_info->bytes_may_use;
3847 spin_lock(&root->fs_info->free_chunk_lock);
3848 avail = root->fs_info->free_chunk_space;
3849 spin_unlock(&root->fs_info->free_chunk_lock);
3852 * If we have dup, raid1 or raid10 then only half of the free
3853 * space is actually useable. For raid56, the space info used
3854 * doesn't include the parity drive, so we don't have to
3857 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3858 BTRFS_BLOCK_GROUP_RAID1 |
3859 BTRFS_BLOCK_GROUP_RAID10))
3862 to_add = space_info->total_bytes;
3865 * If we aren't flushing all things, let us overcommit up to
3866 * 1/2th of the space. If we can flush, don't let us overcommit
3867 * too much, let it overcommit up to 1/8 of the space.
3869 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3875 * Limit the overcommit to the amount of free space we could possibly
3876 * allocate for chunks.
3878 to_add = min(avail, to_add);
3880 if (used + bytes < space_info->total_bytes + to_add)
3885 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3886 unsigned long nr_pages)
3888 struct super_block *sb = root->fs_info->sb;
3891 /* If we can not start writeback, just sync all the delalloc file. */
3892 started = try_to_writeback_inodes_sb_nr(sb, nr_pages,
3893 WB_REASON_FS_FREE_SPACE);
3896 * We needn't worry the filesystem going from r/w to r/o though
3897 * we don't acquire ->s_umount mutex, because the filesystem
3898 * should guarantee the delalloc inodes list be empty after
3899 * the filesystem is readonly(all dirty pages are written to
3902 btrfs_start_delalloc_inodes(root, 0);
3903 if (!current->journal_info)
3904 btrfs_wait_ordered_extents(root, 0);
3909 * shrink metadata reservation for delalloc
3911 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3914 struct btrfs_block_rsv *block_rsv;
3915 struct btrfs_space_info *space_info;
3916 struct btrfs_trans_handle *trans;
3920 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3922 enum btrfs_reserve_flush_enum flush;
3924 trans = (struct btrfs_trans_handle *)current->journal_info;
3925 block_rsv = &root->fs_info->delalloc_block_rsv;
3926 space_info = block_rsv->space_info;
3929 delalloc_bytes = percpu_counter_sum_positive(
3930 &root->fs_info->delalloc_bytes);
3931 if (delalloc_bytes == 0) {
3934 btrfs_wait_ordered_extents(root, 0);
3938 while (delalloc_bytes && loops < 3) {
3939 max_reclaim = min(delalloc_bytes, to_reclaim);
3940 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3941 btrfs_writeback_inodes_sb_nr(root, nr_pages);
3943 * We need to wait for the async pages to actually start before
3946 wait_event(root->fs_info->async_submit_wait,
3947 !atomic_read(&root->fs_info->async_delalloc_pages));
3950 flush = BTRFS_RESERVE_FLUSH_ALL;
3952 flush = BTRFS_RESERVE_NO_FLUSH;
3953 spin_lock(&space_info->lock);
3954 if (can_overcommit(root, space_info, orig, flush)) {
3955 spin_unlock(&space_info->lock);
3958 spin_unlock(&space_info->lock);
3961 if (wait_ordered && !trans) {
3962 btrfs_wait_ordered_extents(root, 0);
3964 time_left = schedule_timeout_killable(1);
3969 delalloc_bytes = percpu_counter_sum_positive(
3970 &root->fs_info->delalloc_bytes);
3975 * maybe_commit_transaction - possibly commit the transaction if its ok to
3976 * @root - the root we're allocating for
3977 * @bytes - the number of bytes we want to reserve
3978 * @force - force the commit
3980 * This will check to make sure that committing the transaction will actually
3981 * get us somewhere and then commit the transaction if it does. Otherwise it
3982 * will return -ENOSPC.
3984 static int may_commit_transaction(struct btrfs_root *root,
3985 struct btrfs_space_info *space_info,
3986 u64 bytes, int force)
3988 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3989 struct btrfs_trans_handle *trans;
3991 trans = (struct btrfs_trans_handle *)current->journal_info;
3998 /* See if there is enough pinned space to make this reservation */
3999 spin_lock(&space_info->lock);
4000 if (space_info->bytes_pinned >= bytes) {
4001 spin_unlock(&space_info->lock);
4004 spin_unlock(&space_info->lock);
4007 * See if there is some space in the delayed insertion reservation for
4010 if (space_info != delayed_rsv->space_info)
4013 spin_lock(&space_info->lock);
4014 spin_lock(&delayed_rsv->lock);
4015 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
4016 spin_unlock(&delayed_rsv->lock);
4017 spin_unlock(&space_info->lock);
4020 spin_unlock(&delayed_rsv->lock);
4021 spin_unlock(&space_info->lock);
4024 trans = btrfs_join_transaction(root);
4028 return btrfs_commit_transaction(trans, root);
4032 FLUSH_DELAYED_ITEMS_NR = 1,
4033 FLUSH_DELAYED_ITEMS = 2,
4035 FLUSH_DELALLOC_WAIT = 4,
4040 static int flush_space(struct btrfs_root *root,
4041 struct btrfs_space_info *space_info, u64 num_bytes,
4042 u64 orig_bytes, int state)
4044 struct btrfs_trans_handle *trans;
4049 case FLUSH_DELAYED_ITEMS_NR:
4050 case FLUSH_DELAYED_ITEMS:
4051 if (state == FLUSH_DELAYED_ITEMS_NR) {
4052 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
4054 nr = (int)div64_u64(num_bytes, bytes);
4061 trans = btrfs_join_transaction(root);
4062 if (IS_ERR(trans)) {
4063 ret = PTR_ERR(trans);
4066 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4067 btrfs_end_transaction(trans, root);
4069 case FLUSH_DELALLOC:
4070 case FLUSH_DELALLOC_WAIT:
4071 shrink_delalloc(root, num_bytes, orig_bytes,
4072 state == FLUSH_DELALLOC_WAIT);
4075 trans = btrfs_join_transaction(root);
4076 if (IS_ERR(trans)) {
4077 ret = PTR_ERR(trans);
4080 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4081 btrfs_get_alloc_profile(root, 0),
4082 CHUNK_ALLOC_NO_FORCE);
4083 btrfs_end_transaction(trans, root);
4088 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4098 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4099 * @root - the root we're allocating for
4100 * @block_rsv - the block_rsv we're allocating for
4101 * @orig_bytes - the number of bytes we want
4102 * @flush - whether or not we can flush to make our reservation
4104 * This will reserve orgi_bytes number of bytes from the space info associated
4105 * with the block_rsv. If there is not enough space it will make an attempt to
4106 * flush out space to make room. It will do this by flushing delalloc if
4107 * possible or committing the transaction. If flush is 0 then no attempts to
4108 * regain reservations will be made and this will fail if there is not enough
4111 static int reserve_metadata_bytes(struct btrfs_root *root,
4112 struct btrfs_block_rsv *block_rsv,
4114 enum btrfs_reserve_flush_enum flush)
4116 struct btrfs_space_info *space_info = block_rsv->space_info;
4118 u64 num_bytes = orig_bytes;
4119 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4121 bool flushing = false;
4125 spin_lock(&space_info->lock);
4127 * We only want to wait if somebody other than us is flushing and we
4128 * are actually allowed to flush all things.
4130 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4131 space_info->flush) {
4132 spin_unlock(&space_info->lock);
4134 * If we have a trans handle we can't wait because the flusher
4135 * may have to commit the transaction, which would mean we would
4136 * deadlock since we are waiting for the flusher to finish, but
4137 * hold the current transaction open.
4139 if (current->journal_info)
4141 ret = wait_event_killable(space_info->wait, !space_info->flush);
4142 /* Must have been killed, return */
4146 spin_lock(&space_info->lock);
4150 used = space_info->bytes_used + space_info->bytes_reserved +
4151 space_info->bytes_pinned + space_info->bytes_readonly +
4152 space_info->bytes_may_use;
4155 * The idea here is that we've not already over-reserved the block group
4156 * then we can go ahead and save our reservation first and then start
4157 * flushing if we need to. Otherwise if we've already overcommitted
4158 * lets start flushing stuff first and then come back and try to make
4161 if (used <= space_info->total_bytes) {
4162 if (used + orig_bytes <= space_info->total_bytes) {
4163 space_info->bytes_may_use += orig_bytes;
4164 trace_btrfs_space_reservation(root->fs_info,
4165 "space_info", space_info->flags, orig_bytes, 1);
4169 * Ok set num_bytes to orig_bytes since we aren't
4170 * overocmmitted, this way we only try and reclaim what
4173 num_bytes = orig_bytes;
4177 * Ok we're over committed, set num_bytes to the overcommitted
4178 * amount plus the amount of bytes that we need for this
4181 num_bytes = used - space_info->total_bytes +
4185 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4186 space_info->bytes_may_use += orig_bytes;
4187 trace_btrfs_space_reservation(root->fs_info, "space_info",
4188 space_info->flags, orig_bytes,
4194 * Couldn't make our reservation, save our place so while we're trying
4195 * to reclaim space we can actually use it instead of somebody else
4196 * stealing it from us.
4198 * We make the other tasks wait for the flush only when we can flush
4201 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4203 space_info->flush = 1;
4206 spin_unlock(&space_info->lock);
4208 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4211 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4216 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4217 * would happen. So skip delalloc flush.
4219 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4220 (flush_state == FLUSH_DELALLOC ||
4221 flush_state == FLUSH_DELALLOC_WAIT))
4222 flush_state = ALLOC_CHUNK;
4226 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4227 flush_state < COMMIT_TRANS)
4229 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4230 flush_state <= COMMIT_TRANS)
4234 if (ret == -ENOSPC &&
4235 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4236 struct btrfs_block_rsv *global_rsv =
4237 &root->fs_info->global_block_rsv;
4239 if (block_rsv != global_rsv &&
4240 !block_rsv_use_bytes(global_rsv, orig_bytes))
4244 spin_lock(&space_info->lock);
4245 space_info->flush = 0;
4246 wake_up_all(&space_info->wait);
4247 spin_unlock(&space_info->lock);
4252 static struct btrfs_block_rsv *get_block_rsv(
4253 const struct btrfs_trans_handle *trans,
4254 const struct btrfs_root *root)
4256 struct btrfs_block_rsv *block_rsv = NULL;
4259 block_rsv = trans->block_rsv;
4261 if (root == root->fs_info->csum_root && trans->adding_csums)
4262 block_rsv = trans->block_rsv;
4265 block_rsv = root->block_rsv;
4268 block_rsv = &root->fs_info->empty_block_rsv;
4273 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4277 spin_lock(&block_rsv->lock);
4278 if (block_rsv->reserved >= num_bytes) {
4279 block_rsv->reserved -= num_bytes;
4280 if (block_rsv->reserved < block_rsv->size)
4281 block_rsv->full = 0;
4284 spin_unlock(&block_rsv->lock);
4288 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4289 u64 num_bytes, int update_size)
4291 spin_lock(&block_rsv->lock);
4292 block_rsv->reserved += num_bytes;
4294 block_rsv->size += num_bytes;
4295 else if (block_rsv->reserved >= block_rsv->size)
4296 block_rsv->full = 1;
4297 spin_unlock(&block_rsv->lock);
4300 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4301 struct btrfs_block_rsv *block_rsv,
4302 struct btrfs_block_rsv *dest, u64 num_bytes)
4304 struct btrfs_space_info *space_info = block_rsv->space_info;
4306 spin_lock(&block_rsv->lock);
4307 if (num_bytes == (u64)-1)
4308 num_bytes = block_rsv->size;
4309 block_rsv->size -= num_bytes;
4310 if (block_rsv->reserved >= block_rsv->size) {
4311 num_bytes = block_rsv->reserved - block_rsv->size;
4312 block_rsv->reserved = block_rsv->size;
4313 block_rsv->full = 1;
4317 spin_unlock(&block_rsv->lock);
4319 if (num_bytes > 0) {
4321 spin_lock(&dest->lock);
4325 bytes_to_add = dest->size - dest->reserved;
4326 bytes_to_add = min(num_bytes, bytes_to_add);
4327 dest->reserved += bytes_to_add;
4328 if (dest->reserved >= dest->size)
4330 num_bytes -= bytes_to_add;
4332 spin_unlock(&dest->lock);
4335 spin_lock(&space_info->lock);
4336 space_info->bytes_may_use -= num_bytes;
4337 trace_btrfs_space_reservation(fs_info, "space_info",
4338 space_info->flags, num_bytes, 0);
4339 space_info->reservation_progress++;
4340 spin_unlock(&space_info->lock);
4345 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4346 struct btrfs_block_rsv *dst, u64 num_bytes)
4350 ret = block_rsv_use_bytes(src, num_bytes);
4354 block_rsv_add_bytes(dst, num_bytes, 1);
4358 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4360 memset(rsv, 0, sizeof(*rsv));
4361 spin_lock_init(&rsv->lock);
4365 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4366 unsigned short type)
4368 struct btrfs_block_rsv *block_rsv;
4369 struct btrfs_fs_info *fs_info = root->fs_info;
4371 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4375 btrfs_init_block_rsv(block_rsv, type);
4376 block_rsv->space_info = __find_space_info(fs_info,
4377 BTRFS_BLOCK_GROUP_METADATA);
4381 void btrfs_free_block_rsv(struct btrfs_root *root,
4382 struct btrfs_block_rsv *rsv)
4386 btrfs_block_rsv_release(root, rsv, (u64)-1);
4390 int btrfs_block_rsv_add(struct btrfs_root *root,
4391 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4392 enum btrfs_reserve_flush_enum flush)
4399 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4401 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4408 int btrfs_block_rsv_check(struct btrfs_root *root,
4409 struct btrfs_block_rsv *block_rsv, int min_factor)
4417 spin_lock(&block_rsv->lock);
4418 num_bytes = div_factor(block_rsv->size, min_factor);
4419 if (block_rsv->reserved >= num_bytes)
4421 spin_unlock(&block_rsv->lock);
4426 int btrfs_block_rsv_refill(struct btrfs_root *root,
4427 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4428 enum btrfs_reserve_flush_enum flush)
4436 spin_lock(&block_rsv->lock);
4437 num_bytes = min_reserved;
4438 if (block_rsv->reserved >= num_bytes)
4441 num_bytes -= block_rsv->reserved;
4442 spin_unlock(&block_rsv->lock);
4447 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4449 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4456 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4457 struct btrfs_block_rsv *dst_rsv,
4460 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4463 void btrfs_block_rsv_release(struct btrfs_root *root,
4464 struct btrfs_block_rsv *block_rsv,
4467 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4468 if (global_rsv->full || global_rsv == block_rsv ||
4469 block_rsv->space_info != global_rsv->space_info)
4471 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4476 * helper to calculate size of global block reservation.
4477 * the desired value is sum of space used by extent tree,
4478 * checksum tree and root tree
4480 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4482 struct btrfs_space_info *sinfo;
4486 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4488 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4489 spin_lock(&sinfo->lock);
4490 data_used = sinfo->bytes_used;
4491 spin_unlock(&sinfo->lock);
4493 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4494 spin_lock(&sinfo->lock);
4495 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4497 meta_used = sinfo->bytes_used;
4498 spin_unlock(&sinfo->lock);
4500 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4502 num_bytes += div64_u64(data_used + meta_used, 50);
4504 if (num_bytes * 3 > meta_used)
4505 num_bytes = div64_u64(meta_used, 3);
4507 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4510 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4512 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4513 struct btrfs_space_info *sinfo = block_rsv->space_info;
4516 num_bytes = calc_global_metadata_size(fs_info);
4518 spin_lock(&sinfo->lock);
4519 spin_lock(&block_rsv->lock);
4521 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4523 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4524 sinfo->bytes_reserved + sinfo->bytes_readonly +
4525 sinfo->bytes_may_use;
4527 if (sinfo->total_bytes > num_bytes) {
4528 num_bytes = sinfo->total_bytes - num_bytes;
4529 block_rsv->reserved += num_bytes;
4530 sinfo->bytes_may_use += num_bytes;
4531 trace_btrfs_space_reservation(fs_info, "space_info",
4532 sinfo->flags, num_bytes, 1);
4535 if (block_rsv->reserved >= block_rsv->size) {
4536 num_bytes = block_rsv->reserved - block_rsv->size;
4537 sinfo->bytes_may_use -= num_bytes;
4538 trace_btrfs_space_reservation(fs_info, "space_info",
4539 sinfo->flags, num_bytes, 0);
4540 sinfo->reservation_progress++;
4541 block_rsv->reserved = block_rsv->size;
4542 block_rsv->full = 1;
4545 spin_unlock(&block_rsv->lock);
4546 spin_unlock(&sinfo->lock);
4549 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4551 struct btrfs_space_info *space_info;
4553 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4554 fs_info->chunk_block_rsv.space_info = space_info;
4556 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4557 fs_info->global_block_rsv.space_info = space_info;
4558 fs_info->delalloc_block_rsv.space_info = space_info;
4559 fs_info->trans_block_rsv.space_info = space_info;
4560 fs_info->empty_block_rsv.space_info = space_info;
4561 fs_info->delayed_block_rsv.space_info = space_info;
4563 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4564 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4565 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4566 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4567 if (fs_info->quota_root)
4568 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4569 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4571 update_global_block_rsv(fs_info);
4574 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4576 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4578 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4579 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4580 WARN_ON(fs_info->trans_block_rsv.size > 0);
4581 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4582 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4583 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4584 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4585 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4588 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4589 struct btrfs_root *root)
4591 if (!trans->block_rsv)
4594 if (!trans->bytes_reserved)
4597 trace_btrfs_space_reservation(root->fs_info, "transaction",
4598 trans->transid, trans->bytes_reserved, 0);
4599 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4600 trans->bytes_reserved = 0;
4603 /* Can only return 0 or -ENOSPC */
4604 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4605 struct inode *inode)
4607 struct btrfs_root *root = BTRFS_I(inode)->root;
4608 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4609 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4612 * We need to hold space in order to delete our orphan item once we've
4613 * added it, so this takes the reservation so we can release it later
4614 * when we are truly done with the orphan item.
4616 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4617 trace_btrfs_space_reservation(root->fs_info, "orphan",
4618 btrfs_ino(inode), num_bytes, 1);
4619 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4622 void btrfs_orphan_release_metadata(struct inode *inode)
4624 struct btrfs_root *root = BTRFS_I(inode)->root;
4625 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4626 trace_btrfs_space_reservation(root->fs_info, "orphan",
4627 btrfs_ino(inode), num_bytes, 0);
4628 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4632 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4633 * root: the root of the parent directory
4634 * rsv: block reservation
4635 * items: the number of items that we need do reservation
4636 * qgroup_reserved: used to return the reserved size in qgroup
4638 * This function is used to reserve the space for snapshot/subvolume
4639 * creation and deletion. Those operations are different with the
4640 * common file/directory operations, they change two fs/file trees
4641 * and root tree, the number of items that the qgroup reserves is
4642 * different with the free space reservation. So we can not use
4643 * the space reseravtion mechanism in start_transaction().
4645 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4646 struct btrfs_block_rsv *rsv,
4648 u64 *qgroup_reserved)
4653 if (root->fs_info->quota_enabled) {
4654 /* One for parent inode, two for dir entries */
4655 num_bytes = 3 * root->leafsize;
4656 ret = btrfs_qgroup_reserve(root, num_bytes);
4663 *qgroup_reserved = num_bytes;
4665 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4666 rsv->space_info = __find_space_info(root->fs_info,
4667 BTRFS_BLOCK_GROUP_METADATA);
4668 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4669 BTRFS_RESERVE_FLUSH_ALL);
4671 if (*qgroup_reserved)
4672 btrfs_qgroup_free(root, *qgroup_reserved);
4678 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4679 struct btrfs_block_rsv *rsv,
4680 u64 qgroup_reserved)
4682 btrfs_block_rsv_release(root, rsv, (u64)-1);
4683 if (qgroup_reserved)
4684 btrfs_qgroup_free(root, qgroup_reserved);
4688 * drop_outstanding_extent - drop an outstanding extent
4689 * @inode: the inode we're dropping the extent for
4691 * This is called when we are freeing up an outstanding extent, either called
4692 * after an error or after an extent is written. This will return the number of
4693 * reserved extents that need to be freed. This must be called with
4694 * BTRFS_I(inode)->lock held.
4696 static unsigned drop_outstanding_extent(struct inode *inode)
4698 unsigned drop_inode_space = 0;
4699 unsigned dropped_extents = 0;
4701 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4702 BTRFS_I(inode)->outstanding_extents--;
4704 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4705 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4706 &BTRFS_I(inode)->runtime_flags))
4707 drop_inode_space = 1;
4710 * If we have more or the same amount of outsanding extents than we have
4711 * reserved then we need to leave the reserved extents count alone.
4713 if (BTRFS_I(inode)->outstanding_extents >=
4714 BTRFS_I(inode)->reserved_extents)
4715 return drop_inode_space;
4717 dropped_extents = BTRFS_I(inode)->reserved_extents -
4718 BTRFS_I(inode)->outstanding_extents;
4719 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4720 return dropped_extents + drop_inode_space;
4724 * calc_csum_metadata_size - return the amount of metada space that must be
4725 * reserved/free'd for the given bytes.
4726 * @inode: the inode we're manipulating
4727 * @num_bytes: the number of bytes in question
4728 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4730 * This adjusts the number of csum_bytes in the inode and then returns the
4731 * correct amount of metadata that must either be reserved or freed. We
4732 * calculate how many checksums we can fit into one leaf and then divide the
4733 * number of bytes that will need to be checksumed by this value to figure out
4734 * how many checksums will be required. If we are adding bytes then the number
4735 * may go up and we will return the number of additional bytes that must be
4736 * reserved. If it is going down we will return the number of bytes that must
4739 * This must be called with BTRFS_I(inode)->lock held.
4741 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4744 struct btrfs_root *root = BTRFS_I(inode)->root;
4746 int num_csums_per_leaf;
4750 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4751 BTRFS_I(inode)->csum_bytes == 0)
4754 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4756 BTRFS_I(inode)->csum_bytes += num_bytes;
4758 BTRFS_I(inode)->csum_bytes -= num_bytes;
4759 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4760 num_csums_per_leaf = (int)div64_u64(csum_size,
4761 sizeof(struct btrfs_csum_item) +
4762 sizeof(struct btrfs_disk_key));
4763 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4764 num_csums = num_csums + num_csums_per_leaf - 1;
4765 num_csums = num_csums / num_csums_per_leaf;
4767 old_csums = old_csums + num_csums_per_leaf - 1;
4768 old_csums = old_csums / num_csums_per_leaf;
4770 /* No change, no need to reserve more */
4771 if (old_csums == num_csums)
4775 return btrfs_calc_trans_metadata_size(root,
4776 num_csums - old_csums);
4778 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4781 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4783 struct btrfs_root *root = BTRFS_I(inode)->root;
4784 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4787 unsigned nr_extents = 0;
4788 int extra_reserve = 0;
4789 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4791 bool delalloc_lock = true;
4795 /* If we are a free space inode we need to not flush since we will be in
4796 * the middle of a transaction commit. We also don't need the delalloc
4797 * mutex since we won't race with anybody. We need this mostly to make
4798 * lockdep shut its filthy mouth.
4800 if (btrfs_is_free_space_inode(inode)) {
4801 flush = BTRFS_RESERVE_NO_FLUSH;
4802 delalloc_lock = false;
4805 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4806 btrfs_transaction_in_commit(root->fs_info))
4807 schedule_timeout(1);
4810 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4812 num_bytes = ALIGN(num_bytes, root->sectorsize);
4814 spin_lock(&BTRFS_I(inode)->lock);
4815 BTRFS_I(inode)->outstanding_extents++;
4817 if (BTRFS_I(inode)->outstanding_extents >
4818 BTRFS_I(inode)->reserved_extents)
4819 nr_extents = BTRFS_I(inode)->outstanding_extents -
4820 BTRFS_I(inode)->reserved_extents;
4823 * Add an item to reserve for updating the inode when we complete the
4826 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4827 &BTRFS_I(inode)->runtime_flags)) {
4832 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4833 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4834 csum_bytes = BTRFS_I(inode)->csum_bytes;
4835 spin_unlock(&BTRFS_I(inode)->lock);
4837 if (root->fs_info->quota_enabled) {
4838 ret = btrfs_qgroup_reserve(root, num_bytes +
4839 nr_extents * root->leafsize);
4844 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4845 if (unlikely(ret)) {
4846 if (root->fs_info->quota_enabled)
4847 btrfs_qgroup_free(root, num_bytes +
4848 nr_extents * root->leafsize);
4852 spin_lock(&BTRFS_I(inode)->lock);
4853 if (extra_reserve) {
4854 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4855 &BTRFS_I(inode)->runtime_flags);
4858 BTRFS_I(inode)->reserved_extents += nr_extents;
4859 spin_unlock(&BTRFS_I(inode)->lock);
4862 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4865 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4866 btrfs_ino(inode), to_reserve, 1);
4867 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4872 spin_lock(&BTRFS_I(inode)->lock);
4873 dropped = drop_outstanding_extent(inode);
4875 * If the inodes csum_bytes is the same as the original
4876 * csum_bytes then we know we haven't raced with any free()ers
4877 * so we can just reduce our inodes csum bytes and carry on.
4879 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
4880 calc_csum_metadata_size(inode, num_bytes, 0);
4882 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
4886 * This is tricky, but first we need to figure out how much we
4887 * free'd from any free-ers that occured during this
4888 * reservation, so we reset ->csum_bytes to the csum_bytes
4889 * before we dropped our lock, and then call the free for the
4890 * number of bytes that were freed while we were trying our
4893 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
4894 BTRFS_I(inode)->csum_bytes = csum_bytes;
4895 to_free = calc_csum_metadata_size(inode, bytes, 0);
4899 * Now we need to see how much we would have freed had we not
4900 * been making this reservation and our ->csum_bytes were not
4901 * artificially inflated.
4903 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
4904 bytes = csum_bytes - orig_csum_bytes;
4905 bytes = calc_csum_metadata_size(inode, bytes, 0);
4908 * Now reset ->csum_bytes to what it should be. If bytes is
4909 * more than to_free then we would have free'd more space had we
4910 * not had an artificially high ->csum_bytes, so we need to free
4911 * the remainder. If bytes is the same or less then we don't
4912 * need to do anything, the other free-ers did the correct
4915 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
4916 if (bytes > to_free)
4917 to_free = bytes - to_free;
4921 spin_unlock(&BTRFS_I(inode)->lock);
4923 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4926 btrfs_block_rsv_release(root, block_rsv, to_free);
4927 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4928 btrfs_ino(inode), to_free, 0);
4931 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4936 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4937 * @inode: the inode to release the reservation for
4938 * @num_bytes: the number of bytes we're releasing
4940 * This will release the metadata reservation for an inode. This can be called
4941 * once we complete IO for a given set of bytes to release their metadata
4944 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4946 struct btrfs_root *root = BTRFS_I(inode)->root;
4950 num_bytes = ALIGN(num_bytes, root->sectorsize);
4951 spin_lock(&BTRFS_I(inode)->lock);
4952 dropped = drop_outstanding_extent(inode);
4955 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4956 spin_unlock(&BTRFS_I(inode)->lock);
4958 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4960 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4961 btrfs_ino(inode), to_free, 0);
4962 if (root->fs_info->quota_enabled) {
4963 btrfs_qgroup_free(root, num_bytes +
4964 dropped * root->leafsize);
4967 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4972 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4973 * @inode: inode we're writing to
4974 * @num_bytes: the number of bytes we want to allocate
4976 * This will do the following things
4978 * o reserve space in the data space info for num_bytes
4979 * o reserve space in the metadata space info based on number of outstanding
4980 * extents and how much csums will be needed
4981 * o add to the inodes ->delalloc_bytes
4982 * o add it to the fs_info's delalloc inodes list.
4984 * This will return 0 for success and -ENOSPC if there is no space left.
4986 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4990 ret = btrfs_check_data_free_space(inode, num_bytes);
4994 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4996 btrfs_free_reserved_data_space(inode, num_bytes);
5004 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5005 * @inode: inode we're releasing space for
5006 * @num_bytes: the number of bytes we want to free up
5008 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5009 * called in the case that we don't need the metadata AND data reservations
5010 * anymore. So if there is an error or we insert an inline extent.
5012 * This function will release the metadata space that was not used and will
5013 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5014 * list if there are no delalloc bytes left.
5016 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5018 btrfs_delalloc_release_metadata(inode, num_bytes);
5019 btrfs_free_reserved_data_space(inode, num_bytes);
5022 static int update_block_group(struct btrfs_root *root,
5023 u64 bytenr, u64 num_bytes, int alloc)
5025 struct btrfs_block_group_cache *cache = NULL;
5026 struct btrfs_fs_info *info = root->fs_info;
5027 u64 total = num_bytes;
5032 /* block accounting for super block */
5033 spin_lock(&info->delalloc_lock);
5034 old_val = btrfs_super_bytes_used(info->super_copy);
5036 old_val += num_bytes;
5038 old_val -= num_bytes;
5039 btrfs_set_super_bytes_used(info->super_copy, old_val);
5040 spin_unlock(&info->delalloc_lock);
5043 cache = btrfs_lookup_block_group(info, bytenr);
5046 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5047 BTRFS_BLOCK_GROUP_RAID1 |
5048 BTRFS_BLOCK_GROUP_RAID10))
5053 * If this block group has free space cache written out, we
5054 * need to make sure to load it if we are removing space. This
5055 * is because we need the unpinning stage to actually add the
5056 * space back to the block group, otherwise we will leak space.
5058 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5059 cache_block_group(cache, 1);
5061 byte_in_group = bytenr - cache->key.objectid;
5062 WARN_ON(byte_in_group > cache->key.offset);
5064 spin_lock(&cache->space_info->lock);
5065 spin_lock(&cache->lock);
5067 if (btrfs_test_opt(root, SPACE_CACHE) &&
5068 cache->disk_cache_state < BTRFS_DC_CLEAR)
5069 cache->disk_cache_state = BTRFS_DC_CLEAR;
5072 old_val = btrfs_block_group_used(&cache->item);
5073 num_bytes = min(total, cache->key.offset - byte_in_group);
5075 old_val += num_bytes;
5076 btrfs_set_block_group_used(&cache->item, old_val);
5077 cache->reserved -= num_bytes;
5078 cache->space_info->bytes_reserved -= num_bytes;
5079 cache->space_info->bytes_used += num_bytes;
5080 cache->space_info->disk_used += num_bytes * factor;
5081 spin_unlock(&cache->lock);
5082 spin_unlock(&cache->space_info->lock);
5084 old_val -= num_bytes;
5085 btrfs_set_block_group_used(&cache->item, old_val);
5086 cache->pinned += num_bytes;
5087 cache->space_info->bytes_pinned += num_bytes;
5088 cache->space_info->bytes_used -= num_bytes;
5089 cache->space_info->disk_used -= num_bytes * factor;
5090 spin_unlock(&cache->lock);
5091 spin_unlock(&cache->space_info->lock);
5093 set_extent_dirty(info->pinned_extents,
5094 bytenr, bytenr + num_bytes - 1,
5095 GFP_NOFS | __GFP_NOFAIL);
5097 btrfs_put_block_group(cache);
5099 bytenr += num_bytes;
5104 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5106 struct btrfs_block_group_cache *cache;
5109 spin_lock(&root->fs_info->block_group_cache_lock);
5110 bytenr = root->fs_info->first_logical_byte;
5111 spin_unlock(&root->fs_info->block_group_cache_lock);
5113 if (bytenr < (u64)-1)
5116 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5120 bytenr = cache->key.objectid;
5121 btrfs_put_block_group(cache);
5126 static int pin_down_extent(struct btrfs_root *root,
5127 struct btrfs_block_group_cache *cache,
5128 u64 bytenr, u64 num_bytes, int reserved)
5130 spin_lock(&cache->space_info->lock);
5131 spin_lock(&cache->lock);
5132 cache->pinned += num_bytes;
5133 cache->space_info->bytes_pinned += num_bytes;
5135 cache->reserved -= num_bytes;
5136 cache->space_info->bytes_reserved -= num_bytes;
5138 spin_unlock(&cache->lock);
5139 spin_unlock(&cache->space_info->lock);
5141 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5142 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5147 * this function must be called within transaction
5149 int btrfs_pin_extent(struct btrfs_root *root,
5150 u64 bytenr, u64 num_bytes, int reserved)
5152 struct btrfs_block_group_cache *cache;
5154 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5155 BUG_ON(!cache); /* Logic error */
5157 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5159 btrfs_put_block_group(cache);
5164 * this function must be called within transaction
5166 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5167 u64 bytenr, u64 num_bytes)
5169 struct btrfs_block_group_cache *cache;
5172 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5177 * pull in the free space cache (if any) so that our pin
5178 * removes the free space from the cache. We have load_only set
5179 * to one because the slow code to read in the free extents does check
5180 * the pinned extents.
5182 cache_block_group(cache, 1);
5184 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5186 /* remove us from the free space cache (if we're there at all) */
5187 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5188 btrfs_put_block_group(cache);
5193 * btrfs_update_reserved_bytes - update the block_group and space info counters
5194 * @cache: The cache we are manipulating
5195 * @num_bytes: The number of bytes in question
5196 * @reserve: One of the reservation enums
5198 * This is called by the allocator when it reserves space, or by somebody who is
5199 * freeing space that was never actually used on disk. For example if you
5200 * reserve some space for a new leaf in transaction A and before transaction A
5201 * commits you free that leaf, you call this with reserve set to 0 in order to
5202 * clear the reservation.
5204 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5205 * ENOSPC accounting. For data we handle the reservation through clearing the
5206 * delalloc bits in the io_tree. We have to do this since we could end up
5207 * allocating less disk space for the amount of data we have reserved in the
5208 * case of compression.
5210 * If this is a reservation and the block group has become read only we cannot
5211 * make the reservation and return -EAGAIN, otherwise this function always
5214 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5215 u64 num_bytes, int reserve)
5217 struct btrfs_space_info *space_info = cache->space_info;
5220 spin_lock(&space_info->lock);
5221 spin_lock(&cache->lock);
5222 if (reserve != RESERVE_FREE) {
5226 cache->reserved += num_bytes;
5227 space_info->bytes_reserved += num_bytes;
5228 if (reserve == RESERVE_ALLOC) {
5229 trace_btrfs_space_reservation(cache->fs_info,
5230 "space_info", space_info->flags,
5232 space_info->bytes_may_use -= num_bytes;
5237 space_info->bytes_readonly += num_bytes;
5238 cache->reserved -= num_bytes;
5239 space_info->bytes_reserved -= num_bytes;
5240 space_info->reservation_progress++;
5242 spin_unlock(&cache->lock);
5243 spin_unlock(&space_info->lock);
5247 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5248 struct btrfs_root *root)
5250 struct btrfs_fs_info *fs_info = root->fs_info;
5251 struct btrfs_caching_control *next;
5252 struct btrfs_caching_control *caching_ctl;
5253 struct btrfs_block_group_cache *cache;
5255 down_write(&fs_info->extent_commit_sem);
5257 list_for_each_entry_safe(caching_ctl, next,
5258 &fs_info->caching_block_groups, list) {
5259 cache = caching_ctl->block_group;
5260 if (block_group_cache_done(cache)) {
5261 cache->last_byte_to_unpin = (u64)-1;
5262 list_del_init(&caching_ctl->list);
5263 put_caching_control(caching_ctl);
5265 cache->last_byte_to_unpin = caching_ctl->progress;
5269 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5270 fs_info->pinned_extents = &fs_info->freed_extents[1];
5272 fs_info->pinned_extents = &fs_info->freed_extents[0];
5274 up_write(&fs_info->extent_commit_sem);
5276 update_global_block_rsv(fs_info);
5279 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5281 struct btrfs_fs_info *fs_info = root->fs_info;
5282 struct btrfs_block_group_cache *cache = NULL;
5283 struct btrfs_space_info *space_info;
5284 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5288 while (start <= end) {
5291 start >= cache->key.objectid + cache->key.offset) {
5293 btrfs_put_block_group(cache);
5294 cache = btrfs_lookup_block_group(fs_info, start);
5295 BUG_ON(!cache); /* Logic error */
5298 len = cache->key.objectid + cache->key.offset - start;
5299 len = min(len, end + 1 - start);
5301 if (start < cache->last_byte_to_unpin) {
5302 len = min(len, cache->last_byte_to_unpin - start);
5303 btrfs_add_free_space(cache, start, len);
5307 space_info = cache->space_info;
5309 spin_lock(&space_info->lock);
5310 spin_lock(&cache->lock);
5311 cache->pinned -= len;
5312 space_info->bytes_pinned -= len;
5314 space_info->bytes_readonly += len;
5317 spin_unlock(&cache->lock);
5318 if (!readonly && global_rsv->space_info == space_info) {
5319 spin_lock(&global_rsv->lock);
5320 if (!global_rsv->full) {
5321 len = min(len, global_rsv->size -
5322 global_rsv->reserved);
5323 global_rsv->reserved += len;
5324 space_info->bytes_may_use += len;
5325 if (global_rsv->reserved >= global_rsv->size)
5326 global_rsv->full = 1;
5328 spin_unlock(&global_rsv->lock);
5330 spin_unlock(&space_info->lock);
5334 btrfs_put_block_group(cache);
5338 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5339 struct btrfs_root *root)
5341 struct btrfs_fs_info *fs_info = root->fs_info;
5342 struct extent_io_tree *unpin;
5350 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5351 unpin = &fs_info->freed_extents[1];
5353 unpin = &fs_info->freed_extents[0];
5356 ret = find_first_extent_bit(unpin, 0, &start, &end,
5357 EXTENT_DIRTY, NULL);
5361 if (btrfs_test_opt(root, DISCARD))
5362 ret = btrfs_discard_extent(root, start,
5363 end + 1 - start, NULL);
5365 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5366 unpin_extent_range(root, start, end);
5373 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5374 struct btrfs_root *root,
5375 u64 bytenr, u64 num_bytes, u64 parent,
5376 u64 root_objectid, u64 owner_objectid,
5377 u64 owner_offset, int refs_to_drop,
5378 struct btrfs_delayed_extent_op *extent_op)
5380 struct btrfs_key key;
5381 struct btrfs_path *path;
5382 struct btrfs_fs_info *info = root->fs_info;
5383 struct btrfs_root *extent_root = info->extent_root;
5384 struct extent_buffer *leaf;
5385 struct btrfs_extent_item *ei;
5386 struct btrfs_extent_inline_ref *iref;
5389 int extent_slot = 0;
5390 int found_extent = 0;
5394 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5397 path = btrfs_alloc_path();
5402 path->leave_spinning = 1;
5404 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5405 BUG_ON(!is_data && refs_to_drop != 1);
5408 skinny_metadata = 0;
5410 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5411 bytenr, num_bytes, parent,
5412 root_objectid, owner_objectid,
5415 extent_slot = path->slots[0];
5416 while (extent_slot >= 0) {
5417 btrfs_item_key_to_cpu(path->nodes[0], &key,
5419 if (key.objectid != bytenr)
5421 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5422 key.offset == num_bytes) {
5426 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5427 key.offset == owner_objectid) {
5431 if (path->slots[0] - extent_slot > 5)
5435 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5436 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5437 if (found_extent && item_size < sizeof(*ei))
5440 if (!found_extent) {
5442 ret = remove_extent_backref(trans, extent_root, path,
5446 btrfs_abort_transaction(trans, extent_root, ret);
5449 btrfs_release_path(path);
5450 path->leave_spinning = 1;
5452 key.objectid = bytenr;
5453 key.type = BTRFS_EXTENT_ITEM_KEY;
5454 key.offset = num_bytes;
5456 if (!is_data && skinny_metadata) {
5457 key.type = BTRFS_METADATA_ITEM_KEY;
5458 key.offset = owner_objectid;
5461 ret = btrfs_search_slot(trans, extent_root,
5463 if (ret > 0 && skinny_metadata && path->slots[0]) {
5465 * Couldn't find our skinny metadata item,
5466 * see if we have ye olde extent item.
5469 btrfs_item_key_to_cpu(path->nodes[0], &key,
5471 if (key.objectid == bytenr &&
5472 key.type == BTRFS_EXTENT_ITEM_KEY &&
5473 key.offset == num_bytes)
5477 if (ret > 0 && skinny_metadata) {
5478 skinny_metadata = false;
5479 key.type = BTRFS_EXTENT_ITEM_KEY;
5480 key.offset = num_bytes;
5481 btrfs_release_path(path);
5482 ret = btrfs_search_slot(trans, extent_root,
5487 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5488 ret, (unsigned long long)bytenr);
5490 btrfs_print_leaf(extent_root,
5494 btrfs_abort_transaction(trans, extent_root, ret);
5497 extent_slot = path->slots[0];
5499 } else if (ret == -ENOENT) {
5500 btrfs_print_leaf(extent_root, path->nodes[0]);
5503 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5504 (unsigned long long)bytenr,
5505 (unsigned long long)parent,
5506 (unsigned long long)root_objectid,
5507 (unsigned long long)owner_objectid,
5508 (unsigned long long)owner_offset);
5510 btrfs_abort_transaction(trans, extent_root, ret);
5514 leaf = path->nodes[0];
5515 item_size = btrfs_item_size_nr(leaf, extent_slot);
5516 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5517 if (item_size < sizeof(*ei)) {
5518 BUG_ON(found_extent || extent_slot != path->slots[0]);
5519 ret = convert_extent_item_v0(trans, extent_root, path,
5522 btrfs_abort_transaction(trans, extent_root, ret);
5526 btrfs_release_path(path);
5527 path->leave_spinning = 1;
5529 key.objectid = bytenr;
5530 key.type = BTRFS_EXTENT_ITEM_KEY;
5531 key.offset = num_bytes;
5533 ret = btrfs_search_slot(trans, extent_root, &key, path,
5536 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5537 ret, (unsigned long long)bytenr);
5538 btrfs_print_leaf(extent_root, path->nodes[0]);
5541 btrfs_abort_transaction(trans, extent_root, ret);
5545 extent_slot = path->slots[0];
5546 leaf = path->nodes[0];
5547 item_size = btrfs_item_size_nr(leaf, extent_slot);
5550 BUG_ON(item_size < sizeof(*ei));
5551 ei = btrfs_item_ptr(leaf, extent_slot,
5552 struct btrfs_extent_item);
5553 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5554 key.type == BTRFS_EXTENT_ITEM_KEY) {
5555 struct btrfs_tree_block_info *bi;
5556 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5557 bi = (struct btrfs_tree_block_info *)(ei + 1);
5558 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5561 refs = btrfs_extent_refs(leaf, ei);
5562 if (refs < refs_to_drop) {
5563 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5564 "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5566 btrfs_abort_transaction(trans, extent_root, ret);
5569 refs -= refs_to_drop;
5573 __run_delayed_extent_op(extent_op, leaf, ei);
5575 * In the case of inline back ref, reference count will
5576 * be updated by remove_extent_backref
5579 BUG_ON(!found_extent);
5581 btrfs_set_extent_refs(leaf, ei, refs);
5582 btrfs_mark_buffer_dirty(leaf);
5585 ret = remove_extent_backref(trans, extent_root, path,
5589 btrfs_abort_transaction(trans, extent_root, ret);
5595 BUG_ON(is_data && refs_to_drop !=
5596 extent_data_ref_count(root, path, iref));
5598 BUG_ON(path->slots[0] != extent_slot);
5600 BUG_ON(path->slots[0] != extent_slot + 1);
5601 path->slots[0] = extent_slot;
5606 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5609 btrfs_abort_transaction(trans, extent_root, ret);
5612 btrfs_release_path(path);
5615 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5617 btrfs_abort_transaction(trans, extent_root, ret);
5622 ret = update_block_group(root, bytenr, num_bytes, 0);
5624 btrfs_abort_transaction(trans, extent_root, ret);
5629 btrfs_free_path(path);
5634 * when we free an block, it is possible (and likely) that we free the last
5635 * delayed ref for that extent as well. This searches the delayed ref tree for
5636 * a given extent, and if there are no other delayed refs to be processed, it
5637 * removes it from the tree.
5639 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5640 struct btrfs_root *root, u64 bytenr)
5642 struct btrfs_delayed_ref_head *head;
5643 struct btrfs_delayed_ref_root *delayed_refs;
5644 struct btrfs_delayed_ref_node *ref;
5645 struct rb_node *node;
5648 delayed_refs = &trans->transaction->delayed_refs;
5649 spin_lock(&delayed_refs->lock);
5650 head = btrfs_find_delayed_ref_head(trans, bytenr);
5654 node = rb_prev(&head->node.rb_node);
5658 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5660 /* there are still entries for this ref, we can't drop it */
5661 if (ref->bytenr == bytenr)
5664 if (head->extent_op) {
5665 if (!head->must_insert_reserved)
5667 btrfs_free_delayed_extent_op(head->extent_op);
5668 head->extent_op = NULL;
5672 * waiting for the lock here would deadlock. If someone else has it
5673 * locked they are already in the process of dropping it anyway
5675 if (!mutex_trylock(&head->mutex))
5679 * at this point we have a head with no other entries. Go
5680 * ahead and process it.
5682 head->node.in_tree = 0;
5683 rb_erase(&head->node.rb_node, &delayed_refs->root);
5685 delayed_refs->num_entries--;
5688 * we don't take a ref on the node because we're removing it from the
5689 * tree, so we just steal the ref the tree was holding.
5691 delayed_refs->num_heads--;
5692 if (list_empty(&head->cluster))
5693 delayed_refs->num_heads_ready--;
5695 list_del_init(&head->cluster);
5696 spin_unlock(&delayed_refs->lock);
5698 BUG_ON(head->extent_op);
5699 if (head->must_insert_reserved)
5702 mutex_unlock(&head->mutex);
5703 btrfs_put_delayed_ref(&head->node);
5706 spin_unlock(&delayed_refs->lock);
5710 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5711 struct btrfs_root *root,
5712 struct extent_buffer *buf,
5713 u64 parent, int last_ref)
5715 struct btrfs_block_group_cache *cache = NULL;
5718 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5719 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5720 buf->start, buf->len,
5721 parent, root->root_key.objectid,
5722 btrfs_header_level(buf),
5723 BTRFS_DROP_DELAYED_REF, NULL, 0);
5724 BUG_ON(ret); /* -ENOMEM */
5730 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5732 if (btrfs_header_generation(buf) == trans->transid) {
5733 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5734 ret = check_ref_cleanup(trans, root, buf->start);
5739 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5740 pin_down_extent(root, cache, buf->start, buf->len, 1);
5744 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5746 btrfs_add_free_space(cache, buf->start, buf->len);
5747 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5751 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5754 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5755 btrfs_put_block_group(cache);
5758 /* Can return -ENOMEM */
5759 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5760 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5761 u64 owner, u64 offset, int for_cow)
5764 struct btrfs_fs_info *fs_info = root->fs_info;
5767 * tree log blocks never actually go into the extent allocation
5768 * tree, just update pinning info and exit early.
5770 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5771 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5772 /* unlocks the pinned mutex */
5773 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5775 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5776 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5778 parent, root_objectid, (int)owner,
5779 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5781 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5783 parent, root_objectid, owner,
5784 offset, BTRFS_DROP_DELAYED_REF,
5790 static u64 stripe_align(struct btrfs_root *root,
5791 struct btrfs_block_group_cache *cache,
5792 u64 val, u64 num_bytes)
5794 u64 ret = ALIGN(val, root->stripesize);
5799 * when we wait for progress in the block group caching, its because
5800 * our allocation attempt failed at least once. So, we must sleep
5801 * and let some progress happen before we try again.
5803 * This function will sleep at least once waiting for new free space to
5804 * show up, and then it will check the block group free space numbers
5805 * for our min num_bytes. Another option is to have it go ahead
5806 * and look in the rbtree for a free extent of a given size, but this
5810 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5813 struct btrfs_caching_control *caching_ctl;
5815 caching_ctl = get_caching_control(cache);
5819 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5820 (cache->free_space_ctl->free_space >= num_bytes));
5822 put_caching_control(caching_ctl);
5827 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5829 struct btrfs_caching_control *caching_ctl;
5831 caching_ctl = get_caching_control(cache);
5835 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5837 put_caching_control(caching_ctl);
5841 int __get_raid_index(u64 flags)
5843 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5844 return BTRFS_RAID_RAID10;
5845 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5846 return BTRFS_RAID_RAID1;
5847 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5848 return BTRFS_RAID_DUP;
5849 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5850 return BTRFS_RAID_RAID0;
5851 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
5852 return BTRFS_RAID_RAID5;
5853 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
5854 return BTRFS_RAID_RAID6;
5856 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
5859 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5861 return __get_raid_index(cache->flags);
5864 enum btrfs_loop_type {
5865 LOOP_CACHING_NOWAIT = 0,
5866 LOOP_CACHING_WAIT = 1,
5867 LOOP_ALLOC_CHUNK = 2,
5868 LOOP_NO_EMPTY_SIZE = 3,
5872 * walks the btree of allocated extents and find a hole of a given size.
5873 * The key ins is changed to record the hole:
5874 * ins->objectid == block start
5875 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5876 * ins->offset == number of blocks
5877 * Any available blocks before search_start are skipped.
5879 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5880 struct btrfs_root *orig_root,
5881 u64 num_bytes, u64 empty_size,
5882 u64 hint_byte, struct btrfs_key *ins,
5886 struct btrfs_root *root = orig_root->fs_info->extent_root;
5887 struct btrfs_free_cluster *last_ptr = NULL;
5888 struct btrfs_block_group_cache *block_group = NULL;
5889 struct btrfs_block_group_cache *used_block_group;
5890 u64 search_start = 0;
5891 int empty_cluster = 2 * 1024 * 1024;
5892 struct btrfs_space_info *space_info;
5894 int index = __get_raid_index(flags);
5895 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
5896 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5897 bool found_uncached_bg = false;
5898 bool failed_cluster_refill = false;
5899 bool failed_alloc = false;
5900 bool use_cluster = true;
5901 bool have_caching_bg = false;
5903 WARN_ON(num_bytes < root->sectorsize);
5904 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5908 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
5910 space_info = __find_space_info(root->fs_info, flags);
5912 btrfs_err(root->fs_info, "No space info for %llu", flags);
5917 * If the space info is for both data and metadata it means we have a
5918 * small filesystem and we can't use the clustering stuff.
5920 if (btrfs_mixed_space_info(space_info))
5921 use_cluster = false;
5923 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5924 last_ptr = &root->fs_info->meta_alloc_cluster;
5925 if (!btrfs_test_opt(root, SSD))
5926 empty_cluster = 64 * 1024;
5929 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5930 btrfs_test_opt(root, SSD)) {
5931 last_ptr = &root->fs_info->data_alloc_cluster;
5935 spin_lock(&last_ptr->lock);
5936 if (last_ptr->block_group)
5937 hint_byte = last_ptr->window_start;
5938 spin_unlock(&last_ptr->lock);
5941 search_start = max(search_start, first_logical_byte(root, 0));
5942 search_start = max(search_start, hint_byte);
5947 if (search_start == hint_byte) {
5948 block_group = btrfs_lookup_block_group(root->fs_info,
5950 used_block_group = block_group;
5952 * we don't want to use the block group if it doesn't match our
5953 * allocation bits, or if its not cached.
5955 * However if we are re-searching with an ideal block group
5956 * picked out then we don't care that the block group is cached.
5958 if (block_group && block_group_bits(block_group, flags) &&
5959 block_group->cached != BTRFS_CACHE_NO) {
5960 down_read(&space_info->groups_sem);
5961 if (list_empty(&block_group->list) ||
5964 * someone is removing this block group,
5965 * we can't jump into the have_block_group
5966 * target because our list pointers are not
5969 btrfs_put_block_group(block_group);
5970 up_read(&space_info->groups_sem);
5972 index = get_block_group_index(block_group);
5973 goto have_block_group;
5975 } else if (block_group) {
5976 btrfs_put_block_group(block_group);
5980 have_caching_bg = false;
5981 down_read(&space_info->groups_sem);
5982 list_for_each_entry(block_group, &space_info->block_groups[index],
5987 used_block_group = block_group;
5988 btrfs_get_block_group(block_group);
5989 search_start = block_group->key.objectid;
5992 * this can happen if we end up cycling through all the
5993 * raid types, but we want to make sure we only allocate
5994 * for the proper type.
5996 if (!block_group_bits(block_group, flags)) {
5997 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5998 BTRFS_BLOCK_GROUP_RAID1 |
5999 BTRFS_BLOCK_GROUP_RAID5 |
6000 BTRFS_BLOCK_GROUP_RAID6 |
6001 BTRFS_BLOCK_GROUP_RAID10;
6004 * if they asked for extra copies and this block group
6005 * doesn't provide them, bail. This does allow us to
6006 * fill raid0 from raid1.
6008 if ((flags & extra) && !(block_group->flags & extra))
6013 cached = block_group_cache_done(block_group);
6014 if (unlikely(!cached)) {
6015 found_uncached_bg = true;
6016 ret = cache_block_group(block_group, 0);
6021 if (unlikely(block_group->ro))
6025 * Ok we want to try and use the cluster allocator, so
6029 unsigned long aligned_cluster;
6031 * the refill lock keeps out other
6032 * people trying to start a new cluster
6034 spin_lock(&last_ptr->refill_lock);
6035 used_block_group = last_ptr->block_group;
6036 if (used_block_group != block_group &&
6037 (!used_block_group ||
6038 used_block_group->ro ||
6039 !block_group_bits(used_block_group, flags))) {
6040 used_block_group = block_group;
6041 goto refill_cluster;
6044 if (used_block_group != block_group)
6045 btrfs_get_block_group(used_block_group);
6047 offset = btrfs_alloc_from_cluster(used_block_group,
6048 last_ptr, num_bytes, used_block_group->key.objectid);
6050 /* we have a block, we're done */
6051 spin_unlock(&last_ptr->refill_lock);
6052 trace_btrfs_reserve_extent_cluster(root,
6053 block_group, search_start, num_bytes);
6057 WARN_ON(last_ptr->block_group != used_block_group);
6058 if (used_block_group != block_group) {
6059 btrfs_put_block_group(used_block_group);
6060 used_block_group = block_group;
6063 BUG_ON(used_block_group != block_group);
6064 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6065 * set up a new clusters, so lets just skip it
6066 * and let the allocator find whatever block
6067 * it can find. If we reach this point, we
6068 * will have tried the cluster allocator
6069 * plenty of times and not have found
6070 * anything, so we are likely way too
6071 * fragmented for the clustering stuff to find
6074 * However, if the cluster is taken from the
6075 * current block group, release the cluster
6076 * first, so that we stand a better chance of
6077 * succeeding in the unclustered
6079 if (loop >= LOOP_NO_EMPTY_SIZE &&
6080 last_ptr->block_group != block_group) {
6081 spin_unlock(&last_ptr->refill_lock);
6082 goto unclustered_alloc;
6086 * this cluster didn't work out, free it and
6089 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6091 if (loop >= LOOP_NO_EMPTY_SIZE) {
6092 spin_unlock(&last_ptr->refill_lock);
6093 goto unclustered_alloc;
6096 aligned_cluster = max_t(unsigned long,
6097 empty_cluster + empty_size,
6098 block_group->full_stripe_len);
6100 /* allocate a cluster in this block group */
6101 ret = btrfs_find_space_cluster(trans, root,
6102 block_group, last_ptr,
6103 search_start, num_bytes,
6107 * now pull our allocation out of this
6110 offset = btrfs_alloc_from_cluster(block_group,
6111 last_ptr, num_bytes,
6114 /* we found one, proceed */
6115 spin_unlock(&last_ptr->refill_lock);
6116 trace_btrfs_reserve_extent_cluster(root,
6117 block_group, search_start,
6121 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6122 && !failed_cluster_refill) {
6123 spin_unlock(&last_ptr->refill_lock);
6125 failed_cluster_refill = true;
6126 wait_block_group_cache_progress(block_group,
6127 num_bytes + empty_cluster + empty_size);
6128 goto have_block_group;
6132 * at this point we either didn't find a cluster
6133 * or we weren't able to allocate a block from our
6134 * cluster. Free the cluster we've been trying
6135 * to use, and go to the next block group
6137 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6138 spin_unlock(&last_ptr->refill_lock);
6143 spin_lock(&block_group->free_space_ctl->tree_lock);
6145 block_group->free_space_ctl->free_space <
6146 num_bytes + empty_cluster + empty_size) {
6147 spin_unlock(&block_group->free_space_ctl->tree_lock);
6150 spin_unlock(&block_group->free_space_ctl->tree_lock);
6152 offset = btrfs_find_space_for_alloc(block_group, search_start,
6153 num_bytes, empty_size);
6155 * If we didn't find a chunk, and we haven't failed on this
6156 * block group before, and this block group is in the middle of
6157 * caching and we are ok with waiting, then go ahead and wait
6158 * for progress to be made, and set failed_alloc to true.
6160 * If failed_alloc is true then we've already waited on this
6161 * block group once and should move on to the next block group.
6163 if (!offset && !failed_alloc && !cached &&
6164 loop > LOOP_CACHING_NOWAIT) {
6165 wait_block_group_cache_progress(block_group,
6166 num_bytes + empty_size);
6167 failed_alloc = true;
6168 goto have_block_group;
6169 } else if (!offset) {
6171 have_caching_bg = true;
6175 search_start = stripe_align(root, used_block_group,
6178 /* move on to the next group */
6179 if (search_start + num_bytes >
6180 used_block_group->key.objectid + used_block_group->key.offset) {
6181 btrfs_add_free_space(used_block_group, offset, num_bytes);
6185 if (offset < search_start)
6186 btrfs_add_free_space(used_block_group, offset,
6187 search_start - offset);
6188 BUG_ON(offset > search_start);
6190 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
6192 if (ret == -EAGAIN) {
6193 btrfs_add_free_space(used_block_group, offset, num_bytes);
6197 /* we are all good, lets return */
6198 ins->objectid = search_start;
6199 ins->offset = num_bytes;
6201 trace_btrfs_reserve_extent(orig_root, block_group,
6202 search_start, num_bytes);
6203 if (used_block_group != block_group)
6204 btrfs_put_block_group(used_block_group);
6205 btrfs_put_block_group(block_group);
6208 failed_cluster_refill = false;
6209 failed_alloc = false;
6210 BUG_ON(index != get_block_group_index(block_group));
6211 if (used_block_group != block_group)
6212 btrfs_put_block_group(used_block_group);
6213 btrfs_put_block_group(block_group);
6215 up_read(&space_info->groups_sem);
6217 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6220 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6224 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6225 * caching kthreads as we move along
6226 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6227 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6228 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6231 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6234 if (loop == LOOP_ALLOC_CHUNK) {
6235 ret = do_chunk_alloc(trans, root, flags,
6238 * Do not bail out on ENOSPC since we
6239 * can do more things.
6241 if (ret < 0 && ret != -ENOSPC) {
6242 btrfs_abort_transaction(trans,
6248 if (loop == LOOP_NO_EMPTY_SIZE) {
6254 } else if (!ins->objectid) {
6256 } else if (ins->objectid) {
6264 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6265 int dump_block_groups)
6267 struct btrfs_block_group_cache *cache;
6270 spin_lock(&info->lock);
6271 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6272 (unsigned long long)info->flags,
6273 (unsigned long long)(info->total_bytes - info->bytes_used -
6274 info->bytes_pinned - info->bytes_reserved -
6275 info->bytes_readonly),
6276 (info->full) ? "" : "not ");
6277 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6278 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6279 (unsigned long long)info->total_bytes,
6280 (unsigned long long)info->bytes_used,
6281 (unsigned long long)info->bytes_pinned,
6282 (unsigned long long)info->bytes_reserved,
6283 (unsigned long long)info->bytes_may_use,
6284 (unsigned long long)info->bytes_readonly);
6285 spin_unlock(&info->lock);
6287 if (!dump_block_groups)
6290 down_read(&info->groups_sem);
6292 list_for_each_entry(cache, &info->block_groups[index], list) {
6293 spin_lock(&cache->lock);
6294 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6295 (unsigned long long)cache->key.objectid,
6296 (unsigned long long)cache->key.offset,
6297 (unsigned long long)btrfs_block_group_used(&cache->item),
6298 (unsigned long long)cache->pinned,
6299 (unsigned long long)cache->reserved,
6300 cache->ro ? "[readonly]" : "");
6301 btrfs_dump_free_space(cache, bytes);
6302 spin_unlock(&cache->lock);
6304 if (++index < BTRFS_NR_RAID_TYPES)
6306 up_read(&info->groups_sem);
6309 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
6310 struct btrfs_root *root,
6311 u64 num_bytes, u64 min_alloc_size,
6312 u64 empty_size, u64 hint_byte,
6313 struct btrfs_key *ins, int is_data)
6315 bool final_tried = false;
6319 flags = btrfs_get_alloc_profile(root, is_data);
6321 WARN_ON(num_bytes < root->sectorsize);
6322 ret = find_free_extent(trans, root, num_bytes, empty_size,
6323 hint_byte, ins, flags);
6325 if (ret == -ENOSPC) {
6327 num_bytes = num_bytes >> 1;
6328 num_bytes = round_down(num_bytes, root->sectorsize);
6329 num_bytes = max(num_bytes, min_alloc_size);
6330 if (num_bytes == min_alloc_size)
6333 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6334 struct btrfs_space_info *sinfo;
6336 sinfo = __find_space_info(root->fs_info, flags);
6337 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6338 (unsigned long long)flags,
6339 (unsigned long long)num_bytes);
6341 dump_space_info(sinfo, num_bytes, 1);
6345 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6350 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6351 u64 start, u64 len, int pin)
6353 struct btrfs_block_group_cache *cache;
6356 cache = btrfs_lookup_block_group(root->fs_info, start);
6358 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6359 (unsigned long long)start);
6363 if (btrfs_test_opt(root, DISCARD))
6364 ret = btrfs_discard_extent(root, start, len, NULL);
6367 pin_down_extent(root, cache, start, len, 1);
6369 btrfs_add_free_space(cache, start, len);
6370 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6372 btrfs_put_block_group(cache);
6374 trace_btrfs_reserved_extent_free(root, start, len);
6379 int btrfs_free_reserved_extent(struct btrfs_root *root,
6382 return __btrfs_free_reserved_extent(root, start, len, 0);
6385 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6388 return __btrfs_free_reserved_extent(root, start, len, 1);
6391 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6392 struct btrfs_root *root,
6393 u64 parent, u64 root_objectid,
6394 u64 flags, u64 owner, u64 offset,
6395 struct btrfs_key *ins, int ref_mod)
6398 struct btrfs_fs_info *fs_info = root->fs_info;
6399 struct btrfs_extent_item *extent_item;
6400 struct btrfs_extent_inline_ref *iref;
6401 struct btrfs_path *path;
6402 struct extent_buffer *leaf;
6407 type = BTRFS_SHARED_DATA_REF_KEY;
6409 type = BTRFS_EXTENT_DATA_REF_KEY;
6411 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6413 path = btrfs_alloc_path();
6417 path->leave_spinning = 1;
6418 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6421 btrfs_free_path(path);
6425 leaf = path->nodes[0];
6426 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6427 struct btrfs_extent_item);
6428 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6429 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6430 btrfs_set_extent_flags(leaf, extent_item,
6431 flags | BTRFS_EXTENT_FLAG_DATA);
6433 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6434 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6436 struct btrfs_shared_data_ref *ref;
6437 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6438 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6439 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6441 struct btrfs_extent_data_ref *ref;
6442 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6443 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6444 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6445 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6446 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6449 btrfs_mark_buffer_dirty(path->nodes[0]);
6450 btrfs_free_path(path);
6452 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6453 if (ret) { /* -ENOENT, logic error */
6454 btrfs_err(fs_info, "update block group failed for %llu %llu",
6455 (unsigned long long)ins->objectid,
6456 (unsigned long long)ins->offset);
6462 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6463 struct btrfs_root *root,
6464 u64 parent, u64 root_objectid,
6465 u64 flags, struct btrfs_disk_key *key,
6466 int level, struct btrfs_key *ins)
6469 struct btrfs_fs_info *fs_info = root->fs_info;
6470 struct btrfs_extent_item *extent_item;
6471 struct btrfs_tree_block_info *block_info;
6472 struct btrfs_extent_inline_ref *iref;
6473 struct btrfs_path *path;
6474 struct extent_buffer *leaf;
6475 u32 size = sizeof(*extent_item) + sizeof(*iref);
6476 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6479 if (!skinny_metadata)
6480 size += sizeof(*block_info);
6482 path = btrfs_alloc_path();
6486 path->leave_spinning = 1;
6487 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6490 btrfs_free_path(path);
6494 leaf = path->nodes[0];
6495 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6496 struct btrfs_extent_item);
6497 btrfs_set_extent_refs(leaf, extent_item, 1);
6498 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6499 btrfs_set_extent_flags(leaf, extent_item,
6500 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6502 if (skinny_metadata) {
6503 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6505 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6506 btrfs_set_tree_block_key(leaf, block_info, key);
6507 btrfs_set_tree_block_level(leaf, block_info, level);
6508 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6512 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6513 btrfs_set_extent_inline_ref_type(leaf, iref,
6514 BTRFS_SHARED_BLOCK_REF_KEY);
6515 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6517 btrfs_set_extent_inline_ref_type(leaf, iref,
6518 BTRFS_TREE_BLOCK_REF_KEY);
6519 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6522 btrfs_mark_buffer_dirty(leaf);
6523 btrfs_free_path(path);
6525 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6526 if (ret) { /* -ENOENT, logic error */
6527 btrfs_err(fs_info, "update block group failed for %llu %llu",
6528 (unsigned long long)ins->objectid,
6529 (unsigned long long)ins->offset);
6535 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6536 struct btrfs_root *root,
6537 u64 root_objectid, u64 owner,
6538 u64 offset, struct btrfs_key *ins)
6542 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6544 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6546 root_objectid, owner, offset,
6547 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6552 * this is used by the tree logging recovery code. It records that
6553 * an extent has been allocated and makes sure to clear the free
6554 * space cache bits as well
6556 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6557 struct btrfs_root *root,
6558 u64 root_objectid, u64 owner, u64 offset,
6559 struct btrfs_key *ins)
6562 struct btrfs_block_group_cache *block_group;
6563 struct btrfs_caching_control *caching_ctl;
6564 u64 start = ins->objectid;
6565 u64 num_bytes = ins->offset;
6567 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6568 cache_block_group(block_group, 0);
6569 caching_ctl = get_caching_control(block_group);
6572 BUG_ON(!block_group_cache_done(block_group));
6573 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6577 mutex_lock(&caching_ctl->mutex);
6579 if (start >= caching_ctl->progress) {
6580 ret = add_excluded_extent(root, start, num_bytes);
6581 } else if (start + num_bytes <= caching_ctl->progress) {
6582 ret = btrfs_remove_free_space(block_group,
6585 num_bytes = caching_ctl->progress - start;
6586 ret = btrfs_remove_free_space(block_group,
6591 start = caching_ctl->progress;
6592 num_bytes = ins->objectid + ins->offset -
6593 caching_ctl->progress;
6594 ret = add_excluded_extent(root, start, num_bytes);
6597 mutex_unlock(&caching_ctl->mutex);
6598 put_caching_control(caching_ctl);
6603 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6604 RESERVE_ALLOC_NO_ACCOUNT);
6605 BUG_ON(ret); /* logic error */
6606 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6607 0, owner, offset, ins, 1);
6609 btrfs_put_block_group(block_group);
6613 static struct extent_buffer *
6614 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6615 u64 bytenr, u32 blocksize, int level)
6617 struct extent_buffer *buf;
6619 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6621 return ERR_PTR(-ENOMEM);
6622 btrfs_set_header_generation(buf, trans->transid);
6623 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6624 btrfs_tree_lock(buf);
6625 clean_tree_block(trans, root, buf);
6626 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6628 btrfs_set_lock_blocking(buf);
6629 btrfs_set_buffer_uptodate(buf);
6631 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6633 * we allow two log transactions at a time, use different
6634 * EXENT bit to differentiate dirty pages.
6636 if (root->log_transid % 2 == 0)
6637 set_extent_dirty(&root->dirty_log_pages, buf->start,
6638 buf->start + buf->len - 1, GFP_NOFS);
6640 set_extent_new(&root->dirty_log_pages, buf->start,
6641 buf->start + buf->len - 1, GFP_NOFS);
6643 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6644 buf->start + buf->len - 1, GFP_NOFS);
6646 trans->blocks_used++;
6647 /* this returns a buffer locked for blocking */
6651 static struct btrfs_block_rsv *
6652 use_block_rsv(struct btrfs_trans_handle *trans,
6653 struct btrfs_root *root, u32 blocksize)
6655 struct btrfs_block_rsv *block_rsv;
6656 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6658 bool global_updated = false;
6660 block_rsv = get_block_rsv(trans, root);
6662 if (unlikely(block_rsv->size == 0))
6665 ret = block_rsv_use_bytes(block_rsv, blocksize);
6669 if (block_rsv->failfast)
6670 return ERR_PTR(ret);
6672 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
6673 global_updated = true;
6674 update_global_block_rsv(root->fs_info);
6678 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6679 static DEFINE_RATELIMIT_STATE(_rs,
6680 DEFAULT_RATELIMIT_INTERVAL * 10,
6681 /*DEFAULT_RATELIMIT_BURST*/ 1);
6682 if (__ratelimit(&_rs))
6684 "btrfs: block rsv returned %d\n", ret);
6687 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6688 BTRFS_RESERVE_NO_FLUSH);
6692 * If we couldn't reserve metadata bytes try and use some from
6693 * the global reserve if its space type is the same as the global
6696 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
6697 block_rsv->space_info == global_rsv->space_info) {
6698 ret = block_rsv_use_bytes(global_rsv, blocksize);
6702 return ERR_PTR(ret);
6705 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6706 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6708 block_rsv_add_bytes(block_rsv, blocksize, 0);
6709 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6713 * finds a free extent and does all the dirty work required for allocation
6714 * returns the key for the extent through ins, and a tree buffer for
6715 * the first block of the extent through buf.
6717 * returns the tree buffer or NULL.
6719 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6720 struct btrfs_root *root, u32 blocksize,
6721 u64 parent, u64 root_objectid,
6722 struct btrfs_disk_key *key, int level,
6723 u64 hint, u64 empty_size)
6725 struct btrfs_key ins;
6726 struct btrfs_block_rsv *block_rsv;
6727 struct extent_buffer *buf;
6730 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6733 block_rsv = use_block_rsv(trans, root, blocksize);
6734 if (IS_ERR(block_rsv))
6735 return ERR_CAST(block_rsv);
6737 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6738 empty_size, hint, &ins, 0);
6740 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6741 return ERR_PTR(ret);
6744 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6746 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6748 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6750 parent = ins.objectid;
6751 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6755 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6756 struct btrfs_delayed_extent_op *extent_op;
6757 extent_op = btrfs_alloc_delayed_extent_op();
6758 BUG_ON(!extent_op); /* -ENOMEM */
6760 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6762 memset(&extent_op->key, 0, sizeof(extent_op->key));
6763 extent_op->flags_to_set = flags;
6764 if (skinny_metadata)
6765 extent_op->update_key = 0;
6767 extent_op->update_key = 1;
6768 extent_op->update_flags = 1;
6769 extent_op->is_data = 0;
6770 extent_op->level = level;
6772 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6774 ins.offset, parent, root_objectid,
6775 level, BTRFS_ADD_DELAYED_EXTENT,
6777 BUG_ON(ret); /* -ENOMEM */
6782 struct walk_control {
6783 u64 refs[BTRFS_MAX_LEVEL];
6784 u64 flags[BTRFS_MAX_LEVEL];
6785 struct btrfs_key update_progress;
6796 #define DROP_REFERENCE 1
6797 #define UPDATE_BACKREF 2
6799 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6800 struct btrfs_root *root,
6801 struct walk_control *wc,
6802 struct btrfs_path *path)
6810 struct btrfs_key key;
6811 struct extent_buffer *eb;
6816 if (path->slots[wc->level] < wc->reada_slot) {
6817 wc->reada_count = wc->reada_count * 2 / 3;
6818 wc->reada_count = max(wc->reada_count, 2);
6820 wc->reada_count = wc->reada_count * 3 / 2;
6821 wc->reada_count = min_t(int, wc->reada_count,
6822 BTRFS_NODEPTRS_PER_BLOCK(root));
6825 eb = path->nodes[wc->level];
6826 nritems = btrfs_header_nritems(eb);
6827 blocksize = btrfs_level_size(root, wc->level - 1);
6829 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6830 if (nread >= wc->reada_count)
6834 bytenr = btrfs_node_blockptr(eb, slot);
6835 generation = btrfs_node_ptr_generation(eb, slot);
6837 if (slot == path->slots[wc->level])
6840 if (wc->stage == UPDATE_BACKREF &&
6841 generation <= root->root_key.offset)
6844 /* We don't lock the tree block, it's OK to be racy here */
6845 ret = btrfs_lookup_extent_info(trans, root, bytenr,
6846 wc->level - 1, 1, &refs,
6848 /* We don't care about errors in readahead. */
6853 if (wc->stage == DROP_REFERENCE) {
6857 if (wc->level == 1 &&
6858 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6860 if (!wc->update_ref ||
6861 generation <= root->root_key.offset)
6863 btrfs_node_key_to_cpu(eb, &key, slot);
6864 ret = btrfs_comp_cpu_keys(&key,
6865 &wc->update_progress);
6869 if (wc->level == 1 &&
6870 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6874 ret = readahead_tree_block(root, bytenr, blocksize,
6880 wc->reada_slot = slot;
6884 * helper to process tree block while walking down the tree.
6886 * when wc->stage == UPDATE_BACKREF, this function updates
6887 * back refs for pointers in the block.
6889 * NOTE: return value 1 means we should stop walking down.
6891 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6892 struct btrfs_root *root,
6893 struct btrfs_path *path,
6894 struct walk_control *wc, int lookup_info)
6896 int level = wc->level;
6897 struct extent_buffer *eb = path->nodes[level];
6898 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6901 if (wc->stage == UPDATE_BACKREF &&
6902 btrfs_header_owner(eb) != root->root_key.objectid)
6906 * when reference count of tree block is 1, it won't increase
6907 * again. once full backref flag is set, we never clear it.
6910 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6911 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6912 BUG_ON(!path->locks[level]);
6913 ret = btrfs_lookup_extent_info(trans, root,
6914 eb->start, level, 1,
6917 BUG_ON(ret == -ENOMEM);
6920 BUG_ON(wc->refs[level] == 0);
6923 if (wc->stage == DROP_REFERENCE) {
6924 if (wc->refs[level] > 1)
6927 if (path->locks[level] && !wc->keep_locks) {
6928 btrfs_tree_unlock_rw(eb, path->locks[level]);
6929 path->locks[level] = 0;
6934 /* wc->stage == UPDATE_BACKREF */
6935 if (!(wc->flags[level] & flag)) {
6936 BUG_ON(!path->locks[level]);
6937 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6938 BUG_ON(ret); /* -ENOMEM */
6939 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6940 BUG_ON(ret); /* -ENOMEM */
6941 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6943 btrfs_header_level(eb), 0);
6944 BUG_ON(ret); /* -ENOMEM */
6945 wc->flags[level] |= flag;
6949 * the block is shared by multiple trees, so it's not good to
6950 * keep the tree lock
6952 if (path->locks[level] && level > 0) {
6953 btrfs_tree_unlock_rw(eb, path->locks[level]);
6954 path->locks[level] = 0;
6960 * helper to process tree block pointer.
6962 * when wc->stage == DROP_REFERENCE, this function checks
6963 * reference count of the block pointed to. if the block
6964 * is shared and we need update back refs for the subtree
6965 * rooted at the block, this function changes wc->stage to
6966 * UPDATE_BACKREF. if the block is shared and there is no
6967 * need to update back, this function drops the reference
6970 * NOTE: return value 1 means we should stop walking down.
6972 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6973 struct btrfs_root *root,
6974 struct btrfs_path *path,
6975 struct walk_control *wc, int *lookup_info)
6981 struct btrfs_key key;
6982 struct extent_buffer *next;
6983 int level = wc->level;
6987 generation = btrfs_node_ptr_generation(path->nodes[level],
6988 path->slots[level]);
6990 * if the lower level block was created before the snapshot
6991 * was created, we know there is no need to update back refs
6994 if (wc->stage == UPDATE_BACKREF &&
6995 generation <= root->root_key.offset) {
7000 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7001 blocksize = btrfs_level_size(root, level - 1);
7003 next = btrfs_find_tree_block(root, bytenr, blocksize);
7005 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7010 btrfs_tree_lock(next);
7011 btrfs_set_lock_blocking(next);
7013 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7014 &wc->refs[level - 1],
7015 &wc->flags[level - 1]);
7017 btrfs_tree_unlock(next);
7021 if (unlikely(wc->refs[level - 1] == 0)) {
7022 btrfs_err(root->fs_info, "Missing references.");
7027 if (wc->stage == DROP_REFERENCE) {
7028 if (wc->refs[level - 1] > 1) {
7030 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7033 if (!wc->update_ref ||
7034 generation <= root->root_key.offset)
7037 btrfs_node_key_to_cpu(path->nodes[level], &key,
7038 path->slots[level]);
7039 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7043 wc->stage = UPDATE_BACKREF;
7044 wc->shared_level = level - 1;
7048 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7052 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7053 btrfs_tree_unlock(next);
7054 free_extent_buffer(next);
7060 if (reada && level == 1)
7061 reada_walk_down(trans, root, wc, path);
7062 next = read_tree_block(root, bytenr, blocksize, generation);
7063 if (!next || !extent_buffer_uptodate(next)) {
7064 free_extent_buffer(next);
7067 btrfs_tree_lock(next);
7068 btrfs_set_lock_blocking(next);
7072 BUG_ON(level != btrfs_header_level(next));
7073 path->nodes[level] = next;
7074 path->slots[level] = 0;
7075 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7081 wc->refs[level - 1] = 0;
7082 wc->flags[level - 1] = 0;
7083 if (wc->stage == DROP_REFERENCE) {
7084 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7085 parent = path->nodes[level]->start;
7087 BUG_ON(root->root_key.objectid !=
7088 btrfs_header_owner(path->nodes[level]));
7092 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7093 root->root_key.objectid, level - 1, 0, 0);
7094 BUG_ON(ret); /* -ENOMEM */
7096 btrfs_tree_unlock(next);
7097 free_extent_buffer(next);
7103 * helper to process tree block while walking up the tree.
7105 * when wc->stage == DROP_REFERENCE, this function drops
7106 * reference count on the block.
7108 * when wc->stage == UPDATE_BACKREF, this function changes
7109 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7110 * to UPDATE_BACKREF previously while processing the block.
7112 * NOTE: return value 1 means we should stop walking up.
7114 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7115 struct btrfs_root *root,
7116 struct btrfs_path *path,
7117 struct walk_control *wc)
7120 int level = wc->level;
7121 struct extent_buffer *eb = path->nodes[level];
7124 if (wc->stage == UPDATE_BACKREF) {
7125 BUG_ON(wc->shared_level < level);
7126 if (level < wc->shared_level)
7129 ret = find_next_key(path, level + 1, &wc->update_progress);
7133 wc->stage = DROP_REFERENCE;
7134 wc->shared_level = -1;
7135 path->slots[level] = 0;
7138 * check reference count again if the block isn't locked.
7139 * we should start walking down the tree again if reference
7142 if (!path->locks[level]) {
7144 btrfs_tree_lock(eb);
7145 btrfs_set_lock_blocking(eb);
7146 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7148 ret = btrfs_lookup_extent_info(trans, root,
7149 eb->start, level, 1,
7153 btrfs_tree_unlock_rw(eb, path->locks[level]);
7154 path->locks[level] = 0;
7157 BUG_ON(wc->refs[level] == 0);
7158 if (wc->refs[level] == 1) {
7159 btrfs_tree_unlock_rw(eb, path->locks[level]);
7160 path->locks[level] = 0;
7166 /* wc->stage == DROP_REFERENCE */
7167 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7169 if (wc->refs[level] == 1) {
7171 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7172 ret = btrfs_dec_ref(trans, root, eb, 1,
7175 ret = btrfs_dec_ref(trans, root, eb, 0,
7177 BUG_ON(ret); /* -ENOMEM */
7179 /* make block locked assertion in clean_tree_block happy */
7180 if (!path->locks[level] &&
7181 btrfs_header_generation(eb) == trans->transid) {
7182 btrfs_tree_lock(eb);
7183 btrfs_set_lock_blocking(eb);
7184 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7186 clean_tree_block(trans, root, eb);
7189 if (eb == root->node) {
7190 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7193 BUG_ON(root->root_key.objectid !=
7194 btrfs_header_owner(eb));
7196 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7197 parent = path->nodes[level + 1]->start;
7199 BUG_ON(root->root_key.objectid !=
7200 btrfs_header_owner(path->nodes[level + 1]));
7203 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7205 wc->refs[level] = 0;
7206 wc->flags[level] = 0;
7210 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7211 struct btrfs_root *root,
7212 struct btrfs_path *path,
7213 struct walk_control *wc)
7215 int level = wc->level;
7216 int lookup_info = 1;
7219 while (level >= 0) {
7220 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7227 if (path->slots[level] >=
7228 btrfs_header_nritems(path->nodes[level]))
7231 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7233 path->slots[level]++;
7242 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7243 struct btrfs_root *root,
7244 struct btrfs_path *path,
7245 struct walk_control *wc, int max_level)
7247 int level = wc->level;
7250 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7251 while (level < max_level && path->nodes[level]) {
7253 if (path->slots[level] + 1 <
7254 btrfs_header_nritems(path->nodes[level])) {
7255 path->slots[level]++;
7258 ret = walk_up_proc(trans, root, path, wc);
7262 if (path->locks[level]) {
7263 btrfs_tree_unlock_rw(path->nodes[level],
7264 path->locks[level]);
7265 path->locks[level] = 0;
7267 free_extent_buffer(path->nodes[level]);
7268 path->nodes[level] = NULL;
7276 * drop a subvolume tree.
7278 * this function traverses the tree freeing any blocks that only
7279 * referenced by the tree.
7281 * when a shared tree block is found. this function decreases its
7282 * reference count by one. if update_ref is true, this function
7283 * also make sure backrefs for the shared block and all lower level
7284 * blocks are properly updated.
7286 * If called with for_reloc == 0, may exit early with -EAGAIN
7288 int btrfs_drop_snapshot(struct btrfs_root *root,
7289 struct btrfs_block_rsv *block_rsv, int update_ref,
7292 struct btrfs_path *path;
7293 struct btrfs_trans_handle *trans;
7294 struct btrfs_root *tree_root = root->fs_info->tree_root;
7295 struct btrfs_root_item *root_item = &root->root_item;
7296 struct walk_control *wc;
7297 struct btrfs_key key;
7302 path = btrfs_alloc_path();
7308 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7310 btrfs_free_path(path);
7315 trans = btrfs_start_transaction(tree_root, 0);
7316 if (IS_ERR(trans)) {
7317 err = PTR_ERR(trans);
7322 trans->block_rsv = block_rsv;
7324 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7325 level = btrfs_header_level(root->node);
7326 path->nodes[level] = btrfs_lock_root_node(root);
7327 btrfs_set_lock_blocking(path->nodes[level]);
7328 path->slots[level] = 0;
7329 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7330 memset(&wc->update_progress, 0,
7331 sizeof(wc->update_progress));
7333 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7334 memcpy(&wc->update_progress, &key,
7335 sizeof(wc->update_progress));
7337 level = root_item->drop_level;
7339 path->lowest_level = level;
7340 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7341 path->lowest_level = 0;
7349 * unlock our path, this is safe because only this
7350 * function is allowed to delete this snapshot
7352 btrfs_unlock_up_safe(path, 0);
7354 level = btrfs_header_level(root->node);
7356 btrfs_tree_lock(path->nodes[level]);
7357 btrfs_set_lock_blocking(path->nodes[level]);
7359 ret = btrfs_lookup_extent_info(trans, root,
7360 path->nodes[level]->start,
7361 level, 1, &wc->refs[level],
7367 BUG_ON(wc->refs[level] == 0);
7369 if (level == root_item->drop_level)
7372 btrfs_tree_unlock(path->nodes[level]);
7373 WARN_ON(wc->refs[level] != 1);
7379 wc->shared_level = -1;
7380 wc->stage = DROP_REFERENCE;
7381 wc->update_ref = update_ref;
7383 wc->for_reloc = for_reloc;
7384 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7387 if (!for_reloc && btrfs_fs_closing(root->fs_info)) {
7388 pr_debug("btrfs: drop snapshot early exit\n");
7393 ret = walk_down_tree(trans, root, path, wc);
7399 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7406 BUG_ON(wc->stage != DROP_REFERENCE);
7410 if (wc->stage == DROP_REFERENCE) {
7412 btrfs_node_key(path->nodes[level],
7413 &root_item->drop_progress,
7414 path->slots[level]);
7415 root_item->drop_level = level;
7418 BUG_ON(wc->level == 0);
7419 if (btrfs_should_end_transaction(trans, tree_root)) {
7420 ret = btrfs_update_root(trans, tree_root,
7424 btrfs_abort_transaction(trans, tree_root, ret);
7429 btrfs_end_transaction_throttle(trans, tree_root);
7430 trans = btrfs_start_transaction(tree_root, 0);
7431 if (IS_ERR(trans)) {
7432 err = PTR_ERR(trans);
7436 trans->block_rsv = block_rsv;
7439 btrfs_release_path(path);
7443 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7445 btrfs_abort_transaction(trans, tree_root, ret);
7449 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7450 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7453 btrfs_abort_transaction(trans, tree_root, ret);
7456 } else if (ret > 0) {
7457 /* if we fail to delete the orphan item this time
7458 * around, it'll get picked up the next time.
7460 * The most common failure here is just -ENOENT.
7462 btrfs_del_orphan_item(trans, tree_root,
7463 root->root_key.objectid);
7467 if (root->in_radix) {
7468 btrfs_free_fs_root(tree_root->fs_info, root);
7470 free_extent_buffer(root->node);
7471 free_extent_buffer(root->commit_root);
7475 btrfs_end_transaction_throttle(trans, tree_root);
7478 btrfs_free_path(path);
7481 btrfs_std_error(root->fs_info, err);
7486 * drop subtree rooted at tree block 'node'.
7488 * NOTE: this function will unlock and release tree block 'node'
7489 * only used by relocation code
7491 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7492 struct btrfs_root *root,
7493 struct extent_buffer *node,
7494 struct extent_buffer *parent)
7496 struct btrfs_path *path;
7497 struct walk_control *wc;
7503 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7505 path = btrfs_alloc_path();
7509 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7511 btrfs_free_path(path);
7515 btrfs_assert_tree_locked(parent);
7516 parent_level = btrfs_header_level(parent);
7517 extent_buffer_get(parent);
7518 path->nodes[parent_level] = parent;
7519 path->slots[parent_level] = btrfs_header_nritems(parent);
7521 btrfs_assert_tree_locked(node);
7522 level = btrfs_header_level(node);
7523 path->nodes[level] = node;
7524 path->slots[level] = 0;
7525 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7527 wc->refs[parent_level] = 1;
7528 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7530 wc->shared_level = -1;
7531 wc->stage = DROP_REFERENCE;
7535 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7538 wret = walk_down_tree(trans, root, path, wc);
7544 wret = walk_up_tree(trans, root, path, wc, parent_level);
7552 btrfs_free_path(path);
7556 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7562 * if restripe for this chunk_type is on pick target profile and
7563 * return, otherwise do the usual balance
7565 stripped = get_restripe_target(root->fs_info, flags);
7567 return extended_to_chunk(stripped);
7570 * we add in the count of missing devices because we want
7571 * to make sure that any RAID levels on a degraded FS
7572 * continue to be honored.
7574 num_devices = root->fs_info->fs_devices->rw_devices +
7575 root->fs_info->fs_devices->missing_devices;
7577 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7578 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7579 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7581 if (num_devices == 1) {
7582 stripped |= BTRFS_BLOCK_GROUP_DUP;
7583 stripped = flags & ~stripped;
7585 /* turn raid0 into single device chunks */
7586 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7589 /* turn mirroring into duplication */
7590 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7591 BTRFS_BLOCK_GROUP_RAID10))
7592 return stripped | BTRFS_BLOCK_GROUP_DUP;
7594 /* they already had raid on here, just return */
7595 if (flags & stripped)
7598 stripped |= BTRFS_BLOCK_GROUP_DUP;
7599 stripped = flags & ~stripped;
7601 /* switch duplicated blocks with raid1 */
7602 if (flags & BTRFS_BLOCK_GROUP_DUP)
7603 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7605 /* this is drive concat, leave it alone */
7611 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7613 struct btrfs_space_info *sinfo = cache->space_info;
7615 u64 min_allocable_bytes;
7620 * We need some metadata space and system metadata space for
7621 * allocating chunks in some corner cases until we force to set
7622 * it to be readonly.
7625 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7627 min_allocable_bytes = 1 * 1024 * 1024;
7629 min_allocable_bytes = 0;
7631 spin_lock(&sinfo->lock);
7632 spin_lock(&cache->lock);
7639 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7640 cache->bytes_super - btrfs_block_group_used(&cache->item);
7642 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7643 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7644 min_allocable_bytes <= sinfo->total_bytes) {
7645 sinfo->bytes_readonly += num_bytes;
7650 spin_unlock(&cache->lock);
7651 spin_unlock(&sinfo->lock);
7655 int btrfs_set_block_group_ro(struct btrfs_root *root,
7656 struct btrfs_block_group_cache *cache)
7659 struct btrfs_trans_handle *trans;
7665 trans = btrfs_join_transaction(root);
7667 return PTR_ERR(trans);
7669 alloc_flags = update_block_group_flags(root, cache->flags);
7670 if (alloc_flags != cache->flags) {
7671 ret = do_chunk_alloc(trans, root, alloc_flags,
7677 ret = set_block_group_ro(cache, 0);
7680 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7681 ret = do_chunk_alloc(trans, root, alloc_flags,
7685 ret = set_block_group_ro(cache, 0);
7687 btrfs_end_transaction(trans, root);
7691 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7692 struct btrfs_root *root, u64 type)
7694 u64 alloc_flags = get_alloc_profile(root, type);
7695 return do_chunk_alloc(trans, root, alloc_flags,
7700 * helper to account the unused space of all the readonly block group in the
7701 * list. takes mirrors into account.
7703 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7705 struct btrfs_block_group_cache *block_group;
7709 list_for_each_entry(block_group, groups_list, list) {
7710 spin_lock(&block_group->lock);
7712 if (!block_group->ro) {
7713 spin_unlock(&block_group->lock);
7717 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7718 BTRFS_BLOCK_GROUP_RAID10 |
7719 BTRFS_BLOCK_GROUP_DUP))
7724 free_bytes += (block_group->key.offset -
7725 btrfs_block_group_used(&block_group->item)) *
7728 spin_unlock(&block_group->lock);
7735 * helper to account the unused space of all the readonly block group in the
7736 * space_info. takes mirrors into account.
7738 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7743 spin_lock(&sinfo->lock);
7745 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7746 if (!list_empty(&sinfo->block_groups[i]))
7747 free_bytes += __btrfs_get_ro_block_group_free_space(
7748 &sinfo->block_groups[i]);
7750 spin_unlock(&sinfo->lock);
7755 void btrfs_set_block_group_rw(struct btrfs_root *root,
7756 struct btrfs_block_group_cache *cache)
7758 struct btrfs_space_info *sinfo = cache->space_info;
7763 spin_lock(&sinfo->lock);
7764 spin_lock(&cache->lock);
7765 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7766 cache->bytes_super - btrfs_block_group_used(&cache->item);
7767 sinfo->bytes_readonly -= num_bytes;
7769 spin_unlock(&cache->lock);
7770 spin_unlock(&sinfo->lock);
7774 * checks to see if its even possible to relocate this block group.
7776 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7777 * ok to go ahead and try.
7779 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7781 struct btrfs_block_group_cache *block_group;
7782 struct btrfs_space_info *space_info;
7783 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7784 struct btrfs_device *device;
7793 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7795 /* odd, couldn't find the block group, leave it alone */
7799 min_free = btrfs_block_group_used(&block_group->item);
7801 /* no bytes used, we're good */
7805 space_info = block_group->space_info;
7806 spin_lock(&space_info->lock);
7808 full = space_info->full;
7811 * if this is the last block group we have in this space, we can't
7812 * relocate it unless we're able to allocate a new chunk below.
7814 * Otherwise, we need to make sure we have room in the space to handle
7815 * all of the extents from this block group. If we can, we're good
7817 if ((space_info->total_bytes != block_group->key.offset) &&
7818 (space_info->bytes_used + space_info->bytes_reserved +
7819 space_info->bytes_pinned + space_info->bytes_readonly +
7820 min_free < space_info->total_bytes)) {
7821 spin_unlock(&space_info->lock);
7824 spin_unlock(&space_info->lock);
7827 * ok we don't have enough space, but maybe we have free space on our
7828 * devices to allocate new chunks for relocation, so loop through our
7829 * alloc devices and guess if we have enough space. if this block
7830 * group is going to be restriped, run checks against the target
7831 * profile instead of the current one.
7843 target = get_restripe_target(root->fs_info, block_group->flags);
7845 index = __get_raid_index(extended_to_chunk(target));
7848 * this is just a balance, so if we were marked as full
7849 * we know there is no space for a new chunk
7854 index = get_block_group_index(block_group);
7857 if (index == BTRFS_RAID_RAID10) {
7861 } else if (index == BTRFS_RAID_RAID1) {
7863 } else if (index == BTRFS_RAID_DUP) {
7866 } else if (index == BTRFS_RAID_RAID0) {
7867 dev_min = fs_devices->rw_devices;
7868 do_div(min_free, dev_min);
7871 mutex_lock(&root->fs_info->chunk_mutex);
7872 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7876 * check to make sure we can actually find a chunk with enough
7877 * space to fit our block group in.
7879 if (device->total_bytes > device->bytes_used + min_free &&
7880 !device->is_tgtdev_for_dev_replace) {
7881 ret = find_free_dev_extent(device, min_free,
7886 if (dev_nr >= dev_min)
7892 mutex_unlock(&root->fs_info->chunk_mutex);
7894 btrfs_put_block_group(block_group);
7898 static int find_first_block_group(struct btrfs_root *root,
7899 struct btrfs_path *path, struct btrfs_key *key)
7902 struct btrfs_key found_key;
7903 struct extent_buffer *leaf;
7906 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7911 slot = path->slots[0];
7912 leaf = path->nodes[0];
7913 if (slot >= btrfs_header_nritems(leaf)) {
7914 ret = btrfs_next_leaf(root, path);
7921 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7923 if (found_key.objectid >= key->objectid &&
7924 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7934 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7936 struct btrfs_block_group_cache *block_group;
7940 struct inode *inode;
7942 block_group = btrfs_lookup_first_block_group(info, last);
7943 while (block_group) {
7944 spin_lock(&block_group->lock);
7945 if (block_group->iref)
7947 spin_unlock(&block_group->lock);
7948 block_group = next_block_group(info->tree_root,
7958 inode = block_group->inode;
7959 block_group->iref = 0;
7960 block_group->inode = NULL;
7961 spin_unlock(&block_group->lock);
7963 last = block_group->key.objectid + block_group->key.offset;
7964 btrfs_put_block_group(block_group);
7968 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7970 struct btrfs_block_group_cache *block_group;
7971 struct btrfs_space_info *space_info;
7972 struct btrfs_caching_control *caching_ctl;
7975 down_write(&info->extent_commit_sem);
7976 while (!list_empty(&info->caching_block_groups)) {
7977 caching_ctl = list_entry(info->caching_block_groups.next,
7978 struct btrfs_caching_control, list);
7979 list_del(&caching_ctl->list);
7980 put_caching_control(caching_ctl);
7982 up_write(&info->extent_commit_sem);
7984 spin_lock(&info->block_group_cache_lock);
7985 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7986 block_group = rb_entry(n, struct btrfs_block_group_cache,
7988 rb_erase(&block_group->cache_node,
7989 &info->block_group_cache_tree);
7990 spin_unlock(&info->block_group_cache_lock);
7992 down_write(&block_group->space_info->groups_sem);
7993 list_del(&block_group->list);
7994 up_write(&block_group->space_info->groups_sem);
7996 if (block_group->cached == BTRFS_CACHE_STARTED)
7997 wait_block_group_cache_done(block_group);
8000 * We haven't cached this block group, which means we could
8001 * possibly have excluded extents on this block group.
8003 if (block_group->cached == BTRFS_CACHE_NO)
8004 free_excluded_extents(info->extent_root, block_group);
8006 btrfs_remove_free_space_cache(block_group);
8007 btrfs_put_block_group(block_group);
8009 spin_lock(&info->block_group_cache_lock);
8011 spin_unlock(&info->block_group_cache_lock);
8013 /* now that all the block groups are freed, go through and
8014 * free all the space_info structs. This is only called during
8015 * the final stages of unmount, and so we know nobody is
8016 * using them. We call synchronize_rcu() once before we start,
8017 * just to be on the safe side.
8021 release_global_block_rsv(info);
8023 while(!list_empty(&info->space_info)) {
8024 space_info = list_entry(info->space_info.next,
8025 struct btrfs_space_info,
8027 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8028 if (space_info->bytes_pinned > 0 ||
8029 space_info->bytes_reserved > 0 ||
8030 space_info->bytes_may_use > 0) {
8032 dump_space_info(space_info, 0, 0);
8035 list_del(&space_info->list);
8041 static void __link_block_group(struct btrfs_space_info *space_info,
8042 struct btrfs_block_group_cache *cache)
8044 int index = get_block_group_index(cache);
8046 down_write(&space_info->groups_sem);
8047 list_add_tail(&cache->list, &space_info->block_groups[index]);
8048 up_write(&space_info->groups_sem);
8051 int btrfs_read_block_groups(struct btrfs_root *root)
8053 struct btrfs_path *path;
8055 struct btrfs_block_group_cache *cache;
8056 struct btrfs_fs_info *info = root->fs_info;
8057 struct btrfs_space_info *space_info;
8058 struct btrfs_key key;
8059 struct btrfs_key found_key;
8060 struct extent_buffer *leaf;
8064 root = info->extent_root;
8067 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8068 path = btrfs_alloc_path();
8073 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8074 if (btrfs_test_opt(root, SPACE_CACHE) &&
8075 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8077 if (btrfs_test_opt(root, CLEAR_CACHE))
8081 ret = find_first_block_group(root, path, &key);
8086 leaf = path->nodes[0];
8087 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8088 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8093 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8095 if (!cache->free_space_ctl) {
8101 atomic_set(&cache->count, 1);
8102 spin_lock_init(&cache->lock);
8103 cache->fs_info = info;
8104 INIT_LIST_HEAD(&cache->list);
8105 INIT_LIST_HEAD(&cache->cluster_list);
8109 * When we mount with old space cache, we need to
8110 * set BTRFS_DC_CLEAR and set dirty flag.
8112 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8113 * truncate the old free space cache inode and
8115 * b) Setting 'dirty flag' makes sure that we flush
8116 * the new space cache info onto disk.
8118 cache->disk_cache_state = BTRFS_DC_CLEAR;
8119 if (btrfs_test_opt(root, SPACE_CACHE))
8123 read_extent_buffer(leaf, &cache->item,
8124 btrfs_item_ptr_offset(leaf, path->slots[0]),
8125 sizeof(cache->item));
8126 memcpy(&cache->key, &found_key, sizeof(found_key));
8128 key.objectid = found_key.objectid + found_key.offset;
8129 btrfs_release_path(path);
8130 cache->flags = btrfs_block_group_flags(&cache->item);
8131 cache->sectorsize = root->sectorsize;
8132 cache->full_stripe_len = btrfs_full_stripe_len(root,
8133 &root->fs_info->mapping_tree,
8134 found_key.objectid);
8135 btrfs_init_free_space_ctl(cache);
8138 * We need to exclude the super stripes now so that the space
8139 * info has super bytes accounted for, otherwise we'll think
8140 * we have more space than we actually do.
8142 ret = exclude_super_stripes(root, cache);
8145 * We may have excluded something, so call this just in
8148 free_excluded_extents(root, cache);
8149 kfree(cache->free_space_ctl);
8155 * check for two cases, either we are full, and therefore
8156 * don't need to bother with the caching work since we won't
8157 * find any space, or we are empty, and we can just add all
8158 * the space in and be done with it. This saves us _alot_ of
8159 * time, particularly in the full case.
8161 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8162 cache->last_byte_to_unpin = (u64)-1;
8163 cache->cached = BTRFS_CACHE_FINISHED;
8164 free_excluded_extents(root, cache);
8165 } else if (btrfs_block_group_used(&cache->item) == 0) {
8166 cache->last_byte_to_unpin = (u64)-1;
8167 cache->cached = BTRFS_CACHE_FINISHED;
8168 add_new_free_space(cache, root->fs_info,
8170 found_key.objectid +
8172 free_excluded_extents(root, cache);
8175 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8177 btrfs_remove_free_space_cache(cache);
8178 btrfs_put_block_group(cache);
8182 ret = update_space_info(info, cache->flags, found_key.offset,
8183 btrfs_block_group_used(&cache->item),
8186 btrfs_remove_free_space_cache(cache);
8187 spin_lock(&info->block_group_cache_lock);
8188 rb_erase(&cache->cache_node,
8189 &info->block_group_cache_tree);
8190 spin_unlock(&info->block_group_cache_lock);
8191 btrfs_put_block_group(cache);
8195 cache->space_info = space_info;
8196 spin_lock(&cache->space_info->lock);
8197 cache->space_info->bytes_readonly += cache->bytes_super;
8198 spin_unlock(&cache->space_info->lock);
8200 __link_block_group(space_info, cache);
8202 set_avail_alloc_bits(root->fs_info, cache->flags);
8203 if (btrfs_chunk_readonly(root, cache->key.objectid))
8204 set_block_group_ro(cache, 1);
8207 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8208 if (!(get_alloc_profile(root, space_info->flags) &
8209 (BTRFS_BLOCK_GROUP_RAID10 |
8210 BTRFS_BLOCK_GROUP_RAID1 |
8211 BTRFS_BLOCK_GROUP_RAID5 |
8212 BTRFS_BLOCK_GROUP_RAID6 |
8213 BTRFS_BLOCK_GROUP_DUP)))
8216 * avoid allocating from un-mirrored block group if there are
8217 * mirrored block groups.
8219 list_for_each_entry(cache, &space_info->block_groups[3], list)
8220 set_block_group_ro(cache, 1);
8221 list_for_each_entry(cache, &space_info->block_groups[4], list)
8222 set_block_group_ro(cache, 1);
8225 init_global_block_rsv(info);
8228 btrfs_free_path(path);
8232 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8233 struct btrfs_root *root)
8235 struct btrfs_block_group_cache *block_group, *tmp;
8236 struct btrfs_root *extent_root = root->fs_info->extent_root;
8237 struct btrfs_block_group_item item;
8238 struct btrfs_key key;
8241 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8243 list_del_init(&block_group->new_bg_list);
8248 spin_lock(&block_group->lock);
8249 memcpy(&item, &block_group->item, sizeof(item));
8250 memcpy(&key, &block_group->key, sizeof(key));
8251 spin_unlock(&block_group->lock);
8253 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8256 btrfs_abort_transaction(trans, extent_root, ret);
8260 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8261 struct btrfs_root *root, u64 bytes_used,
8262 u64 type, u64 chunk_objectid, u64 chunk_offset,
8266 struct btrfs_root *extent_root;
8267 struct btrfs_block_group_cache *cache;
8269 extent_root = root->fs_info->extent_root;
8271 root->fs_info->last_trans_log_full_commit = trans->transid;
8273 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8276 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8278 if (!cache->free_space_ctl) {
8283 cache->key.objectid = chunk_offset;
8284 cache->key.offset = size;
8285 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8286 cache->sectorsize = root->sectorsize;
8287 cache->fs_info = root->fs_info;
8288 cache->full_stripe_len = btrfs_full_stripe_len(root,
8289 &root->fs_info->mapping_tree,
8292 atomic_set(&cache->count, 1);
8293 spin_lock_init(&cache->lock);
8294 INIT_LIST_HEAD(&cache->list);
8295 INIT_LIST_HEAD(&cache->cluster_list);
8296 INIT_LIST_HEAD(&cache->new_bg_list);
8298 btrfs_init_free_space_ctl(cache);
8300 btrfs_set_block_group_used(&cache->item, bytes_used);
8301 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8302 cache->flags = type;
8303 btrfs_set_block_group_flags(&cache->item, type);
8305 cache->last_byte_to_unpin = (u64)-1;
8306 cache->cached = BTRFS_CACHE_FINISHED;
8307 ret = exclude_super_stripes(root, cache);
8310 * We may have excluded something, so call this just in
8313 free_excluded_extents(root, cache);
8314 kfree(cache->free_space_ctl);
8319 add_new_free_space(cache, root->fs_info, chunk_offset,
8320 chunk_offset + size);
8322 free_excluded_extents(root, cache);
8324 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8326 btrfs_remove_free_space_cache(cache);
8327 btrfs_put_block_group(cache);
8331 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8332 &cache->space_info);
8334 btrfs_remove_free_space_cache(cache);
8335 spin_lock(&root->fs_info->block_group_cache_lock);
8336 rb_erase(&cache->cache_node,
8337 &root->fs_info->block_group_cache_tree);
8338 spin_unlock(&root->fs_info->block_group_cache_lock);
8339 btrfs_put_block_group(cache);
8342 update_global_block_rsv(root->fs_info);
8344 spin_lock(&cache->space_info->lock);
8345 cache->space_info->bytes_readonly += cache->bytes_super;
8346 spin_unlock(&cache->space_info->lock);
8348 __link_block_group(cache->space_info, cache);
8350 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8352 set_avail_alloc_bits(extent_root->fs_info, type);
8357 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8359 u64 extra_flags = chunk_to_extended(flags) &
8360 BTRFS_EXTENDED_PROFILE_MASK;
8362 write_seqlock(&fs_info->profiles_lock);
8363 if (flags & BTRFS_BLOCK_GROUP_DATA)
8364 fs_info->avail_data_alloc_bits &= ~extra_flags;
8365 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8366 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8367 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8368 fs_info->avail_system_alloc_bits &= ~extra_flags;
8369 write_sequnlock(&fs_info->profiles_lock);
8372 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8373 struct btrfs_root *root, u64 group_start)
8375 struct btrfs_path *path;
8376 struct btrfs_block_group_cache *block_group;
8377 struct btrfs_free_cluster *cluster;
8378 struct btrfs_root *tree_root = root->fs_info->tree_root;
8379 struct btrfs_key key;
8380 struct inode *inode;
8385 root = root->fs_info->extent_root;
8387 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8388 BUG_ON(!block_group);
8389 BUG_ON(!block_group->ro);
8392 * Free the reserved super bytes from this block group before
8395 free_excluded_extents(root, block_group);
8397 memcpy(&key, &block_group->key, sizeof(key));
8398 index = get_block_group_index(block_group);
8399 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8400 BTRFS_BLOCK_GROUP_RAID1 |
8401 BTRFS_BLOCK_GROUP_RAID10))
8406 /* make sure this block group isn't part of an allocation cluster */
8407 cluster = &root->fs_info->data_alloc_cluster;
8408 spin_lock(&cluster->refill_lock);
8409 btrfs_return_cluster_to_free_space(block_group, cluster);
8410 spin_unlock(&cluster->refill_lock);
8413 * make sure this block group isn't part of a metadata
8414 * allocation cluster
8416 cluster = &root->fs_info->meta_alloc_cluster;
8417 spin_lock(&cluster->refill_lock);
8418 btrfs_return_cluster_to_free_space(block_group, cluster);
8419 spin_unlock(&cluster->refill_lock);
8421 path = btrfs_alloc_path();
8427 inode = lookup_free_space_inode(tree_root, block_group, path);
8428 if (!IS_ERR(inode)) {
8429 ret = btrfs_orphan_add(trans, inode);
8431 btrfs_add_delayed_iput(inode);
8435 /* One for the block groups ref */
8436 spin_lock(&block_group->lock);
8437 if (block_group->iref) {
8438 block_group->iref = 0;
8439 block_group->inode = NULL;
8440 spin_unlock(&block_group->lock);
8443 spin_unlock(&block_group->lock);
8445 /* One for our lookup ref */
8446 btrfs_add_delayed_iput(inode);
8449 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8450 key.offset = block_group->key.objectid;
8453 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8457 btrfs_release_path(path);
8459 ret = btrfs_del_item(trans, tree_root, path);
8462 btrfs_release_path(path);
8465 spin_lock(&root->fs_info->block_group_cache_lock);
8466 rb_erase(&block_group->cache_node,
8467 &root->fs_info->block_group_cache_tree);
8469 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8470 root->fs_info->first_logical_byte = (u64)-1;
8471 spin_unlock(&root->fs_info->block_group_cache_lock);
8473 down_write(&block_group->space_info->groups_sem);
8475 * we must use list_del_init so people can check to see if they
8476 * are still on the list after taking the semaphore
8478 list_del_init(&block_group->list);
8479 if (list_empty(&block_group->space_info->block_groups[index]))
8480 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8481 up_write(&block_group->space_info->groups_sem);
8483 if (block_group->cached == BTRFS_CACHE_STARTED)
8484 wait_block_group_cache_done(block_group);
8486 btrfs_remove_free_space_cache(block_group);
8488 spin_lock(&block_group->space_info->lock);
8489 block_group->space_info->total_bytes -= block_group->key.offset;
8490 block_group->space_info->bytes_readonly -= block_group->key.offset;
8491 block_group->space_info->disk_total -= block_group->key.offset * factor;
8492 spin_unlock(&block_group->space_info->lock);
8494 memcpy(&key, &block_group->key, sizeof(key));
8496 btrfs_clear_space_info_full(root->fs_info);
8498 btrfs_put_block_group(block_group);
8499 btrfs_put_block_group(block_group);
8501 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8507 ret = btrfs_del_item(trans, root, path);
8509 btrfs_free_path(path);
8513 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8515 struct btrfs_space_info *space_info;
8516 struct btrfs_super_block *disk_super;
8522 disk_super = fs_info->super_copy;
8523 if (!btrfs_super_root(disk_super))
8526 features = btrfs_super_incompat_flags(disk_super);
8527 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8530 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8531 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8536 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8537 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8539 flags = BTRFS_BLOCK_GROUP_METADATA;
8540 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8544 flags = BTRFS_BLOCK_GROUP_DATA;
8545 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8551 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8553 return unpin_extent_range(root, start, end);
8556 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8557 u64 num_bytes, u64 *actual_bytes)
8559 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8562 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8564 struct btrfs_fs_info *fs_info = root->fs_info;
8565 struct btrfs_block_group_cache *cache = NULL;
8570 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8574 * try to trim all FS space, our block group may start from non-zero.
8576 if (range->len == total_bytes)
8577 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8579 cache = btrfs_lookup_block_group(fs_info, range->start);
8582 if (cache->key.objectid >= (range->start + range->len)) {
8583 btrfs_put_block_group(cache);
8587 start = max(range->start, cache->key.objectid);
8588 end = min(range->start + range->len,
8589 cache->key.objectid + cache->key.offset);
8591 if (end - start >= range->minlen) {
8592 if (!block_group_cache_done(cache)) {
8593 ret = cache_block_group(cache, 0);
8595 wait_block_group_cache_done(cache);
8597 ret = btrfs_trim_block_group(cache,
8603 trimmed += group_trimmed;
8605 btrfs_put_block_group(cache);
8610 cache = next_block_group(fs_info->tree_root, cache);
8613 range->len = trimmed;